Data and methods

The data come from ERA5-Land, the reanalysis of the Copernicus Climate Change Service: not station records, but the reconstruction of the climate by a model driven by observations. The means are over the thirty years 1996–2025, the climate normal of the World Meteorological Organization. Contains modified Copernicus Climate Change Service information 2026. Contains information from CRU TS 4.10 (Climatic Research Unit, University of East Anglia; Harris et al. 2020), which is made available under the ODbL.

The grid is ERA5-Land's own, 0.1° wide (about 10 km; 3600 × 1801 points), with no remapping: every value on the site comes from the land points of the source. Sea points are absent; means over several points are weighted by the cosine of the latitude, that is by the cell's area.

Places: Wikidata, Natural Earth, geoBoundaries, © OpenStreetMap contributors (ODbL), GeoNames (CC BY 4.0). City elevations are those of Wikidata, fixed by hand in the few wrong cases.

Hurricanes, typhoons and cyclones: IBTrACS v04r01 (Gahtan et al. 2024, doi:10.25921/82ty-9e16; Knapp et al. 2010, Bulletin of the American Meteorological Society), NOAA National Centers for Environmental Information: US government data, free to redistribute and use without restriction.

Correction with CRU TS. CRU TS 4.10 is the 0.5° grid (about 50 km) built from weather stations by the Climatic Research Unit of the University of East Anglia. Against CRU, ERA5-Land's range between day and night is too narrow (where CRU has enough stations, maxima 1.1 °C too low on average, minima 0.9 °C too high), it has 9 % more precipitation and 72 % more days with at least 0.1 mm, from the drizzle the model spreads over many days. For every month and every 0.5° box we take the difference between the two 1996–2025 means (temperatures) or their ratio (precipitation, within a factor 0.5–2: CRU's rain gauges miss part of the snow and of the rain in wind, so CRU is too dry in winter and on mountains); we make it regular over the year (mean and two harmonics); we apply it in full where CRU has at least 4 stations, less where it has fewer, not at all where it has none (nor where CRU has no data, as on small islands); we smooth it between neighbouring boxes and interpolate it to the 10 km cells. So the level of the values changes, not ERA5-Land's detail (relief, coasts) nor the differences between years: the comparison with 1961–1990 and the yearly series of "Effect of climate change" stay ERA5-Land's. The elevation correction of the cities comes after, on the corrected cells.

Monthly means cannot give maxima, minima and rain days: daily or hourly data are needed, reduced upstream to monthly series 1950–2025 (the mean of the daily maxima and minima of every month, the days with at least 1 mm of every month), then handled like temperature and precipitation.

Rain days: the days of the month with at least 1 mm of precipitation, counted on the hourly data of each cell, averaged over the years of 1996–2025 and then corrected with CRU TS. The cell is a mean: in ERA5-Land the precipitation comes from ERA5, at about 31 km, interpolated to the 9 km grid, and the model spreads drizzle over many days. CRU counts days from 0.1 mm, the site from 1 mm: ERA5-Land's days are also counted at 0.1, 0.2, 0.5 and 2 mm, and their number falls regularly with the threshold (ln N = a − b√threshold). We look for the threshold at which ERA5-Land has as many days as CRU from 0.1 mm: the difference is the extra drizzle, and the corrected days from 1 mm are those ERA5-Land counts from 1 mm plus that difference. On average the days fall by about a fifth.

Cloud cover: ERA5-Land does not have it, so it comes from ERA5 (total cloud cover, monthly means) on a 25 km grid interpolated to the 10 km cells: it has less detail than temperature, and in mountains or on coasts it is a mean over a wider area. It is the fraction of the sky covered, not sunshine hours nor radiation: thin cloud counts as much as an overcast sky.

Sea temperature: ERA5-Land has no sea, so it comes from ERA5 (sea surface temperature, monthly means, 25 km grid), averaged over 1996–2025. A city or region is «on the sea» if the coast (Natural Earth) is within 10 km; it takes the mean of the nearest sea cells, which on a 25 km grid are usually 10–30 km off the shore, and only if one lies within 60 km. Lakes do not count (ERA5 has a temperature for the Great Lakes: cells far from every sea are left out). The name is that of the smallest Natural Earth sea that holds the cell.

