Many small island states are planning to invest heavily in solar photovoltaics in an attempt to curb their overreliance on fossil fuels for electricity generation. In order to efficiently exploit the abundant solar energy resource, these islands need reliable solar irradiance data. However, the orographic effects arising from their volcanic origins often result in strong variability and uncertainty in the solar resource. In this context, satellite-based models present an effective alternative to ground-based measurements. Different downscaling approaches have been applied that compensate for the large spatial resolution of satellite images and the terrain-related effects that they disregard. Nevertheless, the accuracy of these methods is influenced by the solar radiation decomposition model used. Moreover, the variogram model used in the kriging process to characterize the spatial dependence of the solar radiation has a significant effect on the results. In this study, we compare the performances of seven radiation decomposition models for the anisotropy analysis and seven variogram models for the spatial interpolation of the solar irradiance. A dense network of ground measurements at 43 stations is used to evaluate the accuracy of the different models. Results reveal that the Yao radiation model coupled with the Matern variogram provide the best results.