Capabilities and limitations of convection-permitting WRF simulations of lake-effect systems over the Great Salt Lake

Although previous studies suggest that the Weather Research and Forecasting (WRF) model can produce physically realistic banded Great Salt Lake-effect (GSLE) precipitation features, the accuracy and reliability of these simulations for forecasting applications remains unquantified. The ability of the WRF to simulate non-banded GSLE features is also unknown. In this study we use subjective, traditional, and object-based verification to evaluate convection-permitting (1.33-km grid spacing) WRF simulations of 11 banded and 8 non-banded GSLE events. In all simulations, the WRF was configured with the Thompson microphysics and the Yonsei University (YSU) planetary boundary layer parameterizations. Subjectively, a majority of the simulations of banded GSLE events produce physically realistic precipitation features. In contrast, simulations of non-banded GSLE events rarely produce physically realistic precipitation features and sometimes erroneously produce banded precipitation features. Simulations of banded GSLE events produce equitable threat scores (ETS) comparable to other convective-storm verification studies, whereas simulations of nonbanded events exhibit lower ETS. Object-based verification shows that the WRF tends to generate precipitation to the right (relative to the flow) and downstream of observed. These results, although based on a specific WRF parameterization suite, suggest that deterministic prediction of GSLE using convection-permitting models will prove challenging in practice with current numerical models. In addition, identifying and addressing the causes of the rightward and downstream precipitation bias is necessary to achieve optimal performance from future probabilistic and/or deterministic high-resolution forecast systems.

We also use the WRF model to examine the sensitivity of simulations of the Great Salt Lake–effect snowstorm of 27 October 2010 to the choice of microphysics parameterization (MP). It is found that the simulated precipitation from four MP schemes examined varies in areal coverage, amount, and position. The Thompson scheme (THOM) verifies best against radar-derived precipitation estimates and gauge observations. The Goddard, Morrison, and WRF double-moment 6-class microphysics schemes (WDM6) produce more precipitation than THOM, with WDM6 producing the largest overprediction relative to radar-derived precipitation estimates and gauge observations. Analyses of hydrometeor mass tendencies show that WDM6 creates more graupel, less snow, and more total precipitation than the other schemes. These results indicate that the rate of graupel and snow production can strongly influence the precipitation efficiency in simulations of lake-effect storms.