would predict water quality variations in distribution systems were examined and developed. ... JOURNAL AWWA .... studies, much work remains to be done.
IRESEARCH & TECHNOLOGY
]
Measuring and Modeling Variations in Distribution System Water Quality Robert M. Clark and Judith A. Coyle
The authors describe a field study that examined the effects of hydraulic mixing on water quality variations in a distribution system. Conducted at the North Penn Water Authority (average production of 5 mgd and 225 mi of distribution pipe), the study incorporated a field sampling program that utilixed customized automated samplers designed to prevent loss of volatile constituents. As a complement to the field sampling program, computer models that would predict water quality variations in distribution systems were examined and developed. A clear need was shown for obtaining more representative monitoring results than are normally acquired from distribution system sampling. Until recently most of the interest in drinking water quality has been focused on finished water leaving the treatment plant. Finished water, however, may undergo substantial changes in quality during transport through the distribution system to the consumer, and the distribution system itself can have a negative effect on water quality. Some factors that influence such changes
TABLE Comparison
1
of mean recovery ratios for volatiles in samples collected manually and by two customized automated samplers Vinyl Chloride
Custom sampler with EMSL capper versus manual sampling 0.91 (0.09)* Custom sampler with solenoidactuated valves versus manual sampling 0.96 (0.16) Custom samplers compared 0.95 (0.16) *Standard deviations in parentheses
46
include (1) chemical and biological quality of source water; (2) effectiveness and efficiency of treatment processes; (3) integrity of the treatment plant, storage facilities, and distribution system: (4) age, type, design, and maintenance of the distribution network; and (5) quality of treated water.* Another factor that may influence water quality in a distribution system is
l,lDichloroethylene
l,l,lTrichloroethane
TCE
0.90 (0.10)
0.94
0.96
0.89 (0.15) 0.99 (0.23)
0.93 (0.13) 0.97 (0.13)
0.94 (0.13) 0.95 (0.17)
the effect of mixing water from different sources. Water distribution systems frequently draw water from multiple sources, such as a combination of wells, surface sources, or both. Mixing of waters from different sources takes place within the distribution system and is a function of complex system hydraulics.z-4 For all of these reasons, the quality of water within the distribution system may vary with both location and time. To understand water quality variations in a distribution system, one must also understand hydraulic behavior as well as the kinetics of contaminant formation and destruction. These parameters can be studied on a limited basis in specially constructed pipe loops, but it is nearly impossible to study them on a full-scale. basis because of the complexity of most distribution systems. Extensive documentation has shown that both microbiological and chemical deterioration in water quality takes place in distribution systems. Little success has been demonstrated in quantifying these effects, and few models have been developed that adequately describe these changes. Hydraulic models are frequently used to analyze water system operations and to understand the highly variable flow patterns that exist in most distribution systems. These models are widely used and well documented. Contaminant propagation models can be coupled with hydraulic models and used
RESEARCH AND TECHNOLOGY
JOURNAL AWWA
Copyright (C) 1990 American Water Works Association
to model chemical, biological, and hydraulic changes that take place and to predict the spatial and temporal distribution of contaminants throughout the system. Several of these models have been discussed elsewhere. This article describes a study that was primarily conducted to examine the effects of hydraulic behavior on water quality variations that take place in distribution systems.5 By using a conservative contaminant as a tracer, this study eliminated the variable of growth or decay of biological and chemical contaminants and documents the effects of hydraulic mixing. A project conducted at the North Penn Water Authority (NPWA) in Lansdale, Pa., was the basis for this analysis.
