The spatiotemporal character of rainfall is particularly important for hydrologic modeling, as well as hydroclimatic risk estimation and impact assessment. Existing atmospheric reanalysis datasets offer extensive record lengths and global coverage, but usually their spatial resolution is coarse for distributed hydrologic simulations at small spatial scales. On the other hand, the temporal coverage of high-resolution radar-based rainfall estimates can be rather short for risk applications. To address these shortcomings, we simultaneously bias-correct and downscale a state-of-the-art atmospheric reanalysis (ERA5) rainfall dataset, using the radar-based Stage IV precipitation product as fine resolution reference, to develop an hourly CONUS-wide precipitation product over a 4-km grid, which extends back to 1979. In this regard, we refine an existing parametric quantile mapping framework based on a two-component theoretical distribution model, where we impose continuity of the parametric forms via optimal threshold selection to transition between higher and lower rain rates. An evaluation over the probability frequency and time domains, using NOAA’s raingauge measurements as benchmark, reveals that the developed product benefits from the strengths of the calibration datasets, demonstrating good performance and robust behavior over all studied time periods and Köppen climate classification zones, including snow-prone regions or areas where mesoscale convective systems become dominant. The accuracy of the yielded high spatial-resolution rain rates, especially in low probability events, shows that the developed product can be effectively used for hydroclimatic risk applications and frequency analysis, while its high temporal and spatial resolution makes it particularly useful for distributed hydrologic modeling.

We refined an existing parametric quantile-quantile (Q-Q) mapping framework based on a two-component theoretical distribution model, where we impose continuity of the parametric forms through a selection of an optimal threshold to transition between higher and lower rain rates. The parametric Q-Q mapping nonlinear transformation is obtained by matching the theoretical quantiles of the incorporated rainfall fields at different spatial resolutions within a certain calibration period, after fitting proper theoretical distribution models to the available data. Higher rain rates are parameterized using a Generalized Pareto (GP) distribution model, fitted above a sufficiently high threshold. For the case of lower rainfall intensities, we initially conducted tests to assess the suitability of the exponential, gamma, and lognormal parametric forms in fitting the empirical rain rates below the selected threshold, using the method of Maximum Likelihood (ML). The lognormal model demonstrated better performance in reproducing rainfall intensities over the whole study domain and was, therefore, adopted. To ensure the best possible distribution fit, the threshold to transition between higher and lower rainfall intensities was selected by minimizing the Anderson-Darling statistic. For more details on the employed techniques the user is referred to the accompanying publication; see Emmanouil et al. (2021). 

A detailed description of the dataset is included in the accompanying readme.txt file. Some important points:

1. Software used for the creation of the dataset: 

   • Data were created using Matlab 2019a (R2019a).

2. Methods for processing the data:

   • Any package and/or software that handles NetCDF files can be employed by the user to process the data.

   • For Matlab, the user is referred to the ncdisp() and ncread() functions.

3. Variable List:

   • precip

     • Size: (8760xN) or (8784xN), where N denotes the number of data points contained in each ERA5 grid cell (see also Grid_cell_IDs.txt). 

     • Units: Millimeters per hour (mm/h).

     • Description: Set of hourly precipitation estimates, corresponding to a selected year and grid cell ID, with a spatial resolution of 4x4 km. 
                                      

     • Missing hourly estimates are included as NaN.

   •    lat                                                                                         

     • Size: (1xN), where N denotes the number of data points contained in each ERA5 grid cell (see also Grid_cell_IDs.txt).      

     • Units: Degrees North (degN).

     • Description: latitude of the downscaled ERA5 precipitation product's estimation points (geographic coordinate specifying the north–south position).
           

   • lon                                                                                        

     • Size : (1xN), where N denotes the number of data points contained in each ERA5 grid cell (see also Grid_cell_IDs.txt). 

     • Units: Degrees East (degE).

     • Description: Longitude of the downscaled ERA5 precipitation product's estimation points (geographic coordinate specifying the east–west position).