Table of Contents Next Section
Contamination of groundwater resources in Dover Township, Ocean County, New Jersey (Figure 1), including the contamination of water-supply wells, was identified in the 1960s (Toms River Chemical Corporation 1966) and subsequently documented in the 1970s (ATSDR 2001a,b,c,d). Water-quality analyses, conducted since the mid-1980s, indicate this contamination has generally consisted of volatile organic compounds such as trichloroethylene and semi-volatile organic compounds such as styrene-acrylonitrile trimer (ATSDR 2001d). Based on public health assessments conducted for the Dover Township area, ATSDR and NJDHSS have determined that completed human exposure pathways to groundwater contaminants have occurred through private and community water supplies (ATSDR 2001a,b,c,d). As a result, NJDHSS and ATSDR are conducting an epidemiologic study of childhood leukemia and nervous system cancers that occurred in Dover Township. The epidemiologic study is exploring a variety of possible risk factors, including environmental exposures. To assist NJDHSS with the environmental exposure assessment component of the epidemiologic study, ATSDR developed a water-distribution model using the EPANET 2 software. Results obtained from the model will be used to assess exposure to drinking water sources that are being investigated as potential risk factors in the epidemiologic investigation.
Because of the lack of appropriate historical data, the EPANET 2 model was calibrated to the present-day (1998) water-distribution system characteristics using data collected during March and August 1998. The reliability of the calibrated model was demonstrated by successfully conducting a water-quality simulation of the transport of a naturally occurring conservative element—barium—and comparing results with data collected at 21 schools and 6 points of entry to the water-distribution system during March and April 1996. Results of the field-data collection activities, model calibration, and reliability testing were described previously (Maslia et al. 2000a,b). Following calibration, the model was used to simulate historical characteristics of the water-distribution system serving the Dover Township area from 1962 through 1996.
This report describes the historical reconstruction analysis of the water-distribution system serving the Dover Township area. It is viewed as a companion document to Maslia et al. (2000a) which describes the analysis of the 1998 water-distribution system. Therefore, the report focuses on these aspects of the historical reconstruction analysis: (1) data sources and requirements, (2) methods of analysis, (3) simulation strategies, (4) selected simulation results, and (5) the use of sensitivity analysis to address issues of uncertainty and variability of historical system operations.
Given the paucity of historical contaminant-specific concentration data during most of the period relevant to the epidemiologic study, ATSDR and NJDHSS decided that modeling efforts should concentrate on estimating the percentage of water that a study subject might have received from each point of entry (well or well field) to the water-distribution system (Plates 3–37). This approach uses the concept of "proportionate contribution" described in Maslia et al. (2000a, p. 4) wherein at any given point in the distribution system, water may be derived from one or more sources in differing proportions.
Databases were developed from diverse sources of information and were used to describe the historical distribution- system networks specific to the Dover Township area. These data were applied to EPANET 2 and simulations were conducted for each month of the historical period—January 1962 through December 1996 (420 simulations or "model runs"). After completing the 420 monthly analyses, source-trace analysis simulations were conducted to determine the percentage of water contributed by each well or well field operating during each month. Results of these analyses—the percentage of water derived from the different sources that historically supplied the water-distribution system—were provided to health scientists for their analysis in assessing the environmental factors being considered by the epidemiologic investigation.
A simulation approach to the historical reconstruction of the water-distribution system in the Dover Township area required knowledge of the functional as well as the physical characteristics of the distribution system. Accordingly, six specific types of information were required: (1) pipeline and network configurations for the distribution system; (2) potable water-production data including information on the location, capacity, and time of operation of the groundwater production wells; (3) consumption at locations throughout the distribution system; (4) storage-tank capacities, elevations, and water-level data; (5) high-service and booster pump characteristic curves; and (6) system-operations information such as the on-and-off cycling schedule of wells and high-service and booster pumps, and the operational extremes of water levels in storage tanks.
