Delayed pipe replacement halves environmental impacts but quadruples water loss

Scheduling pipe replacement is critical for water distribution systems (WDSs) when managing finances and water loss. WDS replacements are often delayed due to high immediate costs without considering long‐term environmental consequences. This study is the first to examine a real‐world WDS using a novel workflow transferrable to other WDSs that integrates GIS, hydraulic modeling, breakage prediction, and life cycle analysis to evaluate environmental impacts and water loss of five replacement schedules (25‐, 50‐, 75‐, 100‐, 150‐year intervals). Environmental impacts were reduced by half when replacement interval changed from 25 year to 150 years, yet volume of water leaked from the system quadrupled. Benefits plateaued beyond 50–75‐year replacement while water loss steadily increased. Lowering water loss through break management enabled one‐sixth pipe replacement without exceeding original leakage at 25‐year replacement. Results were robust to uncertainty parameters and assert the importance of equilibrating environmental impacts and water loss when designing pipe replacement frequency.

Overdue replacement of pipes leads to more breaks in the system, requiring more pumping and water treatment to meet the water demand of the communities served by the system.This impact is significant, as the number of annual breaks in piped systems increases exponentially with time, sometimes from 0 to over 300 in a span of 70 years depending on the characteristics of the WDS (Mailhot et al., 2000;Pelletier et al., 2003;Shamir & Howard, 1979).
As water pumping is the most energy-intensive activity in water treatment systems, accounting for 55%-90% of the total energy consumption (Pabi et al., 2013), increased breaks and water losses in WDSs may accumulate significant environmental impacts over time.In order to appropriately manage challenges such as this, many researchers use life cycle analysis (LCA) as a decision-making tool to allow utilities to make environmentally informed decisions.In drinking water treatment, LCA studies have investigated treatment processes, such as biofiltration (Jones et al., 2018), membrane filtration (Prézélus et al., 2021), water reuse (Ortiz et al., 2007), and the overall treatment process (Bonton et al., 2012;Lundie et al., 2004), but LCA studies on drinking water distribution systems are lacking.While previous studies have revealed the potential importance of pipe replacement in the environmental footprint of WDSs, they are typically inadequate in coverage or sophistication.For example, researchers have used LCA to evaluate the impacts of using different pipe materials or implementing pipe rehabilitation options (Basupi et al., 2014;Hajibabaei et al., 2018;Herstein & Filion, 2011;Parvez, 2019;Vahidi et al., 2015).While these studies show the impacts of specific operational decisions in the context of cost, greenhouse gas emissions, and embodied energy, they neglect timedependent leakage and hydraulic effects in the system, providing no information on pipe replacement's relationship to water loss and resulting environmental impacts.Others have begun to generate time-dependent estimations of the total cost or energy usage of certain aspects of WDS operation (Ambrose et al., 2008;Filion et al., 2004;Lee et al., 2018;Prosser et al., 2013).However, these studies employ oversimplified breakage models that either cannot extrapolate to the entire system or fail to consider variations in pipe material, diameter, length, and other factors (Barton et al., 2022;Vladeanu & Koo, 2015).
This study expands on previous work by taking a novel approach of incorporating a combination of hydraulic modeling, leakage prediction, and real-world WDS information into the LCA (EPA, 2020;ESRI, 2022;GreenDelta, 2018;Mailhot et al., 2000;Python Software Foundation, 2016;Schwaller & van Zyl, 2015).A cradle-to-grave impact analysis was conducted on a WDS in Alabama, USA.To respect the anonymity of the utility, any references to the system will remain deliberately vague throughout this article.To improve upon previous model oversimplifications, this study first used real-world GIS information, which considered elevation, spatial distribution of water users, and WDS construction (e.g., portions covered by paved areas) at a high spatial granularity.The GIS information was then inputted into EPANET 2.2 computational modeling software, to generate time-based discretization allowing for improved accuracy in assessing the time-dependent leakage and hydraulic effects.While there are computationally more efficient models (Paluszczyszyn et al., 2015), EPANET was selected due to its open access, ease of use, and flexibility to model hydraulics according to the Darcy-Weisbach, Hazen-Williams, and Manning's mathematical frameworks (Schmid, 2002).Even so, utilities desiring to implement the methodology of this article in their own system may substitute their existing or preferred hydraulic model in place of EPANET.The demand values used in EPANET were then corrected for leakage as predicted by the Weibull-exponential statistical model, which was the best at predicting the number of discoverable leaks (breaks) as a function of pipe age in systems with leakage history not extending to pipe installation (left-truncated) such as the WDS in question (EPA, 2020;Mailhot et al., 2000).The output from the models was then merged with and validated against additional real-world data from the Alabama WDS to ensure that the findings agree with actual operating data and reflect the variability of system characteristics.

