On Finding Optimal Speckle Filtering for Extraction of Vegetation Biophysical Information Using Sentinel-1 SAR Imagery

The SAR imagery such as Sentinel-1 in general has a major problem with the speckle effects. There are many speckle filtering methods have been developed to reduce the speckle effect. This research aims to test the ability of a number of speckle filtering methods to extract vegetation biophysical information from Sentinel-1. The ground truth of vegetation biophysical information in this research were simulated using Sentinel-2 MSI imagery. That is, Leaf Area Index (LAI), Canopy Water Content (CWC), Canopy Chlorophyll Content (CCC), Fraction of Vegetation Cover (FVC), and Fraction of Absorbed Photosynthetically Active Radiation (FAPAR). The Sentinel-1 imagery was speckle filtered using various methods, namely Lee, Lee Sigma, Refined Lee, IDAN, Boxcar, Frost, Gamma Map, and Median. Some speckle filtering parameters were modified, i.e., the processing windows. The Dual Polarization SAR Vegetation Index (DPSVI) were then extracted from the speckle-filtered Sentinel-1. DPSVI were then tested for correlation with vegetation biophysical information using the Pearson Correlation Coefficient (r). The test results show that Boxcar produces the highest r values for all types of vegetation biophysical information, with values ranging from 0.6s to 0.7s. Followed by Lee, Gamma Map, Median, and Frost. Each with a processing window size of 21x21. Since there are no r values was found which reached 0.8 for processing window sizes up to 21x21, the simulation was then run using the regression method. The simulation results show that to achieve r values of 0.8, it is predicted that window sizes range from 35x35 to 93x93.