Yearly historical network configurations maps the period 1962 through 1996 are presented on Plates through 37. These maps show the complexity of the system increased significantly over the time span of the historical period. For example, the 1962 water-distribution system served nearly 4,300 customers from a population of about 17,200 persons (Board of Public Utilities, State of New Jersey 1962) and was characterized for modeling by (Plate 3):
By contrast, in 1996—the last year of the historical reconstruction period—the water-distribution system served nearly 44,000 customers from a population about 89,300 persons (Board of Public Utilities, State New Jersey 1996) and was characterized for modeling by (Plate 37):
Analysis of production data indicates that the historical distribution systems could be characterized by three typical demand periods each year: (1) a low- or winter-demand period, generally represented by the month of February—designated as the minimum-demand month; (2) a peak- or summer-demand period, represented by one of the months of May, June, July, or August—designated as the maximum-demand month; and (3) an average-demand period, generally-represented by the month of October—designated as the average-demand month.
Water-production data were gathered, aggregated, and analyzed for each well for every month of the historical period (Appendix B). These data were obtained from the water utility (Flegal 1997), the annual reports of the Board of Public Utilities, State of New Jersey (1962–1996), and NJDHSS data searches (Michael P. McLinden, written communication, August 28, 1997). The production data were measured by using in-line flow meters at water-supply wells (George J. Flegal, Manager, United Water Toms River, Inc., oral communication, August 28, 2001).
Monthly production data were represented graphically in a three-dimensional plot (Figure 12). Referring to this plot, the x-axis is the year (1962–96), the y-axis is the month (January–December), and the z-axis is the total monthly production in million gallons. Maximum production is shown to occur in the months of May, June, July, or August. In addition, considerable production increases occurred in 1971, 1988, and 1995. These years are characterized on the plot by sharp peaks.
To simulate the distribution of water for each of the 420 months of the historical period, network configuration, consumption, and operational information were required. Before 1978, operational data were unavailable requiring development of system-operation parameters —designated as "Master Operating Criteria" (Table 4). These are based on hydraulic engineering principles necessary to successfully operate distribution systems similar to the one serving the Dover Township area. From 1978 forward, for selected years, operators of the water utility provided information on the generalized operating practices for a typical "peak-demand" (summer) and "non-peak demand" (fall) day. These guidelines were used in conjunction with the "Master Operating Criteria" to simulate a typical 24-hour daily operation of the water-distribution system for each month of the historical period.
Examples of historical water-distribution system operating schedules for the minimum-, maximum-, average-demand months of 1962, 1965, 1971,1978, 1988, 1995, and 1996 are presented in Appendix (Tables D-1 through D-7). These tables indicate hour-by-hour operation of wells and high-service booster pumps during a typical day of the minimum-, maximum-, and average-demand month for the given year. In 1962 and 1965 (Tables D-1 and D-2, respectively), high-service and booster pumps were not part the distribution system and, therefore, only groundwater wells were operated to supply demand by discharging water directly into the distribution system. In 1968, high-service and booster pumps were added to the distribution system. From that year forward, some wells supplied storage tanks, then high-service and booster pumps were operated to meet distribution-system demands; other wells still discharged directly into distribution system (refer to Tables D-1 through D-7 in Appendix D for details).
In this type of study, the ideal or desired condition is to obtain all data required for model simulations through direct measurement or observation. In reality, however, necessary data are not routinely available direct measurement or observation and must be synthesized using generally accepted engineering analyses methods. Issues of data sources and the methods used obtain data that cannot be directly measured reflect, ultimately, on the credibility of simulation results. address these issues for historical reconstruction analysis, the methods for obtaining the necessary data were grouped into three categories (Table 12):
Of the six specific types of information required for the historical reconstruction analysis, the network pipeline data, groundwater well-location data, groundwater well-production data, and storage-tank data were obtained by direct measurement or observation. These data were available throughout the entire historical period and they could be assessed for quality and verified by independent means such as state reports or field observations. For example, groundwater well-production data were available for every well for every month of the historical period and these data were measured by the water utility using in-line flow-metering devices at groundwater wells (George J. Flegal, Manager, United Water Toms River, Inc., oral communication, August 28, 2001).
