Assessment of combined sewer overflows impacts under flooding in coastal cities

Floods as a natural hazard cause significant economic and social costs, and losses of life (Haq et al. 2012). One of the infrastructures which can be affected by flooding is combined sewer systems. Combined sewer systems are designed to collect domestic and industrial wastewater and rainwater runoff. During periods of heavy rainfall and flood, the volume of wastewater may exceed the capacity of the treatment plants. Combined sewer systems are designed to be able to overflow and discharge excess wastewater to rivers and other water bodies nearby (UEPA 2001). Hence, many researchers have tried to predict and model flood status in the regions such as Fotovatikhah et al. (2018), who presented a comprehensive survey about the application of computational intelligence-based methods in flood management systems. They also introduced the most recent promising approaches with respect to the accuracy and error rate for flood debris forecasting and management. The results showed the model's high efficiency. Burgan & Icaga (2019) applied the Adaptive Hydraulics (AdH) and the Finite Element Surface Water Modeling System models to generate a hydraulic model for flood conditions in the Acarkay basin, Turkey. The results showed that the majority of settled areas would not face flood risk, while agricultural lands in some regions near the banks of streams may experience flood risks. Kaya et al. (2019) presented a flood modeling method for rivers having no upstream gauged stations. They predicted upstream hydrograph by the reverse flood routing method. This hydrograph was utilized as an inflow for the HEC-RAS model. They applied the model to Guneysu Basin in Rize Province in the Eastern Black Sea region of Turkey. The comparison of observed data with the produced flood map showed the model's capability. Al-Ani & Al-Obaidi (2019) used a multiple linear regression model (MLRM) and artificial neural network (ANN) for sewer sediment accumulation calculation in Baghdad city, Iraq. The results showed that the MLRM is acceptable, with an R² value of 89.55%. The ANN model was found to be practical with an R² value of 82.3%. They also applied sensitivity analysis which showed that the flow is the most influential parameter on the depth of sediment deposition. Keum et al. (2020) developed a classification-based real-time flood prediction model for urban areas by combining a numerical analysis model based on a hydraulic theory with a machine learning model. The results were compared with a verified two-dimensional model, which indicated that the goodness of fit of the developed model was 85%. The model showed its capability in predicting floods for risk management. Mirra et al. (2020) used the oxygen uptake rate (OUR) to show its applicability for the management of onsite wastewater treatment plants (WWTPs). The results proved that when there is no wastewater generation in the household, the OUR in the reactor is very low, 0.0007 to 0.0015 mg/l s, and thus does not require a high oxygen supply. These results, if verified with field experiments, will enable the optimization of the energy use during onsite WWTP operation. Also, other researchers tried to develop prediction methods, such as Wu & Chau (2013), Taormina & Chau (2015), and Mosavi et al. (2018).

Due to special situations, coastal cities and their infrastructures in terms of energy, transportation, water, and wastewater are vulnerable due to storm surges, hurricanes, and combined coastal and inland flooding (De Almeida & Mostafavi 2016; Zeynolabedin et al. 2020). As an example, Hurricane Sandy was the largest storm to hit the northeast USA in recorded history, killing 159, knocking out power to millions, and causing $70 billion in damage in eight states (Kenward et al. 2013). Six months after Sandy, data from the eight hardest-hit states showed that 11 billion gallons of untreated and partially treated sewage flowed into rivers, bays, canals, and in some cases, city streets, largely as a result of record storm-surge flooding that swamped the region's major sewage treatment facilities. The vast majority of that sewage flowed into the waters of New York City (NYC) and northern New Jersey in the days and weeks during and after the storm (Kenward et al. 2013).

