ABSTRACT 
 
 
 
 
Title of Document: ASSESSING FOREST BIOMASS AND 
MONITORING CHANGES FROM 
DISTURBANCE AND RECOVERY WITH 
LIDAR AND SAR  
  
 Wenli Huang, Doctor of Philosophy, 2015 
  
Directed By: Professor Guoqing Sun, Department of 
Geographical Sciences  
 
 
 
This dissertation research investigated LiDAR and SAR remote sensing for 
assessing aboveground biomass and monitoring changes from anthropogenic forest 
disturbance and post-disturbance recovery.  First, waveform LiDAR data were 
applied to map forest biomass and its changes at different key map scales for the two 
study sites: Howland Forest and Penobscot Experimental Forest.  Results indicated 
that the prediction model at the scale of individual LVIS footprints is reliable when 
the geolocation errors are minimized.  The evaluation showed that the predictions 
were improved markedly (20% R2 and 10% RMSE) with the increase of plot sizes 
from 0.25 ha to 1.0 ha.  The effect of disturbance on the prediction model was strong 
at the footprint level but weak at the 1.0 ha plot-level.  Errors reached minimum when 
footprint coverage approached about 50% of the area of 1.0 ha plots (16 footprints) 
with no improvement beyond that.  
 
 
Then, a sensitivity analysis was conducted for multi-source L-band SAR 
signatures, to change in forest biomass and related factors such as incidence angle, 
soil moisture, and disturbance type.  The effect of incidence angle on SAR 
backscatter was reduced by an empirical model. A cross-image normalization was 
used to reduce the radiometric distortions due to changes in acquisition conditions 
such as soil moisture.  Results demonstrated that the normalization ensured that the 
derived biomass of regrowth forests was cross-calibrated, and thus made the detection 
of biomass change possible. 
Further, the forest biomass was mapped for 1989, 1994 and 2009 using multi-
source SAR data, and changes in biomass were derived for a 15- and a 20-year 
period.  Results improved our understanding of issues concerning the mapping of 
biomass dynamic using L-ban SAR data.  With the increase of plot sizes, the speckle 
noise and geolocations errors were reduced.  Multivariable models were found to 
outperform the single-term models developed for biomass estimation. 
The main contribution of this research was an improved knowledge 
concerning waveform LiDAR and L-band SAR’s ability in monitoring the changes in 
biomass in a temperate forest.  Results from this study provide calibration and 
validation methods as a foundation for improving the performance of current and 
future spaceborne systems.  
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ASSESSING FOREST BIOMASS AND MONITORING CHANGES FROM 
DISTURBANCE AND RECOVERY WITH LIDAR AND SAR   
 
 
 
By 
 
 
Wenli Huang 
 
 
 
 
 
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 
2015 
 
 
 
 
 
 
 
 
 
 
 
 
 
Advisory Committee: 
Professor Ralph Dubayah, Chair 
Research Professor Guoqing Sun, co-chair 
Professor Eric Kasischke 
Research Scientist Bruce Cook 
Professor Andrew Baldwin  
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
© Copyright by 
Wenli Huang 
2015 
 
 
 
 
ii 
 
 
 
Dedication 
 
To 
 
My parents, 
 Bilong Huang & Guangmin Que,  
 
and  
 
My husband,  
Rui Leng 
  
For their endless love, supports and sacrifices. 
 
  
iii 
 
 
Acknowledgements 
 
This dissertation research was funded by the grants from NASA’s Terrestrial Ecology 
Program (NNX09AG66G) and Carbon Monitoring System (NNX10AT74G), and 
received the financial support from the Chinese Scholarship Council and the Ann G. 
Wylie Dissertation Fellowship. 
 
I would like to thank my advisor and mentor Dr. Guoqing Sun who introduced me to 
lidar and radar remote sensing, and my academic advisor Dr. Ralph Dubayah who gave 
me valuable advices and opportunities in research.  I would also give special thanks to 
my committee member Dr. Bruce Cook for assistance with field data collection, 
conceptual advice and reviews of manuscripts, as well as Dr. Eric Kasischke and Dr. 
Andrew Baldwin for conceptual advice and comments throughout the research.  
 
I would also thank Dr. John Lee for collection forest management maps and assistance in 
field measurements in Maine; Dr. Chengquan Huang for acquisition and processing of 
LTSS-VCT disturbance product over the study area; Dr. Michelle Hofton, Bryan Blair 
and the LVIS team at NASA/GSFC for helping in acquisition and processing waveform 
lidar data; UAVSAR team at NASA/JPL in acquisition and processing the airborne SAR 
data; Alaska SAR Facility in providing the archived SIR-C/XSAR and AIRSAR data, 
and JAXA for providing PALSAR data.  I am thankful to my friends Conrad Berger and 
Dana Watkinson for valuable English edits on this dissertation thesis. 
 
I would like to express my special thanks to Dr. Anu Swantaran, Dr. Wenjian Ni, Dr. 
Zhiyu Zhang, and Dr. Paul Montesano for field planning, data collections, valuable 
discussions and reviews of manuscripts.  Many thanks to Dr. Xiaoguang Cheng, Dr. 
Xiaoqing Wang, Dr. Dongliang Wang, and Dr. Kun Fang for valuable discussions and 
reviews of manuscripts.  I would also thank Bob Crossgrove, Liz Smith, Suzanne Bach, 
Christine Kang, and Dr. Rachel Berndtson for their supports in the department. 
 
Many thanks to my lab mates Dr. Hao Tang, Qiongyu Huang, Dr. Laura Duncanson, Dr. 
Amanda Whitehurst, Rachel Moore, Anika Cartas, Wenlu Qi and all the VCL&GEL lab 
mates for their supports in work and life.  I would also thank all my Chinese friends, 
Feng R. Zhao, Danxia Song, Jie Zhang, Qing Ying, Xin Tao, Xiaopeng Song, Dong Chen, 
Dr. Bo Jiang, Dr. Dongdong Wang, Dr. Tao He, Dr. Lei Wang, Dr. Maosheng Zhao, Dr. 
Feng A. Zhao, Dr. Yi Peng and Dr. Huiran Jin for their suggestions and supports during 
my doctoral program in the department. 
 
 vi 
 
Table of Contents 
 
 
Dedication ..................................................................................................................... ii 
Acknowledgements ...................................................................................................... iii 
List of Tables ............................................................................................................. viii 
List of Figures .............................................................................................................. ix 
Chapter 1 Introduction ............................................................................................... 1 
1.1 Background ................................................................................................... 1 
1.1.1 Relevance of forest biomass to terrestrial ecosystems and carbon budget 1 
1.1.2 Current approaches to estimate forest carbon stocks and changes ........... 4 
1.2 Research questions and objectives ................................................................ 8 
1.3 Howland Forest and Penobscot Experimental Forest ................................... 9 
1.4 Dissertation outline ..................................................................................... 12 
Chapter 2 Mapping Biomass Change after Forest Disturbance using waveform 
LiDAR 14 
2.1 Introduction ................................................................................................. 14 
2.2 Data ............................................................................................................. 18 
2.2.1 Field campaign ........................................................................................ 18 
2.2.2 LiDAR data ............................................................................................. 20 
2.2.3 Auxiliary data.......................................................................................... 20 
2.3 Method ........................................................................................................ 21 
2.3.1 Allometric-based biomass calculation .................................................... 21 
2.3.2 LiDAR data processing ........................................................................... 22 
2.3.3 Forest disturbance identification ............................................................. 25 
2.3.4 Mapping forest biomass and biomass changes ....................................... 26 
2.3.5 Biomass mapping from LVIS data ......................................................... 28 
2.4 Result .......................................................................................................... 29 
2.4.1 Model performance at the LVIS footprint-level ..................................... 29 
2.4.2 Evaluation of prediction model ............................................................... 33 
2.4.3 Biomass mapping from LVIS data ......................................................... 39 
2.4.4 Biomass change mapping ....................................................................... 40 
2.5 Discussion ................................................................................................... 45 
2.5.1 Prediction model at LiDAR footprint-level ............................................ 45 
2.5.2 Disturbance effect ................................................................................... 46 
2.5.3 Map scale and footprint density .............................................................. 47 
2.5.4 Application of footprint-level model for biomass change mapping ....... 49 
2.6 Conclusion .................................................................................................. 50 
Chapter 3 Sensitivity of Multi-Source SAR Backscatter to Changes in Forest 
Aboveground Biomass ................................................................................................ 53 
3.1 Introduction ................................................................................................. 53 
3.2 Data ............................................................................................................. 58 
3.2.1 SAR data ................................................................................................. 58 
3.2.2 Auxiliary data.......................................................................................... 61 
3.3 Method ........................................................................................................ 63 
 vii 
 
3.3.1 Sensitivity of SAR backscatter to biomass ............................................. 64 
3.3.2 Incidence-angle based correction for airborne SAR backscatter ............ 66 
3.3.3 Sensitivity of SAR backscatter to soil moisture and cross-image 
normalization ...................................................................................................... 67 
3.4 Result .......................................................................................................... 70 
3.4.1 Sensitivity of SAR backscatter to incidence angle ................................. 70 
3.4.2 Sensitivity of SAR backscatter to soil moisture ..................................... 73 
3.4.3 Sensitivity of normalized SAR backscatter to forest biomass ................ 74 
3.5 Discussion ................................................................................................... 77 
3.6 Conclusion .................................................................................................. 81 
Chapter 4 Mapping Biomass Change after Forest Disturbance from Multi-source 
SAR 82 
4.1 Introduction ................................................................................................. 82 
4.2 Data ............................................................................................................. 84 
4.3 Method ........................................................................................................ 85 
4.3.1 Multi-source SAR data processing ......................................................... 85 
4.3.2 Modeling approaches for biomass estimation ......................................... 89 
4.3.3 Accuracy assessment of regression models ............................................ 91 
4.3.4 Mapping forest biomass and changes ..................................................... 92 
4.4 Result .......................................................................................................... 92 
4.4.1 SAR backscatter sensitivity to biomass .................................................. 92 
4.4.2 Evaluation of prediction models ............................................................. 93 
4.4.3 Biomass mapping from multi-source SAR data ................................... 102 
4.4.4 Biomass change mapping ..................................................................... 104 
4.5 Discussion ................................................................................................. 104 
4.5.1 Mapping scale and errors in radar data ................................................. 104 
4.5.2 Errors in prediction models ................................................................... 106 
4.5.3 Uncertainty in in-situ estimates ............................................................ 109 
4.6 Conclusion ................................................................................................ 109 
Chapter 5 Summary of Findings, Contributions and Future Research Directions 111 
5.1 Major findings ........................................................................................... 112 
5.2 Major contributions ................................................................................... 113 
5.3 Future research directions ......................................................................... 114 
Glossary .................................................................................................................... 117 
Appendices ................................................................................................................ 119 
Bibliography ............................................................................................................. 124 
 
  
 viii 
 
List of Tables 
 
Table 1-1. Comparison of the forest biomass from field inventory .............................. 4 
Table 1-2. Comparison of the forest biomass from remote sensing-based estimation . 6 
Table 2-1. Correlation between RH metrics of LVIS footprint samples .................... 27 
Table 2-2. Summary of single term regression models at footprint-level .................. 30 
Table 2-3. Combined and disturbance-specific models at footprint-level .................. 33 
Table 2-4. Evaluation of the footprint-level combined RH50 and RH75 models by 
plot-level field data in 2009 and 2003 ........................................................................ 35 
Table 2-5. Evaluation of the footprint-level disturbance-specific RH50 and RH75 
models by plot-level field data in 2009 and 2003 ....................................................... 35 
Table 3-1. Key instrument parameters of SAR system in this study. ......................... 59 
Table 3-2. SAR data selected for sensitivity analysis.  ............................................... 60 
Table 3-3. Correction model parameters and coefficients of determinant for three 
polarizations ................................................................................................................ 70 
Table 4-1. Summary of field estimated biomass in 2009 at different plot-level. ....... 84 
Table 4-2. The calibration factor (CF) for ALOS/PALSAR data ............................... 87 
Table 4-3. Co-registration accuracies for the selected SAR data the test sites ........... 88 
Table 4-4. SAR backscatter coefficient, cross- and full- polairzed indices ................ 90 
Table 4-5. Summary of single term regression models at different plot-level. .......... 95 
Table 4-6. Summary of single index regression models at different plot-level. ......... 96 
Table 4-7. Summary of multi-variable regression models at different plot-level....... 97 
Table 5-1. Allometric equations for biomass estimation in this study ..................... 122 
Table 5-2. DBH-based Height Equations from Field Measurements (1989 to 2010)
................................................................................................................................... 123 
 
 
 ix 
 
List of Figures 
 
Figure 1-1. Conceptual diagram of global carbon dioxide budget. Global Carbon 
Project 2014 (Le Quere et al., 2014). ............................................................................ 3 
Figure 1-2. Conceptual relationship between changes in forest and carbon budget. .... 3 
Figure 1-3. Forest Harvest Trend in Maine, data from Maine Forest Service. Types of 
harvest: partial=harvest where trees are removed individually or in small (<5 acre) 
patches; shelterwood=harvest of mature trees from a forest site in two or more stages 
(5-15 years apart); clear-cut=harvest on a site larger than 5 acres that results in a 
residual basal area of acceptable growing stock trees >4.5" DBH of less than 30". .... 9 
Figure 1-4. Location of study area and field sites: the site to the north is Howland 
Forest (HF), and the site to the south is Penobscot Experimental Forest (PEF). ........ 11 
Figure 2-1. Layout of typical field plot in 2003, 2009, and 2010. (a) stand plot (80 
diameter) in 2003; (b) 1-ha (50m by 200m) plot in 2009; (c) 0.5-ha (50m x 100m) 
plot in 2010. Black dots are LVIS footprint center. .................................................... 19 
Figure 2-2. Year of disturbance from LTSS-VCT product at the HF and PEF study 
sites. Pink polygon is the near matured forest at HF site; and dark blue polygon is the 
outline of reserved area at both sites. .......................................................................... 21 
Figure 2-3. Images of gridded RH (R: RH50, G: RH100, B: RH25) metrics over study 
sites in 2009. Left - HF; right - PEF. Red lines are major roads. Dark blue rectangle in 
HF is the stem-map site. Field measurements are labeled with different colors for 
2003 (red), 2009 (cyan), 2010 (yellow) and 2011 (orange). ....................................... 23 
Figure 2-4. Typical co-incidence waveforms (center within 2m) from LVIS 2003 and 
2009 data. (a) RHs relatively unchanged; (b) RHs have significant negative changes; 
(c) RHs with positive change.  Black solid waveform is from 2009, gold dash 
waveform is from 2003. Red solid line is the detected ground in LGE, dashed straight 
lines are the RH100, dashed with dot lines are the RH50. ......................................... 23 
Figure 2-5. Workflow of model development and evaluation for biomass mapping. 26 
Figure 2-6. Biomass mapping and change detection from LVIS data. ....................... 29 
Figure 2-7. Relationship between field biomass and LVIS relative height metric 
RH50 at footprint-level: (a) Combined model; (b) Disturbance-specific models. 
Different colors indicate different models: rectangle dots (green) represent the 
undisturbed group, and triangle dots (red) represent the disturbed group. ................. 31 
Figure 2-8. Evaluation of combined footprint-level RH50 model in 2009, with the 
solid line for y=x. (a) 0.25 ha field plots; (b) 0.5 ha field plots; (c) 1 ha field plots in 
2009; (d) 0.5 ha field plots in 2003. All results were from the application of combined 
model. Rectangle dots represent the undisturbed points, and triangle dots represent 
the disturbed points. .................................................................................................... 37 
Figure 2-9. Evaluations of disturbance-specific RH50 footprint-level biomass model, 
with solid line for y=x.  (a) 0.25 ha field plots; (b) 0.5 ha field plots; (c) 1.0 ha field 
 x 
 
plots in 2009; (d) ~ 0.5 ha field plots in 2003. Different colors indicate the different 
models were used: rectangle dots (green) represent the undisturbed group, and 
triangle dots (red) represent the disturbed group. ....................................................... 38 
Figure 2-10. Biomass map for HF site (a) and (b), and PEF site (c) and (d) in 2003 
and 2009 at 1.0 ha level by the combined RH50 models. A color of orange to dark 
green indicates an increase of biomass. At HF site, pink polygon is near matured old-
growth forest; and dark blue polygon is the outline of reserve area. .......................... 41 
Figure 2-11. Change in biomass for HF site (a) and PEF site (b) from 2003 to 2009 at 
1.0 ha level by the combined RH50 models. The (c) and (e) are the years of 
disturbances: disturbances prior to 2002 (yellow), between 2003 and 2008 (red), and 
after the 2009 (purple). (d) is the forest management map of HF created from 
information from private owner (international paper company) and Google images. 
The plantation is represented with green solid filled polygons. The clear-, select-, and 
strip-cuts prior to 2002 are outlined with dark red solid lines with gray cross-hatched 
pattern, purple long dotted lines, and red double long dashed lines with gray stripes, 
respectively. The select-cut during 2003 to 2008 is outlined by red double dashed 
lines with irregular dots. At HF site, pink polygon is the near matured forest; and dark 
blue polygon is the outline of reserved area. The select-cut during 2003 to 2008 is 
outlined by red double dashed lines with irregular dots. ............................................ 42 
Figure 2-12. Changes in biomass from LVIS 2003-2009. The (a) and (b) are biomass 
maps at 1.0 ha level by the combined RH50 models using all LVIS from 2003 to 
2009footprints in Maine, non-forest cells are maksed out (in white). (c) is the change 
in biomass from 2003 to 2009, and (d) is the change map that masked out cells (in 
grey) where in two years overlapped in predictions (95% confidence interval). (e) ~ 
(h) are the zoom-in maps over HF sites, where (e) is the year of disturbance: 
disturbance between 1984 and 2002 (dark green), disturbance between 2003 and 2008 
(red), and non-changed forest (grey); (f) is the forest management of HF; (g) is the 
HF biomass change map from 2003 to 2009; and (h) is the map excluding cells with 
overlapped predictions in 2003 and 2009, respectively. ............................................. 44 
Figure 2-13. RMSE (%) of the RH50 biomass prediction models developed at 
footprint-level and the evaluations at 0.25, 0.5 and 1.0 ha scales. Bars from left to 
right represent best models for disturbed (pink), undisturbed (green), disturbance-
specific (light blue), and combined (dark blue). ......................................................... 48 
Figure 2-14. Sensitivity of RMSE% (a) and R2 (b) to the density of footprints within 
the plot (pt ha-1) at 0.25 ha, 0.5 ha and 1.0 ha plot-levels; RMSE (%) and R2 values 
were from the combined RH50 footprint-level biomass prediction model. ............... 49 
Figure 3-1. Map of study sites, and coverage of Landsat ETM+ scene (red dashed line, 
p011/r029), ALOS PALSAR FBD scene (blue solid line) and PLR scene (purple 
dashed line). (a) 0.5 ha FIA style plot in 1992 and 2003, (b) 1.0 ha field plot in 2009.
..................................................................................................................................... 58 
Figure 3-2. Forest management map at the HF study site. Solid dark blue line polygon 
is the outline of the reserved area, and the solid green line is the near mature forest. 
PLT=plantation; CC=clear-cut; SC/STC=strip-cut; SEL=select-cut; SH=shelterwood-
 xi 
 
harvest, 1 and 2 denotes 1st and 2nd entry of shelterwood-harvest; OB/OR= overstory 
removal; IMP=intensive management plot. Number inside the polygon denotes the 
year of management. ................................................................................................... 62 
Figure 3-3. Water fraction by Volume (WFV, %) and precipitation (mm) as a 
function of date. WFV measured at depths of 2 inches (~=5 cm). ............................. 62 
Figure 3-4. Flowchart of SAR data processing and sensitivity analysis. .................... 63 
Figure 3-5. SAR backscatter as a function of forest aboveground biomass from model 
simulation and remote sensing observation. Simulation from ZELIG plotted with (a) 
PALSAR HH, (b) PALSAR HV, (c) UAVSAR HH, and (d) UAVSAR HV. ADAMS 
to WESTBURY are different type of soil based on drainage and taxonomic 
classification (Levine et al., 1994). ............................................................................. 65 
Figure 3-6. Conceptual diagram of the cross-image normalization. Figure units are in 
power domain (m2·m-2) for backscatter (σ), and density per hectare (Mg·ha-1) for 
AGB. S1 is the saturation point in a near-mature forest; S2 is the maximum soil effect 
point in a non-forested area. Dashed blue line (Veg) is the fitted backscatter versus 
AGB received from vegetation canopy without soil influences in theory. Brown 
dashed and dotted lines are backscatter from soil surfaces. Green dotted lines are 
backscatter from vegetation canopy plus soil. ①②③ denotes backscatter from data 
with different surface and soil moisture conditions. ................................................... 68 
Figure 3-7. (a) Selected plots on AIRSAR image (10/07/1994, R: HH, G: HV, B: VV), 
and (b) Forest management map at study site. Pink polygon is the near-mature forest; 
and dark blue polygon is the outline of the reserved area. Solid line polygons with 
labels (i.e., PLT89 = plantation in 1989; CC86 = clear-cut in 1986; STC92 = strip-cut 
in 1992; SHL08 = shelterwood-harvest in 2008) are the plots selected for sensitivity 
analysis. S1 is the saturation point in a near-mature forest  (biomass >200 Mg·ha-1), 
S2 is the maximum soil effect point in a non-forested area (biomass <10 Mg·ha-1). . 70 
Figure 3-8. Mean backscatter in three polarizations (R:HH, G:HV, B: VV) as a 
function of incidence angle. AISSAR data acquired on 10/07/1994before (a) and after 
incidence angle normalization (b), UAVSAR data acquired on 08/05/2009 before (c) 
and after incidence angle normalization (d). ............................................................... 71 
Figure 3-9. Polarized AIRSAR image acquired on 1994/10/07 in three polarizations 
and composite (R: HH, G: HV, B: VV). The above images (a)-(d) show the image 
requested from JPL archive, and the below images (e)-(h) are after correction for 
incident angle. ............................................................................................................. 72 
Figure 3-10. Polarized UAVSAR image acquired on 2009/08/05 in three polarizations 
and composite (R: HH, G: HV, B: VV). The above images (a)-(d) show the calibrated 
image distributed by JPL, and the below images (e)-(h) are after correction for 
incident angle. ............................................................................................................. 73 
Figure 3-11. Mean backscatter (Sigma) plotted as a function of water fraction by 
volume (WFV, %): (a) High biomass stand, (b) Low biomass stand. ........................ 74 
Figure 3-12. Changes in SAR backscatter (HV) from SIR-C/XSAR 1994 at 
represented plots: (a) before normalization, (b) after normalization. ......................... 76 
 xii 
 
Figure 3-13. Original and normalized multi-sources backscatter (1989‒2009) from 
airborne- and spaceborne- SAR systems at represented stands: (a) and (c) are from 
original data, and (c) and (d) are from normalized data. Airborne data include 
AIRSAR 89/09/02, 94/10/07, and UAVSAR 09/08/05. Spaceborne data include SIR-
C/XSAR 94/10/07 and PALSAR 09/08/30................................................................. 77 
Figure 4-1. Timeline of SAR data at HF site. Optical data are acquired during the 
leaf-on season, and SAR data are from L-band. Forest management maps are 
digitized from local management maps. ..................................................................... 84 
Figure 4-2. Flowchart of SAR data preparation, normalization and mapping. ........... 85 
Figure 4-3. Model evaluation across different scales for UAVSAR 2009 data: HH and 
HV polarizations. UAVSAR 0.25 ha, 0.5 ha, and 1.0 ha plot-levels: (a) 0.25 ha HH, 
(b) 0.25 ha HV (c) 0.25 ha Rcross, (d) 0.25 ha RFDI, (e) 0.5 ha HH, (f) 0.5 ha HV (g) 
0.5 ha Rcross, (h) 0.25 ha RFDI, (i) 1.0 ha HH, (j) 1.0 ha HV (k) 1.0 ha Rcross, and (l) 
1.0 ha RFDI.  X axis is the field-measured AGB density (Mg∙ha-1). Y axis is 
predicted AGB density (Mg∙ha-1). Red solid line is 1:1 line. .................................... 98 
Figure 4-4. Model evaluation across different scale for UAVSAR 2009 data: VV 
polarization. UAVSAR 0.25 ha, 0.5 ha, and 1.0 ha plot-levels: (a) 0.25 ha VV, (b) 
0.25 ha BMI (c) 0.25 ha VSI, (d) 0.5 ha VV, (e) 0.5 ha BMI, (f) 0.5 ha VSI, (g) 1.0 ha 
VV, (h) 1.0 ha BMI, and (i) 1.0 ha VSI.  X axis is the field-measured AGB density 
(Mg∙ha-1). Y axis is predicted AGB density (Mg∙ha-1). Red solid line is 1:1 line. . 99 
Figure 4-5. Model evaluation across different scale for PALSAR 2009 data: HH and 
HV polarizations. PALSAR 0.25 ha, 0.5 ha, and 1.0 ha plot-levels: (a) 0.25 ha HH, (b) 
0.25 ha HV (c) 0.25 ha Rcross, (d) 0.25 ha RFDI, (e) 0.5 ha HH, (f) 0.5 ha HV (g) 0.5 
ha Rcross, (h) 0.25 ha RFDI, (i) 1.0 ha HH, (j) 1.0 ha HV (k) 1.0 ha Rcross, and (l) 
1.0 ha RFDI.  X axis is the field-measured AGB density (Mg∙ha-1). Y axis is 
predicted AGB density (Mg∙ha-1). Red solid line is 1:1 line. .................................. 100 
Figure 4-6. Model evaluation across different scale for PALSAR 2009 data: HH and 
HV polarizations. PALSAR 0.25 ha, 0.5 ha, and 1.0 ha plot-levels: (a) 0.25 ha HH, (b) 
0.25 ha HV (c) 0.25 ha Rcross, (d) 0.25 ha RFDI, (e) 0.5 ha HH, (f) 0.5 ha HV (g) 0.5 
ha Rcross, (h) 0.25 ha RFDI, (i) 1.0 ha HH, (j) 1.0 ha HV (k) 1.0 ha Rcross, and (l) 
1.0 ha RFDI.  X axis is the field measured AGB density (Mg∙ha-1). Y axis is 
predicted AGB density (Mg∙ha-1). Red solid line is 1:1 line. .................................. 101 
Figure 4-7. Forest biomass maps at 100 m resolution (1994-2009). ........................ 103 
Figure 4-8. Change in biomass at HF site via cross-image normalized multi-source 
SAR data: (a) 2009PALSAR-1994SIR-C, (b) 2009UAVSAR-1994AIRSAR, and (c) 
2009UAVSAR-1989AIRSAR. Maps are at 100 m spatial resolution, and with 
confidence interval (95%) is applied in predictions. ................................................ 103 
Figure 4-9. Simulated errors from SAR datasets: (a) PALSAR 2009 HV with ±1 dB 
error, (b) UAVSAR 2009 HV with ±1 dB error, (c) PALSAR 2009 HV with ±0.5 dB 
error, and (d) UAVSAR 2009 HV with ±0.5 dB error. Red solid line is the fitted 
regression model, and blue dashed lines show the SAR calibration error interval. .. 107 
 xiii 
 
Figure 4-10. Errors from regression models: (a) PALSAR 2009 HV, (b) UAVSAR 
2009 HV. Y axis is the field measured AGB density (Mg∙ha-1) at 1.0 ha plot-level. X 
axis is SAR backscatter at HV polarization in decibel (dB). Red solid line is the 
prediction curve, and blue dashed lines are the 95% confidence intervals of 
predictions. ................................................................................................................ 108 
Figure 5-1. Fied estimated biomass from national equations as a function of regional 
equations across different plot-levels at HF and PEF sites: (a) 1.0 ha, (b) 0.5 ha, (c) 
0.25 ha, and (d) 0.125 ha........................................................................................... 119 
Figure 5-2. Tree height as a funciton of DBH by: (a) seven species groups from 
Jenkins, and (b) two general species groups. ............................................................ 120 
Figure 5-3. Growth curve with age from LTSS-VCT and field estimated biomass: (a) 
0.25 ha plot, (b) 0.5 ha plot, (c) 1.0 ha plot. X axis represents age classes in an 
interval of 5 year, with classes as 0-5, 5-10, …, 100. Age 100 indicated that the plots 
haven’t been disturbed from 1900s according to management record. Y axis 
represents mean AGB at plot-level. Blue dots show mean values for each age class, 
with dashed bars are variance. .................................................................................. 121 
Figure 5-4. Growth curve with age and biomass from ZELIG model simulation: (a) 
all values, and (b) mean values of each soil classes. ADAMS to WESTBURY are 10 
different type of soil based on drainage and taxonomic classification. .................... 121 
 1 
 
Chapter 1  Introduction       
  
1.1 Background 
1.1.1 Relevance of forest biomass to terrestrial ecosystems and carbon budget  
Terrestrial carbon storage is an critical component in the global carbon budget, 
and is essential to understanding climate change and ecosystem responses to human 
activities, however, current estimates contains high levels of uncertainties (Le Quere 
et al., 2014).  The global carbon dioxide budget is described by emissions and sinks 
divided proportionally among atmosphere, ocean and land:  
EFF + ELULUC = GATM + SOCEAN + SLAND   Equation 1-1 
Where EFF , ELULCC are the carbon sources from fossil fuel & cement, and land use 
and land-cover (LULUC) change; GATM is the carbon growth in the atmosphere; 
SLAND, SOCEAN are the carbon sinks from land and ocean (Figure 1-1).  Note that the 
SLAND and ELULUC have the largest uncertainties of 2.9±0.8 GtC/yr and 0.9±0.5 GtC/yr, 
and both are associated with changes in forest carbon stocks.  
Forest carbon stocks are generally stored in the form of biomass, which 
includes the above- and below- ground biomass (AGB and BGB), such as trees, 
shrubs, vines, roots, and the dead mass of fine and coarse litter associated with the 
soil.  Forest aboveground biomass (hereafter biomass) can be converted into carbon 
stocks by multiplying a simple carbon fraction (about 0.5 in many cases).  It is an 
important component in temperate and tropical forest ecosystems, and is a relatively 
smaller part in boreal forest ecosystems (Malhi et al., 1999).   
 2 
 
Anthropogenic disturbance and management, including deforestation and 
forest degradation from management manipulations, have led to changes in biomass 
and thus the carbon budget (Hall et al., 2011; Houghton et al., 2012).  Yet the loss of 
carbon from forest disturbance and the gain from post-disturbance recovery have not 
been well assessed.  Accurate surface measures of spatial and temporal variations in 
biomass change will support climate treaty frameworks such as Reduced Emissions 
from Deforestation and Forest Degradation Plus (REDD+).   
Forest management might increase or decrease carbon storage (Figure 1-2), 
and applying the former will help to reduce climate disruption as well as increase 
forests’ resilience in the face of climate change (Evans and Perschel, 2009).  There is 
a debate regarding the net effect of forest management on the carbon budget (Nunery 
and Keeton, 2010), because it is difficult to accurately quantify how much of the 
forest carbon changes are due to stand-scale management and landscape-scale 
strategies (Canadell and Raupach, 2008).  Forest management has been incorporated 
into estimates in net flux of carbon in forest only recently (Houghton et al., 2012).  
Therefore, accurately quantifying biomass and changes after forest 
disturbance is desirable in order to reduce the uncertainties in the carbon budget, and 
thus to inform the policy and decision-making community. 
 