Article Impact Statement
Environmental benefits from replacing pipes infrequently plateau beyond 50-75 year replacement; leakage steadily increases.Managing leakage allows significant reduction in pipe replacement.category at different replacement frequencies was juxtaposed via one-at-a-time sensitivity analysis.
To the best of the authors' knowledge, this study is the first to present a comprehensive, time-dependent LCA considering leakage with advanced yet accessible statistical and hydraulic modeling at the scale of a WDS while retaining the accuracy of modeling each pipe as an individual data point.The results have been verified for confidence through validation against real-world operational data and a sensitivity analysis, making them directly usable to support real-world WDS decision-making in pipe replacement, break repair, and tap water production.The methodology is expected to be transferable to other WDSs where lefttruncated leakage history and basic pipe network information are available.

| Scope and functional unit
This study utilized OpenLCA v1.11 software and the TRACI 2.1 impact assessment method.The EcoInvent v3.7 and World Steel 2020 databases supplied the background processes necessary to conduct a cradleto-grave analysis of the WDS in question.Figure 1 illustrates the overall breakdown of the WDS into the flows and sub-processes that were used for impact quantification.
The distribution system in question is a subset of a much larger distribution system.Lack of an existing hydraulic model for the system necessitated the selection of a subsection that could be hydraulically isolated from the rest of the system.For utilities seeking to replicate this study, the methodology is scalable and transferable, but hydraulic models for large systems take significant time to develop if one does not currently exist.The subsection in question has about 25.2 km (15.6 mi.) of cement-lined ductile iron pipe (DIP) mains (>63.5 mm diameter) and about 1.7 km (1.1 mi.) of polyvinylchloride (PVC) service lines (63.5 mm diameter) with an average pipe age of approximately 40 years.Pipes in this section range in diameter from 63.5 to 300 mm (2.5 00 to 12 00 ).
The analysis was based off a functional unit of delivering the required water flow of 387 m 3 /d ($0.102 mgd) through the WDS over 300 years.The flow was the reported base water consumption of the distribution system in Alabama, USA, according to data collected in May of 2022.While population may increase in the area of interest over the 300-year timeframe, the corresponding increase in demand would have to be met with construction of new infrastructure falling outside the scope of analysis.The demand was kept constant in order to isolate the variable of pipe replacement, but the effect of changing this variable is investigated in the sensitivity analysis.The demand value is shared by 189 connections, each consuming approximately 1.42 lpm ($0.376 gpm) for 300 years.The number of connections was determined by overlaying the distribution system with the Microsoft Building Footprints layer in ArcMap and using satellite imagery to determine which structures were served by the system (Microsoft Corporation, 2018).All structures were single-family homes except for a public pool and a duplex, but due to the small physical size of the exceptions, they were deemed to generate equivalent water demands to the rest of the structures.Thus, all structures were assumed to demand the same amount of water per unit of time.
In order to assess the trade-offs between system leakage and pipe replacement, pipe replacement intervals of 25, 50, 75, 100, and 150 years were used.The replacement frequencies were selected based off the national average age of pipe failure, 50 years, the theoretical life span of cement-lined DIP and PVC pipes, 100+ years, and the national average replacement frequency, 125 years (Folkman, 2012(Folkman, , 2014;;Muster et al., 2011).

