Enhanced error correction and spatial downscaling of precipitation and air temperature in the middle and low reaches of the Yangtze River using a random forest model with the Sokol method
摘要
Disease vectors like mosquitoes and snails display distinct point, linear, or patchy distributions shaped by environmental drivers. Fine-scale vector mapping is essential for public health decision making or environmental management. However, coarse remote sensing data (above 5–10 km resolution) often fail to capture fine-scale environmental features. To improve this, current downscaling methods need optimization. Integrating ground measurements for error correction and using locally adapted factors (such as urban heat metrics for temperature and coastal proximity for precipitation) can enhance the resolution of sub-grid variability. This is particularly important in the Yangtze River basin, where characterized by complex monsoon-climate interactions and rapid urbanization processes. This study developed an integrated downscaling framework combining: (1) Sokol-based error correction using ground observations, and (2) random forest algorithms with optimized predictors—including nighttime light (urban heat), vegetation indices (evapotranspiration), and coastal proximity (land-sea interactions)—to enhance temperature and precipitation downscaling. Building upon foundational predictors (geolocation, topography, and land surface temperature), the precipitation downscaling model integrating coastline proximity and enhanced vegetation index achieved a 6.49%