Frost Gap Check
Research preview · 8 US regions
Point-specific frost intelligence

Your forecast is regional.
Your cold pocket isn’t.

In hilly country, valley floors and hollows run colder than the forecast on frost nights. Type your address to see the real weather-station evidence around your field, orchard, or vineyard — and whether you sit in a cold pocket.

or click anywhere on the map

Address lookup uses the U.S. Census geocoder. This static preview does not store your search.

Shaded areas are frost pockets — where the standard forecast runs too warm on frost nights. Everywhere else is fine, so only the valleys that matter show color: 3–4°F too warm 4–5°F 5°F+ · dots = the weather stations behind it. Click a spot or type your address.
How this works — in plain terms

We measured where the forecast is wrong, with real weather data.

No jargon needed. Here is exactly what is behind the map above.

1

We collected real temperatures

Millions of readings from two kinds of weather stations near you: official airport stations, and thousands of backyard “citizen” weather stations. We don’t trust them blindly — every backyard sensor is checked against nearby trusted stations, and the badly-sited or miscalibrated ones are thrown out before anything is used.

2

We compared them to the forecast

For every clear, calm night — the kind when frost forms — we lined up what actually happened at each spot against what the standard public forecast predicted. That shows which spots the forecast gets wrong, and by how much.

3

We found the cold pockets

Some spots — valley floors and hollows — pool cold air and run several degrees colder than the forecast on frost nights. Those are the red dots on the map. Type your address to see the measured evidence around it.

Why it matters: across 137 stations last winter, the standard forecast said “above freezing” on 425 nights that actually froze. In a cold pocket, that gap is the difference between doing nothing and losing a crop.

4.87 millionreal temperature readings

From 2,300+ backyard stations across 8 US regions, 3 winters.

133,200past forecasts checked

The standard public forecast, pulled for each station’s exact spot.

11,254nights compared

Where we had both a real reading and a forecast to line up.

140 of 148stations we trust

We keep the stations with genuinely good data and drop only the 8 that were broken or miscalibrated — while being careful not to throw out real cold pockets.

The technical terms, translated
  • NOAA — the U.S. government weather agency; the source of all this data.
  • Citizen weather stations (a.k.a. CWOP / PWS) — backyard personal weather stations people run; there are thousands, so they cover the valleys airports miss.
  • MADIS — NOAA’s public archive, where we downloaded the historical station readings.
  • NBM (the “standard forecast”) — the National Blend of Models, the official public temperature forecast we check against.
  • Clear, calm night — no wind, no clouds; the condition that lets cold air pool in low spots and cause frost.
  • How we keep the good stations — a badly-sited backyard sensor gives itself away in the daytime (it reads too hot in the sun); a genuine frost pocket only shows up at night. We use that difference to drop bad sensors while keeping the real cold pockets — so a valley station reading cold isn’t mistaken for a broken one.
What goes into the estimate — and what makes it confident
  • Nearby measured stations — the biggest factor. The number is anchored to how warm the forecast actually ran at real stations around you. More stations, closer together and agreeing, = a more confident estimate; far from any station, we say so.
  • Terrain shape. How low and sheltered a spot sits — a valley floor pools cold air, a ridge doesn’t. We use elevation and “topographic position” (how far below the surrounding land the spot sits) and valley depth.
  • The night’s weather. Cold pockets only form on clear, calm nights, so the tool checks the live forecast for that setup.
  • Season. The effect is strongest in the cold months.
  • What matters most (and least). The surrounding measured stations and how deep the valley is drive the estimate. Raw altitude alone is weak — it’s the relative low spot that pools cold, not the height. Least useful: satellite data (it mostly repeated what the terrain already told us).