Predicting forest above-ground biomass using SAR imagery and GEDI data through machine learning in GEE cloud

The estimation of Forest above-ground biomass (AGB) is critical for comprehending forest ecosystems and promoting biodiversity restoration. The study was conducted to develop an effective approach to predict Forest Above-Ground Biomass (AGB) using Machine Learning, Image Classification, and GEE open-source fast processing system in the Similipal Tiger Reserve (STR), India. The study utilized six machine learning models and integrated various data sources, including Sentinel-1 and Sentinel-2 imagery, forest canopy height data from NASA’s GEDI Global Ecosystems Dynamics Investigation-LiDAR, geo-environmental Shuttle Radar Topography Mission (SRTM) data, and Climate Hazards Group Infrared precipitation with Station (CHIRPS) data. Sentinel-based optical and Synthetic Aperture Radar (SAR) signatures were also extracted before and after the monsoon season to evaluate AGB in a subtropical region. The Random forest-based Boruta method was used to examine the importance of multiple factors contributing to the prediction’s accuracy. In addition, the assessment of multicollinearity, which is accomplished by measuring the variance inflation factor (VIF), was carried out to address the issue of interrelatedness among variables that could affect the accuracy of the AGB mapping. The Random Forest model exhibited superior accuracy compared to other models, achieving an R2 of 0.71, MAE of 16.12Mg/ha, RMSD of 22.27Mg/ha, NRMSD of 0.212, and AUROC of 75%. The scatterplot analysis revealed a positive correlation between forest biomass and factors, such as forest canopy height, elevation, normalized differential vegetation index, normalized differential moisture index, and the ratio of VV/VH after the monsoon season. The VIF values ranged between 1.12 and 8.73, with VH_Jan_Mar having the maximum VIF and forest canopy height having the minimum VIF. As per the Boruta algorithm, 16 attributes were deemed important, while 14 attributes had less influence on AGB. The study presented a novel approach for estimating biomass in subtropical regions using remote sensing data set and machine learning models in Google platform. These results can be successfully used by the planners of STR for monitoring variation in AGB ensuring better habitat for wild animals.