Handling the difference in elevation between a model surface and an observation site is always a challenge in surface data assimilation. However, a reasonable assimilation scheme can efficiently assimilate surface automatic weather station (AWS) data into a mesoscale model. In this paper, surface AWS data are first assimilated into a weather research and forecasting (WRF) model through an ensemble Kalman filter using the Guo et al. (2002) scheme. Then an adjusted scheme is proposed that assimilates 10-m wind observations, 2-m potential temperature, 2-m dew point temperature, and surface pressure. This scheme is then validated by mean square root error analysis, simulated result and assimilation increment analysis, and sensitive experiments to check the assimilation response of each AWS meteorological parameter. Results show that the assimilation of surface AWS data through the ensemble square root filter (EnSRF) using the Guo et al. (2002) scheme can improve the simulation results. The separate assimilation of any element of the surface observation data (temperature, humidity, wind, surface pressure) can affect the forecast of 18 h accumulated rainfall. However, different elements have different impacts, and the one having most influence is the dew point temperature. The use of 2-m potential temperature and 2-m dew point temperature, instead of 2-m temperature and 2-m specific humidity, leads to better simulation results.