ANALYSIS OF THE 1998 WATER-DISTRIBUTION SYSTEM
SERVING THE DOVER TOWNSHIP AREA, NEW JERSEY:

Field Data Collection Activities and
Water Distribution System Modeling

Calibration Statistics

In an effort to assess the overall quality and reliability of the model calibration, analyses were conducted using calibration statistics. Because measured data for both March and August 1998 were collected in terms of pressure at the test hydrant locations, analysis of calibration statistics will be presented in terms of measured and simulated pressures. The calibration statistic used for the analyses, referred to as "mean absolute pressure difference(P)," is defined as the mean of the absolute value of the difference between measured pressure values and simulated pressure values in pounds per square inch (psi) over the 48-hour duration of the test. This calibration statistic is defined mathematically as:

(9)

where:

P	= mean absolute pressure difference, (psi),
NHM	= number of hourly measurements,
	= measured pressure at hour i, (psi),
	= simulated pressure at hour i, (psi), and
	= the absolute value of pressure difference, (psi).

Results of the calibration analysis are presented as a series of illustrations (Figures 11, 12, and 13; Plates 10 and 11) and are described below.

Figure 11. Comparison of mean absolute pressure difference (P) for test hydrants, March and August 1998 simulations.

Figure 11

Figure 12. Comparison of measured and simulated hourly pressure data: (A) March 24-25, 1998; and (B) August 14-15, 1998.

    
Figure 12A            Figure 12B

A comparison of P  by hydrant (or measuring point) location is presented in Figure 11. The bars on this graph were computed by applying Equation (9) over the 48-hour test period for each location. The data used for the graph indicate that for March 1998, P  for all locations ranges from 1.4 psi to 5.3 psi, and for August 1998, P  for all locations ranges from 2.9 psi to 6.6 psi. (The absolute pressure differences and P  for all hydrants for the tests are also provided in Appendices I [March 1998] and J [August 1998].) System-wide P  for March 1998 is lower than for August 1998, and with the exception of hydrants H-2 and H-3, P  for March 1998 is less that 5 psi. For August 1998, system-wide P  is also less than 5 psi except for hydrants H-4, AH-5, H-9, and H-13.

To assess the goodness-of-fit of the calibrated model to field conditions, a comparison of measured hourly pressure values with simulated hourly pressure values is presented in Figure 12A (March 1998) and 12B (August 1998). A different plotting symbol is used for each test hydrant location. Ideally, if no difference between measured and simulated pressure data occurred, each hourly data point would coincide with the lines of equality shown in Figure 12. Because a difference exists between measured data and model simulated results, the data scatter about the lines of equality. For the model simulations, the August 1998 data have more scatter than do the March 1998 data, indicating a larger difference between measured and simulated pressure data.

The reliability of the calibrated model can also be judged in terms of the frequency of absolute pressure difference (absolute value of difference between measured and simulated hourly pressure data). Figure 13 is a plot of the frequency of absolute pressure difference for both March and August 1998. The graph indicates that for the March 1998 simulation, 90% or more of the simulated hourly values result in an absolute pressure difference of about 5 psi or less. For the August 1998 simulation, data presented in Figure 13 indicate that 90% or more of the simulated hourly values result in an absolute pressure difference of about 7.5 psi or less.

To view the areal distribution of P  in terms of the spatial locations of the test hydrants, P  for each test hydrant is plotted using the test hydrant location map from Plate 6. The P  for each test hydrant is shown on Plates 10 (March 1998) and 11 (August 1998). Comparison of Plates 10 and 11 indicates that the largest P occur in the southern Dover Township and Berkeley Township areas, and that overall, the March 1998 simulation has a smaller P. Primarily we attribute this characterization of the difference in pressure to the lack of metered consumption data available specifically during the test periods. As discussed above, metered consumption data were available for the Dover Township area serviced by the water-distribution system solely on a quarterly basis; thus, each meter was read four times per year. However, all meters were not read at the same time nor within a few days of each other. For the March 1998 simulation, modelers had quarterly consumption data available for October 1997 through April 1998. These same data were used for the August 1998 simulation because metered data representing consumption conditions for August 14-16, 1998, or for the peak-demand season were not available to investigators for the August 1998 simulation. To further reduce P,  metered consumption data (and consequently, nodal consumption data) during the time of the test, and specifically during the peak-demand test, would be required.

Figure 13. Frequency of absolute pressure difference between March and August 1998 test data and simulations.

