Biomass estimation at ICESat/GLAS footprints using support vector regression algorithm for optimization of parameters

Abstract

Forest aboveground biomass is a great significance for a better understanding of global carbon cycle. Estimation of biomass at ICESat/GLAS footprint was done by integrating datasets from single sensors. The biomass estimation accuracy of support vector machine regression was studied. Multiple linear regression equations were established from some of the most important variables found using support vector machine algorithm. The study also introduced a more accurate method of data collection from ICESat footprints. The results of the study were very encouraging. SVM gives AGB prediction where the best 6 variables had an RMSE of 18.502 t/ha. The study conclusively established that SVM is capable of estimating the biomass single sensor approach in predicting AGB. The outcome of the study was the formulation of an approach that could result in predicted biomass estimation. The study finally analyzed its limitations and suggested improvements that would result in even better estimation accuracies.