Combined sewer overflows (CSOs) contain stormwater and untreated human and industrial waste, toxic materials, and debris, so they are considered to be major water pollution in different aspects (Karamouz et al. 2015). CSOs are a shock loading to the environment that could cause large amounts of pollutants to enter water bodies including rivers and harbors. They are considered a threat to human and aquatic life (Gandhi 2013). According to the United Environmental Protection Agency (UEPA 2017), approximately 772 cities in the USA face this problem. A large amount of CSOs spread the pollutants in aquatic environments, so there will be a high tangible or intangible cost involved to get rid of pollution and restore the water quality (Semadeni-Davies et al. 2008). The intensity of the stormwater/flood event affects the concentrations of CSOs' pollutants (UEPA 2004). The CSOs' discharge, volume, and duration are important in the context of human health and environmental impacts. These impacts are also dependent on antecedent moisture conditions, groundwater levels, and land use which determine the amount of rainfall infiltration and runoff (UEPA 2013). To quantify CSO impacts, different methods and techniques are used by researchers. For instance, Schroeder et al. (2011) analyzed CSO datasets from four catchments of Berlin's combined sewer system in Germany. A comparison of rainfall characteristics (e.g. duration, maximum hourly intensity, and total depth) with CSO volumes showed that total rainfall depth was the best determinant of CSO occurrence. Friedrich & Kretzinger (2012) investigated the vulnerability of wastewater collection and disposal infrastructure such as pipelines and manholes, pumping stations, and wastewater treatment plants caused by sea level rise (SLR). They identified the infrastructural elements within the wastewater system that are most vulnerable to SLR. Hummel et al. (2018) mentioned that across the USA, 394 wastewater treatment plants, serving over 31 million people, were exposed to flooding and CSOs. Bae (2020) investigated the effects of CSOs on the river's water quality throughout three preliminary field tests and three main ones in South Korea. The results indicated that CSOs did not affect water quality on the mainstream since their concentration did not significantly change during rainfall events, although the water quality of tributaries has rapidly deteriorated due to the influence of the CSOs during rainfall events.

Generally, previous studies concentrated on a specific aspect of CSOs, ignoring other impacts. Also, most studies lack the uncertainty analysis of CSOs which is an important subject. In this study, the aforementioned gaps are covered by evaluating the impacts of CSOs on different aspects such as economic, social, environmental, and health by applying the multi-criteria decision-making (MCDM) method. An uncertainty analysis is done to investigate the uncertain nature of CSOs. Also, a distributed hydrologic model named GSSHA (Gridded Surface Subsurface Hydrologic Analysis) is used to estimate the volume of runoff and depth of flooding in different parts of the case study. In the following, the methodology is explained in detail, and the corresponding results are delivered and discussed.

In this study, the impacts of CSOs on different aspects such as economic, social, environmental, and health are investigated in the Coney Island treatment plant which is one of the NYC's 14 treatment plants. To assess the impacts of CSOs, four criteria and 22 sub-criteria are selected. By applying the analytic hierarchy process (AHP) method, a weight is assigned to each sub-criterion in order to consider their importance. Then, for quantifying the impact of CSOs, two different MCDM methods, namely fuzzy Technique for Order Performance by Similarity to Ideal Solution (fuzzy TOPSIS) (Hwang & Yoon 1981) and Preference Ranking Organization Method for Enrichment Evaluation (PROMETHEE) (Morais & De Almeida 2007), are employed. Based on the result of MCDM methods and by considering Superstorm Sandy data, the flood depth and runoff volume are obtained for Coney Island WWTP located in NYC by running a distributed hydrologic model named GSSHA (Downer & Ogden 2004). In the end, the CSO's volume is calculated. The flowchart of methodology is delivered in Figure 1.

The methodology is summarized in five main steps delivered in the following:

• (1)

  Dividing CSO evaluation into four main criteria and corresponding sub-criteria,

• (2)

  Weighting sub-criteria by the AHP method,

• (3)

  Quantifying CSO impact using PROMETHEE and fuzzy TOPSIS,

• (4)

  Identifying critical sub-criteria and their variation by running sensitivity and uncertainty analysis, and

• (5)

  Running the GSSHA model for estimating the volume of runoff, depth of inundation in a WWTP, and the volume of CSO.

