Leakage estimation in developing country water networks based on water balance, minimum night flow and component analysis methods

In many countries significant amounts of drinking water are lost in the water supply system. To reduce such losses, water utilities should first identify how much is lost, where and why the losses are occurring, and how they can be reduced (Knobloch & Klingel 2013). Three main methods are used to determine how much water is being lost: the water balance – or top-down – method (Knobloch et al. 2014), the minimum night flow (MNF) or bottom-up method (García et al. 2006; Tabesh et al. 2009; Xin et al. 2014), and component analysis, also known as the Burst and Background Estimates (BABE) method (Lambert & Morrison 1996; McKenzie 2003; McKenzie & Seago 2005; Fanner & Thornton 2005; Lambert 2009).

Several studies have been conducted to determine water losses (WL), set water loss reduction targets and determine key performance indicators. In a study to develop a method for estimating the components of apparent losses (AL), Xin et al. (2014) used MNF so as to be able to investigate the total AL. Vučijak et al. (2013) studied WL in five water utilities in Bosnia-Herzegovina, and found leakage of between 25% and 60% of system input volume (SIV), using the water balance method. In Pirot, Serbia, leakages make up 80% of non-revenue water (NRW), using the water balance method (Radivojević et al. 2008).

In Blantyre, Malawi, Chiipanthenga (2008) found that 72% of the total losses were leakage, using the water balance method. Akita (2009) showed that 44% of NRW in Kampala, Uganda was from bursts and leaks, using both water balance and MNF methods. Fanner & Thornton (2005) compared the water balance and BABE methods in a network in a city (Salt Lake) in Utah State, USA. They realized that the amount of leakage is sensitive to the infrastructure condition factor (ICF) and emphasized that ICF must be below 2.0 if the best estimate of real losses from the water balance is used.

While some of these studies were carried out in developing countries, none compared all three methods, so they could not point out the challenges of measuring water loss components. Neither did these studies investigate the suitability of the methods for estimating leakage in water networks in developing countries. The estimates discussed above were all aimed at determining leakage as a first step to finding the major components of NRW. The influence on the procedure of some of the parameters used was not reported. For instance, meter inaccuracies due to overhead tanks in high-rise buildings would affect the water balance method, as would customer night use in MNF, or burst flow rates and ICF in the component analysis.

Factors affecting the accurate estimation of leakage in water networks have been investigated in a few studies (Fanner & Thornton 2005; García et al. 2006). Fanner & Thornton (2005) found that the estimated volume of real losses in component analysis is sensitive to the ICF value used. Therefore, the authors recommend that ICF should be determined for every network for accurate leakage estimation. They also showed that the infrastructure leakage index (ILI) could be used in the absence of an ICF factor. Thus, assuming an ICF value, customer night use, _, recommended or determined for other networks does not depict the actual situation. ICF has not been investigated nor ILI accurately estimated for networks in most developing countries.

García et al. (2006) revisited the MNF method and investigated the sensitivity of leakage estimates to the average zone pressure, the leakage exponent and the MNF hour, to ascertain the uncertainty associated with the method. Some uncertainties were found and this led to investigation of the parameter in MNF in developing country water networks (Amoatey et al. 2013). was identified as crucial to accurate leakage estimation, leading to its modification to make it suitable for a water network in Ghana. This was first piloted (Amoatey et al. 2013) and further applied to the entire network (Amoatey et al. 2017).

Some studies combining leakage estimation methods as a basis for understanding local situations have also led to development of location-specific software (McKenzie & Seago 2005; Tabesh et al. 2009; Tsitsifli & Kanakoudis 2010; Knobloch et al. 2014). To determine the water loss components, Knobloch et al. (2014) developed software based on automated zonal water balance for a network in Pforzheim, Germany. Tabesh et al. (2009), and Tabesh & Asadiani Yekta (2005) developed integrated software using water balance and MNF concepts, together with Geographical Information Systems (GIS) modeling, based on their Iranian water network situation.

The concepts behind the water balance, MNF and BABE methods, have been used to develop various software tools by several institutions for determining water balances and computing performance indicators (McKenzie & Seago 2005; Tsitsifli & Kanakoudis 2010). These include BENCHLOSS, BENCHLEAK, FASTCALC, and AQUALIBRE, among others. Such tools and software have been helpful for water networks where accurate data are well documented. Use of them is cumbersome, however, for water networks in developing countries where few data are available or various parameters have not been established. Often, the results of using these tools to estimate leakage may not be a true reflection of the situation in the network.

The data-intensive nature of using any of the three methods cannot be over-emphasized. The water balance method requires accurate documentation of SIV, authorized consumption (billed and unbilled), field investigations of customer accounts (active and inactive), and billed records, as well as meter readers’ records, to establish AL components (Tabesh & Asadiani Yekta 2005; Chiipanthenga 2008; Farley et al. 2008; Radivojević et al. 2008). Real Losses (RL) are estimated by multiplying the number of leaks by their duration and burst flow rates (McKenzie 2003; McKenzie & Seago 2005; Lambert 2009; Tabesh et al. 2009). For networks in most developing countries, burst flow rates have not been established and so the International Water Association (IWA) recommended rates are used, which can influence the accuracy of the leakage estimates significantly.

