← back to the sky · the field guide

DATA & METHODS

This page explains how satellite passes, magnetometer records and 486 star-calibrated photographs become the aurora on your screen: which parts are measured, which are modeled, which are procedural, and how each one is checked. No prior background is assumed. Every number an enthusiast might want to verify is here, with its source named.

Three words, used precisely

Everything the observatory draws belongs to one of three categories, and the app labels which is which:

Each section below is tagged with the category it describes.

The instruments MEASURED

The reconstructions rest on satellites that image or fly through the aurora, ground magnetometers that record its currents, and photographs with enough stars to calibrate. What each contributes:

Source	What it measures	Cadence	Role here
DMSP/SSUSI (polar weather satellites)	Far-ultraviolet images of the aurora: brightness, energy flux and boundaries along each orbit swath	One pass per ~101-min orbit	The primary brightness/energy measurement. Flux, boundaries & timing are taken per pass; no interpolation between passes is presented as data
POES + MetOp TED (5 satellites)	Precipitating electrons, 50 eV–20 keV, sampled in several different local-time planes	Continuous along-track	Fills sectors and moments between imaging passes; cross-calibrated to FUV-equivalent flux (measured median ratio ≈ 8; TED's band misses the storm-time hard tail SSUSI sees)
AMPERE (66-satellite Iridium constellation, JHU/APL)	Field-aligned currents: the electrical circuit between space and the atmosphere, which outlines the auroral oval	Continuous, 10-min product	The oval's position between imaging passes: a continuous measurement that holds the reconstruction when no imager is overhead
TREx RGB + THEMIS all-sky cameras (UCalgary)	Calibrated images of the real sky over western Canada	Minutes / 5-min mosaics	Independent ground truth: the renderer's structure statistics are benchmarked against these frames at the same minute and place
Geomagnetic indices (Kyoto WDC, GFZ)	AE (auroral-zone currents, minute by minute) and SYM-H (the storm's ring current)	1 min	Substorm detection, expansion scaling, and the gates on storm-time dynamics
NOAA SWPC feeds	Real-time solar wind, Kp, OVATION nowcast	Minutes	Drive the live scene through the same pipeline as the validated storms
The photo corpus	486 star-calibrated citizen-science and ISS photographs	—	A measurement (borders, heights, structure spacing) and the renderer's benchmark; details below

How a storm is rebuilt MEASUREDMODELED

No single instrument sees the whole aurora at once: imagers pass overhead every ~101 minutes, particle detectors sample thin tracks, and the currents are measured continuously but coarsely. The reconstruction's job is to fuse them into one coherent sky without inventing anything the instruments didn't record.

The fusion happens in the sun-fixed magnetic frame, the aurora's natural coordinate system. The oval is organized by the sun-Earth direction and Earth's magnetic pole rather than by the rotating ground beneath it, so a measurement made over Alaska constrains what the same magnetic sector looks like when Siberia rotates into it. In that frame:

• Position comes from AMPERE's measured 10-min field-aligned-current boundaries, continuous through the whole storm.
• Brightness and energy are fused from DMSP/SSUSI far-ultraviolet imaging and POES + MetOp TED electron precipitation, the five particle satellites cross-calibrated to FUV-equivalent flux (measured median ratio ≈ 8, because TED's 50 eV–20 keV band misses the storm-time hard tail SSUSI sees).
• Between samples, the fused field is carried in the sun-fixed frame with AE-indexed substorm scaling; the auroral boundary moves along a relation fitted to the measured record.

Where the data is thin, the app says so. Every grid cell carries a confidence value: how directly that sector and moment were sampled. In Source mode, brightness dims with confidence, so a poorly-sampled sector looks uncertain. The Simulation renders the best estimate at near-full brightness and reports confidence in the HUD instead. F18's pre-midnight blind sector and the gaps between passes are constrained by POES crossings where they exist.

Daylight is handled physically. Dayside precipitation is real (the imagers measure it in sunlight), but no eye can see aurora through daylight. All auroral layers therefore fade with ground daylight using the same sun position that draws the terminator: the model fades out fully, while measured snapshots dim to a ghost so you can still see what the instruments saw.

