The object of this paper is to investigate the relationship between polarimetric SAR information and LAI. RADARSAT-
2 Fine Quad-pol SLC data with shallower and steeper incidence angles were programmed throughout the 2008 growing
season. Optical data were acquired using a hyperspectral CASI airborne sensor as well as the SPOT-4 multi-spectral
satellite. The optical data were used to generate LAI map for the entire study site. Backscatter coefficients, ratios of
backscatter intensity, three polarimeric variables and three Cloude-pottier Decomposition parameters were extracted
from the polarimetric data set. Temporal variations of the backscatter coefficient were analyzed. The results show an
increase in backscatter with corn and soybean growth. The statistical analysis quantified the relationship between the
radar parameters and LAI revealing a strong sensitivity for some radar configurations. For both corn and soybean,
RADARSAT-2 cross-polarization (HV) backscatter at either shallow or steep incidence angles was well correlated with
LAI. To avoid sensitivity to sensor calibration and changing target moisture conditions, ratios of backscatter intensity,
polarimetric variables and Cloude-pottier Decomposition parameters were investigated. For corn, the ratio of HV/HH
and HV/VV as well as pedestal height, total power, correlation coefficient, Entropy and alpha angle were highly
correlation with LAI at steeper incidence angle. For soybean, the higher correlations were found with the ratio of HV/HH
as well as pedestal height, total power, Entropy and alpha angle at shallow incidence angle. In general, the best results
were observed for corn using the FQ6 acquisition. For soybean, the FQ20 data provided the most promising results.
The mapping of agricultural-land use systems using single-date information has, in the past, met with limited success. Broad-band multi-spectral sensors have been used primarily for mapping of broad land cover types, but have been less successful for identifying species-level variation. Hyperspectral sensors have had some success for species mapping, but these images often cover a small area and are not appropriate for large-scale land-use assessment. Phenological changes in crop broad-band spectral properties over the growing season offer a promising method of detecting species variation associated with growth rates, plant structure and cropping practices. This paper will present preliminary results of the use of multi-temporal optical imagery for mapping agricultural species. Three SPOT-4 multispectral scenes were acquired during early, middle and late season growth stages over an agricultural region in eastern Ontario, Canada in 2004. Three supervised classification methods were compared: Maximum-Liklihood, Decision Tree and Neural Network approaches. The impact of atmospheric correction was explored to determine if statistical models using multi-sensor, multi-date inputs are sensitive to differences in atmospheric conditions during image acquisition. The success of each method is assessed based on classification accuracies determined using an independent set of ground measurements. Preliminary results indicate that multi-date information is essential to deriving accurate land use information, and that further inputs in addition to remote sensing data may be needed to define specific classes.