Chlorate formation in water distribution systems: a modeling study

Chlorine-based chemicals are commonly employed in water treatment plants (WTPs) to oxidize organic and inorganic pollutants, to remove microbial pathogens, and to improve water organoleptic features. While these compounds are very effective as oxidants/disinfectants, their application is known to promote the production of pollutants defined as disinfection by-products (DBPs), which have potential harmful effects on humans and are, therefore, a main public health concern (US EPA 2011; WHO 2011). The concerns about the presence of DBPs in drinking water and their possible negative effects, also related to congenital anomalies (Righi et al. 2012), have led to the development of regulations (Directive 98/83/CE is in force in EU) and international guidelines to identify threshold values of DBP concentration to avoid possible harmful effects for water consumption (US EPA 2011; WHO 2011). Chlorate, supposed to act as an endocrine disrupting compound, is not yet regulated in the EU. The World Health Organization (WHO) guidelines define 700 μg/L as a limit for chlorate in drinking water and an acceptable daily intake equal to 30 μg/kg of body weight, considering a consumption of 2 L/d of water and a 60 kg adult individual (WHO 2011).

DBPs mostly derive from chlorine reaction with natural organic matter and other precursors (Hua & Reckhow 2007; Goslan et al. 2009), therefore the physicochemical properties of raw water are of main importance (Sorlini & Collivignarelli 2005). Other crucial issues about DBP generation are type and dose of disinfectant. While in small WTPs, particularly the ones dedicated to groundwater treatment, a single-phase disinfection (i.e., by means of hypochlorite) is usually adopted, large installations generally perform multiple-step disinfection processes. A primary disinfection, aimed at the removal of pollutants and pathogens (i.e., by means of ozone, chlorine gas or chlorine dioxide), is commonly followed by a secondary disinfection, which has the main goal of assuring drinking water quality along the water distribution network (WDN) to final users. As a result, both primary and secondary disinfectants ought to be considered as DBP potential sources, and their interaction should be taken into account as well (US EPA 2011).

Potabilization treatments are specifically designed to ensure that DBPs' concentrations in WTPs outflow are below the international guidelines values. The most common approach in DBP control is a minimization of organic and inorganic precursors: coagulation, anion exchange, membranes, activated carbon, ozone, are technical solutions regularly implemented at full scale in WTPs. THMs and chlorite ions are effectively removed by active carbon (Sorlini et al. 2014), usually foreseen in WTPs' upstream secondary disinfection, although chlorate is not. Moreover, DBPs can still form within the WDN as a consequence of disinfectant reaction in pipes and tanks (Baribeau et al. 2002). Therefore, particularly in the case of chlorate, it is necessary to understand whether DBP build-up in WDNs can result in concentration values that significantly exceed guideline values.

The aim of this work is to analyze the net amount of chlorate produced in a water distribution system along the path between WTPs and users, taking into account raw waters having dissimilar physicochemical characteristics and different disinfection operative conditions. Although very few studies have been performed regarding the presence of DBPs in WDNs (e.g., Kim et al. 2002), none of them involved chlorate; therefore, this research is the first one to address the evaluation of chlorate neo-formation and dynamics within a WDN. The EPANET 2.0 model of the benchmark WDN Anytown (Walski et al. 1987) was analyzed as an iconic case study, and water quality simulations have been run to evaluate maximum chlorate concentrations in different flow conditions.

Anytown WDN was chosen in the present research because despite its simple structure it represents a benchmark case that has often been considered in modeling studies of water supply systems (Walski et al. 1987; Farmani et al. 2005; Herstein & Filion 2011; Fu et al. 2012; Atkinson et al. 2014; Basupi et al. 2014). The complete description of the WDN was reported by Walski et al. (1987), and only its main features are summarized here. The network structure is composed of 34 pipes (for a cumulative length of approximately 70 km) and 16 demand nodes with a daily average demand of 404 L/s comprehensive of residential and industrial users, corresponding to a population of about 130,000 inhabitants. Water is provided by a single reservoir connected to a pumping station, and two tanks whose sizes are 15,000 and 1,600 m³, respectively, are included. Water demand values at single nodes ranged between 12.6 and 63.1 L/s, and the highest and lowest nodes had elevations equal to 6.1 and 36.6 m. The minimum and maximum values due to temporal variations in water demand were 0.6 and 1.3 times the mean daily values, respectively.