• is night leakage rate [m³/hr]

• is flow rate into a District Metered Area [m³/hr]

• is customer night use [m³/hr]

  

  (2)

  where

• is average daily leakage rate [m³/day]

• is night leakage rate [m³/hr]

• is the hour-day factor [ - ]

• is night use for customers who use WC [m³/h]

• is night use for customers who do not use WC [m³/h]

In this study, the results of the modified MNF procedure are compared with those from the water balance and component analysis methods.

Finally, component analysis (BABE) splits total leakage into bursts and background leakage. It models leakage volumes based on the nature of leak occurrences and durations (Lambert & Morrison 1996; Lambert 2009). The burst element is the product of the number of bursts or leaks on each pipe category (transmission, distribution mains, and service connections), the burst flow rates for each pipe category and the burst durations. Background losses are estimated by computing the unavoidable background leakage and multiplying by the ICF (Lambert & Morrison 1996; Fanner & Thornton 2005; Lambert 2009).

• is unavoidable background leakage [m³/h]

• is reported bursts [m³/h]

• is average leak runtime [h]

• is background leakage [m³/h]

• is leakage exponent [ - ]

• AZNP is length of mains [km]

• ICF is infrastructure condition factor [ - ]

• is length of mains in [km]

• is number of service connections [no.]

• is length of private pipes [km]

• is operating pressure [m]

While the three methods can be used to complement or verify each other, no study has compared all three, although some have compared two – e.g., Akita (2009), Fanner & Thornton (2005), Tabesh & Asadiani Yekta (2005). Comparison is necessary to ascertain the methods’ suitability for networks in developing countries where very few data exist, and factors such as burst flow rates, ICF, and/or customer night use have not been established. Where these factors have been established – e.g., in developed countries – their use is helpful. In developing countries, however, they might not give a true reflection of the situation in the water networks. A further aim of a three-way comparison is to identify the factors influencing the differences in the methods when applied to networks in developing countries.

The components of losses in the IWA standard water balance were computed using the volumes of water supplied to the network and sold, giving the NRW. In Baifikrom, Ghana – the case study network – billed metered consumption (BMC) is 85% of billed authorized consumption (BAC), with the other 15% being billed unmetered consumption (BUC).

Unbilled Authorized Consumption (UAC) is provided as the volume of water used in treatment plant processes like backwashing, and water for firefighting. UAC has two components unbilled metered (UMC) and unbilled unmetered Consumption (UUC). UAC is equal to UUC. The real loss component is calculated as the product of the number of bursts, burst durations and burst flow rates.

The modified MNF method (Amoatey et al. 2017) accommodates customers who do not use toilet flushing sanitation facilities. The estimated values for the two customer categories were computed from standard water consumption rates in Ghana (Adombire 2007), and sanitation facilities records were obtained for the case study network. The night leakage was measured and the HDF was estimated from pressure measurements. These measured and estimated parameters were used in Equations (1)–(3) to obtain leakage values.

Finally, Equations (4)–(8), which describe the BABE concept, were used to estimate leakage. IWA standard burst flow rates and a recommended ICF were used, as local rates have not yet been established (Akita 2009; Lambert 2009). The number of bursts recorded monthly per pipe category (mains and service connections), and the burst durations were used to estimate background leakage and bursts.

The case study used the Baifikrom Water Supply Network in the central region of Ghana (Figure 1). The network supplies an approximately 250 km² rural-urban area with a population of about 122,000 in about 10,000 households (AVRL 2008). The pipes are made of asbestos cement (AC), high density polyethylene and polyvinyl chloride, with a total length of about 150 km. Pipe sizes range from 75 to 500 mm (AVRL 2008). There are about 5,700 registered customers, of whom 85% were metered and billed monthly in 2013. Some areas are served by public standpipes, which are also metered. Per capita consumption varies between 30 and 75 L/day (Adombire 2007), and the total daily supply is about 5,400 m³.

The proportion of the daily supply volume lost as leakage was determined using the three methods. For the water balance method, monthly records of SIV and water volumes billed were evaluated. For MNF, flow meters and data loggers were installed in the network, and both flow and pressure were logged at 15-minute intervals over 4 months. The MNF rate was calculated from the 15-minute averages over the entire measurement period. In addition, standard average consumption figures for toilet flush, shower and faucet use in Ghana were used to compute (Adombire 2007).

In applying BABE, the numbers of leaks and bursts recorded were obtained from the water utility. In the absence of established burst flow rates, the IWA recommended standard leak flow rates of 6 and 12 m³/h were used respectively for bursts occurring on distribution mains and service connections to estimate annual leakage losses (Fanner & Thornton 2005; Lambert 2009; Melato et al. 2009). An ICF of 1.5 was chosen based on Akita's recommendations (2009). As a matter of policy, burst duration in the case study network is 48 hours, so burst losses were estimated the average number of bursts, the standard burst flow rates and burst durations.

