Combination of hyperspectral and LiDAR for aboveground biomass estimation using machine learning

Abstract

The increase in greenhouse gases in the atmosphere is due to carbon dioxide (CO2), which has affected climate change. Therefore, the forest plays an essential role in carbon storage which absorbs the CO2 and releases oxygen (O2) to stabilize the earth's ecosystem. This research aims to estimate aboveground biomass (AGB) using a combination of airborne hyperspectral and LiDAR data with field observation in a tropical forest. The objective of this study is to test the ability of vegetation indices and topographic features derived from hyperspectral and LiDAR data using machine learning for AGB estimation and to identify the best machine learning algorithms for estimating AGB in tropical forest. In this research, artificial neural network (ANN) and random forest (RF) algorithm were used to predict the AGB using different models with different combinations of variables. During model selection, the best model fit was selected by calculating statistical parameters such as the residual of the coefficient of determination (R2) and root mean square error (RMSE). Based on the statistical indicators, the most suitable model is Model 4 using anRF algorithm with mtry = p, and a combination of field observation, LiDAR, hyperspectral, vegetation indices (VIs), and topography. This model produced R2 = 0.997 and RMSE = 30.653 kg/tree. Therefore, using a combination of field observation and remote sensing data with machine learning techniques is reliable in forest management to estimate AGB in tropical forest.