Performance of the WRF model at the convection-permitting scale in simulating snowfall and lake-effect snow over the Tibetan Plateau

This study investigated the performance of the Weather Research Forecasting (WRF) model at 4-km horizontal grid spacing in simulating precipitation, 2 m air temperature (T2), snowfall, and lake-effect snow (October 4-8, 2018) over the Tibetan Plateau (TP). Multiple simulations with different physical parameterization schemes (PPSs), including two planetary boundary layer schemes (Yonsei University and Mellor-Yamada-Janjic), no cumulus and multi-scale Kain-Fritsch, two land surface models (Noah and Noah-MP), and two microphysics schemes (Thompson and Milbrandt), were conducted and compared. Compared with gauge observations, all PPSs simulate mean daily precipitation with mean relative errors (MREs) of 27.7%-53.6%. Besides, spatial correlation coefficients (SCCs) between simulated and observed mean daily precipitation range from 0.56 to 0.71. For simulations of T2, all PPSs perform similarly well, even though the mean cold biases are up to about 3 degrees C. Meanwhile, all PPSs exhibit acceptable performance in simulating spatial distributions of snow depth, snow cover, and snowfall amount, with SCCs of 0.37-0.65 between simulations and observations. However, the WRF simulations significantly overestimate snow depth (similar to 0.4 cm mean error) and snowfall amount (MREs >372%). The Milbrandt scheme slightly outperforms the other PPSs in simulating snow-related variable magnitudes. Due to their inaccurate temperature and airflow modeling over the lake surface and its surroundings, none of the WRF simulations well reproduce the characteristics that more snow occurs over the lake and downwind area. Overall, this study provides a useful reference for future convection-permitting climate modeling of snow or other extreme events when using the WRF model in the TP and other alpine regions.