Can Thermal Drones Detect Water Leaks?

Thermal drones can find water leaks, but not in the way most people assume. A thermal camera cannot see water and cannot see through the ground. What it reads is surface temperature, and a leak underneath changes that temperature in ways a good survey can pick out. That distinction decides when to fly, what to buy, and how much trust to place in the result.

How A Thermal Drone Finds A Water Leak

Every surface above absolute zero emits infrared energy, and a thermal sensor reads the longwave band from roughly 8 to 14 micrometres. What it produces is a temperature map of the surface, nothing more.

Escaping water changes that map through three mechanisms. Evaporation cools the ground above the leak. Conduction carries heat differently through saturated soil than through dry backfill. And saturated ground has higher thermal inertia, because both its heat capacity and its thermal conductivity rise with moisture content, which damps the daily temperature swing at the surface. That damped swing is often the clearest signature of all.

The differentials involved are small. A documented survey over a 517 m corridor in Tías, Lanzarote, flown pre-dawn on 27 March 2026 at 30 m above ground with around 3 cm ground sample distance, logged anomalies between 0.6°C and 1.3°C above the local pipe baseline. Roughly one degree is what you are hunting.

Why Time Of Day Matters More Than The Camera

Because the signal is that faint, survey conditions dominate results. Every serious practitioner flies at night or before dawn. Solar radiation loads the surface unevenly during the day, creating false warm patches that swamp a one-degree signature. There is also thermal crossover to avoid, the point in the daily cycle where target and background temperatures equalise and contrast disappears entirely.

Season matters in the UK too. Cold water leaks depend on the difference between mains temperature and the surrounding ground, and that gap is largest between October and March. Attempting cold water detection in mild summer weather is largely wasted flight time.

Hot water is a different proposition. Leaks on heated pipework produce strong contrast, which is why the strongest published evidence comes from district heating networks. A study in Pattern Recognition Letters, volume 140, December 2020, analysed roughly 13.4 million image patches from 12 UAV infrared sequences over Danish cities, and a convolutional neural network reached 87.2% average weighted accuracy while detecting about 98.6% of true leakages. Impressive, but on hot water with a large differential.

What UK Water Companies Have Actually Done

The UK context gives this real weight. The Environment Agency's water resources report for 2024 to 2025, published on 18 November 2025, recorded 2,617 million litres a day lost to leakage in England, close to 19% of distribution input and 43.0 litres per person per day. That was the lowest figure in over two decades, against a 2017 to 2018 baseline of 2,987 Ml/d, and companies have committed to halving leakage by 2050.

Several have put drones to work on it. Anglian Water trialled thermal drones at Southery and Wissington in Norfolk in January 2017 as part of a £60 million leakage programme covering nearly 24,000 miles of mains. By October 2022 it had added visual detection of unusual flora growth alongside FLIR thermal sensors, reporting savings of up to £7,000 per flight in water loss.

Severn Trent announced a drones-in-a-box trial in January 2025, flying in Nottinghamshire from a command centre at its Coventry headquarters. The company cited a 6 km rural pipeline that previously took a long time to walk being covered in a few hours, against targets of a 15% leakage reduction by 2025 and halving network losses by 2045.

United Utilities partnered with Sensat in December 2021 to combine thermal data with photogrammetry, scoring vegetation stress, terrain relief and thermal signal across rural trunk mains. A PwC case study on Team UAV, serving large UK water companies, reported 75% leak detection accuracy with a tenfold efficiency gain.

Where Thermal Leak Detection Falls Short

Read those figures carefully and a pattern emerges. Thermal is a targeting layer, not a verdict.

Depth is the hard limit. Detection depends on burial depth, ground cover, soil type, soil moisture and how much water is escaping. The technique works on shallow infrastructure, broadly the first metre or so, and deeper mains often produce no measurable surface change at all. Claims of detection tens of metres down are not credible.

False positives are routine. Reflective surfaces read wrong, dense concrete mutes the signal, and variable-density materials can look damp with no water present. In vegetated ground you cannot separate a subsurface leak from natural root-zone moisture, which is why vegetation stress is a supporting indicator rather than proof. Warm humid air also reduces infrared transmission, so readings come back colder than reality.

Most importantly, thermography does not replace acoustic correlation, pressure logging or targeted excavation. Every anomaly is a candidate until something else verifies it. Where thermal earns its place is in narrowing kilometres of main down to a handful of dig sites, quickly and without closing a road.

Choosing A Thermal Drone For Leak Surveys

Three specifications matter. Radiometric output, meaning a temperature value stored per pixel in a 16-bit R-JPEG rather than a colourised picture, so emissivity can be adjusted afterwards. Thermal resolution, which sets how small an anomaly you can resolve at a given altitude. And NETD, the smallest temperature difference the sensor can distinguish, quoted in millikelvin.

Across DJI's current enterprise thermal range the NETD figure is consistent at 50 mK or better at f/1.0. The DJI Matrice 4T and DJI Mavic 3T both carry 640 by 512 thermal sensors with 12 micrometre pixel pitch, while the DJI Zenmuse H30T offers 1280 by 1024 natively. Against roughly one-degree anomalies, 50 mK is comfortably sensitive enough, so the meaningful differences between these payloads are resolution, endurance and automation rather than raw sensitivity.

FAQs

Can a thermal camera detect a water leak?

Indirectly. A thermal camera reads surface temperature, so it detects the cooling, conduction and thermal inertia changes a leak causes in the ground or structure above it. It does not see the water itself.

How deep can thermal imaging detect water?

Realistically the first metre or so, and only when enough water is escaping to change the surface temperature. Deeper mains frequently produce no detectable signature, and depth performance varies with soil type, ground cover and flow rate.

What is the best time of day for thermal leak detection?

Night or pre-dawn, in calm dry conditions. Minimal solar loading gives the sensor the cleanest temperature contrast and avoids thermal crossover, when target and background temperatures equalise.

Do UK water companies use drones to find leaks?

Yes. Anglian Water, Severn Trent, United Utilities and South West Water have all run thermal drone leak detection trials, with Severn Trent moving towards automated drone-in-a-box deployment.

Final Thoughts

The honest answer is yes, with conditions. A radiometric thermal drone flown before dawn in the right season over shallow mains will narrow a long stretch of pipe down to a short list of dig sites far faster than a crew on foot. It will not confirm a leak on its own, and it will not find deep or well-drained losses. Used as the triage layer in a survey that finishes with acoustic verification, it earns its keep and then some.

For thermal survey work of this kind, the DJI Matrice 4 Thermal (M4T) pairs a radiometric 640 by 512 sensor with the endurance a mains corridor needs, and is available now at the Coptrz official online store.