Abstract Title of Document: A CHEMICAL CLIMATOLOGY OF LOWER TROPOSPHERIC TRACE GASES AND AEROSOLS OVER THE MID-ATLANTIC REGION Jennifer Carrie Hains, Doctor of Philosophy, 2007 Directed By: Professor Russell R. Dickerson, Department of Atmospheric and Oceanic Science Ozone and aerosols affect air quality, visibility and human health. The University of Maryland research aircraft conducted flights over the Mid-Atlantic region between 1995 and 2005 to characterize pollution events. I developed a chemical climatology of trace gases and aerosols that can be used to validate and improve models. O 3 and SO 2 measured aboard the aircraft were compared with O 3 and SO 2 generated with the Community Multiscale Air Quality (CMAQ). In general, CMAQ under-estimates O 3 above 500 m and over-estimates O 3 below 500 m (possible reasons for this include chemistry not being properly represented in the model). A sensitivity test of the rate of photolysis of NO 2 was performed and improving the photochemistry did improve the modeled O 3 . CMAQ over-predicts the SO 2 column content by about 50%, possibly because the model gives SO 2 too long a lifetime. To test this theory I developed a method for calculating the SO 2 lifetime using in-situ measurements. The mean SO 2 lifetime was 19 ? 7 hours for measurements made in the daytime in the summer in the Mid-Atlantic region with in- cloud processes responsible for ~80% of the removal. I made comparisons of three aerosol sampling systems and found the uncertainty of PM 2.5 , sulfate, and ammonium measured with the Speciation Trends Network is larger than what has been reported and is at least 20%. I have developed clustering methodologies to group back trajectories associated with aircraft profiles as well as group trace gas and aerosol profiles by size and shape. The first clustering method produced eight distinct meteorological regimes associated with pollution and haze events. I quantified the amount of O 3 transported for each meteorological regime. Using the second method, I found a strong correlation between O 3 profiles and point source NOx emissions. The comparisons of model and measured profiles, comparisons of surface measurements, and clustering methods are used to explain sources, sinks and distributions of trace gases and aerosols in the mid-Atlantic thus improving the understanding of the lower atmospheric composition in this area. A CHEMICAL CLIMATOLOGY OF LOWER TROPOSPHERIC TRACE GASES AND AEROSOLS OVER THE MID-ATLANTIC REGION By Jennifer Carrie Hains 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 Doctor of Philosophy 2007 Advisory Committee: Professor Russell R. Dickerson, Chair Professor Neil Blough Professor Douglas English Professor John M. Ondov Professor Robert Hudson ? Copyright by Jennifer Carrie Hains 2007 ii Dedication To Julie, Peter, Brendan, Miss Elisabeth and Cindy Rollo. Thank you for all of your love and support. iii Acknowledgements Funding for this work was provided in part by the Maryland Department of the Environment and Constellation Energy Group, Baltimore Gas and Electric Company and Potomac Electric Power Company through the Electric Power Research Institute and Maryland Industrial Partnership. Work related to GOME was supported by the University of Bremen. Special thanks to Russell Dickerson for infinite patience and inspiring creativity. Also many thanks to Brett Taubman, Jeff Stehr, Lung-Wen Chen, Charles Piety and Lackson Marufu for great ideas and support. Many thanks to Dale Allen the IDL master. Also many thanks to Bruce Doddridge, Mian Chin, Andreas Richter, Annette Ladst?tter-Wei?enmayer, John Burrows, Anne Thompson, Peter Mueller, Robert Hudson, Tad Aburn, Diane Franks, Matthew Seybold, David Krask, Can Li, Pedro Bueno, Zahra Chaudry, Bryan Bloomer, and Rob Levy. iv Table of Contents Abstract.....................................................................................................................1 Dedication .................................................................................................................ii Acknowledgements ..................................................................................................iii Table of Contents .....................................................................................................iv List of Tables............................................................................................................vi List of Figures .........................................................................................................vii Chapter 1: Introduction..............................................................................................1 1.1 Background......................................................................................................1 1.2 Chemical Transport Models .............................................................................3 1.3 Meteorology Associated with Elevated O 3 .......................................................5 1.4 O 3 Chemistry ...................................................................................................6 1.5 SO 2 Chemistry .................................................................................................7 1.6 Determination of Meteorological Influences on Pollution Episodes: Clustering Back Trajectories...................................................................................................9 1.7 Determining the influence of Point Source on Pollution Episodes: Clustering species profiles ....................................................................................................12 1.8 Surface Measurements ...................................................................................14 1.9 Overview .......................................................................................................17 Chapter 2: Sampling Platform and Instrumentation..................................................20 2.1 Introduction ...................................................................................................20 2.2 Aircraft ..........................................................................................................20 2.3 SO 2 ................................................................................................................22 2.4 CO.................................................................................................................26 2.5 O 3 ..................................................................................................................27 2.6 Aerosol Absorption........................................................................................28 2.7 Scattering.......................................................................................................30 Chapter 3: Determination of Meteorological Influences on Pollution Episodes: Clustering Back Trajectories....................................................................................33 3.1 Introduction. ..................................................................................................33 3.2. Observations .................................................................................................34 3.2.1 Measurements .........................................................................................34 3.2.2 Trajectory Calculations ...........................................................................39 3.2.3. Cluster Analysis .....................................................................................40 3.3. Results and Discussion..................................................................................44 3.3.1 Cluster Solution ......................................................................................44 3.3.2 Pollution Profiles.....................................................................................50 3.3.3 O 3 Transport............................................................................................57 3.4. Conclusions ..................................................................................................59 Chapter 4: Cluster Analysis of Pollutant Profiles. ....................................................62 4.1 Introduction ...................................................................................................62 4.1.1 Background.............................................................................................62 4.1.2 Cluster Analysis ......................................................................................62 4.2 Results...........................................................................................................64 v 4.2.1 O 3 ............................................................................................................64 4.2.2 SO 2 .........................................................................................................71 4.2.3 Particle Scattering ...................................................................................74 4.2.4 Angstrom Exponent ................................................................................79 4.2.5 CO ..........................................................................................................82 4.2.6 Particle Absorption..................................................................................85 4.3 Discussion .....................................................................................................88 4.4 Conclusions ...................................................................................................92 Chapter 5: Comparisons of University of Maryland Aircraft and Trace Gas Profiles with Models CMAQ and GOCART.........................................................................94 5.1 Introduction ...................................................................................................94 5.1.1 Background.............................................................................................94 5.1.2 Description of Models.............................................................................95 5.2 Comparisons Between Models and Measurements .........................................97 5.2.1 O 3 Comparisons ......................................................................................97 5.2.2 The Effects of Aerosols on the Photolysis Rate of NO 2 and the Production of O 3 ..............................................................................................................105 5.2.3 SO 2 Comparisons ..................................................................................131 5.2.4 Lifetime Calculation..............................................................................148 5.2.5 Verification of Lifetime Equation and Results.......................................149 5.3 Conclusions .................................................................................................167 Chapter 6: A Side by Side Comparison of Filter-based PM 2.5 Measurements at a Suburban Site: A Closure Study.............................................................................169 6.1 Introduction .................................................................................................169 6.1.1 Background...........................................................................................169 6.1.2 Experiment............................................................................................170 6.2 Results and Discussion.................................................................................177 6.2.1 Uncertainty Analysis .............................................................................177 6.2.2 Gravimetric Mass Comparisons.............................................................184 6.2.3 Chemical Compositions.........................................................................189 6.2.4 Mass Closure.........................................................................................193 6.3 Conclusions .................................................................................................198 Chapter 7: Conclusions.........................................................................................201 7.1 Summary .....................................................................................................201 7.2 Recommendations for Future Work..............................................................205 Appendix A...........................................................................................................207 Appendix B ...........................................................................................................215 vi List of Tables Chapter 2. Table 1. Years when trace gases and aerosols were sampled. 22 Chapter 3. Table 1. Cluster median profile ranks and % O 3 transported. 48 Table 2. Cluster median profile values for 5500 ? and AOD. 49 Table 3. Statistical difference among cluster values. 50 Chapter 4. Table 1. Altitude bins used in Equation 1. 64 Table 2. SO 2 /CO ratios for O 3 Clusters. 90 Chapter 5. Table 1. CMAQ and aircraft O 3 column contents. 99 Table 2. Statistics for aerosol optical depth. 107 Table 3. Median O 3 column contents for the test episode. 119 Table 4. The average aircraft and CMAQ SO 2 column content. 133 Table 5. The average aircraft and GOCART SO 2 column content. 135 Table 6. Location of box edges (from Figure 32). 154 Table 7. Statistics of SO 2 lifetimes generated with 8 hr model. 157 Table 8. Statistics of SO 2 lifetimes generated with 16 hr model. 158 Table 9. Statistics of SO 2 lifetimes generated with 24 hr model. 159 Table 10. Statistics of SO 2 lifetimes generated with 32 hr model. 160 Table 11. Statistics for SO 2 lifetime from measurements. 164 Chapter 6. Table 1. Analytical methods for PM 2.5 species. 171 Table 2. Comparison of 2-? uncertainties for PM 2.5 species. 179 Table 3. Average PM 2.5 concentrations and uncertainties. 181-183 Table 4. Percentage of significant differences between STN RS and DRI F . 185 Table 5. Regression statistics for STN S vs. TEOM and the DRI F vs TEOM. 189 Table 6. Average reconstructed mass for STN RS and DRI F. 195 Chapter 7. Table 1. Percentage of transported O 3 . 202 Appendix A. A.4. Comparisons of trace gas column averages between aircraft and surface. 212 vii List of Figures Chapter 1 Figure 1. Counties out of compliance with O 3 standards. 1 Figure 2. Counties out of compliance withPM 2.5 standards. 2 Chapter 2 Figure 1. Locations of aircraft flights. 21 Figure 2. Flow diagram for SO 2 monitor. 24 Figure 3. Flow schematic for the CO detector. 26 Figure 4. Flow schematic for O 3 analyzer. 28 Figure 5. Flow diagram for PSAP 30 Figure 6. Flow schematic for nephelometer. 31 Chapter 3. Figures 1. Median values for O 3 , SO 2 , CO, and aerosols. 35-37 Figure 2. Percent change in the TRMSD. 44 Figure 3. Spaghetti plots of the back trajectories. 45 Figure 4. Trajectory density maps. 46 Figure 5. The median morning profiles for Clusters. 55 Figure 6. The median afternoon profiles for Clusters. 56 Figure 7. Pie chart showing the transport in the region. 58 Chapter 4. Figure 1. Median O 3 profiles for each cluster. 65 Figure 2. Back trajectory density plots for O 3 Clusters. 67 Figure 3. Circles drawn around an example back trajectory. 69 Figure 4. Statistics for NOx emissions for O 3 clusters. 70 Figure 5. Median O 3 column content and NOx emissions. 71 Figure 6. Median SO 2 profiles for each cluster. 72 Figure 7. Back trajectory density plots for SO 2 Clusters. 73 Figure 8. Statistics for SO 2 emissions for SO 2 clusters. 74 Figure 9. Median scattering profiles for each cluster. 76 Figure 10. Back trajectory density plots for scattering clusters. 77 Figure 11. Statistics for SO 2 emissions for scattering clusters. 78 Figure 12. Median Angstrom exponent profiles for each cluster. 80 Figure 13. Back trajectory density plots for Angstrom exponent clusters. 81 Figure 14. Statistics for SO 2 emissions for Angstrom exponent clusters. 82 Figure 15. Median CO profiles for each cluster. 83 Figure 16. Back trajectory density plots for CO Clusters. 84 Figure 17. Median absorption profiles for each cluster. 86 Figure 18. Back trajectory density plots for absorption Clusters. 87 Figure 19. Matching species profiles for O 3 clusters. 89 Figure 20. Matching species profiles for scattering clusters. 91 Figure 21. Matching species profiles for absorption clusters. 92 viii Chapter 5 Figure 1. Median CMAQ and aircraft O 3 profiles. 98 Figure 2. The ratio of CMAQ/Aircraft O 3. 99 Figure 3. CMAQ and aircraft O 3 profiles for smallest differences. 102 Figure 4. CMAQ and aircraft O 3 profiles for median differences. 103 Figure 5. CMAQ and aircraft O 3 profiles for largest differences. 104 Figure 6. The median Angstrom exponent for test episode. 106 Figure 7. The median single scattering albedo for test episode. 109 Figure 8. The median asymmetry parameter. 110 Figure 9. Standard and revised j-NO 2 values used in CMAQ. 111 Figure 10. O 3 profiles from the aircraft, standard and revised CMAQ runs. 114-118 Figure 11. Median CMAQ O 3 differences (standard ? revised). 119 Figure 12. Differences between revised and standard CMAQ O 3 . 121-122 Figure 13. O 3 differences (revised-standard) for a single swath. 124 Figure 14. Swath used in curtain plot (Figure 13). 125 Figure 15. Median CMAQ ozone reductions. 127-128 Figure 16. Changes in O 3 reductions. 129-130 Figure 17. Median CMAQ and aircraft profiles of SO 2 . 132 Figure 18. The ratio of median CMAQ/aircraft SO 2 . 133 Figure 19. Median GOCART and aircraft profiles of SO 2 . 134 Figure 20. The ratio of median GOCART/aircraft SO 2 . 135 Figure 21. CMAQ and aircraft SO 2 profiles for smallest differences. 137 Figure 22. CMAQ and aircraft SO 2 profiles for median differences. 138 Figure 23. CMAQ and aircraft SO 2 profiles for largest differences. 139 Figure 24. GOCART and aircraft SO 2 profiles for smallest differences. 141 Figure 25. GOCART and aircraft SO 2 profiles for median differences. 142 Figure 26. GOCART and aircraft SO 2 profiles for largest differences. 143 Figure 27. US state population and area sources of SO 2 . 145 Figure 28. The SO 2 flux from national emissions and GOCART. 146 Figure 29. CMAQ average H 2 O 2 148 Figure 30. SO 2 generated using a Gausian plume model. 152 Figure 31. The locations of the sampling points used in the model. 153 Figure 32. Boxes used to determine SO 2 flux. 154 Figure 33. Histogram of SO 2 lifetimes from model. 156 Figure 34. Back trajectories associated with the SO 2 profiles. 162 Figure 35. Histogram of SO 2 lifetimes. 164 Figure 36. Lifetime of SO 2 with respect to OH oxidation. 167 Chapter 6. Figure 1. Sampler configuration for STN R , STN S and DRI F . 173 Figure 2. Time series of PM 2.5 concentrations. 186 Figure 3. Comparisons of PM 2.5 among TEOM, DRI F and STN S . 188 Figure 4. Frequency of gravimetric ? reconstructed PM 2.5 differences. 199 Figure 5. Contributions of individual species to PM 2.5 mass. 197 ix Appendix A. A.1. Comparison of aircraft and surface measurements of O 3 . 209 A.2. Comparison of aircraft and surface measurements of SO 2. 210 A.3. Comparison of aircraft and surface measurements of CO. 211 A.5. Comparison of aircraft and surface measurements of CO (outliers removed). 214 Appendix B B.1. SO 2 absorption cross section. 217 B.2. Map of GOME SO 2 . 218 B.3. GOME and aircraft SO 2 . 220 B.4. Light is scattering through absorbing and aerosol layers. 222 B.5. Aircraft and GOME (revised AMF) SO 2 . 223 B.6. Diagram of the steps used to calculate the revised O 3 correction. 225 B.7. Aircraft and GOME (revised AMF and O 3 correction) SO 2 . 226 1 Chapter 1: Introduction 1.1 Background Numerous locations in the Mid-Atlantic US do not comply with National Ambient Air Quality Standards (NAAQS) for O 3 (80 ppb eight hour standard, formerly 120 ppb one hour standard) and PM 2.5 (15 ?g/m 3 annual average standard and 35 ?g/m 3 daily average standard). Figures1 and 2 show counties in the Mid- Atlantic that violate the NAAQS eight hour O 3 standard and the annual PM 2.5 air quality standards. Figure 1. Counties in the Mid-Atlantic Region out of compliance with NAAQS 8- hr O 3 standards. 2 Figure 2. Counties in the Mid-Atlantic Region out of compliance with NAAQS annual PM 2.5 standards. Asthma hospitalizations (Buchdahl et al., 2000; White et al., 1994; Wong et al., 2001), reduced lung function in children (Frischer et al., 1999; Gauderman et al., 2002) and acute myocardial infarcation (Ruidavets et al., 2005) have been associated with exposure to large O 3 concentrations. Fine particulate matter (PM 2.5 with an aerodynamic diameter < 2.5 ?m) alters the radiative balance of the Earth, decreases visibility and acts as cloud condensation nuclei (CCN). Increases in the CCN concentration can impact the global climate (IPCC, 2001). These increases in CCN lead to smaller cloud droplets which make clouds brighter and more reflective. Recent studies (Laden et al., 2000; Schwartz and Neas, 2000; Peters et al, 2001a) 3 have shown that elevated levels of PM 2.5 are associated with cardiovascular and respiratory problems and even increased mortality rates. SO 2 is a major precursor of fine particulate matter in the Mid-Atlantic region of the United States and accounts for 30-60% of PM 2.5 mass (Chen et al., 2002; Malm et al., 2004; Rees et al., 2004; Schwab et al., 2004; Frank et al., 2006; Ondov et al., 2006). 1.2 Chemical Transport Models Models are used to predict pollution events and can be used in regulatory measures (such as when to issue warnings to the public to not drive, pump gas, paint, run electrical appliances, etc.) in order to reduce the pollution impact on the area. To achieve good predictions, accurate initial conditions are needed. A chemical climatology of the vertical and horizontal distribution of trace gases and aerosols can be used to improve model initial conditions and determine how well models generate these three-dimensional distributions. The Community Multiscale Air Quality (CMAQ) system was developed by the EPA to improve predictions of pollution events associated with O 3 , PM 2.5 and reactive nitrogen species. The CMAQ modeling system incorporates outputs from meteorological modeling systems and emissions databases into a chemical transport model. Hogrefe et al. (2004) describe comparisons between O 3 surface measurements and CMAQ model results for 5 years of data (1993-1997) in the Eastern US. They found that CMAQ tends to overestimate small values of one and eight ?hour maximum O 3 , and underestimate large values of 1-hr maximum O 3 . They also found that CMAQ captures the higher eight-hr maximum O 3 . Tesche et al. (2006) compared daily CMAQ sulfate with surface measurements made using six 4 different networks in the United States in 2002. They found that the monthly averages of daily CMAQ simulations overestimated sulfate in the summer and fall and underestimate sulfate in the winter and spring with a fractional bias ? 30%. I found the opposite, in summary, CMAQ overestimates SO 2 (and likely underestimates sulfate) when compared with aircraft profiles (presented in Chapter 5). Levy (2007) found that CMAQ underestimates PM 2.5 in Maryland. Because sulfate accounts for 30-60% of PM 2.5 (Chen et al., 2002; Malm et al., 2004; Rees et al., 2004; Schwab et al., 2004; Frank et al., 2006;Ondov et al., 2006) it is likely that CMAQ underestimates sulfate. Mueller et al. (2006) also found that CMAQ underestimates sulfate. Mueller et al. (2006) found that CMAQ consistently underestimates cloud cover for model simulations in summertime episodes of 1992, 1993 and 1995 in the Eastern US. They also compared surface SO 2 and sulfate measurements to CMAQ model results and they found that CMAQ typically over- estimated SO 2 and underestimated sulfate. They concluded that this is likely because CMAQ does not properly account for in-cloud oxidation of SO 2 . The Georgia Tech/ Goddard Global Ozone Chemistry Aerosol Radiation and Transport (GOCART) model was developed to simulate the atmospheric sulfur cycle (Chin et al., 2000a). Chin et al., (2000b) compared daily surface measurements of SO 2 in the US and Europe with GOCART simulations for 1989 and 1990. They found that GOCART was able to capture daily variations in SO 2 and sulfate, but the model overestimates SO 2 in the summer (by more than a factor of two) and underestimates measured maximum sulfate for the US. 5 1.3 Meteorology Associated with Elevated O 3 Understanding the relationship between meteorology and pollution events can improve forecasting of these events. In the Mid-Atlantic region voluntary measures such as car-pooling, public transportation, refueling after dark, and limiting electrical usage are encouraged on days when pollution events are predicted. These measures can help reduce pollution levels and their effectiveness is determined in part by how well pollution levels can be predicted. Elevated levels of O 3 are generally associated with high pressure systems and weak winds (Vukoich, 1994). Vukovich et al. (1999) and Ryan et al. (1998) found that larger O 3 levels are generally associated with areas with high pressure systems just above the surface as well as high pressure systems to the west or northwest. These high pressure systems are generally associated with little cloud cover, weak winds, subsidence and low-level inversions that allow for local O 3 accumulation. These conditions are also conducive for transport of O 3 and O 3 pre-cursors from the industrialized Mid-West. Vukovich et al. (1999) found that the most O 3 exceedences occur in July. These O 3 exceedences can be reduced if energy saving programs are implemented during this time period. The development of a chemical climatology and determination of meteorological conditions associated with pollution events can aid in the improvement of model predictions and forecasting these events as well as improve the understanding of transport over source regions. 6 1.4 O 3 Chemistry O 3 is formed from oxides of nitrogen (NOx) and volatile organic compounds (VOC). O 3 is produced by the following reaction of NO 2 and light NO 2 + h? (? < 430 nm) ! NO + O (1) O + O 2 + M ! O 3 + M (2) However, the NO formed in Reaction 1, readily reacts with O 3 to form NO 2 as shown below: NO + O 3 ! NO 2 + O 2 (3) This results in a steady state between O 3 and NO 2 . O 3 production is driven by reactions with the hydroxyl radical (OH) and organic compounds that provide a sink for NO, shown in equations 4-7 below: OH + RCH 2 R + O 2 ! H 2 O + RCOOHR (4) RCOOHR + NO ! NO 2 +RCHOR (5) RCHOR + O 2 ! RCOR + HO 2 (6) HO 2 + NO ! NO 2 +OH (7) Here R represents a portion of the organic compound C n H m , where n and m are integers. The major sources of OH include photodissociation of O 3 . O 3 + h? (? > 340nm) ! O( 1 D) + O 2 (8) O( 1 D) + H 2 O ! 2OH (9) In polluted areas, sources can include photodissociation of nitrous acid (HONO) and hydrogen peroxide (H 2 O 2 ). The reaction between the hydroperoxyl radical (HO 2 ) and NO can also form OH. Sources of NOx (NO and NO 2 ) in the United States include transportation (56% for 2002, EPA, 2003) and fuel combustion (37% for 2002, EPA, 7 2003). Though the sources of NOx are generally at or near the surface the peak in O 3 production does not occur at the surface because the amount of light needed for NO 2 photodissociation increases with altitude (Kelley et al., 1995, Taubman et al., 2004a). 1.5 SO 2 Chemistry Annual emissions of SO 2 for 2000 in the US were 1.65 x 10 7 kg (EPA, 2003). Fuel combustion accounted for 86%, industrial processes accounted for 9% and transportation sources accounted for 5%. SO 2 is the pre-cursor for most sulfate; Rees et al.(2004) found that sulfate accounts for 38% of PM 2.5 annually in Pittsburgh, PA, Frank et al. (2006) found that sulfate accounts for 44-53% of PM 2.5 annually in Bronx, NY, Schwab et al. (2004) found that sulfate accounts an average of 30% of PM 2.5 in the summer, at six sites in NY; Ondov et al. (2006) found sulfate accounts for 32-40% of PM 2.5 mass in Baltimore, MD, and Malm et al. (2004) found it accounts for 50-60% of PM 2.5, and Chen et al.(2002) found it accounts for 35% of PM 2.5 . SO 2 is a short lived species that is oxidized quickly with the OH radical to form sulfate; other loss processes include dry and wet deposition. The reaction with OH proceeds as follows: OH . + SO 2 + M! HOSO 2 +M (10) HOSO 2 + O 2 ! HO 2 + SO 3 (11) When sufficient water vapor is available, SO 3 is converted to H 2 SO 4 SO 3 + H 2 O + M ! H 2 SO 4 + M (12) Typical atmospheric concentrations of OH give rise to an atmospheric lifetime for SO 2 of about a week. Seinfeld and Pandis (1998) suggest that by accounting for a typical dry deposition velocity of 1 cm s -1 and a boundary layer of 1km, the lifetime 8 of SO 2 is about one day. I have found that the lifetime of SO 2 is 19 ? 7 hours (presented in Chapter 5) for the Mid-Atlantic for summertime and daytime conditions. The oxidation of SO 2 with OH (determined from CMAQ) accounts for about 11% of SO 2 removal. This lifetime was determined from profiles made when fair weather cumulus clouds were common. The reaction between SO 2 and aqueous H 2 O 2 (found in fair weather cumulus clouds) also account for a significant amount of SO 2 oxidation. Using the calculated SO 2 lifetime (presented in Chapter 5) and assuming OH and H 2 O 2 contribute significantly to the SO 2 loss, it appears that the Seinfeld and Pandis (1998) theoretical estimate of SO 2 dry deposition velocity is too fast for the Mid-Atlantic in the summer during daylight hours. Global model calculations have shown the SO 2 lifetime to vary from, 0.6 ? 2.6 days (Pham et al., 1995; Chin et al., 1996; Rested et al., 1998; Koch et al., 1999; Roelofs et al., 1998; Berglen et al., 2004). SO 2 in the gas phase can also dissolve in water to form the following species depending on the pH: SO 2 + H 2 O ! SO 2 . H 2 O (13) SO 2 . H 2 O ! HSO 3 - + H + (14) HSO 3 - ? ! H + + SO 3 2- (15) Here the bisulfite (HSO 3 - ) form is most often produced at a pH of 2-6, common for atmospheric droplets. SO 2 can also be oxidized by H 2 O 2 in clouds and fogs at a pH less than 4.5 (Finlayson-Pitts and Pitts, 2000). HSO 3 - + H 2 O 2 ?! SO 2 OOH - + H 2 O (16) SO 2 OOH - + H + ! H 2 SO 4 (17) 9 Edgerton et al. (2006) measured hourly SO 2 , sulfate and other trace gases and aerosols at surface stations in the Southeast US in early spring 2002 (as part of the Southeastern Aerosol Research and Characterization Study, SEARCH). Using back trajectories and pollutant ratios they calculated SO 2 to sulfate conversion rates in SO 2 plumes (generally from coal-fired power plants, in the first 10 hours of transit time from the source) corresponding to an e-folding lifetime of 500- 40 hours. 1.6 Determination of Meteorological Influences on Pollution Episodes: Clustering Back Trajectories In order to effectively reduce pollution, major sources and meteorological conditions associated with pollution events need to be accurately determined. Clustering, a statistical technique to group data in space has been used to assess the impacts of emissions and meteorology on pollutant concentrations at receptor sites. This technique has been employed to group back trajectories into different meteorological regimes (Moody and Galloway, 1988; Dorling et al., 1992a; Dorling et al., 1992b; Lee et al., 1994; Moy et al., 1994; Dorling and Davies, 1995; Moody et al., 1995; Harris and Oltmans, 1997; Brankov et al., 1998; Moody et al., 1998; Cape et al., 2000; Eneroth et al., 2003; Berto et al., 2004; Jorba et al., 2004; Russell et al., 2004). The clustered back trajectories can then be used to determine source regions and synoptic regimes that support the regional transport of different atmospheric constituents. The studies listed above differ mainly in methods used to calculate the trajectories and the different techniques used to cluster the trajectories. A major limitation of the published studies cited above arises from the fact that all of the receptor sites were surface-based. This restricts the amount of 10 information available on regional transport and the influences of lower atmospheric dynamics on the pollution measured at the surface. Eneroth et al. (2003) and Jorba et al. (2004) clustered trajectories at multiple altitudes to better describe the general circulation patterns in the troposphere, but the measurements were still fixed at the surface. Taubman et al. (2006) improved upon the previous studies by using similar statistical techniques to analyze several years of data collected from aircraft. Aircraft provide a horizontally and vertically mobile sampling platform. The horizontal mobility allows for deployment to specific areas of interest, while the vertical mobility provides insight into boundary layer dynamics, and allows for measurements representative of a larger area. The ability to deploy to specific locations enables the investigation of multi-day haze and O 3 episodes, the influences of regionally transported pollution on urban and rural areas as well as the impacts large metropolitan areas have on Mid-Atlantic air quality. The vertical profile information presents a more complete picture of the composition and dynamics of the lower atmosphere, and this allows for the investigation of factors influencing the transport and chemical transformations of air pollutants and their precursors. Specifically, the nocturnal emissions from elevated sources and transport of pollution in the residual layer can be calculated from vertical profiles taken before the stable, nocturnal boundary layer has eroded and the pollution mixed down to the surface. The identification of transported pollution allows for a more accurate assessment of the effects of mixed layer development on surface pollution as well as local emissions and photochemical production. 11 The Regional Atmospheric Measurement Modeling and Prediction Program (RAMMPP) (http://www.atmos.umd.edu/~RAMMPP/) was formed to address problems with air pollution over the Mid-Atlantic US. To perform long term air quality studies and analyze tran-boundary pollution transport RAMMPP uses ? measurements (ground-based and airborne), chemical transport modeling (Models- 3/CMAQ), meso-scale modeling (MM5), and air quality forecasting. The airborne measurements have been conducted by the University of Maryland since 1992 with an instrumented light aircraft outfitted for atmospheric research. The aim of the aircraft analyses thus far has been to answer specific questions regarding lower atmospheric CO (Dickerson et al., 1995; Doddridge et al., 1998), pollutant transport and boundary layer dynamics during individual, Mid-Atlantic haze and O 3 episodes (Ryan et al., 1998; Taubman et al., 2004a), and the air quality and radiative impacts of smoke in the Mid-Atlantic from Canadian forest fires (Taubman et al., 2004b). Additionally, a fortuitous experiment demonstrated the regional air quality benefits of the 2003 North American blackout (Marufu et al., 2004). In Chapter 3 a chemical climatology of trace gases and aerosols (some of which was published in Taubman et al., 2006) that answers some of the overarching questions not yet addressed by these case studies will be presented. Typical clustering analyses clustered back trajectories ending at a single location. However, there were limited individual locations over which enough flights were performed to provide statistical meaning using this typical analysis. Furthermore, narrowly focusing on a few locations would fail to take advantage of the regional coverage offered by the dataset. For these reasons I have developed a novel 12 approach to clustering the data using multiple spatially heterogeneous receptor locations (presented in Taubman et al., 2006). Ozone events have long been identified as regional in nature with variability expected on scales of hundreds of km (Logan, 1985), so the use of multiple receptor locations is justified. I modified the standard distance calculation to account for spatial variability in the cluster algorithm. A detailed description of the methods is presented in Chapter 3. The statistical analysis of vertical profiles that I have developed is inherently different from analyses of single measurements at surface-based receptor sites. Using these techniques we were able to quantify the impacts of source regions and transport patterns on Mid- Atlantic air quality. The measurements from this study overlap in time with those from the Mid- Atlantic EPA Supersites in Baltimore, New York, and Pittsburgh. Because of the regional nature of the study, the results presented in Chapter 3 will complement the investigations from those sites, aiding in measurement comparisons, model validation, and understanding the processes that control regional pollutant transport to and between the individual sites. The analyses should also be useful for air quality forecasting and modeling of pollution episodes as well as pollution control strategies. 