Figure 13

Overall, P  for the March and August 1998 simulations are within the calibration criteria of ±5 psi to ±7.5 psi for pressure difference selected as a calibration target by investigators, and within the criteria suggested by Cesario and Davis (1984) and Walski (1983) previously discussed (see "Overview" section above). The March and August 1998 model simulations are accurate characterizations of the water-distribution system for winter-demand and peak-demand conditions. In addition to the analysis of pressure difference (Figures 11-13; Plates 10 and 11), comparison of measured and simulated pressures for the 25 hydrants presented in this report (Appendices I and J), the comparison of measured and simulated hydraulic head at ground-level and elevated storage tanks (Appendices K and L), comparison of measured and simulated booster pump flows (Appendices M and N), and comparison of measured and simulated groundwater well flows (Appendices O and P) support the assertion that the model presented herein is an acceptable and reliable representation of water-distribution system conditions existing during 1998.

Sensitivity Analysis

Model analyses give rise to parameter uncertainty because the models use parameters for which data values may be incomplete, missing, unknown, or contain errors. As previously discussed, up to 10 sources of error may be introduced when calibrating a water distribution system model (AWWA Engineering Computer Applications Committee 1999). Therefore, it is important to understand and acknowledge which parameters may have a substantial effect on model output results if the parameter value is varied. Four methods exist for quantifying parameter uncertainty (EPA 1997): (1) sensitivity analysis, (2) analytical uncertainty propagation, (3) probabilistic uncertainty analysis (Monte Carlo simulation), and (4) classical statistical methods. The sensitivity analysis method will be used for the current study.

For the sensitivity analysis method, a model parameter value is varied from the calibrated value, and the resulting model output is then compared with the calibrated values of model output to assess the sensitivity of the model to that particular parameter. As shown in Table 10, the parameters estimated to have high or moderate sources for error, and thus presumably the parameters with the greatest uncertainty, are pipe roughness values (Hazen-Williams "C-factor"), effects of system demands (variation of consumption), and outdated or unknown pump-characteristic (head-discharge) curves.

Of the three parameters that are presumed to have the greatest source for error, two parameters, the effects of system demand and outdated or unknown pump-characteristic curve data, are dependent on one another. That is, if a modification is made to a pump-characteristic curve so that more or less water is supplied for a given head, a change must also be made to system demands (consumption) to increase or decrease demand, respectively. This is because in water-distribution system models such as EPANET, supply must equal demand if supply is increased, demand must be increased. Thus, it is not possible to change one of these parameters without changing the other parameter. Because of this, it is not possible to determine the sensitivity of the model to a variation in just one of these parameters.

The third of the model parameters with the greatest uncertainty for this investigation is pipe roughness (Hazen-Williams "C-factor") values. The values chosen for this parameter will cause higher or lower friction losses for water flowing through network pipes. Using the Hazen-Williams roughness coefficient to express the head loss due to friction associated with flow through a pipe, the following head loss formula is used in EPANET:

where:

Thus, as the value assigned to C decreases, the head loss due to friction increases. As shown in Equation (10), pipe diameter could have an effect on head loss similar to varying C. This uncertainty is introduced into the model by using nominal pipe diameters rather than the manufacturer's specified internal pipe diameter.

Table 12. Variation of parameter values used to conduct sensitivity analyses using the calibrated water-distribution system model, Dover Township area, New Jersey

For the sensitivity analysis, values assigned for roughness coefficient, C, and pipe diameter, d, were varied from the calibrated values. Internal pipe diameter data were supplied by the water utility. Values for roughness coefficient were obtained from the range of C values generally associated with a particular type of pipe material (Rossman 1994, Walski 1984). Table 12 lists the parameter value variations for C and d and the simulations conducted as part of the sensitivity analysis. For pipe diameter variation, one sensitivity simulation was conducted for all varied diameters (SENS1 in Table 12). For roughness coefficient variation, a sensitivity simulation was conducted for each pipe material type (SENS2, SENS3, and SENS4, in Table 12). A simulation was also conducted that included all the variations in roughness coefficient in one simulation (SENS5 in Table 12). The sensitivity simulations identified in Table 12 were conducted for both the March and August 1998 test conditions.

