Abstract
<title>Abstract</title> <p>Rangelands are vital ecosystems that sustain rural livelihoods and ecological equilibrium; however, they experience significant degradation due to overgrazing and climate change. Field-based techniques for estimating above-ground biomass (AGB) incur substantial labor expenses, face scaling challenges in hilly regions, and encounter many additional complications. Thus, machine learning (ML) and remote sensing may replace extensive monitoring. To fulfil this requirement, our research developed a geospatial framework for estimating AGB in Upper Dir and Swat, Trans-Himalayan rangelands. Field-derived dry matter yield (DMY) assessments from the 2022 and 2023 growing seasons were integrated with Sentinel-2 multispectral imagery. To identify the most effective spectral predictor variables, we initially assessed 50 vegetation indices. We then selected the most significant vegetation indices strongly correlated with biomass: NDVI, REDVI1, REDVI2, NDRE1, GNDVI, and GDR.. Machine learning techniques, including random forest (RF), support vector machine (SVM), and partial least squares regression (PLSR), were evaluated. The Nash–Sutcliffe Efficiency (NSE) coefficient and R² metrics were employed to assess three ML algorithms. Vegetation indices frequently exhibited strong correlations with biomass, as indicated by previous studies; specifically, the Red Edge Difference Vegetation Index 1 (REDVI1) achieved an R² value of 0.86, while NDVI attained an R² value of 0.7376 in 2023. The RFR algorithm surpassed PLSR and SVR, exhibiting the highest predicted accuracy and little variability. Spatial mapping revealed regional variations in forage availability, with an average AGB of 5,896.78 kg/ha across the research area, estimated using REDVI2 regression model. RFR and high-resolution Sentinel-2 data offer an accurate, scalable, and non-invasive method for quantifying rangeland biomass in challenging mountainous environments. This evidence-based methodology offers a fundamental scientific foundation for guiding adaptive land management and safeguarding pastoral livelihoods in vulnerable Trans-Himalayan ecosystems.</p>