Climate-Resilient Sanitary Sewers through Minimized Inflow

We considered five sources of uncertainty for future networks in LMICs: sewage production per capita, population growth, impervious areas, the fraction of stormwater that enters the sanitary network, and changing precipitation intensity due to climate change (IPCC 2023; United Nations 2019b; ISF-UTS and SNV 2019; UN-Habitat 2007; Astaraie-Imani et al. 2012).

The amount of sewage produced per capita is directly influenced by the daily per capita water consumption and the rate of the consumed water that returns to the sewer system as sewage. The amount of water consumed is influenced by water management actions (e.g., the water price and changes in demographic and economic composition) (Ashoori et al. 2017; Astaraie-Imani et al. 2012) and climate change (Wang et al. 2016). The effect of climate change on consumption is unclear. Although climate change–induced increases in temperature would likely increase water demand, the water consumption per capita is expected to decrease in water-scarce locations (Greve et al. 2018).

Sewer designs are based on estimates of the final population to be served by the infrastructure. However, population growth in many LMICs is highly uncertain (United Nations 2019b); cities in LMICs are expected to accommodate 2 billion new urban inhabitants by 2050 (United Nations 2019b). Moreover, UN-Habitat estimates that 95% of urban development is unplanned in LMICs (UN-Habitat 2007), leading to uncertain land-use patterns and likely increases in paved surfaces (i.e., impervious areas; Brelsford et al. 2020), consequently increasing the potential runoff into the sanitary network.

Sanitary sewers are designed to collect some stormwater (Department of Human Settlements 2019). However, varied construction quality and design criteria change how much storm runoff enters the sanitary sewers (Kesik 2015). The ingress of stormwater into sanitary sewers occurs mainly due to cross-connections between the stormwater and sanitary sewage systems and unexpected connections of stormwater sources occurring on private property (e.g., when a home’s roof drainage is connected to the sanitary system; Department of Human Settlements 2019). However, little is known about the degree to which stormwater enters sanitary sewers in LMICs. Given that stormwater inflow continues to be a systemic issue for sanitary sewers in well-regulated and managed sanitary networks, stormwater inflow is likely problematic in many LMICs.

Global climate change is expected to increase extreme weather events worldwide by increasing rainfall intensity and frequency and/or by increasing drought (IPCC 2014). An increase in intense rainfall events due to climate change directly magnifies the relevance of stormwater inflow in sanitary networks; excess inflow in the network can lead to SSOs, compromising the safe operation of sanitary sewers. For Brazil, intense rainfall is expected to become more frequent (Santos et al. 2020). Simulations of how climate change will affect rainfall in Brazil show regionalized effects, with rainfall intensity increasing year-round in the Southeast and South Regions; and increasing only seasonally in the North Region (Santos et al. 2020).

We developed a simple mathematical-hydraulic sanitary sewer model to compare quantitatively the uncertainties related to sewage production per capita, population growth, impervious areas, the fraction of stormwater that enters the network, and climate change. We contrasted two estimates for flow rates (loads): projected flows and possible flows. The projected flow refers to the deterministic load that sanitary sewers are designed to accommodate at the end of their design lifetime. In contrast, we estimate a range of the possible flows based on our quantifications of future uncertainties. Thus, while every deterministically designed sewer has only one projected flow, there is a range of possible flows in that sewer, which varies throughout its lifetime. Over the infrastructure’s design lifetime, if the possible load ends up being the same value as the projected load, the future system behaves as predicted, and the design is likely to perform as required at a minimal cost. If the possible load is higher than the projected one, the infrastructure is underdesigned, and dangerous and costly overflows are likely to occur more often than desired. Conversely, if the possible load is smaller than the projected one, the infrastructure is overdesigned—the sewer cost more to build than it might otherwise have.

