Are Frost Dates Changing? 184 Cities, 1981-2010 vs 2011-2025

Last updated: 21 October 2026. Data: frost seasons 1981-2025, daily temperatures to 31 July 2026.

This page compares frost in two periods: 1981-2010, the 30-year period behind Leaftide’s city frost date tables, and 2011-2025. It covers 219 cities in 21 countries, of which 184 get frost often enough to measure a change. Every number is a change in days, with a 95% interval that shows how sure we can be of it. Where the interval includes zero, the change is too small to tell apart from ordinary year-to-year swings, and this page says so.

For your own frost dates, use the last and first frost dates by city or the Frost Date Finder. The last frost date guide explains how to plan sowing around them.

How much have frost dates changed since 1981-2010?

The table shows the median change across the cities in each region, 2011-2025 minus 1981-2010, with the 95% interval in brackets. Bold values have an interval that excludes zero. A negative frost-day value means fewer frost days. A positive first-frost value means a later first autumn frost. A negative last-frost value means an earlier last spring frost. Source: Leaftide analysis of Copernicus AgERA5 v2.0.

† The Canadian first autumn frost interval excludes zero only narrowly and touches zero when the cities are grouped in larger regions for the bootstrap (see the methodology). It rests on 12 cities in 10 groups, and a bootstrap with that few groups tends to give intervals that are too narrow. We do not count it as a finding.

The median city had 8 fewer frost days per winter in 2011-2025 than in 1981-2010 (184 cities, interval 3 to 12.5 fewer). The drop was larger in Europe, at 13 fewer frost days per winter (79 cities, interval 4 to 19 fewer), and smaller in the US, at 7 fewer (84 cities, interval 1 to 12.8 fewer).

In the median city, the first autumn frost came 5 days later (184 cities, interval 2 to 7.5 days); in the median European city it came 7 days later (79 cities, interval 2 to 12 days). The frost-free season grew by about 6 days in the median city (interval 2.5 to 11 days) and by 10.5 days in the median European city (interval 3 to 19 days).

The last spring frost showed no measurable change: its interval includes zero for all cities together and in every region. The Southern Hemisphere cities showed no measurable change in any measure. Only 9 of the 20 cities there get frost often enough to include, so their intervals are wide.

Is the last frost date getting earlier?

Not by a measurable amount when the two periods are compared. In the median city across 184, the last spring frost in 2011-2025 came 2 days earlier than in 1981-2010 (Copernicus AgERA5, Leaftide analysis). The 95% interval runs from 5 days earlier to half a day later. It includes zero, so the change is within normal variation, but the estimate leans earlier: with larger city groups in the bootstrap (see the methodology), the interval runs from 5 days earlier to exactly zero.

The estimated change was earlier in 117 of the 184 cities, later in 52 and zero in 15, and in all but 17 cities it is within normal variation. The share of 64% with an earlier estimate has a 95% interval of 44% to 75%, which includes the 50% you would expect if nothing had changed.

Seventeen cities have a last-frost interval that excludes zero: 14 earlier and 3 later. With 184 cities tested, about 9 would show that by chance alone. None of the 17 stays significant after correcting for the number of cities tested (see the methodology).

A trend test points earlier in some cities. For each city we also fitted a trend to the last spring frost over 1981-2025 (a Theil-Sen slope with a Mann-Kendall test). In 30 of the 184 cities the trend toward an earlier last frost is statistically significant (p below 0.05), and in none is a trend toward a later one. By chance alone, about 5 per direction would be expected. After the same correction for testing 184 cities, 1 city remains. The trend test treats each city as independent evidence, while neighbouring cities share the same weather, so the count of 30 overstates the evidence. It suggests the last frost may be moving earlier in some places, by too little for the comparison of the two periods to measure.

Should I plant tomatoes earlier because of climate change?

No. Frost-tender crops such as tomatoes, peppers, squash and beans are killed by a light frost, and basil is damaged even above freezing (see frost damage temperatures). When they can go out depends on the last spring frost, and that date showed no measurable change between 1981-2010 and 2011-2025 across 184 cities (Copernicus AgERA5, Leaftide analysis).

