ABSTRACT Title of dissertation: QUANTIFICATION OF THE EFFECTIVENESS OF AGRICULTURAL RIPARIAN BUFFERS TO PROTECT STREAM HEALTH IN MARYLAND'S COASTAL PLAIN AND PIEDMONT REGIONS. Linda Suzann Barker, Doctor of Philosophy, 2004 Dissertation directed by: Assistant Professor Gary K. Felton Department of Biological Resources Engineering This research quantified the effectiveness of agricultural riparian buffers to protect stream health in Maryland's Coastal Plain and Piedmont physiographic regions. Easily-obtainable data were used to develop scientific guidance for buffer management strategies. Three data sets were used: the 1998 University of Maryland Agricultural Buffer Survey, the 1996 Smithsonian Environmental Research Center Water Quality Survey, and the 1995-1997 Maryland Biological Stream Survey. Collectively, these data were used to represent baseflow water quality and landscape conditions in agricultural catchments. A set of landscape characteristics describing the agricultural riparian landscape was developed, from which a classification system for agricultural riparian landscapes was developed. The Agricultural Riparian Classification System can identify a subset of sites for targeted research. Additionally, the distribution of agricultural riparian buffers was characterized. Although over 70% of sites were buffered, because the statewide average buffer width was approximately 49 meters, almost 50% of buffers could not be detected by remotely sensed data with a resolution of 30 meters. Models were developed to predict measures of stream health at a site. Buffers acted differently on instream nitrate concentration, fish IBI (FIBI), benthic IBI (BIBI), and instream physical habitat (PHI). All models indicated that nitrate source terms overshadowed any on-site buffer effects and that Confined Animal Feeding Operations and pastures acted as point sources, overwhelming non- point-source effects. Therefore, livestock best managements practices are critical for the reduction of nitrate to streams. FIBI, BIBI and PHI in the Piedmont region were unaffected by buffer presence, but BIBI and PHI in the Coastal Plain were affected by buffer presence, type and width. Regression tree modeling was able to delineate a range of minimum effective buffer width between 22-38 meters. All measures of stream health in Piedmont systems were controlled by hydrology and geomorphology. Therefore, insofar as buffers can mediate hydrologic effects on flow conditions in a stream, they may indirectly affect FIBI, and BIBI and PHI in Piedmont systems. Because FIBI was not directly affected by buffer presence, use of FIBI to measure success of buffer installation or restoration would give false results. QUANTIFICATION OF THE EFFECTIVENESS OF AGRICULTURAL RIPARIAN BUFFERS TO PROTECT STREAM HEALTH IN MARYLAND'S COASTAL PLAIN AND PIEDMONT REGIONS by Linda Suzann Barker Dissertation submitted to the Faculty of the Graduate School of the University of Maryland, College Park in partial fulfillment of the requirements for the degree of Doctorate of Philosophy 2004 Advisory Committee: Assistant Professor Gary Felton, Chair/Advisor Associate Professor Andrew Baldwin Professor William Lamp Professor Estelle Russek-Cohen Professor Adel Shirmohammadi ii DEDICATION For Daddy. Now you?ll have to address me as Dr. Honey. And Gary We had, perhaps, the perfect mentor-student relationship, and I hope to continue as your professional associate and friend. iii ACKNOWLEDGEMENTS Although there is only one name on a doctoral diploma, a work of this magnitude is necessarily the results of a veritable phalanx of supporters. I would like to thank my parents, William Barker and Doll Watson, for their commitment to my early education, and fostering the attitude that I was capable of learning anything in which I had interest. Daddy - I truly believe that my mathematically based career is the direct result of all those years doing flash cards during the drive to school in the morning! I would like to thank my children, Alan and Edward Robert. Edward, you are my hero. You inspire me to take each day as it comes and make the world a better place with the tools I have at hand, seeking joy that is around and within me always. Alan, our summer doing fieldwork together is one of my best memories. I wish for you the same wonderful lifelong journey of learning that I have known. You are the jewel of my heart. I would like to thank Pat Moran, without whose encouragement and financial support I would not have undertaken this effort. I would like to thank Dr. L. Harold Stevenson of McNeese State University. Your initial encouragement upon my first interest in graduate school, continued support through my Master?s program, and friendship through the years has been one iv of my most treasured academic connections. Just look where that first phone call has led! I would like to thank all my past students. You were the ones who taught me that I was born to teach, and gave me the initial motivation to get my ?union card?! I would like to thank Gustavo Cruz of the Universidad Nacional Aut?noma de Honduras. Your window on the incredible world of the Rio Platano Biosphere piqued my interest in research and directed me to ecosystem analysis. I would like to thank the numerous entities that provided financial support; Society for Rural America, American Water Resources Association, American Women in Science, American Business Women?s Association, Society of Women Engineers, the Graduate School of the University of Maryland as administrators of the Ilene Nagel Travel Grant and the Jacob K. Goldhaber travel grants, and the Department of Biological Resources Engineering. I would like to thank Dr. David Correll, Dr. Tom Jordan and Dr. Don Weller of the Smithsonian Environmental Research Center for the use of their data. I would like to thank Dr. Ron Klauda, Dr. Paul Kazyak and Dan Boward of the Maryland Department of Natural Resources. Without your friendly, open encouragement of a student interested in your data, would I be here? I don?t know. I would like to thank my supervisors at Maryland DNR ? Phil Jones, Harley Speir and especially Harry Hornick. Your support, encouragement and flexibility this past year have been an invaluable help in preventing yet another A.B.D.!!! v I would like to thank the team of professionals (both faculty and staff) at the University of Maryland, who have dedicated their careers to the development of minds. These five years have been unadulterated joy. I will forever carry the ?imprint? of Drs. Andrew Baldwin, William Lamp, Richard McCuen, Margaret Palmer, Karen Prestegaard, Estelle Russek-Cohen, Fred Wheaton, and especially Adel Shirmohammadi. It has been my privilege to study and work with you. vi TABLE OF CONTENTS DEDICATION ...............................................................................................................ii ACKNOWLEDGEMENTS ..........................................................................................iii TABLE OF CONTENTS ..............................................................................................vi LIST OF FIGURES.......................................................................................................xi LIST OF TABLES ......................................................................................................xiii CHAPTER 1 - INTRODUCTION AND LITERATURE REVIEW .............................1 BACKGROUND........................................................................................................1 GOAL OF THE STUDY............................................................................................3 GENERAL BENEFITS OF RIPARIAN BUFFERS .................................................3 ENVIRONMENTAL CONTROLS ON INSTREAM NITRATE CONCENTRATION ..................................................................................................5 MODELS TO PREDICT INSTREAM NITRATE IN AGRICULTURAL AREAS.....................................................................................................................10 ENVIRONMENTAL CONTROLS ON INSTREAM COMMUNITIES ? FISH AND MACROINVERTEBRATES................................................................14 MODELS TO PREDICT INSTREAM COMMUNITIES ? FISH ..........................17 MODELS TO PREDICT INSTREAM COMMUNITIES ? MACROINVERTEBRATES...................................................................................19 THE STATE OF RIPARIAN RESTORATION GUIDANCE - PLACEMENT AND WIDTH ..................................................................................20 RIPARIAN CLASSIFICATION SYSTEMS...........................................................23 MARYLAND?S AGRICULTURAL RIPARIAN BUFFERS - AVAILABLE INFORMATION ..............................................................................25 MBSS 1995-1997 DATA.........................................................................................28 SUMMARY OF LITERATURE FINDINGS AND RESEARCH NEEDS.............31 CITED REFERENCES ............................................................................................33 vii CHAPTER 2 ? CHARACTERIZATION OF AGRICULTURAL RIPARIAN BUFFER LANDSCAPES IN THE COASTAL PLAIN AND PIEDMONT PHYSIOGRAPHIC REGIONS OF MARYLAND ...................................................................................49 ABSTRACT .............................................................................................................49 INTRODUCTION....................................................................................................50 Background...........................................................................................................50 Objectives .............................................................................................................52 METHODS...............................................................................................................53 Overall summary of methods ...............................................................................53 Sample site selection ............................................................................................53 Site characterization .............................................................................................60 Data analysis.........................................................................................................67 RESULTS AND DISCUSSION...............................................................................67 Maryland's agricultural riparian buffers: distribution by type..............................69 Maryland's agricultural riparian buffers: width....................................................70 Maryland's agricultural riparian buffers: distribution of buffer width classes .....71 Maryland's agricultural riparian buffers: fragmentation.......................................72 Maryland's agricultural riparian buffers: vegetation ............................................73 Agricultural riparian landscapes...........................................................................73 Comments on the correlation matrix ....................................................................77 Buffer impact on streambank erosion...................................................................78 SUMMARY .............................................................................................................79 CITED REFERENCES ............................................................................................82 CHAPTER 3 ? CLASSIFICATION OF AGRICULTURAL RIPARIAN ZONE LANDSCAPES IN THE COASTAL PLAIN AND PIEDMONT PHYSIOGRAPHIC REGIONS OF MARYLAND ...................................................................................86 Abstract.....................................................................................................................86 INTRODUCTION....................................................................................................87 Background...........................................................................................................87 Purpose .................................................................................................................90 Objectives .............................................................................................................91 METHODS...............................................................................................................92 Overall summary of methods ...............................................................................92 Data development .................................................................................................93 Cluster analysis and interpretation of results........................................................95 RESULTS AND DISCUSSION...............................................................................98 CONCLUSIONS ....................................................................................................100 CITED REFERENCES ..........................................................................................102 viii CHAPTER 4 - AGRICULTURAL RIPARIAN BUFFER SURVEY ? USE OF TWO DATA SETS: 1995-1997 MBSS AND 1998 UNIVERSITY OF MARYLAND BUFFER SURVEY.........................................................................106 Abstract...................................................................................................................106 INTRODUCTION..................................................................................................107 Background.........................................................................................................107 Purpose ...............................................................................................................108 Objectives ...........................................................................................................108 METHODS.............................................................................................................109 Overall summary of methods .............................................................................109 Description of the data sets.................................................................................109 Differences between the two databases ..............................................................114 Development of modified data sets ....................................................................115 RESULTS AND DISCUSSION.............................................................................116 Comparison of the modified databases...............................................................116 Distribution of buffer types ................................................................................117 Buffer width........................................................................................................119 Usefulness of remotely sensed data....................................................................120 CONCLUSIONS ....................................................................................................122 CITED REFERENCES ..........................................................................................125 CHAPTER 5 ? INITIAL MODELING TO INVESTIGATE THE EFFECTS OF AGRICULTURAL RIPARIAN BUFFERS TO MITIGATE NITRATE STREAM POLLUTION...................................................................................126 INTRODUCTION..................................................................................................126 I. Examination of descriptor variables...................................................................127 METHODS.....................................................................................................127 RESULTS AND DISCUSSION.....................................................................127 II. Modeling Nitrate Change Along a Reach.........................................................134 METHODS.....................................................................................................134 RESULTS AND DISCUSSION.....................................................................135 III. Modeling Nitrate - Land Use Relationships ....................................................138 METHODS.....................................................................................................138 RESULTS AND DISCUSSION.....................................................................139 SUMMARY ...........................................................................................................140 CITED REFERENCES ..........................................................................................141 ix CHAPTER 6 ? MODELING THE EFFECTS OF AGRICULTURAL RIPARIAN BUFFERS TO MITIGATE NITRATE STREAM POLLUTION .................................................................142 Abstract...................................................................................................................142 INTRODUCTION..................................................................................................143 Background.........................................................................................................143 Purpose ...............................................................................................................145 Objectives ...........................................................................................................145 METHODS.............................................................................................................146 Overall summary of methods .............................................................................146 Data development ...............................................................................................146 Modeling.............................................................................................................160 RESULTS AND DISCUSSION.............................................................................164 Graphical data exploration and correlation analysis ..........................................164 Modeling results .................................................................................................172 CONCLUSIONS ....................................................................................................177 CITED REFERENCES ..........................................................................................179 CHAPTER 7 ? USE OF MARYLAND BIOLOGICAL STREAM SURVEY DATA (1995-1997) TO DETERMINE EFFECT OF AGRICULTURAL RIPARIAN BUFFERS ON MEASURES OF BIOLOGICAL STREAM HEALTH.......................................................................182 Abstract...................................................................................................................182 Introduction ............................................................................................................183 Background.........................................................................................................183 Purpose ...............................................................................................................185 Objectives ...........................................................................................................185 METHODS.............................................................................................................186 Description of MBSS database...........................................................................186 Database partitioning..........................................................................................190 Modeling.............................................................................................................193 RESULTS AND DISCUSSION.............................................................................197 Buffer effect on fish IBI .....................................................................................197 Buffer effect on benthic IBI ...............................................................................200 Buffer effect on physical habitat index ..............................................................202 Comparison of analysis techniques. ...................................................................206 CONCLUSIONS ....................................................................................................206 CITED REFERENCES ..........................................................................................208 CHAPTER 8 ? FINAL CONCLUSIONS ................................................................210 CHAPTER 9 ? FUTURE WORK ............................................................................216 x APPENDICES............................................................................................................220 APPENDIX 1: University of Maryland Agricultural Buffer Survey Field Sheet221 APPENDIX 2: University of Maryland Agricultural Buffer Survey Guide to Field and Map Measurements ................................................225 APPENDIX 3: Development of GIS Watershed and Landscape Descriptors ....239 APPENDIX 4: University of Maryland Agricultural Buffer Survey Correlation Matrix for Buffer Survey Data ................................243 APPENDIX 5: SERC* Nitrate Data Development .............................................245 APPENDIX 6: SERC* Data Correlation Matrix.................................................247 APPENDIX 7: MBSS* Data Correlation Matrix ................................................255 APPENDIX 8: SERC* NO3 SAS Modeling.......................................................272 APPENDIX 9: SERC* NO3 S-PLUS Modeling.................................................274 APPENDIX 10: MBSS* NO3 SAS Modeling......................................................279 APPENDIX 11: MBSS* NO3 S-PLUS Modeling................................................282 APPENDIX 12: Fish IBI SAS Modeling ..............................................................297 APPENDIX 13: Fish IBI S-PLUS Modeling ........................................................301 APPENDIX 14: Benthic IBI SAS Modeling.........................................................317 APPENDIX 15: Benthic IBI S-PLUS Modeling...................................................321 APPENDIX 16: PHI SAS Modeling.....................................................................339 APPENDIX 17: PHI S-PLUS Modeling...............................................................343 APPENDIX 18: SAS Programs.............................................................................359 CITED REFERENCES ..............................................................................................360 CURRICULUM VITAE ............................................................................................377 xi LIST OF FIGURES Figure 1. Schematic of the agricultural riparian landscape as defined for the 1998 University of Maryland Agricultural Riparian Buffer Survey. ...54 Figure 2. Map of the University of Maryland 1998 Agricultural Riparian Buffer Survey sample watershed locations superimposed over 16-digit Hydrologic Unit Code watersheds as defined by USGS. .....................55 Figure 3. 1998 University of Maryland Agricultural Riparian Buffer Survey sampled watersheds, streams, forested areas and sample sites. ...........56 Figure 4. Average agricultural riparian buffer widths by type, as determined by the 1998 University of Maryland Agricultural Riparian Buffer Survey..................................................................................................71 Figure 5. Histogram of agricultural riparian buffer widths for forested, grass buffers and all buffer types combined, as determined by the 1998 University of Maryland Agricultural Riparian Buffer Survey. ............72 Figure 6. Statewide and regional land use patterns for the 1998 University of Maryland Agricultural Riparian Buffer Survey watersheds.................74 Figure 7. Agricultural land use patterns, as determined by the 1998 University of Maryland Agricultural Riparian Buffer Survey. ............75 Figure 8. Percent of sites with livestock access to the stream vs. percent of sites with high erosion, as determined by the 1998 University of Maryland Agricultural Riparian Buffer Survey. ...................................................79 Figure 9. Schematic of the agricultural riparian landscape as defined for the 1998 University of Maryland Agricultural Riparian Buffer Survey. ...92 Figure 10. Map of University of Maryland 1998 Agricultural Riparian Buffer Survey sample watershed locations superimposed over 16-digit Hydrologic Unit Code watersheds as defined by USGS. .....................93 Figure 11. Distribution of agricultural riparian buffers by type and physiographic region, as determined by Survey* and MBSS* data. .118 Figure 12. Histogram of agricultural buffer width for all buffer types combined, based on Survey* and MBSS* data....................................................120 Figure 13. Percent of agricultural riparian buffers in the Coastal Plain and Piedmont physiographic regions of Maryland that are narrower than the resolution of remotely sensed data, based on the 1998 University of Maryland Agricultural Riparian Buffer Survey.. ...............................121 Figure 14. Average agricultural riparian buffer width in the Coastal Plain and Piedmont physiographic regions of Maryland, as determined by Survey* and MBSS* data...................................................................122 xii Figure 15. Map of the 1998 University of Maryland Agricultural Riparian Buffer Survey?s German Branch watershed (5,690 ha), with SERC nitrate sampling sites (n = 21), assigned average spring baseflow nitrate values, forest and CAFO areas. ..........................................................131 Figure 16. Map of the 1998 University of Maryland Agricultural Riparian Buffer Survey?s Gunpowder watershed (12,880 ha) with SERC nitrate sampling sites (n = 14), average spring baseflow nitrate values, forest and CAFO areas. ......................................................................132 Figure 17. Change in average spring baseflow nitrate concentration along a reach (ppm/km) vs. estimated average buffer width along the reach, for upstream-downstream sampled segments of the German Branch (n = 4) and Gunpowder (n = 4) watersheds, based on the 1998 University of Maryland Agricultural Riparian Buffer Survey. ................................136 Figure 18. Schematic of the agricultural riparian landscape as defined for the 1998 University of Maryland Agricultural Riparian Buffer Survey. .147 Figure 19. Map of 1998 University of Maryland Agricultural Riparian Buffer Survey watershed locations superimposed over 16-digit Hydrologic Unit Area (HUA) code watersheds as defined by USGS. ..................150 Figure 20. Map of 1998 University of Maryland Agricultural Riparian Buffer Survey German Branch watershed (5,690 ha), with SERC nitrate sampling sites (n = 21), assigned average spring baseflow nitrate values, forest and CAFO areas. ..........................................................151 Figure 21. Map of 1998 University of Maryland Agricultural Riparian Buffer Survey Gunpowder watershed (12,880 ha) with SERC nitrate sampling sites (n = 14), average spring baseflow nitrate values, forest and CAFO areas. ...................................................................................................152 Figure 22. Comparison of land use distribution for agricultural riparian landscapes in Maryland?s Coastal Plain and Piedmont physiographic regions, as determined from MBSS* and SERC* data sets. ..............165 Figure 23. Nitrate (points) vs. percent agriculture (bars) in the site catchment for four data sets with varying degrees of agriculture. .......................167 Figure 24. Instream nitrate vs. percent agriculture in the site contributing watershed, based on MBSS* data. .....................................................168 Figure 25. Instream nitrate for five groups of sites, based on MBSS* data. .......170 Figure 26. Correlation between instream nitrate and buffer width, for all sites, Coastal Plain and Piedmont sites, from MBSS* and SERC* data.....171 Figure 27. Regression tree model predicting instream nitrate (ppm) from MBSS* data (n = 249). .......................................................................174 Figure 28. Regression tree model predicting instream nitrate (ppm) for all sites using SERC* data (n= 35). .................................................................175 Figure 29. ANOVA results for PHI versus physiographic region, stream order, adjacent buffer type, and adjacent agricultural land use (n = 243). ..205 xiii LIST OF TABLES Table 1. Descriptors used to characterize the agricultural riparian landscape in the 1998 University of Maryland Agricultural Riparian Buffer Survey. .64 Table 2. Values of categorical descriptors used to characterize the agricultural riparian landscape in the 1998 University of Maryland Agricultural Riparian Buffer Survey.............................................................................66 Table 3. Statewide and regional values and standard errors for riparian landscape descriptors in the 1998 University of Maryland Agricultural Riparian Buffer Survey. ..........................................................................................68 Table 4. Mean values and standard deviations for riparian landscape descriptors in the 1998 University of Maryland Agricultural Riparian Buffer Survey.......................................................................................................97 Table 5. Number of sites per cluster resulting from the SAS Proc Cluster Procedure. Multiple procedures were performed to produce 4 ? 8 clusters of sites, based on 1998 University of Maryland Agricultural Riparian Buffer Survey data. ....................................................................99 Table 6. Riparian landscape descriptors that differentiated clusters of sites and corresponding descriptor categories, based on data from the 1998 University of Maryland Agricultural Riparian Buffer Survey. .................99 Table 7. Landscape descriptor qualities for clusters of sites produced by the Agricultural Riparian Classification System (ARCS), based on data from the 1998 University of Maryland Agricultural Riparian Buffer Survey.....................................................................................................100 Table 8. Landscape descriptors used to characterize the agricultural riparian landscape in the 1998 University of Maryland Agricultural Riparian Buffer Survey. ........................................................................................111 Table 9. Values of categorical descriptors used to characterize the agricultural riparian landscape in the 1998 University of Maryland Agricultural Riparian Buffer Survey...........................................................................113 Table 10. Comparison of the original 1998 University of Maryland Agricultural Riparian Buffer Survey (Survey), 1995-1997 Maryland Biological Stream Survey (MBSS), modified Survey (Survey*), and modified MBSS (MBSS*) data sets. .....................................................117 Table 11. Least squares means values (and standard deviations) for quantitative landscape descriptors based on mixed models analysis of the significant difference between Coastal Plain and Piedmont sites, based on data from the 1998 University of Maryland Agricultural Riparian Buffer Survey.....................................................................................................128 xiv Table 12. Fisher exact probability indicating significant difference between Coastal Plain and Piedmont values for qualitative landscape descriptors, based on 1998 University of Maryland Agricultural Riparian Buffer Survey data.. ...........................................................................................130 Table 13. Correlation coefficients for landscape descriptors vs. instream nitrate and scale of influence for SERC* data (35 sites from the 1998 University of Maryland agricultural Riparian Buffer Survey that were assigned nitrate values. . .........................................................................133 Table 14. Average values of nitrate decrease along a reach, riparian and land use for sites in the contributing watershed, based on SERC* data...............136 Table 15. Confined Animal Feeding Operation (CAFO) and instream baseflow nitrate information for German Branch and Gunpowder watersheds, based on SERC* data. ............................................................................139 Table 16. Linear regression model predicting instream nitrate at a site using SERC* data of the 1998 University of Maryland Agricultural Riparian Buffer Survey. ........................................................................................140 Table 17. Explanatory variables from the SERC* data set used to develop models predicting instream nitrate. ........................................................149 Table 18. Values for SERC* landscape descriptors: all sites, German Branch and Gunpowder watersheds....................................................................154 Table 19. Riparian landscape descriptors and summary values for MBSS* data. .158 Table 20. Comparison of key parameters of the SERC* and MBSS* data. ..........160 Table 21. Confined Animal Feeding Operation (CAFO) and instream baseflow nitrate information for German Branch and Gunpowder watersheds, as determined from SERC* data. ...........................................................169 Table 22. Instream spring baseflow nitrate values and selected agricultural riparian landscape descriptors in the Coastal Plain and Piedmont physiographic regions of Maryland, based on MBSS* data. .........................................172 Table 23. Comparison of model terms and accuracy for regression tree and multiple linear regression (MLR) models predicting instream spring baseflow nitrate for agricultural riparian sites statewide, Coastal Plain (CP) and Piedmont (Pd) physiographic regions of Maryland, based on data from the SERC* and MBSS* data sets. ................................................................173 Table 24. Comparison of the original 1995-1997 Maryland Biological Stream Survey data (MBSS) and modified 1995-1997 Maryland Biological Stream Survey data (MBSS*). ...............................................................190 Table 25. Criteria and explanatory variables and values for agricultural sites in Coastal Plain and Piedmont physiographic regions of Maryland, based on MBSS* data.............................................................................191 Table 26. Comparison of model terms and accuracy for regression tree and multiple linear regression models predicting fish IBI for agricultural riparian sites in the Coastal Plain and Piedmont physiographic regions of Maryland, based on data from the MBSS* data set. ................................................198 xv Table 27. Comparison of model terms and accuracy for regression tree and multiple linear regression models predicting benthic macroinvertebrate IBI for agricultural riparian sites in the Coastal Plain and Piedmont physiographic regions of Maryland, based on MBSS* data. .........................................201 Table 28. Comparison of model terms and accuracy for regression tree and multiple linear regression models predicting PHI for agricultural riparian sites in Coastal Plain and Piedmont physiographic regions of Maryland, based on MBSS* data.............................................................................................204 1 CHAPTER 1 INTRODUCTION AND LITERATURE REVIEW BACKGROUND Habitat is being altered throughout the world at an alarming rate. Natural regimes are altered by anthropogenic modifications to landscapes. Human-induced changes often alter aquatic environments by changing the amounts and distribution of water, sediment, and nutrients released from the modified landscapes. These changes in turn create responses in physical, chemical, and biological processes in streams and rivers that are ultimately manifest in the biotic community (Imhof et al., 1996). In 1983, concern about worsening water quality in the Chesapeake Bay led to the formation of the Chesapeake Bay Program (CBP), a multi-jurisdictional cooperative effort to restore the nation's largest estuary. The Chesapeake Executive Council set restoration of the Bay's "living resources" as a primary goal in 1987. To achieve that goal, it called for a number of actions to improve water quality, including a 40 percent reduction in the amount of nitrogen and phosphorus entering the Bay (Alliance for the Chesapeake Bay, 1993). Recognition that human alteration of the landscape causes diffuse and significant sources of water pollution has achieved widespread acceptance over the 2 past decade. Accordingly, the general approach to watershed protection has shifted from a focus on point-source pollution control to non-point-source (NPS) pollution control. These efforts emphasize implementation of Best Management Practices (BMPs) for pollution reduction. Agricultural is a significant contributor of NPS nitrogen to the Chesapeake Bay (EPA, 1983). Agricultural riparian buffers are considered important components of these management efforts because of the limitations of other best management practices for NPS pollution control and because they are generally recognized for their abilities to reduce NPS pollution and stabilize stream systems (Chesapeake Bay Program, 1996). Riparian forest buffers (RFB) were singled out in part for their nutrient reduction capabilities, but also because they were reminiscent of the natural landscape in which the stream channel and stream ecosystems of the Bay watershed had evolved. There are several programmatic indications of the trust in agricultural riparian buffers to mitigate pollution in the Chesapeake Bay watershed. In 1994 the Executive Council issued Directive 94-1, a set of goals and actions to increase the focus on riparian stewardship and to enhance efforts to conserve and restore riparian forest buffers (Alliance for the Chesapeake Bay, 1996). In October 1996, the Executive Council of the CBP established a goal of restoring riparian forest buffers on 2010 miles of streams in the Chesapeake Bay watershed by 2010 (Chesapeake Bay Program, 1999). The Conservation Reserve Program (CRP) and its enhancement, the Conservation Reserve Enhancement Program (CREP), were established in Maryland to provide enhanced incentive funds to farmers who restore wetlands and stream 3 buffers on environmentally sensitive lands (Chesapeake Bay Program, 1999). In May of 2002, the CBP published nutrient reduction estimates for various management plans in the Chesapeake Bay watershed. One plan estimated reduction of 62 million pounds of nitrogen per year through "maximum participation" in current voluntary pollution reduction programs (Blankenship, 2002). Projected implementation and effectiveness of agricultural riparian buffers was a major portion of this pollution reduction estimate. GOAL OF THE STUDY The goal of this study is to determine the effectiveness of agricultural riparian buffers to mitigate the effect of agricultural pollution on physical, chemical and biological measures of stream health, using easily- obtainable data. There are numerous government-sponsored data sets that contain a vast amount of information, and can possibly be exploited to answer questions beyond the original scope of the survey. The Maryland Biological Stream Survey data is a prime example: it is a large environmental data set that contains information on riparian settings in Maryland. GENERAL BENEFITS OF RIPARIAN BUFFERS Widespread recognition of riparian areas as unique ecosystems can be traced to the late 1970s. Riparian ecosystems were identified by Ewel (1978) as having two essential characteristics: laterally flowing water that rises and falls at least once within a growing season and a high degree of connectedness with other ecosystems. Since 4 that time, riparian buffers have received much attention for their role in stream ecosystem function, including their capacity to mitigate NPS pollution. Streamside forests are extremely complex ecosystems that help provide optimum food and habitat for stream communities (Vannote et al., 1980; Naiman et al., 1988; Gregory et al., 1991; Naiman and Decamps, 1997; Correll, 2000; Stewart et al., 2000). Whereas energy and material are exchanged between terrestrial and aquatic ecosystems via meteorological, geologic, and biological vectors, the riparian zone forms a permeable, but influential, boundary that serves as a vector interceptor and moderator of sediments, allochthonous nutrients, temperature and bank erosion (Likens and Bormann, 1974). The riparian buffer vegetation impacts stream health in all areas of description ? physical, chemical and biological (Correll, 1996; Tabacchi et al., 1998). Because forested riparian buffers serve as a primary source of nutrients to low-order streams, it can perhaps be said that riparian forest buffers are the single most critical element of these ecosystems. The energy flow that drives these ecosystems is derived from and mediated by the land-based vegetation of the riparian forest buffer. Leaf-fall and woody debris provide the basis of the food chain in the form of detritus. Canopy characteristics can have a strong influence on shredders and detritivores by altering the quality and abundance of detrital inputs (Egglishaw, 1964; Molles, 1982; Corkum, 1992). Overhanging vegetation mediates the amount of sunlight energy reaching the stream, thereby mitigating thermal impacts and limiting growth of instream vegetation. Branches, roots and other tree parts that fall into the stream channel provide shelter for instream inhabitants, retain sediment, and contribute to the maintenance of habitat diversity through the development of riffles and pools. 5 Streamside vegetation also contributes to channel stability and shape by holding bank sediment and reducing erosion during periods of high flow (Ormerod et al., 1993). Offstream riparian vegetation provides additional vegetation that increases the organic content and infiltration capacity of the soil, thereby contributing to denitrifying conditions that reduce nitrogen inputs in groundwater moving to the stream (Lowrance et al., 1984, 1997; Correll, 1996; Barling and Moore, 1994). This offstream vegetation also serves to filter and slow sediment and surface flow moving toward the stream (Osborne and Kovacic, 1993). Thus riparian buffers serve to mediate the energy- and mass- flow interactions between small streams and the extended landscape systems of which they are a part. Careful management of riparian area therefore offers a strategy for buffering streams from a variety of land use impacts while still allowing land use to continue over the broader landscape. ENVIRONMENTAL CONTROLS ON INSTREAM NITRATE CONCENTRATION In a seminal paper on watershed hydrology, Hynes (1975) stated that the behavior of aquatic ecosystems and associated hydrologic characteristics are derived primarily from the basin they drain. The physical and chemical characteristics of a stream are controlled by physical watershed characteristics and the interaction of precipitation, surface runoff, and groundwater with this matrix of watershed characteristics (Larsen et al., 1986; Hughes and Larsen, 1988; McMahon and Harned, 1998). Numerous studies have indicated that land use/cover can play an important 6 role in determining stream water quality. Despite uncertainties in measuring and predicting the effects of agricultural land use, it is clear that agriculture has played an important role in the recent increases in nutrient delivery to coastal waters and the consequent acceleration of eutrophication. Agriculture is considered a major source of nutrient discharge from the watershed of the Chesapeake Bay (EPA, 1983; Alliance for the Chesapeake Bay, 1993). Numerous studies have shown that watersheds with greater proportions of agriculture discharge greater amounts of nitrate-nitrogen in their surface streams (Loehr, 1974; Dillon and Kirchner, 1975; Hill, 1978; Omernik et al., 1981; Beaulac and Teckhow, 1982; US EPA, 1990; Correll, 1983; Peterjohn and Correll, 1984; Roberts and Marsh, 1987; Haith and Shoemaker, 1987; Osborne and Wiley, 1988; Neill, 1989; Correll et al., 1992; Nearing et al., 1993; Tim et al., 1995; Hunsaker and Levine, 1995; Bohlke and Denver, 1995; Taraba et al., 1996; Freifelder et al., 1998; USGS, 1999). Beyond the effect of general agricultural land use, several researchers have found that row-crop agriculture is an important predictor of chemistry variables in groundwater (Correll et al., 1992) and receiving streams (Correll et al., 1994; Richards et al., 1996; Taraba et al., 1996; Jordan et al., 1997a, b, c). In particular, Taraba et al. (1996) found nitrate-nitrogen in discharge water from agricultural watersheds in Kentucky to be linearly related to percent row crops. Jordan et al. (1993, 1997c) found dramatic differences in watershed discharges related to the proportion of cropland. By their estimation, in Coastal Plain settings, cropland discharged 20 kg N /ha/yr, pasture discharged 5.8 kg N /ha/yr and forest discharges 1.4 kg N /ha/yr. In 7 contrast, they found discharges of phosphorus were not correlated with land use. Modeling work performed by Gburek and Folmar (1999) demonstrated that baseflow nitrate concentration could be accurately modeled as a function of agricultural land use categories. Based on a one-time sampling of baseflow in an agricultural watershed of the Valley and Ridge Province, they modeled nitrate suing a multiple linear regression equation as NO3-N (mg/l) = [(1.0)(% forest) +(8.0)( % rotation) +(20.0)( % corn) + (10.0)(% pasture) +(30.0)( % animals)] / 100 (1.1) A relatively small number of researchers found streamwater discharge of nitrate to be poorly correlated with cropland. Lowrance and Leonard (1988) found that nitrate was not correlated to percent of spring-planted row crops, but correlated with land in winter wheat, which is fertilized in late fall or early winter. Jordan and Weller (1996) reported that for larger drainage basins (0.2 - 1.3 x 105 km2), nitrate was poorly correlated with cropland, but highly correlated with a finer analysis considering anthropogenic input of nitrogen from atmospheric deposition, fertilizer application, and nitrogen-fixing crops. Agricultural practices may also contribute to differences in nitrate-nitrogen discharge from agricultural watersheds. Weil et al. (1990) found significantly higher nitrate-nitrogen in groundwater under irrigated Coastal Plain soils treated with poultry manure than under unmanured fields. Angle et al. (1993) found that soil nitrate concentrations were consistently lower under no-tillage fields compared to 8 conventional tillage. These results indicated that the use of no-tillage cultivation might reduce nitrate leaching beyond the crop root zone. Physiographic setting has also been shown to affect differences in nitrate- nitrogen discharge from agricultural watersheds McMahon and Lloyd (1995) found that streams in Coastal Plain agricultural basins had higher nitrogen and phosphorus concentrations than non-Coastal Plain agricultural basins. However, fields and planted acreages were larger, and absolute amounts of fertilizers used were larger. In a later study, McMahon and Harned (1998) reported that agricultural basins in the Coastal Plain with medium drained soils had the highest nitrogen and phosphorus concentrations. Jordan et al. (1997a) also found dramatic differences in watershed discharges related to both land use and location within the Coastal Plain. Concern about worsening water quality in our Nation's waters has prompted considerable research on the ability of buffers to mitigate the effects of agricultural NPS pollution. Two papers focused on the ability of riparian buffers to reduce nitrate in groundwater flowing laterally through the buffer spearheaded these efforts. Peterjohn and Correll (1984) and Lowrance et al. (1984) identified agricultural riparian buffers as critical areas in stream ecosystems, able to retain as much as 90 percent of the nitrate-nitrogen in upland groundwater moving into the forest from an upland agricultural field. This early work did not determine the mechanisms for this reduction. Since that time, the significance of temperate riparian forests as nitrate removal areas, and therefore as barriers to the efficient export of nitrate to the adjacent stream, has been well documented (Lowrance et al., 1984, 1997; Peterjohn and Correll, 1984; Gilliam and Skaggs, 1988; Groffman and Teidje, 1989; Correll, 1991; 9 Correll et al., 1992; Groffman et al., 1993; Haycock and Pinay, 1993; Jordan et al., 1993, 1997 a, b, c; Pinay et al., 1993, 1993; Fennessy and Cronk, 1997; Snyder et al., 1998; Creed and Band, 1998). The vast majority of research reported in the literature has been conducted on Coastal Plain riparian forests, which are typically located on well-drained uplands above poorly drained riparian forests (Jordan et al., 1993). These studies were typically conducted in conditions where shallow aquicludes forced groundwater to flow laterally through forest soils. Under these conditions, riparian forests acted as nutrient sinks, retaining 70 to 90 percent of the total nitrogen inputs, which entered mainly as nitrate in subsurface discharges from adjacent cropland (Jordan et al., 1997a). The mechanisms of nitrogen removal by riparian forests were not completely understood, but much of the nitrogen removal was believed to be due to denitrification, since these forests demonstrated denitrifying conditions, and nutrient retention has been shown to be primarily a belowground process. Riparian effectiveness to reduce nitrate in groundwater has been found to reflect both temporal and physical factors. Groffman and Tiedje (1989) documented differences in excess of a factor of 10 in the rates of denitrification in riparian soils. During brief periods in the spring and autumn, daily rates of denitrification exceeded 0.5 kg N/ha/d. Annual denitrification rates ranged from less than 1 kg N/ha/yr in well- drained sandy soils to over 40 kg N/ha/yr in poorly drained clay loam soils. Bohlke and Denver (1995) found reduced effectiveness of riparian forests on dissolved nitrate in Coastal Plain watersheds where groundwater flow followed relatively deep flow paths before converging and discharging rapidly upward to the stream. 10 MODELS TO PREDICT INSTREAM NITRATE IN AGRICULTURAL AREAS Whereas these studies used an empirical approach to delineate buffer function, another approach to investigate the effect of riparian buffers on water quality has been to develop a physically explicit model. Models allow researchers to test their knowledge about the behavior and functioning of environmental systems. Phillips (1989) generated a physically explicit Riparian Buffer Delineation Equation to predict buffer widths for removal of 90 percent of nitrate from runoff typical of 50 acres of row crops. This model predicted effective widths from 5 to 93m, based on soils and other conditions. The Riparian Ecosystem Management Model (REMM) is another physically explicit riparian model that has been in development and testing for many years (Lowrance and Shirmohammadi, 1985; Inamdar et al., 1999, 1999; Lowrance et al., 2000). This model quantifies groundwater nitrate reductions by riparian buffers under varying site conditions. Both of these models are extremely data intensive and are limited to field scale effects on groundwater passing through a riparian buffer. A landscape approach to wetland function has been suggested in order to make reasonable decisions about how any particular wetland might affect water quality parameters on a broader scale (Whigham et al., 1998) and, in fact, several studies predicting water quality from landscape and riparian buffer characteristics have been conducted. This landscape approach has indicated that location within the watershed appears to play a key role in riparian effectiveness. Riparian areas that border uplands appear to be more important for nitrogen processing and retention of large sediment 11 particles (Whigham et al., 1998). Although Osborne and Wiley (1988) found riparian areas to be important areas for nitrate processing, they cautioned that although riparian forests in the lower part of watershed can limit nutrient inputs from immediate surrounding areas, they have no effect on mitigating inputs from upstream. In many studies, nutrient concentration in streams was more strongly related to catchment-wide land use and geology than riparian corridor characteristics (Omernik, 1976; Omernik et al., 1981; Osborne and Wiley, 1988; Close and Davies-Colley, 1990; Hunsaker and Levine, 1995; Richards et al., 1996; Herlihy and Kaufman, 2000). Some studies attributed their results to the use of low-resolution land use/land cover data (Richards et al., 1996). Most recently, however, the Mid-Atlantic Remote Sensing Atlantic Coast (RESAC) Program and the University of Maryland performed a study to identify and assess riparian forest buffers in Montgomery County, Maryland, using extremely high-resolution remotely sensed (RS) data (resolution of 4 meters). The extent of riparian buffer coverage was compared with watershed water quality. Even at this level of resolution, the relationship between water quality and watershed riparian buffer became significantly different only at an alpha of 0.20 (Zinecker et al., 2001). In other studies, riparian characteristics explained more variation in water chemistry parameters than whole watershed data, indicating that some water quality parameters can be modified by local riparian conditions (Osborne and Kovacic, 1993; Johnson et al., 1990; Tufford et al., 1998). Tufford et al. (1998) conducted a Geographic Information System (GIS) study in which buffer zones of set widths were constructed around the streams and land use/land cover were related to pollution levels 12 in stream. Land use within a 150-m zone around the stream was the best predictor of nitrate-nitrogen concentrations in the stream, with R2 values from 0.25 - 0.63. Basnyat et al. (1999) conducted a similar study, but instead of using set boundary widths for the riparian zone, they identified "contributing zones" using GIS tools and RS data. Models were then constructed and evaluated at three scales: basin scale, contributing- zone scale, and stream buffer/riparian zone scale. Regression results at the stream buffer scale suggested that water quality was highest when passive land uses, such as forests and grasslands are located adjacent to streams. Several models have been developed to evaluate water quality effects of alternative watershed land use and management strategies. The advent of GIS methods has allowed water quality models to be coupled with landscape information to provide a means for describing variable conditions throughout a watershed. The hydrologic submodels of existing distributed watershed models such as the Hydrological Simulation Program - Fortran (HSPF) (Johansen et al., 1983) and the Soil and Water Assessment Tool) (SWAT) (Arnold et al., 1998) are conceptually based and not physically based. These models all require intensive site-specific parameter calibration. Nikolaidis et al. (1993) developed the Nutrient Transport and Transformation Model (NTT-Watershed), a physically based model used to design riparian buffers for NPS pollution control. Heng and Nikolaidis (1998) simulated nitrogen dynamics in a predominantly forested watershed. Osmond et al. (1997) developed WATERSHEDSS, a prototype GIS-assisted model to aid in the management of predominantly agricultural watersheds. Perry et al. (1999) used the ?SPANS? model to determine the water quality impact of different riparian forest 13 management scenarios in a highly agricultural watershed. Riparian effects were modeled by extrapolating field-scale measurements on riparian forest buffer capacity to the watershed scale. While providing results, all of these models assume linear relationships between forces acting at the local and watershed scales, are highly technical, and data-intensive. The differences in conclusions from empirical studies indicate that the influences of landscape factors on streams are complex and probably operate at both site (field scale) and watershed scales. Most studies have focused on reducing sediment and water-borne pollutant loads under controlled field-plot experiments. However, the water quality impact of riparian buffers clearly depends on their location and interaction with other watershed elements as well as on their physical characteristics (Vandervalk and Jolly, 1992; Whigham et al., 1998). The degree of pollution reduction can be expected from the installation of buffers remains unclear; in fact, no research has been done to demonstrate the change of in-stream levels of pollutants in response to conversion of riparian land from crops to buffers (Dosskey, 2000). There are now over 700 publications on the water quality functions of riparian buffers (Correll, 2000) and yet the cumulative effectiveness of riparian buffers in reducing NPS pollution in agricultural watersheds has not been fully established. The problem is twofold. Most research is performed on small areas, at the field or site scale, where as management decisions are made at the watershed scale or larger. Secondly, most research is concerned with the effect of buffers on nitrate in 14 groundwater nitrate, whereas most management problems are concerned with levels of instream nitrate. ENVIRONMENTAL CONTROLS ON INSTREAM COMMUNITIES ? FISH AND MACROINVERTEBRATES Living resources, not nutrients, are not the final concern in the management of ecosystems. Current management strategies are incorporating results of biological assessments into decision-making. Rapid bioassessment techniques that qualify the health of instream communities are being used to assess water conditions and impacts of stream restoration efforts and land use decisions. Therefore, understanding controls in the instream communities that are used to assess stream conditions is critical to properly assess their usefulness as a guidance criterion for management decisions. Significant questions exist concerning the scale of the primary regulating mechanism(s) for instream communities. By the late 1990s, the importance of large- scale land use and catchment characteristics as determinants of stream assemblages was recognized. Johnson and Gage (1997) suggested that upland structure and land uses beyond the riparian zone are the primary regulators of instream community structure and processes. Richards et al. (1996) suggested that the influence of landscape cover characteristics throughout the basin might be as important as riparian vegetation for understanding stream ecosystems. Several studies have sought to identify the predominant environmental factors that control stream macroinvertebrate communities (Richards et al., 1993, 1996, 1997; 15 Collier, 1995; Bunn et al., 1999; Lammert and Allan, 1999) and fish communities (Frenzel and Swanson, 1996; Hall et al., 1996). There is considerable evidence to suggest that both the quality and quantity of available habitat affect the structure and composition of resident biological communities (Maddock, 1999). Several studies demonstrated strong associations between agricultural land use and alterations in stream habitats that cause compositional changes to stream communities (Lenat, 1984; Corkum, 1989, 1990; Quinn and Hickey, 1990; Roth et al., 1998; Allan et al., 1997). Several studies have sought to distinguish the relative effects of riparian influence versus the extended landscape. The influence of stream buffer regions on stream habitats and biota has been shown to vary significantly and does not always follow distinct trends (Johnson and Gage, 1997). Aquatic communities are affected by geologic, geomorphic, hydrologic and riparian conditions, but the relations are aquatic community dependent. Several studies have shown that fish and macroinvertebrates appear to respond differently to landscape and habitat variables (Lammert and Allan, 1999; Fitzpatrick et al., 2000; Herlihy and Kaufman, 2000). Plafkin et al. (1989) suggested that macroinvertebrates are indicative of local habitat conditions while fish reflect conditions over broader spatial areas because of their relative mobility and longevity. The mobility of fish and their possible linkage into larger metapopulations may reduce their sensitivity to the variability of stream habitat (Schlosser and Angermeier, 1995). Thus it might be expected that indexes based on macroinvertebrates give somewhat different indications of stream condition than indexes based on fish. 16 Lammert and Allan (1999) related biotic condition to patterns of land use and channel structure for 18 sites in an agricultural watershed in Michigan. They used multiple linear regression to add instream habitat variables to land use variables in order to explain variation in biotic integrity scores. Fish showed a stronger relationship to flow variability and immediate land use, while macroinvertebrates correlated most strongly with dominant substrate. Their results suggested that riparian land use and instream habitat were not independent variables, supporting the hypothesis of Frissell et al. (1986) that the regional configuration of a watershed constrains the local structure. Other studies have documented influences other than land use that show significant effects on instream biotic communities. Fish assemblages have been found to be affected by hydrology and hydraulics (Poff and Allan, 1995), stream order (Osborne et al., 1992), riparian grazing (Armour et al., 1991; Wohl and Carline, 1996), percent cropland in the watershed (Frenzel and Swanson, 1996), riparian characteristics (Fitzpatrick et al., 2000), and nutrients (Miltner and Rankin, 1998). The benthic macroinvertebrate community has been shown to be affected by riparian grazing (Armour et al., 1991; Wohl and Carline, 1996), stream width ((Richards et al., 1996), nutrients (Miltner and Rankin, 1998), and past land use (Harding et al., 1998). Measures of these environmental determinates are often not included in a readily available GIS data sets that focus on land use. Most recently, a cooperative feasibility study was conducted in 2001 with the US Environmental Protection Agency (EPA) Mid-Atlantic integrated assessment team, Maryland Department of Natural Resources (Power Plant Research Program, 17 Watershed Management and Analysis Division) and Versar, Inc., of Columbia, Maryland. The goal was to develop improved methods for targeting riparian buffer restoration in the Mid-Atlantic. This study used Maryland Biological Stream Survey (MBSS) data from the Patapsco river basin, which is a largely urban area. They found fish Index of Biotic Integrity (IBI) to be significantly related to urban land use, with more variation being explained as the scale of focus was increased from local to riparian to catchment. In contrast, the benthic IBI was found to be strongly affected by local conditions, including riparian buffer width. For a sub-population of sites with greater than 50 percent agricultural land use, there was a 55 percent reduction in the likelihood of a poor or very poor benthic IBI if riparian buffer width was increased from 0 to 30m (Southerland et al., 2002). MODELS TO PREDICT INSTREAM COMMUNITIES ? FISH Studies seeking to determine the scale of critical influence on fish communities have shown mixed results. Several studies have shown that watershed land use was a more important determinate for fish communities than the riparian zone (Steedman, 1988; Gregory et al., 1991; Taylor et al., 1993; D?Angelo et al., 1997; Roth et al., 1998; Allan et al., 1997; Lammert and Allan, 1999; Southerland et al., 2002). Roth et al. (1998) found that regional land use was the primary determinant of local habitat and fish communities, able to overwhelm the ability of local riparian vegetation to support high-quality habitat and biotic communities. Correlations were strongest at the 18 catchment scale (R2 = 0.50 for fish IBI vs. percent agriculture, R2 = 0.76 for habitat index vs. percent agriculture), and tended to become weak and non-significant at local scales. Allan et al. (1997) found that extent of agricultural land at the subcatchment scale was the single best predictor of local stream conditions in a highly agricultural watershed in Michigan, with local riparian vegetation a weak secondary predictor of habitat quality and biotic integrity. In contrast, a set of studies by Goldstein et al. (1996) in Minnesota and North Dakota found fish to be more related to aquatic habitat, riparian, and hydrologic conditions than to watershed agricultural land. Steedman (1988) reported that biotic integrity of Ontario streams was related to the proportion of stream channel length with riparian forest coverage. Fitzpatrick et al. (2000) examined correlations among stream biological quality, watershed and riparian land cover, and hydrological characteristics for 25 agricultural streams in Wisconsin. Several spatial scales were found to influence fish communities, but local riparian conditions appeared to be more important than watershed land cover. Lammert and Allan (1999) also found that land use immediate to the stream was more predictive of fish IBI than regional land use, but was less important than instream habitat variables in explaining the variability observed in fish assemblages. Stewart et al. (2000) correlated habitat, fish and macroinvertebrate data for 38 2nd- and 3rd-order warmwater streams in Wisconsin. They found that fish density was positively correlated to the average length of riparian corridors that were greater than 30m in width (R = 0.42, P < .009), implying that a longer, more continuous, wider riparian corridor is beneficial for fish density in the stream. Herlihy and Kaufman (2000) examined relationships between indicators of instream ecological conditions 19 (chemistry, benthos, and fish) with indicators of local reach riparian condition, watershed condition, and natural drivers (size, elevation, and slope). Multiple regression models showed the fish community to have very weak or no relationship to either watershed or riparian indicators. Wichert and Rapport (1998) performed the only temporal study. In this study they examined the effects of improved agricultural practices and riparian vegetation on stream habitat and fish community structure over 43 years. Water quality data showed that levels of critical substances in the three study watersheds never reached levels harmful to fish life, but degraded habitat and fish communities. Their findings showed that effects of agricultural BMPs were mixed with effects of riparian vegetation in short term studies, and that changes in fish communities are evident only over long periods of comparison. MODELS TO PREDICT INSTREAM COMMUNITIES ? MACROINVERTEBRATES The current interest in the use of fish IBI as a criterion to judge stream restoration or land use management decisions make it of critical interest. Empirical studies have consistently shown that benthic macroinvertebrate communities are strongly influenced by local conditions. D?Angelo et al. (1997) found that local factors had the greatest influence on the invertebrate community, and Richards et al. (1996) suggested that land use features, such as riparian buffers, have strong local influences on stream habitats. Aguiar et al. (2002) also found that riparian features 20 had greater influence than other environmental characteristics on the composition of macroinvertebrate assemblages. Lammert and Allen (1999) found local instream habitat variables to be superior to land use in predicting biotic integrity for macroinvertebrates. Fitzpatrick et al. (2002) examined correlations among stream biological quality, watershed and riparian land cover, and hydrological characteristics for 25 agricultural streams in Wisconsin. They found that, in contrast to fish, benthic communities were influenced more by instream habitat, riparian width and baseflow. They found no significant correlation between measures of macroinvertebrate community and percent of watershed area in row crops. In the study conducted by Stewart et al. (2000) on 38 2nd- and 3rd-order warmwater streams in Wisconsin, agricultural land use located in the 10m to 30m buffer zone was found to be more influential on the macroinvertebrate community than agricultural land use located farther away from the stream. Although Herlihy and Kaufman (2000) found the benthic IBI to be more strongly associated with watershed than riparian conditions for the overall data set, the subset of upland sites showed a stronger correlation between IBI and riparian conditions. THE STATE OF RIPARIAN RESTORATION GUIDANCE - PLACEMENT AND WIDTH Upland vegetated buffers are widely recognized as being necessary to protect wetlands, streams and other aquatic resources, and stream buffer width guidance has been available as far back as the late 1980s (Budd et al., 1987). However, because 21 riparian buffers perform multiple functions, no single recommendation for sizing and management of buffers for practical performance criteria has been found to be suitable for all cases (FISRWG, 1998). In general, narrower buffer widths are considered adequate for chemical and physical functions, while wider buffers are considered necessary for adequate protection of the biological components of stream ecosystems (Brazier and Brown, 1973; Hewlett and Fortson, 1982; Large and Petts, 1992; Osborne and Kovacic, 1993; Castelle et al., 1994; Haycock and Muscutt, 1995; Alliance for the Chesapeake Bay, 1996). Published width recommendations range from a few meters for temperature control to several hundred meters for recreation (Hewlett and Fortson, 1982; Castelle et al., 1994). At least 15 to 30 meters are believed to be necessary to protect wetland and streams under most conditions (Castelle et al., 1994). Stream buffers less than 50 meters wide can be important for defining elements of stream habitat such as woody debris and shoreline protection (Budd et al., 1987). Buffer width recommendations also depend on the size of the stream, ranging from less than 10 meters for small ditches and streams to greater than 50 meters for large rivers (Haycock and Muscutt, 1995). Adequate widths to provide water quality protection through nutrient reduction are widely variable, ranging from 20 meters (Haycock and Muscutt, 1995) to 90 meters (Castelle et al., 1994). If ecological objectives are to be incorporated into riparian buffer guidance, meeting the minimum area needs of a species, guild, or community is especially important. The minimum area is the amount of habitat required to support the expected or appropriate use and can vary greatly across species and seasons (FISRWG, 1998). However, most width guidance values for "species 22 diversity" do not indicate if the species are terrestrial or aquatic. Finally, riparian influence might be expected to vary with stream order (Vannote et al., 1980; Alliance for the Chesapeake Bay, 1996), which will influence width and placement recommendations. Three comprehensive sources of stream/riparian guidance have been published. The U.S. EPA authored a multi-year synthesis of current expertise on stream corridor restoration and published the results as a guidebook entitled ?Stream Corridor Restoration; Principles, Processes, and Practices? (FISRWG, 1998). The United States Department of Agriculture (USDA) Forest Service published the ?Chesapeake Bay Riparian Handbook: a Guide for Establishing and Maintaining Riparian Forest Buffers? (Palone and Todd, 1997). The Maryland Department of Natural Resources (DNR) published a guidance manual on their website entitled ?Riparian Forest Buffers: Function and Design for Protection and Enhancement of Water Resources? (Maryland DNR website, 2002). However, all of these publications failed to provide quantitative guidance for the functions and processes of riparian buffers in different settings. Rather, general guidelines for buffer widths of 100 to 300 feet were suggested. Agricultural systems management and restoration have traditionally focused on site-level strategies (Imhof et al., 1996). Specific guidance for riparian restoration and management has typically been established by political acceptability and not necessarily scientific merit (Castelle et al., 1994; FISRWG, 1998), and often does not incorporate concepts of landscape ecology or clear ecological objectives. Resource agencies are most often responsible for setting buffer size requirements or guidelines. 