Eight Datasets, One Number: How the World Settles on the Global Temperature

by Arjun Mehta · August 27, 2026

Science & Technology Writer, The Globe Ledger

Once a year a single number is issued for the temperature of the planet, and for a few days it is treated as though someone read it off an instrument. Nobody did. The figure is a consolidation of separate scientific products built by separate institutions from overlapping but non-identical observations, and understanding how it is assembled is the difference between reading climate reporting and being at the mercy of it.

For 2025, the World Meteorological Organization put the global mean near-surface temperature at 1.44 C above the 1850-1900 average, with an uncertainty of plus or minus 0.13 C, in the consolidated statement confirming the annual figure. In the 176-year instrumental record, which begins in 1850, two of the underlying datasets ranked 2025 as the second-warmest year and six ranked it third. All eight agree that the eleven years from 2015 to 2025 are the warmest eleven in the record, and that 2023 to 2025 are the warmest three.

The eight records

The WMO consolidation for 2025 drew on eight datasets: ERA5, the reanalysis produced through the European Centre for Medium-Range Weather Forecasts and Copernicus; JRA-3Q from the Japan Meteorological Agency; GISTEMP v4 from NASA; NOAAGlobalTemp v6 from the US National Oceanic and Atmospheric Administration; HadCRUT.5.1.0.0 from the UK Met Office and the University of East Anglia; Berkeley Earth; DCENT; and CMST, the China Merged Surface Temperature dataset.

They are not eight copies of the same thing. Reanalyses such as ERA5 and JRA-3Q run a weather forecasting model over the historical observation stream to produce a physically consistent gridded field everywhere on Earth. The station-and-ship-based products interpolate from measurements taken where measurements exist, and differ in how they treat the enormous areas - polar regions, stretches of open ocean, parts of Africa in the nineteenth century - where they do not. They also differ in how they correct for the long history of changes in how sea surface temperature was measured, from canvas buckets to engine intakes to drifting buoys. That is where most of the spread between them comes from, and it is why the plus or minus 0.13 C is not decorative.

Provisional is not final

The consolidated annual figure also arrives later than the first one does. In its provisional State of the Climate 2025 Update, released for COP30 in November 2025, the WMO reported a January-to-August 2025 anomaly of 1.42 C plus or minus 0.12 C, drawn at that stage from six international datasets and explicitly described as preliminary. The confirmed full-year figure, from eight datasets, came afterwards at 1.44 C plus or minus 0.13 C. Both numbers were correct for what they were. Reports that quote one without the qualifier - partial year, fewer datasets, provisional - generate an apparent revision where there is only a sequence. The edition matters even after confirmation: the fuller State of the Global Climate 2025 report, published on 23 March 2026, gives the year as about 1.43 C plus or minus 0.13 C. The difference between 1.43 and 1.44 is far inside the stated uncertainty, and is itself a product of the consolidation the two documents describe.

Every indicator has its own baseline

The second common confusion is arithmetic rather than scientific. Global surface temperature anomalies are expressed against 1850-1900, the closest usable stand-in for pre-industrial conditions. Other indicators in the same reports use different reference periods: in the same confirmation statement, the WMO gave the 2025 global annual mean sea surface temperature as 0.49 C above the 1981-2010 baseline. That is a much warmer baseline than 1850-1900, so the two numbers cannot be compared, added or ranked against each other. A figure quoted without its baseline is not a small omission; it is an unusable figure.

What the Paris 1.5 C actually refers to

The most consequential distinction is one the WMO states explicitly in its Global Annual to Decadal Climate Update 2025-2029, published in May 2025. The update notes that "the 1.5 C level specified in the Paris Agreement refers to long-term level of warming inferred from global temperatures, typically over 20 years." A single calendar year above 1.5 C is therefore not, on its own, the breach the treaty is written around.

In the same update, the WMO gave an estimate of the current 20-year warming level - combining recent observations with the decadal forecasts the update contains - as 1.44 C above the 1850-1900 average, with a 90 per cent confidence range of 1.22 to 1.54 C. Because that estimate is built partly from forecasts, it is not an observation. Its coincidence with the 2025 annual value is a coincidence; the two are different quantities computed different ways.

That update also contains predictions, and they should be read as predictions rather than outcomes. It gave an 86 per cent probability that global mean near-surface temperature would exceed 1.5 C above the 1850-1900 average for at least one year between 2025 and 2029, and a 70 per cent probability that the five-year mean for that period would exceed it, with annual values predicted to fall between 1.2 C and 1.9 C. The predictions come from an ensemble of 220 members contributed by 14 institutes, including four WMO Global Producing Centres: the Barcelona Supercomputing Centre, the Canadian Centre for Climate Modelling and Analysis, Deutscher Wetterdienst and the Met Office.

The indicators that move more slowly

Surface temperature is the noisiest indicator in the set, because it responds within months to El Nino and La Nina. The slower measurements carry more signal about the underlying system, and they come from a different observing enterprise entirely.

The WMO Greenhouse Gas Bulletin No. 21, published in October 2025 and reporting data for 2024, gave global mean surface mole fractions for 2024 of 423.9 plus or minus 0.2 parts per million for carbon dioxide, 1942 plus or minus 2 parts per billion for methane and 338.0 plus or minus 0.1 parts per billion for nitrous oxide. Relative to 1750 those stand at 152, 266 and 125 per cent respectively. The 2023-to-2024 rise in carbon dioxide, 3.5 ppm, was described in the bulletin as the largest one-year increase in the modern measurement record, exceeding the previous record of 3.3 ppm from 2015 to 2016. The same bulletin reports the NOAA Annual Greenhouse Gas Index at 1.54 for 2024, a 54 per cent increase in total radiative forcing from 1990 to 2024. Those figures rest on a physical network: the bulletin's 2024 analysis used 179 stations for carbon dioxide, 171 for methane and 123 for nitrous oxide, with 56 WMO Members contributing data.

The ocean and cryosphere indicators move slower still. The WMO's confirmation statement reported that heat content of the upper 2,000 metres of the ocean rose by roughly 23 plus or minus 8 zettajoules between 2024 and 2025; the State of the Global Climate 2025 report records ocean heat content at its highest level since records began in 1960, above the previous high set in 2024. That report puts global mean sea level around 11 cm higher than in 1993, rising since 2012 at a faster rate than in the earlier part of the satellite record. Glacier mass loss from the reference glaciers in the 2024/2025 hydrological year was among the five worst on record. Arctic sea ice reached an annual maximum of about 14.19 million square kilometres, the lowest annual maximum in the satellite record that begins in 1979, with the annual average the lowest or second lowest of that era; Antarctic sea ice extent was the third lowest, after 2023 and 2024.

The State of the Global Climate 2025 added a further indicator for the first time: Earth's energy imbalance, the difference between incoming solar radiation and outgoing energy, which is the quantity the whole system is actually responding to. It is a telling addition. The annual headline temperature is the most quoted output of this machinery and close to the least informative part of it - a single noisy year, drawn from eight records that do not perfectly agree, measured against a baseline that has to be reconstructed, and answering a question the Paris Agreement did not ask.