| Pipe replacement
Pipe replacement is the complete replacement of the pipes in the WDS at designed life spans.This regularly scheduled replacement is being evaluated as a proactive approach to leakage mitigation rather than the reactionary alternative of simply repairing pipe breaks as necessary.It includes materials extraction and processing, pipe manufacture, transport, placement of pipes in the ground, and the restoration of the site to original conditions after the pipe has been buried.Smart pipe renewal is an alternative to pipe replacement, but life cycle assessments of pipe rehabilitation strategies are available elsewhere and are outside the scope of this study (Kaushal & Najafi, 2020;Loss et al., 2018).The amount of pipe required is directly dependent on the replacement frequency.For example, replacing pipes every 150 years would require two sets of pipes in the 300-year time period.The amount of pipe required to distribute water for 300 years was calculated using Equation ( 1): where i is a counter for each subset of years included in the replacement period (i.e., 0-25, 25-50, 50-75, 75-100, and 100-150 years), n i is the number of years in the subset of years i, j is a counter for each pipe segment, and l j is the length of each pipe.PVC pipe is transported an average of $133.3 km (82.8 mi.) from the manufacturer and cement-lined DIP is transported an average of $32.2 km (20 mi.) from the manufacturer based on real-world mapping data.A weighted average based on each pipe material's length in the system yielded an average distance of transport of approximately $32.5 km (20.2 mi.).All transport distances are specific to the Alabama WDS, but extremes on the national scale were considered in the sensitivity analysis.
The weight of materials required for manufacturing or recycling of pipe is calculated using Equation (2): where W mat is the weight of material, l mat is the total length of pipe of a certain material and diameter needed in the system over 300 years, δ is the weight per unit length for the pipe in question as provided in Table S1 in the Supplementary Materials (SMs), and ε is the recycling efficiency of the material in question (replace with 1 when not a factor such as in pipe manufacture).
Excavation for pipe replacement or repair is calculated using Equation (3): where V exc is the volume of soil to be excavated in m 3 , D j is the diameter of each pipe in m, and d j is the average burial depth of the pipe in m (a value of $0.91 m (3 ft.) was used as this is the minimum depth of cover required for DIP by the manufacturer, and although PVC requires less cover, connectivity between the two types of pipe must be maintained) (American Cast Iron Company, n.d.).
The trench width is assumed to be $0.91 m (3 ft.) wider than the diameter of the pipe in accordance with trenching details from the Alabama Department of Transportation (ALDOT, 2008).When determining the amount of grassy area or paved area disturbed by excavation, Equation (3) was modified by replacing burial depth with percent grass cover or asphalt cover.From an overlay of the system with roads in the area, 20% of the system was found to be paved.
The amount of asphalt needed to repave the site is calculated using Equation ( 4): where W Asphalt is the weight of asphalt needed in kg, A paved is the percent of total area to be paved in m 2 , d asphalt is the depth in m of asphalt to be paved, and γ is the compacted specific weight of asphalt in kg per m 3 .When determining the amount of seed needed, the depth and specific weight terms in Equation ( 4) were collectively replaced with the amount of seed in kg per m 2 and the paved area replaced with seeded area.The depth and specific weight of asphalt were $0.15 m (0.5 ft.) and $2531 kg per m 3 (158 lb per cu.ft.), respectively (Alabama Asphalt Pavement Association, 2022; ALDOT, 2019).The amount of seed needed was $2.8 Â 10 À3 kg per m 2 (25 lb per ac) (USDA Forest Service Northern Region Engineering, 2020).

