WEEKS 3–4
Weeks 1 to 5 from NOAA's 35-day GEFS ensemble (00 UTC, 31 members): gas-heating HDD and power-burn CDD against the normal, the last five years and last year, with the members' odds and a scored record.
Week by week
Each week’s ensemble-mean total against the 1991–2020 normal, the 5-year average and last year; the members’ 10th–90th percentile range is in the table.
Weeks 3 and 4 by region
Departure from normal for each EIA storage region: where the signal is.
How it has moved
The forecast total for the same dates from each recent run: is week 3 trending colder or warmer?
The winter ahead
Loading the winter outlook…
Is it any good?
Every archived run whose week has since been observed, scored against simply forecasting the normal.
Skill is 1 − (the forecast's mean absolute error) ÷ (the error of forecasting the 1991–2020 normal): above zero beats the normal, 1 is perfect. Anomaly correlation compares the forecast's and the observed departures from normal; "sign right" is how often the forecast called the side of normal correctly. Beyond two weeks, day-to-day weather is not predictable. What can be predicted is a shift in the odds, so treat weeks 3–4 as a lean, never a number, and check the record. The GEFS 35-day runs only at 00 UTC and reaches NOAA's bucket about 28 hours after it starts, so this page updates once a day, early morning UTC.