Data for historical consumption consisted of two components—monthly volumes (quantity) and spatial distribution (location). The monthly volumes were obtained by using a quantitative estimation method. Data were available from metered billing records for October 1997 through April 1998 and verified through the calibration process described in Maslia et al. (2000a,b); the magnitude of monthly historical production was known based on measured flow data. Using these data, estimates of historical consumption were quantified by imposing the requirement that total consumption must equal total production.
Direct measurement or quantitative estimates of the spatial distribution of historical point- demand data (demands at specific pipeline locations) were not available for the Dover Township area. Therefore, qualitative description methods were used to estimate historical data values. In doing so, estimates of the spatial distribution of historical point-demand data were based on two assumptions: (1) historical demand patterns were similar to the present-day demand patterns which are known from available metered billing records; and (2) demand patterns could be inferred from land-use classification using historical land-use classification as a surrogate indicator. To assess the validity of this approach, historical land-use classification or zoning maps for Dover Township were used in conjunction with distribution-system network maps for 1962, 1967, 1978, 1990, and 1996 (network maps like the ones shown on Plates 3, 8, 19, 31, and 37, respectively). Using information obtained from the land-use classification and distribution-system network maps, geospatial and comparative analyses were conducted (Table 3). Results of these analyses indicated that the distribution of land-use classification in Dover Township was relatively static and changed little during the historical period. These analyses substantially validated the qualitative description method used to estimate the spatial distribution of historical demand.
The high-service and booster pump-characteristic data were derived using information obtained from the water utility (Flegal 1997). This information consisted of head values versus flow values which were refined during the model calibration process (Maslia et al. 2000a,b).
The historical system-operation data were obtained using each of the three methods of obtaining data described previously—depending on the time frame (Table 12). For the early historical period (1962–77), investigators relied on hydraulic engineering principles and the "Master Operating Criteria" (Table 4). Because data describing specific operational practices were not available, operating schedules developed for these early historical networks were based on qualitative descriptions of system operations. To maintain a balanced flow condition, however, water-distribution systems of similar configuration and facilities as the historical Dover Township area system generally operate using on-andoff cycling schedules of limited variability. That is, wells and high-service and booster pumps must be cycled on-and-off within a limited or narrow operating range. Simulations conducted on the water-distribution system serving the Dover Township area confirmed the limited variability of the on-and-off cycling operating schedule.
For the 1977–1987 period, system-operation data were developed from quantitative estimates and qualitative descriptions of the operating schedules. These data were derived using hydraulic engineering principles, the "Master Operating Criteria," and from information provided by the water utility that described the general operations of the water-distribution system for a typical "peak" day (summer) and a "non-peak" (fall) day. For some of the years, the water utility also provided estimates of discharge to the distribution system from the high-service and booster pumps (Richard Ottens, Jr., Production Manager, United Water Toms River, Inc., written communication, 1998).
System-operation data for the most recent historical systems (1988–96) were obtained from direct measurement or observation, quantitative estimates, and qualitative descriptions of operating schedules. Data sources used to develop these operating schedules (for example, Tables D-6 and D-7) included the generalized operating notes from the water utility (Richard Ottens, Jr., Production Manager, United Water Toms River, Inc., written communication, 1998), hourly operations data for 1996 (Flegal 1997), notes taken by ATSDR and NJDHSS staff during field-data collection activities in March and April 1998 (Maslia et al. 2000a), and the observation that the distribution system had previously operated in a manner very similar to the present-day system (1998), for which detailed information was available.