Previous research in water quality deterioration Many investigators have attempted to document the possible deterioration of water quality once finished water enters the distribution system. A 1976 study documented that bacteriological changes may cause aesthetic problems involving taste and odor, discolored water, slime
The quality of water within the distribution system may vary with both location and time. To understand water quality variations in a distribution system, one must also understand hydraulic behavior as well as the kinetics of contaminant formation and destruction. growths, and economic problems such as corrosion of pipes and biodeterioration of materials.‘j Bacterial counts tend to increase during distribution and are influenced by a number of factors including bacterial quality of the finished water entering the system, temperature, residence time, presence or absence of a disinfectant residual, construction materials, and availability of nutrients for growth.7 Maul et al investigated variations in spatial and temporal heteroAUGUST
1990
Figure 1. Layout of the North Penn Water Authority
trophic plate count (HPC) bacteria in a distribution system and found that free and total chlorine residuals decreased rapidly with distance from the treatment plant and could not be detected in the peripheral parts of the system where HPC levels were highest.*,gBased on the results from this study, a sampling design for monitoring water quality was developed. The relationship of bacteriological quality to turbidity and particle counts in distribution water was studied by McCoyand Olson.l”Turbiditywasfound to be a linear function of total particle concentration but not of the number of bacterial cells. Degradation of bacterial
distribution
system
water quality was shown to be the result of unpredictable intermittent events that occurred within the system, at which times there was a correlation between turbidity and HPC. While investigating a biofilm problem in a distribution system, LeChevallier et al found that HPC densities increased as the water flowed from the treatment plant through the study area.11 A number of investigators have attempted to quantify the deterioration in water quality in distribution systems. Donlon and Pipes attempted to relate velocity of water in a distribution pipe to total bacterial counts in a sampling device.12 Pisigan and Singley have related ROBERT
Copyright (C) 1990 American Water Works Association
M. CLARK
&JUDITH
A. COYLE
47
‘I I I I c i
Figure 2. Sampling results from six sites in the North Penn distribution system (taken at 4-hour intervals starting at 9 a.m. Nov. 14, 1985, and ending at 5 p.m. Nov. 15, 1985)
Figure 3. Comparison of modeled and measured hardness at selected monitoring sites
Figure 4. Comparison of modeled and measured chloroform at selected monitoring sites
48 RESEARCHANDTECHNOLOGY
Figure 5. Comparison of modeled and measured TTHM at selected monitoring sites
JOURNALAWWA
Copyright (C) 1990 American Water Works Association
fluid velocity and chemical quality to corrosion of water in pipe loopsi3Reiber et al have evaluated corrosive water versus chemical quality in consumer systems by monitoring in distribution systems.14 McClelland and Maury developed a mobile water quality laboratory, and Schimpff used it to monitor water quality in a distribution system.‘5J6 Matson and Characklis have developed both theoretical and empirical models for biofilm growth and attachment in distribution systemsi Gallowin et al have developed theoretical equations for the transport of particles in distribution systems. i*.19 Despite the results obtained from these studies, much work remains to be done. The current state of knowledge of the kinetics of distribution system water quality changes is insufficient to permit the development of a detailed predictive model. In the study that is described in this article, contaminant propagation was related to hydraulic behavior in a Pennsylvania water authority’s distribution system.
in the Lansdale zone, which represents the largest portion of the distribution system. NPWA well pumps operate on time clocks, with a few controlled by water level in the storage tank. Some wells pump continuously, but others are not pumped during the late evening and early morning hours. At these times,
Analysis of historical water quality data
The current state of knowledge of the kinetics of water quality changes is insufficient to permit the development of a detailed predictive model.
North Penn Water Authority The NPWA serves 14,500 customers in 19 municipalities in Pennsylvania’s Montgomery County, which is located north of Philadelphia. NPWA produces an average of 5 mgd. Sources include 1 mgd of treated surface water purchased from the Keystone Water Company and 4 mgd obtained from 40 wells operated by NPWA. The NPWA system is largely contiguous, but there are a few unconnected satellite systems that were not modeled in this study. The contiguous distribution system has five storage tanks and two pumping stations and consists of 225 mi of pipe (Figure 1). Pipe sizes range from 3 to 24 in., with the majority of the pipes being cement-lined ductile iron or unlined cast iron. In the southern part of the system the rate of flow is determined by the elevation of the tank in the Keystone system and by a throttling valve at the tie-in. Flow, which is monitored continuously, is relatively constant; the throttling valve is adjusted seasonally. Water flows into the Lansdale lowpressure zone and from there enters the Lawn Avenue tank, from which it is pumped into the Souderton pressure zone. Additional water from the Hillcrest pressure zone enters the Lansdale zone at the Office Hillcrest transfer point; this water is derived solely from wells in the Hillcrest zone. Except for unusual and extreme circumstances, such as fire or main breaks, water does not flow from the Souderton zone into the Lansdale zone nor from Lansdale into Hillcrest. The emphasis in this study was on modeling water movement and quality AUGUST1990
analysis type, TTHMs, chloroform, bromoform, TCE, chloride, hardness, sulfate, and total solids). These constituents were selected because they generally reflect a significant spatial variability. Only those samples collected at wells and at the 31 routine sampling locations are included in the data base, which contained information on 583 samples when this study was conducted.
