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

Executive Summary

The 15-day outlook from the 00Z models shows a significant divergence, with the GFS forecasting a much warmer scenario than the ECMWF. This translates to the GFS seeing lower national heating demand (90.7 weighted HDDs) and substantially higher cooling demand (48.7 weighted CDDs) compared to the ECMWF (106.9 HDDs / 25.0 CDDs). Low model agreement, driven by major differences in the Midwest and South, warrants low confidence in the overall forecast.

Hub-Level Weather

The national forecast is shaped by major disagreements in the highest-weighted demand regions:

• Midwest (Chicago, 35% weight): This is the primary driver of the heating demand difference. The ECMWF is significantly colder, forecasting 192.5 cumulative HDDs versus the GFS's 160.5 HDDs. The split is most pronounced in the day 7-10 period (May 11-13), where the ECMWF maintains a much cooler pattern.
• South (Houston & Atlanta, 25% total weight): The models diverge dramatically on the start of the cooling season. The GFS is far hotter, particularly in Houston, where it forecasts 224.3 cumulative CDDs versus just 123.2 CDDs from the ECMWF. This GFS heat extends to Atlanta, which shows 124.5 CDDs on the GFS versus only 22.4 CDDs on the ECMWF.
• East (New York & Boston, 40% total weight): The models are in closer agreement on total heating demand over the period (e.g., NYC HDDs are 74.3 vs 75.0). However, they disagree on the day-to-day timing, with the GFS showing more warmth mid-period (May 14) and the ECMWF pushing warmth later (May 16).

Demand Implications (Natural Gas)

The two models present starkly different scenarios for natural gas consumption:

• GFS Scenario: The warmer outlook implies a weaker call on residential/commercial heating demand, particularly in the Midwest. This is offset by a much stronger call on gas-fired power generation to meet early-season cooling load in the South Central and Southeast regions. The latest GFS run trended significantly warmer, adding +8.1 weighted CDDs compared to its prior cycle, reinforcing this higher power burn outlook.
• ECMWF Scenario: This solution points to more persistent late-season heating demand and a much more muted start to the cooling season. This would result in stronger residential/commercial demand and significantly weaker power burns compared to the GFS. The ECMWF was largely stable cycle-over-cycle.

Model Agreement & Caveats

Forecast confidence is low.

• Convergence: The models show poor agreement, quantified by a low convergence score of 0.1946 and a high weighted mean absolute daily HDD difference of 4.14. This indicates a high degree of uncertainty.
• Run-over-run shifts: While both models trended slightly colder on HDDs, the GFS made a large move warmer on CDDs (+8.1), widening the gap with the ECMWF and suggesting its solution may be less stable.
• Source Limitations: This analysis is based on raw operational GFS and ECMWF model output via Open-Meteo. It does not incorporate ensemble guidance (e.g., GEFS, EPS), which is critical for assessing the range of possible outcomes when deterministic models diverge this significantly.