Simulation of water-distribution networks require detailed descriptions of network operations, such as the on-and-off scheduling of high-service and booster pumps and groundwater wells for the entire period of simulation. In order to simplify these rigorous data requirements, a surrogate or alternative method—designated the "source-node-link" or SNL simulation method (Figure 19)—was devised whereby balanced flow conditions were maintained and the measured volumes of monthly water production were used while avoiding the need for detailed network operations data, which were not available for most of the historical period. Comparison of flow results obtained using the surrogate SNL simulation method with measured flow data obtained during August 1998 for the Holly and Parkway treatment plants showed that the SNL method simulated nearly identical flows to those measured (Figure 20 and Table 16).
Analysis of the proportionate contribution of water from wells and well fields to selected network locations in the Dover Township area illustrates the increasing complexity and operational variability of the distribution system throughout the historical period. These results were obtained by conducting source-trace analysis simulations. Results are presented as areal distributions of the simulated proportionate contribution water from active wells or well fields to all locations serviced by the water-distribution system for selected years 1962, 1965, 1971, 1978, 1988, 1995, and 1996 (minimum-, maximum-, and average-demand months; Plates 52–153). The annual variation of the simulated proportionate contribution of water from operating wells and well fields to selected locations in the Dover Township area is shown for the minimum-demand month of February (Figure 23), the maximum-demand months May, June, July, or August (Figure 24), and the average-demand month of October (Figure 25). For each of these examples, five geographically distinct pipeline locations were selected from the historical networks to represent the spatial distribution of proportionate contribution results. These locations are identified as locations A, C, D, and E.
Comparison of the May 1962 results with the June 1996 results (Figure 24), indicates the increasing complexity of the network and distribution-system operations and how such operations influenced the proportionate contribution of water to specific locations. In May 1962, only two well fields (Holly and Brookside) provided water to any one location; whereas, June 1996, as many as seven well fields provided water to the distribution system (for example, pipeline location E in Figure 24).
Simulation results for the maximum-demand months of May 1962, July 1971, June 1978, July 1988, August 1995, and June 1996 for pipeline location exemplify the annual variation in the contribution water to this location and indicate the following (Figure 24):
The simulation results shown in Figures 23 through 25 demonstrate that the contribution of water from wells and well fields varied by time and location. However, the results also show that certain wells provided the predominant amount of water to locations throughout the Dover Township area. The proportionate contribution of water from specific water sources at specified times during the historical period of 1962 through 1996 are provided on Plates 52 through 153.
The proportionate contribution results described above were obtained from trace-analysis simulations conducted on the historical distribution-system networks whereby balanced flow conditions were achieved through the manual refinement of modeling parameters. The adjusted parameters were the on-and-off cycling pattern values (pattern factor values assigned in EPANET 2) of wells and supply nodes representing wells linked to storage tanks and high-service and booster pumps and the operational extremes of water levels in the storage tanks. This modeling approach was designated as the "manual adjustment process." Simulation results presented in Figures 23–25, on Plates 52–153, and in Appendices H and I were obtained using the manual adjustment process and were the bases of comparisons for all sensitivity analyses.
To address the issue of uncertainty and variability of system operations, and specifically to test the sensitivity of the proportionate contribution results to variations in model-parameter values, a technique was required that would "search" for and select a set of alternate operating conditions different from those determined using the manual adjustment process. These alternate operating conditions needed also to result in the satisfactory operation of the historical water-distribution system. Such a technique was found in the Genetic Algorithm optimization (GA) method which refers to a method of optimization that attempts to find the most optimal solution by mimicking (in a computational sense) the mechanics of natural selection and natural genetics. (Details of the methodology and the application of the method to water-distribution system operations is presented in Appendix E.)