| Tap water production
The sum of the water loss of the system and the water consumed by the customers served by the system is used as the volume of tap water production needed.The water loss is based off a statistical prediction of the number of pipe breaks in the system for a given pipe age.The number of breaks at any given replacement frequency is calculated using the Weibull-exponential statistical formula given in Equation ( 5).This statistical model was chosen because the amount of data required to solve physically-based empirical equations is not practical for most utilities to obtain.Moreover, this model accommodates pipe break history datasets that do not include the entire history since pipe installation (Mailhot et al., 2000;Pelletier et al., 2003).
where s is a counter for each pipe segment group (segments are grouped according to diameter (63.5, 100, 150, 200, 250, and 300 mm) and installation period (1975-1985,1985-1995,1995-2005, and 2005-2015), ρ s is the Weibull distribution shape parameter for each pipe segment group, k 1s is the reciprocal of the Weibull distribution scaling parameter for each pipe segment group, k 2s is the exponential distribution rate parameter for each pipe segment group, and x is an arbitrary variable for integration.The three parameters required for the above equation are found by maximizing the natural log of the likelihood function given in Equation ( 6) below.Powell's derivativefree optimization method was used to maximize this equation in Python as shown in Figure S1 in the SM (Python Software Foundation, 2016; Ragonneau & Zhang, 2023).The parameter values obtained for each pipe group are displayed in Table S2 of the SM (Mailhot et al., 2000).
where L(ρ, k 1 , k 2 ) is the likelihood function, β is a parameter specifying the number of breaks recorded for each pipe segment j, T aj is the time in years from installation to the end of the recorded break history for pipe segment j, T bj is the time in years from installation to the beginning of the recorded break history for pipe segment j, and t 1j is the number of years between the installation of pipe j and the first recorded break.
Water loss was calculated based on a leakage flow determined by the orifice equation given by Equation ( 7).Parameter values for a very high leakage index were selected due to the high average pressure in the system (>75 m head) (Schwaller & van Zyl, 2015).
where Q leak is the flow out of each leak in m 3 /h, C d is the discharge coefficient (0.8 for high leakage index), A o is the area of the leak in m 2 when under no pressure, g is the gravitational constant in m/h 2 , and N 1 is the leakage exponent (1.23 for high leakage index and 1.5 for background leakage).The area of pipe breaks at 50 m head was 1.637Â10 À4 m 2 for mains (>63.5 mm dia.), 2.18Â10 À5 m 2 for service lines (63.5 mm dia.), and 1.4Â10 À7 m 2 for background leaks.The leakage area under no pressure was preferred because the pressure effects were already accounted for by the leakage exponent.Thus, the leakage area under no pressure was back calculated using Equation ( 8), an empirical relationship used by Schwaller and van Zyl in calculating the leakage area at 50 m pressure head (Cassa & van Zyl, 2011;Schwaller & van Zyl, 2015).
where A 50 is the area of the leak in m 2 when under 50 m pressure head.
Because the number of background breaks could not be estimated, the International Water Association water loss task force's baselines of 20 L/h/km of mains and 1.25 L/h/connection for a pressure head of 50 m were used.These values were corrected by Schwaller and van Zyl for high leakage systems resulting in values of 28 L/h/km of mains and 1.75 L/h/connection and were then corrected for pressure using Equations ( 7) and ( 8) (Schwaller & van Zyl, 2015).Equation ( 5) was used with Equation ( 9) to determine the amount of water and chlorine discharged by pipe breaks.
where # breaksi is the number of breaks during period i, t break is the time in hours of leaking before water is shut off and the pipe is repaired, Y is the amount of water loss (L), and X i is the concentration of chlorine in the water (mg/L) during period i (replace with 1 when computing water loss).
The time of leakage before water shutoff was determined using Schwaller and Van Zyl's relationship between frequency of leak intervention and leak duration (Schwaller & van Zyl, 2015).The very high leakage category was selected based on personal correspondence with the water utility staff in 2022, indicating that break detection/control is performed passively based on pressure drops and inspections of anomalies in the distribution system.Based on this leakage category, the time of leakage prior to water shutoff was 216 h for mains and 576 h for service lines (Schwaller & van Zyl, 2015).
The environmental impact per unit of tap water production given in OpenLCA included chlorine addition and pumping into the WDS.The impact per unit water was reduced such that disinfection and pumping were no longer included in this process but rather quantified in separate processes.This was done because chlorine residual and pumping are often issues of concern for WDSs, and thus, the impacts of the processes were isolated from the larger process of tap water production.In this way, the increased pumping energy and chlorine required to accommodate leakage can be clearly seen.The resulting unit impacts of tap water production for each impact category are listed in Table S3 in the SM.

