Abstract We present the first comprehensive hourly-scale evaluation and intercomparison of eight widely used gridded precipitation datasets against 1, 568 in situ gauge stations over the contiguous United States for 2002–2021. The datasets include four satellite-based products (IMERGᵥ7, CMORPH, PDIRNow, GSMaP), one radar-based product (Stage IV), two reanalysis-based products (ERA5, CONUS404), and one merged product (NLDAS). At the hourly timescale, average Kling-Gupta efficiency (KGE) scores range from 0. 125 to 0. 407, substantially lower than at the daily timescale (0. 412–0. 754). Stage IV and NLDAS show relatively higher KGE and classification accuracy, as well as better representation of diurnal cycles, although NLDAS markedly underestimates precipitation probability distributions, especially for extremes. Among satellite products, IMERGᵥ7 yields the highest hourly KGE and best captures diurnal cycles, while PDIRNow most closely matches observed probability distributions and GSMaP attains the highest KGE at the daily scale. Dynamic downscaling from ERA5 to CONUS404 improves precipitation probability distributions and performance under freezing conditions but reduces fidelity in capturing hour-to-hour variability. Across all datasets, missed and false precipitation events are the dominant error sources, with hourly-scale errors exceeding daily-scale errors (50% vs. 15%). Hourly precipitation trends also vary widely among datasets; however, seven products (all except CMORPH) exhibit significant increases consistent with gauge observations. Based on these results, we provide guidance for the selection, improvement, and development of hourly gridded precipitation products for different applications.
Wang et al. (Tue,) studied this question.