How the forecast works

Day length is the clock

The single biggest driver of when leaves turn is photoperiod — day length. It starts dropping after the June solstice, and the tree responds by building the abscission layer that eventually severs the leaf. Photoperiod is fixed by latitude and date, which is why peak lands in roughly the same week every year in a given place.

That makes timing genuinely predictable. It also means anyone claiming a dramatically different peak date this year is usually overselling the weather signal.

Latitude and elevation set the offset

We anchor at roughly October 20 for 40°N at sea level, then shift:

• about 2.5 days earlier per degree of latitude north
• about 3.5 days earlier per 1,000 feet of elevation

That reproduces the windows published by state forestry departments and the Park Service: northern Minnesota in the last week of September, Vermont in the first week of October, the high Smokies mid-October while the valleys below them run into November.

The elevation term is why we build location pages rather than county pages. Newfound Gap and Cades Cove sit in the same county and peak about two weeks apart. A county-shaded map physically cannot tell you that.

Weather decides quality, not timing

Yellows and oranges come from carotenoids that are in the leaf all along, revealed as chlorophyll breaks down. They show up every year regardless of weather.

Reds are different. Anthocyanins are manufactured in autumn from sugars trapped in the leaf, and they need warm sunny days followed by cool, frost-free nights. Summer drought, unusually warm nights, or an early hard freeze all suppress them. That's the difference between a year people talk about and a year of muted bronze.

The light layer

Solar position is computed locally from NOAA's equations rather than pulled from an API, so every location gets sunrise and sunset azimuth for its own peak date. Sunrise bearing drifts by more than 8° over a month in autumn, which is enough to decide whether a ridge is backlit or flat when you arrive.

Our sun times agree with the standard NOAA implementation to within about a minute, and azimuths to within half a degree. For reference, the two most widely used public solar APIs disagree with each other by more than that.

What this doesn't do

• It is a climatology, not an observation. We model where peak usually falls. We do not currently ingest live satellite greenness or ground reports, so we cannot tell you that a specific ridge turned early this week.
• It does not yet adjust for this year's weather. Drought and temperature anomalies can move peak by several days and change color quality substantially. Adding that correction is the next piece of work.
• Species vary within a state. Oaks hold color weeks after maples have dropped. Larch and tamarack turn gold after everything else is bare. Our per-state species lists say what to expect, but the dates describe the dominant canopy.

When a number is uncertain, the page says so. A forecast that admits its error bars is more useful than one that doesn't.