1.7 Determining the influence of Point Source on Pollution Episodes: Clustering species profiles Methods that clarify the influence of meteorology and emissions on the vertical distribution of trace gases and aerosols can improve modeling and prediction of pollution events. I developed a method for clustering vertical profiles of trace gases and aerosols to group distinct profile shapes that may be associated with 13 different meteorological patterns or various transport regimes. This complements the previous section that clustered back trajectories and then formed associated trace gas and aerosol profiles. Many previous studies (Dorling et al., 1992a; Dorling et al., 1992b; Lee et al., 1994; Moy et al., 1994; Dorling and Davies, 1995; Moody et al., 1995; Harris and Oltmans, 1997; Brankov et al., 1998; Moody et al., 1998; Cape et al., 2000; Eneroth et al., 2003; Berto et al., 2004; Jorba et al., 2004; Russell et al., 2004, Taubman et al., 2006) were devoted to clustering back trajectories to describe meteorological patterns associated with different trace gas and aerosol values. Moy et al. (1994), Brankov et al. (1998), and Taubman et al. (2006) were able to use back trajectory clusters to describe meteorological patterns associated with smog events. The converse of this method, clustering by O 3 profiles to identify different transport patterns, has been applied to ozonesonde and aircraft data (Diab et al., 2003; Diab et al., 2004, Colette et al., 2005 a; Colette et al., 2005 b). Models used to predict O 3 and PM 2.5 levels have limited ability to describe lower tropospheric transport within the planetary boundary layer (Seigneur, 2001; Mebust et al., 2003; Zhang et al., 2004; Hodzic, et al., 2005). There is inadequate information on the planetary boundary layer distribution of trace gases and aerosols to improve the models. The University of Maryland has conducted summertime aircraft measurement campaigns since 1993 to gain a better understanding of the chemistry and dynamics of the lower troposphere including (Dickerson et al., 1995; Doddridge et al., 1998; Ryan et al., 1998; Taubman et al., 2004a; 2004b; 2006). Species measured aboard the University of Maryland research aircraft include O 3 , SO 2 , CO, particle light absorption at 565 nm, and total particle scattering at 450, 550, and 700 14 nm. I have clustered vertical profiles of the species by shape and absolute value to improve understanding of meteorological and emissions influences on trace gases and aerosols in the lower troposphere. Taubman et al. (2006) grouped 48 hr back trajectories associated with 550 of the University of Maryland profiles into eight distinct meteorological regimes and used these clusters of back trajectories to describe differences among morning and afternoon profiles of O 3 , SO 2 , CO, particle scattering, ?ngstr?m exponent (?) calculated from the 450/700 nm ratio of particle scattering, and particle absorption. In Chapter 4, I will introduce a method for clustering these same profiles by their shape and magnitude (mixing ratio and scattering and absorption coefficients). This allows for separation of profiles based on small-scale structure and these differences may be ascribable to other factors such as emissions. The characterization of the planetary boundary layer and the lower free tropospheric composition of trace gases and aerosols can be used to evaluate and improve chemical transport modeling of these species, and aid in the forecasting of pollution events. It can also improve understanding of the relationship between the meteorology and chemistry of the lower troposphere. 1.8 Surface Measurements The EPA has developed a Speciation Trends Network (STN) which has measured aerosols at the surface for 54 sites in the US since 1999. Precise measurements are critical for PM 2.5 source apportionment tasks based on chemical mass balance and/or multivariate receptor models (Hopke, 1984; Watson et al., 1984; Kim and Hopke, 2005; Kim et al., 2005; Ogulei et al., 2005). 15 NAAQS calls for the use of a Federal Reference Method, FRM, (Code of Federal Regulations (CFR), 1997) for the measurement of filter-based gravimetric PM 2.5 mass to determine compliance. However, other sampling and analytical protocols have been used extensively in air quality monitoring projects, such as the Speciation Trends Network ?STN (US EPA, 1999), the Interagency Monitoring and Protective Visual Environment network ?IMPROVE (Malm et al., 1994; Ames et al., 2001; Malm et al., 2002; 2004; 2005) and the California Regional PM 10 /PM 2.5 air quality study (Chow et al., 2006). Equivalence of PM 2.5 mass determined with different protocols is currently under evaluation (Peters et al., 2001b; Watson and Chow, 2002; Solomon et al., 2003; Chow et al., 2005a). A FRM for PM 2.5 speciation has not yet been established by the United States Environmental Protection Agency. I collected PM 2.5 samples during the 2002 intensive sampling periods at Fort Meade, Maryland (FME). I have used the samples to evaluate the STN speciation samplers and filter analyses under typical and elevated PM 2.5 events. FME, a suburban site located in the Baltimore-Washington urban corridor, approximately 3 km east of the Baltimore-Washington Parkway (I-295) and 10 km east of Interstate 95, was the anchor site for the Maryland Aerosol Characterization (MARCH- Atlantic) study (Chen et al., 2002) and part of the nationwide Speciation Trends Network (STN). It also served as one of the satellite sites for the Baltimore Supersite experiment during 2001 ? 2003 (Lake et al., 2003; Harrison et al., 2004; Lee et al., 2005a; Ogulei et al., 2005; Park et al., 2005a; Park et al., 2005b; Ondov et al., 2006). Previous studies indicate that FME observations often reflect regional haze episodes and local accumulation under stagnant conditions. The annual mean PM 2.5 16 concentration at FME is around 13 ?g/m 3 , and is influenced by local and regional sulfate, wood smoke, industry, mobile sources and secondary nitrate (Chen et al., 2001; 2002; 2003). During January and July 2002, PM 2.5 speciation monitors from two different protocols (Speciation Trends Network-STN and Desert Research Institute-DRI) were installed at FME to concurrently measure atmospheric aerosol on a 24-h basis. Two Sequential Filter Samplers (SFS, Desert Research Institute, Reno, NV) from DRI were deployed in both January and July, while a Reference Ambient Air Sampler (RAAS PM 2.5 , Thermo Scientific, Waltham, MA) and a Met One Speciation Air Sampling System (SASS, Met One Instruments Inc., Grants Pass, OR) represented the STN operation in January and July, respectively. The change of STN sampling systems (from January to July) was made with the understanding that both samplers had been equally approved by EPA for the STN application (US EPA, 1999). However their performances are not the same with respect to the DRI sampler. The SFS samples were analyzed by DRI and the RAAS and SASS samples were analyzed at the Research Triangle Institute (RTI, Research Triangle Park, NC) using methods described in Chow et al. (1996) and US EPA (1999). I will refer to the SFS samplers as DRI F and the RAAS and SASS samplers as STN R and STN S (STN RS denotes both instruments) hereafter. Components quantified by both DRI and RTI include gravimetric PM 2.5 mass, 35 trace elements, elemental carbon (EC), organic carbon (OC), total carbon (TC), and water soluble ions such as sulfate, nitrate and ammonium. DRI and RTI often used different techniques and instruments for the 17 analyses. Continuous measurements of PM 2.5 mass were made in July with a Tapered Element Oscillating Microbalance (TEOM 1400a, Thermo Scientific, Waltham, MA). Field performance of the STN R and performance of the STN RS size-selective inlet was assessed during the early stage of STN RS development (Peters et al., 2001b, 2001c), but up-to-date evaluations of the STN RS speciation data under real-world operation are rather limited. I will compare the STN RS data from FME with collocated DRI measurements and investigate the PM 2.5 chemical composition and mass closure within the context of uncertainty analysis. Approaches and conclusions presented in Chapter 6 can be tested in other studies facilitating a weight of evidence approach (e.g., Burton et al., 2002; Weed, 2005) to improve the design of ambient PM 2.5 networks. The objective and results of this study are coordinated with others in the region including Lee et al., (2005a, 2005b), Flanagan et al., (2006) and the EPA- sponsored Eastern Supersites program (Solomon et al., 2003; Rees et al., 2004; Ondov et al., 2006). 1.9 Overview The chemical climatology was developed with 10 years of summertime measurements of trace gases and aerosols made aboard the UMD research aircraft. Chapter 2 will provide specifics on instrumentation used aboard the aircraft. Most flights were made in the Mid-Atlantic region and Chapter 2 will present locations and times when flights were made. An introduction to the chemical climatology with statistics of trace gas and aerosol measurements will be presented in Chapter 3. Diurnal variations in measurements will also be discussed. I performed a cluster analysis of back 18 trajectories associated with the aircraft flights in order to describe different meteorological regimes associated with the profiles. Results from this cluster analysis will be presented in Chapter 3, and some results from this Chapter were published in Taubman et al. (2006). The typical flight pattern consisted of flying upwind of pollution centers in the morning and downwind of the centers in the afternoon. This allowed for quantification of transported lower tropospheric O 3 . A description of how transported O 3 was calculated and which meteorological regimes were associated with the most transport is presented in Chapter 3. I developed of a methodology to cluster profiles of trace gases and aerosols to separate extreme events and to better understand meteorological and point source influences on aircraft profiles. This clustering methodology is presented in Chapter 4 along with the relationship between trace gas and aerosol profiles with point source emissions of SO 2 and NOx. To better understand how well models predict trace gases in the Mid-Atlantic I have compared O 3 , SO 2 , and CO with a regional model (CMAQ) presented in Chapter 5. I have also compared aircraft measured SO 2 with a global model (GOCART). The SO 2 column content generated from CMAQ and GOCART was 50% larger than that measured aboard the University of Maryland aircraft. As described in Chapter 4 the SO 2 profiles were not well correlated with SO 2 emissions encountered along a 48 hour back trajectory. These findings support the hypothesis that the lifetime of SO 2 in the Mid-Atlantic region in the summer is less than 48 hours. I developed a method for calculating the SO 2 lifetime using the UMD research 19 aircraft profiles and the EPA?s SO 2 emissions database. This lifetime calculation is presented in Chapter 5. In 2002 I collected surface filter samples of aerosols with two different American sampling systems. Comparisons of surface PM 2.5 measurements made and the uncertainties associated with the sampling systems will be presented in Chapter 6. A summary of the research and recommendations for future work will be presented in Chapter 7. 20 Chapter 2: Sampling Platform and Instrumentation 2.1 Introduction The chemical climatology was developed with 10 years of aircraft measurements of trace gases and aerosols. These measurements were made aboard the UMD research aircraft, mostly in the Mid-Atlantic region during regional haze and O 3 episodes. This Chapter will give specifics on where flights were made and how they were made. It will also provide details on the instruments used to collect trace gases O 3 , SO 2 and CO as well as aerosol scattering and absorption. 2.2 Aircraft A twin engine Piper Aztec was used to collect vertical profiles of trace gases and aerosols. Instruments were housed inside the aircraft and the sampling inlets were attached to the upper fuselage. On the upper fuselage there was an aft facing inlet that was attached to the trace gas instruments. There was also a forward facing isokinetic inlet that sampled aerosols. There were line losses of particles due to impaction on the side walls of the forward facing inlet and for this reason only submicron particles (< 1?m) were sampled. Flights were made mostly in the Mid- Atlantic region from 1995-2005. Spirals from 3 m above the surface to about 3000 m were made at small airports shown in Figure 1. 21 Figure 1. Locations of University of Maryland research aircraft flights made in the Eastern US between 1995 and 2005. Spirals were completed within 30 minutes at a vertical climb rate of 100 m min -1 . Flight patterns were generally chosen to capture transport of pollutants to areas downwind of urban areas, and so flights conducted in the morning (before 12 noon EST) were upwind of urban areas in the Mid-Atlantic, while flights conducted in the afternoon (after 12 noon EST) were downwind of urban areas. The full instrument suite was not used every year and Table 1 shows the years in which new instruments were added to the suite. I collected samples aboard the aircraft for 12 of the flights (30 profiles) in 2005. 22 O 3 CO SO 2 PSAP Nephelometer 1996 X 1997 X 1999 X X 2000 X X X X 2001 X X X X X 2002 X X X X X 2003 X X X X X 2004 X X X X X 2005 X X X X X Table 1. Years when O 3 , CO, SO 2 , absorption with the PSAS and scattering with the nephelometer were sampled. 2.3 SO 2 Thermo Scientific SO 2 analyzer (43C, Franklin, MA) measures SO 2 fluorescence from a pulsating UV light. SO 2 absorbs radiation in three wavelength regions, 1) 390-340 nm, 2) 320-250nm and 3) 230-190 nm. SO 2 absorbs weakly in region 1 and is quenched rapidly. SO 2 is also quenched rapidly by O 2 and N 2 in region 2. SO 2 is quenched least in region 3. In this region SO 2 absorbs a quantum of energy (h? 1 ) and forms an electronically excited molecule SO 2 + h? 1 ! SO 2 * (1) The light intensity absorbed by SO 2 , I a , is a function of the incident light, I o , the absorption coefficient, a, the path length, x, and the SO 2 concentration (SO 2 ) and described by beer?s law: I a = I o *{1 ? exp-[a*x*(SO 2 )]} (2) The electronically excited SO 2 molecule can release the excess energy in three ways: fluorescence, quenching and dissociation. Fluorescence of SO 2 can be written as: SO 2 * ! SO 2 + h? 2 (3) 23 Here the quantum of energy released, h? 2 , is at a lower frequency than the quantum of energy absorbed, h? 1. Reaction 3 proceeds with a rate constant k f . Quenching of the excited SO 2 can be written as SO 2 * + M ! SO 2 + M (4) Here M is a molecule of air that absorbs the excess energy. The rate constant for Reaction 4 is kq. Dissociation of SO 2 can be written as: SO 2 * ! SO + O (5) Dissociation occurs with a rate constant of k d . The fluorescent intensity measured by the detector, F, can be written as F = [G*k f *Io*{1 ? exp-[a*x*(SO 2 )]}] / (k f + k d + k q *[M]) (6) Here G accounts for the geometry of the fluorescent chamber design. Assuming the SO 2 concentration and path length are small, this equation reduces to: F = [G*k f *Io*a*x*(SO 2 )] / (k f + k d + k q *[M]) (7) The reaction rates (k f , k d and k q ) are relatively constant over a wide range of temperatures and pressures. The incident light Io can be engineered to remain constant, and G and x are also constants so the fluorescent intensity is directly proportional to the SO 2 concentration. This direct proportionality between fluorescent intensity and SO 2 concentration is the basis of the SO 2 instrument. 24 Figure 2. Schematic of flow diagram for SO 2 pulsed fluorescence monitor (Thermo Scientific, 2004a). Figure 2 shows a schematic of gas flow through the SO 2 monitor. The monitor draws air in through a sample port at a rate of 0.5 L/min. Air then moves through a Nafion hydrocarbon kicker. The hydrocarbon kicker is used because there are significant interferences from polynuclear aromatic hydrocarbons (PAH); the most common of these being naphthalene. The hydrocarbon kicker is a semi permeable membrane, allowing only hydrocarbons through. A differential pressure is established by passing the sample air through a capillary tube. SO 2 enters a fluorescence chamber and is excited to a higher energy state. The excited SO 2 emits in 3 wavelength ranges with the 190-230 nm range being the most easily measured. The instrument runs on a switch able zero or measure mode. During the zero modes sampled air is run through a K 2 CO 3 filter to remove SO 2 . The zero modes are averaged and subtracted from the measure mode to decrease background noise (Thermo Scientific, 2004a). 25 In order to reduce the effects of non-methane hydrocarbons (NMHC) an extra PFA Teflon canister packed with activated charcoal was added to the kicker. The addition of the canister helps to reduce the partial pressure of the NMHCs in the sampling device and increases their diffusion out of the semi-permeable membrane (Luke, 1997). These modifications make interferences from most hydrocarbons negligible, though there are still interferences from NO x that can only be removed through zeroing. A UV lamp source is used to generate light in the 230-190 nm region. The light passes through a condensing lens (to focus the beam) and a series of reflective bandpass filters to stabilize and intensify the beam. The beam is passed through a relay lens and a circular baffle to remove stray light. The beam then enters the reaction volume, which contains the ambient air. The detector is perpendicular to the incident beam of light. Before the fluorescent light reaches the photo multiplier tube, the beam passes through a condensing lens and then a bandpass filter to ensure only light from SO 2 fluorescence enters the detector. Luke (1997) approximated the detection limit to be around 0.3 ppb during the 1994 National Science Foundation-sponsored Gas-Phase Sulfur Intercomparison Experiment. I calculated the detection limit in 2005 for the SO 2 monitor used aboard the Maryland Research aircraft. I measured zero mode SO 2 1-minute averages for 30 minutes. I assume that the detection limit is two times the standard deviation of the zeros and this was 0.25 ppb. Luke (1997) approximated the uncertainty of the instrument to be 16% (at the 95% confidence level) when sampling mixing ratios were greater than 0.5 ppb. 26 2.4 CO A modified (Dickerson and Delany, 1988) Thermo Scientific CO infrared filter correlation analyzer (43C, Franklin, MA) was used to measure CO. An IR source of 4.6 ?m wavelength radiation is used because CO absorbs in this region. Figure 3 shows the schematic of the instrument. Figure 3. Flow schematic for the CO detector (Dickerson and Delany, 1988). Modifications include a collecting lens, cooling chamber, dessicant, chemical zero, and operation at positive pressure. The air sample is drawn into the instrument and enters the sample chamber. The source is chopped and then travels through a gas filter wheel. Half of the filter wheel contains high concentrations of CO and the other half contains N 2 . N 2 does not absorb in this IR region and so light that passes through this half of the wheel becomes the measure beam. The high concentration of CO in the other half of the wheel absorbs all the IR radiation and this produces a reference beam. Both beams pass through the sample cell, with a 30 m effective path length. The CO concentration being measured is derived from the relative intensity of the measure 27 and reference beam (Thermo Scientific, 2004b). The instrument is capable of a 2-5% precision determined from a 1-min mean of 10 s data. 2.5 O 3 Ozone is measured with a Thermo Scientific UV photometric analyzer (49, Franklin, MA). This analyzer operates on the Beer-Lambert law by measuring the attenuation of light due to O 3 absorption at 254 nm. The source is a low pressure mercury vapor lamp, which has 95% output at 254 nm. The detector is a solar blind vacuum photodiode sensitive to UV light only. Figure 4 gives a schematic of the instrument. The sample enters the instrument and is split into two gas streams. One of the streams is scrubbed of O 3 , which allows it to be the reference beam, I o . The reference beam then passes onto the sample solenoid valve. The other gas stream flows directly to the sample solenoid and is the measure beam, I. The solenoid switches the instrument between zero and measure modes (sampling from the I and I o beam). The ratio of I/I o is directly proportional to the concentration of O 3 . I/I o = exp -KlC (8) Where K = 308 cm -1 atm -1 at 0 o C and 1 atm, l is the length of the cell (38 cm), and C is the O 3 partial pressure in atm. In order to register a change of 1ppb concentration, the instrument must be able to detect a change in I/I o of 2 parts in 10 5 . It takes about 10 seconds to measure I and I o , and the source must be stable to 2 parts in 10 5 . Because this stability is difficult to reach, a second detector is used to monitor and correct the changes in light intensity. The instrument employs 2 photometers with a single light source and two absorption cells and detector systems. These photometers operate synchronously but out of phase so that when one is in measure mode, the 28 other is in zero mode. A flush time of about 7 seconds is used to remove the ambient air from the sample cell, and then the measurement is made during the next 3 seconds. By taking the average of the 2 photometer readings, the fluctuations of the lamp intensity are cancelled out. The instrument is capable of 1 ppb precision for 10 s data. Figure 4. Flow schematic for U.V. photometric O 3 analyzer (Thermo Scientific, 2004c). 2.6 Aerosol Absorption A Particle Soot Absorption Photometer (PSAP; Radiance Research, Seattle, WA) was used to measure near real-time aerosol absorption on a filter. An LED light source shines 567 nm light through an opal glass and onto a Pallafax filter. The PSAP measures the absorption coefficient, ? ap (Mm -1 = 10 6 m -1 ), with an integrating technique. The absorption coefficient is determined from the volume of air sampled during an averaging time using Beers law: 29 ? ap = A/V * ln (I o /I) (9) Where A is the area of the filter exposed to the light beam, V is the average volume of air sampled by the instrument, I o is the average filter transmittance for averaging time j and I is the average filter transmittance for averaging time j + 1. The ? ap is then corrected for filter nonlinearities including filter loading and filter characteristics. This correction can be written as: ? ap = ? ap f(Tr) (10) Where f(Tr) is a transfer function based on filter loading (or transmittance, which is recorded by the instrument), for Pallafex filters. Figure 5 shows a diagram of the air flow. Ambient air flows through the first filter (measurement filter), where all particles are removed. The particle-free air then flows through the reference filter. The transmission of light through the filter is measured with a photodiode. A second filter, adjacent to the first is used to ensure that changes in the change in intensity results from buildup of particles on the filter and not fluctuations in the LED source. 30 Figure 5. Flow diagram for PSAP (http://www.cmdl.noaa.gov/aero/instrumentation/psap_desc.html). The detection limit is 0.9x10-6 m -1 (95% confidence level) when 1-min measurement averages are used (Bond et al., 1999). Bond et al. (1999) has recommended corrections for differences in flow rates, spot size, and exaggerations of absorption due to scattering, and these have been applied to PSAP measurements. 2.7 Scattering The TSI integrating nephelometer (Model 3563, TSI, St. Paul, MN) was used to measure light scattering of atmospheric particles at 450, 550, and 700 nm. Figure 6 show a diagram of the instrument. 31 Figure 6. Flow schematic for the integrating nephelometer. The light source travels through a diffuser plate, ensuring a lambertian source. A photomultiplier tube detector is positioned parallel to the incoming light and measures scattering over 7 o to 170 o . The light source flux that reaches the detector is defined by the following: ???? ? ? dyIB o )sin()(/ 2 1 ? = 11 where Io is the intensity of the incident light, y is the distance between the source and the detector, ? is angle between the light reaching the source and the incident light, and ?(?) is the angular scattering function (TSI, 1997). Integrating the scattering function from ? 1 = 0 to ? 2 = ? gives: )2/(*)/( ??scato yIB = (12) Where ? scat is the scattering coefficient (TSI, 1997). A chopper is used to generate an AC signal. The light passes through three different filters to correct for interferences. The first filter lets through all light for the full signal measurement. The second filter is used to remove the dark signal from the photomultiplier. The third filter measures the light source to monitor the lamp stability (TSI, 1997). Corrections were made for truncation errors (where forward 32 scattered light at angles less than 7 o is blocked) and nonlambertian errors as suggested by Anderson and Ogren (1998). The inlet airstream was dried to an RH < 20%. To account for hydroscopic particle growth an estimated growth factor, F(RH), was calculated. F(RH) is the ratio of scattering from ambient particles to scattering from dried particles F(RH) = ? sp (?, RH) / ? sp (ref) = {(1-RH amb )/(1-RH ref )} -? (13) Here ? sp (?, RH) represents light scattering from the ambient particles, ? sp (ref) represents light scattering from the dried particles, RH amb is the ambient RH, RH ref is the RH inside the nephelometer, and ? is derived from parallel nephelometers. A ? value of 0.35 was used, similar to that in Remer et al. (1997), because the region of their study is similar to those presented here. The detection limits for scattering are 0.44x10 -6 , 0.17x10 -6 and 0.26x10 -6 m for scattering at 450, 550, and 700 nm. The instrument is calibrated with CO 2 for Raleigh scattering 33 Chapter 3: Determination of Meteorological Influences on Pollution Episodes: Clustering Back Trajectories. 3.1 Introduction. This chapter will address the two questions: 1) Is there a statistical link between characteristic regional transport patterns in the Mid-Atlantic US during summertime haze and O 3 episodes and specific pollution loadings? 2) Can the local O 3 contributions be differentiated from regionally imported O 3 ; if so, are the regional contributions quantifiable? Much of the work in this chapter was published in Taubman et al. (2006). In this chapter I will present statistics of trace gasses and aerosols from all flights made in June-August for 1995-2005. I collected measurements aboard the aircraft for 12 of the flights (36 profiles) in 2005. I calculated statistics for morning (before 12 noon EST) and afternoon (after 12 noon EST) profiles. Most morning flights were made upwind of pollution centers and most afternoon flights were made downwind of pollution centers. I clustered back trajectories for flights from 1997-2003 to determine meteorological regimes associated with the flights. The method for clustering back trajectories will be described and the resulting meteorological regimes determined by the clusters will be analyzed. I calculated statistics for profiles of trace gases and aerosols associated with the meteorological regimes and will discuss the results. I also calculated transported O 3 using in-situ measurements. The method and the amount of transport associated with each of the meteorological regimes will be discussed. 34 3.2. Observations 3.2.1 Measurements All flights analyzed for this study were conducted in the summertime (June, July, and August) and were specifically designed to characterize episodic pollution events. The observations reported herein represent polluted periods, not background values. Statistics for all flights made between 1995 and 2005 are presented first. The flight locations for statistics of all flights are presented in Figure 1 in Chapter 2. From 1995 through 2005, there were 658 summertime flights, which included 305 morning spirals (before noon EST, average time 09:30 EST) and 353 afternoon spirals (after noon EST, average time 13:30 EST). The median profiles for the morning and afternoon O 3 , CO, SO 2 , scattering at 550 nm and absorption at 550 nm are shown in Figure 1. The single scattering albedo represents the relative contribution of scattering from particles and was calculated using: single scattering albedo = ? 550 / (? 550 + abs 550 ) (1) Where ? 550 is the scattering coefficient at 550 nm and abs 550 is absorption coefficient at 550 nm. The ?ngstr?m exponent (? ) represents the relative size of particles and was calculated using: A =[log(? 450 ) ? log(? 700 )] / [log(450) ? log(700)] (2) Here ? 450 and ? 700 are the scattering coefficients at 450 and 700 nm respectively. Statistics for the single scattering albedo and ? are also presented in Figure 1. The profiles shown were generated by calculating the median value at each altitude layer from all of the measured profiles. 35 Figures 1a-c. Median values calculated every 100 m from all morning (green diamonds, before 12:00 EST) and afternoon (red diamonds, after 12:00 EST) profiles for a) O 3 (305 morning profiles, 353 afternoon profiles), b) CO (134 morning profiles, 178 afternoon profiles), and c) SO 2 (234 morning profiles, 254 afternoon profiles). The solid lines represent the 25 th and 75 th percentiles. 0 0.5 1 1.5 2 2.5 3 3.5 0 2040608010 O 3 (ppbv) A l t i tu d e (k m ) 0 0.5 1 1.5 2 2.5 3 3.5 0 100 200 300 400 500 CO (ppbv) A l titu d e (k m ) 0 0.5 1 1.5 2 2.5 3 3.5 02468 SO 2 (ppbv) A l titu d e (k m ) a b c 36 0 0.5 1 1.5 2 2.5 3 3.5 0 50 100 150 200 Scattering at 550 nm (Mm -1 ) A l ti tu d e (k m) 0 0.5 1 1.5 2 2.5 3 3.5 .60.70.80.9 1 Single scattering albedo at 550 nm A l ti tu de ( k m ) 0 0.5 1 1.5 2 2.5 3 3.5 0 5 10 15 20 25 Absorption at 550 nm (Mm -1 ) A l ti tu d e (k m) d e f 37 Figure 1d-g. Median values calculated every 100 m from all morning (green, before 12:00 EST) and afternoon (red, after 12:00 EST) profiles for d) scattering at 550 nm (189 morning profiles, 185 afternoon profiles), e) single scattering albedo at 550 nm (132 morning profiles, 153 afternoon profiles), f) absorption at 550 nm (175 morning profiles, 214 afternoon profiles), and g) ?ngstr?m exponent (189 morning profiles, 185 afternoon profiles). The solid lines indicate the 25 th and 75 th percentiles. The morning O 3 profile (Figure 1a) shows relatively small values (~45-55 ppbv) within the nocturnal boundary layer (roughly the lowest 500 m), with considerably more O 3 in the residual layer above. This results from surface deposition and titration with freshly emitted NO within the nocturnal boundary layer combined with night-time regional transport from upwind sources within the residual layer. Solar heating induces a more thoroughly mixed afternoon O 3 profile with photochemical production adding to that which was transported overnight. Above approximately 2 km, the morning and afternoon values are nearly identical (~55 ppbv), indicating a summertime continental background value. The morning and afternoon CO profiles (Figure 1b) are nearly identical below ~1 km, with large values near the surface that fall off with altitude. The shape of the 0 0.5 1 1.5 2 2.5 3 3.5 1 1.2 1.4 1.6 1.8 2 2.2 2.4 Angstrom exponent at 575 nm A l ti tu d e (km ) g 38 vertical profiles is similar to those presented by Dickerson et al. (1995) for spring, although the absolute values presented herein are slightly greater than in the previous study. Above 1 km, the afternoon values are slightly larger than the morning, indicating convective outflow from the boundary layer, the preamble to long range transport. The SO 2 profiles (Figure 1c) show little difference between the morning and afternoon. The afternoon profile shows somewhat smaller values near the surface, likely the result of oxidation to sulfate. There is also evidence of vertical mixing in the afternoon; however, both profiles show greater values near the surface that decrease sharply with altitude. Sulfur dioxide is a fairly short-lived species, typically less than a day in summertime (Seinfeld and Pandis, 1998), with sources generally elevated slightly above the surface. The afternoon scattering profiles (Figure 1d) are somewhat larger than the morning profiles between 200 and 2000 m and this may be explained by SO 2 oxidation to sulfate, the primary scattering component in fine particles over the eastern U.S. There is a maximum in the afternoon single scattering albedo near the top of the boundary layer where RH is also at a maximum. Both profiles decrease considerably above 2 km. The single scattering albedo profiles are similar to the scattering profiles and the single scattering albedo in the afternoon boundary layer average is greater than that in the morning profile (0.94 + 0.01 vs. 0.93 + 0.01) (Figure 1e). This increase of single scattering albedo in the afternoon is presumably the result of SO 2 oxidation and RH changes due to the planetary boundary layer growth. The absorption was relatively invariant with altitude (Figure 1f), so the 39 decline in single scattering albedo with altitude is driven by a decrease in scattering values. These observations are consistent with those of Novakov et al. (1997), who reported an increase in the relative amounts of carbonaceous to sulfate species with altitude over the Eastern U.S. coastline. No diurnal pattern in ?ngstr?m exponent was observed (Figure 1g); however a slight decrease with altitude is apparent in both the morning and afternoon profiles. The presence of larger particles aloft may be due to particle growth through preferential aging in the lower free troposphere as articulated in Taubman et al. (2004a). 3.2.2 Trajectory Calculations Back trajectories are a standard tool for determining the source regions and transport patterns of air parcels observed at receptor sites. While they may not represent the exact transport path of an air parcel, back trajectories are good representations of the general 3-dimensional wind flow and are useful in identifying particular synoptic situations. The accuracy and errors associated with the different estimations of air parcel trajectories have been quantified (Stohl et al., 1995; Stunder, 1996; Stohl, 1998). Individual trajectories may be subject to errors; however, clustering multiple trajectories together minimizes errors and uncertainties. I calculated 48 hour, 3-dimensional kinematic back-trajectories ending at the time and location of each aircraft spiral (made from 1997 ? 2003 with the UMD research aircraft) using the NOAA Air Resources Laboratory (ARL) HYbrid Single- Particle Lagrangian Integrated Trajectory (HY-SPLIT) model (Version 4) (R. R. Draxler and G. D. Rolph, 2003, http://www.arl.noaa.gov/ready/hysplit4.html) and 80 km Eta Data Assimilation System (EDAS) 3-hourly archive data. Kinematic back 40 trajectories were used because, due to improvements to the accuracy of the vertical wind component, they have been shown to be more accurate than other methods (e.g., isentropic and isobaric) (Stohl et al., 1995; Stohl, 1998; Jorba et al., 2004). Two-day back trajectories were long enough to capture regional transport patterns and short enough to keep trajectory errors, which accumulate with simulation time, to acceptable levels. The air parcel latitudes, longitudes, and pressures were recorded at 1 h intervals. Trajectories were calculated ending at 1, 2, and 3 km (above ground level). Back trajectories associated with 550 flights made from 1997-2003 were used in the clustering analysis. 3.2.3. Cluster Analysis I performed a separate cluster analysis for back trajectories ending at 1, 2, and 3 km. These ending altitudes describe the vertical range over which the aircraft vertical survey spirals were performed. By clustering the trajectories at each of the three altitudes, any variations in the atmospheric circulation patterns in the lower atmosphere and the impacts on regional transport could be identified. The results of the cluster analysis for the three altitudes were similar. The 2 km trajectory cluster results were used for the remainder of the analysis because this altitude is near the middle of the aircraft spirals and most representative of the entire spiral. The trajectories trace a 3-dimensional path through time to the receptor site. To determine the similarity among individual trajectories, the total variability between each trajectory pair must be quantified. The variability may be calculated as a scalar distance between trajectories. At each time step, the position of the air parcel is defined by its latitude, longitude, and pressure. These data were converted to 41 Cartesian coordinates by treating the Earth as a sphere and calculating their position in 3-dimensional space. The x, y and z distances in km are given as: x = (r e + alt i )*cos(?/180 * lon i )* sin(? /2 ? ? /180*lat i ) (3) y = (r e + alt i )*sin(? /180 * lon i )* sin(? /2 ? ? /180*lat i ) (4) z = (r e + alt i )* cos(? /2 ? ? /180*lat i ) (5) Where r e is the radius of the earth (approximated at 6378 km), alt i , lat i , and lon i are the altitude, latitude and longitude at a specific hour i, in the back trajectory. Vertical variability along the trajectory paths may have a large impact on transport and hence, pollution levels, but the spatial distances described by the variability in the vertical wind component are typically less than the horizontal spatial distances covered by the air parcels. Thus, without normalizing the data, the vertical variability may not have an equal impact on the cluster analysis when examining the similarity among trajectories. To account for this inconsistency I calculated the mean value and standard deviation for each coordinate at every time step (x avg , y avg , z avg and x stdev , y stdev , z stdev ). I then subtracted the mean value from the individual coordinates and normalized them with the standard deviation. x* = (x ? x avg ) / x stdev (6) Normalized differences (like x*) were also calculated for y and z. In this way, the coordinates were all converted to a standardized distance from the mean value of that particular coordinate and equal weighting was given to all three coordinates in the cluster analysis. The Euclidean distance, D, between each trajectory pair was calculated according to the equation: 42 222 1 )**()**()**( jkikjkij n k jkikij zzyyxxD ?