Hourly calibration algorithm for Fengyun-2G quantitative precipitation estimates using spatial random forest and improved temporal disaggregation scheme

Abstract

Fengyun satellite series offers Quantitative Precipitation Estimates (QPEs) with high spatiotemporal resolution, providing coverage at both regional and global scales. However, as indirect estimates of precipitation based on infrared data, Fengyun-2G (FY-2G) QPEs inherently contain considerable regional and seasonal uncertainties. In this study, to mitigate these errors, we propose a novel calibration algorithm that enhances the spatial and temporal performance of FY-2G QPE at both daily and hourly scales. This method consists of two components: first, a daily-scale spatial calibration using a multi-variable-assisted Random Forest (RF) technique, and second, an hourly-scale temporal calibration based on an Improved Proportional Factor (IPF) method, which refines traditional PF approaches by integrating Geographic Difference Analysis (GDA) and Inverse Distance Weighting (IDW) for error correction. The proposed approach was applied to generate a high-quality FY-2G precipitation dataset for the Chinese mainland (0.1°/hr, 2018–2019). Validation against the China Merged Precipitation Analysis (CMPA) dataset demonstrates significant improvements: Daily-scale spatial calibration using RF increases the correlation coefficient (CC) from 0.74 to 0.92 (+24.3 %), reduces BIAS from 15.69 % to 8.04 % (−48.8 %), and decreases RMSE from 5.97 mm to 2.93 mm (−50.9 %). Hourly-scale temporal calibration using RF-IPF improves CC from 0.47 (FY-2G) to 0.77, reduces BIAS from −36.73 % to 8.25 %, and lowers RMSE from 2.98 mm to 2.11 mm during extreme precipitation events. These results highlight the effectiveness of the proposed calibration framework in enhancing both spatial and temporal accuracy of infrared-based QPEs. Unlike existing methods that primarily focus on daily-scale corrections, our approach extends calibration to the hourly level, improving precipitation estimates crucial for short-duration extreme weather events. While this methodology is demonstrated in China, its framework can be adapted to other satellite-based precipitation datasets and applied across diverse climatic and topographic regions, providing valuable insights for hydrological and climatic research worldwide.