The Conditional Bias of Extreme Precipitation in Multi‐Source Merged Data Sets.

Multi‐source data merging via weighted average (WA) is widely applied to enhance large‐scale precipitation estimates. However, these data sets usually contain substantial conditional biases with respect to extreme precipitation (EP) events—undermining their utility for extreme event analysis. Nevertheless, the main source of such EP biases remains unknown. Here, we demonstrate that WA algorithms are responsible for less than 1% of total EP biases. Instead, EP biases originate from the multi‐source precipitation inputs, which are not adequately adjusted prior to WA. Specifically, current data‐merging frameworks only correct the monthly means or statistical distributions of the remote sensing/reanalysis precipitation inputs prior to WA. Such procedures are insufficient for adjusting EP timing uncertainties, which eventually propagate into the WA‐based merged data set as an EP bias. Therefore, developing algorithms that iteratively adjust EP timing and intensity errors should be prioritized in future precipitation merging frameworks. Plain Language Summary: Remote sensing (RS) and reanalysis systems are crucial for estimating large‐scale precipitation. Weighted averaging (WA) of different data sets can enhance overall precipitation estimation accuracy and has been widely applied for generating global precipitation data sets. However, WA algorithms often lead to biases for extreme precipitation (EP). Such issues undermine the usefulness of WA‐based precipitation data sets for flood forecasting. This study investigates the sources of EP biases in WA frameworks, based on surface precipitation gauge observations and numerical experiments. Results show that the WA algorithms themselves contribute less than 1% to EP biases. Instead, most EP bias is related to RS/reanalysis data correction procedures. Specifically, current WA methods only adjust the monthly means or general statistical distributions of the input data. However, EP occurrence errors are often neglected during the precipitation correction. This means that the timing and location of EP as estimated by different data sets are not entirely consistent, leading to substantial biases when they are averaged. Therefore, to improve the accuracy of EP estimates, it is important to develop preprocessing methods that better account for both the timing and intensity errors of extreme events. Key Points: We investigate sources of bias in extreme precipitation (EP) estimates provided by commonly used data merging frameworksWe demonstrate that EP biases arise from the neglect of EP timing error correction and not the merging algorithmAlgorithms that iteratively adjust the EP intensities and timing errors should be prioritized in future merging frameworks [ABSTRACT FROM AUTHOR]