+?+?= ? = (7) In the above equation, D is the 3-dimensional distance between the two trajectories under comparison, represented here by the subscripts i and j. The variables x*, y*, and z* represent the normalized distances from the means of the Cartesian coordinates. The number of time steps used in the analysis is given by k (48 for hourly time steps over 2 days). However, the first six time steps back from the receptor site were given zero weighting to account for the spatial heterogeneity of the aircraft spiral locations. To further discount the spatial variability of the receptor locations and place the emphasis on the source regions, the trajectory time steps were weighted linearly back in time, increasing the weighting for each hour after the initial six zero-weighted time steps. After the distances between individual trajectories were calculated I clustered the trajectories using an agglomerative, hierarchical clustering algorithm in Matlab. The algorithm used an average linkage function, where the average distances between all pairs of objects in clusters i and j are calculated, to determine the distances between the trajectories making up the clusters. Average linkage minimizes the within-cluster variance while maximizing between-cluster variance and has been identified as an effective method for categorizing different synoptic situations (Kalkstein et al., 1987). Newly formed clusters were linked to other trajectories to create successively larger clusters until all of the trajectories were connected by a hierarchical dendrogram. The algorithm has no inherent mechanism for identifying an appropriate terminus for this iterative process. Barring manual intervention, all 43 objects are eventually grouped into one cluster. So, the final number of clusters was specified arbitrarily from 1 to 15 clusters. To determine the appropriate number of clusters I first calculated an ?average? trajectory, or trajectory center, for each cluster. The root mean square deviation (RMSD) of each trajectory within a cluster from the cluster center was quantified. The RMSDs were then summed to give the total root mean square deviation (TRMSD). The percent change in the TRMSD was plotted against the total number of clusters (e.g., Cape et al., 2000; Brankov et al., 1998) (Figure 2). Large changes were interpreted as the merging of significantly different trajectories into the same cluster. Accordingly, the appropriate number of clusters would be found just prior to the large percent change in TRMSD. While this technique lends objectivity to the analysis, a subjective interpretation of the optimal number of clusters based on the meteorology and pollutant profiles as well as what value constitutes a large enough percent change in TRMSD is still required. When eight clusters were merged into fewer clusters the change in TRMSD remained consistently high (~10%) and grew larger upon further agglomeration. After reviewing the meteorology and pollutant profiles associated with each cluster, eight was determined to be the appropriate number of clusters 44 Figure 2. Percent change in the total root mean square deviation (TRMSD) calculated by summing the root mean square deviation of each cluster versus the number of total clusters. 3.3. Results and Discussion 3.3.1 Cluster Solution The eight cluster solution is shown in Figure 3 with trajectory ?spaghetti? plots of each cluster. The relative density of air parcel locations in each cluster, however, is better described by trajectory density plots, given in Figure 4. A linear interpolation method was used to generate values between the original trajectory latitude and longitude points and smooth the density plots. The locations of the largest (top 0.3%) annual NO x and SO 2 emitters in the eastern U.S. (EPA AirData Facility Emissions Report ? Criteria Air Pollutants 1999, http://www.epa.gov/air/data/) are overlaid on the trajectory density maps. The trajectory densities represent the relative amount of time the air parcels from every trajectory in a cluster spent over the areas defined by the spaghetti plots before reaching the receptor location. This is a technique based on the ?residence time 45 analysis? (Ashbaugh, 1983) and it will be shown below that it is an effective means of determining downwind pollutant loadings. Figure 3. Spaghetti plots of the 48 hr HY-SPLIT back trajectories ending at 2km altitude that make up a) Cluster 1, b) Cluster 2, c) Cluster 3, d) Cluster 4, e) Cluster 5, f) Cluster 6, g) Cluster 7, and h) Cluster 8. 46 47 Figure 4. Trajectory density maps for a) Cluster 1, b) Cluster 2, c) Cluster 3, d) Cluster 4, e) Cluster 5, f) Cluster 6, g) Cluster 7, and h) Cluster 8. The plots were generated using a linear interpolation method between the trajectory end points. They indicate the relative density (%) of air parcels over the total area described by the spaghetti plots. Also pictured are the locations of the top 0.3% emitters annually of NO x (diamonds) and SO 2 (crosses) in the eastern U.S. To determine the statistical difference between the constituents associated with each cluster I first subdivided the clusters into morning and afternoon profiles according to the aforementioned criteria. For all trace gas profiles but morning O 3 , I calculated the boundary layer (defined here as the layer between 100 m and 2000 m) column content (in matm-cm). For morning O 3 , the residual layer (defined here as the layer between 500 m and 2000 m) column content (also in matm-cm) was quantified to capture the impacts of regionally transported pollution on the receptor locations. For the aerosol profiles I calculated the extinction weighted single scattering albedo column average, aerosol optical depth at 550 nm between the surface and 3 km (AOD), and scattering weighted ? . The cluster median values were then determined. The cluster median ranks are given in Table 1 and the cluster median values for the extinction weighted single scattering albedo and AOD are given in Table 2. 48 Table 1. Cluster median profile ranks, % O 3 transported, and cluster median column content SO 2 /CO ratios for the morning (upper number) and afternoon (lower number). Values in parentheses under the cluster number are the total profiles that went into that cluster (left) and the percent of the total profiles (right). The grey area indicates insufficient data to calculate statistical values for Cluster 8. clusters Median O 3 rank % O 3 transported Median CO rank Median SO 2 rank SO 2 /CO Median 5500 ? rank Median AOD rank Median ? rank 1 (107,26.3) 1 2 67 + 4 a 4 7 4 2 0.014 0.017 2 1 1 1 6 7 2 (77,19.0) 4 5 67 + 6 8 8 5 4 0.015 0.016 1 2 2 2 7 6 3 (108,26.6) 3 1 54 + 8 3 3 7 3 0.009 0.011 3 4 3 4 4 4 4 (39,9.6) 6 8 82 + 8 7 2 8 6 0.010 0.006 4 3 5 3 2 5 5 (24,5.9) 5 3 62 + 16 5 6 3 7 0.015 0.008 5 5 6 7 1 1 6 (15,3.7) 2 4 73 + 17 2 4 1 1 0.026 0.016 6 7 4 5 3 3 7 (23,5.7) 8 7 56 + 16 6 5 6 8 0.013 0.008 7 6 7 6 5 2 8 (13,3.2) 7 6 55 + 11 1 1 2 5 0.008 0.006 a error estimated by adding in quadrature 1? /?n from the residual layer and afternoon boundary layer mean values. 49 Table 2. Cluster median profile values for the morning and afternoon aerosol extinction weighted 5500 ? and AOD. clusters a am 5500 ? pm 5500 ? am AOD pm AOD 1 0.91 + 0.05 b 0.95 + 0.04 0.37 + 0.19 0.35 + 0.30 2 0.91 + 0.06 0.94 + 0.05 0.31 + 0.23 0.31 + 0.25 3 0.90 + 0.06 0.93 + 0.04 0.30 + 0.28 0.26 + 0.12 4 0.88 + 0.06 0.94 + 0.03 0.22 + 0.06 0.29 + 0.07 5 0.87 + 0.08 0.91 + 0.05 0.17 + 0.10 0.15 + 0.10 6 0.82 + 0.09 0.85 c 0.25 + 0.08 0.25 c 7 0.81 + 0.17 0.85 + 0.12 0.15 + 0.12 0.20 + 0.12 a there were not enough data to calculate statistical values for Cluster 8 b the error represents 1? of the cluster mean value c no error is given because there was only one profile that went into the Cluster 6 pm 5500 ? and AOD values Using the individual profile values, I calculated the statistical difference between the cluster medians using a multiple comparison procedure with statistical values generated from the Kruskal-Wallis test. This test is similar to the standard one-way analysis of variance, but is a non-parametric version. The one-way analysis of variance requires data to be normally distributed whereas in the Kruskal-Wallis test the data must only be continuously distributed. The test ranks the values and performs an analysis of variance on the ranks rather than the values themselves. For this study, the cutoff for the probability value (p-value) was set to 0.05. When the p- value was less than this limit, the null hypothesis was rejected and the cluster medians were declared statistically different with greater than 95% confidence. The results are summarized in Table 3. 50 Table 3. Statistical difference among cluster morning and afternoon profile median values. The > (<) signifies that the median value for the cluster is statistically greater (less) than, with 95% confidence, the median values of the cluster numbers listed after the symbol. Grey areas indicate no statistical difference. Clusters 1 2 3 4 5 6 7 8 am O 3 >7,8 >8 >8 <1 <1,3,6 pm O 3 >4 <3 >2,4,7 <1,3 <3 am CO <8 <8 >2,4 pm CO am SO 2 <6 <6 >3,4 pm SO 2 am 5500 ? pm 5500 ? am AOD >4,5 <1 <1 pm AOD >5 >5 <1,2 am ? <5 <5 >1,2,7 <5 pm ? <5,7 >1 >1 3.3.2 Pollution Profiles Figures 5 and 6 show the morning and afternoon median vertical profiles, respectively, for each constituent. Cluster 1, associated with large amounts of O 3 , a 51 large SO 2 /CO ratio, large, highly scattering particles, and a large AOD (see Figures 5, 6, and Tables 1, 2), shows moderate northwesterly flow. These values are indicative of aged point source pollution. The greatest trajectory density lies over the northern Ohio River Valley where there are several large NO x and SO 2 sources (see Figure 4). Fresh NO x and SO 2 emissions from these sources have had ample opportunity under a moderate flow regime to produce O 3 and secondary aerosols en route to the Mid- Atlantic. Cluster 2 shows similar wind direction to Cluster 1, but with higher wind speeds (see Figure 3). The greatest trajectory density also lies mainly over the northern Ohio River Valley and extends into the Great Lakes region. The particles are also large and highly scattering, but the AOD is lower. The CO is even less than in Cluster 1, the SO 2 /CO ratio is large, and the O 3 values, particularly in the afternoon, are small (see Figures 5, 6, and Tables 1, 2). These values are all consistent with northwesterly flow similar to Cluster 1 that brings northern Ohio River Valley point source pollution, but with higher wind speeds, so that there is less time for local, photochemical O 3 production or mixing with urban, mobile source pollution. Figure 4 shows that, in fact, the greatest trajectory density intersects many large NO x and SO 2 sources. Cluster 3 is typified by stagnant conditions with light, southerly flow (see Figure 3). The greatest air parcel density is found over the central Mid-Atlantic region. Ozone values, particularly in the afternoon, are large, as are CO values, whereas SO 2 values, especially in the morning, are small. Hence, the SO 2 /CO ratio is small. The particle property values are moderate (see Figures 5, 6, and Tables 1, 2). 52 These values, together with the stagnant conditions associated with Cluster 3, indicate local, urban, mobile source-dominated pollution. Figure 4 shows that there are few large NO x and SO 2 sources in the area of greatest trajectory density. Presumably, because there is less hygroscopic sulfate available for particle growth, the particles are smaller and less scattering than in the first two clusters. The transport pattern identified by Cluster 4 is characterized by moderate southwesterly flow and the greatest trajectory density lies over the southern Ohio River Valley (see Figures 3, 4). For the most part, this cluster is associated with little pollution loading and the SO 2 /CO ratio is small (see Figures 5, 6, and Tables 1, 2), suggesting that there are fewer point sources located farther south along the Ohio River. Figure 4 shows no large NO x or SO 2 sources in the area of greatest trajectory density, although many do encircle this area. Also, the afternoon O 3 values are particularly small, and not much larger than the morning values (see Figures 5, 6, and Table 1), indicating there was little photochemical production during the air parcels transport. Cluster 5 shows fairly fast north-northwesterly flow over the northern Great Lakes region into the Mid-Atlantic region (see Figure 3). Generally, this flow pattern seems to transport little pollution into the region. However, the O 3 values are large below ~1200 m in the morning and ~1500 m in the afternoon and only fall off to lower values aloft (see Figures 5, 6). The areas of greatest trajectory density do intersect large NO x and SO 2 sources (see Figure 4), a fact corroborated by a high SO 2 /CO ratio in the morning (Table 1), but the wind speeds, particularly aloft, are too great to allow for pollution to accumulate in the Mid-Atlantic region. The fast wind 53 speeds are also not conducive to particle growth, so the smallest particles are found in this cluster. The wind direction of the trajectories in Cluster 6 is northwesterly as in Clusters 1 and 2, with still faster wind speeds than in Cluster 2 (see Figure 3). The greatest trajectory density again lies over the northern Ohio River Valley and several large NO x and SO 2 sources (see Figure 4). The pollution loadings of this cluster are also consistent with these sources, but because of the higher wind speeds, the pollution appears to be relatively fresher. The O 3 values are moderately large, with smaller values in the afternoon, the CO values are moderate, and the SO 2 values are very large, so that the SO 2 /CO ratio is also very large (see Figures 5, 6, and Table 1). The SO 2 apparently did not have much opportunity for oxidation before entering the Mid-Atlantic region. The particles were smaller and less scattering and the AOD was smaller than in Clusters 1 and 2 (see Figures 5, 6, and Tables 1, 2). Overall, Cluster 7 is associated with the least pollution of any of the clusters (see Figures 5, 6, and Tables 1, 2). The flow is out of the north, bringing relatively cool, dry, continental air to the Mid-Atlantic region. There are no major urban centers, nor are there many NO x or SO 2 sources in the area of greatest trajectory density (see Figure 4). Cluster 8 comprises very few trajectories. The flow is fast and from the southwest, originating near Texas (see Figure 3). There were not enough particle data to generate any statistical values. The O 3 values are small, the SO 2 values are moderate, and the CO is very large (see Figures 5, 6, and Table 1). The areas of greatest trajectory density do no intersect many large NO x or SO 2 sources (see Figure 54 4). Despite the fast wind speeds, the air parcels appear to be picking up a local, mobile source, indicated by the large CO values, small SO 2 /CO ratio, and trajectory densities. Figure 7 summarizes the transport from the areas of greatest trajectory density into the Mid-Atlantic region as a percent of the total number of profiles examined in this study. 55 Figure 5. The median morning profiles for Clusters 1 (brown), 2 (red), 3 (orange), 4 (yellow), 5 (green), 6 (dark blue), 7 (blue), and 8 (violet) of a) O 3 , b) CO, c) SO 2 , d) single scattering albedo (550 nm), and e) ?ngstr?m exponent (575 nm). 56 Figure 6. The median afternoon profiles for Clusters 1 (brown), 2 (red), 3 (orange), 4 (yellow), 5 (green), 6 (dark blue), 7 (blue), and 8 (violet) of a) O 3 , b) CO, c) SO 2 , d) single scattering albedo (550 nm), and e) ?ngstr?m exponent (575 nm). 57 3.3.3 O 3 Transport Thus far, the first of the two original questions posed has been addressed. Namely, a statistical link between characteristic regional transport patterns into the Mid-Atlantic during summertime haze and O 3 episodes and specific pollution loadings has been established. In this section I will quantify the contribution of regionally transported O 3 to afternoon boundary layer O 3 over the Mid-Atlantic. The percent of the afternoon O 3 boundary layer column content for each cluster that can be accounted for by regional transport was estimated with the following equation: % O 3 transported = 100? ? ? ? ? ? ? ? ? ? ? ? ? ? ABL MBL MBL RL , (2) where RL is the residual layer column content, MBL is the morning boundary layer column content, and ABL is the afternoon boundary layer column content. The equation simplifies to the ratio of RL/ABL after the MBLs cancel out. The accuracy of this method depends upon the Lagrangian nature of the morning and afternoon profiles from each cluster. Because flight plans were designed in a Lagrangian manner, where morning flights were upwind of afternoon flights, the estimate should be accurate. The results are shown in Table 1 and Figure 7. The amount of afternoon O 3 that can be accounted for by regional transport ranges from a low of 55% to a high of 82%. One of the smallest contributions from transport (58%) corresponds to Cluster 3. This cluster shows the most stagnant conditions so that transport would not be expected to contribute as much to the afternoon totals (the weak winds allow for transport of only a few hundreds of km in a 24 hr period). The largest contributions from regional transport are seen in Clusters 1(70%), 2(69%), 4(82%) and 6(73%). 58 The trajectory density plots (see Figure 4) show that their greatest air parcel densities are over the Ohio River Valley. Those of Cluster 4 lie over the southern Ohio River Valley whereas those of the other three Clusters all lie over the northern portion of the Ohio River. While the pollution loadings associated with Cluster 4 are relatively small, those in Clusters 1, 2, and 6, particularly with respect to O 3 , SO 2 , and particle pollution, are large. In general, the greatest regional O 3 transport was from the Ohio River Valley, while some of the least transport occurred during clean, northerly flow (Cluster 7) and when stagnant conditions persisted and photochemical production was highest (Cluster 3). Our analysis neglected O 3 produced locally from precursors transported from upwind and may thus be an under-estimate of the role of transport. Figure 7. Pie chart showing the transport from the particular areas, as defined by the trajectory densities in each cluster, into the Mid-Atlantic region as a percent of the total number of profiles examined in the study. The Northern Ohio River slice comprises Clusters 1, 2, and 6. 59 3.4. Conclusions Several years of episodic, summertime aircraft vertical profile trace gas and aerosol data collected as part of the Regional Atmospheric Measurement, Modeling, and Prediction Program (RAMMPP) were analyzed in this study. The data were divided into morning and afternoon profiles to identify diurnal patterns. Little diurnal variation was observed in the CO, SO 2 , or ?ngstr?m exponent but O 3 values were greater in the afternoon than the morning. O 3 above the planetary boundary layer in the lower free troposphere, amenable to long range transport, was consistently ~55 ppbv. The single scattering albedo was larger in the afternoon than the morning, likely the result of VOC and SO 2 oxidation to secondary organic aerosols and sulfate, respectively. A decrease in the single scattering albedo above 2000 m was due to invariant absorption values with altitude combined with scattering values that declined with altitude. This phenomenon could have a large-scale radiative impact, although the aerosol extinction in the lower free troposphere may be too low to have any significant effects. Even so, this occurrence merits further investigation. Characteristic transport patterns and source regions during summertime haze and O 3 episodes were analyzed with an agglomerative hierarchical cluster analysis of back trajectory data. Eight clusters were identified, which were then divided into morning and afternoon profiles. The median profile values were calculated and statistical differences were determined using a nonparametric procedure. When the greatest trajectory density lay over the northern Ohio River Valley and large NO x and SO 2 sources, the result was large O 3 values, a large SO 2 /CO ratio, large, scattering particles, and high AOD over the Mid-Atlantic U.S. In contrast, relatively clean 60 conditions over the Mid-Atlantic occurred when the greatest trajectory density lay over the southern Ohio River Valley and nearly missed many large NO x and SO 2 sources. The greatest afternoon O 3 occurred during periods of near stagnation (when the average wind speed was 4.4m s -1 at 2km) that were most conducive to photochemical production. The least pollution occurred when flow from the north- northwest was too fast for pollution to accumulate and when flow was from the north, where there are few urban or industrial sources. O 3 transport over several hundred kilometers into the Mid-Atlantic U.S. was estimated by calculating the ratio of the residual layer O 3 in the upwind morning profiles to the downwind afternoon boundary layer values. The greatest O 3 transport (69-82%) occurred when the maximum trajectory density lay over the southern and northern Ohio River Valley (~59% of the total profiles). The least O 3 transport (55- 58%) was associated with fast southwesterly flow (~3% of the total profiles), clean northerly flow (~6% of the total profiles), and stagnant, polluted conditions (~27% of the total profiles). Altogether, about 64% of the O 3 during an episode is already present as the air enters the Baltimore/Washington area from the West. In summary, this investigation demonstrated the ability to identify important statistical differences among pollution profiles that resulted from seemingly minor variations of the typical summertime, polluted meteorological regime. When trajectory density plots were overlaid on maps with the largest annual NO x and SO 2 emitters, specific source regions were identified. The results indicate that the areas of maximum trajectory density together with wind speed are effective predictors of regional pollutant loadings. Additionally, due to the Lagrangian nature of the dataset, 61 the regionally transported contribution to the total afternoon boundary layer column O 3 content in each cluster could be quantified. 62 Chapter 4: Cluster Analysis of Pollutant Profiles. 4.1 Introduction 4.1.1 Background Understanding the influences of meteorology and emissions on the vertical distribution of trace gases and aerosols can improve modeling and prediction of pollution events. Some work presented in this chapter is from Hains et al. (2007a). I have developed a method to cluster vertical profiles of trace gases and aerosols. I then examined meteorological conditions as indicated by back trajectories associated with each cluster. Results from this cluster analysis are used to explain meteorological and emission influences on the vertical distribution of trace gases and aerosols. I have clustered over 150 profiles of O 3 , CO, SO 2 , absorption, scattering, and ?ngstr?m exponent (?) collected between 1997-2003 in June, July and August. I developed a method for integrating point source emission sources (from the EPA?s AirData database) encountered by a 48-hr back trajectory. I have employed this method to explain the relationship between point source emissions and the different clustered profiles. 4.1.2 Cluster Analysis Statistical cluster analysis involves determining the differences between the objects being analyzed, and clustering those objects with the smallest differences. The trace gases presented in Chapter 3 showed distinctive profiles; for example, most of the SO 2 was found below 500 m throughout the day, while O 3 was most concentrated 63 above 500 m in the morning and was more uniform from the surface to 2000 m in the afternoon (the largest values in the profile were found near the 1100 m level). For this work, the raw data were averaged into altitude layers of 100 m (gases) or 200 m (aerosols) and then the layers were grouped into bins shown in Table 1. The slope and correlation of the points in each pair of profiles under comparison were considered as well as the total difference in values between the two profiles within each altitude bin. The following equation was used to calculate the differences among the aircraft profiles: ()())1(])1(exp(1[]1[1*)( 4 1 2 2 1 ?? = = = = ? ? ? ? ? ? ???+?+?= k k na a jaia srccabsDij k Here, k is the index for the four different altitude bins for the profiles and a is an index for the n k layers within the k th bin (Table 1). The species value is represented by c for the i th and j th profile. In each altitude bin, k, there are at least four layers of trace gas or aerosol data. A regression was made to obtain the slope, s, and the correlation coefficient, r, for each pair of profiles using the mixing ratio (or aerosol coefficient) within the k bins. The first part of Equation 1 determines the square of the sum of the differences between values at each altitude bin, k. The second part of the equation multiplies the difference by one plus differences associated with the correlation and slope. When the correlation is small or negative the profiles are very different and the 1- r portion increases, which increases the total difference D ij . As the correlation coefficient approaches unity the 1-r portion approaches zero, and this decreases the total difference D ij . The exponent of the slope portion is used to account for the slope between the pairs of profiles. A slope near unity suggests that the profiles are similar and thus should add little to the total difference. The exponent 64 of the slope was used to guarantee that slopes much different from unity will make the exponential term small and thus increase the total difference. Once the difference between each pair of profiles has been calculated, the profiles with the smallest differences are clustered. These clusters are constructed from hierarchical cluster trees generated with an average linkage algorithm in Matlab (described in Chapter 3 section 3.2.3). k bin Altitude bins for trace gases Number of layers in bin (m k ) Altitude bins for aerosols Number of layers in bin (m k ) 1 151-650 m 5 100-900 m 4 2 651-1150 m 5 901-1700 m 4 3 1151-1650 m 5 1701-2500 m 4 4 1651-2450 m 8 Table 1. Altitude bins for trace gases and aerosols used in Equation 1. 4.2 Results 4.2.1 O 3 Figure 1 shows the median profiles for each of the six O 3 clusters calculated in the above manner. The clustering technique identified a small group of outliers, Cluster 6, with large values of O 3 above 2000 m altitude. These profiles were made on 8 and 9 July 2002 when smoke from Canadian forest fires impacted the Mid- Atlantic region (Colarco et al., 2004; Taubman et al., 2004a). The transported O 3 can be seen in the peak (up to 150 ppb) above 2000 m. This shows that the statistical technique employed can separate anomalous episodes. The quartiles for the six clusters rarely overlap, which further exemplifies how the method was able to separate distinct events. 65 Figure 1. Median O 3 profiles for each cluster. Error bars represent the 25 th and 75 th percentiles. The number of profiles in each cluster is shown in parentheses in the key. Clusters 1 and 2 show the smallest O 3 values, while Clusters 3, 4 and 5 show the largest. Cluster 6 profiles were made when the Canadian forest fires impacted the region and the peak above 2000 m shows their influence. I calculated two-day HYSPLIT back trajectories for each spiral, ending at an altitude of 1, 2 and 3 km, at the latitude and longitude of the spiral and at the time the spiral was made. Back trajectories were similar at all altitudes and so I chose those ending at 1 km because these are most likely to be closer to point source emissions. Profiles from Clusters 3, 4 and 5 had large O 3 values and the back trajectory density plots (Figure 2) show passage over the Northern Ohio River Valley, where there is a higher concentration of NO x sources. Taubman et al. (2006) found a similar relationship between back trajectories concentrated over the Northern Ohio River Valley and large mixing ratios of O 3 and suggested that the large concentration of 0 500 1000 1500 2000 2500 0 50 100 150 200 Median O 3 (ppb) A l t i t ude ( m ) Cluster 1 (80) Cluster 2 (52) Cluster 3 (66) Cluster 4 (28) Cluster 5 (4) Cluster 6 (2) 66 power plants in this region contributes to the O 3 in the Mid-Atlantic region. The back trajectory density plots for Clusters 3-5 also show larger densities around the I-95 corridor, which is suggestive of stagnation events that lead to higher O 3 values. Cluster 2 has back trajectories that pass over the Atlantic Ocean, which may explain the smaller O 3 values associated with this cluster. Cluster 1 has the second smallest O 3 values (Cluster 1 column content is 19% less than that of Cluster 3), even though the back trajectories associated with Cluster 1 are similar to those of Cluster 3. To address this discrepancy, the integrated NO x point source emissions along the back trajectories were examined. 67 Figure 2. Back trajectory density plots for O 3 Clusters 1-5. The top 0.3% NOx sources are shown with a + symbol. Clusters 3, 4 and 5 associated with larger O 3 values show larger densities near point sources as well as along the I-95 corridor, suggestive of stagnation. Cluster 5 also has an unusual flow pattern from the northeast. Clusters 1-2, with smaller O 3 values are associated with more variable winds. 68 I integrated the NO x emissions along each back trajectory to explain the influence of upwind emissions on upwind ozone mixing ratios. Emissions from the daily EPA Clean Air Market unit level emissions database (http://cfpub.epa.gov/gdm/index.cfm?fuseaction=emissions.wizard) were used in this study. I drew a circle, centered at each back trajectory position for each hour of the two day back trajectory (Figure 3). The radius of the circle was 80 km to account for uncertainties associated with the back trajectory position and the influence of eddy diffusion and mixing processes. The emissions within each circle were summed. The sum of the emissions for each circle was divided by the area of the circle. I used emissions from the day on which the back trajectory crossed a circle for the date of each back trajectory position. The summed emissions will be referred to as integrated emissions. Statistics (median, 5 th , 25 th , 75 th and 95 th percentiles) for the integrated NO x emissions for each O 3 cluster are shown in Figure 4. 69 Figure 3. Circles drawn around an example back trajectory. The emissions contained in each circle were summed and divided by the area of the circle. Then emissions from each circle were summed. The pink * represent point source locations. 70 Figure 4. Statistics for NOx emissions encountered by back trajectories for each O 3 cluster. The NOx emissions are sums of all emissions (g day -1 ) encountered by a back trajectory divided by the area of the circle drawn around the back trajectory point (m 2 ). Clusters 1 and 2, with the smallest O 3 values are also associated with the smallest NOx emissions. Clusters 3, 4 and 5, with the largest O 3 values are associated with the largest NOx emissions. Clusters 3, 4 and 5 have the largest O 3 column contents and the largest NO x emissions, while Clusters 1 and 2 have the smallest O 3 column contents and the smallest NO x emissions. Even though back trajectory density maps for Clusters 1 and 3 are similar, Cluster 1 is associated with 21% less integrated NOx emissions, explaining the 18% smaller O 3 values. The median O 3 column content and integrated NO x emissions for Clusters 1 through 5 show a positive relationship (Figure 5), suggesting that NO x emissions from point sources play an important role in downwind O 3 production. 0 1 2 3 4 Cluster 1 (80) Cluster 2 (52) Cluster 3 (66) Cluster 4 (28) Cluster 5 (4) O 3 Cluster In t e g r a t e d N O x e m is s i o n s (g m -2 da y -1 ) Q25 Q5 median Q95 Q75 71 Figure 5. Median O 3 column content and median integrated NOx emissions (from point sources) for O 3 Clusters 1 through 5. The O 3 is positively correlated with integrated NOx emissions. Profiles were also analyzed by time of day, where morning profiles are defined as those made before 12 noon EST and afternoon as profiles made after 12 noon EST. Sixty-one and sixty-eight percent of the profiles in Clusters 3 and 4 were measured in the afternoon, whereas only 38% and 46% of the profiles in Clusters 1 and 2 were measured in the afternoon. Greater O 3 values in Clusters 3 and 4 may be partly explained by the increased number of afternoon profiles which were generally made downwind of urban centers, and had more time for O 3 production. 4.2.2 SO 2 I also clustered the SO 2 profiles and generated three distinct SO 2 profile clusters (Figure 6). Of the 192 profiles analyzed, 170 (89%) fell into the relatively clean Cluster 3. The other clusters reflect large values of SO 2 at altitudes from near O 3 and NO x emissions y = 14.671x + 0.313 R 2 = 0.997 0 5 10 15 20 25 0 0.5 1 1.5 NO x emissions (g m -2 day -1 ) O 3 co lu m n co n t en t ( m at m cm ) Y = 14.6x + 0.3 R 2 = 0.997 72 the surface (Cluster 2) to 1000 m (Cluster 1). Back trajectories associated with Cluster 3 show a broader area of origin than the more heavily polluted Clusters 1 and 2 (Figure 7). The median SO 2 profile from Cluster 2 shows large values near the surface that decrease above 500 m. Profiles in Cluster 1 show large values near the surface that do not drop off as rapidly as those in Cluster 1, indicating better mixing in the lower troposphere. 0 500 1000 1500 2000 2500 0 5 10 15 20 25 Median SO 2 (ppb) A l ti tu d e (m ) Cluster 1 (10) Cluster 2 (12) Cluster 3 (170) Figure 6. Median SO 2 profiles for each cluster. Error bars represent the 25 th and 75 th percentiles. The number of profiles in each cluster is shown in parentheses in the key. Cluster 3 is the background Mid-Atlantic summertime SO 2 profile, representing the majority of SO 2 profiles measured. Clusters 1 and 2 represent profiles impacted by chance plume encounters. 73 Figure 7. Back trajectory density plots for SO 2 Clusters 1-3. The top 0.3% SO 2 sources are shown with a circle. Back trajectories associated with Clusters 2 and 3 show more density over SO 2 sources while the back trajectories associated with Cluster 1 show more variable origins. The integrated SO 2 emissions along each back trajectory and statistics for each SO 2 cluster were calculated (Figure 8). The SO 2 emissions do not show as much range as the NO x emissions. The lack of relationship between emissions and the SO 2 profiles, and the small number of meaningful SO 2 clusters generated, suggests profiles with larger values are likely the result of chance encounters with fresh plumes, and that the lifetime of SO 2 in the summer is shorter than 48 hours. The lifetime of SO 2 is addressed in Chapter 5. o Top 0.3% SO 2 sources Cluster 1 Cluster 2 Cluster 3 0.0% 1.0% 2.0% 3.0% 4.0% 5.0% Percent density 74 Figure 8. Statistics for SO 2 emissions encountered by back trajectories for each SO 2 cluster. The SO 2 emissions are sums of all emissions (g day -1 ) encountered by a back trajectory divided by the area of the circle drawn around the back trajectory point (m 2 ). The SO 2 emissions show little relationship with the profiles. 4.2.3 Particle Scattering Figure 9 shows the median scattering coefficients (in units of m -1 ) for all flights conducted between 2001 and 2003 (June through August). The clustering methodology produced four scattering clusters, but two are sparsely populated (Clusters 3 and 4) and associated with the Canadian forest fire episode (Figure 9). The median scattering profile for Cluster 1 is similar to the median profile of all flights. Cluster 1, with 125 profiles, has smaller values than Cluster 2, with 48 profiles (Figure 9). Back trajectories associated with profiles from Cluster 2 show winds from the Northern Ohio River Valley, while Cluster 1 has back trajectories 0 2 4 6 8 10 Cluster 1 (10) Cluster 2 (12) Cluster 3 (170) SO 2 cluster I n t e g r at ed S O 2 em i ssi o n s ( g m -2 da y -1 ) Q25 Q5 median Q95 Q75 75 with more variable winds and greater mean wind speeds (Figure 10). Slower wind speeds or stagnant conditions allow time for the conversion of SO 2 to sulfate. Figure 11 shows statistics of integrated SO 2 emissions for each cluster. The median SO 2 emissions for Cluster 2 are almost a factor of two greater than those for Cluster 1. This suggests that the aerosol loading reflects the SO 2 emitted into the air parcel over the previous 48 hours. 76 Figure 9. Median scattering profiles for each cluster. Error bars represent the 25 th and 75 th percentiles. The number of profiles in each cluster is shown in parentheses in the key. Cluster 2 has profiles with twice the scattering value as Cluster 1. Profiles from Clusters 3 and 4 were measured when the Canadian forest fires impacted the region. 