Trend tests point to an earlier last frost in some cities, but the comparison of the two periods shows no measurable change. Plant out by your own local frost risk and the forecast, not on the assumption that spring has moved.

The first autumn frost did move. In the median city it came 5 days later in 2011-2025 than in 1981-2010 (184 cities), and in the median European city 7 days later (79 cities). A later first frost can leave late crops a few more days before frost ends them. It is a median across many cities and many years, though, and any single autumn can still bring an early frost.

Are there fewer frost days than there used to be?

Yes. A frost day here is a day whose minimum air temperature at 2 metres (about 6.5 feet) above the ground is at or below 0 °C (32 °F). The median city had 8 fewer frost days per winter in 2011-2025 than in 1981-2010 (184 cities, 95% interval 3 to 12.5 fewer, Copernicus AgERA5).

Frost days changed more clearly than any other measure. In 94 of the 184 cities, the frost-day interval excludes zero, and in every one of them frost days fell. In 65 cities the drop stays significant after correcting for the number of cities tested. No city had a significant rise.

Frost days and frost date changes by city

These tables give each city’s change, 2011-2025 minus 1981-2010, in days, with the 95% interval in brackets. Source: Leaftide analysis of Copernicus AgERA5 v2.0, at each city’s coordinates.

• Frost days per winter: a negative value means fewer frost days.
• First autumn frost: a positive value means a later first frost.
• Last spring frost: a negative value means an earlier last frost.
• within normal variation: the interval includes zero, so the change cannot be told apart from ordinary year-to-year swings.
• *: the change stays significant after correcting for testing 184 cities (Benjamini-Hochberg, 5% false discovery rate).
• ‡: the interval excludes zero, but the change does not stay significant after that correction. With 184 cities, about 9 per column would show this by chance, so treat a ‡ value as a weak signal, not a finding.

A single city’s dates swing a lot from year to year, and frost at a city centre depends on the exact grid point. The tables show your region’s pattern; they are not a forecast for your garden.

Frost changes in US cities

84 US cities. Source: Copernicus AgERA5, 1981-2010 vs 2011-2025.

Frost changes in Canadian cities

12 Canadian cities. Source: Copernicus AgERA5, 1981-2010 vs 2011-2025.

Frost changes in European cities

79 European cities. Source: Copernicus AgERA5, 1981-2010 vs 2011-2025.

Frost changes in Australian, New Zealand and South African cities

9 cities in the Southern Hemisphere, where each frost season runs from February to January. Source: Copernicus AgERA5, 1981-2010 vs 2011-2025.

Which cities were left out?

35 of the 219 cities had no frost in more than 20% of their winters from 1981 to 2025 at their AgERA5 grid point. With that many frost-free years, a last or first frost date has no stable meaning, so these cities get no numbers on this page. They are:

• US (11): Phoenix, Los Angeles, San Diego, San Jose, San Francisco, Miami, Orlando, Tampa, Jacksonville, Honolulu, New Orleans.
• Canada (1): Victoria.
• Europe (12): Palermo, Genoa, Bari, Catania, Venice, Nice, Barcelona, Valencia, Seville, Málaga, Lisbon, Porto.
• Australia, New Zealand and South Africa (11): Sydney, Melbourne, Brisbane, Perth, Adelaide, Darwin, Toowoomba, Auckland, Wellington, Dunedin, Cape Town.

For some of these, the result depends on the grid point. A second dataset (ERA5, see the methodology) puts Dunedin’s share of frost-free winters at 0% instead of 78%, and Toowoomba’s at 22% instead of 100%.

How this analysis was done

Data. Daily minimum air temperature at 2 metres from AgERA5 version 2.0, the Copernicus Climate Change Service (C3S) agrometeorological dataset (doi:10.24381/f9c1f09d, doi:10.24381/cds.6c68c9bb), published under CC BY 4.0. AgERA5 is the ERA5 reanalysis, aggregated to local-time days and corrected to a 0.1° grid (about 11 km). We used version 2.0, the current version. We took the series at the grid point nearest each city’s coordinates (from GeoNames), from 1979 to 31 July 2026.