We compared projected and possible loads on the sewer network based on the peak flow in the system: sanitary sewage (accounting for groundwater infiltration) and stormwater inflow [Eq. (1)]. To quantify the fraction of stormwater collected in the system, we created a variable (${\mathrm{F}}_{\mathrm{storm}}$), which represents the percentage of the area’s total stormwater runoff (${\mathrm{Q}}_{\mathrm{stormwater}}$) that ends up in the sanitary system. If ${\mathrm{F}}_{\mathrm{storm}}=0$, no stormwater enters the sanitary sewers; thus, the system behaves like a perfectly separate sanitary system. If ${\mathrm{F}}_{\mathrm{storm}}=1$, the system behaves like a combined sewer, collecting all of the area’s runoff along with its sewage. In practice, some stormwater during rainfall events enters the sanitary system; hence, ${\mathrm{F}}_{\mathrm{storm}}>0$. For most systems, however, the value of ${\mathrm{F}}_{\mathrm{storm}}$ is highly unknown, especially at the design stage. For simplicity, we omit the effects of stormwater infiltration. Hence, we model the flow in a sanitary sewer as:where ${\mathrm{Q}}_{\mathrm{load}}$ $(\mathrm{L}·{\mathrm{s}}^{-1})$ is the total flow in the system; ${\mathrm{Q}}_{\mathrm{inflow}}$ $(\mathrm{L}·{\mathrm{s}}^{-1})$ is the flow due to the fraction of stormwater that is collected in the sanitary system, which is a fraction (${\mathrm{F}}_{\mathrm{storm}}$) of the area’s total stormwater runoff (${\mathrm{Q}}_{\mathrm{stormwater}}$); and ${\mathrm{Q}}_{\mathrm{sewage}}$ $(\mathrm{L}·{\mathrm{s}}^{-1})$ is the sewage flow rate during dry weather conditions, accounting for the peak sewage production per capita, the population served by the network, and the groundwater infiltration flow [Eq. (2)]. We defined the peak sewage production per capita (${\mathrm{S}}_{\mathrm{p},\mathrm{p}}$) as the average sewage production per capita (${\mathrm{S}}_{\mathrm{p}}$) multiplied by a peak factor (${\mathrm{p}}_{\mathrm{f}}$) (i.e., ${\mathrm{S}}_{\mathrm{p},\mathrm{p}}={\mathrm{S}}_{\mathrm{p}}{\mathrm{p}}_{\mathrm{f}}$), the initial population (${\mathrm{P}}_{\mathrm{i}}$), and a total population growth factor (g). We approximated the groundwater infiltration flow ($\mathrm{\mu }$) as the product of infiltration rate and upstream area (A)where in year t, ${\mathrm{S}}_{\mathrm{p},\mathrm{p}}(\mathrm{t})$ $(\mathrm{L}·{\mathrm{capita}}^{-1}·{\mathrm{day}}^{-1})$ is the peak sewage production; ${\mathrm{P}}_{\mathrm{i}}$ (capita) is the initial population; $\mathrm{g}(\mathrm{t})$ is the cumulative population growth factor from the beginning of operation until year t; $\mathrm{\mu }$ $(\mathrm{L}·{\mathrm{s}}^{-1}·{\mathrm{m}}^{-2})$ is the infiltration coefficient; and A (${\mathrm{m}}^2$) is the area served by the infrastructure.

Our mathematical model captures two sources of increases in the population served by the sanitary sewer: densification (serving new people in a serviced area), extensification (serving new people in new areas). In both cases, increased population could represent new people in that area or previously unserved populations. In our model, initial population (${\mathrm{P}}_{\mathrm{i}}$) represents all people in the area who will be immediately connected to the system. The population growth factor $\mathrm{g}(\mathrm{t})$ could be used to capture the area’s densification throughout the lifetime of the infrastructure, the system serving a growing proportion of the residents in the area, and/or a growth of the infrastructure to serve new people in a new area. Our use of a constant area (A) omits extensification as a source of population growth. We estimate $\mathrm{g}(\mathrm{t})$ based on population projections, which limits our analysis to modeling population growth–driven densification.

We estimated the possible stormwater flow rate using the rational method. For separate sanitary sewers, only a fraction (${\mathrm{F}}_{\mathrm{storm}}$) of all available stormwater runoff would flow into the sanitary system. Introducing ${\mathrm{F}}_{\mathrm{storm}}$ allowed us to account for three uncertainties explicitly: (1) ease of stormwater inflow into the sanitary sewer; (2) impervious surfaces (through the runoff coefficient, C); and (3) rainfall intensity in the catchment area (i), which is likely affected by climate change. Therefore, ${\mathrm{Q}}_{\mathrm{stormwater}}$ = CiA.

