Weather forecasting is undergoing a revolutionary transformation with artificial intelligence, and our latest model, EPT-1.5, marks a significant leap forward for the European energy sector. Developed by our team at Jua.ai, EPT-1.5, the newest iteration of our Earth Physics Transformer (EPT) family, introduces new capabilities, particularly in predicting critical energy-related weather variables such as wind speeds and solar radiation.
Unprecedented Performance in Wind Forecasting
EPT-1.5 excels especially in wind forecasting at both 10m and 100m elevations, outperforming leading AI weather models like GraphCast, FuXi, and Pangu-Weather, and even surpassing the renowned numerical model, IFS HRES by ECMWF. These results place EPT-1.5 at the forefront of current AI-driven meteorological technology.
Revolutionary Temporal Resolution
One standout feature of EPT-1.5 is its unprecedented temporal resolution. Unlike existing models that provide forecasts at fixed intervals (typically every 6 hours), EPT-1.5 generates continuous forecasts, capable of minute-by-minute predictions for events days or even weeks into the future. This ability significantly improves planning and response strategies in sectors like renewable energy, where precise timing matters.
Advanced Probabilistic Forecasting
Another innovation includes its probabilistic forecasting capability, which predicts multiple potential future weather scenarios, enhancing reliability and risk assessment for long-term operational planning. Furthermore, EPT-1.5 is trained on a massive 5-petabyte dataset, allowing it to understand and model complex global weather interactions more accurately than previous models.
Operational Excellence
Operationally, EPT-1.5 provides high spatial resolution forecasts (9 km x 9 km at the equator) with hourly updates available up to 20 days ahead, running four times daily to ensure the latest information is always available. The model demonstrates exceptional performance in real-world testing, consistently outperforming ECMWF's operational model, IFS HRES, on key benchmarks like wind speed and solar radiation.
Looking Ahead
EPT-1.5 represents more than just an incremental update; it symbolizes the future of AI weather forecasting. Our ongoing work promises even more substantial advancements in the next major release, set to further enhance the model's capabilities and applications.
To read the full technical details and benchmarks, check out our complete paper available on arXiv.