Discrepancies in changes in precipitation characteristics over the contiguous United States based on six daily gridded precipitation datasets

Variability and spatiotemporal changes in precipitation characteristics can have profound socioenvironmental impacts. Several studies have shown that the frequency and/or magnitude of precipitation events have changed over the contiguous United States (CONUS) in the past decades. Most previous studies used only one precipitation dataset and only investigated mean or extreme precipitation. Here, using 6 gridded daily precipitation datasets, we show that there are substantial discrepancies in the changes in characteristics of both extreme and non-extreme precipitation events from 1983 to 2017. Our results highlight that using a single record to study precipitation changes can potentially lead to biased results. Using different datasets enables examining the overall agreements and discrepancies in precipitation characteristics. For example, we show that almost all datasets agree that some areas show statistically significant changes in the annual precipitation maxima; however, the locations and signs of changes are not consistent across datasets. There is a relative agreement between datasets on changes in the total annual precipitation. When examining other percentiles of the precipitation distribution, including non-extreme values, however, we find widespread discrepancies among different precipitation products (e.g., what part of the precipitation distribution is changing). In fact, depending on the source of data, there exist opposing trends and patterns of change in precipitation characteristics. This highlights the need to further investigate non-extreme precipitation events to unravel potential non-extreme but "unexpected" or "unusual" patterns. Finally, we argue that protocols for data selection are needed to address the issue of inter data variability and to ensure reliability of statistical analysis.