0 500 1000 1500 2000 2500 3000 0 100 200 300 400 500 600 Median particle scattering at 550 nm (Mm -1 ) A l ti tu d e (m ) Median (all flights) Cluster 1 (139) Cluster 2 (34) Cluster 3 (2) Cluster 4 (1) 77 Figure 10. Back trajectory density plots for scattering Clusters 1and 2. The top 0.3% SO 2 sources are shown with a circle. Back trajectories associated with Clusters 2 show more density over SO 2 sources while the back trajectories associated with Cluster 1 show more variable origins. Cluster 1 Cluster 2 o Top 0.3% SO 2 sources 0.0% 1.0% 2.0% 3.0% 4.0% 5.0% Percent density 78 Figure 11. Statistics for SO 2 emissions encountered by back trajectories for each scattering cluster. The SO 2 emissions are sums of all emissions (g day -1 ) encountered by a back trajectory divided by the area of the circle drawn around the back trajectory point (m 2 ). Cluster 2 is associated with almost double the emissions of Cluster 1, explaining why cluster 2 profiles have double the scattering values as Cluster 1 (Figure 9). The conversion from SO 2 to sulfate in the summer is so rapid that there is little discernable relationship between SO 2 emissions from the Ohio River Valley and SO 2 values in the Mid-Atlantic. There is a positive relationship between SO 2 emissions in the Northern Ohio River Valley and particle scattering in the Mid- Atlantic, indicating an important source of sulfate aerosols is from Northern Ohio River Valley coal fired power plants (e.g., Taubman et al. 2006). The stronger relationships between emissions and SO 2 and aerosol profiles suggest that the lifetime of sulfate is longer than 48 hours. 0 2 4 6 8 10 12 Cluster 1 (139) Cluster 2 (34) Cluster I n t e g r at ed S O 2 em i s s i o n s ( g m -2 da y -1 ) Q25 Q5 median Q95 Q75 79 4.2.4 Angstrom Exponent The clustering methodology produced four distinct ? clusters (Figure 12). The median profile for Cluster 1 has relatively small ? values and thus represents larger particles. The back trajectories associated with Clusters 1 and 2 are concentrated over point sources in the Northern Ohio River Valley (Figure 13). Cluster 1 back trajectories however, are slower allowing more time for particle growth. Profiles from Cluster 1 are associated with larger integrated SO 2 emissions than the other clusters (Figure 14). Profiles in Cluster 4 have the largest ? values and the least integrated SO 2 emissions. The ? values are calculated from scattering measurements, explaining why they show a relationship similar to that between SO 2 emissions and scattering. Profiles in Cluster 3 decrease sharply above 2000 m and represent large particles. Four of the profiles in Cluster 3 were measured during the Canadian forest fires and one profile was measured on the 4 th of July, when large particles would also be expected. 80 Figure 12. Median Angstrom exponent profiles for each cluster. Error bars represent the 25 th and 75 th percentiles. The number of profiles in each cluster is shown in parentheses in the key. Clusters 1 and 3 have the smallest angstrom exponents (largest particles), while Clusters 2 and 4 have the largest angstrom exponents (smallest particles). Profiles in Cluster 3 were measured when Canadian forest fires impacted the region, bringing in large particles aloft (above 2000 m). 0 500 1000 1500 2000 2500 0.5 1 1.5 2 2.5 3 Median Angstrom exponent (at 550 nm) A l ti tu d e (m ) Cluster 1 (40) Cluster 2 (112) Cluster 3 (6) Cluster 4 (14) 81 Figure 13. Back trajectory density plots for Angstrom exponent Clusters 1-4. The top 0.3% SO 2 sources are shown with a circle. Back trajectories associated with Clusters 1 and 2 show density over SO 2 sources, however back trajectories associated with Cluster 1 has the weakest winds allowing for more particle growth. 82 Figure 14. Statistics for SO 2 emissions encountered by back trajectories for each Angstrom exponent cluster. The SO 2 emissions are sums of all emissions (g day -1 ) encountered by a back trajectory divided by the area of the circle drawn around the back trajectory point (m 2 ). Cluster 1 is associated with the largest SO 2 emissions, while Clusters 2 and 4 emissions are much smaller. This explains the larger particles seen in Cluster 1 profiles (Figure 12). 4.2.5 CO The clustering methodology produced three CO clusters (Figure 15). Cluster 3 had only one profile which was measured during the Canadian forest fire episode of 2002 and shows the signature peak in CO values above 2000 m (Taubman et al., 2004a). Cluster 1, with 87% of the profiles, represents the background CO measured in the summer months in the Mid-Atlantic region. Cluster 2 values are about twice as large as Cluster 1 values. The back trajectories for Cluster 2 are short and concentrated around the I-95 corridor (a source for CO); while the back trajectories for Cluster 1 are more diffuse (Figure 16). This may explain the difference in CO 0 2 4 6 8 10 Cluster 1 (43) Cluster 2 (112) Cluster 4 (14) Cluster I n t e g r at ed S O 2 em i ssio n s ( g m -2 da y -1 ) Q25 Q5 median Q95 Q75 83 values between Clusters 1 and 2. Many of the profiles in Cluster 2 were made near Philadelphia and Baltimore, where the urban environments likely added to CO values. Most of the other flight locations for Cluster 2 were downwind of the I-95 corridor between Virginia and Pennsylvania. Cluster 2 not only has larger peaks near the surface but also larger values aloft. This suggests that the Eastern US is a major source for regional CO. Figure 15. Median CO profiles for each cluster. Error bars represent the 25 th and 75 th percentiles. The number of profiles in each cluster is shown in parentheses in the key. Cluster 3 profiles have double the CO values in Cluster 1. Profiles in Cluster 4 were made when Canadian forest fires impacted the region. 0 500 1000 1500 2000 2500 0 100 200 300 400 500 600 700 CO median (ppb) A l ti tu d e (m ) Cluster 1 (93) Cluster 2 (12) cluster 3 (1) 84 Figure 16. Back trajectory density plots for CO Cluster1 (clean profile) and Cluster 2 (polluted profile). Back trajectories for Cluster 2 are short and concentrated around the I-95 corridor (a source for CO); while the back trajectories for Cluster 1 are more diffuse. Cluster 1 Cluster 2 0.0% 1.0% 2.0% 3.0% 4.0% 5.0% Percent density 85 4.2.6 Particle Absorption The clustering methodology produced three absorption clusters and the associated median profiles are shown in Figure17 along with the median of all flights made between 2000 and 2003 (June through August). The median profile for Cluster 2 represents 77% of the profiles and is similar to the median of all flights made. The median profile for Cluster 1 has on average twice the absorption values of Cluster 2, and is greater than the 75 th percentile of the median of all profiles. The back trajectory densities for Clusters 1 and 2 both show northwesterly winds, however, Cluster 1 winds are slightly faster (Figure 18). These faster back trajectories associated with Cluster 1 may bring in air from the industrialized Midwest to mix with local, mobile emissions. Cluster 3 contains only profiles measured during the Canadian forest fires and shows the characteristic peak above 2000 m (Taubman et al., 2004a). 86 Figure 17. Median absorption profiles for each cluster. Error bars represent the 25th and 75th percentiles. The number of profiles in each cluster is shown in parentheses in the key. Absorption values in Cluster 1 are double those in cluster 2 below 1200 m. Profiles in Cluster 3 were made when Canadian forest fires impacted the region. 0 500 1000 1500 2000 2500 0 5 10 15 20 25 30 35 Median particle absorption (Mm-1) A l ti tu d e (m) Cluster 1 (28) Cluster 2 (103) Cluster 3 (2) median (all flights) 87 Figure 18. Back trajectory density plots for absorption Clusters 1 and 2. The back trajectory densities for clusters both show northwesterly winds, however, Cluster 1 winds are slightly faster. These faster back trajectories associated with Cluster 1 may bring in air from the industrialized Midwest to mix with local, mobile emissions. Cluster 1 Cluster 2 0.0% 1.0% 2.0% 3.0% 4.0% 5.0% Percent density 88 4.3 Discussion In order to better understand the chemistry associated with each O 3 cluster, the median profiles for SO 2 , particle scattering and CO measured simultaneously with the O 3 profiles from each of the clusters (herein referred to as matching species profiles) were examined (Figure 19). The clusters with the least O 3 (Clusters 1 and 2) are associated with the least SO 2 and scattering particles, while the clusters with more O 3 (Clusters 3 and 4) are associated with the most SO 2 and scattering particles (measurements of scattering and SO 2 were not made for Cluster 5). The SO 2 /CO ratio (Table 2) was also used to determine whether mobile or point source pollution was most influential on the O 3 values. The larger SO 2 /CO ratio for Clusters 3 and 4, with large O 3 values, suggests that these clusters were impacted most by point source emissions. CO profiles were similar for Clusters 1-4, but very large CO was measured in Cluster 5 (only one CO profile was measured for this small cluster), suggesting that localized pollution from mobile sources may affect these profiles. 89 Figure 19. Matching species profiles for O 3 clusters. The O 3 cluster number is in parentheses in the key. Profiles with the smallest O 3 values are associated with the smallest scattering and SO 2 values, while profiles with larger O 3 values are associated with more SO 2 and scattering. The profile with the largest O 3 is associated with the most CO, suggesting that this cluster was influenced by mobile sources. 0 500 1000 1500 2000 2500 0 50 100 150 200 Particle scattering (m -1 ) A l ti tu d e (m ) 02468 SO 2 (ppb) 0204060 CO (ppb) a Ve ry s ma ll O3 (2) Small O3 (1) Moderate O3 (3) Large O3 (4) Very large O3 (5) Matching species profiles for O 3 clusters 90 Table 2. SO 2 /CO ratios for O 3 Clusters 1-4. Clusters 3 and 4, with large O 3 values also have large SO 2 /CO ratios, suggesting that they are most heavily influenced by point source emissions. Median matching profiles of SO 2 , O 3 and ? were examined for the scattering clusters (Figure 20). Cluster 2, with the most scattering particles, was also associated with the most O 3 and SO 2 , as well as the largest particles (small ? values). This suggests that days with more aerosol pollution are often associated with conditions conducive to O 3 production and is consistent with the idea the NOx from elevated sources is more effective at producing O 3 than NOx from mobile sources. Cluster SO 2 /CO ratio n 1 0.010528 38 2 0.005548 18 3 0.014472 25 4 0.019471 10 91 Figure 20. Matching species profiles for particle scattering clusters. Profiles with the most particle scattering are associated with the most O 3 and SO 2 as well as the largest particles (smallest a values). Median matching species of CO were examined for the absorption clusters (Figure 21). Cluster 1 profiles, that are twice as absorbing as Cluster 2 profiles, have matching species profiles of CO that are on average 30% larger. This relationship between absorption and CO suggests that increased levels of absorbing species are likely the result of mobile emissions. 0 500 1000 1500 2000 2500 0 25507510 O 3 (ppb) P r e s s u r e de r i v e d a l t i t ude ( m ) 0246 SO 2 (ppb) 1.8 1.9 2 2.1 ? Matching species profiles for particle scattering clusters Least scattering (1) Most scattering (2) 92 Figure 21. Matching species profiles for particle absorption clusters. The profile with the most particle absorption is associated with the most CO. 4.4 Conclusions Clustering profiles of species allows for separation of distinct pollution events from a large collection of profiles, enabling a better understanding of how meteorology and chemistry affect the shape and size of the profiles. Profiles with the largest O 3 values were associated with larger integrated NO x emissions from point sources. The clustering methodology also separated profiles affected by Canadian forest fires. SO 2 profiles were less influenced by regional emissions than local emissions. The amount of SO 2 emitted into an air parcel over the previous 48 hours did not correlate well with observed SO 2 values. The product of SO 2 oxidation, as evidenced by particle light scattering, does correlate with SO 2 emissions integrated 93 over the previous 48 hours. This suggests the sulfate lifetime is longer than 48 hours. Particle size, calculated using scattering values, shows a similar relationship to emissions as scattering. Profiles with the largest CO values were made downwind of urban regions, and so these profiles appear to be characteristic of local/mobile sources. Profiles with highly absorbing particles are likely representative of urban scale pollution and are strongly influenced by mobile sources because they are associated with increased CO concentrations. 94 Chapter 5: Comparisons of University of Maryland Aircraft and Trace Gas Profiles with Models CMAQ and GOCART 5.1 Introduction 5.1.1 Background I compared O 3 measured aboard the University of Maryland Research aircraft in the summer of 2002 with the EPA Models-3 Community Multiscale Air Quality (CMAQ) modeling system. I examined differences among individual profiles as well as the statistical spread of all profiles. The standard CMAQ algorithm does not account for aerosols in the photochemistry of NO 2 . A revised version of CMAQ was run to account for aerosol effects (typically found in the Mid-Atlantic US) on NO 2 photochemistry. I compared O 3 from the standard and revised model run and results are presented below. Emissions of NO x are expected to be reduced by 2018 because of improved technology in motor vehicles and EPA imposed restrictions on power plant emissions. I have investigated how these reductions would be impacted by including aerosol effects in the photochemistry of NO 2 in CMAQ. I compared SO 2 from the aircraft to CMAQ and the Georgia Tech/Goddard Global Ozone Chemistry Aerosol Radiation and Transport (GOCART) model output; both models over-predict the SO 2 . This suggests the models assume a lifetime that is too long. A method for calculating the lifetime of SO 2 from in-situ measurements is described below as well as results from the calculation. 95 5.1.2 Description of Models CMAQ uses the PSU/NCAR Mesoscale Model (MM5) for meteorological modeling. The MM5 uses a non-hydrostatic model with sigma coordinates that follow the terrain (http://www.mmm.ucar.edu/mm5/). CMAQ uses the Sparse Matrix Operator Kernel Emissions (SMOKE) to represent natural and anthropogenic emissions. There are four processors that account for chemistry in the model; these include the Meteorology-Chemistry Interface Processor (MCIP), the Photolysis rate processor (JPROC), the Initial Conditions Processor (ICON) and the Boundary Conditions Processor (BCON). Transport of emissions is modeled with the CMAQ Chemical Transport Model (CCTM). CMAQ has 172 ? 172 grid cells and the size of each grid cell is 12 km ? 12 km. There are 16 vertical layers in the lower tropospheric boundary layer from the surface to 3400 m. The temporal resolution is 1 hour. The GOCART model uses assimilated meteorology from the Goddard Earth Observing System Data Assimilation System (GEOS-DAS; Schubert et al., 1993). This is an online model which allows for daily results to be compared with measurements. GOCART has a spatial resolution of 2 o latitude by 2.5 o longitude. The GEOS DAS meteorological data uses 30 vertical layers from the surface to 80 km and 7 layers between the surface and 1.8 km. GOCART has a temporal resolution of 6 hours and calculates three dimensional SO 2 , dimethylsulfide, sulfate and methanesulfonic acid. Anthropogenic emissions used in GOCART emissions are from the Emission Database for Global Atmospheric Research (EDGAR) and include DMS from the ocean, SO 2 and sulfate from anthropogenic sources, SO 2 from biomass 96 burning, aircraft, and volcanoes. The anthropogenic emissions include industrial processes (81%), residential and commercial fuel consumption (12%), and transportation (road, rail, shipping, 7%), with an annual rate of 72.8 Tg S yr -1 . Anthropogenic emission rates over the United States are assumed constant for the year for the US (Chin et al., 2000a). Chemical reactions for SO 2 in the GOCART model include oxidation by OH in air and H 2 O 2 in cloud to form sulfate. It is assumed that H 2 O 2 is regenerated to prescribed values every 3 hours. Dry deposition is represented as a function of aerodynamic resistance, sublayer resistance, and surface resistance. Dry deposition velocities of SO 2 over land are usually 0.2-0.4 cm s -1 . In-cloud and below-cloud precipitation are also accounted for (Chin et al., 2000a). I compared CMAQ trace gases from the lowest 16 layers (around 10, 24, 68, 116, 185, 282, 398, 544, 727, 949, 1212, 1523, 1886, 2312, 2820, 3393 m above ground level) of the model to aircraft measurements made in 2002. The CMAQ layers were converted to meters above sea level, by adding the surface elevation of each grid point. The model output was extracted at the location and time closest to the aircraft measurements. SO 2 was interpolated in altitude and time to match aircraft measurements. The same extraction process was performed for CMAQ O 3 . The resolution of GOCART SO 2 is 2 o latitude by 2.5 o longitude, with the first seven altitude layers around 118, 223, 377, 590, 880, 1265, 1768 m above ground level for the Mid-Atlantic US for June-August. GOCART has a 6 hour temporal resolution at 6, 12, 18, and 24 UT. Aircraft SO 2 profiles measured within the latitude longitude box and within the 6 hour time period were compared with GOCART SO 2 . 97 Generally, only one aircraft profile was compared with GOCART model output (70% of the time) although in some cases, up to five aircraft profiles were averaged. 5.2 Comparisons Between Models and Measurements 5.2.1 O 3 Comparisons Figure 1 shows the median aircraft measured and CMAQ O 3 profiles with the 25 th and 75 th percentiles for 136 profiles made in 2002. CMAQ is 10% (~6 ppb) smaller than aircraft O 3 between 600 and 2600 m. The ratio of the CMAQ/Aircraft O 3 mixing ratio is shown in Figure 2 and this shows that CMAQ under-predicts O 3 above 600 m and over-predicts O 3 below 600m. The CMAQ O 3 column content is 3% smaller than the aircraft column content (Table 1). 98 Figure 1. Median CMAQ and aircraft O 3 profiles from 2002 (June ?August). The median was obtained from 136 profiles. The error bars represent the 25 th and 75 th percentiles. Though the error bars overlap, CMAQ under-predicts O 3 by 10% between 600 and 2600m. Median CMAQ and aircraft O 3 136 profiles summer 2002 0 500 1000 1500 2000 2500 3000 3500 20 30 40 50 60 70 80 90 100 O 3 (ppb) A l titud e (m a b ove s ea leve l) Aircraft CMAQ 99 Figure 2. The ratio of CMAQ/Aircraft O 3 as a function of altitude. Below 600 m CMAQ over-predicts O 3 , above 600 m CMAQ under-predicts O 3 . Column content (g m -2 )Aircraft CMAQ average 0.29 0.28 Table 1. CMAQ and aircraft O 3 column contents calculated from near the surface (~3 m above ground level) to the top of the aircraft spiral (~ 2500 m). The CMAQ column content is 3% smaller than the aircraft column content. I also compared individual profiles to better understand the relationship between aircraft and CMAQ profiles. I have developed a method to look at these profiles in an objective manner by sorting the differences (between profiles) into three categories, the smallest, median and largest differences. This method is described below. 0 500 1000 1500 2000 2500 3000 3500 0.6 0.7 0.8 0.9 1 1.1 1.2 CMAQ/Aircraft O 3 ratio A l ti tu d e (m A G L ) 100 The differences between aircraft and model profiles were calculated accounting for shape (location of minima and maxima in the profiles) as well as size (absolute differences in mixing ratio). All aircraft spirals and matching CMAQ modeled outputs were initially averaged into 100 m altitude levels. This allowed for consistent comparisons between each pair of modeled and measured O 3 profiles. The difference between modeled and measured O 3 at each altitude bin accounted for the size. The profiles were examined at four altitude bins (250 ? 650 m, 651-1150 m, 1151-1650 m, and 1651-2150 m). In each altitude bin there were at least five altitude layers examined. A linear regression was made between the O 3 mixing ratios of the two profiles being compared at these altitude layers. The slope and correlation coefficient were used to account for the shapes of the profiles being compared. The difference equation from Chapter 4 (Equation 2) was used. The differences were sorted and profiles associated with three of the 5 th percentile (smallest) differences, three of the median differences and three of the 95 th percentile (largest) differences were examined. Figure 3 shows the three modeled and measured profiles with the 5 th percentile smallest differences. I examined profiles in the grid cell closest to where the airplane flew (the center cell) as well as the profiles in the 8 grid cells surrounding this center cell. Figure 3 shows the CMAQ profiles associated with the center cell in dark blue as well as the smallest and largest profiles from the surrounding grid cells. There are some jumps in the CMAQ profiles as shown for the Louisa, VA profile at 1200 m. The aircraft profiles have a diameter of about 1 km, and sometimes they crossed two different CMAQ grid cells. CMAQ O 3 for the closest grid cell was used for the difference calculation and sometimes 101 more than one grid cell was used in a profile. Figure 4 shows three of the modeled and measured profiles with median differences. The differences between profiles were calculated between 250 and 2150 m and the portions of the profile not included in this calculation are shown in grey. On July 8, 2002 Canadian forest fires impacted the region, and the aircraft profile over Easton, MD shows the signature peak of O 3 above 2200 m. Below the forest fire peak the aircraft profile compared reasonably well with CMAQ O 3 . This example presents a limitation of the difference method. Figure 5 shows three of the modeled and measured profiles with the 95 th percentile largest differences. The comparison of CMAQ and aircraft O 3 over Winchester, VA shows that the model does not always under-predict O 3 aloft. The differences between profiles seem to be independent of altitude and the shape of the profiles. Also shown in Figures 3-5 are the 24 hour back trajectories ending at the location of the aircraft spiral (at 1000, 2000, and 3000 m) for each of the CMAQ-aircraft comparisons. The differences between profiles also seem to be independent of wind speed and direction (from back trajectories). 102 Figure 3. CMAQ and aircraft O 3 profiles associated with the 5 th percentile smallest differences (the best agreement). The dark blue profiles represent CMAQ O 3 from the closest (center) grid cell. The light blue profiles represent the smallest and largest O 3 from the surrounding 8 grid cells. There are some jumps in the Louisa, VA CMAQ profile. The aircraft profiles have a diameter of about 1 km, and sometimes they crossed two different CMAQ grid cells. CMAQ O 3 for the closest grid cell was used in the difference calculation and sometimes more than one grid cell was used in a profile. Back trajectories at three altitudes are also shown. Back trajectories ending at 1km 2km 3km CMAQ (nearest box) CMAQ (range of 9 boxes) Aircraft 103 Figure 4. CMAQ and aircraft O 3 profiles associated with the median differences. The dark blue profiles represent CMAQ O 3 from the closest (center) grid cell. The light blue profiles represent the smallest and largest O 3 from the surrounding 8 grid cells. Canadian forest fires impacted the region on July 8, and this is seen in the aircraft O 3 profile over Easton. The differences were calculated between 250 and 2150 m and so the influence of the fires was not accounted for in the equation. This shows a limitation of the method. Back trajectories ending at 1km 2km 3km CMAQ (nearest box) CMAQ (range of 9 boxes) Aircraft 104 Figure 5. CMAQ and aircraft O 3 profiles associated with the 95 th percentile largest differences (the worst agreement). The dark blue profiles represent CMAQ O 3 from the closest (center) grid cell. The light blue profiles represent the smallest and largest O 3 from the surrounding 8 grid cells. The comparison at Winchester shows that CMAQ sometimes over-predicts O 3 aloft. CMAQ (nearest box) CMAQ (range of 9 boxes) Aircraft Back trajectories ending at 1km 2km 3km 105 5.2.2 The Effects of Aerosols on the Photolysis Rate of NO 2 and the Production of O 3 In general CMAQ under-predicts O 3 aloft. Reasons for this under-prediction include problems with emissions inventories as well as problems with meteorology, cloud cover, and CMAQ?s ability to describe transport. The NO 2 photolysis rates that CMAQ uses also impact how much O 3 is produced by the model. The reaction rate coefficient for the photolysis of NO 2 (hereafter referred to as j-NO 2 value) used by the standard version of CMAQ assumes no aerosol loading. Dickerson et al. (1997) show that an increase of aerosols from an optical depth of 0 to 2 increases the j-NO 2 values by 30% above the boundary layer (1000 m). Park (2001) performed a sensitivity test of CMAQ using j-NO 2 values associated with CMAQ aerosols. He used the aerosol properties generated by CMAQ to develop a program to modify the j-NO 2 values accordingly, and then compared the O 3 generated with the modified CMAQ run to surface measurements. He found that the effects on O 3 production were variable, and there were still numerous disagreements between modeled and measured O 3 . Aerosols generated with CMAQ are often under-predicted (Mebust et al., 2003; Mueller et al., 2006; Tesche et al., 2006) and this may partly explain the mixed results Park (2001) found. I performed a sensitivity study using j-NO 2 values associated with typical aerosols measured in the Mid-Atlantic from July 15-18, 2002 using the Park (2001) program that allows for adjustment of the Angstrom slope and intercept defined in the Angstrom equation: ? = ? ? -? (1) 106 Here, ? is the aerosol optical depth, ? is the Angstrom slope (Angstrom exponent) that represents the size of aerosols, ? is the wavelength in ?m, and ? is the intercept related to the amount of aerosols present in the atmosphere. The Angstrom coefficient intercept can be assigned a value of 0, 0.1, 0.2, 0.3, 0.4, or 0.5. The Angstrom coefficient slope can be assigned a value between 0.5, 1.0, 1.5, 2.0, or 2.5. Figure 6 shows the median of Angstrom exponent measurements made during July 15-18, 2002 for 20 aircraft profiles. Of the values allowed in the Park (2001) model, the median Angstrom exponent (Angstrom slope) is closest to 2.0. The Park (2001) model also allows for the adjustment of single scattering albedo, asymmetry parameter and aerosol layer depth. Angstrom exponents 0 500 1000 1500 2000 2500 3000 3500 1.5 1.7 1.9 2.1 2.3 2.5 Angstrom exponents at 550, 575 and 625 nm A l tit u d e (m ) A625 A575 A500 Figure 6. The median Angstrom exponent calculated with the ratio of scattering at 450 and 550 nm (A500), 450 and 700 nm (A575), and 550 and 700 nm (A625) for flights made between July 15-18, 2002. The error bars represent the 25 th and 75 th percentile. The Angstrom exponent measured aboard the aircraft is closest to the 2.0 input for the Park (2001) model. 107 Aerosol optical depth 550 nm 380 nm average 0.31 0.68 median 0.28 0.63 minimum 0.12 0.26 maximum 0.66 1.41 Table 2. Statistics for aerosol optical depth calculated at 550 nm and 380 nm for 17 flights made between July 15-18, 2002. Table 2 shows that the average optical depth at 550 nm is 0.31. Because the photolysis of NO 2 occurs at wavelengths of 380 nm and not at wavelengths of 550 nm (where the aircraft made measurements), I interpolated the aerosol properties to the 380 nm wavelength. The absorption coefficient (abs, with units of m -1 ) can be approximated at different wavelengths, ?, using the relationship (from Bodhaine, 1995): abs = c a /? (2) Here c a is a constant. From Equation 2, abs 380 can be solved using: abs 380 = abs 550 x 550/380 (3) I also converted the scattering coefficient (with units of m -1 ) at 550 nm to the scattering coefficient at 380 nm, scat380, using the relationship (from Bodhaine, 1995): Scat 550 = c s /550 A550 (4) Scat 380 = c s /380 A380 (5) Here c s is a constant. From Equation 5, scat 380 can be solved using: Scat 380 = Scat 550 *550 A550 /380 A380 (6) 108 Here A 550 represents the Angstrom exponent calculated from the ratio of scattering measurements at two different wavelengths (as shown in Equation 2 in Chapter 3) where the average wavelength is 550 nm. A 380 is just the Angstrom exponent calculated from the ratio of scattering at two different wavelengths where the average wavelength is 380 nm. Figure 6 shows the Angstrom exponent calculated from the ratio of scattering at 450 and 550 nm (A500), 450 and 700 nm (A575), and 550 and 700 nm (A600). There is little variability among the Angstrom exponents calculated from different wavelength ratios and so I assumed that A 550 and A 380 (from Equation 6) are equal. I then calculated aerosol optical depth at 380 nm (Table 2). Using the optical depth at 380 nm or 550 nm in Equation 1, results in an Angstrom intercept closest to 0.1. The Park (2001) model allows values of single scattering albedo of 0.92, 0.96 and 1.0. Figure 7 shows the median profile for single scattering albedo at 550 nm measured during July 15-18, 2002 and composed of 17 profiles. For the values allowed in the Park (2001) model the median single scattering albedo is closest to 0.96. 109 Figure 7. The median single scattering albedo at 550 nm for flights made between July 15-18. The error bars represent the 25 th and 75 th percentile. The single scattering albedo measured aboard the aircraft is closest to the 0.96 input for the Park (2001) model The model allows values of asymmetry parameter to be 0.6, 0.7, or 0.8. The asymmetry parameter is calculated from the backscatter to total scattering ratio using the equation g =-2x +1 (7) Where g is the asymmetry parameter and x is the backscatter to total scattering ratio. The aircraft did not make measurements of backscatter to total scattering in 2002, but measurements were made in 2003-2005. Figure 8 shows the asymmetry parameter for 139 flights measured in 2003-2005. For the values allowed in the Park (2001) model the median profile of asymmetry parameter is closest to 0.8. The Park (2001) model allows for aerosol layer depths to be 0.5, 1.5, or 2.5 km. Figures 1d and 1f in Chapter 3 show the median scattering and absorption profiles for all flights made in the Mid-Atlantic region. The depth of the aerosol layer for these profiles is closest to Single scattering albedo for July 15-18, 2002 (17 profiles) 0 500 1000 1500 2000 2500 3000 3500 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 Single scattering albedo at 550 nm A l ti tu d e (m ) 110 1.5. These aerosol values were used in the Park (2001) model and the resultant j-NO 2 values are shown in Figure 9. Above 1000 m the j-NO 2 values calculated with aerosols were 25% larger than those with no aerosols. Figure 8. The median asymmetry parameter at 550 nm for 139 flights made between 2003 and 2005. The error bars represent the 25 th and 75 th percentile. The asymmetry parameter measured aboard the aircraft is closest to the 0.8 input for the revised CMAQ run. Assymetry parameter for 139 flights from 2003-2005 0 500 1000 1500 2000 2500 3000 3500 0.7 0.710.720.730.740.750.760.770.780.79 0.8 Assyme try pa ra me te r A l t i t ude ( m ) 111 Figure 9. Standard and revised j-NO 2 values used in CMAQ (at 0, 1, 2, 3, 4.05, 5 and 10 km). The standard j-NO 2 values assume there are no aerosols (without) and the revised j-NO 2 values (with) were calculated using aerosol properties presented in Figures 6,7 and 8 and Table 2. The different symbols represent the j-NO 2 values at different times of the day. I ran CMAQ from July 15-18, 2002 with the standard j-NO 2 values (assuming no aerosol) and revised j-NO 2 values (assuming aerosol typical for the episode). The same aerosol values were used throughout the domain. The aircraft flies downwind of urban and suburban areas with large optical depths, and also rural areas with small optical depths. Using the average optical depth from all of the flights should be a reasonable approximation of the Mid-Atlantic average optical depth. Levy (2007) found a correlation coefficient (r 2 ) of 0.26 between MODIS satellite retrievals and aircraft calculated aerosol optical depth. Two emissions scenarios were used; one with 2002 emissions and one with 2018 emissions that are substantially lower than those from 2002. This resulted in four model runs for comparison: j-NO 2 values with and without aerosols 0 2000 4000 6000 8000 10000 12000 0.000 0.100 0.200 0.300 0.400 0.500 0.600 0.700 0.800 j-NO 2 value (min -1 ) A l t i tu d e (m ) 12 noon without 1pm without 2 pm without 3 pm without 4 pm without 12 noon with 1 pm with 2 pm with 3 pm with 4 pm with 112 ? 2002 emissions with standard j-NO 2 values (2002, standard) ? 2002 emissions with revised j-NO 2 values (2002, revised) ? 2018 emissions with standard j-NO 2 values (2018, standard) ? 2018 emissions with revised j-NO 2 values (2018, revised). CMAQ O 3 values were generated during a previous run by the New York State Department of Environmental Quality (NYDEQ) using the standard j-NO 2 values. The NYDEQ runs used TOMS data to determine the stratospheric O 3 influence on radiative forcing and on boundary layer O 3 production. I did not have access to the TOMS data, so I performed the four runs using CMAQ default overhead O 3 (generated from Nicolet et al., 1982). The O 3 I generated from the CMAQ run using 2002 emissions and j-values with no aerosols correlated well with the NYDEQ runs using the same emissions and j-values (but different overhead O 3 ). However, the O 3 generated from my CMAQ run was up to 5 ppb smaller than that generated from the NYDEQ run. In order to make meaningful comparisons between aircraft O 3 and O 3 generated with the revised CMAQ run (2002, revised), I adjusted the CMAQ O 3 using the following: O 3 (2002, revised) = O 3 (NYDEQ) * O 3 (2002, revised without TOMS O3) (8) O 3 (2002, standard with TOMS O3) Figures 10a-e show O 3 generated by CMAQ using the standard j-NO 2 values, O 3 generated using the revised j-NO 2 values (adjusted using Equation 8) and O 3 from the aircraft. The revised CMAQ run generated more O 3 (~1 ppb) above 500 m than the standard run. The revised CMAQ run generated less O 3 (1-4 ppb) below 500 m than the standard run. The revised run did not eliminate measurement/model differences, but brought the CMAQ output closer to observations. Figures 10a-e are 113 limited in space because they only represent a few grid cells. Figure 11 shows the median differences in O 3 between revised and standard runs (revised CMAQ? standard CMAQ) for the 16 profiles. Table 3 compares the average O 3 column contents among the aircraft, standard CMAQ runs, and revised CMAQ runs for the July 15-18, 2002 episode. CMAQ O 3 column content from the standard and the revised runs were ~7% smaller than the aircraft O 3 . The O 3 column content from the revised run was 0.3% larger than the O 3 column content from the standard run. 114 Figure 10a. O 3 profiles from the aircraft (pink), standard CMAQ runs (shown in blue), and revised CMAQ runs (shown in green) for July 15, 2002. Above 500 m the revised CMAQ profiles are about 1 ppb larger than the standard CMAQ profiles shown in blue. Below 500 m the revised CMAQ O 3 is smaller than the standard CMAQ O 3 . Aircraft Standard CMAQ revised CMAQ 115 Figure 10b. O 3 profiles from the aircraft (pink), standard CMAQ runs (shown in blue), and revised CMAQ runs (shown in green) for the morning of July 16, 2002. Above 500 m the revised CMAQ profiles are about 1 ppb larger than the standard CMAQ profiles shown in blue. Below 500 m the revised CMAQ O 3 is smaller than the standard CMAQ O 3 . Aircraft Standard CMAQ revised CMAQ 116 Figure 10c. O 3 profiles from the aircraft (pink), standard CMAQ runs (shown in blue), and revised CMAQ runs (shown in green) for the afternoon of July 16, 2002. Above 500 m the revised CMAQ profiles are about 1 ppb larger than the standard CMAQ profiles shown in blue. Below 500 m the revised CMAQ O 3 is smaller than the standard CMAQ O 3 . Aircraft Standard CMAQ revised CMAQ 117 Figure 10d. O 3 profiles from the aircraft (pink), standard CMAQ runs (shown in blue), and revised CMAQ runs (shown in green) for July 17, 2002. Above 500 m the revised CMAQ profiles are about 1 ppb larger than the standard CMAQ profiles shown in blue. Below 500 m the revised CMAQ O 3 is smaller than the standard CMAQ O 3 . Aircraft Standard CMAQ revised CMAQ 118 Figure 10e. O 3 profiles from the aircraft (pink), standard CMAQ runs (shown in blue), and revised CMAQ runs (shown in green) for July 18, 2002. Above 500 m the revised CMAQ profiles are about 1 ppb larger than the standard CMAQ profiles shown in blue. Below 500 m the revised CMAQ O 3 is smaller than the standard CMAQ O 3 . Aircraft Standard CMAQ revised CMAQ 119 Average O 3 column content (g m -2 ) Aircraft 0.3093 CMAQ 2002 (standard run) 0.2885 CMAQ 2002 (revised run) 0.2888 CMAQ 2018 (standard run) 0.2561 CMAQ 2018 (revised run) 0.2555 Table 3. Median O 3 column contents for the July 15-18, 2002 episode (for locations sampled by the aircraft) from the aircraft, standard CMAQ 2002 run, revised CMAQ 2002 run, standard CMAQ 2018 run, and revised CMAQ 2018 run. 0 500 1000 1500 2000 2500 -5 -4 -3 -2 -1 0 1 2 3 2002 ozone differences (revised - standard) A l ti tu d e ( m ) Figure 11. Median CMAQ O 3 differences (standard ? revised) for 16 profiles generated between July 15-18, 2002. Error bars represent the 25 th and 75 th percentiles. Near the surface the revised CMAQ run generates less O 3 than the standard CMAQ run. Above 500 m the revised CMAQ run generates more O 3 than the standard run. 120 Figure 12 shows the difference between revised CMAQ and standard CMAQ runs for three different levels (1, 8, and 16 that are approximately at the surface, 500, and 2000m) for the Eastern US at 14 UT and 18 UT. The largest O 3 production generally occurs within these times. Here negative values, when the revised CMAQ run generates less O 3 than the standard CMAQ run, are shown with cooler colors. These differences, of up to 10 ppb, are seen mainly at the surface. Positive values, when the revised CMAQ run generates more O 3 than the standard run, are shown with warm colors. These differences, of up to 1 ppb, generally occur above 500 m. 121 Figure 12a. Differences between revised and standard CMAQ O 3 (revised- standard) for a July 2002 smog and haze episode. These plots are for 14 UT (10 EST). The differences are negative at the surface meaning that the revised CMAQ run generates less O 3 than the standard run. Above 500 m the differences are positive and the revised CMAQ run produces more O 3 than the standard CMAQ. 2000 O3 differences (revised CMAQ ? standard CMAQ) Layer 16 (~3400 m) 122 Figure 12b. Differences between revised and standard CMAQ O 3 (revised- standard) for a July 2002 smog and haze episode. These plots are for 18 UT (14 EST). The differences are negative at the surface meaning that the revised CMAQ run generates less O 3 than the standard run. Above 500 m the differences are positive and the revised CMAQ run produces more O 3 than the standard CMAQ. 2000 O3 differences (revised CMAQ ? standard CMAQ) Layer 16 (~3400 m) 123 A curtain plot (Figure 13) was used to examine the diurnal variation in the first 16 layers of CMAQ showing the differences in O 3 generated from: Revised CMAQ - standard CMAQ (9) The x-axis represents a swath made one grid cell wide (East and West) extending from the southernmost grid cell to the northernmost grid cell in the Eastern US shown in Figure 14. The y-axis represents the first 16 layers of CMAQ. Six time periods of 3, 7, 11, 15, 19 and 23 UT are shown for July 17, 2002. In the early morning (from 3 to 11 UT) there are positive differences (where the revised CMAQ generated O 3 is larger than the standard CMAQ O 3 ) above 500 m that are mixed down to the surface. At 15 and 23 UT, right after rush hour, there are negative differences (where the revised CMAQ O 3 is smaller than the standard CMAQ O 3 ) near the surface. 