In-depth scientific activity has been devoted in the past decades to the development of models able to predict THMs and haloacetic acids' formation in drinking water (Sadiq & Rodriguez 2004; Chowdhury et al. 2009), although most models are based on multiple regressions and reaction kinetics are usually not considered. Only few researches are available at the moment about chlorate predictive models (Korn et al. 2002; Zanetti et al. 2008), and they are based on empirical equations as well. As discussed before, many different chemical reactions are involved in chlorate build-up, and their relative influence is not constant over the whole process. Therefore, the evaluation of zero-order or first-order kinetic expressions could be quite difficult (Boano et al. 2014), and the huge variance of raw water pH and its high influence on chlorate generation kinetics (Gordon & Tachiyashiki 1991) should also be considered.

In this study, the overall chlorate generation process was considered. Consequently, an empirical equation based on raw water features and on disinfection boundary conditions was preferred because it is able to describe the temporal dynamics of chlorate production in a more complex and flexible way compared to a kinetic approach. Specifically, the predictive equation developed by Korn et al. (2002) (see Equation (5)) from regression analysis of raw water disinfected with chlorine dioxide in eight Canadian WTPs, was considered in the present research. The Korn model was preferred to the one proposed by Zanetti et al. (2008), because the second was obtained from a single WTP. The Korn model predicts the amount of generated chlorate as a function of water physicochemical properties, reaction time, and ClO₂ dose:

The aim of the work is to evaluate the formation of chlorate only in the WDN, beside the amount formed within the WTP. Hence, the value predicted by Equation (5) at t = 0 (i.e., at the entrance of the WDN) was subtracted from the calculated chlorate concentrations, and the net increments in chlorate concentrations caused by ClO₂ decay in the WDN were thus determined.

The values of the physicochemical characteristics of raw water employed in this study were gathered considering a critical analysis of literature data (see Table 1), in which either NPOC or TOC values are shown according to data availability. Table 2 summarizes the values of the physicochemical and operative parameters considered by the Korn model (see Equation (5)) that were used as boundary conditions for the simulations in the present research. The data reported in Table 2 were derived from the ones shown in Table 1 by means of their aggregation in three categories, defined on the grounds of raw water origin: surface water, groundwater, and mixed water. Cl₂ dose values in Table 1 were converted in ClO₂ ones considering a stoichiometric factor equal to 1.05. Mixed water values were gathered as average values of groundwater and surface water. The data in Table 2 are consistent regarding two aspects. First, they reliably describe the quality of surface raw water towards that of groundwater. Second, this approach consistently describes the real situation about WTPs and WDNs, which may be fed by a single or multiple types of raw water.

No hypotheses were made about the nature of primary and secondary disinfection in the WTP feeding the considered WDN. A single water reservoir and therefore a single disinfectant input in the WDN were hypothesized in the simulations.

In buried pipelines, the presence of soil limits heat exchange and temperature fluctuations of water within pipes. For this reason, in this modeling analysis, water temperature was considered constantly equal to 14 °C throughout the WDN (see Table 1) and the influence of heat exchange on chlorate production rate was hence assumed to be negligible. Chlorate production may decrease with increasing distance from the WTP because production rates are proportional to disinfectant concentration, which in turn progressively decays as water travels through the WDN. However, the effect of this decrease in disinfectant concentration was not considered because (i) it was expected to be limited and (ii) the slight overestimation of DBP production in the WDN resulting from the assumption of constant disinfectant concentration led to a conservative estimate of chlorate exposure from drinking water consumption.