Keystone water moves farther into the NPWA system. Keystone water contains much higher levels of total trihalomethanes (TTHMs) than the NPWA well water. Certain wells show theoccurrence of trichloroethylene (TCE) and c&-1,2-dichloroethylene. Inorganic chemicals also vary from well to well and between well water and Keystone water. The NPWA distribution system is well-instrumented, and detailed continuous records are produced on well pumpage, tank heights, and flow at various locations within the system. In addition, NPWA maintains a detailed water quality sampling program covering sources as well as points in the distribution system.
Measuring variations in water quality NPWA has established a routine distribution system sampling program both to comply with federal requirements and to provide detailed information on water quality for NPWA’s operational and quality control purposes. Samples are taken routinely at operating wells and at 31 designated sampling sites (Figure 1). In addition, samples have been collected at other sites in response to customer complaints or for special studies. Samples are analyzed for chemical constituents, volatile organics, and physical and bacterial parameters (36 parameters are analyzed, although not all parameters are analyzed for each sample). A data base contains key information for each sample (date, primary location code, secondary location code, number,
Minimum, maximum, and average values for each constituent at each sampling location were calculated. Average values were then plotted and hand-contoured on a distribution system map. In addition, TTHM samples taken on two specific dates were also plotted and contoured. Examination of these plots showed clear spatial variations, with the highest TTHM values being in the vicinity of the Keystone tie-in and the lowest values being in the Lansdale zone and in the Harleysville area. The plots representing average values were generally based on from 1 to 5 sample points, although sampling may have been performed at different times of the day or different days of the week, making the statistical validity of the analysis somewhat limited. The plots on two specific days, however, showed a clear pattern of TTHM variation associated with the presence of Keystone water in the distribution system. These data were used later in the study toconfirm steadystate predictive estimation of TTHM variations.
Field sampling program In order to investigate the nature of water quality variability under dynamic conditions within the system, a field sampling program was conducted at six sites, which were selected based on spatial variations determined from historical data and modeling results. Because it was felt that TTHMs from theKeystone water system had reached their ultimate formation potential, TTHMs were assumed to be conservative tracers of surface water. This assumption provided the basis for eliminating the effects of biological and chemical contaminant growth and decay and for studying only the effects of hydraulic behavior on water quality. As shown in Figure 2, wide variations existed both within the 34-h period for given points and within the system itself. Figure 2 also depicts the variation in hardness (also conservative) at these same points (hardness is primarily associated with flow from the wells). The most striking result of this analysis is the dramatic change in TTHM values at the Mainland sampling site and, to a lesser degree, at the Lawn ROBERTM.CLARK&JUDITHA.COYLE
Copyright (C) 1990 American Water Works Association
49
Avenue tank inlet. Mainland water showed the lowest TTHM values in the 1 p.m. and 5 p.m. samples, indicating the lowest percentage of Keystone water during that time period. This suggests that Keystone water is most confined within the central portion of the system during the noon-to-midnight period. The water then travels throughout the system during the early morning hours when well pumps are off. At the Mainland site, a flushing back and forth of surface water and groundwater occurs, with the peaks from the wells indicating that water flow at this point is affected by the surface water and groundwater sources. These results point out the problems in attempting to understand a dynamic system using routine monitoring data. Routine TTHM monitoring data represent samples taken over a number of years at different times of day. The pattern of flow in the NPWAdistribution system varies during the course of a day as well pumps cycle off and on. Late at night, when most well pumps are off, Keystone water is the primary source of supply for the Lansdale zone, and this water passes into the eastern and northern portions of the NPWA system, from which it can enter the Lawn Avenue tank. When water from the Lawn Avenue tank enters the Lansdale zone, it can be expected to contain TTHMs because at least a portion of this water originates from Keystone when the Lawn Avenue tank is filling. Samples ranged from a maximum TTHM level of 36 pg/L to a minimum of 13 pg/L at the inlet to the Lawn Avenue tank. This range is consistent with the historical data of 18.8 pg/L in the neighborhood of Lawn Avenue. Data from the pilot sampling run, taken at the Mainland sampling site, shows TTHM levels ranging from 9 to 34 pg/L (Figure 2). As concluded from the analysis of the pilot sampling run, Mainland receives water from the six wells in the Harleysville area when the well pumps are on and from Keystone in the early morning hours. Modeling, as described later, shows that the zone of blending of Keystone and well water is fairly close to Mainland. Thus, simply from an examination of the model prediction, Mainland would be expected to show the variations that it in fact does show. The Mainland site was selected primarily to test this hypothesis in the pilot sampling run. Data gathered in this sampling program were used in testing various quality models, as described later.