23 Buffers are a voluntary BMP, and width recommendations are driven by the buffer function of interest to the funding agency. For example, USDA funds the CREP program in Maryland, and is primarily concerned with the reduction of nutrients in surface water. There are basically two systems for establishment of buffer width guidance - fixed width and variable width buffer systems. Variable width buffer systems have a stronger scientific base, as they consider a combination of buffer sizing criteria, such as functional value, adjacent land use intensity and other site-specific conditions. Unfortunately, variable-with buffer systems require greater expenditure of resources and higher level of training for agency staff, while offering less predictability for land use planning. RIPARIAN CLASSIFICATION SYSTEMS Because physical variation in riparian areas may aid or prevent the interception and transformation of pollutants (Cirmo and McDonnell, 1997; Weller et al., 1998), understanding the underlying natural variation in riparian buffer landscapes can augment interpretation of the impact of land use patterns on aquatic conditions. As the plenary speaker for the American Water Resources Association International Conference on Riparian Ecology and Management in Multi-Use Watersheds, held in Portland, Oregon, in August 2000, Dr. David Correll of the Smithsonian Environmental Research Center stated that "one of the most urgent needs in riparian research is to develop and utilize an adequate system of classification for riparian buffers" (Correll, 2000). Without a comprehensive description of site geology, 24 vegetation, hydrology, and soils, research findings are hard to generalize or extrapolate to other sites (Correll, 2000). Additionally, riparian characteristics that can be readily examined using a GIS format can be combined with land-use and soil maps in order to determine what riparian management alternatives might contribute to decreasing nonpoint source pollution impacts on surface waters (Delong and Brusven, 1991). A few studies have developed procedures for classifying riparian areas. Most riparian classification systems focus on a few selected attributes of riparian areas, such as hydric soil or hydrophilic plant associations (Dick-Peddie and Hubbard, 1977; Brown et al., 1979; Norton et al., 1980, 1981; Batchelor, 1982; Winward, 1984; Youngblood et al., 1985; Curry and Slater, 1986; Szaro and Patton, 1986; Kovalchik, 1987; Baker, 1989; Padgett et al., 1989; Kovalchik and Chitwood, 1990; Petersen et al., 1992; Durkin et al., 1996; Girard et al., 1997; Alpert and Kagan, 1998; Lyon and Sagers, 1998; Arbuckle et al., 1999). Although these perspectives adequately characterized the terrestrial plant communities, they have provided little understanding of the wide array of ecological processes and communities associated with the land- water interface (Gregory et al., 1991). Additionally, most of these were developed for management agencies such as USDA, US Department of Wildlife, US Bureau of Land Management, and the National Forest System (Cowardin et al., 1979; Youngblood et al., 1985; Pierce and Johnson, 1986; Kovalchik, 1987; Padgett et al., 1989). Therefore, many of these systems were developed for specific settings in the western United States (Colorado, Idaho, Kansas, Montana, Nevada, New Mexico, Oregon, Utah and Wyoming). 25 Very few systems have expanded the focus of the classification system to include the landscape setting. Kovalchik and Chitwood (1990) added geomorphology to a floristic classification system they had developed in 1987. This resulted in a system that classified riparian sites by location in the landscape (such as a mountainous landform), in the stream system (ex: first-order watershed), by gradient, elevation, fluvial surface (such as v-shaped floodplain), and soil type. Delong and Brusven (1991) developed a classification system that identified riparian characteristics using discrete categorical units to identify agricultural areas prone to erosion. Most recently, Quinn et al. (2000) developed a landscape classification system for large-scale riparian settings in New Zealand. They used a subjective knowledge-based approach to develop classes that reflected similarities in morphology, riparian functions, human uses, and likely riparian management options. An applicable classification system could enhance efforts to elucidate buffer effectiveness by providing a technique to reduce error in modeling efforts, with subsequently clearer guidance for buffer restoration (placement and width.) MARYLAND?S AGRICULTURAL RIPARIAN BUFFERS ? AVAILABLE INFORMATION There are currently no studies in the primary literature describing the state of riparian buffers in Maryland. Current estimates of buffer extent, location, distribution and type are largely based on surveys conducted using RS data (Chesapeake Bay Program, 1996; 1999). As part of the Chesapeake Bay Riparian Forest Buffer 26 Initiative, a riparian forest buffer inventory was completed for the Bay watershed in 1997. The inventory used the EPA's Environmental Monitoring Assessment Program land cover data (1989-1991) and included an accuracy assessment protocol using aerial photography to verify its reliability. This GIS technology was used to assess the status of riparian areas and to provide a baseline for information on riparian forest buffers in the watershed for state and federal agencies. Maryland was determined to have 26,966 stream km (16,756 miles), with 12,926 stream km (8,032 miles) possessing 30.5 meter (100-foot) buffers on both sides, 14,565 stream km (9050 miles) possessing 30.5 meter (100-foot) buffers on one side, and 12,402 stream km (7706 miles) possessing less than 30.5 meter (100-foot) buffers on one side (Chesapeake Bay Program, 1999). These remote-sensing studies were used to delineate the extent of riparian buffers, but the resolution of the image (30 m) is greater than the width of many buffers. LandSat earth-orbiting satellites have provided images tailored to resource assessment since 1982. In landscape-level analysis studies, the 30-meter ground resolution has been shown to provide approximately 81 percent accuracy (Hewitt, 1990). However, when the resource to be analyzed is small compared to the pixel size (e.g. narrow stream channels or riparian zones) space data cannot provide accurate information on the resource (Muller et al., 1993). The Mid-Atlantic RESAC Program and the University of Maryland performed a study to identify and assess riparian forest buffers in certain subwatersheds of Montgomery County, Maryland, using high- resolution (4 m) RS data (Zinecker et al., 2002). Whereas the resolution of these data 27 was sufficient for adequate representation of riparian buffers, this imagery was available only for the metropolitan corridor between Washington and Baltimore. Pollution reduction estimates are dependent upon accurate estimates of buffer extent and watershed-scale impact on nutrient reduction. In 1999, the University of Maryland, the Maryland Agricultural Cooperative Extension Service and the Forestry Division of Maryland's Department of Natural Resources conducted a study entitled "Environmental Benefits and Costs of a Voluntary Riparian Forest Buffer Program". This research sought to provide guidance for voluntary incentive programs in Maryland by determining the water-quality benefit of agricultural RFBs and the level of incentive necessary for the implementation of RFBs in different areas of the state. Due to the lack of adequately accurate information, a fieldwork-based survey of agricultural riparian buffers in Maryland's Chesapeake Bay watershed was conducted as part of this project. This research provided the first extensive, ground-truthed attempt to describe a large set of agricultural riparian buffers, the surrounding land, and to a lesser extent, the land and streams within the buffers. The only other data set that currently includes extensive, ground-truthed information on Maryland buffers is the Maryland Biological Stream Survey (MBSS). The MBSS has gained national attention in recent years because of the comprehensive, spatially intense nature of the program. Since 1994, the Maryland Department of Natural Resources (DNR) has conducted the MBSS, a comprehensive program to assess the status of biological resources in Maryland?s streams. This annual survey examines the water quality, physical habitat, and biological conditions in first through third order, non-tidal streams (Klauda et al., 1998). Although buffer characterization 28 is not the primary purpose of the survey, the MBSS includes limited site-specific, ground-truthed information on riparian buffers. The results of the 1995-1997 sampling provide the following statewide statistics about Maryland buffers; about 59 percent of all stream miles have forested riparian buffers, 14 percent are buffered by vegetation other than forest, such as abandoned cropland or lawns, 27 percent of all stream miles in the state are unbuffered, and about 40 percent of the forested stream miles have buffers greater than 50 meters wide (Boward et al., 1999). MBSS 1995-1997 DATA Data from the 1995-1997 sampling round have been made publicly available on the Maryland DNR website as Microsoft Excel files. The Guide to Using 1995- 1997 Maryland Biological Stream Survey Data contains an explanation of the 1995- 1997 sampling design and an overview of laboratory and field methods (Mercurio et al., 1999). More detailed information on methods may be found in the MBSS sampling manual (Kazyak, 1997). The guide describes the contents of each data set. Variables listed in each of the data sets are defined and additional information is provided to assist users in interpreting and analyzing MBSS data. Approximately 300 75-meter stream segments were sampled each year, with biological, chemical, and physical parameters measured at each segment using standardized methods. Biological measurements included abundance and health of fish, and composition of benthic macroinvertebrate communities. Chemical measurements included pH, acid-neutralizing capacity (ANC), sulfate, nitrate- 29 nitrogen, conductance, dissolved oxygen and dissolved organic carbon (DOC). Numerous physical habitat measurements were assessed including flow, stream gradient, maximum depth, thalweg depth, wetted width, temperature, the number of root wads and woody debris, embeddedness, instream habitat, epifaunal substrate, pool and riffle quality, bank stability, channel flow status, shading, and riparian buffer type and width. The presence of storm drains, effluent discharges, and beaver ponds were also recorded. The aesthetic value and remoteness of each site were quantified based on evidence of human activity at each site. Regional land cover data were used to characterize catchment land uses. Several indicators of the biological health of the stream sampled were developed from the data collected. A fish IBI and a benthic IBI (Roth et al., 1998; Stribling et al., 1998) were used to assess the condition of both the fish and benthic macroinvertebrate communities by comparing the species assemblages found at each site to minimally impacted reference sites found throughout the state. IBI scores used for the 1995-1997 MBSS are the mean of several individual metric scores and range from 1 (very poor) to 5 (good). A reference-based Physical Habitat Index (PHI) was also developed as a means of summarizing a variety of important habitat metrics (Hall et al., 1999), such as instream habitat, diversity of velocity and depth, and embeddedness. Klauda et al. (1998) reported the preliminary results of the 1995-1997 MBSS, and concluded that degradation of physical habitat is the primary threat to Maryland?s freshwater streams. Several reports documenting MBSS results for specific river basins are available. The results of the first three years (1995-1997) of sampling were 30 summarized for the general public in ?From the Mountains to the Sea: the State of Maryland?s Freshwater Streams? (Boward et al., 1999). The survey assessed stream conditions as deviation from minimally impacted expectations and identified likely sources of degradation by delineating basic relationships between biological conditions and anthropogenic stresses. Roth et al. (1999, 1999) developed two technical versions of the 1995-1997 data analysis in "State of the Streams: 1995-1997 Maryland Biological Stream Survey Results" and ?Relative and cumulative impacts of stressors on streams: 1995-1997 MBSS results?. Basic comparisons were made between the biological and physical habitat indices and riparian and land use/land cover data at the state- and basin-wide scales. Fish and benthic IBI scores were found to increase slightly with riparian width, but these increases were not significant. Very weak correlations were found between certain measures of the benthic community and measures of instream nitrate. Fish and benthic IBI scores were sensitive to the degree of watershed urbanization, but were less able to detect effects of agriculture at the watershed scale. Limited multiple regression analysis was also performed using each of the IBIs as the response variable and seven pre-determined "stressors" as indicator variables. Width of riparian vegetation at the site was not found to be a significant factor for explaining variation in fish IBI. In contrast, benthic IBI increased significantly with PHI and riparian buffer width. 31 SUMMARY OF LITERATURE FINDINGS AND RESEARCH NEEDS Many riparian functions are well established, but the degree to which riparian buffers can contribute to reduction of pollutant levels in agricultural streams remains to be quantified (Fennessy and Cronk, 1997; Dosskey, 2000). An examination of the effect of agricultural buffers at the watershed scale has not been done in Maryland. Such an analysis requires a trustworthy description of buffers, water quality, and the stream communities in primarily agricultural watersheds. Whereas Roth et al. (1995) looked at riparian effect on water quality and instream communities in Maryland, no work has been done focusing on riparian effect in agricultural settings. In addition, the current body of literature has clearly established that buffer effectiveness varies due to placement and landscape characteristics. Therefore, Maryland policymakers are still without guidance to predict or quantify the effect of riparian buffer management at the watershed-level. Current topics of interest are nutrients, driven by CBP goals, and stream communities, driven by popularity of bioassessment techniques. Thus, existing work has established a strong framework for the general environmental drivers in these systems. However, the effectiveness of riparian buffers in any particular setting has not been clearly shown. The general approach used to predict buffer effect on instream measures of stream health has been empirical analysis of a data set with or without a spatial component, such as GIS information. This research will construct empirical models as well, because the scale of the 32 question is large. Easily-obtainable and readily-available data will be used, to establish a protocol that can be used by management agencies. In particular, buffer effectiveness models will be based on MBSS data, which are publicly-available information based on a rapid assessment technique, and contain ground-truthed information on buffers. 33 CITED REFERENCES Aguiar F.C., M.T. Ferreria and P. Pinto. 2002. Relative influence of environmental variables on macroinvertebrate assemblages from an Iberian basin. J. N. Am. Benthol. Soc. 21(1): 43-53. Allan J.D., D.L. Erickson and J. Fay. 1997. The influence of catchment land use on stream integrity across multiple spatial scales. Freshwater Biology 37(1): 149-161. Alliance for the Chesapeake Bay. 1993. Nutrients and the Chesapeake: Refining the Bay cleanup effort. Chesapeake white paper. Annapolis, Md.: Chesapeake Bay Program. Alliance for the Chesapeake Bay. 1996. Final Report Submitted to the Chesapeake Bay Executive Committee. Baltimore, Md.: Riparian Forest Buffer Panel. Alliance for the Chesapeake Bay. 1996. Riparian Forest Buffers. White paper. Annapolis, Md.: Chesapeake Bay Program. Alpert P. and J. Kagan. 1998. The utility of plant community types: a practical review of riparian vegetation classification in the Intermountain United States. Natural Areas Journal 18(2): 124-128. Angle J.S., C.M. Gross, R.L. Hill and M.S. McIntosh. 1993. Soil nitrate concentrations under corn as affected by tillage, manure, and fertilizer applications. Journal of Environmental Quality 22: 141-147. Arbuckle C.J., A.D. Huryn and S.A. Israel. 1999. A landcover classification in the Taieri River Catchment, New Zealand: a focus on the riparian zone. Geocarto International 14(3): 7-14. Armour C.L., D.A. Duff and W. Elmore. 1991. The effects of livestock grazing in riparian and stream ecosystems. Fisheries 16(1): 7-11. Arnold J.G., R. Srinivasan, R.S. Muttiah and J.R. Williams. 1998. Large area hydrologic modeling and assessment. I. Model development. Journal of the American Water Resources Association 34(1): 73-89. Baker W.L. 1989. Classification of the riparian vegetation of the montane and subalpine zones in western Colorado. The Great Basin Naturalist 49(2): 214-228. 34 Barling R.D. and I.D. Moore. 1994. Role of buffer strips in management of waterway pollution: a review. Environmental Management 18(4): 543-558. Basnyat P., L.D. Teeter, K.M. Flynn and B.G. Lockaby. 1999. Relationships between landscape characteristics and nonpoint source pollution inputs to coastal estuaries. Environmental Management 23(4): 539-549. Batchelor R.F. 1982. A taxonomic classification system for Montana riparian vegetation types: an interagency approach to classifying Montana's riparian ecosystems. Bozeman: Montana: Rural Area Development Committee, Wildlife Subcommittee, Riparian Program Team. Beaulac M.N. and K.H. Teckhow. 1982. An examination of land use - nutrient export relationships. Water Resources Bulletin 18: 1013-1094. Blankenship, K. 2002. Bay Journal, May 2002, Bay Program narrows range for nutrient reductions. Available at: http://www.bayjournal.com. Accessed on 15 June, 2003. Bolke J.K. and J.M. Denver. 1995. Combined use of groundwater dating, chemical and isotopic analyses to resolve the history and fate of nitrate contamination on two agricultural watersheds, Atlantic Coastal Plain, Md. Water Resources Research 31: 2319-2339. Boward D.M., P.F. Kazyak, S.A. Stranko, M.K. Hurd and T.P. Prochaska. 1999. From the mountains to the sea: the state of Maryland?s freshwater streams. EPA 903-R-99- 023. Annapolis, Maryland: Maryland Department of Natural Resources, Monitoring and Non-Tidal Assessment Division. Brazier J.R. and G.W. Brown. 1973. Buffer strips for stream temperature control. Research Paper 15. Corvallis, OR: Forest Research Laboratory, School of Forestry, Oregon State University. Brown S., M.M. Brinson and A.E. Lugo. 1979. Structure and function of riparian wetlands. In Proceedings of the Symposium on Strategies for Protection and Management of Floodplain Wetlands and Other Riparian Ecosystems, 17-31. R.R. Johnson and J.F. McCormick, eds. U.S. Department of Agriculture, Forest Service. Budd W.W., P.L. Cohen, P.R. Saunders and F.R. Steiner. 1987. Stream corridor management in the Pacific Northwest. I. Determination of stream corridor widths. Environmental Management 11: 587-897. Bunn S.E., P.M. Davies and T.D. Mosisch. 1999. Ecosystem measures of river health and their response to riparian and catchment degradation. Freshwater Biology 41(2): 333-345. 35 Castelle A.J., A.W. Johnson and C. Conolly. 1994. Wetland and stream buffer size requirements ? a review. Journal of Environmental Quality 23: 878-882. Chesapeake Bay Program. 1996. Final report of the Riparian Forest Buffer Panel. Publ. CBP/TRS 158/96 or EPA 903-R-96-015. Annapolis, Md.: Chesapeake Bay Program Chesapeake Bay Program. 1999. Riparian forest buffers: linking land and water: the Chesapeake Bay riparian forest buffer initiative. Publ. CBP/TRS 220/99 or EPA 903- R-99-002. Annapolis, Md.: Chesapeake Bay Program. Cirmo C.P. and J.J. McDonnell. 1997. Linking the hydrologic and biogeochemical controls of nitrogen transport in near-stream zones of temperate-forested catchments: a review. J. Hydrology 199: 88-120. Close M.E. and R.J. Davies-Colley. 1990. Baseflow chemistry in New Zealand rivers. 2. Influence of environmental factors. New Zealand Journal of Marine and Freshwater Research 24: 343-356. Collier K.J. 1995. Environmental factors affecting the taxonomic composition of aquatic macroinvertebrate communities in lowland waterways of Northland, New Zealand. New Zealand Journal of Marine and Freshwater Research 29(4): 453-465. Corkum L.D. 1989. Patterns of benthic invertebrate assemblages in rivers of northwestern North America. Freshwater Biology 21: 191-205. _______. 1990. Intrabiome distributional patterns of lotic macroinvertebrate assemblages. Canadian Journal of Fisheries and Aquatic Sciences 47: 2147-2157. _______. 1992. Relationships between density of macroinvertebrates and detritus in rivers. Arch. Hydrobiol. 125: 149-166. Correll DL. 1983. N and P in soils and runoff of three Coastal Plain land uses. In Nutrient Cycling in Agricultural Ecosystem, 207-224. R.R. Lowrance, L. Asmussen and R. Leonard, eds. Athens, Ga.: Univ. of Georgia Press. _______. 1991. Human impact on the functioning of landscape boundaries. In The Role of Landscape Boundaries in the Management and Restoration of Changing Environments, 90-109. M.M. Holland, P.G. Risser and R.J. Naiman, eds. New York, New York: Chapman and Hall. _______. 1996. Buffer zones and water quality protection: general principles. In Proceedings of the International Conference on Buffer Zones. N. Haycock, ed. Oxford, England. 36 _______. 2000. The current status of our knowledge of riparian buffer water quality functions. In Proceedings of the International Conference on Riparian Ecology and Management in Multi-Use Watersheds, 5-10. P.J. Wigington, Jr. and R.L. Beschta, eds. Middleburg, Va.: AWRA. Correll D.L., Jordan T.E. and D.E. Weller. 1992. Nutrient flux in a landscape: Effects of coastal land use and terrestrial community mosaic on nutrient transport to coastal waters. Estuaries 15: 431 -442. _______. 1994. The Chesapeake Bay watershed: effects of land use and geology on dissolved nitrogen concentrations. In Toward a sustainable watershed: The Chesapeake experiment: Proceedings of the 1994 Chesapeake Research Conference. P. Hill and S. Nelson, eds. Norfolk, Va.: Ches. Research Consortium. Cowardin L.M., V. Carter, F.C. Golet and E.D. LaRoe. 1979. Classification of wetlands and deepwater habitats of the United States. Publ. No. FWS/OBS-79/31. Washington, DC: US Department of Interior, Fish and Wildlife Service, Office of Biological Services. Creed I.F. and L.E. Band. 1998. Export of nitrogen from catchments within a temperate forest: evidence for a unifying mechanism regulated by variable source area dynamics. Water Resources Research 34(11): 3105-3120. Curry P. and F.M. Slater. 1986. A classification of river corridor vegetation from four catchments in Wales. Journal of Biogeography 13(2): 119-132. D?Angelo D.J., S.V. Gregory, L.R. Ashkenas and J.L. Meyer. 1997. Physical and biological linkages within a stream geomorphic hierarchy: a modeling approach. J. N. Am. Benthol. Soc. 16(3): 480-502. De'ath G. and K.E. Fabricius. 2000. Classification and regression trees: a powerful yet simple technique for ecological data analysis. Ecology 81(11): 3178-3192. Delong M.D. and M.A. Brusven. 1991. Classification and spatial mapping of riparian habitat with applications toward management of streams impacted by nonpoint source pollution. Environmental Management 15(4): 565-571. Dick-Peddie W.A. and J.P. Hubbard. 1977. Classification of riparian vegetation. In Importance, Preservation and Management of Riparian Habitat: A Symposium, 85-90. R.R. Johnson and D.A. Jones, tech. coordinators. Fort Collins, CO.: U.S. Rocky Mt. Forest and Range Experimental Station. Or USDA Forest Service General Technical Report: RM-43. Dillon P.J. and W.B. Kirchner. 1975. The effects of geology and land use on the export of phosphorus from watersheds. Water Research 9: 135-148. 37 Dosskey M.G. 2000. How much can USDA riparian buffers reduce agricultural nonpoint pollution? In Proceedings of the International Conference on Riparian Ecology and Management in Multi-Use Watersheds, 427-430. P.J. Wigington, Jr. and R.L. Beschta, eds. Middleburg, Va.: AWRA. Durkin P., E. Muldavin and M. Bradley. 1995. A preliminary riparian/wetland vegetation community classification of the Upper and Middle Rio Grande watersheds in New Mexico. In Desired Future Conditions for Southwestern Riparian Ecosystems: Bringing Interests and Concerns Together. Fort Collins, Colo.: U.S. Dept. of Agriculture, Rocky Mountain Forest and Range Experiment Station. Egglishaw H.W. 1964. The distributional relationship between the bottom fauna and plant detritus in streams. Journal of Animal Ecology 33: 463-476. EPA. 1983. Chesapeake Bay: a framework for action. Philadelphia, Pa.: U.S. Environmental Protection Agency. Ewel K.L. 1978. Riparian ecosystem: conservation of their unique characteristics. In Strategies for Protection and Management o Floodplain Wetlands and Other Riparian Ecosystems. R.R. Johnson and J.F. McCormick, eds. Washington, DC: USDA, Forest Service. Or USDA Forest Service Gen. Tech. Rpt. WO-12. Federal Interagency Stream Restoration Working Group (10/1998). Stream Corridor Restoration: Principles, Processes, and Practices. GPO Item No. 0120-A; SuDocs No. A 57.6/2:EN 3/PT.653. Fennessy M.S. and J.K. Cronk. 1997. The effectiveness and restoration potential of riparian ecotones for the management of nonpoint source pollution, particularly nitrate. Critical Reviews in Environmental Science and Technology 27(4): 285-317. Fitzpatrick F.A., E.M.P. Giddings, B.C. Scudder, D.J. Sullivan and B.N. Lenz. 2000. Effects of riparian corridors and hydrology on agricultural stream biota. In Proceedings of the International Conference on Riparian Ecology and Management in Multi-Use Watersheds, 215-220. P.J. Wigington, Jr. and R.L. Beschta, eds. Middleburg, Va.: AWRA. Freifelder R.R., S.V. Smith and R.H. Bennett. 1998. Cows, humans and hydrology in the nitrogen dynamics of a grazed rural watershed. Journal of Environmental Management 52(2): 99-111. Frenzel S.A. and R.B. Swanson. 1996. Relations of fish community composition to environmental variables in streams of central Nebraska, USA. Environmental Management 20: 689-705. 38 Frissel C.A., W.J. Liss, C.E. Warren and M.D. Hurley. 1986. A hierarchical framework for stream habitat classification: viewing streams in a watershed context. Environmental Management 10: 199-214. Gburek W.J. and G.J. Folmar. 1999. Flow and chemical contributions to streamflow in an upland watershed: a baseflow survey. Journal of Hydrology 217(1-2): 1-18. Gilliam J.W. and R.W. Skaggs. 1988. Nutrient and sediment removal in wetland buffers. In Proceedings of the National Wetland Symposium: Wetland Hydrology. J.A. Kusler and G. Brooks, eds. Berme, NY: Association of State Wetland Managers. Girard M.M., D.L. Wheeler, D.L. Mills and B. Stephanie. 1997. Classification of riparian communities on the Bighorn National Forest. Lakewood, Co.: USDA Forest Service, Rocky Mountain Region. Goldstein R.M., J.C. Stauffer, P.R. Larson and D.L. Lorenz. 1996. Relation of physical and chemical characteristics of stream to fish communities in the Red River of the North Basin, Minnesota and North Dakota, 1993-1995. US Geological Survey Water-Resources Investigations Report 96-4227. Mounds View, Minnesota: USGS. Gregory S., V. Swanson, J. Frederick and W.A. McKee. 1991. An ecosystem perspective of riparian zones: focus on links between land and water. BioScience 41: 540-551. Groffman P.M. and J.M. Tiedje. 1989. Denitrification in north temperate forest soils: relationships between denitrification and environmental factors at the landscape scale. Soil Biology and Biochemistry 21: 621-626. Groffman P.M., D.R. Zak, S. Christensen, A. Mosier and J.M. Tiedje. 1993. Early spring nitrogen dynamics in a temperate forest landscape. Ecology 74: 1579-1585. Haith D.A. and L.L. Shoemaker. 1987. Generalized watershed loading functions for stream flow nutrients. Water Resources Bulletin 23: 471-478. Hall L.W., M.C. Scott, W.D. Killen, R.D. Anderson. 1996. The effects of land-use characteristics and acid sensitivity on the ecological status of Maryland Coastal Plain streams. Environmental Toxicology and Chemistry 15(3): 384-394. Hall L.W. Jr., R.P. Morgan, E.S. Perry and A. Waltz. 1999. Development of a provisional physical habitat index for Maryland freshwater streams. CBWP-MANTA- EA-99-12. Annapolis, Md.: Maryland Department of Natural Resources, Resource Assessment Service, Monitoring and Non-Tidal Assessment Division. 39 Harding J.S., E.F. Benfield, P.V. Bolstad, G.S. Helfman and E.B.D. Jones III. 1998. Stream biodiversity: the ghost of land use past. Publ. No. USA 95: 14843-14847. Proceedings of the National Academy of Science. Haycock N.E. and A.D. Muscutt. 1995. Landscape management strategies for the control of diffuse pollution. Landscape and Urban Planning 31: 313-321. Haycock N.E. and G. Pinay. 1993. Groundwater nitrate dynamics in grass and poplar vegetated riparian buffer strips during the winter. Journal of Environmental Quality 22: 273-278. Heng H.H. and N.P. Nikolaidis. 1998. Modeling of nonpoint source pollution of nitrogen at the watershed scale. Journal of the American Water Resources Association 34(2): 359-374. Herlihy A. and P. Kaufman. 2000. Riparian vs. watershed influence on macroinvertebrates, fish and chemical indicators of stream condition. Poster at the International Conference on Riparian Ecology and Management in Multi-Use Watersheds. Portland, Oregon, 8/27-31, 2000. Hewitt M.J. 1990. Synoptic inventory of riparian ecosystems: the utility of Landsat Thematic Mapper data. Forest Ecology and Management 33/34: 605-620. Hewlett J.D. and J.C. Fortson. 1982. Stream temperature under an inadequate buffer strip in the southern Piedmont. Water Resources Bulletin 18(6): 983-988. Hill A.R. 1978. Factors affecting the export of nitrate-nitrogen from drainage basins in southern Ontario. Water Research 12: 1045-1057. Hughes R.M. and D.P. Larsen. 1988. Ecoregions: an approach to surface water protection. Journal of the Water Pollution Control Federation 60(4): 486-493. Hunsaker C.T. and D.A. Levine. 1995. Hierarchical approaches to the study of water quality in rivers. Bioscience 45: 193-201. Hynes H.B.N. 1975. The stream and its valley. Verhandlungen, Internationale Vereinigugn fur Theoretische und Augewandte Limnologie 19: 1-25. Imhof J.G., J. Fitzgibbon, W.K. Annable. 1996. A hierarchical evaluation system for characterizing watershed ecosystems for fish habitat. Canadian Journal of Fisheries and Aquatic Sciences 53(SU1): 312-326. 40 Inamdar S.P., J.M. Sheridan, R.G. Williams, D.D. Bosch, R.R. Lowrance, L.S. Altier and DL Thomas. 1999. Riparian ecosystem management model (REMM): I. testing of the hydrologic component for a Coastal Plain riparian system. Transactions of the ASAE 42(6): 1679-1689. Inamdar S.P., R.R. Lowrance, L.S. Altier, R.G. Williams and R.K. Hubbard. 1999. Riparian ecosystem management model (REMM): II. Testing of the water quality and nutrient cycling component for a coastal plain riparian system. Transactions of the ASAE 42(6): 1691-1707. Johanson R.C. 1983. A new mathematical modeling system Hydrological Simulation Program - Fortran (HSPF). American Chemical Society Symposium Series 225: 125- 147. Johnson L.B. and S.H. Gage. 1997. Landscape approaches to the analysis of aquatic ecosystems. Freshwater Biology 37(1): 113-132. Johnston C.A., N.E. Detenbeck and G.J. Niemi. 1990. The cumulative effect of wetlands on stream water quality and quantity: a landscape approach. Biogeochemistry 10: 105-141. Jordan T.E. and D.E. Weller. 1996. Human contributions to terrestrial nitrogen flux. BioScience 46: 655-664. Jordan T.E., D.L. Correll and D.E. Weller. 1993. Nutrient interception by a riparian forest receiving inputs from adjacent cropland. Journal of Environmental Quality 22: 467-473. _______. 1997a. Effects of agriculture on discharges of nutrients from Coastal Plain watersheds of the Chesapeake Bay. Journal of Environmental Quality 26(3): 836-848. _______. 1997b. Nonpoint source discharges of nutrients from Piedmont watersheds of the Chesapeake Bay. Journal of the American Water Resources Association 33(3): 631-645. _______. 1997c. Relating nutrient discharges from watersheds to land use and streamflow variability. Water Resources Research 33(11): 2579-2590. Kazyak P.F. 1997. Maryland Biological Stream Survey Sampling Manual. Columbia, Md.: Maryland Department of Natural Resources, Monitoring and Non-Tidal Assessment Division. 41 Klauda R., P.F. Kazyak, S.A. Stranko, M.T. Southerland, N.E. Roth and J.C. Chaillou. 1998. Maryland Biological Stream Survey: a state agency program to assess the impact of anthropogenic stresses on stream habitat, quality and biota. Environmental Monitoring and Assessment 51(1-2): 299-316. Kovalchik B.L. 1987. Riparian zone associations of the Deschutes, Ochoco, Fremont and Winema National Forests. Publ. No. R6 ECOL TP-279-87. Portland, Or: USDA Forest Service, Pacific Northwest Region 6. Kovalchik B.L. and L.A. Chitwood. 1990. Use of geomorphology in the classification of riparian plant associations in mountainous landscapes of central Oregon, USA. Forest Ecology and Management 33/34: 405-418. Lammert M. and J.D. Allan. 1999. Assessing biotic integrity of streams: Effects of scale in measuring the influence of land use/cover and habitat structure on fish and macroinvertebrates. Environmental Management 23(2): 257-270. Large A.R.G. and G.E. Petts. 1992. Buffer zones for conservation of rivers and bankside habitats. Publ. No. BS12 2UD, project 340/5/y. Bristol, England: National Rivers Authority. Larsen D.P., R.M. Omernik, R.M. Hughes, C.M. Rohm, T.R. Whitier, A.J. Kinney, A.L. Gallant and D.R. Dudley. 1986. Correspondence between spatial patterns in fish assemblages in Ohio streams and aquatic ecoregions. Environmental Management 10(6): 815-828. Lenat D.R. 1984. Agriculture and stream water quality, a biological evaluation of erosion control practices. Environmental Management 8: 333-344. Likens G.E. and F.H. Bormann. 1974. Linkages between terrestrial and aquatic ecosystems. Bioscience 24: 447-456. Loehr R.C. 1974. Characteristics and comparative magnitude of non-point sources. Journal of the Water Pollution Control Federation 46(8): 1849-1871. Lowrance R.R., L.S. Altier, J.D. Newbold, R.R. Schnabel, P.M. Groffman, J.M. Denver, D.L. Correll, J.W. Gilliam, J.L. Robinson, R.B. Brinsfield, K.W. Staver, W. Lucas and A.H. Todd. 1997. Water quality functions of riparian forest buffers in Chesapeake Bay watersheds. Environmental Management 21(5): 687-712. Lowrance R.R., L.S. Altier, R.G. Williams, S.P. Inamdar, J.M. Sheridan, D.D. Bosch, R.K. Hubbard and D.L. Thomas. 2000. REMM: The riparian ecosystem management model. Journal of Soil and Water Conservation 55(1): 27-34. 42 Lowrance R.R. and R.A. Leonard. 1988. Streamflow nutrient dynamics on Coastal Plain watersheds. Journal of Environmental Quality 18(4): 734-740. Lowrance R.R. and A. Shirmohammadi. 1985. REM: a model for riparian ecosystem management in agricultural watersheds. General technical report RM - 120: 237-240. Fort Collins, Co.: USDA Forest Service, Rocky Mountain Forest and Range Experiment Station. Lowrance R.R., R. Todd, J. Fail Jr., O. Hendrickson Jr., R. Leonard and L. Asmussen. 1984. Riparian forests as nutrient filters in agricultural watersheds. BioScience 34: 374-377. Lyon J. and C.L. Sagers. 1998. Structure of herbaceous plant assemblages in a forested riparian landscape. Plant Ecology 138(1): 1-16. Maddock I. 1999. The importance of physical habitat assessment for evaluating river health. Freshwater Biology 41: 373-391. McMahon G. and O.B. Lloyd. 1995. Water-quality assessment of the Albemarle- Pamlico drainage basin, North Carolina and Virginia ? environmental setting and water-quality issues. US Geological Survey Open-File Report 95-136. McMahon G. and D.A. Harned. 1998. Effect of environmental setting on sediment, nitrogen, and phosphorus concentrations in Albemarle-Pamlico drainage basin, North Carolina and Virginia, USA. Environmental Management 22: 887-903. Maryland Department of Natural Resources. Riparian Forest Buffers: Function and Design for Protection and Enhancement of Water Resources. Available at: www.dnr.state.md.us/forests/publications/buffers.html. Accessed on 20 June 2002. Mercurio G., J.C. Chaillou and N.E. Roth. 1999. Guide to using 1995-1997 Maryland Biological Stream Survey Data. Publ. No. CBWP-MANTA-EA-99-5. Annapolis, Md.: Maryland Department of Natural Resources, Monitoring and Non-Tidal Assessment Division. Miltner R.J. and E.T. Rankin. 1998. Primary nutrients and the biotic integrity of rivers and streams. Freshwater Biology 40(1): 145-158. Molles M.C. 1982. Trichopteran communities of streams associated with aspen and conifer forests: long-term structural change. Ecology 63: 1-6. Muller E., H. Decamps and M.K. Dobson. 1993. Contribution of space remote-sensing to river studies. Freshwater Biology 29(2): 301-312. 43 Naiman R.J. and H. D?camps. 1997. The ecology of interfaces: riparian zones. Annual Review of Ecology and Systematics 28: 621-658. Naiman R.J., H. D?camps, J. Pastor and C.A. Johnston. 1988. The potential importance of boundaries to fluvial ecosystems. J. N. Am. Benthol. Soc. 7(4): 289-306. Nearing M.A., R.M. Risse and L.F. Rogers. 1993. Estimation of daily nutrient fluxes to a large Piedmont reservoir from limited tributary data. Journal of Environmental Quality 22: 666-671. Neill M. 1989. Nitrate concentrations in river waters in the southeast of Ireland and their relationship with agricultural practice. Water Research 23: 1339-1355. Nikolaidis N.P., H. Shen, H.U. Heng and J.C. Clausen. 1993. Movement of nitrogen through an agricultural riparian zone ? 2. Distributed modeling. Water Science and Technology 29(3-5): 613-623. Norton B.E., J. Tuhy and S. Jensen. 1981. An approach to classification of riparian vegetation. In In-Place Resource Inventories: Principles & Practices: Proceedings of a Workshop. T.B. Brann, L.O. House IV, H.G. Lund, eds. Orono, Me: Univ. of Maine. Norton B.E., P.J. Urness and A.R. Southard. 1980. A riparian community classification study: final report. Logan, Utah: Utah State University, Dept. of Range Science. Omernik J.M., A.R. Abernathy and L.M. Male. 1981. Stream nutrient levels and proximity of agricultural and forest land to streams: some relationships. Journal of Soil and Water Conservation 36: 227-231. Omernik J.M. 1976. The influence of land use on stream nutrient levels. Publ. No. EPA-600/3-76-014. Washington, DC: U.S. EPA. Ormerod S.J., S.D. Rundle and E.C. Lloyd. 1993. The influence of riparian management on the habitat structure and macroinvertebrate communities of upland streams draining plantation forests. The Journal of Applied Ecology 30(1): 13-24. Osborne L.L., S.L. Kohler, P.B. Bayley, D.M. Dat, W.A. Bertrand, M.J. Wiley and R. Sauer. 1992. Influence of stream location in a drainage network on index of biotic integrity. Transactions of the American Fisheries Society 121(5): 635-645. Osborne L.L. and D.A. Kovacic. 1993. Riparian vegetated buffer strips in water- quality restoration and stream management. Freshwater Biology 29: 243-258. 44 Osborne L.L. and M.J. Wiley. 1988. Empirical relationships between land use/cover and stream water quality in an agricultural watershed. Journal of Environmental Management 26: 9-27. _______. 1992. Influence of tributary spatial position on the structure of warmwater fish communities. Can. J. of Fish. Aquatic Sci. 49(4): 671-681. Osmond D.L., R.W. Gannon, J.A. Gale, D.E. Line, C.B. Knott, K.A. Phillips, M.H. Turner, M.A. Foster, D.E. Lehning, S.W. Coffey and J. Spooner. 1997. WATERSHEDSS: A decision support system for watershed-scale nonpoint source water quality problems. Journal of the American Water Resources Association 33(2): 327-341. Padgett W.G., A.P. Youngblood and A.H. Winward. 1989. Riparian community type classification of Utah and southeastern Idaho. Ogden, Utah: U.S. Dept. of Agriculture, Forest Service, Intermountain Research Station. Palone R.S. and A.H. Todd. 1998. Chesapeake Bay riparian handbook: a guide for establishing and maintaining riparian forest buffers. U.S. Dept. of Agriculture, Forest Service, Northeastern Area State & Private Forestry: Natural Resources Conservation Services: Cooperative State Research, Education, and Extension Service. Perry C.D., G. Vellidis, R.R. Lowrance and D.L. Thomas. 1999. Watershed-scale water quality impacts of riparian forest management. Journal of Water Resources Planning and Management - ASCE 125(3): 117-125. Peterjohn W.T. and D.L. Correll. 1984. Nutrient dynamics in an agricultural watershed: observations on the role of a riparian forest. Ecology 65(5): 1466-1475. Petersen M.M., D.J. Somerville and R.F. Sennett. 1992. Preliminary riparian classification system for private and state-owned lands in Utah. General technical report INT 289. Ogden, Utah: U.S. Department of Agriculture, Forest Service, Intermountain Research Station. Phillips J.D. 1989. Non-point-source pollution control effectiveness of riparian forests along a Coastal Plain river. Journal of Hydrology 110: 221-237. Pierce J. and J. Johnson. 1986. Wetland community type classification of Utah. Publ. No. R4-Ecol-89-01. Ogden, Ut.: USDA Forest Service, Intermountain Research Station. Pinay G., C. Ruffini and A. Fabre. 1993. Nitrogen cycling in two riparian forests soils under different geomorphic conditions. Biogeochemistry 30: 9-29. 45 Pinay G., L. Roques and A. Fabre. 1993. Spatial and temporal patterns of denitrification in a riparian forest. The Journal of Applied Ecology 30(4): 581-591. Plafkin J.L., M.T. Barbour, K.D. Porter, S.K. Gross and R.M. Hughes. 1989. Rapid bioassessment protocols for use in streams and river: benthic macroinvertebrates and fish. Publ. No. EPA/444/4-89-011. Washington, D.C.: EPA, Office of water regulations and standards. Poff N.L. and J.D. Allan. 1995. Functional organization of stream fish assemblages in relation to hydrologic variability. Ecology 76: 606-627. Quinn J.M. and C.W. Hickey. 1990. Characterization and classification of benthic invertebrate communities in 88 New Zealand rivers in relation to environmental factors. New Zealand Journal of Marine and Freshwater Research 24: 387-409. Quinn J., A. Taylor, W. Boyce and T. Fenton. 2000. Riparian zone classification improves management in multi-land use watersheds. In Proceedings of the International Conference on Riparian Ecology and Management in Multi-Use Watersheds, 215-220. P.J. Wigington, Jr. and R.L. Beschta, eds. Middleburg, Va.: AWRA. Richards C. and G. Host. 1994. Examining land-use influences on stream habitats and macroinvertebrates ? a GIS approach. Water Resources Bulletin 30(4): 729-738. Richards C., R.J. Haro, L.B. Johnson and G.E. Host. 1997. Catchment and reach-scale properties as indicators of macroinvertebrate species traits. Freshwater Biology 37: 219-230. Richards C., G.E. Host and J.W. Arthur. 1993. Identification of predominant environmental factors structuring stream macroinvertebrate communities within a large agricultural catchment. Freshwater Biology 29(2): 285-294. Richards C., L.B. Johnson and G.E. Host. 1996. Landscape-scale influences on stream habitats and biota. Canadian Journal of Fisheries and Aquatic Sciences 53: 295-311. Roberts G. and T. Marsh. 1987. The effects of agricultural practices on the nitrate concentration in the surface water domestic supply sources of Western Europe. IAHS Publication No. 164. Roth N.E., M.T. Southerland and J.C. Chaillou. 1999. Relative and cumulative impacts of stressors on streams: 1995-1997 MBSS results. Publ. No. . Annapolis, Md.: Maryland Department of Natural Resources, Monitoring and Non-Tidal Assessment Division. 46 Roth N.E., M.T. Southerland, G. Mercurio, J.C. Chaillou, P.F. Kazyak, S.S. Stranko, A.P. Prochaska, D.G. Heimbuch and J.C. Seibel. 1999. State of the Streams: 1995- 1997 Maryland Biological Stream Survey Results. Publ. No. CBWP-MANTA-EA-99- 6. Annapolis, Md.: Maryland Department of Natural Resources, Monitoring and Non- Tidal Assessment Division. Roth N.E., M.T. Southerland, J.C. Chaillou, H.T. Wilson, D.G. Heimbuch and J.C. Seibel. 1998. Maryland Biological Stream Survey: Ecological Assessment of Non- Tidal Streams Sampled in 1996. Publ. No. CBWP-MANTA-EA-98-1. Annapolis, Md.: Maryland Department of Natural Resources, Monitoring and Non-Tidal Assessment SAS Institute Inc. 1988. SAS/STAT User?s Guide. Release 6.03 Edition. Cary, N.C.: SAS Institute Inc. Schlosser I.J. and P.L. Angermeier. 1995. Spatial variation in demographic processes of lotic fishes: conceptual models, empirical evidence, and implications for conservation. American Fisheries Society Symposium 17: 392-401. Snyder N.J., S. Mostaghimi, D.F. Berry, R.B. Reneau, S. Hong, P.W. McClellan and E.P. Smith. 1998. Impact of riparian forest buffers in agricultural nonpoint source pollution. Journal of the American Water Resources Association 34(2): 385-395. Southerland M.T., G. Mercurio, N.E. Roth, D. Strebel and J. Volstad. 2002. Using the land use-stream condition relationship to target riparian restoration. Presentation. By Versar, Inc. Columbia, Md. Steedman R.J. 1988. Modification and assessment of an index of biotic integrity to quantify stream quality in Southern Ontario. Can. J. Fish. Aquat. Sci. 45: 492 -501. Stewart J.S., D.M. Downes, L. Wang, J.A. Wier and R. Bannerman. 2000. Influences of riparian corridors on aquatic biota in agricultural watersheds. In Proceedings of the International Conference on Riparian Ecology and Management in Multi-Use Watersheds, 209-214. P.J. Wigington, Jr. and R.L. Beschta, eds. Middleburg, Va.: AWRA. Stribling J.B., B.K. Jessup, J.S. White, D.M. Boward and M. Hurd. 1998. Development of a Benthic Index of Biotic Integrity for Maryland Streams. CBWP- EA-98-3. Annapolis, Md.: Maryland Department of Natural Resources, Resource Assessment Service, Monitoring and Non-Tidal Assessment Division. Szaro R.C. and D.R. Patton. 1986. Riparian habitat classification in the southwestern United States. In Transactions of the 51st North American Wildlife and Natural Resources Conference. Washington, D.C.: Wildlife Management Institute. 47 Tabacchi E., D.L. Correll, R. Hauer, G. Pinay, A. Planty-Tabacchi and R.C. Wissmar. 1998. Development, maintenance and role of riparian vegetation in the river landscape. Freshwater Biology 40: 497-516. Taraba J.L., J.S. Dinger, L.V.A. Sendlein and G.K. Felton. 1996. Land use impacts on water quality in small karst agricultural watersheds. ASAE meeting paper 96-2086. July 14-18, 1996. Phoenix, AZ. Taylor C.M., M.R. Winston and W.J. Matthews. 1993. Fish species ? environment and abundance relationships in a Great Plains river system. Ecography 16: 16-23. Tim U.S., R. Jolly and H.H. Liao. 1995. Impact of landscape feature and feature placement on agricultural non-point-source- pollution control. Journal of Water Resources Planning and Management - ASCE 121(6): 463-470. Tufford D.L., H.N. McKellar Jr. and J.R. Hussey. 1998. In-stream nonpoint source nutrient prediction with land-use proximity and seasonality. Journal of Environmental Quality 27(1): 100-111. US EPA. 1990. The quality of our nation's water: a summary of the 1988 National Water Quality Inventory. Publ. No. EPA 440/4-90/003. Washington, D.C.: EPA, Office of Water. US Geological Survey. 1999. The quality of our nation's waters - nutrients and pesticides. US Geological Survey circular 1225. Vandervalk A.G. and R. Jolly. 1992. Recommendations for Research to Develop Guidelines for Use of Wetlands to Control Rural Non-Point Source Pollution: Created and Natural Wetlands for Controlling Nonpoint Source Pollution. R.K. Olson, ed. Washington, DC: US EPA, Office of Research and Development. Vannote R.L., G.W. Minshall, K.W. Cummins, J.R. Sedell and C.E. Cushing. 1980. The river continuum concept. Can. J. Fish. Aquat. Sci. 37: 130-137. Weil R.R., R.A. Weismiller and R.S. Turner. 1990. Nitrate contamination of groundwater under irrigated Coastal Plain souls. J. Environ. Quality 19: 441-448. Weller D.E., T.E. Jordan and D.L. Correll. 1998. Heuristic models for material discharge from landscapes with riparian buffers. Ecological Applications 8(4): 1156- 1169. Whigham D.F., C. Chitterling and B. Palmer. 1988. Impacts of freshwater wetlands on water quality ? a landscape perspective. Envir. Management 12(5): 663-671. 48 Wichert G.A. and D.J. Rapport. 1998. Fish community structure as a measure of degradation and rehabilitation of riparian systems in an agricultural drainage basin. Environmental Management 22(3): 425-443. Winward A.H. 1984. An ecosystem classification for intermountain riparian areas. In Range Watersheds, Riparian Zones and Economics: Interrelationships in Management and Use: Proceedings, 1984 Pacific Northwest Range Management Short Course. Corvallis, Or: Oregon State University. Wohl N.E. and R.F. Carline. 1996. Relations among riparian grazing, sediment loads, macroinvertebrates, and fishes in three central Pennsylvania streams. Can. J. Fish. Aquat. Sci. 53(Suppl.1): 260-266. Youngblood A.P., W.G. Padgett, G. Wayne and A.H. Winward. 1985. Riparian community type classification of eastern Idaho - western Wyoming. Publ. No. R4- ECOL-85-01. Ogden, Ut.: U.S. Dept. of Agriculture, Forest Service, Intermountain Forest Range Experimental Station. Zinecker E. 2001. Poster at the Chesapeake Bay Watershed Restoration: Riparian and Wetland Stewardship Conference. Baltimore, Maryland. September 24-26, 2002 49 CHAPTER 2 ? CHARACTERIZATION OF AGRICULTURAL RIPARIAN BUFFER LANDSCAPES IN THE COASTAL PLAIN AND PIEDMONT PHYSIOGRAPHIC REGIONS OF MARYLAND ABSTRACT This paper presents the characterization of riparian buffers in agricultural settings within the stream channel, buffer, and adjacent agricultural field on both sides of the stream. The survey was designed to use ?easily available data? - fieldwork followed a rapid assessment approach and other landscape data were obtained from publicly available documents and web-based files. The results indicated that almost all sites (93.2%) had some type of buffer and most buffers were forested. Most grass buffers were in the Piedmont physiographic region. Most unbuffered sites were associated with Piedmont pastures where animals had access to the stream, and demonstrated high streambank erosion. Buffers were as prevalent and as wide on the Eastern Shore as the western shore. Buffer width was widely variable. Although the statewide average buffer width for all buffer types was approximately 49 meters, almost 50% of buffers may not be detected by remotely sensed (RS) data with a resolution of 30 meters. This problem was more pronounced in the Piedmont. These 50 results present the first comprehensive structural characterization of Maryland's Coastal Plain and Piedmont agricultural riparian buffers and their landscapes. INTRODUCTION In 1983, concern about worsening water quality in the Chesapeake Bay led to the formation of the Chesapeake Bay Program (CBP), a multi-jurisdictional cooperative effort to restore the nation's largest estuary. The Chesapeake Executive Council set restoration of the Bay's "living resources" as a primary goal in 1987. To achieve that goal, it called for a number of actions to improve water quality, including a 40% reduction in the amount of nitrogen and phosphorus entering the Bay (Alliance for the Chesapeake Bay, 1996a). Efforts to implement these actions focused first on coastal tributaries of the Bay. By 1994, however, attention had expanded to the complete watershed, with the development of a set of goals and actions to increase the focus on riparian stewardship and to enhance efforts to conserve and restore riparian forest buffers (Alliance for the Chesapeake Bay, 1996b). Riparian forest buffers (RFB) were singled out in part for their nutrient reduction capabilities, but also because they are reminiscent of the natural landscape in which the stream channel and stream ecosystems of the Bay watershed had evolved. Background Although there is general agreement on the qualitative benefits of riparian buffers, attempts to quantitatively relate riparian information, watershed information, aquatic conditions and water quality have met with mixed success and very few 51 studies have been conducted at the watershed scale (Gregory et al., 1991; Whigham et al., 1988; Osborne and Wiley, 1988; Wissmar and Beschta, 1998; Tufford et al., 1998; Perry et al., 1999). Because physical variation in riparian areas may aid or prevent the interception and transformation of pollutants (Hill, 1996; Weller et al., 1998), understanding the underlying natural variation in riparian buffer landscapes can augment interpretation of the impact of land use patterns on aquatic conditions. Without a comprehensive description of site geology, vegetation, hydrology, and soils, research findings are hard to generalize or extrapolate to other sites (Correll, 2000). Riparian systems have received much attention for their role in stream ecosystem function and for their capacity to mitigate non-point-source (NPS) pollution (Lowrance et al., 1984, 1997; Correll, 1996; Barling and Moore, 1994). Streamside forests are complex ecosystems that help provide optimum food and habitat for stream communities. They can be effective in removing excess nutrients and sediment from surface runoff and shallow groundwater and shading streams to optimize light and temperature conditions for aquatic plants and animals. Streamside forests also ameliorate the effects of some pesticides and directly provide dissolved and particulate organic food needed to maintain high biological productivity and diversity in the adjoining stream (Maryland DNR website, 2002). In 1998, the University of Maryland, Maryland Cooperative Extension, and the Forestry Division of Maryland's Department of Natural Resources joined together to conduct a fieldwork-based survey of agricultural riparian buffers in Maryland's Chesapeake Bay watershed. This survey provided the first extensive, ground-truthed data set describing a large set of agricultural riparian buffers, the surrounding land, 52 and to a lesser extent, the land and streams within the buffers. Because agencies are typically restricted to ?rapid assessment?-type surveys, this survey described these landscapes using ?easily obtainable? data. Field surveys were designed to take less than one hour per site, and map and other landscape data were obtained from publicly- available documents and web-based files. Therefore, the methodology used in the survey can be reproduced and the results expanded by state agencies. The purpose of this paper is to present the characterization of riparian buffers and their agricultural settings within the Coastal Plain and Piedmont physiographic regions of Maryland. These results are of theoretical use to scientists investigating the functional role of buffers in water quality protection, as well as practical use to the agencies that administer the effort to establish riparian forest buffers in Maryland. The research also provides information of value to field staff that provide farmers with information about riparian forest buffers and cost-share programs that promote them. While these results are specific to Maryland, the general findings are of use to other locations throughout the United States where the establishment of forest buffers is considered as a water quality measure and remotely sensed data are used to conduct buffer surveys. Objectives 1. To develop a set of easily obtainable landscape characteristics that can describe the agricultural riparian landscape. 2. To characterize agricultural riparian landscapes in the Coastal Plain and Piedmont physiographic regions of Maryland in terms of land use, buffer characterization 53 (distribution, vegetation type, width), topography, surface hydrology, groundwater hydrology and channel morphology. 3. To identify correlations among structural measures of riparian landscape characteristics. 4. To evaluate the usefulness of remotely sensed data for riparian buffer surveys. METHODS Overall summary of methods The agricultural riparian landscape was conceptualized as the stream channel, buffer, and adjacent agricultural field (Figure 1). Characterization of this landscape was performed at 209 sites in eight Maryland watersheds. Two scales of data were considered. Information about the site?s contributing watershed provided insight about cumulative hydrologic impacts and sources of agricultural NPS pollution. These data were developed from publicly available Geographic Information System (GIS) files. Site characterization provided ground-truthed information about the local riparian landscape below the resolution of RS data. These data were obtained through on-site surveys. Sample site selection A sample frame of streams and surrounding properties was developed to characterize riparian landscapes of the Maryland Chesapeake Bay watershed. The survey sample design (Cochran, 1977) required that all watersheds be predominantly agricultural, with minimum urban development. Watersheds were selected from the 54 Coastal Plain and Piedmont physiographic regions. Maryland's Appalachian Plateau was not sampled due to fiscal constraints. Because buffers are agricultural Best Management Practices (Novotny and Olem, 1994), they are voluntary measures implemented by the landowner, and their presence, width, and type are functions of agricultural culture as well as environmental conditions. The Eastern Shore of the Chesapeake Bay has a distinct and unique agricultural community. Therefore, the sampling scheme was stratified by physiographic region and "shore" (i.e., "Eastern" and western shores of the Chesapeake Bay) to reflect both physical and cultural influences. The Pocomoke was deliberately included due to interest in Pfisteria piscicida impacts (Burkholder and Glasgow (1997). Figure 2 shows the locations of the eight watersheds that were selected. Figure 1. Schematic of the agricultural riparian landscape as defined for the 1998 University of Maryland Agricultural Riparian Buffer Survey. 55 Figure 2. Map of the University of Maryland 1998 Agricultural Riparian Buffer Survey sample watershed locations superimposed over 16- digit Hydrologic Unit Code watersheds as defined by USGS. A total of 207 sites were surveyed. Maryland's Department of Natural Resources PropertyView database lists all property owners. PropertyView was parsed to identify agricultural property owners adjacent to blue-line streams. Properties were drawn randomly without replacement from the database. An original list and a replacement list of the same size were drawn at the same time; field workers were instructed to make all reasonable attempts to contact properties on the first list before starting on the replacement list. All sites were in some type of agricultural or forested land use and located on one or both sides of a stream or ditch. Figures 3 a-h present the forested area and research site locations for each of the sampled watersheds. The boundaries of each sample watershed were determined by aggregating all 16-digit hydrologic unit areas that contained at least one survey site in that stream system. Figures are not shown to scale - see Figure 2 for an appreciation of the variation in 56 size of the watersheds (25.2 ha - 218.3 ha). Site locations are approximate and unlabelled to preserve confidentiality of site owners. Figure 3. 