What these numbers are

Every city has four monthly series — temperature, rain, wind and air quality — with one value per month. They're historical averages, not a forecast. They answer "what is a normal September like here?", never "will it rain on Tuesday?". If you need to know about tomorrow, check a forecast; if you're deciding which month to book, these are the right numbers.

Nobody types a climate figure in here, me included. The series are produced by an automated pipeline that reads weather station records and public climate datasets, and they only change when I run it again. That rule exists so that nobody — me least of all — can nudge a city into looking better than it is.

Where each number comes from

The sources are ranked, and the best one available for that city and that metric wins:

• A weather station (Meteostat / GHCN) within 35 km of the city and 250 m of its altitude. First choice whenever one qualifies, because it's the only source that actually measured the weather, with an instrument on the ground.
• The ERA5 and ERA5-Land reanalysis ensemble, when no station qualifies. A reanalysis is a model that reconstructs past weather from observations. ERA5-Land has the finer grid and rescues coastal cities, where a coarse cell averages sea into the land.
• MERRA-2, which never wins. It's there purely to object: when it disagrees loudly with the others, instead of publishing the city I flag it for review.
• CAMS for air quality, which I publish as a monthly AQI.

The decade I use, and why

Temperature and wind come from the last ten years rather than the published 1991–2020 normals. That's deliberate: this site answers "where do I go now", and a thirty-year normal cools every city on the map by averaging in a climate that has already passed. Expect my figures to run roughly 0.3 to 1.0 °C above a published normal. That gap is the baseline I chose, not an error.

Rain is the exception: there I do use the thirty-year normal. Temperature trends one way, so a recent decade estimates it better. Rain, though, swings with the NAO and El Niño and comes back — Florence recorded years from 596 to 1275 mm inside that same 1991–2020 window — so any ten-year rain average is ±25% out on luck alone. For rain, the long normal is the honest one.

Why rain never comes from a model

I checked my own sources against real rain gauges in every city that has both, across 2016–2025. The results decided the rules:

• Temperature from ERA5: median error 0.66 °C, and 61 of 67 cities within 1.5 °C. Good enough to publish.
• Wind from ERA5: 2.4 km/h median error. Good enough as a secondary figure.
• Rain from ERA5: the median ratio looks perfect at 0.98, but the tail runs from 0.67× to 2.85× the real gauge, and only 30 of 38 cities land within ±25%. Not good enough, however good the median looks.

The reason is physical. Convective rain — the thunderstorm that soaks a tropical afternoon — is too small for a model to see, so it approximates it. And a 25 km cell laid over a coast or a small island has most of its area over water, so it ends up reporting the ocean's rain as the city's. Chiang Mai is the clearest case: the model's default setting gave 1816 mm a year against the 1180 mm actually observed.

Two models agreeing doesn't fix it. In Cebu, ERA5 and MERRA-2 agree closely with each other and both sit 58% above the rain gauge. So I only publish rain from a station or a national normal, never from a model on its own.

Median for rain, mean for temperature

Ten years is a short sample, and rain is a lopsided one. Da Nang's ten Octobers run from 638 to 1758 mm: the mean gives 1138 mm, the median 987 mm. The mean lands on a value no October actually had, dragged there by a single wet year. So I summarise rain by its median, and temperature and wind by their mean, which is the right call when the values spread evenly either side.

On top of that, a month needs at least 7 of 10 years on record before a station may speak for it. With six values the average wobbles too much — that rule alone rejected Da Nang's rain and sent it to arbitration.

Rain days, and why most cities don't show them

Where you see rain days, they're days with at least 1 mm — the WMO definition — counted from the station's daily files and summarised as the median across years. Only 90 cities have them, and that restraint is deliberate.

The millimetres come from a thirty-year normal; the days only exist in the recent decade. If that decade didn't rain the way the normal says, the two figures contradict each other in the same sentence. London's Aprils in the decade gave about 28 mm against a normal near 44 — and "44 mm across 5 days" describes something no rain gauge ever measured. So I only show the days if that city's decade lands within ±25% of the normal. That test rejected 28 of 118 candidates, and I'd rather leave them blank than get them wrong.

Daily records have two traps I check for. A gap isn't a dry day — some stations only report when it rains, so a month needs at least 25 reporting days to count. And a gauge stuck at zero reports every day and never rains, which shows up as a month with 40 mm and no rain days.

When the sources disagree

The pipeline is allowed to refuse. When the station, the reanalysis ensemble and MERRA-2 can't be reconciled, I write nothing and the city goes to arbitration: I look for a fourth figure published by people — the national weather service, WMO records, the station's own archive — and that one decides. A city held back is the system working, not failing.

I don't relax these thresholds to cover more cities. An empty field costs you nothing; a believable but false number can cost you the trip.

What I judge by hand

Cost of living in euros, internet speed, the number of coworking spaces and each city's description can't be measured the way temperature is. Those I research by hand, with at least two independent sources per number, and then verify in a second pass against sources I didn't use the first time.

Treat them as estimates for comparing one city against another, not as a quote. A monthly cost assumes a modest one-bedroom flat and cooking most days; live differently and your number will differ.

What these numbers won't tell you

A monthly average hides the individual day. A city averaging 28 °C can still hand you a 35 °C week. A big city has microclimates my single point can't see: the coast isn't the hill behind it. Air quality shifts by the hour and by the season, so a monthly AQI is a tendency, not a promise. And none of this knows anything about a heatwave, a typhoon or an odd year.

Use them to get from a whole world of cities down to a handful, then check a forecast before you book.

Spotted a number that's wrong?

If you live somewhere on this map and a figure doesn't match what you see every day, tell me — and send a source if you have one. Corrections go through the same arbitration as everything else, and someone who lives there beats a model.

Contact