Integrating multi-dimensional features for remote sensing–based drought monitoring and driver analysis during the vegetation growing season: a case study in northern Xinjiang

Drought disasters severely constrain ecological security and agricultural sustainability in Xinjiang, and reliable monitoring across diverse vegetation types remains challenging because of the vegetation saturation and soil background effects in traditional indices. This study integrates the kernel-based NDVI (kNDVI), land surface temperature (LST), and precipitation into the TVDI framework to develop a three-dimensional temperature–vegetation–precipitation drought index (kTVPDI). The spatiotemporal evolution of drought from 2000 to 2024 was assessed using the Sen slope, Mann‒Kendall test, and Loess decomposition, and SHAP-based machine learning was applied to identify dominant drivers. The results show that kTVPDI has a significantly stronger correlation with soil moisture (r = −0.875) than TVDI, demonstrating improved monitoring accuracy. Drought severity increases from peripheral to the central regions and peaks at the beginning and end of the growing season. The mean annual drought intensity follows the order: desert > grassland > cropland > forest, and 66% of northern Xinjiang exhibits a slight upward trend. Driver analysis identifies potential evapotranspiration and wind speed as major contributors to drought intensification, whereas elevation and slope mitigate drought severity. Overall, the kTVPDI offers enhanced sensitivity and stability, providing a more robust tool for drought monitoring and supporting ecological management and risk mitigation in other arid and semi-arid regions globally.