Meteosat Third Generation imagery improves CNN-based SSI retrieval
A study introduces a multi-imager, multi-resolution CNN for 10-minute Surface Solar Irradiance (SSI) retrieval over Estonia using MSG/SEVIRI and MTG/FCI satellite imagery. The hybrid SEVIRI-FCI model outperformed the SEVIRI-only model under overcast and cloudy conditions, reducing RMSE by 8.2 W/m² and 5.7 W/m² respectively.
发展脉络
- 首次出现Meteosat Third Generation imagery improves CNN-based SSI retrievalarXiv cs.LG
- 当前判断This work demonstrates a practical benefit of next-generation satellite data for renewable energy applications. It suggests that as MTG data becomes more available, operational SSI retrieval systems could see improved accuracy, especially in cloudy regions. This could influence the adoption of MTG data in solar energy forecasting services.Agent Pulse · 分析
This research evaluates the benefit of Meteosat Third Generation (MTG) satellite imagery for machine-learning-based Surface Solar Irradiance (SSI) retrieval. The authors propose a multi-imager, multi-resolution convolutional neural network that combines MSG/SEVIRI and MTG/FCI data with solar-geometry and clear-sky irradiance features. Using ground-based pyranometer measurements from eight Estonian stations, they show that the hybrid model significantly reduces RMSE under overcast and cloudy conditions compared to using SEVIRI alone, by 8.2 W/m² and 5.7 W/m² respectively. The model is also compared with the SARAH-3 physics-based product. This suggests that higher-resolution MTG data can improve SSI estimation, which is crucial for photovoltaic energy monitoring and forecasting.
The architecture is a multi-imager, multi-resolution CNN that fuses SEVIRI and FCI data. The performance gain is most pronounced in overcast and cloudy conditions, indicating that higher spatial resolution helps resolve cloud structure. The study uses site-based cross-validation and multiple training seeds, suggesting robust evaluation. A next signal would be whether the model generalizes to other regions and if the improvement persists with longer training data.
This work demonstrates a practical benefit of next-generation satellite data for renewable energy applications. It suggests that as MTG data becomes more available, operational SSI retrieval systems could see improved accuracy, especially in cloudy regions. This could influence the adoption of MTG data in solar energy forecasting services.
For solar energy companies, more accurate SSI retrieval can improve power forecasting, grid integration, and revenue optimization. The demonstrated RMSE reduction could translate into better operational efficiency and reduced imbalance costs.
Future work may extend the approach to other regions and weather conditions, and integrate the model into operational forecasting systems. The improvement under cloudy conditions is particularly relevant for Northern Europe, where solar irradiance variability is high.