We used the ratio of possible and projected flow rates (${\mathrm{Q}}_{\mathrm{possible}}/{\mathrm{Q}}_{\mathrm{projected}}$) as a performance metric for sewer design. The ratio ${\mathrm{Q}}_{\mathrm{possible}}/{\mathrm{Q}}_{\mathrm{projected}}$ reflects the state of the infrastructure’s design. While ${\mathrm{Q}}_{\mathrm{possible}}/{\mathrm{Q}}_{\mathrm{projected}}>1$ implies an underdesigned infrastructure, ${\mathrm{Q}}_{\mathrm{possible}}/{\mathrm{Q}}_{\mathrm{projected}}<1$ reflects an overdesigned one. Thus, ${\mathrm{Q}}_{\mathrm{possible}}/{\mathrm{Q}}_{\mathrm{projected}}=1$ means a well-functioning infrastructure at the lowest possible cost.

We demonstrate the potential utility of the model quantitatively by exploring its predictions about the effects of each source of uncertainty in three Brazilian cities. Specifically, we performed a one-way sensitive analysis to evaluate the relationship between each source of uncertainty and the ratio of the projected and possible flows in the sewer at the end of its design lifetime. We visualized the resulting estimates of the uncertainty-induced threats to sanitary sewer performance with tornado plots.

We also evaluated how uncertainties would evolve over time. While the projected load is fixed, the possible load evolves over time (with uncertainty increasing with time). We explored this temporal variation under three fixed values of ${\mathrm{F}}_{\mathrm{storm}}$: minimum, baseline, and maximum. Given the absence of data on ${\mathrm{F}}_{\mathrm{storm}}$, we estimated its possible range based on the range of projections for it obtained using different design guidelines. For each case, we performed a one-way sensitivity analysis for each uncertainty and year throughout the infrastructure’s design lifetime. We visualized the results as a continuous stack of tornado plots for each year.

We applied the proposed model as a demonstrative example in Brazil, where the government aims to increase the percentage of the population with sanitation services from 55% to 90% by 2033 (Brazil 2020). We applied our model to three of Brazil’s regional capital cities: Manaus (in the heart of the Amazon rainforest), Recife (located on the northeast coast), and Curitiba (in the southern part of the country). Notably, Recife has been ranked as the 16th most vulnerable city to climate change worldwide (IPCC 2014) and has received approximately $89 million of investment since 2018 to expand its sanitary sewer coverage in the next decade (by $\sim 440$ kilometers; Secovipe 2019).

In this demonstrative application, we assumed that the sanitary sewer would be implemented in an area with an initial population (${\mathrm{P}}_{\mathrm{i}}$) without access to a sewer infrastructure with an initial population density (${\mathrm{d}}_{\mathrm{i}}$). We assumed that the area served would remain constant during the infrastructure design’s lifetime and that the population growth would densify the served area. In the model, we took the groundwater infiltration coefficient ($\mathrm{\mu }$) as fixed and modeled the uncertainty associated with: population growth (g), per capita demand (${\mathrm{S}}_{\mathrm{p}}$), peak per capita demand (${\mathrm{S}}_{\mathrm{p},\mathrm{p}}$—affected by ${\mathrm{S}}_{\mathrm{p}}$ and the peak factor, see Table S1), runoff coefficient (C), ${\mathrm{F}}_{\mathrm{storm}}$, and climate-affected rainfall intensity (${\mathrm{i}}_{\mathrm{cc}}$).

To estimate the possible load, we also estimated ${\mathrm{F}}_{\mathrm{storm}\text{-}\mathrm{projected}}$ based on Eq. (4), which requires an estimate of the spare capacity (r). The parameter estimates adopted for our examples in Recife, Curitiba, and Manaus are presented in Table 1 (see Tables S3–S5 for more detailed information). To compare the case study cities, we modeled a neighborhood-sized area of 1.5 square kilometers, with an initial a population served (${\mathrm{P}}_{\mathrm{i}}$) of 10,000, representing an initially dense neighborhood ($\sim \mathrm{7,000}$ inhabitants per square kilometer; Dijkstra et al. 2021). We assumed that the operation started in 2020 and the infrastructure design lifetime was 30 years.

