How these numbers are computed

Every date on this site is read from one dataset: NOAA/NCEI U.S. Climate Normals 1991–2020, annual/seasonal — the National Centers for Environmental Information's official 30-year station statistics. Nothing is modeled, interpolated, or estimated by this site.

The freeze-date percentiles

• For each station NCEI computes, from 30 years of daily minimum temperatures, the probability distribution of the last spring freeze and first fall freeze at thresholds including 36°F, 32°F and 28°F.
• The three columns shown are NCEI's 90/50/10 probability levels. For a last-spring date, "9 in 10 years past by" = only a 90% probability point — one spring in ten still saw a later freeze. "Typical" is the median. The same logic mirrors for fall.
• In mild climates the "9 in 10 years by" first-fall date can land in January — the freeze belongs to that winter, just after New Year.
• Where a station reaches a threshold in fewer than half of years, NCEI publishes no dates. This site says so ("frost-free most years") and shows NCEI's frequency-of-occurrence figure instead of inventing a date.

The frost-free windows

The windows table is arithmetic on those same percentiles, nothing more: the cautious window runs from the 1-in-10-late spring date to the 1-in-10-early fall date; typical between the medians; gambler's between the 9-in-10 points. This site reports dates and odds — it does not tell you what, whether, or when to plant.

Station matching

• Cities are the largest US incorporated places by the Census Bureau's official population estimates (data-driven, no hand-picked list).
• Each city is matched to the nearest NCEI normals station (great-circle distance from the city's census-gazetteer reference point) whose freeze-date record is complete and internally consistent; stations farther than 50 km are never used. The station's name, id, distance and elevation appear on every city page.
• A city with no qualifying station within 50 km is excluded rather than approximated.

Validation this site's pipeline enforces

• Percentile ordering must be internally consistent (a 9-in-10 spring date must not fall after the 1-in-10 date, etc.) or the station is rejected.
• The gap between median freeze dates must agree with NCEI's own median growing-season-length statistic to within 30 days (they are different statistics, so exact equality is not expected) or the station is rejected.
• Dates must parse as real calendar dates; probabilities must lie in 0–100%.

Honest limitations

• A station is a point. Valleys collect cold air, cities hold heat, elevation shifts dates — a few miles can move a frost date by two weeks. The station distance is printed on every page so you can judge.
• Normals are a 30-year rear-view mirror (1991–2020), not a forecast, and recent decades have trended warmer than the full record.
• 380 city records; data pulled 2026-09-05; the normals themselves change only when NOAA issues the next set (roughly 2031).