GFS vs ECMWF Demand-Hub Outlook: 2026-05-20 00Z

Executive Summary

The 15-day outlook presents a significant disagreement between models, reducing overall forecast confidence. The GFS is materially colder than the ECMWF on a national basis (43.8 weighted HDDs vs 29.4), driven almost entirely by a colder forecast in the Midwest. Run-over-run trends show the models diverging, with the ECMWF trending significantly warmer in its latest cycle.

Hub-Level Weather

The primary source of model disagreement is in the high-demand Midwest (Chicago, 35% weight), where the GFS projects 64.5 cumulative HDDs versus only 24.4 HDDs from the ECMWF. This difference is most pronounced in the back half of the forecast (June 1-3).

In contrast, the models show better agreement on heating demand in the East (New York City, 30% weight), with the GFS at 44.0 HDDs and the ECMWF at 46.6 HDDs.

For cooling demand, the GFS is consistently warmer in the south. In South Central (Houston, 15% weight), the GFS forecasts 236.2 cumulative CDDs, a notable 54.4 CDDs more than the ECMWF's 181.8 CDD projection.

Demand Implications (Natural Gas)

The GFS scenario implies lingering late-season heating demand for natural gas in the Midwest, which would be supportive for residential/commercial consumption. This is paired with strong power burn demand from a hot forecast in the South.

The ECMWF paints a more bearish picture for heating demand, showing a rapid transition away from heating needs in the Midwest. While it still shows healthy cooling demand building, its overall temperature-driven demand signal for the 15-day period is weaker than the GFS.

Model Agreement & Caveats

Forecast confidence is low due to poor model agreement, reflected in a convergence_score of just 0.38.

The latest model cycle shows a significant divergence in trends. The ECMWF shifted markedly warmer, shedding 9.5 weighted HDDs and adding 12.5 weighted CDDs compared to its prior run. The GFS, meanwhile, trended slightly colder (+1.0 weighted HDDs). This growing split between the models warrants caution until they show better convergence.

Data is sourced from Open-Meteo. Hub locations serve as proxies for broader demand regions.