Hurricanes, typhoons and cyclones: IBTrACS tracks 1980–2025, all basins, the 3-hourly positions interpolated to one per hour. Only storms with a maximum sustained wind of at least 64 knots count (category 1; tropical storms do not): the 1-minute wind of the US centres, that of the official WMO agency where it is missing. For each month: the share of the 46 years in which at least one passes within 300 km of the place, each storm counted once per month, no smoothing across months. From 20 % the month is a contraindication. The name follows the basin that brings most of them: hurricanes (Atlantic, eastern and central Pacific), typhoons (western Pacific), cyclones (Indian Ocean and southern hemisphere). Only cities and tourist regions within 50 km of the sea: a storm that strong is at sea or just ashore, and for an inland place the storms within the radius stay offshore.

From the monthly series the pipeline also computes the standard deviation and the 10th and 90th percentile of each month's precipitation over the years: they are in the store but not yet on the pages. The standard deviation of the annual values serves the climate-change test.

Were maxima and minima missing from the store, the pages would show the mean ± 5 °C, with a notice: today this happens for no place.

Cities are the core of the site: a city is a point, and its values are the bilinear interpolation of the four land cells around it (renormalised when some are sea), or else the nearest cells. Tourist areas, small, are treated like cities.

Countries and regions have only an indicative mean over all their cells, weighted by area (the cosine of the latitude) times the fraction of the cell inside the border, not weighted by population: within an area the climate varies a lot (coasts, plains, mountains), which is why the three main cities can be shown too, and their pages matter more than the mean.

The "When to go?" strip of countries and regions uses this mean (no snow or aurora, which make no sense on a mean) and the mean daylight of the territory. Where the mean hides too much, the "Very varied climate" badge appears when at least one of three conditions holds: the 10th–90th percentile of the annual mean temperature across the cells is at least 8 °C wide, the area is at least 1,500,000 km², the cells span at least 20° of latitude. It also appears when the mean disagrees with the cities: a best month of the mean that none of its cities shares. The badge sits under the strip, and the sentence of a very varied area has no «Avoid»: the reasons against of the mean do not hold for much of it.

Elevation correction. The cells around a city can lie much higher or lower than the town (valleys, coasts, mountains). Where the cells' mean height and the town's elevation differ by at least 100 m, the temperatures (mean, maximum and minimum) are brought to the town's elevation with the mean lapse rate of the atmosphere, 6.5 °C per 1000 m, constant over the year; with no cap. Precipitation, rain days, countries and regions stay as in the cells; a city with no known elevation is not corrected.

ΔT = (z_cells − z_town) × 6.5 °C / 1000 m

In a narrow valley in winter, with cold air pooling on the valley floor, the correction can overestimate. In the advanced reading the correction applied is written under the temperature chart.

No type of place: the thresholds move with the height. The ideal temperature band follows the height its temperatures refer to, the town's where the correction above moved them, the cells' otherwise: higher up, cooler months count as ideal. The rain days, snow and cloud come from the cells, so the rain-day thresholds follow the cells' height, and skiing is judged on the cells' temperature, without the elevation correction.

Base: the month's mean cloud cover (ERA5 tcc) in four classes, with lower bounds 35, 55, 75 %: clear, partly cloudy, mostly cloudy, overcast. Marks: the rain days of the file, corrected with CRU TS, on the thresholds of the verdicts: 1 drop from the first day past the ideal (11 days in the lowlands), 3 from the often-rain days (15), 2 halfway (13); from 2000 m of the cells 15, 17, 19 days. So a drop never sits on an ideal month and 3 drops are an often rainy one. Below that there is no mark and the icon is the base; the same under 25 mm in the month, however many rain days. If the month has at least 30 mm and at least 15 per rain day, the marks are lightning: 1 up to 100 mm, 2 up to 200 mm, 3 above; lightning wins over drops. With a mean up to 0 °C the marks are snowflakes, by the same rules as drops, and there is no lightning. The marks are added to the base, never in its place: the drawing is the base a little smaller with a row of 1–3 marks under it; under 35 % cloud a month with marks takes the next base (partly cloudy), there is no sun with rain. 31 icons in all. Where rain days are missing they are estimated as 1 day per 10 mm, where cloud cover is missing 45 % is used.

