Residual chlorine behavior in a distribution network of a small water supply system

A water distribution system (WDS) is sensitive, dynamic, and particular, and it can be considered a large-scale chemical and biological reactor with high residence times (Grünwald et al. 2001), so that the quality of drinking water may be affected during its course from the treatment plant to the consumption point (Boulos et al. 1995). Consequently, it is essential to guarantee a certain disinfectant concentration throughout the distribution system, considering that if its concentration is reduced below the guideline value through the network, it can promote bacterial growth with consequent health risks if a minimum concentration is not provided (Vidal et al. 1994). Once chlorinated, water with a short contact time in the supply sources remains a long time (several hours to several days) in the pipelines and tanks before it is used. This time can be considered as a second contact time, also known as water age, which requires a higher chlorine demand compared to the chlorine demand at the point of application. It must be pointed out that water age depends primarily on water demand, operation and system design. Therefore, to track down chlorine decay, it is necessary to know the decay rate and the dependence on its own concentration (Rossman 2000).

This chlorine demand may be due to two reaction mechanisms (Vasconcelos et al. 1997; EPA 2005; Kowalska et al. 2006): one produced in the mass of water due to the reaction of chlorine with oxidizable compounds, particularly dependent on the concentration of total organic carbon (TOC), the initial concentration, and temperature (Hallam et al. 2003); the other due to reactions occurring at the wall of the pipe, which may be due to numerous factors such as age of the pipe, diameter (Ndiongue et al. 2005), surface roughness, and internal corrosion, among others (Munavalli & Kumar 2005; Al-Jasser 2007). Similarly, the decay rate of chlorine may be affected by the microbiological activity, nitrification, exposure to ultraviolet light, and the amount and type of compounds which generate disinfectant demand, such as organic and inorganic compounds (Sánchez et al. 2010). Specifically, in small systems, the behavior of residual chlorine in distribution networks has not been studied sufficiently and there is little evidence of variation and final destination. The aim of this research was to analyze the behavior of residual chlorine in the distribution network of small water supply systems through a full-scale case study using hydraulic and quality modeling, which includes mathematical models and field work to calibrate and validate the model together with lab determinations in order to identify the most influential variables.

The research was divided into a hydraulic modeling phase and a water quality phase which included mathematical models and field work for calibration and validation. Lab measurements were performed in order to understand the behavior of residual chlorine in La Sirena WDS, from a set of criteria for the selection of sampling points, as well as for the model calibration and validation. After the modeling stage, variables that helped understand the traceability and decay kinetics of residual chlorine were analyzed for correlation.

The system has a multi-stage-filtration water treatment plant at an average flow of 12.0 L/s, which supplies 851 subscribers. In Figure 1, the scheme of the distribution network is shown; the total length is 13.5 km, 9.5 of which constitutes the main network. Three main zones were identified: high, medium, and low, where the highest point in the network corresponds to the chlorine contact chamber and the lowest to the end of the network. The difference in elevation between these two points is 170 meters. The distribution network is sectioned from four storage tanks (T1, T2, T3, and T4) and gate valves.

Chlorine application was done from a plastic tank with a volume of 240 L and a constant-head dose dispenser through which the orifice depth was regulated according to the concentration-in-network needed depending on the variability of the outlet flow given to the network. For the daily preparation of the chlorine solution, 12 L of sodium hypochlorite was used, with a weight percentage of 10.84%, which was determined at the lab under the iodometric method according to the 4500 CL-B guide for standard methods (Eaton & Franson 2005).

Billing records, which corresponded to the readings of user meters, were used. To determine the demand on consumption nodes, consumption posted was considered (controlled by both value and location) and the unrecorded consumption was considered (that beyond the control of the same operator), such as measurement errors, leaks in the distribution network, losses in deposits, some clandestine water draw-off, and fraud (Cabrera et al. 1999; Farley 2001). Each user was associated with a node in the model and consumption was assigned to that node, always looking for the one in closest proximity or the node that could be considered as the one supplying the user (Cabrera et al. 1996; Alcocer & Tzatchkov 2007).

