Forecasting global radiation directly from Meteosat SEVIRI spectral observations

Renewable energy forms a major pillar of climate change mitigation. Yet, the inherent volatility of solar irradiance as a primary renewable energy source poses significant challenges. The stochastic dynamics of clouds induce rapid fluctuations in photovoltaic energy production by modulating incoming solar irradiance, which is a challenge for grid operators. More accurate solar irradiance forecasts encompassing the grid area and associated photovoltaic plants will help address this challenge.We will provide an overview of regional forecasting methods for surface solar irradiance (SSI) using satellite observations for forecast lead times of minutes to hours. Existing solar forecast models are based on Level-2 products that comprise satellite-derived estimates of SSI or clear-sky index (e.g., Carpentieri et al., 2023) such as the third edition of the Surface Solar Radiation Data Set - Heliosat (SARAH-3; Pfeifroth et al., 2021).We present a probabilistic regional solar forecast model for lead times of up to several hours that has been trained directly on visible and infrared satellite radiances. We utilise the multispectral Level-1.5 images of the Spinning Enhanced Visible Infra-Red Imager (SEVIRI) imager aboard the Meteosat Second Generation geostationary satellite to this end. Our solar forecast model is based on the architecture of the generative SHADECast model (Carpentieri et al., 2024). Leveraging multiple years of SEVIRI scans, we demonstrate accurate probabilistic forecast of spatially extended surface solar irradiance fields and forecast model runtimes reduced to the order of seconds to minutes. We chacterise the model performance including benchmarks with existing solar forecast models and discuss the Meteosat channel importance for solar forecasting.Carpentieri, A., S. Pulkkinen, D. Nerini, D. Folini, M. Wild, A. Meyer, Intraday probabilistic forecasts of surface solar radiation with cloud scale-dependent autoregressive advection, Applied Energy, 351, 2023. doi:10.1016/j.apenergy.2023.121775Carpentieri, A., D. Folini, J. Leinonen, A. Meyer, 2024, Extending intraday solar forecast horizons with deep generative models, arXiv:2312.11966, doi:10.48550/arXiv.2312.11966Pfeifroth, U., J. Drcke, J. Trentmann, R. Hollmann, 2021, SARAH-3 - a new satellite-based Cimate Data Record for surface radiation parameters from the CM SAF, EMS Annual Meeting 2021, online, 610 Sep 2021, EMS2021-454, doi:10.5194/ems2021-454