Improving the accuracy of reanalysis-based hourly precipitation estimates over CONUS

Abstract:

The quality of hydrologic simulations is constrained by the accuracy of precipitation forcing data. In this context, hydroclimatic risk analysis and impact assessment requires multi-year precipitation datasets at adequately high spatial and temporal resolutions; e.g. on the order of 5 km in space and hourly in time. Recently, numerous precipitation datasets have been made available from atmospheric reanalysis, providing an extensive spatial coverage and record lengths that typically exceed 40 years. However, their spatial resolution (25 – 50 km) is rather coarse for physically-based distributed hydrologic simulations. Theoretically, the latter weakness could be overcome by employing high-resolution remote sensing-based rainfall estimates, but the length of these datasets is in the range of 15 to 18 years; i.e. a significant constraint for flood frequency estimation. In addition, the available blended precipitation products encompass biases and epistemic uncertainties that propagate to the conducted hydrologic simulations, significantly affecting their accuracy. This study investigates the limitations of various available precipitation datasets over CONUS (e.g., ERA5, ERA5-Land, Stage IV), by statistically assessing their accuracy in a time period of common coverage at hourly resolution, using raingauge measurements provided by NOAA as benchmark. The comparison is conducted by: a) using statistical metrics to quantify the deviations of the empirical probability distributions of rainfall derived from the analyzed datasets to those from raingauge measurements, and b) assessing the relative errors of the raingauge-recorded and estimated hourly time series, both in terms of systematic and random components. As a final step, we use state-of-the-art statistical tools to establish a parametric correction, and combine the strengths of reanalysis (i.e. long record length) and radar-based datasets (i.e., high-resolution), to create accurate precipitation estimates over CONUS at hourly resolution, beyond the time range covered by radars.