Frost. A frost day is a day with a minimum temperature at or below 0 °C (32 °F). This is air frost: ground frost can form on nights when the air stays above 0 °C.

Seasons. Each frost season runs from mid-summer to mid-summer, so a whole winter sits in one season: 1 August to 31 July in the Northern Hemisphere, 1 February to 31 January in the Southern Hemisphere. The last spring frost is the last frost day before mid-summer, and the first autumn frost is the first frost day after it. The frost-free season runs from the last spring frost to the next first autumn frost.

Periods. The baseline is 1981-2010 (30 years) and the recent period is 2011-2025 (15 years). Last and first frost dates use the springs and autumns of those years. Frost days count whole seasons by the year they start, so the baseline covers the winters from 1981-82 to 2010-11 and the recent period those from 2011-12 to 2025-26. Counting the winters by the year they end instead gives 7 fewer frost days per winter across all cities, with an interval of 2 to 12 fewer, instead of 8 (3 to 12.5). In the Southern Hemisphere the frost-free season of 2025 ends with the first frost of 2026, which is outside the data, so the recent period there has 14 seasons.

Change per city. For each city and measure, the change is the Hodges-Lehmann estimate: the median of all the differences between a recent year and a baseline year. It is less sensitive to single extreme years than a plain difference of averages. The 95% interval comes from 2,000 bootstrap resamples, each redrawing the 30 baseline years and the 15 recent years with replacement.

Change across cities. The regional figures are the median of the cities’ changes. Their intervals come from a bootstrap that resamples whole years and groups of cities together:

• Whole years. Each resample draws one set of years and uses it for every city. A cold spring in one year hits hundreds of cities at once, so cities are not independent. Drawing a whole year for all cities keeps that link. If each city’s years were drawn separately, the intervals would be 2 to 4 times narrower, and the last spring frost would wrongly look like a measurable change.
• Groups of cities. Cities are grouped in boxes of 5° latitude by 5° longitude, and each resample draws whole boxes. This keeps near-identical neighbours, such as Amsterdam, Rotterdam, The Hague and Utrecht, from counting as four separate pieces of evidence. All 184 cities fall into 67 boxes.

With larger 10° boxes, the intervals for all cities together moved by at most about 1 day, and the Canadian first autumn frost interval touched zero, which is why we do not count it. We also checked how much one year’s frost dates predict the next. The year-to-year correlation was small, between −0.26 and +0.20, so drawing years independently is reasonable.

Testing many cities at once. A 95% interval excludes zero by chance about 1 time in 20, even when nothing has changed. With 184 cities, that is about 9 cities per measure. To allow for this, we tested each city with a Mann-Whitney test (the 15 recent years against the 30 baseline years) and applied the Benjamini-Hochberg correction to those p-values, separately for each measure. That keeps the share of false findings among the cities flagged as significant at 5% on average. Only cities that pass it are marked *: 65 for frost days and none for either frost date.

Trend test. For the last spring frost we also fitted a trend to each city’s 45 years, 1981-2025, with a Theil-Sen slope and a Mann-Kendall test. Each city is tested on its own, so the count of significant cities does not allow for neighbouring cities sharing the same weather. The regional figures on this page all come from the comparison of the two periods.

Cities left out. A city is left out when more than 20% of its winters from 1981 to 2025 had no frost at all. That removed 35 of the 219 cities, listed above. The rule is applied once per city, so a city is either in all four measures or in none. In the cities that stay in, a winter with no frost (at most 9 of the 45) counts as the earliest possible last frost and the latest possible first frost: the day before its season starts and the day after it ends. This is why a few cities have very wide intervals. Leaving out every city with any frost-free winter, which keeps 162 cities, gives nearly the same results: a last spring frost 2 days earlier (4 earlier to 1 later), a first autumn frost 4.2 days later (2 to 7) and 8.8 fewer frost days per winter (3 to 14).