Automated sampler development This limited study clearly showed the value of being able to sample over long periods at specified points in the distri-
-160 1.80 110 -140 130-r 3ot
- - -
r-7
- - -
t
50
40
II
I
ThllC-In,”
Figure 6. Prediction of contaminant North Penn distribution system
movement at three selected stations in the
bution system. The pilot sampling program required labor-intensive sampling at a number of sites at frequent intervals over several days. As the dynamic nature of system water quality became clear, questions arose as to the appropriate sampling frequency, and thedesirability of obtaining longer-term information (over a number of daily cycles) became evident. Because of the obvious difficulties of doing this type of sampling with a manual sampling program, automated samplers were sought. A review of commercially available automated samplers showed that none would be adequate for this application. Most such samplers fall into one of two categories: (1) continuous monitors that perform analysis continuously at a location; and (2) rotary samplers designed primarily to provide periodic samples from open channels. Continuous monitors are generally expensive and limit the types of analysis that can be performed, whereas rotary samplers are not generally designed to sample from a pressurized distribution system. The greatest drawback of these grab samplers, however, is that they are not designed to preserve volatile components such as trihalomethanes. No sample-bottlecappingmechanism exists
that will prevent exposure of the sample to the air with the consequent loss of volatile constituents. Scientists at the US Environmental Protection Agency (USEPA) Environmental Monitoring and Support Laboratory (EMSL) in Cincinnati have developed a prototype capping mechanisn that can be used with a rotary sampler; it seals the sample bottle after sampling is completed. This capper, fabricated of PTFE, has a sliding vane that seals the sample when the vane is closed. A wire arm extends from the capper and allows the vane to be set in theopen position. In use, the rotary arm of the sampler pauses above the sample bottle and discharges the sample to the bottle. The wire trigger mechanism then slides the vane shut as the sampler arm rotates to the next sampling position, thus sealing the filled bottle from air contact. This capper had been tested in the laboratory but not in field situations. Accordingly, with the cooperation of the EMSL, a request for proposals (RFP) was prepared and sent to manufacturers of water samplers. The RFP sought a sampler that employed the EMSL-developed capping system and was capable of preserving volatile organics from a water distribution system. Of five manufacturers contacted, positive
Copyright (C) 1990 American Water Works Association
responses were obtained from two.* One company was willing to provide a 24bottle rotary sampler and work with LJSEPA on fitting the cappers, but it was not willing to fabricate the EMSL cappers. The other proposed an alternative design that did not make use of the EMSL cappers but that would preserve volatileorganics. Thealternativedesign, as proposed, made use of 12 solenoidactuated valves connected to a sample manifold that would fill 12 bottles connected by PTFE tubing to the valves. When a bottle was filled, the solenoid valve would shut, closing off the sample bottle. The solenoid valves were operated by a microprocessor-controlled programmer, which allowed for user control of the interval between samples. In cooperation with EMSL, NPWA commissioned one sampler of each design. For the sampler utilizing the EMSL cappers, the cappers were fabricated locally under the direction of EMSL personnel. Both samplers were tested in the laboratory by USEPA personnel for preservation of volatile organics, and both were found to preserve in excess of 95 percent of volatile organics after 24 h. Both samplers, however, were modified to obtain better mechanical and electrical performance and then were shipped to NPWA for field testing. Results from field tests. Some mechanical problems developed with both samplers, particularly with the EMSL capping system. However, most of the problems were eventually solved. Table 1 compares the two custom-made samplers with manual samples as well as with each other. The cost of developing the sampler using the EMSL capper was approximately $7,300. The rotary sampler cost approximately $3,500, and the 24 cappers were fabricated at a cost of $140 each; additional sample bottles and a stronger stepper motor for the rotary sampler made up the rest of the cost. The other sampler was bid at $6,000, and no additional costs were incurred. It is likely that the cost of both designs could be reduced. The high cost of the cappers was a function of the material (PTFE) and of the custom nature of the work for small quantities. Both samplers, however, show promise as relatively lowcost methods for obtaining information on time-varying water quality for volatile compounds. Modeling water quality variations Based on the results of the field sampling studies, it is clear that there are wide variations in water quality in a distribution system both at a given point in time and at a given location. As a complement to the sampling studies, effort was devoted to examining and
