A forecast for the color of the sky at sunrise and sunset — for the moment you're deciding whether to grab a camera and walk outside.
Type any location in the contiguous US and you get a 0–100 color-quality score for the next sunrise or sunset, with diagnostics that show which physical conditions drove the number. This page is the long-form version of those diagnostics: the data LookoutWX reads, and the model that turns it into a score.
Low-angle sunlight takes a long slant path through the atmosphere, and air scatters blue and green far more strongly than red (Rayleigh scattering goes as λ⁻⁴), so only the warm end of the spectrum survives to your eye. But plain orange isn't a great sunset. The drama needs three things at once: clouds high enough to still catch direct light after the ground is in shadow, air clean enough not to mute the reds, and an unobstructed corridor between the sun and those clouds. LookoutWX scores each — on both the sun side and the anti-solar side — and reports the better of the two phases.
The score is a weighted blend of three physical conditions:
Mid- and high-altitude clouds overhead — the screen that catches warm light.
Cirrus and altostratus around 4–10 km altitude are tall enough that low-angle sunlight still reaches them after the surface is in shadow. They're the canvas for the color show. Too few clouds and there's nothing to light up; too many, and the light can't get through. In the LookoutWX model the sweet spot sits near 45% mid-and-high cover — enough canvas to catch the light, enough gaps to let it through — with the lower atmosphere unobstructed.
The path between the sun and your sky, sampled out to 380 km.
Even with great clouds overhead, you'll see nothing if there's a wall of low cloud between you and the horizon — the warm light never reaches you. The corridor is the volume of air between the setting sun and the clouds you'd otherwise see lit. LookoutWX samples cells at 15, 40, 100, 220, and 380 km along the sun's azimuth — distances that correspond to time-slices through the color window, since the warm light reaches cells progressively further toward the sun as the sun drops below the horizon — and penalizes low cloud that would obstruct the light path along the way.
Humidity, smoke, and haze. Crystalline air gives saturated reds.
The longer the path through the atmosphere, the more chances scattering and absorption have to mute the colors. High humidity, dust, and smoke each desaturate the warm end of the spectrum. The most striking sunsets tend to follow cold fronts: dry, low-aerosol air after a storm has rinsed the atmosphere clear.
A sunset (or sunrise) doesn't happen all at once. There's the instant the sun sits at the horizon, and then there's a window of 10–30 minutes around that instant when the sky can keep changing, sometimes dramatically.
The afterglow includes one of the loveliest atmospheric phenomena, the Belt of Venus — look opposite the sun a few minutes after sunset and you'll often see a soft pink band sitting above a deeper blue-grey strip. The pink is reddened sunlight scattering off the upper atmosphere; the strip below it is Earth's own shadow cast against the air. LookoutWX scores the anti-solar sky alongside the over-sun sky for exactly this reason, and distinguishes the two ways it can light up: the clear-sky shadow-band glow (modeled on Lee & Hernández-Andrés' 2003 twilight-purity measurements) versus reddened light caught on anti-solar clouds. Both lift the afterglow score; the “where to look” note tells you which one to expect.
Your displayed score is whichever phase scores higher. The peak-time line on each forecast tells you when, relative to the event, that color is expected to be at its strongest.
Every location is read as a fan, not a single point. LookoutWX samples the observer (P0), eleven cells at five distances along the sun's azimuth — single points close in, three-cell arcs farther out (P1–P5) — and three points on the anti-solar side (A0–A2) for the afterglow. Each model and the satellite are sampled at every one of those cells, so the corridor, the canvas, and the clarity are evaluated over the actual volume of sky between you and the sun rather than at your feet. Scores refresh as new model cycles land (HRRR is hourly) and as fresh satellite scans arrive, so a forecast seen at noon can legitimately move by evening.
The three ingredients each reduce to a number in 0–1 — clarity, canvas (the cloud screen), and tunnel (the open corridor) — and combine into the displayed score as:
Clarity is weighted slightly above canvas — clean air is the single ingredient most common to vivid skies (Corfidi). The gate is the corridor: 0.4 + 0.6·tunnel at sunset (a partial wall still lets overhead clouds light up) but a much tighter 0.1 + 0.9·tunnel for the afterglow, because Belt-of-Venus pink is direct sunlight reddened through the limb — close the corridor and there's no light left to redden. The square-root expansion spreads real-world scores across the 30–90 range instead of bunching them near zero. Higher is better.
LookoutWX also reports a confidence figure. It starts from how much the 31 ensemble members agree on cloud cover at your location, then folds in whether the independent models agree with each other, whether the satellite corroborates the forecast cloud, how much the forecast has been swinging between recent model cycles, and how much of the light corridor the models actually cover. Low confidence means the number could move — check back closer to the event.
LookoutWX fuses five NOAA forecast models with one real-time satellite feed — each chosen for a specific job, from the high-resolution cloud field to the aerosol load to the live view of the sky right now.
NOAA's CONUS convection-allowing model and the canvas/moisture workhorse. LookoutWX reads ice- and liquid-water path, cloud-base height, low/mid/high cloud cover, relative humidity at five levels (2 m and 250/500/700/850 hPa), total precipitable water, surface precip rate + phase for the rain-washout gate, and near-surface + column smoke (recorded for clarity research). At 3 km it resolves marine-layer edges, mountain-wave clouds, and valley fog that global models smear out.
NOAA's global model, read for the two fields HRRR doesn't expose here — surface visibility and boundary-layer height — plus its own low/mid/high cloud cover as an independent cross-check.
The smoke, dust, and haze inputs to air clarity. The 440 nm channel additionally feeds the Lee-2003 twilight-purity curve that shapes clear-sky Belt-of-Venus vividness.
LookoutWX reads the precomputed total-cloud-cover mean and standard deviation. The spread between members at your location is what becomes the confidence figure — how much the ensemble agrees about the sky.
The RRFS Ensemble Forecast System — a 3 km convection-allowing ensemble. LookoutWX reads NCEP's precomputed mean/spread of cloud cover, cloud base, and visibility. Currently recorded for spread calibration but it does not yet move the score or confidence; the feed becomes NCEP-operational at the end of August 2026.
The geostationary satellite, read across five Level-2 products: cloud-top height, cloud-top phase, clear-sky mask, cloud optical depth, and aerosol optical depth. Within ±30 minutes of the event the observation is blended into the canvas and clarity components, weighted by scan freshness and retrieval quality — and it can veto a forecast outright when the model claims cloud the satellite can see isn't there. The measured cloud-top height also anchors the light-corridor geometry in place of an assumed altitude. The optical retrievals (cloud and aerosol depth) only resolve well above ~10° sun elevation, so near the terminator LookoutWX down-weights them by scan freshness and sun angle rather than trusting them blindly.
LookoutWX serves the contiguous United States only, because HRRR and REFS are CONUS-only; Hawaii, Alaska, and points outside the contiguous US aren't covered.