The solar radiation, aside from being responsible for meteorological conditions and ocean circulations, has also acquired a newfound importance as a renewable energy source. As such, a quantification of this resource is necessary using either observed or simulated data. Models have thus become increasingly important, as they generate irradiance data for places where measurements are lacking. Mesoscale numerical weather prediction models, such as the WRF, can be used to predict radiation data, but not always with enough spatial and temporal detail. An alternative consists of coupling mesoscale meteorological models with local radiation models. This study aims at testing WRF’s sensibility to different grid resolutions and parameterizations, with the goal of suggesting a standard configuration to feed a high resolution clear sky solar radiation model, through the use of solar radiation decomposition techniques. In order to optimize the WRF configuration, horizontal fields of global solar radiation data from measuring stations in the Corvo Island (in Azores) and Madeira were used. Several statistical measures were computed. The results showed that the solar radiation decomposition model with the best performance is the DIRINT model. From the tested schemes, the GWG and DEK schemes appear to be the most robust. Among these combinations, an overall best performance is not presented, as results vary according to the time period at study and the spatial domain. Nevertheless, for the Corvo Island, the DEK combination shows a good result for the month of October, for the D03 domain, but with worse results for the month of July. This study has also shown that the station located in the rural area has the best results. It was observed that there is no set of parameterization schemes that is the best for every situation at study.