Your weather app graded tonight's sunset a B+. Before you carry a camera up a hill for it, it is worth knowing what that letter actually measured.
Grades and star ratings have spread across weather products over the last few years, and most of them are computing something real. The trouble is that few of them publish the rubric, and a letter without a rubric is a mood.
We grade skies ourselves, so this is not an objection to the format. It is an argument that a grade earns trust only when the inputs, the weights, and the failure modes are all visible to the person deciding whether to go outside.
A sky score weights cloud fraction by layer, clearance on the western horizon, humidity, and aerosol load into a single quality grade for the light during the 20 to 40 minutes around sunset.
What A Sky Score Measures At Forecast Time
A sky score is a quality prediction rather than an event probability. It estimates how good the light will be at one location during a narrow window, and it answers on a scale rather than with a yes or a no.
That distinction matters because the two are built differently. A rain probability comes out of ensemble agreement about a binary event, while a sky score comes from weighting several continuous fields against each other — a difference worth understanding alongside how forecast models disagree with one another.
The physical inputs are the ones any forecaster would reach for, ranked by how much each one moves the outcome. The terms that carry real weight include but are not limited to:
- Upwind horizon clearance. The slant path the last direct sunlight travels before it reaches the clouds over your head. A gap between cloud base and horizon roughly 100 to 200 kilometers west of you is the single strongest predictor of color.
- High cloud fraction. Cirrus and cirrostratus between about 20,000 and 40,000 feet stay sunlit long after your street has gone dark. Coverage near 30 to 60 percent grades highest, while a solid sheet at that altitude behaves like a lid.
- Mid-level structure. Altocumulus around 8,000 to 18,000 feet supplies the texture that turns a wash of orange into a sky with edges. Mackerel rows, undulations, and distinct cell boundaries all raise the score.
- Low cloud obstruction. Stratus, stratocumulus, and marine layer below about 6,500 feet mostly subtract. A broken low deck can be extraordinary, and a solid one is simply a ceiling.
- Atmospheric clarity. Precipitable water and aerosol optical depth control saturation. Dry, clean air behind a frontal passage renders cleaner reds than humid, hazy air ahead of one.
- Geometry and duration. Latitude, date, and the sun's setting angle determine how long the color window stays open. That is a handful of minutes near the equator and most of an hour at high latitude in shoulder season.
All of these collapse into one number and then into one letter, which is where the information loss happens. The reason to publish the breakdown is that a reader who disagrees with a single term can overrule the grade with their own eyes.
| Rubric term | Weight | What it answers |
|---|---|---|
| Upwind horizon clearance | 30 | Can light reach the clouds at all? |
| High cloud fraction | 25 | Is there a canvas that holds color? |
| Mid-level structure | 15 | Will the sky have texture or a flat wash? |
| Low cloud obstruction | 15 | Is the view blocked from below? |
| Clarity (moisture + aerosol) | 10 | How saturated will the reds render? |
| Geometry and duration | 5 | How long does the window stay open? |
Why The Western Horizon Outweighs The Sky Overhead
A sky score describes the light path first and the clouds second. At sunset the sun already sits below your local horizon, so everything you see overhead is reflected off cloud bases lit from far away.
That light travels a long, shallow route through the densest part of the atmosphere. It typically crosses the lowest few thousand feet somewhere 100 to 200 kilometers upwind of where you are standing.
If a bank of stratus or a thick frontal deck sits along that route, the light is absorbed before it ever arrives. The result is a textbook cirrus field overhead and a gray, unlit sky beneath it.
The color you see is lit from roughly 100 to 200 kilometers west of you, where the last direct sunlight passes under the cloud deck. A score that ignores that upwind gap is guessing.
Accordingly, clearance carries the largest weight in our rubric, and it is the term most often missing elsewhere. A score that samples only your own grid cell is describing the stage without checking whether the lamp is switched on.
This is also why reading cloud layers by altitude is the most useful skill a photographer can bring to a grade. Keep in mind that you can verify the clearance term yourself, forty-five minutes out, by looking west.
