Leakage is a measurement problem before it is a repair problem. A network loses water through thousands of small, invisible openings and a few large, usually invisible, bursts; nobody sees most of them, so they are inferred from flow, pressure and sound. This guide sets out the inference chain: how minimum night flow becomes a leakage figure, how that figure splits into background and burst components, why pressure matters more than most people expect, how acoustic correlators and transient methods locate individual leaks, and how to decide when a leak is worth finding at all.
Where the water goes: the water balance
Standard accounting starts from system input volume and subtracts authorised consumption (billed and unbilled, metered and unmetered). What remains is non-revenue water, which splits into apparent losses (meter under-registration, unauthorised use, data errors) and real losses: physical leakage from mains, service connections and storage. Everything in this guide is about real losses.
A top-down balance gives a single annual figure with a wide uncertainty band, because it is the small difference between two large numbers. A bottom-up estimate from night flows in district metered areas (DMAs) gives a daily figure for each part of the network. UK practice reconciles the two: the DMA totals are summed and compared with the balance, and the discrepancy is treated as a diagnostic rather than averaged away.
Minimum night flow: the core measurement
Between about 02:00 and 04:00, legitimate demand in a residential DMA falls to a small, fairly predictable value, so the inflow at that hour is mostly leakage. Calculating it is a subtraction and a scaling:
Daily leakage = NNF × HDF, where the hour-day factor HDF = Σ24 h(Pt/Pnight)N1 converts the night rate to a 24-hour volume, allowing for the lower daytime pressure.
Legitimate night use has three parts: household use (a few occupants awake, cistern refills, appliances on timers; UK studies put the allowance at roughly 1.7 L per property per hour), non-household night use (measured or estimated per customer), and any night-filling of tanks or reservoirs inside the DMA. This allowance is the weakest link in the method; a DMA with a bakery or a hospital needs the customer metered at night, not assumed.
Worked example. A DMA of 1,500 properties with 25 km of mains records a minimum night flow of 9.0 m³/h at a night-time average zone pressure of 45 m. Legitimate night use at 1.7 L/prop/h is 2.55 m³/h, so NNF = 6.45 m³/h. From the 24-hour pressure profile the HDF is 20 (daytime pressure is lower, so each daytime hour leaks less than a night hour). Daily leakage is therefore 6.45 × 20 = 129 m³/d, or 129,000 / 1,500 = 86 litres per property per day. That is a plausible figure for an older urban DMA and would be flagged for active leakage control.
Background, bursts and the BABE split
Not all of that 129 m³/d is findable. A component-based approach (Burst And Background Estimates, BABE) distinguishes:
- Background leakage: the sum of thousands of weeps at joints, fittings and service connections, each too small to detect acoustically (below roughly 0.25 m³/h at 50 m). It is a function of pipe length, number of connections, pressure and infrastructure condition, and it is only reduced by pressure management or renewal.
- Reported bursts: visible, short-lived; their volume is run time multiplied by flow rate, and the run time is dominated by repair response.
- Unreported bursts: detectable but invisible; their run time is set by how often the DMA is surveyed. This is the component that active leakage control attacks.
The International Water Association formalised the background term as unavoidable annual real losses (UARL), the leakage a well-run network in good condition would still have:
Lm = mains length (km); Nc = number of service connections; Lp = total length of private pipe from boundary to meter (km); P = average operating pressure (m).
For the example DMA, with private pipes averaging 10 m (Lp = 15 km) and an average pressure of 38 m: UARL = (18 × 25 + 0.8 × 1,500 + 25 × 15) × 38 = (450 + 1,200 + 375) × 38 = 76,950 L/d ≈ 77 m³/d. The Infrastructure Leakage Index is current annual real losses divided by UARL: ILI = 129 / 77 = 1.7. An ILI near 1 means the only remaining lever is pressure or renewal; an ILI of 1.7 says roughly 50 m³/d of findable loss is present, which is what the survey crews should expect to locate.
Why pressure matters: the N1 power law
For a rigid orifice, flow varies with the square root of head (Torricelli), so N1 = 0.5. Real leaks do not behave like rigid orifices. Splits in polyethylene, corrosion holes with flexible surroundings and leaking joints open wider as pressure rises, so the exponent is higher:
N1 is typically 0.5 for a large hole in rigid metal, about 1.0 for a mixed network, and 1.5 or more for small leaks in plastic pipe or background leakage at joints. FAVAD (fixed and variable area discharge) formulations reproduce this by adding a term for leak area proportional to pressure.
One consequence is that pressure reduction is often the cheapest leakage reduction available. Cutting the example DMA's average pressure from 38 m to 30 m with N1 = 1.0 cuts leakage by the ratio 30/38 = 0.79, i.e. from 129 to about 102 m³/d, without finding a single leak. Burst frequency falls as well, usually by a larger proportion, which is covered in the pressure management guide.
Acoustic detection: the physics of leak noise
A leak generates broadband noise from turbulence at the orifice and from cavitation when the pressure drop is large. This noise propagates along the pipe wall and the water column, and is picked up by accelerometers on fittings or hydrophones in the water. Three methods use it:
- Listening sticks and ground microphones: a surveyor hears the leak directly. Cheap, skilled, slow.
- Noise loggers: permanently or semi-permanently deployed sensors record noise level and spread at night; a leak shows as a persistent, narrow-spread signal. They narrow the search to a few hundred metres.
- Leak noise correlators: two sensors bracket the suspected leak and cross-correlate their signals to find the time delay between arrivals, which fixes the leak position.