implementing models that would predict water quality variations in a distribution system. In applying models to a water distribution system, the degree of temporal variation and the specificissues that are being studied determine the types of models that are most applicable. Steadystate modeling represents external forces as constant in time and determines solutions that would occur if the system is allowed to reach equilibrium. In dynamic modeling, demands and supplies are allowed to vary and the resulting timevarying solution is determined. In both steady-state and dynamic modeling, a distribution system is rep resented by a link-node network (i.e., pipes are represented as links and functions of pipes, wells, tanks, or intersection of pipes are represented as nodes). Hydraulic models are used to determine flows and velocities in links. Water quality models are used to determine variations in the concentrations of a contaminant throughout the distribution system. The hydraulic and water quality models may be an integrated package or the result of an application of a hydraulic model that may be input to a water quality model for subsequent analysis. Several attempts have been made to model water quality variations in a distribution system.zO-27Among the more interesting studies are those by Liou et al and Hart et al.28.29The model developed by Liou et al includes both an extended-period simulation capability for modeling the hydraulics of a system and a water contaminant propagation model. Contaminant propagation model Hydraulic models of water distribution systems have been widely used and documented.30.31 In steady-state hydraulic models, demands, well pumpage, and tank flows and heads are constant throughout the simulation, and the resulting steady-state (equilibrium) flow and velocity throughout the network are determined. Most dynamic hydraulic’ models of water distribution systems are more correctly classified as “quasidynamic” models in which demands and inflows are fixed for a time period and the resulting flow patterns and tank response are determined. During the next period, demands and inflows may be changed, and the hydraulicconditions from the previous period are used as initial conditions. Most models that represent water quality in a distribution system use the steady-state hydraulic solution along with constituent concentrations in water sources to determine the resulting steady-state spatial distribution of concentrations throughout the network.
Computer modeling employed in this study resulted in the development of a number of generally applicable models capable of being run on a desktop computer.? Models developed through the course of this study include: l a desktop-computer version of a data base model called the Water Supply Simulation Model (WSSM); l a steady-state flow-tracing model that can predict the contribution of flow from any given source in a multisource network to any node under known steady-state hydraulic conditions (In addition, this model can provide traveltime estimates for water from any source to any node in a water distribution network); and l a dynamic water quality model (DWQM) that makes use of detailed dynamic hydraulic information for a water distribution system to provide time and spatial predictions of water quality in a distribution system. All modeling and data base development work was originally conducted using a large mainframe computert at USEPA’s National Computer Center. As more powerful desktop computers and requisite software became available, computational efforts and model and data base development were converted to an advanced computer.§ The developed models made use of “external” hydraulic models. Originally, work was conducted using theKentucky Pipes model.30 Later on in the study, a public domain version of the WADIS0 model (developed by the US Army Corps of Engineers Waterways Experiment Station) became available and was used in both steady-state and extended-period hydraulic modeling.31 Steady-state prediction of mixing Initial efforts were devoted to obtaining a steady-state prediction of mixing within the distribution system using the existing capabilities of the WSSM.32.33 Solver-module predictions were compared with water quality sampling data to assess the spatial variations in quality.34 Average values for specific water quality variables at defined locations were calculated. It was assumed that purchased surface water for the system had achieved a steady-state relationship with regard to TTHM formation. It was also assumed that TTHM production for other sources in the system was zero. Therefore, TTHMs were utilized as a predictor of mixing and blending of surface water with well water. Reasonable agreement was found between *American Sigma, Middleport, N.Y. tIBM-PC. IBM Corp.. Armonk. SlBM 4381, IBM Corp., Armonk, §IBM-AT. IBM Corp., Armonk,
Copyright (C) 1990 American Water Works Association
N.Y.; N.Y N.Y. N.Y.
N-Con.
Larchmont.
solver-module predictions and the analyzed historical data, in terms of general patterns of blending, using TTHMs. It was observed, however, that “point prediction” capacity-the ability of the model to estimate accurately the level of a pollutant at a specific point in the distribution system-was limited.
Sequential steady-state modeling An initial attempt at modeling the dynamic water quality variations observed during the sampling study utilized successive applications of the steady state solver-module technique. Called sequential steady-state modeling, this technique was implemented by defining different steady-state hydraulic scenarios representative of different time periods within the sampling study and successively predicting hydraulics and the associated water quality for each period. This technique again yielded reasonable results in terms of overall patterns. Theoretically this approach fails to take into account water resident in the pipes, which is obviously in a blended state and is extremely cumbersome to handle logistically.