The Cloud Layers, Ranked By What They Do To A Grade
Not every cloud contributes in the same direction, and the same deck can help or hurt depending on where it sits relative to you. Here is how each layer typically moves the score:
| Layer | Typical height | Effect on the grade |
|---|---|---|
| Cirrus / cirrostratus | 20,000–40,000 ft | Strongest positive. Holds color 10 to 20 minutes past sunset; best near 30–60% cover. |
| Altocumulus / altostratus | 8,000–18,000 ft | Positive when broken, negative as a solid sheet. Supplies most of the visible texture. |
| Stratocumulus / marine layer | 1,500–6,500 ft | Usually negative overhead. Strongly negative when it sits on the upwind light path. |
| Fog / surface stratus | Surface–1,000 ft | Negative for sunset color, and a subject in its own right. |
| Clear sky | — | Neutral to negative. Nothing to catch the light; grades in the C range at best. |
High cirrus between about 20,000 and 40,000 feet holds color longest because it stays sunlit after the surface goes dark. Coverage near 30 to 60 percent grades highest.
A zero-cloud evening deserves a note, because clear skies get graded generously by products that only count sunshine. There is nothing up there to catch the light, though the gradient afterward is still worth staying for — see what blue hour actually delivers and how fog rewrites the same scene.
Humidity, Aerosol, And The Smoke Sweet Spot
Clarity is where most published scores are thinnest. Two evenings with identical cloud fields can grade a full letter apart on moisture and particulates alone.
Water vapor scatters across the spectrum and desaturates the result, which is why a muggy August evening at a 74°F dew point tends to produce a soft peach rather than a hard red. Dry air behind a cold front, with dew points in the 40s and pressure building, gives sharper and more saturated color — the mechanism is covered in our piece on dew point versus relative humidity.
Aerosol behaves nonlinearly, which is what trips up most scoring models. A modest load of fine particles deepens the red end by scattering shorter wavelengths out of that long slant path.
Moderate aerosol deepens red; heavy smoke flattens it. Above roughly 0.5 aerosol optical depth, the sky mutes to a uniform orange-gray and any score built without a smoke term runs too high.
What's more, the sign of the aerosol term flips within a single season in the same city. Any grade computed from cloud fields alone will read optimistically through a Western summer, and the error will be systematic rather than random.
Why A Letter Beats A Percentage Here
A percentage is a promise about frequency. When a product prints 70 percent, it is claiming that across many comparable evenings roughly seven in ten deliver, and that is a claim almost nobody publishes verification for.
The second problem is resolution. The inputs to a sky score carry enough error that 78 and 74 are the same forecast, and printing both invites a reader to act on a difference that does not exist.
A letter grade is an ordinal quality judgment at honest resolution: five bands, not a hundred. Percentages imply a calibrated probability of a binary event, and sunset quality is continuous.
This is the same failure that makes the rain number so widely misread, as we argued in what a chance of rain actually means. Of course, a rubric is also decomposable in a way a percentage is not — you can show the six sub-scores and let the reader audit the one they doubt.
| Dimension | Percentage score | Letter grade |
|---|---|---|
| What it claims | Frequency of a binary event | Quality of a continuous outcome |
| Implied resolution | 100 distinguishable states | 5 distinguishable bands |
| Reader's decision | Requires a mental threshold | Band maps directly to an action |
| Auditability | Opaque single number | Decomposes into weighted terms |
| Failure when wrong | Reads as a broken promise | Reads as a miss by one band |
What The Bands Should Mean
A grade is only useful if each band maps to a decision. Ours are written as instructions rather than adjectives:
- A (85–100). Leave now and be in position 30 minutes before sunset. Expect peak color 8 to 15 minutes after the sun goes down.
- B (70–84). Worth the walk with whatever camera is already on you. Pick a foreground that still works if the sky underdelivers by a band.
- C (55–69). An ordinary sky. Shoot only if the composition carries itself — a reflection, a silhouette, a piece of architecture.
- D (40–54). Flat light with a low ceiling on the outcome. Better spent scouting a location for a stronger night.
- F (below 40). Blocked, overcast, or blown out by smoke. Blue hour may still be worth twenty minutes.
