The following is a summary of global temperature conditions in Berkeley Earth’s analysis of October 2025.
- October 2025 was the third warmest October on record behind 2023 and 2024, with a global average of 1.52 ± 0.08 °C (2.73 ± 0.14 °F) above the 1850-1900 average.
- Land temperatures averaged 2.24 ± 0.15 °C (4.04 ± 0.26 °F) above the 1850-1900 average, ranking as the 3rd warmest October for terrestrial regions.
- Ocean temperatures were 0.95 ± 0.07 °C (1.71 ± 0.13 °F) above the 1850-1900 average, making it the fourth warmest October for ocean surfaces.
- We have entered a La Niña state. However this is likely to be short lived, with a 61% probability to transition back to neutral conditions January-March.
- 4% of Earth’s surface experienced a record high October monthly average, with equal contributions from land and ocean surfaces.
- 2025 now has a 94% likelihood to be the third warmest year on record and almost no chance (0.02%) to have an annual average above 1.5 °C.
Global Summary
October 2025 was the third warmest October on record, with a monthly global average of 1.52 ± 0.08 °C (2.73 ± 0.14 °F) above the 1850-1900 average. Along with 2023 and 2034, the last three Octobers have been sharply warmer than all measured Octobers 1850 to 2022. If not for the extreme warmth in 2023 and 2024, 2025 would have been considered a sharp step up from the previous Octobers.
Month-to-month temperatures continue to tick upward markedly in October 2025, bouncing off the recent lows and continuing to rise above the long-term trend line. However, conditions remain cooler than during the bulk of 2023 and 2024.
October 2025 was cooler than 2023 and 2024, but still significantly warmer than any October prior to 2023. October temperatures were slightly above 1.5 °C.
The 12-month moving average is approaching the long term trend line. The short-term natural variability contributing to anomalous warmth in 2023 & 2024 has peaked, and we expect the 12-month average will soon return to the trend line. Short-term variability is an inherent part of the weather and climate and is to be expected even alongside persistent long-term global warming.
Spatial Variation
In October 2025, elevated temperatures were prevalent across the Arctic, North America, Southern Africa, Australia, Antarctica, and Southern Asia. The North Pacific remains anomalously warm. Cooler temperatures were present in Central Russia, the central Pacific (as we enter La Niña), and the Hudson Bay.
4 % of Earth’s surface experienced a record high October monthly average (4% of both land and ocean surfaces), with record warmth in parts of the North Pacific Ocean, the Arctic Ocean, North East Canada, Southern and West Africa, South East Asia, and Indonesia/Papua New Guinea.
In Berkeley Earth’s estimation, 21 countries and Antarctica set record warm national averages in October, including: Barbados, Botswana, Brunei, Fiji, Guinea, Indonesia, Ivory Coast, Jamaica, Lesotho, Liberia, Malaysia, Mexico, Papua New Guinea, Philippines, Sierra Leone, Singapore, South Africa, South Sudan, Taiwan, Tonga, and Vanuatu
Over land regions, October 2025 was the third warmest October since 1850, with a terrestrial average of 2.24 ± 0.15 °C (4.04 ± 0.26 °F) above the 1850 to 1900 average.
Average ocean temperatures in October 2025 were equal to 0.95 ± 0.07 °C (1.71 ± 0.13 °F) above the 1850 to 1900 average. This was the fourth warmest October on record (again behind 2023 and 2024).
Continuing NOAA Data Problems
The current monthly report is prepared using data available to Berkeley Earth as of mid-November 2025. However, we are working around problems with NOAA input data. Since late August 2025, core NOAA climate archives, including the Global Summary of the Day, Global Historical Climatology Network – Daily, and Global Historical Climatology Network – Monthly (GHCN-M), have been broken or incomplete. Because Berkeley Earth draws from multiple data sources, we are resistant to these problems but not entirely immune to them.
While we ordinarily expect to have weather data from ~20,000 weather stations per month in the modern period, the disruption to NOAA services has cut that number by half in August and September. Due to additional standard delays in reporting, the October data set is reduced ever further to ~7,500 stations. This number of weather stations is still more than adequate for constraining the global average, but regional gaps have left unusual holes on our maps.
In the case of GHCN-M, which is NOAA’s flagship climate monitoring platform, the defects are strongly clustered with NOAA failing to report many observations from China, Kazakhstan, the Philippines, and certain other areas. These weather stations continue to be operated by their local meteorological agencies, but their observations have not been captured by NOAA aggregation services.
Drawing on alternative data sources allows us to fill some of the gaps created by missing NOAA data aggregation services, but we still have less data than we would ordinarily expect. As a result, the maps have unusual gaps in Africa.
It is normal to encounter sporadic delays in weather station reporting, and as a result, it is also normal to revise the analysis as more data becomes available. However, the current disruption resulting from the loss of NOAA services is larger than the typical delays and more spatially clustered. We expect the problems with NOAA data aggregation will ultimately be temporary, as the underlying weather stations still exist, and it is simply a matter of getting suitable access. Future revisions, Incorporating additional data, should reduce the uncertainty to a normal level.
ENSO Conditions
The equatorial Pacific has officially transitioned to a La Niña state. However this is likely to short-lived, with a probable return to ENSO Neutral.
There is a 61% chance for a ENSO Neutral transition between January and March which will likely persist through summer, 2026. See the CPC report here.
2025 Outlook
We now calculate almost no chance that 2025 will be the warmest year on record. There is a small chance (6%) that warming over the next 2 months is sufficient to lift 2025 to be the second warmest year on record, but it is much more likely that 2025 will become the 3rd warmest year (behind 2023 and 2024). Furthermore, there is almost no chance that global average temperature anomalies for 2025 will exceed the 1.5 C (2.7 °F) benchmark above the 1850-1900 average.
Estimated Probability of 2025 annual average final rankings:
- 1st – < 1%
- 2nd – 6%
- 3rd – 94%
- 4th or lower – < 1%
New Data Source: HCLIM
This month marks the inclusion the Global Historical Climate Database (HCLIM) as a new data source in our analysis stack. HCLIM is a static database published in 2023 focused on identifying weather stations with records beginning before 1890. There are 3,623 weather stations reporting temperature data in this collection. As it turns out, the largest single source of these records is the Berkeley Earth Dataset itself. However, after deduplication, we identified ~450 weather stations that were not already present in the Berkeley Earth Dataset. Many of these include weather station files that have been newly digitized as part of HCLIM’s data rescue efforts. Inclusion of these additional time series brings the Berkeley Earth Data set to 57,592 temperature reporting weather stations.
The HCLIM collection is most notable for it’s inclusion of many early weather stations records, often pre-1850. Many of these time series are relatively short (e.g. < 10 years) and most come from areas of Europe and North America where other measurements were already known. As such, they don’t greatly change the completeness of historical record, but they do improve some of the regional uncertainties. However, a few records do provide interesting new constraints in undersampled areas, including a few early time series in Australia, India, and South America.
The end result is that HCLIM has an essentially negligible effect on estimates on land-average and global warming since 1850. However, this additional data helps to refine Berkeley Earth’s gridded land-only dataset prior to 1850. The changes are mostly consistent with the pre-existing very large uncertainties in this sparsely sampled era, but may help improve scientific work that builds upon very early instrumental data, especially in regions like North America and Europe.
