Dr. Lux, Head IT-Solutions & Consulting and LuxActive and Key Researcher at SWISDATA, presented at the Vienna Data Science Group Meetup, as part of the presentation of AI Factory Austria AI:AT, on October 23, 2025, CloudShadingAI. It was developed in collaboration with LuxActive and MetGIS. This already in September 2025 accomplished project, funded by the Austrian Research Promotion Agency (FFG), aims to enhance energy grid stability for solar power by predicting shadow patterns caused by clouds and terrain over a 15-minute to 6-hour timeframe. The project combines ultra-high-resolution terrain data, Earth observation data from ESA (MSG-3), and weather data processed by MetGIS to deliver precise forecasts. Two core services were developed:

• A terrain shading service using ASTER GDEM V3 terrain data sunlight for pixel-precise shadow calculations

• A cloud shading service leveraging AI models to analyze cloud movement beyond traditional visual methods

The development of CloudShadingAI benefited greatly from high performance computing resources. Training of several small Artificial Intelligence models was performed simultaneously on a single graphics‑processing‑unit. Multiple processes ran in parallel, each training one model.

Access to the high performance computing cluster follows a clear step‑by‑step procedure. Interested parties consult the AI Factories access calls listed on the eurohpc-ju.europa.eu website, select a specific call, and prepare a brief proposal. After a short review period, a positive confirmation is issued, followed by user registration, access credentials and login information, and finally the ability to use the resources with extensive support.

Feedback from our development team highlighted strong support from the EuroCC/ASC, fast and easy access, and smooth job submission using the Slurm workload manager. The code ran on the cluster without modification, and data upload was straightforward using rsync. Some limitations were noted, such as the maximum execution time for a single job and slower CPUs for data pre‑processing on the Leonardo HPC. Documentation was described as difficult to navigate in places.

The service combines essential data sources: ultra‑high‑resolution terrain information from the ASTER Global Digital Elevation Model version 3, Earth‑observation imagery supplied by the European Space Agency’s Meteosat‑3 satellite, and weather data prepared by MetGIS. By using the terrain model, the system calculates the exact location of shadows cast by the sun on a pixel‑by‑pixel basis.

In addition, a cloud‑shading service employs Artificial Intelligence models. Many small feed‑forward neural networks are employed, each assigned to a geographic group based on location and topography.  The models output movement vectors for each pixel, providing a vector‑based forecast of cloud motion instead of generating new images. These models go beyond visualizing cloud movement. They integrate weather forecasts with motion‑vector data to predict how clouds will shade, e.g., solar panels. 

The cloud shading service for an 6-hour prediction executes now our own hardware in 5–15 minutes and achieves an average shadow accuracy of 93% based on observations from 388 locations in Austria.

CloudShadingAI gained recognition at the Vienna Data Science Meetup, where 120 attendees witnessed its potential to revolutionize solar energy forecasting. The project’s innovative use of many small feed forward neural networks to analyze cloud images and predict movement vectors was highlighted, alongside its practical applications for energy grid management but also beyond e.g., showing a tourism use case, to plan hikes. Attendees were also introduced to the CloudShadingAI API, available for integration into real-world use cases, and the opportunity to request demos via the platform shadow.swisdata.eu provided by the non profit research organization SWISDATA.

CloudShadingAI, developed by LuxActive and MetGIS and now hosted by SWISDATA, represents a significant advancement in solar energy forecasting and beyond. By combining terrain and cloud data with AI-driven models, the project achieves high accuracy in predicting shadow patterns and for example in supporting the energy grid stability. The project’s success was showcased at the Vienna Data Science Meetup, where its technical and practical innovations were presented. For further details please contact us and for updates, visit luxactive.com/de/blog and swisdata.eu/news. Explore the CloudShadingAI demo at shadow.swisdata.eu.

Previous article about CloudShadingAI