 3 
 
 
Figure 1-1. Conceptual diagram of global carbon dioxide budget. Global Carbon 
Project 2014 (Le Quere et al., 2014). 
 
Figure 1-2. Conceptual relationship between changes in forest and carbon budget. 
 
Height
Shrubland Young Forest Mature Forest
Primary ForestSecondary Forest
  Disturbance 
 
 Recovery 
 Neutral 
 4 
 
1.1.2 Current approaches to estimate forest carbon stocks and changes 
Three approaches summarized by Houghton et al. (2012) have been applied to 
estimating forest biomass density (Mg∙ha-1) and changes in density due to LULUC: 1) 
field inventory, 2) remote sensing-based estimates, and 3) modeled estimates.  
 A. Field inventory 
 
Table 1-1. Comparison of the forest biomass from field inventory 
Category Approach Description Advantages Disadvantages 
Direct Destructive 
sampling 
Harvest of all or 
representative trees  
Accurate and direct 
measure;  
Time-consuming & 
labor-intensive; 
Fresh/dry masses 
weighted 
Essential for 
development of 
allometric equations 
Relationship 
applied for limited 
regions 
Indirect Allometric 
equations 
Species based 
relationship between 
biomass and dbh, tree 
height, and wood density 
Easy to conduct 
measurements in the 
field 
Relationship varied 
due to 
environmental and 
climate conditions 
Component 
ratio method 
Biomass converted from 
volume 
With many sampled 
plots from national 
forest inventory 
Relationship varied 
due to 
environmental and 
species 
composition 
 
Field inventory, either directly or indirectly conducted, is the most accurate 
approach for measuring forest biomass (Table 1-1).  Specifically, direct measurement 
requires harvest of all materials, average stands, or individual representative trees 
(Monk et al., 1970).  Many regional studies usually destructively sample individual 
trees to build up an allometric relationship between biomass and field measured tree 
attributes such as height and dbh (diameter at breast height) through regression 
analysis.  On the other hand, indirect field inventories consist of allometric methods 
(i.e., regional or national allometric equations (Jenkins et al., 2003) and volume based 
methods (i.e., component ratio method (Heath et al., 2008).  These field inventories 
 5 
 
form the basis for many national forest inventories and can be combined with 
LULCC to track changes in carbon using an empirical bookkeeping method which 
assigns averaged values for each type of forest.  However, these ground-only 
inventories are labor and time intensive, and are unable to reflect the spatial variation 
of biomass within forest (Houghton et al., 2012).   
B. Remote sensing  
 
Field inventories are accurate if properly conducted, however they are labor-
intensive and time-consuming.  Moreover, no method is currently capable of 
measuring forest biomass directly in a spatial continuous content, therefore a 
combination of remotely sensed data and a well-established field inventory is 
considered suitable for informing programs such as REDD+ (Langner et al., 2014; 
Næsset et al., 2013).  
A variety of remote sensing techniques have been investigated for estimation 
and monitoring of forest carbon stocks (Goetz and Dubayah, 2011; Lu, 2006; Lu et 
al., 2014), and can be generally classified into three major groups based on the sensor 
systems used: 1) passive multispectral and hyperspectral sensor, 2) LiDAR (Light 
Detection and Ranging), and 3) radar (Radio Detection and Ranging) (Table 1-2).  
Passive multispectral and hyperspectral imagery can provide mapping of spatially 
continuous content, but with limited sensitivity to forest biomass unless associated 
with canopy vertical structure from LiDAR and radar.  The capability of LiDAR in 
estimating forest biomass is promising, yet LiDAR has inherent limited spatial 
coverage and temporal resolution.  Current LiDAR systems either provide data with 
limited spatial coverage and temporal resolutions (airborne systems such as small 
 6 
 
footprint discrete-return and large footprint waveform systems), or sampling data 
with sparse temporal observations (spaceborne large footprint waveform systems).  
Radar such as Synthetic Aperture Radar (SAR) compensates for this limitation with 
global coverage and more frequent repeated observations, showing the ability to 
monitor deforestation and forest degradation (Almeida-Filho et al., 2009; Rosenqvist 
et al., 2003).  But many factors other than the change in forest structure and biomass, 
such as radar incidence angle and looking direction, as well as surface and 
environmental conditions, also affect the radar backscattering.  
Table 1-2. Comparison of the forest biomass from remote sensing-based estimation 
Category Sensor Variables Advantages Disadvantages 
Passive Multi-
spectral 
Spectral bands Wall-to-wall Saturation observed 
  Vegetation index   
 Hyper-
spectral 
Multiple bands Accurate for species 
classification 
Saturation observed 
Active SAR Backscatter 
coefficients 
Wall-to-wall Limited by 
environmental conditions 
(i.e., Soil moisture); 
Saturation observed 
 InSAR Coherence Repeat observations Saturation observed 
  Height of phase 
center 
Canopy height 
  
 LiDAR Sampling of 
target attributes 
No saturation Limited observations 
 
Due to issues with each sensor system, few studies have looked into the 
potential for monitoring changes in biomass after forest disturbance and post-
disturbance recovery, only a limited studies have assessed the net effect on changes in 
carbon budget using remote sensing techniques (Dubayah et al., 2010; Hudak et al., 
2012).   Two sets of methods used for mapping changes in forest biomass using 
remote sensing are: 1) the direct method which relates field-derived changes in 
 7 
 
estimated biomass to temporal changes in remote sensing signatures directly; and 2) 
the indirect method (or time-for-space method) which first develops one or two sets 
of models to retrieve forest biomass from remote sensing observations, and then 
calculates change by subtracting one from another.  The direct method requires two or 
more co-incident sets of estimated biomass from field measurements (Dubayah et al., 
2010), which is rare and limits its applications.  On the other hand, the indirect 
method is less efficient than the direct method in terms of accuracy (Rowland et al., 
2008), but is more applicable with current field measurements once the repeated 
remote sensing observations are available.  Recently, studies have looked into the 
possibility of monitoring biomass dynamic using LiDAR and SAR remote sensing 
(Mitchard et al., 2011b; Rowland et al., 2008).  The potential method and key issues 
of using LiDAR and SAR data for assessing forest biomass and monitoring changes 
in biomass after forest disturbance were investigated by empirical approaches in this 
study.  In addition, various factors affect the relationship of remote sensing 
observations to forest biomass, leading to a large uncertainty in estimates.  Important 
sources of uncertainty are found in mapping biomass and changes in biomass from 
remote sensing: (1) uncertainties in the estimate of remotely sensed measurements,  
(2) uncertainties resulting from the biomass prediction models, (3) uncertainties in the 
in-situ estimates of forest attributes, and (4) uncertainties due to time-shift  between 
in situ estimate and remotely sensed measurements (Weisbin et al., 2013).  
Uncertainties from these sources in using LiDAR and SAR data will be addressed in 
this study. 
C. Modeled estimates  
 
 8 
 
A third approach employs physical-based inversion via Radiative Transfer 
Model (RTM) or process-based ecosystem models that calculate internally the carbon 
density of vegetation and soils in different types of ecosystem initialized by climate 
drivers (Peddle et al., 2011; Ranson et al., 2001).  Results from remote sensing such 
as vegetation types and canopy vertical structure are a critical input for these forest 
ecosystem models, and forest biomass is an essential product for validation of output 
from these models.  
1.2 Research questions and objectives 
This dissertation will address the following two major questions: 1) what is 
the level of sensitivity of remotely sensed signatures (i.e., from waveform LiDAR and 
SAR) to changes in forest biomass caused by disturbance and post-disturbance 
recovery?  And 2) what are the key factors to be considered in monitoring and 
assessing the net change in forest biomass, including loss from forest disturbance and 
gain from post-disturbance recovery, via LiDAR and radar remote sensing?   
The overall goal of this study is to monitor forest biomass and assess its 
change after disturbance using LiDAR and radar remote sensing.  Specifically, 
1) Mapping forest biomass from waveform LiDAR data, and assessing the 
loss from disturbance and gain from post-disturbance recovery. 
2) Analyzing the sensitivity of SAR signatures to forest disturbance and major 
influence factors, including incidence angle, soil moisture, and disturbance type. 
3) Mapping aboveground biomass data, and assessing the changes after 
disturbance using multi-source SAR data from spaceborne and airborne platforms. 
 9 
 
1.3 Howland Forest and Penobscot Experimental Forest 
Forest management strategies are changing over time and have distinct 
impacts on the forest structures and thus on biomass.  Statistics reported by the U.S. 
Forest Service (Figure 1-3) show that total harvest wood volume at the State level 
(solid line with filled dot) is relatively stable, but partial harvest (dark grey filled bars) 
has increased dramatically and becomes a major management strategy over the last 
two decades in Maine (Scott et al., 2004).  The use of clear-cut will result in a 
regrowth similar to natural forest.  On the other hand, shelterwood harvest, which 
allows a natural regeneration by increasing light penetration from an opened forest 
canopy, can maintain the same maximum tree height with decreased total biomass at 
stand-level.   
 
Figure 1-3. Forest Harvest Trend in Maine, data from Maine Forest Service. Types of 
harvest: partial=harvest where trees are removed individually or in small (<5 acre) 
patches; shelterwood=harvest of mature trees from a forest site in two or more stages (5-
15 years apart); clear-cut=harvest on a site larger than 5 acres that results in a residual 
basal area of acceptable growing stock trees >4.5" DBH of less than 30". 
 10 
 
Howland Forest (HF) (45°08’ ~ 45°14’ N, 68°42’ ~ 68°45’ W), and the 
Penobscot Experimental Forest (PEF) (45°49’ ~ 45°52.5’ N, 68°30’ ~ 68°38.5’ W) 
are located in south-central Maine.  These forests belong to a transition zone between 
eastern broadleaf and northern forest.  They are essential ecological research stations 
in support of multi-disciplinary studies such as biodiversity, conservation, forest-
atmosphere-soil carbon exchange, and forest management for timber production. 
Most importantly, they provide ground-truth data dating back to the mid-20th century 
for forest biomass studies.  HF has an AmeriFlux Tower within an intermediate aged 
forest, and the surrounding areas are privately land owned by a timber production 
company (International Paper®, IP) with different forest management manipulations 
over the last three decades, such as clear-cuts and plantation during the 1980s, strip-
cuts during the 1990s, and select-cuts (shelterwood harvest and reestablishment) after 
2000.  Harvests since 2000 mostly employ the shelterwood method which typically 
consists of 3 separate entries spaced 10 years apart, with each entry removing about 
1/3 of the basal area (Scott et al., 2004). 
HF and PEF sites consist of sub-boreal forest with mixed deciduous and 
coniferous tree species (Hollinger et al., 1999; Safford et al., 1969).  The dominant 
species include Tsuga canadensis (eastern hemlock), Picea rubens (red spruce), Acer 
rubrum (red maple), Abies balsamea (balsam fir), Thuja occidentalis (northern white 
cedar), Betula papyrifera (paper birch), Fagus grandifolia (Beech), Betula 
alleghaniensis (yellow birch), and Larix laricina (Tamarack) (see Table 5-1 for 
details).  The region features relatively level and gently rolling topography. 
 11 
 
According to USGS 1/3-Arc Second National Elevation Dataset (NED) published in 
2009, the elevation ranges from 40 m to 178 m at HF, and from 29 m to 83 m at PEF.  
These two sites were selected for this study mostly due to the abundant 
ground measurements as well as the large number of remotely sensed datasets that 
have been collected. Details about the field campaign and remotely sensed datasets 
are given in the following chapters.   
 
Figure 1-4. Location of study area and field sites: the site to the north is Howland 
Forest (HF), and the site to the south is Penobscot Experimental Forest (PEF).   
 
 12 
 
1.4 Dissertation outline 
This dissertation contains 5 chapters.  This, the first chapter, summarizes the 
relevance of biomass estimation to the broader discipline of Climate Change and 
current approaches that are applied for mapping and monitoring forest biomass.  
Chapter 2 demonstrates the ability of repeat LiDAR measurements to map the 
changes in forest biomass, and thereby to evaluate the loss from forest disturbance 
and the gain from recovery.  Within the study sites, regression models developed at 
the footprint scale (~0.06 ha) were evaluated with independent plot-level data (up to 1 
ha), and footprint-level models were applied within the study sites.  The effects of 
forest disturbance on prediction models were investigated.  The influence of footprint 
density on biomass predictions is discussed, and average annual biomass reduction 
rate from forest disturbance are given.  
Chapter 3 analyzes the sensitivity of SAR signatures to forest biomass and 
changes due to disturbance.  Sensitivity analysis was conducted to investigate the 
influence of radiometric distortion caused by incidence angle, and the backscatter 
variation caused by soil moisture.  Feasibility of cross-image normalization between 
multi-temporal and multi-sensor SAR data is demonstrated; and the possibility of 
applying normalized backscatter to detect biomass changes due to forest disturbance 
and post-disturbance recovery is discussed.  
Chapter 4 focuses on the application of multi-temporal and multi-sensor SAR 
data to map biomass and biomass changes after forest disturbance.  Cross-image 
normalization proposed in Chapter 3 is applied to airborne and spaceborne multi-
temporal SAR data, and single-term and multi-variable regression models are 
 13 
 
developed. Results from leave-one-out cross-validation are summarized, comparison 
to lidar-derived biomass map is given, and the changes in biomass from for a 15- and 
20- year period from spaceborne and airborne SAR data are reported.  
Chapter 5 summarizes the main findings and contributions from the previous 
chapters.  Future research directions are also discussed based on the results from this 
dissertation study in relation to subsequent spaceborne LiDAR and SAR missions. 
 
 14 
 
Chapter 2  Mapping Biomass Change after Forest Disturbance 
using waveform LiDAR
1
       
  
2.1 Introduction 
Aboveground biomass (AGB, here after biomass) stock from forest represents 
a significant component of the global carbon cycle (Goetz and Dubayah, 2011).  A 
variety of passive and active remote sensing techniques have been investigated for 
measuring and monitoring forest carbon stocks (Goetz and Dubayah, 2011; Lu, 2006). 
Light detection and ranging (LiDAR) is promising because of its ability to directly 
measure canopy vertical profile, providing canopy height information which is highly 
correlated with the forest biomass.  LiDAR systems are categorized as small- or 
large-footprint based on the size of the illuminated ground area. Small-footprint 
LiDAR systems (5-30 cm diameter) provide dense samples for detailed representation 
of the canopy structure, but their use is restricted to low-altitude airborne platforms. 
Small footprint full waveform systems have appeared in recent years with ability to 
record the complete waveform (Mallet and Bretar, 2009).  Large-footprint laser 
systems (10-70 m diameter) record a continuous, vertical profile of returned signal.  
Although large-footprint LiDAR data is not able to capture the very fine spatial 
details of forest canopies, structural attributes can be derived from vertical profiles of 
return energy for application in ecology studies (Mather, 2004).  LiDAR derived 
metrics from small-footprint discrete return LiDAR (Asner et al., 2010; Gonzalez et 
                                                 
1 The presented material has been previously published in: Huang, W., Sun, G., Dubayah, R., Cook, 
B.D., Montesano, P.M., Ni, W., & Zhang, Z. (2013). Mapping biomass change after forest disturbance: 
applying LiDAR footprint-derived models at key map scales. Remote Sensing of Environment, 134, 
319-332.  http://dx.doi.org/10.1016/j.rse.2013.03.017. 
 15 
 
al., 2010; Næsset and Gobakken, 2008; Nilsson, 1996; Pang et al., 2008; Zhao et al., 
2011) and continuous returned full-waveform LiDAR (Drake et al., 2002; Drake et 
al., 2003; Dubayah et al., 2010; Lefsky, 2010; Lefsky et al., 2002b; Lefsky et al., 1999; 
Lefsky et al., 2005a; Lefsky et al., 2007; Means et al., 1999; Ni-Meister et al., 2010)  
have been used for estimation of forest canopy height and biomass.  Various multi-
sensor fusion (Asner et al., 2012; Asner et al., 2010; Kellndorfer et al., 2010; Lefsky 
et al., 2005b; Nelson et al., 2009; Saatchi et al., 2011b; Sun et al., 2011; Swatantran 
et al., 2011) used LiDAR samples and optical or radar imagery data for regional to 
continental mapping of forest attributes.  
Laser Vegetation Imaging Sensor (LVIS) (Blair and Hofton, 1999) with a 
footprint size of 10-25 m, records the entire profile (waveform) of the return signal in 
~30 cm vertical bins (Dubayah and Drake, 2000; Dubayah et al., 2010).  Because the 
footprint size is larger than the diameter of a tree crown and the laser beam can pass 
gaps between trees, a waveform can capture the tree top and ground surface in a 
forest stand. Studies have confirmed the ability of LVIS-derived metrics to estimate 
biomass, even in dense tropical forests.  Drake et al. (2002) reported that height of 
mean energy (HOME or RH50) is the best single term predictor for estimating 
tropical forest biomass at the LVIS footprint-level (~0.05 ha, 25 m diameter) and the 
plot-level (~0.5 ha).  The issue of sampling sizes has also been discussed by several 
studies with small- to large-footprint LiDAR system.  They compared regression 
models at the footprint-level and the plot-level for a tropical wet forest at La Selva, 
Costa Rica, and found that because of geolocation uncertainties, large tree location, 
and species composition, the prediction model was better at plot-level with the R2 of 
 16 
 
0.73 and RMSE of 60.02 Mg∙ha-1.  Results from Hyde et al. (2005) indicated a strong 
agreement between field data and LVIS measurements for height (R2=0.75; 
RMSD=8.2m) and biomass (R2=0.83; RMSD=73.5 Mg∙ha-1) at Sierra Nevada sites in 
California, but not for canopy cover.  Anderson et al. (2006) found good relationship 
between LVIS metrics and height (R2=0.80), but the relationship is weaker between 
metrics and biomass (R2=0.61, RMSE=58 Mg∙ha-1) at Bartlett Experimental Forest 
(BEF) in New Hampshire, USA.  According to Anderson et al. (2008), the possible 
factors for a weaker correlation include geolocation error, species composition, and 
intensity of disturbance.  Dubayah et al. (2010) applied the LVIS data for mapping 
biomass change.  They found various issues that need to be considered in detecting 
and mapping the biomass change with LVIS data, and suggested using range-distance 
based ΔRHE metrics to develop the uniform biomass change equation at plot-level to 
avoid errors caused by ground detection and two sets of regression models. Asner et 
al. (2010) noted the scaling issue that the small-footprint LiDAR prediction errors 
decrease with the increase of plot size.  Mascaro et al. (2011) proposed a “crown-
distributed” approach to address the plot and edge scaling issues caused by the 
disagreement between LiDAR and field measurements. 
The effects of disturbance on the relationship between biomass and height 
metrics were investigated by field observations and model simulation.  Drake et al. 
(2003) investigated the relationships of simple LiDAR metrics (i.e., RH50) with 
estimated biomass, and indicated that there are significant differences between 
different types of forest (i.e., tropical wet forest and tropical moist forest). Ni-Meister 
et al. (2010) indicated that combined height and gap fraction could improve the 
 17 
 
estimation of biomass particularly for coniferous.  Ranson and Sun (2010) simulated 
the waveforms RH metrics from different stands (disturbed and undisturbed forest) by 
a 3D-LiDAR model, and showed that the relationships between forest biomass and 
LiDAR metrics were distinguishable. Asner et al. (2011) found that the fitted curves 
between forest carbon stocks and LiDAR signals are different from plantations and 
natural regrowth after disturbance because of stocking differences. Inventory data and 
modeling results also demonstrated that young forests accumulated biomass much 
faster than the matured forest for the first 10 to 20 years after disturbance (Chazdon, 
2003).  Vegetation change tracker (VCT) algorithm was designed for detecting forest 
disturbance (Huang et al., 2010) via spectral-temporal information from Landsat time 
series stack (LTSS).  The products of yearly disturbance maps from LTSS-VCT were 
used in this study.  
The biomass prediction models can be developed at the scale of footprints and 
larger plots.  To facilitate regional and global biomass mapping using LiDAR 
waveform data, models at footprint-level are desirable because sampling large plots is 
much more time consuming than footprint-level sampling.  The accuracy of biomass 
estimation at coarser scales will depend on the accuracy of the footprint-level models 
and the number of samples (footprints) at this scale. In this study we will investigate 1) 
if the model at footprint-level can be developed with desirable accuracy in our study 
sites, 2) if the forest management practices in term of disturbances will affect the 
models, and 3) what will be the proper scale with concern of uncertainties for 
mapping biomass from LVIS data in our study sites.  Forest biomass map at 1.0 ha 
 18 
 
pixel size was produced from LVIS acquired in 2003 and 2009.  The changes of 
biomass from 2003 to 2009 were analyzed in this study. 
2.2 Data  
2.2.1 Field campaign 
Field measurements for this chapter were conducted during August 2009 to 
2011. Both footprint-level (~0.03 ha, 20 m diameter) and plot-level (0.25 ha - 1.0 ha) 
plots (see Figure 2-1 for typical layout) were measured.  Differential Global Position 
System (DGPS) instruments were used to locate LVIS footprints and establish 
sampling plots.  
Twenty-four 1.0 ha plots (hectare plot, 200m × 50 m, Figure 2-1a) and ten 0.5 
ha plots (half hectare plot, 100m × 50m, Figure 2-1b) were established in 2009 and 
2010, respectively.  The longer edges of these plots were in the range direction of the 
NASA/JPL Uninhabited Aerial Vehicle Synthetic Aperture Radar (UAVSAR) flight 
lines.  The layout of these plots is illustrated in, where each plot consists of sixteen 
0.25 ha (quarter hectare plot, 25m × 25m) subplots.  Ninety-one circular plots with 
20m diameter (Figure 2-1c) centered at each LVIS footprint were measured in 2010 
and 2011.  Forty-seven footprints were measured in August, 2010 and forty-four were 
measured during January and August of 2011.   
Twenty forest inventory samples across the HF site were collected in October, 
2003 for biomass and other forest parameters.  For each inventory sample, three to 
four plots with radius of 4 m, 7 m or 10 m were arranged in the center, 30 m north, 
south-west, and south-east from the center.  The diameter at breast height (DBH, 
diameter at 1.3 m above ground) for every tree with a DBH ≥ 3 cm were recorded 
 19 
 
along with the species in both years.  The relative canopy position (i.e., dominant, co-
dominant, intermediate, suppressed, or dead), heights of two dominant trees were 
measured for each plot in 1992.  The height, crown length and width of 8 trees in 
each plot were measured in 2003.  Details about collections of field data during the 
October 2003 campaign were described by Sun et al. (2011).  Seventeen forest 
inventory samples within the HF study site were used for the evaluation of 2003 
biomass map.  
From 2009 to 2011, the DBH, species of ever tree with DBH ≥ 10 cm, top 
height, height of crown base and crown width of the three highest trees in each 
subplot were recorded.  A census of stems below the established size threshold (DBH 
< 10 cm) and height ≥1.3 m were sampled within a 2 m transect along the center of 
the rectangular plot, and from north-to-south in footprint-level circular plots.  The 
number of stems falling into four diameter categories (i.e., 0-2 cm; 2-5 cm, 5-8 cm 
and 8-10 cm) was counted and used as a representative sample of all small stems in 
the plot.  Biomass values at HF and PEF span from 0.3 to 361.4 Mg ha-1 for 0.25 ha 
plots, 0.6 to 316 Mg ha-1 for 0.5 ha plots, and from 1.0 to 278.9 Mg ha-1 for 1.0 ha 
plots. 
 
Figure 2-1. Layout of typical field plot in 2003, 2009, and 2010. (a) stand plot (80 
diameter) in 2003; (b) 1-ha (50m by 200m) plot in 2009; (c) 0.5-ha (50m x 100m) 
plot in 2010. Black dots are LVIS footprint center.   
 