The layout of each of the components of the distribution network (elevations, pipe lengths and diameters, storage units, and accessories), were entered into the EPANET computer program since it is login-free and has ease of access for a large number of users. A dynamic model, with a total time of 168 hours, was used given that it allows monitoring the behavior of residual chlorine every day for a week (Sánchez et al. 2010), with a time variation curve of the fractional demand at regular time intervals, specifically for 1 hour according to Alcocer & Tzatchkov (2007). The calibration of the hydraulic model consisted of adjusting the values of the roughness coefficients, behavior patterns, and demands calculated by the system, in order to obtain a satisfactory match between the values of the water levels in the tanks and measurements taken at the control points (demand and pressure), within the specified accuracy levels. Greater detail about the process of hydraulic calibration is available in Méndez et al. (2013).

Once the hydraulic model was calibrated and validated, the water quality was modeled. For the construction of the quality model, lab data were required as well as field measurements. The water quality sampling was conducted on 1 weekday and 1 weekend day according to Sánchez et al. (2010). Chlorine residual samples were taken at checkpoints every 2 hours throughout the day. The location of the sampling points was carried out according to Table 1.

Chlorine determination was done using the DPD titration method according to the 4500 CI-F guide of standard methods (Eaton & Franson 2005) with an accuracy of ±0.05 ppm (mg/L). The discharge time considered was 5 minutes.

The process of reaction of chlorine at the pipe wall, k_w, is a variable, very difficult to estimate in operating water distribution networks. To get the decay constant at pipe walls (k_w), the procedure described by Alcocer et al. (2002) was followed. It involves measuring chlorine and flow at both ends of pipe sections without branches. However, it was not possible to detect chlorine decay since the sections with these features were short. Therefore, we chose to assign the minimum value recommended in the literature for PVC equivalent to 0.01 to the model (Lansey & Boulos 2005).

For the study of decay kinetics, the method described by Walski et al. (2001) was used. Consequently, successive measurements were made over time in each of the bottles, and chlorine concentration decay curves were generated. To obtain a constant representative of the chlorine reaction with the water mass (k_b), the best adjustment regression was considered according to kinetic models (Equations (1)–(3)). These values were used for assignment to the model. Two sample points PD1 and PD2 (see Figure 1) were selected since one had a high initial concentration according to Hallam et al. (2003), Huang & McBean 2007 and Vieira et al. (2004), and the other was close to the chlorination point according to Alcocer et al. (2004) and Sánchez et al. (2010). A total of eight samples were taken and around 12 measurements of chlorine concentration were made for each of them in proportion to the reaction velocity. One hundred-mL Winkler bottles (amber) were used as recommended by Powell & West (2000) and Walski et al. (2001), and a forced convection oven model OFA-110-8 was used to ensure proper temperature. The bottles were washed with a solution of sulfuric acid (20%) and distilled water. Subsequently, they were dried in the oven at 110 °C for 1 hour in order to remove any debris or contaminants that could react with chlorine. Based on an exploratory temperature sampling performed in the network, it was determined that the samples had to be placed in an incubator at an average temperature of 21 °C. Adjustments of each of the curves were made considering the mathematical expressions represented by Equations (1)–(3).

Calibration was performed by adjusting the chlorine coefficient (k_b), the patterns provided to model the dynamic behavior in the source (change of the initial C_o concentration during daytime), and the order of the reaction until the output data approached the real values and reached the desired level of accuracy according to Walski et al. (2001). The model was considered calibrated when the difference between the estimated chlorine concentrations and the actual measurements was less than 0.2 mg/L (Walski et al. 2001; Sánchez et al. 2010). To model the kinetics of order one, two, and mixed, the Multi-Species Extension (MSX) was used (Shang et al. 2011). Then, EPANET and MSX were coupled with PEST software for parameter optimization, given that it is a separate model, thus avoiding changes in the original code, which can be used to estimate parameters and perform different tasks of predictive analysis, including sensitivity analysis, correlation and uncertainty (Méndez et al. 2013). The setting ranges for the chlorine reaction coefficients, in the water water (k_b), considered increases of 0.05 (Wu 2006) and variations lower than 0.10 mg/L were allowed for the chlorine injection factor in the source (Jonkergouw et al. 2008).

