Comparison of multi-source merged precipitation products using independent gauge observations

Abstract

Data merging is widely applied to improve large-scale precipitation estimates. However, traditional merging algorithms rely heavily on gauge observations and suffer from increased uncertainties in data-sparse regions. Statistical uncertainty analysis (SUA) offers a potential solution by estimating merging weights analytically and thereby reducing dependence on gauge data. However, the comparative effectiveness of SUA-merged and traditional gauge-merged precipitation datasets remains unclear, largely due to the scarcity of truly independent validation data. Here, using 268 wholly independent rain gauges, we show that SUA-based merging can effectively suppress random errors and outperform remote sensing and reanalysis products. Notably, compared to traditional gauge-merged datasets, SUA-merged precipitation demonstrates averagely stronger correlation with independent observations and lower root-mean-square errors. These results provide direct evidence for the ability of SUA to mitigate reliance on gauge data, especially in observation-scarce regions. However, SUA-merged products still show limitations with regards to accurately classifying rain/no-rain events, highlighting the need for future enhancements targeting false precipitation detection.