Comprehensive surface observations and soil-vegetation data collected at the Oklahoma ARM central facilities are utilized to test the soil-vegetation model installed in the Navy's Coupled Ocean/Atmospheric Mesoscale Prediction System (COAMPS). Although the early installed version of soil-vegetation model (based on Noilhan 1989) was recently upgraded by introducing a skin temperature layer similar to that in the ECMWF model, the upgraded model does not solve all the previously identified problems. In particular, the upgraded model still overpredicts evaporation in the wet period (June 25-30, 1999) and underpredicts evaporation in the dry period (August 14-19, 1999) at the Oklahoma ARM central facility site as indicated by comparisons of the model's predictions of surface heat fluxes, soil temperature and soil water content against the ARM measurements. The two possible sources of the problems are identified: (i) The overprediction of evaporation for the wet period was caused by the neglect of soil water infiltration into the deep layer in the model (after rain or dew formation); (ii) The underprediction of evaporation for the dry period was caused by the neglect of evaporation within the soil pore spaces in the model. To solve these problems, parameterizations are introduced to represent these processes in the model. Tested with the ARM measurements, the parameterizations are found to be effective in correcting the problems.