The result analysis contemplated the development of the following topics: (i) chlorine dosing in the system, (ii) evaluation of the reaction kinetics for the decay of chlorine in the distribution system, (iii) construction and calibration of the model quality, and (iv) the identification of the most influential variables for residual chlorine variation.

Originally, the chlorine dosing in the system was done with a constant dosage flow throughout the day, regardless of the outflow into the network. Under the basic principles of mass balance, obviously the dosed chlorine concentration in the network suffered significant changes, especially at hours of minimum night flow, when the inlet flow tends to decrease, so that the residual chlorine concentration increased daily in values ranging from 1.66 to 2.36 mg/L (mean = 1.20, SD = 0.47, and N = 192), exceeding the guideline value under Colombian law, Resolution 2115 (2007) of 2 mg/L. In addition, during an exploratory sampling, chlorine accumulation problems were detected near PD2 (see Figure 1), with values up to 12 mg/L between 3 and 5 am, due to its proximity to the dosing point and its isolation from the rest of the network by two closed valves.

The first-order, mixed-order, and second-order kinetics models were analyzed. Good correlations with the three models were obtained (see Table 2), but the best adjustments were for the mixed order. A decay coefficient k_b = 0.05 (day⁻¹) for the first-order kinetics with R² = 0.9 was obtained, while the mixed-order kinetics model yielded k_b1 = 0.022 (day⁻¹) and k_b2 = 0.930 (day⁻¹) with R² = 0.98, and for the second-order model k_b = 0.117 L/mg·day with R² = 0.95 similar to that reported by Powell & West (2000) for second-order kinetics under a single component with an average value of R² = 0.94.

These results suggest that the three models can adequately describe the behavior of chlorine; however, the mixed-order kinetics, characterized by two parameters, k₁ and k₂, whose values depend on the concentration at the start of the reaction as opposed to the first-order kinetics (Alcocer & Tzatchkov 2007) has better correlation. This confirms the proposal of Tzatchkov et al. (2004) in the sense that the mixed-order kinetics may significantly contribute to the understanding of reaction kinetics in situations where first-order kinetics does not fit the data, with the advantage of easy application.

The initial concentration in the bottle test was 1.21 mg/L; after 16 hours, a concentration of 0.9 mg/L was obtained (intense reaction at the onset of the process). Even so, despite having monitored chlorine decay for more than 30 days (see Figure 2), a complete disappearance of chlorine in the bottle was never achieved, where the lowest concentration recorded was 0.2 mg/L.

This can be explained because, in the initial phase, chlorine reacts with readily oxidizable substances such as low concentrations of organic matter measured (0.13 to 0.23 mg/L) as well as some possible reactions with inorganic compounds (Tzatchkov et al. 2004; Deborde & von Gunten 2008). In contrast, during the slower and more prolonged phase, the chlorine was consumed possibly by some humic substances (Vieira et al. 2004). With this in mind, research by Vieira et al. (2004) has reported times of up to 104 days, which is higher than that recommended for the duration of the 5–7 day test (Rossman et al. 1994; Vasconcelos et al. 1997; Powell & West 2000; Walski et al. 2001; Alcocer & Tzatchkov 2007). This explains why, with certain compounds, the chlorine reactivity is low, showing small modifications in the structure of the compound under certain characteristics of the treated water (Deborde & von Gunten 2008).

Modeling with first-order, mixed-order, and second-order kinetics was analyzed. Differences in residuals lower than 0.10 mg/L in 98% of the data with a maximum residual difference of 0.12 mg/L were obtained satisfactorily, thus complying with Sánchez et al. (2010), who argued that the difference should be lower than 0.2 mg/L. The model prediction also complies with Walski et al. (2001), who state that the model must be able to reproduce the concentrations of chlorine with an average error of about 0.1 to 0.2 mg/L, which was close to the average error obtained for the scenario using kinetic models of mixed, first, and second order, with values of 0.0513, 0.0507, and 0.0478 mg/L, respectively. These values are in harmony with the MAE obtained by Jonkergouw et al. (2008), which reports 0.059 mg/L, by Wu (2006) with 0.045 mg/L, and by Munavalli & Kumar (2005) with 0.050 mg/L.

