Different reanalysis data and physical parameterization schemes for the Weather Research and Forecasting (WRF) model are considered in this paper to evaluate their performance in meteorological simulations in the Ili Region. A 72-hour experiment was performed with two domains at the resolution of 27 km with one-way nesting of 9 km. (1) Final Analysis (FNL) and Global Forecast System (GFS) reanalysis data (hereafter, WRF-FNL experiment and WRF-GFS experiment, respectively) were used in the WRF model. For the simulation of accumulated precipitation, both the WRF-FNL (mean bias of 0.79 mm) and WRF-GFS (mean bias of 0.31 mm) simulations can display the main features of the general temporal pattern and geographical distribution of the observed precipitation. For the simulation of the 2-m temperature, the simulation of the WRF-GFS experiment (mean warm bias of 1.81 °C and correlation coefficient of 0.83) was generally better than that of the WRF-FNL experiment (mean cold bias of 1.79 °C and correlation coefficient of 0.27). (2) Thirty-six physical combination schemes were proposed, each with a unique set of physical parameters. Member 33 (with the smallest mean-metric of 0.53) performed best for the precipitation simulation, and member 29 (with the smallest mean-metric of 0.64) performed best for the 2-m temperature simulation. However, member 29 and 33 cannot be distinguished from the other members according to their parameterizations. For this domain, ensemble members that contain the Mellor–Yamada–Janjic (MYJ) boundary layer (PBL) scheme and the Grell–Devenyi (GD) cumulus (CU) scheme are recommended for the precipitation simulation. The Geophysical Fluid Dynamics Laboratory (GFDL) radiation (RA) scheme and the MYJ PBL scheme are recommended for the 2-m temperature simulation.

To obtain high-precision precipitation simulation results, different types of rainfall events in the Ili Region are simulated by using the Weather Research and Forecasting (WRF) model with different physical parameterization schemes. According to the spatiotemporal distribution of rainfall evenness, six rainfall events in the Ili Region are divided into four types. Six microphysical parameterization (MP) schemes, five planetary boundary layer (PBL) schemes, and five cumulus (CU) schemes are combined into 14 parameterization members to simulate the rainfall events. It is worth noting that the simulation result sequence of the WRF model (from best to worst) is as follows: type I (events 3 and 5) > type II (events 1 and 6) > type III (event 2) > type IV (event 4). This finding would imply that the WRF model has the best performance for rainfall events with even spatiotemporal distributions, while it is hard to achieve good simulation results for rainfall events with highly uneven spatial and temporal distributions. The results suggest that no single combination of parameterization members provides the best performance for all rainfall events. According to the overall scheme rankings, d, n, and j are the optimal parameterization combination members that accurately describe the spatiotemporal characteristics of the six rainfall events. The study provides guidance for the selection of the physical parameters for the accurate simulation of different types of rainfall events in the arid region of northwestern China.