We based our projected and possible average per capita sewage production (${\mathrm{S}}_{\mathrm{p}}$) values based on available water data and a return rate of 80%, as recommended by Brazilian guidelines (ABNT 1986). We estimated the baseline (projected) sewage production per capita (${\mathrm{S}}_{\mathrm{p}}$) based on each city’s initial (2020) average water consumption, as reported in Brazil’s National System for Water and Sanitation Data (SNIS) database (SNIS 2023). We took this initial ${\mathrm{S}}_{\mathrm{p}}$ as constant in time as our baseline scenario. Minimum and maximum estimates of per capita water consumption were unavailable for our case study cities; instead, we analyzed the water consumption in all capital cities in the same geographic region of Brazil (Northeast for Recife, North for Manaus, and South for Curitiba) and took the minimum and maximum values as the lowest and highest observed values in all capital cities within the same region. Finally, we modeled the possible per capita sewage production (${\mathrm{S}}_{\mathrm{p}}$) as evolving linearly from the baseline (in 2020) to the minimum and maximum values in 2050 (the end of the infrastructure design’s lifetime).

The baseline (projected) population growth (g) was taken from the Brazilian Institute of Geography and Statistics (IBGE 2023). The lower and upper bounds were based on population growth rate projections for Brazil from the United Nations World Population Prospects (United Nations 2019a). The runoff coefficient, C, varies from 0 to 1, depending on land use. We assumed a constant baseline value of 0.6, simulating a residential area, with a linear increase in uncertainty over time, ultimately resulting in the minimum (maximum) bound being 50% $-50\%$ (+50%) different from the baseline by 2050. While the lower bound ($\mathrm{C}=0.3$) represents a significant increase in green spaces (i.e., creation of parks), the upper bound ($\mathrm{C}=0.9$) represents a significant increase in impermeable surfaces (e.g., roads and parking lots; Chin 2013).

The initial rainfall intensity (i) was calculated using each city’s IDF for the baseline case (Table S2; Weschenfelder et al. 2014; Fendrich 1989; Monteiro and Braga 2018). We used a return period of 2 years and a time of concentration (${\mathrm{t}}_{\mathrm{c}}$) of $20\min$. To understand the impacts of climate change, we assumed that the time of concentration of the rainfall event would not change and the city’s rainfall intensity would increase or decrease by a factor, cc(t). The cc(t) variable was based on forecasts of how each city’s rainfall intensity would change due to climate change. For Recife, the estimations of the rainfall intensity were based on a local report by Júlio et al. (2019), Manaus and Curitiba were based on Santos et al. (2020). These localized estimates of changes in rainfall intensity are based on global models of climate change, which themselves are highly uncertain (Gettelman and Rood 2016).

Our analytical focus on the fraction of stormwater entering the system (${\mathrm{F}}_{\mathrm{storm}}$) is novel, so ${\mathrm{F}}_{\mathrm{storm}}$ values are not yet reported for sanitary sewer networks. Instead, we inferred plausible values of ${\mathrm{F}}_{\mathrm{storm}}$ based on the recommended spare capacity in different design guidelines [${\mathrm{F}}_{\mathrm{storm}\text{-}\mathrm{projected}}$, according to Eq. (4)]. To allow for the possibility that Brazil’s guidelines insufficiently account for stormwater inflow, we assessed the boundaries of ${\mathrm{F}}_{\mathrm{storm}\text{-}\mathrm{projected}}$ using three guidelines: for Brazil (ABNT 1986), South Africa (Department of Human Settlements 2019), and Toronto (Qazi 2021). These guidelines differ as to how they anticipate and distinguish stormwater inflow and spare capacity (for air flow). We analyzed the Brazilian guidelines because of the cities’ locations in the application. South Africa is considered an emerging economy and a region-influencing country, as is Brazil (Kingah and Quiliconi 2016); importantly, South Africa has recently updated its sanitary sewer design guidelines (Department of Human Settlements 2019). Toronto also has recently updated its sanitary sewer guidelines, and, during the past decade, efforts have been made in the region to improve the management of stormwater inflow and infiltration in the sanitation infrastructure (Kesik 2015; Sandink and Robinson 2022).

