Aboveground Biomass Prediction by Fusing Gedi Footprints with Optical and SAR Data Using the Random Forest in the Mixed Tropical Forest, India

Abstract

The objective is to predict forest aboveground biomass density (AGBD) by integrating spaceborne Light detection and Ranging (LiDAR) Global Ecosystem Dynamics Investigation (GEDI) L4A AGBD footprints with optical and synthetic aperture radar (SAR) data using random forest (RF) in the mixed tropical forests of the Shoolpaneshwar wildlife sanctuary (SWLS), Gujarat, India. RF was trained using GEDI L4A AGBD, while 3-fold cross-validation (CV) was used to minimize overfitting or underfitting. RF achieved optimal training accuracy with root mean square error (RMSE) = 35.05 Mg/ha and R-squared (R2) = 0.44, while testing showed that RF had predicted AGBD with RMSE = 30.44 Mg/ha and R <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> = 0.46. GEDI derived predictors correlate most with AGBD and are most important in AGBD prediction. The predicted mean AGBD in SWLS is 41.05 Mg/ha, and AGBD patches of greater than 100 Mg/ha lie in the inner parts of SWLS. Overall, the used approach would help assess and monitor carbon dynamics in forest ecosystems.