Increasingly frequent drought events posed significant threats to vegetation stability in the ecologically sensitive upper Yangtze River basin. This study systematically analyzed vegetation response characteristics to drought based on NDVI and multi-scale SPEI data from 1990 to 2022, and developed an integrated drought sensitivity index that incorporated both response magnitude and temporal scales, and quantified vegetation loss risks under varying drought intensities using a Copula-Bayes framework. The results showed that: (1) Vegetation responses to drought exhibited marked spatial and temporal heterogeneity, with cumulative effects dominating across the basin and average response periods ranging from 5 to 8 months. (2) Drought sensitivity demonstrated a bimodal pattern, with higher sensitivity observed at both the arid and humid ends of the moisture gradient, while some humid regions,particularly forest and grassland areas also showed unexpectedly high sensitivity. (3) Under drought scenarios, the likelihood of slight vegetation loss is relatively high, while severe loss remains low overall. The risk of loss differs markedly across regions and vegetation types, and drought sensitivity does not fully align with actual loss probability. Moreover, this study offers methodological and conceptual insights for future research on vegetation–climate interactions, drought risk assessment, and ecosystem resilience under changing climate conditions.Our findings reveal the nonlinear responses of ecosystems to extreme drought and provide a quantitative framework for assessing vegetation sensitivity and loss risk under extreme drought conditions in regions with atypical drought characteristics. • Unified Copula–Bayesian model captures lag and memory in drought–vegetation risk. • Cumulative drought effects on vegetation were more significant than lag effects. • Probabilistic loss decouples from sensitivity across drought scenarios. • Sensitivity is bimodal across moisture gradients including humid biome.
Xu et al. (Sat,) studied this question.