Four types of operational and hydraulic constraints were varied during sensitivity analyses in order to determine the effects of constraint changes on the simulated proportionate contribution results. The constraints subjected to variations were (Table 20): (1) pattern factors assigned to wells and supply nodes—designated as sensitivity simulations SENS0, SENS1, SENS2, and SENS3; (2) minimum pressure requirements at model nodes—designated as sensitivity simulations SENS4 and SENS5; (3) allowable storage tank water-level differences between the starting time (0 hours) and ending time (24 hours) of a simulation—designated as sensitivity simulations SENS6 and SENS7; and (4) daily system operations represented by a "typical" 24-hour day over a month-long period—designated sensitivity simulation SENS8. For the first three types of constraints (SENS0–SENS7), the GA optimization methods were used to obtain simulation results for the proportionate contribution of water at all pipeline locations, and, these results were compared with results previously obtained using the manual adjustment process. For the fourth type of constraint variation (SENS8), the manual adjustment process was used to obtain simulation results for the sensitivity analysis. Descriptions of parameter variations for the sensitivity analyses are listed in Table 20 and the simulation month and year are listed in Table 21.
Results for the sensitivity analysis simulations using the GA methods representing 1962, 1965, 1971, 1978, 1988, 1995, and 1996 conditions are presented in Appendix I (Tables I-1 through I-7) and Appendix J (Figures J-1 through J-7). Analysis of these results indicate small variations when comparing the proportionate contribution results from the manual adjustment process to results obtained using the GA methods (Figure 27). Furthermore, analyses of differences in the simulation results (Appendix K and Figure 31) show that the simulated proportionate contribution of water from wells and well fields is relatively insensitive to changes in system operational parameters. For a 24-hour period, the average percentage of water over all study locations derived from all wells or well fields using either the manual adjustment process or any of the GA methods does not vary appreciably. Statistical analyses of the differences in simulated proportional contribution results obtained using the manual adjustment process and GA methods showed that differences are normally distributed for study locations characterized by the six selected historical networks for years 1962, 1965, 1971, 1978, 1988, and 1996 (Figure 32). These analyses further indicated that, overall, the difference distributions were characterized by a mean, mode, and median of nearly 0% and a standard deviation of less than 4% (Table 23). The sensitivity analyses indicated that the differences in the proportionated contribution of water—simulated by the exhaustive range of operating conditions and hydraulic constraints (Table 20)—are insensitive to the manner in which the water-distribution system was operated over a 24-hour period. As a consequence, the minor differences in the simulated proportionate contribution of water between the manual adjustment process and the GA simulation approach indicate that there was a narrow range within which the historical water-distribution system could have successfully operated to maintain a balanced flow condition and satisfy the "Master Operating Criteria."
For the historical reconstruction analysis, investigators assumed that daily system operations over a period of one month could be represented by a "typical" 24- hour day for each month of the historical period. To test the validity of this assumption, additional sensitivity analyses (SENS8) using hourly operational data obtained from the water utility were conducted. Monthlong simulations were conducted for February, June, and October which represented, respectively, the minimum-, maximum-, and average-demand months for 1996. Simulations were conducted using the manual adjustment process according to the hourly operational data for 1996 supplied by the water utility. When results for the month-long simulations (averages over the month-long period) were compared with results from the "typical" 24-hour day, differences in the proportionate contribution of water to the five pipeline locations (A–E) showed only slight variations (Figure 33). As an example, for June 1996, the difference in the contribution of water from the Parkway well field for the two methods of simulating the daily system operations were 0% for location A, 1% for location B, 4% for location C, 2% for location D, and 3% for location E. Therefore, sensitivity analysis assisted in confirming that the day-to-day operations of the water-distribution system were highly consistent over a month-long period (based on available 1996 hourly data) and could be represented by a "typical" 24-hour operational pattern.
The sensitivity analyses conducted as part of the historical reconstruction of the water-distribution system serving the Dover Township area indicate that: (1) there was a narrow range within which the historical water-distribution systems could have successfully operated and still satisfy hydraulic engineering principles and the "Master Operating Criteria," and (2) daily operational variations over a month did not appreciably change the proportionate contribution of water from specific sources when compared to a typical 24-hour day representing the month.
