What an aurora alert cannot do, including ours

An aurora alert is a notification that measured space-weather conditions have crossed a threshold. That is all it is, and the gap between that and “you will see the northern lights tonight” contains everything on this page.

We build one of these products, so this is not a neutral document. It is, however, an accurate one: every limitation below applies to Lumavik as much as to anything else, and several of them are the reason we have declined to ship features people ask for.

The four things no alert can know

Every aurora alert in existence is built on the same public instruments, described in where aurora forecast data comes from. Those instruments measure the Sun, the solar wind and Earth’s magnetic response. They do not measure any of the following.

Cloud

Cloud is the most common single reason an aurora night fails and it is invisible to every space-weather product. An alert can be entirely correct about the magnetosphere while you stand under complete overcast.

This cannot be fixed by improving the aurora forecast, because the missing information is not in the space-weather domain at all. It can only be addressed by combining two independent forecasts, which is a harder problem than it sounds — see below on why we have not shipped it.

Longitude

A planetary index summarises disturbance across the whole Earth. A storm concentrated over one range of longitudes produces a high planetary value regardless of where on the planet it happened.

An alert triggered on a planetary threshold will therefore fire for people whose own sky is quiet. Accounting for this properly requires modelling where the oval sits in magnetic local time, which is exactly what the OVATION model attempts and does approximately.

Timing

Aurora arrives in substorms: a growth phase of 30 to 60 minutes, an expansion of 10 to 30 minutes, then recovery. An alert fires on conditions measured at a moment; by the time you have dressed and driven somewhere dark, the expansion may be over.

This is a real constraint and it is why we consider “wait longer than feels reasonable” better advice than any notification setting. What a substorm is covers the cycle, including why a display that fades is often worth staying for.

Brightness

The instruments measure energy deposited into the atmosphere. Human vision in the dark uses rod cells, which are sensitive but nearly colourblind, and a great deal of real aurora sits below the threshold at which an unaided eye registers it as anything but grey sky.

A camera does not have this problem, which is why “the alert was right, the photo is green, and I saw nothing” is a coherent and common outcome rather than a contradiction.

The threshold trade-off, which cannot be solved

Every alert product must pick a threshold, and the choice is a trade-off with no correct answer.

Set it low and you catch marginal displays — but you also notify people on nights when nothing is visible, and the notifications stop being read. An alert that fires when nothing is there is the fastest way to make someone ignore the one that matters.

Set it high and every alert is trustworthy, but the product stays silent through displays that were worth seeing.

There is no setting that is right for everyone. Someone living inside the auroral zone who can look out of a window in ten seconds should want a low threshold: the cost of a false alarm is trivial. Someone facing a two-hour drive to dark sky should want a high one. This is why Lumavik’s paid tier exists at all — custom thresholds are the feature that lets a person position themselves on that curve, rather than a feature we invented to have something to charge for.

The free tier keeps the forecast, the Kp index, the three-day outlook, the auroral oval map and one alert location, and that is not going behind a paywall. The forecast itself is never paywalled.

What we deliberately have not shipped

Three features are requested regularly and are not in the app. In each case the reason is that a weak implementation would make the product less trustworthy rather than more.

Cloud cover overlay. The most requested feature, and the one we are most cautious about. Cloud forecasting is a separate discipline with its own accuracy problems, and bolting a mediocre version onto an honest aurora forecast would undermine the point of the app: a combined score is only as good as its worst input, and users would reasonably read a confident-looking overlay as more reliable than it is. Until we can do it at a standard that does not degrade the rest, the app says when cloud is the limiting factor instead of pretending to forecast it.

Long-range specific predictions. We do not offer a notification that says aurora is likely at your location on a specific date three weeks out, because the decisive variable — the magnetic orientation of the arriving solar wind — is not knowable until it is measured at L1, 20 to 60 minutes ahead. The 27-day outlook identifies dates worth watching, and we treat it as that. How accurate the aurora forecast is by lead time sets out the full picture.

A probability of seeing it. We score whether conditions are plausible at your location against your geomagnetic latitude and whether it is dark. We do not print a percentage chance that you will see aurora, because that number would have to include cloud and visual threshold, and we cannot compute either honestly.

What an alert genuinely is good for

The limitations above are real, and they do not make alerts useless. They make them useful for a narrower job than the marketing in this category implies.

An alert is good at telling you that now is unusually worth checking. That is a real service: geomagnetic activity is not uniformly distributed through the year, most people cannot watch the feeds continuously, and the difference between checking on the right night and checking at random is large.

It is good at catching the nights you would otherwise have slept through, which is the single biggest source of missed displays for people who live inside the auroral zone. This is the case where alerts earn their place most clearly, because the cost of stepping outside is nearly zero.

It is good at removing the need to interpret indices, provided the product does the interpretation honestly against your position rather than notifying on a global number.

It is not good at replacing a look at the sky, and no threshold setting changes that.

How to judge any aurora alert product, including this one

Four questions separate an honest implementation from a marketing one, and they apply to competitors and to us equally.

Does it account for your geomagnetic latitude, or only for planetary activity? A product that fires the same alert for Tromsø and for Chicago is not doing the calculation. Your own coordinate is available from the geomagnetic latitude calculator.

Does it account for darkness? An alert during daylight, or during Arctic midsummer twilight, means the darkness check is missing entirely.

Does it tell you what it does not know? A product that shows a confident number with no statement about cloud is presenting more certainty than its inputs support.

Does it distinguish measurement from forecast? A live Kp reading and a three-day outlook have completely different reliability, and a product that presents them with the same visual weight is misleading regardless of intent.

Our comparison pages apply these questions to specific competitors, including where they beat us.

Bottom line

An aurora alert is a notification that measured space-weather conditions crossed a threshold, and it cannot be more than that. Four things sit between a correct alert and a successful night, and none of them are engineering problems: cloud, which no space-weather instrument measures and which causes most failures; longitude, because a planetary index does not say where on Earth the disturbance occurred; timing, because substorm expansions last 10 to 30 minutes; and brightness, because emission energy is not the same thing as visibility to dark-adapted human vision.

The useful warning time is 20 to 60 minutes, set by the travel time of the solar wind from the L1 point where the decisive magnetic measurement is made. Anything promising specific hours or days of notice is extrapolating from a forecast rather than reporting a measurement.

The threshold trade-off between false alarms and missed displays cannot be eliminated, only positioned, and the right position differs between someone who can glance out of a window and someone facing a two-hour drive. Lumavik does not ship a cloud overlay, a long-range specific prediction or a percentage chance of seeing aurora, because in each case a confident-looking output would exceed what the inputs support. An alert is genuinely good at telling you that tonight is worth checking, and it is not a substitute for looking at the sky.