 20 
 
2.2.2 LiDAR data 
The Laser Vegetation Imaging Sensor (LVIS) is a large footprint airborne 
scanning laser altimeter developed by NASA Goddard Space Flight Center (GSFC) 
(Blair et al., 2006; Blair et al., 1999).  LVIS data of study area were acquired during 
leaf-on season in August of 2003 and 2009.  For both years, the footprints had a 
nominal diameter of 20 m.  The 2003 data was reprocessed in 2008, thus it is more 
consistent with the 2009 data. 
LVIS products of version 1.02 provide three types of datasets: LVIS Canopy 
Elevation (LCE), LVIS Geolocated Elevation (LGE), and LVIS Ground Waveforms 
(LGW) (Blair et al., 2006).  From the waveform, mean elevation of the lowest 
detected mode is defined as ground elevation (zg).  Then, relative heights (RH) to the 
ground elevation are calculated at quartile percentage of cumulative waveform energy 
(i.e., 25%, 50%, 75%, and 100%).  The footprint density varies at different locations 
in the study area because of overlapping flight lines during the campaign.  
2.2.3 Auxiliary data 
LTSS-VCT disturbance products were used in this study to identify disturbed 
forests from undisturbed ones (Huang et al., 2010).  The study area is located in the 
center of Landsat p011/r029 of the World Reference System (WRS) with good 
quality images.  Subset product maps are used in this study to differentiate the year of 
disturbances (Figure 2-2).  The product detects most of clear-cuts events, however, it 
has missed some strip-cutting around 1990-1995 and select-cut (shelterwood harvest) 
after 2000 at HF site.  Similar problem has been noticed and documented at other 
validation sites in USA (Thomas et al., 2011).  Therefore, a further forest 
 21 
 
management map was created from operation information from the private owner 
(International Paper®, IP Company) and Google images at HF site.  Several patches 
of disturbed forest were digitized from Google Earth images and combined with 
digitized version of management operation maps in different year.  These maps were 
used to identify the occurrence of disturbance, and will be explained in details in 
sections of results and discussion. 
 
Figure 2-2. Year of disturbance from LTSS-VCT product at the HF and PEF study 
sites. Pink polygon is the near matured forest at HF site; and dark blue polygon is the 
outline of reserved area at both sites.  
2.3 Method 
2.3.1 Allometric-based biomass calculation 
The diameter-based allometric equations used for large stems (DBH ≥ 10 cm) 
and small stems (DBH < 10 cm) came from the comprehensive report of USDA on 
North American forest (Jenkins et al., 2003; Jenkins et al., 2004).  Biomass of large 
stems was calculated by corresponding species-specific allometric equations.   
Biomass of the small stems was calculated by mixed hardwoods equations using the 
midpoint of the diameter class (i.e., 1.0 cm, 2.5 cm, 6.5 cm and 9.0 cm) as the DBH 
 22 
 
times the number of stems in each category. Biomass was first calculated for each 
stem, and then total biomass was aggregated from subplot to plot levels.  
2.3.2 LiDAR data processing 
Relative height metrics (RH25, RH50, RH75 and RH100) of LVIS waveform 
were retrieved from LVIS LGE datasets for all sampling plots and measured 
footprints.  The LVIS RH metrics of the study sites (HF and PEF) in 2009 were 
shown in Figure 2-3 as false color images (R: RH50, G: RH100, B: RH25).  Images 
were created by interpolating of point data into 15m grid with a Delaunay 
triangulation method (TRIGRID function) provided by IDL Version 7.1 (Exelis, 
Boulder, CO).  
The change in canopy profiles in waveform reveals the biomass change 
between 2003 and 2009.  Waveforms acquired in 2003 and 2009 at HF site are shown 
in Figure 2-4.  The distances between the waveform centers in 2003 and 2009 were 
less than 2 m.  These waveforms represent the disturbed forest with (a) near-mature 
forest with neutral changes, (b) disturbed forest with negative change in RH metrics, 
and (c) forest with positive change in RH metrics, respectively.  
 23 
 
 
Figure 2-3. Images of gridded RH (R: RH50, G: RH100, B: RH25) metrics over study 
sites in 2009. Left - HF; right - PEF. Red lines are major roads. Dark blue rectangle in 
HF is the stem-map site. Field measurements are labeled with different colors for 
2003 (red), 2009 (cyan), 2010 (yellow) and 2011 (orange). 
 
 
 
 
Figure 2-4. Typical co-incidence waveforms (center within 2m) from LVIS 2003 and 
2009 data. (a) RHs relatively unchanged; (b) RHs have significant negative changes; 
(c) RHs with positive change.  Black solid waveform is from 2009, gold dash 
waveform is from 2003. Red solid line is the detected ground in LGE, dashed straight 
lines are the RH100, dashed with dot lines are the RH50. 
 
  
 24 
 
The density of LVIS waveforms in each field plot varied depending on 
number of overlapping flightlines at a given portion of forest.  The nominal spacing 
of LVIS footprint is 20 m both along and cross track.  Ideally, the nominal footprint 
density within a plot would be 9, 18, and 36 at sizes of 0.25 ha (50 m × 50 m), 0.5 ha 
(50 m × 100 m) to 1.0 ha (50 m × 200 m).  However, two factors lead to a varied 
footprint density.  First, there were overlapping LVIS flight lines for our study site. 
Similar issues have been mentioned at other LVIS study sites such as Sierra site in 
California (Hyde et al., 2005), Bartlett site in New Hampshire (Anderson et al., 2008), 
and La Selva site in Costa Rica (Dubayah et al., 2010).  In addition, the long edges of 
50m × 200m field plots were set along the range direction of the UAVSAR data, 
which has a 10-20° angle to the LVIS’s flight direction as shown in Figure 2-4 b-c. 
Therefore, the averaged footprint density within the measured plots in 2009 and 2010 
was not consistent and varied from 14, 27, and 53 footprints per plot from 0.25 ha, 
0.5 ha to 1.0 ha plot-levels.  
While processing the footprint-level field samples, two samples in the near 
matured old-growth forest region were found with wrong ground elevations (zg from 
LVIS LGE product) values.  This discrepancy has been mentioned in LVIS known 
data set issues (http://lvis.gsfc.nasa.gov/DataDisclaimer.html), which is caused by 
insufficient energy returned from the ground and errors associated with the automated 
peak-finding algorithm.  These two points were corrected in this study by finding a 
mean ground elevation of their neighboring footprints. 
 25 
 
2.3.3 Forest disturbance identification 
All sampled plots were classified as undisturbed or disturbed based on LTSS-
VCT yearly disturbance product (1984 to 2010), high-resolution images and field 
notes.  Forest disturbance maps were generated to show the disturbances prior to 2003, 
during 2003 to 2009 and after 2009 using the abovementioned data and forest 
management information from the owner.  Most of the disturbed plots in HF site were 
those disturbed by management activities (Sun et al., 2011), including clear-cut stands 
in the 1980s, strip-cuts in the 1990s and select-cuts (shelter-wood harvest) after 2000.  
A few tree plantations in our study site were also labeled as disturbed forest, as they 
were mainly planted after clear-cut.  While the disturbance data we used relied 
heavily on the LTSS-VCT, we enhanced the classification using visual interpretation 
of high resolution imagery and field notes to refine the boundaries of forest 
disturbance patches.  In addition, National Land Cover Data (NLCD) products in 
2001 and 2006 were used to discriminate forest and non-forest for the entire study 
area.  Woody forestlands and wetlands were included as forests in our analysis.  
The sampled footprints consist of 47 undisturbed (51.6%) and 44 disturbed 
(48.4%) samples.  The mean biomass value of undisturbed field samples (157.1 
Mg∙ha-1) was higher than that of disturbed ones (87.4 Mg∙ha-1).  At 0.25 ha plot-level, 
there were 41 undisturbed and 64 disturbed plots. At 0.5 ha plot-level, there were 18 
undisturbed and 34 disturbed plots.  At 1.0 ha plot-level, there were 10 undisturbed 
and 12 disturbed plots.  
 26 
 
2.3.4 Mapping forest biomass and biomass changes 
The mapping procedure consists of four steps: 1) develop the biomass 
estimation models from 2009 LVIS waveform data at the footprint-level and choose 
the best one; 2) evaluate the model performance with the plot-level observation data 
in 2009 and 2003, and determine the pixel size of the biomass map to be generated; 3) 
apply the selected model to generate biomass maps in both 2009 and 2003 with LVIS 
waveform data; and 4) detect the change in biomass from 2003 to 2009.  The first two 
steps are shown in a conceptual workflow in Figure 2-5. 
 
Figure 2-5. Workflow of model development and evaluation for biomass mapping. 
 
This study employed several widely used statistical indicators to evaluate the 
accuracy of different regression models.  Indicators included coefficient of 
determination (R2), root mean square error (RMSE), and RMSE (%) which is the ratio 
of RMSE to mean observed value:  
y
RMSERMSE (%)
      Equation 2-1  
where y is the mean biomass.  
 27 
 
A. Development of Regression Models 
 
Linear regression models were developed relating field-measured biomass and 
LVIS metrics at footprint-level and evaluated at different plot-level (i.e., 0.25 ha, 0.5 
ha and 1.0 ha rectangular plot). Two types of models were developed, i.e., combined 
model without consideration of disturbances and disturbance-specific models. RHs 
metrics are highly correlated as shown in Table 2-1, so we only develop the single 
term regression models.  
Table 2-1. Correlation between RH metrics of LVIS footprint samples 
 RH25 RH50 RH75 RH100 
RH25 1.00    
RH50 0.90 1.00   
RH75 0.79 0.96 1.00  
RH100 0.72 0.87 0.95 1.00 
 
A dummy variable was introduced into the linear regression model to test the 
effect of disturbances on intercepts and slopes (Solberg, 2010).  This dummy variable 
is used to indicate the occurrence of disturbance, where the truth is represented by a 
numerical value of 1. Then the equation becomes: 
Biomass = β1+ β2X1+ β3X2+ β4X1X2+ei    Equation 2-2 
 
where β1 ~β4 are estimated parameters, X1 is the RH metrics in meters, X2 is the 
dummy variable with values 0 (undisturbed ) or 1(disturbed), and ei is the error item.  
The hypothesis is H0: Bundisturbd - Bdisturbed = 0 or β4 = 0. If β4 = 0 then we will reject the 
H0 which means that the undisturbed and disturbed models are the same, otherwise 
they are different.   
 28 
 
B. Evaluation of prediction models by field biomass data 
 
Biomass of LVIS footprints within field plots was predicted from the 
footprint-level model and aggregated to the plot-levels.  Then it was compared with 
corresponding field measurement.  For 2009 data, the predicted biomass was 
evaluated with field measurements aggregated at three scales of plot-levels: 0.25 ha, 
0.5 ha and 1.0 ha.  For 2003 data, the field biomass was measured with FIA-style 
field measurements at ~0.5 ha plot-level.  The predicted biomass was an average of 
the biomass from the footprints within a circle of 80 m diameter.  The best mapping 
pixel size was determined by the plot-level model evaluation.  Generally the 
aggregation footprint-level samples to larger plots reduced the overall variance and 
impact of geolocation errors (Hall et al., 2011).  Two recently studies have shown 
similar pattern that LiDAR prediction errors were reduced at larger plot sizes (Frazer 
et al., 2011; Mascaro et al., 2011).  The optimized pixel size for biomass mapping 
was selected based on statistical measurements including averaged prediction value, 
R2, RMSE, and bias.   
2.3.5 Biomass mapping from LVIS data 
The mapping procedure is illustrated in Figure 2-6. First, two masks were 
created. A persistent non-forest mask was generated from NLCD product in 2001 and 
2006. Non-forest pixels in both years were excluded from the mapping in this study. 
In addition, a data coverage mask was created for the area common to both LVIS data 
in 2003 and 2009. These masks were gridded into the selected optimized pixel size to 
match the biomass mapping scale. Next, the regression model was applied to the 
 29 
 
LVIS footprints of entire study area in 2003 and 2009. Finally, the mean value of the 
footprint-level biomass within each grid cell was calculated and assigned to the pixel. 
The biomass change map was generated by subtracting 2003 biomass from that in 
2009 within the forested area common to both LVIS data collections.  
 
Figure 2-6. Biomass mapping and change detection from LVIS data. 
 
2.4 Result 
2.4.1 Model performance at the LVIS footprint-level 
 A. Single term regression model 
 
 30 
 
Table 2-2 shows the biomass prediction from single term regression models. 
The biomass models at the footprint-level by all four RH metrics have high R2 values 
ranging from 0.70 to 0.86.  For all groups, there were strong and significant 
correlations (p < 0.005) between biomass and RH metrics.  The RH50 and RH75 
metrics perform similarly in terms of R2, RMSE and RMSE (%).  Averaged field 
biomass of 91 sampled footprints was 123.4 Mg∙ha-1. As mentioned above RH 
metrics are highly correlated.  The correlation is as high as 0.96 between RH75 and 
RH50.  Therefore, the single term regression model using RH50 was selected for the 
combined prediction model at the footprint-level for all data as it explains the greatest 
proportion of variance (R2 = 0.86), and has the lowest residual error (RMSE = 31.0 
Mg∙ha-1, and a relative error 25.1%).   
Table 2-2. Summary of single term regression models at footprint-level 
Model  
(ft-level) 
Intercept Slope R2 
RMSE 
(Mg∙ha-1) 
RMSE  
(%) 
Bio ~ RH25 90.8 22.0 0.70 45.1 36.5 
Bio ~ RH50 30.3 16.1 0.86 31.0 25.1 
Bio ~ RH75 -3.3 13.5 0.84 33.0 26.6 
Bio ~ RH100 -51.9 11.1 0.74 42.1 34.2 
*For 2009 data, Bolded is the selected model for mapping 
  
 31 
 
 
 
Figure 2-7. Relationship between field biomass and LVIS relative height metric 
RH50 at footprint-level: (a) Combined model; (b) Disturbance-specific models. 
Different colors indicate different models: rectangle dots (green) represent the undisturbed group, and 
triangle dots (red) represent the disturbed group. 
 
Figure 2-7 shows the relationship between field biomass and RH50 and the 
contrast between undisturbed and disturbed groups at footprint-level.  From Figure 
2-7 (b), we could visually observe the two groups with different slopes of their trend 
lines.  A statistical measurement was used to test the disturbance effect on intercepts 
and slopes in the next section.  
 B. Disturbance effect test 
 
Following equation is from the regression when a dummy variable “disturb” 
with value 0 for disturbed and 1 for undisturbed forest was added for RH50 model: 
Bio = 33.2 + 16.9*RH50 -3.7*disturb - 4.6*RH50*disturb  Equation 2-3 
 
with Multiple R2: 0.94, standard error: 28.7 Mg∙ha-1, F-statistic: 224.7 on 3 and 87 
degrees of freedom, and a p<0.001. 
 32 
 
Since the coefficient for the dummy variable (β4 = -4.6) is not equal to zero, 
we reject the H0 hypothesis.  The effect of disturbances on biomass estimation model 
from RH50 is significant.  
The dummy variable was also added to RH75 model and the regression 
equation was: 
Bio = -8.6 + 15.4*RH75 -22.7*disturb - 6.0*RH75*disturb  Equation 2-4 
 
with Multiple R2: 0.95, standard error: 24.7 Mg∙ha-1, F-statistic: 309.3 on 3 and 87 
degrees of freedom, and a p<0.001. 
Hence, we also rejected the H0 hypothesis for the RH75 model.  
A student T-test with two tails, unequal sample sizes, and unequal variance 
was employed to measure the disturbance effect on the predicted biomass. The T-test 
showed that disturbance has a significant effect on the predicted biomass from both 
RH50 (p<0.001) model and RH75 (p<0.001) model.  
Thus, footprint-level single term models were developed for undisturbed and 
disturbed forest with RH50 and RH75 models.  As shown in Table 2-3, both the 
RMSE and RMSE (%) were reduced for the disturbance-specific RH50 and RH75 
models.  Even though the R2 and RMSE of the disturbance-specific models were not 
always better than the combined model, the comparisons between the field biomass 
and predicted biomass of all sample footprints showed better results from 
disturbance-specific models.  The third lines in RH50 and RH75 of the disturbance-
specific models are the results of comparing predicted biomass with field biomass of 
all sampling footprints.  Disturbance-specific models explained higher predicted 
variance (RH50, R2=0.89; RH75, R2=0.91) than the combined model (RH50, R2=0.86; 
 33 
 
RH75, R2=0.84). RMSE (RH50, 27.9 from 31.0 Mg∙ha-1; RH75, 24.2 from 32.9 
Mg∙ha-1) as well as RMSE (%) (RH50, 22.6% from 25.1%; RH75, 19.6% from 26.6%) 
were reduced.  The large differences in RMSE (%) for different groups (disturbed, 
undisturbed and all) are partially caused by the differences of mean biomass.  The 
mean biomass of the undisturbed forest (157.1 Mg∙ha-1) is over 50% larger than that 
of the disturbed (87.4 Mg∙ha-1), and also larger than that of the combined plots (123.4 
Mg∙ha-1). 
Table 2-3. Combined and disturbance-specific models at footprint-level 
Model Variable N 
GP
# 
 
Intercept Slope R2 
RMSE 
(Mg∙ha-1) 
RMSE 
(%) 
Combined RH50  91 all  30.3 16.1 0.86* 31.0 25.1 
 RH75  91 all  -3.3 13.5 0.84* 32.9 26.6 
 
 
Disturbance- 
specific 
RH50 
47 1  33.2 16.9 0.88* 32.3 20.5 
44 2  36.9 12.3 0.80* 22.4 25.7 
91 all  - - 0.89* 27.9 22.6 
RH75 
47 1  -8.6 15.4 0.89* 30.0 19.1 
44 2  14.1  9.4 0.90* 15.9 18.2 
 91 all  -  - 0.91* 24.2 19.6 
GP#: 1 is undisturbed plots group, 2 is disturbed plots group; N: number of sample; *P-value < 0.005. 
Bolded are models with best performance by evaluation at corresponding scale; 
 
2.4.2 Evaluation of prediction model 
 A. Evaluation of combined prediction model in 2009 and 2003 
 
The footprint-level RH50 model from the combined data 
(Bio=30.3+16.1*RH50) was applied to 2009 LVIS data and evaluated at three plot-
levels: 1) 0.25 ha plot (50m × 50m), 2) 0.5 ha plot (50m × 100m), and 3) 1.0 ha plot 
(50m × 200m), respectively.  The evaluation plots were also divided into disturbed 
and undisturbed plots using LTSS-VCT and Google Earth images.  
The evaluation of the combined footprint-level models with plot-level field 
data was shown in Table 2-4.  As expected, the best model at three plot-levels in 2009 
 34 
 
is RH50 model with higher explanation of total variance, lower RMSE and lower bias.  
The overall model performance improved with larger plot size from 0.25 ha, 0.5 ha to 
1.0 ha. At 1.0 ha plot-level, the combined RH50 model explained 91% of the total 
variance with a positive bias of 2.0 Mg∙ha-1 (1.4%) and RMSE of 22.4 Mg∙ha-1 
(15.6%).  
Figure 2-8 shows the scatter plot of predictions versus field measurements 
from the best combined footprint-level prediction model.  In 2009, the combined 
RH50 model has better performance than RH75 model from evaluations at all three 
plot-levels.  While in 2003, the combined RH75 model was better than combined 
RH50 model.  There is almost no bias (less than ±1.3%) observed for predictions in 
2009, but the evaluation of 2003 biomass prediction at sampling sites showed worse 
results.  In 2003, the combined RH75 prediction model overestimates the biomass 
with a positive bias of 11.9 Mg∙ha-1 (+7.9%).  It has lower explanation of total 
variance (54%) and higher RMSE of 46.6 Mg∙ha-1.  The combined RH50 model also 
overestimates the biomass and has similar explanation of total variance (53%) and 
higher RMSE.  A part of the reason is that the plot size and shape in 2003 (80m 
diameter circle, ~0.5 ha) were different from 2009 (rectangular) which leads to an 
inconsistency.  The number of sample in 2003 is relatively small compared with that 
of 0.5 ha plot-level samples in 2009.  In addition, the GPS unit used in 2003 for field-
sampled plot wasn’t as good as the one for 2009, which leads to geolocation errors. 
  
 35 
 
Table 2-4. Evaluation of the footprint-level combined RH50 and RH75 models by 
plot-level field data in 2009 and 2003 
N: number of sample; Mean of field: mean biomass averaged over samples at plot-level; Mean of 
predict: mean predicted biomass averaged over samples at plot-level; Bolded are models with best 
performance at corresponding scale and year; 
 
 
Table 2-5. Evaluation of the footprint-level disturbance-specific RH50 and RH75 
models by plot-level field data in 2009 and 2003 
Model Year 
Plot-
size 
N 
Mean 
field 
Mean 
predict 
R2 RMSE Bias 
  ha  Mg∙ha-1 Mg∙ha-1  Mg∙ha-1 % Mg∙ha-1 % 
RH50 
2009 
0.5 52 142.7 139.7 0.80 31.3 21.9 -2.9 -2.1 
1.0 22 143.8 145.8 0.91 23.1 16.1 +0.9 +0.6 
2003 0.5 17 151.0 170.9 0.51 54.6 36.2 +19.9 +13.2 
RH75 
2009 
0.25 105 143.6 142.0 0.73 37.8 26.3 -1.7 -1.2 
0.5 52 142.7 140.9 0.75 35.4 24.8 -1.9 -1.3 
1.0 22 143.8 145.4 0.86 27.9 19.4 +1.6 +1.1 
2003 0.5 17 151.0 171.3 0.48 57.4 38.0 +20.3 +13.4 
N: number of sample; Mean of field: mean biomass averaged over samples at plot-level; Mean of 
predict: mean predicted biomass averaged over samples at plot-level; Bolded are models with best 
performance at corresponding scale and year; 
  
Model Year 
Plot-
size 
N 
Mean 
field 
Mean 
predict 
R2 RMSE Bias 
  ha  Mg∙ha-1 Mg∙ha-1  Mg∙ha-1 % Mg∙ha-1 % 
RH50 
2009 
0.25  105 143.6 144.5 0.79 32.6 22.7 +0.8 +0.6 
0.5 52 142.7 144.4 0.83 28.5 20.0 +2.2 +1.5 
1.0 22 143.8 145.8 0.91 22.4 15.6 +2.0 +1.4 
2003 0.5  17 151.0 168.0 0.53 50.4 33.4 +17.0 +11.2 
RH75 
2009 
0.25 105 143.6 147.1 0.72 37.3 26.0 +3.5 +2.4 
0.5 52 142.7 147.7 0.76 33.8 23.7 +4.9 +3.5 
1.0 22 143.8 144.6 0.84 28.5 19.8 +0.8 +0.5 
2003 0.5 17 151.0 162.9 0.54 46.6 30.9 +11.9 +7.9 
 36 
 
 
 B. Evaluation of disturbance-specific prediction models in 2003 and 2009 
 
Similar steps were conducted to evaluate the disturbance-specific models by 
different plot-level data.  Evaluation results were shown in Table 2-5. It can be seen 
from comparing the R2 listed in the Table 2-4 and Table 2-5 that the disturbance-
specific models preformed slightly better than common models.  The disturbance-
specific RH50 model has almost the same values of explanation of variance and 
RMSE as the combined RH50 model.  At 1.0 ha plot-level, the disturbance-specific 
RH50 model explained 91% of the total variance, with a positive bias of 0.9 Mg∙ha-1 
(0.6%) and RMSE of 23.1 Mg∙ha-1 (16.1%).  The bias was reduced from -3.0 Mg∙ha-1 
to 0.9 Mg∙ha-1 from 0.25 ha to 1.0 ha plot-levels.  
  
 37 
 
 
 
Figure 2-8. Evaluation of combined footprint-level RH50 model in 2009, with the 
solid line for y=x. (a) 0.25 ha field plots; (b) 0.5 ha field plots; (c) 1 ha field plots in 2009; (d) 0.5 
ha field plots in 2003. All results were from the application of combined model. Rectangle dots 
represent the undisturbed points, and triangle dots represent the disturbed points. 
 
 
 38 
 
 
Figure 2-9. Evaluations of disturbance-specific RH50 footprint-level biomass model, 
with solid line for y=x.  (a) 0.25 ha field plots; (b) 0.5 ha field plots; (c) 1.0 ha field 
plots in 2009; (d) ~ 0.5 ha field plots in 2003. Different colors indicate the different 
models were used: rectangle dots (green) represent the undisturbed group, and 
triangle dots (red) represent the disturbed group. 
  
 39 
 
 
Figure 2-9 shows the scatter plots of predictions versus field measurements 
from the best disturbance-specific footprint-level models at various plot sizes.  The 
RH50 model has the best performance at all plot levels and in both 2009 and 2003.  
Again, there is almost no bias (less than ±2.0%) observed for predictions in 2009 and 
a less than +13.5% bias in 2003.  Similar to the combined RH models, the 
disturbance-specific models overestimated the biomass for 2003. 
Even though the statistical tests of prediction models at footprint level showed 
significant effect of disturbance, the evaluation of the predicted biomass 
demonstrated that the disturbance effect is reduced in a larger sampling area.  For 
further biomass mapping and change detection the combined RH50 model was used 
and the biomass maps were generated at 1.0 ha (100 m) spatial resolution.  
2.4.3 Biomass mapping from LVIS data 
Figure 2-10 shows the biomass maps from LVIS data for year 2003 and 2009 
were produced using the RH50 regression model and were averaged to 1.0 ha spatial 
resolution.  The biomass ranges up to 350 Mg∙ha-1, with a color of orange (0˗50 Mg∙
ha-1) to dark green (>300 Mg∙ha-1) indicates an increase of biomass. Grey color 
represents the areas of no data and non-forest.  The overall similar patterns of 
biomass can be seen from both years of 2003 and 2009.  As shown in Figure 2-10, the 
north-west region with biomass less than 100 Mg∙ha-1 at HF site in 2009 was mostly 
caused by select-cut (i.e., shelter-wood harvest, removed large trees accounting for 
about 1/3 of the basal area) and strip-cut (i.e., systematically removed stems in rows). 
The undisturbed forests in the center of the map (outlined by pink polygon in dotted 
 40 
 
line) were with high value of biomass (>300 Mg∙ha-1).  For PEF in 2003 and 2009, 
high biomass regions were observed in the south and west region of the map.  Low 
biomass regions with less than 50 Mg∙ha-1 in the north-east were the woody wetland 
along the Penobscot River. 
2.4.4 Biomass change mapping 
Figure 2-11 (a) and (b) shows the changes of biomass (green to red color) 
from 2003 to 2009 at the two study sites, corresponding to the disturbance maps of 
Figure 2-11 (c), (d), and (e). In addition, a confidence interval (CI) was applied for 
the predictions from LVIS data in 2003 and 2009 as shown in Figure 2-12, where 
only predicted values within the 95% CI were kept (Figure 2-12h). It can be seen that 
most of the changes are consistent with the forest disturbance patterns detected by the 
LTSS-VCT product and the historical management map.  
 
 
 
 
 41 
 
 
Figure 2-10. Biomass map for HF site (a) and (b), and PEF site (c) and (d) in 2003 
and 2009 at 1.0 ha level by the combined RH50 models. A color of orange to dark 
green indicates an increase of biomass. At HF site, pink polygon is near matured old-
growth forest; and dark blue polygon is the outline of reserve area. 
  
 42 
 
 
Figure 2-11. Change in biomass for HF site (a) and PEF site (b) from 2003 to 2009 at 
1.0 ha level by the combined RH50 models. The (c) and (e) are the years of 
disturbances: disturbances prior to 2002 (yellow), between 2003 and 2008 (red), and 
after the 2009 (purple). (d) is the forest management map of HF created from 
information from private owner (international paper company) and Google images. 
The plantation is represented with green solid filled polygons. The clear-, select-, and 
strip-cuts prior to 2002 are outlined with dark red solid lines with gray cross-hatched 
pattern, purple long dotted lines, and red double long dashed lines with gray stripes, 
respectively. The select-cut during 2003 to 2008 is outlined by red double dashed 
lines with irregular dots. At HF site, pink polygon is the near matured forest; and dark 
blue polygon is the outline of reserved area. The select-cut during 2003 to 2008 is 
outlined by red double dashed lines with irregular dots. 
 
