GFS vs ECMWF Demand-Hub Outlook: 2026-11-14 12Z

Executive Summary

GFS and ECMWF agree on a cold start to the 15-day window but split sharply from Nov 23–28, leaving demand-weighted cumulative HDD at 248.2 (GFS) vs 302.2 (ECMWF) — a ~54 HDD gap (~22% more heating load under ECMWF), with roughly three-quarters of it beyond day 6. Convergence is low (0.173), and the days 1–6 ECMWF-colder spread of +12.98 HDD is statistically typical for mid-November (expected 12.77, z = 0.04), so the near-term model gap is seasonal bias, not signal. The actionable uncertainty is the late-period pattern: GFS moderates into a demand lull; ECMWF holds a cold surge.

Hub-Level Weather

• Midwest (Chicago, 0.35 weight): GFS 296.7 vs ECMWF 365.7 cumulative HDD (+69.0 colder; mean abs daily diff 5.89). Both cold through Nov 22, then diverge: GFS means 38–47°F Nov 23–28 vs ECMWF 31–34°F with 30–34 HDD/day.
• East (NYC, 0.30): GFS 288.7 vs ECMWF 320.1 HDD (+31.4; diff 3.20). ECMWF colder Nov 24–28 (means 31.8–41.8°F vs GFS 34.7–46.8°F).
• Northeast (Boston, 0.10): Closest agreement of the majors — GFS 329.6 vs ECMWF 321.6 HDD (GFS colder by 8.0; diff 2.52). Both show a Nov 24–25 shot (means low 30s).
• South Central (Houston, 0.15): Largest structural split (mean abs HDD diff 6.57). GFS turns warm: 15-day CDD 37.2 vs HDD 33.5, peaking at 73.1°F / 8.1 CDD Nov 26. ECMWF flips cold after Nov 22: means 47.0–51.6°F, HDD 13.4–18.0/day, 15-day HDD 123.7 vs CDD 6.6.
• South (Atlanta, 0.10): GFS 197.6 vs ECMWF 274.2 HDD (+76.6; diff 5.11). ECMWF colder nearly every day after Nov 16, with means of 40–43°F late period vs GFS 43–53°F.

Demand Implications (Natural Gas)

• 15-day weighted load: ECMWF implies 302.2 weighted HDD vs GFS 248.2 — ~54 weighted HDD (~22%) more heating demand. CDD is negligible nationally (GFS 5.58 vs ECMWF 0.99 weighted); only the GFS Houston solution carries any cooling load.
• Timing: Both models support firm early-period heating (days 1–6 weighted HDD: ECMWF 105.2, GFS 92.2). The fork is Nov 23–28: GFS implies a late-month demand lull with only a brief Houston cooling bump, while ECMWF implies a broad cold surge — Chicago, Houston, and Atlanta together account for the bulk of the weighted gap in that stretch — sustaining elevated heating load (and storage withdrawals, see EIA data) into month-end.
• This is a weather-demand read only; it is not a directional call for natural gas futures (NG).

Model Agreement & Caveats

• Convergence: 0.173 (low) with weighted mean abs daily HDD diff of 4.77 — the models disagree materially, concentrated in days ~9–15. Run-over-run (ECMWF 12z vs 00z): weighted HDD +47.52 (markedly colder), CDD −1.17, convergence +0.001; the report triggered on this material change. No GFS delta is available (prior GFS cycle not on file).
• Spread significance (days 1–6, status "ok"): ECMWF 105.20 vs GFS 92.22 weighted HDD → spread +12.98 colder. That is almost exactly the seasonal norm of +12.77 (Open-Meteo's ECMWF runs systematically cold Oct–Feb; expected spread scales with GFS HDD at +0.043/HDD). Anomaly is just +0.20 HDD (z = 0.04; 3.7th percentile), label "typical" — the gap is the models' usual bias, not news. Verification (status "ok", n = 97): historically 43.2% of such anomalies verified (SE 0.145, p = 0.003, significant at 0.05), but with today's anomaly near zero there is no correction to apply, and the share is not a backtested forecast — do not convert it into an HDD number.
• Caveats: Point-hub interpolation from ~0.25° fields via Open-Meteo (weather stream); 65°F-base HDD/CDD ignores wind and other load drivers; fixed demand weights are approximations; days beyond ~10 carry low skill; the pipeline flags the GFS 12z run as not yet ingested (ingested: false) despite its point series being present, so GFS figures are provisional. The convergence/spread scores describe weather model agreement only and say nothing about Henry Hub price direction.