conus-record-year-2026

Is 2026 on track to be the warmest year on record for the contiguous U.S.?

Short answer: probably. Using NOAA’s monthly data through August 2026, a set of simple statistical models gives roughly an 80% chance (range across models: about 65–87%) that 2026 finishes as the warmest year in the 132-year contiguous U.S. record, and about a 90% chance it finishes in the top two.

Year-to-date versus annual temperature for the contiguous U.S., with the 2026 projection

Analysis by Zeke Hausfather, 1 October 2026. Data through August 2026; the numbers will shift once NOAA publishes September. Full details, including cross-validation results and caveats, are in METHODS.md.

Key numbers

All values are contiguous U.S. (CONUS) average temperatures from NOAA’s Climate at a Glance (nClimDiv), in °F, with °C in brackets.

Quantity Value
January–August 2026 mean 57.98 °F (14.43 °C), warmest January–August on record; 4.13 °F above the 20th-century average
Previous warmest January–August 2012, 57.90 °F
Current annual record 2024, 55.48 °F (13.04 °C)
Runner-up 2012, 55.27 °F
September–December mean needed to beat 2024 50.50 °F (10.28 °C), about 2.2 °F above the 20th-century average for those months
September–December mean, 2016–2025 average 50.96 °F, about 2.6 °F above the 20th-century average
Years when Sep–Dec exceeded 50.50 °F 12 of 131, 8 of them since 2000
Pooled projection for 2026 annual mean 55.8 °F (13.2 °C), 5–95% range 55.1–56.5 °F
Probability 2026 beats 2024 ~79% (model range 67–87%)
Probability 2026 finishes in the top two ~90%

NOAA’s annual value is the unweighted mean of the 12 monthly values (checked to within rounding), so the only unknown is the September–December mean. A 2012-style finish (Sep–Dec of 50.03 °F) would leave 2026 in second place at 55.33 °F.

How the odds were estimated

Six statistical models predict the September–December mean from what is already known, and the annual value follows arithmetically:

  1. Regression of Sep–Dec on the January–August mean plus a warming trend.
  2. The same with the June–August mean added as a predictor (recent months carry a little extra information).
  3. and 4. Versions of 1 and 2 with a trend that steepens after 1980.
  4. A simple month-to-month persistence (AR(1)) model simulated forward from the observed August anomaly.
  5. The trend value plus the last 30 years of observed Sep–Dec departures from trend.

Each model was tested by rolling-origin cross-validation (predicting each year 2000–2025 using only earlier years). Models fitted with a single linear trend from 1895 or 1950 under-predict recent autumns by 0.5–1 °F, because U.S. warming has accelerated, so the headline pools only the models that proved unbiased with properly sized uncertainty ranges. The spread of the pooled models is a fair representation of the structural uncertainty; see METHODS.md for the full tables.

What the estimate does not use: seasonal forecasts (such as NOAA CPC outlooks) or the current El Niño state. Historically El Niño has a mixed-sign and small effect on the CONUS annual mean.

Reproducing

pip install -r requirements.txt
./fetch_data.sh              # optional: refresh the NOAA series (data/ holds the 1 Oct 2026 snapshot)
python record_odds.py        # model fits, 2026 odds, cross-validation on 1980–2025
python cv2.py                # cross-validation on 2000–2025, hinge-trend models, hindcasts of 2012/2016/2024/2025
python make_figure.py        # pooled estimate, results/model_summary.csv, two-panel figure
python make_figure_simple.py # single-panel figure shown above

Files

License and data

Code is released under the MIT License. NOAA nClimDiv / Climate at a Glance data are in the public domain as a U.S. Government work; please cite NOAA National Centers for Environmental Information, Climate at a Glance: National Time Series, retrieved 1 October 2026, from https://www.ncei.noaa.gov/access/monitoring/climate-at-a-glance/national/time-series.