Natural Gas Weather — Week of November 15, 2026

GFS vs ECMWF Demand-Hub Outlook: 2026-11-15 12:00 UTC

Open-Meteo GFS seamless vs ECMWF IFS 0.25°, 15-day horizon, demand-weighted HDD/CDD at 65°F base. Fetched 21:33 UTC.

Executive Summary

Both models agree on a cold start (Nov 15–17) but diverge sharply thereafter: ECMWF holds a much colder days 5–15 pattern, with weighted 15-day cumulative HDD of 264.0 vs GFS 198.6 — roughly one-third more implied heating load. The gap is concentrated in Chicago and NYC, which carry 65% of the demand weight and explain ~83% of the national divergence. Convergence is low (0.164), and the days 1–6 ECMWF-colder spread of 13.8 HDD is statistically typical for mid-November, meaning most of it reflects the models' usual seasonal bias rather than new information.

Hub-Level Weather

Hub (weight) GFS cum HDD ECMWF cum HDD Δ (ECMWF–GFS) Mean abs HDD diff/day
Midwest / Chicago (0.35) 254.3 329.4 +75.1 5.57
East / NYC (0.30) 206.8 299.7 +92.9 6.88
Northeast / Boston (0.10) 249.7 316.3 +66.6 5.05
South Central / Houston (0.15) 33.3 (CDD 44.6) 55.3 (CDD 8.8) +22.0 HDD / −35.9 CDD 2.02
South / Atlanta (0.10) 175.5 189.3 +13.8 2.72
  • Days 1–3 (Nov 15–17): tight agreement. Chicago HDDs differ by ≤1 HDD/day; both models show mean temps in the upper 30s–low 40s °F across the Midwest and East.
  • Days 4–9 (Nov 18–22): GFS turns milder. GFS lifts Chicago into the mid-50s °F by Nov 19–21 (HDD 8–10/day) and NYC to 54–56°F Nov 21–23; ECMWF stays 3–6°F cooler with HDDs 13–19/day in the East.
  • Days 9–11 (Nov 23–25): the core divergence. ECMWF drives a cold shot — Chicago mean temps 33.3–35.1°F (HDD 29.9–31.7/day) and NYC 32.5–35.4°F (HDD 29.6–32.5/day) vs GFS in the mid-40s °F. This lands Mon–Wed of Thanksgiving week.
  • Late window: partial reconvergence, then a second split. Boston and Chicago re-converge by Nov 28, but NYC Nov 27–28 shows GFS at ~59°F (HDD 5.6–5.7) vs ECMWF 47–49°F (HDD 16–18).
  • Houston is a directional flip. GFS produces 44.6 CDD (Nov 19–26, peaking at 7.5 on Nov 26) and closes the window at 71.5°F on Nov 29; ECMWF limits CDD to 8.8 and ends at 49.7°F (15.3 HDD) — a 21.8°F single-day gap, the largest in the dataset.

Demand Implications (Natural Gas)

  • The national picture is heating-dominated: weighted CDD is negligible in both solutions (GFS 6.7, all Houston; ECMWF 1.3). The demand story is entirely natural gas resi/heat load.
  • The 65.5 weighted-HDD gap between solutions is the key risk: if ECMWF verifies, implied heating load over the window runs ~33% above the GFS solution — a materially colder demand scenario for NG balances, with the Nov 23–25 cold shot the single biggest driver of incremental burn and storage draw (see EIA storage).
  • Under GFS, demand fades quickly after Nov 22 (East HDDs drop to 5–12/day late-month) and Houston adds modest cooling load — a much softer pull profile.
  • Atlanta is the closest-agreed hub (Δ13.8 HDD), so the Southeast demand signal is comparatively well constrained.

Model Agreement & Caveats

  • Convergence is low: 0.164 (weighted mean abs HDD diff 5.09 HDD/day). Treat days 1–4 as the reliable signal and days 5–15 as two distinct scenarios rather than a consensus. The ECMWF cold shot is the low-confidence element.
  • Spread significance (status "ok," days 1–6): today's ECMWF-colder spread is 13.81 HDD (ECMWF 93.27 vs GFS 79.47). What is normal for this season is 12.26 HDD — Open-Meteo's ECMWF runs systematically colder from October through February — so today's anomaly is +1.54 HDD (z = +0.30), labeled "typical," with the absolute anomaly at only the 25th percentile of past cycles. In plain terms: the models' gap is slightly wider than the seasonal norm but well within it; most of the raw spread is known bias, not news.
  • Verification (status "ok"): across 96 comparable cycles, about 45% of the ECMWF-colder anomaly was realized at the 5-day verification lag — a statistically significant persistence (t = 3.11, p = 0.002, Newey-West lag 5). This describes past windows and is not a backtested forecast; it should not be converted into a corrected HDD figure.
  • Run-over-run: the 12Z ECMWF run warmed by ~12.0 weighted HDD vs its 00Z run (CDD −0.22), narrowing the gap to GFS (mean abs diff −0.98; convergence +0.023). No prior GFS cycle was available for comparison.
  • Caveats: the GFS 12Z run is flagged not yet ingested in run metadata despite series being present; hubs are single-grid-point proxies at 0.25°; the expected-spread fit is noisy (R² = 0.15, residual SD ≈ 5.1 HDD). The spread-significance score is weather-only — it is not a forecast and says nothing about Henry Hub price direction. See the full weather model comparison for underlying data.

Arc Research publishes market data and research for information only. It is not investment, trading or financial advice, and Arc Research is not a registered investment adviser or commodity trading advisor. Futures and options trading carries a substantial risk of loss. Hypothetical, backtested and paper results are not actual trading. Read the full disclaimer · Terms of Service · Affiliates