1998 University of Maryland Agricultural Riparian Buffer Survey sampled watersheds, streams, forested areas and sample sites. . survey site, grey = forested area Figure 3a. Calvert watershed in Calvert county (2,520 ha, 10 sites) 57 Figure 3b. Cattail Creek watershed in Howard County (6,740 ha, 20 sites) Figure 3c. Little Elk watershed in Cecil County (5,170 ha, 13 sites) 58 Figure 3d. German Branch watershed in Queen Anne County (5,590 ha, 43 sites) Figure 3e. Gunpowder watershed in Baltimore County (12,880 ha, sites) 59 Figure 3f. Monocacy watershed in Frederick County (21,580 ha, 36 sites) Figure 3g. Pocomoke watershed in Wicomico and Somerset counties (21,830 ha, 30 sites) 60 Figure 3h. St. Mary?s watershed in Calvert County (10,730 ha, 16 sites) Site characterization Each survey site consisted of the stream, riparian zone and adjacent fields along both sides of a 30 to 90 meter stream reach. Field and map measurements fell into six key categories (Giles and Trani, 1999): location, topography, surface and groundwater hydrology, land use and stream channel morphology (Table 4). If a buffer was present, vegetation and width measurements were also taken. Two scales of data were considered. Site characterization provided ground- truthed information about the local riparian landscape below the resolution of remotely sensed data. These data were obtained from on-site surveys. Because one of the directives of the University of Maryland Buffer Survey Project was to collect "easily obtainable data", fieldwork was restricted to a suite of structural landscape measurements that could be obtained in a single visit of less than one hour 61 (Appendices 1 and 2). Information at the site watershed scale provided insight about cumulative impacts and sources of agricultural NPS pollution. These data were developed from soil surveys, USGS topographic maps, and publicly available EPA, USGS and state GIS files (Appendix 3). Land use provided information about the source of agricultural NPS pollutants to the adjacent stream (Basnyat et al., 1999; Richards et al., 1996). Because many decisions are made at the watershed scale using surface water quality as the regulatory criteria of interest, both adjacent land use at the site (crop, pasture, forest, other) and overall land use in the site's contributing watershed (percent forest, percent agriculture, percent crops, Confined Animal Feeding Operation area) were considered. Watershed land use distribution was obtained from Maryland Office of Planning GIS land use files. Land use at the site was recorded from visual observation as buffer presence and adjacent field use on either side of the stream. Buffer characterization included buffer width, fragmentation (or bypass) and vegetation measures (USDA, 1999). The riparian buffer was defined as an area of land between a stream channel and an agricultural land use area. If the adjacent field was cropped up to the side of the stream, or was a pasture allowing full animal access to the stream, the buffer width was recorded as 0 feet. However, in two watersheds, several sites were characterized by row crop extending to the top incision of a broad channel with gently sloping sides thickly covered in vegetation. For these sites, the width of vegetation on each side of the channel was considered as an ?in-channel? buffer. In the Piedmont watersheds, many sites were characterized by pasture with sufficient vegetation to restrict animal access to the stream to specific channelized 62 routes. These sites were considered to have a buffer. The maximum width for a riparian buffer was defined as 305 meters (1000 feet) (personal communication, R. Tjaden). In all cases, buffer width was reported as the average width on one side of the stream. Even a wide, densely vegetated buffer can be rendered ineffective if the polluted water is able to bypass or short-circuit travel through this zone (Skaggs and Chescheir, 1999). Therefore, an attempt was made to ascertain the presence of drain tiles at the survey sites, but this proved unsuccessful. A subjective measure of gullying was estimated to provide a measure of buffer bypass at the site-specific scale. Measures of buffer fragment length (km along the stream) were taken from watershed aerial photos to provide a measure of buffer bypass at the stream reach scale. Buffer vegetation was described by tree type (predominance of hardwood or pines), Diameter at Breast Height (DBH) and Basal Area (square feet of trees in an area of observation). Topography affects the transport of surface runoff, which carries sediment and particle-bound pollutants, such as phosphorus. Site topography was described by riparian buffer and adjacent field slope. These were taken as single measurements in the field. Watershed topography was characterized by a topographic index, calculated as maximum topographic relief divided by the area of the 16-digit hydrologic unit area within which the site was located. Watershed elevation data were obtained from GIS files. Surface water hydrology affects the transport of pollutants to and by the stream. Watershed drainage density (area of land drained per length of stream) was given as a measure of hydraulic loading to the stream complex. Sinuosity (tortuosity) was calculated as the length of the longest continuous channel from the uppermost 63 headwater to the bottom of the watershed, divided by the straight distance between these two points. Stream gradient and order were taken from USGS 7.5-minute topographic quadrangle maps. Stream velocity was reported as the average of three measurements of surface velocity. Groundwater hydrology affects the transport and transformation of soluble pollutants, such as nitrogen. Control of agricultural nitrate pollution is primarily accomplished by plant use, denitrification, and dilution (Cirmo and McDonnell, 1997; Hill, 1996). Denitrification capacity is directly related to water table depth (Gold et al., 2000). Dilution is affected by infiltration capacity, which was characterized by hydraulic conductivity (K) and drainage class (good or poor). Hydraulic conductivity was reported as the median value of the range given in the soil survey. Depth to the layer of minimum hydraulic conductivity was reported to indicate any limiting layer for shallow groundwater movement. Hydraulic conductivity and depth to the layer of minimum hydraulic conductivity were reported as the median value of the range given in soil surveys. Stream channel morphology has been shown to be affected by buffers (Schlosser and Karr, 1981). Channel size, width: depth ratio, and subjective measures of bank erosion were reported to provide indications of the erosive power of the stream. Channelization or straightening of the stream was evidence of deliberate channel disruption. All measures of stream channel morphology were determined by visual observation. Tables 1 and 2 present the site descriptors and range of values obtained through the combination of fieldwork, map work and GIS work. Site survey measurements 64 that were taken as multiple replicates (channel width and depth, bank angles, stream velocity, riparian buffer width) were reduced to average values. Therefore, these 34 parameters represented over 150 individual measurements characterizing 207 sites. Table 1. Descriptors used to characterize the agricultural riparian landscape in the 1998 University of Maryland Agricultural Riparian Buffer Survey. Descriptor Category Riparian Landscape Descriptor Units Range of Values Survey watershed % agriculture2 Survey watershed % forest2 % % 38.2 - 76.1 26.1 ? 45.0 Site contributing watershed % agriculture2 % 36.9 ? 100.0 Site contributing watershed % crops2 % 26.3 ? 100.0 Land Use Adjacent land use1 crop/past/ forest/other Buffer width1 m 0 ? 304.8 Length of buffer fragment3 km 0.02 - 6.9 Forest buffer DBH1 cm 3.8 - 84.3 Forest buffer basal area1 m2 0 - 15.5 Buffer Char- acter- ization Shrub height1 m 0.3 - 6.1 Topo- graphy Watershed topographic index 2 Riparian buffer slope1 Adjacent field slope1 Ratio riparian buffer : adjacent field slope1 m/m2 % % unitless 0.9 - 30.1 -2.5 - 62.3 0 - 37.5 -0.11 - 35.2 Watershed drainage density2 km2/km 1.02 - 1.25 Mainstem sinuosity2 km/km 1. 1- 1.4 Stream gradient1 % 0 ? 4.0 Stream velocity1 m/sec 0 ? 1.0 Surface Hydrology Stream order 1,2 unitless 1 - 4 1 from on-site field work 2 from GIS work 3 from soil survey 65 Table 1. (continued) Descriptors used to characterize the agricultural riparian landscape in the 1998 University of Maryland Agricultural Riparian Buffer Survey. Descriptor Category Riparian Landscape Descriptor Units Range of Values Ground- water Riparian buffer avg. depth to water table Adjacent field avg. depth to water table m m 0 - 10.7 0 - 10.7 Hydrology3 Riparian buffer surface K * Adjacent field surface K * cm/hour cm/hour 0.03 - 15.2 0.1 - 15.2 Riparian buffer minimum K * Adjacent field minimum K * cm/hour cm/hour 0.03 - 15.2 0.05 - 15.2 Riparian buffer soil drainage class Adjacent field soil drainage class variable/poor/ good/ excellent variable/poor/ good/ excellent Riparian buffer depth of minimum K * Adjacent field depth of minimum K * m m 0.03 - 15.5 1.7 - 15.1 Channel width m 0.4 - 21.6 Channel depth m 0.2 - 5.6 Channel cross-sectional area Mm 0.1 ? 50.5 Channel width : depth ratio m/m 0.27 ? 30.0 Substrate mud/sand/ gravel/ cobble Channel shape natural/ triangular Channel straightened? yes/no Presence of berms yes/no Degree of gullying low/moderate /severe Average bank angle (degrees) degrees 5.7 ? 90.0 Degree of bank erosion low/moderate/ severe Channel Morphology1 Livestock access yes/no * K = hydraulic conductivity 1 from on-site field work 2 from GIS work 3 from soil survey 66 Table 2. Values of categorical descriptors used to characterize the agricultural riparian landscape in the 1998 University of Maryland Agricultural Riparian Buffer Survey. Buffer Characterization Buffer tree type hardwood 85.94 mix 13.28 pine 0.78 Adjacent field soil drainage class excellent 11.06 good 58.65 poor 13.46 variable 16.83 excellent good poor variable Groundwater Hydrology Riparian buffer soil drainage class 5.77 18.75 67.31 8.17 boulder cobble sand/ gravel sand mud Channel substrate 2.5 24 32 13.5 28 natural triangularChannel shape 91.22 8.78 yes no Channel straightened 27.18 72.86 yes no Berms 10.73 89.27 high medium low noneGullying 3.11 7.77 54.92 34.2 high medium low noneBank erosion 21.52 53.8 22.15 2.53 yes no Channel Morphology Livestock access 20.39 79.61 67 Data analysis The data were analyzed to provide univariate measures (mean, standard deviation, and coefficient of variation) and correlation measures to describe the state of Maryland's agricultural buffers and for use in further research. All analysis was performed using SAS software. Watersheds were of unequal sizes and did not have the same number of sites. Therefore the SAS Proc Mixed procedure (SAS/STAT User?s Guide, 1988) was used to develop ANOVA to identify regional variation among quantitative descriptors. RESULTS AND DISCUSSION Through a combination of fieldwork, map work and GIS work, over 150 individual measurements and values were obtained for each survey site. Many of these values were aggregated into average values, so that analysis was performed on a suite of approximately 30 descriptors (Tables 1 and 2). These data provided information about the state of agricultural riparian buffers in Maryland and their landscape settings, as well as relationships between buffer characteristics and physiographic and cultural settings. Average statewide values and regional values were determined for all quantitative riparian landscape descriptors by means of ANOVA (Table 3). Because there were unequal numbers of sites among watersheds, a mixed model was used. All F-tests and t-tests performed on watershed and site values of landscape descriptors 68 used  = 0.05. All statewide average values are presented in Table 3, and regional values are presented if the F-test showed a significant difference between regions. Table 3. Statewide and regional values and standard deviations for riparian landscape descriptors in the 1998 University of Maryland Agricultural Riparian Buffer Survey. Descriptor Category Riparian Landscape Descriptor State ** mean (SD) Coastal Plain mean (SD) Piedmont mean (SD) Survey watershed % agriculture 55.8 (14.0) Survey watershed % forest 32.2 (15.9) 54.9 (7.3) 23.3 (5.4) Adjacent land use % crop 44.2 (21.2) Adjacent land use % pasture 29.4 (23.2) Adjacent land use % forest 15.3 (10.6) Land Use Adjacent land use % other 11.1 (7.4) Buffer Width Total buffer width (m) 48.8 (61.6) Zoned buffer width (m) 57.3 (64.0) Forested buffer width (m) 58.2 (61.9) Grass buffer width (m) 25.9 (42.7) 13.4 36.0 Length buffer fragment (m) 856 (710) 1263 (274) 469(269) Forest buffer DBH 36 (11) Buffer Condition Forest buffer basal area 5.1 (2.4) 6.5 (0.4) 3.8 (0.4) Topo- graphy Watershed topographic index 12.1 (9.7) 3.9 (3.9) 19.6 (3.0) Riparian buffer slope 8.7 (9.3) 6.5 (1.3) 10.0 (1.3) Adjacent field slope 5.3 (4.3) Watershed drainage density 1.1 (0.3) Mainstem sinuosity 1.2 (0.1) Stream gradient (%) 1.1 (0.9) Stream velocity (mps) 0.3 (0.2) 0.2 (0.1) 0.3 (0.1) Surface Hydrology Stream order 1.8 (0.9) 1.6 (0.1) 2.0 (1.0) 69 Table 3. (continued) Statewide and regional values and standard errors for riparian landscape descriptors in the 1998 University of Maryland Agricultural Riparian Buffer Survey. Riparian water table depth (m) 1.1 (1.3) Field water table depth (m) 3.2 (1.6) Riparian surface K (cm/hr) * 4.1 (1.3) Field surface K (cm/hr) * 4.1 (1.3) Riparian minimum K (cm/hr) * 3.1 (2.8) 1.5 (0.5) 4.3 (0.5) Field minimum K (cm/hr) * 4.1 (7.6) Riparian minimum K depth (m) * 7.8 (3.2) 8.8 (0.3) 7.56 (0.3) Ground- water Hydrology Field minimum K depth (m) * 6.1 (3.2) 7.2 (0.7) 4.6 (0.6) Channel width: depth ratio 4.9 (6.0)Channel Morph- ology Average bank angle 46.4 (14.7) * K = hydraulic conductivity ** Where there was no significant difference in regional values, only the statewide average value is given. Maryland's agricultural riparian buffers: distribution by type Riparian buffers were found to be composed of trees, shrubs, grass, and in- channel vegetation. Although shrub buffers were distinguished from forest buffers by visual inspection, shrub buffers had an average height of 1.9 meters. Therefore, although the differentiation between shrub and forest buffers was subjective, it was clear. In-channel vegetation was considered part of the buffer if there was very little off-channel buffer and the bank vegetation was a large proportion of the land between the field and the stream. This could be a considerable width in sites on the lower Eastern Shore with wide channels and gently sloping banks (average width of in- channel vegetation was 2.1 m, with a maximum value of 8.8 m). Therefore four types 70 of buffers were recognized: forested, "zoned" (forest and grass, as described in Lowrance et al., 1997), grass, and "other" (any other combination of trees, grass, shrubs and in-channel vegetation). Almost all sites (93.2%) had some type of buffer. Forest buffers were the predominant type (52%) and zoned buffers were fairly rare (7%). Statewide, there were approximately three times as many forested buffers as grass buffers. The distribution of buffer type varied by physiographic region. Most Coastal Plain buffers were forested (50%) and most Piedmont buffers were grass (50 %). These results were in marked contrast to the popular belief that people on the Eastern Shore of Maryland are reticent to install forested buffers. Ninety-six percent of buffered sites were buffered on both sides of the stream. Unbuffered Piedmont sites were primarily associated with pastures allowing livestock access to the stream. Maryland's agricultural riparian buffers: width The average statewide buffer width (one side of the stream) over all buffer types was 48.8 meters and buffer widths ranged between 0.6 m and the defined maximum of 305 meters. An F-test performed on buffer width vs. physiographic region and shore demonstrated no significant difference in buffer width for overall buffer width, forested buffer width and zoned buffer width (Figure 5). This result was in contrast to the popular belief that buffers on the Eastern Shore are narrower than buffers elsewhere in the state. (Despite associated water quality problems, the Pocomoke watershed had the widest buffers on the Eastern Shore, and the largest amount of forested area of any sampled watershed.) Results of the F-test performed 71 on buffer width vs. physiographic region and shore demonstrated that Piedmont grass buffers were significantly wider than Coastal Plain grass buffers (Coastal Plain: 13.4 + 6.7 m, Piedmont: 36.0 + 6.0 m). Figure 4. Average agricultural riparian buffer widths by type, as determined by the 1998 University of Maryland Agricultural Riparian Buffer Survey. * Pd = Piedmont, CP = Coastal Plain Maryland's agricultural riparian buffers: distribution of buffer width classes A significant number of buffers surveyed were quite narrow, and this was especially true for grass buffers (Figure 5). Sixteen percent of all buffers and 38% of grass buffers had a total width (both sides of the stream) of less than 10 meters. Sixty- five percent of these buffers were in the Piedmont. This could be problematic for buffer surveys performed using remotely sensed data, which often have a resolution of 10 or 30 meters (Hewitt, 1990; Muller et al., 1993). At a satellite photo resolution of 0 50 100 m et er s buffer width 48.8 57.9 57.3 36 13.4 all buffers forest zoned grass - Pd grass - CP 72 10 meters, 28% of all buffers, 20% of Piedmont buffers and 38% of grass buffers would not be recognized. For RS data with a resolution of 30 meters, almost 50% of buffers would not be included. These results indicate that surveys done with RS data (Day et al., 1996) may drastically underestimate the number and extent of agricultural buffers in Maryland. Figure 5. Histogram of agricultural riparian buffer widths for forested, grass buffers and all buffer types combined, as determined by the 1998 University of Maryland Agricultural Riparian Buffer Survey. Maryland's agricultural riparian buffers: fragmentation Riparian buffer fragmentation is a concern throughout the Chesapeake Bay watershed (Alliance for the Chesapeake Bay, 1998). The average buffer fragment length along the stream channel was 885 m. A t-test performed on buffer fragment 0-9 20 - 29 40 - 49 60 - 69 80 - 89 10 0+ 0 10 20 30 40 50 60 % length groups (m) total forest grass 73 length indicated that Coastal Plain Eastern Shore buffers were significantly longer than Central Maryland Piedmont Maryland's agricultural riparian buffers: vegetation The vegetative condition of forest buffers was quantified using DBH, basal area, and tree type. The statewide average DBH was 36 + 15 cm. A t-test performed on DBH vs. physiographic region indicated no significant difference. However, a t- test performed on basal area vs. physiographic region showed the Coastal Plain buffers to have significantly larger basal area (Coastal Plain: 6.5 + 0.4 m2, Piedmont: 3.8 + 0.4 m 2). Therefore, buffers were seen to have fairly small trees across the state, but buffers had higher biomass per area in the Coastal Plain. This may reflect native fertility or species differences. Both measures of forest vegetation showed low correlations with buffer width. Buffer tree types were distinguished as hardwood, pine, and mixed hardwood and pines, but no pine buffers were found. Hardwood buffers predominated in both the Piedmont and Coastal Plain (78% of Coastal Plain sites and 93% of Piedmont sites) but most mixed forests were found in the Coastal Plain. The Pocomoke watershed was the only representative of the lower Easter Shore, and had a distinctly different tree composition from all other sampled watersheds, with 73% of its forested buffers as mixed hardwood and pines. All other watersheds had less than 25% mixed forests. Agricultural riparian landscapes Land use. The sampling protocol deliberately selected watersheds with predominantly agricultural land use. The average extent of agricultural land use among sampled watersheds was 56% with average forest area of 32%, as determined 74 from the Maryland Office of Planning land use/land cover GIS files (Figure 6). An F- test performed on percent agriculture of the survey watershed vs. physiographic region (n = 8) indicated that there was no regional difference in amount of agriculture for surveyed watersheds. Visual inspection of the land use maps suggested that most forested area in each watershed was associated with riparian areas. In heavily agricultural watersheds, such as the German Branch, riparian buffers constitute essentially all of the forested area. A t-test performed on the forested area vs. shore did not indicate a significant difference between forested area on the western and Eastern shores. Figure 6. Statewide and regional land use patterns for the 1998 University of Maryland Agricultural Riparian Buffer Survey watersheds. There were strong regional differences in the agricultural practices within Maryland (Figure 7). Although on a statewide basis there was more cropping (63% of sites) than pasture (28% of sites); most pasture sites were found in the Piedmont (79% of pasture sites). Whereas the Coastal Plain sites were primarily row crops (80% of 0 50 100 other 12.0 7.38 16.62 % forest 32.2 40.93 23.56 % ag 55.8 51.69 59.82 State C.Pl. Pd 75 sites), Piedmont watersheds had an almost equal split of crop and pasture land usage (48% crops: 42% pasture). Sixty-four percent of sites had the same type of land use on both sides of the stream. Figure 7. Agricultural land use patterns, as determined by the 1998 University of Maryland Agricultural Riparian Buffer Survey. * P = Piedmont, CP = Coastal Plain Topography. Topographic differences between the Coastal Plain and Piedmont physiographic regions are well documented. However, at the watershed scale, only topographic index showed a significant regional difference in landscape relief (Coastal Plain = 3.9 + 3.9 m/m2x106, Piedmont = 19.6 + 3.0 m/m2x106). At the site scale, only buffer slope showed a significant difference between physiographic regions (Coastal Plain = 6.5, Piedmont = 10.0). Surface Hydrology. At the watershed scale, no regional differences were seen. However, the topographic difference between the regions was reflected at the site 0% 25% 50% 75% 100% forest 9 7 10 pasture 28 13 42 crop 63 80 48 state CP P 76 scale by significant difference in average regional values for stream gradient (Coastal Plain = 0.91, Piedmont = 1.42) and velocity (Coastal Plain = 0.2, Piedmont = 0.3). Channel morphology. The F-test (n = 8) indicated no significant difference between the sinuosity of the longest continuous stream channels in the two physiographic regions. However, at the site scale, 87% of sites with straightened streams were in the Coastal Plain. The apparent conflict of these results illustrates the effect of scale. Many larger agricultural streams on Maryland?s Eastern Shore were straightened, or channelized, in the 1930's and 40's to improve drainage and reduce the threat of flooding. However, survey sample sites were primarily on low-order streams. Therefore, survey data reflect straightened low-order streams and constructed drainage ditches. Another indication of channel engineering was a triangular shape. Although triangular channels were rare (9% of all sites), they were concentrated in two watersheds (25% of German Branch sites and 13% of Monocacy sites). This result suggests that straightened stream reaches are concentrated in local areas of both physiographic regions. Groundwater hydrology. No F-tests on measures of groundwater hydrology in the field adjacent to the riparian buffer showed a significant difference between regions. However, F-test on riparian buffer water table vs. physiographic region did not show a significant difference and a t-test indicated that average riparian buffer water table was significantly higher than average field water table. The average water table within the buffer (1.1 m) was at the lower limit of the rooting depth for trees, and lower than the rooting depth for grass. These results suggest that the primary mechanism of reduction of groundwater pollution in these agricultural buffers may not 77 be plant uptake. The average depth to the riparian water table in the Pocomoke watershed (10 cm) was higher than all other watersheds (32 cm to 4 m). An F-test indicated that buffer K was significantly higher in the Coastal Plain than in the Piedmont. This suggests that one may expect Coastal Plain buffers to infiltrate a greater volume of water and infiltrate more rapidly, reducing surface runoff to a greater extent. The F-test on depth of minimum K for buffers did not show a regional difference. Buffer soil drainage was poor for 67% of sites, which was consistent with riparian wetland hydrology. Field soil drainage was essentially the opposite, with 69% of sites in the good/excellent categories. These results are consistent with the agricultural engineering history of this land to improve drainage in order to increase production. Poor productivity in riparian areas due to poor drainage is one historical reason buffers have been allowed to grow in an agricultural region. Comments on the correlation matrix Correlations were analyzed for the full data set and for each physiographic region separately (Appendix 4). Many measures showed different correlation values for the two physiographic regions, but plots of the data showed that these values were misleading due to extreme scatter in the data. The strongest correlation value was that of the relationship between buffer width and length of buffer fragment (R = 0.23), but this was almost entirely due to Coastal Plain sites (R = 0.47). Overall correlations for buffer width and topographic measures were weak (all R < 0.23) but the Piedmont showed a fairly strong relationship between buffer width and field slope (R = 0.40). There was no strong correlation between buffer width and stream order (the strongest 78 value was R = 0.31 for forested buffers in the Piedmont). Relationships between buffer width and measures of groundwater hydrology were also fairly weak. The strongest relationship was between width of Piedmont forested buffers and the depth of the least hydraulically conductive soil in the riparian buffer (R = 0.31). Correlation coefficients between buffer width and measures of buffer vegetation (Rbasal area = -0.02, RDBH = 0.005) indicated that they were not associated. These indicators of buffer condition were only weakly correlated with all other buffer and landscape measures (all R < 0.23). The lack of strong correlations indicates that landscape measures are fairly independent at the site-level riparian landscape scale. These results indicated that although strong correlations may be seen at larger scales, such as basin and physiographic region, these small landscapes are highly variable. Buffer impact on streambank erosion A strong correlation (R = 0.90) was seen between livestock access to the stream and channel bank erosion (Figure 8). Almost all unbuffered sites were pasture sites where animals were allowed direct access to the stream and showed high streambank erosion. Areas with dense cropping that had buffers showed low streambank erosion. Statewide, 20% of the sampled sites allowed livestock access to the stream. Ninety-five percent of these sites were in the Piedmont. Statewide, 90% of the sites with high channel bank erosion were in the Piedmont. Only 5% of sites with high erosion were in the Coastal Plain, even though there was heavy cropping (80% of Coastal Plain sites). These data indicate that current cropping best management practices (BMPs) are effectively minimizing streambank erosion but that current pasturing practices are not. 79 Figure 8. Percent of sites with livestock access to the stream vs. percent of sites with high erosion, as determined by the 1998 University of Maryland Agricultural Riparian Buffer Survey. SUMMARY Most sampled sites had some sort of riparian buffer and most buffers were forested. There were very few zoned buffers in Maryland, and most grass buffers were in the Piedmont physiographic region. Buffer width was widely variable. The correlation between buffer width and buffer fragment length was quite low, which suggested that buffer width may be a function of the individual property owner. The statewide average buffer width for all buffer types was approximately 49 meters. A t- test performed on buffer width vs. type indicated that forested buffers were significantly wider than grass buffers. A t-test performed on buffer width vs. shore indicated that buffers were as wide on the Eastern Shore as the western shore. A t-test 0 10 20 30 40 50 60 0 10 20 30 40 50 60 % sites with high erosion in each of the 8 sampled watersheds % sit es w ith liv es to ck a cc ess to st re a m in ea ch o f t he 8 sa m pl ed w a te rs he ds 80 performed on buffer fragment length vs. shore indicated that the Eastern Shore had longer buffer fragments along the stream. These results present the first comprehensive structural characterization of Maryland's Coastal Plain and Piedmont agricultural riparian buffers and their landscapes. There are two major strengths of these results. Ground-truthed information is provided at the site-specific scale. The site-specific scale of these field- truthed data revealed that a significant portion of buffers were less than 10 meters wide, especially in the Piedmont. With an overall statewide average buffer width of approximately 49 meters, almost 50% of buffers may not be detected by RS data with a resolution of 30 meters. It can be reasonably concluded that surveys using remotely sensed data have significantly underestimated the degree of buffering in Maryland. This also has implications for the accuracy of modeling studies of buffer impact on nutrient mitigation to agricultural streams, such as Richards et al., (1996), Tufford et al., (1998), and Basnyat et al., (1999), in which buffer impact on instream nitrate was modeled using a GIS approach based on remotely-sensed data of buffer extent. In addition to accurate representation of the site-specific buffer landscape, this research has presented information at the 16-digit watershed scale in order to discuss these buffers in the context of a larger landscape. Weaknesses of the data include the subjective nature of some of the qualitative variables (level of erosion, degree of buffer gullying). In addition, although characterization of buffer bypass is critical for watershed-level analysis of buffer function, the only measure successfully obtained was a measure of buffer fragmentation (length of buffer fragment). 81 These results also provide objective information for further analysis of buffer implementation patterns among different farming cultures. The belief has long been held that Maryland's Eastern Shore has few forested buffers and a significant percentage of unbuffered sites. These results refute this belief, in that the number and width of forested buffers on the Eastern Shore were not significantly different from buffers elsewhere in the state. 82 CITED REFERENCES Alliance for the Chesapeake Bay. 1996a. Final Report Submitted to the Chesapeake Bay Executive Committee. Baltimore, Md.: Riparian Forest Buffer Panel. Alliance for the Chesapeake Bay. 1996b. Riparian Forest Buffers. White paper. Annapolis, Md.: Chesapeake Bay Program. Alliance for the Chesapeake Bay. 1998. Readiness to sacrifice forest for sprawl is almost stumping. Bay Journal 8(1). Baker M., M. Wiley and P. Seelbach. 2000. Predicting spatial variation in riparian ecosystem structure and function. Presented at the International Conference on Riparian Ecology and Management in Multi-Land Use Watersheds, Portland, Oregon. August 28-31. American Water Resources Association. Barling R.D. and I.D. Moore. 1994. Role of buffer strips in management of waterway pollution: a review. Environmental Management 18(4): 543-558. Basnyat P., L.D. Teeter, K.M. Flynn and B.G. Lockaby. 1999. Relationships between landscape characteristics and nonpoint source pollution inputs to coastal estuaries. Environmental Management 23(4): 539-549. Burkholder, J.M. and H.B. Glasgow Jr. 1997. Pfiesteria piscicida and other toxic Pfiesteria-like dinoflagellates: behavior, impacts, and environmental controls. Limnology & Oceanography 42:1052-1075 Castelle A.J., A.W. Johnson and C. Conolly. 1994. Wetland and stream buffer size requirements ? a review. Journal of Environmental Quality 23: 878-882. Cirmo C.P. and J.J. McDonnell. 1997. Linking the hydrologic and biogeochemical controls of nitrogen transport in near-stream zones of temperate-forested catchments: a review. Journal of Hydrology 199: 88-120. Cochran W.G. 1977. Sampling Techniques. 3rd ed. Somerset, NJ: John Wiley and Sons. Correll D.L. 1996. Buffer zones and water quality protection: general principles. In Proceedings of the International Conference on Buffer Zones. N. Haycock, ed. Oxford, England. 83 Correll D.L. 2000. The current status of our knowledge of riparian buffer water quality functions. In Proceedings of the International Conference on Riparian Ecology and Management in Multi-Use Watersheds, 5-10. P.J. Wigington, Jr. and R.L. Beschta, eds. Middleburg, Va.: AWRA. Day R.L., P.L. Richards and R.L. Brooks. 1996. Chesapeake Bay forest buffer inventory, final report. University Park, Pa.: Pennsylvania State University, College of Agricultural Sciences and School of Forest Resources. Frissel C.A., W.J. Liss, C.E. Warren and M.D. Hurley. 1986. A hierarchical framework for stream habitat classification: viewing streams in a watershed context. Environmental Management 10: 199-214. Giles, R.H. Jr. and M.K. Trani. 1999. Key elements of landscape pattern measures. Environmental Management 23: 477-481. Gold AJ, PM Groffman, K Addy, DQ Kellogg, and AE Rosenblatt. 2000. The role of landscape setting in riparian groundwater nitrate removal. In Proceedings of the International Conference on Riparian Ecology and Management in Multi-Use Watersheds, 113-118. P.J. Wigington, Jr. and R.L. Beschta, eds. Middleburg, Va.: AWRA. Gregory S., V. Swanson, J. Frederick and W.A. McKee. 1991. An ecosystem perspective of riparian zones: focus on links between land and water. BioScience 41: 540-551. Hewitt M.J. 1990. Synoptic inventory of riparian ecosystems: the utility of Landsat Thematic Mapper data. Forest Ecology and Management 33/34: 605-620. Hill A.R. 1996. Nitrate removal in stream riparian zones. Journal of Environmental Quality 25: 743-755. Lowrance R.R., L.S. Altier, J.D. Newbold, R.R. Schnabel, P.M. Groffman, J.M. Denver, D.L. Correll, J.W. Gilliam, J.L. Robinson, R.B. Brinsfield, K.W. Staver, W. Lucas and A.H. Todd. 1997. Water quality functions of riparian forest buffers in Chesapeake Bay watersheds. Environmental Management 21(5): 687-712. Lowrance R.R., R. Todd, J. Fail Jr., O. Hendrickson Jr., R. Leonard and L. Asmussen. 1984. Riparian forests as nutrient filters in agricultural watersheds. BioScience 34: 374-377. Maddock I. 1999. The importance of physical habitat assessment for evaluating river health. Freshwater Biology 41: 373-391. 84 Maryland Department of Natural Resources. Riparian Forest Buffers: Function and Design for Protection and Enhancement of Water Resources. Available at: www.dnr.state.md.us/forests/publications/buffers.html. Accessed on 20 June 2002. Muller E., H. Decamps and M.K. Dobson. 1993. Contribution of space remote-sensing to river studies. Freshwater Biology 29(2): 301-312. Novotny V. and H. Olem. 1994. Water quality: prevention, identification, and management of diffuse pollution. Van Nostrand Reinhold, New York, 1054 pp. Osborne L.L. and M.J. Wiley. 1988. Empirical relationships between land use/cover and stream water quality in an agricultural watershed. Journal of Environmental Management 26: 9-27. Perry C.D., G. Vellidis, R.R. Lowrance and D.L. Thomas. 1999. Watershed-scale water quality impacts of riparian forest management. Journal of Water Resources Planning and Management - ASCE 125(3): 117-125. Richards C., L.B. Johnson and G.E. Host. 1996. Landscape-scale influences on stream habitats and biota. Canadian Journal of Fisheries and Aquatic Sciences 53: 295-311. SAS Institute Inc. 1988. SAS/STAT User?s Guide. Release 6.03 Edition. Cary, N.C.: SAS Institute Inc. Schlosser I.J. and J.R. Karr. 1981. Riparian vegetation and channel morphology impact in spatial patterns of water quality in agricultural watersheds. Environmental Management 5(3): 233-243. Skaggs R.W. and G.M. Chescheir. 1999. Drains in depth. ASAE Resource Sept 1999:13-14. Tim U.S., R. Jolly, and H.H. Liao. 1995. Impact of landscape feature and feature placement on agricultural non-point-source-pollution control. Journal of Water Resources Planning and Management-ASCE 121(6): 463-470. Tufford D.L., H.N. McKellar Jr. and J.R. Hussey. 1998. In-stream nonpoint source nutrient prediction with land-use proximity and seasonality. Journal of Environmental Quality 27(1): 100-111. USDA. 1999. Assessing conditions of riparian-wetland corridors at the areawide level. Watershed Science Institute Technical Report. Washington, D.C.: USDA NRCS. 85 Weller D.E., T.E. Jordan and D.L. Correll. 1998. Heuristic models for material discharge from landscapes with riparian buffers. Ecological Applications 8(4): 1156- 1169. Whigham D.F., C. Chitterling and B. Palmer. 1988. Impacts of freshwater wetlands on water quality ? a landscape perspective. Environmental Management 12(5): 663- 671. Wissmar R.C. and R.L. Beschta. 1998. Restoration and management of riparian ecosystems: a catchment perspective. Freshwater Biology 40(3): 571-585. 86 CHAPTER 3 ? CLASSIFICATION OF AGRICULTURAL RIPARIAN ZONE LANDSCAPES IN THE COASTAL PLAIN AND PIEDMONT PHYSIOGRAPHIC REGIONS OF MARYLAND ABSTRACT The purpose of this study is to present a riparian classification system for the agricultural riparian landscapes of the Coastal Plain and Piedmont physiographic regions of Maryland. These landscapes form a continuum from the typical Coastal Plain setting on the Eastern Shore, characterized by a high water table, flat land, row crops, and a channelized stream, to the typical Piedmont setting in central Maryland, characterized by a deeper water table, rolling land, pasture, and a natural stream. In 1998, the University of Maryland, Maryland Cooperative Extension, and the Forestry Division of Maryland's Department of Natural Resources conducted a fieldwork-based survey of agricultural riparian buffers in Maryland's Chesapeake Bay watershed. Characterization of the riparian landscape was performed at 207 agricultural sites in eight agricultural watersheds in the Coastal Plain and Piedmont physiographic regions of Maryland. Categorization of riparian landscapes was performed using the CLUSTER procedure (average linkage method) in SAS, partitioning the riparian landscapes into six groups. Most sites were classified in a group with "average" 87 values of all landscape descriptors. Other groups described sites that had at least one extreme landscape parameter. These results present the first comprehensive structural characterization of Atlantic seaboard agricultural riparian buffers and their landscapes, and provide a method for quickly ascertaining a subset of sites for targeted analysis. While these are empirical results specific to Maryland, the general findings and approach are of use to other locations throughout the United States where the establishment of forest buffers is considered as a water quality improvement measure. INTRODUCTION Background Riparian buffers have received much attention for their role in stream ecosystem function (Vannote et al., 1980) and for their capacity to mitigate non-point- source (NPS) pollution (Lowrance et al., 1984, 1997; Correll, 1996; Barling and Moore, 1994). Streamside forests are complex ecosystems that help provide optimum food and habitat for stream communities. They can be effective in removing excess nutrients and sediment from surface runoff and shallow groundwater and shading streams to optimize light and temperature conditions for aquatic plants and animals. Streamside forests also ameliorate the effects of some pesticides and directly provide dissolved and particulate organic food needed to maintain high biological productivity and diversity in the adjoining stream. In October 1996, the Executive Council of the Chesapeake Bay Program established a goal of restoring riparian forest buffers on 88 2010 miles of streams in the Chesapeake Bay watershed by 2010 (Chesapeake Bay Program, 1996). This action exemplified the widespread acceptance of riparian buffers as an important tool for the reduction of non-point-source pollution. Because physical variation in riparian areas may aid or prevent the interception and transformation of pollutants (Hill, 1996; Weller et al., 1998), understanding the underlying natural variation in riparian buffer landscapes can augment interpretation of the impact of land use patterns on aquatic conditions. Without a comprehensive description of site geology, vegetation, hydrology, and soils, research findings are hard to generalize or extrapolate to other sites (Correll, 2000). Management would often like to be able to readily examine riparian characteristics using Geographic Information Systems (GIS), land-use and soil maps in order to determine what riparian management alternatives might contribute to decreasing nonpoint source pollution impacts on surface waters (Delong and Brusven, 1991). However, a system for conducting these comparisons in a logical manner has yet to be developed. As the plenary speaker for the American Water Resources Association International Conference on Riparian Ecology and Management in Multi-Use Watersheds, held in Portland, Oregon, in August 2000, Dr. David Correll of the Smithsonian Environmental Research Center stated that "one of the most urgent needs in riparian research is to develop and utilize an adequate system of classification for riparian buffers" (Correll, 2000). Most riparian classification systems have focused on a few selected attributes of riparian areas, such as hydric soil or hydrophilic plant associations (Dick-Peddie and Hubbard, 1977, cited in Szaro and Patton, 1986; Brown et al., 1979, cited in Hauer 89 and Smith, 1998; Norton et al., 1980, 1981; Batchelor, 1982; Winward, 1984; Youngblood et al., 1985a, 1985b, cited in Kovalchik and Chitwood, 1990; Curry and Slater, 1986; Szaro and Patton, 1986; Kovalchik, 1987, cited in Kovalchik and Chitwood, 1990; Baker, 1989, cited in Imhof et al., 1996; Padgett et al., 1989, cited in Kovalchik and Chitwood, 1990; Kovalchik and Chitwood, 1990; Durkin et al., 1996; Girard et al., 1997; Alpert and Kagan, 1998; Lyon and Sagers, 1998; Arbuckle et al., 1999). Many of these riparian classification systems were developed for management agencies such as United States Department of Agriculture Forest Service, US Department of Wildlife, US Bureau of Land Management, and the National Forest System. Therefore, many of these systems were developed for specific settings in the western United States (Colorado, Idaho, Kansas, Montana, Nevada, New Mexico, Oregon, Utah and Wyoming). Although these perspectives adequately characterized the terrestrial riparian plant communities, they provided little understanding of the wide array of ecological processes associated with the land-water interface (Gregory et al., 1991). Few studies have expanded the focus of the classification system to include the landscape setting. Kovalchik and Chitwood (1990) added geomorphology to a floristic classification system they developed in 1987. Their work was conducted in the western United States, and sites were assigned to groups using geomorphic features (location in the watershed, gradient, elevation, fluvial surface, and soil type). Whereas this was a landscape study, it was conducted at a large scale, and the settings are specific to the western United States. Delong and Brusven (1991) developed a classification system to identify areas at risk for erosion. Six riparian zone 90 characteristics (including vegetation type and height) were assigned discrete categorical values. Therefore, this study was conducted without quantification of any landscape characteristics. Quinn et al. (2000) developed a third classification system including the landscape setting. This landscape classification system was developed for large-scale riparian settings in New Zealand. The aim of this study was to evaluate the potential for classification of stream riparian zones by accounting for the differences in function of riparian areas in different parts of the catchment. Twenty-nine sites covering a range of stream and riparian characteristics were grouped into broad classes that reflected similarities in morphology, riparian functions, human uses, and likely riparian management options. A subjective knowledge-based approach was used rather than a statistical clustering method, due to the wide variety of attributes that were included and the small number of sites. Although this classification system focused on the riparian landscape rather than the vegetation, it was conducted at a much broader scale and did not focus on agricultural sites. Purpose The purpose of this paper is to present a riparian classification system for the agricultural riparian landscapes found in the Coastal Plain and Piedmont physiographic regions of Maryland. In addition to focusing on agricultural systems, these landscapes are distinctively different from the landscapes in which all other riparian buffer classification systems have been developed. Maryland?s agricultural riparian landscapes form a continuum from the typical Coastal Plain agricultural setting on the Eastern Shore, characterized by a high water table, flat land, row crops, 91 and a channelized stream, to the typical Piedmont agricultural setting in central Maryland, characterized by a deeper water table, rolling land, pasture, and a natural stream. This study investigated a method to discretize this continuum into objective, quantitative categories. This classification system is intended for use as a management tool for incentive programs funding the installation of agricultural buffers at sites that may be without any buffer, or have an inadequate buffer. Therefore, the focus is the physical riparian landscape, not existing riparian vegetation. Additionally, this classification system is intended as a technique to reduce error when modeling buffer effectiveness, by identifying homogeneous sites. While the data used here are specific to Maryland, the general findings and approach are of use to other locations throughout the United States where the establishment of forest buffers is considered as a water quality improvement measure. Objectives 1. To discretize the continuum of agricultural riparian landscapes of the Coastal Plain and Piedmont physiographic regions of Maryland's Chesapeake Bay watershed. 2. To develop a classification system for agricultural riparian buffer landscapes in the Coastal Plain and Piedmont regions of Maryland. 92 METHODS Overall summary of methods The agricultural riparian landscape was conceptualized as the stream channel, buffer, and adjacent agricultural field (Figure 9). The 1998 University of Maryland Agricultural Riparian Buffer Survey characterized 207 of these landscapes in eight primarily agricultural watersheds (>38% agriculture) of the Coastal Plain and Piedmont physiographic regions of Maryland (Figure 10). The SAS clustering procedure was then used with these data to create discrete categories of landscapes. (SAS is an integrated suite of software facilities for exploratory data modeling, statistical analysis and graphical display.) Figure 9. Schematic of the agricultural riparian landscape as defined for the 1998 University of Maryland Agricultural Riparian Buffer Survey. 93 Figure 10. Map of University of Maryland 1998 Agricultural Riparian Buffer Survey sample watershed locations superimposed over 16-digit Hydrologic Unit Code watersheds as defined by USGS. Data development In 1998, the University of Maryland, the Maryland Agricultural Cooperative Extension Service, and the Forestry Division of Maryland's Department of Natural Resources conducted a fieldwork-based survey of agricultural riparian buffers in the Coastal Plain and Piedmont physiographic regions of Maryland's Chesapeake Bay watershed. (The Appalachian Plateau was not surveyed due to budget restraints.) Because agencies are typically restricted to ?rapid assessment?-type surveys, this survey described these landscapes using ?easily obtainable? data. Field surveys were designed to take less than one hour per site, and map and other landscape data were obtained from county soil surveys and publicly-available GIS files. Therefore, the methodology can be reproduced by state agencies. This survey provided the first 94 extensive, ground-truthed data set describing a large set of agricultural riparian buffers, the surrounding land, and to a lesser extent, the land and streams within the buffers. Site characterization. Each survey site consisted of the stream, riparian zone and adjacent fields along both sides of a 30 to 90 meter stream reach. Field and map measurements fell into six key categories (Giles and Trani, 1999): location, topography, surface and groundwater hydrology, land use and stream channel morphology (Table 4). If a buffer was present, vegetation and width measurements were also taken. However, buffer data were not pertinent to this classification exercise because the riparian landscape was the focus of study, not the buffer within the landscape. Two scales of data were considered. Site characterization provided ground- truthed information about the local riparian landscape below the resolution of remotely sensed data. These data were obtained from on-site surveys. Because one of the directives of the University of Maryland Buffer Survey Project was to collect "easily obtainable data", fieldwork was restricted to a suite of structural landscape measurements that could be obtained in a single visit of less than one hour (Appendices 1 and 2). Information at the site watershed scale provided insight about cumulative impacts and sources of agricultural NPS pollution. These data were developed from soil surveys, USGS topographic maps, and publicly available EPA, USGS and state GIS files (Appendix 3). Preparation of data for cluster analysis. The original data set included 207 sites with three adjacent land use categories (crop, pasture and forest). Sites with 95 adjacent forest or unknown land use were eliminated for this analysis. Because clustering programs will not consider sites with missing values, those sites were also eliminated, resulting in a final set of 96 sites. Site characterization included both quantitative and qualitative descriptors. Both quantitative and categorical landscape descriptors were used in the clustering procedure. A Shapiro-Wilks test for normality indicated that no quantitative landscape descriptors were normally distributed (all Pr < W were less than 0.04). Therefore, in order to reduce the effect of skewness, quantitative descriptors were log-transformed prior to analysis. The clustering program uses only quantitative variables, so categorical variables were coded as ?dummy? values of ?0? or ?1?; region (Coastal Plain, Piedmont), straightening of the stream channel (yes/no), adjacent field use (crop/pasture), soil drainage class (good/poor), channel shape (natural/triangular), presence of levees or berms (yes/no) and stream order (1st and 2nd, 3rd and 4th order). Mean values and standard deviations for site descriptors are given in Table 4. Cluster analysis and interpretation of results Categorization was performed using the CLUSTER procedure (average linkage method) in SAS. SAS code for cluster analysis is presented in Appendix 18. This is an aggregation method that hierarchically clusters the observations. Each observation begins in a cluster by itself. The two closest clusters are merged to form a new cluster that replaces the two old clusters. Merging of the two closest clusters is repeated until only one cluster is left. The standardization option was used, in order to give all variables equal weight, a common procedure in cluster analysis. Following 96 the Cluster procedure, sites were partitioned into a specific number of groups. The ultimate grouping of sites was determined by repeated trials with 4 to 8 groups of sites and examination of the univariate statistics of the resultant groups. 97 Table 4. Mean values and standard deviations for riparian landscape descriptors in the 1998 University of Maryland Agricultural Riparian Buffer Survey. Landscape Descriptor n Mean Std. Dev. Location Physiographic region * 207 0.55 0.50 Distance downstream (km) 206 3.25 4.86 Site watershed area (ha) 207 3.46 9.23 Land Use Adjacent crop? * 205 0.62 0.49 Adjacent pasture? * 207 0.36 0.48 Riparian Zone Characterization Length buffer fragment (km) 174 0.88 1.18 Topography Field slope (%) 196 4.78 5.40 Riparian slope (%) 193 8.17 9.65 Surface Hydrology Stream order < 3? * 207 0.79 0.41 Stream gradient (%) 207 1.12 0.86 Channel straightened? * 205 0.27 0.45 Groundwater Hydrology Riparian soil drainage class* 199 0.24 0.43 Field soil drainage class* 203 0.79 0.41 Depth to riparian water table (m) 207 1.2 2.6 Depth to field water table (m) 207 3.7 3.2 Riparian surface K (cm/hr) 207 3.8 2.8 Field surface K (cm/hr) 205 4.8 3.0 Riparian minimum K (cm/hr) 207 3.0 2.8 Field minimum K (cm/hr) 205 3.2 2.2 Depth to riparian minimum K (m) 201 8.1 3.0 Depth to field minimum K (m) 205 6.1 3.2 * coded descriptors have values of ?0? or ?1? Physiographic region: Coastal Plain = 0, Piedmont = 1 Adjacent crop or pasture: no = 0, yes = 1 Stream order <3: yes = 0, no = 1 Channel straightened: no = 0, yes = 1 Riparian and field soil drainage class: poor = 0, good = 1 98 RESULTS AND DISCUSSION By repeatedly performing the SAS Proc Cluster procedure on the data using 4 to 8 final numbers of clusters indicated that for this group of sites, there were four optimal number of site clusters. Most sites were classified in a cluster with "average" values of all landscape descriptors, and other clusters contained small numbers of sites (Table 5). The landscape descriptors that differentiated the site clusters represented physiographic region, adjacent land use, surface hydrology and channel morphology (Table 6). Sites were differentiated from the ?average-values? cluster by being described by at least one landscape parameter with an extreme value (Table 7). The four clusters were designated as follows: 1. ?Average sites? - sites with any land use with low drainage density. 2. Coastal Plain crop sites on channelized high-order streams. 3. Coastal Plain or Piedmont crop sites on low-order streams with high drainage density catchments. 4. Piedmont pasture sites on 2nd-3rd-4th -order streams. 99 Table 5. Number of sites per cluster resulting from the SAS Proc Cluster Procedure. Multiple procedures were performed to produce 4 ? 8 clusters of sites, using the 1998 University of Maryland Agricultural Riparian Buffer Survey data. Number of Clusters NCluster 1 NCluster 2 NCluster 3 NCluster 4 NCluster 5 NCluster 6 NCluster 7 NCluster 8 4 87 10 4 1 5 83 10 4 4 1 6 83 9 4 4 1 1 7 80 9 4 4 3 1 1 8 80 9 4 4 2 1 1 1 Table 6. Riparian landscape descriptors that differentiated clusters of sites and corresponding descriptor categories, based on data from the 1998 University of Maryland Agricultural Riparian Buffer Survey. Descriptor Category Descriptor Location Region Land Use Adjacent land use Surface Hydrology Stream order Drainage density1 Channel Morphology Straightened? 1 Drainage density was described by the product of site watershed area and distance downstream along the main channel of the stream. 100 Table 7. Landscape descriptor qualities for clusters of sites produced by the Agricultural Riparian Classification System (ARCS), based on data from the 1998 University of Maryland Agricultural Riparian Buffer Survey. Cluster Number of sites Straightened stream? Drainage density 1 Stream order Region Adjacent land use 1 83 mix low all mix all 2 10 yes low high C. Plain crop 3 4 no high low mix crop 4 4 no low high Piedmont pasture 1 Drainage density was described by the product of site watershed area and distance downstream along the main channel of the stream. An example of the Agricultural Riparian Classification System (ARCS). An agricultural site in Montgomery County (Piedmont physiographic region) was classified with the ARCS as an example of the application of the system. A topographic map provided stream order. Fieldwork consisted of observation of the adjacent land use and whether the stream was channelized (straightened) or natural. This information was gathered by a drive-by observation. Because this was a Piedmont site with adjacent pasture on a natural (unchannelized) 3rd order stream, it was placed in ARCS category 4. CONCLUSIONS This riparian classification system provides the first method for categorizing agricultural riparian landscapes of for the Coastal Plain and Piedmont physiographic regions of Maryland. The study was based on structural characteristics. Sites were characterized with easily obtainable information and represented a gradient of settings 101 from cropland on channelized streams in the Coastal Plain to pastures on natural streams in the Piedmont. The results indicated that, for the Coastal Plain and Piedmont physiographic regions of Maryland, the continuum of agricultural riparian landscapes could be categorized into discrete groupings with recognizable characteristics. Although there are distinct structural differences between representative landscapes of the Coastal Plain and Piedmont physiographic regions, at the riparian landscape scale, most sites in both regions demonstrated average structural characteristics. Other groups of sites deviated from this set of average values in some way. The elements associated with deviance from average values can be obtained from soil surveys, USGS topographic maps and minimal site work. Although minimal, fieldwork is necessary to perform the ARCS because Remote Sensing data do not currently provide the necessary resolution for this work. Data with finer resolution may be available from GIS data sets in the future. 102 CITED REFERENCES Alpert P. and J. Kagan. 1998. The utility of plant community types: A practical review of riparian vegetation classification in the Intermountain United States. Natural Areas Journal 18(2): 124-128. Arbuckle C.J., A.D. Huryn and S.A. Israel. 1999. A landcover classification in the Taieri River Catchment, New Zealand: A focus on the riparian zone. Geocarto International 14(3): 7-14. Baker W.L. 1989. Classification of the riparian vegetation of the montane and subalpine zones in western Colorado. The Great Basin Naturalist 49(2): 214-228. Barling R.D. and I.D. Moore. 1994. Role of buffer strips in management of waterway pollution: a review. Environmental Management 18(4): 543-558. Batchelor R.F. 1982. A taxonomic classification system for Montana riparian vegetation types: an interagency approach to classifying Montana's riparian ecosystems. Rural Area Development Committee, Wildlife Subcommittee, Riparian Program Team. Bozeman: Montana. Brown S., M.M. Brinson and A.E. Lugo. 1979. Structure and function of riparian wetlands. In Proceedings of the Symposium on Strategies for Protection and Management of Floodplain Wetlands and Other Riparian Ecosystems, 17-31. R.R. Johnson and J.F. McCormick, eds. U.S. Department of Agriculture, Forest Service, General Technical Report. Chesapeake Bay Program. 1996. Final report of the Riparian Forest Buffer Panel. October 1996. CBP/TRS 158/96, EPA 903-R-96-015. Correll D.L. 1996. Buffer zones and water quality protection: general principles. In Proceedings of the International Conference on Buffer Zones. N. Haycock, ed. Oxford, England. _______. 2000. The current status of our knowledge of riparian buffer water quality functions. In Proceedings of the International Conference on Riparian Ecology and Management in Multi-Use Watersheds, 5-10. P.J. Wigington, Jr. and R.L. Beschta, eds. Middleburg, Va.: AWRA. Curry P. and F.M. Slater. 1986. A classification of river corridor vegetation from four catchments in Wales. Journal of Biogeography 13(2): 119-132. 103 Delong M.D. and M.A. Brusven. 1991. Classification and spatial mapping of riparian habitat with applications toward management of streams impacted by nonpoint source pollution. Environmental Management 15(4): 565-571. Dick-Peddie W.A. and J.P. Hubbard. 1977. Classification of riparian vegetation. In Importance, preservation and management of riparian habitat: a symposium, 85-90. R.R. Johnson and D.A. Jones, tech. coordinators. USDA Forest Service General Technical Report: RM-43. U S Rocky Mt. Forest and Range Experimental Station, Fort Collins, CO. Durkin P., E. Muldavin and M. Bradley. 1995. A preliminary riparian/wetland vegetation community classification of the Upper and Middle Rio Grande watersheds in New Mexico. In: Desired future conditions for Southwestern riparian ecosystems: bringing interests and concerns together. Albuquerque, New Mexico. U.S. Dept. of Agriculture, Rocky Mountain Forest and Range Experiment Station, Fort Collins, Colo. Giles, R.H. Jr. and M.K. Trani. 1999. Key elements of landscape pattern measures. Environmental Management 23: 477-481. Girard M.M., D.L. Wheeler, D.L. Mills and B. Stephanie. 1997. Classification of riparian communities on the Bighorn National Forest. USDA Forest Service, Rocky Mountain Region, Lakewood, Colo. Gregory S., V. Swanson, J. Frederick and W.A. McKee. 1991. An ecosystem perspective of riparian zones: focus on links between land and water. BioScience 41: 540-551. Hauer F.R. and R.D. Smith. 1998. The hydrogeomorphic approach to functional assessment of riparian wetlands: evaluating impacts and mitigation on river floodplains in the U.S.A. Freshwater Biology 40(3): 517-530. Hill A.R. 1996. Nitrate removal in stream riparian zones. Journal of Environmental Quality 25: 743-755. Imhof J.G., J. Fitzgibbon and W.K. Annable. 1996. A hierarchical evaluation system for characterizing watershed ecosystems for fish habitat. Canadian Journal of Fisheries and Aquatic Sciences 53(SU1): 312-326. Kovalchik B.L. 1987. Riparian zone associations of the Deschutes, Ochoco, Fremont and Winema National Forests. Publ. No. R6 ECOL TP-279-87. Portland, Oregon. USDA Forest Service, Pacific Northwest Region 6. 104 Kovalchik B.L. and L.A. Chitwood. 1990. Use of geomorphology in the classification of riparian plant associations in mountainous landscapes of central Oregon, USA. Forest Ecology and Management 33/34: 405-418. Lowrance R.R., L.S. Altier, J.D. Newbold, R.R. Schnabel, P.M. Groffman, J.M. Denver, D.L. Correll, J.W. Gilliam, J.L. Robinson, R.B. Brinsfield, K.W. Staver, W. Lucas and A.H. Todd. 1997. Water quality functions of riparian forest buffers in Chesapeake Bay watersheds. Environmental Management 21(5): 687-712. Lowrance R.R., R. Todd, J. Fail Jr., O. Hendrickson Jr., R. Leonard and L. Asmussen. 1984. Riparian forests as nutrient filters in agricultural watersheds. BioScience 34: 374-377. Lyon J. and C.L. Sagers. 1998. Structure of herbaceous plant assemblages in a forested riparian landscape. Plant Ecology 138(1): 1-16. Norton B.E., P.J. Urness and A.R. Southard. 1980. A riparian community classification study: final report. Dept. of Range Science, Utah State University, Logan, Utah. Norton B.E., J. Tuhy and S. Jensen. 1981. An approach to classification of riparian vegetation. In In-place resource inventories: principles & practices: proceedings of a workshop. T.B. Brann, L.O. House IV and H.G. Lund, eds. August 9-14, 1981, Univ. of Maine, Orono. Padgett W.G., A.P. Youngblood and A.H. Winward. 1989. Riparian community type classification of Utah and southeastern Idaho. U.S. Dept. of Agriculture, Forest Service, Intermountain Region, Ogden, Utah. Quinn J.M., A. Taylor, W. Boyce, and T. Fenton. 2000. Riparian zone classification improves management of stream water quality and aquatic ecosystems. In Proceedings of the International Conference on Riparian Ecology and Management in Multi-Use Watersheds, 451-455. P.J. Wigington, Jr. and R.L. Beschta, eds. Middleburg, Va.: AWRA. Szaro R.C. and D.R. Patton. 1986. Riparian habitat classification in the southwestern United States. In Transactions of the 51st North American Wildlife and Natural Resources Conference, 215-221. Wildlife Management Institute, Washington, D.C. Vannote R.L., G.W. Minshall, K.W. Cummins, J.R. Sedell and C.E. Cushing. 1980. The river continuum concept. Canadian Journal of Fisheries and Aquatic Sciences 37: 130-137. 105 Weller D.E., T.E. Jordan and D.L. Correll. 1998. Heuristic models for material discharge from landscapes with riparian buffers. Ecological Applications 8(4): 1156- 1169. Winward A.H. 1984. An ecosystem classification for intermountain riparian areas. In Range watersheds, riparian zones and economics: interrelationships in management and use. 1984 Pacific Northwest Range Management Short Course, Oregon State University, Corvallis, Or. Youngblood A.P., W.G. Padgett, G. Wayne and A.H. Winward. 1985. Riparian community type classification of eastern Idaho - western Wyoming. Publ. No. R4- ECOL-85-01. U.S. Dept. of Agriculture, Forest Service, Intermountain Forest Range Experimental Station, Ogden, Utah. Youngblood A.P., W.G. Padgett and A.H. Winward. 1985. Riparian community type classification in the intermountain region. General technical report RM 120 (1985): 510-512. Rocky Mountain Forest and Range Experiment Station, United States, Forest Service. Fort Collins, Colo. 106 CHAPTER 4 ? AGRICULTURAL RIPARIAN BUFFER SURVEY ? USE OF TWO DATA SETS: 1995-1997 MBSS AND 1998 UNIVERSITY OF MARYLAND BUFFER SURVEY ABSTRACT The purpose of this paper is to present two field-based data sets that were used to characterize the distribution of riparian buffers in agricultural settings within the Coastal Plain and Piedmont physiographic regions of Maryland. The first data set is the University of Maryland Agricultural Riparian Buffer Survey, an extensive field- based survey conducted in the summer of 1998. This survey provided extensive, ground-truthed information about agricultural riparian buffers, the surrounding land, and to a lesser extent, the land and streams within the buffers, within the Coastal Plain and Piedmont physiographic regions of Maryland. The second data set is an altered version of the 1995-1997 Maryland Biological Stream Survey (MBSS), conducted by Maryland Department of Natural Resources. This annual survey is conducted to assess water quality, physical habitat, and biological conditions in first through third order, non-tidal streams. Although buffer characterization is not the primary purpose of the survey, MBSS data include limited site-specific, ground-truthed information on riparian buffers as well as land use information for catchments upstream of the 107 sampled sites. A subset of the MBSS data set was selected to represent agricultural sites in primarily agricultural watersheds of the Coastal Plain and Piedmont physiographic regions. The original University of Maryland Survey data were censored to reflect the same defined maximum buffer width used in the MBSS. Comparison of the two data sets indicated excellent agreement. Comparison of the censored data sets to the original UM Survey demonstrated the critical effect of defined maximum buffer width: a defined maximum buffer width of 50 meters yielded an average buffer width of approximately 30 meters, whereas original UM Survey with a defined maximum buffer width of 305 meters (1000 feet) yielded an average width of approximately 49 meters. Since average buffer width is a key value for management, this effect has important implications for policymakers. These two data sets provide extensive, ground-truthed, field-based information on riparian buffers in the state of Maryland. INTRODUCTION Background In 1998, the University of Maryland, the Maryland Agricultural Cooperative Extension Service, and the Forestry Division of Maryland's Department of Natural Resources joined together to conduct a fieldwork-based survey of agricultural riparian buffers in Maryland's Chesapeake Bay watershed. This research provided the first extensive, ground-truthed attempt to describe a large set of agricultural riparian buffers, the surrounding land, and to a lesser extent, the land and streams within the 108 buffers. The research has been summarized in Chapter 2. The only other data set that includes extensive, ground-truthed information on Maryland's riparian buffers is the Maryland Biological Stream Survey (MBSS). The Maryland Department of Natural Resources (DNR) conducts this annual survey to assess water quality, physical habitat, and biological conditions in first through third order, non-tidal streams. Although buffer characterization is not the primary purpose of the survey, MBSS data includes limited site-specific, ground-truthed information on riparian buffers as well as land use information for catchments upstream of the sampled sites. Purpose The purpose of this chapter is to compare and contrast these two field-based data sets in order to characterize the distribution of riparian buffers in agricultural settings within the Coastal Plain and Piedmont physiographic regions of Maryland. The MBSS data enabled an objective validation and critique of the results obtained in Chapter 2. One of the key findings of Chapter 2 was the inability of remotely sensed data to distinguish a large percentage of Maryland?s agricultural buffers. Therefore, the development of data sets that accurately portray the width and distribution of Maryland?s agricultural buffers is critical for the development of empirical models of buffer effectiveness. Objectives 1. To develop a parsed data set from the MBSS 195-1997 data that will be representative of agricultural riparian landscapes in Maryland?s Coastal Plain and Piedmont regions, and suitable for model development. 