Ask any two aurora apps for tonight’s odds and you’ll sometimes get different answers — not because one is wrong, but because “aurora forecast” actually covers several distinct layers of prediction with very different levels of confidence and lead time. Understanding those layers is the difference between blindly trusting a single number and reading a forecast the way people who chase the aurora seriously actually do.
The three layers of an aurora forecast
Multi-day outlook. Based on watching the Sun for coronal mass ejections (see what is a CME) and tracking coronal holes as they rotate into an Earth-facing position (see coronal holes and recurrent storms), NOAA issues Kp and G-scale outlooks reaching one to three days ahead, viewable on this site’s live Kp index page alongside the current reading. This layer answers “is something worth watching for this week,” but carries real uncertainty about exact timing and strength.
Real-time solar wind data. Once a CME or fast stream reaches the L1 monitoring point roughly 1.5 million kilometres upstream of Earth, its speed, density, and — most importantly — its magnetic field orientation (Bz) can finally be measured directly (see reading a solar wind plot, or this site’s live solar wind page). This is where forecast confidence jumps sharply, but the lead time shrinks to roughly 30 to 90 minutes, simply because that’s how long it takes the solar wind to travel the remaining distance to Earth.
Observed geomagnetic response. Once the solar wind actually reaches Earth’s magnetosphere, ground-based magnetometers and satellites start recording the response directly — the observed Kp index, the OVATION model’s live aurora probability map, and eventually the visual reports of people actually looking at the sky. NOAA’s formal watches, warnings and alerts (see the NOAA G-scale explained, or this site’s live alerts page) are issued at various points across all three layers, which is why the same event can generate a multi-day watch, a real-time warning, and then a confirming alert in sequence.
A worked example: how a forecast evolves over one evening
Concretely, imagine a Tuesday. NOAA’s multi-day outlook, issued that morning, forecasts a possible G1-G2 period late Wednesday into early Thursday, based on a coronal hole stream expected to arrive — this is enough information to mention the possibility on a city page and to suggest keeping Wednesday night free, but not enough to promise anything. By Wednesday afternoon, the leading edge of that stream reaches the L1 spacecraft: speed has risen to 550 km/s from a 350 km/s background, and Bz has turned to -8 nT and is holding there. This is now a real-time signal with 30-90 minutes of lead time, and a flagship destination like Fairbanks or Reykjavík would likely see its live verdict jump toward “good” within that window. By Wednesday evening, ground magnetometers confirm the estimated Kp has risen to around 5, and OVATION’s grid shows elevated probability specifically over northern Scandinavia and Alaska — at this point the forecast has moved from “estimate” to “observed,” and the verdict reflects current, not predicted, conditions.
Why the 30-90 minute window is a hard physical limit, not a technology gap
It’s tempting to assume better satellites or smarter models would extend the forecast window further out. They help at the margins, but the core limitation is physical: the most reliable, direct measurement of the solar wind that is about to hit Earth can only be taken from a spacecraft positioned between the Sun and Earth, and that spacecraft is only so far away. Move it further from Earth for more warning time, and it starts sampling solar wind that may evolve — speed up, slow down, or change magnetic orientation — before it actually reaches Earth, making the measurement less representative of what actually arrives. In practice, the L1 point represents close to the best available trade-off, and it’s why every serious space weather agency, not just this site, treats 30-90 minutes as the honest ceiling on high-confidence aurora forecasting.
What NOAA’s OVATION model adds
The Kp index is a single global number, which makes it a blunt tool for a phenomenon — the auroral oval — that is genuinely lumpy, shifts with solar wind conditions, and can be considerably more active on one side of the planet than the other at any given moment. NOAA’s OVATION Prime model addresses this by estimating the energy flux of precipitating particles across a geographic grid, roughly one degree of latitude and longitude in resolution, refreshed roughly every five minutes. It’s built as a statistical/empirical model — calibrated from years of particle-detector measurements taken by low-Earth-orbit satellites passing through the auroral zone, correlated against the solar wind conditions present at the time of each pass — rather than as a first-principles physics simulation solved from scratch on each update, which is part of why it can run fast enough to refresh every few minutes rather than requiring hours of computation. When available, this site’s live verdict for each city cross-references OVATION’s local probability alongside the Kp-derived threshold, which is why two cities at a similar geomagnetic latitude can sometimes show meaningfully different verdicts on the same night — the oval genuinely isn’t a perfect, evenly expanding circle, and OVATION is what captures that.