To compare sensitivity simulation results with calibrated model results (March and August 1998), two statistics were computed for each simulation. The statistics, the root-mean square (RMS) and the relative simulation error (RSE), are defined as follows:

where:

RMS	= root-mean-square of pressure difference (psi)
RSE	= relative simulation error (percent),
NH	= number of test hydrants (25),
NTS	= number of hourly time steps in the simulation (48)
	= measured pressure data for hydrant i, at hour j, (psi),
	= simulated pressure for hydrant i, at hour j, (psi), and
	= the absolute value of pressure difference (psi).

These two statistics were calculated for the simulations using the calibrated parameter values (March and August 1998) and for each sensitivity simulation where parameter values were varied. Results are shown in Table 13. For both the March and August 1998 conditions, results in Table 13 clearly indicate that the model of the distribution system is insensitive to variations in diameter and roughness coefficient. The calculated statistics for the two demand conditions (March and August 1998) vary in tenths of psi for the RMS and tenths of a percent for the RSE. Thus, obtaining additional information on roughness coefficient (through additional field testing) or using actual pipe diameters (instead of nominal diameters) would not significantly change or reduce parameter uncertainty for our situation.

Table 13. Comparison of simulation statistics for the water-distribution system model using calibrated and varied parameters, Dover Township area, New Jersey

¹Refer to Equation (11) in text for definition of RMS
²Refer to Equation (12) in text for definition of RSE
³Refer to Table 12 for definition of sensitivity simulations SENS1 - SENS5

Water-Quality Simulation

The panel of expert peers who reviewed the ATSDR modeling effort in December 1998 (ATSDR 1999), recommended that ATSDR obtain a set of water-quality calibration data to assess the reliability of the model calibration for fate and transport simulations. There are two methods that could be used to accomplish this:

1. Introducing a tracer into the water-distribution system and collecting data simultaneously for pressure, flow, and concentration of the tracer at different sampling locations; or



2. Collecting water samples at different sampling locations and analyzing the samples for a naturally occurring tracer already present in the water throughout the distribution system.

We chose the second method because: (1) it was less intrusive, (2) the data were readily available, and (3) at the time the pressure data were collected (March and August 1998), it was not possible to collect flow and water-quality data because of institutional, operational, and budgetary constraints. ATSDR investigators obtained water-quality data from sampling events that occurred on March 28, April 4, and April 24, 1996 (NJDHSS 1999c). These data are used to compare with results of a water-quality simulation described below.

On March 28, 1996, NJDHSS collected water samples from taps at 21 schools located throughout Dover Township and serviced by the water utility (Plate 12, Table 14). The samples were analyzed for, among other constituents, the naturally occurring element, barium. Naturally occurring barium is a conservative constituent in groundwater and as such, the concentration will not vary significantly over time. Therefore, a one-time collection of water samples is sufficient for our purposes of testing the reliability of the model calibration for use in fate and transport simulations. On April 4, 1996, NJDHSS also collected water samples from 5 points of entry to the distribution system (the assumed sources for the barium). These points of entry were (Plate 12; Table 14): well 32 (South Toms River), well 20 (Indian Head), well 31 (Route 70), and the Parkway ground-level storage tank. An additional point of entry, the Holly Plant ground-level storage tank, was sampled on April 24, 1996. This tank was sampled at a later date than were the other points of entry because all wells supplying the tank (wells 21, 30, and 37) were off-line on the day the other points of entry were sampled (April 4, 1996), but well 30 was on-line when the tap samples were obtained at the 21 schools (March 28, 1996). These points of entry were the only sources of water to the distribution system during this time period. The measured concentrations of the barium samples at the 21 school locations and the 6 points of entry are shown on Plate 12 and listed in Table 14.

To simulate the spatial distribution of barium, information on the operating schedule for pumps and wells (cycling on/of schedules) and on tank levels for March 28, 1996, was obtained from the water utility. In addition, data for the daily production for March 28 were also obtained. The system-demand factors based on the March 1998 field-test data were modified so that the average daily production for March 28, 1996, (which was greater than the calibration period of March 24-25, 1998) could be distributed on an hourly basis. The concentrations for barium measured at the points of entry were used as the source concentration and assigned to the model node associated with a specific point of entry (Table 14). For the EPANET simulation, the hydraulic time step was set at 1 hour and the water-quality time step was set at 5 minutes. Initial conditions must be "flushed out" of the distribution system before retrieving fate and transport simulation results. Information from the calibrated model of the distribution system indicates that approximately 1,000 hours (42 days) are needed for the entire system to reach a dynamic equilibrium condition in terms of the mixing of a chemical constituent (such as barium) in the storage tanks (e.g., Holiday City ground-level storage tank shown in Appendix Q). Graphs showing the concentration of barium reaching a dynamic equilibrium condition in the storage tanks operating during this time are provided in Appendix Q. Once dynamic equilibrium was reached, EPANET was run for an additional 24 hours. The concentration at model nodes corresponding to the 21 school locations is reported for 08:00 hours, the approximate time that sample collection was completed. Comparison of measured and simulated barium concentrations is presented in Table 14 and shown spatially on Plate 12.