The network model was analyzed with the EPANET 2.0 software (Rossman 2000). Numerical simulations were run considering three scenarios representing a wide array of different mixing conditions of water in the two tanks. Tank mixing is a crucial issue because tanks are often characterized by high water residence times that enhance disinfectant decay and formation of DBPs. The three considered scenarios were useful to evaluate the impact on chlorate production of three different mixing models, i.e., fully mixed, first-in-first-out (FIFO) plug flow, and last-in-first-out (LIFO) plug flow. The fully mixed model is most suited for tanks with frequent filling and emptying that enhance water mixing, while FIFO and LIFO plug flow models describe tanks with negligible mixing potential and either with separate inlet/outlet pipes (FIFO) or with a single pipe for both inflow and outflow (LIFO). Further mixing scenarios were not considered because alternative descriptions of water mixing would require the introduction of model parameters which are difficult to estimate (e.g., Clark et al. 1996). Moreover, the three considered scenarios already represent very different ways of water recirculation, and hence provide different estimates of water residence times in tanks.

Values of water age (t parameter in Equation (5)) were derived from the hydraulic model of the WDN. A water quality simulation with duration of 120 days was performed because it was seen that after this duration values of water age attained stable periodic conditions that were no longer affected by the initial conditions. Values of water age were hence obtained discarding this initial transient. Water age in the WDN followed a periodic daily pattern induced by the temporal demand of network users, and the maximum value for each node was thus extracted to represent the conditions that led to the highest chlorate production. Maximum chlorate concentrations were then considered for critical points that exhibited the maximum concentrations within the WDN.

The simulations presented in the foregoing sections have been performed, and the main results are reported below focusing on chlorate concentrations in nodes and tanks of Anytown network.

Temporal patterns of water age shown in Figure 1 result in variations over time of chlorate concentration at each point of the WDN. In the remainder of this analysis, only the peak value of the water age for each node and tank has been taken into account because these peak values are associated with the maximum exposure of WDN users to chlorate.

Results of the maximum water ages obtained from the EPANET simulations are displayed in Table 3 for the three different scenarios with respect to mixing conditions in tanks (Mixed, FIFO, and LIFO) and for the three different types of raw water. As anticipated by Figure 2, for Mixed and FIFO modes, water age values are highest in Tank 65 and Node 160 (the nearest to Tank 65), while in LIFO mode, the values in nodes are more homogeneous and very high in the two tanks. The longest time spent by water in the WDN is predicted for the LIFO case and it corresponds to 514 and 655 h (i.e., about 3–4 weeks) for the two tanks. These values are lower for the Mixed and FIFO scenarios, but even in these cases the long permanence in Tank 65 results in water ages that are higher than 1 week (229 and 290 h for Mixed and FIFO, respectively).

Table 3 also displays the chlorate concentration increments calculated with the Korn model (Equation (5)) with respect to the chlorate concentration calculated in Node 10. As stated in the section ‘Evaluation of chlorate neo-formation’, the main goal of the present analysis is the evaluation of the chlorate production within the WDN. To this aim, the concentration of chlorate for water entering the network from the WTP in Node 10 is set to zero to exclusively focus on the chlorate build-up in the WDN. Hence, we have considered the chlorate concentration in Node 10 as the reference level against which the increments in the rest of the network are determined. The increments listed in Table 2 range between a few micrograms per liter for nodes that are close to the WTP up to approximately 40 μg/L for tanks and nearby nodes. Such values are comparable with the amount of chlorate that is generated in WTPs when groundwater is disinfected (50–100 μg/L) and much smaller than chlorate concentrations that are found at the entrance of the WDN fed with treated surface water (200–1,000 μg/L), which typically requires a higher disinfectant dose (Fiore et al. 2012; Sorlini et al. 2014).

