In: Wagner W., Székely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B 
320 
Direct scattering from the canopy and the soil, as well as 
multiple interactions between the vegetation components and 
the soil, all contribute to the magnitude and scattering 
characteristics of the SAR response. Simple linear or non-linear 
expressions fail to adequately express the interaction of 
microwaves with a complex vegetation-over-soil target. A 
physically-based modeling approach is essential for analyzing 
the interaction of crop biological variables and SAR backscatter 
over a wide range of crop conditions and sensor configurations. 
In this study, L- and C-band backscatter at certain polarizations 
exhibited a strong correlation with LAI. Some of the 
unexplained error in these simple correlations may be 
attributable to contributions from the soil moisture. Therefore, 
the water cloud model was used to model the effect of LAI and 
surface soil moisture on SAR backscatter. 
Data needed to parameterize soil moisture in the water cloud 
model were available from the in situ measurements taken 
coincident with the RADARSAT-2 and ALOS overpasses. No 
in situ measurements were available for the August 21 
TerraSAR-X acquisition. Thus TerraSAR-X data were not 
implemented into the water cloud model. 
For the com and soybean crops, the mean backscatter was 
extracted for a 70 x 70 metre area centred on the soil moisture 
sampling sites. A similar approach was taken to calculate 
average LAI for each site, from the LAI maps derived from the 
optical data. 
To overcome instability problems caused by possible 
correlations between parameters, a two-step procedure was 
taken. The model parameter D defining the radar sensitivity to 
soil moisture was first determined using an independent data 
set. Once parameter D was fixed, the remaining parameters A, 
B, E, and C were then simultaneously determined. Three of the 
soybean fields were seeded late because of an unusually rainy 
spring season, and thus for eleven soybean sites, the soybean 
crop had not yet emerged at the time of the June 12 FQ6 and 
June 15 FQ20 RADARSAT-2 acquisitions. For the 
PALSAR/ALOS acquisition on May 23, 2007, soil moisture 
measurements were taken during the satellite overpass. At that 
time, most of fields were bare as crops had not yet emerged. 
Based on these data, a linear regression model was developed to 
describe the relationship between SAR backscatter and soil 
moisture in the absence of vegetation. This process was used to 
determine and fix the parameter D. Next, the remaining 
parameters in the model, A, B, E and C, were determined using 
a non-linear least squares method in the Matlab Curve Fitting 
Toolbox environment, based on the Levenberg-marquardt 
algorithm. 
The degree of model fit was indicated by the coefficient of 
determination (R 2 ) and RMSE, and these statistics are provided 
in Table 3. 
SAR 
Com 
Soybeans 
Backscatter 
Coefficient oi 
determination 
(R 2 ) 
RMSE 
(power) 
Coefficient of 
determination 
(R ! ) 
RMSE 
(power) 
PALSAR/ALOS 
HH 
0.78 
0.013 
0.44 
0.019 
HV 
0.81 
0.002 
0.38 
0.005 
RADARSAT-2 FQ20 
HH 
0.63 
0.026 
0.10 
0.022 
HV 
0.78 
0.004 
0.07 
0.003 
RADARSAT-2 FQ6 
HH 
0.26 
0.038 
0.43 
0.019 
HV 
0.71 
0.004 
0.38 
0.005 
Table 3. Statistics describing the fit of the water cloud model to 
SAR backscatter 
A good model fit was achieved for most SAR configurations for 
com, with coefficients of determination (R; reaching 0.63- 
0.81. The one exception was the poor correlation for C-HH 
backscatter (RADARSAT-2 FQ6). For soybeans, the water 
cloud model provided only weak correlations for all SAR 
frequencies and polarizations. In figure 3, the fitted models 
using the L-HV backscatter are plotted against the observed 
data for com. 
Corn 
Soil moisture (%) ' . ... 2 -2, 
v ' LAI (m m ) 
Figure 3. Modeled and observed L-HV backscatter expressed as 
a function of soil moisture and LAI for com. 
For com, the highest correlations were again reported for L- 
band backscatter at HH and HV polarizations. Slightly lower 
correlations were reported at most C-band polarizations except 
for C-HH backscatter (RADARSAT-2 FQ6 mode). Backscatter 
from soybeans, regardless of frequency or polarization, were 
not significantly correlated with LAI. 
5. CONCLUSION 
This study investigated the relationship between multi 
frequency SAR backscatter and LAI for com and soybean 
crops. TerraS AR-X dual-polarized stripmap data (X-band), 
RADARSAT-2 Fine beam quad-polarized data (C-band) and 
ALOS PALSAR dual-pol data (L-band), as well as optical data 
including the Compact Airborne Spectragrahic Imager (CASI) 
and SPOT-4 multi-spectral data were acquired during the 2008 
crop growing season. SAR backscatter was extracted from each 
SAR image. LAI maps were derived from the optical images at 
a detailed pixel level. Object-based segmentation of the LAI 
maps defined the basic sampling unit upon which mean LAI 
and SAR responses were calculated. 
A statistical correlation analyses quantified the relationship 
between the SAR parameters and LAI. High correlation 
coefficients with com LAI were found for L-band and C-band. 
The highest correlation coefficients (r=0.90—0.96) were 
reported for L-HH, L-HV and C-HV (RADARSAT-2 FQ6