Abstract: In order to more efficiently assimilate surface Automatic Weather Station (AWS) data, a new scheme based on the Ensemble Square Root Filter (EnSRF) is proposed for further improvement via solving the negative impact of assimilation results caused by elevation differences between observation sites and the model surface. Terrain Error of Representativeness (TER) for potential temperature and dewpoint temperature are added to temperature and dewpoint temperature errors of surface observation data assimilation in the WRF-EnSRF system, and a numerical simulation of a heavy rain event in Beijing-Tianjin-Hebei region in 2016 has been conducted. Results show that the root mean square error (RMSE), the threat score (TS) and various elements simulated in the first 13 h generally have been improved. With the TER being added, the RMSE of the wind field is improved in general, whereas the RMSE of potential temperature and dewpoint temperature are unstable in the earlier stage but they are improved in the later stage; TS of the first 24 h and the 24-48 h accumulated rainfall are overall improved compared with the results without TER. Thereby, the new scheme is able to reduce the negative impact of assimilation results caused by elevation differences.