Above-ground Biomass Estimation and Validation Using Multisource Remote Sensing Data, and Linear and Machine Learning Approaches Over High-altitudinal Himalayan Landscape

Estimating the aboveground biomass of forests (also known as AGB) is an important source of information on the carbon cycle. Understanding the potential for biomass supply, determining whether or not there are any possible growth constraints, and making management decisions all require a system that can promptly and correctly assess above-ground biomass (AGB). Therefore, the purpose of this study as a whole is to offer a method that has the potential to improve the precision of the AGB estimation. The primary goals are to 1) investigate the performance of remote sensing data sources to improve AGB predictions, including optical, SAR, and their combination; 2) examine the capability of various statistical and tree-based machine learning models; and 3) compare the performance of machine learning model predictions. Multiple tree-based algorithms were fitted to predictors that were developed from Landsat, Sentinel-2, and Sentinel-1 SAR data that was obtained over the high altitudinal Himalayan region. According to the findings of the predictor importance analysis, the Transformed Normalized Difference Vegetation Index (TNDVI), normalized difference vegetation index (NDVI), the ratio vegetation index (RVI), and the difference vegetation index (DVI) had the most significant impact with correlation coefficients of 0.72, 0.71, 0.68, and 0.65 respectively on the assessment of the aboveground biomass. TNDVI is used to access the performance of machine learning and linear regression models. The accuracy of the model was increased by combining the results of remote sensing data collected from different sources (RMSE: 52.14 Mg ha1 and R2: 0.62). The results of the random forest (RF), support vector regression (SVR) and linear regression (LR) models were not significantly different from one another. Other approaches were not as accurate as the RF model's estimation of the AGB, which had a root mean square error (RMSE) of 56.73 Mg/ha and an R2 value of 0.81. SVR was the approach that produced the second-best results, with an RMSE of 62.69 Mg/ha and an R2 of 0.78.