GFS vs ECMWF Demand-Hub Outlook: 2026-11-12 12:00 UTC

Executive Summary

Both models agree on a cold Nov 13–14 blast across the Midwest and East, then split sharply in week 2: ECMWF holds the eastern half of the country much colder, with weighted 15-day cumulative HDD of 252.7 vs GFS 196.8 — a 55.9 HDD (~28%) gap. The days 1–6 ECMWF-colder spread of 14.7 HDD is statistically typical for mid-November (z = +0.17), and in past seasons only ~31% of such anomalies verified (p = 0.027) — so elevated heating demand is the better lean, but the week-2 gap should be expected to partially wash out. Convergence is low (0.164), so neither solution deserves strong confidence beyond day 6.

Hub-Level Weather

• Chicago and NYC carry 65% of the weight and drive the national divergence. GFS warms both sharply Nov 19–24 (Chicago mid-50s °F, ~6–11 HDD/day; NYC ~9–13 HDD/day), while ECMWF keeps Chicago in the mid-40s (~18–21 HDD/day) and NYC in the low-to-mid 40s (~18–23 HDD/day). GFS then flips cold again Nov 25–26 (Chicago 30.7 HDD on Nov 26) — a cold finish ECMWF does not show.
• Houston is the only cooling hub: GFS builds CDD Nov 19–23 (peaking at 73.6 °F on Nov 23; 15-day CDD 35.5 vs ECMWF 23.0), while ECMWF flips to modest heating Nov 25–26 (~60–61 °F).
• Atlanta shows the largest relative gap (+48% HDD): GFS near-zero HDD Nov 23–24 vs ECMWF ~10/day.

Demand Implications (Natural Gas)

• Weighted 15-day HDD of 252.7 (ECMWF) vs 196.8 (GFS) implies roughly 28% more heating load under the ECMWF solution, concentrated Nov 19–26 — a material swing for natural gas residential/commercial demand.
• Days 1–6 (Nov 12–17) are cold in both solutions (GFS 108.0 vs ECMWF 122.7 weighted HDD), supportive of near-term heating demand.
• Week 2 is the battleground: GFS's warm-up would collapse eastern heating load (Chicago ~5.8–10.7 HDD/day, Atlanta ~0.2–0.7), while ECMWF sustains 17–21 HDD/day in the Midwest and ~23/day in NYC.
• Cooling load is negligible nationally (weighted CDD 5.3 GFS / 3.5 ECMWF), confined to Texas under GFS.
• vs. prior cycle: the new ECMWF run came in 2.24 weighted HDD colder and 1.44 CDD warmer (slightly more southern cooling load); no prior GFS cycle was available for a GFS delta. Report was generated on a flagged material change.

Model Agreement & Caveats

• Convergence is low (0.164) and slipped −0.011 vs the prior cycle; weighted mean absolute daily HDD difference is 5.11. Lean lightly on either model, especially days 7–15.
• Spread significance (days 1–6, status "ok"): the seasonally expected ECMWF-minus-GFS spread is 13.8 HDD given GFS coldness (fit: 9.35 + 0.041 × GFS HDD); today's actual spread is 14.7, an anomaly of +0.86 HDD (z = 0.17, 16th percentile) — labeled typical. This is the models' usual cold-season bias, not new information. Verification (n = 97): historically 31.4% of such anomalies verified (SE 0.142, t = 2.21, p = 0.027, significant), meaning the ECMWF-colder signal has been directionally real but heavily damped. Do not rescale HDD by that share — it describes past windows and is not backtested as a forecast. The score covers days 1–6 only; it does not grade the week-2 divergence, which is this outlook's main uncertainty.
• Caveats: the GFS 12Z cycle was flagged as not yet ingested at fetch time (ECMWF ingested); Open-Meteo's ECMWF runs systematically colder than observed station weather from October–February, which is consistent with the expected-spread baseline rather than a genuine anomaly; and the significance score is weather-only — it is not a forecast and says nothing about NG direction. Track updates on the weather stream.