| Chlorine dosage and pumping energy consumption
After water is treated, it must be pumped through the WDS to supply water at sufficient pressure and chlorine must be added for disinfection of harmful biological contaminants that may be growing within the WDS (Figure 1).The chlorine dose and pumping energy were modeled based off the Alabama distribution system modeled in EPANET 2.2, as shown in Figure 2.
The input pumps of the system are Goulds pump models 3045 and 3410 with impellers sized $22.9 cm (9.0 in.) and $21.6 cm (8.5 in.) and speeds of 3550 and 3560 rpm, respectively.These pumps are the models used by the Alabama WDS.The efficiency and pump curves for the models are given in Equations S1 and S2 and Figures S2  and S3 in the SM.The mains of the distribution system are comprised of cement-lined DIP.According to the Ductile Iron Pipe Research Association, cement-lined DIP is minimally impacted by tuberculation and pipe roughening, retaining a relatively constant Hazen-Williams roughness coefficient of 140 (Bonds, 2017).While manufacturer guidelines were favored in the analysis of each product, this constant roughness coefficient has been notably contested (Houle, 2018).However, the sensitivity analysis conducted in this study indicated the results were robust toward pumping efficiencies well under what would be yielded by a worst-case scenario of using a roughness coefficient of 2.3 times less than the 140 value (C = 60) over the lifetime of cement-lined DIP (Hudson, 1966).The fact that pipe tuberculation and subsequent is embedded in the sensitivity analysis for pump efficiency is consistent with past research indicating pump management is more significant for largediameter pipe while tuberculation most significantly affects small diameter pipe (Saeed et al., 2020).Service lines in the system are made of PVC pipes.PVC pipe's roughness coefficient is also largely preserved as the pipe ages, with a coefficient of around 150 (Parvez, 2019).The pipe materials and diameters match those used in the Alabama WDS.
Chlorine concentration was dosed at 1.35 mg/L at the pumping station serving the system for all pipe replacement scenarios because this value represents the system's average chlorine usage and insufficient data are available to predict chlorine decay as the pipes age.The average chlorine concentration satisfies the minimum chlorine requirements of 0.2 mg/L at each connection set forth by the World Health Organization (World Health Organization (WHO), 2016).
The EPANET model was run iteratively while the pump speed was adjusted in increments of 1%.The model was deemed appropriate for the lowest pump speed that would meet the flow demand of the system and maintain the minimum pressure requirement of 137,895 Pa (20 PSI) at each connection delivering water to a customer (Power et al., 2008).
The required pump energy from the model simulation and the chlorine dosage were then input into Equation ( 10) to obtain the total pump energy and chlorine needed over 300 years for each replacement schedule.
where X tot is the total pump energy required in kWh or the total chlorine needed in mg over 300 years, X i is the energy in kWh required per liter during period i or the chlorine in mg required per liter during period i; and Q toti is the total flow demand (including leakage) in liters per year during period i.

| Pipe maintenance
Pipe maintenance is the process of repairing the breaks predicted in Equation ( 5).It includes excavation for pipe repair, installation of the new pipe (including upstream processes required to obtain this pipe), the reburial of the pipe, and site restoration.Alternative methods of pipe maintenance are compared in other life cycle analyses and are outside the scope of this study (Kaushal & Najafi, 2020;Loss et al., 2018).The pipe maintenance method laid out in this study precludes the need for other maintenance methods.When calculating excavation and material quantities, the number of breaks from Equation ( 5) was used with Equation (3) to calculate the amount of excavation needed to repair breaks and with Equation (2) to determine material needed to repair breaks.Materials for break repair were quantified based on personal correspondence with Alabama WDS personnel in 2022, which stated that each break would be repaired with an $6.1-m-long replacement pipe (20 ft.) of the same material and diameter as the compromised pipe.Thus, materials for pipe repair may be quantified with Equation ( 2) if the length of repair ($6.1 m) is substituted for the length of pipe.Site disturbance and the amount of asphalt or seed needed to restore the site can be calculated using Equations ( 3) and ( 4), respectively, but the number of breaks provided in Equation ( 5) must be multiplied with the length per leak and used as the length dimension.