For cities and tourist areas: the 1996–2025 normal minus the previous normal, 1961–1990, in the same ERA5-Land reanalysis. Temperature: the difference of the annual means (the mean of the months weighted by their days), in °C with one decimal. Precipitation: the difference of the annual totals, in % of the reference period. Rain days: the difference of the days with at least 1 mm in a year, in % of the reference period. Snow: the difference of the mean snow depth in the winter months (December–February north of the equator, June–August south of it), in % of the reference period, only where the reference winter had at least 2 cm; the depth is that of the ERA5-Land cell (9 km), not moved to the city's elevation.

To tell a signal from noise, Welch's two-sample t-test on the annual values of the two thirty-year periods: the difference counts when it exceeds 2 times the standard error of the difference, from the interannual standard deviation of each period (s₁, s₂); with 58 degrees of freedom this is the 5 % two-sided critical value. Otherwise the badge is grey with ≈.

Caveats. Temperature nearly always passes the test, precipitation often does not. ERA5-Land before 1979 is the preliminary back extension, without satellites: the change in rain is less reliable than the change in temperature. It is the observed change, natural variability included, not an attribution to causes. Approximations: the autocorrelation of the annual values is ignored, a trend inside a period inflates its standard deviation, the city's standard deviation is the mean of the cells'.

Yearly series of "Effect of climate change". The series 1950–2025 come from the same ERA5-Land monthly means: the mean temperature of the year (weighted by the days of the months) minus the 1961–1990 mean, and the rain of the year in % of the 1961–1990 mean (not shown where that mean is almost nil). Months: the same Welch test, month by month, on the standard deviations of the two 30-year periods (2 standard errors). Trend of the rain: the Theil-Sen slope (the median of the slopes between all pairs of years), stated only if the Mann-Kendall test tells it from chance (5 %, two-sided); it assumes no straight line and one exceptional year does not move it. Warm season: the days above the 1961–1990 annual mean on a smooth curve (harmonics) through the twelve monthly means; its start and end in the two periods are compared. Only where the warmest and the coldest month are at least 10 °C apart and the curve crosses the threshold once each way: not in the tropics. The years before 1979 carry the caveats given above.

Sunrise, sunset and day length are computed for the 15th of each month from the position of the sun (NOAA algorithm) at the city's point, sunrise and sunset being the moment the sun's upper limb touches the horizon. The times are in the legal time of the place on that day: the zone from the IANA boundaries, daylight saving from tzdata, hence standard time in the months it does not apply. In the polar night there are no times; under the midnight sun the day lasts 24 hours.

Three foreign ministries publish their advice as open data: the United Kingdom (FCDO, Open Government Licence v3.0), Canada (Global Affairs Canada, Open Government Licence – Canada) and the United States (State Department, public domain). There is no common European Union advice and no public United Nations data; Germany (Auswärtiges Amt) has an interface, but its terms of use do not allow reducing the advice to a level. The label links the UK Foreign Office page of the country (on the Italian pages, the Italian Foreign Ministry's viaggiaresicuri.it, which writes its advice as free text with no machine-readable level).

Each source gives a country a level: do not travel when it advises against all travel to the whole country (UK «avoid all travel to the whole country», Canada level 3 of 3, US level 4 of 4); travel not advised when it advises against non-essential travel to the whole country (UK «all but essential travel», Canada level 2, US level 3). The country takes the highest level that at least 2 sources give it; one source alone is not enough. Advice for some areas only gives no label: no source publishes the borders of the areas, so they cannot reach the cities.

The advice is downloaded every week; last check on 9 November 2026. A source that does not answer keeps its previous values; after 30 days without an answer it stops counting. Advice changes fast: always check before leaving.