Retrieve Monthly Averages of Ambient Air Temperature Daily Profile from Satellite-Derived Land Surface Temperature Data for Sizing Off-Grid PV Systems in Djibouti's Arid Climate

27th European Photovoltaic Solar Energy Conference and Exhibition; 4249-4253 As many other governments of sub-Saharan Africa, the Republic of Djibouti has focused its energy policy onto renewable power supply utilities. With a 30% electrification rate, a lot of Djiboutian people don’t access energy easily. Thereby, PV systems could be used for supplying the dispatched remote populations across the country. If sizing these systems requires estimating the amount of solar energy reaching the surface, the authors think that ambient air temperature is also a very important critical environment parameter. Indeed, this parameter acts fundamentally into photovoltaic conversion as well as into battery chemical reactions and, so, interacts with the 2 most important parts of an off-grid PV system by altering its efficiency and lifetime. This study presents the method we developed to take into consideration this parameter for sizing future off-grid PV systems in Djibouti. Ambient temperature was retrieved from 15 minutes Land Surface Temperature (LST) free accessible data, deriving from Meteosat 9 satellite images and developed by the Land Surface Analysis (LSA), a Satellite Application Facility (SAF) of the European Organization for the Exploitation of Meteorological Satellites (EUMETSAT). Different methods have been developed for retrieving air temperature data from LST. However, the number of available in situ weather stations and the need of night-time air temperature data incited us to develop and train a 3- layer feed-forward pattern associator neural network. The input variables were the LST pixel value, the related error value, the Solar Zenith Angle (SZA) and the year fraction. The output target was the ambient temperature retrieved by the different local temporary weather stations we have deployed in the country during the year 2010. Furthermore, in order to overcome the significant lack of LST data and the restricted number of available ground weather stations, we chose to estimate, in the first instance, monthly averages of ambient air temperature daily profile from nominal LST data, and smooth the possible extreme values in post processing. In order to assess performance of the complete process, we used statistical indices Root Mean Squared Error (RMSE) and Correlation Coefficient (CC). Final results show good robustness of the procedure, which could be generalized to other arid African developing countries.