  
 43 
 
At HF site, biomass changes in the undisturbed near-mature forests (center of 
the map, highlighted by pink polygon in dotted line) were mostly positive (5 to 15 
Mg∙ha-1) or near neutral (-5 to 5 Mg∙ha-1).  The average annual biomass accumulation 
from undisturbed forest and regrowth is +4.4 Mg∙ha-1.  The area surrounding this 
undisturbed forest shows strong negative change (in red, represent < -50 Mg∙ha-1). 
Most areas along the roads are the degraded forests from select-cut between 2003 and 
2008 (filled irregular dots) as shown in Figure 2-11 (d).  The average annual biomass 
reduction rate from forest disturbance is -7.0 Mg∙ha-1.  Several patches highlighted 
with yellow in Figure 2-11 (c) show an increasing biomass during 2003 to 2009, due 
to the forest regrowth after clear-cut in the 1980s or strip-cut in the 1990s.  At PEF 
site, patches of strong negative biomass changes with red color in Figure 2-11 (e) 
were sparsely distributed over the study region.  Most of them were detected by 
LTSS-VCT disturbance product in Figure 2-11 (d).  The average annual biomass 
reduction rate from forest disturbance is -6.2 Mg∙ha-1.  The woody wetlands with low 
biomass along the Penobscot River were regrowth from clear-cut prior to 2002.  The 
average annual biomass accumulation from regrowth is +4.4 Mg∙ha-1.  
Regression model developed in 2009 were applied to two years LVIS data, 
and change in biomass were mapped from 2003 to 2009 (Figure 2-12).  Prediction 
intervals (i.e., 95% confidence interval) were calculated for estimated biomass in 
2003 and 2009, and cells with overlapped predictions were masked out (Figure 2-12 d) 
as neutral (statistically indistinguishable from zero).  A detailed comparison is given 
for HF site.  The patterns of biomass change (Figure 2-12 h) are consistent with VCT 
product (Figure 2-12 e) and local forest management (Figure 2-12 d), where most of 
 44 
 
the near matured area as neutral and positive change in biomass are mostly from 
plantation as well as forest recovery after disturbance (i.e., disturbed between 1984 
and 2002). 
 
Figure 2-12. Changes in biomass from LVIS 2003-2009. The (a) and (b) are biomass 
maps at 1.0 ha level by the combined RH50 models using all LVIS from 2003 to 
2009footprints in Maine, non-forest cells are maksed out (in white). (c) is the change 
in biomass from 2003 to 2009, and (d) is the change map that masked out cells (in 
grey) where in two years overlapped in predictions (95% confidence interval). (e) ~ 
(h) are the zoom-in maps over HF sites, where (e) is the year of disturbance: 
disturbance between 1984 and 2002 (dark green), disturbance between 2003 and 2008 
(red), and non-changed forest (grey); (f) is the forest management of HF; (g) is the 
HF biomass change map from 2003 to 2009; and (h) is the map excluding cells with 
overlapped predictions in 2003 and 2009, respectively.  
  
 45 
 
 
2.5 Discussion 
Our results highlight four important issues concerning biomass mapping from 
waveform LiDAR: (1) the feasibility of the prediction model at the LiDAR footprint, 
(2) the effect of forest disturbances on the biomass prediction model, (3) the effect of 
map scale and footprint density on the biomass estimation, and (4) the application of 
the footprint-level model for biomass change detection. 
2.5.1 Prediction model at LiDAR footprint-level 
The results in our study sites demonstrate that LiDAR footprint-level models 
could be developed and applied to map biomass, with 91% explanation of total 
variance, a RMSE of 22.4 Mg∙ha-1 (15.6%) for the combined RH50 model at 1.0 ha 
plot-level.  Two main factors lead to this conclusion.  First, the accurate location 
provided by DGPS and high quality LVIS data reduced the geolocation errors.  On 
the one hand, in our study, all footprint-level field measurements and sampling plots 
at HF in 2010 and 2011 and at PEF in 2009 were located using DGPS with a 
measurement error of 0.5-3.0 m (best case was 0.5-1.0 m).  On the other hand, 
increased accuracy of the geolocation has been reported for the LVIS product 
released after 2003 by improved post data processing.  Reprocessed 2003 LVIS data 
which using the same waveform analysis method are more consistent with 2009 LVIS 
data. Therefore geolocation errors are mostly avoided for the data used in the study. 
Secondly, the footprint size of LVIS facilitates the application of models at the 
footprint-level. LVIS data has a 20m diameter footprint, which generally corresponds 
 46 
 
to the largest tree crowns, and thus can capture canopy characteristics (Dubayah and 
Drake, 2000).  
2.5.2 Disturbance effect 
Footprint-level regression models differed between undisturbed and disturbed 
forest, different species composition.  Previous study (Anderson et al., 2006, 2008) 
has mentioned the effect of species composition to the biomass estimation.  Forest 
disturbances change the spatial structure as well as the species composition.  The 
statistical tests reveal a significant difference at 95% confidence level between 
models for the disturbed and undisturbed forests.  The disturbance-specific models 
performed slightly better for biomass estimation than the combined model (Table 2-3).   
But the evaluations results (Table 2-4 and Table 2-5) at different plot-levels from 0.25 
ha, 0.5 ha to 1.0 ha show almost the same biomass estimation accuracies for the 
combined and disturbance-specific models.  The evaluation of biomass estimation 
shows that the combined RH50 model overpasses the combined RH75 model, and the 
disturbance-specific RH50 and RH75 models regardless of scale.  These results 
weaken the importance of introducing disturbance factor into footprint-level model.  
It is reasonable for the effect of disturbance weaker at plot levels because averaging 
of LVIS footprints in plot will reduce the disturbance effect.  The combined model 
was used for biomass mapping in this study.  However, we still recommend 
considering the disturbance effect in area with more complicated species composition. 
On the other hand, it is important to note the potential error that can be 
introduced by the classification of disturbance from LTSS-VCT.  Based on field notes, 
recent Google Earth imagery and LTSS-VCT product, we are confident in the 
 47 
 
accurate classification of disturbance for our field samples.  For a broader region 
more efforts are need to guarantee the accuracy in identification of disturbance.  
Careful application of a disturbance dataset is recommended when conducting forest 
biomass change assessments. 
2.5.3 Map scale and footprint density 
Biomass maps were produced by application of prediction models developed 
at the footprint-level (~0.03 ha; 20 m diameter circle plot) and re-sample the footprint 
biomass into map grids.  The field biomass samples at different plot-levels (0.25 ha, 
0.5 ha, and 1.0 ha; rectangular plot) served as independent data for evaluation of the 
accuracies of biomass maps.  
Figure 2-13 shows the RMSE (%) of the biomass prediction models 
developed at footprint level, and the evaluations at 0.25 ha, 0.5 ha and 1.0 ha plot-
level.  We can see a decreasing trend for RMSE (%) with increasing size of plots.  At 
1.0 ha plot, the RMSE (%) from disturbance-specific and combined models is similar 
and is much lower than at other smaller plot levels.  Recent studies (Frazer et al., 
2011; Mascaro et al., 2011) have indicated that 1.0 ha plots could capture biomass 
with low and stable errors close to 10%.  Therefore, we used a 1.0 ha scale for the 
biomass mapping and change detection using the combined RH50 models.  
 48 
 
 
Figure 2-13. RMSE (%) of the RH50 biomass prediction models developed at 
footprint-level and the evaluations at 0.25, 0.5 and 1.0 ha scales. Bars from left to 
right represent best models for disturbed (pink), undisturbed (green), disturbance-
specific (light blue), and combined (dark blue). 
 
Figure 2-14 shows the sensitivity of (a) RMSE (%) and (b) R2 to the footprint 
density (pt·ha-1) from 0.25 ha, 0.5 ha and 1.0 ha plot-level.  With the increasing of 
footprint density, there is a decreasing trend for RMSE (%), and an increasing trend 
of R2 at 0.25 ha, 0.5 ha and 1.0 ha plot-levels.  At all three plot-levels, a critical 
inflection point where point density equals to 16 pt·ha-1 was observed regardless of 
scale.  The relationship between RMSE (%) and point density become stable after 
reaching this inflection point. This is the same to the trend of R2.  Therefore, optimal 
point density of 16 is suggested for a high quality of biomass estimation at 1.0 ha 
plot-level.  This is feasible because the average footprint density within our field 
 49 
 
sampled plots was over 50 pt·ha-1 with a 20 m nominal spacing both along and across 
track. 
 
Figure 2-14. Sensitivity of RMSE% (a) and R2 (b) to the density of footprints within 
the plot (pt ha-1) at 0.25 ha, 0.5 ha and 1.0 ha plot-levels; RMSE (%) and R2 values 
were from the combined RH50 footprint-level biomass prediction model. 
 
2.5.4 Application of footprint-level model for biomass change mapping 
The footprint-level models were developed using 2009 data in this study and 
then applied to both 2009 and 2003 data for mapping forest biomass and its change.  
Repeat acquisition of LiDAR data has been used for detecting changes of canopy 
height and biomass. Dubayah et al. (2010) recommended using the relationship 
between the biomass change and change in LVIS-derived range-based forest canopy 
height metrics for biomass change studies to avoid using two sets of biomass 
estimation models.  However, due to limited co-incidence field measurements at 
either footprint or plot level, we couldn’t develop the similar equations.  Instead, 
footprint-level plots were selected to develop a uniform biomass equation and then 
this equation was applied to LVIS data in both 2003 and 2009.  In addition to the 
forest spatial structural variations from disturbances, other factors such as species 
composition, seasonal changes of leaf area index, reflectance of ground surface, etc. 
will also affect the LiDAR waveform metrics and the biomass prediction model.  This 
 50 
 
should be considered in applying prediction model developed at a place at certain 
time to other places or data acquired at different seasons.  The LiDAR waveform data 
used in this study were acquired using the same instrument (LVIS) and at the same 
season (August) in 2003 and 2009.  The processing of the LVIS data is also the same 
for 2003 and 2009 data.  Therefore the common relationship between biomass and 
LiDAR waveform metrics should keep consistent from 2003 to 2009. 
2.6 Conclusion 
1) Prediction model at the scale of individual LVIS footprints is reliable when the 
geolocations of the measured footprints were determined by DGPS with a best 
accuracy of 0.5-1.0 m.  
2) The differences between biomass prediction models for disturbed and undisturbed 
forests were statistically significant (P<0.001) at the scale of footprint, and the 
disturbance-specific models performed slightly better (R2=0.89, RMSE=27.9 Mg ha-1, 
and relative error of 22.6%) than the combined model (R2=0.86, RMSE=31.0 Mg ha-1, 
25.1%).  
3) The evaluation using field plot data showed that the predictions of biomass were 
improved markedly with the increase of plot sizes from 0.25 ha to 1.0 ha and that the 
effect of disturbance was not strong. At 1.0 ha plot-level, both disturbance-specific 
and combined models agreed well with field estimates (R2=0.91, 23.1 Mg∙ha-1, 16.1%; 
and R2=0.91, 22.4 Mg∙ha-1, 15.6%).  
4) Sensitivity analysis on levels of variation and error to footprint density suggests 
that a certain density of LVIS footprints is required for biomass mapping. The errors 
 51 
 
were minimized when footprint coverage approached about 50% of the area of 1.0 ha 
plots (16 footprints).  
5) By applying the footprint-level models developed from 2009 LVIS data to both 
2009 and 2003 LVIS data, the change in biomass from 2003 to 2009 could be 
assessed. The average annual biomass reduction rate from forest disturbance at two 
sites is -7.0 Mg∙ha-1 and -6.2 Mg∙ha-1, the average annual biomass accumulation from 
regrowth is +4.4 Mg∙ha-1 and +5.2 Mg∙ha-1, respectively.  
Estimating biomass dynamics over relatively short time scales is a difficult 
task, yet is central to obtaining a better understanding of the effects of disturbance 
and subsequent regrowth on the terrestrial carbon cycle.  There is additionally a 
strong and growing need to develop effective mapping and monitoring in support of 
climate treaty frameworks such as REDD+ (Goetz and Dubayah, 2011).  The work 
presented here gives some examples of LiDAR remote sensing approach to this 
problem.  
Change studies using LiDAR remote sensing are also dependent on having 
sufficient LiDAR coverage to develop spatially meaningful maps.  Ideally such 
coverage would be wall-to-wall, but practically that may not be achievable nor may it 
be entirely necessary.  Our errors were minimized when footprint coverage 
approached about 50% of the area of 1.0 ha plots (16 footprints) with no 
improvement beyond that.  This is a particularly important point when considering 
monitoring from space-based LiDAR, which is unlikely to have swath-mapping 
capability in the near term. 
 52 
 
While our experiment showed improvement in biomass prediction when 
disturbance was included, the results were not compelling.  We doubt that this 
conclusion is generalizable beyond the present study.  While canopy metrics must 
implicitly include the effects of disturbance as reflected in height, other factors also 
control height, most notably climate and edaphic factors.  Thus more work is needed 
to untangle the relationships between these factors, disturbance, and their 
manifestation in height metrics.  That said, the fusion of Landsat disturbance products 
with time series of LiDAR data is a powerful approach to quantifying landscape level 
changes in vegetation structure and will certainly be exploited with increasing 
frequency in future studies. 
Ultimately, there is distance to travel before we can confidently monitor 
biomass and canopy structure dynamics at policy-relevant scales with the requisite 
accuracy in a consistent and transparent framework from remote sensing.  LiDAR 
remote sensing is so new that only now are we able to evaluate data sets with 
sufficient time intervals between them, and for which contemporaneous field 
estimates are available.  We anticipate that as more investigations undertake such 
studies, rapid progress will result in this important capability. 
 53 
 
Chapter 3  Sensitivity of Multi-Source SAR Backscatter to 
Changes in Forest Aboveground Biomass
2
    
  
3.1 Introduction 
The carbon budget of terrestrial ecosystems contains large uncertainties at 
both global and regional scales (Goetz and Dubayah, 2011).  Aboveground biomass 
(AGB, hereafter biomass) stock from forest represents an important component of the 
global carbon cycle and related carbon policy (Houghton et al., 2009).  
Anthropogenic disturbance including deforestation and forest degradation due to 
management has led to significant changes in biomass and thus the carbon budget 
(Hall et al., 2011).  However, the loss of carbon due to deforestation and forest 
degradation, and the gain from post-disturbance recovery have not been sufficiently 
assessed.  The use of active remote sensing techniques such as Synthetic Aperture 
Radar (SAR) is a promising approach for measuring and monitoring the spatial and 
temporal variation of forest carbon stock (Hall et al., 2011; Kasischke et al., 1997; Lu, 
2006).  
Imaging with SAR has advantages over optical imagery in its capacity to 
penetrate clouds, rain, smoke, and haze, which are known problem for optical sensors.  
The ability to penetrate the forest canopy makes it possible to retrieve the forest 
structure as a function of backscatter mechanisms (Kasischke et al., 1997).  Generally, 
studies have reported that SAR backscatter is more sensitive to canopy biomass 
(especially tree trunks) at longer wavelengths.  Full polarimetric SAR (PolSAR) 
                                                 
2 The presented material has been previously published in: Huang, W., Sun, G., Ni, W., Zhang, Z., 
Dubayah, R. (2015). Sensitivity of Multi-Source SAR Backscatter to Changes in Forest Aboveground 
Biomass, Remote Sensing. 7(8): 9587-9609; doi: 10.3390/rs70809587. 
 54 
 
provide four combinations of transmitted and received polarizations: co-polarized 
bands HH (horizontal transmitted and horizontal received) and VV (vertical 
transmitted and vertical received), as well as cross-polarized bands HV (horizontal 
transmitted and vertical received) and VH (vertical transmitted and horizontal 
received).  Longer SAR wavelengths such as P- (30-100 cm) and L- (15-30 cm) 
bands penetrate farther into the forest canopy and capture more vertical structure than 
C- (4.8-7.7 cm) and X- (2.8-5.2 cm) bands, while HV backscatter from dual-
polarization (e.g., HH, HV) or full-polarization (i.e., HH, VV, HV, and VH) are more 
sensitive to woody biomass. 
A series studies suggested that a widely applicable relationship exists between 
biomass and backscatter from L-band SAR for woody vegetation with lower levels of 
biomass (≤150 Mg·ha-1) in tropical (Collins et al., 2009; Englhart et al., 2011; 
Mitchard et al., 2011a; Mitchard et al., 2009b; Pope et al., 1994; Saatchi et al., 
2011b), temperate and boreal biomes (Botkin and Simpson, 1990; Kurvonen et al., 
1999; Ranson et al., 1995; Sandberg et al., 2011).  Both airborne and spaceborne 
systems were involved in these studies. These including airborne instruments such as 
AIRborne SAR (AIRSAR) and Uninhabited Aerial Vehicle SAR (UAVSAR) 
developed by NASA, Experimental-SAR (E-SAR) operated by the German 
Aerospace Center (DLR), as well as spaceborne instruments such as Spaceborne 
Imaging Radar-C and X-B and SAR (SIR-C/XSAR), and Phased Array type L-band 
SAR (PALSAR) on board the Advanced Land Observing Satellite (ALOS).  Ranson 
et al. (1995) investigated the use of multi-frequency, multi-polarization and multi-
season image data from SIR-C/XSAR to map forest cover type and estimate 
 55 
 
aboveground biomass for a Boreal Ecosystem-Atmosphere Study (BOREAS) site in 
Saskatchewan, Canada.  Santos et al. (2002) utilized the L-HH channel of JERS-1 
data in tropical forest-savanna contact zones, and found that the logarithmic and 
sigmoid functions were adequate to explain the SAR backscatter as a function of 
forest biomass.  Collins et al. (2009) indicated that the L-HV channel of polarimetric 
SAR backscatter intensity from AIRSAR was best suited (R2=0.92) for modeling 
biomass (both above- and below-ground) of the tropical savannahs in North Australia.  
Mitchard et al. (2009a) examined the relationships between field-measured biomass 
at four study sites in Cameroon, Uganda and Mozambique and data from 
ALOS/PALSAR, and found that biomass estimates based on these relationships were 
highly significant and similar among sites.  Sandburg et al. (2011) explored the 
relationship between SAR backscatter from E-SAR data and forest biomass in 
southern Sweden, and found that for L-band data the best results were obtained from 
HV-polarized backscatter. Robinson et al.(2013) studied the variation in field 
estimated biomass at different scales (0.0625, 0.25, 0.5, and 1.0 ha) in a temperate to 
boreal transitional region (Howland, ME), and found that the cross-polarized HV had 
the highest sensitivity to field estimated biomass (R2=0.68).  
Nevertheless, a number of important factors can affect the relationship 
between radar backscatter and forest biomass, including change in forest structure 
(Dobson et al., 1995), radar incidence angle (Sun et al., 1991; Wang et al., 1993) and 
look direction (Sun and Ranson, 1998), surface and environmental conditions (i.e., 
soil moisture) (Harrell et al., 1997; Kasischke et al., 2011), trunk dielectric properties 
(Way et al., 1991) and saturation (Dobson et al., 1992; Kasischke et al., 1997; Ranson 
 56 
 
and Sun, 1997).  Backscatter from vegetated areas depends on a combination of 
canopy structure and underlying surface conditions (Ranson and Sun, 2000).  The 
former is related to water content of the vegetation, while the latter is linked to soil 
moisture (SM) changes.  Both (vegetation and soil) are affected by environmental 
factors such as temperature and moisture that will affect the dielectric constants of the 
target. These effects (i.e., temperature and moisture) need to be removed before any 
further analysis of changes caused by forest disturbances. 
However, few studies have been focused on analyzing sensitivity of multi-
source SAR backscatter to changes of forest biomass after forest disturbances.   
Ranson and Sun (2000) reported up to a 7 dB change in backscatter coefficient 
between frozen and non-frozen conditions and suggested multi-year comparisons 
under similar temperature conditions should be made such during in the summer 
growing season.  In addition, saturation in SAR backscatter varies among different 
bands.  Dobson et al. (1992) indicated biomass saturation levels of ~200 Mg·ha-1 and 
~100 Mg·ha-1 for P-band and L-band polarimetric SAR (PolSAR), respectively.   
Imhoff (1995) reported that the radar signal saturation was ~100 Mg·ha-1 for the P-
band, ~40 Mg·ha-1 for the L-band, and ~20 Mg·ha-1 for the C-band by comparing 
SAR backscatter and forest biomass relationships over a tropical broadleaf evergreen 
forest in Hawaii and coniferous forest stands in North America and Europe.  Ranson 
and Sun (1997) indicated that SAR backscatter from AIRSAR and SIR-C/XSAR 
could provide estimates of biomass up to ~150 Mg·ha-1 with an average of 97 Mg·ha-
1.  Lucas et al. (2010) studied the relationship between PALSAR data and biomass in 
Australia, and concluded that PALSAR data acquired when surface moisture and 
 57 
 
rainfall are minimal allow better estimation of the biomass of woody vegetation and 
that retrieval algorithms ideally need to consider differences in surface moisture 
conditions and vegetation structure. 
This chapter will describe the analysis of the sensitivity of multi-sensor and 
multi-temporal SAR signatures to changes in forest biomass.  The objectives were to 
investigate: 1) a procedure to reduce the distortion in SAR backscatter caused by 
incidence angle and soil moisture; 2) the feasibility of cross-image calibration 
between multi-temporal and multi-sensor SAR data; and 3) the possibility of applying 
normalized backscatter to detect changes in biomass due to forest disturbance and 
post-disturbance recovery.  This chapter is organized as follows. First, the study area 
and data are described.  Then the factors that influence changes in SAR backscatter 
are described and discussed.  A cross-image normalization method is proposed to 
reduce the offset between multi-temporal and multi-sensor data.  The performance of 
our correction is evaluated and applied to selected stands after forest disturbance.  A 
biomass-backscatter regression model is developed from field measurements, and 
applied to map changes in forest biomass in the study area.  Finally, the significance 
and limitations of the results are discussed and conclusions are drawn. 
 58 
 
3.2 Data 
 
Figure 3-1. Map of study sites, and coverage of Landsat ETM+ scene (red dashed line, 
p011/r029), ALOS PALSAR FBD scene (blue solid line) and PLR scene (purple 
dashed line). (a) 0.5 ha FIA style plot in 1992 and 2003, (b) 1.0 ha field plot in 2009. 
 
3.2.1 SAR data 
All SAR data used in this study were collected during middle April to early 
October and by systems operating at the L-band (23.6 cm wavelength or 1.3 GHz 
frequency).  Key parameters for SAR data used in this study site (HF, Figure 3-1) are 
given in Table 3-1.  Airborne Synthetic Aperture Radar (AIRSAR) L-band data with 
full polarizations (i.e., HH, VV, HV, VH) and multi-look angles (i.e., target at 25°, 
35°, 45°) were collected in 1989 and 1994. Uninhabited Aerial Vehicle SAR 
(UAVSAR) data were gleaned for the study area on August 5th, 6th, 7th and 14th in 
2009, with looking angle ranges from 20° to 60°, and headings of 167⁰ and 347⁰.  
Meanwhile, Spaceborne Imaging Radar C- and X-Band SAR (SIR-C/XSAR) data 
with full polarizations at L-band and C-band were collected in April and October 
1994. Phased Array type L-band SAR (PALSAR) on board the Advanced Land 
 59 
 
Observing Satellite (ALOS) is an enhanced version of JERS-1 SAR instruments 
launched in 2006 with a 46-day recurrent period. PALSAR data were collected for 
study area from 2007 to 2010 with fine beam dual mode (FDB: HH, HV) and 
polarimetric mode (PLR: HH, HV, VH, VV).  The Japan Space Exploration Agency 
(JAXA) has reported a geometric accuracy of 70 m and a radiometric accuracy of 
0.22 dB for measurements done in the Amazon forest areas (Shimada et al., 2009).   
Detail information about the SAR data selected for this study are listed in Table 3-2, 
including acquisition date, center incidence angle, and environmental conditions such 
as temperature and precipitation. 
 
Table 3-1. Key instrument parameters of SAR system in this study  
Parameter AIRSAR UAVSAR 
SIR-C/   
XSAR 
JERS-1 
PALSAR 
FBD 
PALSAR 
PLR 
Platform Airborne Airborne Spaceborne Spaceborne Spaceborne 
Available date 1994 2009 1994 1995, 1998 2007-2010 
Frequency 1.26GHz 1.26GHz 1.26GHz 1.3 GHz 1.27 GHz 
Polarization HH, HV, 
VH, VV 
HH, HV, 
VH, VV 
HH, HV, 
VH, VV 
HH HH, HV HH, HV, 
VH, VV 
Look angle 25-64 25-65 17-63 35 10-50 
Swath width 15km 16km 15km 75km 70km 30km 
Pixel size 10m 6m 12.5m 18m 20m 30m 
*Pixel size for multi-looked ground-projected product. FBD=Fine beam dual; PLR=Polarimetric. 
Technical specs from https://www.asf.alaska.edu/sar-data  
 
 60 
 
 
Table 3-2. SAR data selected for sensitivity analysis  
Sensor Scene  
id  
Acquisition  
date 
Pixel  
size* 
  
(m)  
Incidence  
angle#  
 
(◦) 
Environmental conditions 
Temperature 
(°C) 
Precipitation  
(mm) 
3day 7day 14day 
SIR-C 
/XSAR 
PR12331 04/13/1994 12.5  31.7 T ~ 5.5 °C  17.3 42.1 50.9 
SIR-C 
/XSAR 
PR47494 10/04/1994 12.5  31.7 T ~ 10.1 °C  0 0.2 76 
AIRSAR 
 
/ 09/02/1989 10  35.0 T ~ 10.1 °C  16.8 18.6 27.6 
AIRSAR 
 
CM6221 10/07/1994 10  35.0 T ~ 10.1 °C  0 0.2 76 
UAVSAR 16702_ 
09054_016 
08/05/2009 6  48.0 T ~ 21.6 °C  11.5 49.6 94.4 
PALSAR 
/FBD 
ALPSRP 
191680890 
08/30/2009 20  34.3 T ~ 14.8 °C  20.4 32 34.2 
 
*Pixel size for the multi-looked ground-projected product. #incidence angles are at scene center; for technical specifications see 
https://www.asf.alaska.edu/sar-data  
 61 
 
 
3.2.2 Auxiliary data 
Local forest management maps, Landsat derived disturbance maps, and 
climate data were prepared as referenced dataset for the study site.  Specifically, 
forest management maps were digitized and composited from scanned American 
Forest Management (AFM) maps for the study site during 1980 to 2011 (Figure 3-2).  
Three management maps were used: 1) AFM forest property map (1980-2011), 2) 
Edinburg, and 3) Howland cut map (manage unit map with harvest dates) in 2000 
rectified to road map and satellite images with UTM/WGS84 map projection.  The 
products of yearly forest disturbance map from Landsat time series stack vegetation 
change tracker (LTSS-VCT) (Huang et al., 2010) were obtained for the study area 
(see details in Chapter 2). 
In addition, with an AmeriFlux tower in the HF study site, various climate 
data were collected from 1987 including daily air temperature, total precipitation, 
wind speed, etc.  The soil near the AmeriFlux tower consists with 50.35% sand, 33.75% 
silt, and 15.9% clay.  From autumn of 1999, soil moisture, temperature and salinity 
were collected by five Hydra-Vitel probes and thermistors buried at depths of ~5 cm, 
10 cm, 20 cm, 50 cm, and 100 cm.  Figure 3-3 shows the variation of water fraction 
by content (WFC, %) versus precipitation (mm) in 2009. 
 
 62 
 
 
Figure 3-2. Forest management map at the HF study site. Solid dark blue line polygon 
is the outline of the reserved area, and the solid green line is the near mature forest. 
PLT=plantation; CC=clear-cut; SC/STC=strip-cut; SEL=select-cut; SH=shelterwood-
harvest, 1 and 2 denotes 1st and 2nd entry of shelterwood-harvest; OB/OR= overstory 
removal; IMP=intensive management plot. Number inside the polygon denotes the 
year of management. 
 
Figure 3-3. Water fraction by Volume (WFV, %) and precipitation (mm) as a 
function of date. WFV measured at depths of 2 inches (~=5 cm). 
 63 
 
 
3.3 Method 
Sensitivity analysis in this study includes three steps: incidence angle effect 
reduction, reduction of soil moisture effect, and determination of sensitivity to forest 
biomass (Figure 3-4).  First, a theoretical model and field measurements are 
employed to illustrate relationship between SAR backscatter and forest biomass.  
Then, an empirical model is applied to reduce incidence angle effects on SAR 
backscatter .  The influence of soil moisture on SAR backscatter is analyzed through 
in-situ soil moisture and field biomass studies.  A cross-image normalization 
procedure is proposed to lessen the environmental effect on SAR backscatter.  Lastly, 
the sensitivity of SAR backscatter to biomass is determined using normalized SAR 
data and field measurements. 
 