Overall, the simulation results for the proportionate contribution of water from wells and wells fields indicate variation by time and location. However, the results also show that certain wells provided the predominant amount of water to locations throughout the Dover Township area. In summary, therefore, the reconstructed historical water-distribution systems and operating criteria—based on applying the "Master Operating Criteria" and using generalized water-utility information—are believed to be plausible and realistic scenarios under which the historical 1962–96 water-distribution system was operated.
AVAILABILITY OF MODEL INPUT DATA AND PROPORTIONATE CONTRIBUTION RESULTS FILES
EPANET 2 compatible input data sets developed to conduct the monthly historical simulations for January 1962–December 1996, using the manual adjustment process, are provided with this report in a computer disc-read only memory (CD-ROM) format. The CD-ROMs contain the INP file formats described in the EPANET 2 Users Manual. Additionally, each CD-ROM contains a fully executable copy of the public-domain EPANET 2 water-distribution system model (Version 2.0, Build 2.00.08) that was used to conduct the historical monthly simulations, and the EPANET 2 Users Manual.
Also included on the CD-ROMs are data files that contain digital (electronic) results shown on Plates 52 through 153. These data files contain the nodal values of simulated proportionate contribution of water from each operating well or well field to all water-distribution system pipeline locations—obtained using the manual adjustment process—for the minimum-, maximum-, and average-demand months for seven selected years 1962, 1965, 1971, 1978, 1988, 1995, and 1996. The files are prepared in "text," "Excel," and "DBF" formats.
Readers desiring information about the model input data files or the proportionate contribution result data files contained on the CD-ROMs may also contact the senior author of the report at the following address:
Morris L. Maslia, P.E.
Agency for Toxic Substances and Disease Registry
1600 Clifton Road, Mail Stop E-32
Atlanta, Georgia 30333
Telephone: (404) 498-0415
Facsimile: (404) 498-0069
E-mail: mmaslia@cdc.gov
The authors acknowledge their colleagues at ATSDR for providing assistance and advice with all aspects of this investigation: John E. Abraham, Patrick Brady, James A. Clark, Amy B. Funk, Amanda J. Gonzalez, Virginia Lee, Susan Metcalf, Thomas A. Mignone, Jennifer Noack, Mary G. Odom, Sven E. Rodenbeck, Kevin A. Ryan, Kathy Skipper, Allan S. Susten, Gregory V. Ulirsch, and Ann Walker.
The following organizations and their staff provided suggestions, assistance, or information: New Jersey Department of Health and Senior Services (NJDHSS); Ocean County Health Department; Dover Township Municipal Government; Citizens Action Committee on Childhood Cancer Cluster; Multimedia Environmental Simulations Laboratory at the Georgia Institute of Technology; New Jersey Department of Environmental Protection; New Jersey Geological Survey; U.S. Environmental Protection Agency, Region II; U.S. Geological Survey; United Water Toms River, Inc.; Union Carbide Corporation; and Ciba-Geigy Corporation.
The authors would like to specifically acknowledge Jerald A. Fagliano, Michael Berry, and Jonathan E. Savrin of NJDHSS for suggestions and advice from the epidemiologic perspective; Michael J. McLinden, formerly with NJDHSS, for assistance in obtaining historical production data; Lewis A. Rossman, U.S. Environmental Protection Agency, National Risk Management Research Laboratory, for assistance with and requested modifications to the EPANET 2 water-distribution system model; S. Jack Alhadeff and Thomas R. Dyar, U.S. Geological Survey, Center for Spatial Analysis Technologies, for assistance with preparing historical aerial photographs; and Robert E. Faye, for assistance with data collection and computer simulation.
The authors also acknowledge Steven G. Buchberger, Donald V. Chase, Robert E. Faye, Walter Grayman, Frederick Hart, Lewis A. Rossman, and James G. Uber for providing review of the this report. Naida Gavrelis, Eastern Research Group, Inc., provided logistical support in coordinating review of the report.
Carolyn A. Casteel, Caryl J. Wipperfurth, and Bonnie J. Turcott, U.S. Geological Survey, Atlanta, Georgia, assisted with the preparation of text and illustrations for final printing.