Integrating grassland height for Enhanced aboveground biomass estimation in northern China

Accurate estimation of grassland aboveground biomass (AGB) is crucial for terrestrial carbon cycling, global climate change research, degradation assessment, and sustainable land management. This study employs XGBoost model, combined with feature selection via Random Forest & Pearson correlation, alongside SHapley Additive exPlanations (SHAP), to enhance AGB predictions across diverse grassland ecosystems in China. Results indicate that incorporating vegetation height significantly improves model performance, increasing test R2 values by 0.01–0.07 (final range: 0.59 to 0.68), and reducing the errors nRMSE to ≤ 0.04. This underscores the critical role of vegetation height in improving biomass estimation accuracy. SHAP analysis further reveals the relative importance of key predictors, offering insights into their individual contributions to model performances. Spatiotemporal analysis (2001–2021) reveals rising AGB trends in highly productive regions, whereas arid and degraded grasslands exhibit stability or continue to decline, highlighting their vulnerability to climatic changes and anthropogenic pressures. Although the model demonstrates strong predictive capability, regional heterogeneity and complex feature interactions warrant further investigation. This research highlights the effectiveness of machine learning combined with remote sensing in monitoring grassland degradation, providing valuable insights for ecosystem restoration, carbon sequestration strategies, and policy-driven conservation efforts.