Figure 3-4. Flowchart of SAR data processing and sensitivity analysis. 
 
 64 
 
3.3.1 Sensitivity of SAR backscatter to biomass 
We employed a theoretical model to show the sensitivity of L-band SAR 
backscatter to forest biomass.  ZELIG (Urban, 1990) is an individual tree-based 
forest gap model with the ability to simulate the forest ecosystem with complex 
species composition over a long-term period.  Driven by temperature and 
precipitation data, as well as growth and environmental response parameters of each 
tree species and soil characteristics (Levine et al., 1994), the model was 
parameterized for Howland Forest (Ranson et al., 2001).  The output of ZELIG model 
are forest stand parameters, including biomass density, forest height and leaf area 
index (LAI), and the size of each tree.  Then, a radar model that simulated L-band 
radar backscatter of PolSAR data was implemented based on the output of ZELIG 
model (Ni et al., 2013a; Sun et al., 1991).  A look-up table (LUT) was generated from 
the physical-based forest backscatter model and was used in this study.  
Based on model simulations as well as field measurements, the relationship 
between SAR backscatter and forest biomass is strong (Figure 3-5).  However, the 
SAR backscatter has a wider dynamic range when the level of biomass density is high. 
The simulated SAR attributes for co-polarization (HH) and cross-polarization (HV) 
were plotted as a function of simulated forest biomass from the look-up table (Figure 
3-5), where different soil types reflected differences in hydrological condition related 
to soil moisture and ground roughness (Levine et al., 1994; Ni et al., 2013a).  The 
SAR backscatter data from airborne (i.e., UAVSAR) and spaceborne observations 
(i.e., PALSAR) were plotted as a function of field measured biomass (Figure 3-5, in 
blue circles).  The figure shows that when biomass increases from 0 to 50 Mg·ha-1, 
 65 
 
the backscatter ranges from -20 dB to -13 dB (~7 dB) at HV polarization, and from -
14 dB to -9 dB at HH polarization (~5 dB).  This is consistent with conclusions in the 
aforementioned literatures that HV is more sensitive to forest disturbance than HH; 
and that HV is sensitive to structures of post-disturbance tree regrowth, while HH is 
more sensitive to moisture content of the soils.  
 
Figure 3-5. SAR backscatter as a function of forest aboveground biomass from model 
simulation and remote sensing observation. Simulation from ZELIG plotted with (a) 
PALSAR HH, (b) PALSAR HV, (c) UAVSAR HH, and (d) UAVSAR HV. ADAMS 
to WESTBURY are different type of soil based on drainage and taxonomic 
classification (Levine et al., 1994).  
 
The above analysis indicates that changes in forest biomass after disturbance 
are detectable from SAR backscatter.  However, many factors other than the change 
in forest structure and biomass, such as radar incidence angle and surface conditions, 
 66 
 
also affect the radar backscattering.  The influence of these factors will be discussed 
in the following sections. 
3.3.2 Incidence-angle based correction for airborne SAR backscatter 
Numerous studies have indicated that incident angle (IA) lead to changes in 
backscatter (Sun et al., 1991; Wang et al., 1993), and model-based correction was 
effective in reducing IA influence on SAR backscatter (Menges et al., 2001; Sun et 
al., 2002).  Thus for airborne SAR data (AIRSAR and UAVSAR), two set of 
empirical models are developed for IA correction:   
HH and HV polarization: ncocoraw )cos/(cos     Equation 3-1 
    VV polarization: ncocoraw )/(  
 
where oraw is the raw backscatter before correction, c  is the radar incidence angle at 
the center of the image, and   is the local incidence angle, and n is the power 
exponent coefficient with a range of [1, 2] determined by target attributes.  The local 
incidence angle is defined as the angle between the radar line-of-sight and the local 
vertical with respect to geoid. If topography is gentle (i.e., slope near to zero), then 
local incidence angle can be assumed equal to incidence angle.  
Then, the correction models are applied to all pixels: 
HH and HV polarization:  ncorawocorr )cos/(cos     Equation 3-2 
         VV polarization:  noocorr )/( craw  
  
where ocorr is the corrected radar backscatter coefficients. 
 
 67 
 
3.3.3 Sensitivity of SAR backscatter to soil moisture and cross-image normalization 
To study the effect of surface environments such as soil moisture (SM) on 
changes in SAR backscatter, relationship between SAR signatures and soil moisture 
for the period of 2007-2010 over the selected stands was analyzed.  Based on the 
forest management map, SAR signatures near the Main AmeriFlux tower (S1, near 
mature forest with biomass >200 Mg·ha-1) and near the non-forested area (S2, clear-
cut with biomass <10 Mg·ha-1) were extracted from PALSAR collected during the 
growth season (May to October in 2009).  The relationship between SAR signatures 
and soil moisture is analyzed for the period of 2007 to 2010 (Figure 3-6).  
A conceptual diagram of the cross-image normalization is shown in Figure 
3-6.  The existence of radar “saturation” and different mechanisms from soil and 
canopy vegetation were utilized to conduct the cross-image normalization for multi-
temporal and multi-sensor SAR data.  Model simulations (Ni et al., 2013a) have  
indicated that SAR backscatter from soil is stronger than that from canopy in low 
biomass areas (non-forest areas), but weaker in high biomass areas (near mature 
forest).  The objective of the normalization was to produce SAR signatures, acquired 
from different time and/or by different sensor, with similar patterns with regard to 
their SAR backscatter- biomass.   
 68 
 
 
 
The following two-step normalization was developed for both airborne and 
spaceborne SAR data.  First, the one-point normalization makes use the saturation 
point at higher levels of biomass (i.e., near-mature forest).  The first-step of  the 
normalization can be expressed as: 
ooo oSo tSoo   )/( 11       Equation 3-3 
where oo  (sigma naught) is the original SAR backscatter before normalization.  
Both o tS1 and o oS1 are the SAR backscatter of a near-mature forest (S1) from the 
target (S1t) or reference image and the original image (S1o) that need to be 
normalized.  The variables o tS2 and o oS2 are the SAR backscatter of the non-forested 
area (S2) from the target (S2t) or reference image and the image (S2o) that need to be 
normalized. 
Figure 3-6. Conceptual diagram of the cross-image normalization. Figure units are in 
power domain (m2·m-2) for backscatter (σ), and density per hectare (Mg·ha-1) for 
AGB. S1 is the saturation point in a near-mature forest; S2 is the maximum soil effect 
point in a non-forested area. Dashed blue line (Veg) is the fitted backscatter versus 
AGB received from vegetation canopy without soil influences in theory. Brown 
dashed and dotted lines are backscatter from soil surfaces. Green dotted lines are 
backscatter from vegetation canopy plus soil. ①②③ denotes backscatter from data 
with different surface and soil moisture conditions. 
 
 69 
 
Then, the two-point normalization is applied to reduce the environmental 
effect of the soil component by subtracting it from the total signature.  Normalized 
SAR backscatter after the second-step normalization can be expressed as:  
ba ooo
oo
o
o
o




      Equation 3-4 
where o is the difference between SAR backscatter at two different soil moisture 
conditions, and the coefficients a and b are derived from linear fitting of the data: 
oStSSoStSS
oSoSoSSSS
oSoSSS
b
a
222111
121211
2121
,
)/()(
)/()(






  Equation 3-5  
where 
t1S  and oS1  are the target and the original SAR backscatter, respectively, of 
mature forest (S1) from Equation 3-3.  The variables
t2S  and oS2  are the target and 
the original SAR backscatter of the non-forested area (S2) from Equation 3-3.  Finally, 
1S and 2S are the differences between SAR backscatter of near-mature forest 
and non-forested area at two different soil moisture conditions. 
Stands with select-cut and clear-cut trees were selected with assistance from 
forest management map (Figure 3-7a), to test the proposed normalization algorithm, 
and to analyze the  sensitivity of normalized changes in backscatter to forest biomass 
over time.  Figure 3-7a shows representative plots on the AIRSAR image (October 
7,1994, R: HH, G: HV, B: VV).  Figure 3-7b shows forest management map of 
logging activities from 1984 to 2011, where patches with different color denote types 
and years of disturbance.  To make the management map comparable to the SAR 
images, the disturbances were reclassified into plantation before 1994 (1984–1993), 
 70 
 
plantation after 1994 (1994–1995), natural regrowth after prior clear-cut, select-cut 
and strip-cut events (1984–1993), and recent select-cut (1994–2008). 
 
Figure 3-7. (a) Selected plots on AIRSAR image (10/07/1994, R: HH, G: HV, B: VV), and (b) 
Forest management map at study site. Pink polygon is the near-mature forest; and dark blue 
polygon is the outline of the reserved area. Solid line polygons with labels (i.e., PLT89 = 
plantation in 1989; CC86 = clear-cut in 1986; STC92 = strip-cut in 1992; SHL08 = 
shelterwood-harvest in 2008) are the plots selected for sensitivity analysis. S1 is the 
saturation point in a near-mature forest  (biomass >200 Mg·ha-1), S2 is the maximum soil 
effect point in a non-forested area (biomass <10 Mg·ha-1). 
3.4 Result 
3.4.1 Sensitivity of SAR backscatter to incidence angle 
Table 3-3. Correction model parameters and coefficients of determinant for three 
polarizations 
Polarization Correction Model n R
2
 
HH 
ncorawocorr )cos/(cos raw   
1.5940 0.9733 
HV 1.5250 0.9665 
VV nrawoocorr )/( craw    -1.3293 0.9777 
 
The correction models and coefficients of determination derived from 
UAVSAR data for three polarizations are given in Table 3-3.  Cosine models were 
 71 
 
applied to normalize backscatter at HH and HV polarizations, and linear model is 
selected for VV polarization.  
 
Figure 3-8. Mean backscatter in three polarizations (R:HH, G:HV, B: VV) as a function of 
incidence angle. AISSAR data acquired on 10/07/1994before (a) and after incidence angle 
normalization (b), UAVSAR data acquired on 08/05/2009 before (c) and after incidence 
angle normalization (d).   
 
A plot of SAR backscatter data as a function of incidence angle (Figure 3-8) 
illustrates the trend of SAR backscatter before and after the incidence angle based 
radiometric correction.  Before correction, there are significant decreasing trends 
along the incidence angle for both AIRSAR (Figure 3-8a) and UAVSAR data (Figure 
10 20 30 40 50 60 70
-25
-20
-15
-10
-5
0
Incident angle ()
S
i
g
m
a
 
(
d
B
)
 
 
HH
HV
VV
10 20 30 40 5 6 7
-25
-20
-15
-10
-5
0
Incident angle ()
S
i
g
m
a
 
(
d
B
)
 
 
HH
HV
VV
(d)  
10 20 30 40 50 60 70
-25
-20
-15
-10
-5
0
Incident angle ()
S
ig
m
a
 
(
d
B
)
 
 
HH
HV
VV
10 20 30 40 5 6 7
-25
-20
-15
-10
-5
0
Incident angle ()
S
ig
m
a
 
(
d
B
)
 
 
HH
HV
VV
(a)  (b)  
(c)  (d)  
 72 
 
3-8c).  After incidence angle correction, the overall trends of backscatter along the 
incidence angle were mostly removed (Figure 3-8b and Figure 3-8d).  Although there 
is slightly decreasing trend from 20° to 30°, the corrected values keep relatively 
stable (less than 1 dB variation) from 30° to 60° in all three polarizations (i.e., R: HH, 
G: HV, B: VV).  
The incidence angle contamination from both AIRSAR (Figure 3-10) and 
UAVSAR (Figure 3-9) can be seen clearly in the near range (i.e., left edge of images 
in the upper row), and there is a systematic decreasing from near range to far range.  
The effect of correction on AIRSAR image is very similar to that of UAVSAR 
(Figure 3-10 and Figure 3-9).  
 
 
Figure 3-9. Polarized AIRSAR image acquired on 1994/10/07 in three polarizations 
and composite (R: HH, G: HV, B: VV). The above images (a)-(d) show the image 
requested from JPL archive, and the below images (e)-(h) are after correction for 
incident angle. 
 73 
 
 
3.4.2 Sensitivity of SAR backscatter to soil moisture 
Two plots with field measurements taken in 2009 were selected to show the 
sensitivity of backscatter to soil moisture (Figure 3-11).  Soil water fraction by 
volume (WFV) data was collected at 5 cm (2 inches) depth from the Main Tower, 
Howland.  The SAR backscatter (sigma) was extracted from PALSAR data collected 
during the 2007 to 2010 grown seasons with incidence angle ranging from 37° to 40°.  
Scatter plots show an increase in backscatter with increasing soil moisture at HH 
polarization for the low biomass plot (Figure 3-11b).  A correction of radiometric 
distortion due to different soil moisture conditions is possible.  
Figure 3-10. Polarized UAVSAR image acquired on 2009/08/05 in three polarizations 
and composite (R: HH, G: HV, B: VV). The above images (a)-(d) show the calibrated 
image distributed by JPL, and the below images (e)-(h) are after correction for 
incident angle. 
 
 
 74 
 
 
3.4.3 Sensitivity of normalized SAR backscatter to forest biomass 
A two-step normalization was conducted to both airborne and spaceborne 
SAR data.  First, spaceborne SIR-C/XSAR data obtained on April 13 (PR12331) and 
October 4 (PR47494), 1994 were chosen to test the effectiveness of the proposed 
algorithm. Specifically, the SIR-C/XSAR data acquired on April was normalized to 
the data acquired in October.  Changes in backscatter from the two SIR-C/XSAR 
scenes were illustrated for the selected plots (Figure 3-12).  The patch of forest near 
the flux tower was selected as the saturation point (S1), and the clear-cut (CC86) with 
no canopy in 1994 was selected as the low biomass point (S2).  The two selected 
scenes had similar sensor conditions (i.e., same looking direction and incidence 
angle).  However, environmental conditions changed from April to October, including 
temperature increased from 5.5 °C to 21.6 °C, and an increase in surface soil moisture.  
The soil surface was very wet in April not only due to cumulative rainfall that fell 
three to seven days prior to the SAR acquisition, but also due to the presence of snow.   
While rainfall totals were higher in October two weeks prior to the SAR acquisitions 
(76 mm vs. 50.9 mm, Table 3-2), melting snow and lower air temperature in April 
(a) High biomass stand (b) Low biomass stand 
Figure 3-11. Mean backscatter (Sigma) plotted as a function of water fraction by volume 
(WFV, %): (a) High biomass stand, (b) Low biomass stand. 
 75 
 
would have resulted in higher surface moisture.   In fact, field crews in 1994 reported 
running water (small streams and ponds) at numerous locations in the forest.  Finally, 
deciduous trees had not yet leafed out in April, while deciduous trees had retained 
their leaves in October, suggesting that sap flow (and thus stem dielectric properties) 
would have been lower in April than in October. 
Before the normalization, we saw an increasing trend of backscatter (~2 dB) 
for all plots from April to October due to the changes in environmental conditions 
(Figure 3-12a).  After normalization, the trend was removed for most plots (i.e., S1, 
S2, and PLT89), and reduced for the select-cut plots (i.e., SHL08 and STC89) with 
subtle changes (<1dB).  The proposed two-step normalization algorithm successfully 
removed 50% to 100% of the effects of environmental conditions on SAR backscatter 
(Figure 3-13b).  Therefore, the following section analyzes the sensitivity of 
normalized changes in backscatter to forest biomass over time using multi-sources 
SAR data. 
Changes in multi-sensor SAR backscatter over time were analyzed for 
selected plots (Figure 3-13).  Airborne AIRSAR data from 1989 and 1994 were 
normalized to UAVSAR data in 2009 (Figure 3-13a).  Spaceborne SIR-C/XSAR data 
in Oct 1994 were normalized to PALSAR data in 2009 (Figure 3-13b).  As expected, 
the change in biomass with time varies for different types of forest disturbance. 
Whole-stand disturbance (plantation after clear-cut) before the 1990s could be 
effectively detected its backscatter signature.  Plantation (PLT89) and natural 
recovery (CC86) are good examples to show post-disturbance recovery from clear-cut 
events before the 1990s.   
 76 
 
On the other hand, changes after partial-stand disturbance (i.e., shelterwood-
harvest and strip-cut) could be detected by SAR signatures only under certain 
conditions.  The strip-cut stand, occurred in 1990s (STC92), shows a relatively slow 
recovery rate, which is not sensitive to SAR signatures.  However, the decrease of 
biomass from shelterwood-cut between 1994 and 2009 (SHL08) is detectable. For 
example, SHL08 means the 1st entry was in 1998 and the 2nd entry was in 2008, when 
each time ~1/3 of the total biomass was removed.  Previous model simulation (Sun 
and Ranson, 1998) showed that when radar looks perpendicular to the strip-cut 
direction, the radar signature doesn’t change significantly.  Meanwhile, the reason for 
radar not being able to detect the regrowth may be that radar signal after the strip-cut 
was still strong.  
 
Figure 3-12. Changes in SAR backscatter (HV) from SIR-C/XSAR 1994 at 
represented plots: (a) before normalization, (b) after normalization. 
 
 77 
 
 
3.5 Discussion 
Our objectives were to investigate the influence of incidence angle (IA), soil 
moisture (SM), and changes in forest biomass on SAR backscatter.  Knowledge from 
these analyzes forms the base for the cross-image normalization.  Results from both 
spaceborne- and airborne- systems demonstrated that normalization ensured the 
derived biomass of regrowth forests was cross-calibrated, making the detection of 
Figure 3-13. Original and normalized multi-sources backscatter (1989‒2009) from 
airborne- and spaceborne- SAR systems at represented stands: (a) and (c) are from 
original data, and (c) and (d) are from normalized data. Airborne data include AIRSAR 
89/09/02, 94/10/07, and UAVSAR 09/08/05. Spaceborne data include SIR-C/XSAR 
94/10/07 and PALSAR 09/08/30. 
 
 
 78 
 
biomass changes from multi-source SAR data possible. Several main factors lead to 
this conclusion. 
Results from airborne systems indicated that the influence of IA on SAR 
backscatter could be reduced to less than 1 dB using an empirical model (Figure 3-8).  
We chose an empirical method for incidence angle correction because it is efficient 
and only needs basic sensor parameters (i.e., near range distance, instrument height, 
incident angle at near range and slant range resolution).  Thus, this procedure was 
suitable for this study area, which has with a relatively flat topography and limited 
information in terms of sensor parameters in the archived data (AIRSAR).  However, 
a more advanced radiometric correction that integrated terrain variations (Small, 2011) 
should be applied in regions with large terrain variation to improve the precision of 
the correction. 
Second, the sensitivity of SAR backscatter to soil moisture was analyzed via 
spaceborne PALSAR data and in situ measurements from April 2007 to October in 
2010.  A cross-image normalization procedure was used to reduce the influence of 
environmental and acquisition conditions among multi-source SAR data.  This 
procedure successfully reduced the temporal changes in backscatter for SIR-C/XSAR 
data collected between April and October 1994 by 50% to 100% (Figure 3-12).  
Assuming that surface roughness was the same in April and October, increased 
surface wetness would have increased direct backscattering from soil and double-
bounce backscattering.  More sap flow and higher temperatures in October increased 
the dielectric constant of the tree trunks, increasing direct backscatter from the trunk 
and the trunk-ground double bounce.  A further application to multi-sensor SAR data 
 79 
 
indicated that the proposed procedure successfully reduced radiometric distortions 
due to different acquisition conditions (Figure 3-13).  We chose near-mature forest 
stands and non-forested area because they met the requirements for high and low 
biomass in the conceptual diagram (Figure 3-6Figure 3-6).  Future work needs to 
incorporate physical-based models and develop according normalization algorithms. 
In addition, the influence of incidence angle and soil moisture depends on the 
area and location of study area.  For instance, previous studies suggests that 
correction of incidence angle effect is essential for study area with a range of 
incidence angle greater than 10° across the landscape (Menges et al., 2001).  While 
the effects of soil moisture will become more important as a study area approaching 
national to continental scale, because of the increasing spatial variation in soil 
moisture. However, current soil moisture products derived from interpolation of 
meteorological station data or available satellite observation have limited spatial and 
temporal resolution.  For example, the Aqua Advanced Microwave Scanning 
Radiometer-EOS (AMSE-E) is with a nominal spatial resolution of 25 km (Lucas et 
al., 2010), which is beyond the scale of our study site.  The relationship between SAR 
backscatter and soil moisture for the forested and non-forested areas is still under 
exploration because of limited observational as well as high-temporal SAR data.  
Thirdly, the ability to detect forest biomass reduction and regrowth using SAR 
backscatter also depends on numerous factors such as the level of disturbance, model 
uncertainties, radar looking directions, forest structures, trunk dielectric properties, 
and tree species. Forest recovery from whole-stand disturbance (>50% removal), such 
as natural regrowth and plantation after clear-cut events, could effectively be detected 
 80 
 
by backscatter signatures.  Depending on the magnitude of change and radar looking 
direction, the recovery from partial-stand disturbance (shelterwood harvest and strip-
cut with <50% removal) was not always captured by SAR signatures.  Uncertainty in 
biomass estimation from single-date models will affect this ability.  We were able to 
detect changes for pixels with changes greater than 100 Mg·ha-1 or above 50% of 150 
Mg·ha-1, although with a ~50 Mg·ha-1 prediction error from the model.  This is 
consistent with our analysis of selected plots (Figure 3-13), indicating that the level of 
disturbance has an influence on SAR detection ability.  A previous study (Robinson et 
al., 2013) had mentioned the effect of forest structure and tree species composition on 
the biomass estimation.  The model developed for biomass mapping in this study did 
not take these effects into account because of a relatively simple species composition 
in the study area.  However, we still recommend considering structure and species 
effects in areas with more complicated composition. 
Lastly, previous studies have indicated that a combination of multiple 
observations could improve the accuracy of biomass estimation from SAR data 
because multiple images provide more samples and using them together could reduce 
the radar speckle noise (Englhart et al., 2011; Zhang, 2011).  For example, Zhang 
(2011) indicated that models developed using multi-direction SAR backscatter (i.e., 
left and right look) achieved the best performance in estimation of forest biomass.   
Moreover, Englhart et al. (2011) demonstrated that regression models developed 
from multiple observations with multiple frequencies such as SAR data from 
PALSAR (L-band) and TerraSAR-X (X-band) are more accurate than regression 
models developed from single observation.  There is a trade-off between pixel size 
 81 
 
and the ability to detect disturbance.  Therefore for biomass estimation, the data were 
resampled to a 100 m resolution, and a mean value was calculated for each cell so 
that it was large enough to reduce the speckle noise in the SAR data used in this study.   
Airborne SAR data (i.e., UAVSAR and AIRSAR) with higher spatial resolution 
revealed more detail than spaceborne data (i.e., PALASAR and SIR-C/XSAR) in 
terms of changes in forest biomass. 
3.6 Conclusion 
This chapter analyzed the sensitivities of multi-sensor and multi-temporal 
SAR signatures to changes in forest biomass.  Knowledge from these sensitivity 
analyses and corresponding correction is the base to the cross-image radiometric 
normalization.  Results from both spaceborne- and airborne- systems demonstrated 
the normalization ensured the derived biomass of regrowth forests were cross-
calibrated, and thus make the detection of biomass changes possible. 
Findings from this study indicate that the use of a correction model can reduce 
incidence angle effects on SAR backscatter to less than 1 dB, and that cross-image 
normalization can reduce the effect of soil moisture on changes in backscatter to less 
than 50%.  Thus, the changes in forest biomass greater than 100 Mg·ha-1 or above 
50% of 150 Mg·ha-1 are detectable using cross-normalized SAR data. 
 82 
 
Chapter 4  Mapping Biomass Change after Forest Disturbance 
from Multi-source SAR
3
                        
4.1 Introduction 
 A series of studies have utilized airborne instruments such as AIRSAR (Ranson 
and Sun, 1997; Saatchi et al., 2011a) and UAVSAR (Robinson et al., 2013) to estimate 
forest biomass; and spaceborne instruments such as PALSAR (Mitchard et al., 2009a; 
Santoro et al., 2010), to detect clear-cut areas, evaluate forest biomass (Lucas et al., 
2010), and retrieve soil moisture for bare land (Guo et al., 2013) and post-burned forest 
(Bourgeau-Chavez et al., 2013).  Specifically, L-band SAR imagery can be used to 
monitor the changes in biomass and carbon stocks for early stages of secondary 
succession (Ustin et al., 1991) and post-disturbance recovery (Hall et al., 2011).  Ustin et 
al. (1991) indicated that changes in biomass during the early stages of secondary 
succession could be monitored by L-band airborne SAR data for two datasets with an 
interval of five years.  Their study also demonstrated that cross-image normalization is 
feasible for radar data obtained through stable systems under similar conditions.  Ranson 
and Sun (1997c) developed a cube root regression model to estimate biomass density for 
a temperate forest using SIR-C/XSAR and AIRSAR data.  Salas et al. (2002) evaluated 
the spatial and temporal variability in JERS-1 data and characterized the overall structure 
of clearings and secondary vegetation with age information derived from Landsat.  
Baltzer et al. (2003) utilized multi-sensor L-band SAR data (i.e., spaceborne SEASAT 
and JERS-1) to map regrowth of a conifer plantation in the Thetford Forest, UK.  
Luckman et al. (1997) compared biomass estimated from spaceborne SAR instruments 
                                                 
3 The presented material is under preparation for publication: Huang, W., Sun, G., Ni, W., Zhang, Z., 
Dubayah, R. Mapping Biomass Change after Forest Disturbance using Multi-source SAR data.  
 83 
 
(SIR-C/XSAR and JERS-1) and concluded that L-band SAR data were suitable for 
detecting regeneration in tropical regions. Santoro et al. (2010) utilized airborne and 
spaceborne data such as PALSAR to detect clearcutting of a boreal forest from July 2007 
to October 2008 and found a temporal consistency in the time series of L-HV backscatter 
measurements and developed a detection algorithm for clear-cut areas.  
However, few studies quantify the changes in forest biomass using multi-source 
L-band SAR data, because of issues such as limited SAR data acquisition over time, 
changes in forest structure (Dobson et al., 1995), radar incidence angle (Sun et al., 1991; 
Wang et al., 1993) and look direction (Sun and Ranson, 1998), variations in surface 
conditions (Harrell et al., 1997; Kasischke et al., 2011; Lucas et al., 2010) and trunk 
dielectric properties (Way et al., 1991), and saturation (Dobson et al., 1992; Kasischke et 
al., 1997; Ranson and Sun, 1997).  These issues have been discussed in the last chapter.  
Sandberg et al. (2014) estimated biomass change in a hemi-boreal forest from 2007 to 
2010 using airborne P-band SAR data, and concluded that growth and thinning in a forest 
can be measured with 64 equivalent number of look (ENL) in SAR imagery and a 50% 
change in biomass.  Ahmed et al. (Ahmed et al., 2014) indicated that the combined error 
from field measurements and remote sensing could be as much as 100% at sub-plot level 
(0.25 ha) for Howland and Harvard Forest sites, and recommended at least a 1 ha (100 m) 
spatial resolution for mapping of forest biomass.   
In this chapter we will investigate: 1) what will be the proper scale for mapping 
biomass from SAR data, 2) what are the uncertainties in the mapping of forest biomass 
using data from spaceborne and airborne systems, and 3) at which level predictions are 
reliable in terms of the changes in biomass.  The changes in forest biomass for the study 
 84 
 
sites were analyzed using SAR data from spaceborne (1994‒2009) and airborne (1989‒
1994‒2009) SAR systems.  
4.2 Data 
Spaceborne and airborne SAR data (Figure 4-1) are employed to map the change 
in biomass over a 15-year period (1994‒2009) and a 20-year period (1989‒2009).   
Details about the study site, SAR data, and auxiliary data have been introduced in the last 
two chapters, and are not described herein.  The field data used for training and validation 
are summarized in Table 4-1. 
LTSS-VCT year of disturbance 
map (1984-2010)
SIRC/XS (94/10/04) PALSAR(2007-2010)
UAVSAR(09/08/05)AIRSAR (89/09/02) AIRSAR(94/10/07)
Clear-cut (1980s)
Stripe-cut (1990s)
Selective-cut (2000s)
JERS1(98/08/30)
SAR Data Timeline
Forest Disturbance
(Local management maps)    
Optical Observation
Airborne
Spaceborne      
 
Figure 4-1. Timeline of SAR data at HF site. Optical data are acquired during the leaf-on 
season, and SAR data are from L-band. Forest management maps are digitized from local 
management maps. 
 
