Data fusion of satellite imagery and downscaling for generating highly fine-scale precipitation

Due to the frequent occurrence of extreme precipitation events on a global scale, accurate estimation of regional precipitation has emerged as a critical concern. Specifically, region-scale hydrological modeling demands precipitation data with high spatiotemporal resolution and precision. Existing research has primarily concentrated on the correction and spatial downscaling of precipitation products. However, a considerable challenge persists in concurrently generating precipitation data with three key characteristics: high precision, high spatiotemporal resolution, and high spatial coverage (termed '3H'). This entails the provision of daily precipitation data at no more than a 1 km resolution, encompassing a full spatial extent. To address this challenge and obtain 3H precipitation data for regional hydrology research, this study introduces a multi-source precipitation data fusion and downscaling approach known as the 'Generate high Resolution, Accurate, Seamless data using Point-Surface (GRASPS) fusion method. This method combines the strengths of several satellite and model data sources to produce a more precise precipitation dataset at a daily scale and 1 km resolution, covering the Wuhan Urban Agglomeration. These sources include the Integrated Multi-satellitE Retrievals for Global Precipitation Measurement (IMERG) from the Global Precipitation Measurement (GPM) mission, Tropical Rainfall Measuring Mission (TRMM) Multi-satellite Precipitation Analysis (TMPA), and Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks – Climate Data Record (PERSIANN-CDR). Validation against precipitation data from 36 ground gauges yielded a Pearson Correlation Coefficient of 0.77, with Root Mean Squared Error, Mean Absolute Error, and Bias reduced to 6.08 mm, 2.20 mm, and −0.13 mm, respectively. Compared to prior studies, this research not only improved the spatial resolution of the precipitation dataset to 1 km but also enhanced the accuracy of extreme precipitation, resulting in an accuracy increase from 76.92 % to 91.67 %. Additionally, the generated precipitation dataset exhibited excellent performance at both daily and monthly scales. In terms of different land-cover types, the proposed method displayed improved performance in urban areas. Furthermore, the data obtained was subjected to testing across different input variables, precipitation levels, and downscaling algorithms. In conclusion, this study successfully obtained 3H precipitation data to bridge the gap in high-quality and fine-scale precipitation data. The proposed method and the generated dataset hold substantial implications for regional hydrology research and its practical applications.