Satellite-Based Solar Irradiance Forecasting: Replacing Cloud Motion Vectors by Deep Learning

Abstract

Satellite-based (SAT) methods are widely used to forecast surface solar irradiance up to several hours ahead. Herein, a cloud index-based version of the Heliosat method is applied to infer irradiance from Meteosat Second Generation images. The cloud index (CI) is derived from images in the visible range and quantifies the impact of clouds on surface solar irradiance. Conventional SAT methods utilize cloud motion vectors (CMVs) from consecutive CI images to predict future cloud conditions and subsequently retrieve irradiance. In this study, HelioNet is introduced—a convolutional neural network (CNN) with UNet architecture designed to predict future CI situations from sequences of preceding CI images. Forecasts of two HelioNet configurations are benchmarked against CMV and persistence over a full year (2023), with lead times (LT) up to 4 h. HelioNet_15 min recursively generates forecasts at 15 min resolution. HelioNet_hybrid begins with forecasts at 15 min resolution for $\text{LT}\le 45\text{\hspace{0.17em}min}$, then uses a 45 min resolved model to forecast all remaining LT steps. HelioNet_15 min achieves root mean square error (RMSE) improvements of >15% over the CMV model within the first hour on image level. HelioNet_hybrid shows superior performance for all LT across all metrics considered, with an average RMSE improvement of >11% on image and 8% at irradiance level.