Photographs as measurement instruments MEASURED

A photograph with visible stars can be turned into a calibrated instrument. Star patterns give a full pointing solution: exactly where the camera stood, where it aimed, and what angle of sky each pixel covers. 486 citizen-science and ISS photographs from the 2024 storms were star-calibrated this way, and they do three jobs:

They measure the aurora

• The visible red border and column tops. Ensemble fits across the corpus measure where the soft red emission reached and how tall the columns stood, epoch by epoch: border heights of 110–360 km across the storms, with Gannon-night column tops at 790–870 km. The reconstruction extends soft red emission equatorward to that measured border and lifts red-column tops to the measured heights.
• Fine-structure spacing. 19 star-calibrated photographs resolve the folding of real curtains: a median fold-strand spacing of 15 km. The renderer draws 17 km, a fitted and disclosed number.
• Stereo triangulation has begun. Two same-minute corpus photographs (Muszyna & Kłonów, Poland, May 11 2024; a 122 km baseline) stereoscopically measure the auroral lower border at 99 ± 5 km, independently confirming the model's ~100 km base height.

They grade the result

Rendered brightness is benchmarked camera-by-camera against the real frames, and the renderer's structure statistics are compared against calibrated all-sky research cameras (TREx RGB) at the same minute and place. In the app, click any calibrated camera dot to stand at its exact spot and moment with the plate-solved frame outlined on the sky; gold dots blend the actual photograph over the render for direct validation.

One disclosed assumption

Frames are auto-leveled using the photo's own horizon (the level-camera assumption that also powers the star clock). Untick "level" in a photo-POV to see the raw solved orientation at the assumed exposure time; the residual tilt there mostly measures clock error from parallactic rotation rather than camera tilt.

The physics in the colors and altitudes MODELED

Auroral color follows atomic physics, and the renderer computes it rather than paints it:

• Color from energy. The mix of emissions follows the Rees & Luckey (1974) energy–ratio relation: harder electron spectra drive green low, softer spectra leave red high. (The physics of why is in the field guide.)
• Altitude from energy. Peak emission heights follow the Fang et al. (2008) Maxwellian energy-deposition parameterization (its P_ij coefficients verified against three independent sources), then epoch-scaled by our photo-measured heights. Storm-time heating expands the upper atmosphere and lifts the red: Kataoka et al. (2024) measured Gannon-night red to ~1000 km, consistent with our independent 790–870 km ensemble tops.
• The red line's slow lifetime. The 630 nm red line takes ~110 s to radiate, so fine ray structure physically cannot exist in deep red; the renderer fades ray texture with altitude accordingly.
• Sunlit blue tops. Where a column top rises above Earth's shadow into sunlight, it turns N₂⁺ blue-violet (the Gannon signature), using real shadow geometry per location and time, matching the −24° solar-elevation threshold (Grandin et al. 2024). Deep-night sectors self-gate: the shadow there sits above every measured top.
• SAR arcs. Below the measured red border, soft red renders as SAR-regime diffuse glow: smooth, structureless, pure 630 nm (Kozyra et al. 1997), at ~400 km (altitude per FRIPON Europe observations), with latitude fitted to the SAR-classified reports in Chris Wicklund's sightings map and drawn only over the longitude sectors that reported it, night side only.
• The background sky. A faint flat airglow floor, the 557.7 nm nightglow at ~250 R (Leinert et al. 1998, Table 13), is always present; it is a real sky glow, not part of the auroral model. City lights from NASA's Black Marble add the amber that real night sides have.

Substorms and motion MEASUREDMODELED

The aurora brightens in discrete substorm cycles. The reconstruction detects each one from the measured magnetometer record, and its motion follows published morphology:

