water · infrastructure · family: the wrong ruler
leak metrics that liewhen taps run dry
Under intermittent water supply, the standard water-loss metrics reward utilities for supplying less — so nobody can tell whether loss reduction is working
Problem statement
Non-revenue water (NRW) — the difference between water put into a distribution network and water billed to customers — is treated as an indicator of how efficiently a water utility operates. In cities where water is supplied intermittently (a few hours a day or a few days a week), the metric breaks: leakage only happens while pipes are pressurized, so the volume of NRW rises and falls with how much water the utility manages to supply. A utility that pumps less (because of fuel shortages, dwindling water resources or war) reports a "better" NRW percentage; a utility that succeeds in extending supply hours toward continuous service reports "worse" NRW. The performance indicator therefore cannot distinguish genuine loss reduction from a shrinking service, and no accepted normalization yet lets an intermittently supplied system be benchmarked against others with reasonable accuracy.
Why this matters
Global NRW is estimated at 126 billion cubic metres per year, costing roughly US$39 billion annually, and the level of leakage is described in the source as "likely the most important single indicator of the efficiency of water utilities perceived by regulators, the public and the media." When the yardstick is wrong, money and blame are misallocated: the paper's own case study, Sana'a (1.32 million consumers, water table falling 6–8 m a year), showed reported NRW "improving" from a ten-year average of 35% of system input (2005–2015) to 22% in 2015 — a year in which conflict-driven fuel and electricity shortages cut the amount of water supplied. Conversely, utilities that do the right thing — moving toward 24/7 supply — see their indicators climb, and the paper says political pressure may be "subjecting NRW management to failure in situations where certain measures are not at fault." Utilities and projects are typically evaluated on preset criteria that include the NRW level, so a broken metric can end good programs and reward bad ones.
What’s been tried and why it hasn’t worked
The IWA standard water balance and its performance indicators (NRW %, litres per connection per day, Infrastructure Leakage Index) are the recommended best practice, but their volume-based values move with the amount of water supplied; the authors note the supply-volume dependence "appears critical and intuitive" yet "has not been recognised in the literature." The obvious fix — the "when-system-is-pressurised" (w.s.p.) adjustment that scales losses by average supply hours — has been used for real-loss indicators, and the paper extends it to NRW and apparent losses, but finds three failure modes: it over-estimates apparent losses (once demand is met, extra supplied water becomes leakage, not theft or metering error, so scaling both alike is wrong); it is acutely sensitive to the average supply time, whose uncertainties "significantly undermine the accuracy" — and "for water systems with a Tavg of less than 8 h/day, the results of this approach become more uncertain"; and it is likely biased toward systems with rising supply. Regression of NRW volume against system input volume, the paper's alternative, tracks a single system's trajectory well but "can only be used for monitoring the NRW for individual systems, and not for a comparison of different systems." The authors conclude that "comparing and benchmarking a water supply system to other systems with reasonable accuracy does not appear to be possible" today. Underneath all of this sits a data problem: average supply time is itself uncertain to calculate (in Sana'a it would have to be estimated for each of 369 distribution areas from valve-closure and pumping records that are not currently available), and customer meters themselves misread under intermittent supply (air expelled through meters when pipes refill, and low-flow filling of customers' storage tanks), so both the numerator and the denominator of the indicator are uncertain.
What would unlock progress
Progress needs a benchmarking framework built for intermittency rather than patched onto continuous-supply metrics: a correction-factor curve for average supply time (the paper's explicit open item), an apparent-loss treatment that does not scale with pressurized hours, and cheap ways to measure actual pressurized time per zone (pressure loggers, smart-meter timestamps) rather than infer it. The adjacent precedent is normalization in other utility benchmarking — e.g., weather-normalized energy consumption or capacity-factor-adjusted plant performance — where an uncontrolled driver is modeled out before comparison. Regulators and lenders then need to adopt the normalized indicator so that transitioning to 24/7 supply is not punished.
Entry points for student teams
A team could take published or partner-utility monthly data (system input, billed consumption, supply hours by zone) and build an open normalization tool that reports both raw and w.s.p./regression-adjusted NRW, propagates uncertainty in supply time, and shows where the <8 h/day cliff makes results unreliable — the paper's Sana'a analysis gives a template (its monthly data were obtained from the Sana'a water utility and appear in the paper as charts and summary tables). A second team could design and field-test a low-cost pressurized-hours logger for district metered areas and quantify how much better-measured supply time tightens the indicator. A policy-analysis team could review how development-bank NRW covenants and national regulators score intermittently supplied utilities and propose a benchmarking protocol. Skills: hydraulic engineering, statistics/uncertainty analysis, embedded sensing, water-utility regulation.
Genes
Structural cousins: stuck the same way, in other fields
Sources
"Monitoring Nonrevenue Water Performance in Intermittent Supply," T. AL-Washali, S. Sharma, F. AL-Nozaily, M. Haidera, M. Kennedy (IHE Delft / TU Delft / Sana'a University), *Water* 11(6):1220, 2019, (open-access copy: ), accessed 2026-08-17 go to source 1 ↗ go to source 2 ↗
Working notes
Notes made while writing and checking this brief, shown as written. They use working shorthand.
