Natural Gas Weather — Week of November 16, 2026
GFS vs ECMWF Demand-Hub Outlook: 2026-11-16 00Z
Executive Summary
Both models open the period cold (Nov 16–17) and then diverge sharply: ECMWF finishes the 15-day window at 237.7 demand-weighted HDD vs GFS at 205.3 — a 32.4 HDD (+15.8%) colder national picture driven mainly by the East. The days 1–6 ECMWF-colder spread of 12.6 HDD is statistically "typical" for mid-November (z = −0.15), so it reads as the models' usual seasonal bias rather than a fresh cold signal. ECMWF trended 26.3 weighted HDD warmer run-over-run, which is the material change that triggered this report.
Hub-Level Weather
| Hub (weight) | GFS cum HDD | ECMWF cum HDD | ECMWF − GFS | Mean abs daily HDD diff |
|---|---|---|---|---|
| Midwest / Chicago (0.35) | 273.2 | 289.7 | +16.5 | 2.53 |
| East / NYC (0.30) | 199.6 | 261.0 | +61.4 | 5.01 |
| South Central / Houston (0.15) | 72.9 | 83.0 | +10.2 | 2.28 |
| Northeast / Boston (0.10) | 248.6 | 289.3 | +40.7 | 4.33 |
| South / Atlanta (0.10) | 139.9 | 166.3 | +26.4 | 2.29 |
- Cold start, high agreement (Nov 16–17): models are within ~0.4–0.8 HDD/day at every hub; both show 20+ HDD in the Midwest and Northeast.
- Contested mid-period warm-up (Nov 18–22): GFS is markedly milder in the East — NYC averages ~10.4 HDD/day under GFS vs ~15.2 under ECMWF — and shows genuine cooling demand in Houston (32.2 cumulative CDD, peaking at 7.4 CDD on Nov 22 at 72.4°F, vs 0.4 CDD / 65.4°F for ECMWF).
- Flipped cold risk (Nov 23–26): ECMWF drives a much colder East Coast surge; the single largest daily gap is NYC on Nov 25 (26.5 vs 10.5 HDD; 38.5°F vs 54.5°F). GFS instead holds its cold for the Midwest at the end of the window (Chicago Nov 30: 28.0 vs 22.4 HDD).
- The East hubs (NYC + Boston, 0.40 combined weight) account for the bulk of the national model gap.
Demand Implications (Natural Gas)
- Heating dominates: weighted cumulative CDD is negligible in both models (GFS 4.8, ECMWF 0.8), so the natural gas demand picture is effectively all HDD.
- ECMWF implies ~15.8% more weighted heating load than GFS over 15 days. The gap is concentrated in the high-weight East: NYC alone contributes a 61.4 HDD cumulative divergence (18.5 weighted HDD).
- Near term (days 1–6), ECMWF is 12.6 weighted HDD colder (80.3 vs 67.6) — supportive of heating load early, but per the significance scoring below, that gap is not unusual for the season.
- The demand-relevant uncertainty is timing of the late-month cold: ECMWF's Nov 24–26 East cold shot vs GFS's Nov 28–30 Midwest/East cold changes which days carry peak load, even where totals converge.
- This is a weather-demand signal only; it says nothing about direction for NG.
Model Agreement & Caveats
- Convergence: score of 0.228, up 0.052 from the prior cycle — modest agreement that is improving. Models align well through day 2 and diverge materially from day 3 onward (weighted mean abs HDD diff 3.39/day). Lean on the day 1–2 signal strongly; treat 15-day cumulative totals as a wide range.
- Spread significance (days 1–6, status "ok"): today's ECMWF-minus-GFS spread is 12.65 HDD vs a seasonal expected spread of 13.45 — an anomaly of −0.81 HDD (z = −0.147, 12th percentile), labeled "typical." In other words, the "ECMWF colder" gap is the well-documented Open-Meteo ECMWF cold bias normal from October to February, not new information.
- Verification (status "ok"): across 95 past cycles, 41.4% of the spread anomaly was realized (SE 0.145, t = 2.851, p = 0.004 — statistically significant at the 5% level). With today's anomaly essentially at zero, this adds little incremental signal. Note the share describes past windows and is not backtested as a forecast — it is not a corrected HDD number.
- Run-over-run: ECMWF 00Z came in 26.3 weighted HDD warmer (and convergence +0.05) vs its prior 12Z Nov 15 run; no prior GFS cycle was available for comparison. CDD delta (−0.5) is immaterial.
- Data caveats: the GFS 00Z run is flagged
ingested: falseat fetch time (08:43Z) despite series being present — treat GFS values with some caution. Underlying data are ~0.25° gridded Open-Meteo fields for five point hubs, two deterministic runs only; see the weather model history for run-over-run context. Degree days use a 65°F base; the convergence score concerns weather agreement, not price.