124 Figure 13. O 3 differences (revised-standard) for a single swath in the CMAQ grid. The y-axis represents the first16 altitude layers used in CMAQ. The x-axis represents a swath of the grid cells examined (Figure 14), where 1 is the Southernmost grid cell and 172 is the Northernmost grid cell. Here negative differences mean that O 3 generated with the revised CMAQ is smaller than the standard CMAQ O 3 and these are seen closer to the surface. 125 Figure 14. The curtain plot in Figure 13 was made from a vertical swath, shown in red. This swath represents the x-axis in Figure 13. I also examined how the emissions reductions scenario, expected for 2018, would be impacted by incorporating aerosols into CMAQ. Figures 15 a and b show the CMAQ O 3 reductions (CMAQ 2018 ? CMAQ 2002) for profiles made during the July15-18 episode (at locations where the aircraft made spirals) for the standard CMAQ runs (Figure 15 a) and the revised CMAQ runs (Figure 15 b). The revised CMAQ reductions and standard CMAQ reductions are similar, with the largest 126 reductions (10 ppb) near the surface and smaller reductions (7 ppb) at 2000 m. Figure 15 c shows the difference between: [(2018 revised ? 2002 revised) ? (2018 standard ? 2002 standard)] (10) The differences between O 3 reductions using revised CMAQ and standard CMAQ (Equation 10) are small (Figure 15 c) for the locations where the UMD research aircraft made spirals. However, the revised reductions are smaller than the standard reductions, and this means the standard CMAQ run slightly overestimates reductions at the surface (by 0.6ppb). Above 1000 m the standard CMAQ run underestimates reductions. The column contents in Table 3 suggest that the reductions using the revised CMAQ runs are 3% larger than the reductions using the standard CMAQ runs. 127 a 0 500 1000 1500 2000 2500 -14 -12 -10 -8 -6 -4 -2 0 2018-2002 ozone reductions (standard j-values) A l ti tu d e ( m ) b 0 500 1000 1500 2000 2500 -14 -12 -10 -8 -6 -4 -2 0 2018-2002 ozone reductions (revised j-values) A l ti tu d e ( m ) Figure 15 a b. Median CMAQ ozone reductions (CMAQ 2018 ? CMAQ 2002) using a) standard j-values and b) revised j-values. Error bars represent the 25 th and 75 th percentiles. The largest reductions occur near the surface. 128 c 0 500 1000 1500 2000 2500 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 O 3 changes (ppb) (2018 revised - 2002 revised) - (2018 standard - 2002 standard) Al ti tu d e ( m ) Figure 15 c. Changes in O 3 reductions of revised CMAQ? standard CMAQ. Near the surface standard CMAQ overestimates the reductions and above 500 m the standard CMAQ underestimates the reductions. Figure 16 shows the results of Equation 10 (differences in O 3 reductions for revised and standard CMAQ runs) for the Eastern US. These differences are plotted at three levels (approximately the surface, 500 m and 3400 m) at 14 UT and 18 UT. The positive changes show that the standard model over-predicts O 3 reductions (because the revised CMAQ reductions are smaller than the standard CMAQ reductions) by up to 2 ppb near the surface. Above 500 m the standard model under- predicts O 3 reductions by up to 2 ppb. 129 Figure 16a. O 3 differences of [(2018, revised ? 2002, revised) ? (2018, standard ? 2002, standard)] for the 1st, 8th, and 16th layers at 14 UT. The standard CMAQ runs over-predict O 3 reductions near the surface (warm colors). Above 500 m the standard model under-predicts O 3 reductions. 2000 O3 differences (2018 revised ? 2002 revised) - (2018 standard ? 2002 standard) Layer 16 (~3400 m) 130 Figure 16b. O 3 differences of [(2018, revised ? 2002, revised) ? (2018, standard ? 2002, standard)]. For the 1st, 8th, and 16th layers at 18 UT. The standard CMAQ runs over-predict O 3 reductions near the surface (warm colors). Above 500 m the standard model under-predicts O 3 reductions. This study has important policy implications. The National Ambient Air Quality Standards (NAAQS) regulate surface O 3 and the EPA requires states to use the CMAQ model to determine future compliance, for surface sites. Above I have shown that accounting for aerosols in the photolysis rates of NO 2 decreases O 3 production near the surface. State agencies that are not in compliance with NAAQS O 3 standards can use this model bias to their advantage when developing the State 2000 O3 differences (2018 revised ? 2002 revised) - (2018 standard ? 2002 standard) Layer 16 (~3400 m) 131 Implementation Plans. The reductions in O 3 (incurred by reductions point and mobile NOx emissions) generated with the standard CMAQ model are overestimated at the surface. This must be accounted for when state agencies develop plans to reduce O 3 . 5.2.3 SO 2 Comparisons Modeled SO 2 from CMAQ and GOCART was compared to aircraft profiles. Figure 17 shows the median of 118 CMAQ and aircraft SO 2 profiles (ppb) for 2002 (June ? August), with error bars representing the 25 th and 75 th percentiles. These profiles were made in the area contained by 37.18 to 44.53 o latitude and -79.44 to - 68.36 o longitude. CMAQ over-predicts SO 2 by a factor of 1.2 at 200 m and by a factor of 4.6 at 2300 m (Figure 18). CMAQ over-predicts the column content by 55% (Table 4). 132 Figure 17. Median CMAQ and aircraft profiles of SO 2 from 2002 (June ?August). The median was obtained from 118 profiles. The error bars represent the 25 th and 75 th percentiles. Though the error bars overlap, CMAQ over-predicts SO 2 throughout the profile. Aircraft and CMAQ SO 2 profiles 2002 (118 profiles) 0 500 1000 1500 2000 2500 3000 3500 012345678 Median SO 2 (ppb) Al t i t u de ( m ab o v e s ea l eve l ) CMAQ Aircraft 133 Figure 18. The ratio of median CMAQ SO 2 mixing ratios divided by median aircraft SO 2 mixing ratios. CMAQ over-predicts SO 2 by a factor of 1.2 at 200 m and a factor of 4.6 at 2300 m. Table 4. The average aircraft and CMAQ SO 2 column content (g m -2 ) for 118 profiles. The average CMAQ column content is 55% larger than the average aircraft column content. Figure 19 shows median SO 2 profiles (?g/m 3 ) from 223 GOCART and aircraft averaged profiles for 2000 to 2003 (April ? August) with error bars representing the 25 th and 75 th percentiles. These profiles were made in the area contained by 34 to 44 o Column content (g m -2 )Aircraft CMAQ average 0.009 0.014 CMAQ/aircraft SO 2 ratio 0 500 1000 1500 2000 2500 3000 3500 012345 CMAQ/aircraft SO 2 ratio A l t i t u d e ( m ab o ve sea l evel) 134 latitude and -82.5 to -67.5 o longitude. Figure 20 shows the ratio of GOCART/aircraft SO 2 at each of the seven altitudes examined. GOCART over-predicts SO 2 by a factor of 1.4 at 100 m and by a factor of 2 at 1250 m. The GOCART column content is 50% larger than the aircraft column content (Table 5). Although CMAQ and GOCART are representative of different times and locations and are not strictly comparable with each other, they show a consistent high bias relative to observations. Figure 19. Median GOCART and aircraft profiles of SO 2 from 2000-2002 (April ?August). The median was obtained from 223 profiles. The error bars represent the 25 th and 75 th percentiles. Though the error bars overlap, GOCART over- predicts SO 2 up to 1800 m. GOCART and aircraft median SO 2 2000 - 2003 (223 flights) 0 200 400 600 800 1000 1200 1400 1600 1800 2000 02468101214161820224 SO 2 (?g/m 3 ) A l t i t u d e ( m ab o ve g r o u n d l evel) aircraft GOCART 135 Figure 20. The ratio of median GOCART SO 2 mixing ratios divided by median aircraft SO 2 mixing ratios. GOCART over-predicts SO 2 by a factor of 1.4 at 100 m and a factor of 2 at 1250 m. Table 5. The average aircraft and GOCART SO 2 column content (g m -2 ) for 223 profiles. The average GOCART column content is 50% larger than the average aircraft column content. Differences between CMAQ and measured SO 2 were calculated and sorted as those for O 3 . The profile differences are smallest for SO 2 with small mixing ratios (around 2 ppb at the surface). Figures 21-23 show CMAQ and aircraft profiles with the 5 th percentile smallest differences, median differences and 95 th percentile largest Column content (g m -2 ) Aircraft GOCART average 0.012 0.018 GOCART/aircraft SO 2 ratio 0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 0.5 1 1.5 2 2.5 GOCART/aircraft SO 2 ratio A l t i tude ( m a b o v e gr o und l evel) 136 differences. Back trajectories (from HYSPLIT) are also shown. Profiles associated with the median differences show that the model over-predicts SO 2 above 1500 m. The profile over Easton, MD on June 25, 2002 shows that the model does under- predict SO 2 sometimes. Profiles associated with the 95 th percentile largest differences show that the model over-predicts SO 2 by a factor of two to five throughout the profile. There does not appear to be a relationship among wind speeds and direction and SO 2 profiles. 137 Figure 21. CMAQ and aircraft SO 2 profiles associated with the 5 th percentile smallest differences (best agreement). The red profiles represent CMAQ SO 2 from the closest (center) grid cell. In general CMAQ over-predicts SO 2 . Back trajectories at 3 altitudes are also shown. Back trajectories ending at 1km 2km 3km CMAQ (nearest box) Aircraft 138 Figure 22. CMAQ and aircraft SO 2 profiles associated with the median differences. The red profiles represent CMAQ SO 2 from the closest (center) grid cell. Back trajectories ending at 1km 2km 3km CMAQ (nearest box) Aircraft 139 Figure 23. CMAQ and aircraft SO 2 profiles associated with the 95 th percentile largest differences (worst agreement). The red profiles represent CMAQ SO 2 from the closest (center) grid cell. Back trajectories ending at 1km 2km 3km CMAQ (nearest box) Aircraft 140 Differences between GOCART simulations and aircraft observations of SO 2 were calculated in a manner similar to the differences between CMAQ and aircraft. GOCART has coarse vertical resolution and there were only seven altitude layers to compare with the aircraft profiles. For this reason only one altitude bin with the seven layers (k=1 from Equation 1 in chapter 4) was analyzed for differences. Figures 24-26 show profiles associated with the 5 th percentile smallest differences, median differences, and 95 th percentile largest differences. Profiles associated with the median differences show that the model tends to over-predict SO 2 , however profiles associated with the 95 th percentile largest differences show the model sometimes over-predicts and sometimes under-predicts SO 2 . 141 Figure 24. GOCART and aircraft SO 2 profiles associated with the 5 th percentile smallest differences (best agreement). The red profiles represent GOCART SO 2 from the closest (center) grid cell. GOCART (nearest box) Aircraft 142 Figure 25. GOCART and aircraft SO 2 profiles associated with the median differences. The red profiles represent GOCART SO 2 from the closest (center) grid cell. GOCART gets the right shape, but the magnitude is too large. GOCART (nearest box) Aircraft 143 Figure 26. GOCART and aircraft SO 2 profiles associated with the 95 th percentile largest differences (worst agreement). The red profiles represent GOCART SO 2 from the closest (center) grid cell. GOCART gets the right shape, but the magnitude is too large. The model over-prediction could be a result from: 1. Emissions that are too large, with losses modeled correctly. 2. Correct emissions, but the loss in the model is too slow. To test possibility 1 (that the emissions are too large), I calculated the flux of SO 2 using national inventories of point and area SO 2 sources (area sources are composed of mobile emissions) as well as accounting for a small contribution from biogenic sources. I compared these SO 2 fluxes with those used in GOCART. For my SO 2 flux GOCART (nearest box) Aircraft 144 calculation I used the EPA?s AirData (http://epa.gov/air/data/geosel.html) database for the United States point and area SO 2 sources (emissions are in g/hr). I used the National Pollutant Release Inventory (NPRI) database for Canadian point source emissions. The NPRI database does not report area emissions and therefore I estimated them from the EPA?s AirData database. I calculated the slope between the US state populations ? x-axis, (ESRI data and maps 2000) and US state area SO 2 emissions- y-axis, by forcing the line through zero (Figure 27). This slope of area emissions/population was used to approximate area emissions from Canadian municipalities using Canadian populations (ESRI data and maps 2000). I also estimated the small contribution of SO 2 from biogenic sources using sulfur fluxes presented in Wayne (2000); the biogenic contribution of SO 2 is 0.7% of the contribution from anthropogenic emissions. I calculated the total flux of SO 2 for half of the United states and some Canadian municipalities by adding the point and area source emissions for each state (or municipality) to the biogenic contribution and dividing this by the total area of the state (or municipality). Figure 28 shows a comparison of the SO 2 fluxes I calculated from the national databases and the SO 2 flux used in the GOCART model. The average flux from the national inventories (Figure 28 a) is 2.8 x10 -4 g m -2 hr -1 and the average flux from GOCART (Figure 28 b) is 2.5 x10 -4 g m -2 hr -1 , only 16% smaller than the average flux from the national inventories. The SO 2 emissions used in CMAQ were generated with SMOKE which converts the resolution of the national inventories into a resolution that can be used in CMAQ. Because the SO 2 fluxes from the models (CMAQ and GOCART) are similar to those I calculated using the national inventories, it is unlikely that model emissions 145 are too large by a substantial amount. The likely explanation for why the model SO 2 is larger than measured is that the model removal of SO 2 is too slow. Figure 27. US state population and area sources of SO 2 . y = 0.0075x R 2 = 0.4194 0 40000 80000 120000 160000 200000 0.E+00 5.E+06 1.E+07 2.E+07 2.E+07 3.E+07 Population A r ea so u r c es ( t o n s / y e a r ) 146 Figure 28. a) The SO 2 flux (g hr -1 m -2 ) calculated using national emission inventories and b) the SO 2 (g hr -1 m -2 ) flux used by GOCART. The model may underestimate reactions that oxidize SO 2 to sulfate properly and this could explain a model loss of SO 2 that is too slow (possibility 2). SO 2 oxidation to sulfate via H 2 O 2 in clouds is well understood and in the Mid-Atlantic region with an acidic environment this is the most probable reaction pathway for a b 0.0014 ? 0.0016 < 0.0002 0.0002 ? 0.0004 0.0004 ? 0.0006 0.0006 ? 0.0008 0.0008 ? 0.0010 0.0010 ? 0.0012 0.0012 ? 0.0014 0.0016 ? 0.0018 > 0.0018 SO2 flux (g hr -1 m 2 ) 147 sulfate formation (Seinfeld and Pandis, 1998). Figure 29 shows the mixing ratio of H 2 O 2 over the average CMAQ domain and at a specific rural location (Big Meadows, VA) for July 1, 2002. H 2 O 2 is 2-4 times greater than SO 2 , from 1000 ? 2000 m, (Figure 17) and large enough to oxidize completely the SO 2 to sulfate. It is therefore likely that the model generates enough H 2 O 2 to react with SO 2 . Heterogeneous oxidation on mineral aerosols is not as well understood (Detener et al., 1996; Zhang et al., 2006), and may be more difficult to account for in the models. The models may also not fully account for wet and dry deposition in the Mid-Atlantic region, thus increasing the lifetime of SO 2 . The models may also under-represent clouds (as described by Mueller et. al., 2006), where SO 2 is oxidized to sulfate with H 2 O 2 and therefore produce an SO 2 lifetime that is too long. There are no NAAQS exceedences of SO 2 in the Mid-Atlantic region, but there are exceedences of PM 2.5 for which sulfate (with an SO 2 precursor) is a major component, accounting for 30% of PM 2.5 mass (Rees et al., 2004; Schwab et al., 2004; Frank et al., 2006; Ondov et al., 2006). 148 Figure 29. The CMAQ domain average H 2 O 2 (ppbv) for a day in July 2002 (a) and the CMAQ H 2 O 2 for a rural site (Big Meadows at Shenandoah National park) for the same day. The H 2 O 2 is 2-4 times greater than CMAQ SO 2 (from Figure 17) at 1000-2000 m. The altitudes presented are above ground level. 5.2.4 Lifetime Calculation The CMAQ and GOCART overestimation of SO 2 in the atmosphere suggests that the models do not properly simulate the lifetime of SO 2 ; specifically the models usually overestimate the lifetime. I have calculated the SO 2 lifetime for the conditions when aircraft observations were made (in the daytime for June through August in the Mid-Atlantic region see Figure 1 in Chapter 2 for map of locations). The mean profile of SO 2 shows a rapid decrease in mixing ratio with increasing altitude. This shows that SO 2 is destroyed on time scales fast relative to mixing in the planetary boundary layer. If SO 2 is on average destroyed before it is advected away a b 149 from the source region, the Eastern US, then we can assume that the rate of emissions into the atmosphere is equal to the rate of loss in the atmosphere (i.e., production equals loss). For an air column the production is the flux, F, in g m -2 s -1 . Because the steady state approximation can be applied to the system the rate of loss (L g m -2 s -1 ) of SO 2 in an air column is ? ? = 0 2 ][' dzSOkL (1) Where k? is the effective first order rate constant, s -1 , [SO 2 ] is the concentration of SO 2 in g m -3 as a function of altitude, z. The product of k? and [SO 2 ] must be integrated to an altitude where the concentration of SO 2 is much less than at the surface. At steady state the flux is equal to the loss and can be written as: ? ? = 0 2 ][' dzSOkF (12) The effective first order rate constant is the sum of all losses, including dry deposition, attack by OH, and oxidation by H 2 O 2 in cloud droplets. Equation 12 can be rearranged to separate the integral of the effective first order rate constant k? that is the inverse of the mean lifetime, ? (s). ? ? == 0 2 ][ 1 ' 1 dzSO Fk ? (13) 5.2.5 Verification of Lifetime Equation and Results If Equation 13 is valid then we can use it to calculate the lifetime from measurements of SO 2 made aboard the UMD research aircraft. I developed a method to test Equation 13 using the Gaussian plume equation multiplied by a lifetime factor (exp -t/? , where ? is a user-defined lifetime) to generate SO 2 profiles from a known 150 source. I calculated the average lifetime by averaging the integrated profiles and dividing them by the source flux. If the lifetime I calculated using Equation 13 was the same as the user-defined lifetime then Equation 13 is valid. The Gaussian plume equation is: ? ? ? ? ? ? + ?+ ? ??= 2 2 2 2 2 2 2 )( exp() 2 )( exp() 2 exp( 2 ),,( zzyzy hzhzyq zyxC ??????? (14) Here C is the concentration at an altitude z, a distance x downwind of a source and a distance y that is perpendicular to the x-axis. The emission rate is given by q, ? represents the wind speed and was assumed to be 6 m s -1 (the average wind speed for all 48 hr back trajectories ending at 1 km, associated with flights the UMD research aircraft made in 2002), h represents the effective stack height, assumed to be 200 m, and ? y and ? z are functions of x and represent the standard deviation of the plume distribution in the horizontal and vertical directions respectively. To determine the ? y and ? z values I assumed that the stability class was D, which is a neutral stability class and associated with winds greater than 6 m s -1 and moderate incoming solar radiation during the day (Schnelle and Dey, 2000). The equations for ? y and ? z are given below: ? y = ax b (15) ? z = cx d (16) Where a = 44 and b = 0.51 for stability class D (Schnelle and Dey, 2000) and c = 68 and d= 0.89 for stability class D (Seinfeld and Pandis, 1998). The effective stack height, h from Equation 14, is the sum of the actual height of the stack, H plus the plume rise ?h. The plume rise can be calculated using the Holland plume rise 151 formula (Schnelle and Dey, 2000), however, for simplicity, I assumed that all stack heights were 200 m. To test the method for determining SO 2 lifetime (using Equation 13) I generated SO 2 profiles from a single source of 4.5x10 7 g/day, in 14400 grid cells of 0.01 o latitude by 0.01 o longitude at 24 altitude levels from 0 ? 30 km. To account for the lifetime (due to chemical or physical removal of SO 2 ) I multiplied the Gaussian plume dispersion Equation 13 by exp -t/? , where t is the time it takes to get to the sampling point from the source and ? is a user-defined input lifetime. Figure 30 shows the SO 2 column contents in this box generated from this one source. The flux of SO 2 from the one source was 1.35 x10 -4 g m -2 day -1 for the chosen domain. US EPA (2003) states that 86% of SO 2 is generated from fuel combustion and the rest (14%) is generated from transportation and industrial sources. To account for these transportation and industrial sources I added 509 g day -1 to each grid cell. This added a 2.2x10 -5 g day -1 m -2 to the flux and 2.5x10 -5 gm -2 was added to the column content in each grid cell. Figure 30 shows the SO 2 column contents generated by one SO 2 source. In order to calculate the lifetime, the column contents are divided by the flux (the emission rate/ area of the box). Sampling any single point will probably not return the lifetime that is input into the model. However, an average of lifetimes from all sampling points in the box must equal the lifetime put into the model. 152 Figure 30. SO 2 column contents generated using a Gaussian plume dispersion model with one source. To calculate the lifetime of SO 2 from UMD aircraft profiles I am limited to the locations over which the aircraft flew; the number of locations is a small fraction of the 14400 grid of 0.01 o latitude by 0.01 o longitude described above. I performed a test to determine if the locations and number of spirals made by the UMD research aircraft were sufficient to calculate the average lifetime using Equation 13. In 2002 the UMD airplane flew at 17 different locations and made a total of 90 different spirals sampling SO 2 . Even though there were only 17 locations, the SO 2 profiles were independent because the winds changed between sampling days. To represent these locations in the model the 17 locations were shifted by 0.1 degree latitude or longitude, North, South, East, and West for a total of 85 sampling points. To test the effects of nudging the points on lifetime, the 85 points were shifted 0.15 degrees latitude or longitude, North, South, East, and West. This resulted in 5 sets of 85 153 samples that should have similar average lifetimes. Figure 31 shows the locations of these 425 sampling points (5 x 85) with sources of SO 2 in green. The sampling locations are represented by pink circles. I also adjusted the area of the box used to calculate the flux in order to determine how that affected the resulting lifetime. Lifetimes of 8, 16, 24 and 32 hours were tested using three different sized boxes to calculate the flux of SO 2 . Figure 32 shows the boxes used in this study and Table 6 gives the locations of the boxes and the distances and times between the westernmost sampling point and the western edge of the box. Figure 31. The locations of the 425 sampling points to be used in the simplified Gaussian plume dispersion model are shown in pink. The green circles represent power plants emitting SO 2 and the size of the circle represents the relative size of SO 2 emissions. 154 Figure 32. Boxes used to determine SO 2 flux from point sources. Box 1 Box 2 Box 3 Box 4 Box 5 Initial latitude 36.1 36.1 36.1 36.1 36.1 Final latitude 41.2 41.2 41.2 41.2 41.2 Initial longitude -83.5 -85.3 -88.5 -93.5 -98.5 Final longitude -75.5 -75.5 -75.5 -75.5 -75.5 Distance from western most sample point (m) to western edge of box 3.6E+05 5.3E+05 8.0E+05 1.2E+06 1.7E+06 Time (hours) from western most sample point to western edge of box 16.7 24.7 36.8 6.9 77.0 Table 6. Location of box edges (from Figure 32) in degrees latitude and longitude. Also shown is the time needed for sources from the western most Box 1 Box 2 Box 3 Box 4 Box 5 Power plants 155 points of the box to arrive at the western most sampling point. Table 7a shows the lifetimes generated using an input lifetime of 8 hours and using 3 different fluxes (from Box 1, Box 2 and Box 3) and 5 different sets of 85 sampling locations for a total of 15 groups of 85 measurements of lifetimes. The lifetimes and SO 2 column contents appear to be lognormally distributed (Figure 33). The mean ? x and variance ? x 2 for a lognormal distribution are given by the following (Wilks, 1995) ? x = exp[? y + 0.5*? y 2 /] (17) ? x 2 = (exp[? y 2 ] ? 1) * exp[2? y + ? y 2 ] (18) Where ? y and ? y 2 are the mean and variance of the transformed variable y = ln(x). The lognormal statistics for each of the sets of lifetimes calculated with different fluxes (from Box 1, Box 2 and Box 3) are shown in Table 7b. Table 7c shows the average lifetime of all 15 groups with the standard deviation and the standard error (the standard deviation / ?15). Statistics for lifetimes calculated using inputs of 16, 24 and 32 hours are shown in Tables 8-10. I calculated the 2-? uncertainty associated with the lifetimes generated using the method, by accounting for the accuracy (the difference between the median and the 95 th percentile of the 15 average lifetimes) and the precision (the standard error). I added these in quadrature and determined there was a 30% uncertainty associated with the method assuming a normal distribution, the uncertainty associated with the method assuming a lognormal distribution was 20%. 156 Figure 33. Histogram of SO 2 lifetimes calculated using Gaussian plume dispersion model and an input lifetime of 8 hours. 0 100 200 300 400 500 600 700 2 4 6 8 101214161820470 SO 2 lifetime (hours) F r eq ue nc y 157 A C Box 1 Center North South West East Normal Distribution Log-normal Distribution Mean 9.43 6.65 7.09 5.35 5.87 Mean 7.32 6.25 Standard Error 5.12 2.70 3.59 0.93 1.52 Standard Error 0.44 0.29 Standard Deviation 47.22 24.86 33.14 8.53 14.05 Standard Deviation 1.70 1.11 Box 2 Center North South West East Mean 10.74 7.76 8.28 6.34 7.08 Standard Error 5.37 2.82 3.77 0.97 1.60 Standard Deviation 49.48 26.03 34.74 8.97 14.77 Box 3 Center North South West East Mean 11.31 8.22 8.75 6.74 7.55 Standard Error 5.53 2.91 3.88 1.00 1.65 Standard Deviation 50.99 26.82 35.79 9.22 15.22 B Box 1 Center North South West East Mean 5.87 5.33 4.69 6.98 5.49 Standard Error 3.19 2.60 2.12 4.06 2.96 Standard Deviation 29.41 23.95 19.58 37.46 27.26 Box 2 Center North South West East Mean 7.70 6.87 6.42 8.13 7.53 Standard Error 3.15 2.56 2.35 3.20 3.06 Standard Deviation 29.07 23.57 21.65 29.49 28.17 Box 3 Center North South West East Mean 7.99 7.19 6.74 8.05 7.79 Standard Error 2.78 2.32 2.18 2.59 2.64 Standard Deviation 25.60 21.37 20.06 23.87 24.36 Table 7. a) Statistics of SO 2 lifetimes (hours) from profiles generated using a Gaussian plume model assuming a lifetime of 8 hours for boxes 1, 2 and 3 for a) an assumed normal distribution b) an assumed lognormal distribution. c) The average lifetime for all sets (Center, North, South, West and East) for boxes 1, 2 and 3 as well as the standard deviation and standard error (standard deviation / ?15). 158 A C Box 3 Center North South West East Normal Distribution Log-normal Distribution Mean 16.88 13.78 14.49 12.25 13.04 Mean 15.30 14.76 Standard Error 5.90 3.33 4.18 1.45 2.05 Standard Error 0.74 0.42 Standard Deviation 54.43 30.67 38.52 13.38 18.93 Standard Deviation 2.87 1.63 Box 4 Center North South West East Mean 18.54 15.30 16.09 13.72 14.62 Standard Error 6.08 3.42 4.30 1.50 2.11 Standard Deviation 56.05 31.53 39.62 13.82 19.44 Box 5 Center North South West East Mean 22.71 18.80 19.76 16.89 17.99 Standard Error 7.35 4.13 5.19 1.82 2.55 Standard Deviation 67.72 38.11 47.88 16.73 23.50 B Box 3 Center North South West East Mean 15.14 14.16 15.51 14.55 14.75 Standard Error 3.86 3.40 4.52 3.45 3.71 Standard Deviation 35.58 31.34 41.71 31.79 34.18 Box 4 Center North South West East Mean 15.39 14.41 16.57 14.85 15.02 Standard Error 2.65 2.34 3.58 2.47 2.52 Standard Deviation 24.46 21.56 32.99 22.74 23.24 Box 5 Center North South West East Mean 18.63 17.50 18.45 18.03 18.10 Standard Error 3.00 2.67 3.23 2.82 2.80 Standard Deviation 27.67 24.62 29.75 25.96 25.82 Table 8 a) Statistics of SO 2 lifetimes (hours) from profiles generated using a Gaussian plume model assuming a lifetime of 16 hours for boxes 3, 4 and 5 for a) an assumed normal distribution b) an assumed lognormal distribution. c) The average lifetime for all sets (Center, North, South, West and East) for boxes 3, 4 and 5 as well as the standard deviation and standard error (standard deviation / ?15). 159 A C Box 3 Center North South West East Normal Distribution Log-normal Distribution Mean 20.56 17.51 18.34 15.84 16.65 Mean 20.23 20.71 Standard Error 6.06 3.52 4.31 1.70 2.28 Standard Error 1.04 0.73 Standard Deviation 55.86 32.43 39.71 15.69 20.98 Standard Deviation 4.01 2.82 Box 4 Center North South West East Mean 23.73 20.52 21.49 18.77 19.76 Standard Error 6.22 3.60 4.41 1.76 2.33 Standard Deviation 57.37 33.19 40.64 16.22 21.46 Box 5 Center North South West East Mean 29.58 25.68 26.86 23.59 24.81 Standard Error 7.52 4.36 5.33 2.14 2.82 Standard Deviation 69.35 40.15 49.15 19.72 26.01 B Box 3 Center North South West East Mean 19.72 18.66 21.70 18.71 19.19 Standard Error 4.52 4.03 6.21 3.99 4.37 Standard Deviation 41.68 37.20 57.22 36.77 40.25 Box 4 Center North South West East Mean 21.15 20.01 24.64 20.08 20.66 Standard Error 3.00 2.65 4.95 2.74 2.87 Standard Deviation 27.69 24.46 45.62 25.22 26.45 Box 5 Center North South West East Mean 25.94 24.68 26.33 24.81 25.14 Standard Error 3.31 2.97 3.72 3.08 3.06 Standard Deviation 30.52 27.42 34.26 28.39 28.23 Table 9. a) Statistics of SO 2 lifetimes (hours) from profiles generated using a Gaussian plume model assuming a lifetime of 24 hours for boxes 3, 4 and 5 for a) an assumed normal distribution b) an assumed lognormal distribution. c) The average lifetime for all sets (Center, North, South, West and East) for boxes 3, 4 and 5 as well as the standard deviation and standard error (standard deviation / ?15). 160 A C Box 4 Center North South West East Normal Distribution Log-normal Distribution Mean 27.58 24.40 25.48 22.50 23.56 Mean 29.17 28.67 Standard Error 6.30 3.71 4.47 1.93 2.48 Standard Error 1.48 1.26 Standard Deviation 58.10 34.18 41.21 17.83 22.83 Standard Deviation 5.75 4.88 Box 5 Center North South West East Mean 34.86 30.99 32.30 28.72 30.04 Standard Error 7.62 4.49 5.41 2.36 3.01 Standard Deviation 70.28 41.41 49.88 21.76 27.75 Box 6 Center North South West East Mean 41.35 36.80 38.34 34.16 35.70 Standard Error 8.95 5.28 6.36 2.79 3.54 Standard Deviation 82.52 48.65 58.60 25.68 32.67 B Box 4 Center North South West East Mean 25.44 24.19 30.80 23.95 24.84 Standard Error 3.30 2.91 6.03 2.96 3.16 Standard Deviation 30.43 26.87 55.63 27.29 29.16 Box 5 Center North South West East Mean 31.54 30.21 32.28 30.01 30.50 Standard Error 3.58 3.23 4.07 3.32 3.29 Standard Deviation 33.03 29.80 37.56 30.57 30.36 Box 6 Center North South West East Mean 37.46 35.89 38.16 35.65 36.23 Standard Error 4.22 3.81 4.73 3.89 3.88 Standard Deviation 38.91 35.14 43.64 35.89 35.75 Table 10. a) Statistics of SO 2 lifetimes (hours) from profiles generated using a Gaussian plume model assuming a lifetime of 32 hours for boxes 3, 4 and 5 for a) an assumed normal distribution b) an assumed lognormal distribution. c) The average lifetime for all sets (Center, North, South, West and East) for boxes 3, 4 and 5 as well as the standard deviation and standard error (standard deviation / ?15). 161 The above study was used to determine the uncertainty associated with the method. I calculated the actual lifetime of SO 2 for 180 daytime profiles made in June, July and August from 2000-2003 in the Mid-Atlantic region using equation 13. I integrated all aircraft profiles of SO 2 from the lowest altitude where measurements were made (usually 3 m above ground) to 5000 m. The aircraft generally measured SO 2 up to 3000 m. I assumed SO 2 was 0.07 ppb between 5000 m and the highest altitude the aircraft sampled (Thornton et al. 1987; Andronache et al. 1997). Extrapolating the SO 2 to 5000 m added 9% on average to the column measured by the aircraft. I calculated the flux using national inventories and an estimate of the biogenic contribution as described in section 5.2.3. Back trajectories of 12, 24, 32, 40 and 48 hours (with one hour interval outputs) were used to determine which states to include in the flux calculation. The flux associated with each state (or municipality) was weighted by the number of back trajectory points in the state divided by the total number of back trajectory points. The weighted fluxes were then summed. Figure 34 shows 24 hr back trajectories for all 180 profiles. 162 Figure 34. Back trajectories of 24 hr (with one hr intervals) associated with the 180 SO 2 profiles used to calculate the SO 2 lifetime. Statistics for lifetimes are shown in Table 11 and a histogram of the lifetimes is shown in Figure 35. The fluxes used to calculate the lifetime were determined using 24 hour back trajectories. I calculated the uncertainty associated with the lifetime by accounting for four factors: 1. The uncertainty associated with the method (Equation 13), determined from the simplified Gaussian plume model to be 30% assuming a normalized distribution (the uncertainty associated with the method was 20% assuming a lognormal distribution). 163 2. The standard error (Table 11), calculated from the standard deviation of all 180 lifetimes and divided by the square root of the number of independent days flights made (in this case 60 days). 3. Uncertainties associated with area and point source emissions (area source emission uncertainties were estimated to be 50% and point source emission uncertainties were estimated to be 16% using the Luke et al. (1997) reported uncertainty). 4. Uncertainties associated with SO 2 measured aboard the University of Maryland research aircraft (assumed to be 16% from Luke et al. (1997)). To determine the uncertainty associated with area sources I recalculated the lifetime assuming a 50% uncertainty associated with area source emissions and this resulted in a 6% uncertainty associated with the lifetime. I also recalculated the lifetime assuming a 16% uncertainty associated with the point source emissions and this resulted in a 14% uncertainty associated with the lifetime. Therefore the total uncertainty associated with emissions was 20% (14% + 6%). I then added the four factors listed above in quadrature to get the 2-? uncertainty of 7 hours. The average lifetime is 19 ? 7 hours (at the 95 percent confidence level). The lognormal average lifetime is 20 ? 6 hours. These lifetimes are within the range of model results (for the global average SO 2 lifetime) of 0.6 to 2.6 days (Pham et al., 1995; Chin et al., 1996; Rested et al., 1998; Koch et al., 1999; Roelofs et al., 1998; Berglen et al., 2004) and are on the shorter side of the lifetime estimates. 164 Lifetime statistics (hours) Mean 19 Standard Error 1.7 Median 17 Standard Deviation 13 Minimum 1.5 Maximum 63 Count 180 lognormal distribution Lifetime statistics (hours) Mean 20 Standard Deviation 17 Standard Error 2.2 Table 11. Statistics for SO 2 lifetime. The standard error is the standard deviation divided by the square root of 60 (the number of days sampled). 