Optical flow-based forecasting of surface irradiance using cloud motion vectors in Brazil

Abstract

The energy transition is driving the integration of renewable energy sources with an intrinsic relationship with meteorological and climatic conditions. In this context, it is essential to develop tools that support the growth of the share of intermittent renewable sources, such as solar energy, in the electrical grid. This research developed a short-term forecasting model to predict surface solar irradiance at four different locations in Brazil. The forecasting model uses the Farnebäck optical flow method to track cloud motion in GOES-16 satellite images. We compared the predicted results with actual data collected at SONDA network stations during different seasons of the year. The methodology showed the lowest RMSE values during winter in Cachoeira Paulista (SP), with 0.13 for a 30-minute forecast horizon and 0.11 for a 120-minute horizon. The highest RMSE deviation was observed during autumn, with 0.39 in 30 minutes and 0.41 in 120 minutes at the same site. These findings are consistent with research in tropical regions of South America, indicating the potential utility of this approach in the field.