October set a heat record. The 2026 forecast answers a different question.

October 2026 was the warmest October in the ERA5 record, according to Copernicus’s November update. Global surface air temperature averaged 15.35°C: 0.90°C above the 1991–2020 October average and an estimated 1.75°C above the pre-industrial October baseline.

The same update says 2026 is virtually certain to finish above 1.5°C. That is a forecast for a calendar-year average, with November and December still to come. October’s result, the annual outlook and the Paris Agreement’s long-term warming level answer different questions.

The newsflash also expects the four-year average for 2023–2026 to exceed 1.5°C above pre-industrial levels. Sustained warmth across several years matters, but a four-year average is still a shorter window than a long-term climate assessment.

What remains to be predicted

The monthly bulletin gives a greater than 99% probability that the full-year 2026 anomaly will exceed 1.5°C. Its 95% prediction interval is 1.56–1.61°C above the estimated 1850–1900 average. Those endpoints describe a forecast range, not two observed annual results.

For the remaining months, the statistical model uses linear regression trained on ERA5 data from 1979–2025. Recent land-air and extra-polar sea-surface temperature anomalies inform the November–December prediction. Copernicus puts the chance of 2026 ranking second-warmest at about 80%, and the chance of tying or exceeding 2024 at about 20%.

A very high probability of clearing one threshold can coexist with uncertainty about the final ranking. Neither probability turns the last two months into measurements. The bulletin is also one dataset-based assessment, not several independent estimates merely because it appears on multiple Copernicus pages.

Editorial visualization

Why the reference period changes the number

Copernicus’s methods use 1991–2020 as the standard baseline for recent climate. An anomaly compares a period with the corresponding period in that reference—October with past Octobers, for example. The estimated 1850–1900 baseline instead provides a pre-industrial comparison.

That is why 0.90°C and 1.75°C can describe the same October without contradicting each other. Changing the reference changes the zero point, not the month’s weather. The global average also does not imply that every location warmed by the same amount.

ERA5 does not extend into 1850–1900. The conversion methodology therefore relates its modern reference to older datasets. Annual global anomalies use a 0.88°C offset; monthly offsets vary with the season. Applying the annual offset mechanically to a single month would give the wrong comparison.

A consistent record is constructed, not read from one thermometer

ECMWF’s reanalysis explanation describes a blend of historical observations and model information. Observations are unevenly distributed and contain errors; data assimilation helps produce a physically consistent global record. Using the same modelling system through the reanalysis period reduces artificial changes caused by switching systems.

This is why the record can cover places without a dense network of instruments. It also retains uncertainty: ECMWF notes that periods and areas with sparse observations tend to be less certain. Reanalysis is more than a collection of thermometer readings, and less than a perfect measurement everywhere.

Editorial visualization

For the longer horizon, the World Meteorological Organization explains that Paris temperature levels concern sustained warming, typically assessed over 20 years. Its May 2026 outlook is background context, not the source of this November forecast. An individual year above 1.5°C does not itself establish that the long-term level has been crossed.

The next annual assessment will replace the remaining forecast months with data. A claim about the long-term threshold will still need a longer view, even after the calendar-year result is known.