GFS vs ECMWF Demand-Hub Outlook: 2026-09-28 00:00 UTC

Executive Summary

A substantial divergence exists across the 15-day outlook between the GFS and ECMWF models, reflected in a low convergence score of 0.3093. The ECMWF projects a significantly colder pattern across key consuming regions, generating 72.94 demand-weighted cumulative HDDs compared to 56.72 HDDs in the GFS. Run-over-run deltas indicate the GFS warmed sharply by shedding 9.81 weighted HDDs from the prior cycle, whereas the ECMWF held stable (+0.66 HDDs), amplifying forecast risk for natural gas space heating demand.

---

Hub-Level Weather

The divergence is most pronounced in high-weight northern heating hubs during the Day 10–15 window (October 7–12):

+--------------------------+--------+-------------+---------------+-------------+---------------+
| Hub                      | Weight | GFS Cum HDD | ECMWF Cum HDD | GFS Cum CDD | ECMWF Cum CDD |
+--------------------------+--------+-------------+---------------+-------------+---------------+
| Midwest (Chicago)        | 0.35   | 83.30       | 113.00        | 2.19        | 5.71          |
| East (New York City)     | 0.30   | 54.35       | 72.20         | 25.94       | 14.02         |
| South Central (Houston)  | 0.15   | 0.00        | 0.00          | 241.83      | 224.10        |
| Northeast (Boston)       | 0.10   | 109.04      | 91.71         | 10.69       | 8.99          |
| South (Atlanta)          | 0.10   | 3.61        | 25.57         | 95.61       | 57.98         |
+--------------------------+--------+-------------+---------------+-------------+---------------+

• Midwest (Chicago - 35% weight): The primary driver of aggregate HDD divergence (mean absolute difference of 3.07 HDD/day). The ECMWF advertises a sharp cold shot on October 10 (mean 47.52°F, 17.48 HDDs) and October 11–12 (14.22 and 13.88 HDDs), whereas the GFS holds temperatures milder (October 10 mean 58.97°F, 6.03 HDDs; October 12 at 62.44°F, 2.56 HDDs).
• East (New York City - 30% weight): ECMWF is colder overall (72.20 HDDs vs. 54.35 HDDs GFS). The GFS sustains warmer early-period temperatures (October 2 CDD of 12.61 vs. 7.97 ECMWF) and resists the late-period chill (October 12 ECMWF 12.56 HDDs vs. GFS 1.90 HDDs).
• South (Atlanta - 10% weight) & South Central (Houston - 15% weight): GFS preserves late-season power burn cooling load (Atlanta cumulative CDD of 95.61 vs. 57.98 ECMWF; Houston CDD of 241.83 vs. 224.10 ECMWF). The ECMWF pushes the cold front deep into the South by October 10–12, delivering 25.57 HDDs to Atlanta compared to only 3.61 HDDs in the GFS.
• Northeast (Boston - 10% weight): Boston is the sole hub where the GFS is colder (109.04 HDDs vs. 91.71 HDDs ECMWF), primarily driven by an earlier cold plunge on October 7–8 (17.90 GFS HDDs vs. 10.55 ECMWF HDDs on Oct 7).

---

Demand Implications (Natural Gas)

• Heating Load Ramp-Up: If the ECMWF verifies (72.94 weighted HDDs), residential/commercial (ResCom) heating demand will experience an early seasonal acceleration across the Midwest and Mid-Atlantic, supporting prompt NG contracts. The GFS path (56.72 weighted HDDs) implies a looser balance with delayed storage withdrawals.
• Cooling Load Fade: With national cooling demand waning seasonally, GFS's higher weighted CDDs (55.45 vs. 46.52 ECMWF) offer limited incremental power burn support relative to the potential heating load gains offered by the ECMWF.
• Net Implied Load: In aggregate, the ECMWF represents a net-bullish load scenario for natural gas relative to the GFS, with Chicago and New York driving ~70% of the total 16.21 weighted HDD model gap.

---

Model Agreement & Caveats

• Convergence Score (0.3093): Low model agreement reflects major structural synoptic differences regarding the timing and southern penetration of the Canadian trough after Day 8. Confidence in deterministic heating load totals beyond October 6 remains low.
• Run Volatility: The GFS shed 9.81 weighted HDDs and added 2.72 CDDs run-over-run, highlighting elevated operational model instability. The ECMWF displayed strong cycle continuity (+0.66 HDDs).
• Open-Meteo & Catalog Limitations: Degree-day aggregates rely on 5 demand-weighted metropolitan centroids. While effective for capturing broad population-weighted weather demand, point data may smooth out localized pipeline freeze-offs or microclimate heating spikes in non-hub production and market basins. Lean on ensemble spreads until convergence improves toward 0.70.