Table 4-1. Summary of field estimated biomass in 2009 at different plot-level 
Plot-size N Mean Min Max 
ha 
 
Mg∙ha-1 Mg∙ha-1 Mg∙ha-1 
0.25 ha 115 141.1 0.3 361.4 
0.5 ha 57 140.3 0.6 316.0 
1.0 ha 24 143.3 1.0 278.9 
All 196 141.2 0.3 361.4 
 85 
 
 
4.3 Method 
The overall procedure consists of multi-source data processing, radiometric 
normalization, model development, and biomass mapping (Figure 4-2).    
 
Figure 4-2. Flowchart of SAR data preparation, normalization and mapping. 
4.3.1 Multi-source SAR data processing 
Current radar derived products for surface conditions such as soil moisture are 
still under development and are often with large uncertainties in estimates.  In order to 
limit the influence of surface conditions on the retrieval of forest biomass from radar 
backscatter attributes, SAR images with very low or no precipitation within three days 
around acquisition date were selected for further analysis.  
Archived AIRSAR data in the 1990s were provided by JPL in the compressed 
strokes matrix (CM) format (JPL, 2003).  The 4 x 4 Stokes matrix data was compressed 
into 10-byte format, to reduce the original 1.92 GB data to 37.5 Mbytes per scene 
(Dubois and Norikane, 1987).  SIR-C/XSAR data in 1994 were provided in the 
compressed SIR-C format (JPL, 1994) with a CEOS header file.  For both AIRSAR and 
Normalization and Mapping
Regression 
models
Local relationship 
developed between 
AGB and SAR 
backscatter
Two-steps
Radiometric
Normalization
Normalized 
Airborne SAR 
{HH, HV, VV}
(89-94-09)
Normalized 
Spaceborne 
SAR {HH, HV, 
/VV} (94-09)
Stacked 
Airborne 
SAR
{HH, HV, VV}
Stacked 
Spaceborne 
SAR 
{HH,HV,/VV}
Geometric 
co-registration
PALSAR* 
HH, HV, /VV
AIRSAR*
HH, HV, VV
SIRC* 
HH, HV, /VV
UAVSAR* 
HH, HV, VV
* All SAR data are Geocoded and Radiometric calibrated;
** AGB range is [0, 250]
*** Forest disturbance and regrowth (<150 Mg ha
-1
) 
AGB maps**
AGB change 
maps***
 86 
 
SIR-C/XSAR systems, multi-looked complex (MLC) data were used in this study.  The 
compressed AIRSAR and SIR-C/XSAR data were processed to coherency elements (T3) 
from quad polarimetric data and to covariance elements (C3) from dual polarimetric data.  
These datasets were processed using the PolSARPro Polarimetric SAR Data Processing 
and Educational Tool developed by the European Space Agency (Version 4.2.0)  (Pottier 
et al., 2009).  
Slant range AIRSAR layers were converted to ground range images using “Slant 
Range Correction” function in ENVI 4.8.  This step converts the distance from the radar 
(slant range, ϒSR) to distance along the ground (i.e., ground range, ϒGR) and samples data 
to a desired output pixel spacing as follows:  
șsin/SRGR           Equation 4-1 
where θ is the incidence angle.  
Then, the ground range layers of AIRSAR and SIR-C were manually registered to a 
Landsat7 ETM+ image (path011/row029) acquired on Sept 9, 2000, using a first-degree 
polynomial model. 
PALSAR operates in several modes, including two fine-beam modes: single 
polarization (FBS) and dual polarization (FBD); and a full polarization mode (PLR). The 
FBD mode with HH and HV data at the 34.3° incidence angle (image center) were used 
in this study.  PALSAR FBD and PLR data from 2007 to 2009 were acquired at Level 
1.1/1.5 and were processed to normalized radar cross section (NRCS). These datasets 
were processed using the MapReady Remote Sensing Toolkit (Version 3.1.22) (Gens and 
Logan, 2003) supported by Alaska Satellite Facility (ASF) and the terrain information 
 87 
 
from National Elevation DEM (NED).  NRCS of the polarization component is derived 
from following equations with a single calibration factor (CF): 
L1.5 product:  NRCS (dB) = 10log10(<DN
2>) + CF   Equation 4-2 
 
L1.1 product:  NRCS (dB) = 10log10(<R
2 + I2>) + CF - 32.0  Equation 4-3 
 
where DN is the digital number value (amplitude) in the image; R and I are intensity and 
phase values; and CF is listed in Table 4-2 provide by JAXA. 
(http://www.eorc.jaxa.jp/en/about/distribution/info/alos/20090109en_3.html).  
Table 4-2. The calibration factor (CF) for ALOS/PALSAR data 
Processing  
date 
Before  
Jan. 8, 2009 
After  
Jan. 9, 2009 
FBD34.3 HH -83.2 -83 
FBD34.3 HV -80.2 -83 
PLR21.5 -83.4 -83 
 
UAVSAR polarimetric (PolSAR) data in 2009 were acquired from JPL in ground 
range grid (GRD) format in geographic coordinate system (WGS 84).  These datasets 
were then projected into a common frame of reference (UTM 19N WGS 84) used in this 
study.  
 
 88 
 
Table 4-3. Co-registration accuracies for the selected SAR data the test sites  
   
RMS 
Sensor Scene id 
Acquisition 
date #Point X Y Resolution (m) 
PALSAR/FBD ALPSRP198390890 10/15/2009 31 0.4510 0.4045 12.5 
PALSAR/FBD ALPSRP191680890 8/30/2009 29 0.4005 0.3541 12.5 
PALSAR/PLR ALPSRP179280900 6/6/2009 57 0.4575 0.3957 12.5 
PALSAR/PLR ALPSRP172570900 4/21/2009 76 0.4240 0.4826 12.5 
PALSAR/FBD ALPSRP131290890 7/12/2008 31 0.4985 0.4066 12.5 
PALSAR/FBD ALPSRP124580890 5/27/2008 46 0.4179 0.3250 12.5 
PALSAR/FBD ALPSRP120350890 4/28/2008 38 0.3237 0.3979 12.5 
PALSAR/FBD ALPSRP091030890 10/10/2007 30 0.3866 0.3505 12.5 
PALSAR/FBD ALPSRP086800890 9/11/2007 41 0.4251 0.4156 12.5 
PALSAR/FBD ALPSRP084320890 8/25/2007 41 0.4193 0.4483 12.5 
PALSAR/FBD ALPSRP080090890 7/27/2007 30 0.4840 0.4593 12.5 
PALSAR/FBD ALPSRP077610890 7/10/2007 39 0.4495 0.4778 12.5 
PALSAR/FBD ALPSRP073380890 6/11/2007 37 0.4806 0.4431 12.5 
PALSAR/PLR ALPSRP071920900 6/1/2007 47 0.4185 0.4019 12.5 
PALSAR/FBD ALPSRP068932710 5/11/2007 65 0.4345 0.4079 12.5 
PALSAR/FBD ALPSRP065210900 4/16/2007 71 0.4477 0.4872 12.5 
SIR-C/XSAR PR12331 4/13/1994 32 0.2816 0.4851 30 
SIR-C/XSAR PR47494 10/4/1994 22 0.3878 0.4410 30 
AIRSAR CM6220 10/7/1997 31 0.3549 0.3759 10 
*PALSAR data used in this study involved two modes: FBD= Fine beam dual mode, and PLR=Polarimetric mode. 
 
 
 
 89 
 
 
Finally, all SAR data including AIRSAR, SIR-C/XSAR, PALSAR were co-
registered to the same Landsat image (September 09, 2000, path011/row029,) using a 
modified ROI_PAC script for precise co-registrations (Ni et al., 2013b).  The registration 
errors (RMS) were minimized (<1 pixel) to reduce the influence of geolocation errors in 
SAR data (Table 4-3).  All conversions were processed in power domain (m2 m-2), and 
the arithmetic mean values were used when aggregating SAR data to different scales.   
Backscatter coefficients were converted into dB domain for biomass estimation: 
)m(mı(dB)ı 2210log10        Equation 4-4  
 
Airborne SAR data (i.e., AIRSAR and UAVSAR) were normalized by the 
correction models as described in Chapter 3 to reduce the effect of incidence angle. 
4.3.2 Modeling approaches for biomass estimation 
SAR backscatter coefficients, band ratios, and relevant indices are found to be 
sensitive to canopy structural variables and biomass (Gonçalves et al., 2011; Henderson 
and Lewis, 1998; Pope et al., 1994; Saatchi et al., 2010).  Specifically, Table 4-4 
summarizes the indices used in this study as predictors in biomass modeling.  We utilized 
indices including single polarization backscattering coefficients (i.e., HH, HV, and VV), 
cross-polarized indices, and full-polarized indices.  The single-polarized indices are 
applicable for all SAR data in this study, while the full-polarized indices are not available 
for either PALSAR FBS or PALSAR FBD due to the lack of full-polarization (absent HV 
and/or VV).  All statistical calculations were performed using R with packages ‘caTools’, 
‘forecast’ and ‘caret’ (Liaw and Wiener, 2002).  
 90 
 
Table 4-4. SAR backscatter coefficient, cross- and full- polarized indices 
Attributes Equation Function and reference 
ok  
)(10log20][
10 10
][
jk
o
k
dB
FSdB
o
k


 Backscattering coefficient*  
Henderson and Lewis (1998)  
   
R
cross
 
HH
HV
0
0


 
Cross-polarized ratio 
Henderson and Lewis (1998) 
RFDI 
HVHH
HVHH
00
00




 
Radar forest degradation index  
Saatchi et al. (2010)  
   
BMI 
2
00 VVHH  
 
Biomass index 
Pope et al. (1994) 
VSI 
BMIoHV
o
HV

  
Volume scattering index 
Pope et al. (1994) 
   
*Expressed in the intensity format. |Sk| is the amplitude response (k = HH, VV, or HV) 
and Fj is a calibration factor for column j, determined from data collected over the 
trihedral corner reflectors (Gonçalves et al., 2011). 
  
 91 
 
Forest biomass values are related to SAR backscatter ( op
) as: 
)( oe pbaAGB      Equation 4-5 
where op
 represents backscatter value (i.e., sigma naught) at different polarizations (i.e., 
HH and HV) and a and b are the exponential coefficients.  To improve the fitting process, 
we estimated coefficients using linear regression: 
oy pba       Equation 4-6 
where y=ln(AGB) and a and b are the linear coefficients.  
4.3.3 Accuracy assessment of regression models 
We assessed the regression models using leave-one-out cross-validation (LOOCV) 
performed with R (Version 2.14.2) (Kuhn, 2008).  The predicted values were regressed 
against the field-measured values to quantify the accuracy of the model using statistical 
indicators such as root mean square error from cross-validation (RMSEcv), relative error 
of RMSEcv (RMSEcv%), and coefficient of determination (R2).  
The LOOCV is an effective solution to evaluate the regression models when there 
are a small number of samples to test the model (Zhao and Popescu, 2009). The general 
procedure of LOOCV is described as follows:  
1) Select observation i from a test set (i.e., n independent observations y1, …, yn) 
and fit the model using the remaining data. Then compute the predicted residual for the 
omitted observation: 
ii*i yye        Equation 4-7 
2) Repeat step 1 for i=1, …, n. 
3) Compute the RMSE from ** ee n1 ，， , which is called cvRMSE . 
 92 
 
The ratio of 
cvRMSE to the mean observed value is also calculated:  
y
RMSEcvRMSEcv (%)
    Equation 4-8 
where y is the mean biomass. 
 
4.3.4 Mapping forest biomass and changes 
The mapping procedure consists of four steps (Figure 4-2): 1) normalize the SAR 
data; 2) develop the prediction model; 3) evaluate the model performance; and 4) map the 
biomass and changes. 
To normalize the SAR data, a two-step radiometric normalization was conducted 
to spaceborne and airborne SAR data.  Specifically, airborne AIRSAR data in 1989 and 
1994 were normalized to UAVSAR data in 2009, and spaceborne SIR-C/XSAR data in 
1994 were normalized to PALSAR data in 2009.  Then, the developed biomass-
backscatter regression models were applied to multi-year SAR data to generate biomass 
maps for spaceborne and airborne data from 1989, 1994 and 2009 for the study area.  
Finally, maps of the changes in biomass were created for 15- and 20-year periods from 
airborne SAR data, and for a 15-year period from spaceborne SAR data. 
 
4.4 Result 
4.4.1 SAR backscatter sensitivity to biomass 
The backscatter-biomass regression models show strong correlation across 
different spatial scales (Table 4-5).  The SAR backscatter values at all polarizations (HH, 
HV, and VV) were regressed to field-measured biomass at three scales (0.25 ha, 0.5 ha, 
 93 
 
and 1.0 ha).  At all polarizations, the R2 of this relationship increased as the plot size 
increased, where the least variability was observed at 1.0 ha plot-scale.  This biomass-HV 
relationship had the highest coefficient of determinations (R2) at all spatial scales for 
UAVSAR (0.47, 0.48, and 0.60) and PALSAR (0.50, 0.57, and 0.68) as plot size 
increased.  The biomass-backscatter relationship derived from airborne UAVSAR data is 
stronger than that from spaceborne PALSAR data.  This is not unexpected, given the 
finer spatial resolution and the higher signal to noise ratio in airborne data.  Since more 
equivalent ‘looks’ average values over field plots, there was a better agreement at 1.0 ha 
plot-scale despite the variability at 0.25 ha plot-scale. 
 
4.4.2 Evaluation of prediction models 
The predictions models from PALSAR 2009 data were evaluated across three 
plot-levels (Table 4-5, Table 4-6, and Table 4-7).  The best model at all three plot-levels 
is the HV model, and overall model accuracy improved with larger plot size.  The best 
single-term model is the HV model at 1.0 ha plot-level with the highest explanation of 
total variance (68%) and the lowest RMSECV of 45.5 Mg∙ha
-1 (31.7%).   
The predictions models from UAVSAR data in 2009 were evaluated across three 
plot-levels: 1) 0.25 ha, 2) 0.5 ha, and 3) 1.0 ha, respectively (Figure 4-3, Figure 4-4, 
Figure 4-5, and Figure 4-6).  This is similar to the trends observed from the PALSAR 
data.  As expected, the best model at all three plot-levels is the HV model with higher 
explanation of total variance (R2), and lower RMSECV.  The overall model performance 
improved as the plot-size increased.  The HV model at 1.0 ha (equivalent to 100 m pixel 
size) explained 60% of the total variance and had a RMSECV of 49.0 Mg∙ha
-1 (34.2%).  
 94 
 
Moreover, the best multi-variable model is the HH+HV+VV model at 1.0 ha plot-level 
with the highest explanation of total variance (72%) and the lowest RMSECV of 40.9 
Mg∙ha-1 (28.5%).  
Therefore, the multi-variable models (Table 4-7) derived from PALSAR data (HH 
and HV) and UAVSAR data (HH, HV, and VV) at 1.0 ha plot-level were selected to 
predict biomass.  The selected models were applied to all SAR data (i.e., UAVSAR, 
AIRSAR, PALSAR, and SIR-C/XSAR) and the biomass maps were generated at 100 m 
spatial resolution. 
  
 95 
 
 
Table 4-5. Summary of single term regression models at different plot-level 
Instrument 
/Year 
Variable 
Plot-size 
a b R2 
RMSEcv RMSEcv 
(ha) (Mg∙ha-1) (%) 
PALSAR 
2009 
HH 
0.25 8.37±0.88 0.56±0.13 0.24 75.9 53.8% 
0.50 8.65±1.25 0.60±0.18 0.26 72.0 51.3% 
1.0 9.26±2.06 0.69±0.31 0.34 65.6 45.8% 
HV  
0.25 11.64±0.83 0.57±0.07 0.50 63.0 44.6% 
0.50 11.79±1.10 0.59±0.09 0.57 53.8 38.4% 
1.0 12.37±1.54 0.64±0.13 0.68 45.5 31.7% 
UAVSAR 
2009 
HH 
0.25 8.11±0.48 0.63±0.08 0.20 91.2 64.6% 
0.50 8.24±0.66 0.66±0.12 0.19 84.4 60.2% 
1.0 8.63±0.95 0.72±0.17 0.27 78.1 54.5% 
HV 
0.25 11.16±0.50 0.57±0.04 0.47 59.5 42.2% 
0.50 11.27±0.62 0.57±0.05 0.48 55.8 39.7% 
1.0 11.42±0.82 0.59±0.07 0.60 49.0 34.2% 
VV 
0.25 10.26±0.58 0.75±0.08 0.44 67.2 47.6% 
0.50 10.44±0.75 0.78±0.10 0.45 62.0 44.2% 
1.0 10.87±1.03 0.83±0.14 0.64 46.9 32.7% 
* bolded is the best models; equations in form of AGB=exp(a+bx); 
  
 96 
 
 
Table 4-6. Summary of single index regression models at different plot-level 
Instrument 
/Year 
Variable 
Plot-size 
a b R2 
RMSEcv RMSEcv 
(ha) (Mg∙ha-1) (%) 
PALSAR 
2009 
Rcross  
0.25 8.90±0.94 0.78±0.17 0.15 103.6 73.4% 
0.50 9.84±1.33 0.94±0.24 0.27 90.3 64.4% 
1.0 10.81±1.98 1.13±0.36 0.30 93.1 64.9% 
RFDI 
0.25 1.98±0.82 -1.03±0.31 0.09 132.7 94.0% 
0.50 1.03±1.24 -1.41±0.47 0.23 107.1 76.3% 
1.0 -0.08±2.03 -1.82±0.77 0.25 119.5 83.4% 
UAVSAR 
2009 
Rcross 
0.25 11.32±1.18 1.10±0.19 0.30 92.7 65.6% 
0.50 11.84±1.64 1.19±0.27 0.37 97.9 69.8% 
1.0 13.20±2.42 1.42±0.40 0.41 92.3 64.4% 
RFDI 
0.25 0.22±0.95 -1.98±0.42 0.26 116.8 82.7% 
0.50 -0.03±1.33 -2.10±0.59 0.33 121.6 86.7% 
1.0 -1.44±2.13 -2.72±0.94 0.35 126.0 87.9% 
BMI 
0.25 9.26±0.51 0.73±0.08 0.28 85.4 60.5% 
0.50 9.43±0.67 0.76±0.10 0.28 77.2 55.0% 
1.0 9.80±0.92 0.81±0.14 0.40 66.4 46.3% 
VSI 
0.25 16.02±1.38 1.79±0.22 0.28 74.1 52.5% 
0.50 16.39±1.86 1.85±0.29 0.32 66.2 47.2% 
1.0 17.57±2.41 2.04±0.38 0.42 62.3 43.5% 
* bolded is the best models; equations in form of AGB=exp(a+bx); 
Rcross=HV/HH, RFDI=(HH+HV)/(HH-HV), BMI=(HH+VV)/2, VSI=HV/(HV+BMI). 
 
 
 
 
 
 
 
  
 97 
 
 
 
 
Table 4-7. Summary of multi-variable regression models at different plot-level 
Instrument 
/Year 
Variable 
Plot-
size a b c d R2 
RMSEcv RMSEcv 
(ha) (Mg∙ha-1) (%) 
PALSAR 2009 HH, HV 
0.25 11.84±0.82 -0.20±0.15 0.70±0.12 
 
0.46 68.2 48.3% 
0.5 12.14±1.07 -0.32±0.22 0.79±0.17 
 
0.56 59.2 42.2% 
1.0 12.77±1.41 -0.42±0.33 0.90±0.24 
 
0.62 55.8 38.9% 
UAVSAR 2009 
HH, HV 
0.25 11.92±0.61 -0.26±0.14 0.75±0.11 
 
0.56 53.2 37.7% 
0.5 12.10±0.77 -0.29±0.18 0.79±0.14 
 
0.61 47.5 33.8% 
1.0 12.34±1.08 -0.34±0.29 0.83±0.22 
 
0.70 41.9 29.2% 
HH, 
HV,VV 
0.25 11.90±0.62 -0.26±0.14 0.72±0.16 0.04±0.16 0.56 52.9 37.5% 
0.5 12.05±0.77 -0.28±0.18 0.70±0.20 0.11±0.20 0.62 46.7 33.3% 
1.0 12.28±1.13 -0.32±0.31 0.76±0.35 0.09±0.32 0.72 40.9 28.5% 
*bolded is the best models; equations in form of AGB=exp(a+bx1+cx2+dx3); 
 
 
 98 
 
 
Figure 4-3. Model evaluation across different scales for UAVSAR 2009 data: HH and 
HV polarizations. UAVSAR 0.25 ha, 0.5 ha, and 1.0 ha plot-levels: (a) 0.25 ha HH, 
(b) 0.25 ha HV (c) 0.25 ha Rcross, (d) 0.25 ha RFDI, (e) 0.5 ha HH, (f) 0.5 ha HV (g) 
0.5 ha Rcross, (h) 0.25 ha RFDI, (i) 1.0 ha HH, (j) 1.0 ha HV (k) 1.0 ha Rcross, and 
(l) 1.0 ha RFDI.  X axis is the field-measured AGB density (Mg∙ha-1). Y axis is 
predicted AGB density (Mg∙ha-1). Red solid line is 1:1 line.  
 
 99 
 
 
Figure 4-4. Model evaluation across different scale for UAVSAR 2009 data: VV 
polarization. UAVSAR 0.25 ha, 0.5 ha, and 1.0 ha plot-levels: (a) 0.25 ha VV, (b) 
0.25 ha BMI (c) 0.25 ha VSI, (d) 0.5 ha VV, (e) 0.5 ha BMI, (f) 0.5 ha VSI, (g) 1.0 ha 
VV, (h) 1.0 ha BMI, and (i) 1.0 ha VSI.  X axis is the field-measured AGB density 
(Mg∙ha-1). Y axis is predicted AGB density (Mg∙ha-1). Red solid line is 1:1 line. 
 100 
 
 
 
 
Figure 4-5. Model evaluation across different scale for PALSAR 2009 data: HH and 
HV polarizations. PALSAR 0.25 ha, 0.5 ha, and 1.0 ha plot-levels: (a) 0.25 ha HH, (b) 
0.25 ha HV (c) 0.25 ha Rcross, (d) 0.25 ha RFDI, (e) 0.5 ha HH, (f) 0.5 ha HV (g) 0.5 
ha Rcross, (h) 0.25 ha RFDI, (i) 1.0 ha HH, (j) 1.0 ha HV (k) 1.0 ha Rcross, and (l) 
1.0 ha RFDI.  X axis is the field-measured AGB density (Mg∙ha-1). Y axis is 
predicted AGB density (Mg∙ha-1). Red solid line is 1:1 line. 
 101 
 
   
Figure 4-6. Model evaluation across different scale for PALSAR 2009 data: HH and 
HV polarizations. PALSAR 0.25 ha, 0.5 ha, and 1.0 ha plot-levels: (a) 0.25 ha HH, (b) 
0.25 ha HV (c) 0.25 ha Rcross, (d) 0.25 ha RFDI, (e) 0.5 ha HH, (f) 0.5 ha HV (g) 0.5 
ha Rcross, (h) 0.25 ha RFDI, (i) 1.0 ha HH, (j) 1.0 ha HV (k) 1.0 ha Rcross, and (l) 
1.0 ha RFDI.  X axis is the field measured AGB density (Mg∙ha-1). Y axis is predicted 
AGB density (Mg∙ha-1). Red solid line is 1:1 line. 
 102 
 
 
4.4.3 Biomass mapping from multi-source SAR data 
Exponential regression models were derived from field-measured biomass at the 
1.0 ha plot-level and SAR backscatter at different polarizations.  Specifically, two sets of 
models were developed: the PALSAR models for spaceborne systems (i.e., SIR-C and 
PALSAR), and the UAVSAR models for airborne systems (i.e., AIRSAR and UAVSAR).  
The models were applied to the corresponding SAR data, and biomass maps were 
generated at 100 m spatial resolution for the study site.  
Biomass maps from SAR data were produced using the selected regression model 
(Figure 4-7a-d).  The biomass map from LVIS data (Figure 4-7f) was produced using the 
RH50 regression model as described in Chapter 2.  The biomass ranges up to 250 Mg∙ha-
1, with a color of orange (0 - 50 Mg∙ha-1) to dark green (>250 Mg∙ha-1) indicating an 
increase of biomass.  The overall patterns are similar in maps generated from spaceborne 
and airborne SAR data in 1994 and 2009.  In both 1994 and 2009, high values of biomass 
(>200 Mg∙ha-1) were located in the near mature forests (outlined by pink polygon).  
Regions north of this undisturbed forest had lower values of biomass in 2009, 
corresponding to the disturbance detected by LTSS-VCT product.  These patterns in 2009 
were similar to the map derived from waveform LiDAR data (i.e., LVIS in Chapter 2). 
 
 103 
 
 
Figure 4-7. Forest biomass maps at 100 m resolution (1994-2009). 
 
 
Figure 4-8. Change in biomass at HF site via cross-image normalized multi-source SAR 
data: (a) 2009PALSAR-1994SIR-C, (b) 2009UAVSAR-1994AIRSAR, and (c) 
2009UAVSAR-1989AIRSAR. Maps are at 100 m spatial resolution, and with confidence 
interval (95%) is applied in predictions. 
 
