Polarimetric decomposition of SAR data for forest structure assessment

Abstract

One of the prime factors in the global carbon budget is forest biomass, and it is estimated that about 20% of the carbon flux to the atmosphere comes from deforestation and disturbance. Thus, it is important to quantify the amount of biomass present on Earth for climate change studies and for the estimation of global carbon stock. Synthetic Aperture Radar (SAR) polarimetry finds a useful application in remote sensing of forests as it can help with the assessment of forest structure and above-ground forest biomass. This thesis presents MATLAB implementation of three polarimetric target decomposition theorems, Pauli decomposition, Freeman-Durden decomposition, and H/A/alpha decomposition. The potential of air-borne SAR polarimetry for forest structure assessment is then evaluated using P- and L-band SAR data acquired with the ONERA SETHI system under the BioSAR 2010 campaign over the boreal forest of Remningstorp in southern Sweden. Results show that Freeman-Durden decomposition provides better visual classification of the forested area than Pauli and H/A/alpha decomposition techniques. Results also show that at L-band, most scattering occurs in forest canopy, while at P-band, forest canopy is penetrated to a higher degree and ground-locked mechanism takes over. Also, a biomass estimation model is developed based on a linear regression analysis of different decomposition products. Results for P-band data show that of all studied parameters, the parameter |δ|, which is a sensitive indicator for the amount of ground scattering visible, is best correlated with reference biomass. On the other hand, results for L-band data show that of all studied parameters, the parameter |HH-VV| is best correlated with reference biomass. Biomass estimation model corresponding to parameter |δ|, in case of P-band, and parameter |HH-VV|, in case of L-band, shows the lowest root mean square error of the order of 35-50 tons/ha, for all flight headings.