Why darkness gates everything
No amount of geomagnetic activity produces a visible aurora if the sky itself is too bright to see it against — either because the sun hasn’t set far enough below the horizon, or because of the midnight sun at very high latitudes in summer. This site’s verdict logic gates on solar elevation dropping below about -6° (civil twilight) before considering the aurora “potentially visible” at all, which is also why the same Kp and OVATION numbers produce a “daylight” or “midnight sun” verdict rather than a probability during the wrong part of the day or year — see the best time of night to see the aurora for more on this, and why a city like Dunedin in the Southern Hemisphere runs on an entirely different seasonal calendar to Tromsø or Yellowknife in the north.
Why different apps disagree
If you compare this site’s verdict against another aurora app on the same night, occasional disagreement is normal rather than a sign one of them is broken. Most of the difference comes down to which layer of the forecast an app leans on most heavily and how recently its data refreshed: an app built mainly around the multi-day Kp outlook will read differently from one built around live OVATION data, and both can differ from what people are actually reporting from outside at that moment, since visual reports also depend on local cloud cover and light pollution that no space weather product measures directly.
Putting it together
A genuinely useful aurora forecast, then, isn’t one number — it’s a layered read: check the multi-day outlook to decide which nights to keep free, watch the real-time solar wind data as the evening approaches for the actual Bz signal, and treat the Kp/OVATION-based verdict as the best available real-time answer, always cross-checked against whether it’s actually dark and whether the sky is clear where you’re standing. None of these layers, alone or combined, can promise a sighting — see what is the Kp index for why “high enough Kp” is a threshold check, not a guarantee.
Frequently asked questions
Why can aurora forecasts only reliably look 30-90 minutes ahead?
Because the most direct measurement of the solar wind that will hit Earth comes from a spacecraft stationed upstream at the L1 point, and depending on wind speed, that solar wind takes roughly 30 to 90 minutes to travel from there to Earth. Beyond that window, forecasts rely on estimates rather than direct measurement.
Are multi-day aurora forecasts useless, then?
Not useless, but different in kind — a multi-day outlook tells you whether a storm-causing event (a CME or a fast coronal hole stream) is expected to arrive at all, which is genuinely useful for planning which nights to keep free. It just can't tell you the precise strength or exact timing that a 30-90 minute forecast can.
What is the OVATION model, and how is it different from Kp?
OVATION is a NOAA model that estimates the probability of visible aurora across a geographic grid, driven by real-time solar wind measurements rather than the single planetary Kp number. It captures the oval's actual shape and local variation in a way Kp, being a single global average, cannot.
Why does this site combine Kp, OVATION and darkness into one verdict?
Because none of the three alone is sufficient: Kp gives a rough threshold, OVATION gives a more detailed live probability, and neither says anything about whether it is actually dark where you are. Combining them is the only way to get a genuinely useful yes/no answer for a specific place and moment.
How is the OVATION model actually built?
OVATION Prime estimates the energy flux of precipitating particles across a geographic grid (roughly one degree of latitude and longitude) using statistical relationships derived from years of particle-detector observations aboard low-Earth-orbit satellites, calibrated against real-time solar wind inputs. It is a statistical/empirical model rather than a first-principles physics simulation, which is part of why it is fast enough to update every few minutes.
Why do two different aurora apps sometimes show different odds for the same night?
Most disagreement comes down to which layer an app is emphasising and how it weights them: an app leaning heavily on the multi-day Kp forecast will read differently from one leaning on live OVATION data, and both can differ from ground-truth reports once people are actually outside looking. None of these are 'wrong' exactly -- they are answering slightly different questions.
Does this site's verdict use anything other than Kp, OVATION, darkness and moon?
Those four are the core inputs, cross-checked against each other rather than used independently -- the goal is a single, defensible answer rather than a wall of separate numbers. Cloud cover is deliberately excluded from the automated verdict, since no NOAA product measures it; checking a local weather forecast (see reading a cloud forecast) is left to the visitor.