Heating and cooling degree days from eleven weather models.
Liberal Weather computes HDD and CDD for the contiguous United States from nine global weather models — NOAA GEFS (14- and 34-day) and AIGEFS, NOAA CFSv2; ECMWF ENS, AIFS ENS and EC46; and Environment Canada's CMC GEPS (14- and 39-day) — plus weighted blends of them (the Liberal Ensembles). Degree days are gas- and population-weighted on a 0.25° CONUS grid built from county-level U.S. Census data — not a sample of cities — and compared against 5, 10 and 30-year ERA5 climatology, with a daily record of how each model has actually performed against observed reanalysis at every lead time.
No warranty · Not trading advice
This is an experimental weather-data project. It is provided as is, with no warranty of any kind, express or implied, and no guarantee of accuracy, completeness or timeliness.
It is not trading, investment, financial or operational advice, and nothing here is a recommendation to buy or sell any commodity, security or product. Degree-day forecasts can and do go wrong — always verify against official sources before acting on anything.
Not affiliated with, sponsored by, or endorsed by NOAA, ECMWF, Environment and Climate Change Canada, Copernicus, the U.S. Energy Information Administration or the U.S. Census Bureau.
The models
Every model on the dashboard. Range is the forecast horizon; members is the ensemble size.
| Model | Range | Cycles | Members | 14-day blend | 34-day blend |
|---|---|---|---|---|---|
| loading… | |||||
How a degree day is defined
A day is integrated over its four synoptic samples (00, 06, 12 and 18 Z): the base is subtracted from each sample and the four results are averaged — not the base applied to a daily mean (that understates the index, because max(0, ·) is convex), and deliberately not the (Tmax+Tmin)/2 convention some services use. The same estimator is applied to every model so they stay directly comparable, and the AI models publish no daily extremes at all.
HDD = how far each sample falls below 60 °F (the residential gas-heating balance point); CDD = how far it rises above 65 °F. A day's value is the mean over its samples; values are summed over each model's forecast window and expressed in °F-days.
Weighting is county-level. Each model's grid-cell degree days are averaged with a spatial weight grid built from county-level U.S. Census data — county total population (Census PEP vintage 2024, anchored to 2020 Census block groups) and the share of homes heated with gas (ACS B25040) — anchored to state-level EIA residential and commercial gas consumption, and snapped onto the 0.25° CONUS grid at each block group's centre of population — no smoothing and no ocean mask. CDD is population-weighted, and HDD is 60% gas / 40% population — because heating demand follows the gas-burning population, not the grid-cell average.
Data & licensing
All sources permit commercial use and redistribution; the obligations are attribution only.