0 5 10 15 20 25 30 35 0 5 10 15 20 25 30 35 40 45 50 55 60 65 SO 2 lifetime (hours) F r eq ue nc y Figure 35. Histogram of SO 2 lifetimes calculated using 24 hour back trajectories to weight the flux. These lifetimes were calculated from 180 profiles measured in the daytime in the summer from 2000-2003. 165 As discussed above, GOCART uses similar emissions to that presented in national inventories and CMAQ emissions are derived from these national inventories. Therefore it is unlikely that the models over-estimate SO 2 because the emissions of SO 2 are too large. The model over-prediction is likely explained by inadequate oxidation of SO 2 to sulfate in clouds. Mueller et al. (2006) found the CMAQ has difficulty generating typical cloud cover, and reduced cloud cover results in less oxidation of SO 2 by H 2 O 2 . UMD CMAQ runs may also underestimate cloud cover. Future work should include a verification of CMAQ and GOCART cloud cover. To investigate the effects of OH on the lifetime of SO 2 , the EPA?s Community Multi-scale Air Quality (CMAQ) was used to generate OH. CMAQ version 4.5.1 was used with CBIV chemistry, 2002 base year emissions supplied by a regional planning organization, and MM5 version 3 meteorology that was nudged back to observations using data assimilation (Zhang and Anthes, 1982; Grell et al., 1995; Zhang and Zheng, 2004). We have examined OH profiles (from the surface to 645 mbar) from CMAQ for days in 2002 when the University of Maryland Research aircraft made spirals. The CMAQ OH profiles along 24-hr Hysplit back trajectories (ending at 1 km and the location of the UMD aircraft spiral) were averaged to get the 24 hour average OH profile. All of the 24 hour average OH profiles associated with aircraft profiles (made in June through August 2002) were then averaged. This average OH profile represents the daily average OH likely encountered by the SO 2 plumes measured aboard the University of Maryland Research aircraft in 2002. The effective second order rate constant, for the SO 2 + OH reaction, changes by only 2% 166 between the surface and 645 mbar and so it can be approximated with the high pressure rate constant of 9.5 ? 10 -13 molecules cm -3 s -1 (JPL, 2006). The approximate lifetime of SO 2 (with respect to OH oxidation), ? OH (seconds), can be calculated as shown below: ? OH = (k OH ? [OH]) -1 (5) Here k OH is the high pressure rate constant (molecules cm -3 s -1 ) and [OH] is the concentration of OH (molecules cm -3 ). The approximate average SO 2 lifetime, with respect to OH oxidation, for days and locations where the University of Maryland research aircraft made flights is shown as a function of altitude in Figure 36. The average SO 2 lifetime (with respect to OH oxidation) between the surface and 950 mbar, is seven days, and this suggests that OH accounts for only 11% of SO 2 removal. 167 600 700 800 900 1000 1100 02468 Lifetime of SO 2 with respect to OH oxidation (days) P r esss u r e ( mb a r ) Figure 36. The lifetime of SO 2 with respect to OH oxidation, where OH was generated from CMAQ. Between the surface and 950 mbar the average lifetime of SO 2 (with respect to OH oxidation) is seven days. 5.3 Conclusions Aircraft measurements of O 3 were compared with CMAQ. CMAQ over- predicts O 3 from the surface to 600 m, and under-predicts O 3 by 10% between 600 and 2600 m. The CMAQ column content is 3% smaller than the aircraft column content. Possible explanations for the modeled and measured differences include misrepresentation of clouds and aerosols, especially in how they affect photchemistry. I made adjustments in the photochemistry of CMAQ by accounting for aerosol properties measured during a four day event in July 2002. The aerosol properties affected the photolysis of NO 2 and this affected the O 3 production. In general the revised CMAQ model runs over-predicted O 3 above 500 m (~1 ppb) and under- 168 predicted O 3 below 500 m (1-4 ppb). O 3 reductions are expected for 2018 because stricter regulations on power plant emissions will be implemented and motor vehicles should have cleaner emissions. I tested how the expected decreases in O 3 would be affected if CMAQ accounted for aerosol properties in the NO 2 photochemistry. At the surface, I found that the standard CMAQ runs over-predict O 3 reductions up to 2 ppb and above 500 m the standard CMAQ runs under-predict O 3 reductions up to 2 ppb. SO 2 from CMAQ and GOCART were also compared with aircraft profiles. The models tend to over-predict the SO 2 column content by 50-55% (GOCART and CMAQ respectively). This over-prediction may result from an over-prediction of the lifetime by either including too large emission sources of SO 2 or not accounting for destruction processes properly. I calculated the summertime lifetime of SO 2 in the Mid-Atlantic region to be 19 ?7 hours from in-situ measurements of SO 2 . This is on the short side of typical global model estimates of the SO 2 lifetime. The emissions used in CMAQ and GOCART do not appear to be overestimated and thus it is likely that these models underestimate the rate of removal of SO 2 . I examined the CMAQ profiles of OH to determine the lifetime of SO 2 with respect to oxidation by OH. Oxidation by OH roughly accounts for 25% of the SO 2 lifetime. This suggests that CMAQ underestimates oxidation of SO 2 in clouds. 169 Chapter 6: A Side by Side Comparison of Filter-based PM 2.5 Measurements at a Suburban Site: A Closure Study 6.1 Introduction 6.1.1 Background As shown in Chapter 1, numerous counties in the Mid-Atlantic region violated the NAAQS PM 2.5 standards. Models can be effective tools to determine sources and methods for reducing PM 2.5 , but this requires accurate measurements of PM 2.5 . In this Chapter I will give results from ambient measurements an uncertainty analysis of PM 2.5 samplers used in the Speciation Trends Network. There are no NAAQS standards for speciated mass; however, understanding the PM 2.5 composition can aid states in determining sources of PM 2.5 . This is one reason why data is collected from monitors in the Speciation Trends Network. Accurate and precise measurements of the speciated mass are necessary to determine sources and develop strategies to reduce PM 2.5 . Some work presented in this chapter is from Hains et al. (2007b). A part of the Maryland Aerosol Research and Characterization study (MARCH- Atlantic) was conducted in Maryland in the Baltimore-Washington corridor. Experiments were carried out during 2002 at a suburban site in Maryland, United States, where two samplers from the U.S. Environmental Protection Agency (USEPA) Speciation Trends Network: Met One Speciation Air Sampling System ? STN S and Thermo Scientific Reference Ambient Air Sampler ? STN R , two Desert Research Institute Sequential Filter Samplers ? DRI F , and a continuous TEOM monitor (Thermo Scientific Tapered Element Oscillating Microbalance, 1400a), all run in parallel. These monitors differ not only in sampling configuration but also in 170 protocol-specific sample analysis procedures. I will present statistics for PM 2.5 mass and speciated mass as well as an uncertainty analysis for the different samplers. This Chapter addresses PM 2.5 concentration and composition as a function of time for summer and winter and the uncertainty associated with PM 2.5 measurements. 6.1.2 Experiment STN RS and DRI F differ in filter types used to collect aerosol as well as flow rates required by the specific cyclone to maintain a stable cut-point at 2.5 ?m. Figure 1 illustrates all the sampler configurations and Table 1 summarizes the specifications of the samplers along with analytical methods for determining all species reported. STN R samplers are considered FRM equivalent (Solomon et al., 2003) and have been compared with other samplers (Peters et al., 2001b, 2001c; Solomon et al., 2003), while DRI F has been successfully deployed in many air quality studies since 1988 (Chow et al., 1992, 1996; Chen et al., 2002; Watson and Chow, 2002). 171 DRI analysis * RTI analysis ** PM 2.5 mass gravimetry mass gravimetry Trace elements x-ray fluorescence x-ray fluorescence Sulfate ion chromatography ion chromatography Nitrate ion chromatography ion chromatography Ammonium automated colorimetry ion chromatography Chloride ion chromatography chlorine is measured with XRF Sodium ion atomic absorption ion chromatography Potassium ion atomic absorption ion chromatography EC thermal optical reflectance (IMPROVE) thermal optical transmittance (NIOSH***) OC thermal optical reflectance (IMPROVE) thermal optical transmittance (NIOSH***) Instrument specifications DRI F STN R STN S Flow (L min -1 ) 20 ? 0.8 16.7 ? 0.3 (mass and elements) 7.3 ? 0.1(ions and carbon) 6.7 ? 0.1 Cyclone Bendex 240 AN 3.68 SC 2.141 Nitric acid denuder coating Aluminum oxide Magnesium oxide Magnesium oxide Sample inlet height (m) 10 15 15 Filter diameter (mm) 47 47 47 Table 1. Analytical methods for species collected by DRI F (analyzed by DRI) and STN RS (analyzed by RTI) and instrument specifications. Flow rate uncertainties are ? 1-?. * DRI operating procedure, 1990; Chow et al., 1993c; Chow et al., 2001. ** US EPA, 2001; Thermo Anderson, 2001. *** National Institute for Occupational Safety and Health. 172 Ai r F l o w 16 . 7 L / m i n MgO denuder Mass, Elements NO 3 - , SO 4 2- NH 4 + , K + , Na + OC, EC, TC Not analyzed 1 6 .7 L / m i n 7. 3 L / m i n 24 L / m i n 24 L / m i n 7. 3 L / m i n Pump Nylon Filter Teflon Filter PM 2.5 Inlet Manifold Manifold Critical orifice flow controller Cellulose Filter Quartz Filter PM 2.5 Cyclone PM 2.5 Cyclone Ai r F l o w 16 . 7 L / m i n 1 6 .7 L / m i n 7. 3 L / m i n 24 L / m i n 24 L / m i n 7. 3 L / m i n 16 . 7 L / m i n 1 6 .7 L / m i n 7. 3 L / m i n 24 L / m i n 24 L / m i n 7. 3 L / m i n 16 . 7 L / m i n 1 6 .7 L / m i n 7. 3 L / m i n 24 L / m i n 24 L / m i n 7. 3 L / m i n 16 . 7 L / m i n 1 6 .7 L / m i n 7. 3 L / m i n 7. 3 L / m i n 24 L / m i n 24 L / m i n 24 L / m i n 24 L / m i n 7. 3 L / m i n Electronic Flow Sensor 6. 7 L / m i n 6. 7 L / m i n Mass, Elements by XRF NO 3 - , SO 4 2- , NH 4 +, K + , Na + OC, EC MgO Denuder Flow Controller PM 2.5 Cyclone Flow Controller Electronic Flow Sensor Flow Controller Electronic Flow Sensor Pump Teflon Filter 6. 7 L / m i n PM 2.5 Cyclone Nylon Filter PM 2.5 Cyclone Quartz Filter Manifold 6. 7 L / m i n 6. 7 L / m i n 6. 7 L / m i n 6. 7 L / m i n 6. 7 L / m i n 6. 7 L / m i n 6. 7 L / m i n 6. 7 L / m i n 6. 7 L / m i n 6. 7 L / m i n 6. 7 L / m i n 6. 7 L / m i n 6. 7 L / m i n (a) STN R (b) STN S 173 Figure 1. Sampler configuration for a) STN R (Anderson RAAS) b) STN S (Met- One SASS) c) DRI F for elements and ions d) DRI F for carbonaceous material. STN RS use a critical orifice to set the flow rate and monitors it with a mass flow sensor. STN RS record ambient temperature and pressure and this is used to convert the mass flow to volumetric flow. The average volumetric flow rate and total volume sampled are recorded for every 24-hr sampling period (Thermo Anderson, Pump 73 L/ m i n ma k e - u p f l o w 20 L/min Mass, Elements 113 L/min Al coated HNO 3 Denuder PM 2.5 Cyclone 20 L/min Teflon Filter SO 4 2- , NH 4 + Na+, K + , Cl - , NO 3 - NO 3 - Quartz Filter NaCl impregnated cellulose Filter Flow Controller Flow Controller 73 L/ m i n ma k e - u p f l o w 73 L/ m i n ma k e - u p f l o w 73 L/ m i n ma k e - u p f l o w 73 L/ m i n ma k e - u p f l o w EC, OC EC, OC 73 L/ m i n m a ke- up f l ow 20 L/min 113 L/min Pump 20 L/min Teflon Filter Quartz Filter Quartz Filter Flow Controller Quartz Filter PM 2.5 Cyclone Flow Controller 73 L/ m i n m a ke- up f l ow 73 L/ m i n m a ke- up f l ow (c) DRI F (elements and ions) (d) DRI F (carbon) 174 2001; US EPA, 2001). The STN R flow was calibrated with a flow audit device (BGI deltaCal) and the STN S flow was calibrated with a bubble meter (Sensidyne/Gilian Gilibrator 2). The DRI F also uses a critical orifice to maintain constant flow, but the flow was measured and adjusted only once every third day using a rotameter (calibrated against a NIST-traceable Roots meter). The flow rate is recorded before and after each three-day sampling period for the DRI F , and it can drop by 4% due to buildup of water and particles on the filter. DRI uses the average flow rate (from the initial and final flow) to calculate the total volume sampled and the resultant mass concentration. STN RS record the total volume sampled, which is calculated from the mass flow sensor, temperature and pressure readings. The sample flow rates for PM 2.5 mass were 20, 16.7, and 6.7 L min -1 in DRI F , STN R , and STN S , respectively. Since all the samplers used 47-mm filters, DRI F imposed an approximately 17% larger face velocity than the STN R and an 82% larger face velocity than the STN S around the filter. The STN R sample flow rate was 7.3 L/min for ions and carbon (similar to the STN S ) and the DRI F imposed a 64% larger face velocity than the STN R . Cyclones used by STN R and STN S (Table 1) exhibit different size-selection curves at their specified flow, but Peters et al. (2001c) found that only sites dominated by crustal material had significantly different PM 2.5 mass collected by the two samplers. Chen et al. (2002) showed a minor crustal material contribution at FME, ~3% of PM 2.5 mass on average, and therefore strong biases resulting from imperfect size cut are not expected in this study. There may also be diffusion losses of ultrafine particles between the sampler inlet and filter which vary with the different flow rates 175 used by DRI F , STN R and STN S . Ultrafine particles (< 0.1 ?m in diameter) typically contribute little to PM 2.5 mass in this environment (e.g., Tolocka et al., 2005; Ondov et al., 2006) and strong biases resulting from diffusion losses are unlikely. The DRI F used a front quartz-fiber filter with a sodium-chloride-impregnated cellulose backup filter to collect nitrate. The backup filter captured nitrate volatized from the front filter (Zhang and McMurry, 1992). These filters were located behind a bundle of aluminum-oxide-coated denuders to remove gaseous nitric acid. Specifications of the denuders are described in Chow et al. (1993a). The STN R and STN S collected nitrate particles behind a magnesium-oxide denuder on a single nylon filter (Figure 1). Specifications of the denuders are described in Research Triangle Institute (2000). Frank (2006) found that denuded nylon filters captured more nitrate than undenuded Teflon filters. The different denuders and filter types used by the STN RS and DRI F in this study likely affect the nitrate collection efficiency as suggested by Solomon et al. (2003) and Frank (2006). Quartz-fiber filters were used in all the samplers to collect carbonaceous material, and DRI F used backup filters to account for known sampling artifacts from volatile organic compounds (McDow and Huntzicker, 1990; Turpin et al., 1994; Chow et al., 1996; Chow et al., 2001). For carbon analysis, RTI adopted the Speciation Trends Network-Thermal Optical Transmission (STN-TOT) method (Peterson and Richards, 2002; OC/EC Laboratory, 2003), while DRI used the Interagency Monitoring of Protected Visual Environments-Thermal Optical Reflectance (IMPROVE-TOR) method (Chow et al., 1993b). The IMPROVE-TOR and STN-TOT differ in temperature steps used to extract OC and EC and in optical 176 charring corrections. They usually yield equivalent total carbon (TC) but different OC and EC concentrations (Chow et al., 2001; Schmid et al., 2001; Chow et al., 2004; Subramanian et al., 2004; Chow et al., 2005a). The IMPROVE-TOR method generally assigns less OC and more EC to a filter sample than the STN-TOT method. DRI quantified water-soluble potassium (K + ) and sodium (Na + ) with atomic absorption spectroscopy (AAS) and RTI quantified the species with ion chromatography (IC). AAS has a lower detection limit (Chow et al., 1993c; Technology Transfer Network Air Quality System, 2006). There were also differences in blank collection. A field blank was collected every third day for the DRI F sampler and once every two weeks for the STN S sampler. Only one field blank was collected for the STN R sampler. DRI corrected for field blanks as part of their analysis (Watson et al., 1989a; 1989b), but RTI did not. To correct STN RS samples for field blanks, we averaged all STN RS blank values, converted them from mass/filter to mass/m 3 using the volume sampled by the instrument, and then subtracted the blanks from the mass measurement. Sample recovery was scheduled for different time periods. The DRI F filters were collected from the site every three days, so that used filters remained in the sampler for up to 2.5 days (an average of 1.5 days). The STN R filters were collected every day, immediately after the sampling finished, so that used filters remained in the sampler for less than 30 minutes. The STN S filters were collected every other day, so that used filters remained in the sampler for about 12 hours. Chen (2002) performed an audit experiment in summer 2001 at FME with the DRI F samplers, to determine how filters left in the sampler may be affected by volatile losses and/or 177 passive collection. He found that OC and TC mass (measured on the front quartz- fiber filters) decreased (by 38% and 29%, respectively) during a 2.5-day period after sampling. Total PM 2.5 mass and sulfate mass varied less than their respective uncertainties. A TEOM measures near real-time continuous PM 2.5 mass. The TEOM at FME drew ambient air in at 3 L/min through a PM 2.5 cyclone inlet. A constant volumetric flow was achieved using a mass flow controller corrected for ambient temperature and pressure. The air stream was heated to 50 o C to maintain a low, relatively constant relative humidity. This heating likely increased volatilization of nitrate and semi-volatile organic compounds. The TEOM measurements were adjusted with scaling factors of 1.03 ? TEOM mass + 3.0 to account for loss of semi- volatile material and to be compatible with FRM measurements as recommended by Patashnick and Rupprecht (1991). The mean mass concentration was recorded every 30 minutes, every hour, and every eight hours. All one-hour measurements made in a day were averaged to compare with the DRI F and STN S data. 6.2 Results and Discussion 6.2.1 Uncertainty Analysis Uncertainties associated with flow control and sample analysis need to be accounted for to determine the uncertainty in total PM 2.5 and each reported species concentration. For STN RS , the species concentration (with units of mass m -3 at ambient temperature and pressure) is calculated using the equation below: 178 Species concentration = m ? (t ?mass flow ? MM -1 ? R ? T ? P -1 ) -1 (1) Here m is the mass of a given species on the filter, t is the time over which sampling occurred, mass flow has units of mass time -1 , MM is the molar mass of the air sampled, R is the gas constant (0.08314 L atm K -1 mol -1 ), T is ambient temperature and P is the ambient pressure. Uncertainties in the calculated concentration reflect uncertainties in the laboratory analysis, the mass flow sensor reading, the temperature reading and the pressure reading. Uncertainties associated with the integration time appear to be less than 1% and are therefore not included in the error analysis. US EPA (2001) states that STN RS temperature readings must be within ?4 K of the actual temperature and pressure readings must be within ?0.013 atm of the actual pressure. These ranges represent part of the uncertainty associated with the measurements. The precision associated with a commercial mass flow sensor for the maximum allowable mass flow, i.e., ?2% at the 1-? level, is used as an estimate of the mass flow sensor uncertainty (Table 1). Flanagan et al. (2006) report the percentage difference in laboratory replicates of PM 2.5 and speciated masses. I adopted their values of laboratory uncertainty to calculate the total uncertainty. The resultant ?2-? uncertainty, u, (i.e., the 95% confidence level) associated with PM 2.5 mass, sulfate, ammonium, OC or elemental concentration is given by: u = mass concentration ? [(?A/A) 2 + (?mf/mf) 2 + (?T/T) 2 + (?P/P) 2 ] ? (2) Here ?A/A represents fractional uncertainty associated with the laboratory determination of the mass of a species (uncertainties from Flanagan et al., 2006 were used), ?mf/mf represents the fractional uncertainty associated with the mass flow meter measurements, and ?T/T and ?P/P represent the fractional uncertainty 179 associated with temperature and pressure measurements, respectively. RTI did not report uncertainties for samples analyzed in 2002, however they did report uncertainties for samples measured in the U.S. in 2005 to the EPA?s Air Quality System database (AQS, Technology Transfer Network Air Quality System, 2006). The uncertainties reported by RTI include laboratory analysis (?1-? uncertainty) and a 5% uncertainty associated with flow control and shipment of the samples (RTI, 2004). Using their uncertainties associated with concentrations that were similar to (within ?1% of) the FME samples, and multiplying them by two to obtain the ?2-? uncertainties, I found the resultant uncertainties are on average 2.5 times larger than those calculated from Equation (2) for most species except PM 2.5 mass (Table 2). For this Chapter I adopt the RTI reported ?2-? uncertainties. Kim et al., (2005) report fractional uncertainty associated with measurements made in New York, New Jersey and Vermont. Uncertainties they reported for sulfate, ammonium and calcium agreed within 20% of the uncertainties used in this paper. Calculated 2 ? uncertainty (%) RTI reported 2 ? uncertainty (%) PM 2.5 10 10 OC 12 27 Sulfate 9 16 Ammonium 4 14 Iron 6 16 Table 2. Comparison of 2-? uncertainty in concentration calculated using Equation 2 and RTI reported 2-? uncertainty (from 2005 AQS database). The DRI F measures the flow rate using a pressure drop across a critical orifice. Ambient temperature and pressure can alter this flow rate. DRI calculates the uncertainty for each measurement by accounting for the variability between the initial 180 and final flow tests through 24-hr sampling (typically ?4%), as well as precision in laboratory analyses (Chow et al., 1993c). The monthly average concentration of species and the average uncertainty (i.e., the average of all 2-? uncertainty values for the month) for STN RS versus DRI F are shown in Table 3 along with the signal-to- MDL (minimum detection limit) ratio, where the MDL was obtained from Chow et al. 1993c) for the DRI samplers and the median of all 2005 MDL values reported by RTI (to the EPA?s AQS database) for the STN samplers. The signal-to-noise ratio for each species can be calculated from Table 3 by dividing the species average by the 2- ? uncertainty. 181 DRI F spe c i e s ave r age (? 2 ? u n cert a i n t y ) 7. 30 ? 0 . 98 2. 04 ? 0 . 27 0. 98 ? 0 . 12 0. 97 ? 0 . 11 0. 61 ? 0 . 04 2. 00 ? 0 . 53 0. 75 ? 0 . 18 2. 74 ? 0 . 61 3. 56 ? 1 . 66 14. 6 ? 9 . 40 49. 0 ? 10 . 5 41. 7 ? 5 . 63 36. 6 ? 10 . 7 1. 76 ? 52 . 2 DRI F S i gn al -to- MD L 8. 58 240 115 115 72 58. 7 22. 2 80. 7 21. 0 19. 2 64. 5 245. 1 24. 4 3. 5 ST NR S i gn al -to- MD L 11. 9 193 60. 2 171 171 7. 82 1. 45 9. 28 5. 46 2. 40 6. 59 22. 12 2. 98 0. 57 ST N R spe c i e s ave r age (? 2 ? u n cert a i n t y) 8. 77 ? 0 . 94 2. 32 ? 0 . 40 1. 02 ? 0 . 14 1. 49 ? 0 . 29 1. 49 ? 0 . 29 1. 88 ? 0 . 63 0. 35 ? 0 . 45 2. 23 ? 0 . 78 2. 95 ? 1 . 52 17. 8 ? 4 . 79 47. 4 ? 7 . 94 46. 4 ? 6 . 68 29. 9 ? 12 . 7 2. 84 ? 2 . 72 RM S D i ffe r e n c e 1. 66 0. 35 0. 14 0. 57 0. 93 0. 73 0. 49 0. 97 1. 10 4. 72 5. 06 9. 51 16. 71 1. 95 A v era g e D i ffe r e n c e (S TN R - DRI F ) 1. 47 0. 28 0. 05 0. 52 0. 88 -0. 12 -0. 40 -0. 52 -0. 61 3. 20 -1. 59 4. 78 -6. 72 1. 08 C o rr ela t io n (r ) 0. 98 0. 94 0. 92 0. 97 0. 93 0. 80 0. 65 0. 80 0. 88 0. 85 0. 97 0. 95 0. 50 0. 48 De m i ng I n t erc ep t 0. 62 -0. 01 -0. 03 -0. 22 -0. 71 0. 72 -0. 20 0. 60 1. 62 1. 25 5. 46 3. 45 -25. 01 -0. 25 De m i ng Sl o p e 0. 76 0. 89 0. 98 0. 80 0. 89 0. 68 2. 72 0. 91 0. 66 0. 75 0. 92 0. 82 2. 06 0. 71 PM 2. 5 S u l fate Am m o ni um N i tr at e (w i t h ba c kup* ) N i tr at e (n o ba c kup) OC EC TC Br om i n e C a lciu m P o t a ssi um I r on S ili co n T i t a ni um 182 Table 3a. January average concentrations and uncertainties for PM 2.5 , sulfate, ammonium, nitrate, OC, EC, TC, bromine, calcium, potassium, iron, silicon and titanium measured with the STNRS and DRIF. The ?2-? uncertainty is just the average of all uncertainties for the month. Deming slope, intercept, correlation coefficient, monthly average difference and RMS difference for species measured with STN R and DRI F in January and STN S and DRI F in July are presented. Slopes and intercepts were calculated with the y-axis = DRI F and the x- axis = STN RS . Bromine, calcium, potassium, iron, silicon and titanium are reported in units of ng/m 3 and shaded in grey. All other species are reported in units of ?g/m 3 . *Only DRI F collected nitrate with a front and backup filter. 183 DRI F spe c i e s ave r age (? 2 ? u n cert a i n t y) 24. 4 ? 1 . 29 8. 44 ? 0 . 43 2. 93 ? 0 . 24 0. 54 ? 0 . 05 0. 03 ? 0 . 04 6. 33 ? 0 . 63 0. 98 ? 0 . 33 7. 32 ? 0 . 72 4. 00 ? 0 . 52 41. 1 ? 3 . 79 128 ? 7 . 64 83. 2 ? 5 . 85 157 ? 1 7 . 0 4. 95 ? 18 . 6 DRI F S i gn al -to- MD L 28. 3 992 344 63. 5 3. 5 186 29. 0 215 23. 5 54. 1 168. 9 489. 3 104. 6 9. 7 ST N S S i gn al -to- MD L 37. 62 810. 10 153. 51 68. 89 68. 89 29. 05 1. 98 31. 03 6. 63 7. 57 18. 87 43. 35 17. 65 1. 54 ST N S spe c i e s ave r age (? 2 ? u n cert a i n t y) 27. 8 ? 2 . 82 9. 72 ? 1 . 43 2. 61 ? 0 . 36 0. 60 ? 0 . 17 0. 60 ? 0 . 17 6. 97 ? 1 . 35 0. 47 ? 0 . 48 7. 45 ? 1 . 45 3. 58 ? 1 . 45 56. 1 ? 8 . 39 135 ? 1 2 . 3 91. 0 ? 15 . 7 175 ? 2 6 . 7 7. 68 ? 3 . 21 RM S D i ffe r e n c e 5. 59 2. 28 0. 73 0. 25 0. 63 1. 14 0. 62 1. 03 1. 15 26. 73 54. 82 44. 62 148. 55 8. 68 A v era g e D i ffe r e n c e (S TN S - DRI F ) 3. 75 1. 29 -0. 32 0. 06 0. 57 0. 64 -0. 51 0. 13 -0. 42 14. 95 7. 51 7. 85 19. 61 2. 73 C o rr ela t io n (r ) 0. 96 0. 97 0. 95 0. 54 0. 13 0. 99 0. 58 0. 98 0. 88 0. 89 0. 94 0. 88 0. 88 0. 58 De m i ng I n t erc ep t -0. 50 -0. 20 0. 10 0. 17 0. 03 -0. 14 0. 03 0. 12 1. 06 -2. 92 3. 00 -6. 39 9. 51 1. 47 De m i ng Sl o p e 0. 88 0. 89 1. 08 0. 62 0. 01 0. 93 2. 02 0. 97 0. 82 0. 79 0. 92 0. 98 0. 84 0. 45 PM 2. 5 S u l fate Am m o ni um N i tr at e (w i t h ba c kup* ) N i tr at e (n o ba c kup) OC EC TC Br om i n e C a lciu m P o t a ssi um I r on S ili co n T i t a ni um Table 3 b Same as Table 3 a, but for July. 184 6.2.2 Gravimetric Mass Comparisons Comparisons of daily STN R and STN S PM 2.5 with DRI F PM 2.5 are shown in Figure 2 and their error bars (representing the ? 2-? uncertainty) overlap only part of the time. Table 3 shows the Deming slope and intercept, which reduces variance in both independent (x) and dependant (y) variables (Cornbleet and Gochman, 1979), as well as the correlation coefficient, monthly average difference and monthly RMS difference between the two pairs of measurements. Good correlations (r ~ 0.95) are found between STN R and DRI F and between STN S and DRI F with respect to PM 2.5 mass, though both the STN R and STN S measurements are generally larger than the DRI F measurements. The percentage differences ([STN RS -DRI F ] / [STN RS +DRI F ]/2 ? 100) ranged from 8 to 31% between daily PM 2.5 from STN R and DRI F and from -38 to 67% between STN S and DRI F . To determine whether the daily differences were statistically significant I calculated the z-test values for each day using the standard formula (Wilks, 1995): z = {(xbar 1 - xbar 2 ) - E[xbar 1 - xbar 2 ]} / (s 1 2 /n 1 + s 2 2 /n 2 ) 1/2 (3) Here xbar 1 and xbar 2 are the individual measurement of PM 2.5 from STN RS and DRI F , respectively. The s 1(2) represents the STN RS (DRI F ) ?1-? uncertainty value for the specified day. It is assumed that n = 1 and the expected value of the difference between xbar 1 and xbar 2 , i.e., E[xbar 1 ? xbar 2 ], is zero. A z-value less than 1.96 indicates the two measurements are significantly different at the 95% confidence level. Table 4 shows the percentage of days when the paired measurements were significantly different under this test. In January 62% of the daily measurements of 185 PM 2.5 were significantly different, and in July this percentage was lowered slightly to 50%. Percentage of significantly different values January Percentage of significantly different values July PM 2.5 62% 50% Nitrate 100% 0% Sulfate 15% 33% Ammonium 15% 38% OC 36% 8% EC NA NA TC 69% 8% Bromine 0% 5% Calcium NA 65% Potassium 0% 26% Iron 15% 29% Silicon 29% 30% Titanium NA NA Table 4. Percentage of days when the species measured with STN RS and DRI F were significantly different at the 95% confidence level. Only species with concentrations greater than three times the MDL were compared. Comparisons could not be made for EC, calcium (January), nitrate (July) or titanium because over half of the measurements were too small. 186 Figure 2. Time series of PM 2.5 concentrations measured with STN RS and DRI F for January (a) and July (b). Error bars represent ?2-? uncertainty. Watson and Chow (2002) compared mass concentrations obtained with the STN R and DRI F (both analyses were performed at DRI) and found similar results. January 2002 0 5 10 15 20 1/7 1/10 1/13 1/16 1/19 1/22 1/25 1/28 1/31 2/3 Date PM 2.5 m ass ( ? g/ m 3 ) STNR DRIF July 2002 0 10 20 30 40 50 60 6/29 7/2 7/5 7/8 7/11 7/14 7/17 7/20 7/23 7/26 7/29 Date PM 2.5 m ass ( ? g/ m 3 ) STNS DRIF (a) (b) 187 They attribute the discrepancies between the DRI F and the STN R to different instrument inlet designs, flow controls, and resulting cyclone cutoff efficiencies. As discussed in the experimental section above, large particle intrusion is not expected to be a major issue at FME despite the uncertainty in the flow and size cut. Other reasons for the inter-sampler discrepancies include differences in face velocity which may result in losses of volatile material. For submicrometer particles, the overall filter collection efficiency decreases with increasing face velocity (Liu et al. 1983; Lippmann 1995; McDow and Hutzicker, 1990). The overall efficiency of membrane filters, however, is close to 100% for particles larger than the pore size (Lippmann 1995), which is ~0.2 ?m in this study. The TEOM data are available for half of July 2002, and comparisons were made between it and the DRI F and STN S . The DRI F and STN S versus TEOM have r- values of 0.95 and slopes within 11% of unity (Table 5). These results agree with prior studies (Chen et al., 2003; Rees et al., 2004; Lee et al., 2005a; Lee et al., 2005b). The RMS difference is greater for STN S -TEOM than DRI F -TEOM. The STN S - TEOM average difference is positive and about half of the RMS difference, while the DRI F -TEOM average difference is slightly negative and about 1/8 of the RMS difference (Table 5). The magnitude of these differences indicates a systematic bias (in addition to random noise) between the STN S and TEOM measurements. In contrast, deviations between the DRI F and TEOM appear to be random in nature (Figure 3a) and generally fall within 10% of the Deming regression line. Chen (2002) and Chen et al. (2003) found similar results when comparing the DRI F to the TEOM in summer months from 1999-2001. The addition of the 3.0 ?g/m 3 offset added to 188 TEOM measurements may not fully compensate for volatile losses from the heated inlet. Figure 3. Comparisons of PM 2.5 total mass between TEOM and (a) DRI F and (b) STN S . Deming regression line shown in black, ?10% (of the regression line) shown in broken grey. The TEOM and DRI F generally agree within experimental error. TEOM and STN S 0 5 10 15 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 40 45 50 55 PM 2.5 (?g/m 3 ) STN S PM 2. 5 ( ? g/ m 3 ) T E OM y = 0.97x -2.43 r = .95 TEOM and DRI F 0 5 10 15 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 40 45 50 55 PM 2.5 (?g/m 3 ) DRI F PM 2. 5 ( ? g/ m 3 ) T E O M y = 1.11x -2.24 r = .95 (a) (b) 189 M o nt hl y Averag e y 21. 10 21. 10 M o nt hl y Averag e x 24. 06 20. 62 RM S D i ffe r e n c e 5. 35 4. 28 Averag e D i ffe r e n c e (x -y ) 2. 96 - 0 . 48 N 16 16 Co rrel a t i o n (r ) 0. 95 0. 95 I n te r c e p t - 2 . 64 - 2 . 24 Sl o p e 0. 97 1. 11 y TE O M TE O M x ST N S DRI F Table 5. Deming slope, intercept, correlation, and average and RMS difference (?g/m 3 ) for the STN S versus TEOM, and the DRI F versus TEOM as well as N, number of days comparisons were made. The averages (?g/m 3 ) for each sampler for the 2 nd half of July are also given. 6.2.3 Chemical Compositions Besides gravimetric mass, Tables 3 and 4 include the statistics and comparisons of major contributing species to PM 2.5 including sulfate, ammonium, 190 nitrate, OC, EC, TC, bromine and potassium as well as crustal mass made of calcium, iron, silicon and titanium. In January, 15% of the paired sulfate measurements were found to be significantly different, but in July this fraction increased to 33%. Although sulfate measurements from the different instruments are well correlated with r-values greater than 0.94, the STN RS consistently report higher values than the DRI F . Since the average deviation is 14 to 17% for both PM 2.5 and sulfate (Table 3), there appears to be a bias in the flow control, allowing more or less sample volume than specified. It should be noted that sulfate concentration is not sensitive to a small difference in the size cut. Chen (2002) show that sulfate mass from DRI F increases by 4% when filters are exposed for 72 hours after sampling while total mass may either increase (by 1%) or decrease (by 3%). This suggests that the different filter exposure times may have minimal effects on the differences between DRI F and STN RS for sulfate and mass. DRI F and STN RS measure nitrate on different filter substrates behind different denuder configurations (Figure 1). Comparisons between the front only DRI F filters and front plus backup DRI F filters with STN RS have both been made. The nitrate concentrations are well correlated in the winter (without or with backup filter concentrations added), although DRI F measures only 3 to 65% of the average STN R nitrate (without or with backup filter concentration added; see Table 3). All differences were found statistically significant (Table 4). The nylon filters used by STN R appear to retain much more nitrate than single quartz-fiber filters. Moreover, the DRI F filters remained in the field for up to 2.5 days longer, and this led to more nitrate loss through volatilization. The DRI F July average nitrate (on the front filter) 191 is below its 2-? uncertainty and most of the nitrate (above the 2-? uncertainty) was found on the backup filter. The July measurements of nitrate do not correlate well (r = 0.13 front filter only, r = 0.54 front and backup filter), and the DRI F nitrate accounts for 6 to 90% of the STN S (without or with backup filters added). When the DRI F front and backup nitrate are compared with STN S , there are no significant differences (Table 4). Ammonium shows good inter-sampler correlation with r-values greater than 0.92, and significant differences in 15 to 38% of the daily measurements in January and July. In January the average difference as well as the RMS difference between the DRI F and the STN R measured ammonium is negligible. In July the DRI F monthly average is slightly greater than the STN S average, but within 11% (Table 3). Like nitrate, ammonium can also be volatilized readily (Appel and Tokiwa, 1981; Appel et al., 1984; Chow et al., 2005b; Pathak et al., 2004). Pathak et al., 2004 found that there were substantially less losses of ammonium than nitrate on filter samplers possibly resulting from chemical reactions on the filter. For total carbon (TC) that is independent of thermal/optical method, the STN S concentration is similar to that of the DRI F . The STN R concentration is less than DRI F , but within 20%. Inter-sampler differences of TC were significant 8% of the time in July and 69% in January (Table 4). Correlation between the DRI F and STN S is good in July with an r-value of 0.98, much better than the r-value of 0.80 between the DRI F and STN R in January. Since the TC concentration was low in January (<1/3 of that in July) and close to the MDL, more scatter could be expected. The OC/EC ratio was 5.4 in -January, compared with 14.8 in July (based on STN RS ). This reflects 192 larger secondary organic aerosol contributions in the summer (Polidori et al., 2006). OC correlation was similar to that of TC with an r-value of 0.99 in July and an r-value of 0.80 in January. OC is the dominant fraction of TC in both seasons and this explains the similar relationship. EC correlation is poor between the paired measurements both in winter and summer and the STN RS EC are generally only ~50% of the DRI F EC, likely because of the different ways STN-TOT and IMPROVE-TOR define EC (Chow et al., 1993b; Peterson and Richards, 2002; OC/EC Laboratory, 2003). STN RS EC concentrations were generally less than 3 times the MDL and for this reason the z-test comparison was not performed. McDow and Hutzicker (1990) demonstrate that increases in face velocity increase volatilization of organic species. The DRI F and STN RS all use 47-mm filters. Assuming that the filter holder has negligible effects on the area of the filter impacted by the flow, the face velocity can be approximated by the flow rates such that the DRI F has the largest face velocity (with a flow rate of 20 L min -1 ) for OC collection, followed by STN R and STN S (with flow rates of ~ 7 L min -1 ). In July the average DRI F OC and TC are smaller than the STN S , and these differences may be partly attributed to the effects of face velocity. The higher temperatures in July might facilitate OC volatilization, especially from the DRI F filters that were left in the field for a longer time period. However, in January the DRI F TC is larger than the STN R . This is explained neither by flow control differences nor by face velocity. A problem specific to the TC and OC measurement is the blank correction and the only field blank collected for the STN R sampler showed relatively high OC. The STN R field blank OC was on average 50% of the non-blank corrected OC, while the STN S and 193 DRI F field blank OC was on average 20% of the non-blank corrected OC. The winter STN R TC and OC might have been overcorrected. The quantification of OC mass might also be affected by different thermal analysis protocols that define the OC and EC split differently. Inter-sampler comparisons of crustal species, including silicon (in July), calcium and iron, as well as trace elemental species that are > 3 times the MDL (bromine and potassium) all have r-values greater than 0.85. STN S generally reports larger crustal species concentrations than DRI F does, consistent with the situation for PM 2.5 mass and sulfate. The smaller DRI F concentration could be reflected by either a small DRI F /STN S slope (< 1) or a negative intercept (Table 3). STN RS and DRI F differences for silicon, calcium, iron and potassium concentrations were significant 0 to 30% of the time in January and 25 to 65% of the time in July. Calcium (in January), and Titanium, were below three times the MDL and thus the z-test was not performed for these species. 