How we verify temperature data
We publish daylight figures for five hundred cities and we compute every one of them ourselves from solar geometry. We publish far fewer temperature figures, and the reason is not laziness. It is that a temperature number is only meaningful if you know which years it averages and which agency measured it, and for most cities on the internet you cannot find out either.
This page is the rule we apply before a temperature figure ships, written down so you can hold us to it.
The two questions a temperature number has to survive
Before any average temperature appears on this site, its source record has to state both of the following explicitly. Not implicitly, not on a different page, not in a footer that says “data from various sources”.
- The averaging period. Which run of years the average covers. “January average 14°C” is not a fact until you know whether that is 1961–1990, 1981–2010, 1991–2020, or the last four years scraped off a forecast API.
- The measuring agency. A named national meteorological service or an equivalent institutional record — the Met Office, AEMET, DWD, the Thai Meteorological Department, NOAA's National Centers for Environmental Information, the Hong Kong Observatory. Not an aggregator that republishes numbers without saying where it got them.
If either is missing, the figure does not ship. We do not estimate it, we do not average two sources that disagree, and we do not fill the gap from a nearby city.
Why we only use 1991–2020 normals
The World Meteorological Organization defines a climatological standard normal as a thirty-year average, and it recommends that the reference period be updated every decade rather than every thirty years, precisely because the climate is moving fast enough that an old baseline misleads. The current standard period is 1991–2020.
This matters more than it sounds. If one city's average comes from a 1961–1990 record and another's comes from 1991–2020, the older one will read colder — not because the place is colder, but because the baseline is older. Put those two numbers in the same table and you have manufactured a difference that does not exist. So we do not mix periods. Every temperature comparison on this site is 1991–2020 against 1991–2020, or it is not published.
What we compute instead, and why it is safe
Daylight is a different kind of quantity from temperature. Sunrise, sunset, solar noon and day length are not measurements at all — they are consequences of latitude, longitude, the date, and the tilt of the Earth's axis. Given those, the answer is arithmetic, and the same arithmetic an almanac uses.
So we implement the standard solar position algorithm directly: solar declination and the equation of time for the date, then sunrise and sunset at the moment the sun's centre sits 0.833 degrees below the horizon. That 0.833 is the conventional correction for atmospheric refraction plus the sun's own angular radius, and it is why day length is a few minutes longer than a naive geometric calculation gives. Local clock times come from the IANA time zone database with daylight saving applied for the specific date, not for today.
The result is reproducible. If you take the latitude and longitude printed on any city page, put them into any competent solar calculator with the same date, you will get our numbers back to within a minute. That is the standard we hold daylight to, and it is the standard temperature cannot meet without a sourced record behind it.
What this rule does not fix
A properly sourced 1991–2020 normal is still an average of thirty years at one weather station. It tells you nothing about variance, and variance is often what actually ruins a month. A city with a January mean of 16°C and a January mean of 16°C can be a place where every day is 16°C or a place that swings between 8°C and 24°C, and those are different places to live in. Where we have the figures, we prefer to show daily highs and lows rather than a single mean for exactly this reason.
It also says nothing about rain, humidity, wind or air quality, which between them decide most of what people actually mean when they ask whether a month is any good. Two of those get their own pages: what the northeast monsoon actually does to Malaysia, and how bad the Chiang Mai burning season really is.
And it says nothing about how our own annual weather rating behaves, which is a separate and larger trap — see why annual weather scores mislead.
Common follow-up questions
Why does timedatelive show daylight for 500 cities but temperature for far fewer?
Because daylight is computed from latitude, longitude and date using standard solar geometry, so it is reproducible for any city whose coordinates we have. Temperature has to come from a measured record, and a measured record only counts if it states both its averaging period and the agency that made it. Most published city temperature figures state neither.
Why only 1991–2020 climate normals?
Because the WMO standard reference period is 1991–2020 and it recommends updating every decade. Mixing a 1961–1990 average with a 1991–2020 average in the same comparison makes the older city look colder than it is, which is an artefact of the baseline rather than a real difference in climate.
What is the 0.833 degree correction in your sunrise times?
It is the standard allowance for atmospheric refraction bending sunlight over the horizon, combined with the fact that sunrise is measured from the sun's upper edge rather than its centre. It makes the day a few minutes longer than pure geometry would predict, and every almanac applies it.