Aurora Advisor
Contents
Concept
Decision tool for Australian aurora observers that answers “should I drive 60 minutes to a dark site tonight?” Combines real-time solar wind data (NOAA SWPC), substorm trigger detection (Bz drops + hemispheric power jumps), and local weather forecasts (ACCESS-G model via Open-Meteo) into a single Go/No-Go score that accounts for both space weather potential and terrestrial conditions (cloud cover, moon phase, travel time).
Features
Shipped (v1.0):
- Real-time substorm trigger detection from NOAA solar wind data
- Multi-criteria site scoring (activity, weather, travel time, moon)
- Telegram bot with automated aurora alerts
- Historical playback engine with parameter tuning infrastructure
- Hybrid Rust/TypeScript architecture — Rust CLI for fast offline analysis (10k configs in <1s)
In progress (v2.0):
- Longer lead-time forecasting (6–12 hours using L1 solar wind from ACE/DSCOVR)
- ML-based probabilistic predictions replacing binary heuristics
- Event timeline forecasting (intensity curve, peak timing, optimal viewing windows)
Quick Facts
| Status | Active |
| Stack | TypeScript |
What This Is
A specialized tool for Australian aurora observers that solves the “should I drive 60 minutes?” problem. It combines real-time solar wind data (NOAA), substorm trigger logic (Bz/HP trends), and local weather (ACCESS-G model) to provide actionable advice.
Why We’re Building It
Existing tools like the Glendale app are powerful but difficult to use, and many sources lack the localized Australian context and travel-time “trade-off” logic required for confident planning. This tool aims to maximize observation success by providing a clear, advice-driven indicator of when to leave for a site.
Core Value
Providing a single, definitive “Go/No-Go” score that accounts for both space weather potential and local terrestrial conditions (travel time, clouds, moon).
Current Milestone: v2.0 - Advanced Forecasting System
Goal: Transform from binary heuristic predictions to ML-based probabilistic forecasting with longer lead time and event timeline predictions.
Target features:
- Longer lead time forecasting (6-12 hours using L1 solar wind data from ACE/DSCOVR)
- ML models for improved accuracy (ensemble methods, feature engineering, trained on historical events)
- Probabilistic predictions (confidence intervals, risk scores instead of binary Go/No-Go)
- Event timeline forecasting (predict intensity curve, peak timing, optimal viewing windows)
Key Decisions
- Architecture: Node.js/TypeScript prototype (currently CLI).
- Primary Data: NOAA SWPC JSON products (Solar Wind) and Text products (Hemispheric Power).
- Weather Model: ACCESS-G via Open-Meteo for Australian accuracy.
- Future Expansion: Design hooks for “Astro Predictor” (seeing, transparency) and Telegram alerts.
Requirements
# Validated (v1.0 shipped)
- ✓ Real-time substorm trigger detection (Bz drops + HP jumps) — Phase 1
- ✓ Multi-criteria site scoring (activity, weather, travel time, moon) — Phase 2
- ✓ Telegram bot with automated alerts — Phase 3
- ✓ Historical playback engine with tuning infrastructure — Phase 4
- ✓ Hybrid Rust/TypeScript architecture —
- ✓ Fast offline analysis (Rust CLI, 10k configs in <1s) — Phase 6
# Active (v2.0 in progress)
- Integrate L1 solar wind data for 6-12 hour lead time
- Build ML training pipeline with feature engineering
- Implement ensemble models for probabilistic predictions
- Add event timeline forecasting (intensity curve, peak timing)
# Out of Scope
- [Exclusion 1] — Historical Substorm Analysis: Detailed correlation study of past solar events to refine trigger math (moved to future milestone).
- [Exclusion 2] — Astro Predictor (Seeing/Transparency): Captured as a future milestone/separate project to maintain focus on Aurora.
- [Exclusion 2] — Mobile Native App: Initial focus is CLI/Web-based alerts.
Last updated: 2026-02-21 after completing v1.0 and starting v2.0 milestone
Roadmap
- Scope (not prioritized):