6.2.4 Mass Closure Reconstructed mass from the sum of individual species determines the degree to which the gravimetrically measured total mass is explained by the measured species (Chow et al., 1996; Andrews et al., 2000; Malm et al., 2005; Frank et al., 2006). To reconstruct the PM 2.5 mass, the crustal mass, organic mass and mass of all other species are added together. The crustal mass is the sum of silicon, calcium, iron and titanium multiplied by factors to account for oxygen associated with them (Frank, 2006) as shown below: Crustal mass = 3.73 ? silicon + 1.63 ? calcium + 2.42 ? iron + 1.94 ? titanium (4) 194 There is much debate over what factor should be used to determine the oxygen, nitrogen and hydrogen associated with organic carbon, and this factor can range from 1.2 to 2.5 (Turpin and Lim 2001; Rees et al., 2004; El-Zanan et al., 2005). We multiply the organic carbon by a factor of 1.8, similar to Rees et al. (2004), because the area is highly influenced by regional sources. Front and backup filter nitrate are included in the DRI F reconstructed mass. The reconstructed mass from the DRI F samplers is well correlated with the measured gravimetric mass in both January and July (r = 0.94 ? 0.99, see Table 6), and a good correlation is also found for STN S . The July DRI F reconstructed PM 2.5 mass overestimates the gravimetric mass by 6% while the STN S reconstructed mass underestimates the gravimetric mass by just 3%. For STN R in January, the average measured and reconstructed mass differ by less 2%, although their correlation is not as good (r = 0.80). Histograms of the difference between the gravimetric and reconstructed masses (i.e., the residuals) are shown in Figure 4. In January, the DRI F residuals are shifted negatively from the normal distribution, with a mode at -1 ?g m - 3 . The STN R residuals have a mode at zero and an apparent outlier, which explains the poorer correlation. There is better overlap between the DRI F and STN S residuals in July, but the DRI F residuals are still less than STN S residuals. 195 Correl at i o n (r ) 0. 94 0. 80 0. 99 0. 98 In t ercep t 0. 38 0. 73 1. 5 - 0 . 37 Slope 1. 2 0. 93 0. 99 0. 99 Average d i f f eren ce (g r a v i me tr ic - recon s t r u c t e d ) -1 .5 - 0 . 12 -1 .4 0. 57 RMS d i f f eren ce 1. 7 1. 7 2. 1 3. 2 Average recon s t r u c t e d m a ss 8. 8 8. 9 25. 5 27. 3 Average gravim e t r ic m a ss 7. 3 8. 8 24. 1 27. 8 J a nuar y DRI F STN R Ju l y DRI F STN S Table 6. Average reconstructed mass for STN RS and DRI F for January and July (units are in ?g/m 3 ). Also shown is the Deming slope, intercept, and correlation for the gravimetric (x-axis) and reconstructed mass (y-axis). The DRI reconstructed mass is generally larger than the gravimetric mass and the STN reconstructed mass is generally smaller than the gravimetric mass. 196 Figure 4. Frequency distribution of gravimetric ? reconstructed differences (residuals), for January DRI F and STN R and July DRI F and STN S . January 0 2 4 6 8 10 12 14 -5 -4 -3 -2 -1 0 1 2 3 4 5 Gravimetric - reconstructed PM 2.5 mass (?g/m 3 ) Fr e q u e nc y DRI STN July 2002 0 2 4 6 8 10 12 14 -6 -4.5 -3 -1.5 0 1.5 3 4.5 6 7.5 Gravimetric - reconstructed PM 2.5 mass (?g/m 3 ) Fr e q u e nc y DRI STN (a) (b) 197 Figure 5. Contributions of individual species to PM 2.5 mass (relative contribution) for (a) January and (b) July. Numbers in boxes are the DRI F relative contribution divided by STN RS relative contribution. Error bars represent the standard deviation of the relative contributions. Figure 5 shows the contributions of sulfate, organic matter (OM = OC ? 1.8), EC, ammonium, nitrate, crustal mass and the sum of all other species, to total mass (the relative contribution) as well as the ratios of DRI F /STN RS relative contribution. January 0 10 20 30 40 50 60 70 80 July 0 10 20 30 40 50 60 70 80 Sulfate OM EC Ammonium Nitrate Crustal mass Others DRI F STN R DRI F STN S 1.04 1.31 2.52 1.13 0.77 1.25 2.87 1.00 1.10 2.34 1.31 0.92 1.08 0.91 R e l a t i v e c ont r i bu t i on t o m a s s ( % ) (a) (b) Sulfate OM EC Ammonium Nitrate Crustal mass Others 198 Here nitrate from the front and backup filter of DRI F was used. In January and July STN RS report larger sulfate concentrations, but the relative contribution of sulfate to total mass is similar for STN RS and DRI F (shown by the ratios of relative contribution [DRI F /STN RS ] being close to unity in Figure 5). A systematic bias could explain why the difference between the sulfate concentrations does not show up in the relative contributions. This bias can result from differences in how the two instruments record volume as described in the experimental section. In January DRI F reports more OM concentration than STN R and the relative contribution of OM to total mass from DRI F is greater than that from STN R . In July DRI F reports less OM concentration than STN S and the relative contribution of OM to total mass from DRI F is greater than that from STN S . This should not negate the above argument that there is a systematic bias between the two instruments. The relative contribution of OM to total mass is affected by artifacts in both mass and OC measurements. The differences in OM relative contribution are not the same as the differences in sulfate relative contribution because of issues related to organic sampling artifacts, blank correction and analysis protocols. The mass closure of DRI F usually exceeds 100%, consistent with an uncorrected positive organic sampling artifact. For STN R , however, the problem associated with organic sampling artifacts has been offset by a relatively high blank subtraction in this study. The organic sampling artifact is a major issue regarding PM 2.5 mass closure, particularly for low PM-loaded samples. 6.3 Conclusions Measurements from the DRI and RTI analyzed samplers (DRI F versus STN R and DRI F versus STN S ) at Fort Meade, MD were generally well correlated. 199 ? PM 2.5 , sulfate, OC, TC and ammonium all had r-values in excess of 0.8. ? The STN method reported larger PM 2.5 mass than the DRI method by 14 ? 17% and generally showed larger concentrations than the DRI F . ? Possible causes for the bias between STN RS and DRI F include different flow monitoring strategies, DRI F losses of volatile species because used filters remained in the field for a longer time and/or because face velocities were larger than those for the STN RS . ? With the current state of ambient monitoring it is reasonable to expect uncertainties of at least 20% (at the 95% confidence level) for PM 2.5 , sulfate, ammonium, and organic matter. Even though the PM 2.5 mass measurements were well correlated, differences between the measurements were statistically significant more than 50% of the time under the current uncertainty estimates. The uncertainty associated with PM 2.5 mass must be raised from 10% to 20% for January measurements, and from 10% to 28% for July measurements, to make the differences statistically significant only 5% of the time (using a z-test and assuming only random errors). Even though the measurements of speciated mass were well correlated, the differences between the samplers are statistically significant at the 95% confidence level from 5 to 100% of the time. Particularly, measurements of EC did not compare well. Two different analysis methods, IMPROVE-TOR and STN-TOT, were used, and these two methods are known to define EC differently. Nitrate correlated well between the two samplers in January, however the DRI F measurements were substantially smaller than those 200 from the STN R and all the measurements were significantly different using a z-test. In July the nitrate correlation was weaker, possibly because of the increased volatility and lower concentration of the nitrate aerosol. It is likely that the STN RS nylon filters retained more nitrate than the DRI F quartz filters (e.g. Frank 2006). At FME this problem was mitigated somewhat because DRI F used backup filters. Residuals of gravimetric ? reconstructed mass were generally small and negative for both DRI F and STN RS . The differences possibly result from the organic sampling artifact and/or conversion factor between the mass of organic carbon and organic matter. Overall, the error estimates used in the current STN network (i.e., from AQS) may be too low to account for the actual uncertainty in the measurements, and to some extent this may impact the conclusions of trend analyses and receptor modeling based on the STN data. With the current state of ambient monitoring it is reasonable to expect uncertainties of at least 20% (at the 95% confidence level) for PM 2.5 , sulfate, ammonium, and organic matter and larger uncertainties for EC and nitrate. Further evaluation for these sampling systems is recommended through side-by-side measurements at multiple locations for longer periods of time. 201 Chapter 7: Conclusions 7.1 Summary In Chapters 3-6 I presented work from clustering back trajectories and profiles of trace gases and aerosols, comparisons of model and measured profiles of trace gases and surface comparisons of aerosols to explain sources, sinks and distributions of aerosols and trace gases in the Mid-Atlantic region. From 1995 - 2002, airborne measurements of O 3 , CO, SO 2 , and aerosol properties were made during summertime air pollution episodes over the Mid-Atlantic U.S. (34.7? to 44.6?N, 68.4? to 81.6?W) as part of the Regional Atmospheric Measurement, Modeling, and Prediction Program (RAMMPP). In Chapter 3, I presented statistics for all profiles made. Little diurnal variation was identified in the CO, SO 2 , and ?ngstr?m exponent profiles, although the ?ngstr?m exponent profiles decreased with altitude. Boundary layer O 3 was greater in the afternoon, while lower free tropospheric O 3 was invariant at ~55 ppbv. The single scattering albedo increased from morning to afternoon (0.93 + 0.01 - 0.94 + 0.01); however, both profiles decreased with altitude. A cluster analysis of back trajectories in conjunction with the vertical profile data was used to identify source regions and characteristic transport patterns during summertime pollution episodes. When the greatest trajectory density lay over the northern Ohio River Valley, the result was large O 3 values, large SO 2 /CO ratios, highly scattering particles, and large aerosol optical depths. Maximum trajectory density over the southern Ohio River Valley resulted in little pollution. The greatest afternoon O 3 values occurred during periods of stagnation. North-northwesterly and northerly flow brought the least pollution overall. The contribution of regional transport to 202 afternoon boundary layer O 3 was quantified. When the greatest cluster trajectory density lay over the Ohio River Valley (~59% of the profiles), transport accounted for 69-82% of the afternoon boundary layer O 3 . Under stagnant conditions (~27% of the profiles), transport only accounted for 58% of the afternoon boundary layer O 3 . On average transported O 3 accounts for 64% of the O 3 measured in the aircraft profiles (this is a weighted averaged shown in Table 1). This transported O 3 may be an underestimate because we were unable to account for O 3 precursors produced by upwind sources. The results from this study provide a description of regional chemical and transport processes that will be valuable to investigators from the Baltimore, New York, and Pittsburgh EPA Supersites. Cluster % of flights made for this cluster % O 3 transported weighted % O 3 transported 1 26 67 18 2 19 67 13 3 27 54 14 4 10 82 8 5 6 62 4 6 4 73 3 7 6 56 3 8 3 55 2 weighted average 64 Table 1. The percent of O 3 transported for each back trajectory cluster and the weighted average of O 3 transported from upwind sources for all clusters. Upwind emission sources of NO x and SO 2 play a crucial role in the amount of O 3 and aerosols in the lower troposphere in the Mid-Atlantic region. In Chapter 4 a hierarchical clustering method was used to separate distinct chemical and meteorological events from over 150 aircraft vertical profiles in the lower troposphere 203 measuring O 3 , SO 2 , CO, and particle absorption and scattering in the Mid-Atlantic US. Forty-eight-hour back trajectories were run for each profile and the integrated NO x and SO 2 point source emissions encountered by each trajectory were calculated using data from the EPA Clean Air Market Division?s database. Greater integrated point source NO x emissions along the back trajectories were correlated with greater O 3 mixing ratios measured during the flights, indicating that O 3 mixing ratios are strongly influenced by and can be predicted with point source emissions. The amount of CO observed depended on where the profiles were made, and larger CO values were found in areas with larger mobile source emissions. Profiles with greater particle absorption were associated with greater CO values. There is a pervasive ?background? SO 2 profile over the eastern US with mixing ratios decreasing smoothly from about 3.5 ppb near the surface to 0.2 ppb at 2400 m. Most SO 2 measured fit this clean profile, but there were exceptions and the clustering method was able to separate these profiles with larger SO 2 values. Profiles with larger, more scattering particles, were correlated with greater integrated SO 2 emissions. The clustering technique also separated profiles made during the 2002 Canadian forest fires. The UMD aircraft measurements of O 3 have also been compared with EPA?s Community Multiscale Air Quality (CMAQ) model. CMAQ under-predicts O 3 by 10% above 500 m altitude. I performed a sensitivity test of the model to determine how including aerosols with NO 2 photolysis rate coefficients affected O 3 production using a revised CMAQ run. These adjustments of the chemistry had modest impacts on CMAQ calculated profiles. In general the revised CMAQ run generated more O 3 204 above 500 m (~1 ppb), and generated less O 3 (1-4 ppb) below 500 m and brought them into closer agreement with observations. Improvements in the model?s ability to describe clouds might increase the oxidation of SO 2 to sulfate and thereby bring the modeled O 3 in closer agreement with measurements. The UMD aircraft SO 2 measurements were also compared with CMAQ and GOCART. Both models over-predicted SO 2 aloft by ~50%. Possible reasons for this include problems with the emissions inputs and the difficulty the models have resolving clouds. Because the models over-predict SO 2 , they likely over-predict the lifetime of SO 2 . This has far-reaching policy implications on the ability of the models to describe the oxidation product of SO 2 (sulfate) and the ability of the models to describe PM 2.5 accurately. Some locations in the Mid-Atlantic are not in compliance with PM 2.5 standards, and improvement of the models ability to replicate the oxidation of sulfate will aid in the development of state implementation plans for the reduction of PM 2.5 . Assessing the effects of air quality on public health and the environment requires reliable measurement of PM 2.5 mass and the individual chemical components of fine aerosols. In Chapter 6 PM 2.5 measurements that are part of a newly- established national network were compared with more conventional sampling systems. Experiments were carried out during 2002 at a suburban site in Maryland, United States, where two samplers from the U.S. Environmental Protection Agency Speciation Trends Network: Met One Speciation Air Sampling System ? STN S and Thermo Scientific Reference Ambient Air Sampler ? STN R , two Desert Research Institute Sequential Filter Samplers ? DRI F , and a continuous TEOM monitor 205 (Thermo Scientific Tapered Element Oscillating Microbalance, 1400a) were sampling air in parallel. These monitors differ not only in sampling configuration but also in protocol-specific sample analysis procedures. Measurements of PM 2.5 mass and major contributing species were well correlated among the different methods with r- values > 0.8. Despite the good correlations, daily concentrations of PM 2.5 mass and major contributing species were significantly different at the 95% confidence level from 5 to 100% of the time. Larger values of PM 2.5 mass and individual species were generally reported from STN R and STN S . The January STN R average PM 2.5 mass (8.8 ?g m -3 ) was 1.5 ?g m -3 larger than the DRI F average mass. The July STN S average PM 2.5 mass (27.8 ?g m -3 ) was 3.8 ?g m -3 larger than the DRI F average mass. These differences can only be partially accounted for by known random errors. Variations in flow control, face velocity, and sampling artifacts possibly influence the measurement of PM 2.5 speciation and mass closure. Statistical tests indicate that the current uncertainty estimates used in the STN network may underestimate the actual uncertainty. 7.2 Recommendations for Future Work The chemical climatology has been used to evaluate modeled O 3 and SO 2 . In Chapter 5, I showed that CMAQ modeled O 3 responds to radiative changes due to aerosols and so CMAQ would likely also respond to changes in clouds. A rigorous analysis of how well MM5 and CMAQ represent cloud cover should be performed. Model improvements would be useful. A determination of how changes in cloud cover affect O 3 and aerosol production would be enlightening. A detailed comparison 206 of measured aerosols with those generated by CMAQ would also be useful for improving forecasting of PM 2.5 events. Appendix A presents comparisons between aircraft and surface measurements of trace gases made for morning and afternoon flights carried out over Ft. Meade, Maryland (1999-2002). Morning aircraft measurements were averaged from 100 ? 500 m and afternoon measurements were averaged from 100 ? 2000 m. O 3 measurements compared better in the afternoon, likely because O 3 is better mixed in the atmosphere later in the day. CO and SO 2 measurements compared better in the morning. They both have peaks below 500m which is consistent with the expectation of CO coming from ground level combustions and SO 2 emissions from point sources. Extension of this work to all EPA surface sites near aircraft profiles may prove interesting. Satellites can be powerful tools to monitor the movement of atmospheric pollutants and may have future uses in the prediction of pollution events. Appendix B shows results from comparisons of the Global Ozone Monitoring Experiment (GOME) satellite with UMD aircraft profiles of SO 2 . Because of the coarse resolution of GOME and high level of noise, the comparison was poor. This provides an understanding of the limitations of satellite measurements of SO 2 . The chemical climatology presented here can be used for validation and improvement of other satellite measurements. 207 Appendix A Surface and Aircraft Measurements Understanding sources of pollution can aid in prevention of pollution events. The transport of pollutants can often been seen in vertical profiles (made with aircraft) as described in Chapter 3. Vertical profiles are expensive and are limited in space and time. Surface networks measuring trace gases and aerosols on a continuous or near-continuous basis have been set up by the EPA throughout the US to monitor pollution levels. I compared surface measurements of O 3 , SO 2 , and CO with average columns measured aboard the UMD research aircraft to assess how well surface measurements represent the mixed layer and how they might be influenced by transported pollutants. During the 1999 ?2002 intensive sampling period at Fort Meade, the University of Maryland research aircraft made flights over Fort Meade measuring O 3 , SO 2 , and CO. The shapes of O 3 profiles are affected by the breakdown of the nocturnal boundary layer and to account for this I divided the flights into morning and afternoon. Morning flights were flown between 6:00 and 12:00 EST, with an average time of 9:30 EST. Afternoon flights were flown between 12:00 and 19:00 EST, with an average time of 14:30 EST. In order to compare the aircraft measurements with surface measurements I assumed the afternoon boundary layer extended from 100 m to 2000 m and I calculated a boundary layer average for all of the trace gases in this layer. For morning flights I assumed that the residual nocturnal boundary layer was 208 between 500 m and 2000 m, and I calculated an average below that, from 100 - 500 m, to compare with surface measurements. I used a three-hour average of the surface measurements around the time of the flight to compare with aircraft measurements. Comparisons between surface and aircraft measurements of O 3 , SO 2 , and CO, for morning and afternoon flights, are shown in A.1-A.3. O 3 measurements compare better in the afternoon (r 2 = 0.6) than in the morning (r 2 = 0.5). The average difference (surface ? aircraft) is smaller in the afternoon (8 ppb) than in the morning (9 ppb) and the RMS difference is also smaller in the afternoon than in the morning (A4). In the afternoon the profile is generally well mixed and this explains the better correlation and smaller differences between surface and aircraft measurements in the afternoon. SO 2 measurements compared better in the morning, when the comparison was made between the surface and the aircraft 100 ? 500 m average, than in the afternoon when the comparison was made between the surface and the aircraft 100 - 2000 m average. Morning SO 2 comparisons had an r 2 of 0.8 and this dropped to 0.5 in the afternoon. The average difference (surface ? aircraft) increased from 0.4 to 1.6 ppb between morning and afternoon, though the RMS difference was similar (3 ppb). The SO 2 generally peaks below 500 m (at elevations where it is emitted) and concentrations drop off substantially above this level, so the average SO 2 from 100 m to 2000 m is smaller than the average SO 2 from 100 to 500 m. Afternoon surface SO 2 was compared with aircraft average SO 2 from 100 ? 500 m (A.2.c) and the r 2 of 0.7 was better than that from A.2.b which had an r 2 of 0.5. 209 A.1. Comparison of aircraft and surface measurements of O 3 for a) morning and b) afternoon flights. The afternoon shows better correlation between surface and measurements aloft, likely due to improved mixing in the afternoon. Morning y = 1.10x + 4.39 R 2 = 0.46 0 20 40 60 80 100 120 2040608010120 Aircraft O 3 ppb (100m -500m average) S u r f ac e O 3 pp b av er age ( 3 h o u r av er age) Afternoon y = 0.97x + 9.98 R 2 = 0.59 0 20 40 60 80 100 120 140 0 2040608010120140 Aircraft O3 ppb (100m - 2000m average) S u r f ac e O 3 p p b av er ag e ( 3 h o u r av er ag e) a b 210 Morning y = 1.29x - 1.69 R 2 = 0.78 0 5 10 15 20 25 0 5 10 15 20 25 Aircraft SO 2 ppb (100m -500m average) S u r f ace S O 2 ppb aver ag e (3 h o u r aver ag e) Afternoon y = 1.60x - 0.33 R 2 = 0.53 0 5 10 15 20 25 0 5 10 15 20 25 Aircraft SO 2 ppb (100m - 2000m average) S u rfa ce S O 2 ppb av e r a g e ( 3 ho u r aver ag e) Afternoon y = 1.00x - 0.32 R 2 = 0.70 0 5 10 15 20 25 0 5 10 15 20 25 Aircraft SO 2 ppb (100m - 500m average) S u r face S O 2 ppb aver ag e (3 hour av e r age ) a b c 211 A.2. Comparison of aircraft and surface measurements of SO 2 for a) morning and b) afternoon flights (with aircraft averages from 100 -2000 m) and b) afternoon flights (with aircraft averages from 100 ? 500 m). Surface measurements compare well with aircraft averages in the lower boundary layer (100 -500 m). A.3. Comparison of aircraft and surface measurements of CO for a) morning and b) afternoon flights. Both morning and afternoon show poor correlation and this could be because of spikes in surface CO data. Morning y = 0.23x + 175.70 R 2 = 0.12 0 100 200 300 400 500 0 100 200 300 400 500 Aircraft CO ppb (100m -500m average) S u r f a c e C O p p b av er ag e ( 3 h o u r av er ag e ) Afternoon y = 0.41x + 138.75 R 2 = 0.10 0 100 200 300 400 500 600 700 0 100 200 300 400 500 600 700 Aircraft CO ppb (100m - 2000m average) S u r f a c e C O ppb a v e r a g e ( 3 hour a v e r a g e ) a b 212 Aircraft column average (ppb) Aircraft standard deviation (ppb) Surface 3-hr average (ppb) Surface standard deviation (ppb) RMS difference (ppb) Average difference (surface- aircraft) (ppb) O 3 am 47.0 17.8 56.1 29.0 22.6 9.1 O 3 pm 74.6 17.0 82.5 21.6 15.8 8.0 SO 2 am 7.2 4.8 7.7 7.0 3.4 0.4 SO 2 pm 3.2 2.0 4.8 4.4 3.5 1.6 CO am 250 125 233 83 118 -16.2 CO pm 204 95 222 124 127 17.7 A.4. Comparisons of O 3 , SO 2 , and CO aircraft column averages with surface measurements. Morning aircraft measurements were averaged between 100-500 m and afternoon aircraft measurements were averaged between 100-2000 m. The RMS difference, average difference (surface ? aircraft), and standard deviation is also shown. All data have units of ppb. CO morning and afternoon aircraft measurements do not compare well with surface measurements (A.3). This is likely because of some peaks at the surface not seen aloft. When one outlier (June 24, 1999) is removed from the morning data, the correlation between morning surface and aircraft measurements improves from an r 2 of 0.1 to an r 2 of 0.6 (A.5a). The June 24, 1999 spiral was made at 6:00 EST, and shows CO around 100 ppb, from the surface to 3km. The small CO values are seen at nearby locations of Gaithersburg, MD and Manassas, VA. O 3 is also extremely low below 150 m (around 30 ppb for all three locations). The CO surface measurement shows 300-400 ppb from 5:00 to 7:00 EST, with a standard deviation of 64 ppb (for this specific day, this differs somewhat from the standard deviation for all days in the analysis shown in A.4). No peaks like this are seen in the aircraft profile, even below 100 m, suggesting that this is a very local plume (perhaps a vehicle was idling near 213 the instrument). The RH is about 90% at the surface and drops off to 50% at 300 m, suggesting that the lower level aircraft measurements were made in the inversion layer. When an outlier is removed from the afternoon measurements (A.5.b) the correlation improves from an r 2 of 0.1 to an r 2 of 0.5. This outlier occurred on June 24, 2002. The aircraft made a spiral at 15:00 pm EST, and shows 400 ppb of CO at 200 m, which decreases quickly aloft. The surface measurements show a three-hour average CO of 611 ppb, with at standard deviation of 66 ppb. Since this was an afternoon profile, I calculated the average from 100 m to 2000 m, and the peak near the surface was washed out. Though the correlation did improve when the outliers were removed, the correlation was still not as good as that for O 3 and SO 2 . Thus, surface CO measurements may not be representative of the mixed layer measurements. 214 A.5. Comparison of aircraft and surface measurements of CO for a) morning and b) afternoon flights with outliers removed. The correlation improves when the outliers are removed, however the correlation is not as good as that for O 3 and SO 2 . Morning y = 0.47x + 94 R 2 = 0.60 0 100 200 300 400 500 0 100 200 300 400 500 Aircraft CO ppb (100m -500m average) S u r f ac e CO p p b a v e r ag e ( 3 ho ur a v e r a g e ) Afternoon y = 0.43x + 106 R 2 = 0.45 0 100 200 300 400 500 600 700 0 100 200 300 400 500 600 700 Aircraft CO ppb (100m - 2000m average) S u rf ac e C O p p b a v e r ag e ( 3 ho ur a v e r a g e ) a b 215 Appendix B Satellite Measurements Surface measurements can be made continuously to show the diurnal and seasonal variability of SO 2 . A network of surface stations like those in the Sulfate Regional Experiment and the Atmospheric Integrated Research Monitoring Network provide information about regional SO 2 distributions (EPRI, 1981; Hicks, 2001) but this can be expensive and difficult to implement everywhere. Aircraft measurements provide altitude profiles and some spatial information including possible transport of pollutants. However, aircraft observations are also expensive and very limited in space and time. Satellites show SO 2 distributions around the world year round. These measurements are usually taken once daily and thus do not offer information on the diurnal variability of SO 2 ; nor do they provide information about the vertical distribution of SO 2 . Because of their spatial coverage, satellites are great tools to monitor mesoscale and synoptic scale atmospheric events. A combination of surface, aircraft, and satellite measurements can be a powerful tool for describing the SO 2 distribution. Satellites could also be used in conjunction with models to predict and characterize pollution events. Understanding the uncertainty in satellite measurements is a key step in the advancement of satellites into the tropospheric air quality monitoring ensemble. In this Appendix I will present comparisons of SO 2 from the UMD research aircraft with those retrieved from The Global Ozone Monitoring Experiment (GOME) instrument aboard the European Research Satellite (ERS-2). 216 GOME Instrumentation GOME measures scattered and reflected light from the Earth and the atmosphere. Data collected in the wavelength range of 315.5-327 nm are used to determine SO 2 column content with Differential Optical Absorption Spectroscopy (DOAS) (Eisinger et al. 1998). Beer?s law enables the quantification of the concentration of a species from the absorption spectrum using measurements of attenuated and unattenuated light. It is difficult to measure the true unnattenuated light, I o , coming from the Earth because of Mie and Rayleigh scattering, as well as absorption by atmospheric species that attenuate light. B.1 shows the absorption cross section of SO 2 in the wavelength region from 317 to 325. There is more structure in the spectrum in the smaller wavelength region of 318 to 320. DOAS fits a curve to the absorption spectrum in the larger wavelength region (317 to 325 nm) to describe the ?unattenuated? beam I o ?. This ?unattenuated? beam is only unattenuated by SO 2 and accounts for the difficult to measure attenuation from scattering and other atmospheric species (Platt, 1994). The differential absorption, D? is just I o ? ? I. 217 B.1. SO 2 absorption cross section. In the smaller box I o is the true unattenuated beam of light that cannot be measured because of atmospheric scattering and absorption. I o ? is fit over the larger wavelength region (317-325) and is only attenuated by the species of interest. The differential absorption (D?) is then just I o ? ? I (Finlayson-Pitts and Pitts, 2000; Platt, 2003). GOME measures trace gasses with a nadir scanning double monochromator. The resolution of the monochromator is 0.17-0.33 nm. The incoming light is split into 4 channels and is recorded with a 1024 reticon photodiode array. SO 2 absorbs at 317 to 325, and the monochromator channel corresponding to this region detects the gas. The SO 2 integrated column content is then calculated using a differential optical absorption spectrometry algorithm (Eisinger et. al. 1998). There is an overlap in the absorption signal between SO 2 and O 3 from the Huggins bands in the 300-360 nm region (Finlayson-Pitts and Pitts, 2000). This O 3 interference can be removed from the GOME data by subtracting the SO 2 signal in areas with little SO 2 (like areas over the ocean) from the SO 2 signal in an area of interest. 318 nm 320 nm 218 Comparisons between In-situ and Satellite Measurements: Implications for Revisions to the Air Mass Factor. I have compared the aircraft column contents of SO 2 with those measured with GOME to test the sensitivity of GOME SO 2 measurements. B.2 shows a map of GOME SO 2 retrievals over North America. Plumes of SO 2 are visible over Mexico City as well as the Eastern and Midwestern US. Plumes in these areas are most likely located in the troposphere. B.2. Map of GOME SO 2 . The SO 2 plume over the Eastern US and Mexico City can clearly be seen with a column content of about 0.4 DU. B.3 shows comparisons I made between the default GOME retrievals and aircraft column contents. The correlation between GOME and aircraft measurements is poor and GOME retrieves much less SO 2 than the aircraft measures. To improve the GOME retrievals I used in-situ data to modify the Air Mass Factor (AMF) used in 219 the GOME retrieval algorithm. The AMF is used to convert the measured slant columns of trace gases into vertical columns and can be written as: AMF = ? slant /? vert (1) Where ? slant is the optical density along the slant path (this is what is measured by GOME) and ? vert is the optical density along the vertical path. To determine the AMF, a radiative transfer model is run with and without the absorbing species to calculate respective intensites I(?) w and I(?) w/o along the slant path s s . ln (I(?) w/o / I(?) w ) = ss TOA x dssCs sx )(),( 0 ?? ? (2) Here x is some absorbing species, s s is the slant path length, ? x is the extinction cross section, and C x is the concentration of the species of interest. The left side of Equation 2 is the slant optical density. The vertical optical density, ? vert , can be calculated as: vvv TOA xvert dssCs x )(),( 0 ??? ? = (3) Here s v is the vertical path length. The air mass factor can be written as: AMF = ln (I(?) w/o / I(?) w ) / vvv TOA x dssCs x )(),( 0 ?? ? (4) (Perliski et al., 1993). The extinction cross section can be measured in the lab or determined from literature. The radiative transfer model SCIATRAN solves the following equation to determine I(?) w and I(?) w/o . '')',',()',,',,( 4 )( ),,()( ),,( 1 1 2 0 ?????????? ?? ? ? ddzIzp zb zIzc dz zdI ?? ? ? +?