GFS vs ECMWF Demand-Hub Outlook: 2026-04-21 00Z

Executive Summary

The 15-day outlook presents a highly uncertain demand picture, characterized by significant disagreement between the GFS and ECMWF models. While the GFS maintains a colder national forecast (160.7 weighted HDDs), the ECMWF trended sharply warmer in this cycle (-19.5 HDDs run-over-run). This divergence is primarily driven by a major split in the high-demand Midwest hub, where the GFS forecasts a late-season cold shot not seen in the ECMWF.

Hub-Level Weather

The national demand-weighted forecast is a tale of two competing regions:

• Midwest (Chicago, 35% weight): This is the primary point of disagreement. The GFS is substantially colder, projecting 204.5 cumulative HDDs versus just 135.7 HDDs from the ECMWF. The divergence is most pronounced in the May 1-4 period, where GFS daily HDDs are 7-10 higher than the ECMWF.
• East (NYC, 30% weight & Boston, 10% weight): The models are reversed in the East, where the ECMWF is colder. It forecasts 216.9 HDDs for NYC and 302.7 for Boston, compared to the GFS's 188.3 and 270.7, respectively. This partially offsets the GFS's cold Midwest forecast.
• South (Houston, 15% weight): Cooling demand is emerging, with the GFS showing a slightly warmer outlook (155.2 CDDs) than the ECMWF (139.8 CDDs).

Demand Implications (Natural Gas)

The competing model solutions imply very different scenarios for shoulder-season natural gas demand.

• The GFS forecast (160.7 weighted HDDs) suggests lingering heating demand that is stronger than seasonal norms, which would provide price support.
• The ECMWF forecast (145.4 weighted HDDs) points to a faster decline in heating load. Critically, the ECMWF's substantial warmer trend in this cycle (-19.5 weighted HDDs) is a bearish signal, suggesting demand expectations could be revised lower.
• Total weighted cooling demand is nearly identical between models (26.8 CDDs), indicating that while cooling is beginning, it is not yet the primary driver of forecast disagreement.

Model Agreement & Caveats

Confidence in this forecast is low due to significant and growing model disagreement.

• Convergence: The weighted mean absolute HDD difference is 3.14 per day, a substantial margin indicating poor model agreement. The overall convergence score is 0.24.
• Run-over-Run Trends: The models are trending in opposite directions. The GFS added a modest amount of heating demand (+2.8 weighted HDDs), while the ECMWF shed a significant amount (-19.5 weighted HDDs). This growing divergence increases forecast uncertainty.
• Source: This analysis is based on raw GFS and ECMWF model data from Open-Meteo. Ensemble guidance should be consulted to gauge the full range of potential outcomes.