The conditional bias of extreme precipitation in multi-source merged datasets

Authors: KANG, X - Tianjin University; DONG, J - Tianjin University; Crow, Wade; WEI, L - Nanjing University; ZHANG, H - Tianjin University

Submitted to: Geophysical Research Letters
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 11/1/2024
Publication Date: 11/22/2024
Citation: Kang, X., Dong, J., Crow, W.T., Wei, L., Zhang, H. 2024. The conditional bias of extreme precipitation in multi-source merged datasets. Geophysical Research Letters. 51(22). Article e2024GL111378. https://doi.org/10.1029/2024GL111378.
DOI: https://doi.org/10.1029/2024GL111378

Interpretive Summary: Remote sensing is a crucial tool for monitoring large-scale precipitation patterns and tracking the extent and severity of water cycle extremes (i.e., floods and droughts). However, remotely sensed precipitation datasets often contain substantial biases with regards to the intensity of extreme precipitation events. To minimize these errors, remotely sensed datasets are commonly merged with other precipitation datasets to derive a unified product with minimized errors. However, new results in this paper illustrate that such merging is not generally an effective way to remove extreme precipitation biases. Instead, we describe alternative data processing techniques that are likely to be much more effective and should, therefore, be prioritized when designing optimal precipitation merging approaches. As such, this study describes an improved set of priorities for efforts to better utilize remotely sensed precipitation for water resource and hazard applications.

Technical Abstract: Multi-source data merging is widely applied to enhance large-scale precipitation estimates. However, these merged datasets usually contain substantial conditional biases with respect to extreme precipitation (EP) events – undermining their utility for flood forecasting and related analyses. EP biases are typically attributed to merging algorithms, which attenuate precipitation variability to minimize merging error. Here, we demonstrate that merging algorithms are responsible for less than 1% of total EP bias in merged precipitation datasets. Instead, EP biases arise mainly from the remote sensing (RS) and reanalysis precipitation pre-processing prior to data merging. Specifically, current data-merging frameworks only correct the monthly means or statistical distribution of the RS/reanalysis precipitation inputs during pre-processing. Such procedures are insufficient for adjusting EP timing uncertainties, which eventually propagate into the merged dataset as an EP bias. Therefore, developing algorithms that iteratively adjust EP timing errors and intensity should be prioritized in future precipitation merging frameworks.