About Jua

Jua's EPT (Earth Physics Transformer) weather model learns historical weather patterns and fundamental atmospheric physics principles through its end-to-end transformer architecture. It is trained on public and proprietary observed data and physics-based simulations, such as ERA5. This dual approach allows it to capture complex atmospheric dynamics and physical laws while maintaining rapid processing speeds — unlike traditional NWP models, which solve physical equations sequentially by dividing the atmosphere into 3D grid cells and requiring precise initial conditions.

Further reading: EPT-2 Launch Announcement & EPT-2 Technical Report

Key differences vs. traditional numerical modelling

End-to-end AI models learn both historical weather patterns and atmospheric physics through advanced neural networks trained on observed data and physics-based simulations. Jua's models can predict weather events outside their training dataset by learning fundamental physics. Processing raw observational input directly into forecasts provides key advantages:

1. Accuracy - Generate accurate forecasts for hyper-local predictions

2. Resource efficiency - Less computational power needed during inference

3. Adaptation capability - Quick incorporation of new data patterns

4. Resolution flexibility - Seamless scaling across different resolutions

5. Uncertainty handling - Direct probability distribution outputs

Why does it matter for energy trading?

Our AI weather model's speed and accuracy directly impact energy trading decisions by:

1. Enabling faster reactions to weather shifts that affect renewable generation and demand

2. Adapting quickly to emerging weather patterns that influence market pricing

3. Delivering consistent accuracy across multiple forecast horizons critical for different trading products

Ready to get started?

Dive into our getting started guide now!