Optimizing spaceborne LiDAR and very high resolution optical sensor parameters for biomass estimation at ICESat/GLAS footprint level using regression algorithms

Abstract

Forests sequester large quantity of carbon in their woody biomass and hence accurate estimation of forest biomass is extremely crucial. The present study aims at combining information from spaceborne LiDAR (ICESat/GLAS) and high resolution optical data to estimate forest biomass. Estimation of aboveground biomass (AGB) at ICESat/GLAS footprint level was done by integrating data from multiple sensors using two regression algorithms, viz. random forest (RF) and support vector machine (SVM). The study used forest height and canopy return ratio (rCanopy) for determination of effective size of ICESat/GLAS footprints for field data collection. The forest height was predicted with root mean square error (RMSE) of 1.35 m. The study showed that six most important parameters derived from LiDAR, and passive optical data were able to explain 78.7% (adjusted) variation in the observed AGB with an RMSE of 13.9 Mg ha–1. It was also observed that 15 most important parameters were able to explain 83% (adjusted) variation in the observed AGB. It was found that SVM regression algorithm explained 88.7% of variation in AGB with an RMSE of 13.6 Mg ha–1 on the combined datasets while RF regression algorithm explained 83.5% of variation in AGB with an RMSE of 20.57 Mg ha–1. The study demonstrated that RF regression algorithm performs equally well on datasets irrespective of the correlation of underlying variables with the predicted variable whereas SVM regression was found to perform well on those datasets which had a subset of underlying variables that are correlated with the predicted variable. The study highlighted that sensor integration approach is more accurate than single sensor approach in predicting the AGB.