  
 104 
 
4.4.4 Biomass change mapping  
Changes in biomass (green to red color) from 1994 to 2009 were derived at 100 m 
spatial resolution (Figure 4-8a & b). Confidence intervals (CI, 95%) of prediction were 
generated to exclude the predictions with overlapped values in two periods.  Spatial 
patterns of changes in biomass derived from spaceborne and airborne systems are 
consistent with the forest disturbance patterns detected by the LTSS-VCT product 
(Figure 4-7c).  Specifically, both change maps show the forest disturbance as well as the 
post-disturbance regrowth (i.e., highlighted by red and green colors in Figure 4-7e).  In 
addition, results from airborne systems (UAVSAR 2009 and AIRSAR 1994) reveal more 
detailed changes in biomass than those from spaceborne systems (PALSAR 2009 and 
SIR-C 1994).   
4.5 Discussion 
Our results highlight three important issues concerning the mapping of biomass 
and changes in biomass using L-band SAR data: (1) the effect of map scale on model 
accuracy and the errors from radar backscatter, (2) the effect of modeling approaches on 
the biomass estimation, and (3) the effect of errors in in-situ estimates of biomass. 
4.5.1 Mapping scale and errors in radar data  
Speckle noise in radar data is a critical factor affecting the application of SAR 
data in mapping of forest biomass.  The improved accuracy in prediction models with the 
increase of plot sizes from 0.25 to 1.0 ha, was largely due to the reduction in geolocation 
errors and speckle noise.  The decreasing trends in RMSECV with increase of plot size are 
consistent in both spaceborne and airborne SAR systems.  Previous studies (Luckman et 
 105 
 
al., 1998; Saatchi et al., 2011a; Sumantyo and Amini, 2008) have indicated that to reduce 
the effect of speckle noise on biomass estimation, it is required to filter or aggregate the 
data to a certain map scale.  The optimal pixel size or the equivalent number of multi-
look (ENL) for biomass mapping from radar data is system dependent because nominal 
spatial resolutions vary among different SAR systems.  For example, to develop a robust 
biomass prediction model, spaceborne PALSAR with an nominal ground range resolution 
of 12.5 m, is suggested to be aggregated to at least 75 m (~0.5 ha) (Sun et al., 2011). 
While airborne UAVSAR, with a finer ground range resolution of 5 m, could be used at 
50 m spatial resolution (0.25 ha) (Zhang, 2011).  
Multi-temporal and multi-looking-angle SAR data are found to be efficient in 
improving the accuracy in biomass estimates, because the averaging process reduces 
sensor calibration errors and influences from environment.  However, the variances from 
multi-temporal data are uncertain and the underlining mechanisms are complicated 
because the changes in SAR backscattering depend on soil moisture and dielectric 
conditions which have not been well quantified spatially or temporally. 
In addition, simulated errors (i.e., 1 dB and 0.5 dB) from SAR data show that 
most of the estimated biomass from field measurement is within the error zones of HV 
model for PALSAR (Figure 4-9 & c) and UAVSAR (Figure 4-9 b & d).  Results indicate 
that 0.5 dB or better radiometric calibration accuracy is required to generate a reliable 
HV model for biomass estimation.  Recent spaceborne SAR systems such as PALSAR 
with a radiometric accuracy of 0.22 dB in the Amazon forest areas (Shimada et al., 2009), 
would meet for this requirement.  However, the archived AIRSAR and SIR-C/XSAR data 
does not have a high enough radiometric accuracy.  Therefore, mapping changes in 
 106 
 
biomass using a combination of these datasets would have inherent errors from the 
remotely sensed data. 
Due to limited co-incidence field measurements in our study site, we could not 
develop the direct change model as suggested in the previous study (Dubayah et al., 
2010).  Instead, 0.25 to 1.0 ha plots were selected to develop uniform biomass equations 
using LOOCV method for spaceborne and airborne SAR data.  The L-band airborne SAR 
data employed in this study were acquired using similar instruments (AIRSAR and 
UAVSAR).  The sensor calibration accuracy improved over time for airborne as well as 
spaceborne instruments.   
4.5.2 Errors in prediction models 
Modeling approaches have influence on the values and distributions of predicted 
biomass.  A 95% confidence interval of prediction is used here to present a range for the 
mean in the exponential HV models (blue dashed line in Figure 4-10).  The confidence 
intervals are narrow at the lower range of biomass (i.e., 0˗150 Mg∙ha-1), however getting 
wider at the higher level of biomass (i.e., >150 Mg∙ha-1). 
 
 107 
 
  
Figure 4-9. Simulated errors from SAR datasets: (a) PALSAR 2009 HV with ±1 dB 
error, (b) UAVSAR 2009 HV with ±1 dB error, (c) PALSAR 2009 HV with ±0.5 dB 
error, and (d) UAVSAR 2009 HV with ±0.5 dB error. Red solid line is the fitted 
regression model, and blue dashed lines show the SAR calibration error interval. 
 
±1 dB ±1 dB 
±0.5 dB ±0.5 dB 
 108 
 
 
Figure 4-10. Errors from regression models: (a) PALSAR 2009 HV, (b) UAVSAR 2009 
HV. Y axis is the field measured AGB density (Mg∙ha-1) at 1.0 ha plot-level. X axis is 
SAR backscatter at HV polarization in decibel (dB). Red solid line is the prediction curve, 
and blue dashed lines are the 95% confidence intervals of predictions.  
 
The model evaluations show that HV models are the best among single-term 
models; and that HH+HV+VV models are the best among multi-variable models; and 
models that use dual or full polarization perform better than single polarization in 
estimation of biomass.  The HV model is the best single-term prediction model for both 
spaceborne (i.e., PALSAR) and airborne (i.e., UAVSAR) SAR systems. The performance 
of all models were improved markedly with increase of plot sizes from 0.25 ha to 1.0 ha.  
At 1.0 ha plot-level, HV models derived from both spaceborne and airborne data agreed 
well with field estimates (R2=0.68, RMSECV=45.5 Mg∙ha
-1, RMSECV%=31.7%; and 
R2=0.60, RMSECV=49.0 Mg∙ha
-1, RMSECV%=34.2%).  The HH+HV+VV model derived 
from airborne data corresponds better with field estimates at 1.0 ha plot-level than that of 
single-term models in terms of R2 as well as RMSE (R2=0.72, RMSECV=40.9 Mg∙ha
-1, 
RMSECV%=28.5%).  
(a)  (b)  
 109 
 
4.5.3 Uncertainty in in-situ estimates 
In this study, the same year (2009) in-situ data and SAR backscatter were selected 
to develop regression models.  Therefore, the time-shift between in-situ and RS 
measurements was minimized.  Careful selection of suitable allometric equations for 
specific study sites is critical for a robust estimate of biomass.  The choice of DBH based 
allometric equations have influences on the estimated field biomass (Appendix Figure 
5-1).  The difference between two allometric equations (Jenkins and Young) for our study 
sites is within 5% at 0.25 ha, 0.5 ha, and 1.0 ha plot-level.  However, larger uncertainties 
are found in a higher range of biomass, and this could potentially increase the uncertainty 
in the produced biomass map.  
4.6 Conclusion 
Exponential regression models were used for mapping changes in forest biomass 
at a medium spatial resolution (100 m) using multi-source spaceborne and airborne SAR 
data over 15- and 20- year periods.  Results from this study demonstrated the ability of 
multi-source SAR data to monitor changes in forest biomass, and assess the changes in 
forest carbon budget.  
Changes mapped using archived SAR dataset (i.e., AIRSAR and SIR-C/XSAR) 
have uncertainties from sensor calibration error, and the timing of data acquisition limits 
the application of multi-temporal monitoring.  When the period between the dates of two 
or multiple sets is too short, the loss of carbon from forest disturbance is not measurable 
(above the measurement error).  Conversely, the gain of carbon from forest stand 
regrowth will be beyond the detectable range due to saturation.  Results from this study 
 110 
 
improved our knowledge about L-band SAR’s capability to monitor changes in biomass 
from forest disturbance and post-disturbance recovery.  
Several current and future spaceborne SAR systems will continue to provide data 
for these types of studies.  The ALOS PALSAR-2 launched in 2014 provides full 
polarizations at high spatial resolution (<10 m) and has a 14-day repeat cycle to monitor 
forest carbon changes (Shimoda and Xiong, 2012).  The NASA-ISRO Synthetic Aperture 
Radar (NISAR, targeted to launch in 2020) mission is designed to carry an L-band and S-
band SAR to make global measurements.  Future soil moisture products with a better 
spatial resolution (~3 km) will be available from Soil Moisture Active Passive (SMAP, 
launched in 2015) mission. This will improve the biomass predictions from SAR data by 
reducing the effects of surface wetness.  Findings from this study provide calibration and 
validation methods as the foundation for improving the performance of current and future 
spaceborne system. 
 111 
 
Chapter 5  Summary of Findings, Contributions and Future 
Research Directions        
 
Mapping biomass and changes in biomass over relatively short time scales is a 
challenging task, yet it is crucial in order to obtain a better understanding of the 
effects of forest disturbance and post-disturbance regrowth on the terrestrial carbon 
cycle.  Current approaches using active remote sensing for biomass dynamic studies 
explored mostly local to regional scales due to data limitations.  There is additionally 
a strong and growing need to develop effective mapping and assessment at a national 
to continental scale in support of climate treaty frameworks such as REDD+ (Goetz 
and Dubayah, 2011).  The work presented in this dissertation investigates LiDAR and 
SAR remote sensing approaches to this problem at selected study sites.  
In this dissertation, multi-temporal waveform LiDAR and multi-source SAR 
data were applied to map forest biomass and its changes at different key map scales 
for the two study sites: Howland Forest and Penobscot Experimental Forest.  The 
main goals were to map changes in forest biomass from waveform LiDAR data and 
assess the loss from disturbance and the gain from post-disturbance recovery, analyze 
the sensitivity of SAR signatures to changes in forest biomass and the major influence 
factors including incidence angle, soil moisture and disturbance type, and finally map 
changes in biomass using multi-source SAR data from spaceborne and airborne 
platforms. 
Major findings and contributions of this dissertation are summarized in 
sections 5.1 and 5.2, and future research directions are described in section 5.3. 
 
 112 
 
5.1 Major findings  
Chapter 2 to Chapter 4 of this dissertation explored approaches for biomass 
estimation from LiDAR and SAR data, and revealed the limitations of using a single-
sensor approach.  Discussion of the problems that need to be dealt with led to key 
insights into LiDAR and SAR data acquisition and processing relevant to future 
projects to support climate treaties such as REDD+.  Specifically, the main findings 
and significance are summarized as follows. 
Firstly, biomass prediction models at waveform LiDAR footprint-level were 
developed and then applied to assess the change in biomass at 1.0 ha map scale.  The 
disturbance-specific models were found to be significantly different at footprint-level 
and the effects of disturbance on the biomass estimations gradually diminished with 
the increase of plot sizes from 0.25 ha to 1 ha.  Evaluation of models indicated an 
improved accuracy with increasing plot sizes, and agreed well with field estimates at 
1 ha plot-level (R2=0.91, 22.4 Mg∙ha-1, 15.6%).  Moreover, prediction errors from 
waveform LiDAR were minimized when footprint coverage approached about 50% 
of the area of 1 ha plots (16 footprints) with no improvement beyond that.  
Secondly, the sensitivity analysis of forest biomass to SAR backscattering was 
conducted to investigate: 1) the influence of factors other than the change in forest 
biomass (i.e., radiometric distortion caused by incidence angle and soil moisture) on 
SAR backscatter; 2) the feasibility of cross-image normalization between multi-
temporal and multi-sensor SAR data; and 3) the possibility of applying normalized 
backscatter to detect changes in biomass due to forest disturbance and post-
disturbance recovery.  Results from the cross-image normalization indicated that: 1) 
 113 
 
the effect of incidence angle on SAR backscatter could be reduced to less than 1 dB 
by the correction model for airborne SAR data, and 2) over 50% of the changes in 
SAR backscatter due to soil moisture could be eliminated by the proposed cross-
image normalization procedure, and 3) changes in forest biomass above 100 Mg·ha-1 
or greater than 50% (biomass limited to 250 Mg·ha-1) are detectable using the cross-
normalized SAR data. 
Lastly, the changes in forest biomass over the study sites were mapped at 100 
m spatial resolution for periods of a 15-year period using spaceborne data and a 20-
year period using airborne data.  Results highlighted important issues concerning the 
mapping process using L-band SAR data: 1) speckle noise in radar data as well as 
geolocation errors were reduced with the increase of plot sizes from 0.25 ha to 1 ha, 
and 2) multi-variable models were found to outperform the single-term models 
developed for biomass estimation using either spaceborne or airborne data. At 1 ha 
plot-level, multi-variable models using spaceborne and airborne data agreed better 
with field estimates (R2=0.62, RMSEcv=55.8 Mg∙ha-1, 38.9%; and R2=0.72, 
RMSEcv=40.9 Mg∙ha-1, 28.5%) than single-term HV models (R2=0.68, 
RMSEcv=45.5 Mg∙ha-1, 31.7%; and R2=0.60, RMSEcv=49.0 Mg∙ha-1, 34.2%). 
5.2 Major contributions 
This study mapped changes in biomass from LiDAR data in Chapter 2, 
analyzed sensitivity of SAR backscatter to biomass in Chapter 3, and mapped 
changes in forest biomass using multi-source SAR data in Chapter 4. The major 
contributions are:  
 114 
 
 No reliable models at waveform LiDAR footprint-level models for biomass and 
change mapping have been developed previously.  This study indicates that 
prediction models at the scale of individual LVIS footprints are reliable when 
geolocation errors are minimized, and suggests a certain density of LVIS 
footprints is required for mapping biomass dynamics. 
 This study attempts to minimize the influence of incidence angle and soil 
moisture on SAR backscatter for estimation of biomass.  The normalization 
algorithms to reduce effects of incidence angle and soil moisture are not bound 
by a specific sensor. 
 No previous research has been conducted to map the changes in forest biomass 
from forest disturbance and post-disturbance recovery using multi-source L-band 
SAR data.  This dissertation study provides a detailed case study and may 
stimulate new interest in studies on forest biomass dynamics from SAR data. 
5.3 Future research directions 
This dissertation is a preliminary study in mapping changes in biomass after 
forest disturbance over a temperate forest using LiDAR and SAR remote sensing.  
Results from this study improve our knowledge concerning waveform LiDAR and L-
band SAR’s capability in monitoring and assessment of net carbon budget from forest 
disturbance and post-disturbance recovery.  Findings from this study provide 
calibration and validation methods as a foundation for improving the performance of 
current and future spaceborne systems. 
Experiments with repeat LiDAR in Chapter 2 have demonstrated the ability of 
LiDAR to monitor and assess changes in biomass after forest disturbance.  The 
 115 
 
sensitivity analysis and discussions in Chapter 3 and Chapter 4 have indicated the 
importance of integrating multi-sensor and multi-source remote sensing data to 
produce accurate estimates at a local to regional scale.  Although radar data 
outperform optical data by penetrating through clouds for forest applications, it also 
has limitations and complexities.  Combined multi-temporal SAR data will improve 
single date estimates for biomass.  Also, recent studies have indicated that synergic 
use of LiDAR, SAR and InSAR has great potential in reducing uncertainties in 
biomass estimates.  The vertical vegetation structure (i.e., canopy height) can also be 
derived from InSAR or optical stereo images. More studies are going on in this area, 
and ecosystem models which provide the relationships between vegetation attributes 
are being used for attribute estimates and predictions under different scenarios. 
Moreover, several ongoing spaceborne LiDAR and SAR missions guarantee 
the continuation of this work in terms of mapping forest biomass and the changes in 
biomass. For example, the NASA Earth Venture Program recently has selected the 
Global Ecosystem Dynamics Investigation (GEDI) Lidar with multiple laser beams, 
which is based on the high-SNR waveform profile LVIS, to measure forest structures 
and surface topography at high resolution from the International Space Station (ISS) 
(Blair et al., 2014; Dubayah et al., 2014).  GEDI will be launched as early as 2018 to 
provide billions of vegetation height and structure measurements for the precise 
estimation of biomass within the orbital coverage provided by ISS (+/- 51.6 degrees 
latitude).  On the other hand, several spaceborne SAR missions provide more 
opportunities in mapping forest biomass using SAR remote sensing.  The potential for 
multi-source L-band SAR to estimate forest biomass at global level will be increased 
 116 
 
as ALOS/PALSAR-2 and NISAR become operational and provide data comparable 
to existed ALOS-PALSAR and SIR-C/XSAR data archives.  
Ultimately, there still exist areas for continued investigation before we can 
confidently map biomass dynamics using remotely sensed data at policy-relevant 
scales (e.g., regional, national to continental scales) with the required accuracy in a 
consistent and transparent framework.  LiDAR and SAR remote sensing are relatively 
new for the vegetation community, and the lack of consistent and continuous 
observation leads to large uncertainties in different studies.  We anticipate more 
researches will improve our ability in reducing these uncertainties.    
 117 
 
Glossary 
 
AFM  American Forest Management 
AGB  Aboveground Biomass 
AIRSAR AIRborne Synthetic Aperture Radar 
ALOS  Advanced Land Observing Satellite 
ALTM  Airborne Laser Terrain Mappers 
ASF  Alaska Satellite Facility  
BGB  Belowground Biomass 
BOREAS  Boreal Ecosystem-Atmosphere Study 
ESA  European Space Agency 
REDD+ Reduced Emissions from Deforestation and Forest Degradation Plus 
DBH  Diameter at Breast Height 
FIA  Forest Inventory and Analysis 
GEDI  Global Ecosystem Dynamics Investigation 
InSAR  Interferometric Synthetic Aperture Radar  
JAXA  Japan Aerospace Exploration Agency 
JERS-1 Japanese Earth Resources Satellite 1   
JPL  Jet Propulsion Laboratory 
LiDAR  Light Detection and Ranging  
LULCC Land Use and Land-Cover Change 
LUT   Look Up Table   
LVIS  LiDAR Vegetation an Imaging Sensor 
 118 
 
NASA  National Aeronautics and Space Administration 
NED  National Elevation Dataset 
NISAR  NASA-ISRO Synthetic Aperture Radar  
NLCD  National Land Cover Data  
PALSAR Phased Array type L-band Synthetic Aperture Radar 
PolSAR Polarimetric Synthetic Aperture Radar 
Radar  Radio Detection and Ranging 
SAR   Synthetic Aperture Radar  
SIR-C/XSAR Spaceborne Imaging Radar-C and X-Band SAR 
RTM   Radiative Transfer Model  
UAVSAR Uninhabited Aerial Vehicle Synthetic Aperture Radar 
LTSS-VCT Landsat Series Stack- Vegetation Change Tracker  
 119 
 
Appendices 
 
 
Figure 5-1. Fied estimated biomass from national equations as a function of regional 
equations across different plot-levels at HF and PEF sites: (a) 1.0 ha, (b) 0.5 ha, (c) 0.25 ha, 
and (d) 0.125 ha. 
 
 
 120 
 
 
 
Figure 5-2. Tree height as a funciton of DBH by: (a) seven species groups from Jenkins, and (b) two 
general species groups. 
Species group: 
aa = aspen / alder / cottonwood / willow;  cl = cedar / larch;  mo = hard maple / oak / hickory / beech; 
mb = soft maple / birch;  pi = pine;  sp = spruce;  tf = true fir / hemlock;  
B = broadleaf; N = needleleaf; 
 121 
 
 
Figure 5-3. Growth curve with age from LTSS-VCT and field estimated biomass: (a) 0.25 ha 
plot, (b) 0.5 ha plot, (c) 1.0 ha plot. X axis represents age classes in an interval of 5 year, with 
classes as 0-5, 5-10, …, 100. Age 100 indicated that the plots haven’t been disturbed from 
1900s according to management record. Y axis represents mean AGB at plot-level. Blue dots 
show mean values for each age class, with dashed bars are variance. 
 
 
 
 
 
Figure 5-4. Growth curve with age and biomass from ZELIG model simulation: (a) all values, 
and (b) mean values of each soil classes. ADAMS to WESTBURY are 10 different type of 
soil based on drainage and taxonomic classification. 
 
 122 
 
Table 5-1. Allometric equations for biomass estimation in this study 
Jenkins* Code Genus Species Common name 
Allometric Equations 
Young1   Jenkins2   
b0 b1 b0 b1 
tf ABBA Abies balsamea Balsam fir 0.5958 2.4017 -2.5384 2.4814 
mb ACRU Acer rubrum Red maple 0.9392 2.3804 -1.9123 2.3651 
mb BEAL Betula alleghaniensis Yellow birch 1.1297 2.3376 -1.9123 2.3651 
mb BEPA Betula papyrifera Paper birch 0.4792 2.6634 -1.9123 2.3651 
mb BEPO Betula populifolia Grey birch 1.0931 2.3146 -1.9123 2.3651 
mo FAGR Fagus grandifolia Beech 1.3303 2.2988 -2.0127 2.4342 
mh FRAM Fraxinus americana White ash / / -2.4800 2.4835 
sp PIRU Picea rubens Red spruce 0.8079 2.3316 -2.0773 2.3323 
pi PIST Pinus strobus White pine 0.4080 2.4490 -2.0773 2.3323 
aa POGR Populus grandidentata Big-toothed aspen 0.4689 2.6087 -2.2094 2.3867 
aa POTR Populus tremuloides Trembling aspen 0.4689 2.6087 -2.2094 2.3867 
cl THOC Thuja occidentalis 
Northern white 
cedar 
1.1182 1.9269 -2.0336 2.2592 
tf TSCA Tsuga canadensis Hemlock 0.6803 2.3617 -2.5384 2.4814 
mh UNK / / / / / -2.4800 2.4835 
1 Young, Ribe, and Wainwright 1980 (Young et al., 1980) weight tables for tree and 
shrub species in Maine 
B (kg) = exp(b0+b1*ln(dbh/2.54))*0.4536; where 2.54 and 0.4536 are the factors to convert 
inch to cm, ln = natural log base "e" (2.718282) 
2 Jenkins equations (Jenkins et al., 2003) for national-scale biomass estimation 
B (kg) = exp(b0+b1*ln(dbh)) 
*Jenkins abbreviations:  
aa = aspen / alder / cottonwood / willow; cl = cedar / larch; df = douglas-fir; mb = soft maple 
/ birch; mh = mixed hardwood;  
mo = hard maple / oak / hickory / beech; pi = pine;  sp = spruce; tf = true fir / hemlock; wo = 
woodland species (juniper / oak / mesquite); 
 
 123 
 
 
Table 5-2. DBH-based Height Equations from Field Measurements (1989 to 2010) 
Research site Code Species group 
Parameter DBH Height 
n R2 RMSE p 
a b min max min max 
Howland DB Deciduous 
Broadleaf 
-3.4299 6.0729 2.5 60.9 2.0 39.3 1346 0.5606 3.83 <0.001 
Howland EN Evengreen 
Needleleaf 
-3.9934 5.6339 1.5 108.0 0.8 44.1 7035 0.5677 3.48 <0.001 
Howland aa Aspen / alder / 
cottonwood / 
willow 
-7.4545 8.2228 3.1 58.0 4.0 39.3 132 0.6243 4.45 <0.001 
Howland mb Soft maple / 
birch 
-1.7171 5.2598 2.5 56.2 2.0 32.4 1145 0.5259 3.42 <0.001 
Howland mo Hard maple / 
oak / hickory / 
beech  
-16.0939 10.4568 3.2 60.7 4.2 35.5 55 0.7106 4.92 <0.001 
Howland cl Cedar / larch -1.9222 4.0313 3.1 53.0 1.0 24.0 830 0.4828 2.50 <0.001 
Howland tf Douglas-fir  -3.1102 5.1018 2.6 71.4 0.8 32.1 2841 0.6248 3.10 <0.001 
Howland pi Pine -10.8191 8.5601 3.0 108.0 1.6 44.1 402 0.6956 4.90 <0.001 
Howland sp Spruce   -3.6442 5.7850 1.5 53.0 2.0 25.8 2962 0.5310 3.27 <0.001 
Diameter at breast height (dbh) equation: h = a*ln(dbh)+b 
Where dbh = diameter at breast height (cm, at height = 1.37 m), h = height (m) 
n = number of trees in sample; 
R2 = coefficient of determination; 
RMSE = root mean squared error or estimate of the standard deviation of the regression error term; 
p = significant level; 
Include both hardwood and softwood species from dryland forest 
 124 
 
Bibliography 
 
Ahmed, R., Siqueira, P., & Hensley, S. (2014). Analyzing the Uncertainty of Biomass 
Estimates From L-Band Radar Backscatter Over the Harvard and Howland Forests. 
IEEE Transactions on Geoscience and Remote Sensing, 52, 3568-3586 
Almeida-Filho, R., Shimabukuro, Y.E., Rosenqvist, A., & Sánchez, G.A. (2009). Using 
dual-polarized ALOS PALSAR data for detecting new fronts of deforestation in the 
Brazilian Amazônia. International Journal of Remote Sensing, 30, 3735-3743 
Anderson, J., Martin, M.E., Smith, M.L., Dubayah, R.O., Hofton, M.A., Hyde, P., 
Peterson, B.E., Blair, J.B., & Knox, R.G. (2006). The use of waveform lidar to 
measure northern temperate mixed conifer and deciduous forest structure in New 
Hampshire. Remote Sensing of Environment, 105, 248-261 
Anderson, J.E., Plourde, L.C., Martin, M.E., Braswell, B.H., Smith, M.-L., Dubayah, R.O., 
Hofton, M.A., & Blair, J.B. (2008). Integrating waveform lidar with hyperspectral 
imagery for inventory of a northern temperate forest. Remote Sensing of Environment, 
112, 1856-1870 
Asner, G.P., Mascaro, J., Muller-Landau, H.C., Vieilledent, G., Vaudry, R., Rasamoelina, 
M., Hall, J.S., & van Breugel, M. (2012). A universal airborne LiDAR approach for 
tropical forest carbon mapping. Oecologia, 1-14 
Asner, G.P., Powell, G.V.N., Mascaro, J., Knapp, D.E., Clark, J.K., Jacobson, J., 
Kennedy-Bowdoin, T., Balaji, A., Paez-Acosta, G., Victoria, E., Secada, L., Valqui, 
M., & Hughes, R.F. (2010). High-resolution forest carbon stocks and emissions in the 
Amazon. Proceedings of the National Academy of Sciences, 107, 16738-16742 
Balzter, H., Skinner, L., Luckman, A., & Brooke, R. (2003). Estimation of tree growth in 
a conifer plantation over 19 years from multi-satellite L-band SAR. Remote Sensing 
of Environment, 84, 184-191 
Blair, J., Dubayah, R., Hofton, M., Luthcke, S., Rabine, D., Wake, S., Coyle, B., Stysley, 
P., & Salerno, C. (2014). Full-waveform Airborne and Spaceborne Laser Altimetry 
for Mapping and Sampling the Earth's Forests, Cryosphere, and Land surfaces. In, 
AGU. San Francisco, USA 
 125 
 
Blair, J.B., & Hofton, M.A. (1999). Modeling laser altimeter return waveforms over 
complex vegetation using high-resolution elevation data. Geophys. Res. Lett., 26, 
2509-2512 
Blair, J.B., Hofton, M.A., & Rabine, D.L. (2006). Processing of NASA LVIS elevation 
and canopy (LGE, LCE and LGW) data products, version 1.02. In G. NASA Goddard 
Space Flight Cent., Md (Ed.) 
Blair, J.B., Rabine, D.L., & Hofton, M.A. (1999). The Laser Vegetation Imaging Sensor: a 
medium-altitude, digitisation-only, airborne laser altimeter for mapping vegetation 
and topography. ISPRS Journal of Photogrammetry and Remote Sensing, 54, 115-122 
Botkin, D.B., & Simpson, L.G. (1990). Biomass of the North American Boreal Forest: A 
Step toward Accurate Global Measures. Biogeochemistry, 9, 161-174 
Bourgeau-Chavez, L.L., Leblon, B., Charbonneau, F., & Buckley, J.R. (2013). Evaluation 
of polarimetric Radarsat-2 SAR data for development of soil moisture retrieval 
algorithms over a chronosequence of black spruce boreal forests. Remote Sensing of 
Environment, 132, 71-85 
Canadell, J.G., & Raupach, M.R. (2008). Managing Forests for Climate Change 
Mitigation. Science, 320, 1456-1457 
Collins, J.N., Hutley, L.B., Williams, R.J., Boggs, G., Bell, D., & Bartolo, R. (2009). 
Estimating landscape‐scale vegetation carbon stocks using airborne multi‐frequency 
polarimetric synthetic aperture radar (SAR) in the savannahs of north Australia. 
International Journal of Remote Sensing, 30, 1141-1159 
Dobson, M.C., Ulaby, F.T., LeToan, T., Beaudoin, A., Kasischke, E.S., & Christensen, N. 
(1992). Dependence of radar backscatter on coniferous forest biomass. IEEE 
Transactions on Geoscience and Remote Sensing, 30, 412-415 
Dobson, M.C., Ulaby, F.T., Pierce, L.E., Sharik, T.L., Bergen, K.M., Kellndorfer, J., 
Kendra, J.R., Li, E., Lin, Y.C., Nashashibi, A., Sarabandi, K., & Siqueira, P. (1995). 
Estimation of forest biophysical characteristics in Northern Michigan with SIR-C/X-
SAR. IEEE Transactions on Geoscience and Remote Sensing, 33, 877-895 
Drake, J.B., Dubayah, R.O., Clark, D.B., Knox, R.G., Blair, J.B., Hofton, M.A., Chazdon, 
R.L., Weishampel, J.F., & Prince, S. (2002). Estimation of tropical forest structural 
characteristics using large-footprint lidar. Remote Sensing of Environment, 79, 305-
319 
 126 
 