Table 14. Identification of barium sampling locations and comparison of measured and simulated barium concentration values, Dover Township area, New Jersey

¹See Plate 12 for sampling and point of entry locations
²Data from NJDHSS (1999c)
³Simulation results for 08:00 hours on March 28, 1996
⁴Relative difference in percent, computed using Equation (14). Mean relative difference for the 21 school samples is 13.6%
⁵Model bias = simulated value / measured value
⁶Date of sampling is March 28, 1996
⁷Date of sampling is April 4, 1996 except for Holly Plant storage tank whose sampling date is April 24, 1996

To assess the accuracy of the water-quality simulation with respect to the transport of barium, 5 statistics are computed: (1) the concentration difference (column 5 in Table 14); (2) the percent absolute difference (column 6 in Table 14); (3) the model bias (column 7 in Table 14); (4) the geometric bias; and (5) the correlation coefficient. The mathematical definitions for these statistics are provided below:

where:

where:

C	= the relative difference, (percent), and
| . |	= the absolute value of a function;

where:

where:

where:

and

	,and	(18)
		(19)

where:

The values in column 5 of Table 14 were computed using Equation (13) and indicate a difference between measured and simulated barium concentrations ranging from -6.6 micrograms per liter (µ/L) to 12.4 µ/L and an absolute difference ranging from 0.2 µ/L to 12.4 µ/L. Using Equation (14) to compute the relative difference (column 6 in Table 14) yields a range of 0.6% to 25.6 % with a mean relative difference of 13.6%. This analysis indicates that model-simulated values are acceptable when compared with measured values. However, a more rigorous analysis of evaluating the accuracy of the model with respect to the water-quality simulation can be conducted using the concept of model bias and the correlation coefficient as described below.

Model bias allows us to test the accuracy of the model with respect to the transport of barium by expressing the bias in terms of a simulated-to-measured ratio (Rogers et al. 1999). This ratio (C_s/C_m), computed using Equation (15) and listed in Table 14 (column 6), has the following properties:

The data we have available, which are spatially disparate, are best suited for an analysis using the geometric bias. The geometric bias is the geometric mean of the individual C_s/C_m ratios, and is computed using Equation (16). The geometric mean is used because the distribution of C_s/C_m ratios is skewed like a lognormal distribution. That is, the values are restricted for under prediction (0-1), but are unrestricted for over prediction (anything greater than 1). For the data presented in Table 14, the geometric bias is 0.93, which indicates that the model slightly under predicts the measured values. To assess the correlation of simulated values with the measured data, the correlation coefficient was computed using Equations (17-19). For the data in Table 14, the correlation coefficient is 0.81, indicating a high degree of correlation. These statistical analyses (model bias, geometric bias, and correlation coefficient) provide further evidence that the model of the water-distribution system presented herein (both hydraulic and water quality components): (1) is reasonably calibrated, and (2) provides an acceptable representation of the 1998 water-distribution system characteristics for our intended use.

Use of the Model for Epidemiologic Investigations

In the next phase of this investigation, ATSDR will provide to NJDHSS epidemiologists information on the percentage of water that residences of study subjects may have received from each of the points of entry (wells or well fields) to the water-distribution system-the concept of "proportionate contribution" previously discussed in the "Method of Analysis" section. The application of this methodology can be demonstrated using the trace analysis option available within EPANET in conjunction with the characterization of the present-day (1998) water-distribution system. A trace analysis was conducted for each water source point of entry to the water-distribution system operating during the time of interest (Table 15). Once dynamic equilibrium was reached (as described in the section on "Water-Quality Simulation"), a 24-hour trace analysis simulation was conducted by assigning a trace node to coincide with a known and operating point of entry. Locations of the points of entry for the present-day distribution system are listed in Table 15. Results of the trace analysis were obtained in terms of the percentage of water that any location of interest receives from the trace node or location (water source) as an average over a 24-hour time period. These results are presented as a series of maps displaying the areal distribution of simulated proportionate contribution of water from points of entry to model nodes in the distribution-system network for March 1998 winter-demand conditions (Plates 13-16) and August 1998 peak-demand conditions (Plates 17-22). In addition, results are presented in terms of the proportionate contribution of water to 5 selected nodes (Figure 14; Table 15) coinciding with: (1) test hydrant H-1 representing, the southwestern portion of the study area; (2) test hydrant H-11, representing the south-central portion of the study area; (3) test hydrant H-17, representing the central portion of the study area; (4) test hydrant H-20 representing the northwestern portion of the study area; and (5) test hydrant AH-22, representing the northeastern portion of the study area.