Primary source is a peer-reviewed methods paper (tier 1 research paper, open access) from IHE Delft's water-loss group, with the limitation and open-item statements taken from its Conclusions; the global NRW figure is the paper's citation of Liemberger & Wyatt (2018) and should be verified against that source. `failure:ignored-context` (data/information sub-pattern) chosen because the IWA indicators were designed without accounting for intermittently pressurized networks — not `failure:wrong-problem`, since the objective (reduce losses) is right; only the measurement basis ignores context. `constraint:data` is the binding constraint (supply time and meter accuracy are unmeasured), and `constraint:coordination` was ruled out — utilities and regulators agree on the goal, and no coordination failure blocks a better indicator; the method simply does not exist. Related collection briefs: `water-distribution-gradual-leak-detection` and `water-aging-pipe-network-failure-prediction` address finding physical losses; this brief is the distinct problem of measuring loss performance under intermittency. The customer-meter accuracy strand (air over-registration during network filling; under-registration from low-flow filling of customer storage tanks) is documented in the IWA journal AQUA (Ferrante et al. 2022, vol. 71 no. 11) but full text was not accessible at intake — flagged for follow-up as a possible companion brief.
Source type: Self-articulated (researchers naming an unresolved methodological gap in their own field).
Verified at intake 2026-08-17: gate (net) + adversarial source check + contested-tag second coding.
Related collection briefs (distinct sub-problems, cross-referenced at intake 2026-08-17): `water-south-africa-missing-wastewater-reticulation-metric`.
Reconciliation 2026-10-01: The whole write-up was read against its sources. Read in full: the paper on the Source line, from the TU Delft repository copy at https://pure.tudelft.nl/ws/files/55333433/water_11_01220_v4.pdf (fetched 2026-10-01, all 14 pages; the publisher page behind the DOI refused the request, so the citation was checked against the Crossref record at https://api.crossref.org/works/10.3390/w11061220). Also read in full: Ferrante, Rogers, Mugabi and Casinini, "Impact of intermittent supply on water meter accuracy," AQUA 71(11):1241, 2022, https://doi.org/10.2166/aqua.2022.091, from the Internet Archive copy of the publisher's PDF (fetched 2026-10-01, 10 pages); the note above says its full text was not accessible at intake. For Liemberger and Wyatt only the Crossref abstract was read (https://api.crossref.org/works/10.2166/ws.2018.129). Changes: (1) The Problem Statement said the metric is judged by "regulators, lenders and the public" and "was defined for networks that are pressurized 24 hours a day," and spoke of "the many cities" with intermittent supply. The paper's sentence about regulators, the public and the media is about the level of leakage, not NRW as a whole (it is quoted correctly under Why This Matters), and it does not mention lenders; what the paper says of NRW is that it is considered an indicator of operational efficiency. The paper says nothing about how the metric was originally defined and gives no count of cities. The sentence now says NRW is treated as an indicator of how efficiently a utility operates, and the other two statements were removed. (2) "Drought" was listed as a reason a utility pumps less. The paper speaks of dwindling water resources in the Sana'a Basin, not drought; the text now says that. (3) The text said no accepted normalization lets an intermittent system "be tracked over time or benchmarked against others." The paper says regression analysis does track a single system over time; it is comparison between systems that "does not appear to be possible." The sentence now speaks only of benchmarking. (4) "Intermittent supply is the norm across much of South Asia, the Middle East and Africa" is not in any source named here and was removed. (5) The text said the 2015 shortages "simply cut production, not losses." The paper says they reduced the amount of water supplied; it does not say losses were unchanged (its regression puts 2015 losses, at 2009's production level, 4% above 2009, and it also reports the utility's loss-control actions in 2015). The sentence now says the shortages cut the amount of water supplied. (6) The text said "utilities and donor projects" moving to 24/7 supply "are penalized," as a fact. The paper does not mention donor projects, and says political pressure "may be" subjecting loss management to failure; the sentence now follows the paper, and the quotation is unchanged. (7) "Loss-reduction programs are typically evaluated on preset NRW criteria" became the paper's statement: utilities and projects are typically evaluated on preset criteria, among which the NRW level is important. (8) The text said the IWA indicators "are widely used but assume continuous pressurization." The paper calls them recommended best practice and shows that the volume-based ones vary with the amount of water supplied; it does not say they are widely used or what they assume. The sentence now says what the paper says. (9) "It is biased toward systems with rising supply" became "likely biased," the paper's wording. (10) The text said average supply time "varies by zone and week, is rarely metered." The paper says calculating the average supply time has uncertainties for intermittent systems and that in Sana'a it would have to be estimated for each of 369 distribution areas from valve-closure and pumping records that are not currently available; the sentence now says that, with the 369 areas and the missing records kept to Sana'a. (11) The meter errors were described as "air passage and trickle-filling of roof tanks," and in the notes as "float-valve roof tanks." Ferrante et al. describe air expelled through customer meters when pipes refill (over-reading) and, citing earlier work, under-reading from low-flow filling of the storage tanks customers install; they do not mention roofs or float valves. Both places now use the paper's terms. (12) "Regulators and lenders (who set the NRW targets)" lost the parenthesis, which no source here supports. (13) The entry point said "the paper's Sana'a dataset gives a template." The paper says its monthly figures were obtained from the Sana'a water utility, and it shows them in charts and summary tables; no separate dataset was found in the copy read. The sentence now says the paper's analysis is the template and says where the data came from and how they appear. Confirmed in the sources and left as written: the Source line (authors, title, journal, volume, issue, article number, year, affiliations); 126 billion cubic metres and US$39 billion a year (in the paper and in the Liemberger and Wyatt abstract); 1.32 million consumers; the water table falling 6–8 m a year; 35% of system input as the 2005–2015 average and 22% in 2015; the three limits of the pressurised-hours adjustment; the 8 h/day threshold; the correction-factor curve as the paper's open item; and all eight quotations, each found word for word in the paper (two begin with a capital letter there). Not checked: nothing in the body rests on a source that could not be read. The titles were not changed. Genome tags untouched. A second reader went over these repairs against the same sources on the same day; all thirteen stand, four of them (1, 10, 11 and 13) with the wording tightened as described above.