| End of life
End of life includes the recycling of recyclable pipe materials and the disposal of unrecyclable materials as landfill waste.In OpenLCA, credit was allocated to pipe recycling for the avoided use of virgin pipe.Recycling was calculated using recycling efficiencies and the amount of pipe materials calculated for pipe replacement.The amount of pipe input into the recycling process may be calculated using Equation (1).Equation ( 2) was used in calculating the amounts of materials needed for this process.Equation ( 2) was modified by replacing ε with 1 À ε when calculating the materials needed for the unrecycled portion of a partly recycled process or the disposal of unrecyclable materials.Recycling efficiencies for cement-lined DIP and PVC were taken to be 98% and 60%, respectively (DIPRA, 2016;Mantia & Paolo, 1996).Equations ( 3) and ( 4) were once again used to calculate the excavation volume, disturbed area, and materials for restoring the site when pipes were removed for endof-life management.The transport distance to end-of-life treatment after pipes are retired is $35.4 km (22 mi.) for both PVC pipe and cement-lined DIP.This distance was plugged into Equation ( 11) to determine the total transport quantity required to transport retired or broken pipes to end-of-life treatment.
where θ is the transportation quantity required in kg Â km; l break is the length per break in m; and σ is the distance to be traveled in km.
The variables l total and σ were set to zero and the average distance to manufacturer, respectively, when calculating the transport quantity required for transporting replacement pipe from the manufacturer to the distribution system in order to conduct break repair.
As utilities continue to adapt to an age of digitized records and web-based hydraulic models, the framework of this study offers an accessible approach to optimizing pipe replacement even for utilities with sparse leakage data and basic hydraulic information.The authors recommend the use of AI for assistance in applying the methodology to their own systems.Many of the myriad AI language models gaining widespread popularity are capable tools for translating the presented mathematical equations and descriptions of steps into formats suitable for spreadsheets, coding languages, or online management systems already used by water utilities.

| Data used
See Tables S4-S10 in the SM for a detailed list of flows involved in each process, each flow type, respective providers, and equations used for calculations.

| Sensitivity analysis
To investigate the effects of different system characteristics on the results of this study, the methodology was repeated while adjusting system variables considered to vary from one WDS to the next.The variables are listed along with the original WDS value and the minimum and maximum values in Table 1.
Pipe diameters were varied by changing the percentage of the system pipe that is 300-mm diameter, increasing or decreasing the percentages of the other diameters proportionally.Hydraulic considerations of the diameter changes were neglected because hydraulic impacts were sufficiently evaluated via the manipulation of other variables.In the absence of standard minimum and maximum values, pipe diameter and average household demand (excluding leakage) were varied by plus or minus 10%.The leakage parameters of the orifice equation were grouped into categories of very low, low, medium, high, and very high leakage (Schwaller & van Zyl, 2015).Maximum transportation distance was estimated using trial and error in plotting distances between municipalities and the nearest pipe manufacturers.

