Many Photovoltaic (PV) systems are connected to electric power grids, and the grids get risk of instability due to the fluctuations of PV output. Numerical Weather Prediction (NWP) models are used for forecasting the solar irradiance and proper grid management. NWPs usually have many physical parameterization options, and appropriate schemes of the options should be selected for accurate forecasting. The options should be changed by regional and climatic conditions and other factors. The target country is Thailand which is in the tropics. At there, cumulus and cumulonimbus appear frequently, and their behavior makes weather forecasting difficult. The optimal combination of schemes in the tropics is found through the sensitivity analysis of the options. By the optimization, the forecasting accuracy increases from 0.773 to 0.814 in correlation coefficient with the observation. It’s also found that contributions of surface layer and planetary boundary layer processes are significant for improvement of accuracy.