Check against a second dataset. We repeated the calculation with ERA5 at its original 0.25° resolution (doi:10.24381/1cf1ad76) for 30 cities spread across the four regions. For the 23 cities included in both, the medians agreed within 1.5 days. ERA5 gave 7 fewer frost days per winter where AgERA5 gave 7.5 fewer, a first autumn frost 4 days later where AgERA5 gave 3, and a last spring frost 2 days earlier where AgERA5 gave 3. The two datasets ranked the cities’ frost-day changes almost identically (Spearman correlation 0.96), but agreed much less on individual cities’ date changes (0.46 for the last spring frost). The date changes are therefore more reliable for a region than for a single city.

Weather stations. In an earlier check of five cities, the station records at Chicago O’Hare and Toronto Pearson airports warmed faster from 1981 to 2025 than the reanalysis at the same spot. Most nearby stations did not show this, so the airport records may be the outliers. Either way, where a station has warmed faster than the reanalysis, this page may understate the local change.

Why these numbers differ from the city frost date tables

This page and the last and first frost dates by city use different datasets, so their numbers cannot be compared date for date.

• This page uses AgERA5 daily minimum temperatures at 0.1° resolution and counts a minimum at or below 0 °C as frost. It reports only changes in days, never dates.
• The city tables and the Frost Date Finder use the C3S frost-day grid for 1981-2010 (WFDEI-based, 0.5° cells, 10-day steps) and count a minimum below 0 °C as frost.

For the same city the two can differ by weeks. In an earlier check of Chicago, the AgERA5 point at the city centre and the grid cell put the typical 1981-2010 first autumn frost more than 3 weeks apart. Use the city tables for your dates, and this page for how much they have moved.

Sources

• Copernicus Climate Change Service (C3S) (2026): Agrometeorological indicators time-series from 1979 to present derived from reanalysis (AgERA5 v2.0). Climate Data Store, doi:10.24381/f9c1f09d, accessed 21 October 2026. Licence: CC BY 4.0.
• Boogaard, H., Schubert, J., De Wit, A., Lazebnik, J., Hutjes, R. and Van der Grijn, G. (2020): Agrometeorological indicators from 1979 to present derived from reanalysis. Copernicus Climate Change Service (C3S) Climate Data Store, doi:10.24381/cds.6c68c9bb, accessed 21 October 2026.
• Copernicus Climate Change Service (C3S) (2025): ERA5 hourly time-series data on single levels from 1940 to present. Climate Data Store, doi:10.24381/1cf1ad76, accessed 21 October 2026. Licence: CC BY 4.0. Used for the cross-check only.
• GeoNames: cities5000 extract, downloaded 21 October 2026. geonames.org. Licence: CC BY 4.0.
• Benjamini, Y. and Hochberg, Y. (1995): “Controlling the false discovery rate: a practical and powerful approach to multiple testing.” Journal of the Royal Statistical Society, Series B 57(1), 289-300.

Contains modified Copernicus Climate Change Service information 2026. Neither the European Commission nor ECMWF is responsible for any use that may be made of the Copernicus information or data it contains.

Cite this data

Leaftide’s tables and the article text are published under Creative Commons Attribution 4.0 (CC BY 4.0). The underlying temperature data is provided by the Copernicus Climate Change Service under CC BY 4.0.

How to cite: Leaftide (2026). “Are Frost Dates Changing? 184 Cities, 1981-2010 vs 2011-2025.” Retrieved from https://leaftide.com/learn/are-frost-dates-changing/. Computed from Copernicus Climate Change Service AgERA5 v2.0 daily minimum temperature, doi:10.24381/f9c1f09d and doi:10.24381/cds.6c68c9bb.

Last updated: 21 October 2026.

To look up the frost dates for your own garden, use the Frost Date Finder. For what frost does to plants, see frost damage temperatures, and for planning around your dates, the last frost date guide.