Uncertainty and outliers in high‐resolution gridded precipitation products over eastern North America

Several observational precipitation products that provide high temporal (≤3 h) and spatial (≤0.25°) resolution gridded estimates are available, although no single product can be assumed worldwide to be closest to the (unknown) “reality.” Here, we propose and apply a methodology to quantify the uncertainty of a set of precipitation products and to identify, at individual grid points, the products that are likely wrong (i.e., outliers). The methodology is applied over eastern North America for the 2015–2019 period for eight high‐resolution gridded precipitation products: CMORPH, ERA5, GSMaP, IMERG, MSWEP, PERSIANN, STAGE IV and TMPA. Four difference metrics are used to quantify discrepancies in different aspects of the precipitation time series, such as the total accumulation, two characteristics of the intensity‐frequency distribution, and the timing of precipitating events. Large regional and seasonal variations in the observational uncertainty are found across the ensemble. The observational uncertainty is higher in Canada than in the United States, reflecting large differences in the density of precipitation gauge measurements. In northern midlatitudes, the uncertainty is highest in winter, demonstrating the difficulties of satellite retrieval algorithms in identifying precipitation in snow‐covered areas. In southern midlatitudes, the uncertainty is highest in summer, probably due to the more discontinuous nature of precipitation. While the best product cannot be identified due to the lack of an absolute reference, our study is able to identify products that are likely wrong and that should be excluded depending on the specific application.