109 2. To evaluate the parsed MBSS data set and confirm that is as representative of agricultural riparian landscapes in Maryland's Coastal Plain and Piedmont physiographic regions as data from the 1998 University of Maryland Agricultural Riparian Buffer Survey. METHODS Overall summary of methods Two ground-truthed databases were examined, compared and contrasted to characterize the extent and nature of agricultural riparian buffers in the Coastal Plain and Piedmont physiographic regions of Maryland. The first data set was the University of Maryland (UM) Agricultural Riparian Buffer Survey, an extensive field- based survey conducted in the summer of 1998. The second data set was the 1995- 1997 MBSS, conducted by Maryland DNR. The data sets were examined and censored to create two new data sets with comparable characteristics. Univariate analysis on the data sets provided contrast and validation of their effectiveness to represent the Maryland's agricultural riparian landscapes. Description of the data sets The 1995-1997 MBSS used a rapid bioassessment protocol to collect site-level data on the stream channel and adjacent land during the spring and early summer (Kazyak, 1997). Although buffer characterization was not the primary purpose of the survey, limited information was taken on the riparian zone (buffer width, vegetation type). Land use information for the site catchment was obtained from Geographic 110 Information System (GIS) data. About 300 sites (75 meter stream segments) were sampled each year during the three-year period 1995-97, producing a data set with over 900 sites intended to be representative of all basins in the state. An MBSS data collection site was represented by approximately 160 variables. Numerous location variables and time variables identified exactly where and when the samples were collected. Not only were the immediate sites described but some adjacent land use data were also recorded. Variables describing site landscapes are described briefly: Percent agriculture and rowcrops in catchment: Percent agricultural land use in the site catchment and % rowcrops in catchment were based on the 1996 Multi-Resolution Land Use Classification (MRLC) land cover data base for EPA Region III (MRLC, 1996a; MRLC, 1996b). Maximum depth: maximum depth within the 75-meter segment (cm). Area: the site catchment area, reported in acres. Stream order: represents the Strahler (1957) convention used for ranking stream reaches by order. Stream order determination was based on a stream reach file digitized from 1:250,000 scale topographic maps for the MBSS in 1987. Riparian width: the width of the vegetated riparian buffer was estimated (m), to a maximum of 50 m. If the buffer was greater than or equal to 50 m, a value of 50 was entered. This measure was the width of the riparian buffer on the side of the stream with the smallest buffer. 111 The 1998 University of Maryland Agricultural Riparian Survey characterized riparian landscapes in eight primarily agricultural watersheds of Maryland's Coastal Plain and Piedmont physiographic regions. Site survey measurements that were taken as multiple replicates (channel width and depth, bank angles, stream velocity, riparian buffer width) were reduced to average values. Therefore, these 34 parameters represented over 150 individual measurements characterizing 207 sites. Tables 8 and 9 present the site descriptors and range of values obtained through the combination of fieldwork, map work and GIS work. Table 8. Landscape descriptors used to characterize the agricultural riparian landscape in the 1998 University of Maryland Agricultural Riparian Buffer Survey. Descriptor Category Riparian Landscape Descriptor Units Range of Values Survey watershed % agriculture2 Survey watershed % forest2 % % 38.2 - 76.1 26.1 ? 45.0 Site contributing watershed % agriculture2 % 36.9 ? 100.0 Site contributing watershed % crops2 % 26.3 ? 100.0 Land Use Adjacent land use1 crop/pasture/ forest/other Buffer width1 m 0 ? 304.8 Length of buffer fragment3 km 0.02 - 6.9 Forest buffer DBH1 cm 3.8 - 84.3 Forest buffer basal area1 m2 0 - 15.5 Buffer Char- acter- ization Shrub height1 m 0.3 - 6.1 Topo- graphy Watershed topographic index 2 Riparian buffer slope1 Adjacent field slope1 Ratio riparian buffer:adjacent field slope1 m/ m2 % % unitless 0.9 - 30.1 -2.5 - 62.3 0 - 37.5 -0.11 - 35.2 Watershed drainage density2 km2/km 1.02 - 1.25 Mainstem sinuosity2 km/km 1. 1- 1.4 Stream gradient1 % 0 ? 4.0 Stream velocity1 m/sec 0 ? 1.0 Surface Hydrology Stream order 1,2 unitless 1 - 4 1 from on-site field work 2 from GIS work 3 from soil survey 112 Table 8. (continued) Descriptors used to characterize the agricultural riparian landscape in the 1998 University of Maryland Agricultural Riparian Buffer Survey. Descriptor Category Riparian Landscape Descriptor Units Range of Values Ground- water Riparian buffer avg. depth to water table Adjacent field avg. depth to water table m m 0 - 10.7 0 - 10.7 Hydrology3 Riparian buffer surface K * Adjacent field surface K * cm/hour cm/hour 0.03 - 15.2 0.1 - 15.2 Riparian buffer minimum K * Adjacent field minimum K * cm/hour cm/hour 0.03 - 15.2 0.05 - 15.2 Riparian buffer soil drainage class Adjacent field soil drainage class variable/poor/ good/ excellent variable/poor/ good/ excellent Riparian buffer depth of minimum K * Adjacent field depth of minimum K * m m 0.03 - 15.5 1.7 - 15.1 Channel width m 0.4 - 21.6 Channel depth m 0.2 - 5.6 Channel cross-sectional area m2 0.1 ? 50.5 Channel width : depth ratio m/m 0.27 ? 30.0 Substrate mud/sand/ gravel/ cobble Channel shape natural/ triangular Channel straightened? yes/no Presence of berms yes/no Degree of gullying low/moderate/ severe Average bank angle (degrees) degrees 5.7 ? 90.0 Degree of bank erosion low/moderate/ severe Channel Morphology1 Livestock access yes/no * K = hydraulic conductivity 1 from on-site field work 2 from GIS work 3 from soil survey 113 Table 9. Values of categorical descriptors used to characterize the agricultural riparian landscape in the 1998 University of Maryland Agricultural Riparian Buffer Survey. Buffer Characterization Buffer tree type hardwood 85.94 mix 13.28 pine 0.78 Adjacent field soil drainage class excellent 11.06 good 58.65 poor 13.46 variable 16.83 excellent good poor variable Groundwater Hydrology Riparian buffer soil drainage class 5.77 18.75 67.31 8.17 boulder cobble sand/ gravel sand mud Channel substrate 2.5 24 32 13.5 28 natural triangularChannel shape 91.22 8.78 yes no Channel straightened 27.18 72.86 yes no Berms 10.73 89.27 high medium low noneGullying 3.11 7.77 54.92 34.2 high medium low noneBank erosion 21.52 53.8 22.15 2.53 yes no Channel Morphology Livestock access 20.39 79.61 114 Differences between the two databases Because the two databases were developed for different purposes, they contained different landscape information on two different populations of sites. The MBSS data included a far wider range of site types than the University of Maryland Buffer Survey. MBSS data included non-agricultural sites, watersheds that were not primarily agricultural, and physiographic regions other than the Coastal Plain and Piedmont. In addition, the two databases contained different site descriptors (land uses adjacent to the site, land uses in the site watershed, physiographic regions) and buffer descriptors. Buffer vegetation was described slightly differently in the two data sets. In the MBSS data, riparian buffer type was designated by one of the land use/land cover types used for adjacent land cover. Although there was one descriptor for "forest" buffer, there were multiple adjacent land use descriptors describing "grass" buffers (tall grass, old field, and lawn). In the UM Buffer Survey, buffer types were divided into forest, grass and "other". Although the width of buffer at a site was defined in both data sets as the average buffer width on one side of the stream, this value was derived slightly differently. The MBSS reported the width of the narrowest buffer on one side of the stream. The UM Buffer Survey reported the average of 6 readings, 3 on each side of the stream. Maximum buffer width was defined differently, as well. MBSS defined maximum buffer width as 50 m (Mercurio et al., 1999). The UM Buffer Survey defined maximum buffer width as 305 m. 115 Development of modified data sets Two new data sets (designated MBSS* and Survey*) were created by modifying the original data sets. The MBSS* data set was created by taking a subset of the original MBSS data to include only agricultural sites in primarily agricultural watersheds of the Coastal Plain and Piedmont physiographic regions. A "primarily agricultural watershed" was defined as one whose percent agriculture was greater than 35%, consistent with the minimum site percent agriculture found in the UM Buffer Survey. These restrictions reduced the MBSS data from 904 sites to 279 sites. Because censorship of the MBSS data removed sites, univariate values for the MBSS* data set were different from the original MBSS data set. The Survey* data set was created, not by removing any sites, but by re-defining maximum buffer width. All buffer widths between 50 meters and 305 meters in the Survey data set were defined as 50 meters. Therefore maximum buffer width was the only significant change between the UM Buffer Survey data set and the Survey* data set. The data were analyzed to provide univariate measures (mean, standard deviation, and percents), distribution measures (statewide distribution of buffer types, regional distribution of buffer types, width histograms) and correlation (Appendices 6 and 7) for agricultural buffers in Maryland. Values from the two data sets were compared to determine if they represented the same population of sites. All data were analyzed using SAS software. 116 RESULTS AND DISCUSSION Comparison of the modified databases Comparison of the original and altered data sets is summarized in Table 10. Survey* and MBSS* data sets both represented approximately the same number of sites. Univariate analysis of several landscape descriptors indicated that the two databases represented the same population of sites. All sites were limited to the Coastal Plain and Piedmont physiographic regions of Maryland. Both data sets showed approximately 60% watershed agriculture. Both data sets showed approximately 17% grass buffers, and between 55 to 70% forest buffers. Average buffer width using the MBSS* and Survey* data sets was approximately 30 meters and were not statistically different at the 95% level of confidence. These results indicated that the two censored data sets portrayed the same population of sites. 117 Table 10. Comparison of the original 1998 University of Maryland Agricultural Riparian Buffer Survey (Survey), 1995-1997 Maryland Biological Stream Survey (MBSS), modified Survey (Survey*), and modified MBSS (MBSS*) data sets. Survey MBSS Survey* MBSS* Number sites 207 954 207 279 Number physiographic regions 2 6 2 2 % forest buffers 70.5 56.0 70.5 55.4 % grass buffers 16.4 10.0 16.4 17.4 % unbuffered sites 6 6 24 Site % ag Mean 56.0 44.2 56.0 64.7 SD 14.0 26.52 14.0 14.0 Max 100.0 96.8 100.0 96.8 Min 38.2 0 38.2 37.3 Site % crops 44.2 41.8 Site % pasture 29.4 21.84 Buffer width (m) Mean 48.8 24.2 24.9 31.1 SD 57.7 20.4 18.3 20.2 Max 304.8 45.7 45.7 45.7 Min 0 0 0 0 Distribution of buffer types Distribution of buffer type was depicted as the relative statewide number of sites with no buffer, forested buffer, or grass buffer. Both data sets showed the same relative distribution of forested to grass buffers in the Coastal Plain (Figure 11). The data sets differed in the amount of unbuffered agricultural sites. Survey* data showed 6% unbuffered sites, whereas MBSS* data showed approximately 24% unbuffered sites. Users may find that this is the primary difference in the two data sets. 118 Both data sets showed the same pattern for distribution of buffers across physiographic region (Figure 12), although the MBSS* data reflected a higher percentage of unbuffered sites in the Piedmont. The Piedmont (P) had more unbuffered sites and grassed buffers than the Coastal Plain (CP). The Coastal Plain had more forested buffers than grass buffers. These results are in marked contrast to the belief that farmers on the Eastern Shore of Maryland (in the Coastal Plain) are less likely to install buffers at all, and forested buffers in particular. Figure 11. Distribution of agricultural riparian buffers by type and physiographic region, as determined by Survey* and MBSS* data. Survey* and MBSS* data represent riparian agricultural sites in catchments with > 35% agriculture, in the Coastal Plain and Piedmont physiographic regions of Maryland. (RFB = riparian forest buffer, CP = Coastal Plain, P = Piedmont). 0% 20% 40% 60% 80% 100% none 2.33 9.3 10.75 50.9 Grass 20.93 8.6 24.73 22.7 RFB 76.74 82.1 64.52 26.4 Survey* MBSS* Survey* MBSS* CP CP P P 119 Buffer width Although the average buffer widths developed from Survey* and MBSS* data were not significantly different, average buffer width developed from the original UM Survey data was approximately 49 meters, and was significantly wider. This effect demonstrated the critical effect of maximum buffer width. The Survey data with defined maximum buffer width of 305 meters produced a greater average value than the Survey* data, which censored buffer widths to no wider than 46 meters. Since average buffer width is a key value for management, this effect has important implications for policymakers. Buffer width histograms were constructed using MBSS* and Survey* data (Figure 12). Most sites fell into one of two width groups; the narrowest width group of buffers less than 10 meters (30 ft) wide and the widest width group of buffers greater than 40 meters (120 ft). Further resolution of buffer width is impossible within these categories. Therefore, a larger defined maximum buffer width, such as used in the UM Buffer Survey, would enable finer delineation of the distribution of buffer widths. 120 Figure 12. Histogram of agricultural buffer width for all buffer types combined, based on Survey* and MBSS* data. * Buffer widths greater than 50 meters have been censored to 50 meters. Usefulness of remotely sensed data The buffer width histograms for the MBSS* data verified the implications concerning the usefulness of RS data for buffer surveys determined from the 1998 University of Maryland Agricultural Riparian Buffer Survey. Most buffer surveys have been performed with remotely sensed (RS) data because of the exhaustive nature of the fieldwork. RS data are generally available in 10-meter and 30-meter resolution. Figure 13 illustrates the inaccuracy inherent in RS data used for this work. Both data sets indicated that approximately 30% of agricultural riparian buffers were less than 10 meters wide, and approximately 50% of buffers were less than 30 meters wide. Even considering the total width of buffer on both sides of a stream, surveys performed with RS data would drastically underestimate the number and amount of 0-9 10-19 20-29 30-39 40-50 0 10 20 30 40 50 % buffer width length groups (m) Survey* MBSS* 121 agricultural buffers in Maryland. This would be especially problematic in the Piedmont, where grass buffers were more prevalent, since remote sensing technologies currently have difficulty distinguishing grass from low-level crops or pasture. Figure 13. Percent of agricultural riparian buffers in the Coastal Plain and Piedmont physiographic regions of Maryland that are narrower than the resolution of remotely sensed data, based on the 1998 University of Maryland Agricultural Riparian Buffer Survey. Average buffer width over all buffer types was compared between the two physiographic regions (Figure 14). Analysis of the MBSS* data set indicated that Coastal Plain buffers were the same width as Piedmont buffers and the Survey* data indicated that Coastal Plain buffers were wider than Piedmont buffers. These results were not surprising, since the data indicated more forested buffers in the Coastal Plain 0% 25% 50% 75% 100% what's seen 54 50 width <30m 46 50 width <10m 29 36 Survey MBSS 122 and that forested buffers were wider than grass buffers. However, the result that Coastal Plain buffers were as wide or wider than Piedmont buffers was in direct contrast to the popular but anecdotal impressions that Eastern Shore (Coastal Plain) has narrow buffers. Figure 14. Average agricultural riparian buffer width in the Coastal Plain and Piedmont physiographic regions of Maryland, as determined by Survey* and MBSS* data. CONCLUSIONS Both the UM Agricultural Buffer Survey and the MBSS provided extensive, ground-truthed, field-based information on riparian buffers in the state of Maryland. These two independent data sets, created for different purposes, were subset and 0 10 20 30 40 50 60 70 average buffer width (m) MBSS* 32.3 35.4 Survey* 57.9 37.5 Coastal Plain Piedmont 123 censored to provide information about a common population of sites. The two modified data sets produced same or similar descriptions of the nature and extent of riparian buffers at agricultural sites in highly agricultural watersheds of Maryland's Coastal Plain and Piedmont physiographic regions. Because of the similarity of results, these data sets are considered trustworthy for use in development of models of agricultural buffer effectiveness. The use of MBSS* data confirmed the accuracy of the following results reported in Chapter 2: (1) Most agricultural sites had buffers. (2) The Coastal Plain was dominated by forested buffers, whereas the Piedmont had a higher proportion of grass buffers, (3) Forested buffers were wider than grass buffers, (4) Average width of Coastal Plain buffers was the same or greater than Piedmont buffers, (5) A significant portion of buffers were narrower than the current resolution of RS data, indicating that buffer surveys performed with RS data drastically underestimate the number and amount of agricultural buffers in Maryland. MBSS* data differed from the UM Survey data in the estimate of the number of unbuffered agricultural sites. Further work is necessary to determine a more accurate estimate of the distribution of unbuffered sites in Maryland. This use of two independent data sets also provided insight into interpretation of MBSS data. The use of a 150-foot defined maximum buffer width drastically censored interpretation of actual buffer width and caused a significant underestimation 124 of average buffer width. Therefore, buffer width values reported from MBSS data (Boward et al., 1999; Roth et al., 1999) should be viewed with caution. 125 CITED REFERENCES Alliance for the Chesapeake Bay. 1996. Riparian Forest Buffers. White paper. Annapolis, Md.: Chesapeake Bay Program. Barker L.S. 2003. Quantification of the Effectiveness of Agricultural Riparian Buffers to Protect Stream Health in Maryland's Coastal Plain and Piedmont Regions. Unpublished PhD Dissertation. College Park, MD: University of Maryland, Biological Resources Engineering Department. Boward DM, PF Kazyak, SA Stranko, MK Hurd and TP Prochaska. 1999. From the mountains to the sea: the state of Maryland?s freshwater streams. EPA 903-R-99-023. Maryland Department of Natural Resources, Monitoring and Non-tidal Assessment Division, Annapolis, Maryland. Chesapeake Bay Program. 1996. Final report of the Riparian Forest Buffer Panel. Publ. CBP/TRS 158/96 or EPA 903-R-96-015. Annapolis, Md.: Chesapeake Bay Program Day RL, Richards PL and RL Brooks. 1996. Chesapeake Bay Forest Buffer Inventory. Final Report, College of Agricultural Sciences and School of Forest Resources, Pennsylvania State University, University Park, December 31, 1996. 30pp. Giles RH Jr. and MK Trani. 1999. Key elements of landscape pattern measures. Environmental Management 23: 477-481. Kazyak, P.F. 1997. Maryland Biological Stream Survey. Sampling Manual. Maryland Department of Natural Resources, Chesapeake Bay Research and Monitoring Division, Annapolis, Maryland. Klauda R, Kazyak PF, Stranko SS, Southerland MT, Roth NE and JC Chaillou. 1998. Maryland Biological Stream Survey: A state agency program to assess the impact of anthropogenic stresses on stream habitat, quality and biota. Environmental Monitoring and Assessment 51(1-2): 299-316. Mercurio G, JC Chaillou and NE Roth. 1999. Guide to using 1995-1997 Maryland Biological Stream Survey Data. Prepared by Versar, Inc., Columbia, Md. for Maryland Department of Natural Resources, Monitoring and Non-Tidal Assessment Division. Annapolis, Maryland. CBWP-MANTA-EA-99-5. Roth NE, MT Southerland, G Mercurio, JC Chaillou, PF Kazyak, SS Stranko, AP Prochaska, DG Heimbuch and JC Seibel. 1999. State of the Streams: 1995-1997 Maryland Biological Stream Survey Results. Versar, Inc., Columbia, MD and Post, Buckley, Schuh and Jernigan, Inc., Bowie, MD. CBWP-MANTA-EA-99-6. 126 CHAPTER 5 ? INITIAL MODELING TO INVESTIGATE THE EFFECTS OF AGRICULTURAL RIPARIAN BUFFERS TO MITIGATE NITRATE STREAM POLLUTION INTRODUCTION This chapter presents the initial exploratory work to model the effect of agricultural riparian buffers on instream nitrate. The effect of buffers on instream nitrate was prompted by the regulatory interest in agriculture?s contribution of nitrate to the Chesapeake Bay. These exploratory procedures were performed on the data developed by combining landscape data from the 1998 University of Maryland Buffer Survey and nitrate data provided by the Smithsonian Environmental Research Center. The work is presented in three parts: I Examination of explanatory variables II Modeling nitrate change along a reach III Modeling nitrate - land use relationships 127 I. EXAMINATION OF DESCRIPTOR VARIABLES METHODS The UM Buffer Survey data included over 30 landscape descriptors that were candidates as explanatory variables for multiple linear regression models. Environmental descriptors are well known to cause collinearity problems in modeling, so work was performed to attempt to reduce the number of explanatory variables used by reducing the number of correlated variables. The SAS Proc Mixed procedure (SAS Institute Inc., 1988) was used to test the quantitative landscape descriptors for regional differences, using one average value per descriptor per watershed. The Fisher exact test was used to determine regional differences for qualitative landscape descriptors. Finally, the overall correlation matrix with all descriptors and all sites was developed and examined (Appendix 4). RESULTS AND DISCUSSION Proc mixed results. The SAS Proc mixed regression procedure was used to test the hypothesis that Coastal Plain sites showed differences from Piedmont sites. These models demonstrated that some variables did show a statistically significant difference between physiological regions (Table 11). All log values were tested, but only log values that showed Pr < 0.05 are presented. Regional differences were seen for all categories of landscape descriptors except channel morphology. This exercise was done to reduce the number of explanatory variables used in multiple regression 128 modeling. However, the use of the SAS R-square search procedure and collinearity diagnostics in the final modeling procedure eliminated the need to reduce initial variables. Fisher exact test results. Regional differences were seen for all qualitative landscape descriptors except drain tiles and channel shape (Table 12). Closer examination of these data revealed that there were very few data for ?presence of drain tiles?, so this descriptor was eliminated in later modeling work. Table 11. Least squares means values (and standard deviations) for quantitative landscape descriptors based on mixed models analysis of the significant difference between Coastal Plain and Piedmont sites, based on data from the 1998 University of Maryland Agricultural Riparian Buffer Survey. All data Coastal Plain Piedmont Pr>F Topography Riparian slope (%) 8.17 (9.65) 6.85 (1.13) 9.18 (1.02) 0.18 LOG riparian slope 0.79 (0.37) 0.69 (0.05) 0.88 (0.06) 0.06 Field slope (%) 4.78 (5.40) 2.82 (1.16) 5.66 (1.15) 0.13 Riparian slope: Field slope 2.66 (4.62) 3.15 (0.80) 2.79 (0.76) 0.76 Topographic index (m/m2) 3.87 (3.94) 19.56 (3.03) 0.03 LOG Topographic index .43 (.20) 1.23 (.14) 0.03 Surface Hydrology Drainage density (km2/km) 1.22 (.09) 1.12 (.08) 0.44 Site watershed area: stream distance (km2/km) 1.58 (1.99) 1.94 (0.37) 1.40 (0.36) 0.35 Stream gradient (%) 1.11 (0.86) 1.42 (0.22) 0.91(0.22) 0.15 Velocity (m/sec) 0.3 (0.2) 0.2 (0.05) 0.3 (0.05) 0.78 Tortuosity (km/km) 1.16 (.07) 1.28 (.02) 0.26 Stream order 1.83 (0.89) 1.62 (0.11) 1.99 (0.99) 0.04 LOG Stream order 0.21 (0.20) 0.16 (0.03) 0.25 (0.03) 0.03 Land Use % agriculture 55.75 (7.16) 39.54 (8.12) 60.79 (6.09) 0.10 % forest 54.9 (7.3) 23.3 (5.4) 0.03 %forest:%agriculture 1.9 (0.4) 0.4 (0.3) 0.04 129 Table 11. (continued) Least squares means values (and standard deviations) for quantitative landscape descriptors based on mixed models analysis of the significant difference between Coastal Plain and Piedmont sites (1998 University of Maryland Agricultural Riparian Buffer Survey). All data Coastal Plain Piedmont Pr>F Groundwater Hydrology Riparian water table (m) 1.2 (2.6) 0.16 Field water table (m) 3.7 (3.2) 0.26 Riparian K at surface (cm/hr) 3.8 (2.8) 0.06 Log riparian K at surface 0.36 (0.17) 0.29 (0.03) 0.42 (0.03) 0.02 Riparian minimum K (cm/hr) 3.0 (2.8) 1.6 (0.4) 4.2 (0.4) 0.01 LOG riparian minimum K 0.29 (0.20) 0.18 (0.03) 0.38 (0.03) 0.002 Field avg. K at surface (cm/hr) 4.8 (4.8) 0.28 Field minimum K (cm/hr) 3.2 (2.2) 0.17 LOG field minimum K (cm/hr) 0.33 (0.14) 0.29 (0.02) 0.36 (0.02) 0.03 Depth of riparian minimum K (m) 8.1 (3.0) 8.8 (0.3) 7.5 (0.3) 0.02 Depth of field minimum K (m) 6.1 (3.2) 7.2 (0.7) 4.6 (0.7) 0.03 LOG depth of field minimum K (m) 1.27 (0.22) 1.35 (0.06) 1.14 (0.06) 0.04 Channel Morphology Channel width (m) Channel depth (m) 3.3 (2.3) 1.00 Channel cross-sectional area (m2) 6.1 (8.0) 0.53 Channel width:depth ratio 5.8 (3.9) 0.51 Average bank angle 45.2 (18.4) 0.67 Buffers Total buffer width (m) 65.5 0.85 Log total buffer width 1.77 0.03 Average tree buffer width (m) 55.8 0.85 Log average tree width 1.09 0.03 Average grass buffer width (m) 7.4 0.73 Average shrub buffer width 1.0 0.09 Average in-channel buffer width (m) 0.4 0.05 Log average in-channel buffer width -0.4 0.04 DBH (cm) 36.1 0.22 Basal area (m2) 5.2 0.78 Buffer width:slope 1175 0.69 Shrub height (m) 1.9 0.99 Average length of buffer fragments (m) 885 (1183) 901 (319) 779 (328) 0.80 130 Table 12. Fisher exact probability indicating significant difference between Coastal Plain and Piedmont values for qualitative landscape descriptors, based on 1998 University of Maryland Agricultural Riparian Buffer Survey data. Groundwater Hydrology Units Fisher Exact Prob Riparian zone drainage class variable/poor/good/excellent 9.60E-07 Drain tiles presence yes/no 0.0692 Field drainage class variable/poor/good/excellent 9.10E-07 Land Use Adjacent field use crop/pasture/forest 4.70E-06 Channel Morphology Substrate mud/sand/gravel/cobble 7.83E-27 Rosgen score unitless 3.30E-07 Channel shape natural/triangular 0.8717 Channelization yes/no 7.53E-14 Streambank erosion high/medium/low 2.40E-05 Correlation results: The UM Buffer Survey provided at least one buffer/landscape description on 8 reaches of first through third order streams with upstream/downstream nitrate values from the SERC nitrate survey (Figures 15 and 16). There were 4 reaches in the German Branch watershed and 4 reaches in the Gunpowder watershed. These data provided a set of 35 sites with both riparian landscape and nitrate information, identified as the SERC* data set. 131 Figure 15. Map of the 1998 University of Maryland Agricultural Riparian Buffer Survey?s German Branch watershed (5,690 ha), with SERC nitrate sampling sites (n = 21), assigned average spring baseflow nitrate values, forest and CAFO areas. survey sites, grey = forested area, dark grey = CAFO area 132 Figure 16. Map of the 1998 University of Maryland Agricultural Riparian Buffer Survey?s Gunpowder watershed (12,880 ha) with SERC nitrate sampling sites (n = 14), average spring baseflow nitrate values, forest and CAFO areas. survey sites, grey = forested area, dark grey = CAFO area A strong scale effect was discovered among the landscape descriptors. Intercorrelation values among landscape and land use descriptors that were developed at the watershed scale were often as high as 0.7-0.8. Absolute values for correlation coefficients of site-level scale descriptors vs. instream nitrate were all less than 0.35. Absolute values for correlation coefficients of watershed-scale descriptors vs. instream nitrate were all greater than or equal to 0.44 (Table 13). These high correlation values were not, in fact, due to a strong relationship between the landscape descriptors and instream nitrate, but were an artifice of censoring the data and causing repeated values 133 for different sites. Therefore, in order to more accurately reflect physical relationships, all final models were limited to site catchment data. Table 13. Correlation coefficients for landscape descriptors vs. instream nitrate and scale of influence, based on SERC* data (35 sites from the 1998 University of Maryland agricultural Riparian Buffer Survey that were assigned nitrate values). Landscape descriptor R scale % agriculture in watershed -0.46 w % forest:ag in watershed 0.46 w Topographic index 0.46 w Stream tortuosity 0.45 w % forest in site catchment 0.44 w Field water table 0.34 s CAFO area in site catchment 0.34 s Field slope 0.31 s Basal area -0.26 s Riparian K minimum depth -0.26 s Riparian slope 0.25 s Riparian K @ minimum depth -0.24 s Forest buffer width -0.24 s % ag in site catchment 0.24 s Riparian K @ surface -0.23 s Stream gradient 0.18 s Grass buffer width 0.18 s Buffer tree portion width -0.18 s DBH 0.17 s Riparian water table depth 0.16 s % crops in site catchment 0.15 s Buffer slope:field slope -0.13 s Total buffer width -.006 s w = watershed scale: values calculated for survey watersheds s = site scale: values calculated for site catchments or field site location 134 II. MODELING NITRATE CHANGE ALONG A REACH The 1998 University of Maryland Agricultural Buffer Survey was part of a larger study funded by the Fund for Rural America, under the United State Department of Agriculture. Because of the focus on reduction of agricultural contributions to nitrate in the Chesapeake Bay, one of the objectives of this project was to determine buffer impact on instream nitrate. METHODS Multiple linear regression modeling was performed in SAS (SAS Institute Inc., 1988) using the R-square search method. Explanatory variables were the landscape and land use descriptors that had shown highest correlations from the Proc Mixed work. The criteria variable was the change along the reach in average spring baseflow nitrate concentration, expressed in ppm/km. These values were developed as the difference in average spring nitrate values for the upstream and downstream sites, divided by the distance between sites. Development of site nitrate values for the SERC* data set is presented in Appendix 5. 135 RESULTS AND DISCUSSION The overall average change of nitrate along a reach (from upstream to downstream) was an increase of 0.26 ppm/km. This was surprising, since the average buffer width was 67 m (193 ft) and no sites were completely without a buffer. This was primarily due to behavior in the German Branch watershed (Figure 17). Since three of the four German Branch reaches showed a nitrate gain, the overall nitrate change in German Branch was nitrate gain (Table 14). The land use was examined to explain this rather surprising result. Both watersheds were heavily agricultural areas, with an average of 62% agriculture and 57% crops in the site catchments. However, all German Branch sites had higher % agriculture and % crops than the Gunpowder sites, 3 sites had smaller buffers than those in the Gunpowder, and all German Branch sites had less % forest. Correlation values of broad land use terms were strong and reasonable; % agriculture was related to increase in nitrate (0.83) and % forest was related to reduction in nitrate (-0.83). However, correlation values of landscape descriptors suggested senseless relationships. Increasing buffer width was related to increase in nitrate (0.29) and increase in confined animal feeding operation (CAFO) area was positively related to reduction in nitrate (-0.13). 136 Figure 17. Change in average spring baseflow nitrate concentration along a reach (ppm/km) vs. estimated average buffer width along the reach, for upstream-downstream sampled segments of the German Branch (n = 4) and Gunpowder (n = 4) watersheds, based on the 1998 University of Maryland Agricultural Riparian Buffer Survey. -2 0 2 0 100 200 300 400 buffer width n itr a te dr o p ga in Gunpowder German Branch Table 14. Average values of nitrate decrease along a reach, riparian and land use for sites in the contributing watershed, based on SERC* data. DATA NO3 (ppm/k m) Ag % Crop % Pasture % Forest % CAFO (ha) Buffer width (m) Basal area (m2) all data -0.26 62. 57 5 38 4 67 4.2 Gunpowder 0.06 50 39 11 50 6.5 86 3.6 German Branch -0.57 74 74 0 26 1.5 48 4.8 The SAS R-square search produced the following model for change in nitrate as a function of the explanatory variables:  NO3 = -1.71 + 0.02 (% crop) + 0.003 (buffer width, ft.) (5.1) 137 The model was significant (Pr > F = 0.005) and the adjusted R2 was 0.84. No collinearity problems were evident and the residuals were normally distributed. However, this model contained an illogical sign indicating that wider buffers were related to nitrate gain along a reach. This result was concluded to be an artifice of the data. The relationship in the model reflected the individual sites in this very small data set, not larger trends. Whereas this study was investigating non-point-source (NPS) effects, these data suggested that point sources might be effectively overwhelming NPS effects in this small watershed (5,590 ha). 138 III. MODELING NITRATE - LAND USE RELATIONSHIPS METHODS This portion of the research investigated more closely the nitrate ? land use relationships in the German Branch and Gunpowder watersheds. Spring baseflow nitrate values were considered in relation to site location, adjacent land use, watershed, and stream order. Visual examination of Figures 16 and 17 led to the suspicion that confined animal feeding operations (CAFO) might be acting as point sources and overwhelming NPS effects. To study this relationship further, CAFO factors were developed for the site catchments: CAFO area = CAFO area in site catchment (ha) (5.2) CAFO factor = CAFO area / distance from site (ha/ km) (5.3) CAFO factor2 = CAFO area / distance from site2 (ha2/km2) (5.4) Correlation analysis was conducted and models were then developed in SAS (multiple linear regression) and S-PLUS (regression tree analysis). SAS and S-PLUS are both integrated suites of software facilities for exploratory data modeling, statistical analysis and graphical display. S-PLUS is an enhanced version of S distributed by Statistical Sciences, Inc., Seattle, Washington, that includes a regression tree function. 139 RESULTS AND DISCUSSION Whereas visual examination of the maps indicated ?hot spots? near CAFOs, the correlation analysis verified this effect. German Branch, a Coastal Plain watershed, had fewer and smaller CAFOs than the Gunpowder watershed (Table 15). The correlation of CAFO area in the site catchment vs. nitrate was fairly strong for all sites (R = 0.34). The complete correlation matrix for the SERC* data set is presented in Appendix 6. However, when the correlation was conducted on the individual watersheds, German Branch showed a much stronger response (R = 0.51) than the Gunpowder (0.12). Table 15. Confined Animal Feeding Operation (CAFO) and instream baseflow nitrate information for German Branch and Gunpowder watersheds, based on SERC* data. # of CAFOs CAFO area (ha) Ave. NO3 German Branch 3 3-6 0.58 Gunpowder 5 8-17 3.57 Linear regression modeling indicated that the primary controls on instream nitrate were agricultural land use. A two-term model used percent agriculture and CAFO area in the site catchment as the explanatory variables (R2 = 0.17) (Table 16). Percent crops have been shown in past studies to be a strong influence on instream nitrate (Taraba et al., 1996) These data showed percent agriculture to be a stronger influence than percent crops. CAFO presence was found to be a profound influence. 140 Therefore the inclusion of CAFOs in the general agriculture designation may have caused percent agriculture to show a stronger control over nitrate than percent crops. Table 16. Linear regression model predicting instream nitrate at a site using SERC* data of the 1998 University of Maryland Agricultural Riparian Buffer Survey. Variable Parameter estimate b0 0.449 b1 (CAFO area) 0.038 b2 (% agriculture in site catchment) 0.215 SUMMARY This preliminary work provided guidance that was used in all subsequent modeling exercises. Models were successfully developed to predict nitrate at a site from onsite and catchment descriptor variables. Based on these results, land use was expected to be the strongest predictor of nitrate. Detailed site characteristics were not shown to be important predictors for instream nitrate. These models were developed from the 1998 University of Maryland Agricultural Riparian Buffer Survey data. Because there was no measurement of CAFO area in the MBSS, models developed from these data will be unable to capture CAFO effect. 141 CITED REFERENCES SAS Institute Inc. 1988. SAS/STAT User?s Guide. Release 6.03 Edition. Cary, N.C.: SAS Institute Inc. Taraba J.L., J.S. Dinger, L.V.A. Sendlein and G.K. Felton. 1996. Land use impacts on water quality in small karst agricultural watersheds. ASAE meeting paper 96-2086. July 14-18, 1996. Phoenix, AZ. 142 CHAPTER 6 ? MODELING THE EFFECTS OF AGRICULTURAL RIPARIAN BUFFERS TO MITIGATE NITRATE STREAM POLLUTION ABSTRACT Whereas numerous studies have considered buffer effect on groundwater at the field scale, most decisions are made at the watershed scale, using surface water quality as the regulatory criteria of interest. The purpose of this study was to quantify the effect of agricultural riparian buffers on instream nitrate utilizing readily available data. Multiple linear regression models and regression tree models were developed to predict instream nitrate concentration from landscape characteristics at both the site and the site catchment scales. Two sets of models were developed. In the first modeling exercise, nitrate, landscape and buffer information were all obtained from the 1995-1997 Maryland Biological Stream Survey data set. In the second modeling exercise, nitrate data were provided by the Smithsonian Environmental Research Center in Edgewater, Maryland, and landscape and land use information were obtained from the 1999 University of Maryland Agricultural Buffer Survey. In both approaches, the original data were partitioned to develop a data set that represented 143 spring baseflow conditions at agricultural sites in heavily agricultural watersheds of the Coastal Plain and Piedmont physiographic regions. It was demonstrated that catchment characteristics exert a much stronger influence on instream nitrate than riparian buffers. Extent of agricultural land use showed a strong relationship to instream nitrate (R = 0.55). Confined animal feeding operations (CAFOs) were also often associated with elevated instream nitrate levels, or ?hot spots?. Multiple linear regression, regression tree models and ANOVA all indicated that agricultural land use, especially the presence of CAFOs, overshadowed any on-site buffer effects in control over instream nitrate at that site. These results were repeated for a subset of first-order stream sites. Although previous studies have reported significant reduction of nitrate in groundwater passing laterally to a stream, buffers have little effect on total instream nitrate adjacent to the site. INTRODUCTION Background In 1983, concern about worsening water quality in the Chesapeake Bay led to the formation of the Chesapeake Bay Program (CBP), a multi-jurisdictional cooperative effort to restore the nation's largest estuary. The Chesapeake Executive Council set restoration of the Bay's "living resources" as a primary goal in 1987. To achieve that goal, it called for a number of actions to improve water quality, including a 40% reduction in the amount of nitrogen and phosphorus entering the Bay (Chesapeake Bay Program, 1996). 144 Efforts to implement these actions focused first on coastal tributaries of the Bay. By 1994, however, attention had expanded to the complete watershed, with the development of a set of goals and actions to increase the focus on riparian stewardship and to enhance efforts to conserve and restore riparian forest buffers (Alliance for the Chesapeake Bay, 1996). Forest buffers were singled out in part for their nutrient reduction capabilities, but also because they were reminiscent of the natural landscape in which the stream channel and stream ecosystems of the Bay watershed had evolved. In October 1996, the Executive Council of the CBP established a goal of restoring riparian forest buffers on 2010 miles of streams in the Chesapeake Bay watershed by 2010. This action exemplified the widespread acceptance of riparian buffers as an important tool for the reduction of non-point-source pollution. In May of 2002, the CBP published nutrient reduction estimates for various management plans in the Chesapeake Bay watershed. One plan estimated reduction of 62 millions pounds of nitrogen per year through "maximum participation" in current voluntary pollution reduction programs (Alliance for the Chesapeake Bay, 2002). Projected implementation and effectiveness of agricultural riparian buffers was a major portion of this pollution reduction estimate. There are now over 700 publications on the water quality functions of riparian buffers (Correll, 2000) and there is general agreement on the qualitative capacity of riparian buffers to mitigate non-point-source (NPS) pollution (Peterjohn and Correll, 1984; Lowrance et al., 1984, 1997; Barling and Moore, 1994; Correll, 1996; Hill, 1996). However, attempts to quantitatively relate riparian information, watershed information and water quality have met with mixed success and very few studies have 145 been conducted at the watershed scale (Gregory et al., 1991; Whigham et al., 1988; Osborne and Wiley, 1988; Wissmar and Beschta, 1998; Tufford et al., 1998; Perry et al., 1999; Herlihy and Kaufman, 2000). These studies have not resulted in a tool that can enable policymakers to predict or quantify the water quality effect of a given riparian buffer scenario at the watershed level. Purpose Whereas numerous studies have considered buffer effect on groundwater at the field scale, most decisions are made at the watershed scale, using surface water quality as the regulatory criteria of interest. The purpose of this study was to quantify the effect of agricultural riparian buffers on instream nitrate concentration under spring baseflow conditions. Multiple linear regression models and non-parametric regression tree models were developed to predict instream nitrate based on riparian landscape and site catchment characteristics. The models were then examined to elucidate the nature and strength of buffer effect on instream nitrate. These results are of theoretical use to scientists investigating the functional role of buffers in water quality protection, as well as practical use to the agencies that administer the funds to establish riparian forest buffers in Maryland. While these are empirical results specific to Maryland, the general findings are of use to other locations throughout the United States where the establishment of forest buffers is considered a water quality management tool. Objectives 1. To determine the relative importance of riparian buffer vs. catchment characteristics on instream nitrate. 146 2. To verify results through the use of two independent data sets and models developed from two different algorithms (multiple linear regression and regression tree analysis). METHODS Overall summary of methods Empirical models were developed to predict instream nitrate concentration at a site from landscape characteristics at both watershed and riparian landscape scales. Instream nitrate concentration data were combined with easily obtainable watershed and site data to develop physio-chemical profiles of agricultural sites in the Coastal Plain and Piedmont physiographic regions of Maryland's Chesapeake Bay watershed. Both multiple linear regression and regression tree modeling were used to investigate the water quality impact of agricultural land use and riparian buffer management. Data development Two modeling approaches were utilized, using different data sets to develop the models. For each approach, the original data were either modified or partitioned in order to describe a population of agricultural sites. Site characterization. In both data sets, each survey site consisted of the stream, riparian zone and adjacent land along both sides of a 30 to 90 meter stream reach. Two scales of data were considered. Site characterization provided ground- truthed information about the local riparian landscape (Figure 18). These data were 147 obtained from on-site surveys using a rapid assessment approach. Information at the site watershed scale provided insight about cumulative impacts and sources of NPS pollution. These data were developed from soil surveys, USGS topographic maps, and publicly available EPA, USGS and state GIS files (Appendix 3). Figure 18. Schematic of the agricultural riparian landscape as defined for the 1998 University of Maryland Agricultural Riparian Buffer Survey. Data for the SERC* models. In the first modeling approach, two data sets were combined. Landscape and buffer information were obtained from the University of Maryland (UM) Agricultural Buffer Survey, a field survey of over 200 sites. Details of this research were presented in Chapter 2. The purpose of the survey was to characterize the nature and extent of riparian buffers in agricultural watersheds of Maryland's Coastal Plain and Piedmont physiographic regions. Site-level data were obtained by fieldwork performed in the summer of 1998. A suite of structural riparian buffer landscape measurements was taken to describe the agricultural riparian 148 landscape for each site. Detailed procedures are presented in Appendices 1-2. Measurements fell into six key categories: land use, buffer characterization, topography, surface hydrology, groundwater hydrology, and stream channel morphology. Data for each site?s catchment were derived from GIS work using land use/land cover files available from the Maryland Department of Planning Appendix 3). The explanatory variables from the SERC* data set that were used in developing models to explain instream nitrate are presented in Table 17. The Smithsonian Environmental Research Center (SERC) in Edgewater, Maryland provided the nitrate data used in this modeling approach. In 1996, SERC conducted a sampling survey of nitrate in selected stream networks of Maryland's Chesapeake Bay watershed. Nitrate concentration was measured along the mainstem and at upstream-downstream locations on low-order feeder streams within these stream networks. Grab samples were taken approximately every six weeks to produce eight samples for each site. 149 Table 17. Explanatory variables from the SERC* data set used to develop models predicting instream nitrate. Descriptor Category Riparian Landscape Descriptor Units Land Use Site contributing watershed % agriculture2 Site contributing watershed % crops2 % % Buffer width1 m Length of buffer fragment3 km Buffer Character- ization Forest buffer DBH1 cm Forest buffer basal area1 m2 Stream gradient1 %Surface Hydrology Stream velocity1 m/sec Ground- water Hydrology3 Riparian water table depth Field water table depth m m Riparian buffer surface K * Adjacent field surface K * cm/hour cm/hour Riparian buffer minimum K * Adjacent field minimum K * cm/hour cm/hour Riparian buffer depth of minimum K * Adjacent field depth of minimum K * m m Channel Morph- ology1 Channel width m Channel depth m Channel cross-sectional area m2 Channel width : depth ratio unitless * K = hydraulic conductivity 1 from on-site field work 2 from GIS work 3 from soil survey 150 Two watersheds were common to both the UM Agricultural Buffer Survey and the SERC nitrate survey. Locations of the German Branch watershed (Eastern Shore, Coastal Plain physiographic region) and Gunpowder watershed (western shore, Piedmont physiographic region) are shown in Figure 19. Watersheds, forested area and site locations are presented in Figures 20 and 21 (not to scale; the Gunpowder watershed (128.8 ha) was roughly twice as large as the German Branch watershed (56.9 ha)). Figure 19. Map of 1998 University of Maryland Agricultural Riparian Buffer Survey watershed locations superimposed over 16-digit Hydrologic Unit Area (HUA) code watersheds as defined by USGS. 151 Figure 20. Map of 1998 University of Maryland Agricultural Riparian Buffer Survey German Branch watershed (5,690 ha), with SERC nitrate sampling sites (n = 21), assigned average spring baseflow nitrate values, forest and CAFO areas. survey sites, grey = forested area, dark grey = CAFO area 152 Figure 21. Map of 1998 University of Maryland Agricultural Riparian Buffer Survey Gunpowder watershed (12,880 ha) with SERC nitrate sampling sites (n = 14), average spring baseflow nitrate values, forest and CAFO areas. survey sites, grey = forested area, dark grey = CAFO area Information from the UM Buffer Survey and SERC nitrate survey was used to develop physio-chemical characterization of 35 sites for use in modeling. This data set was designated "SERC*". SERC* data parameters and summary statistics are presented in Table 18, and the correlation matrix is presented in Appendix 6. All sites were on low-order streams, either within 20 meters of a SERC nitrate sampling site or 153 between two SERC sampling sites. An average spring nitrate value was developed for each site. SERC sample values for March through May were averaged to develop an average spring value for each SERC sampling site. If a UM Buffer site was within 20 meters of a SERC sampling site, the SERC nitrate value was assigned to the UM site. If the UM Buffer site was between two SERC sampling sites, the nitrate values were assigned by linear interpolation. SERC site values provided the upper and lower values for interpolation, and the distance along the stream between the sites was the measure used for interpolation. Development of site nitrate values is presented in Appendix 5. Landscape information was obtained from UM Buffer Survey data. Because these models were to be compared with models derived from MBSS data, the maximum defined buffer width was reduced from the UM Buffer Survey value of 1000 feet to 50 meters, the maximum value used in the MBSS. Visual examination of Figures 21 and 22 led to the suspicion that confined animal feeding operations (CAFO) might be acting as point sources and overwhelming NPS effects. To study this relationship further, CAFO factors were developed for the site catchments: CAFO area = CAFO area in site catchment (ha) (6.1) CAFO factor = CAFO area / distance from site (ha/ km) (6.2) CAFO factor2 = CAFO area / distance from site2 (ha2/km2) (6.3) 154 Table 18. Values for SERC* landscape descriptors: all sites, German Branch and Gunpowder watersheds. All sites (n = 35) German Branch (n = 21) German Branch (n = 14) Variable Mean S.D. Mean S.D. Mean S.D. NO3 (ppm) 3.38 2.03 2.63 2.25 4.50 0.88 % agriculture 66.42 17.30 70.62 19.29 60.12 11.78 % crops 63.46 18.81 70.15 19.74 53.42 12.05 CAFO (ha) 1.78 3.86 0.58 1.55 3.57 5.43 CAFOfctr 1.55 3.73 0.74 1.47 5.51 5.51 CAFOfcsq 1.21 4.55 0.51 1.28 7.04 7.04 Forest buffer width (m) 40.8 67.9 36.6 67.1 47.5 71.9 Grass buffer width (m) 0.7 1.8 0.7 1.9 0.4 1.6 Total buffer width (m) 57.9 71.6 53.6 71.3 64.3 74.4 DBH (in) 31.4 18.5 30.5 19.1 33.0 18.3 Basal area (m2) 4.7 3.1 5.7 3.2 3.1 2.0 Riparian slope (%) 9.06 6.92 5.73 3.88 14.08 7.55 Adjacent field slope (%) 5.87 5.45 3.17 2.31 9.93 6.33 Riparian slope:field slope 2.23 1.71 2.33 1.78 2.09 1.64 Stream gradient (%) 0.88 0.45 0.81 0.42 0.99 0.48 Stream order 2.11 0.68 2.19 0.81 0.39 0.39 Riparian water table depth(m) 0.6 0.5 0.6 0.6 0.7 0.5 Riparian surface K (cm/hr) 4.3 2.0 4.8 2.5 3.3 0.00 Riparian minimum K (cm/hr) 4.1 2.3 4.6 2.8 0.00 0.00 Riparian min. K depth(m) 7.1 3.0 6.7 3.8 7.7 0.8 Field water table depth (m) 2.9 1.6 2.2 1.5 3.8 1.0 Field surface K (cm/hr) 3.3 0.2 3.2 0.2 3.3 0.1 Field minimum K (cm/hr) 3.0 0.7 2.8 0.9 3.3 0.1 Field minimum K depth (m) 3.6 1.8 3.2 2.1 0.15 0.15 Channel width (m) 4.6 2.7 5.2 2.4 3.6 2.9 Channel depth (m) 1.2 0.6 0.9 0.4 0.5 0.7 Channel cross-section (m2) 57.45 69.25 53.19 42.08 64.32 0.76 Channel width:depth 5.01 4.08 6.24 4.15 2.50 2.50 Bank angle (degrees) 35.98 17.48 39.19 15.64 27.42 20.69 155 Data for the MBSS models. In the second modeling exercise, instream nitrate concentrations, landscape and buffer information were obtained from the 1995-1997 Maryland Biological Stream Survey (MBSS) data set. The MBSS is an ongoing effort by the Monitoring and Non-Tidal Assessment Division of Maryland's Department of Natural Resources (DNR). The purpose of the MBSS is to assess water quality, physical habitat, and biological conditions in first through third order, non-tidal streams throughout the state of Maryland (Klauda et al., 1998). About 300 sites (75 meter stream segments) were sampled each year during the three-year period 1995-97. These data have been made available to the public through the Maryland DNR website. MBSS data collection sites were described by approximately 160 variables. Although buffer characterization was not the primary purpose of the survey, limited information was taken on the riparian buffer at the site (presence, vegetation type and width). Site-level data (including instream nitrate) was obtained by field surveys performed during the spring and early summer. Land use information for the site watersheds was obtained by GIS work. The explanatory variables from the MBSS* data set that were used in developing models to explain instream nitrate are described briefly: ANC: Acid neutralizing capacity was given in micro-eq/L. Area: the site catchment area, reported in acres. Average thalweg depth: the deepest portion of the lateral transect of the stream, was measured in centimeters at the 0, 25, 50, and 75 meter 156 points of the sample segment. Average thalweg depth was the average of these measures (cm). Blackwater conditions: blackwater was coding indicating whether the site was located in a blackwater stream. A value of ?1? indicated that the site was blackwater, a value of ?0? indicated that it was not. Conductance: reported in micro-mho/cm. Embeddedness: embeddedness was a given as the percentage that gravel, cobble, and boulder particles were surrounded by sediment or flocculent material. Flow, or streamflow, was calculated from raw data collected at each stream segment and reported as cfs. At most sites, a standard transect method was employed. The field crew constructed a velocity/depth profile of the segment using a current meter to measure stream velocity and recording stream depth at 5 to 20 regular intervals across the stream. At each location along the transect, velocity was measured at a point 0.6 of the distance from the water surface to the bottom. Calculation of discharge from raw velocity, depth, and lateral location data followed standard procedures. Instream habitat and bank stability: Instream habitat structure and bank stability were assigned qualitatively, based on visual observations within the 75-m stream ample segment. Scores ranged from 0 to 20 in four categories from poor to optimal. Instream habitat was based on 157 the percentage of stable habitat. Bank stability was based on the percent of banks with erosional scars or erosion potential. Maximum depth: maximum depth within the 75-meter segment (cm). Percent land use: land use in the site catchment was based on the 1996 Multi-Resolution Land Use Classification (MRLC) land cover data base for EPA Region III (MRLC, 1996a; MRLC, 1996b). High intensity urban land use: a land use classification describing heavily built up urban centers with very little vegetation and high population densities. pH: given in standard pH units. Riparian width: the width of the vegetated riparian buffer was estimated (m), to a maximum of 50 m. If the buffer was greater than or equal to 50 m, a value of 50 was entered. This measure was the width of the riparian buffer on the side of the stream with the smallest buffer. SO4: Sulfate, nitrate nitrogen, and dissolved organic carbon concentrations were given as mg/L. Stream order: represents the Strahler (1957) convention used for ranking stream reaches by order. Stream order determination was based on a stream reach file digitized from 1:250,000 scale topographic maps for the MBSS in 1987. Average velocity: thalweg velocity was measured with a flow meter at the deepest portion of the lateral transect at the 0, 25, 50, and 75 meter points of the sample segment and presented in meters per second. 158 Water temperature: given in degrees C. Woody debris: the number of pieces of woody debris at each site was recorded. The 1995-1997 MBSS data set contains 907 sites representing all physiographic regions and random land uses in Maryland. Sites with the following characteristics were partitioned from the original data set: agricultural land use, catchment percent agricultural greater than 35%, located in the Coastal Plain and Piedmont physiographic regions. The resultant reduced data set of 279 sites was designated "MBSS*". MBSS* stream and landscape descriptors and summary statistics for agricultural sites in the Coastal Plain and Piedmont physiographic regions of Maryland are presented in Table 19. The MBSS* correlation matrix is presented in Appendix 7. Table 19. Riparian landscape descriptors and summary values for MBSS* data. Variable n Mean Std Dev Min Max NO3 (ppm) 278 3.49 2.28 0.26 16.16 Riparian width (m) 279 34.6 21.28 0 50.00 Water temperature (oC) 279 19.79 3.36 12.00 30.90 % agriculture 279 63.67 13.39 35.09 91.92 % urban land use 279 2.25 4.05 0 31.80 % forest 279 33.33 12.68 7.19 61.53 Stream order 279 1.94 0.80 1.00 3.00 Dissolved oxygen (ppm) 278 8.47 1.92 1.00 14.00 pH 279 7.09 0.48 5.27 8.82 Conductance, field (?-mho/cm) 279 176.10 91.14 54.00 980.00 Acid Neut. Capacity (?-eq/L) 278 435.30 285.54 -26.90 1722.2 SO4 (ppm) 278 11.28 6.00 1.43 33.91 Adjacent pasture presence (no = 0, yes = 1) 279 0.14 0.35 0 1.00 159 Table 19. (continued) Riparian landscape descriptors and summary values for MBSS* data. Variable n Mean Std Dev Min Max Instream habitat 279 12.86 4.06 1.00 19.00 Epifaunal substrate 279 11.33 4.64 1.00 19.00 Velocity depth (cm) 279 11.54 4.30 1.00 20.00 Pool quality 279 12.66 3.89 1.00 19.00 Riffle quality 279 11.97 4.88 0 20.00 Channel alteration 279 9.93 4.44 0 18.00 Bank stability 279 10.36 4.14 1.00 19.00 Embeddedness 279 55.32 32.26 0 100 Channel flow (m3/s) 279 7.50 1.56 3.48 34.80 Shading 279 66.36 23.21 5.00 98.00 Remoteness 279 11.11 4.63 1.00 19.00 Aesthetics 279 14.15 3.44 1.00 19.00 Woody debris 279 3.73 3.78 0 21.00 Number rootwads 279 1.91 2.39 0 25.00 Max. channel depth (cm) 279 66.45 28.92 9.00 200.00 Stream gradient (%) 279 0.90 1.17 0 10.50 Avg. stream width (m) 279 5.68 4.04 0.05 23.03 Avg. thalweg (m/s) 279 10.05 5.18 0.60 25.66 Avg. velocity (m/s) 279 0.19 0.14 0 1.72 Site catchment area (ha) 279 3000 3816 31 29069 % wetlands 279 0.28 0.33 0 2.50 % barren 279 0.08 0.40 0 4.35 % water 279 0.32 0.48 0 4.64 % high urban land use 279 0.30 0.51 0 3.82 % high urban land use 279 1.94 3.63 0 27.98 % pasture 279 21.84 10.50 1.82 62.07 % probable row crops 279 29.86 10.42 3.51 85.47 % row crops 279 11.96 8.65 0.39 34.43 % coniferous forest 279 3.06 4.61 0 25.41 % deciduous forest 279 21.50 10.48 2.00 55.73 % mixed forest 256 3.71 2.74 0 12.29 % emergent wetlands 203 0.19 0.34 0 2.50 % woody wetlands 123 3.36 6.31 0 29.92 Blackwater conditions (no = 0, yes = 1) 50 0.04 0.19 0 1.00 160 Comparison of the data sets. The two data sets used for modeling (MBSS* and SERC*) were parsed from the original data sets in order to provide description of the same population of sites. Both the MBSS* and SERC* data sets represented agricultural sites in agricultural watersheds of the Coastal Plain and Piedmont regions of the Central and Eastern portions of the state. Average percent agriculture, average NO3, and average buffer width were not significantly different between the two data sets (t-test,  = 0.05). Therefore, the two data sets were considered comparable. Table 20 provides a summary of the site characterization descriptors for the two data sets. Table 20. Comparison of key parameters of the SERC* and MBSS* data. Data set MBSS* SERC* # Sites 279 35 Physiographic regions Coastal Plain, Piedmont Coastal Plain, Piedmont State areas Eastern Shore, Central Md. Eastern Shore, Central Md. Adjacent land use Forest, Crop, Pasture Forest, Crop, Pasture Average % agriculture 63.7 66.4 Range % agriculture 35-92 37-100 Average NO3 , ppm 3.5 3.4 Range NO3 0.3-16.2 0.2-7.8 Average buffer width, m 34.6 30.2 Max buffer width, m 50.0 50.0 Modeling Correlation analysis was conducted and models were then developed in SAS (multiple linear regression) and S-PLUS (regression tree analysis). SAS and S-PLUS are both integrated suites of software facilities for exploratory data modeling, statistical analysis and graphical display. S-PLUS is an enhanced version of S distributed by Statistical Sciences, Inc., Seattle, Washington, that includes a regression tree function. 161 All models were expected to take the general form that instream nitrate was a function of source terms (such as agricultural land use) and mitigation factors. Any buffer term in the models would be a quantification of buffer effectiveness. Validation of MBSS* models was conducted using a data set of 10 percent of sites withheld from the MBSS* data (n = 27p). The accuracy of these nitrate predictions was assessed using an R2 value developed from the correlation of observed and predicted values. SERC* models were not validated because of the small number of sites (35 sites). Multiple linear regression models. Multiple linear regression models were constructed using landscape measures as explanatory variables to predict instream nitrate. Therefore the model structure was Y = B0 + B1X1 + B2X2 + ? BnXn + error (6.4) where Y = average spring nitrate concentration for a site, X1...Xn = site descriptor measurements. Linear regression modeling was performed with SAS software, using a Proc Reg R2 search procedure (SAS, 1988). The R2 search procedure in SAS is more comprehensive than traditional stepwise regression methods for multiple regression model development, because every possible combination of terms is analyzed. This procedure constructs multiple linear regression models from all combinations of explanatory variables. Search results are presented as a list of model terms and associated adjusted R2 values. In order to compare the explanatory power of models 162 with different numbers of terms, the adjusted R2 statistic was used. This statistic is adjusted for the degrees of freedom of the sums of squares associated with R2 and was calculated as Adj R2 = 1-[(SSE / (n-p))/ (CSS / (n-1))] = 1-[(n-1)/ (n-p)] (1- R2). (6.5) where SSE = sum of squares of error, n = sample size, p = degrees of freedom, CSS = corrected sum of squares, R2 = square of the correlation between response variable and predicted values. [As a cautionary note, this procedure must be used with a data set without missing values to obtain consistent Adj-R2 values.] Suggested model terms were then used in a standard regression procedure to determine parameter coefficients and significance. Regression tree models. Like multiple linear regression models, regression trees explain variation of a single response variable by one or more explanatory variables. The tree is constructed by repeatedly splitting the data into two mutually exclusive groups, each of which is as homogeneous as possible. The objective is to partition the response into homogenous groups, but also to keep the tree reasonably small. Each group is characterized by the mean value of the response variable, group size, and the values of the explanatory variables that define it (De'ath and Fabricius, 2000). Regression trees do not have non-significant terms. However, the terms are arranged in descending order of explanatory power. Regression tree models cannot 163 predict a continuous distribution of nitrate values (as do linear regression models), but group sites into ?terminal nodes? with an average nitrate value for each group of sites. Therefore, regression tree models visually present the range of nitrate values found among the data, distributed among different conditions. For example, in Figure 28, average nitrate values ranged from 1.5 ppm nitrate at the group of sites with lowest percent agriculture in the site catchment to 5.90 at sites with highest percent agriculture. The efficacy of the overall model and its parts is described by Residual Mean Deviance (RMD), which is the sums of squares of the error corrected for the degrees of freedom. It should be noted that R2 is a measure of explained variation and RMD is a measure of unexplained variation, and that although the range of R2 is constrained to values between 0 and 1, RMD values can take values greater than 1. Because the model is in the form of a ?decision tree?, these models are more effective than multiple linear regression in elucidating threshold effects. Comparison of linear and regression tree models. The linear models and regression tree models developed from the two data sets were compared between and among one another. The primary explanatory variables identified by the models were compared, as well as their scales of effect. Diagnostics of several models indicated that, although the SAS R-square search provided a selection of linear models with 2 - 20 terms, no more than 5 terms were consistently significant ( of 0.05%) for models constructed from the MBSS* data, and no more than 3 terms were consistently significant for models constructed from the SERC* data. Therefore, the terms and explanatory power of the 3- or 5-term SAS multiple linear regression models were compared with the terms and explanatory power of 3- or 5-term regression tree 164 models. Developing a common measure of accuracy enabled comparison of the relative accuracy of comparable linear and regression tree models. A coefficient of determination (R2) was created for the regression tree models that could be compared to the R2 of the linear regression models. R2 for the pruned regression tree models was developed as follows: R2 = 1 ? (Residual SSE / Corrected Total SSE) (6.6) where Residual SSE =  (NO3i ? NO3predicted)2 =  (deviance for terminal node)2 Corrected Total SSE =  (NO3i ? NO3mean)2 . RESULTS AND DISCUSSION Graphical data exploration and correlation analysis Land use. Land use distribution was compared for the two modeling data sets, as shown in Figure 22. The original 16 land use categories in the MBSS data set were collapsed to the 3 categories used in the SERC data set (agriculture, forest and other). The MBSS* and SERC* data sets were very similar in their depiction of land use distribution. 