Table 15. Simulated proportionate contribution of water for 1998 conditions from points of entry to selected locations, Dover Township area, New Jersey

¹Based on calibrated model of the water-distribution system reaching dynamic equilibrium.
²Refer to Plate 6 for locations.
³Well(s) supplying distribution system or storage tanks not in operation during this time period.

Simulated results obtained from the trace analysis for each point of entry to the water-distribution system, operating under winder-demand (March 1998) and peak-demand (August 1998) conditions, are now discussed in terms of generalized areal distributions of the proportionate contribution of water to locations throughout the Dover Township area. Plate 13 shows the areal distribution of the simulated proportionate contribution of water from the Berkeley wells (33, 34, and 35) for March 1998 conditions. Under these conditions, the wells supply 90% to100% of the demanded water to the southwestern portion of the study area, 25% to 50% of the water to the southeastern portion, and from 1% to 10% to the central and northeastern portions of the study area. As a comparison, the simulated proportionate contribution of water from the Berkeley wells under peak-demand conditions for August 1998 (Plate 17) indicates a supply of water ranging from 50% to 100% for the southwestern portion of the study area and a 1% to 10% contribution of water to the south-central area. The simulations for the Berkeley wells indicate that they supply 1% to 10% of the water for the borough of South Toms River in August but do not supply any water in March (compare Plates 13 and 17), and they do not supply any water to the central, southeastern, north-central, and northern portions of the study area under the August 1998 conditions (Plate 17).

The reason the Berkeley wells do not supply water to the borough of South Toms River under March 1998 conditions is evident by comparing Plates 13 and 14. Plate 14 presents the areal distribution of the simulated proportionate contribution of water from the South Toms River well 32 under March 1998 conditions. This simulation indicates that well 32 supplies from 90% to 100% of the water demanded in the borough of South Toms River during March; however, for the August 1998 conditions, well 32 was not operating. Therefore, part of the water demand for the borough of South Toms River was supplied by the Berkeley wells (Plate 17).

As discussed above, under the simulated August 1998 conditions, the Berkeley wells do not supply any water to the southeastern, eastern, and northeastern portions of the Dover Township area (Plate 17) when compared with simulated March 1998 conditions (Plate 13). The reason for this is clear by reviewing Plate 18 (Brookside well 43) and Plate 22 (Windsor ground-level storage tank). Under simulated August 1998 conditions, these two points of entry are operating, and therefore, they supply the water demand in the southeastern, eastern, and northeastern portions of the study area. Specifically, the Windsor ground-level storage tank (Plate 22) supplies 90% to 100% of the water demand in the southeastern portion of the area, from 1% to 75% of the water in the eastern portion of the area, and 1% to 25% of the water in the northeastern portion of the study area. Brookside well 43, under August 1998 simulated conditions (Plate 18), supplies from 1% to 50% of the water in the northeastern portion of the study area and from 1% to 100% of the demand for water in an elliptical north-to-south oriented area (Plate 18).

The areal distribution of the simulated proportionate contribution of water from the Parkway well field tank is shown on Plate 16 for winter-time conditions (March 1998) and on Plate 20 for peak-demand conditions. Comparison of these two maps shows that the amount of water contributed by the Parkway ground-level storage tank in August 1998 to the southeastern and eastern parts of the distribution system (Plate 20) is considerably reduced or eliminated when compared with the March 1998 conditions (Plate 16). This is a result of well 40 and the Windsor ground-level tank being in operation for the peak-demand conditions existing in August 1998 (Plate 22). As a consequence of this operation, the Windsor tank meets the demand for water in the southeastern and eastern parts of Dover Township serviced by the distribution system. The trace analysis also shows the contribution of water from the Parkway well field tank to the northern area of Dover Township increases from 50% to 75% in March (Plate 16) to 75% to 100% in August (Plate 20). In addition, in the northern area of Dover Township between the Holiday City ground-level tank and the North Dover elevated tank, the Parkway ground-level tank does not supply any water in March 1998 (Plate 16), but supplies between 10% and 50% of the water demand in August 1998 (Plate 20).

