Résumé
The current worldwide transition from conventional to renewable energy goes along with an expected increase of the installed photovoltaic (PV) capacity. Since renewable energies like PV are intermittent, solar forecasting is required in order to ensure grid stability and enables a massive injection of solar power into the electricity grids.The high availability of solar irradiance in the tropics allows high yields of PV power. Nevertheless, enhanced convection, homogenous air masses and strong thermal contrasts between land and sea make cloud evolution highly uncertain and cause a pronounced variability of irradiance in the tropics. Especially in non-interconnected tropical areas this leads to the fact that the injection of renewable energies currently has to be capped in order to ensure grid stability. From this, arises the necessity for accurate irradiance forecasts in these areas. Compared to global weather models which typically provide irradiance in intervals of three hours, regional NWP models allow irradiance modelling on temporal and spatial scales which are much higher and therefore more appropriate for solar power forecasting needs. The basis for accurate irradiance and cloud evolution forecasts is an accurate analysis of the atmospheric state for the initialisation of the regional NWP model. Such analyses are obtained by data assimilation methods which statistically combine observations and background information. Geostationary satellites provide reliable radiometric signatures of clouds and moreover high spatio-temporal resolution and coverage, which is especially beneficial in data-sparse regions like islands. The shortcoming is that a great number of observations have to be neglected in the course of satellite data assimilation since optical and thermal channels of satellite sensors do not provide cloud properties from inside clouds and since the single observations are highly correlated. This work aims at finding an assimilation strategy which makes optimal use of the available satellite observations in the assimilation process. We evaluate for the first time the potential of satellite data assimilation in regional NWP regarding short-term solar irradiance forecasts in the tropics. Therefore, numerical experiments using the models COSMO and WRF in conjunction with state-of-the-art data assimilation methods are performed. Special attention is drawn to the interplay of nesting and satellite data assimilation, satellite channel selection, bias correction and observation error treatment. The regions of study are two French tropical oversea territories: Reunion Island (21°S, 55°E), and French Guiana (5°N, 52°W).