New to LNG? Start with the basics

EU gas storage forecast: winter stress test

This EU gas storage forecast projects how full European storage will be on 31 March. It starts from today's GIE AGSI+ level, uses the 16-day weather forecast, then normal weather. Pick a preset or move the sliders to test a cold winter, an LNG shortfall or a supply cut.

Projected 31 March storage: 19.6% (222 TWh) under this scenario, likely range 16.9 to 22.4%

Data as of gas day 19 Nov 2026: 69.6% full (788 TWh of 1132 TWh). Projected change to 31 March: −566 TWh.

Likely range = projection ± 2.7 points, the back-test error for the same number of days to 31 March. The model under-predicted withdrawals last winter (2025/26): run from 1 October, it left 5.5 points too much gas in storage at 31 March. The range is not bias-corrected.

EU gas storage, % full (1 Oct 2026 to 31 Mar 2027)
  • This winter
  • Projection
  • Average 2021/22 to 2025/26
  • Range 2021/22 to 2025/26

Source: GIE AGSI+ (storage), GIE ALSI (LNG send-out). Weather data: Open-Meteo (CC BY 4.0). Projection: this site's model.

31 March under each preset

PresetAssumptions31 Mar (% full)Likely range (% full)31 Mar (TWh)
Normal0 °C, LNG 0%19.616.9 to 22.4222
Cold winter−2.0 °C, LNG 0%5.22.5 to 7.959
Cold + LNG shortfall−2.0 °C, LNG −20%3.60.8 to 6.340
Mild+2.0 °C, LNG 0%34.131.3 to 36.8386

Weather: forecast to 5 Dec 2026, then the average of winters 2016/17 to 2025/26, shifted by the scenario. LNG baseline: 4.00 TWh/d, the last 30 days' average. Likely range: projection ± 2.7 points, from the back-test. Not bias-corrected.

Scenario

Weather and supply
0 °C
Applies after the 16-day forecast.
0%
Recent average 4.00 TWh/d.
0%
Share of average EU demand (9.6 TWh/d) saved.
Supply disruption
0.00 TWh/d
For example, a pipeline outage.

How the projection works

Daily EU net withdrawal is modelled from heating degree days (HDD) and LNG send-out. The model is fitted on daily data for the winters 2021/22 to 2025/26 (1 October to 31 March).

Net withdrawal (TWh/d) = a + b × HDD + c × (LNG send-out − winter average)

TermEstimateStd. errorUnit
a, intercept−2.8140.108TWh/d
b, per HDD+0.7040.012TWh/d per HDD
c, LNG deviation−0.1750.099TWh/d per TWh/d

R² 0.80, root mean square error 1.36 TWh/d, 911 days.

Back-test: winter 2025/26

The model was refitted on 2020/21 to 2024/25 and run through winter 2025/26. Actual storage on 31 March was 313.8 TWh (27.7% full).

Projection fromError, actual weather (TWh)Error, actual weather (pp)Error, normal weather (TWh)
1 Oct 2025+58.8+5.2+62.2
1 Nov 2025+37.0+3.3+31.3
1 Jan 2026+16.8+1.5+10.3

Error = projected minus actual storage on 31 March. Positive means the model left too much gas in storage. With actual weather the model still under-predicted withdrawals that winter, so treat projections as a stress test, not a forecast.

Likely range

The headline and the preset table show a likely range of projection ± the back-test error, in percentage points of working gas volume. For each back-test start date the larger of the two errors (actual or normal weather) is used. The width grows linearly with the days left to 31 March, from zero on 31 March to 5.5 points from 1 October. The range is symmetric: the projection is not bias-corrected, so the low bias seen in 2025/26 is stated, not adjusted for. It rests on one winter's back-test and is a guide, not a confidence interval.

Limitations

Data: GIE AGSI+ and GIE ALSI, updated daily. Weather data by Open-Meteo.com (CC BY 4.0), daily mean 2 m temperature.