Aurora

ML backend for real-time aurora borealis visibility predictions.

Live API

Endpoint: https://aurora.arryan.xyz/predict

Request

curl -X POST https://aurora.arryan.xyz/predict \
  -H "Content-Type: application/json" \
  -d '{"latitude": 64.1, "longitude": -21.9}'

Response

{
  "probability": 0.335,
  "confidence": "high",
  "gb_probability": 0.346,
  "xgb_probability": 0.325,
  "conditions": {
    "is_dark": true,
    "cloud_cover": 89,
    "kp_index": 4.0,
    "geomagnetic_storm": false,
    "moon_interference": true
  }
}

Parameters

Field	Type	Description
latitude	float	Location latitude (-90 to 90)
longitude	float	Location longitude (-180 to 180)

Response Fields

Field	Description
probability	Aurora visibility probability (0-1)
confidence	Model agreement level (high, moderate, low)
gb_probability	Gradient Boosting model prediction
xgb_probability	XGBoost model prediction
conditions	Current environmental conditions

Model

Dual-model ensemble using Gradient Boosting and XGBoost trained on 5,293 samples.

• Test accuracy: 95.2%
• AUC: 0.99

Confidence scoring based on model agreement:

• High: models agree within 10%
• Moderate: models disagree 10-20%
• Low: models disagree more than 20%

Data Sources

• Weather: Apple WeatherKit (with Open-Meteo fallback)
• Space Weather: NOAA Space Weather Prediction Center
• Astronomy: Solar/lunar position calculations

Infrastructure

• FastAPI backend
• Kubernetes with Horizontal Pod Autoscaler (2-20 pods)
• Redis caching for NOAA data
• CI/CD via GitHub Actions and Ansible

Local Development

# Install dependencies
pip install -r requirements.txt

# Run API
uvicorn inference.app:app --host 0.0.0.0 --port 8080

# Retrain models
python training/train.py

License

Proprietary. Part of the Cosmofy iOS app.