165 Figure 22. Comparison of land use distribution for agricultural riparian landscapes in Maryland?s Coastal Plain and Piedmont physiographic regions, as determined from MBSS* and SERC* data sets. The SERC* data set (n = 35) was constructed from landscape data from the 1998 University of Maryland Agricultural Riparian Buffer Survey and the 1996 SERC Agricultural Watersheds Stream Chemistry Survey. MBSS is the 1995-1997 Maryland Biological Stream Survey (MBSS*) data set. MBSS* is a subset of the MBSS data, representing riparian agricultural sites in catchments with > 35% agriculture, in the Coastal Plain and Piedmont physiographic regions of Maryland. MBSS lo ag is a subset of the 1995-1997 MBSS data representing riparian agricultural sites in catchments with < 35% agriculture, in the Coastal Plain and Piedmont regions of Maryland. Nitrate. The MBSS nitrate data were obtained during the spring and early summer, the same time frame used to develop the average spring nitrate values from the SERC data. ANOVA performed on the data sets indicated that the SERC* and MBSS* average nitrate values were not significantly different at the 95% level of 0 10 20 30 40 50 60 70 MBSS* SERC* ag forest other 166 confidence. Although both the SERC and MBSS data were derived from grab samples and did not capture the transient effect of rain events, both data sets captured the increased transport capacity of spring precipitation (Felton, 1996). Average site values using only spring values from the SERC data set were statistically higher than average annual site values (t-test,  = 0.05). Detailed information on the development of nitrate values is given in Appendix 5. These spring baseflow values were not significantly different from the nitrate values derived from MBSS data collected in spring and early summer (t-test,  = 0.05). Therefore, the nitrate values presented in this research can be considered representative of the upper range of baseflow conditions. Average instream nitrate showed a strong relationship to percent agriculture in the site's catchment (Figures 23 and 24). These results were similar to Taraba et al. (1996). Four data sets were used to represent low, medium and high levels of agriculture: MBSS sites in watersheds with less than 35% agriculture in the Coastal Plain and Piedmont regions (?MBSS lo ag?), all MBSS sites in the Coastal Plain and Piedmont regions (MBSS), and the MBSS* and SERC* data sets. The average nitrate value of the MBSS* set with 65% agriculture (3.5 ppm) was significantly higher than the average nitrate value of the original MBSS data set with 43% agriculture (2.35 ppm). The correlation coefficient between instream nitrate and percent agriculture in the site's catchment for the data sets represented in Figure 24 was 0.995. 167 Figure 23. Nitrate (points) vs. percent agriculture (bars) in the site catchment for four data sets with varying degrees of agriculture. The relationship between instream nitrate and agriculture was also investigated by developing correlations using all sites in the data sets. The correlation coefficients for percent agriculture vs. nitrate for the two SERC* watersheds were similar to each other (German Branch R = 0.43, n = 21; Gunpowder R = 0.55, n = 14). Figure 25 shows the relationship between instream nitrate and percent agriculture in the site catchment for the MBSS* data set. The correlation coefficient (R = 0.55, n = 279) was similar in strength to those seen in the SERC* data set. As a comparison, Taraba (1996) found that percent row crops explained 79% of variation in average instream nitrate concentration. However, this relationship was confined to karst watersheds with well and moderately drained soils, and used only 15 sites to develop the regression. 2.35 3.50 3.39 1.19 0 10 20 30 40 50 60 70 MBSS MBSS* SERC MBSS lo ag % a gr ic u ltu re in th e sit e w a te rs he d 0 1 2 3 4 a v er a ge in st re a m sp ri n g n itr a te (p pm ) 168 Figure 24. Instream nitrate vs. percent agriculture in the site contributing watershed, based on MBSS* data. * y = 0.0214x + 25.503, R2 = 0.1759, n = 161 Visual examination of the land use and nitrate values on the watershed maps indicated elevated instream nitrate levels, or ?hot spots?, near CAFOs. Eight of the thirty-five sites had a CAFO in their catchments. These were often associated with German Branch, the Coastal Plain watershed, had fewer and smaller CAFOs than the Gunpowder watershed (Table 21). Correlation analysis verified the ?hot spot? effect. The correlation of CAFO area in the site catchment vs. nitrate was fairly strong for all sites (R = 0.34). When the correlation was conducted on the individual watersheds, the German Branch watershed showed a much stronger response (R = 0.51) than the Gunpowder watershed (0.12). 0 5 10 15 20 35 45 55 65 75 85 95 percent agriculture in the site catchment in s tr e a m n itr a te (p pm ) 169 Table 21. Confined Animal Feeding Operation (CAFO) and instream baseflow nitrate information for German Branch and Gunpowder watersheds, as determined from SERC* data. # of CAFOs CAFO area (ha) Ave. NO3 (ppm) German Branch 3 3-6 0.58 Gunpowder 5 8-17 3.57 Although these data were insufficient to develop a strong analysis, this information does serve to illustrate the potential that point sources have to modify NPS contributions. Another potential point source for agricultural nitrate is the tile drain system common in Coastal Plain areas with high water tables. These under-field drainage systems collect drainage water that may be discharged directly to the stream, bypassing the riparian zone and modifying NPS values. Instream nitrate vs. location. The data were partitioned by physiographic and geographic region to investigate instream nitrate as a function of location. Figure 25 presents the results of ANOVA. When sites were partitioned only by physiographic region, there was no significant difference in average instream nitrate values. Although the common belief may be otherwise, the Eastern Shore did not have significantly a higher instream nitrate than the overall statewide average value. However, when sites were partitioned by both physiographic and geographic regions, Southern Maryland (central Maryland geographic region, Coastal Plain physiographic region) had a significantly lower nitrate value than the other three areas using an F test ( = 0.05). 170 Figure 25. Instream nitrate for five groups of sites, based on MBSS* data. Instream nitrate vs. buffer presence. Only MBSS* data were used to investigate the relationship between instream nitrate and buffer presence, because there was only one SERC* site without a buffer. ANOVA indicated that buffer presence and type had no significant effect on instream nitrate at that site ( = 0.05). Nor was there a significant difference when the data were partitioned by physiographic region ( = 0.05). These results indicate that buffer presence has no effect on instream nitrate adjacent to that site. Instream nitrate vs. buffer characteristics. Correlation analyses were conducted between average instream nitrate and all available buffer measures in both data sets. For example, Figure 26 presents the correlation coefficients for buffer width vs. nitrate using all buffer types. The correlation values for the two overall data sets were comparable at -0.01 to -0.07. When the data were partitioned by physiographic region, both data sets showed a slightly stronger signal in the Piedmont. Not only 0 1 2 3 4 5 a v e. n itr a te (p pm ) ave. nitrate 3.5 3.82 3.94 3.48 1 statewide E. Shore, Piedmont E. Shore, C Plain central Md, Piedmont southern Md 171 buffer width, but also all buffer characteristics showed weak relationships to instream nitrate at the adjacent site (absolute value of all correlation coefficients less than 0.2). Correlation matrices for SERC* and MBSS* data are given in Appendices 6 and 7. Figure 26. Correlation between instream nitrate and buffer width, for all sites, Coastal Plain and Piedmont sites, from MBSS* and SERC* data. * CP = Coastal Plain, Pd = Piedmont Regional differences. Univariate analysis was performed to investigate differences between physiographic regions. Table 22 presents a summary of statewide and regional statistics for the MBSS* data set. Although the difference in average nitrate values between Piedmont and Coastal Plain sites was not significantly different, variance of nitrate values was lower among Piedmont sites than Coastal Plain sites. This was true of SERC* data, as well. The Coastal Plain had a higher percentage of buffered sites than the Piedmont. Piedmont sites demonstrated a significant difference between nitrate and buffered and unbuffered sites, and showed all CP Pd SERC MBSS*0 0.05 0.1 0.15 0.2 n eg a tiv e co rr el a tio n co ef fic ie n t data set 172 stronger correlation strength between nitrate and width, % agriculture, and channelization than did Coastal Plain sites. These results suggested that models might show regional differences in the effect of buffers, or show greater accuracy if developed with regional data. Therefore, models were developed for both statewide and regional analysis. Table 22. Instream spring baseflow nitrate values and selected agricultural riparian landscape descriptors in the Coastal Plain and Piedmont physiographic regions of Maryland, based on MBSS* data. All sites Coastal Plain Piedmont NO3 avg. (mg/l) 3.5 3.4 3.6 NO3 Std. Dev. (mg/l) 2.28 2.89 1.65 NO3 avg. buffered sites (mg/l) 3.5 3.4 3.5 NO3 avg. unbuffered sites (mg/l) 3.8 3.3 4.0 Avg. buffer width (m) 34.6 34.4 39.0 % Buffered sites 84 93 78 R between %ag: %forest 0.95 0.95 0.94 R between width: NO3 -0.07 -0.05 -0.10 R between NO3: channelization 0.19 0.14 0.29 R between %ag: NO3 0.52 0.48 0.68 Modeling results Although linear regression and regression trees use very different algorithms, this work showed strong agreement between the two model types. The difference in the models was not found in the informational content of the models, but rather in the structure of that information. Summary documents of linear regression and regression tree modeling are given in Appendices 8 - 11. Table 23 presents the terms and R2 values for both multiple linear regression and regression tree models. All models consistently designated agricultural land use in the site catchment as the dominant explanatory variables (Table 23). These results were consistent with 173 past studies, in which agriculture (Allan et al., 1997), and percent crops in particular (Taraba et al., 1996), were shown to exert a strong influence on instream nitrate. The particular influence of percent crops was demonstrated in the MBSS* regression tree model, in which it enters the model twice as an explanatory variable (Figure 27). Table 23. Comparison of model terms and accuracy for regression tree and multiple linear regression (MLR) models predicting instream spring baseflow nitrate for agricultural riparian sites statewide, Coastal Plain (CP) and Piedmont (Pd) physiographic regions of Maryland, based on data from the SERC* and MBSS* data sets. Regression Tree MLR Model terms Channel width: depth < 3.7 CAFO area < 1.5 ha % ag in site catchment < 71% CAFO area % crops in site catchment SERC* All n = 35 R2 0.52 0.17 Model terms % ag in site catchment < 58% stream order < 2.5 SERC* CP n = 21 R2 0.41 Model terms % crop in site catchment < 51%SERC* Pd n = 14 R2 0.52 Model terms % ag in site catchment < 55% % crop in site catchment < 40% % deciduous forest < 19% conductance < 145 % crop in site catchment < 16% % ag in site catchment adjacent crop presence geographic region conductance acid neutralizing capacity MBSS* All ncal = 254 nval = 25 R2 0.48 (Validation R2 = 0.58) 0.41 (Validation R2 = 0.55) Model terms % pasture < 20% % deciduous forest < 20% conductance < 224 % forest % pasture embeddedness MBSS* CP n = 124 R2 0.51 0.65 Model terms % ag in site catchment < 73% % ag in site catchment < 57% % crop in site catchment < 34 MBSS* Pd n = 155 R2 0.56 174 Figure 27. Regression tree model predicting instream nitrate (ppm) from MBSS* data (n = 249). * true values are to the left of the decision, values at the terminal nodes of the decision tree are average instream baseflow nitrate (ppm) for the sites in that grouping. This study was somewhat unique, because CAFO area in the site catchment was an explanatory variable in the SERC* data. Both the SERC* multiple linear regression and regression tree models identified CAFO area as the primary explanatory variable to predict instream nitrate, indicating the particular influence of this type of agriculture. Models also showed physiographic region to have an influence. The SERC* regression tree model used the ratio of channel width:depth to perform the initial ?split? of the data (Figure 28), but examination of the actual grouping of sites revealed 175 that this created an almost perfect split by physiographic region. The MBSS* linear regression model designated physiographic region as the third explanatory variable. These results were consistent with preliminary results presented in Chapter 5, indicating that models developed from regional data may be more powerful. Figure 28. Regression tree model predicting instream nitrate (ppm) for all sites using SERC* data (n= 35). |CWvsCD<3.73462 siteag<71.2148 CAFO<1.5 BAS<77.5 CA<44 4.2660 5.8500 0.5958 1.9850 2.5070 4.5870 * true values are to the left of the decision, values at the terminal nodes of the decision tree are average instream baseflow nitrate (ppm) for the sites in that grouping. 176 Regression tree models developed to predict instream nitrate by physiographic region also consistently identified percent agriculture and percent crops in the site catchment as primary explanatory variables. The influence of CAFO area in the site catchment was again demonstrated by the SERC* Coastal Plain regression tree model. Although percent agriculture in the site catchment and stream order were the explanatory variables used, examination of the actual grouping of sites revealed that this created an almost perfect grouping of CAFO sites. This group had the highest average nitrate value of all groups (4.49 ppm vs. 1.0 and 2.25 ppm). There were no buffer terms in any regional models. To investigate the effect of buffer location within the stream network, a multiple linear regression model was developed for first order sites using MBSS* data (n = 108). Model accuracy was approximately equal for the model using all sites (Adj-R2 = 0.40 for 1st-order sites and 0.42 for all sites), and the terms were essentially the same. There were no buffer terms in this model. These results were consistent with ANOVA, which showed that instream nitrate was not significantly lower at buffered first-order sites. With the exception of the SERC* multiple regression model, the R2 statistic for all models was fairly high (0.41 to 0.56), indicating that approximately half of the variation in nitrate data was explained by the ?easily obtainable? site-level landscape data contained in these data sets. R2 values for the MBSS* calibration models (n = 254) were remarkably similar (0.48 and 0.41), indicating that the regression tree and multiple linear regression algorithms were equally effective for these data. MBSS* 177 model validation R2 values were quite high and also remarkably similar (0.58 and 0.55), indicating that these models captured true signals in the data. CONCLUSIONS There have been numerous studies published on the effects of land use and riparian buffers on water quality. This study is unique, both in the scale and nature of the data and the technical approach used. Most studies have focused on the effect of buffers on reduction of nitrate in groundwater. However, most management decisions are concerned with much larger and more complex systems. This study included agricultural riparian settings with crop and pasture land use, on 1st- through 4th-order streams, in both Coastal Plain and Piedmont physiographic regions, covering almost 2/3 of the state of Maryland. The environmental issue of interest was distribution of instream nitrate concentrations within a watershed. Two data sets were used, both of which represented the same specific population of sites across a large geographic area. The SERC* data were unique because they contained detailed site landscape information and CAFO area in the site catchment. The MBSS* data were unique by containing general site landscape and land use information for over 250 sites. The SERC* and MBSS* data sets both confirmed the primary importance of agricultural land use to predict instream nitrate. The two data sets also provided different and unique insights into the behavior of the system. The SERC* data set provided information on the significance of CAFOs. CAFO presence was a primary explanatory variable for instream nitrate at a site as 178 determined by both linear regression and regression tree algorithms. Thus CAFOs exerted an overwhelming influence on instream nitrate where present, acting essentially as point sources of nitrate to the stream. These results were supported by correlation analysis, in that the correlation coefficient between CAFO presence and instream nitrate was three times that of CAFO presence and percent of crops in the site catchment. Whereas models developed from both data sets identified essentially the same explanatory variables, the SERC* model R2 values were inconsistent (0.17 to 0.65). MBSS* models demonstrated relatively consistent R2 values (0.41- 0.56), and the data set was large enough (n = 279 sites) to enable model validation. Although both data sets were essentially ?snapshot? data, the selection of structural explanatory variables renders these models less sensitive to changes over time. The technical approach to this study was not to find a single "best" model for the prediction of instream nitrate, but to develop multiple models and consider the pattern of primary explanatory descriptor variables and goodness-of-fit. All models demonstrated that land-based sources of nitrate exert a much stronger influence on instream nitrate than riparian buffers. All models indicated that, although buffers may be actively removing nitrate from groundwater at a site, buffer mitigating effects are overwhelmed by agricultural activity. These results are not surprising, in that instream nitrate at a site is an integration of upstream and lateral effects, and this study indicated that adjacent buffer effect was not strong enough to predominate over catchment-level effects. 179 CITED REFERENCES Allan J.D., D.L. Erickson and J. Fay. 1997. The influence of catchment land use on stream integrity across multiple spatial scales. Freshwater Biology 37(1): 149-161. Alliance for the Chesapeake Bay. 1996. Final Report Submitted to the Chesapeake Bay Executive Committee. Baltimore, Md.: Riparian Forest Buffer Panel. Alliance for the Chesapeake Bay. 1996. Riparian Forest Buffers. White paper. Annapolis, Md.: Chesapeake Bay Program. Barker L.S. 2003. Quantification of the Effectiveness of Agricultural Riparian Buffers to Protect Stream Health in Maryland's Coastal Plain and Piedmont Regions. Unpublished PhD Dissertation. College Park, MD: University of Maryland, Biological Resources Engineering Department. Barling R.D. and I.D. Moore. 1994. Role of buffer strips in management of waterway pollution: a review. Environmental Management 18(4): 543-558. Cirmo C.P. and J.J. McDonnell. 1997. Linking the hydrologic and biogeochemical controls of nitrogen transport in near-stream zones of temperate-forested catchments: a review. J. Hydrology 199: 88-120. Correll DL. 1996. Buffer zones and water quality protection: general principles. In Proceedings of the International Conference on Buffer Zones. N. Haycock, ed. Oxford, England. _______. 2000. The current status of our knowledge of riparian buffer water quality functions. In Proceedings of the International Conference on Riparian Ecology and Management in Multi-Use Watersheds, 5-10. P.J. Wigington, Jr. and R.L. Beschta, eds. Middleburg, Va.: AWRA. De'ath G. and K.E. Fabricius. 2000. Classification and regression trees: a powerful yet simple technique for ecological data analysis. Ecology 81(11): 3178-3192. Felton G.K 1996. Agricultural chemicals at the outlet of a shallow carbonate aquifer. Transactions of the ASAE 39(3): 873-887. Gregory S., V. Swanson, J. Frederick and W.A. McKee. 1991. An ecosystem perspective of riparian zones: focus on links between land and water. BioScience 41: 540-551. 180 Herlihy A. and P. Kaufman. 2000. Riparian vs. watershed influence on macroinvertebrates, fish and chemical indicators of stream condition. Poster at the International Conference on Riparian Ecology and Management in Multi-Use Watersheds. Portland, Oregon, 8/27-31, 2000. Hill AR 1996. Nitrate removal in stream riparian zones. Journal of Environmental Quality 25: 743-755. Klauda R., P.F. Kazyak, S.A. Stranko, M.T. Southerland, N.E. Roth and J.C. Chaillou. 1998. Maryland Biological Stream Survey: a state agency program to assess the impact of anthropogenic stresses on stream habitat, quality and biota. Environmental Monitoring and Assessment 51(1-2): 299-316. Lowrance R.R., L.S. Altier, J.D. Newbold, R.R. Schnabel, P.M. Groffman, J.M. Denver, D.L. Correll, J.W. Gilliam, J.L. Robinson, R.B. Brinsfield, K.W. Staver, W. Lucas and A.H. Todd. 1997. Water quality functions of riparian forest buffers in Chesapeake Bay watersheds. Environmental Management 21(5): 687-712. Lowrance R.R., R. Todd, J. Fail Jr., O. Hendrickson Jr., R. Leonard and L. Asmussen. 1984. Riparian forests as nutrient filters in agricultural watersheds. BioScience 34: 374-377. Osborne L.L. and M.J. Wiley. 1988. Empirical relationships between land use/cover and stream water quality in an agricultural watershed. Journal of Environmental Management 26: 9-27. Perry C.D., G. Vellidis, R.R. Lowrance and D.L. Thomas. 1999. Watershed-scale water quality impacts of riparian forest management. Journal of Water Resources Planning and Management - ASCE 125(3): 117-125. Peterjohn W.T. and D.L. Correll. 1984. Nutrient dynamics in an agricultural watershed: observations on the role of a riparian forest. Ecology 65(5): 1466-1475. SAS Institute Inc. 1988. SAS/STAT User?s Guide. Release 6.03 Edition. Cary, N.C.: SAS Institute Inc. Taraba J.L., J.S. Dinger, L.V.A. Sendlein and G.K. Felton. 1996. Land use impacts on water quality in small karst agricultural watersheds. ASAE meeting paper 96-2086. July 14-18, 1996. Phoenix, AZ. Tufford D.L., H.N. McKellar Jr. and J.R. Hussey. 1998. In-stream nonpoint source nutrient prediction with land-use proximity and seasonality. Journal of Environmental Quality 27(1): 100-111. 181 Whigham D.F., C. Chitterling and B. Palmer. 1988. Impacts of freshwater wetlands on water quality ? a landscape perspective. Envir. Management 12(5): 663-671. Wissmar R.C. and R.L. Beschta. 1998. Restoration and management of riparian ecosystems: a catchment perspective. Freshwater Biology 40(3): 571-585. 182 CHAPTER 7 ? USE OF MARYLAND BIOLOGICAL STREAM SURVEY DATA (1995-1997) TO DETERMINE EFFECT OF AGRICULTURAL RIPARIAN BUFFERS ON MEASURES OF BIOLOGICAL STREAM HEALTH ABSTRACT The purpose of this study was to determine the ability of agricultural riparian buffers to mitigate the effects of agricultural pollution on measures of biological stream health, indicated by fish and benthic macroinvertebrate Index of Biotic Integrity (IBI) and Physical Habitat Index. The Maryland Biological Stream Survey (MBSS) 1995-1997 data set was partitioned to represent a population of agricultural sites in Maryland's Coastal Plain and Piedmont regions. ANOVA, multiple linear regression models and regression tree models were developed using riparian and site catchment landscape characteristics. Results indicated that fish, benthic macroinvertebrates, and instream habitat are controlled by environmental drivers at different scales, and that riparian buffers vary in their impact on different measures of biological stream health. Fish IBI at all sites, and benthic IBI and physical habitat index (PHI) in the Piedmont region were unaffected by buffer presence. Because FIBI was not directly affected by buffer presence, use of FIBI to measure success of buffer installation or restoration would give false results. Benthic IBI and PHI in the Coastal 183 Plain were affected by buffer presence, type and width and regression tree modeling was able to delineate a range of minimum effective buffer width between 22-38 meters. While these are empirical results specific to Maryland, the general findings are of use to other locations where the establishment of forest buffers is considered as an aquatic ecosystem restoration measure. INTRODUCTION Background In 1983, concern about worsening water quality in the Chesapeake Bay led to the formation of the Chesapeake Bay Program (CBP), a multi-jurisdictional cooperative effort to restore the nation's largest estuary. The Chesapeake Executive Council set restoration of the Bay's "living resources" as a primary goal in 1987. To achieve that goal, it called for a number of actions to improve water quality, including a 40% reduction in the amount of nitrogen and phosphorus entering the Bay (Alliance for the Chesapeake Bay, 1996a). Efforts to implement these actions focused first on coastal tributaries of the Bay. By 1994, however, attention had expanded to the complete watershed, with the development of a set of goals and actions to increase the focus on riparian stewardship and to enhance efforts to conserve and restore riparian forest buffers (Alliance for the Chesapeake Bay, 1996b). Forest buffers were singled out in part for their nutrient reduction capabilities, but also because they were 184 reminiscent of the natural landscape in which the stream channel and stream ecosystems of the Bay watershed had evolved. In October 1996, the Executive Council of the CBP established a goal of restoring riparian forest buffers on 2010 miles of streams in the Chesapeake Bay watershed by 2010. This action exemplified the widespread acceptance of riparian buffers as an important tool for the reduction of non-point-source pollution. The Conservation Reserve Program (CRP) and its enhancement (the Conservation Reserve Enhancement Program CREP) are additional programmatic indications of the trust in agricultural buffers to mitigate pollution. There are now over 700 publications on the water quality functions of riparian buffers (Correll, 2000) and there is general scientific agreement on the qualitative capacity of riparian buffers to mitigate non- point-source (NPS) pollution (Lowrance et al., 1984, 1997; Correll, 1996; Barling and Moore, 1994). However, attempts to quantitatively relate riparian and watershed information with aquatic conditions have met with mixed success and very few studies have been conducted at the watershed scale. Herlihy and Kaufman (2000) conducted an empirical analysis of Pacific and East Coast North American Water Quality Assessment (NAWQA) data sets, looking at the effect of landscape and habitat variables on fish, macroinvertebrates and instream nitrate. Roth et al. (1998) used the Maryland Department of Natural Resources? (DNR) 1995-1997 Maryland Biological Stream Survey (MBSS) data to investigate the effects of stressors on stream health, but this study was much broader in scope, and did not focus on agricultural sites. 185 Purpose State agencies justify programs, focus resources, allocate funds for projects, and control where money gets spent and work gets done. Decisions are usually made on a state wide or regional basis. State agencies seldom perform in-depth data collection studies to guide or support decisions, but rather, depend on readily available data. Two current regulatory questions concern the use of biocriteria in the development of TMDLs and the effectiveness of riparian buffers to mitigate NPS pollution. The purpose of this study was to determine the ability of agricultural riparian buffers to mitigate the effects of agricultural pollution on stream health, indicated by fish and benthic macroinvertebrate Indices of Biotic Integrity (IBI) and Physical Habitat Index (PHI). The results are of theoretical use to scientists investigating the functional role of buffers in aquatic ecosystem protection, as well as practical use to the agencies that develop Total Maximum Daily Loads (TMDL) and administer the money to establish riparian forest buffers in Maryland. While these are empirical results specific to Maryland, the general findings are of use to other locations throughout the United States where the establishment of forest buffers is considered as an aquatic ecosystem restoration measure. Objectives 1. To determine the effectiveness of agricultural riparian buffers to mitigate the effects of agricultural pollution on measures of biological stream health. 2. To use the MBSS 1995-1997 database to represent a population of agricultural sites. 186 3. To provide scientific guidance for buffer management strategies for maximum ecosystem benefit. 4. To assess applicability of IBIs for local and regional decision-making. METHODS The 1995-1997 MBSS data set was partitioned to create a data set representative of agricultural sites in heavily agricultural watersheds of Maryland's Coastal Plain and Piedmont physiographic regions (Barker, 2003). Statistical analysis techniques were then applied to the data to determine buffer effect on criteria of interest. Empirical models were developed to predict instream index values at a site from landscape characteristics at both watershed and riparian landscape scales. ANOVA, multiple linear regression and Classification and Regression Tree (CART) analysis were used to model the water quality impact of agricultural land use and riparian buffer management. The criterion variables were Fish IBI, Benthic IBI and PHI. Overall comparisons of model structure and terms enabled determination of buffer effect, controlling factors, and the scale of controlling factors. Description of MBSS database All information was obtained from the 1995-1997 MBSS data set. The MBSS is an ongoing effort by the Monitoring and Non-Tidal Assessment Division of Maryland's Department of Natural Resources. The purpose of the MBSS is to assess water quality, physical habitat, and biological conditions in first through third order, non-tidal streams throughout the state of Maryland (Klauda et al., 1998). A rapid 187 bioassessment protocol was followed (Kazyak, 1997) to collect site-level data on the stream channel and adjacent land during the spring and early summer. Although buffer characterization was not the primary purpose of the survey, limited information was taken on the riparian zone (buffer width, vegetation type). The MBSS recorded the width of buffer on the side of the stream with the smallest buffer, with a defined range of values from 0 to 50 meters and land use information for the site catchment was obtained from GIS land cover data sets (Mercurio et al., 1999). Several stream health indicators were developed and included in the MBSS data. This study used three indicators of biological stream health as criteria variables; fish and benthic macroinvertebrate Indices of Biotic Integrity and Physical Habitat Index (PHI). The fish IBI, benthic IBI and PHI were calibrated separately for low-gradient Coastal Plain streams and for higher gradient non-Coastal-Plain streams. An MBSS data collection site was represented by approximately 160 variables. Numerous location variables and time variables identified exactly where and when the samples were collected. Not only were the immediate sites described but some adjacent land use data were also recorded. Those variables that were used in developing models in this paper are described briefly. ANC: Acid neutralizing capacity was given in ?-eq/L. Area: the site catchment area, reported in acres. Average thalweg depth: the deepest portion of the lateral transect of the stream, was measured in centimeters at the 0, 25, 50, and 75 meter points of the sample segment. Average thalweg depth was the average of these measures (cm). 188 Blackwater conditions: blackwater was coding indicating whether the site was located in a blackwater stream. A value of ?1? indicated that the site was blackwater, a value of ?0? indicated that it was not. Conductance: reported in micro-mho/cm. Embeddedness: embeddedness was a given as the percentage that gravel, cobble, and boulder particles were surrounded by sediment or flocculent material. Flow, or streamflow, was calculated from raw data collected at each stream segment and reported as cfs. At most sites, a standard transect method was employed. The field crew constructed a velocity/depth profile of the segment using a current meter to measure stream velocity and recording stream depth at 5 to 20 regular intervals across the stream. At each location along the transect, velocity was measured at a point 0.6 of the distance from the water surface to the bottom. Calculation of discharge from raw velocity, depth, and lateral location data followed standard procedures. High intensity urban land use: a land use classification describing heavily built up urban centers with very little vegetation and high population densities. Instream habitat and bank stability: Instream habitat structure and bank stability were assigned qualitatively, based on visual observations within the 75-m stream ample segment. Scores ranged from 0 to 20 in four categories from poor to optimal. Instream habitat was based on 189 the percentage of stable habitat. Bank stability was based on the percent of banks with erosional scars or erosion potential. Maximum depth: maximum depth within the 75-meter segment (cm). Percent land use: Percent land use in the site catchment were based on the 1996 Multi-Resolution Land Use Classification (MRLC) land cover database for EPA Region III (MRLC, 1996a; MRLC, 1996b). pH: given in standard pH units. Riparian width: the width of the vegetated riparian buffer was estimated (m), to a maximum of 50 m. If the buffer was greater than or equal to 50 m, a value of 50 was entered. This measure was the width of the riparian buffer on the side of the stream with the smallest buffer. SO4: Sulfate, nitrate nitrogen, and dissolved organic carbon concentrations were given as mg/L. Stream order: represents the Strahler (1957) convention used for ranking stream reaches by order. Stream order determination was based on a stream reach file digitized from 1:250,000 scale topographic maps for the MBSS in 1987. Average velocity: thalweg velocity was measured with a flow meter at the deepest portion of the lateral transect at the 0, 25, 50, and 75 meter points of the sample segment and presented in meters per second. Water temperature: given in degrees C. Woody debris: the number of pieces of woody debris at each site was recorded. 190 Database partitioning In order to focus on the effects of agricultural riparian buffers, a new data set was created by parsing and censoring the original MBSS data (Table 24). The MBSS sampled approximately 300 sites (75 meter stream segments) each year during the three-year period 1995-97, producing a data set with 907 sites representative of all basins in the state. For this study, only agricultural sites in primarily agricultural catchments of the Coastal Plain and Piedmont physiographic regions were considered. An agricultural site in a "primarily agricultural" catchment was defined as one whose adjacent land use was some form of agriculture and percent agriculture in that catchment exceeded 35 percent. Riparian buffer vegetation is described in the original MBSS data set using one "forest" buffer descriptor and multiple "grass" buffer descriptors (tall grass, old field, and lawn). These descriptors were collapsed to "forest", "grass" and "other". The resulting reduced data set was designated "MBSS*" to distinguish it from the original data set (Barker, 2003). Table 24. Comparison of the original 1995-1997 Maryland Biological Stream Survey data (MBSS) and modified 1995-1997 Maryland Biological Stream Survey data (MBSS*). Variable MBSS MBSS* Sites 954 279 No. physiographic regions 6 2 No. geographic regions 3 2 Site land use types 16 5 Minimum site %ag 0 35 Average site %ag 44.2 63.7 Average site area (ha) 14,988 18,315 % buffers 66.6 84.2 Average buffer width (m) 24.2 34.6 191 Table 25 summarizes the population characteristics of the MBSS* data set, which represented sites in the Coastal Plain and Piedmont physiographic regions, in Eastern Shore and central geographic regions, with adjacent agricultural land use (crop, pasture, forest, orchard, grass), and with a minimum of 35% agriculture in the site catchment. These restrictions produced a data set of 279 sites representing a population of sites with more buffered sites and wider buffers than the statewide averages. Table 25. Criteria and explanatory variables and values for agricultural sites in Coastal Plain and Piedmont physiographic regions of Maryland, based on MBSS* data. Variable n Mean Std Dev Min Max % barren 279 0.08 0.40 0 4.35 % water 279 0.32 0.48 0 4.64 % high urban land use 279 0.30 0.51 0 3.82 % high urban land use 279 1.94 3.63 0 27.98 % pasture 279 21.84 10.50 1.82 62.07 % probable row crops 279 29.86 10.42 3.51 85.47 % row crops 279 11.96 8.65 0.39 34.43 % coniferous forest 279 3.06 4.61 0 25.41 % deciduous forest 279 21.50 10.48 2.00 55.73 % mixed forest 256 3.71 2.74 0 12.29 % emergent wetlands 203 0.19 0.34 0 2.50 % woody wetlands 123 3.36 6.31 0 29.92 Blackwater conditions* 50 0.04 0.19 0 1.00 * no = 0, yes = 1 192 Table 25. (continued) Variables used in modeling and values for agricultural sites in highly agricultural the Coastal Plain and Piedmont physiographic regions of Maryland. Variable n Mean Std Dev Min Max NO3 (ppm) 278 3.49 2.28 0.26 16.16 BIBI 276 3.07 0.760.87 1.00 4.78 FIBI 260 3.71 0.76 1.00 5.00 PHI 279 60.05 28.21 1.27 98.47 Riparian width (m) 279 31.02 21.28 0 50.00 Water temperature (oC) 279 19.79 3.36 12.00 30.90 % agriculture 279 63.67 13.39 35.09 91.92 % urban land use 279 2.25 4.05 0 31.80 % forest 279 33.33 12.68 7.19 61.53 Stream order 279 1.94 0.80 1.00 3.00 Dissolved oxygen (ppm) 278 8.47 1.92 1.00 14.00 pH 279 7.09 0.48 5.27 8.82 Conductance, field (?-mho/cm) 279 176.10 91.14 54.00 980.00 ANC (?-eq/L) 278 435.30 285.54 -26.90 1722.2 SO4 (ppm) 278 11.28 6.00 1.43 33.91 Adjacent pasture presence 279 0.14 0.35 0 1.00 Instream habitat 279 12.86 4.06 1.00 19.00 Epifaunal substrate 279 11.33 4.64 1.00 19.00 Velocity depth (cm) 279 11.54 4.30 1.00 20.00 Pool quality 279 12.66 3.89 1.00 19.00 Riffle quality 279 11.97 4.88 0 20.00 Channel alteration 279 9.93 4.44 0 18.00 Bank stability 279 10.36 4.14 1.00 19.00 Embeddedness 279 55.32 32.26 0 100.00 Channel flow (m3/s) 279 7.50 1.56 3.48 34.80 Shading 279 66.36 23.21 5.00 98.00 Remoteness 279 11.11 4.63 1.00 19.00 Aesthetics 279 14.15 3.44 1.00 19.00 Woody debris 279 3.73 3.78 0 21.00 Number rootwads 279 1.91 2.39 0 25.00 Max. channel depth (cm) 279 66.45 28.92 9.00 200.00 Stream gradient (%) 279 0.90 1.17 0 10.50 Avg. stream width (m) 279 5.68 4.04 0.05 23.03 Avg. thalweg 279 33.46 16.77 2.00 84.25 Avg. velocity 279 0.19 0.14 0 1.72 Site catchment area (ha) 279 3000 3816 31 29069 % wetlands 279 0.28 0.33 0 2.50 193 Modeling Both multiple linear regression and regression tree models were developed. Each modeling procedure provided different insights into the data. SAS multiple linear regression procedures provide extensive diagnostic information about the relative importance of explanatory variables in a model. Regression tree analysis allows the inclusion of quantitative and qualitative variables, and provides information about threshold effects. Three criteria were used to represent measures of biological stream health (fish IBI, benthic IBI and Physical Habitat Index). All models were expected to take the general form that the criterion variable was a function of source terms (such as agricultural land use) and mitigation factors. All MBSS land use descriptors are developed at the scale of the site catchment. Therefore, source terms included percent agriculture and crops in the site catchment. Examples of mitigation factors were percent forest and wetlands in the site catchment and buffer descriptors. These were limited to buffer width and type. Model verification was conducted by determining the accuracy of predictions for a set of 10 percent of sites withheld from the MBSS* data. Values for fish IBI, benthic IBI and PHI were not assigned for every site. The number of sites with all criteria variable values assigned was only 36. Therefore, models were developed using individual data sets (n = 233, 247, 233). The intent of this analysis was not to produce a single predictive model, but to elucidate patterns of buffer effect and the controlling factors and scales of effect for the criteria of interest. Therefore, ANOVA was also performed to determine effect of 194 buffer presence and type. A significant effect was demonstrated by a difference at the 95% confidence level. Multiple linear regression models. Multiple linear regression models were constructed using landscape measures as explanatory variables. Therefore the multiple linear regression model structure was Y = B0 + B1X1 + B2X2 + ? BnXn + error (7.1) where Y = fish IBI, benthic IBI, or PHI, X1...Xn = site descriptor measurements. Modeling was performed with SAS software. The initial step was an R-square search performed within the PROC REG procedure (SAS, 1988). This procedure constructs multiple linear regression models from all combinations of explanatory variables. The R-square search procedure is more comprehensive than traditional stepwise regression methods for multiple regression model development, because every possible combination of terms is analyzed. Results are presented as a list of model terms and associated adjusted R-square values. This statistic is adjusted for the degrees of freedom of the sums of squares associated with R2 and is calculated as Adj R2 = 1-[(SSE / (n-p))/ (CSS / (n-1))] = 1-[(n-1)/ (n-p)] (1- R2). (7.2) where SSE = sum of squares of error, n = sample size, p = degrees of freedom, CSS = corrected sum of squares, 195 R2 = square of the correlation between response variable and predicted values. This procedure must be used with a data set without missing values to obtain consistent Adj-R2 values. Suggested model terms are then used in a standard regression procedure to determine parameter coefficients, significance and collinearity. Regression tree models. Like multiple linear regression models, regression trees explain variation of a single response variable by one or more explanatory variables. The tree is constructed by repeatedly splitting the data into two mutually exclusive groups, each of which is as homogeneous as possible. The objective is to partition the response into homogenous groups, but also to keep the tree reasonably small. Each group is characterized by the mean value of the response variable, group size, and the values of the explanatory variables that define it (De'ath and Fabricius, 2000). Regression trees do not have non-significant terms. However, the terms are arranged in descending order of explanatory power. Because the model is in the form of a ?decision tree?, these models are more effective than multiple linear regression in elucidating threshold effects. Regression tree models cannot predict a continuous distribution of criteria variable values (as do linear regression models), but group sites into ?terminal nodes? with an average value for the criteria variable. Therefore, regression tree models visually present the range of values for the criteria variable, distributed among different conditions. For example, in Figure 30, fish IBI values ranged from 2.56 in shallow streams with poor instream habitat to 4.05 in high-order, deeper streams with low urban land use. The efficacy of the overall model and its 196 parts is described by Residual Mean Deviance (RMD), which is the sums of squares of the error corrected for the degrees of freedom. It should be noted that R2 is a measure of explained variation and RMD is a measure of unexplained variation, and that although the range of R2 is constrained to values between 0 and 1, RMD values can take values greater than 1. Comparison of linear and regression tree models. The linear models and regression tree models developed from the two data sets were compared between and among one another. The primary explanatory variables identified by the models were compared, as well as their scales of effect. Diagnostics of several models indicated that, although the SAS R-square search provided a selection of linear models with 2 - 20 terms, no more than 5 terms were consistently significant ( of 0.05%). Therefore, the terms and explanatory power of the 5-term SAS multiple linear regression models were compared with the terms and explanatory power of 5-term regression tree models. Developing a common measure of accuracy enabled comparison of the relative accuracy of comparable linear and regression tree models. A coefficient of determination (R2) was created for the regression tree models that could be compared to the R2 of the linear regression models. R2 for the pruned regression tree models was developed as follows: R2 = 1 ? (Residual SSE / Corrected Total SSE) (7.3) where Residual SSE =  (NO3i ? NO3predicted)2 =  (deviance for terminal node)2 Corrected Total SSE =  (NO3i ? NO3mean)2. 197 RESULTS AND DISCUSSION Buffer effect on fish IBI Because fish IBI was stratified by physiographic region, modeling was performed on all sites, Coastal Plain and Piedmont sites. There was strong agreement among all models. Differences between the models were not found in the informational content of the models, but rather in the structure of that information. Summary documents of fish IBI linear regression and regression tree modeling are given in Appendices 12 - 13. Table 26 presents the terms and R2 values for both multiple linear regression and regression tree models. All models identified measures of instream habitat and location in the stream network as dominant explanatory factors. ANOVA analysis demonstrated increase in fish IBI with stream order, as well. These results were consistent with past studies, in which stream order and instream habitat were highly correlated to fish IBI (Roth et al., 1998). No models included any buffer terms or any other land-based terms at the site-level scale. Because instream hydrology and geomorphology are reflective of factors at the fairly broad scale, all models indicated that the site watershed scale is the dominant scale of influence. There were no buffer terms in the models. 198 Table 26. Comparison of model terms and accuracy for regression tree and multiple linear regression models predicting fish IBI for agricultural riparian sites in the Coastal Plain and Piedmont physiographic regions of Maryland, based on data from the MBSS* data set. Regression Tree MLR Model terms maximum depth < 38.5 cm instream habitat < 13.5 acreage < 2235.5 ac average thalweg < 37.5 cm high urban land use < 1.8% instream habitat stream order pH conductance "blackwater" conditions* All sites n = 233 R2 0.51 (Validation R2 = 0.26) 0.46 (Validation R2 = 0.32) Model terms maximum depth < 38.5 cm instream habitat < 8.5 % pasture < 13.1% acreage < 12467 ac epifaunal substrate < 13.5 Coastal Plain n = 116 R2 0.74 Model terms velocity depth < 10.5 cm % row crops < 10.9% ANC < 993 instream habitat < 14.5 maximum depth < 76.5 cm Piedmont n = 117 R2 0.53 * ?blackwater? refers to water with high tannin content, giving it a brown appearance. Regression tree models developed to predict fish IBI were very similar. They identified measures of velocity, depth and structural variation within the channel as primary explanatory variables, and included a land use term. Models developed by physiographic region identified the same hydrologic and geomorphologic factors, but in different order. There were no buffer terms in any regional models. The R2 value for the Coastal Plain model (0.74) showed stronger goodness-of-fit than the general model (0.51) and the Piedmont model (0.53). 199 Land use was different in all models. The regression model using all sites indicated percent high urban land use as a primary explanatory variable. This result was consistent with other published studies of the entire MBSS data set (Roth et al., 1998; Boward et al., 1999). However, they used linear regression to identify urban land use as the most influential land use on fish IBI, and this linear regression model using agricultural sites did not identify any type of land use an a primary explanatory factor. Regression tree models developed form data partitioned into physiographic regions identified agricultural land use as an important influence on fish IBI. Instream habitat, which appeared in all models, is a complex explanatory variable, assigned based on subjective judgment of several instream characteristics. It was highly correlated several other explanatory variables in the models: velocity depth (R = 0.72), epifaunal substrate (R = 0.78), maximum depth (R = 0.41), and average thalweg (R = 0.40). The R2 statistic for all models was fairly high (0.46 to 0.74), indicating that approximately half of the variation in fish IBI was explained by the ?easily obtainable? site-level landscape data contained in these data sets. R2 values for the MBSS* calibration models (n = 233) were similar (0.51 and 0.46), indicating that the regression tree and multiple linear regression algorithms were equally effective for these data. The linear regression model had the lowest R2 value (0.46). Whereas the regression tree models employed stable structural characteristics as explanatory variables, the linear regression models included pH and conductance as explanatory variables. The ?snapshot? nature of these data may have added ?noise? and reduced the explanatory power of the model. Validation R2 values for the regression tree and 200 linear regression models (using all sites) were low but reasonable (0.26 and 0.32), indicating that these models captured true signals in the data. All models indicated strong watershed-level control on fish IBI, mediated through the hydrology and geomorphology of the stream. These results are consistent with fish biology, in that fish are highly dependent on instream features, and can swim away from a specific site (field scale). Overall analysis indicated that riparian buffers have no effect on fish IBI that could be captured by these data. Buffer effect on benthic IBI Because benthic IBI was stratified by physiographic region, modeling was performed on all sites, Coastal Plain and Piedmont sites. Models indicated that there are different scales of controlling influences on benthic IBI between regions. Summary documents of benthic IBI linear regression and regression tree modeling are given in Appendices 14 - 15. Table 27 presents the terms and R2 values for both multiple linear regression and regression tree models. 201 Table 27. Comparison of model terms and accuracy for regression tree and multiple linear regression models predicting benthic macroinvertebrate IBI for agricultural riparian sites in the Coastal Plain and Piedmont physiographic regions of Maryland, based on MBSS* data. Regression Tree MLR Model terms embeddedness < 85 woody debris < 0.5 riparian width < 37.5 m SO4 < 10.25 mg/l bank stability < 3.5 embeddedness woody debris riparian width SO4 mg/l pH All sites n = 247 R2 0.53 (Validation R2 = 0.02) 0.40 (Validation R2 = 0.10) Model terms epifaunal substrate < 6.5 adjacent cover = crop, pasture SO4 < 12.5 ppm riparian width < 37.5 m woody debris < 0.5 Coastal Plain n = 94 R2 0.53 Model terms SO4 < 12.5 ppm % crops < 52.2% avg. velocity < 0.14 m/s conductance < 268 adjacent cover = grass Piedmont n = 153 R2 0.64 Both regression tree and linear models using all sites indicated that forest buffers were influential on benthic IBI (the regression tree model used riparian width and linear regression identified woody debris). ANOVA analysis demonstrated the influence of buffers, as well. Benthic IBI was higher at forested buffer sites (3.3) than other sites (2.8). These results were consistent with the studies of Roth et al. (1998), which were conducted on the entire MBSS data set. For Coastal Plain sites, explanatory variables reflected non-hydrologic instream conditions (epifaunal substrate, woody debris), adjacent landscape influence (crop and pasture presence, 202 adjacent cover, riparian width), and chemistry (SO4). Thus conditions at the site were dominant for benthic IBI in the Coastal Plain. For Piedmont sites, explanatory variables reflected hydrologic instream conditions (average velocity), chemistry (SO4, conductance) and landscape influence (percent crops in the site catchment, adjacent cover). These explanatory variables indicated weak watershed-level control on benthic IBI in the Piedmont, mediated through the hydrology and chemistry of the stream. The R2 statistic for all models were very consistent (0.40 to 0.64), demonstrating that approximately half of the variation in benthic IBI was explained by the ?easily obtainable? site-level landscape data contained in these data sets. R2 values for the MBSS* calibration models (n = 247) were similar (0.53 and 0.40), indicating that the regression tree and multiple linear regression algorithms were equally effective for these data. Validation R2 values for models using all sites were very low (0.02 and 0.10), further verifying that the factors influencing benthic IBI differ by physiographic region. Buffer effect on physical habitat index PHI is a mathematical function of six explanatory variables (Hall et al., 1999). Instream habitat, embeddedness, velocity/depth diversity, and aesthetic rating were used in both the Coastal Plain, and non-Coastal-Plain PHI. Pool quality and maximum depth were used in the Coastal Plain PHI. Riffle/run quality and number of rootwads were used in the non-Coastal-Plain PHI. Several of these variables used to develop the PHI were also used as explanatory variables in fish IBI and benthic IBI models. Due to collinearity effects, PHI models could not be developed using all 203 possible explanatory variables. Therefore, PHI models were developed withholding the independent variables used to develop the PHI. Because PHI was stratified by physiographic region, modeling was performed on all sites, Coastal Plain and Piedmont sites. Because PHI was stratified by physiographic region, modeling was performed on all sites, Coastal Plain and Piedmont sites. Models indicated that there are different scales of controlling influences on PHI between regions. Summary documents of linear regression and regression tree modeling are given in Appendices 16 - 17. Table 28 presents the terms and R2 values for both multiple linear regression and regression tree models. 204 Table 28. Comparison of model terms and accuracy for regression tree and multiple linear regression models predicting PHI for agricultural riparian sites in Coastal Plain and Piedmont physiographic regions of Maryland, based on MBSS* data. Regression Tree MLR Model terms average thalweg < 25.6 cm average velocity < 0.11 m/s flow < 0.20 cfs water temperature <21.65 oC riparian width < 3.5 m average thalweg average velocity acreage % rowcrops in site catchment riparian width All sites n = 243 R2 0.57 (Validation R2= 0.48) 0.45 (Validation R2= 0.52) Model terms avg. thalweg < 26.4 cm stream gradient < 0.95 flow < 0.49 avg. velocity < 0.1 riparian width < 22.5 m Coastal Plain n = 122 R2 0.71 Model terms stream width < 2.2 m avg. thalweg < 10.9 cm stream gradient < 0.22 avg. velocity < 0.13 m/s avg. thalweg < 26.1 cm Piedmont n = 153 R2 0.51 The regression tree and linear models for all sites selected almost identical explanatory variables, indicating hydrology and riparian width were dominant. Both models indicated that forest buffers were influential on PHI. ANOVA demonstrated the influence of buffers, as well (Figure 29). PHI was higher for sites with forested buffers (60) than for sites with grass buffers (53), and PHI was higher for sites with forest as the adjacent cover than sites with adjacent crops (68 vs. 44). 205 Figure 29. ANOVA results for PHI versus physiographic region, stream order, adjacent buffer type, and adjacent agricultural land use (n = 243). Both Coastal Plain and Piedmont models used several hydrologic factors (average thalweg, average surface velocity, stream gradient, flow), emphasizing the dominant control of hydrology on PHI. There were no land use or chemistry variables in either model. The riparian buffer influence seen in both regression tree and linear regression models with all sites was limited to the Coastal Plain, and was the only difference between the regional models. Thus watershed-influenced hydrologic conditions are dominant in both regions, and riparian buffers are also important for PHI in the Coastal Plain. The R2 statistic for all models were very consistent (0.45 to 0.71), demonstrating that slightly more than half of the variation in PHI was explained by the ?easily obtainable? site-level landscape data contained in these data sets. R2 values for the MBSS* calibration models (n = 243) were similar (0.57 and 0.45), indicating that the regression tree and multiple linear regression algorithms were equally effective for 40 45 50 55 60 65 70 PHI Pie dm on t Co as tal Pla in 2-3 or der 1st or der Fo re st b uff er Gra ss buf fer Ad jace nt for es t Ad jace nt cr op 206 these data. Validation R2 values for models using all sites were reasonable (0.48 and 0.52), indicating that these models captured true signals in the data. Comparison of analysis techniques. There was overall excellent agreement among ANOVA, multiple linear regression and regression tree models. Both types of models showed the same overall patterns and scales of effect. ANOVA was able to show the effect of buffer presence and type, whereas multiple linear regression and regression tree models were able to show buffer width effect and indirect forest buffer effects (woody debris, bank stability). The combination of techniques worked in a synergistic manner to provide a good overall picture of buffer effect on the system. CONCLUSIONS The MBSS data were successfully partitioned to represent the population of sites. The data showed four scales of effect - instream, site-level, watershed, and regional. The analyses were able to provide answers to questions about buffer effect. The general agreement among ANOVA, multiple linear regression and regression tree analyses provided confidence that the answers are reflecting genuine behavior of the system. Forest buffers were seen to be important, although secondary controls for benthic IBI and PHI in the Coastal Plain. Because critical forested buffer widths of 22.5 and 37.5 meters were identified by these models, a grass buffer or forested buffer less than 10 meters can be reasonably assumed to be fairly ineffective for protection of stream health. These results are consistent with the general values reported for buffer 207 width effectiveness in the review article on buffer size requirements by Castelle et al. (1994). Hydrology and stream geomorphology were the controlling factors for fish IBI at all sites (and for BIBI and PHI in Piedmont systems). Insofar as buffers can mediate hydrologic effects on flow conditions in a stream, they may indirectly affect fish IBI. Because fish are not found in very small headwater streams, the installation of buffers in areas of strong hydrologic impact under storm conditions in larger 2nd order streams may have more effect on fish IBI, if that is the primary goal. Because FIBI was not directly affected by buffer presence, use of FIBI to measure success of buffer installation or restoration would give false results. These results can provide significant guidance for agencies seeking to restore or plant buffers. Depending on the endpoint, placement and width of buffers can be an effective tool to mitigate the effects of agricultural non-point-source pollution on biological measures of stream health. 208 CITED REFERENCES Alliance for the Chesapeake Bay. 1996. Final Report Submitted to the Chesapeake Bay Executive Committee. Baltimore, Md.: Riparian Forest Buffer Panel. Alliance for the Chesapeake Bay. 1996. Riparian Forest Buffers. White paper. Annapolis, Md.: Chesapeake Bay Program. Barker L.S. 2003. Quantification of the Effectiveness of Agricultural Riparian Buffers to Protect Stream Health in Maryland's Coastal Plain and Piedmont Regions. Unpublished PhD Dissertation. College Park, MD: University of Maryland, Biological Resources Engineering Department. Barling R.D. and I.D. Moore. 1994. Role of buffer strips in management of waterway pollution: a review. Environmental Management 18(4): 543-558. Boward D.M., P.F. Kazyak, S.A. Stranko, M.K. Hurd and T.P. Prochaska. 1999. From the mountains to the sea: the state of Maryland?s freshwater streams. EPA 903-R-99- 023. Annapolis, Maryland: Maryland Department of Natural Resources, Monitoring and Non-Tidal Assessment Division. Correll DL. 1996. Buffer zones and water quality protection: general principles. In Proceedings of the International Conference on Buffer Zones. N. Haycock, ed. Oxford, England. _______. 2000. The current status of our knowledge of riparian buffer water quality functions. In Proceedings of the International Conference on Riparian Ecology and Management in Multi-Use Watersheds, 5-10. P.J. Wigington, Jr. and R.L. Beschta, eds. Middleburg, Va.: AWRA. De'ath G. and K.E. Fabricius. 2000. Classification and regression trees: a powerful yet simple technique for ecological data analysis. Ecology 81(11): 3178-3192. Hall, L.W., R.P. Morgan, E.S. Perry and A. Waltz. 1999. Development of a provisional physical habitat index for Maryland freshwater streams. Publ. No. CBWP-MANTA-EA-99-12. Annapolis, Md.: Maryland Department of Natural Resources, Monitoring and Non-Tidal Assessment Division. Herlihy A. and P. Kaufman. 2000. Riparian vs. watershed influence on macroinvertebrates, fish and chemical indicators of stream condition. Poster at the 209 International Conference on Riparian Ecology and Management in Multi-Use Watersheds. Portland, Oregon, 8/27-31, 2000. Kazyak, P.F. 1997. Maryland Biological Stream Survey. Sampling Manual. Maryland Department of Natural Resources, Chesapeake Bay Research and Monitoring Division, Annapolis, Maryland. Klauda R., P.F. Kazyak, S.A. Stranko, M.T. Southerland, N.E. Roth and J.C. Chaillou. 1998. Maryland Biological Stream Survey: a state agency program to assess the impact of anthropogenic stresses on stream habitat, quality and biota. Environmental Monitoring and Assessment 51(1-2): 299-316. Lowrance R.R., L.S. Altier, J.D. Newbold, R.R. Schnabel, P.M. Groffman, J.M. Denver, D.L. Correll, J.W. Gilliam, J.L. Robinson, R.B. Brinsfield, K.W. Staver, W. Lucas and A.H. Todd. 1997. Water quality functions of riparian forest buffers in Chesapeake Bay watersheds. Environmental Management 21(5): 687-712. Lowrance R.R., R. Todd, J. Fail Jr., O. Hendrickson Jr., R. Leonard and L. Asmussen. 1984. Riparian forests as nutrient filters in agricultural watersheds. BioScience 34: 374-377. Mercurio G., J.C. Chaillou and N.E. Roth. 1999. Guide to using 1995-1997 Maryland Biological Stream Survey Data. Publ. No. CBWP-MANTA-EA-99-5. Annapolis, Md.: Maryland Department of Natural Resources, Monitoring and Non-Tidal Assessment Division. Roth N.E., M.T. Southerland, J.C. Chaillou, H.T. Wilson, D.G. Heimbuch and J.C. Seibel. 1998. Maryland Biological Stream Survey: Ecological Assessment of Non- Tidal Streams Sampled in 1996. Publ. No. CBWP-MANTA-EA-98-1. Annapolis, Md.: Maryland Department of Natural Resources, Monitoring and Non-Tidal Assessment. SAS Institute Inc. 1988. SAS/STAT User?s Guide. Release 6.03 Edition. Cary, N.C.: SAS Institute Inc. 210 CHAPTER 8 ? FINAL CONCLUSIONS The purpose of the research was to quantify the effectiveness of agricultural riparian buffers in protecting stream health in Maryland's Coastal Plain and Piedmont physiographic regions. State and federal agencies charged with development and implementation of riparian buffer management cannot use highly complex, data- intensive models. This research used easily obtainable data to develop scientific guidance for buffer management strategies. Three data sets were used; the 1998 University of Maryland Agricultural Buffer Survey, the 1996 Smithsonian Environmental Research Center (SERC) Water Quality Survey and the 1995-1997 Maryland Biological Stream Survey. Collectively, these data represented baseflow water quality and landscape conditions in small agricultural watersheds characteristic of the Coastal Plain and Piedmont physiographic regions of Maryland. The 1998 University of Maryland Agricultural Buffer Survey was conducted to develop a set of easily obtainable landscape characteristics describing the agricultural riparian landscape. Six key data categories were addressed: land use, buffer characterization, topography, surface and groundwater hydrology and stream channel morphology. All information was successfully obtained through a combination of 211 rapid assessment fieldwork, map work, and publicly available Geographic Information System (GIS) files. From these structural descriptors, a classification system of the riparian landscape was developed. This classification system provided the first method for understanding and categorizing the structure of the agricultural riparian landscapes of for the Coastal Plain and Piedmont physiographic regions of Maryland. Although there are distinct structural differences between the Coastal Plain and Piedmont physiographic regions, most riparian landscapes demonstrated similar structural characteristics. Results indicated that classification of a site can be accomplished with minimal site work (in most cases, a drive-by can provide the on-site information necessary), map work, and soil survey information. Therefore, if a research question focuses on one subset of sites, the classification system can provide assistance identifying sites. Both the University of Maryland Agricultural Buffer Survey and the Maryland Biological Stream Survey (MBSS) data provided extensive, ground-truthed, field- based information. This research successfully parsed and censored these two independent data sets, created for different purposes, to provide information about a common population of sites. The two modified data sets were used to develop the first comprehensive structural characterization of riparian landscapes in agricultural catchments in Maryland's Coastal Plain and Piedmont physiographic regions. Most agricultural sites had buffers. The Coastal Plain was dominated by forested buffers, whereas the Piedmont had a significant portion of grass buffers. Because forested buffers were wider than grass buffers, the average width of Coastal Plain buffers was 212 the same or greater than Piedmont buffers. Comparison of these data demonstrated the effect of censored data. For example, average buffer width was found to be strongly influenced by maximum defined buffer width. Therefore, use of censored data sets should be viewed with caution. There are two major strengths of these results. Ground-truthed information was provided at the site-specific scale. Additionally, this research presented information at the 16-digit watershed scale in order to discuss these buffers in the context of a larger landscape. The site-specific scale of these field-truthed data revealed that a significant portion of buffers was less than 10 meters wide, especially in the Piedmont. With an overall statewide average buffer width of approximately 160 feet, almost 50% of buffers would not be detected by RS data with a resolution of 30 meters. It can be reasonably concluded that surveys depending solely on remotely sensed data have significantly underestimated the degree of buffering in Maryland. Once the nature and distribution of Maryland?s agricultural riparian landscapes were elucidated, modeling was used to investigate buffer effectiveness. The classification system was not used in final modeling work for two reasons. The SERC* data set was too small to be further subdivided by classification for nitrate modeling. Although the MBSS* data set was of adequate size, further field work and soil survey was needed to provide the on-site slopes and water table information necessary to use the classification system with the MBSS* data set. Neither data set was compatible with a paired watershed approach. Therefore, due to limitations in these easily obtainable data, all models predicted some measure of stream health at a site as a function of landscape characteristics. 