AuroraData - Aurora Planning & Substorm Advisor

Phase 4/4

← Back to Projects

Roadmap

• Scope (not prioritized):

Concept

Aurora hunting in Australia requires a 60+ minute drive to dark sites, but most tools do not help answer the critical question: “Should I leave now?” AuroraData combines real-time solar wind monitoring, substorm trigger detection, local weather forecasts, and travel time to produce a single “Go/No-Go” score with site-specific recommendations.

Quick Facts

Roadmap

• Scope (not prioritized):

The Problem

Existing tools like the Glendale app are powerful but complex. Australian observers face unique challenges:

• Geographic disadvantage: Need strong events (Kp 6-7+) to see aurora at southern Australian latitudes
• Travel time investment: 60-90 minute drives to dark sites required
• Local weather gaps: Most tools use global models, not Australian-specific forecasts
• Decision paralysis: Multiple data sources provide conflicting signals

The question this tool answers: “If I leave now, will I see aurora when I arrive in 90 minutes?”

Architecture

┌──────────────────────────────────────────────────┐
│                 Data Sources                      │
│  NOAA SWPC (Bz, Bt, Speed, Density, HP)          │
│  Open-Meteo (ACCESS-G cloud cover, precipitation)│
│  Moon Phase Calculator                            │
└──────────────┬───────────────────────────────────┘
               │
      ┌────────▼────────────┐
      │  Substorm Analyzer  │
      │  Bz southward turns │
      │  HP step detection   │
      │  120-min trend buffer│
      └────────┬────────────┘
               │
      ┌────────▼────────────┐
      │  Site Scoring Engine│
      │  Activity (40%)     │
      │  Cloud cover (30%)  │
      │  Travel time (20%)  │
      │  Moon phase (10%)   │
      └────────┬────────────┘
               │
      ┌────────▼────────────┐
      │  Weather Reliability│
      │  MAE-based forecast │
      │  confidence scoring │
      └────────┬────────────┘
               │
      ┌────────▼────────────┐
      │   Output Layer      │
      │  CLI Dashboard      │
      │  LLM Advice (GPT)   │
      │  Telegram Bot        │
      │  Historical Playback │
      └─────────────────────┘

Phase 1: Substorm Trigger Engine

Real-time space weather monitoring with substorm detection:

• Solar Wind Buffer: 120-minute rolling window with deduplication by time tag, sorted order
• Substorm Triggers: Bz southward turns (> 5 nT shift) combined with HP step changes (30-minute rolling window)
• TOA Estimation: Solar wind arrival time calculated from speed: (1,500,000 / speed) / 60 minutes
• Opportunity Scoring: (abs(Bz) * 5) + (Speed / 10), capped at 100, blended with terrestrial potential at 60/40

Phase 2: Localized Planning & Site Selection

Multi-criteria site scoring with Australian-specific weather:

• Scoring Weights: Activity 40%, cloud cover 30%, travel time 20%, moon illumination 10%
• Cloud Forecast: Near-term weighting [0.4, 0.3, 0.2, 0.1] for hours 0-3 at arrival time, with fallback to 4-hour basic forecast beyond 12 hours
• Travel Penalty: Exponential decay (rate=0.02), 45-minute threshold for required activity increase
• Weather Reliability: MAE thresholds — < 10 = HIGH, 10-20 = MEDIUM, > 20 = LOW reliability
• Decision Output: >= 70 = GO, >= 40 = MAYBE, < 40 = NO-GO with contextual warnings

Phase 3: Intelligent Advice & Remote Access

LLM-generated recommendations and automated alerting:

• Pipeline Architecture: Analysis logic extracted into pure pipeline.ts function — no console output, no dotenv imports, fully reusable
• LLM Advice: GPT-4o-mini generates natural-language recommendations (max 300 tokens) with graceful degradation when API key is absent
• Telegram Bot: Automated alerts with 15-minute check interval, 1-hour cooldown between broadcasts, JSON file persistence for subscribers, auto-removal on send failure

Phase 4: Historical Playback Engine

Backtesting and validation against real storm events:

• Historical Data Loader: Loads NOAA mag/plasma data with time-tag join for robust matching
• HP History Support: Uses last two injected HP points for jump detection
• Playback Runner: Simulates pipeline decisions against historical events with parallel site scoring
• Analysis Report: 3-hour prediction window for realistic accuracy evaluation

Example Output

Location: Adelaide, SA
Target Site: Victor Harbor (75 min drive)

AuroraData Analysis:
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Space Weather:
  Bz: -12 nT (southward, favorable)
  HP: 45 GW → 68 GW (building substorm)
  Speed: 580 km/s (elevated)
  Arrival: ~43 minutes

Site Recommendations:
  1. Victor Harbor — Score: 82 (GO)
     Cloud: 15%, Travel: 75 min, Moon: 12%
     Confidence: HIGH
  2. Port Germein — Score: 71 (GO)
     Cloud: 22%, Travel: 45 min, Moon: 12%

LLM Advice:
  "Strong substorm building with favorable Bz.
   Leave within 30 minutes for Victor Harbor
   to arrive during peak activity window."

Telegram: Alert sent to 12 subscribers
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
RECOMMENDATION: GO NOW

Tech Stack

• Language: TypeScript (Node.js)
• Data Sources: NOAA SWPC (solar wind, hemispheric power), Open-Meteo API (ACCESS-G weather model)
• LLM: OpenAI GPT-4o-mini for advice generation
• Alerting: Telegram Bot API with subscriber management
• Testing: Jest with ts-jest (ESM support)

Integration with AuroraPhoto

AuroraData handles the planning side (when to go, where to go), while AuroraPhoto handles the capture side (automated camera control, HFR focus):

1. AuroraData: “Strong event predicted, leave now for Victor Harbor”
2. AuroraPhoto: Automated multi-camera capture once on-site

Roadmap

• Scope (not prioritized):

← Back to Projects