Vanessa Speight, Jim Uber, Walter Grayman, Kathy Martel, Melinda Friedman, Phil Singer, and Fran DiGiano

OBJECTIVES:

The objective of this project was to present a rigorous and quantitative method, using hydraulic and water quality models in a Monte Carlo simulation framework, to characterize the underlying variability and uncertainty in water quality parameters resulting in better sampling program design for those parameters. The sampling objectives to be evaluated included regulatory, public health, and operational process control elements.

BACKGROUND:

Water quality sampling of drinking water distribution systems is generally conducted for (1) compliance purposes, (2) to characterize ambient conditions (e.g., quarterly disinfection by-product sampling at four locations), or (3) to detect poor water quality conditions that could potentially pose a threat to public health (e.g., detection of total coliforms or intentional breaches of security). Distribution system water quality sampling plans are typically designed to fulfill a specific objective, using minimal guidance to select locations and methods for collection and analysis of water quality data.

The design of any water quality monitoring program will depend on the specific questions that need to be answered and on the water quality goals that have been set by the utility during the planning process. In considering a sampling program, the utility should address the following questions:

1. What is the purpose of the sampling program?

2. What water quality and system parameters need to be measured to meet the program objectives?

3. How much spatial and temporal variation is there in these parameters? Is this variability likely to be significant?

4. What accuracy and precision are required in the sampling process?

5. Where, and perhaps when, should data be collected?

6. How often should samples be collected?

7. Are manual grab samples or online monitoring most appropriate, considering the expected level of temporal variation? Should samples be composited in order to measure time averaged water quality conditions?

8. What monetary and personnel resources are available?

9. How effective will the sampling program be in meeting our goals? Can sampling goals be quantified using well defined metrics?

Because water quality sampling is conducted on a very small percentage of the total water delivered to consumers, sample locations and frequency should be carefully selected to serve multiple monitoring objectives such as regulatory, security-related, and operational process control objectives. However, water quality sampling locations and schedules are frequently selected on an ad hoc basis and without consideration of information about distribution system dynamics and connectivity. In the absence of quantitative metrics and methods, informal sampling guidelines focusing on ease of access, spatial distribution, and population weighting of monitoring locations have taken precedence.

APPROACH:

The research approach developed a probabilistic modeling framework using several steps. First, a thorough literature review was conducted to identify water quality reactions and parameters of interest. Next, expert interviews were conducted to select the appropriate model relationships and to provide background on parameter distributions and values. The selected mechanisms were incorporated into a distribution system modeling framework using EPANET-MSX and Monte Carlo simulation tools. The ensemble of model results was then stored and analyzed to evaluate the effectiveness of different sampling programs.

Monte Carlo simulation is a general and conceptually simple technique that characterizes uncertainty through a large population of scenarios. Each scenario is a simulation of one plausible real event. In this case, a scenario is comprised of the water quality characteristics in a distribution system for a particular set of boundary and operating conditions. Each separate scenario is defined by randomly drawing parameter values from probability distributions that describe the variability and uncertainty associated with parameters. As a result, an ensemble (collection) of scenarios will approximate the underlying probability distributions of the parameters defining the scenario and the outcomes of the scenario simulations will define the statistical distributions of the results. For example, the distribution of total coliforms in a distribution system could be predicted for one set of hydraulic conditions, assuming that the coliform source was a particular pipe break or a soil water intrusion caused by a power outage. By considering many such plausible events (an ensemble), the statistical distribution of total coliforms in a distribution system can be estimated and sampling plans assessed.

RESULTS/CONCLUSIONS:

The extensive literature search indicated that there is a rich database of general information on the entry, movement, and reaction of water quality constituents in the distribution system. However, the same search showed that there are significant limitations in terms of quantitative information on these processes and severe limitations on the quantitative uncertainty associated with these parameters.

The modeling framework was demonstrated for three different water quality parameters of concern: disinfectant residual, disinfection by-products, and microbial contaminants. As proof of concept, the Monte Carlo modeling framework was applied to two small distribution system models. It is based on the widely used EPANET network modeling software and extensions to support Monte Carlo simulation and a wide range of user defined water quality transformations and reactions in the distribution system. Working with subject matter experts, the process of determining the appropriate models for water quality parameters revealed that simplified models of reactions may be sufficient to use for model simulation because of the lack of detailed system data to support the use of more sophisticated mechanistic models. The application of the modeling methodology provided a database of plausible scenarios from which to evaluate the ability of different sampling programs to meet a given set of metrics. To capture events of short duration and small magnitude, grab sampling may be insufficient and continuous monitors provide a better opportunity to detect events. None of the sampling programs that were evaluated was able to detect all events.

APPLICATIONS/RECOMMENDATIONS:

This project provides an important demonstration of the application of probabilistic modeling to water quality reactions in drinking water distribution systems. The project approach successfully extended techniques used in water system security and risk assessment research to a broad consideration of multiple water quality parameters in distribution systems and is already serving as the fundamental basis for a methodology being used by USEPA for risk modeling and research needs identification. For utilities, this approach provides a way to incorporate case-specific data for water quality, account for system hydraulics, and develop a customized sampling program.

While the small example problem applied in this project clearly illustrates a significant variation in the effectiveness of alternative monitoring options, it would be imprudent to try to generalize the results to cover all situations. However, some conclusions can be drawn from the investigation of the literature and the example simulation performed as part of this study:

* Though there has been extensive research in the area of microbial contamination of distribution systems, there remains a great deal that must be learned in order to effectively design and assess alternative sampling programs. Additional research is needed within these areas in order to expand the knowledge base so that more intelligent decisions can be made on monitoring microbial activity.

* The primary research need is in the quantitative definition of contamination events in terms of event duration, magnitude, and frequency.

* A clear metric for judging the effectiveness of alternative sampling programs must be defined.

* It appears that reliance upon infrequent, small volume samples will not result in the detection of many microbial events. As the duration of contaminant events become smaller, the likelihood of their detection through grab samples become probabilistically less likely. Mechanisms for taking larger volume samples in conjunction with more frequent (or continuous) sampling appears to be the best possibility for detecting contamination events.

* The research team suggests that the probabilistic based Monte Carlo simulation approach developed in this report is a promising and appropriate method for designing sampling programs for microbial contamination in distribution systems. The water industry would benefit from further research and application in this specific area.

RESEARCH PARTNER:

U.S. Environmental Protection Agency

PARTICIPANTS:

Seven water utilities from the United States and Australia participated in this project.