In the following, the methodology is explained based on the aforementioned steps.

CSOs can result in high concentrations of microbial pathogens, solids, debris, and toxic pollutants. Also, CSOs could result in severe oxygen deficits that cause water quality concerns by putting stress on public health and aquatic life (House et al. 1993; Walsh et al. 2005; Holeton et al. 2011; Passerat et al. 2011; Madoux-Humery et al. 2013). The impacts of CSOs can be categorized as short-term and long-term. The short-term impacts appear immediately after a storm. They are due to dissolved contaminants, bacteria, and viruses causing the death of fishes and aquatic life (Boët et al. 1994; Kenward et al. 2013). Long-term impacts such as oxygen concentrations and ecological damage cause environmental changes slowly over time and affect organisms over generations (Harremoes 1982). In this study, different impacts of CSOs are classified into four main criteria, public health, environmental, economic, and social, with 22 sub-criteria, as summarized in Table 1.

In order to quantify the impact of CSOs, fuzzy TOPSIS (Zadeh 1996; Hwang & Yoon 1981) and PROMETHEE (Vincke & Brans 1985) methods are applied which are explained in the following.

The main purpose of the fuzzy TOPSIS method is to estimate the closeness of alternatives to ideal solutions in the decision-making process. To evaluate the chance of an alternative, the closeness coefficient will be estimated during the process; the higher the value of the closeness coefficient, the more closeness to ideal solutions. The main steps of the proposed fuzzy TOPSIS method can be described as follows:

• 1.
  Determination of decision matrix (Equation (4)):

  

  (4)
• 2.
  Normalizing (Equation (5)):

  

  (5)
• 3.
  Determination of the weighted normalized decision matrix using the weights obtained from the AHP method output (Equation (6)):

  

  (6)
• 4.
  Calculation of the distance of each alternative from fuzzy positive and fuzzy negative ideal solutions (Equation (7)):

  

  (7)
• 5.
  Calculations of closeness coefficient index (Equation (8)):

  

  (8)

Conservation scales are applied to transform the linguistic terms into fuzzy numbers. Table 2 illustrates linguistic variables and related triangular fuzzy numbers to assess the criteria weights importance. According to this table, we have seven different importance levels, and each sub-criterion has a specific importance level. As an example, IM4 indicates equal importance with negative or positive triangle fuzzy numbers.

In this study, six factors are considered for sensitivity analysis to investigate how they affect the CSO impact variation. For this purpose, two sensitivity analysis methods are applied namely OAT (one factor at a time) and GSA (global sensitivity analysis). In the OAT method, the changes in model outputs are only related to a specific parameter. It does not consider the effects of varying two or more parameters simultaneously which is the limit of this method. To apply OAT, first, all parameters are set at their default values and the model is simulated. Next, each parameter individually is varied between its maximum and minimum values, while all others are held at their default values. In the end, the percentage change in each model output is calculated to determine which parameters cause the greatest and lowest variations in model outputs.

It should be noted that higher first-order sensitivity index values indicate a parameter that provides a large contribution to output variance, while lower values indicate a parameter whose interactions result in significant output variance but individually has little effect.

According to the nature of sub-criteria defined for the evaluation of CSO impacts, there are uncertainties associated with the selected and estimated values. In this study, a new code is developed in MATLAB software based on the Monte Carlo simulation to apply uncertainty analysis. The Monte Carlo simulation is a numerical procedure to reproduce random variables that preserve the specified distributional properties (Tung & Yen 2006; Moya et al. 2013). In the Monte Carlo simulation, the response of the system of interest is repeatedly measured under various system parameter sets generated from the known or assumed probabilistic distributions. Not all of the sub-criteria could be treated as random numbers; therefore, a number of them are selected for this part of the analysis. The following steps are taken for uncertainty analysis based on the Monte Carlo method:

• 1.

