Opposite spatiotemporal sensitivity of terrestrial vegetation to intensified extreme precipitation across China's aridity gradients

Extreme precipitation has intensified in many regions under global warming, yet how vegetation sensitivity to extreme precipitation varies along aridity gradients and changes over time remains poorly understood. Here we quantified vegetation sensitivity to extreme precipitation ( β EP ) across China during 1982–2020 using a Bayesian dynamic linear model (DLM) based on long-term NDVI and precipitation datasets, and disentangled its spatiotemporal drivers using Random Forest and SHAP attribution. Vegetation exhibited significantly positive β EP values across 72% of dryland grids (mean = 0.03), with grasslands contributing the largest positive β EP values, whereas only 37% of humid grids had positive β EP and 43% showed negative values (mean = −0.005), with forests exhibiting weak negative responses. Temporally, β EP increased significantly in arid and semi-arid regions but declined in humid ecosystems, with widespread directional turning of sensitivity observed after 1999 across all aridity zones. The long-term mean β EP was primarily controlled by water availability in drylands and solar radiation in humid regions. In contrast, elevated CO 2 exerted positive effects on the interannual fluctuations of β EP in dry regions but negative effects in humid regions. These gradient-driven contrasting responses require differentiated ecological policies for vegetation dynamics projection and extreme precipitation risk mitigation.