• Detection. Substorm flags (amber ▼ on the timeline, ⚡ badge while active) come from the AE index: onset = a rise of ≥150 nT within 20 minutes from a non-rising baseline; the expansion lasts until AE stops making new highs.
• Expansion. During flagged expansions the simulation brightens a bulge using the published statistical template (Gjerloev et al. 2007), expanding westward/eastward at 8.8 / 5.3 km/s (Ogasawara et al. 2011; Craven et al. 1989). Timing and amplitude come from our measured AE; onset location from the model's own brightest near-midnight sector.
• Arc dynamics. Omega-band brightness waves ride the poleward edge eastward at ~0.9 km/s during substorm-active hours (Opgenoorth et al. 1983). Giant-undulation waves ride the equatorward diffuse edge westward when SYM-H drops below −150 nT (Lui et al. 1982: the Kelvin–Helmholtz instability of the SAID shear). Procedural fine structure drifts along the arc at the published curl and fold scales (Hallinan 1976; Partamies et al. 2001).
• The bounds. All of this motion is brightness-led and mean-neutral, with any displacement bounded inside one grid cell. Animation never moves the aurora somewhere the measurements didn't put it.

What is procedural — and how it stays honest PROCEDURAL

Real auroral rays are 0.1–10 km wide, below the resolution of every instrument in this reconstruction. Drawing a measured 100-km grid cell as a flat slab would look like nothing anyone saw, so in Simulation mode the fine ray structure inside the curtains is generated procedurally. The rules:

• Position, brightness, extent and heights stay measured. Procedural texture only fills in detail inside cells the instruments lit.
• Its statistics are fitted. Strand spacing is tuned to our own corpus measurement (15 km measured median; 17 km drawn). Ray widths follow published distributions (Maggs & Davis 1968 → Yang et al. 2024), which our photo resolution cannot yet independently verify.
• Column tops vary per cell: the photo-measured epoch envelope, scaled by each cell's own measured flux and energy through the Rees range–energy curve: bright soft cells reach the measured maximum; weaker cells stand shorter.
• Tall-ray placement follows the observed soft-precipitation flank (illustrative until photo triangulation replaces it), but ray altitudes (500–1100 km) and the red→blue sunlight transition use real shadow geometry per location and time.
• In ground view, cells near you resolve into 3D field-aligned columns at the local dipole inclination, so looking straight up produces the true corona convergence at the magnetic zenith. Column placement, heights and brightness are the measured field; only sub-cell ray multiplicity is procedural.

Which auroral forms render, partially render, or are left out (dunes, STEVE, the picket fence) is tagged form-by-form in the shapes of the aurora.

Eyes versus camera MODELED

The default 📷 long-exposure view shows what cameras record, calibrated against the corpus photographs. The 👁 naked-eye view applies the physiology of night vision to the whole scene, not just the aurora:

• Dark-adapted rod vision has a visibility floor near ~1 kR of green: below it, nothing; just above it, a colorless gray veil.
• Rod sensitivity peaks at 507 nm and collapses toward 630 nm (the Purkinje shift), so faint aurora renders dim gray-green, and red appears only where it is bright enough to engage cone vision, matching the naked-eye versus photograph color reports collected by Grandin et al. (2024).

This is the rendering side of "why didn't it look like the pictures?" The practical side, with camera advice, lives in how to see the aurora.

The live sky and the forecast MODELED

LIVE renders NOAA SWPC's real-time feeds (the OVATION nowcast and solar-wind data) through the same pipeline as the validated storms, with alert conditions checked against the sky every 15 minutes. FORECAST takes the WSA–Enlil model's inbound solar wind, drives the border relation fitted to our measured storms, and simulates three nights ahead complete with synthetic substorms. It is labeled in the app as what it is: a simulation of what the predicted numbers would look like. What forecasting physically can and cannot deliver is covered in the field guide.

The historical simulations MODELED

Nothing before the satellite era can be a measured reconstruction, so the historical scenes are labeled historical simulations and anchored to the strongest records that exist:

• 2003 Halloween runs on the actual measured geomagnetic record — hourly OMNI and Kyoto indices (Dst −353/−383 nT), with the ACE instrument-saturation gap bridged using documented values.
• 1859 Carrington (Aug 28 – Sept 3: precursor, great storm, second night) is anchored to 33 verified eyewitness sightings with documented hours, and to the timed reconstructions in the modern literature. Where the record is uncertain, the app tells you.

The storms themselves, with primary sources, live in the great storms.