  Six sub-criteria in the environmental impact category are selected for uncertainty assessment, i.e. Biochemical Oxygen Demand (BOD) (B29), DO (B21), algal growth (B24), Total Suspended Solids (TSS) (B22), and nutrients (B26). These sub-criteria are also related to each other, and any change in their values will alter the final index noticeably. Therefore, these six factors were selected for uncertainty assessment.

• 2.

  For the aforementioned sub-criteria, the proper distribution should be chosen for generating random values for each of the selected sub-criteria. Different distributions such as gev, rician, gamma, nakagami, weibull, uniform, and normal are examined to determine the best distribution function to fit the observed values.

• 3.

  For each factor, 200 values were generated based on its distribution, and by utilizing the PROMETHEE approach, the range of variation of CSO impact index is obtained.

When a flood happens, a treatment plant's structure and its operation are subject to severe challenges that could result in partial or total operational failure. Here, for quantifying CSO volumes in different situations, three scenarios are defined that can be used to quantify the treatment plant response to flooding: (1) full operation of the treatment plant, (2) total failure after flooding, and (3) partial failure.

Old urban areas with combined sewer systems, especially coastal cities, are subject to the overflow of billion cubic meters of CSOs to the nearby water bodies every year (Karamouz & Abbasi 2016). It is necessary to study the functionality of sewer systems in these areas for extreme conditions such as storms and hurricanes. NYC is the most populated city in the USA with over 8.3 million people (2019 estimation) distributed over 784 km². It is located in 40°43′N and 73°56′W at the southern tip of the State of New York. The average annual rainfall and temperature values are 12.1 °C and 1,144mm, respectively.

NYC is a coastal city that is in danger of hurricanes and tropical cyclones causing storm surges and flooding. Also, SLR and coastal erosion have made much of the NYC coastline vulnerable to flooding (Gornitz et al. 2002; Bowman et al. 2008). Since the 17th century, several severe hurricanes, storms, and cyclones have affected NYC causing serious damage to critical infrastructures, such as transportation, and energy systems which were inundated and shut down for days (Kenward et al. 2013). The characteristics of different hurricanes that occurred in NYC are delivered in Appendix B. Hurricane Sandy, which occurred in 2012, was the largest super storm ever recorded with the speed of 160.9 km/h, the highest peak of water level (3.49 m) and damage amounting to $75 billion. Table 3 indicates data information for the NYC basin in 2012.

After Hurricane Sandy, more than 41 million m³ (11 billion gallons) of overflows were discharged into the rivers, harbors, and other surrounding water bodies of the USA. From this amount, one-third of 13 million m³ (3.45 billion gallons) was untreated sewage, and the remaining 28 million m³ (7.55 billion gallons) was partially treated with just chlorination or filtration (Kenward et al. 2013). NYC had a total amount of 20 million m³ (5.2 billion gallons) of untreated or partially treated sewage overflow after Hurricane Sandy (Kenward et al. 2013). Raw sewages generated in NYC combined with stormwater are first collected by the drainage system and then directed to the city's 14 WWTPs. The locations of NYC's 14 WWTPs and their covered areas are shown in Figure 2. Table 4 shows the volume of untreated and partially treated overflow after Hurricane Sandy in all WWTPs in NYC. Also, the location of CSO's outfalls is shown in Appendix C.

Among different treatment plants in NYC, Coney Island has the lowest resiliency and is vulnerable to flooding (Kokoszka 2013). This plant has the highest amount of CSO overflow. Therefore, it is selected as the focus area of this study. During Superstorm Sandy, Coney Island WWTP was inundated and shut down for 2 h in order to minimize damage to the personnel and electrical equipment. The characteristics of Coney Island WWTP are summarized in Appendix D.

The estimated weights for sub-criteria based on AHP pairwise comparison from the questionnaire outputs are shown in Table 5. These weights are normalized, so that the summation of the weights for each criterion would be one. Also, the corresponding CR is 3% (less than 10%) which is considered acceptable. Based on Table 5, it can be concluded that environmental impacts are the most important aspect of CSOs based on experts' opinions.