| RESULTS AND DISCUSSION
While changing the replacement interval from 25 years to 150 years does yield environmental benefit, the corresponding increase in leakage far outpaces the environmental benefits.In this replacement schedule, environmental impacts are generally halved, with the exception of carcinogenics, which are quartered (23%) and noncarcinogenics and ozone depletion which are cut to one third (35% and 32% of 25 years, respectively).However, the average volume of water leaked from the system per year more than quadruples from 160 kL/d (0.042 mgd) at the 25-year replacement interval to 700 kL/d (0.185 mgd) at the T A B L E 1 Sensitivity analysis input variable limits.150-year replacement interval.In addition, the environmental impacts of the replacement intervals begin to plateau after 50-75 years such that relatively little environmental benefit is gained by extending pipe replacement from 75 to 100-year replacement or from 100 to 150-year replacement.For example, global warming impacts decrease by 38.9% when moving from 25 to 50-year replacement but only by 10.0% when moving from 75 to 150-year replacement.Meanwhile, the average volume of water loss per year increases by 75.5% when moving from 25 to 50-year replacement and by 77.8% when moving from 75 to 150-year replacement.Tap water production and pipe replacement dominate all impact categories with combined contributions ranging from 79% to 99% (Figure 3).Tap water production impacts increase with replacement frequency due to increased water loss, whereas pipe replacement decreases with replacement frequency as expected.Decreases in pipe replacement-related environmental impacts generally exceeded those from increases in tap water production.Thus, the overall impacts of the WDS decrease with longer pipe replacement frequencies across all impact categories.However, as mentioned before, the degree to which these environmental impacts decrease becomes less significant at higher replacement intervals.

Variable
So far, only six other studies have conducted an LCA of WDSs in the context of pipe aging and optimal replacement frequency with a comparable scope to the current study (Haidery & Bas, 2020;Hajibabaei et al., 2018Hajibabaei et al., , 2020;;Khan, 2019;Parvez, 2019;Sanjuan-Delm as et al., 2014).As noted in Table 2, there is some variation in the reported impacts normalized to length of pipe placed.There are multiple reasons for this.All studies except Khan's and the present study are cradle-to-gate analyses focused on production, transportation, and installation of the water lines with end-of-life impacts sometimes included.Meanwhile, the studies also differ in impact assessment methodology.Parvez and the current study employ TRACI's method while the others use one of the CML baselines.While both methodologies are mid-point approaches focusing on human health and environmental impact categories, the CML baselines were developed by the University of Leiden in the Netherlands largely drawing from European or global datasets while TRACI was developed by the USEPA with input data specific to the United States (Acero et al., 2016).The exception is that Khan uses a self-formulated impact assessment using literature review and converting energy impacts to CO 2 equivalent emissions.Finally, it is also notable that the current study implements the Weibull-exponential model, which excels in the areas where the former models lack.The Weibull-exponential model accommodates left-truncated leakage data while avoiding oversimplification of system characteristics.In addition, when used with the modified orifice equation shown in Equation ( 7), the model yields leakage estimates agreeing with typical leakage percentages reported in WDSs (10%-40%) (Barton et al., 2022;Folkman, 2012).Also, the model's prediction for leakage loss at the system's current average pipe age was 37.5% of total water production, just 0.2% difference from the 37.7% leakage estimate obtained via personal correspondence with one of the utility's contractors in 2022.This model combined with GIS mapping and hydraulic modeling gives the current study a distinct advantage over the other studies listed in Table 2.
The findings of the current study listed in Table 2 are consistent with previous literature as the range of impact values fit well with the range given in previous literature when considering the variability in scope and methodology between studies.Even so, a sensitivity analysis is needed to evaluate the significance of variations of input parameters.One study, Parvez, 2019, showed that PVC had one third of the embodied energy of DIP (Parvez, 2019).The sensitivity analysis given by Filion et al., 2004, showed the embodied energy of the WDS was sensitive to the embodied energy of pipe fabrication (Filion et al., 2004).Another study, Khan, 2019, reported pipe maintenance and installation make up 34% of the WDS's global warming impacts with open-cut installation resulting in a 35%-40% decrease in impact from trenchless installation.The same study also reported pumping contributed 62% of the WDS's global warming impacts (Khan, 2019).
The sensitivity analysis results shown in Figure 4 were representative of the results of the entire sensitivity analysis.The results were robust against variations in all 11 parameters studied with the partial exceptions of pipe material and paved area.Paved area has a significant influence on the magnitude of acidification, eutrophication, fossil fuel depletion, global warming, and ozone depletion, respiratory effects, and smog.However, this influence becomes insignificant for some categories when replacing pipes infrequently (i.e., 100-or 150-year replacement), with only fossil fuel depletion and ozone depletion being significant for 150-year replacement.Meanwhile, pipe material follows a similar pattern of reducing significance, having a significant impact on carcinogenics, ecotoxicity, and noncarcinogenics at frequent pipe replacement but only significantly impacting carcinogenics at 150-year replacement.While paved area and pipe material have significant influence over the magnitude of some categories of impact, the overarching trends seen in Figure 3 are robust to all 11 parameters.Complete sensitivity charts are provided in Figures S4-S8 of the SM.Another noteworthy result of the sensitivity analysis was the magnitude of change in water loss resulting from changing the leakage flow per leak, as shown in Figure 5.This parameter is practically lowered by water systems lowering the system pressure and detecting and repairing pipe breaks more quickly.Lowering pressure is appropriate for systems with high pressure from elevation change rather than design considerations.Leakage flow per leak F I G U R E 4 Sensitivity of carcinogenics and fossil fuel depletion impacts to system variables under different replacement schedules.
was the only parameter to have an on water loss.While the results of the environmental impact analysis were not significantly affected by changing the leakage flow per leak, the overall leakage decreased significantly, as shown in Figure 5.As a result, the water loss at 150-year replacement using low leakage parameters was nearly equal to the 29% water loss at 25-year replacement for the original system (Figure 3).In other words, if the system took measures to prevent prolonged or high flow leakage from pipe breaks, then the system could either enjoy the cost savings of reduced nonrevenue water or replace pipes one-sixth as often without increasing nonrevenue water.Depending on the monetary value the distribution system attaches to nonrevenue water and replacement pipe, reducing nonrevenue water, reducing pipe replacement, or a mixture of both may be desirable.Many strategies exist for lowering water loss per pipe break including advanced break detection, fast repair mobilization, effective pressure management, or a combination of these.