The results of the water balance and modified MNF were similar, while BABE yielded higher values. As most of the parameters used in the modified MNF method were measured or estimated based on field data, it is considered the most suitable for developing country networks, provided that flow and pressure can be logged in the network.

The amount of water produced and amount of water sold (authorized consumption) was used to determine NRW. As noted above, 85% of BAC gave the BMC. The utility's burst records were used to determine the value of RL as no data exist on invisible or background leakage. The details of the computed components of the water balance, with the data available, are presented in Table 1.

RL, mainly visible leaks repaired by water utility staff, make up 10% of the SIV. As there is no active leakage control to identify and locate undetectable leaks, background leakage is not accounted for in the standard water balance table above – i.e., for this study. This implies that leakage may be under-estimated.

Night leakage was obtained from measured night flows (Figure 2). HDF was estimated from network pressure readings, while was estimated for the two night user categories identified. The MNF for the measurement period is 40 m³/hr and occurs between 03:00 and 04:00. The area under the curve is approximately 5,400 m³ and represents the average total daily flow volume. The estimated values for the two categories and the total leakage – approximately 30.2 m³/hr – are shown (solid red lines) in Figure 2.

The MNF results are summarized in Table 2. Leakage is estimated to be about 11% of the daily input volume. The method accounts for background losses as well as bursts as the minimum night measurement should capture all leaks running in the network. Comparing the MNF and water balance results, it can be seen that background losses are 1% of the SIV. These relatively low losses suggest that the network is in quite good condition. Whether this is realistic for water networks in developing countries is a subject for further investigation – e.g., a similar comparison in a network with similar characteristics, so that background losses are not underestimated.

The method used to compute bursts for the BABE method is described above. Unavoidable background leakage was estimated and multiplied by the ICF (1.5). The leakage estimate is presented in Table 3 and suggests that approximately 18% of SIV is leakage (Table 3). This is considerably higher, 7 or 8%, than estimated by the other two methods, and further studies are needed to ensure that it is not an overestimate.

The BABE method is expected to be quite accurate as it relies on network parameters such as numbers of customers, main and private pipe lengths, ICF, burst flow rate and the system pressure in the network. This implies that inadequate or inaccurate network parameter records influence the leakage estimate, so accurate records must be kept if leakage reduction targets are to be set (Fanner & Thornton 2005; Melato et al. 2009). The relatively high levels of background loss may arise because parameters such as burst flow rates and ICF were not determined for the case study network but assumed from IWA recommended ranges.

However, since burst flow rates were also used in the water balance method and the difference in the leakage estimate compared to that from the modified MNF method was quite small, ICF is likely to be the most sensitive parameter in the BABE methodology. Equally, as the Equations (4)–(8) for estimating unavoidable background leakage and subsequently background leakage were developed empirically from well-maintained water networks in developed countries with different network operating conditions and usually higher operating pressures (Mutikanga 2012), the equations may need calibration to improve their accuracy and applicability to water networks in developing countries.

While the objective of this paper is to compare leakage estimates, it is clear that AL is relatively high in the case study network and this requires urgent attention to save water for areas not served. High AL in developing countries’ water networks have also been reported by Mutikanga et al. (2010).

The study was an investigation of how well three leakage estimation methods – water balance, modified MNF and BABE – compute RL in water networks, especially in developing countries. Since all parameters used in the modified MNF method are measured and estimated, it should yield the most representative estimate for the case study network. This must be repeated for similar water networks in Ghana to establish whether the water balance method is reasonably accurate and whether, in all cases, the percentage difference when compared with the MNF will be as low. On the other hand, the difference between the water balance and modified MNF methods as against the BABE method is significantly higher and this can be ascertained when further studies from similar water networks is carried out.

Burst flow rates and ICF have been identified as the factors accounting for the difference when the methods are compared. The extent to which burst flow rates and ICF influence the amount of leakage estimated in the case study network will be studied further. As BABE is based on empirical equations, it is suggested that the equations should be calibrated using local data. This is likely to improve the accuracy of the leakage estimate.

It can be concluded from this study that the modified MNF method is suitable for developing country networks. In networks where flows and pressures cannot be logged, the water balance method can be used. Network-specific burst flow rates and ICFs should be established for developing country water networks where it is intended to use BABE, and the equations used should be calibrated.

The findings of this study have implications for further research, as well as for water utility policy and practice. For research, the sensitivities of all factors should be investigated, to support water utility policies. Similar comparative studies should be carried out in similar networks, to establish the background loss behavior in the three estimation methods and determine their suitability for water networks in developing countries.

With respect to water utility policy and practice, proper documentation of network details as well as investment in the establishment of local burst flow rates should be encouraged. Burst duration should be reduced to minimize leakage. Moreover, improvement in metering and billing policy is required to reduce the high levels of AL.