Ground surface energy fluxes and the characteristic of soil temperature are simulated by a mesoscale atmospheric model, RAMS (the regional atmospheric model system), with up to 40 different sets of parameters, and the applicability of RAMS to arid and semi-arid regions is explored. The simulation lasted 30 days to prove its accuracy and stability. Both sunny and rainy synoptic processes are included in the simulation period. The result shows that RAMS is impossible to give sound outcome with its default parameters even using valid input data when applied in the Heihe region. Only when the model runners adjust manually soil parameters, can RAMS gain results that are consonant with observational data. Soil moisture content, soil thickness, and soil temperature are the most important in all the initial soil parameters.