| CONCLUSIONS AND RECOMMENDATIONS
While frequent pipe replacement has always been expected to mitigate water loss, this study is one of the few to systematically quantify the magnitude of water saved and the associated environmental benefits/costs.The results suggest that the magnitude of water saved significantly outpaces the environmental benefits of reducing pipe installation.A 25-year replacement interval approximately doubles the environmental impacts of 150-year replacement but cuts the volume of water loss to approximately one fourth the 25-year value.When considering the percentage of water production that becomes nonrevenue water, this is a reduction from 64% to 29% including background loss.While these results alone may seem to favor the abandonment of environmental benefit in view of the potential cost benefit, the diminishing environmental returns for pipe replacements after 50 years suggest 50-year replacement is the sweet spot for obtaining high environmental benefit while still reducing water loss from the status quo 125-year replacement.In addition, the sensitivity analysis provides further suggestions as to how water utilities may get the best of both worlds.The increase in environmental impacts with frequent replacement is primarily due to the increased amount of pipe manufacture, transport, and installation, which outpace the environmental benefits of reduced water losses.However, systems that achieve low water loss per pipe break through leak management reduce their nonrevenue water loss, allowing the operators to replace pipes up to one-sixth as often without exceeding previous non-revenue water volumes.The corresponding reduction in environmental impacts without affecting F I G U R E 5 Low leakage impacts of the WDS in Alabama, USA, normalized to 50-year replacement total of original system; leakage is not normalized.
nonrevenue water may also allow water utilities take advantage of current USEPA funding for the reduction of greenhouse gases.Accordingly, WDS operators are recommended to develop funding requests for leakage detection and management systems centered on the premise of reducing pipe replacement while also capitalizing on financial savings via nonrevenue water reduction and reduced pipe costs.
While sensitivity analysis indicated the results of this study were robust, the transferability of the general conclusion needs to be further investigated.Utilities are encouraged to use the scalable methodology, guidance, and Python code presented in this paper and their WDS-specific data to obtain more accurate results for their own systems.

F
I G U R E 3 Impacts of the WDS in Alabama, USA, normalized to 50-year replacement total; leakage is not normalized (labels at the top of the figure are categories of environmental impact while color-coded indicators are categories of operational management).
T A B L E 2 Summary of comparable LCA studies on water distribution systems.