Seven downscaled climate models, 16 cities, the twenty years to 2010 against the twenty to 2050. Every model warms every city — the coolest projection anywhere in the set is still +0.03°C. What they disagree about is the amount, by a median 0.51°C, which is 43% of the warming itself.
Across 16 cities and 7 models there is no disagreement about direction at all: 16 of 16 cities are warmed by every single model, and the most conservative projection anywhere in the set is +0.03°C. The disagreement is entirely about how much.
This has to come first, because everything after it is about disagreement and would read differently otherwise. Across 111 city-model pairs, not one projects cooling.
Not most models, and not on average. Every model, every city.
The single coolest of all 111 city-model pairs. The bottom of this ensemble is still warming.
Ensemble mean across the 16 cities, 1991–2010 against 2031–2050.
A projection is usually quoted as one number. Underneath it are 7 numbers that disagree, and the width of that disagreement is a median 0.51°C.
Between the warmest and coolest model for the same city. As a share of the warming itself: 43%.
The widest in the set, around a mean of 0.72°C. Which number you quote depends on which model you opened.
Places where the models differ from each other by more than the warming they are projecting. Sydney is the opposite case, tightest at 0.28°C.
This site already has a Philippine climate page. It compared two reconstructions of the observed past and found them 1.09°C apart on Manila’s annual mean. That was history. This is the future.
From 26.82°C to 27.72°C on the ensemble mean.
A 0.93-degree spread on one city. The observed-past disagreement was 1.09°C, so this is not a new problem appearing in projections.
Per year, baseline against future, ensemble mean. The models range from 7.2 to 26.6 for that future figure.
The far north is supposed to warm fastest, and across these 16 cities latitude and warming correlate at only 0.4. The disagreement between models is larger than the geographical signal it is supposed to reveal.
Across all 16 cities. A relationship exists and it is weak enough that the ordering is not what a map would predict.
At 40.7 degrees north — warming more than Reykjavik, which sits at 64.2 degrees and warms 1.12°C.
Tied with Dhaka on exactly that figure — and also the city with the widest disagreement in the set, which is the next section and is not a coincidence.
Delhi is tied with Dhaka for the lowest mean warming in this set, and has by far the widest range behind it. The models are not slightly apart; they describe two different futures.
Equal lowest of the 16 cities. Read alone it looks like the least-affected place in the set.
One model projects 0.03°C across twenty years — almost nothing, and still positive. Another projects 1.57°C.
A 1.54°C spread around a 0.72°C mean. Quoting the mean alone would be the single most misleading number on this page.
The number this page reports is the range across models of a single scenario. Three things it does not measure, and one it does.
Every model here is run on the same emissions pathway. The spread is what remains after that choice is fixed: different physics, different resolution, different treatment of clouds and land. Adding scenarios would widen it further, most at longer horizons.
None of these numbers is a measurement. They are what several independent models produce when asked the same question, and the agreement between them is evidence about the models rather than about the atmosphere.
A wide spread around a robust sign is the ordinary condition of this field, and it is why 16 of 16 cities here are warmed by every member of the ensemble while none of the members agree on the amount. Reporting the second without the first would be a different and dishonest page.
A city plan built on a single downscaled projection is built on one draw from a distribution 0.51°C wide at the median. The useful output is the range.
Four limits, and the first two bound how far anything above travels.
These are the high-resolution members Open-Meteo redistributes, downscaled to a 10km grid: 7 models, not the several dozen the IPCC assesses. A different subset would give a different spread, and this one is not weighted by skill.
Points, not regions, and picked to span from Reykjavik at 64.2 degrees north to Sao Paulo at 24 south rather than sampled at random. Nothing here is a global average and no global figure is quoted. Collection is limited by a daily cap on the source and this set is what a day of it buys.
1991–2010 against 2031–2050, because the source's coverage ends at 2050. Near-term is where model choice matters most relative to scenario choice, which flatters the finding and is worth saying.
Each value is a model grid cell interpolated to a coordinate, not a thermometer in that city. Absolute temperatures carry a bias that mostly cancels in the difference between two windows, which is why this page reports warming rather than temperature.
Every figure on this page traces to one of these, through a CSV in
data/global-climate-spread/. Each is checked against its source query on every
build.
Daily mean and maximum temperature for a point from seven downscaled CMIP6 models, 1991-2050, for 30 cities — Free, no key. Coverage ends at 2050 and a request past it returns 400 rather than a truncated series, which is why the future window here is 2021-2050. The parameter is `models`; sending `model` as well is also a 400.
The model intercomparison project the underlying projections come from — Cited for provenance. The models here are the high-resolution HighResMIP members Open-Meteo redistributes, downscaled to a 10km grid; they are a subset of CMIP6 and not the full ensemble the IPCC assesses.
The observed-past counterpart: two ERA5 reanalyses of the same Philippine record disagreeing by 1.09 C on Manila's annual mean — This page is the projection-side extension of that one. The comparison is deliberate: if two reconstructions of what already happened differ by a degree, a spread between projections of what has not happened yet is the expected shape rather than a surprise.