Scoring & Confidence Metrics
Behind every probability percentage is a chain of calculations that weigh up the evidence and estimate how reliable it is. Here's how Lumina scores the current space weather and how it decides whether to be confident about the result.
The space weather score
Lumina calculates a 0–100 overall space weather score that feeds into the aurora label you see on the dashboard (Low, Moderate, High, Very High). The score is a weighted blend:
| Factor | Weight | Why |
|---|---|---|
| IMF Bz | 50% | Southward Bz is the single most important aurora driver — it's what opens the gate |
| Solar wind speed | 25% | Faster wind carries more kinetic energy into the system |
| Dynamic pressure | 15% | Higher pressure compresses the magnetosphere, intensifying coupling |
| Hemispheric power | 10% | HP confirms that energy is actually reaching the atmosphere |
Each input is soft-saturated before being weighted — extreme values help but with diminishing returns, so a single wild reading can't dominate the score.
Confidence — how reliable is this forecast?
Lumina provides two separate confidence numbers that multiply together:
Data confidence (0–1)
How good is the incoming data right now? This combines:
- Completeness — are all the expected data sources reporting, or are some missing?
- Cadence — is the data fresh? Stale data reduces confidence.
- Stability — is the IMF fluctuating wildly? High volatility makes any single reading less trustworthy.
Model confidence (0–1)
How reliable is the probability calculation given the current conditions? This combines:
- Duration score — has the current state persisted long enough to be meaningful? A 5-minute southward spike is less reliable than an hour of sustained southward Bz.
- Volatility dampening — if the IMF clock angle is flipping around, the model knows conditions are unstable and adjusts accordingly.
What confidence warnings actually mean
When you see a confidence warning on the dashboard, here's how to interpret it:
- Stable — data is fresh, conditions are steady. The probability is as reliable as it gets. Trust it.
- Volatile — data is current but the IMF is jumping around. The probability could swing quickly — check back frequently.
- Uncertain — data is stale or incomplete. The probability is a best guess but shouldn't be treated as gospel. Use the Events page and magnetometer signals as a cross-check.
Magnetotail energy state
The magnetotail energy state is a separate diagnostic that feeds into substorm detection and storm memory. It models how much energy has accumulated in the tail and whether the system is:
- Loading — energy is building. The tail is stretching. Substorm likely in the next 30–60 minutes if loading continues.
- Stable — the system is in equilibrium. No imminent substorm expected.
- Releasing — energy is being discharged. This is the substorm expansion phase — aurora is happening now.
You'll find the magnetotail state in the activity drivers accordion on the Aurora Probability card, and on the Space Weather Visualiser as the plasma sheet brightness.
Limitations to be aware of
Lumina's scoring is physically motivated — it's built on established space physics relationships, not machine learning — but it has limitations:
- A single L1 satellite can't capture all the spatial structure in the solar wind. A southward IMF patch might miss Earth entirely, or be bigger than expected.
- Substorms are inherently somewhat unpredictable — the exact moment of onset depends on tail stability thresholds that vary with conditions.
- Regional effects (like the South Atlantic Anomaly or local ground conductivity) aren't modelled — these can affect where aurora actually appears within the oval.