Estimating Forest Above-ground Biomass Using Sentinel-2 and SDGSAT-1 with Predictive Regression Models

Abstract

Forests are vital for maintaining terrestrial biodiversity, sequestering carbon in vegetation and soil, regulating atmospheric carbon levels, and mitigating global warming. Determining above-ground biomass in forests is essential for sustainable forest resource management, effective carbon management, monitoring forest health, conserving biodiversity, managing resources, and enhancing climate resilience. This study aims to explore the effectiveness of Sentinel-2 and SDGSAT-1 optical images, vegetation indices, Gray-Level Co-Occurrence Matrix (GLCM) texture metrics, vegetation biophysical variables, and topographical variables for estimating Above-Ground Biomass (AGB) in deciduous, coniferous, and mixed forest areas located within the buffer zone of Kure Mountains National Park (KMNP) in Turkey. The research utilized three types of regression models: Multiple Linear Regression (MLR), Partial Least Squares (PLS) and Theil-Sen regression. The Least Absolute Shrinkage and Selection Operator (LASSO) method was employed for feature selection. Shapley Additive Explanations (SHAP) were used to interpret the model outputs and assess the importance of each feature. The Theil-Sen regression model emerged as the most effective AGB estimation, with R-2 = 0.79, MAE = 22.97 t/ha, RMSE = 28.38 t/ha, and rRMSE = 9.95%. The analysis indicated that integrating Sentinel-2 and SDGSAT-1 satellite data, especially Sentinel-2 band 11 and SDGSAT-1 band 6, along with Sentinel-2 Chlorophyll Red Edge (CIRed-Edge), Green Difference Vegetation Index (GDVI), and Weighted Difference Vegetation Index (WDVI), Fcover vegetation biophysical variables, SDGSAT GLCM dissimilarity, and altitude topographical variables, improved the precision and reliability of AGB estimation.