Correlation turns on the speed of sound in the pipe, which is far below the free-water value because the pipe wall expands under the pressure wave:
c0 ≈ 1,480 m/s (water at 20 °C); K = bulk modulus of water ≈ 2.2 GPa; D = internal diameter; E = pipe material modulus; e = wall thickness.
For a 300 mm ductile iron main (e = 7.2 mm, E = 170 GPa) the correction term is 2.2 × 0.3 / (170 × 0.0072) = 0.54, giving c = 1,480 / √1.54 = 1,190 m/s. For a 180 mm SDR17 polyethylene main (e = 10.6 mm, E ≈ 0.9 GPa) the term is 41.5, giving c = 1,480 / √42.5 = 227 m/s. The leak position measured from sensor A, with sensor spacing L and measured arrival delay Δt (A earlier than B), is:
An error in c maps directly into position error: using the iron value on a PE main would put the leak hundreds of metres out. Good practice is to measure c on site by correlating a known noise source, and to use hydrophones rather than accelerometers on plastic because the wall transmits so little energy. Plastic pipe also attenuates the signal strongly, so sensor spacing has to come down from 200-300 m on iron to under 100 m on PE.
Transient and inverse methods
A leak also reflects pressure waves. If a small, controlled transient is injected (a fast valve movement) and the pressure trace recorded at high frequency, the reflection from the leak arrives after a delay 2x/c, where x is the distance to the leak, and its amplitude scales with the leak's effective area. Joukowsky's relation, Δh = c ΔV / g, governs it, applied to the fraction of the wave the leak diverts. Inverse transient analysis fits a hydraulic transient model to the recorded trace with leak location and size as unknowns.
These methods cover long trunk mains quickly and are not defeated by plastic pipe, but they need high-rate pressure logging (hundreds of Hz), a clean injection and a well-characterised wave speed. In practice they complement acoustics on trunk mains and bulk transfers; acoustics still dominate in distribution.
The economics: when is a leak worth finding?
Active leakage control has a cost per DMA survey, and the unreported burst volume it recovers depends on how long bursts have been running. If bursts arise at a rate r (per year) with average flow q and the DMA is surveyed every T years, the average run time of an unreported burst is T/2 and the annual unreported loss is roughly r q T/2. Survey cost per year is Cs/T. Minimising total cost (survey cost plus water value V per m³) gives the economic intervention interval:
With Cs = £2,000 per survey, r = 4 bursts/yr, q = 1.5 m³/h (13,140 m³/yr each if run all year) and V = £0.50/m³ (marginal cost of water including energy and treatment), Topt = √(4,000 / (4 × 13,140 × 0.5)) = √0.152 = 0.39 yr, so the DMA should be surveyed about every five months. Where the marginal value of water is higher, for instance where it defers a resource scheme, the interval shortens; a DMA with a low burst rate can be left a year or more. This same arithmetic, run across all DMAs, is what sets a company's economic level of leakage.
Repair quality is the last term in the chain. A repair clamp on a corroding iron main buys time; the burst rate of the pipe is unchanged and the next failure is usually within metres. Once a main is generating repeat bursts, the economic answer moves from repair to renewal, which is the subject of the pipe condition guide and, for the construction method, the trenchless rehabilitation guide.
How to run a leakage assessment on a DMA
- Confirm the DMA boundary. Check boundary valves are shut (zero-pressure test or step test) and that all inflow and outflow meters are logged at 15-minute intervals or finer.
- Extract minimum night flow. Take the lowest sustained hourly flow between 02:00 and 04:00 over several nights, discarding nights with bursts, refilling or exceptional use.
- Subtract legitimate night use. Apply the household allowance and add measured non-household night use. Meter large night users rather than assuming them.
- Scale to daily leakage with the hour-day factor. Build the 24-hour pressure profile at the average zone point and apply the N1 power law to convert night leakage to a daily volume.
- Compute UARL and ILI. Use mains length, connections, private pipe length and average pressure to get UARL, and compare with current losses to see how much is findable.
- Decide the intervention. Compare the economic survey interval with the time since the last survey; consider pressure management if the night pressure is high and the ILI is near 1.
Frequently asked questions
What is a good Infrastructure Leakage Index?
An ILI below 2 indicates a well-managed network where further reductions depend on pressure management or renewal rather than finding leaks. Values of 2 to 4 suggest findable bursts are present and active leakage control pays. Above 4, the network has systemic problems and the economics usually favour both intensive detection and pressure reduction.
Why does leakage rise faster than the square root of pressure?
Because most leak openings are not rigid. Splits in plastic pipe, gaps at joints and holes surrounded by corroded metal widen as pressure increases, so the discharge area grows with pressure as well as the velocity through it. The combined effect gives N1 exponents of around 1 for mixed networks and 1.5 or more for plastic pipe and background leakage.
Why is leak noise correlation harder on plastic pipe?
Polyethylene is roughly 200 times less stiff than iron, so the speed of sound in a PE main falls to around 200-300 m/s and the wall transmits very little of the leak noise. Signals attenuate quickly, so sensors must be closer together, hydrophones in the water are preferred to accelerometers on fittings, and the wave speed should be measured on site rather than taken from tables.
What is legitimate night use and how is it estimated?
Genuine customer demand during the minimum night flow period. UK practice allows about 1.7 litres per household per hour for domestic use, adds measured night consumption for non-household customers, and accounts for any tank filling. It is subtracted from minimum night flow to give net night flow, which is attributed to leakage.