Hybrid model from cloud motion vector and spatio-temporal autoregressive technics for hourly satellite-derived irradiance in a complex meteorological context

Abstract

Islands in tropical regions have high potential for solar energy, but the weather conditions in these areas are complex, with high fluctuations in the amount of sunlight received over time and across different locations, making it difficult to predict solar irradiance accurately.In a preliminary study, two spatio-temporal technics STVAR (spatio-temporal autoregressive) and CMV (cloud motion vector) showing a good predictive performance in literature, were assessed in this challenging environment. The strengths and the weaknesses of different models for different conditions/locations were presented. In this paper, we focus on the validation STVAR/CMV blends for the same satellite-derived irradiance dataset. In a first step, the research of the equation defining the blended model is investigated, highlighting a linear combination of irradiance predicted from CMV and STVAR by least-squares fit, as being optimal. A benchmarking illustration as a function of the orographic context exhibits the reduction of their respective gaps forced by their separate application. Then, the analysis of spatial evolution of the linear combination coefficients, led us to propose a model that quantifies coefficients of the blended model as a function of site elevation that represents an effective proxy for the microclimatological/topographical nature of the considered location. The proposed model shows good performance with an averaged relative RMSE of 16.50% in the entire study area. This model can be an appropriate choice for short-term forecasting even under complex orography conditions.

Keywords

• satellite-derived irradiance
• complex meteorological context
• cloud motion vector
• spatio-temporal forecasting
• blended model