Estimation of grassland aboveground biomass in the Three-Rivers Source Region with explainable machine learning

Abstract

Grassland aboveground biomass (AGB) is a key indicator of ecosystem function. While machine learning (ML) has improved AGB estimation from remote sensing, limited interpretability restricts its application in management. This study applied the SHapley Additive exPlanations (SHAP) method with the optimal ML model in the Three-River Source Region (TRSR) to quantify the main and interactive effects of climatic drivers on AGB and reveal their nonlinear responses. Results showed that vegetation indices contributed most to AGB estimation. AGB showed threshold responses to temperature and precipitation, with peak positive effects at 10–12 °C for current-month temperature. Warming enhanced AGB under low antecedent temperature or moderate precipitation but had diminishing or negative effects under extreme hydrothermal conditions. From 2003 to 2022, AGB increased in 56.4 % and declined in 35.15 % of the area. This study provides an interpretable AGB model and insights into climate-biomass relationships, supporting adaptive grassland management in alpine regions.