Will European gas storage last the season?
Every morning: where European gas storage is heading over the next weeks and at the end of the season, the probability of finishing below the threshold, and what moved that answer since the previous run. Built on 35 past winters, the 16-day weather forecast and the observed supply balance, and scored in public against a daily replay of 2019-2026. No price forecast.
Read this before using the numbers
- This is a projection of physical storage, not of prices, and not advice to hedge or trade.
- The model does not simulate how the market reacts to a shortage (extra LNG cargoes, demand cuts). In winter 2021-22 it projected a much lower end-of-winter level than the market delivered. In a tight winter, read the low tail as "what happens if nobody reacts".
- Uncertainty bands were too narrow in our 7-winter hindcasts (the raw P10-P90 band held 63-72 % of outcomes instead of 80 %). They are widened by a factor measured on those same hindcasts, and every published run is scored below.
- Where the model earns its keep: 1 to 4 weeks ahead (20-29 % smaller errors than the 5-year average in the 2019-2026 replay). On the end-of-season level it is only as accurate as that average; what it adds there is a calibrated range and the daily attribution.
- Low storage is not a supply cut. Finishing a winter at 20-30 % is tight, not a shortage.
EU storage fill: observed and projected
Uncertainty fan: ProDashed red line: 30 % threshold. Model run 13 Oct 2026, data to 11 Oct 2026.
Country projections
Including Germany, France, Italy.
Pro adds Germany, France and Italy with their own fans, probabilities and revision attribution, plus import and refill needs, weather analogues, the live track record and CSV downloads.
Track record
The same model replayed as if it had been published every day from October 2019, each day using only the storage and weather data available then (2569 EU runs). Mean absolute error in fill points, against the "5-year average" baseline on the same days.
| Horizon | Model | 5-yr average | Gain | Inside P10-P90 |
|---|---|---|---|---|
| 7 days | 0.48 | 0.68 | +29 % | 82 % |
| 14 days | 0.92 | 1.23 | +25 % | 84 % |
| 30 days | 1.90 | 2.37 | +20 % | 85 % |
| End of season | 6.90 | 6.83 | -1 % | 80 % |
Target for the band is 80 %. The widening of the band was set on these same winters, so this coverage is in-sample; the live record below is the out-of-sample test.
Season by season: gain over the 5-year average
Bad seasons stay on the list. The worst, 2021-22, is the regime change when Russian flows stopped and the market reacted in ways the model does not simulate.
Sources: GIE AGSI+ (storage, 18 EU countries), Open-Meteo (observed and forecast temperatures, 18 cities weighted by gas demand, 1991 onwards). Model: daily storage change explained by heating degree days, season, weekday and fill level, plus a Kalman-filtered supply balance; uncertainty from replaying every past winter since 1991, adjusted for warming. Runs are published daily around 07:00 CET and are never edited after publication.