213 Both multiple linear regression and regression tree modeling were used. SAS regression procedures provided diagnostic information about relative importance of the explanatory variables. Regression tree analysis allowed inclusion of quantitative and qualitative variables, and provided information about threshold effects. The general agreement among ANOVA, multiple linear regression and regression tree analysis provided confidence that the models accurately portrayed the system. All models indicated that nitrate source terms overshadowed any on-site buffer effects. Although no direct buffer effect on instream nitrate was seen in this study, as buffers are installed or widened, land may be taken out of agricultural use, reducing nitrate input to the stream. Whereas this study focused on agricultural non-point- source (NPS) effects, confined animal feeding operations (CAFOs) and pastures were found to act as point sources, overwhelming and confounding NPS effects. Therefore, livestock best management practices (BMPs) are critical for the reduction of nitrate and sediment input to streams. Forest buffers were not seen to be a controlling factor for fish IBI at a site, but hydrology and stream geomorphology were. Therefore, insofar as buffers can mediate hydrologic effects on flow conditions in a stream, they may indirectly affect fish IBI. Because fish are not found in very small headwater streams, the installation of buffers in areas of strong hydrologic impact under storm conditions in larger 2nd order streams may have more effect on fish IBI, if that is the primary goal. This research showed that use of this index to measure success of a buffer installation or restoration project had the potential to give false results. 214 Forest buffers were seen to be secondary controls for benthic IBI and PHI in the Coastal Plain. Regression tree modeling elucidated a minimum effective width between 22 ? 38 meters. Because riparian forest buffers directly affected the habitat for benthic invertebrates, there is an indirect relationship between benthic IBI and PHI. Although these easily-obtainable data could not capture the strength of this indirect relationship, results indicate that installation of buffers could stabilize stream systems and affect the benthic communities. A functional instream ecosystem may be critical for nitrate reduction as it moves from the headwaters to the Chesapeake Bay. Forested riparian buffers serve as a primary source of nutrients to low-order streams, fueling and mediating the energy base of instream ecosystems. Branches, roots and other tree parts that fall into the stream channel provide shelter for instream inhabitants, retain sediment, and contribute to the maintenance of habitat diversity through the development of riffles and pools. Streamside vegetation also contributes to channel stability and shape by holding bank sediment and reducing erosion during periods of high flow. Therefore, riparian zones may act more importantly as source zones to protect habitat and provide food for the instream community that process nitrate, than as buffers that mitigate the input of nitrate to the stream. This research indicated that instream nitrate concentration, fish, benthic macroinvertebrates, and instream habitat were controlled by environmental drivers at different scales, and that riparian buffers varied in their importance. Because buffers act differently on these measures of biological stream health, clear definition of the criteria used to define buffer effectiveness is critical. There was no one recommended 215 "minimum width" of buffer to protect stream health. Depending on the objective, however, buffers can be an effective tool to mitigate the effects of agricultural non- point-source pollution on biological measures of stream health. Specific management suggestions from this research include: Installation of buffers at least 4 m wide on as many low-order stream miles as possible Caution against using instream nitrate or fish IBI to measure effectiveness of site level land use management strategies, such as buffer installation. Caution against using benthic IBI or PHI to measure effects of regional management strategies. While this research provided empirical results specific to Maryland, the general findings are of use to other locations throughout the United States where the establishment of forest buffers is considered as an aquatic ecosystem restoration measure. 216 CHAPTER 9 ? FUTURE WORK Attempts to model comprehensive buffer effects, such as change in nitrate along a reach were unsuccessful. Ground-truthed data were too limited to develop trustworthy models of integrative buffer effects. This research determined that resolution of remotely sensed data used in publicly available GIS files was insufficient to adequately characterize distribution and width of agricultural buffers. Should adequate, higher resolution data become available, the methods developed in this research might be used to model integrative buffer effects. Although all models indicated that buffers were secondary or ineffective in controlling measures of stream health, there are clearly more complex relationships at work than could be captured by these easily obtainable data. Future research should focus on elucidating the nature and strength of the relationship between riparian forests and instream biological communities. A better understanding of this relationship may well provide the key for effective land use decisions to maximize the effectiveness of the instream biological processing of unavoidable nitrate inputs from the headwaters of the Chesapeake Bay. The weakness of regression models developed in this research may also be due to missing key explanatory variables. These data did not contain some measures that 217 would be expected to explain significant portions of the variation of buffer effectiveness. There were no measures of agricultural BMP implementation, such as tillage and nutrient management. There were no good measures of buffer bypass, such as fragmentation and agricultural drain tiles. This research used easily obtainable measures such as gradient, tortuosity, drainage density, water table depth, and drainage to indirectly describe the transport efficiency of the site catchment. A better description (index) of transport efficiency could provide a way to explain more of the variation in buffer effectiveness, especially for measures that are strongly influenced by hydrology, such as nitrate and fish IBI. Should better measures of these landscape and land use descriptors become available, the methods developed in this research could be used to create better models of buffer effect. There were inherent limits to the usefulness of easily obtainable data as the criteria variables. Use of the indices developed by the MBSS as criteria variables injected variation into the modeling relationships, because they were themselves blended measures. If the species are partitioned into more specific sub-groupings, according to chemical or hydrological sensitivity, buffer effect on specific endpoints might be better elucidated. Use of instream nitrate as a criteria variable is also, in a sense, a blended measure. Stream water is a mix of surface runoff and groundwater of different ages that have arrived in the stream by various pathways. The mix of ages in streamwater may affect these relationships, for if the majority of the water in the stream is old, having traveled through deep pathways, there may be negligible riparian effects. Further work considering the age of the streamwater or the dominance of a 218 particular flow path may provide better guidance for placement of buffers for maximum effectiveness. This research focused on nitrate as the measure of stream chemistry, due to interest in reduction of agricultural nitrate to the Chesapeake Bay. However, phosphate has become a topic of regulatory concern, as well. These data sets contain information on phosphate and can be analyzed using the same techniques to provide information on the effectiveness of agricultural buffers to control instream phosphate. This research focused on agricultural riparian buffers. Urban buffers are quickly becoming a topic critical interest in land use planning. The techniques developed here could be easily employed to investigate the effects of stormwater BMPs, zoning and land use management. The data sets developed through this work can be further explored. Although the Survey* data were not particularly useful to model buffer effectiveness, they provide a unique and comprehensive description of Maryland?s agricultural landscapes. These data have not been fully investigated to ascertain all that can be said about these landscapes. Similarly, the MBSS* data set provides a different, but unique and comprehensive description of Maryland?s agricultural riparian landscapes. Neither of these data sets has not been fully investigated. Finally, the data and results of this work can be integrated with the socioeconomic information developed from the University of Maryland project entitled "Environmental Benefits and Costs of a Voluntary Riparian Forest Buffer Program", funded by USDA?s Fund for Rural America in 1998. The project had a number of research objectives, including environmental and sociological 219 characterization of agricultural riparian settings in Maryland. From these data, future work can match the physiographic information about riparian buffer sites developed in this research with information about agronomic land use and socio-economic information about site owner. Results should provide proximate level and type of environmental benefits one might reasonably expect from a voluntary incentive-based program to establish riparian buffers in the Maryland portion of the Chesapeake Bay watershed. 220 APPENDICES 221 APPENDIX 1: UNIVERSITY OF MARYLAND AGRICULTURAL BUFFER SURVEY FIELD SHEET Riparian Buffer Project Field Data Site number: Date: Time: Team type: Team number: Names: Watershed: Water station #: Rain past 24 hours: Buffer Continuous length: Site length: Number of sides: 1 or 2 Type: T / T-G / G / S Trees - H / HP / PH / P Shrubs/u'growth ht.: Grass - fallow / planted N fixers? Y / N Gullying - L / M / S Berms? N / L / R / L&R Transect Buffer width slope DBH bas.area L R L R L R L R T1 T2 T3 222 Riparian Buffer Project Field Data Stream Sketch / Comments Channelization -Y/ N Velocity: (depth): Channel shape: TRI RECT ROUND Substrate: MS / S / SG / SGC / G / C Adjacent Land (facing upstream) Right side land use: Field slope: Apparent slope length: Left side land use: Field slope: Apparent slope length: Tiling? Y / N Stream Top width Incision RB Angle/ LB Angle/ Water to depth erosion erosion depth R P R P R P 223 Riparian Buffer Project Map Information Site number: Nearest water station: Team type: Team number: Names: Watershed: County: Physiographic region: Latitude: Longitude: Stream order: Stream gradient (%): Stream distance upstream of buffer (ft): Stream drainage area (sq mi): Upstream water station: Downstream water station: Distance between water stations (ft): % of distance represented by buffer: 224 Riparian Buffer Project Map Information Soil info LB soil type(s): LB Wtr Tbl (ft): LB Drainage: Poor / Good / Well / Excessive LB Surface "Permeability" (ft): LB Min "Permeability" (ft): depth range at min: RB soil type(s): RB Wtr Tbl (ft): RB Drainage: Poor / Good / Well / Excessive RB Surface "Permeability" (ft): RB Min "Permeability" (ft): depth range at min: LF soil type(s): LF Wtr Tbl (ft): LF Drainage: Poor / Good / Well / Excessive LF Surface "Permeability" (ft): LF Min "Permeability" (ft): depth range at min: RF soil type(s): RF Wtr Tbl (ft): RF Drainage: Poor / Good / Well / Excessive RF Surface "Permeability" (ft): RF Min "Permeability" (ft): depth range at min: 225 APPENDIX 2: UNIVERSITY OF MARYLAND AGRICULTURAL BUFFER SURVEY GUIDE TO FIELD AND MAP MEASUREMENTS The purpose for taking the following measurements is to quantify the characteristics of an agricultural riparian buffer zone. Field Sheet ID Info: Site number: in the order that you visit them Date: date measurements were taken Time: time of day (military time) Team Type: Linda or Jim Team Number: your team number Names: of team members Watershed: Watershed name (German Branch, Cattail Creek) Water station #: corresponding water quality data site for this buffer (Linda?s team will fill this in.) Rain: approximate inches in the past 24 hours Field Measurements: Stream: channelization velocity depth at velocity measurement channel shape channel bed substrate width incision to stream bed bank angles bank erosion water depth (in riffles and pools) Buffer: continuous length of buffer site length number of sides type tree types height of undergrowth type of grass presence of nitrogen fixers gullying berms 226 width slope DBH (diameter of trees at breast height) basal area Adjacent fields: land use slope and slope length presence of drain tiles Instruments and gear: Measuring equipment: 100-foot tape measure, (stiff) 8-foot measuring tape, basal area prism, DBH (diameter at breast height) stick, Abbey level, fishing float attached to a 20-foot string, fishing pole with weight pencils, covered clipboard, lanyards, map Personal equipment: rubber boots, extra socks poison ivy block, soap-water-cloth, poison ivy gel, antibiotic salve sunscreen, hat tick spray light colored pants and shirt light-weight field jacket or long-sleeved shirt rain gear and towel lots of liquids tissue and bag cell phone, emergency telephone numbers University ID, copy of project introductory letter 1st aid kit, small pocket knife 227 Description of measurements: Stream: Stream overall measurements: Channelization. Channelization has occurred when the path of the stream has been straightened. Berms (levees, or mounds of dirt parallel to the stream) can be evidence of channelization. (Note: DO NOT use the angle formed by the berm as the slope angle.) Record as: Yes / No Velocity (depth). The velocity of the stream can be measured by the use of a second hand watch and a device made of a bobber and twenty feet of line. By tossing the bobber upstream, the velocity can be calculated by timing how long the bobber takes to travel the distance of the line and dividing the length by the time. Record as units of x feet per y seconds, and calculate feet per second later. Try to measure in the center of the stream at a riffle, and take note of the depth of the water where velocity was measured. Record as: x feet per y seconds Channel shape. The shape of the channel or stream between banks. Record as: rectangular, round, or triangular Stream channel bed substrate. Look in both riffles and pools. Record as: M (mud/silt), S (sand), SG (sand/gravel), G (gravel - up to 1.5 inches diameter), C (cobbles - up to 4 inches in diameter), SGC. ROSGEN score. Only applicable to Linda?s teams. This is a method of categorizing streams. A Rosgen score is a letter/number combination. The letter references the channel shape, sinuosity, incision and apparent instability (erosion of banks). The number references the substrate material. Agricultural streams OFTEN do not fit neatly into the Rosgen system, yet it is often used. Part of the work of this project is to consider the usefulness of the Rosgen system here in Maryland. Therefore, put the stream in the closest Rosgen category, even though it may not be a good fit. Channel form: For low gradient streams (<2%): C = point bars present, entrenched, rectangular shape, low sinuosity, fairly stable. F = no point bars, entrenched, rectangular shape, low sinuosity, unstable (erosion) E = not very entrenched, triangular shape, highly sinuous, vegetated banks For moderate gradient streams (2?4%) B = moderately entrenched, triangular/round shape, stable banks, low sinuosity 228 Stream transect measurements: Top width. The top width is the measured distance from one bank to the opposite bank. The stream bank is determined as the ?slope break? between the buffer and the stream. Use a measuring tape or cast a weight across the stream to determine the distance and record its measurement in feet. Record as: Top width is measured to the nearest half foot. Incision to depth. The incision to depth is the distance from the bottom of the stream to the level of the banks. This measurement can be calculated by standing in the stream and marking the level at which the top width was taken and then measuring from this height to the bottom of the stream. Record as: Incision is measured to the nearest half foot. Bank angles and bank erosion. The bank angles and the erosion are taken on each side. All measures are referenced to facing upstream. The angles can be either calculated by taking the rise and run of the banks to the nearest half foot or by approximation or by using a protractor. After obtaining the rise and run the bank angle can be calculated on a calculator as tan-1 of rise/run. Record as: rise/run or as degrees. The banks? erosion are ranked using the following categories; none, low (these will probably be vegetated), high (lots of slumped dirt), and moderate (in between ?minimum? and ?high?). Record as: N, L, M, H Water depth. Average depth from the surface of the water to the bottom of the stream is taken in both a riffle area where the water is shallower and in a pool area where the water is still. These depths can be taken using a small measuring tape. Record as: to the nearest inch. Buffer: Buffer overall measurements: Continuous length. Determine how long the buffer appears to run ? either from the aerial photo or from visual observation. Record as: feet Site length. Measure the length of the site at which you take measurements by adding the distance between transects. Use the average value paced by two people. Record as: feet Number of sides. Note if the buffer is present on both sides of the stream or just one. Record as: 1 or 2 Type. There are basically 4 types of buffers ? trees, trees and grass, grass, and shrub/undergrowth. Record as: T, T-G, G, S 229 Trees. For forested buffers, record the type of trees. Record as: H (hardwood), HP (hardwood and pine mix, predominantly hardwoods), PH (hardwood and pine mix, predominantly pine), P (pines). Grass. For grass buffers, record whether the buffer is planted grass or wild. Record as: fallow or planted Shrub buffers. Record the height of the undergrowth. Record as: feet. Presence of nitrogen fixers. Some plants, such as black locust trees, do not use nitrate from the soil water, but ?fix? nitrogen from the air by means of symbiotic bacteria. A large proportion of these plants will affect the buffer?s uptake of nitrate. A ?yes? means there is a significant proportion of nitrogen fixers in the buffer. Record as: Yes / No. Gullying. Small gullies or side ditches will sometimes run through the buffer and act as an effective by-pass through the buffer. These gullies can be filled with water or dry and full of vegetation. Record as: Low, Moderate, Severe. Berms. Berms or levees are sometimes found along the one or both banks of the stream, especially if there has been channelization. Record their presence as you face upstream. Record as: N (none), L (left), R (right), LR, left and right). Buffer transect measurements: Buffer width. Width of the buffer (between the stream and the adjacent farm field) is measured by pacing perpendicular to the stream. It is important to occasionally re- measure your average stride with the 100-foot tape or compare the team members? paced-off values to ensure accuracy. (If the stream is very sinuous, guess an average stream direction and measure perpendicularly to the average stream path.) The buffer width is measured for each individual type of buffer: - trees, grass, and shrubs. Combined these values give the total buffer width. The streamside end of the buffer is at the slope break between the stream bank and the buffer, and is the demarcation between buffer and stream bank. You need to pay attention here ? sometimes the demarcation between buffer and bank is not obvious. The field-edge end of the buffer is at the edge of cultivation. All measurements are referenced facing upstream. If there is no buffer on one side, record ?none?. If there is a buffer but you do not have access to it, record "no access", estimate the appropriate measurements and note them in the transect comments section. Record as: feet to the nearest ? foot. 230 Buffer slope. Slope is measured with the abbey level. Align the level line in the sight with a point at the same height as the observer?s eye. Move the level to center the bubble on the level line. The slope is read from the % scale. Take the average, representative value of buffer slope. Record as: %. DBH. The diameter of the dominant trees in the buffer is measured at breast height with the DBH stick. Note the values in the margin of the field sheet, and record the average. Record as: inches. Basal area. Basal area is a measurement of square footage of trees in an area. Use a basal area prism to determine the square footage by holding the instrument at arm's length from the observer and at chest height. Pivot about a fixed point while looking though the prism and count the amount of trees that do not become ?disconnected? through the prism. The number of trees is multiplied by ten to give the square footage. Record as: square feet. Adjacent fields: (All are overall measurements) Adjacent land use. Record the land use on each side of the buffer. Record as: crop, pasture, livestock, houses, and forest. (Call us if you have anything else.) Slope and Slope length. Use the same procedure used to determine the buffer slope. Record as: %. Also record the distance to any apparent slope change ? either pace off the distance or visually estimate it. Record as: yards. Presence of drain tiles. Tiling is a rerouting of water from the fields to the stream through the use underground pipes (tiles) under the root zone. This method of farming can sometimes be detected by looking for a small vent structure in the field - this tower allows a constant atmospheric pressure to be maintained in the system. Record as: Yes / No. 231 Map Measurements: General information: Site number Team Type Team Number Names Watershed County Physiographic region Latitude/Longitude or Maryland Grid XY coordinates Stream order Stream gradient Upstream distance Stream site drainage area The following measurements will be filled in by Linda?s team: Nearest water station Upstream water station Downstream water station Distance between water stations % of distance represented by buffer Soil information: Left and Right buffer - soil type water table drainage surface hydraulic conductivity ?permeability? minimum hydraulic conductivity ?permeability? depth at minimum hydraulic conductivity Left and Right adjacent field - soil type water table drainage surface hydraulic conductivity ?permeability? minimum hydraulic conductivity ?permeability? depth at minimum hydraulic conductivity 232 Instruments and gear: County street map topographic map County soil map aerial photos of watershed ruler pencil thin string grid chart Description of measurements: General information: Latitude. The latitude is the vertical measure of a fixed position; this coordinate is expressed in degrees, minutes, and seconds Longitude. Longitude is the horizontal measure of a fixed position. OR Maryland grid: x coordinate. The Maryland grid x coordinate is the vertical position as is latitude on a map, but this measurement is in feet. The division of this is an indicator every one hundred thousand feet. Maryland grid: y coordinate. The Maryland grid y coordinate is the horizontal position. Stream order. Stream order as determined by the Strahler system will be between 1 - 4. A 1st order stream has no tributaries. Two 1st-order streams run together to make a 2nd-order stream. A 2nd-order stream does not increase order if joined by a 1st-order stream. Record as: 1,2,3,4. Stream gradient. The stream gradient is a calculation of the slope of the stream. This gradient can be calculated by using a topographic map to determine the drop in elevation per length of stream for a length of stream around the buffer. The drop is then divided by the length to give the stream gradient. Record as: %. Upstream distance. The total distance of stream that is traversed to get to the data sampling site can be calculated using a thin line and the map scale. Arrange the line on the map so that it traces the path of the stream from its source to the site of the buffer. Next take the length of measured line and place it so that the distance can be calculated using the legend on the map. Do not use the lengths of side tributaries above the buffer site, only the main path of the stream from the buffer back to the stream source. Record as: feet. 233 Stream drainage area. The stream drainage area is the approximate area of the surrounding lands that drain into the stream at the point of interest. This measurement is determined from the USGS topographic map of the watershed, and is calculated by using a ruler to measure the length and width of the land that drains into the stream at the buffer, then multiplying the two figures to give area. Record as: square miles. Soil information: All values are recorded for left/right buffer and left/right field. All values and the soil survey map are located in the county soil survey book. Soil type. Soil type is determined from the soil survey map. After mapping the location of the field site, obtain the soil type code from the map. The name of the soil types are arranged alphabetically in the tables in the soil survey book. Record as: name of soil type. Water table. The depth to the water table is located in two possible places. There will either be a column for ?depth to seasonally high water table? in the ?engineering properties of the soils? table, or the information will be given in a description of the soil type. Record as: a range of feet. Drainage. Drainage for the soil type is designated as poor, good, well or excessive. This information is located in two possible places, either as a designated column in the ?engineering properties of the soils? table, or in the description of the soil type. You may have to interpret the language to designate the drainage as one of the categories. Record as: poor, good, well or excessive. Surface permeability. This is actually a measure of hydraulic conductivity. The soil table will have a column to the right of the soil description titled ?depth to surface?. The top range will always be (0 ? some value). Further to the right, there is a column titled ?permeability?. Take the value corresponding to this top depth range for surface permeability. Record as: a range of inches. Minimum permeability. This is also a measure of hydraulic conductivity. In the column titled ?permeability?, there may be more than one range of permeabilities at different depths. Use the minimum value. Record as: a range of inches. Depth at minimum. This is the range in the column to the right of the soil description titled ?depth to surface? that corresponds to the minimum value for permeability. Record as: a range of feet. 234 Description of Survey variables Name Description Units ACCESS Did livestock have access to the stream? Y/N AREA Site watershed area sq.mi. B Was there a buffer at all? Y/N BAS1O Basal area: overall measurement for both stream sides: 1st sq.ft. BAS2O Basal area: overall measurement for both stream sides: 2nd sq.ft. BAS3O Basal area: overall measurement for both stream sides: 3rd sq.ft. BAS1L Basal area: L stream side: 1st measuremt sq.ft. BAS2L Basal area: L stream side: 2nd measuremt sq.ft. BAS3L Basal area: L stream side: 3rd measuremt sq.ft. BAS1R Basal area: R stream side: 1st measuremt sq.ft. BAS2R Basal area: R stream side: 2nd measuremt sq.ft. BAS3R Basal area: R stream side: 3rd measuremt sq.ft. BDL Buffer soil type: L side BDR Buffer soil type: R side BDRL Buffer drainage class: L side BDRR Buffer drainage class: R side BERM Were there berms at the site? BETW Distance between SERC sampling stations ft BG1L Buffer grass width: L side: 1st measuremt ft BG2L Buffer grass width: L side: 2nd measuremt ft BG3L Buffer grass width: L side: 3rd measuremt ft BG1R Buffer grass width: R side: 1st measuremt ft BG2R Buffer grass width: R side: 2nd measuremt ft BG3R Buffer grass width: R side: 3rd measuremt ft BGRASS Buffer grass type 235 Name Description Units BIN1L Buffer in-channel vegetation width: L side: 1st measuremt ft BIN2L Buffer in-channel vegetation width: L side: 2nd measuremt ft BIN3L Buffer in-channel vegetation width: L side: 3rd measuremt ft BIN1R Buffer in-channel vegetation width: R side: 1st measuremt ft BIN2R Buffer in-channel vegetation width: R side: 2nd measuremt ft BIN3R Buffer in-channel vegetation width: R side: 3rd measuremt ft BKMDLM Buffer minimum K depth: L side: min value of range ft BKMDRM Buffer minimum K depth: R side: min value of range ft BKMLM Buffer minimum K : L side: min value of range in/hr BKMRM Buffer minimum K: R side: min value of range in/hr BKSLM Buffer surface K : L side: min value of range in/hr BKSRM Buffer surface K : R side: min value of range in/hr BS1L Buffer slope: L side: 1st measuremt % BS2L Buffer slope: L side: 2nd measuremt % BS3L Buffer slope: L side: 3rd measuremt % BS1R Buffer slope: R side: 1st measuremt % BS2R Buffer slope: R side: 2nd measuremt % BS3R Buffer slope: R side: 3rd measuremt % BSH1L Buffer shrub width: L side: 1st measuremt ft BSH2L Buffer shrub width: L side: 2nd measuremt ft BSH3L Buffer shrub width: L side: 3rd measuremt ft BSH1R Buffer shrub width: R side: 1st measuremt ft BSH2R Buffer shrub width: R side: 2nd measuremt ft BSH3R Buffer shrub width: R side: 3rd measuremt ft BT1L Buffer tree width: L side: 1st measurement ft BT2L Buffer tree width: L side: 2nd measurement ft BT3L Buffer tree width: L side: 3rd measurement ft BT1R Buffer tree width: R side: 1st measurement ft BT2R Buffer tree width: R side: 2nd measurement ft BT3R Buffer tree width: R side: 3rd measurement ft BWLM Buffer water table: L side: minimum value of range ft BWLX Buffer water table: L side: maximum value of range ft BWRM Buffer water table: R side: minimum value of range ft BWRX Buffer water table: R side: maximum value of range ft 236 Name Description Units CA1L Channel bank angle: L side: 1st measurement degrees CA2L Channel bank angle: L side: 2nd measurement degrees CA3L Channel bank angle: L side: 3rd measurement degrees CA1R Channel bank angle: R side: 1st measurement degrees CA2R Channel bank angle: R side: 2nd measurement degrees CA3R Channel bank angle: R side: 3rd measurement degrees CD1 Channel depth: 1st measurement ft CD2 Channel depth: 2nd measurement ft CD3 Channel depth: 3rd measurement ft CE1L Channel bank erosion: L side: 1st measurement CE3L Channel bank erosion: L side: 3rd measurement CE1R Channel bank erosion: R side: 1st measurement CE2R Channel bank erosion: R side: 2nd measurement CE3R Channel bank erosion: R side: 3rd measurement CHNZ Is the stream channelized? Y/N CO County CP1 Pool depth: 1st measurement in CP2 Pool depth: 2nd measurement in CP3 Pool depth: 3rd measurement in CR1 Riffle depth: 1st measurement in CR2 Riffle depth: 2nd measurement in CR3 Riffle depth: 3rd measurement in CW1 Channel width: 1st measurement ft CW2 Channel width: 2nd measurement ft CW3 Channel width: 3rd measurement ft DATE Date field work was conducted in DBH1O Overall Diameter at Breast Height: both stream sides: 1st in DBH2O Overall Diameter at Breast Height: both stream sides: 2nd in DBH3O Overall Diameter at Breast Height: both stream sides: 3rd in DBH1L Diameter at Breast Height: L side: 1st measurement in DBH2L Diameter at Breast Height: L side: 2nd measurement in DBH3L Diameter at Breast Height: L side: 3rd measurement in DBH1R Diameter at Breast Height: R side: 1st measurement in DBH2R Diameter at Breast Height: R side: 2nd measurement in DBH3R Diameter at Breast Height: R side: 3rd measurement in 237 Name Description Units DIST Distance from this site to the headwater ft FDL Field soil class: L side FDR Field soil class: R side FDRL Field drainage class: L side FDRR Field drainage class: R side FKMDLM Field minimum K depth: L side: min value of range in/hr FKMDLX Field minimum K depth: L side: max value of range in/hr FKMDRM Field minimum K depth: R side: min value of range in/hr FKMDRX Field minimum K depth: R side: max value of range in/hr FKMLM Field minimum K : L side: min value of range in/hr FKMLX Field minimum K : L side: max value of range in/hr FKMRM Field minimum K : R side: min value of range in/hr FKMRX Field minimum K : R side: max value of range in/hr FKSLM Field surface K : L side: min value of range in/hr FKSLX Field surface K : L side: max value of range in/hr FKSRM Field surface K : R side: min value of range in/hr FKSRX Field surface K : R side: max value of range in/hr FSL Field slope: L side % FSR Field slope: R side % FSL Length of field or length used to calculate slope: L side yds FSR Length of field or length used to calculate slope: R side yds FUL Field use: L side FUR Field use: R side FWLM Field water table: L side: minimum value of range ft FWLX Field water table: L side: maximum value of range ft FWRM Field water table: R side: minimum value of range ft FWRX Field water table: R side: maximum value of range ft GRA Stream gradient % GULLY What degree of gullying is there through th ebuffer? INOUT Is the buffer in-channel or on the land? LAT Latitude of site LBUF Length of buffer fragment ft LONG Longitude of site 238 Name Description Units LSITE Length of the site ft N/A Team member names that conducted the site visit. NFIX Is the buffer vegetation predominated by N-fixers? Y/N ORD Strahler stream order RAIN Has it rained at this site in the past 24 hours? Y/N REG Physiographic region REP % of the distance between SERC sites that this site represents ROS Rosgen classification of the stream SHAPE Channel shape (rectangular, triangular, round) SHRUBHT For shrub vegetation - height ft SIDES How many sides of the stream have buffers? SITENUM Site sampling number STADN Number of downstream SERC sampling station STAUP Number of upstream SERC station SUB Stream substrate SVEG Site vegetation TEAMNO Team number TEAMTYPE Linda or Jim's team TIL Is there evidence of drain tiles? TIME Starting time of site visit TREE What is/are the dominant type of trees in the buffer? VELD Depth of channel used to calculate velocity in VELL Length used to calculate velocity ft VELT Time used to calculate velocity sec VELW Width of channel used to calculate velocity ft WSHED Name of watershed X X coordinate Y Y coordinate 239 APPENDIX 3: DEVELOPMENT OF GIS WATERSHED AND LANDSCAPE DESCRIPTORS Drainage density Scale:12-digit watershed wshed sub-wshed Total watershed area Total stream length Drainage density (m2) (m) (m2/m) /1000 GB 556 56915564 54199.0 1.05 P 916 55902604 29142.0 1.92 938 11244993 9077.0 1.24 953 13424478 19926.0 0.67 952 30835410 59029.0 0.52 974 17950372 9864.0 1.82 972 24469914 28261.0 0.87 985 14281884 7205 1.98 977 24166924 26414 0.91 992 26041828 22539.0 1.16 1.2 StM 908 21876292 16751.00 1.31 930 16815686 15335.0 1.10 948 9576821 8471.0 1.13 949 35845358 28750 1.25 1.19 Cal 835 15293606 31563 0.48 E 39 14324647 9576 1.50 106 37354528 37096 1.01 1.25 G 65 20772786 17942 1.16 66 27002656 23049 1.17 127 64961724 60248 1.08 150 16078304 17503 0.92 128815470 1.08 M UL 34566940 30147 1.15 U2L 19687981 14282 1.38 U3L 23428182 27606 0.85 UR 29735160 22792 1.30 BL 18982308 13815 1.37 B2L 25133874 24352 1.03 B3L 31699351 30550 1.04 B4L 16752963 15900 1.05 BR 15850822 14344 1.11 1.14 Cat 421 6125971 4827 1.27 427 10679008 14544 0.73 435 50582976 48408 1.04 1.02 240 Tortuosity Scale:12-digit watershed sub-wshed strm m wshed length tortuosity (m) (m) (strm m/wshed m) 556 13892 12645 1.1 916 11938 12405 1.0 938 5001 4953 1.0 953 7514 5148 1.5 952 12663 9847 1.3 974 6832 5957 1.1 972 13056 9014 1.4 985 5473 5493 1.0 977 8035 6040 1.3 992 8315 5758 1.4 1.2 908 6453 5575 1.2 930 6233 5431 1.1 948 2906 2816 1.0 949 4337 4295 1.0 1.09 835 11714 8511 1.4 39 3968 3656 1.1 106 17201 12308 1.4 1.24 65 9180 6111 1.5 66 12627 7552 1.7 127 16606 11189 1.5 150 4864 4709 1.0 1.42 UL 13323 8077 1.6 U2L 5740 4636 1.2 U3L 8170 5709 1.4 UR 8155 7626 1.1 BL 8490 8710 1.0 B2L 10195 8633 1.2 B3L 13353 11586 1.2 B4L 7892 6245 1.3 BR 5768 4460 1.3 1.3 421 4827 4400 1.1 427 4986 4550 1.1 435 13299 10331 1.3 1.16 241 Topographic IndexScale:12-digit watershed wshed sub-wshed max elev min elev delta index (strm m/wshed m) (m) (m) (m/m2)E6 GB 556 61.6 12.3 49.3 0.87 P 916 86.3 12.3 74 1.32 938 36.9 12.3 24.6 2.19 953 36.9 12.3 24.6 1.83 952 24.6 12.3 12.3 0.40 974 36.9 12.3 24.6 1.37 972 36.9 12.3 24.6 1.01 985 36.9 12.3 24.6 1.72 977 36.9 12.3 24.6 1.02 992 36.9 12.3 24.6 0.94 1.3 StM 908 160.2 36.9 123 5.64 930 147.9 12.3 136 8.06 948 123.2 12.3 111 11.58 949 135.6 12.3 123 3.44 7.18 Cal 835 147.9 12.3 136 8.87 E 39 456.1 234.2 222 15.49 106 431.48 36.98 395 10.56 13.03 G 65 899.9 468.5 431 20.77 66 949.2 468.5 481 17.80 127 813.6 295.9 518 7.97 150 825.9 406.8 419 26.07 18.15 M UL 650.4 325 325 9.41 U2L 690.1 404.7 285 14.50 U3L 737.5 427.7 310 13.22 UR 914.1 487.8 426 14.34 BL 693.7 393.8 300 15.80 B2L 773.2 307.1 466 18.54 B3L 881.6 437.2 444 14.02 B4L 849.1 451.6 398 23.73 BR 885.2 404.7 481 30.31 17.10 Cat 421 801.3 493.1 308 50.31 427 813.6 480.8 333 31.16 435 813.6 369.8 444 8.77 30.08 242 Land Use Scale:12-digit watershed wshed sub-wshed ag % ag forest % forest % Land Use (m2) (m2) Explained GB 556 4.1E+07 72.88 14857613 26.1 99.0 P 916 2.3E+07 40.30 31112891 55.7 938 5419200 48.19 5624172 50.0 953 6435769 47.94 6693087 49.9 952 1E+07 32.90 20455166 66.3 974 1863530 10.38 16036882 89.3 972 7850681 32.08 16412816 67.1 985 7490371 52.45 6679140 46.8 977 8612412 35.64 14580260 60.3 992 1.2E+07 44.24 14160430 54.4 38.2 60.0 98.2 StM 908 5469504 25.00 12566745 57.4 930 6733723 40.04 8583656 51.0 948 4574044 47.76 4545859 47.5 949 1.8E+07 51.58 14584291 40.7 41.10 49.16 90.3 Cal 835 8339525 54.5 4357106 28.5 83.0 E 39 1E+07 72.08 1800763 12.6 106 1.4E+07 38.38 12603072 33.7 55.23 23.16 78.4 G 65 8481172 40.83 9338777 45.0 66 1.7E+07 62.35 7589699 28.1 127 3.3E+07 50.78 27182097 41.8 150 4821351 29.99 6232361 38.8 45.98 38.42 84.4 M UL 3E+07 86.77 2871521 8.31 U2L 1.5E+07 77.01 2223716 11.29 U3L 2E+07 83.41 2678773 11.43 UR 1.5E+07 50.38 4872916 16.39 BL 1.6E+07 84.02 1998227 10.53 B2L 2.1E+07 85.30 2187812 8.70 B3L 2.3E+07 72.18 5764327 18.18 B4L 1.3E+07 77.50 1876505 11.20 BR 1.1E+07 68.15 2377833 15.00 76.08 12.3 88.4 Cat 421 3633599 59.31 1278561 20.9 427 7524421 70.46 1594412 14.9 435 2.8E+07 56.22 12728429 25.2 62.00 20.32 82.3 243 APPENDIX 4: UNIVERSITY OF MARYLAND AGRICULTURAL BUFFER SURVEY CORRELATION MATRIX FOR BUFFER SURVEY DATA (See Appendix 2 for description of parameters) LBUF ORD Log ORD GRA VEL CWavg CDavg CXSavg LBUF 1.00 0.24 0.24 -0.16 0.03 0.13 0.22 0.22 ORD 0.24 1.00 0.98 -0.33 0.46 0.37 0.34 0.42 logORD 0.24 0.98 1.00 -0.36 0.42 0.32 0.31 0.36 GRA -0.16 -0.33 -0.36 1.00 -0.12 -0.07 -0.19 -0.11 VEL 0.03 0.46 0.42 -0.12 1.00 0.27 0.38 0.35 CWavg 0.13 0.37 0.32 -0.07 0.27 1.00 0.39 0.83 CDavg 0.22 0.34 0.31 -0.19 0.38 0.39 1.00 0.72 CXSavg 0.22 0.42 0.36 -0.11 0.35 0.83 0.72 1.00 logCXS CW vsCD CA avg Log CA BS avg FSavg BSvs FS BTavg LBUF 0.22 0.02 0.18 0.17 -0.07 -0.03 -0.03 0.47 ORD 0.38 0.04 0.18 0.18 -0.04 0.10 0.00 0.09 logORD 0.34 0.01 0.18 0.19 -0.03 0.09 0.01 0.11 GRA -0.22 0.05 -0.17 -0.24 0.06 0.11 0.02 -0.08 VEL 0.40 0.00 0.24 0.21 0.01 0.15 -0.05 0.07 CWavg 0.81 0.56 0.10 0.07 -0.11 -0.02 -0.09 -0.02 CDavg 0.67 -0.28 0.22 0.18 -0.07 0.18 -0.13 0.16 CXSavg 0.76 0.16 0.11 0.10 -0.10 0.15 -0.13 0.06 logCXSav 1.00 0.20 0.19 0.17 -0.06 -0.02 -0.07 0.07 CWvsCD 0.20 1.00 -0.18 -0.26 -0.08 -0.11 0.00 -0.08 CAavg 0.19 -0.18 1.00 0.95 -0.13 -0.04 -0.15 0.15 logCA 0.17 -0.26 0.95 1.00 -0.08 -0.04 -0.11 0.13 BSavg -0.06 -0.08 -0.13 -0.08 1.00 0.36 0.56 -0.04 FSavg -0.02 -0.11 -0.04 -0.04 0.36 1.00 -0.21 0.23 BSvsFS -0.07 0.00 -0.15 -0.11 0.56 -0.21 1.00 -0.13 BTavg 0.07 -0.08 0.15 0.13 -0.04 0.23 -0.13 1.00 244 BTvar DBH BAS BGav SHRht BSH BIN BTOT BTvar 1.00 -0.01 0.00 -0.04 -0.05 0.10 -0.06 0.98 DBHavg -0.01 1.00 0.19 -0.04 -0.11 -0.18 -0.05 0.00 BASavg 0.00 0.19 1.00 0.06 -0.30 0.03 -0.04 -0.02 BGav -0.04 -0.04 0.06 1.00 -0.04 -0.05 -0.03 0.07 SHRht -0.05 -0.11 -0.30 -0.04 1.00 0.00 -0.05 -0.03 BSHav 0.10 -0.18 0.03 -0.05 0.00 1.00 -0.06 0.09 BINav -0.06 -0.05 -0.04 -0.03 -0.05 -0.06 1.00 -0.07 BTOT 0.98 0.00 -0.02 0.07 -0.03 0.09 -0.07 1.00 Tree G SH IN BW FW BKS Log BKS1 TTOT 1.00 -0.23 -0.05 -0.29 -0.25 -0.31 -0.05 0.01 GTOT -0.23 1.00 -0.03 -0.07 0.05 0.00 -0.03 0.09 SHTOT -0.05 -0.03 1.00 -0.03 -0.04 -0.07 0.08 0.20 INTOT -0.29 -0.07 -0.03 1.00 0.32 0.24 -0.11 -0.13 BWav -0.25 0.05 -0.04 0.32 1.00 0.57 -0.13 -0.13 FWav -0.31 0.00 -0.07 0.24 0.57 1.00 -0.10 -0.26 BKSav -0.05 -0.03 0.08 -0.11 -0.13 -0.10 1.00 0.34 logBKS1 0.01 0.09 0.20 -0.13 -0.13 -0.26 0.34 1.00 FKS BKM logBKM1 FKM BKMD FKMD Log FKDM FKS 1.00 0.01 -0.26 0.78 0.49 0.06 -0.50 BKMav 0.01 1.00 0.23 0.06 -0.33 -0.02 0.12 logBKM -0.26 0.23 1.00 -0.15 -0.51 0.01 0.30 FKM 0.78 0.06 -0.15 1.00 0.48 -0.10 -0.45 BKMD 0.49 -0.33 -0.51 0.48 1.00 -0.05 -0.68 245 APPENDIX 5: SERC* NITRATE DATA DEVELOPMENT station March 1-4 April 5-7 May 3-6 mean/stn std/stn German Branch Mainstem stations 325 3.90 4.56 3.95 4.13 0.37 322 3.80 3.06 3.54 3.47 0.38 327 4.16 3.59 3.76 3.84 0.29 310 4.38 3.82 4.38 4.19 0.32 Upper watershed 332 2.18 1.73 2.04 1.98 0.23 312 0.10 0.34 0.16 0.20 0.13 311 4.25 3.46 3.41 3.71 0.47 306 2.45 1.91 2.31 2.22 0.28 313 5.34 4.77 5.21 5.11 0.30 326 7.67 6.93 7.25 7.28 0.37 314 4.78 4.91 5.49 5.06 0.38 328 4.12 3.74 4.25 4.04 0.26 Lower watershed 323 2.34 1.32 1.38 1.68 0.57 321 1.34 1.95 0.06 1.12 0.96 320 3.43 2.45 2.10 2.66 0.69 319 0.21 0.76 0.01 0.33 0.39 317 0.82 0.61 0.51 0.65 0.16 331 2.93 3.20 3.72 3.28 0.40 318 7.95 6.48 2.74 5.72 2.68 330 8.98 8.82 8.84 8.88 0.08 246 station March 1-4 April 5-7 May 3-6 mean/stn std/stn Gunpowder 411 4.05 3.19 3.34 3.53 0.46 412 0.82 0.67 0.74 0.74 0.08 413 4.60 3.52 3.70 3.94 0.58 414 5.63 4.38 4.75 4.92 0.64 415 5.09 4.15 4.48 4.58 0.48 416 5.60 4.67 4.78 5.02 0.51 417 3.95 3.05 3.11 3.37 0.50 418 4.22 3.21 3.18 3.53 0.59 419 3.91 3.35 3.44 3.56 0.30 420 3.50 2.76 2.84 3.03 0.41 421 3.16 2.59 2.64 2.80 0.31 422 3.00 3.47 3.64 3.37 0.34 423 3.95 2.04 2.22 2.74 1.05 424 3.12 2.11 2.15 2.46 0.57 425 4.77 4.09 4.36 4.40 0.34 426 4.14 3.33 3.31 3.59 0.47 427 2.98 2.14 2.23 2.45 0.46 428 6.70 5.95 5.87 6.17 0.46 429 8.05 7.46 7.54 7.68 0.32 430 4.79 4.50 4.32 4.54 0.24 431 8.74 7.95 8.08 8.26 0.42 247 APPENDIX 6: SERC* DATA CORRELATION MATRIX (See Appendix 2 for description of parameters) NO3 pctAG pctFOR forag siteag sitecrop CAFO NO3 1.00 -0.46 0.44 0.46 0.24 0.15 0.34 siteag 0.24 0.29 -0.31 -0.29 1.00 0.97 -0.22 sitecrop 0.15 0.42 -0.43 -0.40 0.97 1.00 -0.28 CAFO 0.34 -0.38 0.40 0.37 -0.22 -0.28 1.00 CAFOfctr 0.19 -0.26 0.29 0.26 -0.17 -0.23 0.73 CAFOfcsq 0.12 -0.19 0.20 0.19 -0.21 -0.26 0.69 ZONED 0.13 -0.13 0.14 0.13 -0.28 -0.32 0.49 RFB -0.24 -0.05 0.06 0.04 -0.32 -0.28 0.08 Gbuf 0.18 0.07 -0.06 -0.07 0.40 0.37 0.00 BTOT -0.01 -0.05 0.06 0.04 -0.30 -0.28 0.38 BT -0.18 -0.12 0.13 0.11 -0.43 -0.41 0.33 CAFOfctr CAFOfcsq ZONED RFB Gbuf BTOT NO3 0.19 0.12 0.13 -0.24 0.18 -0.01 siteag -0.17 -0.21 -0.28 -0.32 0.40 -0.30 sitecrop -0.23 -0.26 -0.32 -0.28 0.37 -0.28 CAFO 0.73 0.69 0.49 0.08 0.00 0.38 CAFOfctr 1.00 0.96 0.75 -0.07 0.02 0.29 CAFOfcsq 0.96 1.00 0.79 -0.08 -0.04 0.29 ZONED 0.75 0.79 1.00 -0.21 -0.12 0.22 RFB -0.07 -0.08 -0.21 1.00 -0.21 0.80 Gbuf 0.02 -0.04 -0.12 -0.21 1.00 0.02 BTOT 0.29 0.29 0.22 0.80 0.02 1.00 BT 0.28 0.29 0.18 0.92 -0.24 0.89 248 BT BG BSH BIN DBH BAS shrubHt NO3 -0.18 0.03 -0.22 0.13 -0.17 -0.26 0.01 siteag -0.43 -0.05 -0.08 0.00 -0.36 -0.16 -0.24 sitecrop -0.41 -0.06 -0.03 -0.06 -0.30 -0.09 -0.28 CAFO 0.33 -0.09 -0.11 -0.08 0.07 -0.17 0.32 CAFOfctr 0.28 0.03 -0.05 0.08 -0.06 -0.04 0.21 CAFOfcsq 0.29 0.01 -0.04 -0.02 -0.04 0.01 0.16 ZONED 0.18 0.56 -0.08 -0.06 0.06 0.10 0.11 RFB 0.92 -0.25 -0.11 -0.10 0.13 0.23 0.09 Gbuf -0.24 0.07 -0.08 0.53 -0.59 -0.54 -0.19 BTOT 0.89 0.03 -0.16 -0.12 0.00 0.13 0.10 BT 1.00 -0.12 -0.14 -0.12 0.14 0.25 0.15 topo RZSlope Fslope Bwidslo BSvsFS tortuous NO3 0.45 0.25 0.31 0.04 -0.15 -0.36 siteag -0.17 0.00 -0.23 0.44 -0.21 -0.16 sitecrop -0.28 -0.08 -0.34 0.44 -0.19 -0.13 CAFO 0.21 0.20 0.56 0.08 -0.13 -0.05 CAFOfctr 0.13 0.15 0.51 0.01 -0.13 0.03 CAFOfcsq 0.10 0.03 0.45 0.01 -0.10 0.04 ZONED 0.05 -0.01 0.45 -0.08 0.08 -0.06 RFB 0.03 0.21 0.23 -0.17 0.44 0.40 Gbuf -0.10 0.02 -0.10 0.63 -0.18 -0.11 BTOT 0.00 0.12 0.40 0.07 0.40 0.33 BT 0.07 0.21 0.43 -0.19 0.44 0.38 249 ORDER RZwtr RZKsurf RZKmin RZKminD NO3 -0.06 0.16 -0.16 -0.27 0.07 siteag 0.09 0.07 -0.07 -0.11 -0.07 sitecrop 0.13 0.05 0.00 -0.04 -0.07 CAFO -0.03 0.29 -0.08 -0.06 0.06 CAFOfctr 0.06 0.14 -0.12 -0.04 0.10 CAFOfcsq 0.05 0.03 -0.11 -0.05 0.11 ZONED -0.08 -0.07 0.11 0.02 -0.03 RFB -0.30 0.12 -0.14 0.00 -0.39 Gbuf -0.18 0.20 0.02 0.05 0.20 BTOT -0.40 0.27 -0.12 -0.01 -0.43 BT -0.31 0.10 -0.11 -0.01 -0.38 Fkmin FKminD CW CD CXS NO3 0.07 0.23 -0.01 -0.16 0.12 siteag -0.10 0.12 -0.01 -0.13 -0.06 sitecrop -0.14 0.13 -0.04 -0.11 -0.10 CAFO 0.15 0.01 0.15 0.21 0.27 CAFOfctr 0.18 0.01 -0.01 0.12 0.16 CAFOfcsq 0.15 0.01 0.01 0.06 0.16 ZONED 0.08 -0.13 -0.08 0.21 0.01 RFB -0.17 -0.39 -0.10 0.47 -0.25 Gbuf -0.12 0.23 -0.13 -0.16 -0.04 BTOT -0.06 -0.44 -0.06 0.50 -0.18 BT -0.12 -0.41 -0.12 0.51 -0.22 250 CWvsCD CA Fwtr Fksurf gradient NO3 -0.04 0.07 -0.06 -0.06 0.19 siteag -0.10 0.08 -0.13 -0.17 -0.16 sitecrop -0.05 0.05 -0.17 -0.19 -0.20 CAFO 0.23 0.18 0.09 0.02 0.31 CAFOfctr 0.04 0.29 0.15 -0.03 0.35 CAFOfcsq -0.01 0.29 0.12 -0.07 0.37 ZONED 0.09 0.09 0.14 0.14 0.38 RFB 0.55 -0.45 0.54 0.49 0.22 Gbuf -0.14 0.07 -0.14 -0.22 0.05 BTOT 0.53 -0.40 0.56 0.47 0.38 BT 0.55 -0.37 0.57 0.50 0.38 NO3 pctAG pctFOR forag siteag BG 0.03 0.11 -0.09 -0.10 -0.05 BSH -0.22 0.19 -0.19 -0.19 -0.08 BIN 0.13 -0.21 0.22 0.20 0.00 DBH -0.17 -0.02 0.05 0.01 -0.36 BAS -0.26 0.44 -0.42 -0.44 -0.16 shrubHt 0.01 -0.24 0.27 0.24 -0.24 topo 0.45 -0.85 0.76 0.81 -0.17 RZSlope 0.25 -0.58 0.58 0.57 0.00 Fslope 0.31 -0.57 0.59 0.55 -0.23 Bwidslo 0.04 0.29 -0.28 -0.28 0.44 BSvsFS -0.15 -0.09 0.09 0.08 -0.21 tortuous -0.36 0.11 -0.11 -0.11 -0.16 gradient 0.19 -0.27 0.24 0.24 -0.16 ORDER -0.06 0.16 -0.15 -0.14 0.09 RZwtr 0.16 -0.07 0.08 0.07 0.07 251 sitecrop CAFO BG -0.06 -0.09 BSH -0.03 -0.11 BIN -0.06 -0.08 DBH -0.30 0.07 BAS -0.09 -0.17 shrubHt -0.28 0.32 topo -0.28 0.21 RZSlope -0.08 0.20 Fslope -0.34 0.56 Bwidslo 0.44 0.08 BSvsFS -0.19 -0.13 tortuous -0.13 -0.05 gradient -0.20 0.31 ORDER 0.13 -0.03 RZwtr 0.05 0.29 CAFOfctr CAFOfcsq ZONED RFB Gbuf BG 0.03 0.01 0.56 -0.25 0.07 BSH -0.05 -0.04 -0.08 -0.11 -0.08 BIN 0.08 -0.02 -0.06 -0.10 0.53 DBH -0.06 -0.04 0.06 0.13 -0.59 BAS -0.04 0.01 0.10 0.23 -0.54 shrubHt 0.21 0.16 0.11 0.09 -0.19 topo 0.13 0.10 0.05 0.03 -0.10 RZSlope 0.15 0.03 -0.01 0.21 0.02 252 BTOT BG 0.03 BSH -0.16 BIN -0.12 DBH 0.00 BAS 0.13 shrubHt 0.10 topo 0.00 RZSlope 0.12 CAFOfctr CAFOfcsq ZONED RFB Gbuf BTOT Fslope 0.51 0.45 0.45 0.23 -0.10 0.40 Bwidslo 0.01 0.01 -0.08 -0.17 0.63 0.07 BSvsFS -0.13 -0.10 0.08 0.44 -0.18 0.40 tortuous 0.03 0.04 -0.06 0.40 -0.11 0.33 gradient 0.35 0.37 0.38 0.22 0.05 0.38 ORDER 0.06 0.05 -0.08 -0.30 -0.18 -0.40 RZwtr 0.14 0.03 -0.07 0.12 0.20 0.27 BT BG BSH BIN DBH BAS shrubHt BG -0.12 1.00 -0.10 0.04 -0.03 0.07 -0.09 BSH -0.14 -0.10 1.00 -0.04 -0.19 -0.13 0.57 BIN -0.12 0.04 -0.04 1.00 -0.30 -0.27 -0.10 DBH 0.14 -0.03 -0.19 -0.30 1.00 0.43 -0.01 BAS 0.25 0.07 -0.13 -0.27 0.43 1.00 -0.03 shrubHt 0.15 -0.09 0.57 -0.10 -0.01 -0.03 1.00 topo 0.07 -0.13 -0.17 0.12 -0.10 -0.44 0.10 RZSlope 0.21 -0.03 -0.19 0.45 -0.09 -0.14 0.16 Fslope 0.43 0.03 -0.20 0.04 0.16 -0.05 0.22 Bwidslo -0.19 -0.01 -0.07 -0.07 -0.38 -0.28 -0.14 BSvsFS 0.44 0.18 -0.15 -0.07 0.30 0.22 0.05 tortuous 0.38 -0.05 0.34 -0.02 -0.11 -0.11 0.12 gradient 0.38 0.12 -0.13 -0.16 -0.11 -0.25 0.24 ORDER -0.31 -0.23 0.13 0.07 0.13 0.20 0.03 RZwtr 0.10 -0.07 -0.23 -0.05 0.16 -0.14 -0.15 253 topo RZSlope Fslope Bwidslo BSvsFS tortuous gradient BG -0.13 -0.03 0.03 -0.01 0.18 -0.05 0.12 BSH -0.17 -0.19 -0.20 -0.07 -0.15 0.34 -0.13 BIN 0.12 0.45 0.04 -0.07 -0.07 -0.02 -0.16 DBH -0.10 -0.09 0.16 -0.38 0.30 -0.11 -0.11 BAS -0.44 -0.14 -0.05 -0.28 0.22 -0.11 -0.25 shrubHt 0.10 0.16 0.22 -0.14 0.05 0.12 0.24 topo 1.00 0.49 0.40 -0.27 0.07 -0.08 0.31 RZSlope 0.49 1.00 0.54 -0.26 0.03 0.06 0.09 Fslope 0.40 0.54 1.00 -0.15 0.32 -0.23 0.39 Bwidslo -0.27 -0.26 -0.15 1.00 -0.23 -0.17 -0.03 BSvsFS 0.07 0.03 0.32 -0.23 1.00 -0.06 0.31 tortuous -0.08 0.06 -0.23 -0.17 -0.06 1.00 0.09 gradient 0.31 0.09 0.39 -0.03 0.31 0.09 1.00 ORDER -0.16 -0.09 -0.27 0.02 -0.46 -0.30 -0.55 RZwtr 0.02 -0.01 0.15 0.16 0.02 0.02 0.00 ORDER RZwtr RZKsurf RZKmin RZKminD Fwtr Fksurf BG -0.23 -0.07 0.15 0.14 -0.15 0.19 0.27 BSH 0.13 -0.23 -0.12 -0.10 -0.21 -0.10 -0.16 BIN 0.07 -0.05 -0.08 -0.06 0.15 0.01 -0.11 DBH 0.13 0.16 0.41 0.22 -0.12 0.16 0.24 BAS 0.20 -0.14 0.17 0.31 -0.19 0.16 0.19 shrubHt 0.03 -0.15 -0.09 -0.12 -0.23 0.15 0.23 topo -0.16 0.02 -0.20 -0.24 -0.01 0.22 0.19 RZSlope -0.09 -0.01 -0.21 -0.16 -0.25 0.36 0.36 ORDER RZwtr RZKsurf RZKmin RZKminD Fwtr Fksurf Fslope -0.27 0.15 0.12 -0.14 -0.23 0.30 0.46 Bwidslo 0.02 0.16 0.06 0.13 0.09 -0.30 -0.26 BSvsFS -0.46 0.02 0.32 0.08 -0.56 0.55 0.77 tortuous -0.30 0.02 -0.38 -0.10 -0.32 0.38 0.09 gradient -0.55 0.00 -0.03 -0.21 -0.13 0.36 0.48 ORDER 1.00 -0.04 -0.02 0.13 0.29 -0.50 -0.59 RZwtr -0.04 1.00 0.13 0.18 -0.17 0.18 0.05 254 Fkmin FKminD CW CD CXS CWvsCD CA BG -0.04 -0.22 -0.14 0.38 -0.17 0.22 -0.19 BSH 0.07 -0.14 -0.03 -0.10 -0.01 -0.09 0.10 BIN 0.14 0.05 -0.28 -0.03 -0.11 -0.08 0.15 DBH 0.19 -0.22 0.12 0.25 0.12 0.18 -0.04 BAS 0.05 -0.30 0.17 0.17 0.03 0.15 0.03 shrubHt 0.09 -0.25 0.02 0.18 -0.04 0.13 0.06 topo 0.15 0.03 -0.21 -0.01 -0.16 0.05 -0.19 RZSlope 0.14 -0.22 -0.39 0.37 -0.36 0.44 -0.19 Fslope 0.05 -0.23 -0.11 0.38 -0.08 0.50 -0.16 Bwidslo 0.01 0.12 0.28 -0.22 0.13 -0.10 0.10 BSvsFS -0.18 -0.52 -0.30 0.53 -0.45 0.44 -0.63 tortuous -0.08 -0.30 -0.29 0.26 -0.27 0.07 -0.34 gradient -0.47 -0.08 -0.13 0.27 -0.32 0.27 -0.45 ORDER 0.55 0.18 0.22 -0.48 0.27 -0.33 0.76 RZwtr 0.13 -0.06 -0.08 0.09 -0.02 0.03 -0.12 NO3 pctAG pctFOR forag siteag sitecrop CAFO RZKsurf -0.16 0.24 -0.23 -0.23 -0.07 0.00 -0.08 RZKmin -0.27 0.27 -0.27 -0.26 -0.11 -0.04 -0.06 RZKminD 0.07 -0.03 0.05 0.06 -0.07 -0.07 0.06 Fwtr -0.06 -0.23 0.22 0.20 -0.13 -0.17 0.09 Fksurf -0.06 -0.24 0.24 0.22 -0.17 -0.19 0.02 Fkmin 0.07 -0.22 0.23 0.23 -0.10 -0.14 0.15 FKminD 0.23 -0.05 0.06 0.08 0.12 0.13 0.01 CW -0.01 0.27 -0.27 -0.29 -0.01 -0.04 0.15 CD -0.16 -0.10 0.12 0.10 -0.13 -0.11 0.21 CXS 0.12 0.07 -0.04 -0.06 -0.06 -0.10 0.27 CWvsCD -0.04 -0.14 0.16 0.14 -0.10 -0.05 0.23 CA 0.07 0.08 -0.04 -0.06 0.08 0.05 0.18 255 APPENDIX 7: MBSS* DATA CORRELATION MATRIX Description of MBSS* variables MBSS Variable Name Description NO3_LAB NO3 FIBI_98 FIBI BIBI_98 BIBI PHI PHI REGION_E Geographic region PHYS_CP Physiographic region ORDER_1 Is the stream 1st order? TEMP_FLD Water temperature DO_FLD Dissolved oxygen PH_FLD Water pH (field and lab) COND_FLD Water conductivity (field and lab) PH_LAB Water pH (field and lab) COND_LAB Water conductivity (field and lab) ANC_LAB Water ANC SO4_LAB Water SO4 DOC_LAB Water DOC ACID_NO Presence of acid source PASTURE Presence of adjacent pasture CHANNEL Channeliztion? INSTRHAB Instream habitat score EPI_SUB Epifaunal substrate score POOLQUAL Pool quality score RIFFQUAL Riffle quality score CHAN_ALT Channel alteration score BANKSTAB Bank stability score EMBEDDED Embeddedness score SHADING Shading score REMOTE Remoteness score AESTHET Aesthetics score WOOD_DEB Woody debris score NUMROOT Number of rootwads RIP_WID Riparian buffer width 256 MBSS Variable Name Description FOR_BUFF Presence of forest buffer ADJ_CROP Presence of adjacent crop AGRI Percent agriculture in site catchment PASTUR Percent pasture in site catchment ROWCROP Percent rowcrops in site catchment PROBCROP Percent probable crops in site catchment cropmix sum of probcrop & rowcrop URBAN Percent urban land use in site catchment HIGHURB Percent high urban land use in site catchment LOWURB Percent low urban land use in site catchment FOREST Percent forest in site catchment CONIFER Percent coniferous forest in site catchment DECIDFOR Percent deciduous forest in site catchment MIXEDFOR Percent mixed forest in site catchment WETLANDS Percent wetlands in site catchment EMERGWET Percent emergent wetlands in site catchment WOODYWET Percent woody wetlands in site catchment BARREN Percent barren land in site catchment WATER Percent water in site catchment COALMINE Percent coal mines in site catchment MAXDEPTH Channel maximum depth VEL_DPTH Velocity depth ST_GRAD Stream gradient AVGWID Channel average width AVGTHAL Average thalweg velocity AVG_VEL Average water velocity CH_FLOW Channel flow ACREAGE Site catchment acreage TRANS Percent transitional land use in site catchment PROBCROP % probable crops in site catchment ROWCROP % rowcrops in site catchment cropmix sum of probcrop & rowcrop BKTRFLAG Presence of brook trout BLACKWAT Presence of blackwater conditions 257 Correlation Matrix NO3_LAB FIBI_98 BIBI_98 PHI RIP_WIDm NO3_LAB 1.00 0.13 0.02 0.13 -0.10 FIBI_98 0.13 1.00 0.12 0.47 0.00 BIBI_98 0.02 0.12 1.00 0.23 0.35 PHI 0.13 0.47 0.23 1.00 0.18 RIP_WIDm -0.10 0.00 0.35 0.18 1.00 TEMP_FLD -0.22 0.08 0.05 -0.28 0.17 AGRI 0.54 0.19 0.04 0.21 -0.14 URBAN -0.25 -0.02 0.05 -0.23 0.15 FOREST -0.39 -0.19 -0.09 -0.05 0.02 TEMP_FLD AGRI URBAN FOREST NO3_LAB -0.22 0.54 -0.25 -0.39 FIBI_98 0.08 0.19 -0.02 -0.19 BIBI_98 0.05 0.04 0.05 -0.09 PHI -0.28 0.21 -0.23 -0.05 RIP_WIDm 0.17 -0.14 0.15 0.02 TEMP_FLD 1.00 -0.37 0.67 -0.15 AGRI -0.37 1.00 -0.48 -0.71 URBAN 0.67 -0.48 1.00 -0.28 FOREST -0.15 -0.71 -0.28 1.00 258 ORDER DO_FLD PH_FLD COND_FLD PH_ LAB NO3_LAB -0.20 0.22 -0.18 0.18 -0.19 FIBI_98 0.08 0.05 0.11 0.00 0.00 BIBI_98 0.18 0.09 0.18 -0.17 0.14 PHI -0.17 0.40 0.01 0.03 -0.24 RIP_WIDm 0.21 -0.13 0.15 -0.06 0.12 TEMP_FLD 0.74 -0.55 0.41 -0.20 0.69 AGRI -0.38 0.39 -0.16 0.04 -0.37 URBAN 0.87 -0.34 0.47 -0.15 0.86 FOREST -0.31 -0.13 -0.20 0.08 -0.31 ORDER 1.00 -0.36 0.40 -0.23 0.95 DO_FLD -0.36 1.00 0.06 -0.03 -0.35 PH_FLD 0.40 0.06 1.00 -0.08 0.32 COND_FLD -0.23 -0.03 -0.08 1.00 -0.23 PH_LAB 0.95 -0.35 0.32 -0.23 1.00 COND_LAB -0.30 0.23 0.01 0.49 -0.29 ANC_LAB -0.11 0.05 0.30 0.27 -0.11 SO4_LAB 0.93 -0.37 0.25 -0.20 0.98 DOC_LAB 0.09 -0.53 -0.32 0.07 0.08 COND_LAB ANC_LAB SO4_LAB DOC_LAB NO3_LAB 0.26 -0.14 -0.20 -0.20 FIBI_98 -0.18 -0.11 -0.04 -0.20 BIBI_98 -0.19 0.01 0.10 -0.26 PHI 0.09 0.06 -0.26 -0.32 RIP_WIDm -0.11 -0.02 0.12 0.00 TEMP_FLD -0.29 -0.06 0.67 0.26 AGRI 0.19 0.01 -0.39 -0.42 URBAN -0.09 0.00 0.83 0.03 FOREST -0.12 0.00 -0.27 0.42 ORDER -0.30 -0.11 0.93 0.09 DO_FLD 0.23 0.05 -0.37 -0.53 PH_FLD 0.01 0.30 0.25 -0.32 COND_FLD 0.49 0.27 -0.20 0.07 PH_LAB -0.29 -0.11 0.98 0.08 COND_LAB 1.00 0.54 -0.28 -0.22 ANC_LAB 0.54 1.00 -0.12 -0.21 SO4_LAB -0.28 -0.12 1.00 0.13 DOC_LAB -0.22 -0.21 0.13 1.00 259 PASTURE INSTR HAB EPI_ SUB VEL_ DPTH POOL QUAL NO3_LAB -0.14 0.26 0.15 0.10 0.08 FIBI_98 -0.03 0.41 0.35 0.44 0.42 BIBI_98 0.06 0.19 0.37 0.24 0.20 PHI -0.18 0.83 0.62 0.77 0.72 RIP_WID -0.23 0.07 0.18 0.16 0.22 TEMP_FLD 0.36 -0.38 -0.08 -0.10 -0.02 AGRI -0.16 0.33 0.19 0.21 0.13 URBAN 0.46 -0.37 0.06 0.03 0.08 FOREST -0.21 -0.06 -0.26 -0.25 -0.21 ORDER 0.57 -0.32 0.16 0.10 0.10 DO_FLD -0.20 0.49 0.36 0.46 0.22 PH_FLD 0.05 -0.07 0.19 0.22 0.17 COND_FLD -0.16 0.05 -0.14 -0.04 0.06 PH_LAB 0.68 -0.37 0.10 0.02 -0.02 COND_LAB -0.21 0.18 -0.05 0.07 0.13 ANC_LAB -0.08 0.06 -0.09 0.09 0.14 SO4_LAB 0.70 -0.38 0.05 -0.01 -0.05 DOC_LAB 0.05 -0.35 -0.38 -0.45 -0.31 PASTURE 1.00 -0.25 0.00 -0.03 -0.14 INSTRHAB -0.25 1.00 0.67 0.64 0.60 EPI_SUB 0.00 0.67 1.00 0.64 0.51 VEL_DPTH -0.03 0.64 0.64 1.00 0.72 POOL QUAL -0.14 0.60 0.51 0.72 1.00 RIFF QUAL 0.03 0.55 0.69 0.70 0.38 CHAN _ALT -0.02 0.46 0.55 0.48 0.41 BANKSTAB -0.07 0.25 0.18 0.13 0.08 EMBEDDED -0.07 -0.41 -0.70 -0.58 -0.35 260 RIFFQUAL CHAN_ALT BANKSTAB EMBEDDED NO3_LAB 0.08 0.20 0.19 -0.11 FIBI_98 0.33 0.19 0.12 -0.26 BIBI_98 0.30 0.25 -0.04 -0.41 PHI 0.60 0.41 0.16 -0.47 RIP_WIDm 0.13 0.07 -0.06 -0.13 TEMP_FLD -0.05 -0.17 -0.02 0.03 AGRI 0.11 0.25 0.02 -0.21 URBAN 0.06 0.03 -0.03 -0.13 FOREST -0.17 -0.29 0.01 0.34 ORDER 0.14 0.02 0.02 -0.20 DO_FLD 0.39 0.28 0.08 -0.34 PH_FLD 0.24 0.09 -0.14 -0.29 COND_FLD -0.10 -0.11 -0.02 0.18 PH_LAB 0.07 -0.03 -0.02 -0.19 COND_LAB -0.01 0.07 -0.13 0.04 ANC_LAB -0.03 0.00 -0.23 -0.02 SO4_LAB 0.03 -0.04 -0.01 -0.13 DOC_LAB -0.33 -0.41 0.08 0.56 PASTURE 0.03 -0.02 -0.07 -0.07 INSTRHAB 0.55 0.46 0.25 -0.41 EPI_SUB 0.69 0.55 0.18 -0.70 VEL_DPTH 0.70 0.48 0.13 -0.58 POOLQUAL 0.38 0.41 0.08 -0.35 RIFFQUAL 1.00 0.38 0.17 -0.58 CHAN_ALT 0.38 1.00 0.34 -0.51 BANKSTAB 0.17 0.34 1.00 0.01 EMBEDDED -0.58 -0.51 0.01 1.00 261 FLOW SHADE REMOTE AESTHET WOOD_ DEB NO3_LAB 0.22 -0.11 -0.12 0.06 -0.06 FIBI_98 0.10 -0.20 0.03 0.12 0.07 BIBI_98 0.10 0.17 0.20 0.27 0.19 PHI 0.29 -0.06 -0.06 0.33 0.03 RIP_WIDm -0.11 0.26 0.32 0.29 0.25 TEMP_FLD -0.22 -0.03 0.50 -0.09 0.29 AGRI 0.27 -0.04 -0.29 0.06 -0.21 URBAN -0.18 0.12 0.63 -0.09 0.29 FOREST -0.16 -0.06 -0.20 -0.01 -0.02 ORDER -0.15 0.01 0.66 0.00 0.31 DO_FLD 0.21 -0.07 -0.24 0.09 -0.29 PH_FLD -0.19 0.17 0.33 0.11 0.03 COND_FLD 0.04 0.01 -0.20 -0.12 -0.09 PH_LAB -0.16 0.05 0.70 0.01 0.28 COND_LAB 0.09 0.09 -0.24 -0.14 -0.15 ANC_LAB -0.07 0.13 -0.07 -0.03 -0.08 SO4_LAB -0.16 0.03 0.64 -0.02 0.28 DOC_LAB -0.12 -0.13 0.01 -0.21 0.12 PASTURE 0.00 -0.11 0.39 0.00 0.07 INSTRHAB 0.30 -0.08 -0.17 0.24 -0.04 EPI_SUB 0.16 0.07 0.15 0.23 -0.01 VEL_DPTH 0.25 -0.05 0.11 0.23 -0.02 POOLQUAL 0.25 -0.01 0.09 0.18 0.11 RIFFQUAL 0.21 -0.02 0.13 0.24 -0.08 CHAN_ALT 0.35 0.06 0.01 0.15 -0.02 BANKSTAB 0.05 -0.20 -0.01 0.09 -0.01 EMBEDDED -0.06 -0.11 -0.17 -0.23 0.03 CH_FLOW 1.00 -0.10 -0.11 -0.03 -0.01 SHADING -0.10 1.00 0.27 0.10 -0.16 REMOTE -0.11 0.27 1.00 0.32 0.09 AESTHET -0.03 0.10 0.32 1.00 -0.17 WOOD_DEB -0.01 -0.16 0.09 -0.17 1.00 NUMROOT 0.07 0.11 0.07 -0.03 0.42 MAXDEPTH 0.25 0.25 0.07 0.27 -0.35 ST_GRAD -0.02 -0.43 -0.02 -0.20 0.64 AVGWID 0.06 0.02 0.01 -0.01 -0.11 262 NUMROOT MAXDEPTH ST_GRAD AVGWID NO3_LAB -0.09 0.04 -0.06 -0.05 FIBI_98 0.07 0.17 0.18 0.10 BIBI_98 0.18 0.02 0.12 0.03 PHI 0.23 0.45 0.08 0.12 RIP_WIDm 0.22 0.13 0.04 0.18 TEMP_FLD 0.17 -0.17 0.23 0.23 AGRI -0.14 0.10 -0.09 -0.11 URBAN 0.24 -0.16 0.23 0.25 FOREST -0.05 0.02 -0.09 -0.10 ORDER 0.25 -0.15 0.27 0.36 DO_FLD -0.18 0.14 -0.07 -0.01 PH_FLD 0.04 0.01 0.09 0.25 COND_FLD -0.07 0.12 -0.15 -0.06 PH_LAB 0.23 -0.21 0.26 0.19 COND_LAB -0.10 0.10 -0.16 -0.04 ANC_LAB 0.01 0.10 -0.10 0.01 SO4_LAB 0.25 -0.22 0.22 0.24 DOC_LAB 0.02 -0.15 0.00 -0.01 PASTURE 0.11 -0.15 0.09 0.00 INSTRHAB 0.07 0.33 0.01 0.06 EPI_SUB 0.09 0.19 0.11 0.20 VEL_DPTH 0.16 0.40 0.11 0.28 POOLQUAL 0.24 0.53 0.13 0.28 RIFFQUAL 0.03 0.21 0.07 0.22 CHAN_ALT 0.11 0.10 0.06 0.23 BANKSTAB -0.02 -0.03 -0.01 0.19 EMBEDDED -0.06 -0.05 -0.18 -0.11 CH_FLOW 0.07 0.25 -0.02 0.06 SHADING 0.11 0.25 -0.43 0.02 REMOTE 0.07 0.07 -0.02 0.01 AESTHET -0.03 0.27 -0.20 -0.01 WOOD_DEB 0.42 -0.35 0.64 -0.11 NUMROOT 1.00 0.01 0.22 0.15 MAXDEPTH 0.01 1.00 -0.54 0.31 ST_GRAD 0.22 -0.54 1.00 -0.25 AVGWID 0.15 0.31 -0.25 1.00 263 THAL VEL FLOW ACREAGE WETLANDS NO3_LAB 0.02 -0.05 0.01 0.01 -0.04 FIBI_98 0.15 0.17 0.19 0.16 0.13 BIBI_98 -0.07 0.10 -0.02 0.05 0.08 PHI 0.42 0.06 0.27 0.17 0.05 RIP_WIDm 0.10 0.03 0.07 0.16 0.02 TEMP_FLD -0.11 0.33 -0.02 0.05 0.31 AGRI 0.04 -0.12 0.03 -0.09 -0.12 URBAN -0.08 0.31 0.11 -0.08 0.29 FOREST 0.03 -0.12 -0.12 0.15 -0.10 ORDER -0.06 0.36 0.17 0.04 0.35 DO_FLD 0.12 -0.09 0.18 -0.05 -0.07 PH_FLD 0.00 0.14 0.19 -0.04 0.12 COND_FLD 0.13 -0.16 0.00 0.01 -0.15 PH_LAB -0.16 0.34 0.05 -0.09 0.34 COND_LAB 0.14 -0.19 0.10 -0.02 -0.17 ANC_LAB 0.12 -0.12 0.05 -0.05 -0.12 SO4_LAB -0.17 0.30 0.06 -0.08 0.30 DOC_LAB -0.04 0.04 -0.14 0.20 0.03 PASTURE -0.13 0.12 -0.02 -0.12 0.14 INSTRHAB 0.32 -0.01 0.21 0.12 -0.01 EPI_SUB 0.15 0.11 0.28 0.08 0.11 VEL_DPTH 0.39 0.11 0.45 0.13 0.09 POOLQUAL 0.51 0.13 0.32 0.19 0.08 RIFFQUAL 0.18 0.07 0.31 0.21 0.12 CHAN_ALT 0.14 0.03 0.28 0.06 -0.02 BANKSTAB 0.06 0.04 0.14 0.13 0.07 EMBEDDED 0.04 -0.15 -0.19 0.09 -0.16 CH_FLOW 0.34 -0.02 0.20 0.23 -0.02 SHADING 0.02 -0.44 -0.08 -0.10 -0.38 REMOTE 0.10 0.06 0.03 0.07 0.12 AESTHET 0.18 -0.16 0.03 0.10 -0.07 WOOD_DEB -0.26 0.65 -0.17 -0.17 0.55 NUMROOT -0.01 0.21 0.02 -0.08 0.14 MAXDEPTH 0.82 -0.53 0.33 0.45 -0.43 ST_GRAD -0.43 0.95 -0.19 -0.23 0.77 AVGWID 0.42 -0.21 0.69 0.51 -0.20 AVGTHAL 1.00 -0.41 0.52 0.60 -0.32 264 BARREN WATER HIGHURB LOWURB NO3_LAB -0.09 -0.08 -0.16 -0.15 FIBI_98 -0.07 0.02 -0.12 -0.10 BIBI_98 0.06 0.02 -0.16 -0.15 PHI -0.12 0.08 -0.01 0.04 RIP_WIDm 0.14 0.02 -0.02 -0.03 TEMP_FLD 0.35 0.09 -0.01 -0.05 AGRI -0.18 0.07 -0.28 -0.26 URBAN 0.39 -0.02 0.30 0.36 FOREST -0.15 -0.10 0.09 0.03 ORDER 0.48 -0.01 -0.03 -0.04 DO_FLD -0.16 -0.03 -0.13 0.02 PH_FLD 0.26 0.07 0.01 0.13 COND_FLD -0.11 0.00 0.15 0.18 PH_LAB 0.36 -0.02 -0.04 -0.05 COND_LAB -0.14 0.02 0.34 0.47 ANC_LAB -0.06 0.12 0.14 0.29 SO4_LAB 0.43 -0.04 -0.02 -0.04 DOC_LAB 0.02 -0.11 0.05 -0.13 PASTURE 0.11 0.02 -0.06 -0.06 INSTRHAB -0.17 0.06 -0.06 -0.01 EPI_SUB 0.07 0.03 -0.12 -0.11 VEL_DPTH 0.05 0.13 -0.06 0.02 POOLQUAL 0.07 0.11 0.06 0.14 RIFFQUAL 0.04 0.08 -0.12 -0.05 CHAN_ALT 0.09 -0.01 -0.01 0.06 BANKSTAB 0.11 -0.09 -0.08 -0.12 EMBEDDED -0.07 -0.06 0.15 0.09 CH_FLOW -0.17 0.13 0.04 0.06 SHADING 0.07 0.12 0.05 0.17 REMOTE -0.03 0.01 -0.06 -0.03 AESTHET -0.13 0.09 -0.16 -0.09 WOOD_DEB 0.13 -0.06 0.13 -0.02 NUMROOT 0.29 -0.01 0.13 0.10 MAXDEPTH -0.10 0.20 0.04 0.18 ST_GRAD -0.02 -0.16 0.02 -0.14 AVGWID 0.74 0.03 0.04 0.08 AVGTHAL -0.10 0.18 0.13 0.21 265 PASTUR PROBCROP ROWCROP crop CONIFER NO3_LAB 0.40 0.22 0.02 0.21 -0.10 FIBI_98 0.10 -0.01 0.14 0.02 0.06 BIBI_98 0.07 -0.18 0.16 -0.05 0.03 PHI 0.10 -0.10 0.15 0.03 -0.04 RIP_WIDm -0.04 -0.08 -0.03 -0.10 0.07 TEMP_FLD -0.26 -0.01 0.06 -0.19 0.26 AGRI 0.52 0.36 0.26 0.49 -0.19 URBAN -0.26 -0.15 0.13 -0.20 0.23 FOREST -0.35 -0.27 -0.39 -0.37 0.03 ORDER -0.22 -0.09 0.14 -0.19 0.26 DO_FLD 0.25 -0.07 0.18 0.14 -0.17 PH_FLD -0.04 -0.24 0.22 -0.02 0.00 COND_FLD 0.08 0.06 -0.15 0.07 -0.13 PH_LAB -0.19 -0.08 0.13 -0.19 0.25 COND_LAB 0.18 -0.01 -0.02 0.13 -0.21 ANC_LAB 0.12 -0.20 0.05 0.00 -0.19 SO4_LAB -0.20 -0.07 0.09 -0.21 0.23 DOC_LAB -0.24 0.15 -0.38 -0.25 0.23 PASTURE -0.11 0.06 0.06 -0.04 0.08 INSTRHAB 0.18 -0.06 0.14 0.08 -0.07 EPI_SUB 0.11 -0.14 0.28 0.09 0.02 VEL_DPTH 0.09 -0.09 0.25 0.09 -0.04 POOLQUAL 0.04 -0.09 0.21 0.04 -0.01 RIFFQUAL 0.06 -0.12 0.21 0.06 -0.01 266 DECIDFOR MIXEDFOR EMERGWET WOODYWET NO3_LAB -0.32 -0.20 -0.12 -0.06 FIBI_98 -0.18 0.02 0.02 -0.02 BIBI_98 0.06 0.06 0.08 -0.17 PHI 0.06 -0.11 -0.08 -0.17 RIP_WIDm 0.02 0.10 0.11 0.01 TEMP_FLD -0.17 0.22 0.39 0.16 AGRI -0.41 -0.38 -0.19 -0.30 URBAN -0.06 0.25 0.37 -0.13 FOREST 0.50 0.21 -0.11 0.42 ORDER -0.12 0.24 0.48 -0.04 DO_FLD 0.19 -0.20 -0.19 -0.42 PH_FLD 0.22 0.01 0.11 -0.27 COND_FLD 0.01 -0.11 -0.17 0.11 PH_LAB -0.12 0.26 0.39 -0.08 COND_LAB 0.07 -0.20 -0.21 -0.20 ANC_LAB 0.27 -0.14 -0.15 -0.21 SO4_LAB -0.11 0.24 0.46 -0.05 DOC_LAB -0.22 0.20 0.11 0.60 PASTURE -0.06 0.09 0.15 -0.07 INSTRHAB 0.05 -0.16 -0.15 -0.21 EPI_SUB 0.02 -0.05 0.04 -0.30 VEL_DPTH 0.08 -0.11 0.04 -0.35 POOLQUAL 0.03 -0.08 0.06 -0.24 RIFFQUAL 0.05 -0.07 0.00 -0.19 267 AVGTHAL AVG_VEL FLOW ACREAGE WET LANDS AVG_VEL -0.41 1.00 -0.16 -0.22 0.86 FLOW 0.52 -0.16 1.00 0.53 -0.15 ACREAGE 0.60 -0.22 0.53 1.00 -0.18 WETLANDS -0.32 0.86 -0.15 -0.18 1.00 BARREN -0.10 0.03 0.28 -0.04 -0.02 WATER 0.18 -0.14 0.11 0.00 -0.11 HIGHURB 0.13 -0.02 0.10 0.07 -0.03 LOWURB 0.21 -0.14 0.13 0.03 -0.12 BARREN WATER HIGHURB LOWURB AVG_VEL 0.03 -0.14 -0.02 -0.14 FLOW 0.28 0.11 0.10 0.13 ACREAGE -0.04 0.00 0.07 0.03 WETLANDS -0.02 -0.11 -0.03 -0.12 BARREN 1.00 -0.02 -0.04 -0.03 WATER -0.02 1.00 -0.03 0.01 HIGHURB -0.04 -0.03 1.00 0.76 LOWURB -0.03 0.01 0.76 1.00 268 PASTUR PROBCROP ROWCROP crop CONIFER CHAN_ALT 0.09 -0.07 0.26 0.14 -0.02 BANKSTAB 0.03 0.10 -0.08 -0.01 0.06 EMBEDDED -0.11 0.24 -0.40 -0.09 0.01 CH_FLOW 0.04 0.16 0.09 0.16 -0.02 SHADING 0.26 -0.07 -0.11 0.25 -0.41 REMOTE 0.04 0.05 -0.09 0.01 -0.04 AESTHET 0.20 0.03 -0.11 0.25 -0.29 WOOD_DEB -0.43 -0.21 0.34 -0.55 0.65 NUMROOT -0.16 -0.15 0.12 -0.24 0.23 MAXDEPTH 0.30 0.14 -0.25 0.47 -0.58 ST_GRAD -0.50 -0.23 0.52 -0.59 0.85 AVGWID 0.06 -0.03 -0.13 0.14 -0.23 AVGTHAL 0.22 0.18 -0.29 0.37 -0.46 AVG_VEL -0.50 -0.22 0.55 -0.60 0.84 FLOW 0.11 0.00 -0.05 0.16 -0.21 ACREAGE 0.04 0.20 -0.32 0.14 -0.19 WETLANDS -0.41 -0.19 0.45 -0.49 0.69 BARREN -0.10 -0.16 0.06 -0.04 0.02 WATER 0.09 0.00 0.04 0.19 -0.18 HIGHURB -0.23 0.02 -0.15 -0.12 0.03 LOWURB -0.12 -0.13 -0.06 0.00 -0.16 PASTUR 1.00 -0.05 -0.28 0.27 -0.59 PROBCROP -0.05 1.00 -0.40 0.59 -0.20 ROWCROP -0.28 -0.40 1.00 -0.16 0.46 crop 0.27 0.59 -0.16 1.00 -0.66 CONIFER -0.59 -0.20 0.46 -0.66 1.00 DECIDFOR 0.07 -0.37 -0.18 0.05 -0.53 MIXEDFOR -0.56 -0.28 0.24 -0.66 0.84 EMERGWET -0.36 -0.27 0.37 -0.39 0.48 WOODYWET -0.09 0.27 -0.53 0.00 -0.13 269 DECIDFOR MIXEDFOR EMERGWET WOODYWET CHAN_ALT 0.02 -0.08 0.07 -0.35 BANKSTAB -0.10 -0.04 0.09 0.14 EMBEDDED -0.07 0.01 -0.07 0.44 CH_FLOW -0.14 -0.13 -0.12 -0.07 SHADING 0.34 -0.34 -0.15 -0.18 REMOTE 0.01 0.01 -0.09 0.03 AESTHET 0.18 -0.26 -0.30 0.01 WOOD_DEB -0.35 0.61 0.46 -0.05 NUMROOT -0.09 0.16 0.37 -0.04 MAXDEPTH 0.30 -0.54 -0.37 0.10 ST_GRAD -0.46 0.79 0.45 -0.22 AVGWID 0.19 -0.26 0.44 0.09 AVGTHAL 0.19 -0.43 -0.30 0.18 AVG_VEL -0.45 0.71 0.46 -0.20 FLOW 0.15 -0.24 0.11 -0.06 ACREAGE -0.02 -0.18 -0.12 0.42 WETLANDS -0.38 0.57 0.32 -0.16 BARREN 0.11 -0.03 0.76 -0.02 WATER 0.09 -0.18 -0.11 -0.10 HIGHURB -0.03 0.12 -0.01 -0.02 LOWURB 0.17 -0.08 -0.11 -0.13 PASTUR 0.07 -0.56 -0.36 -0.09 PROBCROP -0.37 -0.28 -0.27 0.27 ROWCROP -0.18 0.24 0.37 -0.53 crop 0.05 -0.66 -0.39 0.00 CONIFER -0.53 0.84 0.48 -0.13 DECIDFOR 1.00 -0.40 -0.18 -0.15 MIXEDFOR -0.40 1.00 0.37 -0.06 EMERGWET -0.18 0.37 1.00 -0.10 WOODYWET -0.15 -0.06 -0.10 1.00 270 COALMINE TRANS BKTRFLAG BLACKWAT NO3_LAB 0.02 0.21 0.05 -0.07 FIBI_98 0.13 -0.10 0.07 -0.14 BIBI_98 0.03 -0.12 0.08 -0.08 PHI 0.13 -0.13 0.00 -0.17 RIP_WIDm -0.03 0.11 0.08 0.00 TEMP_FLD 0.12 0.08 0.06 0.01 AGRI -0.06 -0.10 0.00 -0.16 URBAN 0.03 0.07 -0.01 -0.05 FOREST 0.03 0.03 0.01 0.22 ORDER 0.11 0.07 0.01 -0.03 DO_FLD -0.02 -0.19 -0.04 -0.19 PH_FLD -0.06 0.01 0.01 -0.20 COND_FLD -0.04 0.23 0.01 0.01 PH_LAB 0.11 0.06 -0.01 -0.04 COND_LAB -0.07 0.08 -0.01 -0.15 ANC_LAB -0.06 -0.04 -0.01 -0.21 SO4_LAB 0.13 0.07 -0.01 -0.02 DOC_LAB 0.09 0.13 -0.01 0.40 PASTURE 0.08 -0.01 -0.01 -0.07 INSTRHAB 0.08 -0.16 -0.03 -0.11 EPI_SUB -0.01 -0.14 -0.01 -0.13 VEL_DPTH 0.07 -0.12 -0.03 -0.23 POOLQUAL 0.04 -0.03 -0.05 -0.18 RIFFQUAL 0.02 -0.12 0.04 -0.13 CHAN_ALT -0.06 -0.15 -0.05 -0.12 BANKSTAB 0.08 -0.04 0.06 0.15 EMBEDDED -0.06 0.10 -0.01 0.27 CH_FLOW 0.01 -0.08 -0.05 -0.01 SHADING -0.33 0.04 0.01 0.00 REMOTE -0.11 0.05 0.05 -0.01 AESTHET -0.09 -0.01 0.08 -0.05 WOOD_DEB 0.46 0.04 0.03 -0.03 NUMROOT 0.17 -0.04 -0.02 -0.02 MAXDEPTH -0.35 -0.03 -0.01 -0.07 ST_GRAD 0.58 0.00 -0.03 -0.05 AVGWID -0.16603 -0.02964 -0.00376 -0.02541 271 COALMINE TRANS BKTRFLAG BLACKWAT AVGTHAL -0.28 -0.04 -0.01 -0.05 AVG_VEL 0.67 0.01 -0.03 -0.04 FLOW -0.11 -0.06 -0.03 -0.07 ACREAGE -0.14 -0.02 0.09 0.12 WETLANDS 0.67 0.01 -0.02 -0.03 BARREN -0.01 0.00 -0.01 -0.01 WATER -0.10 0.04 0.01 -0.06 LOWURB -0.09 0.00 0.01 -0.07 PASTUR -0.34 -0.04 0.02 -0.13 PROBCROP -0.17 0.03 0.05 0.10 ROWCROP 0.39 -0.08 -0.06 -0.16 CONIFER 0.53 0.04 -0.04 0.17 DECIDFOR -0.29 -0.06 -0.03 -0.14 MIXEDFOR 0.37 0.06 -0.03 0.08 EMERGWET 0.34 0.00 -0.02 0.03 WOODYWET -0.14 0.05 0.10 0.24 COALMINE 1.00 -0.03 -0.02 -0.04 TRANS -0.03 1.00 0.02 0.10 BKTRFLAG -0.02 0.02 1.00 -0.02 BLACKWAT -0.04 0.10 -0.02 1.00 272 APPENDIX 8: SERC* NO3 SAS MODELING Model Parameter Search SAS language DATA one ; INFILE 'C:\Documents and Settings\lbarker\Desktop\apr5cafofctr.dat' ; INPUT Sitenum$ NO3 pctAG pctFOR forag siteag sitecrop CAFO CAFOfctr CAFOfcsq ZONED RFB test Gbuf BTOT BT BG BSH BIN DBH BAS shrubHt topo RZSlope Fslope Bwidslo BSvsFS tortuous gradient test ORDER RZwtr RZKsurf RZKmin RZKminD Fwtr Fksurf test Fkmin FKminD CW CD CXS CWvsCD CA ; proc print; proc corr ; proc reg ; model NO3 = siteag sitecrop CAFO CAFOfctr CAFOfcsq ZONED RFB Gbuf BTOT BT BG BSH DBH BAS shrubHt RZSlope Fslope Bwidslo BSvsFS gradient ORDER RZwtr RZKsurf RZKmin RZKminD Fwtr Fksurf Fkmin FKminD CW CD CXS CWvsCD CA / selection=rsquare adjrsq best=3; run; quit; Results R-Square Selection Method Number in Adjusted Model R2 R-Square Variables in Model 1 0.1125 0.0856 CAFO 1 0.0982 0.0708 Fslope 1 0.0707 0.0425 RZKmin 2 0.2179 0.1690 BT Fslope 2 0.2141 0.1650 siteag CAFO 2 0.2032 0.1534 CAFO BT 3 0.3210 0.2553 BTOT BT Fslope 3 0.2915 0.2229 BT Fslope RZKsurf 3 0.2844 0.2152 siteag Fslope FKminD 4 0.3938 0.3130 BTOT BT Fslope FKminD 4 0.3850 0.3030 BTOT BT Fslope RZKsurf 4 0.3709 0.2870 BTOT BT RZSlope FKminD 273 Model Calibration SAS language *Using the parameters: siteag CAFO; model NO3 = siteag CAFO / collinoint tol VIF R influence ; output out=resNO3 residual = NO3res; proc univariate normal plot; var NO3res ; run; quit; Results Analysis of Variance Sum of Mean Source DF Squares Square F Value Pr > F Model 2 30.20774 15.10387 4.36 0.0212 Error 32 110.85382 3.46418 Corrected Total 34 141.06155 Root MSE 1.86123 R-Square 0.2141 Dependent Mean 3.38686 Adj R-Sq 0.1650 Coeff Var 54.95453 Parameter Estimates Parameter Standard Variance Variable DF Estimate Error t Value Pr > |t| Tolerance Inflation Intercept 1 0.44894 1.33552 0.34 0.7389 . 