On hand available predictors for operational satellite-based forecast

Photovoltaic production is highly variable with a limited predictability due to stochastic variation of cloud attenuation of surface solar irradiance. Forecasting this irradiance helps to manage the safe and stable operation of the grid and thus enables to increase solar electricity penetration for a lower carbonated energy mix. In particular, intraday horizon forecasts (typically up to 6 h) are more and more required for the grid power reserve management, the increase of intra-day auctions in the electricity market and the current worldwide development of micro-grids.

Forecast irradiance using images from geostationary meteorological satellite is particularly appropriate for intraday horizon. It gives better performance than NWP models, does not require instrumentation, or costly computing resource. However, the accuracy of such methods is very sensitive to the cloud cover state and its short-term evolution. For instance, methods based on the temporal extrapolation of cloud motion present better results for passing cloud events than for sudden cloud appearance or disappearance. In several studies, reliability predictors have been identified for satellite-based irradiance forecast. They showed clear signals when uncertainty is computed as a function of season, solar zenith angle, cloud albedo and more recently synoptic weather regimes. This anticipation of error range would help grid managers to prepare the sizing of storage capacity or ancillary electricity resource.

A pertinent predictor must have a significant influence on forecast error. It also must be easy to obtain in operational forecast conditions. In this work, we propose to exploit the only observation source required in satellite-based forecast: the satellite image itself. Using 5 years (2017-2021) of image-derived forecasts at 15 min time step over Palaiseau (France), we computed the forecast uncertainties as a function of multiple parameters derived from the HRV channel of Meteosat Second Generation satellite. We highlighted the influence of cloud albedo spatial variance and cloud motion vector field spatial-temporal features in the forecast uncertainties. In addition to the error range provided to users, this work can help forecasters to better characterize their sources of error and to select new predictors for machine-learning approaches.