Climate extremes increasingly shape vegetation dynamics in drylands, yet most modelling frameworks still treat spatial heterogeneity and nonlinear feedbacks separately. Here, we propose Geographically Weighted XGBoost (GW-XGBoost), a hybrid and interpretable framework that injects local coefficients from geographically weighted regression into an Extreme Gradient Boosting ensemble and explains predictions with SHapley Additive exPlanations (SHAP). The model is calibrated for the Middle East using a 30-year Landsat NDVI time series (1995–2024), ERA5-derived temperature and precipitation extreme indices, and topographic predictors. Using four baseline models (OLS, GWR, XGBoost, GW-XGBoost), our hybrid approach consistently delivers the best seasonal performance, with R2 = 0.80–0.85 and RMSE ≤ 0.02, and a mean cross-validated R2 of 0.82 (RMSE = 0.037), improving explained variance by 8–15% over benchmark models. SHAP diagnostics indicate a region-wide ecological transition. Winter and early-spring greenness, formerly constrained by cold extremes and multi-day rainfall, is now increasingly controlled by warm-percentile temperature thresholds. Summer and autumn vegetation have become more dependent on short-lived moisture pulses superimposed on intensifying heat. Spatially, compound-sensitive zones expand markedly in spring and summer, whereas comparatively resilient areas persist mainly along the Persian Gulf margins, the southern Arabian Peninsula and high-altitude belts. Emerging hotspots are identified along the Zagros and Taurus ranges, coastal corridors and hyper-arid interiors where concurrent heat, cold and precipitation anomalies converge. By integrating spatially adaptive regression, nonlinear ensemble learning and explainable AI on multi-source Earth observation data, GW-XGBoost provides high-resolution sensitivity layers that can support drought early-warning, water-resource planning and ecosystem-based adaptation in one of the world’s most climate-vulnerable regions.
Khosravi et al. (Tue,) studied this question.