Read together, the letter answers whether to go, the sub-scores answer why, and the timing line answers when. That last piece matters more than most readers expect, which is why we treat the shape of golden hour light as part of the grade rather than a separate field.
When Not To Trust The Letter
A grade can be built on a sound rubric and still be wrong for reasons the rubric never touches. The failure modes worth knowing include:
- Grid resolution too coarse. Global models run at roughly 9 to 13 kilometers, which cannot resolve a marine layer edge or a coastal clearing line. Those features routinely decide the entire evening in San Francisco or Seattle.
- A stale model cycle. Cirrus rides the jet at 50 to 100 knots, so a run initialized 14 hours ago is describing a cloud field that has already left. Re-check the grade against the current jet stream pattern before committing to a drive.
- No local horizon model. Astronomical sunset assumes a flat sea horizon, and a ridge sitting 12 degrees up to your west ends the show early. No national dataset knows that ridge is there.
- No aerosol input. Smoke season breaks any score without an optical-depth term, and it breaks it in the optimistic direction. That is the worst kind of error for someone deciding whether to drive an hour.
- Single-source input. One deterministic run with no ensemble spread yields a confident letter with no confidence attached. The distinction is the subject of our piece on forecast confidence.
- No published verification. If nobody grades the grader, the letter is decoration. Track record is the only thing that converts a rating into a reason to trust it.
A grade older than the last model cycle is decoration. Cirrus moves at 50 to 100 knots aloft, so an early-morning run rating an 8 p.m. sunset describes a cloud deck that has since moved on.
Note that these failures are not evenly distributed. Coastal cities, mountain valleys, and anywhere with a smoke season will see a well-built score miss far more often than the flat interior will.
The forty-five-minute rule: whatever the letter says, look west forty-five minutes before sunset. If the horizon has a bright strip under the cloud base, the grade is probably low; if the western sky is uniformly gray to the ground, it is probably high.
How We Grade, And How We Check
We publish the sub-scores under every grade because a reader with a west-facing window is a better sensor than a 9-kilometer grid cell. The letter is a starting position, and your own look at the horizon is the tiebreaker.
We also publish the result. Sunset Verify compares each forecast letter against what the sky actually did, which is the only mechanism that turns a grade from a decoration into a record you can audit — the same accountability we apply to measuring forecast accuracy honestly.
If you want the reasoning rather than the number, that is what the daily brief is for. You can read how we write a brief or start with the rest of the journal.
Common Questions
High cirrus between roughly 20,000 and 40,000 feet, covering about 30 to 60 percent of the sky, with mid-level altocumulus underneath for texture. The requirement people forget is a clear gap on the western horizon, 100 to 200 kilometers upwind, so the last direct light can reach those clouds from below. Without that gap, a perfect cirrus field stays gray.What cloud cover makes the best sunset?
A percentage implies a calibrated probability of a binary event, and sunset quality is neither binary nor commonly verified. The inputs also carry enough error that 78 and 74 are the same forecast. Five ordinal bands report only the resolution the model can actually support, and each band maps cleanly to a decision — go now, walk over, or stay in.Why do weather apps use letter grades instead of percentages?
Both, depending on load. A light aerosol layer deepens the red end by scattering shorter wavelengths out of the long slant path, which is why hazy evenings can be spectacular. Above roughly 0.5 aerosol optical depth, usually thick smoke, the sky flattens into a uniform orange with no structure and very little dynamic range for a camera to work with.Does wildfire smoke improve or ruin a sunset?
Roughly 12 to 24 hours for the cloud-layer terms, and only 3 to 6 hours for the western-horizon clearance that decides most evenings. Cirrus travels with the jet at 50 to 100 knots, so a grade computed from an overnight run is describing a sky that has already moved. Re-check after the newest model cycle lands.How far in advance is a sunset grade reliable?
Your own view, once you are inside forty-five minutes. A model cannot see the ridge on your western horizon, the exact edge of a marine layer, or a hole in a deck that a 9-kilometer grid cell averages away. Use the letter to decide whether to get ready, then verify the clearance yourself before you commit to the drive.Should I trust the grade or my own view of the sky?