Abstract Grassland aboveground biomass density (AGBD) is a key indicator for assessing grassland carbon sinks and ecosystem functioning. With the rapid expansion of satellite observations, remote sensing has been widely applied to estimate grassland AGBD. AGBD is highly sensitive to environmental gradients such as precipitation, temperature, and topography; however, this contextual dependence remains insufficiently assessed in remote-sensing estimates. Focusing on grasslands in China’s Ili River Basin, this study uses GEDI L4A footprint-level AGBD as the response variable and integrates multi-source predictors from Sentinel-2 optical data, Sentinel-1 SAR, GLO-30 topography, and TerraClimate. A LASSO-screened, LightGBM-based model for AGBD retrieval was developed, and its robustness and feature mechanisms were evaluated across elevation, slope, precipitation, and temperature gradients. Results show an overall accuracy of R²=0. 445, RMSE = 54. 62 Mg/ha, and MAE = 27. 90 Mg/ha. Along topographic gradients, the model fits best at elevations of 2000–2500 m (R²=0. 543) and on slopes of 0–10 ^ (R²=0. 529) ; along climatic gradients, performance is higher where annual precipitation <300 mm (R²=0. 476) and is optimal at mean annual temperatures of 0–5 ^ C (R²=0. 523). SHAP interpretations indicate that optical reflectance and textures dominate in low-elevation, gentle-slope areas; where terrain is complex or precipitation is higher, the importance of optical textures and radar features increases; above 3000 m, the contribution of optical features declines markedly while texture/topography/radar contributions rise. This study provides a basis for context-aware AGBD mapping in the Ili River Basin and similar regions.
Erzheng Liang (2026) studied this question.