A downscaling-calibrating framework for generating gridded daily precipitation estimates with a high spatial resolution.

• Geodetector model offers a promising method for downscaling variables selection. • GWR is suitable for downscaling before calibrateing satellite using gauge observations. • Newly proposed GWR_Bi-LSTM framework outperformed the comparative. • Calibrated precipitation demonstrated better accuracy than the original IMERG. Accurate precipitation datasets with a high spatiotemporal resolution are crucial for regional hydrological simulation and water resources management. Although various global satellite precipitation products (SPPs) are available, they cannot be directly applied to small-scale research due to their low spatial resolution and significant bias. To solve these problems, based on the consideration of spatiotemporal relationships, this paper presented a comprehensive downscaling-calibrating framework that consists of spatial downscaling procedures and calibrating processes. Specifically, a geographically weighted regression (GWR) model, coupled with the environmental variables selected by the Geodetector model, was applied to generate uncorrected daily precipitation data with high-spatial resolution (1 km × 1 km). Then, the preliminary downscaled daily precipitation was calibrated using a bidirectional long short-term memory (Bi-LSTM) network. The proposed framework was applied to downscale and calibrate the Integrated Multi-satellitE Retrievals for Global Precipitation Measurement (IMERG) product of the Jiulong River Basin in southeast China from 2010 to 2016. The results indicate that: (1) the spatial downscaling procedures improved the spatial resolution of the SPPs while maintaining its accuracy; (2) the calibrating processes can effectively correct the bias of preliminary downscaled daily precipitation. Meanwhile, the accuracy evaluation showed that the proposed GWR_Bi-LSTM framework outperformed the comparative methods, and the generated precipitation dataset exhibited a better performance under various evaluation conditions. Generally, based on the proposed framework, gridded daily precipitation data with high quality and spatial resolution can be generated for hydrological and meteorological research. [ABSTRACT FROM AUTHOR]