Back to Main Dashboard

🔮 Aurora ML Predictions

Generate AI-powered air pollution forecasts using Microsoft's Aurora model

✨ Enhanced Aurora Features

  • Dual Time Input: Uses both T-1 (00:00) and T (12:00) timestamps for better accuracy
  • Forward Predictions: Generate 1-4 steps forward, each covering 12 hours
  • Organized Storage: Results saved in dated folders for easy management
  • Multiple Variables: Predicts PM1, PM2.5, PM10, O₃, NO₂, CO, SO₂ and meteorological variables
  • Enhanced Visualization: Step-by-step analysis with time progression

⚠️ Performance Notes

CPU Mode: Aurora will run on CPU for local testing. This is slower but doesn't require GPU.

GPU Mode: If CUDA GPU is available, Aurora will use it for faster predictions.

Processing Time: CPU: 5-15 minutes per step | GPU: 1-3 minutes total

Memory: CPU mode automatically limits to 2 steps to prevent memory issues.

Coverage: Each step predicts 12 hours forward (max 48 hours with 4 steps).

Aurora will download CAMS data for this date and generate forecasts
Each step represents 12 hours forward from the initial conditions. Aurora uses T-1 (00:00) and T (12:00) as input, then predicts forward.
🔮

Aurora AI Processing

Generating atmospheric predictions using Microsoft's Aurora model...

Initializing Aurora pipeline...
1
Downloading CAMS atmospheric data
2
Loading Aurora ML model
3
Processing initial conditions
4
Running AI predictions
5
Saving results and preparing visualization

Estimated time: 2-5 minutes
This may take longer on CPU-only systems.

📊 What You'll Get:

  • Interactive visualization of predicted air pollution concentrations
  • Step-by-step forecast evolution over time
  • Downloadable NetCDF files with all prediction data
  • Support for all major pollutants and meteorological variables