= (5) Here ? and ?' denote the cosine of the zenith angle, z represents altitude, ? and ?? represent the azimuthal angles in relation to the line-of-site projection on the earth?s 220 surface, c is the total extinction coefficient, b is the total scattering coefficient (the sum of the trace gas and particle scattering coefficients), and p is the total scattering phase function (Rozonov et al., 1997). -0.5 0 0.5 1 1.5 2 2.5 SO 2 col u m n c o n t ent ( m a t m cm ) Aircraft GOME default June 2000 July 2001 May 2001 June 2002 August 2002 B.3. Comparison between default GOME SO 2 and UMD aircraft SO 2 (matm cm). GOME SO 2 is smaller than that measured aboard the aircraft and the correlation between the measurements is poor (r 2 = 0.20). The AMF depends on the altitude of the absorbing species of interest and where the most absorption and scattering occur in a vertical column. B.4 shows how light is scattered when the absorbing and scattering layers are near the surface (Example 1) and when the absorbing and scattering layers are at higher altitudes (Example 2). Light at 320 nm is mostly attenuated by the time it reaches the surface. The path length is smaller in example 1 than in example 2 and thus the intensity reaching the satellite in example 1 will be greater than in example 2. Because the intensity is greater, the AMF for example 1 will be smaller than that for example 2. The default AMF calculation assumes the column concentration of SO 2 has a peak 221 above the boundary layer. I have recalculated the AMF using SO 2 profiles measured aboard the UMD research aircraft. B.5 shows results of the comparison between aircraft SO 2 columns and GOME SO 2 when a revised AMF was used. These adjustments to the AMF made the GOME retrieved SO 2 larger but did not improve the correlation between the aircraft and GOME. 222 B.4. Light is scattered differently when an absorbing and scattering layer is near the surface (example 1) than when the layer is above the planetary boundary layer (example 2). Red arrows denote light that is not scattered through the absorbing layer and green arrows denote light that is scattered through the absorbing layer. The length and space between the arrows represent the generalized degree of scattering (as altitude increases there are less scattering species). There is more scattering when the layer is above the planetary boundary layer (example 2) than when the layer is near the surface. This is because the light scatters on its way to the surface and then again on its way to the satellite for example 2. Absorbing and scattering layer Scattering layer Sun beam Satellite Example 1 Low level absorbing layer. Example 2 Upper level absorbing layer. Absorbing and scattering layer Scattering layer Sun beam Satellite 223 -0.5 0 0.5 1 1.5 2 2.5 SO 2 co l u m n co nt ent ( m at m cm ) Aircraft GOME revised AMF June 2000 July 2001 May 2001 June 2002 August 2002 B.5. Comparison between GOME SO 2 , using a revised AMF, and UMD aircraft SO 2 (matm cm). The revised AMF made the retrieved GOME SO 2 larger than the default retrieval but the correlation between the measurements is poor (r 2 = 0.16). GOME Interference Corrections Because the SO 2 and O 3 absorption band overlap in the UV, GOME SO 2 retrievals must be corrected for O 3 interference (this process will be referred to as an O 3 correction). The default O 3 correction subtracts SO 2 retrieved columns over the Pacific Ocean (where SO 2 should be small) from the SO 2 column at the point of interest. To account for the latitudinal gradient of O 3 , only the ocean SO 2 at latitudes matching that of the point of interest are used in the correction. For annual averages this correction works well but for daily GOME retrievals this correction sometimes gives negative SO 2 . I have developed a method to improve the O 3 correction and this is described below. The method to improve the O 3 correction involves finding regions over the ocean with O 3 column contents similar to those over the area of interest and is 224 diagrammed in B.6. For this method I first generated a map of O 3 . Because O 3 has a shape different from SO 2 I had to use a different AMF. For O 3 a good approximation of the AMF can be made with the solar zenith angle (SZA): AMF = 1 + 1/cos (SZA) (6) I made a grid of 1 o latitude by 1 o longitude and averaged the O 3 in each grid box. I then searched for an O 3 box over the ocean that was within 5% of the O 3 box over the area of interest (Steps 1 and 2 in B.6). The latitude and longitude of this box over the ocean was saved. Next I made a map of SO 2 using the revised AMF, described in the previous section, and found the SO 2 at the area of interest and the SO 2 at the same location as the ocean box (Steps 3 and 5). To correct for the O 3 interference at an area of interest I subtracted the SO 2 at the ocean box (that had O 3 that matched O 3 over the area of interest) from the SO 2 over the area of interest. B.7 shows a comparison of SO 2 from the aircraft with GOME retrieved SO 2 (using the revised AMF and O 3 correction). The revised O 3 correction did decrease the number of negative values, but the correlation between aircraft and GOME SO 2 is still poor. In Chapter 5 I calculated the average lifetime of SO 2 in the summer in the daytime in the Mid-Altantic to be short, ~ 19 hours (Chapter 5, Table 11). GOME has coarse spatial resolution, and only makes measurements once a day. The short lifetime and the coarse resolution partly explain why the correlation between aircraft and GOME SO 2 was poor. The SO 2 in the Mid-Atlantic may also be below the GOME detection limit. This analysis provides an understanding of the limitations of the GOME SO 2 retrievals. 225 B.6. Diagram of the steps used to calculate the revised O 3 correction. 1.) Find O 3 over area of interest. 2.) Find matching O 3 over ocean (where SO 2 is minimal). 3).Find SO 2 in same location as in 2. 4). Find SO 2 column (over spiral location with no O 3 correction. 5.) Subtract SO 2 in step 4 from SO 2 in step 3. 1 2 4 3 226 -0.5 0 0.5 1 1.5 2 2.5 SO 2 co l u m n co n t en t ( m at m cm ) Aircraft GOME revised AMF and O3 correction June 2000 July 2001 May 2001 June 2002 August 2002 B.7. Comparison between UMD aircraft SO 2 (matm cm) GOME SO 2 , using a revised AMF and revised O 3 corrections. The revised O 3 corrections decreased the number of negative GOME retrievals but the correlation between the measurements is still poor (r 2 = 0.15). 227 Bibliography Andronache, C., W.L. Chameides, D.D. Davis, B.E. Anderson, R.F. Pueschel, A.R. Bandy, D.C.Thornton, R.W. Talbot, P. Kasibhatla, and C.S. Kiang, Gas-to-particle conversion of tropospheric sulfur as estimated from observations in the western North Pacific during PEM-West B. J.Geo. Phys. Res., 102, 28511-28538, 1997. Appel, B.R. and Y.Tokiwa, Atmospheric particulate nitrate sampling errors due to reactions with particulate and gaseous strong acids. Atmos. Environ., 15, 1087-1089, 1981. Appel, B.R., Y. Tokiwa, M. Haik, and E.L. Kothny, Artifact particulate sulfate and nitrate formation on filter media. Atmos. Environ., 18, 409-416, 1984. Anderson, T.L., and J.A. Ogren, Determining aerosol radiative properties using the TSI 3563 Integrating Nephelometer, Aerosol Sci. Technol., 29, 57-69, 1998 Andrews, E., P. Saxena, S. Musarra, L.M. Hildemann, P. Koutrakis, P.H. McMurry, I. Olmes, and W.H. White, Concentration and composition of atmospheric aerosols from the 1995 SEAVS experiment and a review of the closure between chemical and gravimetric measurements. J. Air & Waste Manage. Assoc., 50, 648-664, 2000. Ames R.B. and W.C. Malm, Comparison of sulfate and nitrate particle mass concentrations measured by IMPROVE and the CDN. Atmos. Environ., 35, 905-916, 2001. Ashbaugh, L.L., A statistical trajectory technique for determining air pollution source regions, J. Air Pollut. Control Ass., 33, 1096-1098, 1983. Berglen, T.F., T.K. Berntsen, S.A. Isaksen, and J.K. Sundet, A global modle of the coupled sulfur/oxidant chemistry in the troposphere: The sulfur cycle, J. Geophys. Res., 109, D19310, doi:10.1029/2003JD003948, 2004. Berto, A., A. Buzzi, and D. Zardi, Back-tracking water vapour contributing to a precipitation event over Trentino: a case study, Meteorol. Z., 13(3), 189-200, 2004. Bond, T.C., T.L. Anderson, and D. Campbell, Calibration and intercomparison of filter-based measurements of visible light absorption by aerosols, Aerosol Sci. Technol., 30, 582-600, 1999. Bodhaine, B.A., Aerosol absorption measurements at Barrow, Mauna Loa and the south pole, J. Geophys. Res., 100, 8967-8975, 1995. Brankov, E., S.T. Rao, and P.S. Porter, A trajectory-clustering-correlation methodology for examining the long-range transport of air pollutants, Atmos. Environ., 32(9), 1525-1534, 1998. 228 Buchdahl R, C.D. Willems, M, Vander, and A. Babiker Associations between ambient ozone, hydrocarbons, and childhood wheezy episodes: a prospective observational study in south east London. Occup. Environ. Med., 57, 86?93, 2000. Burton, G.A., G.E. Batley, P.M. Chapman, V.E. Forbes, E.P. Smith, T. Reynoldson, C.E. Schlekat, P.J. den Besten, A.J. Bailer, A.S. Green, and R.L. Dwyer, A weight- of-evidence framework for assessing sediment (or other) contamination: Improving certainty in the decision-making process. Human Eco. Risk Assess., 8, 1675-1696, 2002. Cape, J.N., J. Methven, and L.E. Hudson, The use of trajectory cluster analysis to interpret trace gas measurements at Mace Head, Ireland, Atmos. Environ., 34, 3561- 3663, 2000 Chen, L.-W. A., B.G. Doddridge, R. R. Dickerson, P. K. Mueller, J. C. Chow, and W.A. Butler, Seasonal variations in elemental carbon aerosol, carbon monoxide, and sulfur dioxide: Implications for sources. Geophys. Res. Let., 28, 1711-1714, 2001. Chen, L.-W.A., Urban fine particulate matter: Chemical composition and possible origins. Doctoral dissertation, Chemical Physics, University of Maryland, College Park, 2002. Chen, L.-W.A., B.G. Doddridge, R.R. Dickerson, J.C. Chow, and R.C. Henry, Origins of fine aerosol mass in the Baltimore-Washington corridor: implications from observation, factor analysis, and ensemble air parcel back trajectories. Atmos. Environ., 36, 4541-4554, 2002. Chen, L.-W. A., J.C. Chow, B. G. Doddridge, R. R. Dickerson, W. F. Ryan, and P.K. Mueller, Analysis of a summertime PM2.5 and Haze Episode in the Mid-Atlantic Region. J. Air & Waste Manage. Assoc., 53, 946-956, 2003. Chin, M., D. J. Jacob, G. M. Gardner, M. S. Foreman-Fowler, P. A. Spiro, and D. L. Savoie, A global three-dimensional model of tropospheric sulfate, J. Geophys. Res., 101, 18,667?18,690, 1996. Chin, M., R.B. Rood, S.J. Lin, J.F. Muller, and A.M. Thomson, Atmospheric sulfur cycle simulated in the global model GOCART: Model description and global properties, J. Geophys. Res., 105(D20), 24671-24687, 2000a. Chin, M., D.L., Savoie, B.J. Huebert, A.R. Bandy, D.C. Thornton, T.S.Bates, P.K. Quinn, E.S. Saltzman, and W.J. DeBruyn, Atmospheric sulfur cycle simulated in the global model GOCART: Comparison with field observations and regional budgets. J. Geophys. Res., 105, 24689-24712, 2000b. 229 Chow, J.C., J.G. Watson, D.,H. Lowenthal, P.A. Solomon, K.L. Magliano S.D. Ziman, and L.W. Richards, PM10 Source Apportionment in California San-Joaquin Valley. Atmos. Environ., 26, 3335-3354, 1992. Chow, J.C., J.G. Watson, J.L. Bowen, C.A. Frazier, A.W. Gertler, K.F. Kochy, D. Landis, and L. Ashbaugh, A sampling system for reactive species in the Western United States. In Sampling and analysis of airborne pollutants, Winegar, E.D., Keith, L.H. eds, 212-214, 1993a. Chow, J.C., Watson, J.G., Pritchett, L.C., Pierson, W.R., Frazier, C.A., Purcell, R.G., The DRI thermal/optical reflectance carbon analysis system: Description, evaluation, and applications in U.S. air quality studies. Atmos. Environ., 27, 1185-1201, 1993b. Chow, J.C., J.G. Watson, D.H. Lowenthal, P.A.. Solomon, K.L. Magliano, S.D. Ziman, and L.W. Richards, PM10 and PM2.5 Compositions in California?s San Joaquin Valley. Aerosol Sci. and Tech., 18, 105-128, 1993c. Chow, J.C., J.G. Watson, Z.Q. Lu, D.H. Lowenthal, C.A. Frazier, P.A. Solomon, R.H. Thullier, and K. Magliano, Descriptive Analysis of PM2.5 and Pm10 at Regionally Representative Locations during SJVAC/AUSPEX. Atmos. Environ., 30, 2079-2112, 1996. Chow, J.C., J.G. Watson, D. Crow, D.H. Lowenthal, and T. Merrifield, Comparison of IMPROVE and NIOSH carbon measurements. Aerosol Sci. and Tech., 34, 23-34, 2001. Chow, J.C., J.G. Watson, L.-W.A. Chen, W.P. Arnott, and H. Moosmuller,. Equivalence of elemental carbon by thermal/optical reflectance and transmittance with different temperature protocols. Enviro. Sci. and Tech., 38, 4414-4422, 2004 Chow, J.C., J.G. Watson, L.-W.A. Chen, G. Paredes-Miranda, M.C.O. Chang, D. Trimble, K.K. Fung, H. Zhang, and J.Z. Yu, Refining temperature measures in thermal/optical carbon analysis. Atmos. Chem. Phys. 5, 4477-4505, 2005a. Chow, J.C., J.G. Watson, D.H. Lowenthal, and K.L Magliano, Loss of PM 2.5 nitrate from filter samples in central California. J. Air & Waste Manage. Assoc., 55, 1158- 1168, 2005b. Chow, J.C., L.-W.A. Chen, J.G. Watson, D.H. Lowenthal, K.A. Magliano, K. Turkiewicz, and D.E. Lehrman, PM2.5 chemical composition and spatiotemporal variability during the California Regional PM10/PM2.5 Air Quality Study (CRPAQS). J. Geophys. Res. Let. 111, D10S04, 2006. Code of Federal Regulations (CFR), National Primary and Secondary Ambient Air Quality Standards, Final Rules, Title 40, Parts 50-53 and 58, 18 July 1997. 230 Colarco, P.R., M.R. Schoeberl, B.G. Doddridge, L.T. Marufu, O. Torres, and E.J. Welton, Transport of smoke from Canadian forest fires to the surface near Washington, D.C.: Injection height, entrainment, and optical properties, J. Geophys. Res., 109, D06203, 2004. Colette, A., G. Ancellet, and F. Borchi, Impact of vertical transport processes on the tropospheric ozone layering above Europe. Part I: Study of air mass origin using multivariate analysis, clustering and trajectories, Atmos. Environ., 39, 5409-5422, 2005a. Colette, A., and G. Ancellet, Impact of vertical transport processes on the tropospheric ozone layering above Europe. Part II: Climatological analysis of the past 30 years, Atmos. Environ., 39, 5423-5435, 2005b. Cornbleet, P., and N. Gochman, Incorrect least-squares regression coefficients in method-comparison analysis. Clin. Chem., 25, 432-438, 1979. Dentener, F. J., G.R. Carmichael, Y. Zhang, J. Lelieveld, and P.J. Crutzen, Role of mineral aerosol as a reactive surface in the global troposphere, J. Geophys.Res.,101 (D17), 22869-22889, 1996. Diab, R.D., A. Raghunandan, A.M. Thompson, and V.Thouret, Classification of tropospheric ozone profiles over Johannesburg based on mosaic aircraft data, Atmos. Chem. Phys., 3, 713-723, 2003. Diab, R.D., A.M., Thompson, K. Mari, L. Ramsay, and G.J.R. Coetzee, Tropospheric ozone climatology over Irene, South Africa, from 1990 to 1994 and 1998 to 2002, J. Geophys. Res., 109, D20301, 2004. Dickerson, R.R., and A.C. Delany, Modification of a commercial gas filter correlation CO detector for increased sensitivity, J. Atmos. Oceanic Technol., 5(3), 424-431, 1988. Dickerson, R.R., B.G. Doddridge, P.K. Kelley, and K.P.Rhoads, Large-scale pollution of the atmosphere over the North Atlantic Ocean: Evidence from Bermuda, J.Geophys. Res., 100(5), 8945-8952, 1995. Dickerson, R.R., S. Kondragunta, G. Stenchikov, K.L. Civerolo, B.G. Doddridge, and B.N. Holben, The impact of aerosols on solar ultraviolet radiation and photochemical smog. Science 278, 827- 830, 1997. Doddridge, B.G., R.M. Morales, K.P. Rhoads, J.T. Merrill, P.C. Novelli, R.R. Dickerson, V.S. Connors, and H.G. Reichle, Jr, Ground-based and airborne observations of carbon monoxide during NASA/MAPS missions SRL-1 and SRL-2, J. Geophys. Res., 103, 19305-19316, 1998. 231 Dorling, S.R., T.D. Davies, and C.E. Pierce, Cluster Analysis: A technique for estimating the synoptic meteorological controls on air and precipitation chemistry ? method and applications, Atmos. Environ., 26A(14), 2575-2582, 1992a. Dorling, S.R., T.D. Davies, and C.E. Pierce, Cluster Analysis: A technique for estimating the synoptic meteorological controls on air and precipitation chemistry ? results from Eskdalemuir, S. Scotland, Atmos. Environ., 26A(14), 2583-2602, 1992b. Dorling, S.R., and T.D. Davies, Extending cluster analysis ? synoptic meteorology links to characterise chemical climates at six northwest European monitoring stations, Atmos. Environ., 29(2), 145-167, 1995. Draxler, R.R., and G.D. Rolph, HYSPLIT (HYbrid Single-Particle Lagrangian Integrated Trajectory) Model access via NOAA ARL READY website (http://www.arl.noaa.gov/ready/hysplit4.html), NOAA Air Resources Laboratory, Silver Spring, MD, 2003. DRI Operating Procedure, Sequential Filter Sampler: Operation, Maintenance, and Field Calibration. Desert Research Institute, 1990. Edgerton, E.S., B.E. Hartsell, R.D. Saylor, J.J. Jansen, D.A. Hansen, and G.M. Hidy, The Southeastern Aerosol Research and Characterization study, part 3: Continuous measurements of fine particulate matter mass and composition, J. Air & Waste Manage. Assoc.,56, 1325-1341, 2006. Eisinger, M. and J. P. Burrows, Tropospheric sulfur dioxide observed by the ERS-2 GOME instrument, Geophys. Res. Lett., 25, 4177-4180, 1998. El-Zanan, H.S., D.H. Lowenthal, B. Zielinska, J.C. Chow, and N. Kumar, Determination of the organic aerosol mass to organic carbon ratio in IMPROVE samples. Chemosphere 60, 485-496, 2005. Eneroth, K., E. Kjellstrom, and K. Holmen, A trajectory climatology for Svalbard; investigating how atmospheric flow patterns influence observed tracer concentrations, Phys. Chem. Earth, 28, 1191-1203, 2003. EPRI, Aircraft Measurements of Pollutants and Meteorological Parameters During the Sulfate Regional Experiment (SURE) Program: Research Triangle Institute, Inc., 1981. ESRI Data and Maps, Arcview 8.0, 2000. Finlayson-Pitts, B., and J. Pitts, Chemistry of the Upper and Lower Atmosphere, Academic Press, San Diego, 2000. 232 Galloway, J. N. and D. M. Whelpdale, An Atmospheric Sulfur Budget for Eastern North-America, Atmos. Environ., 14(4): 409-417, 1980. Flanagan, J., R. Jayanty, E. Rickman, and M. Peterson, PM2.5 Speciation Trends Network: Evaluation of whole-system uncertainties using data from sites with collocated samplers. J. Air & Waste Manage. Assoc., 56, 492-499, 2006. Frank, N. H., Retained nitrate, hydrated sulfates, and carbonaceous mass in Federal Reference Method fine particulate matter for six Eastern U.S. cities. J. Air & Waste Manage. Assoc., 56, 500-511, 2006. Frischer T, M. Studnicka, C. Gartner, E. Tauber, F .Horak, A. Veiter, J. Spengler, J. Kuhr, and R. Urbanek, Lung function growth and ambient ozone: a 3-year population study in schoolchildren. Am. J. Respir. Crit. Care Med.,160, 390?396, 1999. Gauderman W.J., G.F.Gilliland, H .Vora, E. Avol, D. Stram, R. McConnell, D. Thomas, F. Lurmann, H.G. Margolis, E.B. Rappaport, K. Berhane, and J.M., Peters, Association between air pollution and lung function growth in southern California schoolchildren: results from a second cohort. Am. J. Respir. Crit. Care Med.,166, 76? 84, 2002. Grell, G. A., J. Dudhia, and D. R. Stauffer, A description of the fifth-generation Penn State/NCAR Mesoscale Model (MM5). NCAR Tech. Note NCAR/TN-398 STR, 1995. Hains, J.C., B.F. Taubman, A.M. Thompson, L.T. Marufu, B.G. Doddridge, J.W. Stehr, D.Allen, and R.R. Dickerson, Origins of chemical pollution derived from Mid- Atlantic aircraft profiles using a clustering technique, J. Geophys. Res., submitted 2007a. Hains, J.C., L.W.A. Chen, B.F. Taubman, B.G. Doddridge, and R.R. Dickerson, A side by side comparison of filter-based PM2.5 measurements at a suburban site: A closure study, Atmos. Environ., in press 2007b. Harris, J.M., and S.J. Oltmans , Variations in tropospheric ozone related to transport at American Samoa, J. Geophys. Res., 102(D7), 8781-8791, 1997. Harrison, D., S.S. Park, J. Ondov, T. Buckley, S.R. Kim, and R.K.M. Jayanty, Highly time resolved fine particle nitrate measurements at the Baltimore Supersite. Atmos. Environ., 38, 5321-5332, 2004. Hicks, B. B., T. P. Meyers, R.P. Hosker, R.S. Artz, Climatological features of regional surface air quality from the Atmospheric Integrated Research Monitoring Network (AIRMoN) in the USA , Atmos. Environ., 35(6): 1053-1068, 2001. 233 Hodzic, A., R. Vautard, B. Bessagnet, M. Lattuaati, F. Moreto, Long-term urban aerosol simulation versus routine particulate matter observations, Atmos. Environ., 39, 5851-5864, 2005. Hogrefe, C., J. Biswas, B.Lynn, K. Civerelo, J.Y. Ku, J. Rosenthal, C. Rosenzweig, R. Goldberg, and P.L. Kinney, Simulating regional-scale ozone climatology over the eastern United States: model evaluation results, Atmos. Environ, 38, 2627-2638, 2004. Hopke, P.K., Receptor models in air-pollution. Trac-Trends in Analytical Chemistry 3, R6-R7, 1984. IPCC Climate Change 2001: The scientific Basis, 2001. Jorba, O., C. Perez, F. Rocadenbosch, and J.M. Baldasano, Cluster analysis of 4-day back trajectories arriving in the Barcelona area, Spain, from 1997 to 2002, J. Appl. Meteor., 43(6), 887-901, 2004 JPL Publication, Chemical kinetics and photochemical data fro use in atmospheric studies. Evaluation Number 15, JPL Publication 06-2, 2006. Kalkstein, L.S., G. Tan, and J.A. Skindlov, An evaluation of three clustering procedures for use in synoptic climatological classification, J. Appl. Meteor., 26, 717- 730, 1987. Kelley, P., R.R. Dickerson, W.T. Luke, and G.L, Kok, Rate of NO2 photolysis from the surface to 7.6-km altitude in clear-sky and clouds, Geophys. Res. Lett., 22, 2621- 2624, 1995. Kim, E., and P.K. Hopke, Identification of fine particle sources in mid-Atlantic US area. Water Air and Soil Poll. 168, 391-421, 2005. Kim, E., P.K. Hopke, and Y. Qin, Estimation of organic carbon blank values and error structures of the speciation trends network data for source apportionment. Journal J. Air & Waste Manage. Assoc., 55, 1190-1199, 2005. Koch, D., D. J. Jacob, I. Tegen, D. Rind, and M. Chin , Tropospheric sulfur simulation and sulfate direct radiative forcing in the Goddard Institute for Space Studies general circulation model, J. Geophys. Res.,104, 23,799? 23,822, 1999. Laden, F., L.M. Neas, D.W. Dockery, and J. Schwartz, Association of fine particulate matter from different sources with daily mortality in six US cities. Enviro. Health Perspect. 108, 941-947, 2000. 234 Lake D.A., M.P. Tolocka, M.V., Johnston, and A.S. Wexler, Mass spectrometry of individual particles between 50 and 750 nm in diameter at the Baltimore supersite. Enviro. Sci. & Tech. 37, 3268-3274, 2003. Lee, G., J.T. Merrill, and B.J. Huebert, Variation of free tropospheric total nitrate at Mauna Loa Observatory, Hawaii, J. Geophys. Res., 99(D6), 12821-12831, 1994. Lee, J.H., P.K. Hopke, T.M. Holsen, A.V. Polissar, D.W. Lee, E.S. Edgerton, J.M. Ondov, and G. Allen, Measurement of fine particle mass concentrations using continuous and integrated monitors in eastern US cities. Aerosol Sci. and Tech., 39, 261-275, 2005a. Lee, J.H., P.K. Hopke, T.M. Holsen, and A.V. Polissar, Evaluation of Continuous and filter-based methods for measuring PM2.5 mass concentration. Aerosol Sci. and Tech., 39, 290-303, 2005b. Levy, R.C., Second generation retrieval of tropospheric aerosol properties over land from inversion of visible and near-infrared spectral reflectance: Application over Maryland, Doctoral Dissertation, Atmospheric and Oceanic Science, University of Maryland, 2007. Lippmann M., Filters and filter holders. In: Cohen BS, Hering SV, eds. Air sampling instruments for evaluation of atmospheric contaminants, 8th ed. Cincinnati, OH: American Conference of Governmental Industrial Hygienists, Inc., pp. 247-279. 1/15/98 110 NIOSH Manual of Analytical Methods, 1995. Liu B.Y.H., D.Y.H. Pui, and K.L. Rubow, Characteristics of air sampling filter media. In: Marple VA, Liu BYH, eds. Aerosols in the mining and industrial work environments, vol. 3, instrumentation. Ann Arbor, MI: Ann Arbor Science, pp. 989- 1038, 1983. Logan, J.A., Tropospheric ozone: Seasonal behavior trends and anthropogenic influence, J. Geophys. Res., 90, 10,463-10,482, 1985. Luke, W.T., Evaluation of a commercial pulsed fluorescence detector for the measurement of low-level SO 2 concentrations during the Gas-Phase Sulfur Intercomparison Experiment, J. Geophys. Res., 102(D13), 16255-16265, 1997. Malm, W.C., J.F. Sisler, D. Huffman, R.A. Eldred, and T.A. Cahill, Spatial and seasonal trends in particle concentration and optical extinction in the United States. J. Geophys. Res., 99, 1347-1370, 1994. Malm, W.C., B.A. Schichtel, R.B. Ames, and K.A. Gebhart, A 10-year spatial and temporal trend of sulfate across the United States. J. Geophys. Res., 107, 4627, 2002. 235 Malm, W.C., B.A. Schichtel, M.L. Pitchford, L.L. Ashbaugh, and R.A. Eldred, Spatial and monthly trends in speciated fine particle concentration in the United States. J. Geophys. Res., 109, D03306, 2004. Malm, W.C., D.E. Day, C. Carrico, S.M. Kreidenweis, J.L. Collett, G. McMeeking, T. Lee, J. Carrillo, and B. Schichtel, Intercomparison and closure calculations using measurements of aerosol species and optical properties during the Yosemite Aerosol Characterization Study. J. Geophys. Res., 110, D14302, 2005. Marufu, L.T., B.F. Taubman, B. Bloomer, C.A. Piety, B.G. Doddridge, and R.R. Dickerson, The 2003 North American electrical blackout: An accidental experiment in atmospheric chemistry, Geophys. Res. Lett., 31, L13106, doi:10.1029/2004GL019771, 2004 McDow, S.R., and J.J. Huntzicker, Vapor adsorption artifact in the sampling of organic aerosol : Face velocity effects. Atmos. Environ., 24, 2563-2571, 1990. Mebust, M., B. Eder, F. Binhowski, and S. Roselle, Models-3 Community Multiscale Air Qualtiy (CMAQ) model aerosol component 2. Model evaluation, J.Geophys. Res., 108(D6), 4184, doi:10.1029/2001JD001410, 2003. Moody, J.L., and J.N. Galloway, Quantifying the relationship between atmospheric transport and the chemical composition of precipitation on Bermuda, Tellus 40B, 463- 479, 1988. Moody, J.L., J.W. Munger, A.H. Goldstein, D.J. Jacob, and S.C.Wofsy, Harvard Forest regional-scale air mass composition by Patterns in Atmospheric Transport History (PATH), J. Geophys. Res., 103(D11), 13181-13194, 1998 Moody, J.L., S.J. Oltmans, H. Levy II, and J.T. Merrill, Transport climatology of tropospheric ozone: Bermuda, 1988-1991, J. Geophys. Res., 100(D4), 7179-7194, 1995 Moy, L.A., R.R. Dickerson, and W. F. Ryan, Relationship between back trajectories and tropospheric trace gas concentrations in rural Virginia, Atmos. Environ., 28(17), 2789-2800, 1994. Mueller, S.F., E.M. Bailey, T.M. Cook, and Q. Mao, Treatment of clouds and the associated response of atmospheric sulfur in the Community Multiscale Air Quality (CMAQ) modeling system, Atmos. Environ., 40, 6804-6820, 2006. Nicolet, M. R.R. Meier, and D.E. Anderson, Radiation field in the troposphere and stratosphere-II. Numerical Analysis. Planet. Space. Sci., 30, 935-983, 1982. 236 Novakov, T., D.A. Hegg, and P.V. Hobbs, Airborne measurements of carbonaceous aerosols on the East Coast of the United States, J. Geophys. Res., 102, 30,023? 30,030, 1997. OC/EC Laboratory Environmental and Industrial Sciences Division, Research Triangle Institute, Standard Operating Procedure for the Determination of Organic, Elemental, and Total Carbon in Particulate Matter Using a Thermal/Optical Transmittance Carbon Analyzer, 2003. Ogulei, D., P.K. Hopke, L.M. Zhou, P. Paatero, S.S. Park, and J. Ondov, Receptor modeling for multiple time resolved species: The Baltimore supersite. Atmos. Environ., 39, 3751-3762, 2005. Ondov, J.M., T.J. Buckley, P.K. Hopke, D. Ogulei, M.B. Parlange, W.F. Rogge, K.S. Squibb, M.V. Johnston, and A.S. Wexler, Baltimore Supersite Highly time- and size- resolved concentrations of urban PM2.5 and its constituents for resolution of sources and immune responses. Atmos. Environ., 40, S224-A237, 2006. Park, R., The interaction of regional- and global-scale atmospheric chemistry, transport, and climate processes. Doctoral dissertation, Meteorology, University of Maryland, College Park, 2001. Park S.S., D. Harrison, J.P. Pancras, and J.M. Ondov. Highly time-resolved organic and elemental carbon measurements at the Baltimore Supersite in 2002. J. Geophys. Res., 110, D07S06, 2005a. Park S.S., J.M. Ondov, D. Harrison, and N.P. Nair, Seasonal and shorter-term variations in particulate atmospheric nitrate in Baltimore. Atmos. Environ., 39, 2011- 2020, 2005b. Patashnick, H. and E.G. Rupprecht, Continuous PM10 measurement using the Tapered Element Oscillating Microbalance. J. Air & Waste Manage. Assoc., 41, 1079-1083, 1991. Pathak, R.K., X. Yao, and C.K .Chan, Sampling artifacts of acidity and ionic species in PM 2.5 . Enviro. Sci. & Tech., 38, 254-259, 2004. Perliski, L. and S. Solomon, On the Evaluation of Air Mass Factors for Atmospheric Near-Ultraviolet and Visible Absorption Spectroscopy. J. Geophys. Res., 98 (D6) 10363-10374, 1993. Peterson, M.R., and M.H. Richards, Thermal-Optical-Transmittance Analysis for Organic, Elemental, Carbonate, Total Carbon, and OCX2 in PM2.5 by the EPA/NIOSH Method. RTI Conference proceedings, 2002. 237 Peters, A., D.W. Dockery, J.E. Muller, and M.A. Mittleman,. Increased particulate air pollution and the triggering of myocardial infarction. Circulation 103, 2810-2815, 2001a. Peters, T.M., G.A. Norris, R.W. Vanderpool, D.B.Gemmill, R.W.Wiener, R.W. Murdoch, F.F. McElroy, and M. Pitchford, Field performance of PM2.5 Federal Reference Method samplers. Aerosol Sci. and Tech., 34, 433-443, 2001b. Peters, T.M., R.A. Gussman, L.C. Kenny, and R.W. Vanderpool,. Evaluation of PM2.5 size selectors used in speciation samplers. Aerosol Sci. and Tech., 34, 422- 429, 2001c. Pham, M., J.F. Muller, G. Brasseur, C. Granier, and G. Megie, A three-dimensional study of the tropospheric sulfur cycle, J. Geophys. Res., 100, 26061-26092, 1995. Platt, U., Differential Optical Absorption Spectroscopy (DOAS) in Air Monitoring by Spectroscopic Techniques (M.W. Sigrist. Ed.), Chemical Analysis Series, Vol 127, pp.27-84, Wiley, New York, 1994. Platt, U., Recent Developments in DOAS: an Overview. Poster presentation, Second International DOAS Workshop 2003. Polidori, A., B.J. Turpin, H-J. Lim, J.C. Cabada, R. Subramanian, S.N. Pandis, and A.L. Robinson, Local and regional secondary organic aerosol: Insights from a year of semi-continuous carbon measurements at Pittsburgh. Aerosol Sci. and Tech., 40, 861- 872, 2006. Rees, S.L., A.L. Robinson, A. Khlystov, C.O. Stanier, and S.N. Pandis, Mass balance closure and the Federal Reference Method for PM2.5 in Pittsburgh, Pennsylvania. Atmos. Environ., 38, 3305-3318, 2004. Remer, L.A., S.Gasso, D.A. Heff, Y.J. Kaufmann, and B.N. Holben, Urban/industrial aerosol: Ground gased Sun/sky radiometer and airborne in-situ measurements, J. Geophys. Res., 102, 16849-46859, 1997. Research Triangle Institute, Final data summary report for year 1 of the chemical speciation of PM2.5 filter samples project, 2000. Restad, K., I. S. A. Isaksen, and T. K. Berntsen, Global distribution of sulfate in the troposphere: A three-dimensional model study, Atmos. Environ., 32, 3593?3609, 1998. Roelofs, G.-J., J. Lelieveld, and L. Ganzeveld, Simulation of global sulfate distribution and the influence on effective cloud drop radii with a coupled photochemistry-sulfur cycle model, Tellus, Ser. B, 50, 224? 242, 1998 238 Rozanov,V.V., D. Diebel, R.J.D. Spurr, and J.P. Burrows, GOMETRAN: A radiative transfer model for the satellite project GOME, the plane parallel version, J. Geophys. Res., 102(D14):16683-16695, 1997. Ruidavets J.B., M.Cournot, S. Cassadou, M.Giroux, M. Meybeck, and J.Ferrieres, Ozone air pollution is associated with acute myocardial infarction. Circulation, 111, 563?569, 2005. Russell, A., G.R. McGregor, and G.J. Marshall, An examination of the precipitation delivery mechanisms for Dolleman Island, eastern Antarctic Peninsula, Tellus 56A, 501-513, 2004 RTI, Quality Assurance Projects Plan Chemical Speciation of PM2.5 Filter Samples. Research Triangle Park, NC, 2004. http://www.epa.gov/ttn/amtic/files/ambient/pm25/spec/rtiqap.pdf Ryan, W.F., B.G. Doddridge, R.R. Dickerson, R.M. Morales, K.A. Hallock, P.T. Roberts, D.L. Blumenthal, J.A. Anderson, and K.L. Civerolo, Pollutant transport during a regional O 3 episode in the Mid-Atlantic states, J. Air & Waste Manage. Assoc., 48, 786-797, 1998. Schmid, H., L. Laskus, H.J. Abraham, U. Baltensperger, V. Labanchy, M. Bizjak, P. Burba, H. Cachier, D. Crow, J. Chow, T. Gnauk, A. Even, H.M. Brink, K.P. Giesen, R. Hitzenberger, E. Hueglin, W. Maenhaut, C. Pio, A. Carvalho, J.P. Putaud, D.Toom-Sauntry, and H. Puxbaum,. Results of the "carbon conference" international aerosol carbon round robin test stage I. Atmos. Environ., 35, 2111-2121, 2001. Schnelle, K.B., and Dey, P.R., Atmospheric dispersion modeling compliance guide, New York: McGraw-Hill, 2000. Schubert, S.D., R.B. Rood, and J. Pfaendtner, An assimilated data set for earth science applications, Bul. Am. Met. Soc. 74, 2331-2342, 1993. Schwab, J., H. Felton, and K. Demerjian, Aerosol chemical composition in New York state from integrated filter samples: Urban/rural and seasonal contrasts J. Geophys. Res., 109,(D16) D16S05, 2004. Schwartz, J. and L.M. Neas, Fine particles are more strongly associated than coarse particles with acute respiratory health effects in schoolchildren. Epidemiology, 11, 6- 10, 2000. Seigneur, C., Current Status of air quality models for particulate matter, J. Air & Waste Manage. Assoc., 51, 1508-1521, 2001. Seinfeld, J.H. and S.N. Pandis, Atmospheric chemistry and physics: From air pollution to climate change, John Wiley and Sons, Inc. New York, 1998. 239 Solomon, P., K.Baumantt, E. Edgerton, R. Tanner, D. Eatough, W. Modey, H. Maring, D. Savoie, S. Natarajan, M.B. Meyer, and G. Norris, Comparison of integrated samplers for mass and composition during the 1999 Atlanta Supersites project. J. Geophys. Res. 108, 8423, 2003. Stohl, A., G. Wotawa, P. Seibert, and H. Kromp-Kolb, Interpolation errors in wind fields as a function of spatial and temporal resolution and their impact on different types of kinematic trajectories, J. Appl. Meteor., 34, 2149-2165, 1995. Stohl, A., Computation, accuracy and applications of trajectories ? a review and bibliography, Atmos. Environ., 32(6), 947-966, 1998. Stunder, B.J.B, An assessment of the quality of forecast trajectorie,. J. Appl. Meteor., 35, 1319-1331, 1996. Subramanian, R., A.Y. Khlystov, J.C. Cabada, and A.L. Robinson, Evaluation of measurement methods: Positive and negative artifacts in particulate organic carbon measurements with denuded and undenuded sampler configurations. Aerosol Sci. and Tech., 38, 27-48, 2004. Taubman, B.F., L.T. Marufu, C.A. Piety, B.G. Doddridge, J.W. Stehr, and R.R. Dickerson, Airborne characterization of the chemical, optical, and meteorological properties, and origins of a combined ozone/haze episode over the eastern U.S., J. Atmos. Sci., 61(14), 1781-1793, 2004a. Taubman, B.F., L.T. Marufu, B.L. Vant-Hull, C.A. Piety, B.G. Doddridge, R.R. Dickerson, and Z. Li, Smoke over haze: Aircraft observations of chemical and optical properties and the effects on heating rates and stability, J. Geophys. Res., 109, D02206, doi:10.1029/2003JD003898, 2004b. Taubman, B.F., J.C. Hains, A.M. Thompson, L.T. Marufu, B.G. Doddridge, J.W. Stehr, C.A. Piety, and R.R. Dickerson, Aircraft vertical profiles of trace gas and aerosol pollution over the mid-Atlantic United States: Statistics and meteorological cluster analysis, J. Geophys. Res., 111, D10S07, 2006 Technology Transfer Network Air Quality System, 2006. http://www.epa.gov/ttn/airs/airsaqs/detaildata/downloadaqsdata.htm Tesche, T.W., R. Morris, G. Tonnesen, D. McNally, J. Boylan, and P. Brewer, CMAQ/CAMx annual 2002 performance evaluation over the eastern US, Atmos. Envrion., 40, 4906-4919, 2006. Thermo Anderson, RAAS Operator?s Manual. Chemical speciation monitor, 2001. 240 Thermo Scientific, Model 43C Pulsed fluorescence SO 2 analyzer instruction manual, 2004a. Thermo Scientific, Model 48C UV gas filter correlation CO analyzer instruction manual, 2004b. Thermo Scientific, Model 49C UV photometric O 3 analyzer instruction manual, 2004c. Thornton, D.C., A.R. Bandy, A.R. Drieger, Sulfur dioxide over the western Atlantic ocean. Global Biogeochem. Cycles, 1, 317-328, 1987. Tolocka, M.P., D.A. Lake, V. Johnston, and A.S. Wexler, Size-resolved fine and ultrafine particle composition in Baltimore, Maryland. J. Geophys. Res., 110, D07S04, 2005. TSI, Model 3550/3560 Series Integrating nephelometer, Instruction manual, 1997. Turpin, B.J., J.J. Huntzicker, and S.V. Hering, Investigation of organic aerosol sampling artifacts in the Los Angeles Basin. Atmos. Environ., 28, 3061-3071, 1994. Turpin, B. J. and H. Lim, Species contributions to PM2.5 mass concentrations: Revisiting common assumptions for estimating organic mass. Aerosol Sci. and Tech., 35, 602-610, 2001. US EPA Emissions Monitoring and Analysis Division Monitoring & Quality Assurance Group, Strategic plan. Development of the Particulate Matter (PM2.5) quality system for the chemical speciation monitoring trends sites, 1999. US EPA Office of Air Quality and Planning Standards, National Air Pollution Emission Trends, 2000. US EPA, Model SASS & SuperSASS PM2.5 Ambient Chemical Speciation Samplers Field Operation Manual, 2001. US EPA, National air quality and emissions trends report. 2003 Special Studies Edition, 2003. Vukovich, F.M., Boundary-layer ozone variations in the eastern United-States and their association with meteorological variations ? long-term variations, J. Geophys. Res., 99, 16839-16850, 1994. Vukovich, F.M., R. Wayland, and J.Sherwell, Characteristics of ozone in the Baltimore-Washington area as established from one-hour average concentrations, J. Air & Waste Manage. Assoc., 49, 794-803, 1999. 241 Watson, J. G., J.A. Cooper, and J.J. Huntzicker, The Effective Variance Weighting for Least-Squares Calculations Applied to the Mass Balance Receptor Model. Atmos. Environ., 18, 1347-1355, 1984. Watson, J.G., J.L. Bowen, J.C. Chow, C.F. Rogers, M.G. Ruby, M.J. Rood, and R.T. Egami, High volume measurements of size classified suspended particulate matter, in: Methods of Air Sampling and Analysis 3rd edition (J.P. Lodge), Lewis Publishers, Inc., Chelsea, MI, 1989a. Watson, J.G., P.J. Lioy, and P.K. Mueller, The measurements process: Precision, accuracy, and validity in: Air sampling instruments for evaluation of atmospheric contaminants 7th edition, (S.V. Hering), American Conference of Governmental Industrial Hygienists, Cincinnati, OH, pp. 51-57, 1989b. Watson, J.G. and J.C. Chow, Comparison and evaluation of in situ and filter carbon measurements at the Fresno Supersite. J. Geophys. Res.107, 8341, 2002. Wayne, R.P., Chemistry of Atmospheres Third Edition, Oxford University Press, New York, 2000. Weed, D.L., Weight of evidence: A review of concept and methods. Risk Analysis 25, 1545-1557. 2005. White, M.C., R.A. Etzel, W.D. Wilcox, and C. Lloyd, Exacerbations of childhood asthma and Ozone pollution in Atlanta. Environ Res., 65, 56?68, 1994 Wilks, Daniel, Statistical Methods in the Atmospheric Sciences. Academic Press, San Diego, 1995. Wong G.W., F.W. Ko, T.S.Lau, S.T. Li, D. Hui, S.W. Pang, R.Leung, T.F. Fok, and C.K.W. Lai, Temporal relationship between air pollution and hospital admissions for asthmatic children in Hong Kong. Clin Exp. Allergy, 31, 565?569, 2001. Zhang, D. and R.A. Anthes, A high-resolution model of the planetary boundary layer - Sensitivity tests and comparisons with SESAME-79 data. J.App. Meteo., 21, 1594- 1609, 1982. Zhang, X. Q., and P.H. McMurry, Evaporative Losses of Fine Particulate Nitrates During Sampling. Atmos. Environ., 26, 3305-3312, 1992. Zhang, Y., B. Pun, S. Wu, K. Vijayaraghavan, and C. Seigneur, Application and evaluation of two air quality models for particulate matter for a Southeastern U.S. episode. J. Air & Waste Manage. Assoc., 54, 1478-1493, 2004. 242 Zhang, D.-L. and W.-Z. Zheng, Diurnal cycles of surface winds and temperatures as simulated by five boundary-layer parameterizations, J.App. Meteo., 43, 157-169, 2004. Zhang, X., G. Zhuang, J. Chen, Y. Wang, X. Wang, Z. An, and P. Zhang, Heterogeneous reactions of sulfur dioxide on typical mineral particles, J. Phys. Chem. B, 110, 12588-12596, 2006.