Fog and Snow Detection to Improve Regional Photovoltaic Power Prediction

The work presented here aims at improving photovoltaic (PV) power forecasting in the presence of fog and snow. Under both conditions large errors in local as well as regional PV power predictions may occur. Predicting the formation and dissipation of fog is challenging for current numerical weather prediction (NWP) models. As fog fields can persistently cover larger regions, large uncertainties also in regionally aggregated PV power predictions can arise. Snow cover on PV systems can reduce their power output significantly and, by this, also lead to large errors in PV power predictions. To improve intraday PV power predictions in such situations we propose a two steps approach: First, fog and snow situations are identified based on satellite data and, secondly, PV power forecasts are adapted integrating this information. Meteorological satellite imagery in different spectral channels provides a suitable base to distinguish fog or low stratus (FLS) and snow from other clouds and clear sky conditions. Themodelling approach deploys machine learning methods for the classification of FLS and snow and is trained with labels derived from ground observations. An evaluation of the newly developed classification models shows advantages over existing methods. In particular, the classification of FLS is applicable seamlessly over day and night providing information on the presence of fog before sunrise. This is useful to adapt intraday PV power forecasts already in the early morning hours during winter. To exploit this new information for PV power forecasting, its integration into an existing PV power forecasting system is investigated. Thereto new models for the adaption of PV power forecasts in fog and snow situations were developed, integrating the satellite-based information in combination with fog dissipation climatologies for different seasons and weather situations. The new model reduces intraday forecast errors of the aggregated PV power in Germany during the winter half year 2023/2024 compared to the previously existing fog and snow correction.