London water-level forecast available · cautious warning in public preview

How the Thames London forecast was tested

The simple warning spotted all four recorded London-wide restrictions, but it also raised false alarms in 11 of 25 seasons without a restriction. That is an intentional trade-off: it is designed to flag possible danger early, not to promise that a ban will happen.

Latest London reservoir level

69.5%capacity-weighted Lower Thames and Lower Lee groups

Lower Thames 68% · Lower Lee 76%

Measured 11 Aug 2026
Early warning preview · not an official restriction

Season watch triggered 1 Mar 2026

The latest 60-day forecast is no longer in the lowest fifth for this time of year, but the earlier seasonal signal remains visible. The official Thames Water restriction is separately verified as effective from 23 July 2026.

Open the official legal notice ↗

Recent reservoir levels

A year of London reservoir storage.

This chart shows the latest 12 month-end readings. The forecast uses the full daily history from 1989.

Simple 60-day forecast

Today’s level plus what usually happened next in past years.

Season watchActive
Current combined69.5%11 Aug 2026
Typical change in past years-8.5 pp34 non-event reference years
Forecast reservoir level61%10 Oct 2026
Compared with this time of year15th percentilelatest forecast; warning level is the lowest 20%

The matched change is the net historical outcome of rainfall, customer use, river abstraction, transfers and operations. It does not claim to isolate those drivers. A season watch stays visible for up to 180 days after a threshold crossing so an early signal cannot silently disappear after reservoir refill.

Water-level forecast test

The forecast is available across London’s main supply area.

We chose the method using 2018–2021 data, then tested it on 209 later weekly starting points from 2022–2025. This is separate from the less reliable hosepipe-warning test.

14-day test result1.84 percentage-point typical errorlikely range contained the answer 89% of the time · seasonal change median
30-day test result3.12 percentage-point typical errorlikely range contained the answer 85% of the time · seasonal change median
60-day test result5.38 percentage-point typical errorlikely range contained the answer 77% of the time · absolute seasonal mean
90-day test result5.36 percentage-point typical errorlikely range contained the answer 74% of the time · absolute seasonal mean

Every time period passes the water-level accuracy check. The forecast starts with the Environment Agency weekly reading for 11 Aug 2026. It is available only for postcodes in London’s main supply area; the other five Thames areas do not yet have this forecast.

Past restriction test

The warning spotted all four recorded London restrictions.

4 of 4 spotted

We chose the rule using the 2006, 2012 and 2022 restrictions, then tested it against 2026. This was a test using past data, not a warning issued at the time. Restriction years are not used to calculate the normal reservoir change.

Event starts
4/4
Quiet seasons
14/25
False alarms
11
No-ban seasons correctly left alone
56%

First signal before each announcement

2006
176 days · 9 Oct 2005 · earlier test
2012
158 days · 30 Oct 2011 · earlier test
2022
10 days · 7 Aug 2022 · earlier test
2026
177 days · 25 Jan 2026 · later test

How to read this warning

Useful as a cautious early warning.

Public preview

The later 2026 test was signalled 177 days before announcement. That is encouraging. It is still only one later test, and the warning also appeared in 11 seasons without a restriction. We accept those early signals because missing a developing ban is more costly here than asking users to keep an eye on the situation.

  • Later 2026 restriction spotted: 1 of 1.
  • Later no-ban seasons correctly left alone: 2 of 3.
  • No numerical hosepipe-ban probability is published.
  • The official restriction layer remains authoritative.

Every recorded London restriction start was flagged. The warning remains clearly labelled as a cautious preview because only one of those starts happened after the rule was fixed.

The uncomfortable result

Perfect recall does not mean precise prediction.

Eleven false-alarm seasons is the price of catching all four recorded events with this simple reservoir-only rule. The page exposes that cost because hiding it would make the result look much stronger than it is.

Source and scope

Daily official history, one precise geography.

The London Datastore publishes Thames Water’s daily Lower Lee and Lower Thames percentages from 1989 under the Open Government Licence. The current anchor comes from the Environment Agency’s weekly reservoir chart.

Open the reservoir dataset ↗
  • 13,711 usable historical daily observations through 30 June 2026, plus the 11 Aug 2026 weekly anchor.
  • Lower Thames and Lower Lee are capacity-weighted using approximately 165,000 Ml and 37,000 Ml respectively.
  • The 22 July 2025 Thames Valley restriction is not labelled as a London event because it applied to Oxfordshire, Gloucestershire, Wiltshire and selected Berkshire postcodes, not the London supply area represented by this reservoir series.
  • Each restriction date links to a Thames Water notice or regulatory document; none is inferred from reservoir levels.