MULTISENSOR SATELLITE DATA FUSION AND MACHINE LEARNING FOR ESTIMATING AND EXTRAPOLATING ABOVE-GROUND BIOMASS IN ARKANSAS FORESTS

Forests play a crucial role in biodiversity conservation, climate change mitigation, natural education, scientific research, and carbon sequestration. Advancements in satellite technology and machine learning have enabled us to understand forest dynamics at a regional scale. This study leverages a fusion of optical, radar, and lidar data from open-source satellite sources to estimate the Above Ground Biomass (AGB) of the Ozark and Ouachita forests at a high resolution of 10 meters. Using GEDI biomass data and integrating optical data from Sentinel-2, SAR data from Sentinel-1, and relative forest height data from GEDI's Level 2A product on the Google Earth Engine platform, the study employed Random Forest (RF) regression to model AGB. The study highlights the importance of integrating topographical, spectral, and textural data to enhance AGB model accuracy and interpretability. A total of 34 out of 154 variables were selected for the final model, demonstrating strong correlations with measured biomass. The RF model achieved an R-squared value of 0.95 and an RMSE of 18.46 for the training dataset, and an R-squared value of 0.75 and an RMSE of 34.52 for the validation dataset, indicating robust performance. The key predictors included elevation, forest height metrics such as RH100, RH98, and RH95, and vegetation indices including Enhanced Vegetation Index (EVI), Normalized Difference Vegetation Index (NDVI), and Leaf Area Index (LAI). The study also extrapolated historical biomass from 2014 to 2023 using Landsat-8 data and the image normalization technique. The extrapolated biomass data effectively identified key spatial patterns, demonstrating the model's efficacy. This research features the use of open-source cloud computing platforms and the importance of integrating topographical, spectral, and textural data to enhance AGB model accuracy and interpretability at a regional scale. The results support precise estimation of fire-related emissions and strategic planning to improve forest health and sustainability, contributing significantly to biodiversity conservation and carbon sequestration efforts.