Precipitation Properties in a Multi-Year Database of Convection-Allowing WRF Simulations

In recent years, the Weather Research and Forecasting (WRF) model has been used to forecast

convective precipitation with low enough grid spacing to avoid convective parameterization.

These models have gained popularity due to their ability to simulate precipitation that closely

resembles the observed structure of high impact phenomena such as topographically induced

precipitation, mesoscale convective systems, and supercell thunderstorms. High resolution WRF

models have also been recently utilized to downscale regional climate models to understand

the impacts on changes of precipitation in a changing climate. In both of these uses, an

accurate representation of precipitation events is important for climate mitigation and

adaptation strategies by policy makers. Do these simulations correctly represent climatological

precipitation? If not, then what benefit is there to using them in climate model downscaling? If

not, then why are they useful for forecasting the timing and location of high impact

precipitation events? This study examines the climatology of simulated precipitation over the

U.S. Central Plains by two WRF models; one operated by the National Severe Storms Laboratory

(NSSL) and the other by the National Center for Environmental Prediction (NCEP). Both WRF

simulations have a 4-km grid spacing and were initiated at 00-UTC every day during the period

of interest. Forecast hours 12-36 during the spring and summer months are used for analysis.

Hourly precipitation forecasts are analyzed to determine whether the WRF simulated

climatology is similar to observational climatology from the NCEP Stage-IV precipitation

database. Total precipitation in the spatial domain and the diurnal distribution of precipitation

are evaluated using Hovmöller diagrams. Simulated and observed precipitation “objects” were

also created at each forecast hour and tracked through time. The distribution of object

properties such as size, intensity, shape, longevity, start/stop time, and location were all used

to determine any model biases. Sensitivity testing was also done to ensure the chosen

parameters in the creation of precipitation “objects” did not create any biases in the

evaluation.