Forest biomass estimation using synthetic aperture radar polarimetric features

Polarimetric synthetic aperture radar (POLSAR) images have many applications in forest studies, especially for biomass estimation. An algorithm was proposed to extract optimized features from POLSAR images that are required for estimation. The algorithm included three main steps: feature extraction including radar backscatters and Pope’s, Cloude–Pottier’s, Freeman–Durden’s, and Touzi’s parameters; feature selection using a particle swarm optimization (PSO); and forest biomass estimation using multivariate relevance vector regression (MVRVR) and support vector regression. Based on the PSO, a combination of features was selected. The estimation based on the PSO selection was the most accurate, with the MVRVR model showing the highest coefficient of determination (R2, 0.86) and the lowest errors, with a root-mean square error of 39.17, a mean absolute error of 36.50, and a mean error of 11.59.