Dynamic water quality model To better represent observed water quality variations, the DWQM was developed; it is based on a routing technique that continually accounts for the quality of water resident in the distribution system.35 As with the steady-state model, the DWQM relies on externally available hydraulic information, in this case generated by using the WADIS0 model in extended-period simulation mode (i.e., with detailed hydraulic information relating to flows in pipes and at nodes available for each time step in the quality simulation). The DWQM was tested against the conditions of the November 1985 sampling study, for which good information on systemwide water use and tank levels was available. First, an extended-period hydraulic simulation was developed; it showed good agreement between predicted and actual tank levels and pressures in the system. This was then used to model the quality variations, and it again showed good agreement over the 34-h duration of the sampling run.
~&di~li~
of the dynamic model
The dynamic water quality algorithm was implemented as a microcomputer= based DWQM written in FORTRAN. Versions for two computers* were developed, and the model was applied in a full-scale demonstration on the NPWA distribution system.34-36 The DWQM was used to model the movement of constituents in the NPWA 52
RESEARCH
AND
system. A 34-h period was simulated, corresponding to conditions present during the pilot sampling program that had been conducted Nov. 14-15, 1985. Hydraulic conditions in the distribution system were determined using the WADIS0 hydraulic model, parameters of which were adjusted so that predicted tank levels and flows at selected sites represented those measured during the sampling period. Examination of the results of the application of the DWQM to the NPWA system revealed both close agreement between predicted and observed results as well as some anomalous behavior. For hardness, chloroform, and TTHMs, general concentrations compare favorably with observed values at three selected sampling stations (Figures 3-5). In each case there are some differences in the timingof peak or minimum values. When the spatial variation of predicted TTHM concentrations are compared with the historical average TTHM level, the same general patterns are apparent. Additionally, the predicted patterns bracket the pattern corresponding to the longterm historical average-a result that would be expected because the two selected times correspond to the extreme spatial patterns during the sampling period. There are several known factors that could contribute to the variation between predicted and observed values. These factors include the following: 0 Observed temporal variations’in the concentration of TTHM, chloroform, and hardness at Keystone were not represented in the simulation. l Field data on TTHM concentrations at the wells was not available, so assumed values were not based on observed values. l Because field data on hardness in well water were not availableduring the sampling period, representative values that were based on sampling at other times were used. l The nonconservative aspects of total trihalomethanes and chloroform were not represented in the model application. l Flows from the hydraulic model were aggregated into periods of various lengths, thus losing some accuracy in representing high-frequency hydraulic occurrences in the model.
Dynamic model and flow tracing The dynamic model was combined with the flow-tracing model to predict the movement of contaminants over time in the NPWA system. Based on the earlier analysis (Figure 2), contaminants can move back and forth in the pipes under varying demand and hydraulic scenarios. To extend the contaminant
propagation modeling capability, the tracking model was coupled with the dynamic hydraulic scenarios to predict contaminant movement. To illustrate this effect, an initial influent of 100 units of contaminants was assumed injected at the Keystone tie-in at time equal to zero. Assuming time periods that were representative of a typical day, the contaminant was tracked in the NPWA system. After 300 min, the contaminant was localized in the southeastern portion of the system. At 540 min, the contaminant had spread to the central portion of the system. By 930 min, the contaminant was well dispersed. At 1,440 and 1,560 min, the contaminant began to disappear, and by 2,040 min (nearly a day and a half), the contaminant remained at very low levels but was virtually gone from the system. To more clearly illustrate this effect, a time trace of contaminant movement was calculated at three points in the system (Stations A, B, C as shown in Figure 1). Figure 6 shows the predicted time traceof thecontaminant at stations A, B, and C. Station A is where the contaminant moves into the system, station B is a point of maximum mixing in the system in which water tends to move back and forth, and station C represents the retention of contaminant in the pipe as water moves back and forth past a given point.
Conclusions and recommendations A clear result from this research is the need toobtain morerepresentative monitoringresults than are normally acquired from distribution system sampling. Contaminant values can vary greatly over a relatively short time at a given point, and weekly and yearly cycles can combine with hydraulic and mixing variations to affect contaminant levels at a given point in a distribution system. To provide insight into these variations, several automated samplers were tested for installation in the NPWA system. A rotary sampler fitted with a USEPAdesigned capping system was field tested. This sampler allowed samples to be taken at specified intervals for several days, and the approach appears promising. Steady-state predictive modeling of water quality can provide insight into overall water quality variations and patterns within a distribution system. Interpretation of predictive modeling results must be made in light of an appreciation of system hydraulics, in particular an understanding of the flow patterns and directions that create the gradients of concentration. Quality modeling is based on hydraulic modeling and *MacIntosh,AppleComputers, IBM Corp., Armonk, N.Y.
