Analysis of spatiotemporal variations and driving forces of NDVI based on random forest and Geodetector: A case study of the Heihe river basin in China

Against the backdrop of global climate change, ecosystems in arid inland river basins are experiencing increasing environmental pressures. Long-term, high-resolution monitoring of vegetation dynamics is therefore essential for assessing ecological conditions and identifying potential risks. Based on the annual maximum NDVI time series from 1990 to 2024 in the Heihe River Basin, this study systematically characterizes the spatiotemporal patterns of vegetation change by integrating trend analysis, the Geodetector model, residual diagnostics, and time-lag zoning methods. Furthermore, it quantitatively examines the relative influences of climatic factors and human activities on vegetation dynamics and their phased adjustment characteristics. The results reveal a significant overall greening trend across the Heihe River Basin during the study period, with NDVI exhibiting a clear spatial gradient characterized by higher values in the upstream regions and lower values downstream. Geodetector analysis indicates that hydrological factors—particularly precipitation and soil moisture—consistently exert strong explanatory power on vegetation change, with q values generally exceeding 0.59 throughout the study period. Meanwhile, the explanatory power of aridity indices for NDVI shows a long-term declining trend, suggesting phased adjustments in the statistical relationship between vegetation dynamics and aridity conditions. Residual diagnostics further demonstrate marked differences in the relative contributions of non-climatic factors to vegetation change around the year 2000. Hurst index analysis suggests that approximately 96% of the basin exhibits stable or persistently improving vegetation trends, while potential degradation risks are mainly concentrated in desert fringe areas. Overall, this study elucidates the pronounced spatiotemporal heterogeneity of vegetation dynamics and their driving mechanisms in the Heihe River Basin. • From 1990 to 2024, the Heihe River Basin exhibited a greening trend, with a mean NDVI increase of 0.00032 a⁻¹. • Long-term vegetation dynamics showed a clear multi-stage evolution of degradation–stagnation–recovery–stabilization. • Hydrological variables—particularly precipitation and soil moisture—dominate the spatial differentiation of NDVI. • Vegetation sensitivity to drought weakened, with the drought index q value declining from 0.393 in 1990 to 0.066 in 2024. • Residual analysis indicates an increasing trend in the positive relative contribution of non-climatic factors to NDVI.