Based on a 3 km-grid WRF (Weather Research and Forecasting) model and high-density surface Automatic Weather Station(AWS)observations in East China, the sensitivity of R_HZSCL of GSI-3DVAR for assimilation of temperature and wind of the surface AWS were studied. Results show that the value of R_HZSCL is rather sensitive for assimilating the surface automatic weather station observations by using GSI-3DVAR. With an appropriate R_HZSCL, an accurate surface temperature and wind analysis can be achieved. The temperature and Wind Vector Difference (VWD) RMSE can be reduced by more than 35％ compared to the background fields. However, if R_HZSCL is too large, an excessive expansion on the high and low temperature centers occur and the error is increased. For the surface wind analyses, the default R_HZSCL in GSI-3DVAR produced overly smooth wind and little mesoscale structures. Furthermore, a too small R_HZSCL would cause falsely strong local winds. By carrying out observation data-thinning experiments, it is found that the optimal R_HZSCL varies are opposite to different surface temperature and wind observation density.