TECHNOLOGY
Cupertino,
Calif.;IBM-PC,
JOURNAL
Copyright (C) 1990 American Water Works Association
AWWA
is thus highly sensitive to hydraulic modelingassumptions and results. Field data on water quality is important in developing, verifying, and understanding predictive models. Such data should be available on a time interval sufficient to reflect daily changes in system dynamics. Having the tools to predict time-of-travel between points in a system and to estimate the quality of water provided to any point from any source will allow for realistic water quality monitoring strategies. Future research should include examination of the sensitivity of the steadystate models to changes in hydraulic assumptions. Further algorithmic development should be devoted to additional development and testing of dynamic models.
Acknowledgment This study was conducted as a cooperative agreement between USEPA and the NPWA. Harry Borchers, executive director of NPWA, encouraged and supported the project throughout and provided the knowledge of the NPWA system necessary for successful project completion. Dale Reichenbach of NPWA provided much of the detailed technical information and system data and played a major role in testing the automated samplers. Joseph Roessler, formerly of USEPA’s EMSL, provided many hours of advice, assistance, and testing in development of the automated samplers. Richard Males, RNM Technical Services; Walter Grayman, consulting engineer; and James Goodrich, USEPA Drinking Water Division, assisted throughout all phases of the analysis. Diane Routledge and Patricia Pierson of USEPA helped prepare this article.
References M.T. & GELDREICH, E.E. Distribution Line Sediments and Bacterial Regrowth. Proc. 1977AWWA WQTC. Kansas City, MO. 2. CLARK,R.M. & MALES,R.M. Simulating Cost and Quality in Water Distribution. 1. ALLEN,
JOUY. Water Resources Planning & Management--AXE, 11:454(1985).
3. CLARK,R.M. & MALES,R.M. Developing and Applying the Water Supply Simulation Model.Jour. A WWA, 7&l&61 (1986). 4. MALES,R.M. ETAL.Algorithm for Mixing Problems in Water Systems.]our. Hyde. Engm.-AXE,
111:206 (1985).
5. CLARK,R.M. ET AL. Development, Application and Calibration of Models for Predicting Water Quality in Distribution Systems. Proc. 1988 AWWA WQTC, St. Louis, MO. 6. Res. Rept., Water Research Centre. Deterioration of Bacteriological Quality of Water During Distribution. Water Research No. 6 (Oct. 1976). 7. GELDREICH,E.E. ETAL. The Necessity of Controlling Bacterial Populations in PoAUGUST 1990
24. TANNER,T.L. Adopting a Pipe Network Analysis Computer Program. Proc. 1985 ASCE Water Resources Planning and 8. MAUL, A.; EL-SHAARAWI,A.H.; & BLOCK, Management Div. Swcialty Conference J.C. Heterotrophic Bacteria in Water on Cokputer Appli&tions-in the Water Distribution Systems. I. Spatial and Industry, New York. Temporal Variation. Sci. of the Total 25. CLARK,R.M.; LYKINS,B.W.; & GOODRICH, En&., 44:201(1985a). J.A. Infrastructure and Maintenance of A.H.; & BLOCK, 9. MAUL, A.; EL-SHAARAWI, Water Quality. Proc. 1985 AWWA DisJ.C. Heterotrophic Bacteria in Water tribution System Symposium, Seattle, Distribution Sistems. II. Sampling Design for Monitoring. Sci. of the Total Wash. Envir., 44:215 (1985b). 26. KROON,J.R. & HUNT, W.A. Modeling 10. MCCOY,W.F. & OLSON,B.H. Relationship Water Quality in the Distribution NetAmone Turbiditv Particle Counts and work. Proc. 1989 AWWA Distribution Bactehological duality Within Water System Symposium, Dallas, Texas. Distribution Lines. Water Res., 20:1023 27. CHARACKLIS, W.G. Bacterial Regrowth in (1986). Distribution Systems. AWWARF, Den11. LE~HEVALLIER,M.W.; BABCOCK,T.M.; & ver, Colo. uan. 1988). LEE, R.G. Examination and Characteri28. Lrou, C.P. & KROON,J.R. Modeling the zation of Distribution Svstem Biofilms. Propagation of Waterborne Substances Appl. & Envir. Microbiol.