Global vegetation resilience prediction model based on stacked machine learning

In response to the challenges associated with the accurate quantification and dynamic monitoring of vegetation drought resilience under global climate change, this study proposes a Stacking-RS model integrating multi-source remote sensing data and ensemble learning. Firstly, a Synthetic Standardized Vegetation Index (SA) was constructed to sensitively characterize vegetation status under drought stress. Secondly, a resilience index was quantified based on the degree of vegetation status loss and the corresponding recovery time. Finally, a high-precision resilience assessment model was established, with XGBoost, Random Forest, and LightGBM serving as base learners, and CatBoost as the meta-learner. Validation results demonstrated that the model achieved a coefficient of determination (R2) of 0.8195, a mean absolute error (MAE) of 0.05034, and a root mean square error (RMSE) of 0.05975. Accuracy evaluation revealed significant spatial heterogeneity in model performance, which increased from low latitudes to high latitudes. Among different vegetation types, the model exhibited the highest accuracy for coniferous forests and the lowest for grasslands. Factor analysis indicated that water conditions (including precipitation and soil moisture) and temperature are the dominant environmental factors regulating vegetation drought resilience.