Drake, J.B., Knox, R.G., Dubayah, R.O., Clark, D.B., Condit, R., Blair, J.B., & Hofton, M. 
(2003). Above-ground biomass estimation in closed canopy Neotropical forests using 
lidar remote sensing: factors affecting the generality of relationships. In, Global 
Ecology and Biogeography (pp. 147-159): Wiley-Blackwell 
Dubayah, R., Goetz, S., Blair, J.B., Luthcke, S., Healey, S., Hansen, M., Hofton, M., Hurtt, 
G., Kellner, J., & Fatoyinbo, T. (2014). The Global Ecosystem Dynamics 
Investigation (GEDI) Lidar. In, ForestSAT2014 Open Conference System 
Dubayah, R.O., & Drake, J.B. (2000). Lidar Remote Sensing for Forestry. Journal of 
Forestry, 98, 44-46 
Dubayah, R.O., Sheldon, S.L., Clark, D.B., Hofton, M.A., Blair, J.B., Hurtt, G.C., & 
Chazdon, R.L. (2010). Estimation of tropical forest height and biomass dynamics 
using lidar remote sensing at La Selva, Costa Rica. Journal of Geophysical Research, 
115, G00E09 
Dubois, P.C., & Norikane, L. (1987). Data volume reduction for imaging radar 
polarimetry. In, IGARSS Symposium 
Englhart, S., Keuck, V., & Siegert, F. (2011). Aboveground biomass retrieval in tropical 
forests -- The potential of combined X- and L-band SAR data use. Remote Sensing of 
Environment, 115, 1260-1271 
Evans, A., & Perschel, R. (2009). A review of forestry mitigation and adaptation strategies 
in the Northeast U.S. Climatic Change, 96, 167-183 
Frazer, G.W., Magnussen, S., Wulder, M.A., & Niemann, K.O. (2011). Simulated impact 
of sample plot size and co-registration error on the accuracy and uncertainty of 
LiDAR-derived estimates of forest stand biomass. Remote Sensing of Environment, 
115, 636-649 
Gens, R., & Logan, T. (2003). Alaska Satellite Facility Software Tools: Manual In G. 
Institution (Ed.). Fairbanks, AK, USA: University of Alaska  
Goetz, S., & Dubayah, R. (2011). Advances in remote sensing technology and 
implications for measuring and monitoring forest carbon stocks and change. Carbon 
Management, 2, 231-244 
Gonçalves, F.G., Santos, J.R., & Treuhaft, R.N. (2011). Stem volume of tropical forests 
from polarimetric radar. International Journal of Remote Sensing, 32, 503-522 
 127 
 
Gonzalez, P., Asner, G.P., Battles, J.J., Lefsky, M.A., Waring, K.M., & Palace, M. (2010). 
Forest carbon densities and uncertainties from Lidar, QuickBird, and field 
measurements in California. Remote Sensing of Environment, 114, 1561-1575 
Guo, Y., Shi, Z., Zhou, L.-q., Jin, X., Tian, Y.-f., & Teng, H.-f. (2013). Integrating 
Remote Sensing and Proximal Sensors for the Detection of Soil Moisture and Salinity 
Variability in Coastal Areas. Journal of Integrative Agriculture, 12, 723-731 
Hall, F.G., Bergen, K., Blair, J.B., Dubayah, R., Houghton, R., Hurtt, G., Kellndorfer, J., 
Lefsky, M., Ranson, J., Saatchi, S., Shugart, H.H., & Wickland, D. (2011). 
Characterizing 3D vegetation structure from space: Mission requirements. Remote 
Sensing of Environment, 115, 2753-2775 
Harrell, P.A., Kasischke, E.S., Bourgeau-Chavez, L.L., Haney, E.M., & Christensen Jr, 
N.L. (1997). Evaluation of approaches to estimating aboveground biomass in 
Southern pine forests using SIR-C data. Remote Sensing of Environment, 59, 223-233 
Heath, L.S., Hansen, M., Smith, J.E., Miles, P.D., & Smith, B.W. (2008). Investigation 
into calculating tree biomass and carbon in the FIADB using a biomass expansion 
factor approach. McWilliams, WH, Moisen, GG & Czaplewski, RL (eds.) 
Henderson, F.M., & Lewis, A.J. (1998). Manual of Remote Sensing: Principles and 
Applications of Imaging Radar. New York: John Wiley and Sons 
Hollinger, D.Y., Goltz, S.M., Davidson, E.A., Lee, J.T., Tu, K., & Valentine, H.T. (1999). 
Seasonal patterns and environmental control of carbon dioxide and water vapour 
exchange in an ecotonal boreal forest. Global Change Biology, 5, 891-902 
Houghton, R., House, J., Pongratz, J., Van Der Werf, G., DeFries, R., Hansen, M., Quéré, 
C.L., & Ramankutty, N. (2012). Carbon emissions from land use and land-cover 
change. Biogeosciences, 9, 5125-5142 
Houghton, R.A., Hall, F., & Goetz, S.J. (2009). Importance of biomass in the global 
carbon cycle. Journal of Geophysical Research: Biogeosciences, 114, G00E03 
Huang, C., Goward, S.N., Masek, J.G., Thomas, N., Zhu, Z., & Vogelmann, J.E. (2010). 
An automated approach for reconstructing recent forest disturbance history using 
dense Landsat time series stacks. Remote Sensing of Environment, 114, 183-198 
Hudak, A.T., Strand, E.K., Vierling, L.A., Byrne, J.C., Eitel, J.U.H., Martinuzzi, S., & 
Falkowski, M.J. (2012). Quantifying aboveground forest carbon pools and fluxes 
from repeat LiDAR surveys. Remote Sensing of Environment, 123, 25-40 
 128 
 
Hyde, P., Dubayah, R., Peterson, B., Blair, J.B., Hofton, M., Hunsaker, C., Knox, R., & 
Walker, W. (2005). Mapping forest structure for wildlife habitat analysis using 
waveform lidar: Validation of montane ecosystems. Remote Sensing of Environment, 
96, 427-437 
Imhoff, M.L. (1995). Radar backscatter and biomass saturation: ramifications for global 
biomass inventory. IEEE Transactions on Geoscience and Remote Sensing, 33, 511-
518 
Jenkins, J.C., Chojnacky, D.C., Heath, L.S., & Birdsey, R.A. (2003). National-scale 
biomass estimators for United States tree species. Forest Science, 49, 12-35 
Jenkins, J.C., Chojnacky, D.C., Heath, L.S., & Birdsey, R.A. (2004). Comprehensive 
Database of Diameter-based Biomass Regressions for North American Tree Species. 
In, General Technical Report NE-319. Newtown Square, PA: U.S. Department of 
Agriculture, Forest Service, Northeastern Research Station 
JPL (1994). SIR-C data compression software user guide. In J.P. Laboratory (Ed.). 
Pasadena, California: California Institute of Technology 
JPL (2003). AIRSAR Integrated Processor Documentation. In, AIRSAR data format. 
Pasadena, CA: Jet Propulsion Laboratory Publication 
Kasischke, E.S., Melack, J.M., & Craig Dobson, M. (1997). The use of imaging radars for 
ecological applications--A review. Remote Sensing of Environment, 59, 141-156 
Kasischke, E.S., Tanase, M.A., Bourgeau-Chavez, L.L., & Borr, M. (2011). Soil moisture 
limitations on monitoring boreal forest regrowth using spaceborne L-band SAR data. 
Remote Sensing of Environment, 115, 227-232 
Kellndorfer, J.M., Walker, W.S., LaPoint, E., Kirsch, K., Bishop, J., & Fiske, G. (2010). 
Statistical fusion of lidar, InSAR, and optical remote sensing data for forest stand 
height characterization: A regional-scale method based on LVIS, SRTM, Landsat 
ETM+, and ancillary data sets. J. Geophys. Res., 115, G00E08 
Kuhn, M. (2008). Building predictive models in R using the caret package. Journal of 
Statistical Software, 28, 1-26 
Kurvonen, L., Pulliainen, J., & Hallikainen, M. (1999). Retrieval of biomass in boreal 
forests from multitemporal ERS-1 and JERS-1 SAR images. IEEE Transactions on 
Geoscience and Remote Sensing, 37, 198-205 
 129 
 
Langner, A., Achard, F., & Grassi, G. (2014). Can recent pan-tropical biomass maps be 
used to derive alternative Tier 1 values for reporting REDD+ activities under 
UNFCCC? Environmental Research Letters, 9 
Le Quere, C., Peters, G.P., Andres, R.J., Andrew, R., Boden, T., Ciais, P., Friedlingstein, 
P., Houghton, R., Marland, G., & Moriarty, R. (2014). Global carbon budget 2013. 
Earth System Science Data, 6, 235-263 
Lefsky, M.A. (2010). A global forest canopy height map from the Moderate Resolution 
Imaging Spectroradiometer and the Geoscience Laser Altimeter System. Geophys. 
Res. Lett., 37, L15401 
Lefsky, M.A., Cohen, W.B., Harding, D.J., Parker, G.G., Acker, S.A., & Gower, S.T. 
(2002b). Lidar remote sensing of above ground biomass in three biomes. Global 
Ecology and Biogeography, 11, 393-399 
Lefsky, M.A., Harding, D., Cohen, W.B., Parker, G., & Shugart, H.H. (1999). Surface 
Lidar Remote Sensing of Basal Area and Biomass in Deciduous Forests of Eastern 
Maryland, USA. Remote Sensing of Environment, 67, 83-98 
Lefsky, M.A., Hudak, A.T., Cohen, W.B., & Acker, S.A. (2005a). Geographic variability 
in lidar predictions of forest stand structure in the Pacific Northwest. Remote Sensing 
of Environment, 95, 532-548 
Lefsky, M.A., Keller, M., Pang, Y., De Camargo, P.B., & Hunter, M.O. (2007). Revised 
method for forest canopy height estimation from Geoscience Laser Altimeter System 
waveforms. J. Appl. Remote Sens, 1, 013537 
Lefsky, M.A., Turner, D.P., Guzy, M., & Cohen, W.B. (2005b). Combining lidar 
estimates of aboveground biomass and Landsat estimates of stand age for spatially 
extensive validation of modeled forest productivity. Remote Sensing of Environment, 
95, 549-558 
Levine, E.R., Knox, R.G., & Lawrence, W.T. (1994). Relationships between soil 
properties and vegetation at the Northern Experimental Forest, Howland, Maine. 
Remote Sensing of Environment, 47, 231-241 
Liaw, A., & Wiener, M. (2002). Classification and Regression by randomForest. R news, 
2, 18-22 
Lu, D. (2006). The potential and challenge of remote sensing-based biomass estimation. 
International Journal of Remote Sensing, 27, 1297-1328 
 130 
 
Lu, D., Chen, Q., Wang, G., Liu, L., Li, G., & Moran, E. (2014). A survey of remote 
sensing-based aboveground biomass estimation methods in forest ecosystems. 
International Journal of Digital Earth, 1-64 
Lucas, R., Armston, J., Fairfax, R., Fensham, R., Accad, A., Carreiras, J., Kelley, J., 
Bunting, P., Clewley, D., Bray, S., Metcalfe, D., Dwyer, J., Bowen, M., Eyre, T., 
Laidlaw, M., & Shimada, M. (2010). An Evaluation of the ALOS PALSAR L-Band 
Backscatter -- Above Ground Biomass Relationship Queensland, Australia: Impacts 
of Surface Moisture Condition and Vegetation Structure. IEEE Journal of Selected 
Topics in Applied Earth Observations and Remote Sensing, 3, 576-593 
Luckman, A., Baker, J., Honzák, M., & Lucas, R. (1998). Tropical forest biomass density 
estimation using JERS-1 SAR: Seasonal variation, confidence limits, and application 
to image mosaics. Remote Sensing of Environment, 63, 126-139 
Luckman, A., Baker, J., Kuplich, T.M., Yanasse, C.d.C.F., & Frery, A.C. (1997). A study 
of the relationship between radar backscatter and regenerating tropical forest biomass 
for spaceborne SAR instruments. Remote Sensing of Environment, 60, 1-13 
Malhi, Y., Baldocchi, D., & Jarvis, P. (1999). The carbon balance of tropical, temperate 
and boreal forests. Plant, Cell & Environment, 22, 715-740 
Mallet, C., & Bretar, F. (2009). Full-waveform topographic lidar: State-of-the-art. ISPRS 
Journal of Photogrammetry and Remote Sensing, 64, 1-16 
Mascaro, J., Detto, M., Asner, G.P., & Muller-Landau, H.C. (2011). Evaluating 
uncertainty in mapping forest carbon with airborne LiDAR. Remote Sensing of 
Environment, 115, 3770-3774 
Mather, P.M. (2004). Computer Processing of Remotely Sensed Images. Chichester, West 
Sussex, England: John Wiley & Sons Ltd. 
Means, J.E., Acker, S.A., Harding, D.J., Blair, J.B., Lefsky, M.A., Cohen, W.B., Harmon, 
M.E., & McKee, W.A. (1999). Use of Large-Footprint Scanning Airborne Lidar To 
Estimate Forest Stand Characteristics in the Western Cascades of Oregon. Remote 
Sensing of Environment, 67, 298-308 
Menges, C., Van Zyl, J., Hill, G., & Ahmad, W. (2001). A procedure for the correction of 
the effect of variation in incidence angle on AIRSAR data. International Journal of 
Remote Sensing, 22, 829-841 
Mitchard, E., Saatchi, S., White, L., Abernethy, K., Jeffery, K., Lewis, S., Collins, M., 
Lefsky, M., Leal, M., & Woodhouse, I. (2011a). Mapping tropical forest biomass 
 131 
 
with radar and spaceborne LiDAR: overcoming problems of high biomass and 
persistent cloud. Biogeosciences Discuss, 8, 8781-8815 
Mitchard, E.T.A., Saatchi, S.S., Lewis, S.L., Feldpausch, T.R., Woodhouse, I.H., Sonké, 
B., Rowland, C., & Meir, P. (2011b). Measuring biomass changes due to woody 
encroachment and deforestation/degradation in a forest-savanna boundary region of 
central Africa using multi-temporal L-band radar backscatter. Remote Sensing of 
Environment, 115 
Mitchard, E.T.A., Saatchi, S.S., Woodhouse, I.H., Nangendo, G., Ribeiro, N.S., Williams, 
M., Ryan, C.M., Lewis, S.L., Feldpausch, T.R., & Meir, P. (2009a). Using satellite 
radar backscatter to predict above-ground woody biomass: A consistent relationship 
across four different African landscapes. Geophys. Res. Lett., 36, L23401 
Mitchard, E.T.A., Saatchi, S.S., Woodhouse, I.H., Nangendo, G., Ribeiro, N.S., Williams, 
M., Ryan, C.M., Lewis, S.L., Feldpausch, T.R., & Meir, P. (2009b). Using satellite 
radar backscatter to predict above-ground woody biomass: A consistent relationship 
across four different African landscapes. Geophysical Research Letters, 36, L23401 
Monk, C.D., Child, G.I., & Nicholson, S.A. (1970). Biomass, Litter and Leaf Surface 
Area Estimates of an Oak-Hickory Forest. Oikos, 21, 138-141 
Næsset, E., & Gobakken, T. (2008). Estimation of above- and below-ground biomass 
across regions of the boreal forest zone using airborne laser. Remote Sensing of 
Environment, 112, 3079-3090 
Næsset, E., Gobakken, T., Bollandsås, O.M., Gregoire, T.G., Nelson, R., & Ståhl, G. 
(2013). Comparison of precision of biomass estimates in regional field sample 
surveys and airborne LiDAR-assisted surveys in Hedmark County, Norway. Remote 
Sensing of Environment, 130, 108-120 
Nelson, R., Ranson, K.J., Sun, G., Kimes, D.S., Kharuk, V., & Montesano, P. (2009). 
Estimating Siberian timber volume using MODIS and ICESat/GLAS. Remote Sensing 
of Environment, 113, 691-701 
Ni-Meister, W., Lee, S., Strahler, A.H., Woodcock, C.E., Schaaf, C., Yao, T., Ranson, 
K.J., Sun, G., & Blair, J.B. (2010). Assessing general relationships between 
aboveground biomass and vegetation structure parameters for improved carbon 
estimate from lidar remote sensing. J. Geophys. Res., 115, G00E11 
Ni, W., Sun, G., Guo, Z., Zhang, Z., He, Y., & Huang, W. (2013a). Retrieval of Forest 
Biomass From ALOS PALSAR Data Using a Lookup Table Method. IEEE Journal 
of Selected Topics in Applied Earth Observations and Remote Sensing, V6, 875-886 
 132 
 
Ni, W., Sun, G., Zhang, Z., Guo, Z., & He, Y. (2013b). Co-registration of two DEMs: 
impacts on forest height estimation from SRTM and NED at Mountainous Areas. 
IEEE Geoscience and Remote Sensing Letter 
Nilsson, M. (1996). Estimation of tree heights and stand volume using an airborne lidar 
system. Remote Sensing of Environment, 56, 1-7 
Nunery, J.S., & Keeton, W.S. (2010). Forest carbon storage in the northeastern United 
States: Net effects of harvesting frequency, post-harvest retention, and wood products. 
Forest Ecology and Management, 259, 1363-1375 
Pang, Y., Lefsky, M., Andersen, H.E., Miller, M.E., & Sherrill, K. (2008). Validation of 
the ICEsat vegetation product using crown-area-weighted mean height derived using 
crown delineation with discrete return lidar data. Canadian Journal of Remote 
Sensing, 34, S471-S484 
Peddle, D.R., Huemmrich, K.F., Hall, F.G., Masek, J.G., Soenen, S.A., & Jackson, C.D. 
(2011). Applications of the BIOPHYS Algorithm for Physically-Based Retrieval of 
Biophysical, Structural and Forest Disturbance Information. IEEE Journal of Selected 
Topics in Applied Earth Observations and Remote Sensing, 4, 971-982 
Pope, K.O., Rey-Benayas, J.M., & Paris, J.F. (1994). Radar remote sensing of forest and 
wetland ecosystems in the Central American tropics. Remote Sensing of Environment, 
48, 205-219 
Pottier, E., Ferro-Famil, L., Allain, S., Cloude, S.R., Hajnsek, I., Papathanassiou, K., 
Moreira, A., Williams, M., Minchella, A., Lavalle, M., & Desnos, Y.L. (2009). 
Overview of the PolSARpro v4.0: The open source toolbox for polarimetric and 
interferometric polarimetric SAR data processing. In, International Geoscience and 
Remote Sensing Symposium (pp. 936-939). Capetown, South Africa 
Ranson, K.J., Saatchi, S., & Guoqing, S. (1995). Boreal forest ecosystem characterization 
with SIR-C/XSAR. IEEE Transactions on Geoscience and Remote Sensing, 33, 867-
876 
Ranson, K.J., & Sun, G. (1997). An evaluation of AIRSAR and SIR-C/X-SAR images for 
mapping northern forest attributes in Maine, USA. Remote Sensing of Environment, 
59, 203-222 
Ranson, K.J., & Sun, G. (2000). Effects of environmental conditions on boreal forest 
classification and biomass estimates with SAR. IEEE Transactions on Geoscience 
and Remote Sensing, 38, 1242-1252 
 133 
 
Ranson, K.J., Sun, G., Knox, R.G., Levine, E.R., Weishampel, J.F., & Fifer, S.T. (2001). 
Northern Forest Ecosystem Dynamics Using Coupled Models and Remote Sensing. 
Remote Sensing of Environment, 75, 291-302 
Ranson, K.J., Sun, G., Lang, R.H., Chauhan, N.S., Cacciola, R.J., & Kilic, O. (1997c). 
Mapping of boreal forest biomass from spaceborne synthetic aperture radar. J. 
Geophys. Res., 102, 29599-29610 
Robinson, C., Saatchi, S., Neumann, M., & Gillespie, T. (2013). Impacts of Spatial 
Variability on Aboveground Biomass Estimation from L-Band Radar in a Temperate 
Forest. Remote Sensing, 5, 1001-1023 
Rosenqvist, Å., Milne, A., Lucas, R., Imhoff, M., & Dobson, C. (2003). A review of 
remote sensing technology in support of the Kyoto Protocol. Environmental Science 
& Policy, 6, 441-455 
Rowland, C.S., Balzter, H., Dawson, T.P., Luckman, A., Patenaude, G., & Skinner, L. 
(2008). Airborne SAR monitoring of tree growth in a coniferous plantation. 
International Journal of Remote Sensing, 29, 3873-3889 
Saatchi, S., Dubayah, R., Clark, D., Chazdon, R., & Hollinger, D. (2010). Estimation of 
Forest Biomass Change From Fusion of Radar and Lidar. In, IEEE International 
Geoscience and Remote Sensing Symposium. Honolulu, Hawaii, USA 
Saatchi, S., Marlier, M., Chazdon, R.L., Clark, D.B., & Russell, A.E. (2011a). Impact of 
spatial variability of tropical forest structure on radar estimation of aboveground 
biomass. Remote Sensing of Environment, 115, 2836-2849 
Saatchi, S.S., Harris, N.L., Brown, S., Lefsky, M., Mitchard, E.T.A., Salas, W., Zutta, 
B.R., Buermann, W., Lewis, S.L., Hagen, S., Petrova, S., White, L., Silman, M., & 
Morel, A. (2011b). Benchmark map of forest carbon stocks in tropical regions across 
three continents. Proceedings of the National Academy of Sciences, 108, 9899-9904 
Safford, L.O., Frank, R.M., & Little, E.L. (1969). Trees and shrubs of the Penobscot 
Experimental Forest, Penobscot County, Maine: Northeastern Forest Experiment 
Station 
Salas, W., Ducey, M., Rignot, E., & Skole, D. (2002). Assessment of JERS-1 SAR for 
monitoring secondary vegetation in Amazonia: I. Spatial and temporal variability in 
backscatter across a chrono-sequence of secondary vegetation stands in Rondonia. 
International Journal of Remote Sensing, 23, 1357-1379 
 134 
 
Sandberg, G., Ulander, L.M.H., Fransson, J.E.S., Holmgren, J., & Le Toan, T. (2011). L- 
and P-band backscatter intensity for biomass retrieval in hemiboreal forest. Remote 
Sensing of Environment, 115, 2874-2886 
Sandberg, G., Ulander, L.M.H., Wallerman, J., & Fransson, J.E.S. (2014). Measurements 
of Forest Biomass Change Using P-Band Synthetic Aperture Radar Backscatter. 
IEEE Transactions on Geoscience and Remote Sensing, 52, 6047-6061 
Santoro, M., Fransson, J.E.S., Eriksson, L.E.B., & Ulander, L.M.H. (2010). Clear-Cut 
Detection in Swedish Boreal Forest Using Multi-Temporal ALOS PALSAR 
Backscatter Data. IEEE Journal of Selected Topics in Applied Earth Observations 
and Remote Sensing, 3, 618-631 
Santos, J.R., Lacruz, M.S.P., Araujo, L.S., & Keil, M. (2002). Savanna and tropical 
rainforest biomass estimation and spatialization using JERS-1 data. International 
Journal of Remote Sensing, 23, 1217-1229 
Scott, N.A., Rodrigues, C.A., Hughes, H., Lee, J.T., Davidson, E.A., Dail, D.B., Malerba, 
P., & Hollinger, D.Y. (2004). Changes in carbon storage and net carbon exchange one 
year after an initial shelterwood harvest at Howland Forest, ME. Environmental 
Management, 33, 9-22 
Shimada, M., Isoguchi, O., Tadono, T., & Isono, K. (2009). PALSAR radiometric and 
geometric calibration. IEEE Transactions on Geoscience and Remote Sensing, 47, 
3915-3932 
Shimoda, H., & Xiong, X. (2012). Earth Observing Missions and Sensors: Development, 
Implementation, and Characterization II. In, Proc. of SPIE Vol (pp. 852801-852801) 
Small, D. (2011). Flattening Gamma: Radiometric Terrain Correction for SAR Imagery. 
Geoscience and Remote Sensing, IEEE Transactions on, 49, 3081-3093 
Solberg, S. (2010). Mapping gap fraction, LAI and defoliation using various ALS 
penetration variables. International Journal of Remote Sensing, 31, 1227 - 1244 
Sumantyo, J.T.S., & Amini, J. (2008). A model for removal of speckle noise in SAR 
images (ALOS PALSAR). Canadian Journal of Remote Sensing, 34, 503-515 
Sun, G., & Ranson, K.J. (1998). Radar modelling of forest spatial patterns. International 
Journal of Remote Sensing, 19, 1769-1791 
 135 
 
Sun, G., Ranson, K.J., Guo, Z., Zhang, Z., Montesano, P., & Kimes, D. (2011). Forest 
biomass mapping from lidar and radar synergies. Remote Sensing of Environment, 
115, 2906-2916 
Sun, G., Ranson, K.J., & Kharuk, V.I. (2002). Radiometric slope correction for forest 
biomass estimation from SAR data in the Western Sayani Mountains, Siberia. Remote 
Sensing of Environment, 79, 279-287 
Sun, G., Simonett, D.S., & Strahler, A.H. (1991). A radar backscatter model for 
discontinuous coniferous forests. IEEE Transactions on Geoscience and Remote 
Sensing, 29, 639-650 
Swatantran, A., Dubayah, R., Roberts, D., Hofton, M., & Blair, J.B. (2011). Mapping 
biomass and stress in the Sierra Nevada using lidar and hyperspectral data fusion. 
Remote Sensing of Environment, 115, 2917–2930 
Thomas, N.E., Huang, C., Goward, S.N., Powell, S., Rishmawi, K., Schleeweis, K., & 
Hinds, A. (2011). Validation of North American Forest Disturbance dynamics 
derived from Landsat time series stacks. Remote Sensing of Environment, 115, 19-32 
Urban, D. (1990). A versatile model to simulate forest pattern: a user’s guide to ZELIG 
version 1.0. Environmental Sciences Department, The University of Virginia, 
Charlottesville, Virginia 
Ustin, S., Wessman, C., Curtiss, B., Kasischke, E., Way, J.B., & Vanderbilt, V. (1991). 
Opportunities for using the EOS imaging spectrometers and synthetic aperture radar 
in ecological models. Ecology, 72, 1934-1945 
Wang, Y., Day, J.L., Davis, F.W., & Melack, J.M. (1993). Modeling L-band radar 
backscatter of Alaskan boreal forest. IEEE Transactions on Geoscience and Remote 
Sensing, 31, 1146-1154 
Way, J., Paris, J., Dobson, M.C., McDonals, K., Ulaby, F.T., Weber, J., Ustin, L., 
Vanderbilt, V.C., & Kasischke, E.S. (1991). Diurnal change in trees as observed by 
optical and microwave sensors: the EOS synergism study. IEEE Transactions on 
Geoscience and Remote Sensing, 29, 807-821 
Weisbin, C.R., Lincoln, W., & Saatchi, S. (2013). A Systems Engineering Approach to 
Estimating Uncertainty in Above-Ground Biomass (AGB) Derived from Remote-
Sensing Data. Systems Engineering 
 136 
 
Young, H.E., Ribe, J.H., & Wainwright, K. (1980). Weight tables for tree and shrub 
species in Maine. In, Miscellaneous Report 230. , University of Maine at Orono: Life 
Sciences and Agriculture Experiment Station 
Zhang, Z. (2011). Biomass retrieval based on lidar and SAR data. In, School of 
Geography and Remote Sensing (p. 140). Beijing: Beijing Normal University 
Zhao, K., & Popescu, S. (2009). Lidar-based mapping of leaf area index and its use for 
validating GLOBCARBON satellite LAI product in a temperate forest of the southern 
USA. Remote Sensing of Environment, 113, 1628-1645 
Zhao, K., Popescu, S., Meng, X., Pang, Y., & Agca, M. (2011). Characterizing forest 
canopy structure with lidar composite metrics and machine learning. Remote Sensing 
of Environment, 115, 1978-1996 
Zink, M., Olivier, P., & Freeman, A. (1993). Cross-calibration between airborne SAR 
sensors. IEEE Transactions on Geoscience and Remote Sensing, 31, 237-245