High-resolution humid heat stress indices, characterizing the interactive physiological effects of temperature and humidity, are vital for precision heat-health assessment and climate adaptation. However, existing coarse-resolution datasets smooth fine-scale spatial heterogeneity and may therefore underestimated localized heat extremes. Statistical downscaling can enhance spatial resolution, but its reliance on empirical relationships provides limited constraints on nonlinear land-atmosphere interactions, potentially introducing physically inconsistent biases. Here, we develop a physics-informed spatiotemporal (PIST) downscaling model that integrate surface energy balance and horizontal advection into a statistical framework. By fusing multisource data, our model reconstructs for humid heat stress indices at hourly and 1-km resolution. Applied to the Yangtze River Delta during the warm season (May–October) of 2020, the model performs consistently well across the four indices, with root mean square errors (RMSEs) ranging from 2.24 to 3.98 °C and R2 values exceeding 0.88. Compared with conventional statistical downscaling, PIST reduces RMSE by 53–62%. The incorporation of physical constraints also improves the representation of extreme heat, increasing upper-tail dependence from 0.4 to above 0.8 and reducing systematic biases over urban areas. Overall, PIST exemplifies Digital Earth methodology by transforming geospatial observations into high-resolution, decision-relevant climate information, supporting heat health assessment and urban climate adaptation.

正在加载阅读记录…