Thames Water reports 800-plus leaks found in a 19-week satellite trial

Thames Water says a 19-week satellite-assisted trial helped identify more than 800 leaks and estimates the resulting water saving at 8.7 million litres a day. The August announcement gives an operational example of AI narrowing a search across a buried network.[1][2]

The figures are the utility’s estimates. The satellite system produces locations to investigate; field confirmation and repair turn a detection lead into water retained in the network.

What the satellite can and cannot see

Satellite analysis searches for surface signals associated with underground water. Thames Water’s provider, Origin Orbit, uses that information to prioritise possible leaks for investigation. The image is a search aid rather than a direct view of a broken buried pipe.[1][2]

A second layer comes from acoustic sensors attached to the network. Escaping pressurised water generates characteristic noise that can travel through pipes and surrounding material. Comparing repeated acoustic observations helps identify likely leaks, while crews verify and repair them. Thames Water reported more than 75,000 acoustic sensors deployed, compared with 21,000 in 2025. The growth in sensors creates much denser monitoring; it also means improvements cannot fairly be attributed to one satellite algorithm alone.[1][2]

The 8.7-million-litre estimate

The leak count and saving measure different stages of the process. More than 800 is a count of identified leaks; 8.7 million litres per day is Thames Water’s estimate of the avoided flow. The release does not provide a repair-by-repair measurement ledger.[1][2]

Using 800 as the divisor would give 10,875 litres per leak per day, but the actual count is greater than 800 and leak sizes vary. That arithmetic cannot stand in for a measured average. The useful next detail would be how the estimated saving was established and checked after repair.

Putting the performance claim in context

Thames Water also said overall leakage was down more than 15% from 2019/20 and that it was repairing more than a thousand leaks each week. Those are broader system-wide figures spanning different technology investments, routine crews, weather, replacement pipes and operations. They do not mean the particular 19-week satellite trial produced the whole decline or was responsible for every weekly repair.[1][2]

A 2026 review in the Journal of Hydroinformatics makes the general point that AI leak detection must operate across different scales: whole networks, monitored districts and localised pipe sections. Models trained on laboratory recordings may struggle with real acoustic interference, uneven sensor coverage and changing pressure. A programme combining satellite leads, acoustic information and verified repairs is closer to the operational setting that actually matters, although its evaluation still needs transparent comparisons.

An independent assessment could compare discovery and repair times, confirmed leaks per alert and measured flows against similar network areas using ordinary monitoring. Those results would show which part of the reported gain comes from the new search method.[1][2]

The practical value is the chain from a satellite signal to a confirmed leak and a completed repair. AI can help crews decide where to look; water-flow measurements show what the work ultimately saved.

Report an error or suggest a correction