Comparison of Polarimetric Filters to Retrieve Forest Biomass

There are several polarimetric filters in the literature developed for the most diverse applications. Here, we present an evaluation of different filter sizes and types for improving the AGB estimates in the Brazilian Amazon, varying from none to 21×21 px filter size in full polarimetric ALOS/PALSAR-2. The optimal window size was chosen by the highest coefficient of determination (R2) between forest AGB and backscattering coefficient (σ°). After that, six polarimetric filters were evaluated: BoxCar, Refined Lee, Improved Sigma Lee, Intensity Driven Adaptive Neighbourhood (IDAN), Scattering Model-Based (SMB), and Model-based (MB). The comparison criterion was a set of statistics aimed at preserving the three basic principles of the filtering process. The optimal window size was 11×11 px. The best was Refined Lee, followed by BoxCar and SMB. The R2 differences in the filter choice can be up to 15% for retrieving forest AGB.