0 siteag 1 0.03848 0.01892 2.03 0.0503 0.95172 1.05073 CAFO 1 0.21487 0.08476 2.54 0.0163 0.95172 1.05073 Collinearity Diagnostics (intercept adjusted) Condition --Proportion of Variation- Number Eigenvalue Index siteag CAFO 1 1.21974 1.00000 0.39013 0.39013 2 0.78026 1.25030 0.60987 0.60987 274 APPENDIX 9: SERC* NO3 S-PLUS MODELING Regression tree model with all parameters ? all sites |CWvsCD<3.73462 siteag<71.2148 CAFO<1.5 BAS<77.5 CA<44 4.2660 5.8500 0.5958 1.9850 2.5070 4.5870 275 Regression Tree with all Parameters tree(formula = NO3 ~ siteag + sitecrop + CAFO + CAFOfctr + CAFOfcsq + ZONED + RFB + Gbuf + BTOT + BT + BG + BSH + BIN + DBH + BAS + shrubHt + topo + RZSlope + Fslope + Bwidslo + BSvsFS + tortuous + gradient + ORDER + RZwtr + RZKsurf + RZKmin + RZKminD + Fwtr + Fksurf + Fkmin + FKminD + CW + CD + CXS + CWvsCD + CA, data = Apr5.CAFOfctr, na.action = na.exclude, mincut = 5, minsize = 10, mindev = 0.01) Variables actually used in tree construction: "CWvsCD""siteag""CAFO""BAS""CA" Number of terminal nodes: 6 Residual mean deviance: 1.491 = 43.23 / 29 node), split, n, deviance, yval * denotes terminal node 1) root 35 141.100 3.3870 4) siteag<71.2148 9 5.246 4.2660 * 5) siteag>71.2148 5 3.508 5.8500 * 3) CWvsCD>3.73462 21 75.590 2.4230 12) BAS<77.5 10 9.636 1.2900 24)CA<44 5 3.014 0.5958 * 25) CA>44 5 1.796 1.9850 * 13) BAS>77.5 6 9.782 2.5070 * Total deviance for 2-term model is sum of RMD for all terminal nodes RMD for 5-term model = total deviance/ (total n - number of terminal nodes) 2) CWvsCD<3.73462 14 16.820 4.8320 6) CAFO<1.5 16 24.970 1.7470 7) CAFO>1.5 5 19.880 4.5870 * total deviance = 61.67 total n = 35 number of terminal nodes = 3 RMD = 1.93 276 Development of equivalent R2 for regression tree model Phys NO3 delta square corrected total SS CP 5.72 3.08 9.50 101.65 CP CP 1.12 -1.52 2.32 10.22 Pd CP 2.15 -0.49 0.24 111.88 ALL CP 0.24 -2.40 5.78 CP 2.22 -0.42 0.17 Resid SS CP 0.20 -2.44 5.96 deviance CP 7.28 4.64 21.57 CP 4.13 CP 1.98 -0.66 0.43 46.86 CP 1.55 -1.09 1.19 9.18 CP 3.71 1.07 1.14 CP 3.28 0.64 0.41 60.17 CP 1.68 -0.96 0.92 Pd 2.88 CP 5.06 2.42 5.84 2.07 CP 4.04 1.40 1.95 CP 2.62 -0.02 0.00 4.95 CP 0.65 -1.99 3.97 ALL 5.25 CP 1.14 -1.50 2.24 3.51 CP 2.66 0.02 0.00 24.97 CP 0.20 -2.44 5.96 19.88 CP 7.75 5.11 26.09 CP 0.20 -2.44 5.96 53.61 P 6.17 1.67 2.78 P 2.80 -1.71 2.92 Res/Cor SS P 4.92 0.41 0.17 CP 0.59 P 3.94 -0.56 0.32 Pd 0.48 P 4.87 0.36 0.13 ALL 0.48 P 4.64 0.13 0.02 P 5.05 0.55 0.30 R2 P 5.02 0.51 0.26 CP 0.41 P 4.54 0.03 0.00 Pd 0.52 P 5.05 0.54 0.29 ALL 0.52 P 3.53 -0.97 0.95 P 3.15 -1.36 1.84 P 4.99 0.49 0.24 P 4.42 -0.08 0.01 mean CP 2.64 P 4.51 ALL 3.39 277 SERC* NO3 S-PLUS Modeling Coastal Plain = German Branch |siteag<57.9429 ORDER<2.5 0.9937 4.4910 2.2550 Regression tree: tree(formula = NO3 ~ siteag + sitecrop + CAFO + CAFOfctr + CAFOfcsq + ZONED + RFB + Gbuf + BTOT + BT + BG + BSH + BIN + DBH + BAS + shrubHt + topo + RZSlope + Fslope + Bwidslo + BSvsFS + tortuous + gradient + ORDER + RZwtr + RZKsurf + RZKmin + RZKminD + Fwtr + Fksurf + Fkmin + FKminD + CW + CD + CXS + CWvsCD + CA, data = SERC.CP, na.action = na.exclude, mincut = 5, minsize = 10, mindev = 0.01) Variables actually used in tree construction: [1] "siteag" "ORDER" Number of terminal nodes: 3 Residual mean deviance: 3.346 = 60.22 / 18 Distribution of residuals: Min. 1st Qu. Median Mean 3rd Qu. Max. -4.25400 -0.78220 -0.03148 0.00000 1.02700 3.25700 node), split, n, deviance, yval * denotes terminal node 1) root 21 101.700 2.6400 2) siteag<57.9429 6 4.193 0.9937 * 3) siteag>57.9429 15 74.700 3.2980 6) ORDER<2.5 7 46.860 4.4910 * 7) ORDER>2.5 8 9.177 2.2550 * 278 SERC* NO3 S-PLUS Modeling Piedmont= Gunpowder |sitecrop<50.858 3.683 4.964 Regression tree: tree(formula = NO3 ~ siteag + sitecrop + CAFO + CAFOfctr + CAFOfcsq + ZONED + RFB + Gbuf + BTOT + BT + BG + BSH + BIN + DBH + BAS + shrubHt + topo + RZSlope + Fslope + Bwidslo + BSvsFS + tortuous + gradient + ORDER + RZwtr + RZKsurf + RZKmin + RZKminD + Fwtr + Fksurf + Fkmin + FKminD + CW + CD + CXS + CWvsCD + CA, data = SERC.P, na.action = na.exclude, mincut = 5, minsize = 10, mindev = 0.01) Variables actually used in tree construction: [1] "sitecrop" Number of terminal nodes: 2 Residual mean deviance: 0.4126 = 4.951 / 12 Distribution of residuals: Min. 1st Qu. Median Mean 3rd Qu. Max. -0.88680 -0.40190 -0.07044 0.00000 0.08862 1.31100 node), split, n, deviance, yval * denotes terminal node 1) root 14 10.220 4.507 2) sitecrop<50.858 5 2.877 3.683 * 3) sitecrop>50.858 9 2.074 4.964 * 279 APPENDIX 10: MBSS* NO3 SAS MODELING Model Search SAS language DATA one ; INFILE 'd:\f nitrate paper\app\sascaldata.dat' ; INPUT NO3 RIP_WIDm TEMP_FLD For_Buff CROP Reg_East AGRI PHYS_CP URBAN FOREST ORDER_1 DO_FLD PH_FLD COND_FLD PH_LAB COND_LAB ANC_LAB SO4_LAB DOC_LAB ACID_no PASTURE INSTRHAB EPI_SUB VEL_DPTH POOLQUAL RIFFQUAL CHAN_ALT BANKSTAB EMBEDDED CH_FLOW SHADING REMOTE AESTHET WOOD_DEB NUMROOT MAXDEPTH ST_GRAD AVGWID AVGTHAL AVG_VEL FLOW ACREAGE WETLANDS BARREN WATER HIGHURB LOWURB PASTUR PROBCROP ROWCROP cropmix CONIFER DECIDFOR MIXEDFOR EMERGWET WOODYWET COALMINE TRANS BKTRFLAG BLACKWAT ; *proc print; * proc corr ; proc reg ; *model NO3 = RIP_WIDm TEMP_FLD For_Buff CROP Reg_East AGRI PHYS_CP URBAN FOREST ORDER_1 DO_FLD PH_FLD COND_FLD PH_LAB COND_LAB ANC_LAB SO4_LAB DOC_LAB ACID_no PASTURE INSTRHAB EPI_SUB VEL_DPTH POOLQUAL RIFFQUAL CHAN_ALT BANKSTAB EMBEDDED CH_FLOW SHADING REMOTE AESTHET WOOD_DEB NUMROOT MAXDEPTH ST_GRAD AVGWID AVGTHAL AVG_VEL FLOW ACREAGE WETLANDS BARREN WATER HIGHURB LOWURB PASTUR PROBCROP ROWCROP cropmix CONIFER DECIDFOR MIXEDFOR EMERGWET WOODYWET COALMINE TRANS BKTRFLAG BLACKWAT / selection=rsquare adjrsq best=3; 280 Results R-Square Selection Method Number in Adjusted Model R2 R-Square Variables in Model 1 0.2444 0.2413 AGRI 1 0.1956 0.1924 FOREST 1 0.1250 0.1215 DECIDFOR -------------------------------------------------------------------------------------------------- 2 0.2984 0.2927 Reg_East AGRI 2 0.2895 0.2837 AGRI BARREN 2 0.2817 0.2758 AGRI BANKSTAB -------------------------------------------------------------------------------------------------- 3 0.3618 0.3540 AGRI COND_LAB ANC_LAB 3 0.3361 0.3279 Reg_East AGRI COND_LAB 3 0.3327 0.3245 Reg_East AGRI BARREN -------------------------------------------------------------------------------------------------- 4 0.4044 0.3946 Reg_East AGRI COND_LAB ANC_LAB 4 0.3945 0.3845 AGRI COND_LAB ANC_LAB BANKSTAB 4 0.3852 0.3751 AGRI COND_LAB ANC_LAB BARREN -------------------------------------------------------------------------------------------------- 5 0.4213 0.4094 Reg_East AGRI COND_LAB ANC_LAB BARREN 5 0.4181 0.4061 Reg_East AGRI COND_LAB ANC_LAB BANKSTAB 5 0.4174 0.4054 CROP Reg_East AGRI COND_LAB ANC_LAB Model Calibration SAS language * Using the variables CROP Reg_East AGRI COND_LAB ANC_LAB ; model NO3 = CROP Reg_East AGRI COND_LAB ANC_LAB / collinoint tol VIF R influence ; output out=resNO3 residual = NO3res; proc univariate normal plot; var NO3res ; run; quit; 281 Results Analysis of Variance Sum of Mean Source DF Squares Square F Value Pr > F Model 5 543.27044 108.65409 34.82 <.0001 Error 243 758.31690 3.12065 Corrected Total 248 1301.58734 Root MSE 1.76653 R-Square 0.4174 Dependent Mean 3.48863 Adj R-Sq 0.4054 Coeff Var 50.63686 Parameter Estimates Parameter Standard Variable DF Estimate Error t Value Pr > |t| Tolerance Intercept 1 -2.79256 0.59976 -4.66 <.0001 . CROP 1 0.67960 0.29186 2.33 0.0207 0.97686 Reg_East 1 0.89173 0.22946 3.89 0.0001 0.95248 AGRI 1 0.07470 0.00818 9.13 <.0001 0.95556 COND_LAB 1 0.01257 0.00211 5.96 <.0001 0.71594 ANC_LAB 1 -0.00258 0.000494 -5.21 <.0001 0.72175 Model Validation SAS Language DATA one ; INFILE 'd:\f nitrate paper\app\sasvaldata.dat' ; INPUT NO3 RIP_WIDm TEMP_FLD For_Buff CROP Reg_East AGRI PHYS_CP URBAN FOREST ORDER_1 DO_FLD PH_FLD COND_FLD PH_LAB COND_LAB ANC_LAB SO4_LAB DOC_LAB ACID_no PASTURE INSTRHAB EPI_SUB VEL_DPTH POOLQUAL RIFFQUAL CHAN_ALT BANKSTAB EMBEDDED CH_FLOW SHADING REMOTE AESTHET WOOD_DEB NUMROOT MAXDEPTH ST_GRAD AVGWID AVGTHAL AVG_VEL FLOW ACREAGE WETLANDS BARREN WATER HIGHURB LOWURB PASTUR PROBCROP ROWCROP cropmix CONIFER DECIDFOR MIXEDFOR EMERGWET WOODYWET COALMINE TRANS BKTRFLAG BLACKWAT ; NO3pred = -2.79 + 0.68*CROP + 0.89*Reg_East +0.075*AGRI +0.013*COND_LAB -0.002*ANC_LAB; res = NO3 - NO3pred ; proc print; Proc corr; run; quit; Results: Correlation of NO3 and NO3pred 0.74371 so R2 = 0.553 282 APPENDIX 11: MBSS* NO3 S-PLUS MODELING Regression Tree with all Parameters | AGRI<54.84 PROBCROP<39.865 MAXDEPTH<42.5WOOD.DEB<7.5 DECIDFOR<19.295 COND.LAB<145.2 CHAN.ALT<12COND.LAB<115.45ACREAGE<5226.23 PH.LAB<6.645 AVG.VEL<0.0925EPI.SUB<4.5 SHADING<42.5WATER<0.6CH.FLOW<92.5CONIFER<0.45WOODYWET<0.42 PROBCROP<16.4 MIXEDFOR<1.68 AVGTHAL<34.125FLOW<0.45WATER<0.275ST.GRAD<0.9 1.015 2.010 1.031 3.160 2.030 4.822 3.184 4.815 5.72611.360 2.688 6.539 3.522 5.084 5.198 6.952 3.804 5.898 3.610 4.687 1.498 3.252 1.749 2.939 283 Tree Model with all Parameters tree(formula = NO3.LAB ~ RIP.WIDm + TEMP.FLD + BUFF.TYP + ADJ.COVR + REGION + AGRI + PHYCPIO + URBAN + FOREST + ORDER + DO.FLD + PH.FLD + COND.FLD + PH.LAB + COND.LAB + ANC.LAB + SO4.LAB + DOC.LAB + ACIDSRC + PASTURE + INSTRHAB + EPI.SUB + VEL.DPTH + POOLQUAL + RIFFQUAL + CHAN.ALT + BANKSTAB + EMBEDDED + CH.FLOW + SHADING + REMOTE + AESTHET + WOOD.DEB + NUMROOT + MAXDEPTH + ST.GRAD + AVGWID + AVGTHAL + AVG.VEL + FLOW + ACREAGE + WETLANDS + BARREN + WATER + HIGHURB + LOWURB + PASTUR + PROBCROP + ROWCROP + crop + CONIFER + DECIDFOR + MIXEDFOR + EMERGWET + WOODYWET + COALMINE + TRANS + BKTRFLAG + BLACKWAT, data = Dec13.MBSS.NO3.model.data, na.action = na.exclude, mincut = 5, minsize = 10, mindev = 0.01) Variables actually used in tree construction: "AGRI" "PROBCROP" "MAXDEPTH" "WOOD.DEB" "DECIDFOR" "COND.LAB"] "CHAN.ALT" "ACREAGE" "PH.LAB" "AVG.VEL" "EPI.SUB" "SHADING" ] "ADJ.COVR" "RIP.WIDm" "SO4.LAB" "REMOTE" "MIXEDFOR" "AVGTHAL" "FLOW" "WATER" "ST.GRAD" Number of terminal nodes: 25 Residual mean deviance: 1.478 = 331.1 / 224 node), split, n, deviance, yval * denotes terminal node 1) root 249 1290.000 3.500 2) AGRI<54.84 64 64.270 1.689 4) PROBCROP<39.865 56 30.730 1.479 8) MAXDEPTH<42.5 22 5.701 1.015 * 9) MAXDEPTH>42.5 34 17.220 1.779 18) WOOD.DEB<7.5 26 7.902 2.010 * 19) WOOD.DEB>7.5 8 3.458 1.031 * 5) PROBCROP>39.865 8 13.750 3.160 * 3) AGRI>54.84 185 943.200 4.126 6) DECIDFOR<19.295 99 573.800 4.898 12) COND.LAB<145.2 31 58.140 3.608 24) CHAN.ALT<12 23 34.180 3.189 48) COND.LAB<115.45 7 4.870 2.030 * 49) COND.LAB>115.45 16 15.800 3.696 98) ACREAGE<5226.23 5 2.531 4.822 * 99) ACREAGE>5226.23 11 4.040 3.184 * 25) CHAN.ALT>12 8 8.262 4.815 * 13) COND.LAB>145.2 68 440.500 5.486 26) PH.LAB<6.645 12 199.900 8.072 52) AVG.VEL<0.0925 7 77.790 5.726 * 284 53) AVG.VEL>0.0925 5 29.630 11.360 * 27) PH.LAB>6.645 56 143.300 4.932 54) EPI.SUB<4.5 5 5.915 2.688 * 55) EPI.SUB>4.5 51 109.700 5.152 110) SHADING<42.5 8 10.420 6.539 * 111) SHADING>42.5 43 81.040 4.894 222) ADJ.COVR:LN,PA 13 14.100 4.045 444) COND.LAB<184.3 5 2.691 4.990 * 445) COND.LAB>184.3 8 4.148 3.454 * 223) ADJ.COVR:CP,FR,OF,SL 30 53.480 5.263 446) RIP.WIDm<45 10 19.430 6.176 892) SO4.LAB<10.975 5 7.263 7.156 * 893) SO4.LAB>10.975 5 2.561 5.196 * 447) RIP.WIDm>45 20 21.540 4.806 894) REMOTE<14 5 6.846 3.694 * 895) REMOTE>14 15 6.448 5.177 * 7) DECIDFOR>19.295 86 242.500 3.237 14) PROBCROP<16.4 5 89.380 5.898 * 15) PROBCROP>16.4 81 115.500 3.073 30) MIXEDFOR<1.68 24 18.330 4.148 60) AVGTHAL<34.125 12 8.852 3.610 * 61) AVGTHAL>34.125 12 2.523 4.687 * 31) MIXEDFOR>1.68 57 57.750 2.620 62) FLOW<0.45 10 5.052 1.498 * 63) FLOW>0.45 47 37.430 2.859 126) WATER<0.275 26 12.270 3.252 * 127) WATER>0.275 21 16.160 2.372 254) ST.GRAD<0.9 10 2.598 1.749 * 255) ST.GRAD>0.9 11 6.141 2.939 * 285 Model Validation NO3 obs pred NO3 obs pred 0.43 1.498 3.23 3.25231 0.76 2.00962 3.3 4.815 1.48 1.03125 3.39 1.749 1.52 2.93909 3.64 3.18364 1.54 1.498 4.05 3.16 1.82 2.03 4.11 4.68667 2.03 2.00962 4.17 3.25231 2.16 2.00962 4.4 3.18364 2.41 2.00962 5.28 5.17667 2.42 2.93909 5.35 2.688 2.56 3.16 6.25 4.815 2.65 3.16 8.27 6.53875 2.66 2.00962 9.01 4.815 2.91 3.25231 9.07 3.45375 Splus model NO3 observed vs predicted R2 = 0.5346 0 2 4 6 8 10 0 2 4 6 8 10 NO3 observed (ppm) N O 3 pr ed ic te d (p pm ) 286 Development of equivalent R2 for regression tree model PHYCPIO NO3 delta square mean CP 0.52 -2.87 8.23 CP 3.39 CP 1.12 -2.27 5.15 P 3.59 CP 1.92 -1.47 2.16 ALL 3.50 CP 1.94 -1.45 2.10 SD CP 2.56 -0.83 0.69 CP 2.89 CP 2.86 -0.53 0.28 P 1.65 CP 4.53 1.14 1.30 corrected total SS CP 4.82 1.43 2.05 CP 1018.62 CP 9.59 6.20 38.46 Pd 418.90 CP 3.18 -0.21 0.04 ALL 1437.51 CP 4.73 1.34 1.80 Resid SS CP 5.21 1.82 3.32 CP deviance CP 12.92 9.53 90.85 102.8 CP 1.63 -1.76 3.09 9.1 CP 1.88 -1.51 2.28 271.1 CP 4.57 1.18 1.40 116.6 CP 0.26 -3.13 9.79 499.6 CP 0.54 -2.85 8.12 Pd deviance CP 0.79 -2.60 6.75 10.67 CP 2.23 -1.16 1.34 65.22 CP 3.62 0.23 0.05 54.2 CP 6.67 3.28 10.77 52.24 CP 1.38 -2.01 4.03 182.33 CP 1.98 -1.41 1.98 ALL deviance CP 3.53 0.14 0.02 30.73 CP 4.05 0.66 0.44 13.75 CP 4.06 0.67 0.45 58.14 CP 4.52 1.13 1.28 440.5 CP 5.29 1.90 3.61 89.38 CP 0.79 -2.60 6.75 115.5 CP 5.96 2.57 6.61 748 CP 0.43 -2.96 8.75 Res/Cor SS CP 2.1 -1.29 1.66 CP 0.49 CP 3.65 0.26 0.07 Pd 0.44 CP 0.88 -2.51 6.29 ALL 0.52 CP 1.02 -2.37 5.61 CP 1.65 -1.74 3.02 R2 CP 2.9 -0.49 0.24 CP 0.51 CP 6.1 2.71 7.35 Pd 0.56 CP 3.62 0.23 0.05 ALL 0.48 CP 1.63 -1.76 3.09 287 CP 4.7 1.31 1.72 CP 8.9 5.51 30.37 CP 10.44 7.05 49.72 CP 0.39 -3.00 8.99 CP 5.41 2.02 4.09 CP 0.98 -2.41 5.80 CP 1.44 -1.95 3.80 CP 2.55 -0.84 0.70 CP 3.39 0.00 0.00 CP 3.72 0.33 0.11 CP 1.41 -1.98 3.92 CP 2.77 -0.62 0.38 CP 4.29 0.90 0.81 CP 14.26 10.87 118.19 CP 7.64 4.25 18.07 CP 1.46 -1.93 3.72 CP 1.82 -1.57 2.46 CP 6.27 2.88 8.30 CP 0.52 -2.87 8.23 CP 1.54 -1.85 3.42 CP 2.47 -0.92 0.84 CP 2.85 -0.54 0.29 CP 0.42 -2.97 8.81 CP 0.44 -2.95 8.69 CP 0.46 -2.93 8.58 CP 0.49 -2.90 8.40 CP 0.61 -2.78 7.72 CP 0.67 -2.72 7.39 CP 0.69 -2.70 7.28 CP 0.71 -2.68 7.18 CP 0.76 -2.63 6.91 CP 0.76 -2.63 6.91 CP 0.83 -2.56 6.55 CP 0.93 -2.46 6.05 CP 0.98 -2.41 5.80 CP 1.11 -2.28 5.19 CP 1.12 -2.27 5.15 CP 1.12 -2.27 5.15 CP 1.16 -2.23 4.97 CP 1.24 -2.15 4.62 CP 1.38 -2.01 4.03 CP 1.48 -1.91 3.64 CP 1.55 -1.84 3.38 CP 2.01 -1.38 1.90 288 CP 2.05 -1.34 1.79 CP 2.09 -1.30 1.69 CP 2.09 -1.30 1.69 CP 2.22 -1.17 1.37 CP 2.25 -1.14 1.30 CP 2.28 -1.11 1.23 CP 2.34 -1.05 1.10 CP 2.58 -0.81 0.65 CP 2.65 -0.74 0.55 CP 2.67 -0.72 0.52 CP 2.79 -0.60 0.36 CP 2.81 -0.58 0.33 CP 2.87 -0.52 0.27 CP 2.88 -0.51 0.26 CP 2.92 -0.47 0.22 CP 3.2 -0.19 0.04 CP 3.25 -0.14 0.02 CP 3.64 0.25 0.06 CP 3.76 0.37 0.14 CP 3.77 0.38 0.15 CP 4.4 1.01 1.02 CP 4.59 1.20 1.44 CP 4.64 1.25 1.57 CP 4.96 1.57 2.47 CP 5.12 1.73 3.00 CP 5.17 1.78 3.17 CP 5.2 1.81 3.28 CP 5.24 1.85 3.43 CP 5.26 1.87 3.50 CP 5.28 1.89 3.58 CP 5.35 1.96 3.85 CP 5.45 2.06 4.25 CP 5.45 2.06 4.25 CP 6.25 2.86 8.19 CP 8.82 5.43 29.50 CP 9.88 6.49 42.14 CP 10.71 7.32 53.60 CP 16.16 12.77 163.11 P 0.5 -3.09 9.53 P 0.89 -2.70 7.27 P 1.07 -2.52 6.33 P 1.35 -2.24 5.00 P 1.74 -1.85 3.41 P 2.16 -1.43 2.04 289 P 2.28 -1.31 1.71 P 2.28 -1.31 1.71 P 2.3 -1.29 1.66 P 2.58 -1.01 1.01 P 3.08 -0.51 0.26 P 3.4 -0.19 0.03 P 3.47 -0.12 0.01 P 3.47 -0.12 0.01 P 3.53 -0.06 0.00 P 3.82 0.23 0.05 P 3.9 0.31 0.10 P 4.05 0.46 0.21 P 4.06 0.47 0.22 P 4.09 0.50 0.25 P 4.11 0.52 0.27 P 4.15 0.56 0.32 P 4.47 0.88 0.78 P 4.61 1.02 1.05 P 4.71 1.12 1.26 P 4.77 1.18 1.40 P 5.26 1.67 2.80 P 5.44 1.85 3.43 P 5.74 2.15 4.64 P 5.83 2.24 5.03 P 6.06 2.47 6.12 P 6.5 2.91 8.49 P 6.81 3.22 10.39 P 6.94 3.35 11.24 P 9.01 5.42 29.41 P 2.04 -1.55 2.39 P 5.06 1.47 2.17 P 1.84 -1.75 3.05 P 1.58 -2.01 4.03 P 1.93 -1.66 2.74 P 3.39 -0.20 0.04 P 3.73 0.14 0.02 P 4.88 1.29 1.67 P 1.23 -2.36 5.55 P 3.07 -0.52 0.27 P 3.25 -0.34 0.11 P 4.26 0.67 0.45 P 5.88 2.29 5.26 P 3.26 -0.33 0.11 P 3.49 -0.10 0.01 290 P 2.03 -1.56 2.42 P 5.89 2.30 5.31 P 2.41 -1.18 1.38 P 2.44 -1.15 1.31 P 4.82 1.23 1.52 P 4.84 1.25 1.57 P 4.49 0.90 0.82 P 3.38 -0.21 0.04 P 4.03 0.44 0.20 P 2.42 -1.17 1.36 P 3.53 -0.06 0.00 P 3.01 -0.58 0.33 P 3.17 -0.42 0.17 P 4.77 1.18 1.40 P 2.34 -1.25 1.55 P 5.25 1.66 2.77 P 2.18 -1.41 1.98 P 4.74 1.15 1.33 P 5.74 2.15 4.64 P 2.21 -1.38 1.90 P 2.32 -1.27 1.60 P 8.14 4.55 20.73 P 9.07 5.48 30.07 P 2.62 -0.97 0.93 P 3.26 -0.33 0.11 P 4.85 1.26 1.60 P 0.76 -2.83 7.99 P 0.91 -2.68 7.16 P 1.31 -2.28 5.18 P 1.42 -2.17 4.69 P 1.49 -2.10 4.40 P 1.52 -2.07 4.27 P 1.53 -2.06 4.23 P 1.77 -1.82 3.30 P 1.77 -1.82 3.30 P 1.83 -1.76 3.09 P 1.85 -1.74 3.02 P 1.91 -1.68 2.81 P 1.99 -1.60 2.55 P 2.03 -1.56 2.42 P 2.05 -1.54 2.36 P 2.06 -1.53 2.33 P 2.09 -1.50 2.24 P 2.11 -1.48 2.18 291 P 2.13 -1.46 2.12 P 2.16 -1.43 2.04 P 2.17 -1.42 2.01 P 2.42 -1.17 1.36 P 2.43 -1.16 1.34 P 2.5 -1.09 1.18 P 2.55 -1.04 1.07 P 2.6 -0.99 0.97 P 2.64 -0.95 0.90 P 2.66 -0.93 0.86 P 2.69 -0.90 0.80 P 2.71 -0.88 0.77 P 2.71 -0.88 0.77 P 2.72 -0.87 0.75 P 2.76 -0.83 0.68 P 2.77 -0.82 0.67 P 2.77 -0.82 0.67 P 2.82 -0.77 0.59 P 2.88 -0.71 0.50 P 2.91 -0.68 0.46 P 2.96 -0.63 0.39 P 3.07 -0.52 0.27 P 3.12 -0.47 0.22 P 3.14 -0.45 0.20 P 3.15 -0.44 0.19 P 3.22 -0.37 0.13 P 3.23 -0.36 0.13 P 3.25 -0.34 0.11 P 3.29 -0.30 0.09 P 3.3 -0.29 0.08 P 3.31 -0.28 0.08 P 3.41 -0.18 0.03 P 3.61 0.02 0.00 P 3.69 0.10 0.01 P 3.86 0.27 0.07 P 4.1 0.51 0.26 P 4.16 0.57 0.33 P 4.17 0.58 0.34 P 4.19 0.60 0.36 P 4.34 0.75 0.57 P 4.38 0.79 0.63 P 4.44 0.85 0.73 P 4.58 0.99 0.99 P 4.59 1.00 1.01 292 P 4.62 1.03 1.07 P 4.67 1.08 1.17 P 4.7 1.11 1.24 P 4.76 1.17 1.38 P 4.81 1.22 1.50 P 4.91 1.32 1.75 P 4.91 1.32 1.75 P 4.95 1.36 1.86 P 4.95 1.36 1.86 P 5.07 1.48 2.20 P 5.11 1.52 2.32 P 5.27 1.68 2.83 P 6.4 2.81 7.91 P 6.58 2.99 8.96 P 6.78 3.19 10.20 P 6.95 3.36 11.31 P 8.27 4.68 21.93 293 MBSS* NO3 S-PLUS Modeling Coastal Plain Data |PASTUR<20.45 DECIDFOR<20.01 DO.FLD<7.05 WOODYWET<10.14WOODYWET<4.79 COND.LAB<223.7 ANC.LAB<415 WETLANDS<0.045 PH.FLD<6.705 PROBCROP<37.325 BANKSTAB<6.5 1.382 2.492 5.630 3.545 1.021 9.038 3.231 6.528 5.192 1.651 3.728 10.950 % pasture < 20.45 Deciduous forest < 20.01 Cond lab < 223.7 ANC lan < 415 Wetlands < 0.045 bank stability < 6.5 294 Regression tree: tree(formula = NO3.LAB ~ RIP.WIDm + TEMP.FLD + RIP.WID + BUFF.TYP + ADJ.COVR + REGION + AGRI + PHYCPIO + URBAN + FOREST + ORDER + DO.FLD + PH.FLD + COND.FLD + PH.LAB + COND.LAB + ANC.LAB + SO4.LAB + DOC.LAB + ACIDSRC + PASTURE + INSTRHAB + EPI.SUB + VEL.DPTH + POOLQUAL + RIFFQUAL + CHAN.ALT + BANKSTAB + EMBEDDED + CH.FLOW + SHADING + REMOTE + AESTHET + WOOD.DEB + NUMROOT + MAXDEPTH + ST.GRAD + AVGWID + AVGTHAL + AVG.VEL + FLOW + ACREAGE + WETLANDS + BARREN + WATER + HIGHURB + LOWURB + PASTUR + PROBCROP + ROWCROP + crop + CONIFER + DECIDFOR + MIXEDFOR + EMERGWET + WOODYWET + COALMINE + TRANS + BKTRFLAG + BLACKWAT, data = NO3.MBSS.CP, na.action = na.exclude, mincut = 5, minsize = 10, mindev = 0.01) Variables actually used in tree construction: [1] "PASTUR" "DECIDFOR" "DO.FLD" "WOODYWET" "COND.LAB" "ANC.LAB" [7] "WETLANDS" "PH.FLD" "PROBCROP" "BANKSTAB" Number of terminal nodes: 12 Residual mean deviance: 1.829 = 201.2 / 110 Distribution of residuals: Min. 1st Qu. Median Mean 3rd Qu. Max. -8.368000 -0.468800 0.003655 0.000000 0.375200 5.212000 node), split, n, deviance, yval * denotes terminal node 1) root 122 1010.0000 3.413 2) PASTUR<20.45 73 182.2000 2.230 4) DECIDFOR<20.01 44 102.8000 3.027 8) DO.FLD<7.05 23 19.5100 2.010 16) WOODYWET<10.14 10 2.7150 1.382 * 17) WOODYWET>10.14 13 9.8290 2.492 * 9) DO.FLD>7.05 21 33.4000 4.140 18) WOODYWET<4.79 6 6.7170 5.630 * 19) WOODYWET>4.79 15 8.0480 3.545 * 5) DECIDFOR>20.01 29 9.1020 1.021 * 3) PASTUR>20.45 49 573.2000 5.176 6) COND.LAB<223.7 44 271.1000 4.520 12) ANC.LAB<415 30 157.2000 5.512 24) WETLANDS<0.045 5 25.7400 9.038 * 25) WETLANDS>0.045 25 56.8600 4.807 50) PH.FLD<6.705 9 6.6820 3.231 * 51) PH.FLD>6.705 16 15.2600 5.693 102) PROBCROP<37.325 6 7.7620 6.528 * 103) PROBCROP>37.325 10 0.8058 5.192 * 13) ANC.LAB>415 14 21.0600 2.393 26) BANKSTAB<6.5 9 4.1370 1.651 * 27) BANKSTAB>6.5 5 3.0550 3.728 * 7) COND.LAB>223.7 5 116.6000 10.950 * 295 MBSS* NO3 S-PLUS Modeling Piedmont Data |AGRI<72.665 AGRI<57.13 AVGTHAL<19.375SO4.LAB<11.94 WATER<0.55 MIXEDFOR<2.235 COND.FLD<167 PH.LAB<7.425 ADJ.COVR:bgVEL.DPTH<12ROWCROP<10.965 ST.GRAD<0.9 PROBCROP<34.135 EPI.SUB<16.5 DOC.LAB<1.3 HIGHURB<0.02 AGRI<80.5 FLOW<6.425TEMP.FLD<17.6DO.FLD<9.55 1.2892.2451.714 3.782.9364.7971.9202.8352.9853.948 1.8002.827 4.4612.5143.586 5 330 4.0624.9025.8287.2075.672 agri < 72.665 agri < 57.13 probable crops < 34.135 epifaunal substrate < 16.5 agriculture < 80.5 Regression tree: tree(formula = NO3.LAB ~ RIP.WIDm + TEMP.FLD + RIP.WID + BUFF.TYP + ADJ.COVR + REGION + AGRI + URBAN + FOREST + ORDER + DO.FLD + PH.FLD + COND.FLD + PH.LAB + COND.LAB + ANC.LAB + SO4.LAB + DOC.LAB + ACIDSRC + PASTURE + INSTRHAB + EPI.SUB + VEL.DPTH + POOLQUAL + RIFFQUAL + CHAN.ALT + BANKSTAB + EMBEDDED + CH.FLOW + SHADING + REMOTE + AESTHET + WOOD.DEB + NUMROOT + MAXDEPTH + ST.GRAD + AVGWID + AVGTHAL + AVG.VEL + FLOW + ACREAGE + WETLANDS + BARREN + WATER + HIGHURB + LOWURB + PASTUR + PROBCROP + ROWCROP + crop + CONIFER + DECIDFOR + MIXEDFOR + EMERGWET + WOODYWET + COALMINE + TRANS + BKTRFLAG + BLACKWAT, data = NO3.MBSS.P, na.action = na.exclude, mincut = 5, minsize = 10, mindev = 0.01) Variables actually used in tree construction: [1] "AGRI" "AVGTHAL" "SO4.LAB" "WATER" "MIXEDFOR" "COND.FLD" 296 [7] "PH.LAB" "ADJ.COVR" "VEL.DPTH" "ROWCROP" "ST.GRAD" "PROBCROP" [13] "EPI.SUB" "DOC.LAB" "HIGHURB" "FLOW" "TEMP.FLD" "DO.FLD" Number of terminal nodes: 21 Residual mean deviance: 0.4683 = 62.75 / 134 Distribution of residuals: Min. 1st Qu. Median Mean 3rd Qu. Max. -1.3910 -0.3455 0.0240 0.0000 0.3035 3.7400 node), split, n, deviance, yval * denotes terminal node 1) root 155 418.9000 3.587 2) AGRI<72.665 96 104.1000 2.755 4) AGRI<57.13 28 10.6700 1.911 8) AVGTHAL<19.375 7 2.7650 1.289 * 9) AVGTHAL>19.375 21 4.2880 2.119 18) SO4.LAB<11.94 16 2.0240 2.245 * 19) SO4.LAB>11.94 5 1.1900 1.714 * 5) AGRI>57.13 68 65.2200 3.103 10) WATER<0.55 53 41.0600 3.336 20) MIXEDFOR<2.235 23 12.9500 3.907 40) COND.FLD<167 16 4.3270 3.518 80) PH.LAB<7.425 11 1.6540 3.782 * 81) PH.LAB>7.425 5 0.2141 2.936 * 41) COND.FLD>167 7 0.6523 4.797 * 21) MIXEDFOR>2.235 30 14.8500 2.898 42) ADJ.COVR:FR,TG 19 5.0180 2.594 84) VEL.DPTH<12 5 0.9716 1.920 * 85) VEL.DPTH>12 14 0.9624 2.835 * 43) ADJ.COVR:CP,LN,OF,PA 11 5.0510 3.423 86) ROWCROP<10.965 6 1.2190 2.985 * 87) ROWCROP>10.965 5 1.3030 3.948 * 11) WATER>0.55 15 11.1100 2.279 22) ST.GRAD<0.9 8 0.7266 1.800 * 23) ST.GRAD>0.9 7 6.4460 2.827 * 3) AGRI>72.665 59 140.5000 4.940 6) PROBCROP<34.135 30 54.2000 4.193 12) EPI.SUB<16.5 21 19.1000 3.706 24) DOC.LAB<1.3 9 3.7630 4.461 * 25) DOC.LAB>1.3 12 6.3550 3.139 50) HIGHURB<0.02 5 2.3270 2.514 * 51) HIGHURB>0.02 7 0.6774 3.586 * 13) EPI.SUB>16.5 9 18.4800 5.330 * 7) PROBCROP>34.135 29 52.2400 5.712 14) AGRI<80.5 15 12.3500 4.931 28) FLOW<6.425 5 0.9739 4.062 * 29) FLOW>6.425 10 5.7130 5.365 58) TEMP.FLD<17.6 5 0.6591 4.902 * 59) TEMP.FLD>17.6 5 2.9100 5.828 * 15) AGRI>80.5 14 20.9200 6.549 30) DO.FLD<9.55 8 9.0900 7.207 * 31) DO.FLD>9.55 6 3.7410 5.672 * 297 APPENDIX 12: FISH IBI SAS MODELING Model Parameter Search SAS language DATA one ; INFILE 'D:\g MBSS paper\SAS\SAS split data\FIBIcaldata.dat' ; INPUT FIBI RIP_WIDm TEMP_FLD BUFF_FOR CROP REGION_E PHYS_CP ACID_NO AGRI URBAN FOREST ORDER DO_FLD PH_FLD COND_FLD PH_LAB COND_LAB ANC_LAB SO4_LAB DOC_LAB PASTURE INSTRHAB EPI_SUB VEL_DPTH POOLQUAL RIFFQUAL CHAN_ALT BANKSTAB EMBEDDED CH_FLOW SHADING REMOTE AESTHET WOOD_DEB NUMROOT MAXDEPTH ST_GRAD AVGWID AVGTHALAVG_VEL FLOW ACREAGE WETLANDS BARREN WATER HIGHURB LOWURB PASTUR PROBCROP ROWCROP cropmix CONIFER DECIDFOR MIXEDFOR EMERGWET WOODYWET COALMINE TRANS BKTRFLAG BLACKWAT; proc print; * proc corr ; *proc reg ; *model BIBI = RIP_WIDm TEMP_FLD For_Buff CROP Reg_East AGRI PHYS_CP URBAN FOREST ORDER_1 DO_FLD PH_FLD COND_FLD PH_LAB COND_LAB ANC_LAB SO4_LAB DOC_LAB ACID_no PASTURE INSTRHAB EPI_SUB VEL_DPTH POOLQUAL RIFFQUAL CHAN_ALT BANKSTAB EMBEDDED CH_FLOW SHADING REMOTE AESTHET WOOD_DEB NUMROOT MAXDEPTH ST_GRAD AVGWID AVGTHAL AVG_VEL FLOW ACREAGE WETLANDS BARREN WATER HIGHURB LOWURB PASTUR PROBCROP ROWCROP cropmix CONIFER DECIDFOR MIXEDFOR EMERGWET WOODYWET COALMINE TRANS BKTRFLAG / selection=rsquare adjrsq best=3; run; quit; 298 Results Dependent Variable: FIBI R-Square Selection Method Number in Adjusted Model R-Square R-Square Variables in Model 1 0.2420 0.2387 INSTRHAB 1 0.1791 0.1755 BUFF_FOR 1 0.1769 0.1733 SO4_LAB -------------------------------------------------------------------------------------------------- 2 0.3365 0.3308 BUFF_FOR INSTRHAB 2 0.3246 0.3188 PH_LAB INSTRHAB 2 0.3197 0.3138 SO4_LAB INSTRHAB -------------------------------------------------------------------------------------------------- 3 0.4008 0.3930 ORDER PH_LAB INSTRHAB 3 0.3771 0.3689 TEMP_FLD BUFF_FOR INSTRHAB 3 0.3754 0.3672 ORDER PH_LAB VEL_DPTH -------------------------------------------------------------------------------------------------- 4 0.4369 0.4270 ORDER PH_LAB COND_LAB INSTRHAB 4 0.4225 0.4124 ORDER PH_LAB INSTRHAB TRANS 4 0.4213 0.4112 ORDER PH_LAB INSTRHAB REMOTE -------------------------------------------------------------------------------------------------- 5 0.4685 0.4567 ORDER PH_LAB COND_LAB INSTRHAB BLACKWAT 5 0.4615 0.4497 ACID_NO ORDER PH_LAB COND_LAB INSTRHAB 5 0.4557 0.4437 RIP_WIDm ORDER PH_LAB COND_LAB INSTRHAB Model Calibration SAS language * Using the model variables ORDER PH_LAB COND_LAB INSTRHAB BLACKWAT ; *model FIBI = ORDER PH_LAB COND_LAB INSTRHAB BLACKWAT / collinoint tol VIF R influence ; *output out=resFIBI residual = FIBIres; *proc univariate normal plot; *var FIBIres ; 299 Results Dependent Variable: FIBI Analysis of Variance Sum of Mean Source DF Squares Square F Value Pr > F Model 5 72.77296 14.55459 40.01 <.0001 Error 227 82.57319 0.36376 Corrected Total 232 155.34615 Root MSE 0.60312 R-Square 0.4685 Dependent Mean 3.69103 Adj R-Sq 0.4567 Coeff Var 16.34026 Parameter Estimates Parameter Standard Variance Variable DF Estimate Error t Value Pr > |t| Tolerance Inflation Intercept 1 3.29201 0.17654 18.65 <.0001 . 0 ORDER 1 0.31572 0.05269 5.99 <.0001 0.06860 4.57734 PH_LAB 1 -0.08316 0.01092 -7.62 <.0001 0.06566 5.22911 COND_LAB 1 -0.00284 0.000630 -4.50 <.0001 0.91897 .08818 INSTRHAB 1 0.06751 0.01041 6.49 <.0001 0.85680 1.16714 BLACKWAT 1 -0.67212 0.18308 -3.67 0.0003 0.95352 .04875 Variable: FIBIres (Residual) N 233 Sum Weights 233 Mean 0 Sum Observations 0 Std Deviation 0.59658941 Variance 0.35591893 Skewness -0.1911917 Kurtosis -0.2596319 Uncorrected SS 82.5731915 Corrected SS 82.5731915 Coeff Variation . Std Error Mean 0.03908387 Tests for Normality Test --Statistic--- -----p Value------ Shapiro-Wilk W 0.992425 Pr < W 0.2766 300 Model Validation SAS language DATA one ; INFILE 'D:\g MBSS paper\SAS\SAS split data\FIBIvaldata.dat' ; INPUT NO3 RIP_WIDm TEMP_FLD For_Buff CROP Reg_East AGRI PHYS_CP URBAN FOREST ORDER_1 DO_FLD PH_FLD COND_FLD PH_LAB COND_LAB ANC_LAB SO4_LAB DOC_LAB ACID_no PASTURE INSTRHAB EPI_SUB VEL_DPTH POOLQUAL RIFFQUAL CHAN_ALT BANKSTAB EMBEDDED CH_FLOW SHADING REMOTE AESTHET WOOD_DEB NUMROOT MAXDEPTH ST_GRAD AVGWID AVGTHAL AVG_VEL FLOW ACREAGE WETLANDS BARREN WATER HIGHURB LOWURB PASTUR PROBCROP ROWCROP cropmix CONIFER DECIDFOR MIXEDFOR EMERGWET WOODYWET COALMINE TRANS BKTRFLAG BLACKWAT ; FIBIpred = 3.29 +.316*ORDER - .0832*PH_LAB -.0028*COND_LAB + .0675*INSTRHAB - .6721*BLACKWAT ; res = FIBI - FIBIpred ; proc print; Proc corr; run; quit; Results: Correlation Coefficient for FIBI vs FIBIpred = 0.56446 so R2 = 0.32 301 APPENDIX 13: FISH IBI S-PLUS MODELING Regression Tree with all Parameters ? all sites |MAXDEPTH<38.5 INSTRHAB<13.5 ACREAGE<1657.87 FOREST<38.855COND.LAB<125.85 ACREAGE<2235.49 AVGTHAL<37.5 FOREST<47.445 COND.FLD<127 ACREAGE<1570.9ST.GRAD<1.11 ADJ.COVR:df CHAN.ALT<9 ST.GRAD<1.1 HIGHURB<1.795 WETLANDS<0.435 ADJ.COVR:abcd ORDER<2.5 MAXDEPTH<71 SO4.LAB<9.82ROWCROP<1.795BANKSTAB<9.5 FOREST<28.52 COND.FLD<221.5PH.LAB<7.335 HIGHURB<0.55 AVG.VEL<0.08375 TEMP.FLD<18.15PH.LAB<7.255 1.9622.3973.4162.650 3.612 4.250 3.3223.8833.108 3.7324.135 2.9033.444 2.766 3.0853.7463.7003.8884.515 3.6104.0704.5654.056 4.5293.3583.4883.8584.223 4.468 2.866 302 Tree Model with all Parameters tree(formula = FIBI.98 ~ RIP.WIDm + TEMP.FLD + BUFF.TYP + ADJ.COVR + REGION +AGRI + PHYCPIO + URBAN + FOREST + ORDER + DO.FLD + PH.FLD + COND.FLD + PH.LAB + COND.LAB + ANC.LAB + SO4.LAB + DOC.LAB + ACIDSRC + PASTURE + INSTRHAB + EPI.SUB + VEL.DPTH + POOLQUAL + RIFFQUAL + CHAN.ALT + BANKSTAB + EMBEDDED + CH.FLOW + SHADING + REMOTE + AESTHET + WOOD.DEB + NUMROOT + MAXDEPTH + ST.GRAD + AVGWID +AVGTHAL + AVG.VEL + FLOW + ACREAGE + WETLANDS + BARREN + WATER +HIGHURB + LOWURB + PASTUR + PROBCROP + ROWCROP + crop + CONIFER +DECIDFOR + MIXEDFOR + EMERGWET + WOODYWET + COALMINE + TRANS + BKTRFLAG + BLACKWAT, data = FIBI.cal.data, na.action = na.exclude, mincut = 5, minsize = 10, mindev = 0.01) Variables actually used in tree construction: "MAXDEPTH" "INSTRHAB" "ACREAGE" "FOREST" "COND.LAB" "AVGTHAL" "COND.FLD" "ST.GRAD" "ADJ.COVR" "CHAN.ALT" "HIGHURB" "WETLANDS" "ORDER" "SO4.LAB" "ROWCROP" "BANKSTAB" "PH.LAB" "AVG.VEL" "TEMP.FLD" Number of terminal nodes: 30 Residual mean deviance: 0.1284 = 26.06 / 203 node), split, n, deviance, yval * denotes terminal node 1) root 233 128.5000 3.719 2) MAXDEPTH<38.5 31 14.8300 2.767 4) INSTRHAB<13.5 25 9.3160 2.564 8) ACREAGE<1657.87 15 1.3980 2.252 16) FOREST<38.855 5 0.4263 1.962 * 17) FOREST>38.855 10 0.3408 2.397 * 9) ACREAGE>1657.87 10 4.2580 3.033 18) COND.LAB<125.85 5 1.4660 3.416 * 19) COND.LAB>125.85 5 1.3250 2.650 * 5) INSTRHAB>13.5 6 0.2041 3.612 * 3) MAXDEPTH>38.5 202 81.2600 3.865 6) ACREAGE<2235.49 68 26.1500 3.582 12) AVGTHAL<37.5 61 20.3600 3.675 24) FOREST<47.445 50 13.4400 3.791 48) COND.FLD<127 9 2.1190 4.250 * 49) COND.FLD>127 41 9.0060 3.690 98) ACREAGE<1570.9 27 6.1180 3.549 196) ST.GRAD<1.11 19 2.5080 3.735 392) ADJ.COVR:OF,PA 5 0.2491 3.322 * 393) ADJ.COVR:CP,FR,LN 14 1.1000 3.883 * 197) ST.GRAD>1.11 8 1.3910 3.108 * 303 99) ACREAGE>1570.9 14 1.3160 3.962 198) CHAN.ALT<9 6 0.4345 3.732 * 199) CHAN.ALT>9 8 0.3240 4.135 * 25) FOREST>47.445 11 3.2060 3.149 50) ST.GRAD<1.1 6 1.9120 2.903 * 51) ST.GRAD>1.1 5 0.4973 3.444 * 13) AVGTHAL>37.5 7 0.5954 2.766 * 7) ACREAGE>2235.49 134 46.8900 4.008 14) HIGHURB<1.795 129 37.4900 4.053 28) WETLANDS<0.435 88 25.2800 4.149 56) ADJ.COVR:CP,FR,LN,OF 72 21.4100 4.064 112) ORDER<2.5 33 10.7900 3.831 224) MAXDEPTH<71 14 3.7540 3.463 448) SO4.LAB<9.82 6 0.8976 3.085 * 449) SO4.LAB>9.82 8 1.3570 3.746 * 225) MAXDEPTH>71 19 3.7320 4.103 450) ROWCROP<1.795 5 0.3000 3.700 * 451) ROWCROP>1.795 14 2.3320 4.246 902) BANKSTAB<9.5 6 0.3759 3.888 * 903) BANKSTAB>9.5 8 0.6098 4.515 * 113) ORDER>2.5 39 7.3110 4.262 226) FOREST<28.52 11 1.4520 3.861 452) COND.FLD<221.5 5 0.2986 3.610 * 453) COND.FLD>221.5 6 0.5766 4.070 * 227) FOREST>28.52 28 3.3960 4.419 454) PH.LAB<7.335 20 1.1600 4.565 * 455) PH.LAB>7.335 8 0.7602 4.056 * 57) ADJ.COVR:PA,TG 16 1.0430 4.529 * 29) WETLANDS>0.435 41 9.6370 3.846 58) HIGHURB<0.55 35 6.4790 3.739 116) AVG.VEL<0.08375 9 1.2960 3.358 * 117) AVG.VEL>0.08375 26 3.4210 3.871 234) TEMP.FLD<18.15 5 0.5387 3.488 * 235) TEMP.FLD>18.15 21 1.9740 3.962 470) PH.LAB<7.255 15 0.8318 3.858 * 471) PH.LAB>7.255 6 0.5699 4.223 * 59) HIGHURB>0.55 6 0.4351 4.468 * 15) HIGHURB>1.795 5 2.6250 2.866 * 304 Model Validation FIBI_98 Pred FIBI_98 Pred 1.89 4.13 3.89 4.08 2.33 1.96 3.89 4.13 3 3.86 4 3.60 3 3.36 4 4.62 3 3.11 4.11 3.53 3 3.86 4.11 4.16 3.25 3.42 4.25 3.36 3.44 3.28 4.25 4.62 3.67 2.77 4.33 4.62 3.67 3.77 4.75 4.62 3.67 1.96 4.75 4.62 3.75 4.13 4.78 4.72 3.75 2.90 4.78 4.16 Splus model FIBI observed vs predicted R2 = 0.0117 0 1 2 3 4 5 1 1.5 2 2.5 3 3.5 4 FIBI observed FI BI pr ed ic te d 305 Development of equivalent R2 for regression tree model REGION FIBI_98 delta square corrected total SS CP 1.75 -1.94 3.78 CP 67.01 CP 2 -1.69 2.87 Pd 75.65 CP 2 -1.69 2.87 ALL 142.66 CP 2.25 -1.44 2.08 CP 2.25 -1.44 2.08 Resid SS CP 2.25 -1.44 2.08 deviance CP 2.25 -1.44 2.08 CP 4.54 CP 2.5 -1.19 1.43 0.17 CP 2.5 -1.19 1.43 12.34 CP 2.5 -1.19 1.43 0.61 CP 2.5 -1.19 1.43 17.66 CP 2.5 -1.19 1.43 Pd 0.27 CP 2.5 -1.19 1.43 8.67 CP 2.75 -0.94 0.89 0.52 CP 2.75 -0.94 0.89 11.1 CP 2.75 -0.94 0.89 13.1 CP 2.75 -0.94 0.89 1.89 CP 2.75 -0.94 0.89 35.55 CP 3 -0.69 0.48 ALL 9.3 CP 3 -0.69 0.48 0.2 CP 3 -0.69 0.48 20.3 CP 3 -0.69 0.48 0.59 CP 3 -0.69 0.48 37.4 CP 3 -0.69 0.48 2.62 CP 3 -0.69 0.48 70.41 CP 3 -0.69 0.48 CP 3 -0.69 0.48 CP 3 -0.69 0.48 CP 3 -0.69 0.48 CP 3.25 -0.44 0.20 306 CP 3.25 -0.44 0.20 CP 3.25 -0.44 0.20 Res/Cor SS CP 3.25 -0.44 0.20 CP 0.26 CP 3.25 -0.44 0.20 Pd 0.47 CP 3.25 -0.44 0.20 ALL 0.49 CP 3.25 -0.44 0.20 CP 3.25 -0.44 0.20 R2 CP 3.25 -0.44 0.20 CP 0.74 CP 3.5 -0.19 0.04 Pd 0.53 CP 3.5 -0.19 0.04 ALL 0.51 CP 3.5 -0.19 0.04 CP 3.5 -0.19 0.04 CP 3.5 -0.19 0.04 CP 3.5 -0.19 0.04 CP 3.5 -0.19 0.04 CP 3.67 -0.02 0.00 CP 3.75 0.06 0.00 CP 3.75 0.06 0.00 CP 3.75 0.06 0.00 CP 3.75 0.06 0.00 CP 3.75 0.06 0.00 CP 3.75 0.06 0.00 CP 3.75 0.06 0.00 CP 3.75 0.06 0.00 CP 3.75 0.06 0.00 CP 3.75 0.06 0.00 CP 3.75 0.06 0.00 CP 3.75 0.06 0.00 CP 3.75 0.06 0.00 CP 3.75 0.06 0.00 CP 3.75 0.06 0.00 CP 3.75 0.06 0.00 CP 3.75 0.06 0.00 CP 3.75 0.06 0.00 CP 3.75 0.06 0.00 CP 3.89 0.20 0.04 307 CP 4 0.31 0.09 CP 4 0.31 0.09 CP 4 0.31 0.09 CP 4 0.31 0.09 CP 4 0.31 0.09 CP 4 0.31 0.09 CP 4 0.31 0.09 CP 4 0.31 0.09 CP 4 0.31 0.09 CP 4 0.31 0.09 CP 4 0.31 0.09 CP 4 0.31 0.09 CP 4 0.31 0.09 CP 4 0.31 0.09 CP 4.11 0.42 0.17 CP 4.25 0.56 0.31 CP 4.25 0.56 0.31 CP 4.25 0.56 0.31 CP 4.25 0.56 0.31 CP 4.25 0.56 0.31 CP 4.25 0.56 0.31 CP 4.25 0.56 0.31 CP 4.25 0.56 0.31 CP 4.5 0.81 0.65 CP 4.5 0.81 0.65 CP 4.5 0.81 0.65 CP 4.5 0.81 0.65 CP 4.5 0.81 0.65 CP 4.5 0.81 0.65 CP 4.5 0.81 0.65 CP 4.5 0.81 0.65 CP 4.5 0.81 0.65 CP 4.5 0.81 0.65 CP 4.5 0.81 0.65 CP 4.5 0.81 0.65 CP 4.56 0.87 0.75 CP 4.75 1.06 1.12 308 CP 4.75 1.06 1.12 CP 4.75 1.06 1.12 CP 4.75 1.06 1.12 CP 4.75 1.06 1.12 CP 4.75 1.06 1.12 CP 4.75 1.06 1.12 CP 4.75 1.06 1.12 CP 4.75 1.06 1.12 CP 4.75 1.06 1.12 CP 4.75 1.06 1.12 CP 4.75 1.06 1.12 CP 4.75 1.06 1.12 CP 5 1.31 1.71 P 1.67 -2.08 4.32 P 1.89 -1.86 3.45 P 1.89 -1.86 3.45 P 1.89 -1.86 3.45 P 2.11 -1.64 2.68 P 2.11 -1.64 2.68 P 2.11 -1.64 2.68 P 2.33 -1.42 2.01 P 2.33 -1.42 2.01 P 2.33 -1.42 2.01 P 2.56 -1.19 1.41 P 2.56 -1.19 1.41 P 2.56 -1.19 1.41 P 2.78 -0.97 0.94 P 2.78 -0.97 0.94 P 2.78 -0.97 0.94 P 2.78 -0.97 0.94 P 3 -0.75 0.56 P 3 -0.75 0.56 P 3 -0.75 0.56 P 3 -0.75 0.56 P 3 -0.75 0.56 P 3 -0.75 0.56 P 3 -0.75 0.56 309 P 3 -0.75 0.56 P 3 -0.75 0.56 P 3 -0.75 0.56 P 3 -0.75 0.56 P 3.22 -0.53 0.28 P 3.22 -0.53 0.28 P 3.22 -0.53 0.28 P 3.22 -0.53 0.28 P 3.22 -0.53 0.28 P 3.22 -0.53 0.28 P 3.22 -0.53 0.28 P 3.22 -0.53 0.28 P 3.22 -0.53 0.28 P 3.44 -0.31 0.10 P 3.44 -0.31 0.10 P 3.44 -0.31 0.10 P 3.44 -0.31 0.10 P 3.44 -0.31 0.10 P 3.44 -0.31 0.10 P 3.44 -0.31 0.10 P 3.44 -0.31 0.10 P 3.44 -0.31 0.10 P 3.67 -0.08 0.01 P 3.67 -0.08 0.01 P 3.67 -0.08 0.01 P 3.67 -0.08 0.01 P 3.67 -0.08 0.01 P 3.67 -0.08 0.01 P 3.67 -0.08 0.01 P 3.67 -0.08 0.01 P 3.67 -0.08 0.01 P 3.67 -0.08 0.01 P 3.67 -0.08 0.01 P 3.67 -0.08 0.01 P 3.67 -0.08 0.01 P 3.67 -0.08 0.01 P 3.67 -0.08 0.01 310 P 3.67 -0.08 0.01 P 3.89 0.14 0.02 P 3.89 0.14 0.02 P 3.89 0.14 0.02 P 3.89 0.14 0.02 P 3.89 0.14 0.02 P 3.89 0.14 0.02 P 3.89 0.14 0.02 P 3.89 0.14 0.02 P 3.89 0.14 0.02 P 3.89 0.14 0.02 P 3.89 0.14 0.02 P 3.89 0.14 0.02 P 3.89 0.14 0.02 P 3.89 0.14 0.02 P 3.89 0.14 0.02 P 3.89 0.14 0.02 P 3.89 0.14 0.02 P 3.89 0.14 0.02 P 3.89 0.14 0.02 P 3.89 0.14 0.02 P 3.89 0.14 0.02 P 3.89 0.14 0.02 P 3.89 0.14 0.02 P 4.11 0.36 0.13 P 4.11 0.36 0.13 P 4.11 0.36 0.13 P 4.11 0.36 0.13 P 4.11 0.36 0.13 P 4.11 0.36 0.13 P 4.11 0.36 0.13 P 4.11 0.36 0.13 P 4.11 0.36 0.13 P 4.11 0.36 0.13 P 4.11 0.36 0.13 P 4.11 0.36 0.13 P 4.11 0.36 0.13 311 P 4.11 0.36 0.13 P 4.33 0.58 0.34 P 4.33 0.58 0.34 P 4.33 0.58 0.34 P 4.33 0.58 0.34 P 4.33 0.58 0.34 P 4.33 0.58 0.34 P 4.33 0.58 0.34 P 4.33 0.58 0.34 P 4.33 0.58 0.34 P 4.33 0.58 0.34 P 4.33 0.58 0.34 P 4.33 0.58 0.34 P 4.33 0.58 0.34 P 4.33 0.58 0.34 P 4.33 0.58 0.34 P 4.33 0.58 0.34 P 4.33 0.58 0.34 P 4.33 0.58 0.34 P 4.33 0.58 0.34 P 4.33 0.58 0.34 P 4.56 0.81 0.66 P 4.56 0.81 0.66 P 4.56 0.81 0.66 P 4.56 0.81 0.66 P 4.56 0.81 0.66 P 4.56 0.81 0.66 P 4.56 0.81 0.66 P 4.56 0.81 0.66 P 4.56 0.81 0.66 P 4.56 0.81 0.66 P 4.56 0.81 0.66 P 4.56 0.81 0.66 P 4.56 0.81 0.66 P 4.78 1.03 1.06 P 4.78 1.03 1.06 P 4.78 1.03 1.06 312 P 4.78 1.03 1.06 P 4.78 1.03 1.06 P 4.78 1.03 1.06 P 4.78 1.03 1.06 P 4.78 1.03 1.06 P 5 1.25 1.57 P 5 1.25 1.57 mean CP 3.69 P 3.75 ALL 3.72 313 FIBI Regression Tree Model Coastal Plain Data |MAXDEPTH<38.5 INSTRHAB<8.5 SO4.LAB<14.43 AVGWID<1.615 PASTUR<13.145 EPI.SUB<13.5 TEMP.FLD<20.5 WATER<0.145 ACREAGE<12467.4 URBAN<1.795 TEMP.FLD<18.85 RIP.WIDm<22.5 AESTHET<15.5 EPI.SUB<10.5 TEMP.FLD<23.4COND.FLD<162 2.4503.0832.143 3.500 2.9693.8333.321 .200 4.159 4.0423.1673.694 4.650 3.6344.0784.350 4.592 Max depth < 38.5 Instream habitat < 8.5 % pasture < 13.145 Epifaunal substrate < 13.5 acreage < 12467.4 314 Regression tree: tree(formula = FIBI.98 ~ RIP.WIDm + TEMP.FLD + BUFF.TYP + ADJ.COVR + REGION + AGRI + URBAN + FOREST + ORDER + DO.FLD + PH.FLD + COND.FLD + PH.LAB + COND.LAB + ANC.LAB + SO4.LAB + DOC.LAB + ACIDSRC + PASTURE + INSTRHAB + EPI.SUB + VEL.DPTH + POOLQUAL + RIFFQUAL + CHAN.ALT + BANKSTAB + EMBEDDED + CH.FLOW + SHADING + REMOTE + AESTHET + WOOD.DEB + NUMROOT + MAXDEPTH + ST.GRAD + AVGWID + AVGTHAL + AVG.VEL + FLOW + ACREAGE + WETLANDS + BARREN + WATER + HIGHURB + LOWURB + PASTUR + PROBCROP + ROWCROP + crop + CONIFER + DECIDFOR + MIXEDFOR + EMERGWET + WOODYWET + COALMINE + TRANS + BKTRFLAG + BLACKWAT, data = FIBI.CP, na.action = na.exclude, mincut = 5, minsize = 10, mindev = 0.01) Variables actually used in tree construction: [1] "MAXDEPTH" "INSTRHAB" "SO4.LAB" "AVGWID" "PASTUR" "EPI.SUB" [7] "TEMP.FLD" "WATER" "ACREAGE" "URBAN" "RIP.WIDm" "AESTHET" [13] "COND.FLD" Number of terminal nodes: 17 Residual mean deviance: 0.08351 = 8.268 / 99 Distribution of residuals: Min. 1st Qu. Median Mean 3rd Qu. Max. -0.63400 -0.17200 -0.01588 0.00000 0.15820 0.78130 node), split, n, deviance, yval * denotes terminal node 1) root 116 67.0100 3.694 2) MAXDEPTH<38.5 25 9.1600 2.810 4) INSTRHAB<8.5 18 4.1560 2.542 8) SO4.LAB<14.43 11 1.9770 2.795 16) AVGWID<1.615 5 0.0500 2.450 * 17) AVGWID>1.615 6 0.8333 3.083 * 9) SO4.LAB>14.43 7 0.3571 2.143 * 5) INSTRHAB>8.5 7 0.3750 3.500 * 3) MAXDEPTH>38.5 91 32.9600 3.937 6) PASTUR<13.145 26 7.7500 3.500 12) EPI.SUB<13.5 21 4.5420 3.333 24) TEMP.FLD<20.5 8 0.9297 2.969 * 25) TEMP.FLD>20.5 13 1.8940 3.558 50) WATER<0.145 6 0.3333 3.833 * 51) WATER>0.145 7 0.7143 3.321 * 13) EPI.SUB>13.5 5 0.1750 4.200 * 7) PASTUR>13.145 65 18.2700 4.111 14) ACREAGE<12467.4 48 12.3400 3.941 28) URBAN<1.795 28 6.1790 3.772 56) TEMP.FLD<18.85 7 0.8361 4.159 * 57) TEMP.FLD>18.85 21 3.9460 3.643 114) RIP.WIDm<22.5 6 0.8021 4.042 * 115) RIP.WIDm>22.5 15 1.8080 3.483 230) AESTHET<15.5 6 0.2083 3.167 * 231) AESTHET>15.5 9 0.5972 3.694 * 29) URBAN>1.795 20 4.2370 4.178 58) EPI.SUB<10.5 5 0.0750 4.650 * 59) EPI.SUB>10.5 15 2.6770 4.021 118) TEMP.FLD<23.4 10 1.2890 3.856 236) COND.FLD<162 5 0.5641 3.634 * 237) COND.FLD>162 5 0.2317 4.078 * 119) TEMP.FLD>23.4 5 0.5750 4.350 * 15) ACREAGE>12467.4 17 0.6104 4.592 * 315 FIBI Regression Tree Model Piedmont Data |VEL.DPTH<10.5 ROWCROP<10.86 INSTRHAB<14.5 ST.GRAD<0.95 ANC.LAB<993.35 MAXDEPTH<76.5 ANC.LAB<279.2 AGRI<61.52 DECIDFOR<23.565 DO.FLD<9.15 SHADING<65 ACREAGE<1422.74 WETLANDS<0.435 ANC.LAB<435.1 INSTRHAB<14.5crop<39.685 PH.LAB<7.26WETLANDS<0.58 2.222 3.3712.447 4.073 3.253 3.477 4.0713.4423.889 4.244 3.490 4.2584.6974.3323.813 3.5783.9634.271 2.901 Vel depth < 10.5 Row crop < 10.86 ANC lab < 993.35 Instream habitat < 14.5 Max depth < 76.5 316 Regression tree: tree(formula = FIBI.98 ~ RIP.WIDm + TEMP.FLD + BUFF.TYP + ADJ.COVR + REGION + AGRI + URBAN + FOREST + ORDER + DO.FLD + PH.FLD + COND.FLD + PH.LAB + COND.LAB + ANC.LAB + SO4.LAB + DOC.LAB + ACIDSRC + PASTURE + INSTRHAB + EPI.SUB + VEL.DPTH + POOLQUAL + RIFFQUAL + CHAN.ALT + BANKSTAB + EMBEDDED + CH.FLOW + SHADING + REMOTE + AESTHET + WOOD.DEB + NUMROOT + MAXDEPTH + ST.GRAD + AVGWID + AVGTHAL + AVG.VEL + FLOW + ACREAGE + WETLANDS + BARREN + WATER + HIGHURB + LOWURB + PASTUR + PROBCROP + ROWCROP + crop + CONIFER + DECIDFOR + MIXEDFOR + EMERGWET + WOODYWET + COALMINE + TRANS + BKTRFLAG + BLACKWAT, data = FIBI.P, na.action = na.exclude, mincut = 5, minsize = 10, mindev = 0.01) Variables actually used in tree construction: [1] "VEL.DPTH" "ROWCROP" "INSTRHAB" "ST.GRAD" "ANC.LAB" "MAXDEPTH" [7] "AGRI" "DECIDFOR" "DO.FLD" "SHADING" "ACREAGE" "WETLANDS" [13] "crop" "PH.LAB" Number of terminal nodes: 19 Residual mean deviance: 0.1305 = 16.05 / 123 Distribution of residuals: Min. 1st Qu. Median Mean 3rd Qu. Max. -0.81110 -0.18090 0.03909 0.00000 0.22140 1.22300 node), split, n, deviance, yval * denotes terminal node 1) root 142 75.6500 3.748 2) VEL.DPTH<10.5 27 19.9000 3.066 4) ROWCROP<10.86 6 0.2729 2.222 * 5) ROWCROP>10.86 21 14.1200 3.308 10) INSTRHAB<14.5 15 8.6730 3.001 20) ST.GRAD<0.95 9 2.6550 3.371 * 21) ST.GRAD>0.95 6 2.9420 2.447 * 11) INSTRHAB>14.5 6 0.5243 4.073 * 3) VEL.DPTH>10.5 115 40.2500 3.908 6) ANC.LAB<993.35 106 28.4500 3.994 12) MAXDEPTH<76.5 50 11.1000 3.782 24) ANC.LAB<279.2 7 1.1230 3.253 * 25) ANC.LAB>279.2 43 7.6980 3.868 50) AGRI<61.52 7 1.0340 3.477 * 51) AGRI>61.52 36 5.3880 3.944 102) DECIDFOR<23.565 26 3.6250 3.828 204) DO.FLD<9.15 11 0.6697 4.071 * 205) DO.FLD>9.15 15 1.8340 3.651 410) SHADING<65 8 0.4973 3.442 * 411) SHADING>65 7 0.5941 3.889 * 103) DECIDFOR>23.565 10 0.5158 4.244 * 13) MAXDEPTH>76.5 56 13.1000 4.183 26) ACREAGE<1422.74 5 0.5378 3.490 * 27) ACREAGE>1422.74 51 9.9200 4.251 54) WETLANDS<0.435 33 5.4180 4.401 108) ANC.LAB<435.1 22 1.9740 4.577 216) INSTRHAB<14.5 6 0.6557 4.258 * 217) INSTRHAB>14.5 16 0.4791 4.697 * 109) ANC.LAB>435.1 11 1.3980 4.049 218) crop<39.685 5 0.1013 4.332 * 219) crop>39.685 6 0.5633 3.813 * 55) WETLANDS>0.435 18 2.3970 3.976 110) PH.LAB<7.26 11 0.8278 3.788 220) WETLANDS<0.58 5 0.2615 3.578 * 221) WETLANDS>0.58 6 0.1613 3.963 * 111) PH.LAB>7.26 7 0.5703 4.271 * 7) ANC.LAB>993.35 9 1.8920 2.901 * 317 APPENDIX 14: BENTHIC IBI SAS MODELING Model Parameter Search SAS language DATA one ; INFILE 'D:\g MBSS paper\SAS\SAS split data\BIBIcaldata.dat' ; INPUT BIBI RIP_WIDm TEMP_FLD BUFFER F0R_BUF CROP REG_EAST PHYS_CP ACID_no AGRI URBAN FOREST ORDER DO_FLD PH_FLD COND_FLD PH_LAB COND_LAB ANC_LAB SO4_LAB DOC_LAB PASTURE INSTRHAB EPI_SUB VEL_DPTH POOLQUAL RIFFQUAL CHAN_ALT BANKSTAB EMBEDDED CH_FLOW SHADING REMOTE AESTHET WOOD_DEB NUMROOT MAXDEPTH ST_GRAD AVGWID AVGTHAL AVG_VEL FLOW ACREAGE WETLANDS BARREN WATER HIGHURB LOWURB PASTUR PROBCROP ROWCROP cropmix CONIFER DECIDFOR MIXEDFOR EMERGWET WOODYWET COALMINE TRANS BKTRFLAG BLACKWAT ; *proc print; * proc corr ; proc reg ; *model BIBI = RIP_WIDm TEMP_FLD For_Buff CROP Reg_East AGRI PHYS_CP URBAN FOREST ORDER_1 DO_FLD PH_FLD COND_FLD PH_LAB COND_LAB ANC_LAB SO4_LAB DOC_LAB ACID_no PASTURE INSTRHAB EPI_SUB VEL_DPTH POOLQUAL RIFFQUAL CHAN_ALT BANKSTAB EMBEDDED CH_FLOW SHADING REMOTE AESTHET WOOD_DEB NUMROOT MAXDEPTH ST_GRAD AVGWID AVGTHAL AVG_VEL FLOW ACREAGE WETLANDS BARREN WATER HIGHURB LOWURB PASTUR PROBCROP ROWCROP cropmix CONIFER DECIDFOR MIXEDFOR EMERGWET WOODYWET COALMINE TRANS BKTRFLAG BLACKWAT / selection=rsquare adjrsq best=3; run; quit; 318 Results R-Square Selection Method Number in Adjusted Model R2 R-Square Variables in Model 1 0.1440 0.1405 RIP_WIDm 1 0.1371 0.1336 INSTRHAB 1 0.1311 0.1275 SO4_LAB -------------------------------------------------------------------------------------------------- 2 0.2583 0.2522 SO4_LAB EMBEDDED 2 0.2488 0.2426 BUFFER EMBEDDED 2 0.2408 0.2345 RIP_WIDm SO4_LAB -------------------------------------------------------------------------------------------------- 3 0.3433 0.3352 RIP_WIDm BUFFER EMBEDDED 3 0.3426 0.3345 RIP_WIDm SO4_LAB EMBEDDED 3 0.3270 0.3187 BUFFER F0R_BUF EMBEDDED -------------------------------------------------------------------------------------------------- 4 0.3865 0.3764 RIP_WIDm PH_LAB SO4_LAB EMBEDDED 4 0.3701 0.3597 RIP_WIDm BUFFER URBAN EMBEDDED 4 0.3693 0.3589 RIP_WIDm ORDER SO4_LAB EMBEDDED -------------------------------------------------------------------------------------------------- 5 0.4120 0.3998 RIP_WIDm PH_LAB SO4_LAB EMBEDDED WOOD_DEB 5 0.3996 0.3871 RIP_WIDm BUFFER URBAN EMBEDDED WOOD_DEB 5 0.3991 0.3866 RIP_WIDm PH_LAB SO4_LAB EMBEDDED NUMROOT -------------------------------------------------------------------------------------------------- Model Calibration SAS language *Using the model variables RIP_WIDm PH_LAB SO4_LAB EMBEDDED WOOD_DEB ; model BIBI = RIP_WIDm PH_LAB SO4_LAB EMBEDDED WOOD_DEB / collinoint tol VIF R influence ; output out=resBIBI residual = BIBIres; proc univariate normal plot; var BIBIres ; 319 Results Dependent Variable: BIBI Analysis of Variance Sum of Mean Source DF Squares Square F Value Pr > F Model 5 87.41057 17.48211 33.77 <.0001 Error 241 124.74992 0.51763 Corrected Total 246 212.16049 Root MSE 0.71947 R-Square 0.4120 Dependent Mean 3.02745 Adj R-Sq 0.3998 Coeff Var 23.76483 Parameter Estimates Parameter Standard Variable DF Estimate Error t Value Pr > |t| Tolerance Intercept 1 3.08441 0.12524 24.63 <.0001 . RIP_WIDm 1 0.01139 0.00223 5.10 <.0001 0.92659 PH_LAB 1 0.03017 0.00641 4.71 <.0001 0.08485 SO4_LAB 1 -0.02156 0.00322 -6.71 <.0001 0.08385 EMBEDDED 1 -0.00879 0.00144 -6.11 <.0001 0.94668 WOOD_DEB 1 0.03095 0.00958 3.23 0.0014 0.90725 Variable: BIBIres (Residual) N 247 Sum Weights 247 Mean 0 Sum Observations 0 Std Deviation 0.71211902 Variance 0.5071135 Skewness -0.4123476 Kurtosis -0.0985222 Uncorrected SS 124.74992 Corrected SS 124.74992 Coeff Variation . Std Error Mean 0.04531105 Tests for Normality Test --Statistic--- -----p Value------ Shapiro-Wilk W 0.985269 Pr < W 0.0119 320 Model Validation SAS language DATA one ; INFILE 'D:\g MBSS paper\SAS\SAS split data\BIBIvaldata.dat' ; INPUT BIBI RIP_WIDm TEMP_FLD For_Buff CROP Reg_East AGRI PHYS_CP URBAN FOREST ORDER_1 DO_FLD PH_FLD COND_FLD PH_LAB COND_LAB ANC_LAB SO4_LAB DOC_LAB ACID_no PASTURE INSTRHAB EPI_SUB VEL_DPTH POOLQUAL RIFFQUAL CHAN_ALT BANKSTAB EMBEDDED CH_FLOW SHADING REMOTE AESTHET WOOD_DEB NUMROOT MAXDEPTH ST_GRAD AVGWID AVGTHAL AVG_VEL FLOW ACREAGE WETLANDS BARREN WATER HIGHURB LOWURB PASTUR PROBCROP ROWCROP cropmix CONIFER DECIDFOR MIXEDFOR EMERGWET WOODYWET COALMINE TRANS BKTRFLAG BLACKWAT ; BIBIpred = 3.084 + 0.0114*RIP_WIDm + 0.0302*PH_LAB - .022*SO4_LAB - .0088*EMBEDDED +.0309*WOOD_DEB ; res = BIBI - BIBIpred ; proc print; Proc corr; run; quit; Results: Correlation Coefficients for BIBI vs BIBIpred = 0.312 so R2 = 0.097 321 APPENDIX 15: BENTHIC IBI S-PLUS MODELING Regression tree with all parameters |EMBEDDED<85 WOOD.DEB<0.5 SO4.LAB<10.92 crop<48.9 AVGWID<3.965 INSTRHAB<11.5CH.FLOW<65 BANKSTAB<3.5 FLOW<8.805 PH.LAB<7.435 PASTUR<32.425 crop<52.975INSTRHAB<10.5EMBEDDED<12.5NUMROOT<3.5WOOD.DEB<6.5 POOLQUAL<14.5INSTRHAB<13.5CHAN.ALT<10.5 COND.LAB<178.65 WOODYWET<1.35 DO.FLD<8.6 TEMP.FLD<21.05 RIP.WIDm<37.5 ORDER<1.5 WOOD.DEB<3WATER<0.155 AVGTHAL<39.875 AVGWID<2.59 COND.FLD<238 WATER<0.205 3.336.9782.729 2.8301.7442. 061.952 3.5374.2303.9204.2913.714 3.0463.7102.9373.9583.014 3.420.3742.6673.5573.073 2.0 6 1.5332.1432.9982.4102.003 2.226 2.7143.3813.914 322 Regression Tree Model with all Parameters tree(formula = BIBI.98 ~ RIP.WIDm + TEMP.FLD + BUFF.TYP + ADJ.COVR + REGION + AGRI + PHYCPIO + URBAN + FOREST + ORDER + DO.FLD + PH.FLD + COND.FLD + PH.LAB + COND.LAB + ANC.LAB + SO4.LAB + DOC.LAB + ACIDSRC + PASTURE + INSTRHAB + EPI.SUB + VEL.DPTH + POOLQUAL + RIFFQUAL + CHAN.ALT + BANKSTAB + EMBEDDED + CH.FLOW + SHADING + REMOTE + AESTHET + WOOD.DEB + NUMROOT + MAXDEPTH + ST.GRAD + AVGWID + AVGTHAL + AVG.VEL + FLOW + ACREAGE + WETLANDS + BARREN + WATER + HIGHURB + LOWURB + PASTUR + PROBCROP + ROWCROP + crop + CONIFER + DECIDFOR + MIXEDFOR + EMERGWET + WOODYWET + COALMINE + TRANS + BKTRFLAG + BLACKWAT, data = BIBI.cal.data, na.action = na.exclude, mincut = 5, minsize = 10, mindev = 0.01) Variables actually used in tree construction: "EMBEDDED" "WOOD.DEB" "SO4.LAB" "crop" "AVGWID" "INSTRHAB" "CH.FLOW" "BANKSTAB" "FLOW" "PH.LAB" "PASTUR" "NUMROOT" "POOLQUAL" "CHAN.ALT" "COND.LAB" "WOODYWET" "DO.FLD" "TEMP.FLD" "RIP.WIDm" "ORDER" "WATER" "AVGTHAL" "COND.FLD" Number of terminal nodes: 32 Residual mean deviance: 0.1751 = 37.64 / 215 node), split, n, deviance, yval * denotes terminal node 1) root 247 187.5000 3.076 2) EMBEDDED<85 179 102.9000 3.309 4) WOOD.DEB<0.5 36 21.6300 2.801 8) SO4.LAB<10.92 19 8.4090 3.217 16) crop<48.9 10 2.2880 3.657 32) AVGWID<3.965 5 0.4997 3.336 * 33) AVGWID>3.965 5 0.7579 3.978 * 17) crop>48.9 9 2.0400 2.729 * 9) SO4.LAB>10.92 17 6.2460 2.336 18) INSTRHAB<11.5 5 1.8890 2.830 * 19) INSTRHAB>11.5 12 2.6270 2.130 38) CH.FLOW<65 5 0.7507 1.744 * 39) CH.FLOW>65 7 0.5994 2.406 * 5) WOOD.DEB>0.5 143 69.6400 3.436 10) BANKSTAB<3.5 5 2.4230 1.952 * 11) BANKSTAB>3.5 138 55.8000 3.490 22) FLOW<8.805 101 34.4800 3.624 44) PH.LAB<7.435 87 22.0900 3.716 88) PASTUR<32.425 70 13.7800 3.813 176) crop<52.975 60 9.0160 3.885 352) INSTRHAB<10.5 12 1.6930 3.537 * 323 353) INSTRHAB>10.5 48 5.5010 3.972 706) EMBEDDED<12.5 9 1.0340 4.230 * 707) EMBEDDED>12.5 39 3.7310 3.913 1414) NUMROOT<3.5 25 2.3260 4.024 2828) WOOD.DEB<6.5 18 1.1840 3.920 * 2829) WOOD.DEB>6.5 7 0.4471 4.291 * 1415) NUMROOT>3.5 14 0.5443 3.714 * 177) crop>52.975 10 2.5620 3.378 354) POOLQUAL<14.5 5 1.2170 3.046 * 355) POOLQUAL>14.5 5 0.2430 3.710 * 89) PASTUR>32.425 17 4.9890 3.320 178) INSTRHAB<13.5 6 0.6057 2.937 * 179) INSTRHAB>13.5 11 3.0200 3.529 358) CHAN.ALT<10.5 6 0.2633 3.958 * 359) CHAN.ALT>10.5 5 0.3251 3.014 * 45) PH.LAB>7.435 14 6.9770 3.046 90) COND.LAB<178.65 9 2.1090 3.420 * 91) COND.LAB>178.65 5 1.3510 2.374 * 23) FLOW>8.805 37 14.6000 3.126 46) WOODYWET<1.35 32 7.2250 3.299 92) DO.FLD<8.6 6 1.4500 2.667 * 93) DO.FLD>8.6 26 2.8210 3.445 186) TEMP.FLD<21.05 20 1.2920 3.557 * 187) TEMP.FLD>21.05 6 0.4517 3.073 * 47) WOODYWET>1.35 5 0.2603 2.016 * 3) EMBEDDED>85 68 49.4700 2.463 6) RIP.WIDm<37.5 41 16.8500 2.065 12) ORDER<1.5 22 5.6150 1.755 24) WOOD.DEB<3 14 1.7680 1.533 * 25) WOOD.DEB>3 8 1.9550 2.143 * 13) ORDER>1.5 19 6.6650 2.424 26) WATER<0.155 11 2.5940 2.731 52) AVGTHAL<39.875 6 0.9919 2.998 * 53) AVGTHAL>39.875 5 0.6578 2.410 * 27) WATER>0.155 8 1.6140 2.003 * 7) RIP.WIDm>37.5 27 16.2300 3.068 14) AVGWID<2.59 5 1.2710 2.226 * 15) AVGWID>2.59 22 10.6100 3.260 30) COND.FLD<238 17 7.6100 3.067 60) WATER<0.205 8 4.0780 2.714 * 61) WATER>0.205 9 1.6450 3.381 * 31) COND.FLD>238 5 0.2263 3.914 * 324 Model Validation Pred Obs Pred Obs 2.1425 1.29 2.405714 3.29 2.405714 1.57 3.536667 3.44 3.5565 2.11 3.5565 3.44 2.016 2.11 2.728889 3.44 3.5565 2.11 3.914 3.57 3.92 2.14 4.23 3.67 3.714286 2.56 3.92 3.67 3.5565 2.56 1.952 3.89 2.016 2.78 2.374 4.11 3.5565 2.78 2.71375 4.14 2.1425 3 3.5565 4.33 2.016 3 2.226 4.43 2.71375 3 3.92 4.56 3.92 3.22 Splus BIBI model observed vs predicted R2 = 0.0147 0 1 2 3 4 1 2 3 4 5 BIBI observed BI BI p re di ct ed 325 Development of equivalent R2 for regression tree model REGION BIBI delta square corrected total SS CP 1 -1.83 3.36 115.08 CP CP 1 -1.83 3.36 78.93 Pd CP 1.29 -1.54 2.38 194.01 ALL CP 1.29 -1.54 2.38 CP 1.29 -1.54 2.38 Resid SS CP 1.29 -1.54 2.38 deviance CP 1.29 -1.54 2.38 CP 6.67 CP 1.29 -1.54 2.38 0.72 CP 1.29 -1.54 2.38 3.75 CP 1.29 -1.54 2.38 5.69 CP 1.57 -1.26 1.59 CP 1.57 -1.26 1.59 6.62 CP 1.57 -1.26 1.59 30.73 CP 1.57 -1.26 1.59 54.18 CP 1.57 -1.26 1.59 Pd 25.75 CP 1.57 -1.26 1.59 0.59 CP 1.57 -1.26 1.59 3.54 CP 1.57 -1.26 1.59 9.76 CP 1.86 -0.97 0.95 39.64 CP 1.86 -0.97 0.95 ALL 8.24 CP 1.86 -0.97 0.95 6.24 CP 1.86 -0.97 0.95 2.4 CP 1.86 -0.97 0.95 55.8 CP 1.86 -0.97 0.95 16.85 CP 1.86 -0.97 0.95 16.23 CP 1.86 -0.97 0.95 105.76 CP 1.86 -0.97 0.95 CP 1.86 -0.97 0.95 Res/Cor SS CP 1.86 -0.97 0.95 CP 0.47 CP 1.86 -0.97 0.95 Pd 0.50 CP 1.86 -0.97 0.95 ALL 0.55 CP 1.86 -0.97 0.95 326 CP 1.86 -0.97 0.95 R2 CP 1.86 -0.97 0.95 CP 0.53 CP 2.14 -0.69 0.48 Pd 0.50 CP 2.14 -0.69 0.48 ALL 0.45 CP 2.14 -0.69 0.48 CP 2.14 -0.69 0.48 CP 2.14 -0.69 0.48 CP 2.14 -0.69 0.48 CP 2.14 -0.69 0.48 CP 2.14 -0.69 0.48 CP 2.14 -0.69 0.48 CP 2.43 -0.40 0.16 CP 2.43 -0.40 0.16 CP 2.43 -0.40 0.16 CP 2.43 -0.40 0.16 CP 2.43 -0.40 0.16 CP 2.43 -0.40 0.16 CP 2.71 -0.12 0.02 CP 2.71 -0.12 0.02 CP 2.71 -0.12 0.02 CP 2.71 -0.12 0.02 CP 2.71 -0.12 0.02 CP 2.71 -0.12 0.02 CP 2.71 -0.12 0.02 CP 2.71 -0.12 0.02 CP 2.71 -0.12 0.02 CP 2.71 -0.12 0.02 CP 3 0.17 0.03 CP 3 0.17 0.03 CP 3 0.17 0.03 CP 3 0.17 0.03 CP 3 0.17 0.03 CP 3 0.17 0.03 CP 3 0.17 0.03 CP 3 0.17 0.03 CP 3 0.17 0.03 CP 3 0.17 0.03 327 CP 3 0.17 0.03 CP 3 0.17 0.03 CP 3 0.17 0.03 CP 3.29 0.46 0.21 CP 3.29 0.46 0.21 CP 3.29 0.46 0.21 CP 3.29 0.46 0.21 CP 3.29 0.46 0.21 CP 3.57 0.74 0.54 CP 3.57 0.74 0.54 CP 3.57 0.74 0.54 CP 3.57 0.74 0.54 CP 3.57 0.74 0.54 CP 3.57 0.74 0.54 CP 3.57 0.74 0.54 CP 3.57 0.74 0.54 CP 3.57 0.74 0.54 CP 3.57 0.74 0.54 CP 3.57 0.74 0.54 CP 3.57 0.74 0.54 CP 3.57 0.74 0.54 CP 3.57 0.74 0.54 CP 3.86 1.03 1.06 CP 3.86 1.03 1.06 CP 3.86 1.03 1.06 CP 3.86 1.03 1.06 CP 3.86 1.03 1.06 CP 3.86 1.03 1.06 CP 3.86 1.03 1.06 CP 3.86 1.03 1.06 CP 3.86 1.03 1.06 CP 3.86 1.03 1.06 CP 3.86 1.03 1.06 CP 3.86 1.03 1.06 CP 3.86 1.03 1.06 CP 3.86 1.03 1.06 CP 3.86 1.03 1.06 328 CP 3.86 1.03 1.06 CP 4.14 1.31 1.71 CP 4.14 1.31 1.71 CP 4.14 1.31 1.71 CP 4.14 1.31 1.71 CP 4.14 1.31 1.71 CP 4.14 1.31 1.71 CP 4.14 1.31 1.71 CP 4.14 1.31 1.71 CP 4.14 1.31 1.71 CP 4.43 1.60 2.55 CP 4.43 1.60 2.55 CP 4.43 1.60 2.55 CP 4.71 1.88 3.52 CP 4.71 1.88 3.52 P 1 -2.28 5.18 P 1.44 -1.84 3.37 P 1.67 -1.61 2.58 P 1.67 -1.61 2.58 P 1.67 -1.61 2.58 P 1.67 -1.61 2.58 P 1.89 -1.39 1.92 P 1.89 -1.39 1.92 P 1.89 -1.39 1.92 P 2.11 -1.17 1.36 P 2.11 -1.17 1.36 P 2.11 -1.17 1.36 P 2.11 -1.17 1.36 P 2.11 -1.17 1.36 P 2.33 -0.95 0.89 P 2.33 -0.95 0.89 P 2.33 -0.95 0.89 P 2.33 -0.95 0.89 P 2.33 -0.95 0.89 P 2.33 -0.95 0.89 P 2.56 -0.72 0.51 P 2.56 -0.72 0.51 329 P 2.56 -0.72 0.51 P 2.56 -0.72 0.51 P 2.56 -0.72 0.51 P 2.56 -0.72 0.51 P 2.56 -0.72 0.51 P 2.56 -0.72 0.51 P 2.56 -0.72 0.51 P 2.56 -0.72 0.51 P 2.56 -0.72 0.51 P 2.56 -0.72 0.51 P 2.56 -0.72 0.51 P 2.78 -0.50 0.25 P 2.78 -0.50 0.25 P 2.78 -0.50 0.25 P 2.78 -0.50 0.25 P 2.78 -0.50 0.25 P 2.78 -0.50 0.25 P 2.78 -0.50 0.25 P 3 -0.28 0.08 P 3 -0.28 0.08 P 3 -0.28 0.08 P 3 -0.28 0.08 P 3 -0.28 0.08 P 3 -0.28 0.08 P 3 -0.28 0.08 P 3 -0.28 0.08 P 3 -0.28 0.08 P 3 -0.28 0.08 P 3 -0.28 0.08 P 3 -0.28 0.08 P 3 -0.28 0.08 P 3 -0.28 0.08 P 3 -0.28 0.08 P 3 -0.28 0.08 P 3.22 -0.06 0.00 P 3.22 -0.06 0.00 P 3.22 -0.06 0.00 330 P 3.22 -0.06 0.00 P 3.22 -0.06 0.00 P 3.22 -0.06 0.00 P 3.22 -0.06 0.00 P 3.22 -0.06 0.00 P 3.22 -0.06 0.00 P 3.22 -0.06 0.00 P 3.22 -0.06 0.00 P 3.22 -0.06 0.00 P 3.22 -0.06 0.00 P 3.22 -0.06 0.00 P 3.22 -0.06 0.00 P 3.44 0.16 0.03 P 3.44 0.16 0.03 P 3.44 0.16 0.03 P 3.44 0.16 0.03 P 3.44 0.16 0.03 P 3.44 0.16 0.03 P 3.44 0.16 0.03 P 3.44 0.16 0.03 P 3.44 0.16 0.03 P 3.44 0.16 0.03 P 3.44 0.16 0.03 P 3.44 0.16 0.03 P 3.44 0.16 0.03 P 3.44 0.16 0.03 P 3.44 0.16 0.03 P 3.44 0.16 0.03 P 3.44 0.16 0.03 P 3.44 0.16 0.03 P 3.44 0.16 0.03 P 3.44 0.16 0.03 P 3.44 0.16 0.03 P 3.44 0.16 0.03 P 3.44 0.16 0.03 P 3.67 0.39 0.16 P 3.67 0.39 0.16 331 P 3.67 0.39 0.16 P 3.67 0.39 0.16 P 3.67 0.39 0.16 P 3.67 0.39 0.16 P 3.67 0.39 0.16 P 3.67 0.39 0.16 P 3.67 0.39 0.16 P 3.67 0.39 0.16 P 3.67 0.39 0.16 P 3.67 0.39 0.16 P 3.67 0.39 0.16 P 3.67 0.39 0.16 P 3.67 0.39 0.16 P 3.67 0.39 0.16 P 3.89 0.61 0.38 P 3.89 0.61 0.38 P 3.89 0.61 0.38 P 3.89 0.61 0.38 P 3.89 0.61 0.38 P 3.89 0.61 0.38 P 3.89 0.61 0.38 P 3.89 0.61 0.38 P 3.89 0.61 0.38 P 3.89 0.61 0.38 P 3.89 0.61 0.38 P 3.89 0.61 0.38 P 3.89 0.61 0.38 P 3.89 0.61 0.38 P 3.89 0.61 0.38 P 3.89 0.61 0.38 P 3.89 0.61 0.38 P 3.89 0.61 0.38 P 3.89 0.61 0.38 P 3.89 0.61 0.38 P 3.89 0.61 0.38 P 3.89 0.61 0.38 P 3.89 0.61 0.38 332 P 4.11 0.83 0.70 P 4.11 0.83 0.70 P 4.11 0.83 0.70 P 4.11 0.83 0.70 P 4.11 0.83 0.70 P 4.11 0.83 0.70 P 4.11 0.83 0.70 P 4.11 0.83 0.70 P 4.11 0.83 0.70 P 4.33 1.05 1.11 P 4.33 1.05 1.11 P 4.33 1.05 1.11 P 4.33 1.05 1.11 P 4.33 1.05 1.11 P 4.33 1.05 1.11 P 4.33 1.05 1.11 P 4.33 1.05 1.11 P 4.56 1.28 1.65 P 4.56 1.28 1.65 P 4.78 1.50 2.26 mean CP 2.83 P 3.28 ALL 3.08 333 Regression Tree Model of BIBI For Coastal Plain Data |EPI.SUB<6.5 ADJ.COVR:aef PH.LAB<7.195 FLOW<0.32AVGTHAL<8.25 RIP.WIDm<37.5 SHADING<65FLOW<0.24WETLANDS<0.145 SO4.LAB<12.455 ADJ.COVR:cd COND.FLD<167.5 URBAN<0.605PASTUR<16.58 FLOW<8.84 WOOD.DEB<0.5 DECIDFOR<22.35AESTHET<17.5 RIFFQUAL<16.5HIGHURB<0.005BANKSTAB<9.5AGRI<59.315COND.LAB<115 DO.FLD<9.25 CONIFER<1.365 WOOD.DEB<0.5 ANC.LAB<236.6PH.FLD<6.8 DECIDFOR<15.98 ADJ.COVR:acfTEMP.FLD<19.9 AVG.VEL<0.31875 1.9401.3492.071 2.8801.8082. 