The South Toms River well 32 supplies 90% to 100% of the water demand for March 1998 conditions for the borough of South Toms River (Plate 14). For August 1998 conditions, however, well 32 was not operating. Instead, water was obtained from the Holly Plant ground-level tank, supplied by wells 21, 30, and 37 (Plate 21). Results of the August 1998 analysis show that the Holly Plant tank supplied 10% to 100% of the water demand in a fan-shaped region in the south-central portion of the study area, and from 1% to 75% of the water in a narrow band along the southwestern portion of the area. In addition, Holly Plant tank supplied from 1% to 10% of the water demand for the remainder of the Dover Township area serviced by the distribution system with the exception of the southeastern and extreme northwestern portions of the study area.

The final point of entry for which a trace analysis was conducted is the Route 70 well 31 and results are shown for March 1998 (Plate 15) and August 1998 (Plate 19) conditions. Under March 1998 conditions, the well supplies from 90% to 100% of the water demand for the northwestern portion of the study area and from 10% to 50% of the water demand for the north-central and northeastern portions of the study area (Plate 15). This contrasts with August 1998 conditions (Plate 19) where the Route 70 well 31 supplies 75% to 100% of the water demand in the northwestern portion of the study area, 75% to 90% in the north-central portion, and 1% to 10% percent in the northeastern portion of the study area. The difference in supply from the March to August 1998 conditions (compare Plates 15 and 19) is made up by supply from the Parkway ground-level storage tank for the north-central portion of the study area (Plate 20) and the Windsor ground-level storage tank for the northeastern portion of the study area (Plate 22).

Selected Locations, Dover Township Area, New Jersey

Five test hydrant locations are chosen as examples of the proportionate contribution of water to selected points of interest in the Dover township area and these results are now presented. The test hydrants represent the southwest (test hydrant H-1), south-central (test hydrant H-11), central (test hydrant H-17), northwest (test hydrant H-20), and northeast (test hydrant AH-22) portions of the study area. (See Plate 6 for hydrant locations; see Figure 14 and Table 15 for the ensuing discussion.) At the location represented by test hydrant AH-22 (northeast area of Dover Township), under winter-demand conditions (March 1998), the Berkeley wells (33, 34, and 35) supply 7% of the water, the Route 70 well 31 supplies 26% of the water, and the Parkway well field ground-level storage tank supplies 66% of the water. Under peak-demand conditions (August 1998), the Berkeley wells (33, 34, and 35) supply less than 1% of the water, Route 70 well 31 and the Holly plant ground-level storage tank supply about 1% of the water, Brookside well 43 supplies about 8% of the water, the Parkway well field ground-level storage tank supplies 83% of the water, and the Windsor ground-level storage tank supplies 6% of the water to the area of test hydrant AH-22. Thus, in terms of an exposure assessment, persons residing in an area represented by test hydrant AH-22 would receive, over an average day, more than 80% of their potable water from the Parkway well field tank during summer time (peak-demand) conditions, and about 66% of their potable water from the Parkway well field tank during winter-demand conditions. Similar observations can be made from results presented in Figure 14 and Table 15 for persons residing in areas represented by the other test hydrant locations.

Figure 14. Comparison of simulated percentage of water from points of entry to selected test hydrant locations, Dover Township area, New Jersey: (A) March 1998 conditions, and (B) August 1998 conditions. (Refer to Plate 6 for a map of locations.)

Figure 14

Generalized Approach for Historical Reconstruction

To help NJDHSS with the case-control investigation, water-distribution system networks representing the location of pipelines from 1962 through 1996, based on historical information, will be derived. Using historical information on: (1) system demand conditions, (2) locations of wells, storage tanks, and booster pumps, and (3) system operations, trace analyses will be conducted for each month from January 1962 through December 1996. This will enable ATSDR investigators to provide information to epidemiologists and health scientists that can be used to relate study subject addresses to areas historically served by the water-distribution system using spatial analysis and address-matching techniques. ATSDR investigators will be blinded to case and control status of study participants. In the next phase of this investigation, historical model trace-simulation results, based on residence histories, will be used to estimate exposure to specific water sources by determining the percentage of water both cases and controls may have received from each of the points of entry (i.e., well fields) to the water-distribution system.

Next Section     Table of Contents