https://doi.org/10.1175/JHM-D-24-0092.1

In recent decades, there has been a surge in the availability of precipitation datasets, each leveraging diverse data sources and sophisticated algorithms. However, the density of the gauged station network presents a substantial challenge in thoroughly assessing these datasets, particularly in regions with sparse coverage. This study seeks to quantify the observational uncertainty of precipitation across North America by introducing a novel framework. The framework is designed to detect potential outliers, categorize datasets based on their level of agreement, and operate at both monthly and daily time steps. Eleven datasets spanning from 2002 to 2017 are investigated: ANUSPLIN, CaPA, CHIRPS, CMORPH, DAYMET, ERA5, GSMaP, IMERG, MSWEP, PERSIANN and PRISM. Analyses at the grid-cell scale revealed that on average and across the entire domain, excluding outlier datasets reduced observational uncertainty by ~17% for the monthly time step and ~39% for the daily time step. Aggregating results over 24 hydrological entities provided a broader perspective and revealed that certain datasets were consistently classified in the same group. Notably, MSWEP was always in the high-consistency group, while PRISM and CaPA were predominantly in that group. Satellite-based datasets (CHIRPS, CMORPH, GSMaP, IMERG) were mainly classified in the high-consistency group during summer, reflecting their ability to capture convective precipitation. ERA5 appeared in both the high- and low-consistency groups, depending on the season and region. Although our framework identified sets of consistent datasets and potentially reduces uncertainties, we emphasize that uncertainties remained and encourage the scientific community, to consider these observational uncertainties in their analyses.

https://doi.org/10.1175/JHM-D-24-0092.1