Small honest compressions MODELED

A few things are deliberately drawn out of scale or simplified for readability. They are listed here:

• The Moon renders at its true ephemeris direction and phase for every scene's clock (truncated Meeus series, verified to <1′ against the reference example); each historical storm shows its historically correct moon, and moonlight lightens the night hemisphere in proportion to the illuminated fraction. Its distance, however, is compressed to 9 R⊕ and the disk drawn ~35% over-size for readability.
• The magnetic-latitude grid and dipole field lines (drawn at L = 2, 3, 5, 8) use the same centered/tilted-dipole transform as the border rings. That is adequate at auroral latitudes; the real IGRF field diverges most near the South Atlantic Anomaly.
• Sighting pins are event-integrated: a pin shows that a place reported aurora at some point during the storm, not at the instant on the clock.
• Ground-camera mosaics (TREx/THEMIS) are projected at an assumed 110 km, a product limitation that smears red aurora above 110 km toward footprint edges; rims are feathered by view-elevation confidence.

How changes are validated MEASURED

The renderer is treated like a scientific instrument: it must pass validation before any change ships. Rendered brightness is benchmarked camera-by-camera against the 486 calibrated photographs. Structure statistics are compared against TREx research cameras at the same minute and place. The physics relations above are implemented from the cited papers. When something is illustrative, fitted, or below instrument resolution, the app and this page say so in plain words; when you catch a place where they don't, the in-app 🛟 help button files straight to the maintainer.

Sources & data provenance

Works cited on this page, as implemented in the renderer. The claims this page shares with the field-guide chapters carry DOI-verified numbered references there: field guide · seeing guide · great storms. The implementation set also includes the Newell and Robinson coupling/conductance relations.

• Rees & Luckey (1974) — auroral electron energy from spectroscopic emission ratios; the color–energy relation. (DOI on the field guide.)
• Fang et al. (2008) — Maxwellian energy-deposition parameterization; peak emission altitudes.
• Kozyra et al. (1997) — SAR-arc physics: ring-current heat conducted into the subauroral ionosphere.
• Kataoka et al. (2024) — Gannon-storm red emission measured to ~1000 km.
• Grandin et al. (2024) — Gannon citizen-science observations: naked-eye color thresholds, sunlit-blue −24° gate. (DOI on the field guide.)
• Gjerloev et al. (2007) — statistical substorm brightness template.
• Ogasawara et al. (2011); Craven et al. (1989) — substorm westward/eastward expansion speeds (8.8 / 5.3 km/s).
• Opgenoorth et al. (1983) — omega-band eastward drift (~0.9 km/s).
• Lui et al. (1982) — giant undulations: Kelvin–Helmholtz instability of the SAID shear.
• Hallinan (1976); Partamies et al. (2001) — auroral curl and fold scales.
• Maggs & Davis (1968); Yang et al. (2024) — auroral ray width distributions.
• Leinert et al. (1998), Table 13 — the 557.7 nm airglow floor (~250 R).
• Spanswick, E. & Donovan, E. — Transition Region Explorer (TREx) RGB Dataset [Data set], University of Calgary — the calibrated all-sky ground truth behind structure and color validation. doi:10.11575/4P8E-1K65
• Wicklund et al. (2024) — the 2024 crowd-sightings catalogs, including SAR classifications. doi:10.17605/osf.io/e87cs

Data: NASA SPDF (SSUSI, OMNI) · NASA GIBS (VIIRS DNB · Blue Marble) · AMPERE Science Data Center (JHU/APL, NSF; PI B. Anderson) · UCalgary AuroraX (TREx RGB + THEMIS ASI mosaics) · NOAA SWPC · Kyoto WDC · GFZ Potsdam · sightings: Chris Wicklund's maps (Wicklund et al. 2024, OSF e87cs) · 1859 catalogs: Green & Boardsen 2006, Hayakawa et al. 2018/19, Moreno Cárdenas et al. 2016, González-Esparza et al. 2018 · ground view: map © OpenStreetMap contributors © CARTO, terrain Mapzen/AWS Terrain Tiles. Photographs belong to their photographers.

In memory of Jennifer Gannon. The May 2024 superstorm carries her name — this observatory began as its reconstruction.

the sky · the sun · field guide · about · privacy · pro · ☁️ seasonmap · ☕ support