For each studied city in Brazil, we computed what its ${\mathrm{F}}_{\mathrm{storm}\text{-}\mathrm{projected}}$ would be if it were designed using each of these guidelines. To calculate ${\mathrm{F}}_{\mathrm{storm}\text{-}\mathrm{projected}}$, two types of variables are needed: (1) variables intrinsic to the guideline being used; and (2) variables inherent to the city that the system is being designed for. First, the values of spare capacity (r), the sewage peak factor (${\mathrm{p}}_{\mathrm{f}}$) and the infiltration coefficient ($\mathrm{\mu }$) can be found in the design guidelines (see Supplemental Materials Section S.1.1 and Table S1 for detailed information). Second, the runoff coefficient (C), the rainfall intensity (${\mathrm{i}}_{\mathrm{cc}}$), the population density (${\mathrm{d}}_{\mathrm{i}}$), the sewage production (${\mathrm{S}}_{\mathrm{p}}$), and the population growth rate (g) rely on the cities’ unique characteristics. Therefore, the values of ${\mathrm{F}}_{\mathrm{storm}\text{-}\mathrm{projected}}$ vary based on both the city and the design guideline used.

While Brazil and South Africa have country-level guidelines, Canadian guidelines vary by city. Both Brazilian and South African guidelines include a range of possible infiltration coefficients ($\mathrm{\mu }$), resulting in a range of possible ${\mathrm{F}}_{\mathrm{storm}\text{-}\mathrm{projected}}$ values (Fig. 2). South Africa’s guidelines also have a peak factor range, magnifying the range of the possible values of ${\mathrm{F}}_{\mathrm{storm}\text{-}\mathrm{projected}}$ (Fig. 2; see Supplemental Materials Section S.1.1 for more detailed information). Comparing the countries, Brazilian guidelines resulted in the lowest value of ${\mathrm{F}}_{\mathrm{storm}\text{-}\mathrm{projected}}$. Absent direct comparisons between construction quality in these contexts, we included all three guidelines in our range of possible values of ${\mathrm{F}}_{\mathrm{storm}\text{-}\mathrm{projected}}$. We based our analysis of uncertainty on the baseline case for ${\mathrm{F}}_{\mathrm{storm}\text{-}\mathrm{projected}}$ in the scenario where the network would be designed following Brazilian guidelines, with $\mathrm{\mu }=0.01\mathrm{L}·{\mathrm{s}}^{-1}·{\mathrm{km}}^{-1}$, ${\mathrm{p}}_{\mathrm{f}}=1.8$, and $\mathrm{C}=0.6$. We took the maximum and minimum values of ${\mathrm{F}}_{\mathrm{storm}\text{-}\mathrm{projected}}$ for each city, based on the highest and lowest possible values of ${\mathrm{F}}_{\mathrm{storm}\text{-}\mathrm{projected}}$ resulting from applying any of the three guidelines to that city for all $\mathrm{C}\in [\mathrm{0.3,0.9}]$ (Fig. 2).

Our computation of guideline-projected ${\mathrm{F}}_{\mathrm{storm}\text{-}\mathrm{projected}}$ values ranged from 0.005% (in Recife when $\mathrm{C}=0.9$ using Brazil’s guideline) to 0.26% (in Curitiba when $\mathrm{C}=0.3$ using South Africa’s guideline); see Fig. 2. This wide range of ${\mathrm{F}}_{\mathrm{storm}\text{-}\mathrm{projected}}$ is in part due to our use of three guidelines. If we had used only Brazilian guidelines, the uncertainty range of ${\mathrm{F}}_{\mathrm{storm}\text{-}\mathrm{projected}}$ would have been 11-fold instead of 52-fold. Based on this sizable difference, we include a supplementary analysis based only on Brazil’s guidelines in the Fig. S1. While both Recife [Fig. 2(a)] and Manaus [Fig. 2(c)] had similar ranges of guideline-projected ${\mathrm{F}}_{\mathrm{storm}\text{-}\mathrm{projected}}$, Curitiba’s lower rainfall intensity resulted in the highest values of ${\mathrm{F}}_{\mathrm{storm}\text{-}\mathrm{projected}}$ [Fig. 2(b)]. The intensity of projected rainfall (${\mathrm{i}}_{\mathrm{cc}}$) is inversely correlated with ${\mathrm{F}}_{\mathrm{storm}\text{-}\mathrm{projected}}$ [Eq. (4)]: a lower projected rainfall intensity results in lower projected stormwater runoff, which in turn increases the projected fraction of stormwater inflow that can safely be collected in the system.