302.7352.9863.776 2.9601.8612.7343.6722.8123.725 3.6672.9983.6373.6673.976 3.1484.187 2.6193.8903.284 2.7431.8502.380 2.1183.118 2.8833.4874.093 Epifaunal substrate < 6.5 Adjacent cover :aef SO4 lab < 12.455 PH lab Rip width < 37.5 adjacent cover:cd woody debris < 0.5 334 Regression tree: tree(formula = BIBI.98 ~ RIP.WIDm + TEMP.FLD + BUFF.TYP + ADJ.COVR + REGION + AGRI + URBAN + FOREST + ORDER + DO.FLD + PH.FLD + COND.FLD + PH.LAB + COND.LAB + ANC.LAB + SO4.LAB + DOC.LAB + ACIDSRC + PASTURE + INSTRHAB + EPI.SUB + VEL.DPTH + POOLQUAL + RIFFQUAL + CHAN.ALT + BANKSTAB + EMBEDDED + CH.FLOW + SHADING + REMOTE + AESTHET + WOOD.DEB + NUMROOT + MAXDEPTH + ST.GRAD + AVGWID + AVGTHAL + AVG.VEL + FLOW + ACREAGE + WETLANDS + BARREN + WATER + HIGHURB + LOWURB + PASTUR + PROBCROP + ROWCROP + crop + CONIFER + DECIDFOR + MIXEDFOR + EMERGWET + WOODYWET + COALMINE + TRANS + BKTRFLAG + BLACKWAT, data = BIBI.CP, na.action = na.exclude, mincut = 5, minsize = 10, mindev = 0.01) Variables actually used in tree construction: [1] "EPI.SUB" "ADJ.COVR" "PH.LAB" "FLOW" "AVGTHAL" "RIP.WIDm" [7] "SHADING" "WETLANDS" "SO4.LAB" "COND.FLD" "URBAN" "PASTUR" [13] "WOOD.DEB" "DECIDFOR" "AESTHET" "RIFFQUAL" "HIGHURB" "BANKSTAB" [19] "AGRI" "COND.LAB" "DO.FLD" "CONIFER" "ANC.LAB" "PH.FLD" [25] "TEMP.FLD" "AVG.VEL" Number of terminal nodes: 33 Residual mean deviance: 0.163 = 39.27 / 241 Distribution of residuals: Min. 1st Qu. Median Mean 3rd Qu. Max. -0.940000 -0.227100 -0.000556 0.000000 0.222900 1.480000 node), split, n, deviance, yval * denotes terminal node 1) root 274 207.30000 3.080 2) EPI.SUB<6.5 58 41.14000 2.402 4) ADJ.COVR:CP,OR,PA 29 13.03000 1.914 8) PH.LAB<7.195 24 6.67300 1.713 16) FLOW<0.32 16 2.90200 1.534 32) AVGTHAL<8.25 5 1.35400 1.940 * 33) AVGTHAL>8.25 11 0.34690 1.349 * 17) FLOW>0.32 8 2.23100 2.071 * 9) PH.LAB>7.195 5 0.72420 2.880 * 5) ADJ.COVR:FR,LN,OF,SL 29 14.32000 2.889 10) RIP.WIDm<37.5 11 3.75800 2.365 20) SHADING<65 5 0.07148 1.808 * 21) SHADING>65 6 0.83820 2.830 * 11) RIP.WIDm>37.5 18 5.69700 3.209 22) FLOW<0.24 6 1.24600 2.735 * 23) FLOW>0.24 12 2.42500 3.447 46) WETLANDS<0.145 5 0.32510 2.986 * 47) WETLANDS>0.145 7 0.28100 3.776 * 3) EPI.SUB>6.5 216 132.30000 3.262 6) SO4.LAB<12.455 157 74.47000 3.418 12) ADJ.COVR:LN,OF 27 17.33000 2.763 24) COND.FLD<167.5 14 5.30800 2.254 48) URBAN<0.605 5 0.97360 2.960 * 49) URBAN>0.605 9 0.45290 1.861 * 25) COND.FLD>167.5 13 4.47200 3.312 50) PASTUR<16.58 5 0.33710 2.734 * 51) PASTUR>16.58 8 1.42500 3.672 * 13) ADJ.COVR:CP,FR,PA,SL,TG 130 43.16000 3.554 26) FLOW<8.84 100 27.24000 3.665 52) WOOD.DEB<0.5 16 5.40000 3.154 104) DECIDFOR<22.35 10 1.69200 2.812 * 105) DECIDFOR>22.35 6 0.58150 3.725 * 53) WOOD.DEB>0.5 84 16.87000 3.762 106) AESTHET<17.5 72 13.24000 3.692 212) RIFFQUAL<16.5 66 9.86900 3.741 335 424) HIGHURB<0.005 17 3.65500 3.470 848) BANKSTAB<9.5 12 1.38600 3.667 * 849) BANKSTAB>9.5 5 0.69090 2.998 * 425) HIGHURB>0.005 49 4.52900 3.835 850) AGRI<59.315 14 1.21600 3.637 * 851) AGRI>59.315 35 2.54300 3.915 1702) COND.LAB<115 7 0.20250 3.667 * 1703) COND.LAB>115 28 1.80500 3.976 * 213) RIFFQUAL>16.5 6 1.43400 3.148 * 107) AESTHET>17.5 12 1.11800 4.187 * 27) FLOW>8.84 30 10.56000 3.184 54) DO.FLD<9.25 10 2.02600 2.619 * 55) DO.FLD>9.25 20 3.75400 3.466 110) CONIFER<1.365 6 0.69980 3.890 * 111) CONIFER>1.365 14 1.51300 3.284 * 7) SO4.LAB>12.455 59 43.86000 2.848 14) WOOD.DEB<0.5 18 6.62300 2.346 28) ANC.LAB<236.6 10 1.70900 2.743 * 29) ANC.LAB>236.6 8 1.37000 1.850 * 15) WOOD.DEB>0.5 41 30.73000 3.068 30) PH.FLD<6.8 9 4.90200 2.380 * 31) PH.FLD>6.8 32 20.37000 3.261 62) DECIDFOR<15.98 11 5.40600 2.664 124) ADJ.COVR:CP,LN,PA 5 1.35600 2.118 * 125) ADJ.COVR:FR,OF 6 1.32000 3.118 * 63) DECIDFOR>15.98 21 8.97600 3.574 126) TEMP.FLD<19.9 12 3.57300 3.185 252) AVG.VEL<0.31875 6 1.89900 2.883 * 253) AVG.VEL>0.31875 6 0.58190 3.487 * 127) TEMP.FLD>19.9 9 1.16000 4.093 * 336 Regression Tree Model of BIBI For Coastal Plain Data |SO4.LAB<12.45 crop<52.515 COND.LAB<268.05 CONIFER<1.26 AESTHET<17.5 ST.GRAD<0.315 URBAN<1.515 ANC.LAB<325.8 AVG.VEL<0.34 SHADING<77.5DECIDFOR<27.18MIXEDFOR<2.195 CONIFER<1.565 ROWCROP<15.505SO4.LAB<9.8 ADJ.COVR:cd ANC.LAB<282.765 FLOW<2.57 PROBCROP<36.81 AVG.VEL<0.13875 ROWCROP<10.21 ST.GRAD<1.9 3.3803.7984.154 4.370 4.2883.8393.3543.815 3.132 2.518 3.3093.6942.998 2.555 2.223 2.424 3.8263.222 2.927 1.978 2.3362.963 3.413 SO4 lab Crop < 52.515 avg vel < 0.13875 Cond lab < 268.05 adj cover: 337 Regression tree: tree(formula = BIBI.98 ~ RIP.WIDm + TEMP.FLD + BUFF.TYP + ADJ.COVR + REGION + AGRI + URBAN + FOREST + ORDER + DO.FLD + PH.FLD + COND.FLD + PH.LAB + COND.LAB + ANC.LAB + SO4.LAB + DOC.LAB + ACIDSRC + PASTURE + INSTRHAB + EPI.SUB + VEL.DPTH + POOLQUAL + RIFFQUAL + CHAN.ALT + BANKSTAB + EMBEDDED + CH.FLOW + SHADING + REMOTE + AESTHET + WOOD.DEB + NUMROOT + MAXDEPTH + ST.GRAD + AVGWID + AVGTHAL + AVG.VEL + FLOW + ACREAGE + WETLANDS + BARREN + WATER + HIGHURB + LOWURB + PASTUR + PROBCROP + ROWCROP + crop + CONIFER + DECIDFOR + MIXEDFOR + EMERGWET + WOODYWET + COALMINE + TRANS + BKTRFLAG + BLACKWAT, data = BIBI.P, na.action = na.exclude, mincut = 5, minsize = 10, mindev = 0.01) Variables actually used in tree construction: [1] "SO4.LAB" "crop" "COND.LAB" "CONIFER" "AESTHET" "ST.GRAD" [7] "URBAN" "ANC.LAB" "AVG.VEL" "SHADING" "DECIDFOR" "MIXEDFOR" [13] "ROWCROP" "ADJ.COVR" "FLOW" "PROBCROP" Number of terminal nodes: 23 Residual mean deviance: 0.1311 = 17.04 / 130 Distribution of residuals: Min. 1st Qu. Median Mean 3rd Qu. Max. -0.9780 -0.1883 0.0420 0.0000 0.2220 1.2170 node), split, n, deviance, yval * denotes terminal node 1) root 153 78.9300 3.276 2) SO4.LAB<12.45 130 60.3600 3.369 4) crop<52.515 95 32.4800 3.533 8) COND.LAB<268.05 89 25.7500 3.599 16) CONIFER<1.26 30 4.8290 3.874 32) AESTHET<17.5 24 2.5470 3.750 64) ST.GRAD<0.315 7 0.3710 3.380 * 65) ST.GRAD>0.315 17 0.8200 3.903 130) URBAN<1.515 12 0.2380 3.798 * 131) URBAN>1.515 5 0.1355 4.154 * 33) AESTHET>17.5 6 0.4394 4.370 * 17) CONIFER>1.26 59 17.4900 3.458 34) ANC.LAB<325.8 30 5.8750 3.710 68) AVG.VEL<0.34 25 3.6160 3.826 136) SHADING<77.5 14 1.4610 3.999 272) DECIDFOR<27.18 5 0.1409 4.288 * 273) DECIDFOR>27.18 9 0.6719 3.839 * 137) SHADING>77.5 11 1.2000 3.605 274) MIXEDFOR<2.195 5 0.2579 3.354 * 275) MIXEDFOR>2.195 6 0.3624 3.815 * 69) AVG.VEL>0.34 5 0.2517 3.132 * 35) ANC.LAB>325.8 29 7.7400 3.198 70) CONIFER<1.565 6 1.2260 2.518 * 71) CONIFER>1.565 23 3.0190 3.375 142) ROWCROP<15.505 18 1.4190 3.480 284) SO4.LAB<9.8 10 0.4137 3.309 * 285) SO4.LAB>9.8 8 0.3478 3.694 * 143) ROWCROP>15.505 5 0.6909 2.998 * 9) COND.LAB>268.05 6 0.5986 2.555 * 5) crop>52.515 35 18.4200 2.924 10) ADJ.COVR:LN,OF 8 3.5470 2.223 * 11) ADJ.COVR:CP,FR,PA,TG 27 9.7630 3.132 22) ANC.LAB<282.765 5 1.3380 2.424 * 23) ANC.LAB>282.765 22 5.3480 3.293 46) FLOW<2.57 13 2.4300 3.547 338 92) PROBCROP<36.81 7 0.2704 3.826 * 93) PROBCROP>36.81 6 0.9813 3.222 * 47) FLOW>2.57 9 0.8712 2.927 * 3) SO4.LAB>12.45 23 11.0700 2.750 6) AVG.VEL<0.13875 5 1.6380 1.978 * 7) AVG.VEL>0.13875 18 5.6310 2.964 14) ROWCROP<10.21 11 1.9010 2.678 28) ST.GRAD<1.9 5 0.4005 2.336 * 29) ST.GRAD>1.9 6 0.4275 2.963 * 15) ROWCROP>10.21 7 1.4210 3.413 * 339 APPENDIX 16: PHI SAS MODELING Model Parameter Search SAS language DATA one ; INFILE 'D:\g MBSS paper\SAS\SAS split data\PHIcaldata.dat' ; INPUT PHI RIP_WIDm TEMP_FLD BUFF_FOR CROP REGION_E PHYS_CP ACID_NO AGRI URBAN FOREST ORDER DO_FLD PH_FLD COND_FLD PH_LAB COND_LAB ANC_LAB SO4_LAB DOC_LAB PASTURE ST_GRAD AVGWID AVGTHAL AVG_VEL FLOW ACREAGE WETLANDS BARREN WATER HIGHURB LOWURB PASTUR PROBCROP ROWCROP cropmix CONIFER DECIDFOR MIXEDFOR EMERGWET WOODYWET COALMINE TRANS BKTRFLAG BLACKWAT ; *proc print; * proc corr ; proc reg ; *model PHI = RIP_WIDm TEMP_FLD BUFF_FOR CROP REGION_E PHYS_CP ACID_NO AGRI URBAN FOREST ORDER DO_FLD PH_FLD COND_FLD PH_LAB COND_LAB ANC_LAB SO4_LAB DOC_LAB PASTURE ST_GRAD AVGWID AVGTHAL AVG_VEL FLOW ACREAGE WETLANDS BARREN WATER HIGHURB LOWURB PASTUR PROBCROP ROWCROP cropmix CONIFER DECIDFOR MIXEDFOR EMERGWET WOODYWET COALMINE TRANS BKTRFLAG BLACKWAT / selection=rsquare adjrsq best=3; run; quit; 340 Results R-Square Selection Method Number in Adjusted Model R2 R-Square Variables in Model 1 0.2568 0.2537 AVGTHAL 1 0.1618 0.1584 AVG_VEL 1 0.1442 0.1407 ORDER -------------------------------------------------------------------------------------------------- 2 0.3843 0.3792 AVGTHAL AVG_VEL 2 0.3427 0.3372 DO_FLD AVGTHAL 2 0.3387 0.3332 DOC_LAB AVGTHAL -------------------------------------------------------------------------------------------------- 3 0.4183 0.4110 RIP_WIDm AVGTHAL AVG_VEL 3 0.4158 0.4085 DOC_LAB AVGTHAL AVG_VEL 3 0.4096 0.4022 BUFF_FOR AVGTHAL AVG_VEL -------------------------------------------------------------------------------------------------- 4 0.4458 0.4365 RIP_WIDm DOC_LAB AVGTHAL AVG_VEL 4 0.4447 0.4353 RIP_WIDm DO_FLD AVGTHAL AVG_VEL 4 0.4431 0.4337 RIP_WIDm AVGTHAL AVG_VEL ACREAGE -------------------------------------------------------------------------------------------------- 5 0.4635 0.4522 RIP_WIDm AVGTHAL AVG_VEL ACREAGE ROWCROP 5 0.4634 0.4521 RIP_WIDm DO_FLD AVGTHAL AVG_VEL FLOW 5 0.4626 0.4513 RIP_WIDm DOC_LAB AVGTHAL AVG_VEL FLOW Model Calibration SAS language * Using the model variables RIP_WIDm AVGTHAL AVG_VEL ACREAGE ROWCROP model PHI = RIP_WIDm AVGTHAL AVG_VEL ACREAGE ROWCROP / collinoint tol VIF R influence ; output out=resPHBI residual = PHIres; proc univariate normal plot; var PHIres ; 341 Results Dependent Variable: PHI Analysis of Variance Sum of Mean Source DF Squares Model 5 90102 18020 40.95 <.0001 Error 237 104281 440.00401 Corrected Total 242 194382 Root MSE 20.97627 R-Square 0.4635 Dependent Mean 59.53605 Adj R-Sq 0.4522 Coeff Var 35.23289 Parameter Estimates Parameter Standard Variable DF Estimate Error t Value Pr > |t| Tolerance Intercept 1 4.25150 4.28229 0.99 0.3218 . RIP_WIDm 1 0.27104 0.06452 4.20 <.0001 0.95939 AVGTHAL 1 0.95185 0.09817 9.70 <.0001 0.67232 AVG_VEL 1 74.59470 9.65187 7.73 <.0001 0.98151 ACREAGE 1 -0.000636 0.000173 -3.67 0.0003 0.65173 ROWCROP 1 0.38342 0.12763 3.00 0.0029 0.97658 Variable: PHIres (Residual) N 243 Sum Weights 243 Mean 0 Sum Observations 0 Std Deviation 20.7584446 Variance 430.913023 Skewness -0.3351753 Kurtosis 0.2586999 Uncorrected SS 104280.952 Corrected SS 104280.952 Coeff Variation . Std Error Mean 1.33165484 Tests for Normality Test --Statistic--- -----p Value------ Shapiro-Wilk W 0.984446 Pr < W 0.0093 342 Model Validation SAS language DATA one ; INFILE 'D:\g MBSS paper\SAS\SAS split data\PHIvaldata.dat' ; INPUT PHI RIP_WIDm TEMP_FLD BUFF_FOR CROP REGION_E PHYS_CP ACID_NO AGRI URBAN FOREST ORDER DO_FLD PH_FLD COND_FLD PH_LAB COND_LAB ANC_LAB SO4_LAB DOC_LAB PASTURE ST_GRAD AVGWID AVGTHAL AVG_VEL FLOW ACREAGE WETLANDS BARREN WATER HIGHURB LOWURB PASTUR PROBCROP ROWCROP cropmix CONIFER DECIDFOR MIXEDFOR EMERGWET WOODYWET COALMINE TRANS BKTRFLAG BLACKWAT ; PHIpred = 4.251 + 0.271*RIP_WIDm + .952*AVGTHAL +74.595*AVG_VEL - .00063*ACREAGE +.383*ROWCROP; res = PHI - PHIpred ; proc print; Proc corr; run; quit; Results: Correlation Coefficient for PHI vs PHIpred = 0.722 so R2 = .52 343 APPENDIX 17: PHI S-PLUS MODELING Regression tree with all parameters |AVGTHAL<25.625 FLOW<0.195 WATER<0.12WETLANDS<0.045 TEMP.FLD<20.3 PH.FLD<7.36ADJ.COVR:acf ROWCROP<2.6 AVGWID<2.255 PH.LAB<7.155SO4.LAB<7.095 AVG.VEL<0.105 TEMP.FLD<21.55 RIP.WIDm<20 SO4.LAB<11.325 SO4.LAB<11.035 PROBCROP<25.105AVG.VEL<0.065 RIP.WIDm<3.5ANC.LAB<517.85WATER<0.255CONIFER<1.065crop<48.885DECIDFOR<31.085FLOW<14.205ANC.LAB<253.95PH.LAB<7.17 11.2831.21 60.3780.0449.03 32.6218.60 28.4960.7550.0433.05 54.0083.7267.9951.3826.007 .5048.22 70.4985.8554.7090.7889.3271.1579.7288.4571.97 65.65 344 Regression Tree Model with all Parameters tree(formula = PHI ~ RIP.WIDm + TEMP.FLD + BUFF.TYP + ADJ.COVR + REGION + AGRI + PHYCPIO + URBAN + FOREST + ORDER + DO.FLD + PH.FLD + COND.FLD + PH.LAB + COND.LAB + ANC.LAB + SO4.LAB + DOC.LAB + ACIDSRC + PASTURE + ST.GRAD + AVGWID + AVGTHAL + AVG.VEL + FLOW + ACREAGE + WETLANDS + BARREN + WATER + HIGHURB + LOWURB + PASTUR + PROBCROP + ROWCROP + crop + CONIFER + DECIDFOR + MIXEDFOR + EMERGWET + WOODYWET + COALMINE + TRANS + BKTRFLAG + BLACKWAT, data = PHI.cal.data, na.action = na.exclude, mincut = 5, minsize = 10,mindev = 0.01) Variables actually used in tree construction: "AVGTHAL" "FLOW" "WATER" "WETLANDS" "TEMP.FLD" "PH.FLD" "ADJ.COVR" "ROWCROP" "AVGWID" "PH.LAB" "SO4.LAB" "AVG.VEL" "RIP.WIDm" "PROBCROP" "ANC.LAB" "CONIFER" "crop" "DECIDFOR" Number of terminal nodes: 28 Residual mean deviance: 184.3 = 39620 / 215 node), split, n, deviance, yval * denotes terminal node 1) root 243 194400.0 59.54 2) AVGTHAL<25.625 92 62680.0 37.51 4) FLOW<0.195 28 8439.0 17.69 8) WATER<0.12 19 1812.0 11.28 * 9) WATER>0.12 9 4201.0 31.21 * 5) FLOW>0.195 64 38420.0 46.19 10) WETLANDS<0.045 26 12860.0 59.66 20) TEMP.FLD<20.3 21 5292.0 66.10 40) PH.FLD<7.36 16 2921.0 71.43 80) ADJ.COVR:CP,LN,PA 7 887.5 60.37 * 81) ADJ.COVR:FR,OF 9 511.2 80.04 * 41) PH.FLD>7.36 5 459.1 49.03 * 21) TEMP.FLD>20.3 5 3040.0 32.62 * 11) WETLANDS>0.045 38 17610.0 36.96 22) ROWCROP<2.6 10 1122.0 18.60 * 23) ROWCROP>2.6 28 11910.0 43.52 46) AVGWID<2.255 9 1525.0 28.49 * 47) AVGWID>2.255 19 7388.0 50.64 94) PH.LAB<7.155 9 2547.0 60.75 * 95) PH.LAB>7.155 10 3093.0 41.54 190) SO4.LAB<7.095 5 2144.0 50.04 * 191) SO4.LAB>7.095 5 226.7 33.05 * 3) AVGTHAL>25.625 151 59900.0 72.95 345 6) AVG.VEL<0.105 50 27040.0 60.34 12) TEMP.FLD<21.55 26 8911.0 69.20 24) RIP.WIDm<20 9 3551.0 54.00 * 25) RIP.WIDm>20 17 2181.0 77.24 50) SO4.LAB<11.325 10 733.8 83.72 * 51) SO4.LAB>11.325 7 427.1 67.99 * 13) TEMP.FLD>21.55 24 13890.0 50.75 26) SO4.LAB<11.035 11 5624.0 37.54 52) PROBCROP<25.105 5 964.8 51.38 * 53) PROBCROP>25.105 6 2903.0 26.00 * 27) SO4.LAB>11.035 13 4717.0 61.93 54) AVG.VEL<0.065 8 964.6 70.50 * 55) AVG.VEL>0.065 5 2224.0 48.22 * 7) AVG.VEL>0.105 101 20960.0 79.20 14) RIP.WIDm<3.5 24 7685.0 71.65 28) ANC.LAB<517.85 17 3301.0 78.62 56) WATER<0.255 8 1890.0 70.49 * 57) WATER>0.255 9 412.2 85.85 * 29) ANC.LAB>517.85 7 1549.0 54.70 * 15) RIP.WIDm>3.5 77 11480.0 81.55 30) CONIFER<1.065 13 288.1 90.78 * 31) CONIFER>1.065 64 9863.0 79.68 62) crop<48.885 57 7764.0 81.40 124) DECIDFOR<31.085 48 4536.0 83.17 248) FLOW<14.205 30 3200.0 79.99 496) ANC.LAB<253.95 8 238.1 89.32 * 497) ANC.LAB>253.95 22 2014.0 76.60 994) PH.LAB<7.17 8 970.4 71.15 * 995) PH.LAB>7.17 14 670.3 79.72 * 249) FLOW>14.205 18 530.9 88.45 * 125) DECIDFOR>31.085 9 2276.0 71.97 * 63) crop>48.885 7 553.8 65.65 * 346 Model Validation Obs Pred Obs Pred 14.06 11.28 83.36 71.97 50.47 60.75 79.33 83.72 83.73 83.72 57.46 85.85 66.59 90.78 81.18 67.99 50.47 26.00 49.92 89.32 14.06 18.60 84.47 85.85 90.61 88.45 89.89 65.65 94.26 54.70 95.75 89.32 67.95 90.78 93.21 90.78 94.26 66.12 73.95 90.78 59.18 70.49 98.47 83.72 51.29 48.22 42.23 88.45 63.33 83.72 8.62 11.28 82.19 88.45 55.45 49.03 Splus model PHI observed vs predicted R2 = 0.5592 0 20 40 60 80 100 0 20 40 60 80 100 PHI observed PH I p re di ct ed 347 Development of equivalent R2 for regression tree model REGION PHI delta square corrected total SS CP 2.3 -51.15 2616.47 107843.12 CP CP 50.47 -2.98 8.89 99529.10 Pd CP 33.24 -20.21 408.50 207372.22 ALL CP 14.06 -39.39 1551.68 CP 89.62 36.17 1308.17 Resid SS CP 94.81 41.36 1710.53 deviance CP 2.76 -50.69 2569.62 CP 3491 CP 66.12 12.67 160.49 5936 CP 6.89 -46.56 2167.96 4493 CP 15.73 -37.72 1422.90 4721 CP 9.72 -43.73 1912.43 7278 CP 23.66 -29.79 887.53 5775 CP 64.37 10.92 119.22 31694 CP 61.55 8.10 65.59 Pd 75.8 CP 50.47 -2.98 8.89 818 CP 10.83 -42.62 1816.58 5225 CP 40.66 -12.79 163.62 4908 CP 81.68 28.23 796.85 15980 CP 78.15 24.70 610.02 22010 CP 58.64 5.19 26.92 49017 CP 3.31 -50.14 2514.16 ALL 8439 CP 76.02 22.57 509.34 38420 CP 57.57 4.12 16.96 8911 CP 43.89 -9.56 91.42 13890 CP 7.48 -45.97 2113.37 7685 CP 63.87 10.42 108.55 11480 CP 53.22 -0.23 0.05 88825 CP 11.49 -41.96 1760.76 CP 15.16 -38.29 1466.23 Res/Cor SS CP 26.13 -27.32 746.46 CP 0.29 CP 81.52 28.07 787.85 Pd 0.49 CP 41.73 -11.72 137.39 ALL 0.43 CP 34.47 -18.98 360.29 348 CP 66.37 12.92 166.89 R2 CP 83.73 30.28 916.79 CP 0.71 CP 50.47 -2.98 8.89 Pd 0.51 CP 66.86 13.41 179.79 ALL 0.57 CP 56.49 3.04 9.23 CP 41.73 -11.72 137.39 CP 12.06 -41.39 1713.25 CP 14.06 -39.39 1551.68 CP 6.15 -47.30 2237.42 CP 69.02 15.57 242.38 CP 2.73 -50.72 2572.66 CP 5.9 -47.55 2261.14 CP 88.33 34.88 1216.52 CP 64.12 10.67 113.82 CP 78.34 24.89 619.44 CP 25.71 -27.74 769.58 CP 55.13 1.68 2.82 CP 77.59 24.14 582.67 CP 90.32 36.87 1359.29 CP 93.48 40.03 1602.29 CP 68.78 15.33 234.97 CP 67.1 13.65 186.28 CP 16.94 -36.51 1333.08 CP 87.63 34.18 1168.18 CP 59.18 5.73 32.82 CP 59.44 5.99 35.86 CP 51.29 -2.16 4.67 CP 59.44 5.99 35.86 CP 19.91 -33.54 1125.03 CP 76.02 22.57 509.34 CP 87.98 34.53 1192.22 CP 32.27 -21.18 448.65 CP 56.49 3.04 9.23 CP 83.12 29.67 880.23 CP 95.33 41.88 1753.82 CP 70.64 17.19 295.45 CP 38.81 -14.64 214.37 349 CP 28.99 -24.46 598.36 CP 95.42 41.97 1761.36 CP 90.61 37.16 1380.76 CP 8.79 -44.66 1994.64 CP 25.92 -27.53 757.98 CP 94.18 40.73 1658.82 CP 90.61 37.16 1380.76 CP 81.18 27.73 768.88 CP 91.59 38.14 1454.55 CP 88.21 34.76 1208.16 CP 90.42 36.97 1366.68 CP 16.63 -36.82 1355.82 CP 49.92 -3.53 12.47 CP 34.72 -18.73 350.87 CP 46.89 -6.56 43.05 CP 95.75 42.30 1789.17 CP 41.46 -11.99 143.79 CP 21.92 -31.53 994.23 CP 87.02 33.57 1126.85 CP 84.18 30.73 944.25 CP 47.71 -5.74 32.96 CP 64.88 11.43 130.61 CP 70.64 17.19 295.45 CP 1.27 -52.18 2722.90 CP 10.62 -42.83 1834.53 CP 86.64 33.19 1101.48 CP 44.16 -9.29 86.33 CP 97.85 44.40 1971.24 CP 31.07 -22.38 500.93 CP 63.1 9.65 93.10 CP 53.76 0.31 0.10 CP 93.21 39.76 1580.75 CP 25.5 -27.95 781.28 CP 73.95 20.50 420.19 CP 14.33 -39.12 1530.48 CP 93.48 40.03 1602.29 CP 94 40.55 1644.19 350 CP 15.3 -38.15 1455.53 CP 33.97 -19.48 379.52 CP 98.47 45.02 2026.67 CP 69.25 15.80 249.60 CP 10.73 -42.72 1825.12 CP 66.61 13.16 173.15 CP 82.65 29.20 852.56 CP 89.42 35.97 1293.74 CP 64.63 11.18 124.96 CP 76.22 22.77 518.41 CP 24.47 -28.98 839.92 CP 87.75 34.30 1176.39 CP 49.64 -3.81 14.53 CP 44.7 -8.75 76.59 CP 26.13 -27.32 746.46 P 2.29 -63.13 3985.33 P 73.05 7.63 58.22 P 71.82 6.40 40.97 P 36.37 -29.05 843.87 P 75.4 9.98 99.61 P 47.29 -18.13 328.68 P 40.74 -24.68 609.08 P 52.4 -13.02 169.51 P 60.44 -4.98 24.80 P 34.04 -31.38 984.67 P 66.59 1.17 1.37 P 94.26 28.84 831.78 P 11.57 -53.85 2899.77 P 40.24 -25.18 634.01 P 5.25 -60.17 3620.37 P 35.9 -29.52 871.40 P 76.52 11.10 123.22 P 80.32 14.90 222.03 P 96.68 31.26 977.22 P 87.65 22.23 494.20 P 58.46 -6.96 48.43 P 90.25 24.83 616.55 351 P 93.54 28.12 790.76 P 90.61 25.19 634.56 P 47.8 -17.62 310.45 P 44.74 -20.68 427.64 P 62.38 -3.04 9.24 P 68.39 2.97 8.82 P 71.41 5.99 35.89 P 93.05 27.63 763.45 P 94.26 28.84 831.78 P 89.7 24.28 589.54 P 60.54 -4.88 23.81 P 5.25 -60.17 3620.37 P 87.65 22.23 494.20 P 67.05 1.63 2.66 P 66.14 0.72 0.52 P 20.13 -45.29 2051.14 P 46.27 -19.15 366.70 P 83.64 18.22 331.99 P 67.95 2.53 6.40 P 83.07 17.65 311.54 P 94.26 28.84 831.78 P 28.34 -37.08 1374.89 P 87.2 21.78 474.39 P 91.76 26.34 693.82 P 80.32 14.90 222.03 P 68.83 3.41 11.63 P 91.12 25.70 660.52 P 88.3 22.88 523.52 P 19.48 -45.94 2110.44 P 63.33 -2.09 4.37 P 74.63 9.21 84.83 P 18.84 -46.58 2169.65 P 51.38 -14.04 197.11 P 77.25 11.83 139.96 P 57.96 -7.46 55.64 P 74.24 8.82 77.80 P 93.54 28.12 790.76 352 P 51.38 -14.04 197.11 P 79.33 13.91 193.50 P 82.19 16.77 281.25 P 87.42 22.00 484.02 P 94.15 28.73 825.44 P 5.78 -59.64 3556.87 P 84.73 19.31 372.90 P 71.82 6.40 40.97 P 86.97 21.55 464.42 P 70.56 5.14 26.42 P 81.28 15.86 251.56 P 58.46 -6.96 48.43 P 67.95 2.53 6.40 P 83.36 17.94 321.86 P 76.15 10.73 115.14 P 80.32 14.90 222.03 P 83.92 18.50 342.27 P 60.44 -4.98 24.80 P 24.37 -41.05 1685.06 P 79.33 13.91 193.50 P 84.47 19.05 362.92 P 89.12 23.70 561.71 P 57.46 -7.96 63.35 P 93.44 28.02 785.15 P 68.83 3.41 11.63 P 72.64 7.22 52.14 P 25.91 -39.51 1561.00 P 87.87 22.45 504.03 P 84.2 18.78 352.71 P 22.89 -42.53 1808.76 P 37.81 -27.61 762.28 P 77.25 11.83 139.96 P 43.73 -21.69 470.43 P 48.92 -16.50 272.23 P 92.21 26.79 717.73 P 92.78 27.36 748.60 P 74.63 9.21 84.83 353 P 92.21 26.79 717.73 P 52.81 -12.61 159.00 P 30.9 -34.52 1191.59 P 96.19 30.77 946.82 P 84.47 19.05 362.92 P 86.97 21.55 464.42 P 84.47 19.05 362.92 P 89.89 24.47 598.81 P 7.99 -57.43 3298.15 P 78.31 12.89 166.17 P 77.61 12.19 148.61 P 86.5 21.08 444.39 P 76.89 11.47 131.57 P 84.73 19.31 372.90 P 88.92 23.50 552.27 P 37.81 -27.61 762.28 P 70.56 5.14 26.42 P 77.61 12.19 148.61 P 67.95 2.53 6.40 P 89.51 24.09 580.35 P 50.87 -14.55 211.69 P 48.31 -17.11 292.73 P 22.89 -42.53 1808.76 P 67.5 2.08 4.33 P 51.38 -14.04 197.11 P 89.12 23.70 561.71 P 85.51 20.09 403.63 P 12 -53.42 2853.64 P 16.47 -48.95 2396.05 P 39.75 -25.67 658.92 P 90.43 25.01 625.53 P 80.64 15.22 231.66 P 86.26 20.84 434.33 P 39.26 -26.16 684.32 P 70.56 5.14 26.42 P 43.23 -22.19 492.37 P 79 13.58 184.43 354 P 68.39 2.97 8.82 P 71.41 5.99 35.89 P 93.44 28.02 785.15 P 86.01 20.59 423.97 P 27.11 -38.31 1467.62 P 86.5 21.08 444.39 P 92.47 27.05 731.73 P 76.15 10.73 115.14 P 57.46 -7.96 63.35 P 5.15 -60.27 3632.41 P 84.73 19.31 372.90 P 67.5 2.08 4.33 P 82.19 16.77 281.25 P 83.64 18.22 331.99 P 85.61 20.19 407.66 P 13.36 -52.06 2710.19 P 50.35 -15.07 227.09 P 91.12 25.70 660.52 P 69.7 4.28 18.32 P 42.23 -23.19 537.75 P 8.62 -56.80 3226.18 P 55.45 -9.97 99.39 mean CP 53.45 P 65.42 ALL 60.15 355 PHI Regression Tree Model Coastal Plain Data |AVGTHAL<26.375 ST.GRAD<0.95 FLOW<0.49 ORDER<1.5BUFF.TYP:bc MIXEDFOR<5.215 AVG.VEL<0.10875 RIP.WIDm<22.5 crop<37.8 ANC.LAB<804.4 crop<41.7 HIGHURB<0.81 DOC.LAB<7.45 WETLANDS<0.23 6.62 18.19 31.01 54.87 29.53 57.56 33.76 55.75 84.89 74.42 57.67 50.31 91.60 77.59 64.81 Avg Thalweg < 26.375 Stream gradient < 0.95 Avg Velocity < 0.1085 Flow < 0.49 rip width < 22.5 DOC < 7.45 356 Regression tree: tree(formula = PHI ~ RIP.WIDm + TEMP.FLD + BUFF.TYP + ADJ.COVR + REGION + AGRI + URBAN + FOREST + ORDER + DO.FLD + PH.FLD + COND.FLD + PH.LAB + COND.LAB + ANC.LAB + SO4.LAB + DOC.LAB + ACIDSRC + PASTURE + ST.GRAD + AVGWID + AVGTHAL + AVG.VEL + FLOW + ACREAGE + WETLANDS + BARREN + WATER + HIGHURB + LOWURB + PASTUR + PROBCROP + ROWCROP + crop + CONIFER + DECIDFOR + MIXEDFOR + EMERGWET + WOODYWET + COALMINE + TRANS + BKTRFLAG + BLACKWAT, data = PHI.CP, na.action = na.exclude, mincut = 5, minsize = 10, mindev = 0.01) Variables actually used in tree construction: [1] "AVGTHAL" "ST.GRAD" "FLOW" "ORDER" "BUFF.TYP" "MIXEDFOR" [7] "AVG.VEL" "RIP.WIDm" "crop" "ANC.LAB" "HIGHURB" "DOC.LAB" [13] "WETLANDS" Number of terminal nodes: 15 Residual mean deviance: 158 = 16910 / 107 Distribution of residuals: Min. 1st Qu. Median Mean 3rd Qu. Max. -50.670 -5.638 0.111 0.000 7.305 36.590 node), split, n, deviance, yval * denotes terminal node 1) root 122 107800.0 53.45 2) AVGTHAL<26.375 52 27020.0 30.15 4) ST.GRAD<0.95 44 15420.0 25.17 8) FLOW<0.49 28 3491.0 16.35 16) ORDER<1.5 23 1680.0 13.16 32) BUFF.TYP:G,none 10 208.1 6.62 * 33) BUFF.TYP:FR 13 715.9 18.19 * 17) ORDER>1.5 5 501.3 31.01 * 9) FLOW>0.49 16 5936.0 40.62 18) MIXEDFOR<5.215 7 525.7 54.87 * 19) MIXEDFOR>5.215 9 2883.0 29.53 * 5) ST.GRAD>0.95 8 4493.0 57.56 * 3) AVGTHAL>26.375 70 31630.0 70.76 6) AVG.VEL<0.10875 40 18020.0 61.59 12) RIP.WIDm<22.5 13 4721.0 43.91 24) crop<37.8 7 1453.0 33.76 * 25) crop>37.8 6 1706.0 55.75 * 13) RIP.WIDm>22.5 27 7278.0 70.11 26) ANC.LAB<804.4 22 4447.0 74.61 52) crop<41.7 16 1365.0 80.96 104) HIGHURB<0.81 10 719.5 84.89 * 105) HIGHURB>0.81 6 234.7 74.42 * 53) crop>41.7 6 714.7 57.67 * 27) ANC.LAB>804.4 5 424.4 50.31 * 7) AVG.VEL>0.10875 30 5775.0 82.98 14) DOC.LAB<7.45 24 1917.0 87.52 28) WETLANDS<0.23 17 378.5 91.60 * 29) WETLANDS>0.23 7 565.1 77.59 * 15) DOC.LAB>7.45 6 1384.0 64.81 * 357 PHI Regression Tree Model Piedmont Data |AVGWID<2.205 AVGTHAL<10.875 AVG.VEL<0.1375ADJ.COVR:cde ST.GRAD<0.225 AVGTHAL<26.125 TEMP.FLD<18.65 CONIFER<1.46TEMP.FLD<20.4 ANC.LAB<457.7 ROWCROP<10.08 ANC.LAB<313.8AVGWID<3.325CONIFER<1.525FLOW<6.84 ROWCROP<14.375 ADJ.COVR:acdPH.FLD<7.065FLOW<6.955 7.4825.08 33.1563.10 36.51 64.5988.3053.9029.30 76.3360.2171.8987.2876.4585.8466.1470.3085.9671.4850.73 Buffer width < 2.205 Avg thalweg < 10.85 stream gradient < 0.225 Avg velocity < avg thalweg < 26.125 358 Regression tree: tree(formula = PHI ~ RIP.WIDm + TEMP.FLD + BUFF.TYP + ADJ.COVR + REGION + AGRI + URBAN + FOREST + ORDER + DO.FLD + PH.FLD + COND.FLD + PH.LAB + COND.LAB + ANC.LAB + SO4.LAB + DOC.LAB + ACIDSRC + PASTURE + ST.GRAD + AVGWID + AVGTHAL + AVG.VEL + FLOW + ACREAGE + WETLANDS + BARREN + WATER + HIGHURB + LOWURB + PASTUR + PROBCROP + ROWCROP + crop + CONIFER + DECIDFOR + MIXEDFOR + EMERGWET + WOODYWET + COALMINE + TRANS + BKTRFLAG + BLACKWAT, data = PHI.P, na.action = na.exclude, mincut = 5, minsize = 10, mindev = 0.01) Variables actually used in tree construction: [1] "AVGWID" "AVGTHAL" "AVG.VEL" "ADJ.COVR" "ST.GRAD" "TEMP.FLD" [7] "CONIFER" "ANC.LAB" "ROWCROP" "FLOW" "PH.FLD" Number of terminal nodes: 20 Residual mean deviance: 179 = 24160 / 135 Distribution of residuals: Min. 1st Qu. Median Mean 3rd Qu. Max. -31.2600 -6.6340 -0.3113 0.0000 7.5440 53.2000 node), split, n, deviance, yval * denotes terminal node 1) root 155 99530.0 65.42 2) AVGWID<2.205 25 13530.0 33.76 4) AVGTHAL<10.875 11 1739.0 15.48 8) AVG.VEL<0.1375 6 75.8 7.48 * 9) AVG.VEL>0.1375 5 818.4 25.08 * 5) AVGTHAL>10.875 14 5225.0 48.13 10) ADJ.COVR:LN,OF,PA 7 1460.0 33.15 * 11) ADJ.COVR:CP,FR,TG 7 624.1 63.10 * 3) AVGWID>2.205 130 56130.0 71.51 6) ST.GRAD<0.225 6 4908.0 36.51 * 7) ST.GRAD>0.225 124 43510.0 73.20 14) AVGTHAL<26.125 27 15980.0 60.55 28) TEMP.FLD<18.65 15 4756.0 74.08 56) CONIFER<1.46 9 2580.0 64.59 * 57) CONIFER>1.46 6 152.6 88.30 * 29) TEMP.FLD>18.65 12 5057.0 43.65 58) TEMP.FLD<20.4 7 2632.0 53.90 * 59) TEMP.FLD>20.4 5 660.2 29.30 * 15) AVGTHAL>26.125 97 22010.0 76.72 30) ANC.LAB<457.7 60 8880.0 80.44 60) ROWCROP<10.08 12 2653.0 69.61 120) ANC.LAB<313.8 7 965.9 76.33 * 121) ANC.LAB>313.8 5 928.9 60.21 * 61) ROWCROP>10.08 48 4468.0 83.15 122) AVGWID<3.325 5 950.2 71.89 * 123) AVGWID>3.325 43 2811.0 84.46 246) CONIFER<1.525 24 565.1 87.28 * 247) CONIFER>1.525 19 1813.0 80.89 494) FLOW<6.84 10 843.5 76.45 * 495) FLOW>6.84 9 551.6 85.84 * 31) ANC.LAB>457.7 37 10950.0 70.69 62) ROWCROP<14.375 23 5228.0 76.52 124) ADJ.COVR:CP,LN,OF 7 1256.0 66.14 * 125) ADJ.COVR:FR,PA,TG 16 2886.0 81.06 250) PH.FLD<7.065 5 1420.0 70.30 * 251) PH.FLD>7.065 11 623.5 85.96 * 63) ROWCROP>14.375 14 3655.0 61.10 126) FLOW<6.955 7 657.3 71.48 * 127) FLOW>6.955 7 1490.0 50.73 * 359 APPENDIX 18: SAS PROGRAMS *Program to look at qualitative variables Is there a difference between regions? ; DATA one ; INFILE 'd:/ProcFreq/shape.dat' ; INPUT sitenum$ wshed$ reg$ shape$ ; PROC freq ; TABLES shape reg *shape ; RUN ; QUIT ; *Proc mixed for GIS variables; DATA GIS ; INFILE 'd:/ProcMixed/mar19gis.dat' ; INPUT wshed$ reg$ shore$ DrDens sinuous topo ag forest ; PROC print ; PROC mixed ; CLASS reg wshed ; MODEL DrDens = reg ; RANDOM wshed(reg) / CL; LSMEANS reg ; RUN ; PROC mixed ; CLASS shore wshed ; MODEL forest = shore ; RANDOM wshed(shore) / CL; LSMEANS shore ; RUN ; QUIT ; SAS Code for Proc Cluster Analysis DATA one ; INFILE 'd:/f Classification/Nov25SAS.dat' ; INPUT SITENUM$ RZSlostd FSloSTD BSFSstd RZwtrSTD LBufSTD CXSstd pctAGstd ShapeSTD ChanSTD FieldSTD RZdraSTD OrderSTD ; proc cluster data=one method=average std outtree=outtree2 pseudo ccc simple; var RZSlostd FSloSTD BSFSstd RZwtrSTD LBufSTD CXSstd pctAGstd ShapeSTD ChanSTD FieldSTD RZdraSTD OrderSTD ; run; goptions reset=all; goptions ; /* vertical axis */ axis1 value=(font=swiss color=blue height=.2) label=(height=.2 color=green); /* horizontal axis */ axis2 value=(font=swiss color=red height=.2) label=(height=.2 color=black); title h=.5 'proc tree for average method - with standardized data'; data outtree2; set outtree2; if _name_ ne 'CL1' and _parent_ eq '' then delete; run; proc tree data=outtree2 horizontal vpages=3 hpages=3 vaxis=axis1 haxis=axis2; run; 360 CITED REFERENCES Aguiar F.C., M.T. Ferreria and P. Pinto. 2002. Relative influence of environmental variables on macroinvertebrate assemblages from an Iberian basin. J. N. Am. Benthol. Soc. 21(1): 43-53. Allan J.D., D.L. Erickson and J. Fay. 1997. The influence of catchment land use on stream integrity across multiple spatial scales. Freshwater Biology 37(1): 149-161. Alliance for the Chesapeake Bay. 1993. Nutrients and the Chesapeake: Refining the Bay cleanup effort. Chesapeake white paper. Annapolis, Md.: Chesapeake Bay Program. Alliance for the Chesapeake Bay. 1996. Final Report Submitted to the Chesapeake Bay Executive Committee. Baltimore, Md.: Riparian Forest Buffer Panel. Alliance for the Chesapeake Bay. 1998. Readiness to sacrifice forest for sprawl is almost stumping. Bay Journal 8(1). Alliance for the Chesapeake Bay. 1996. Riparian Forest Buffers. White paper. Annapolis, Md.: Chesapeake Bay Program. Alpert P. and J. Kagan. 1998. The utility of plant community types: a practical review of riparian vegetation classification in the Intermountain United States. Natural Areas Journal 18(2): 124-128. Angle J.S., C.M. Gross, R.L. Hill and M.S. McIntosh. 1993. Soil nitrate concentrations under corn as affected by tillage, manure, and fertilizer applications. Journal of Environmental Quality 22: 141-147. Arbuckle C.J., A.D. Huryn and S.A. Israel. 1999. A landcover classification in the Taieri River Catchment, New Zealand: a focus on the riparian zone. Geocarto International 14(3): 7-14. Armour C.L., D.A. Duff and W. Elmore. 1991. The effects of livestock grazing in riparian and stream ecosystems. Fisheries 16(1): 7-11. Arnold J.G., R. Srinivasan, R.S. Muttiah and J.R. Williams. 1998. Large area hydrologic modeling and assessment. I. Model development. Journal of the American Water Resources Association 34(1): 73-89. Baker W.L. 1989. Classification of the riparian vegetation of the montane and subalpine zones in western Colorado. The Great Basin Naturalist 49(2): 214-228. 361 Barker L.S. 2003. Quantification of the Effectiveness of Agricultural Riparian Buffers to Protect Stream Health in Maryland's Coastal Plain and Piedmont Regions. Unpublished PhD Dissertation. College Park, MD: University of Maryland, Biological Resources Engineering Department. Barling R.D. and I.D. Moore. 1994. Role of buffer strips in management of waterway pollution: a review. Environmental Management 18(4): 543-558. Basnyat P., L.D. Teeter, K.M. Flynn and B.G. Lockaby. 1999. Relationships between landscape characteristics and nonpoint source pollution inputs to coastal estuaries. Environmental Management 23(4): 539-549. Batchelor R.F. 1982. A taxonomic classification system for Montana riparian vegetation types: an interagency approach to classifying Montana's riparian ecosystems. Bozeman: Montana: Rural Area Development Committee, Wildlife Subcommittee, Riparian Program Team. Beaulac M.N. and K.H. Teckhow. 1982. An examination of land use - nutrient export relationships. Water Resources Bulletin 18: 1013-1094. Blankenship, K. 2002. Bay Journal, May 2002, Bay Program narrows range for nutrient reductions. Available at: http://www.bayjournal.com. Accessed on 15 June, 2003. Bolke J.K. and J.M. Denver. 1995. Combined use of groundwater dating, chemical and isotopic analyses to resolve the history and fate of nitrate contamination on two agricultural watersheds, Atlantic Coastal Plain, Md. Water Resources Research 31: 2319-2339. Boward D.M., P.F. Kazyak, S.A. Stranko, M.K. Hurd and T.P. Prochaska. 1999. From the mountains to the sea: the state of Maryland?s freshwater streams. EPA 903-R-99- 023. Annapolis, Maryland: Maryland Department of Natural Resources, Monitoring and Non-Tidal Assessment Division. Brazier J.R. and G.W. Brown. 1973. Buffer strips for stream temperature control. Research Paper 15. Corvallis, OR: Forest Research Laboratory, School of Forestry, Oregon State University. Brown S., M.M. Brinson and A.E. Lugo. 1979. Structure and function of riparian wetlands. In Proceedings of the Symposium on Strategies for Protection and Management of Floodplain Wetlands and Other Riparian Ecosystems, 17-31. R.R. Johnson and J.F. McCormick, eds. U.S. Department of Agriculture, Forest Service. 362 Budd W.W., P.L. Cohen, P.R. Saunders and F.R. Steiner. 1987. Stream corridor management in the Pacific Northwest. I. Determination of stream corridor widths. Environmental Management 11: 587-897. Bunn S.E., P.M. Davies and T.D. Mosisch. 1999. Ecosystem measures of river health and their response to riparian and catchment degradation. Freshwater Biology 41(2): 333-345. Burkholder, J.M. and H.B. Glasgow Jr. 1997. Pfiesteria piscicida and other toxic Pfiesteria-like dinoflagellates: behavior, impacts, and environmental controls. Limnology & Oceanography 42:1052-1075 Castelle A.J., A.W. Johnson and C. Conolly. 1994. Wetland and stream buffer size requirements ? a review. Journal of Environmental Quality 23: 878-882. Chesapeake Bay Program. 1996. Final report of the Riparian Forest Buffer Panel. Publ. CBP/TRS 158/96 or EPA 903-R-96-015. Annapolis, Md.: Chesapeake Bay Program Chesapeake Bay Program. 1999. Riparian forest buffers: linking land and water: the Chesapeake Bay riparian forest buffer initiative. Publ. CBP/TRS 220/99 or EPA 903- R-99-002. Annapolis, Md.: Chesapeake Bay Program. Cirmo C.P. and J.J. McDonnell. 1997. Linking the hydrologic and biogeochemical controls of nitrogen transport in near-stream zones of temperate-forested catchments: a review. J. Hydrology 199: 88-120. Close M.E. and R.J. Davies-Colley. 1990. Baseflow chemistry in New Zealand rivers. 2. Influence of environmental factors. New Zealand Journal of Marine and Freshwater Research 24: 343-356. Cochran W.G. 1977. Sampling Techniques. 3rd ed. Somerset, NJ: John Wiley and Sons. Collier K.J. 1995. Environmental factors affecting the taxonomic composition of aquatic macroinvertebrate communities in lowland waterways of Northland, New Zealand. New Zealand Journal of Marine and Freshwater Research 29(4): 453-465. Corkum L.D. 1989. Patterns of benthic invertebrate assemblages in rivers of northwestern North America. Freshwater Biology 21: 191-205. _______. 1990. Intrabiome distributional patterns of lotic macroinvertebrate assemblages. Canadian Journal of Fisheries and Aquatic Sciences 47: 2147-2157. _______. 1992. Relationships between density of macroinvertebrates and detritus in rivers. Arch. Hydrobiol. 125: 149-166. 363 Correll DL. 1983. N and P in soils and runoff of three Coastal Plain land uses. In Nutrient Cycling in Agricultural Ecosystem, 207-224. R.R. Lowrance, L. Asmussen and R. Leonard, eds. Athens, Ga.: Univ. of Georgia Press. _______. 1991. Human impact on the functioning of landscape boundaries. In The Role of Landscape Boundaries in the Management and Restoration of Changing Environments, 90-109. M.M. Holland, P.G. Risser and R.J. Naiman, eds. New York, New York: Chapman and Hall. _______. 1996. Buffer zones and water quality protection: general principles. In Proceedings of the International Conference on Buffer Zones. N. Haycock, ed. Oxford, England. _______. 2000. The current status of our knowledge of riparian buffer water quality functions. In Proceedings of the International Conference on Riparian Ecology and Management in Multi-Use Watersheds, 5-10. P.J. Wigington, Jr. and R.L. Beschta, eds. Middleburg, Va.: AWRA. Correll D.L., Jordan T.E. and D.E. Weller. 1992. Nutrient flux in a landscape: Effects of coastal land use and terrestrial community mosaic on nutrient transport to coastal waters. Estuaries 15: 431-442. _______. 1994. The Chesapeake Bay watershed: effects of land use and geology on dissolved nitrogen concentrations. In Toward a sustainable watershed: The Chesapeake experiment: Proceedings of the 1994 Chesapeake Research Conference. P. Hill and S. Nelson, eds. Norfolk, Va.: Ches. Research Consortium. Cowardin L.M., V. Carter, F.C. Golet and E.D. LaRoe. 1979. Classification of wetlands and deepwater habitats of the United States. Publ. No. FWS/OBS-79/31. Washington, DC: US Department of Interior, Fish and Wildlife Service, Office of Biological Services. Creed I.F. and L.E. Band. 1998. Export of nitrogen from catchments within a temperate forest: evidence for a unifying mechanism regulated by variable source area dynamics. Water Resources Research 34(11): 3105-3120. Curry P. and F.M. Slater. 1986. A classification of river corridor vegetation from four catchments in Wales. Journal of Biogeography 13(2): 119-132. D?Angelo D.J., S.V. Gregory, L.R. Ashkenas and J.L. Meyer. 1997. Physical and biological linkages within a stream geomorphic hierarchy: a modeling approach. J. N. Am. Benthol. Soc. 16(3): 480-502. 364 Day R.L., P.L. Richards and R.L. Brooks. 1996. Chesapeake Bay forest buffer inventory, final report. University Park, Pa.: Pennsylvania State University, College of Agricultural Sciences and School of Forest Resources. De'ath G. and K.E. Fabricius. 2000. Classification and regression trees: a powerful yet simple technique for ecological data analysis. Ecology 81(11): 3178-3192. Delong M.D. and M.A. Brusven. 1991. Classification and spatial mapping of riparian habitat with applications toward management of streams impacted by nonpoint source pollution. Environmental Management 15(4): 565-571. Dick-Peddie W.A. and J.P. Hubbard. 1977. Classification of riparian vegetation. In Importance, Preservation and Management of Riparian Habitat: A Symposium, 85-90. R.R. Johnson and D.A. Jones, tech. coordinators. Fort Collins, CO.: U.S. Rocky Mt. Forest and Range Experimental Station. Or USDA Forest Service General Technical Report: RM-43. Dillon P.J. and W.B. Kirchner. 1975. The effects of geology and land use on the export of phosphorus from watersheds. Water Research 9: 135-148. Dosskey M.G. 2000. How much can USDA riparian buffers reduce agricultural nonpoint pollution? In Proceedings of the International Conference on Riparian Ecology and Management in Multi-Use Watersheds, 427-430. P.J. Wigington, Jr. and R.L. Beschta, eds. Middleburg, Va.: AWRA. Durkin P., E. Muldavin and M. Bradley. 1995. A preliminary riparian/wetland vegetation community classification of the Upper and Middle Rio Grande watersheds in New Mexico. In Desired Future Conditions for Southwestern Riparian Ecosystems: Bringing Interests and Concerns Together. Fort Collins, Colo.: U.S. Dept. of Agriculture, Rocky Mountain Forest and Range Experiment Station. Egglishaw H.W. 1964. The distributional relationship between the bottom fauna and plant detritus in streams. Journal of Animal Ecology 33: 463-476. EPA. 1983. Chesapeake Bay: a framework for action. Philadelphia, Pa.: U.S. Environmental Protection Agency. Ewel K.L. 1978. Riparian ecosystem: conservation of their unique characteristics. In Strategies for Protection and Management o Floodplain Wetlands and Other Riparian Ecosystems. R.R. Johnson and J.F. McCormick, eds. Washington, DC: USDA, Forest Service. Or USDA Forest Service Gen. Tech. Rpt. WO-12. Federal Interagency Stream Restoration Working Group (10/1998). Stream Corridor Restoration: Principles, Processes, and Practices. GPO Item No. 0120-A; SuDocs No. A 57.6/2:EN 3/PT.653. 365 Felton G.K 1996. Agricultural chemicals at the outlet of a shallow carbonate aquifer. Transactions of the ASAE 39(3): 873-887. Fennessy M.S. and J.K. Cronk. 1997. The effectiveness and restoration potential of riparian ecotones for the management of nonpoint source pollution, particularly nitrate. Critical Reviews in Environmental Science and Technology 27(4): 285-317. Fitzpatrick F.A., E.M.P. Giddings, B.C. Scudder, D.J. Sullivan and B.N. Lenz. 2000. Effects of riparian corridors and hydrology on agricultural stream biota. In Proceedings of the International Conference on Riparian Ecology and Management in Multi-Use Watersheds, 215-220. P.J. Wigington, Jr. and R.L. Beschta, eds. Middleburg, Va.: AWRA. Freifelder R.R., S.V. Smith and R.H. Bennett. 1998. Cows, humans and hydrology in the nitrogen dynamics of a grazed rural watershed. Journal of Environmental Management 52(2): 99-111. Frenzel S.A. and R.B. Swanson. 1996. Relations of fish community composition to environmental variables in streams of central Nebraska, USA. Environmental Management 20: 689-705. Frissel C.A., W.J. Liss, C.E. Warren and M.D. Hurley. 1986. A hierarchical framework for stream habitat classification: viewing streams in a watershed context. Environmental Management 10: 199-214. Gburek W.J. and G.J. Folmar. 1999. Flow and chemical contributions to streamflow in an upland watershed: a baseflow survey. Journal of Hydrology 217(1-2): 1-18. Giles, R.H. Jr. and M.K. Trani. 1999. Key elements of landscape pattern measures. Environmental Management 23: 477-481. Gilliam J.W. and R.W. Skaggs. 1988. Nutrient and sediment removal in wetland buffers. In Proceedings of the National Wetland Symposium: Wetland Hydrology. J.A. Kusler and G. Brooks, eds. Berme, NY: Association of State Wetland Managers. Girard M.M., D.L. Wheeler, D.L. Mills and B. Stephanie. 1997. Classification of riparian communities on the Bighorn National Forest. Lakewood, Co.: USDA Forest Service, Rocky Mountain Region. Gold AJ, PM Groffman, K Addy, DQ Kellogg, and AE Rosenblatt. 2000. The role of landscape setting in riparian groundwater nitrate removal. In Proceedings of the International Conference on Riparian Ecology and Management in Multi-Use Watersheds, 113-118. P.J. Wigington, Jr. and R.L. Beschta, eds. Middleburg, Va.: AWRA. 366 Goldstein R.M., J.C. Stauffer, P.R. Larson and D.L. Lorenz. 1996. Relation of physical and chemical characteristics of stream to fish communities in the Red River of the North Basin, Minnesota and North Dakota, 1993-1995. US Geological Survey Water-Resources Investigations Report 96-4227. Mounds View, Minnesota: USGS. Gregory S., V. Swanson, J. Frederick and W.A. McKee. 1991. An ecosystem perspective of riparian zones: focus on links between land and water. BioScience 41: 540-551. Groffman P.M. and J.M. Tiedje. 1989. Denitrification in north temperate forest soils: relationships between denitrification and environmental factors at the landscape scale. Soil Biology and Biochemistry 21: 621-626. Groffman P.M., D.R. Zak, S. Christensen, A. Mosier and J.M. Tiedje. 1993. Early spring nitrogen dynamics in a temperate forest landscape. Ecology 74: 1579-1585. Haith D.A. and L.L. Shoemaker. 1987. Generalized watershed loading functions for stream flow nutrients. Water Resources Bulletin 23: 471-478. Hall L.W., M.C. Scott, W.D. Killen, R.D. Anderson. 1996. The effects of land-use characteristics and acid sensitivity on the ecological status of Maryland Coastal Plain streams. Environmental Toxicology and Chemistry 15(3): 384-394. Hall L.W. Jr., R.P. Morgan, E.S. Perry and A. Waltz. 1999. Development of a provisional physical habitat index for Maryland freshwater streams. CBWP-MANTA- EA-99-12. Annapolis, Md.: Maryland Department of Natural Resources, Resource Assessment Service, Monitoring and Non-Tidal Assessment Division. Harding J.S., E.F. Benfield, P.V. Bolstad, G.S. Helfman and E.B.D. Jones III. 1998. Stream biodiversity: the ghost of land use past. Publ. No. USA 95: 14843-14847. Proceedings of the National Academy of Science. Hauer F.R. and R.D. Smith. 1998. The hydrogeomorphic approach to functional assessment of riparian wetlands: evaluating impacts and mitigation on river floodplains in the U.S.A. Freshwater Biology 40(3): 517-530. Haycock N.E. and A.D. Muscutt. 1995. Landscape management strategies for the control of diffuse pollution. Landscape and Urban Planning 31: 313-321. Haycock N.E. and G. Pinay. 1993. Groundwater nitrate dynamics in grass and poplar vegetated riparian buffer strips during the winter. Journal of Environmental Quality 22: 273-278. 367 Heng H.H. and N.P. Nikolaidis. 1998. Modeling of nonpoint source pollution of nitrogen at the watershed scale. Journal of the American Water Resources Association 34(2): 359-374. Herlihy A. and P. Kaufman. 2000. Riparian vs. watershed influence on macroinvertebrates, fish and chemical indicators of stream condition. Poster at the International Conference on Riparian Ecology and Management in Multi-Use Watersheds. Portland, Oregon, 8/27-31, 2000. Hewitt M.J. 1990. Synoptic inventory of riparian ecosystems: the utility of Landsat Thematic Mapper data. Forest Ecology and Management 33/34: 605-620. Hewlett J.D. and J.C. Fortson. 1982. Stream temperature under an inadequate buffer strip in the southern Piedmont. Water Resources Bulletin 18(6): 983-988. Hill A.R. 1978. Factors affecting the export of nitrate-nitrogen from drainage basins in southern Ontario. Water Research 12: 1045-1057. Hughes R.M. and D.P. Larsen. 1988. Ecoregions: an approach to surface water protection. Journal of the Water Pollution Control Federation 60(4): 486-493. Hunsaker C.T. and D.A. Levine. 1995. Hierarchical approaches to the study of water quality in rivers. Bioscience 45: 193-201. Hynes H.B.N. 1975. The stream and its valley. Verhandlungen, Internationale Vereinigugn fur Theoretische und Augewandte Limnologie 19: 1-25. Imhof J.G., J. Fitzgibbon, W.K. Annable. 1996. A hierarchical evaluation system for characterizing watershed ecosystems for fish habitat. Canadian Journal of Fisheries and Aquatic Sciences 53(SU1): 312-326. Inamdar S.P., J.M. Sheridan, R.G. Williams, D.D. Bosch, R.R. Lowrance, L.S. Altier and DL Thomas. 1999. Riparian ecosystem management model (REMM): I. testing of the hydrologic component for a Coastal Plain riparian system. Transactions of the ASAE 42(6): 1679-1689. Inamdar S.P., R.R. Lowrance, L.S. Altier, R.G. Williams and R.K. Hubbard. 1999. Riparian ecosystem management model (REMM): II. Testing of the water quality and nutrient cycling component for a coastal plain riparian system. Transactions of the ASAE 42(6): 1691-1707. Johanson R.C. 1983. A new mathematical modeling system Hydrological Simulation Program - Fortran (HSPF). American Chemical Society Symposium Series 225: 125- 147. 368 Johnson L.B. and S.H. Gage. 1997. Landscape approaches to the analysis of aquatic ecosystems. Freshwater Biology 37(1): 113-132. Johnston C.A., N.E. Detenbeck and G.J. Niemi. 1990. The cumulative effect of wetlands on stream water quality and quantity: a landscape approach. Biogeochemistry 10: 105-141. Jordan T.E. and D.E. Weller. 1996. Human contributions to terrestrial nitrogen flux. BioScience 46: 655-664. Jordan T.E., D.L. Correll and D.E. Weller. 1993. Nutrient interception by a riparian forest receiving inputs from adjacent cropland. Journal of Environmental Quality 22: 467-473. _______. 1997a. Effects of agriculture on discharges of nutrients from Coastal Plain watersheds of the Chesapeake Bay. Journal of Environmental Quality 26(3): 836-848. _______. 1997b. Nonpoint source discharges of nutrients from Piedmont watersheds of the Chesapeake Bay. Journal of the American Water Resources Association 33(3): 631-645. _______. 1997c. Relating nutrient discharges from watersheds to land use and streamflow variability. Water Resources Research 33(11): 2579-2590. Kazyak P.F. 1997. Maryland Biological Stream Survey Sampling Manual. Columbia, Md.: Maryland Department of Natural Resources, Monitoring and Non-Tidal Assessment Division. Klauda R., P.F. Kazyak, S.A. Stranko, M.T. Southerland, N.E. Roth and J.C. Chaillou. 1998. Maryland Biological Stream Survey: a state agency program to assess the impact of anthropogenic stresses on stream habitat, quality and biota. Environmental Monitoring and Assessment 51(1-2): 299-316. Kovalchik B.L. 1987. Riparian zone associations of the Deschutes, Ochoco, Fremont and Winema National Forests. Publ. No. R6 ECOL TP-279-87. Portland, Or: USDA Forest Service, Pacific Northwest Region 6. Kovalchik B.L. and L.A. Chitwood. 1990. Use of geomorphology in the classification of riparian plant associations in mountainous landscapes of central Oregon, USA. Forest Ecology and Management 33/34: 405-418. Lammert M. and J.D. Allan. 1999. Assessing biotic integrity of streams: Effects of scale in measuring the influence of land use/cover and habitat structure on fish and macroinvertebrates. Environmental Management 23(2): 257-270. 369 Large A.R.G. and G.E. Petts. 1992. Buffer zones for conservation of rivers and bankside habitats. Publ. No. BS12 2UD, project 340/5/y. Bristol, England: National Rivers Authority. Larsen D.P., R.M. Omernik, R.M. Hughes, C.M. Rohm, T.R. Whitier, A.J. Kinney, A.L. Gallant and D.R. Dudley. 1986. Correspondence between spatial patterns in fish assemblages in Ohio streams and aquatic ecoregions. Environmental Management 10(6): 815-828. Lenat D.R. 1984. Agriculture and stream water quality, a biological evaluation of erosion control practices. Environmental Management 8: 333-344. Likens G.E. and F.H. Bormann. 1974. Linkages between terrestrial and aquatic ecosystems. Bioscience 24: 447-456. Loehr R.C. 1974. Characteristics and comparative magnitude of non-point sources. Journal of the Water Pollution Control Federation 46(8): 1849-1871. Lowrance R.R., L.S. Altier, J.D. Newbold, R.R. Schnabel, P.M. Groffman, J.M. Denver, D.L. Correll, J.W. Gilliam, J.L. Robinson, R.B. Brinsfield, K.W. Staver, W. Lucas and A.H. Todd. 1997. Water quality functions of riparian forest buffers in Chesapeake Bay watersheds. Environmental Management 21(5): 687-712. Lowrance R.R., L.S. Altier, R.G. Williams, S.P. Inamdar, J.M. Sheridan, D.D. Bosch, R.K. Hubbard and D.L. Thomas. 2000. REMM: The riparian ecosystem management model. Journal of Soil and Water Conservation 55(1): 27-34. Lowrance R.R. and R.A. Leonard. 1988. Streamflow nutrient dynamics on Coastal Plain watersheds. Journal of Environmental Quality 18(4): 734-740. Lowrance R.R. and A. Shirmohammadi. 1985. REM: a model for riparian ecosystem management in agricultural watersheds. General technical report RM - 120: 237-240. Fort Collins, Co.: USDA Forest Service, Rocky Mountain Forest and Range Experiment Station. Lowrance R.R., R. Todd, J. Fail Jr., O. Hendrickson Jr., R. Leonard and L. Asmussen. 1984. Riparian forests as nutrient filters in agricultural watersheds. BioScience 34: 374-377. Lyon J. and C.L. Sagers. 1998. Structure of herbaceous plant assemblages in a forested riparian landscape. Plant Ecology 138(1): 1-16. Maddock I. 1999. The importance of physical habitat assessment for evaluating river health. Freshwater Biology 41: 373-391. 370 McMahon G. and O.B. Lloyd. 1995. Water-quality assessment of the Albemarle- Pamlico drainage basin, North Carolina and Virginia ? environmental setting and water-quality issues. US Geological Survey Open-File Report 95-136. McMahon G. and D.A. Harned. 1998. Effect of environmental setting on sediment, nitrogen, and phosphorus concentrations in Albemarle-Pamlico drainage basin, North Carolina and Virginia, USA. Environmental Management 22: 887-903. Maryland Department of Natural Resources. Riparian Forest Buffers: Function and Design for Protection and Enhancement of Water Resources. Available at: www.dnr.state.md.us/forests/publications/buffers.html. Accessed on 20 June 2002. Mercurio G., J.C. Chaillou and N.E. Roth. 1999. Guide to using 1995-1997 Maryland Biological Stream Survey Data. Publ. No. CBWP-MANTA-EA-99-5. Annapolis, Md.: Maryland Department of Natural Resources, Monitoring and Non-Tidal Assessment Division. Miltner R.J. and E.T. Rankin. 1998. Primary nutrients and the biotic integrity of rivers and streams. Freshwater Biology 40(1): 145-158. Molles M.C. 1982. Trichopteran communities of streams associated with aspen and conifer forests: long-term structural change. Ecology 63: 1-6. Muller E., H. Decamps and M.K. Dobson. 1993. Contribution of space remote-sensing to river studies. Freshwater Biology 29(2): 301-312. Naiman R.J. and H. D?camps. 1997. The ecology of interfaces: riparian zones. Annual Review of Ecology and Systematics 28: 621-658. Naiman R.J., H. D?camps, J. Pastor and C.A. Johnston. 1988. The potential importance of boundaries to fluvial ecosystems. J. N. Am. Benthol. Soc. 7(4): 289-306. Nearing M.A., R.M. Risse and L.F. Rogers. 1993. Estimation of daily nutrient fluxes to a large Piedmont reservoir from limited tributary data. Journal of Environmental Quality 22: 666-671. Neill M. 1989. Nitrate concentrations in river waters in the southeast of Ireland and their relationship with agricultural practice. Water Research 23: 1339-1355. Nikolaidis N.P., H. Shen, H.U. Heng and J.C. Clausen. 1993. Movement of nitrogen through an agricultural riparian zone ? 2. Distributed modeling. Water Science and Technology 29(3-5): 613-623. 371 Norton B.E., J. Tuhy and S. Jensen. 1981. An approach to classification of riparian vegetation. In In-Place Resource Inventories: Principles & Practices: Proceedings of a Workshop. T.B. Brann, L.O. House IV, H.G. Lund, eds. Orono, Me: Univ. of Maine. Norton B.E., P.J. Urness and A.R. Southard. 1980. A riparian community classification study: final report. Logan, Utah: Utah State University, Dept. of Range Science. Novotny V. and H. Olem. 1994. Water quality: prevention, identification, and management of diffuse pollution. Van Nostrand Reinhold, New York, 1054 pp. Omernik J.M., A.R. Abernathy and L.M. Male. 1981. Stream nutrient levels and proximity of agricultural and forest land to streams: some relationships. Journal of Soil and Water Conservation 36: 227-231. Omernik J.M. 1976. The influence of land use on stream nutrient levels. Publ. No. EPA-600/3-76-014. Washington, DC: U.S. EPA. Ormerod S.J., S.D. Rundle and E.C. Lloyd. 1993. The influence of riparian management on the habitat structure and macroinvertebrate communities of upland streams draining plantation forests. The Journal of Applied Ecology 30(1): 13-24. Osborne L.L., S.L. Kohler, P.B. Bayley, D.M. Dat, W.A. Bertrand, M.J. Wiley and R. Sauer. 1992. Influence of stream location in a drainage network on index of biotic integrity. Transactions of the American Fisheries Society 121(5): 635-645. Osborne L.L. and D.A. Kovacic. 1993. Riparian vegetated buffer strips in water- quality restoration and stream management. Freshwater Biology 29: 243-258. Osborne L.L. and M.J. Wiley. 1988. Empirical relationships between land use/cover and stream water quality in an agricultural watershed. Journal of Environmental Management 26: 9-27. _______. 1992. Influence of tributary spatial position on the structure of warmwater fish communities. Can. J. of Fish. Aquatic Sci. 49(4): 671-681. Osmond D.L., R.W. Gannon, J.A. Gale, D.E. Line, C.B. Knott, K.A. Phillips, M.H. Turner, M.A. Foster, D.E. Lehning, S.W. Coffey and J. Spooner. 1997. WATERSHEDSS: A decision support system for watershed-scale nonpoint source water quality problems. Journal of the American Water Resources Association 33(2): 327-341. Padgett W.G., A.P. Youngblood and A.H. Winward. 1989. Riparian community type classification of Utah and southeastern Idaho. Ogden, Utah: U.S. Dept. of Agriculture, Forest Service, Intermountain Research Station. 372 Palone R.S. and A.H. Todd. 1998. Chesapeake Bay riparian handbook: a guide for establishing and maintaining riparian forest buffers. U.S. Dept. of Agriculture, Forest Service, Northeastern Area State & Private Forestry: Natural Resources Conservation Services: Cooperative State Research, Education, and Extension Service. Perry C.D., G. Vellidis, R.R. Lowrance and D.L. Thomas. 1999. Watershed-scale water quality impacts of riparian forest management. Journal of Water Resources Planning and Management - ASCE 125(3): 117-125. Peterjohn W.T. and D.L. Correll. 1984. Nutrient dynamics in an agricultural watershed: observations on the role of a riparian forest. Ecology 65(5): 1466-1475. Petersen M.M., D.J. Somerville and R.F. Sennett. 1992. Preliminary riparian classification system for private and state-owned lands in Utah. General technical report INT 289. Ogden, Utah: U.S. Department of Agriculture, Forest Service, Intermountain Research Station. Phillips J.D. 1989. Non-point-source pollution control effectiveness of riparian forests along a Coastal Plain river. Journal of Hydrology 110: 221-237. Pierce J. and J. Johnson. 1986. Wetland community type classification of Utah. Publ. No. R4-Ecol-89-01. Ogden, Ut.: USDA Forest Service, Intermountain Research Station. Pinay G., C. Ruffini and A. Fabre. 1993. Nitrogen cycling in two riparian forests soils under different geomorphic conditions. Biogeochemistry 30: 9-29. Pinay G., L. Roques and A. Fabre. 1993. Spatial and temporal patterns of denitrification in a riparian forest. The Journal of Applied Ecology 30(4): 581-591. Plafkin J.L., M.T. Barbour, K.D. Porter, S.K. Gross and R.M. Hughes. 1989. Rapid bioassessment protocols for use in streams and river: benthic macroinvertebrates and fish. Publ. No. EPA/444/4-89-011. Washington, D.C.: EPA, Office of water regulations and standards. Poff N.L. and J.D. Allan. 1995. Functional organization of stream fish assemblages in relation to hydrologic variability. Ecology 76: 606-627. Quinn J.M. and C.W. Hickey. 1990. Characterization and classification of benthic invertebrate communities in 88 New Zealand rivers in relation to environmental factors. New Zealand Journal of Marine and Freshwater Research 24: 387-409. 373 Quinn J., A. Taylor, W. Boyce and T. Fenton. 2000. Riparian zone classification improves management in multi-land use watersheds. In Proceedings of the International Conference on Riparian Ecology and Management in Multi-Use Watersheds, 215-220. P.J. Wigington, Jr. and R.L. Beschta, eds. Middleburg, Va.: AWRA. Richards C. and G. Host. 1994. Examining land-use influences on stream habitats and macroinvertebrates ? a GIS approach. Water Resources Bulletin 30(4): 729-738. Richards C., R.J. Haro, L.B. Johnson and G.E. Host. 1997. Catchment and reach-scale properties as indicators of macroinvertebrate species traits. Freshwater Biology 37: 219-230. Richards C., G.E. Host and J.W. Arthur. 1993. Identification of predominant environmental factors structuring stream macroinvertebrate communities within a large agricultural catchment. Freshwater Biology 29(2): 285-294. Richards C., L.B. Johnson and G.E. Host. 1996. Landscape-scale influences on stream habitats and biota. Canadian Journal of Fisheries and Aquatic Sciences 53: 295-311. Roberts G. and T. Marsh. 1987. The effects of agricultural practices on the nitrate concentration in the surface water domestic supply sources of Western Europe. IAHS Publication No. 164. Roth N.E., M.T. Southerland and J.C. Chaillou. 1999. Relative and cumulative impacts of stressors on streams: 1995-1997 MBSS results. Publ. No. . Annapolis, Md.: Maryland Department of Natural Resources, Monitoring and Non-Tidal Assessment Division. Roth N.E., M.T. Southerland, G. Mercurio, J.C. Chaillou, P.F. Kazyak, S.S. Stranko, A.P. Prochaska, D.G. Heimbuch and J.C. Seibel. 1999. State of the Streams: 1995- 1997 Maryland Biological Stream Survey Results. Publ. No. CBWP-MANTA-EA-99- 6. Annapolis, Md.: Maryland Department of Natural Resources, Monitoring and Non- Tidal Assessment Division. Roth N.E., M.T. Southerland, J.C. Chaillou, H.T. Wilson, D.G. Heimbuch and J.C. Seibel. 1998. Maryland Biological Stream Survey: Ecological Assessment of Non- Tidal Streams Sampled in 1996. Publ. No. CBWP-MANTA-EA-98-1. Annapolis, Md.: Maryland Department of Natural Resources, Monitoring and Non-Tidal Assessment SAS Institute Inc. 1988. SAS/STAT User?s Guide. Release 6.03 Edition. Cary, N.C.: SAS Institute Inc. 374 Schlosser I.J. and P.L. Angermeier. 1995. Spatial variation in demographic processes of lotic fishes: conceptual models, empirical evidence, and implications for conservation. American Fisheries Society Symposium 17: 392-401. Skaggs R.W. and G.M. Chescheir. 1999. Drains in depth. ASAE Resource Sept 1999:13-14. Snyder N.J., S. Mostaghimi, D.F. Berry, R.B. Reneau, S. Hong, P.W. McClellan and E.P. Smith. 1998. Impact of riparian forest buffers in agricultural nonpoint source pollution. Journal of the American Water Resources Association 34(2): 385-395. Southerland M.T., G. Mercurio, N.E. Roth, D. Strebel and J. Volstad. 2002. Using the land use-stream condition relationship to target riparian restoration. Presentation. By Versar, Inc. Columbia, Md. Steedman R.J. 1988. Modification and assessment of an index of biotic integrity to quantify stream quality in Southern Ontario. Can. J. Fish. Aquat. Sci. 45: 492-501. Stewart J.S., D.M. Downes, L. Wang, J.A. Wier and R. Bannerman. 2000. Influences of riparian corridors on aquatic biota in agricultural watersheds. In Proceedings of the International Conference on Riparian Ecology and Management in Multi-Use Watersheds, 209-214. P.J. Wigington, Jr. and R.L. Beschta, eds. Middleburg, Va.: AWRA. Stribling J.B., B.K. Jessup, J.S. White, D.M. Boward and M. Hurd. 1998. Development of a Benthic Index of Biotic Integrity for Maryland Streams. CBWP- EA-98-3. Annapolis, Md.: Maryland Department of Natural Resources, Resource Assessment Service, Monitoring and Non-Tidal Assessment Division. Szaro R.C. and D.R. Patton. 1986. Riparian habitat classification in the southwestern United States. In Transactions of the 51st North American Wildlife and Natural Resources Conference. Washington, D.C.: Wildlife Management Institute. Tabacchi E., D.L. Correll, R. Hauer, G. Pinay, A. Planty-Tabacchi and R.C. Wissmar. 1998. Development, maintenance and role of riparian vegetation in the river landscape. Freshwater Biology 40: 497-516. Taraba J.L., J.S. Dinger, L.V.A. Sendlein and G.K. Felton. 1996. Land use impacts on water quality in small karst agricultural watersheds. ASAE meeting paper 96-2086. July 14-18, 1996. Phoenix, AZ. Taylor C.M., M.R. Winston and W.J. Matthews. 1993. Fish species ? environment and abundance relationships in a Great Plains river system. Ecography 16: 16-23. 375 Tim U.S., R. Jolly and H.H. Liao. 1995. Impact of landscape feature and feature placement on agricultural non-point-source-pollution control. Journal of Water Resources Planning and Management - ASCE 121(6): 463-470. Tufford D.L., H.N. McKellar Jr. and J.R. Hussey. 1998. In-stream nonpoint source nutrient prediction with land-use proximity and seasonality. Journal of Environmental Quality 27(1): 100-111. USDA. 1999. Assessing conditions of riparian-wetland corridors at the areawide level. Watershed Science Institute Technical Report. Washington, D.C.: USDA NRCS. US EPA. 1990. The quality of our nation's water: a summary of the 1988 National Water Quality Inventory. Publ. No. EPA 440/4-90/003. Washington, D.C.: EPA, Office of Water. US Geological Survey. 1999. The quality of our nation's waters - nutrients and pesticides. US Geological Survey circular 1225. Vandervalk A.G. and R. Jolly. 1992. Recommendations for Research to Develop Guidelines for Use of Wetlands to Control Rural Non-Point Source Pollution: Created and Natural Wetlands for Controlling Nonpoint Source Pollution. R.K. Olson, ed. Washington, DC: US EPA, Office of Research and Development. Vannote R.L., G.W. Minshall, K.W. Cummins, J.R. Sedell and C.E. Cushing. 1980. The river continuum concept. Can. J. Fish. Aquat. Sci. 37: 130-137. Weil R.R., R.A. Weismiller and R.S. Turner. 1990. Nitrate contamination of groundwater under irrigated Coastal Plain souls. J. Environ. Quality 19: 441-448. Weller D.E., T.E. Jordan and D.L. Correll. 1998. Heuristic models for material discharge from landscapes with riparian buffers. Ecological Applications 8(4): 1156- 1169. Whigham D.F., C. Chitterling and B. Palmer. 1988. Impacts of freshwater wetlands on water quality ? a landscape perspective. Envir. Management 12(5): 663-671. Wichert G.A. and D.J. Rapport. 1998. Fish community structure as a measure of degradation and rehabilitation of riparian systems in an agricultural drainage basin. Environmental Management 22(3): 425-443. Winward A.H. 1984. An ecosystem classification for intermountain riparian areas. In Range Watersheds, Riparian Zones and Economics: Interrelationships in Management and Use: Proceedings, 1984 Pacific Northwest Range Management Short Course. Corvallis, Or: Oregon State University. 376 Wissmar R.C. and R.L. Beschta. 1998. Restoration and management of riparian ecosystems: a catchment perspective. Freshwater Biology 40(3): 571-585. Wohl N.E. and R.F. Carline. 1996. Relations among riparian grazing, sediment loads, macroinvertebrates, and fishes in three central Pennsylvania streams. Can. J. Fish. Aquat. Sci. 53(Suppl.1): 260-266. Youngblood A.P., W.G. Padgett, G. Wayne and A.H. Winward. 1985. Riparian community type classification of eastern Idaho - western Wyoming. Publ. No. R4- ECOL-85-01. Ogden, Ut.: U.S. Dept. of Agriculture, Forest Service, Intermountain Forest Range Experimental Station. Zinecker E. 2001. Poster at the Chesapeake Bay Watershed Restoration: Riparian and Wetland Stewardship Conference. Baltimore, Maryland. September 24-26, 2002 CURRICULUM VITAE Linda S. Barker barkerlinda@yahoo.com RESEARCH INTERESTS Stock assessment studies of finfish and other aquatic species of the Chesapeake Bay and its tributaties. Quantification of the impact of riparian buffers on channel morphology, streamflow patterns, water-quality and instream benthic communities of low-order non-tidal streams. Data mining and statistical analysis of existing environmental databases to determine environmental patterns and answer questions beyond the scope of the original design of the data. Development of biological engineering and environmental education techniques and curricula. FORMAL EDUCATION 1998-2003. University of Maryland, College Park, MD Ph.D. in Biological Resources Engineering ? Water Resources Specialty. GPA 4.0 1997-1998. University of Maryland at Baltimore County, Baltimore, MD One year of graduate classes in the Department of Biological Sciences. GPA 4.0 1988-1991. McNeese State University, Lake Charles, LA M.S., Environmental Science. GPA 4.0 1974-1978. Tulane University, New Orleans, LA B.S., Chemical Engineering and Pre-Med Studies. GPA 3.0 PROFESSIONAL EXPERIENCE - Technical Research Statistician VI Maryland Department of Natural Resources, Fisheries Program. November 2003 ? Conduct statistical analyses and modeling, design sampling plans, review reports and proposals to support management of Chesapeake Bay resident and migratory species. Research Statistician VI Maryland Department of Natural Resources, Fisheries Service, Striped Bass Stock Assessment. January ? October 2003. Conduct statistical analyses, design sampling plans, and prepare reports and presentations to support management of the Chesapeake Bay and coastal migratory stock of Striped Bass (Morone saxatilis). Independent Research In association with the Biology Department, U.N.A.H., Tegucigalpa, Honduras. 2002. Bio-hydrology and aquatic invertebrate studies of tropical low-order streams in the Rio Platano Biosphere Reserve of Honduras. Research Assistant and Doctoral Candidate University of Maryland, Department of Biological Resources Engineering. Designed and coordinated the first large-scale, ground-truthed agricultural riparian buffer survey in Maryland. Designed fieldwork protocols and training materials, conducted fieldwork, trained and coordinated oversight of 8 other field teams, and helped to network with the appropriate agricultural agencies in the targeted agricultural watersheds. Managed data QA/QC & conducted statistical analysis using SAS & Splus. Used this database and the Md DNR MBSS database to interpret land use patterns, physical and biological systems, and hydrology in agricultural watersheds. Developed environmental models for the prediction of instream nitrate, biological and physical stream health parameters. Developed and presented technical and scientific papers at national conferences and for state agencies and committees. Expert Witness Eggleston and Wolf, Baltimore, Maryland. 1997. Provided information that led to the out-of-court settlement of a drycleaner case. Director ? Small Business Assistance Program. Maryland Department of the Environment. 1993-1997. Integrated air- with water- and waste- compliance requirements at both the state and federal levels with pollution prevention opportunities. Created outreach programs for small businesses affected by environmental regulations. Initiated and coordinated cooperative partnerships with affected industries, conducted outreach and training for industry and state personnel. Handbooks were used as blueprints for similar programs in other states. Two awards of appreciation from local business organizations. Environmental Specialist Louisiana Dept. of Environmental Quality, Lake Charles, LA. 1991 Supervised ambient air quality monitoring for LA?s southwest district. Provided technical training for air quality compliance personnel. Technical Specialist Dearborn Chemical Division, Houston District. 1990. Provided technical support for water-quality and industrial equipment. Developed and conducted training for engineers, operators and maintenance personnel. Environmental Project Engineer W.R. Grace Chemical Plant, Lake Charles, LA. 1989. As part of a cooperative program with McNeese State University. Analyzed industrial processes & made recommendations for minimization of wastewater discharge and effluent toxicity. Designed, constructed and operated an on-site industrial wastewater treatment pilot plant & designed a whole-plant WWT facility. Family leave-of-absence. 1980-1988. Provided care for my two young sons, one of whom is severely handicapped. Chemical Engineer PPG Industrial Chemical Plant, Lake Charles, LA. 1979-1980. Performed industrial process and wastewater discharge analysis projects. PROFESSIONAL EXPERIENCE - Teaching Departmental Instructor University of Maryland, Dept. of Biological Resources Engineering. 2000 - 2001. Taught ENBE 100 "Introduction to Biological Resources Engineering", a multi- disciplinary introduction to engineering principles core course for non-engineers. Teaching Assistant University of Maryland at Baltimore County. Fall 1998. Oversight for 5 "Discussion Sections" (100 students) of Biology 101, the introductory Biology course for Biology majors. This course is taught under a NSF grant to explore "active learning" and interactive teaching. Departmental Instructor Essex Community College. 1994-1997. Designed and taught the introductory Biology lecture and lab courses for students in the Med- Vet- Tech program and Introduction to Biology for non-science majors. Departmental Instructor Loyola College of Baltimore, Biology Department. 1993-1994. Taught 4 sections (100 students) per semester of Biology lab for Biology majors. Teaching Assistant and Lab Coordinator. McNeese State University, Biology Department. 1988-1991. Taught 4 sections (100 students) per semester of the Zoology lab course (Bio 102L). PROFESSIONAL PUBLICATIONS Barker, L.S. and H. Speir. 2003. Effects of size limit increases on the harvest of striped bass in the Chesapeake Bay. Fisheries Technical Report No. 43. Maryland Dept. of Natural Resources, Annapolis. Md. Barker L.S. and G.K. Felton 2002. Modeling the effects of agricultural riparian buffers to mitigate nitrate stream pollution. ASAE Paper 02-2044. ASAE Annual International Meeting, Chicago, Ill. July 28-31. Barker, L.S., G.K. Felton, and E. Russek-Cohen. 2001. Use of a classification system to evaluate the impact of agricultural riparian buffers on instream nitrate concentrations. Presented at the annual conference of the American Water Resources Association. Albuquerque, New Mexico. November 12-15. p. 27. Montas, H.H., L.B. Moran and G.K. Felton. 2000. Water-quality impact of agricultural riparian buffers in a small Maryland watershed. Presented at the annual conference of ASAE. Chicago, Illinois. July 2000. Moran, L.B. 1997. Guide to Compliance and P2 for Small Maryland Printers. MDE. Baltimore, Md. Moran, L.B. 1996. Guide to Compliance and P2 for Chrome Platers in Maryland. MDE. Baltimore, Md. Taddeo, L. and L.B. Moran. 1995. Guide to Compliance and P2 for Autobody Shops in Maryland. MDE. Baltimore, Md. Moran, L.B. 1994. Guide to Compliance and P2 for Maryland Drycleaners. MDE. Baltimore, Md. PROFESSIONAL PRESENTATIONS Barker, L.S. 2003. Statistical analysis in stock assessment of Chesapeake Bay striped bass management. Presented as an ENBE departmental seminar. College Park, Md. April 2, 2003. Barker, L.S. 2003. Use of the Maryland Biological Stream Survey (1995-1997) data to quantify the effectiveness of agricultural riparian buffers. Presented at the Mid- Atlantic Water Pollution Biology Workshop. Cacapon, West Virginia. Barker, L.S. 2002. The effectiveness of agricultural riparian buffers to mitigate nitrate stream pollution. Presented at the Chesapeake Bay Watershed Conference. Baltimore, Maryland. September 24-26, 2002. Barker, L.S. 2002. Agricultural buffers in Maryland and their effect on chemical, biological and physical measures of stream health. Presented by invitation to members of the Maryland CREP Advisory Panel. Annapolis, Maryland. August 29, 2002. Barker, L.S. and G.K. Felton. 2002. Modeling the effects of agricultural riparian buffers to mitigate nitrate stream pollution. Presented at the annual conference of the American Society of Agricultural Engineers. Chicago, Ill. July 26-31, 2002. Barker, L.S. 2002. Agricultural buffers in Maryland. Presented by invitation to the Maryland State Soil Conservation Committee. Annapolis, Maryland. June 20, 2002. Barker, L.S. 2002. Agricultural buffers in Maryland and their effect on in-stream nitrate levels. Presented by invitation to the Maryland ReLeaf Committee. Annapolis, Maryland. February 7, 2002. 1994-1996. Over 50 presentations for Maryland's small business professional organizations - environmental regulatory compliance and pollution prevention opportunities. 1995-1997. Yearly presentations representing MDE at National Small Business Assistance conferences. HONORS AND AWARDS - Distinguished Teaching Assistant. Dept. of Biol. Resources Engineering. 2002. - Ilene Nagel Travel Grant for research in Honduras. U. Md. Graduate School. 2001. - J.K. Goldhaber Travel Grant for research in Honduras. U. Md. Grad. School. 2002. - Invited Reviewer. American Women in Science. 2001. - Olive Lynn Salembier Re-entry Scholarship. Society of Women Engineers. 2000. - S. Bufton Memorial Education Scholarship. Amer. Business Women?s Assn. 2000. - Ruth Satter Fellowship. American Women in Science Educational Fdn. 2000. - Elected to Sigma Xi (international scientific research honor society). 2000. - Richard A Herbert Graduate Scholarship. American Water Resources Assn. 1999. - Elected to Alpha Epsilon (ASAE honor society). 1999. - Award of Appreciation. Korean Drycleaners Association of Maryland. 1996.