How We Score the Sky

Step 1: Thirteen Layers of Sky

We pull cloud cover, humidity, visibility, and rain rate at 13 pressure levels from NOAA's GFS weather model, from 950mb near the surface up to 200mb near the tropopause. Those levels are grouped into four bands: mid-low, mid-high, high, and very high cloud, since clouds at different altitudes behave very differently at sunset.

Step 2: Is It Actually Lit?

Cloud cover alone doesn't tell you whether a sky will be beautiful. A cloud can sit directly overhead and stay grey, or be thin and glow orange, depending entirely on whether sunlight can reach its underside. For the two lowest bands, we detect the cloud base and cast rays back toward the sun at 40, 80, and 130 kilometers to check whether that light path is actually clear.

Bands that are lit use a different beauty curve than bands sitting in shadow, and the two are blended by how much of the cloud is actually illuminated. This is the main reason two forecasts with the same "cloud cover %" can look completely different in real life.

Step 3: The Moisture Penalty

Some of the least rewarding skies aren't obviously cloudy at all, just hazy, milky, and low-contrast. We look for elevated humidity at levels where cloud cover itself is reported low, sampled at the observer's position and 50 and 100 kilometers further along the sunset direction. Stacked haze across several levels can reduce the final score by up to 40%, even on a technically "clear" forecast.

Step 4: Combining Into One Score

Each band's beauty score is combined with fixed weights: mid-low 23%, mid-high 37%, high 27%, very high 13%, then adjusted by the moisture penalty. A clear sky scores around 30 (a safe bet, not a great one), scattered mid and high cloud with cirrus can reach 100, and a fully overcast low sky scores close to 1.