Artificial Intelligence weather forecasting company based in Switzerland claims superiority over Microsoft and Google's weather prediction systems
In a significant breakthrough for the field of AI weather forecasting, Swiss startup Jua has launched its AI weather model named EPT-2. The model has been reported to outperform Microsoft's Aurora and Google DeepMind's Graphcast, as well as the European Centre for Medium-Range Weather Forecasts' (ECMWF) ENS and IFS HRES forecasts.
EPT-2 is a native physics simulation that learns patterns from massive datasets. Unlike other AI-based forecasters, Jua took a step further by building a physics simulation model from scratch, integrated with AI. This approach allows EPT-2 to more efficiently and accurately learn how the atmosphere behaves without needing massive supercomputers.
The model ran forecasts 25% faster than other models tested and used 75% less computing power. In key variables like 10-metre wind speed and 2-metre air temperature, EPT-2 showed lower root mean square error (RMSE) than Aurora beyond 6 days. The accuracy advantage on temperature was maintained throughout the forecast range, and EPT-2 consistently outperformed the ECMWF's IFS HRES deterministic forecasts in both variables.
The research results, which will be published on the open-access archive arXiv next week, show that EPT-2 captures important weather features that other models miss, such as temperature peaks and brief wind fluctuations critical for energy applications.
Jua's CEO and co-founder, Marvin Gabler, is confident that EPT-2 can beat all competition, despite DeepMind's Graphcast not being included in the study. The new AI weather model is potentially thousands of times faster on less energy-intensive machines than traditional weather models, which use complex physics equations run on billion-dollar supercomputers.
Jua has raised a total of $27mn in funding from backers including 468 Capital, Future Energy Ventures, and Promus Ventures. As AI-based weather forecasting continues to gain traction due to demand for more accurate and cheaper ways to predict the Earth's climate, Jua's EPT-2 model is poised to lead the way in this exciting field.
Artificial-intelligence technology plays a crucial role in Jua's AI weather model, EPT-2, as it learns patterns from vast datasets. In comparison to other AI-based forecasters, EPT-2's unique approach involves integrating physics simulation with AI, enabling it to learn the atmosphere's behavior more accurately while requiring less computing power.
