Spatiotemporal Dynamics and Drivers of Agricultural Drought in the Huang-Huai-Hai Plain Based on Crop Water Stress Index and Spatial Machine Learning

Agricultural drought poses a critical constraint to food security and regional sustainable development, particularly in the Huang-Huai-Hai Plain, a major grain-producing region characterized by high spatial heterogeneity in drought risk. Previous studies have demonstrated that the Crop Water Stress Index (CWSI) outperforms traditional meteorological indices in detecting agricultural droughts in various regions. However, there is limited research specifically focusing on its spatiotemporal dynamics and the complex relationships with environmental factors, particularly in the Huang-Huai-Hai Plain. To fill this gap, this study first estimated CWSI using remote sensing evapotranspiration data and systematically assessed the spatiotemporal dynamics of agricultural drought in the Huang-Huai-Hai Plain from 2005 to 2020. Then, an integrated analytical framework that combines Local Indicators of Spatial Association (LISA) with Random Forest (RF) modeling has been proposed to identify primary environmental drivers. Results revealed a general downward trend in CWSI over the study period, with drought hotpots primarily concentrated in the central plains and along the eastern foothills of the Taihang Mountains. LISA identified four distinct spatial cluster types and revealed significant spatial associations between CWSI and six environmental variables. The major driving factors of CWSI included vegetation conditions (NDVI), land surface temperature (LST), rainfall, and temperature-related factors (SAT, DSR), with LST and SAT exhibiting the strongest correlations with CWSI in multiple regions. Among these, LST and SAT exhibited strong positive correlations with CWSI in multiple regions. By integrating spatial clustering and variable importance analysis, we found that agricultural drought patterns are shaped by interacting environmental factors, with region-specific dominant mechanisms. This study provides a novel analytical framework that bridges remote sensing, spatial statistics, and machine learning, offering valuable insights and tools for drought monitoring and attribution at regional scales.