The CSO impact is assessed by two different methods. Based on the PROMETHEE method developed in MATLAB, an average value of 0.475 was obtained for the CSO impact, which indicates an average condition for the CSO impacts followed by Sandy. According to the fuzzy TOPSIS index, the relative closeness to the ideal solutions C_i (TOPSIS index) is determined, and the results are delivered in Table 6. Based on the results, depth and volume sub-criteria have the highest rank while temperature has the lowest.

OAT sensitivity results show that the CSO impact index will be varied between 0.48 and 0.54 due to changes in input variables (Figure 3). The detailed results indicate that between water quality parameters, BOD, dissolved oxygen (DO), and TSS have the highest impact on CSO, while algae and nutrient have the lowest impact (Figure 4). The histogram of the CSO impact index is presented in Figure 5.

On the other hand, the GSA sensitivity analysis shows that TSS, BOD, and DO have the highest impact on CSO, respectively, while nutrients and algae have the lowest impact. The results are summarized in Table 7.

Five sub-criteria of BOD, DO, TSS, nutrients, and algal growth are selected for uncertainty analysis due to their uncertain nature. In this regard, 200 random numbers for each factor are generated using different distributions between their minimum and maximum values. The results show a relatively wide range of 0.15–0.55 (Figure 6(a)). For estimating the probability density function (PDF) and cumulative distribution function, a non-parametric method called Kernel density estimation is used. Figure 6(b) shows the values for the histogram of the CSO impact index and empirical PDF fitted to the data using Kernel.

Based on the 200 generated values of CSO impacts, generalized extreme value (gev) distribution is the most appropriate fitted distribution among other distributions mentioned in the Methodology. The PDF of fitted distribution functions is shown in Figure 7.

Therefore, at times of severe flooding, there are more uncertainties associated with other elements of CSOs generated such as runoff volume into dependencies of the different parts of WWTP as far as their failure is concerned and its water quality characteristics.

The result of the delineated floodplain by the hydrologic model (WMS-GSSHA) during Superstorm Sandy is shown in Figure 8. Superstorm Sandy was an event that lasted for 2 days. This period was considered in floodplain mapping, which is based on the idea that, during the 2-day event, the flood inundation changes with time. The most critical inundation depth is estimated at 3.78 m (12.40 ft) in the whole study area. Figure 8 shows the most critical inundation depth in different parts of the area for each cell. In the next step, the location of the treatment plant is specified and flood depth is observed in all of the cells of WWTP's location. The critical depth of flooded inundation in the treatment plant was observed as 3.23 m (10.6 ft) during Superstorm Sandy.

The runoff volume obtained from the GSSHA hydrologic model is 10.5 MCM (2.385 billion gallons) for the entire area. Considering the Coney Island WWTP drainage area (32% of the entire area), the runoff estimation is about 32% of the whole volume, i.e., 3.3 MCM (763 million gallons). According to the information obtained from the official NYC website (NYC 2012b), the water consumption in NYC is estimated at 0.476 m³/day (125.8 gallons/day) for each person. Considering this amount, approximately 80% of water is converted into wastewater; therefore, the wastewater rate is estimated at 0.380 m³/person (100.56 gallons/person). Considering the population in the drainage area of Coney Island WWTP (596,326 people), the produced volume of wastewater is 0.28 MCM (60 million gallons/day in this area; hence, 0.52 MCM for 2 days of Sandy) during the modeling time period.

