How these dates are calculated

The source

Daily minimum and maximum temperatures come from NASA POWER, which reanalyses satellite observations and model output into a continuous daily record on a global grid running from 1981 to the present. It is free, it is in the public domain, and it covers every land surface on earth, which is the only reason a worldwide frost dataset is possible at all.

For each location this site uses the 30-year window from 1995 to 2024. That is long enough for percentiles to mean something and recent enough that a warming trend has not been averaged away by decades of colder records.

Place names, coordinates and populations come from GeoNames, used under CC BY 4.0.

The growing year

A calendar year splits the southern hemisphere's growing season in half, so every calculation here runs on a frost year instead: it starts on 1 January north of the equator and 1 July south of it. Within that year, day 1 to day 196 is spring and everything after is autumn.

The last spring frost is the last day on or before day 196 that fell to the threshold; the first autumn frost is the first day after it. One code path serves both hemispheres, which is why a page for Christchurch reads exactly like a page for Chicago.

The percentiles

For each year in the record the site finds that year's last spring frost and first autumn frost, then takes percentiles across the resulting distribution. Three are published:

• Cautious — the 90th percentile of last frost. Frost has finished by this date in nine years out of ten. This is the date to work to if you cannot protect a tender plant.
• Balanced — the median. Right about half the time, which is what an "average last frost date" actually means.
• Bold — the 10th percentile. Frost has finished by this date in only one year in ten. Worth a gamble if you have fleece and a spare packet of seed.

Three temperature thresholds are computed for each: light frost at 0°C (32°F), moderate at -2.2°C (28°F) and hard at -4.4°C (24.1°F). The planting calendar uses light frost, because that is the one that kills a tomato.

The planting windows

Crop timing is stored the way a seed packet states it: in weeks before or after the last spring frost. Resolving a window is then a matter of adding that offset to the frost date computed for your location, and clipping the result against the far end of the season so that a crop is not sown too late to ripen. Harvest estimates add days-to-maturity to the planting date, counted from sowing or from planting out depending on the crop.

Where a crop cannot fit the season at all, the page says so rather than printing a date that would waste a packet of seed.

Hardiness zones

The USDA zone is computed the way the USDA computes it: take the coldest night of each year, average those across the record, and place the result in a 5 °F band. Outside the United States there is no official zone, so the figure is labelled "equivalent" — it is directly comparable with a nursery rating on the same scale, and nothing more.

What this cannot tell you

These are statistics, not forecasts. They describe what happened across 30 years at a grid cell; they say nothing about what this particular spring will do. A single warm March does not move them and a single late frost does not invalidate them.

The grid resolves roughly half a degree of latitude, so the figures describe your district rather than your garden. Cold air pools in hollows and drains off slopes; a walled town garden can run a fortnight ahead of the open fields a mile away. Treat every date here as the baseline your own site varies from, and if you keep your own records, trust those instead.

Frost is also only one limit. Heat, drought, day length and soil will decide plenty of outcomes that a frost date has no opinion about.

Coverage

1,801 locations are currently computed. Every one of them was processed individually — no page on this site is interpolated from a neighbouring town.