- University: University of Copenhagen
- Place of defence: Copenhagen
Abstract
Accurate above-ground biomass (AGB) estimation supports sustainable forest management, carbon accounting, and climate mitigation. This study evaluated Landsat-derived vegetation indices (NDVI, EVI, NBR) from different seasonal windows with linear, quadratic, and Random Forest models to estimate AGB in Nepal’s Kankali Community Forest using field data from 2005–2025. Biomass increased substantially over two decades, reflecting effective community forestry management. Post-monsoon (November) indices, particularly EVI, exhibited the strongest correlations with AGB, while dry season (March) indices showed weakened relationships. Among the tested models, the Random Forest algorithm with November-only vegetation indices achieved the highest overall accuracy in this study (test R² = 0.66, RMSE = 79.49), marginally ahead of multi-season data (R² = 0.62). Incorporating elevation and coarse climate variables produced negligible accuracy gains, likely due to the site’s narrow environmental gradients and the strong predictive power of spectral indices alone. Findings confirm the value of optimally timed optical imagery and robust nonlinear models for scalable biomass monitoring, with clear application potential for REDD+, forest inventories, and carbon stock assessment in similar lowland tropical systems.