The level of risk for the users that is associated with chlorate formation in the WDN is strictly related to the amount of chlorate that is already present in drinking water from the WTP, i.e., to the input chlorate concentration. Most countries have not imposed any restrictions on maximum chlorate concentration in drinking water yet. Hence, the value of 700 μg/L suggested by the WHO guidelines for drinking water quality (WHO 2011) is here considered as a reference level to assess the relevance of chlorate build-up. The calculated increments in chlorate concentrations were hence compared to the WHO reference level, and the resulting values are reported in Table 4 for the different raw water types and tank mixing scenarios. The relative increments vary within a relatively narrow range, comprising between 0.4 and 6.7% of the WHO threshold value. The highest values always occur at the tanks in LIFO scenarios, and correspond to 5.9, 6.7, and 3.3% when the raw water source is represented by groundwater, surface water, and mixed water, respectively. Beside tanks, the maximum relative increments at WDN nodes can be found in the FIFO and, to a slightly lower extent, in the Mixed scenario (4.0%, 4.6%, and 2.9% for groundwater, surface water, and mixed water, respectively).

The previous results show that the increments of chlorate concentration within the network resulting from disinfectant reaction in tanks are very limited (<7%) compared to the concentration limit proposed by WHO. Since tanks do not represent points of water consumption by users, it is worth stressing that this increment is even lower (<5%) for WDN nodes where the actual user demand of water is located.

The production of DBPs in WDNs is expected after the application of chlorine-based disinfection agents. International guidelines (WHO 2011) suggest limiting the concentrations of DBP values in order to guarantee the safety of drinking water. Among possible disinfection methods, chlorine-based ones are widely employed because of their cost-effectiveness ratio. It is thus necessary to evaluate the production of DBPs in WDNs and their dynamical behavior.

This study has focused on the formation of chlorate in a benchmark WDN. Chlorate build-up in the WDN has been evaluated by coupling an EPANET-based model of the network to determine water residence times in the WDN and an empirical equation (Korn et al. 2002) to calculate the amount of produced chlorate over the wide range of residence times in the network. The results show that the increments in chlorate concentrations are limited (i.e., lower than concentrations typically found in WTPs) and that, in the investigated conditions, the DBP production is much lower (<5–7%) of the values suggested by commonly accepted international guidelines (WHO 2011). Even though these values are specific for the benchmark network that was considered here, it is expected that the results are also representative of similar medium-sized WDNs, for which the neo-formation of chlorate in WDN pipes and tanks should not represent a health concern provided that chlorate concentration is kept below suggested guideline values at treatment plants.

A simplifying assumption that has been made in the present analysis is related to the influence of pipe conditions on chlorate dynamics. In real networks, spatial variations of the pipe properties such as type of material, age, and roughness influence chlorine dynamics (Al-Jasser 2007; Fuchigami et al. 2012) and may thus also affect chlorate formation within the WDN. However, these effects are typically neglected in modeling studies because detailed information about pipe characteristics are usually unavailable and including the influence of pipe wall conditions would introduce a high level of uncertainty that would make results less reliable. Moreover, the predictive equation employed in the present analysis is already one of the most detailed ways to estimate chlorate formation, and it does not include the influence of wall materials which were considered less important than the other factors considered here. Another practical issue that has been disregarded is the presence of water leakages. While this assumption is not realistic, it is important to stress that leakages would result in a less critical situation for DBP formation. On the one hand, low water velocities generally occur during the night when user demands are low, resulting in higher water age that favors DBP production. On the other hand, leakage intensity is also highest during the night because of high pressure levels. The inclusion of leakage in the model would hence lead to a small increase in velocities at night that would reduce water age in the system and also chlorate production, partially balancing the effect of low night-time velocities. These changes in water age are expected to be small and probably not relevant compared to the residence time in tanks, but they would, anyway, make the situation less critical for chlorate production. Nonetheless, future insights on pipe wall influence on chlorine-based disinfectants and the development of more realistic and refined WDN models may lead to more precise predictions of DBP production in water networks.

Given the increasing attention received by DBP control to ensure safety of drinking water, it is expected that the formation of DBPs in water within WDNs will represent an active research topic for future studies about drinking water quality. Future developments will likely better analyze the controlling factors for the kinetics of formation of chlorate and other DBPs. In this context, our study has shown the importance of an accurate description of the multiple factors that affect DBP formation. In particular, the residence time of drinking water in WDNs should not be neglected when DBP formation is assessed because of the remarkably long times spent by water in storage tanks.