According to the MCDM results, environmental impacts are more critical among the CSOs' impact criteria. Based on the WWTP operational state, three scenarios of full operation, total and partial failures are defined. The main assumption is that the extreme event and runoff volume are the same for all scenarios. Since the extreme event dominated the second scenario, in reality, it is assumed that some structural and nonstructural measures are done to mitigate flooding impacts. Depending on the applied measures, in different situations, different scenarios can occur. For the first scenario, the WWTP works well without any disruption. The second one refers to a situation in which the main components of WWTP fail due to flood inundation leading to total failure of the system. In the third scenario, some parts of WWTP are being flooded and then fail. It is assumed that WWTP continues its performance with the parts that are not flooded. Since different parts of the Coney Island treatment plant are located at different heights, in a flooding event with a particular flood depth, some parts of WWTP fail and others may work as before. Table 8 shows the proximity of different unit operations from the ground level of Coney Island WWTP (NYCDEP 2013) and the assigned percentage of treatment share.

Although the numbers presented in Table 8 may not be an indicator of the actual percentage of the plant's operation under different flooding depth, they are estimated based on the information published by NYC-DEP, consultation with experts, and engineering judgement for the purpose of developing the methodology based on the AHP method to show the importance of the units. Even though the percentage of treatment is a reflection of the importance of each unit, it should be noted that the failure of one component may yield to the failure of the entire system. As an example, at the flood depth of 2.743 m (9 ft), most of the system's components below that depth fail because of the way different unit operations are positioned, but it can still partially operate at 30.7% of its capacity. The calculated CSO volumes in these scenarios are shown in Table 9.

Combined sewer treatment systems in coastal areas have a significant role in urban systems' serviceability and are vulnerable to heavy rain and flooding events. This vulnerability is intensified at the time of coastal storms. During an intense flood, runoff volume may exceed the capacity of a WWTP causing overflows and sewage to bypass into natural water bodies. This leads to different impacts such as environmental predicaments of significant proportions. In this study, a framework is proposed to quantify the impacts of CSOs in four different criteria such as human health, economics, social, and the environment by defining 22 different sub-criteria for each aspect. Because of the low resiliency, a high volume of CSO, and high damages during Hurricane Sandy, Coney Island WWTP is used as the case study. The AHP method was used for weighting the criteria and sub-criteria based on expert's opinions by using online questionnaires. Considering the actual, maximum, and minimum values of the sub-criteria, the PROMETHEE and fuzzy TOPSIS were used to quantify the impacts of CSO. The PROMETHEE result with an account of 0.475 shows the average impact condition after Superstorm Sandy in NYC. The results show that the environmental sub-criteria such as volume and depth have the most severe impacts. Then, five important sub-criteria with an uncertain nature were selected for further analysis, based on the Monte Carlo simulation, using PROMETHEE methods. Also, the results of sensitivity analysis indicate that TSS, BOD, and DO have the highest impact on CSO output.

For floodplain delineation, determining flood depth and runoff volume, the GSSHA model was used considering the Hurricane Sandy information. The results indicate that flood inundation depth at the most critical point was 3.78 m, and the maximum flooded inundation at the Coney Island WWTP location was estimated at 3.24 m. To evaluate changes in the CSOs volume affecting the environmental conditions, the volume of runoff is estimated using GSSHA as 763 million gallons. The produced wastewater in the drainage area of Coney Island WWTP was 60 million gallons per day. Finally, according to the WWTP response to flooding, three scenarios of the total, partial, and no failures were defined and the volumes of CSO were estimated in different scenarios. Not accessing to wastewater network of NYC and lack of data in the location of CSO's outfalls was the major limitation of this study. This topic would provide a worst-case flood scenario that could affect wastewater infrastructure. The results show that the proposed approach can be used to quantify the CSO impacts and also to identify which regions have the most exposure to prioritize investments and for better cost assessment of preparation, remediation, and recovery during floods. This will promote more effective use of funding opportunities for better future planning of wastewater infrastructure. It gives benefit points to operators of WWTPs to manage the CSOs during storms. For future investigation, researchers can develop uncertainty of hydrological models in flood estimation, and/or develop methods to ensure recovery and quick return of the treatment plant to its required function. Moreover, one can evaluate the vulnerability of the case study WWTP to the storm-driven waves in addition to the CSOs as a future study.

All relevant data are included in the paper or its Supplementary Information.