At the end of the infrastructure design lifetime ($\mathrm{T}=30\mathrm{year}$; year 2050), a one-way sensitivity analysis showed that the fraction of stormwater entering the sanitary sewer is the most relevant uncertainty for all three cities. In each city, stormwater inflow could cause the possible load to exceed the projected load by more than 2.6 times [Fig. 3(a)], dominating the range of uncertainty. For each city, ${\mathrm{F}}_{\mathrm{storm}}$ was at least 1.7 times more important than the per capita sewage production and at least 2.4 times more important than climate change [Fig. 3(a)].

The dominance of ${\mathrm{F}}_{\mathrm{storm}}$ in the output of our uncertainty analysis is the direct result of our projected uncertainty. For each city, the ratio of possible maximum and minimum ${\mathrm{F}}_{\mathrm{storm}}$ values in 2050 is $>20$, an order of magnitude larger than the next largest range [$\sim 1.5$ for $\max /\min$ sewage production in 2050; see Fig. 3(b)]. This large range is in part a result of our use of multiple guidelines (Fig. 2). Since ${\mathrm{F}}_{\mathrm{storm}}$ is a novelty of the study, measured values of ${\mathrm{F}}_{\mathrm{storm}}$ are not available; instead, we used Brazil’s design guidelines, as well as two others, to allow for the possibility that ${\mathrm{F}}_{\mathrm{storm}}$ is underestimated in Brazil’s guidelines. A supplemental analysis based only on Brazil’s guidelines found that the importance of ${\mathrm{F}}_{\mathrm{storm}}$ had a similar magnitude as (and was slightly smaller than) the per capita sewage production (Fig. S1).

Both the climate change factor and the runoff coefficient resulted in a similar variation in the ratio ${\mathrm{Q}}_{\mathrm{possible}}/{\mathrm{Q}}_{\mathrm{projected}}$ in 2050 [Fig. 3(a)], which suggests a concerning limit to our ability to control stormwater flows through increased green space control. For example, Júlio et al. (2019) predicted that climate change will increase rainfall by 68% in Recife, which our model suggests would increase the sanitary sewer flows rate by approximately 5% in 2050. Mitigating this 5% increase in sewer flows would require a reduction of more than 68% in the C factor, which is likely impossible in an urban environment (baseline $\mathrm{C}=0.6$, a 68% reduction would be $\mathrm{C}=0.19$).

Uncertainties associated with the sewage flow (per capita sewage production and population growth factor) affect the flow rate more than the climate change and runoff coefficients. In 2050, the range of possible per capita sewage production rates could increase (decrease) the possible flow rate by 37% to 47% ($-2\%$ to $-38\%$), depending on the city. The range of possible population growth rates could increase (decrease) flows by 6% to 8% ($-8\%$ to $-10\%$), again depending on the city. Controlling both variables represents an important opportunity to limit sewer flow rates. For example, decreasing water consumption (perhaps by regulating the water price) reduces sewage production, thus decreasing the flow rate. Alternatively, controlling a city’s master plan can result in different rates for the population growth in the sewer-served area, also affecting the sewer’s flow rate.

While many parameters are known in 2020 and become uncertain as 2050 approached, ${\mathrm{F}}_{\mathrm{storm}}$ is uncertain at time zero (2020). In each of the three cities (Recife, Curitiba, and Manaus), uncertainty evolves asymmetrically due to the asymmetrical nature of the estimators used for population growth, sewage production per capita, and climate change (Fig. 4 and Table 1). While sewage production per capita and population growth effects are independent of ${\mathrm{F}}_{\mathrm{storm}}$ [Eq. (1) and Fig. 4], larger initial values of ${\mathrm{F}}_{\mathrm{storm}}$ magnify the effects of climate change and runoff coefficient uncertainties [Fig. 4(c)].

Controlling the amount of stormwater that enters the system (${\mathrm{F}}_{\mathrm{storm}}$) is the most important variable for reducing the inflow and the flow rate in the sanitary sewer throughout the sewer’s lifespan. Decreasing ${\mathrm{F}}_{\mathrm{storm}}$ reduces the total inflow in the system, offering a way to mitigate the possible impacts of other uncertainties or the baseline. Conversely, if ${\mathrm{F}}_{\mathrm{storm}}$ is equal to or less than the value projected in our baseline design [as in Figs. 4(a and b)], reducing sewage production becomes the most effective way to reduce sewer flows during the sewer’s lifespan. In these cases, the uncertainty of sewage production per capita and population growth has a larger effect than uncertainty of climate change, similar to the results shown in Fig. 3.