~53:2714(1987). in Distribution Networks. Jour. A WWA, 12. DONLON,R.M. & PIPES,W.O. Pipewall 79:11:54 (Nov. 1987). Biofilm in Drinking Water Mains. Proc. 29. HART, F.i.; MEADER,J.L.; & CHIANG,S.1986 AWWA WQTC, Portland, Ore. M. CLNET-A Simulation Model for Tracing Chlorine Residuals in a Potable 13. PISIGAN,R.A. JR.& SINGLEY,E J. Influence of Buffer Capacity, Chlorine Residual, Water Distribution Network. Proc. 1986 and Flow Rate on Corrosion of Mild Steel AWWADistribution System Symposium, and Copper. Jour. A WWA, 79:2:62 (Feb. Minneapolis. Minn. 1987). 30. WOOD,DJ. & RAVES,A.G. Reliability of T.F.; & BENJAMIN, 14. REIBER,S.H.; FERGUSON, Algorithms for Pipe Network Analysis. M.R. Corrosion Monitoring and Control Jour. Hydr. Div.-ASCE, 107:1145(1981). in the Pacific Northwest. Jour. A WWA, 31. GESSLER,T. & WALSKI, T.M. Water 79:2:71 (Feb. 1987). Distribution System Optimization. TREL15. MCCLELLAND,N.I. & MALJRY,K.H. Water 35-11, Waterways Experiment Station, Quality Monitoring in the Distribution US Army Corps of Engineers, Vicksburg, System. EPA-600/2-77-074(Mar. 1977). Miss. (Oct. 1985). 16. SCHIMPFF,W.K. ET AL. Water Quality 32. CLARK,R.M. & MALES,R.M. Developing Effects Related to Blending Waters in the and Applying the Water Simulation Distribution System, USEPA Rept. Model./our. A WWA, 78:8:61(Aug. 1985). EPA-600/2-80-132(1980). 33. MALES:R.M. ETAL.Algorithm for Mixing 17. MATSON,T.V. & ~HA~CKLIS, W.G. DifProblems in Water Svstems. Tour. Hvdr. fusion Into Microbial Aggregates, Water Engrg:AXE, lll:iO6 (198%). Res., 10~877(1976). 34. MALES,R.M.; GRAYMAN,W.M.; & CLARK, L.S. SeparatedFlow Conditions 18. GALOWIN, R.M. Modeling Water Quality in Distriat Pipe Wells of Distribution Mains. Risk bution Systems. Jour. Water Resource Reduction Engrg. Lab., USEPA Office of Planning&Management-AXE, 114:197 Res. & Devel. Cincinnati, Ohio. (1988). 19. EILERS,R.G. & CLARK,R.M. Flow Separ- 35. GRAYMAN, W.M.; CLARK, R.M.; & MALES, ation Conditions at Pipe Wells of Water R.M. Modeling Distribution System WaDistribution Mains. Proc. 1988 AWWA ter Quality: A Dynamic Approach. Jour. Ann. Conf., Orlando, Fla. Water Resource Planning & Manaae20. CHUN,D.G. & SELZNICK,H.L. Computer ment-ASCE, 114:295(1988). Modeling of Distribution Svstem Water 36. CLARK, R.M. ET AL. Modeling ContamQuality.-Proc. 1985 ASCI? Water Reinant Propagation in Drinking Water sources Planning and Management Div. Distribution Systems. Aqzta, 3:137(1988). Specialty Conference on Computer Applications in the Water Industry, New About the authors: York. Robert M. Clark is 21. METZGER,I. Water Quality Modeling of Distribution Svstems. Proc. 1985 ASCE director of the DrinkWater Resources Planning and ManageWater Research ment Div. Specialty Conference on Computer Applications in the Water Industry, ronmental Protection New York. Agency (USEPA), 26 22. SARIKELLE,S.; MEHRFAR,K.E.; & CHUANG, W. Martin Luther Y.T. Incorporating Graphics in Water Distribution Svstems. Proc. 1985 ASCE Water Resources Planning and Manage- OH 45268. Clark has been a US Public ment Div. Specialty Conference on Com- Health Serviceofficersince 1961, detailed puter Applications in the Water Industry, to the USEPA in 1970. He has been New York. director of the D WRD since 1985 and is a 23. SHAWCROSS, J.F. Modeling Complex Water Distribution Systems. Proc. 1985 ASCE ?nQmbQY of A WWA, ASCE, and AAEE. Water Resources Planning and Man- At the time of this study, Judith A. Coyle agement Div. Specialty Conference on was water quality manager, North Penn Water Authority, 200 N. Chestnut St., Computer Applications in the Water tinsdale, PA 19446, Industry, New York. table Water: Community Water Supply.
Jour. A WWA, 64:9:596 (Sept. 1972).
ROBERTM. CLARK &JUDITH A. COYLE 53 Copyright (C) 1990 American Water Works Association