Evaluation of rainstorm characteristics in a regional climate model

It is well understood that choices of parametrization and model settings in global and regional climate simulations may affect mean precipitation. In this work we evaluate their effect on the characteristics of individual precipitating events, using a novel rainstorm identification and tracking algorithm (Chang et al 2016). As a test case we use high-resolution dynamically downscaled simulations (most at 12 km) over a relatively large domain (much of North America), with WRF as the regional climate model (RCM) and boundary conditions from NCEP-R2 reanalysis for summer 2005. For reference we compare to the NCEP stage IV observational precipitation data product, which is based on combined radar and gauge data. In a series of experiments we vary the convective parameterization, microphysical scheme, spectral nudging approach, spin-up time, and resolution, with the highest spatial resolution of 4 km being convection-permitting. The identification and tracking algorithm identifies individual rainstorms, including low-intensity precipitation events with complicated features and evolution patterns. The method allows us to decompose changes in total mean precipitation into their causes: changes in precipitation event size, intensity, number, and duration. (Previous identification and tracking methods focus on only very large and intense events.) The "control" 12 km resolution simulation shows substantial biases in total precipitation (> 50% bias) resulting primarily from storms that are too large in spatial extent. These biases are relatively insensitive, with most experiments showing no significant improvement in simulated rainstorm size. The higher-resolution simulations at 4 km (with cumulus parameterization turned off) do however significantly reduce model bias both in rainstorm size. These results suggest that higher resolution and resolving rather than parametrizing convection may be the most likely path for increasing the fidelity of model precipitation fields.