Results show that 98% of the time, concentrations below 1.2 mg/L are presented with an average value of 0.985 mg/L and only 2.41% of the data do not meet the standard because they show concentrations lower than 0.3 mg/L. Hence, the system, despite being run by a community, can adequately meet the standards of water quality guidelines. It was evident that the hydraulic paths and resulting retention times are affected by the configuration of the valves in the water distribution network, where their closing (whether on purpose or accidentally), can lead to a dead point where minimum velocity and chlorine concentrations lower than the reference value predominate. Furthermore, as can be seen in Figure 3, the variation of water level in the tanks is low, which means that some masses of water do not exit the tank at a specific time, thus generating a potential dead zone that explains the increase in water age and chlorine reaction. This proves an inefficient operation and indicates that the tank should be analyzed as a hydraulic reactor, since maybe the complete mix assumption is not met, suggesting further research in the future, both for modeling criteria and for storage systems design.

According to the EPANET model results, 94.97% of the reactions occur in the tanks and 5.03% in the mass of water in the network. This behavior of residual chlorine may be related to the age of the water because its decay depends on the time spent in contact with the different substances in the water (Sánchez et al. 2010). It was evident that the chlorine concentrations decrease when water age shows the highest values with full tanks, precisely in the early hours of the morning, when consumption is minimal. This is consistent with reports from EPA (2002) and Sánchez et al. (2010) and suggests that the reaction at the pipe walls may be negligible. Such results are consistent with studies by Lu et al. (1999), which showed that consumption of chlorine, due to the relatively new or in-very-good-condition plastic pipes (PVC), is insignificant compared to the demand in water volume. This fact reveals the good conditions for water treatment through multi-stage filtration; therefore, good treatment is essential for keeping water quality in the network. Despite the variability of C_o due to the system dynamics, the chlorine dosing was optimal, because 95% of the time, concentrations were below 1 mg/L in each of the sectors serving subscribers with minimum values of 0.7 mg/L for 25%, improving the values initially found according to the scheme with constant application of chlorine for 24 hours as described in Figure 3.

This paper presents the behavior of residual chlorine using computer modeling for a small water supply system located in La Sirena, Cali, Colombia. The results show that the kinetics of first and second order together with that of mixed order, can adequately describe the behavior of residual chlorine in this type of water distribution network. It showed that the most influential variables for the behavior of residual chlorine in the network are: the quality of input water, chlorine dosing, the effect of storage that increases water age, and the presence of dead points. It revealed that 95% of the reactions occur in the tanks due to small variations of water level, and the negligible reactions at the pipe walls.

The study found that the modeling of residual chlorine in this small network is a valuable tool for monitoring the water quality in the network, useful for compliance with water quality guidelines.

Similarly, we found that modeling allowed a detailed and comprehensive knowledge of the hydraulic performance and water quality in the distribution network. The disinfection process was optimized because the potentially problematic parts in the system hydraulics were identified, such as the dead zones in pipes due to closed valves, sections directly connected to the injection point where chlorine had accumulated (near PD2), as well as problems in disinfection due to dosing changes. Moreover, it allowed the analysis of the behavior of the disinfectant in the network, determining the optimal chlorine dosage in the plant, thus optimizing the location of chlorine sampling points in the network. This enables better control and consistent operation, constituting a tool for planning processes, design, repair, and redevelopments. Future research should consider the influence of network size through comparative studies with other small networks, as well as storage tank analysis as a hydraulic reactor to ensure that the assumption of complete mixture or piston flow is met, and thereby adjust modeling criteria.

The authors would like to thank the Administrative Board of La Sirena water supply system, the community of La Sirena, the Puerto Mallarino Research and Technology Transfer Station, Cinara-Universidad del Valle, Dr Velitchko Tzatchkov from the Instituto Mexicano de Tecnología del Agua (IMTA) [Mexican Institute of Water Technology] and Professor Maikel Méndez of the Instituto Tecnológico de Costa Rica (ITCR) [Costa Rica Institute of Technology].