x>EPA

United States
Enviranmental
Protection Agency

epa.gov/researcfi

nnovative Science for a Sustainable Future

MODELING STRATEGIES TO IDENTIFY WATER DISTRIBUTION SYSTEM SAMPLING LOCATIONS

Background

The delivery of safe, potable water to communities is the
primary objective of drinking water utilities. However,
the quality of the water can deteriorate as it is
transported from the treatment plant through the
distribution system to the customers due to interactions
with the pipe walls and constituents in the water itself.
Additionally, drinking water distribution systems can be
vulnerable to intentional or accidental contamination
incidents. Since these systems consist of thousands of
pipes and service connections over large geographic
regions, utilities cannot financially afford to install
continuous online monitoring sensors everywhere in the
system. Thus, drinking water utilities collect samples
throughout the distribution system to evaluate the quality
of the water to protect the health of the community.
Sampling can be conducted to meet regulatory needs, to
aid in the operations and maintenance of the system, to
respond to customer complaints, and to investigate
possible contamination incidents.

If a water contamination incident is suspected following
an alert from a water quality monitoring sensor or
customer complaint, samples could be taken to support
response actions. Sampling could be used to confirm that
a contamination incident has occurred or is occurring in
the distribution system and to identify the type or
concentration of a contaminant. The contamination
injection location (or source), the extent of the
contamination plume, and the required decontamination
area could also be identified by sampling. Identifying the
contamination source is important to stop more
contamination from entering the system. Knowledge of
the source would also help define the extent of
contamination. By determining the extent of
contamination, the percentage of the population exposed
to contamination, or unaffected areas of the system can
be identified. Accurate determination of the source and
extent of contamination is challenging because there can
be limited available measurements and significant
uncertainty in system hydraulics, contaminant reaction

dynamics, and incident details. Decision makers might
have a difficult time implementing an effective response
action if there is a large uncertainty associated with the
contamination incident. Thus, it is important to develop
techniques that can characterize a contamination incident
quickly to mitigate the effects.

Modeling Uncertainties

Decision makers can use water distribution modeling
tools to plan and inform sampling strategies. In order to
achieve confidence in the modeling results, model
uncertainty must be addressed. Water distribution system
models have various sources of uncertainty in the
hydraulic modeling parameters, including

•	infrastructure representation (e.g., incorrect pipe
diameters, missing pipes)

•	customer demands (e.g., changes due to public health
notices - do not use, boil water)

•	operational controls (e.g., valve settings, pump
curves)

•	initial conditions (e.g., tank levels, pump statuses)

Additionly, water quality modeling parameters also have
uncertainty associated with them, including

•	contaminant type (e.g., biological, chemical)

•	contaminant reaction dynamics

•	amount of contaminant injected

•	contaminant injection location

•	time of the contaminant injection

•	duration of the contaminant injection

Identifying the parameters that affect contaminant
transport within the distribution system can be useful to
select sampling locations to decrease the uncertainty in
the extent of contamination.

Hart et al. (2019) investigated uncertainty in the
contamination plume when modifying the following
water quality modeling parameters:

•	customer demands

•	isolation valve status

Connect with us Online



You
trim

EPA/600/S-23/268


-------
v>EPA

•	bulk reaction rate coefficient

•	contaminant injection start time

•	contaminant injection duration

•	contaminant injection location

•	contaminant injection rate

The uncertainty in the injection location parameter had
the most effect on the extent of contamination compared
to the other parameters. More than half of the locations
contaminated only small areas of the system and only a
few locations contaminated a large area of the system.
Figure 1 shows an example of how the extent of
contamination is affected by the uncertainty in the
injection location. The contaminant injection rate,
reaction coefficient, and injection duration were the next
most significant parameters after the injection location,
since they also affected the total area contaminated in the
system. Increasing the injection rate provided more
contaminant that could spread further into the system
before becoming too diluted. High reaction rate
coefficients decreased the concentration below the
contamination threshold more quickly. Increasing the
injection duration ensured that the contaminant remained
in the system longer and therefore increased the area
affected.

Figure 1. Map of the water distribution system in which the nodes are
sized by the average extent of contamination of an injection
occurring at the node and colored by the uncertainty. Nodes outlined
with red diamonds are used as example injection location [from Hart
etal. (2019)].

Sampling for Extent of Contamination

Affected drinking water utilities can take samples within
the distribution system to determine the source and
extent of a contamination. Rodriguez et al. (2021)
presented an optimization framework to reduce the
uncertainty about the contamination incident as quickly
as possible by selecting the best sampling locations to

determine the source and extent of the contamination.

Step 1 built a database of simulation results from
potential contamination scenarios with different
characteristics (e.g., injection location, amount, duration,
customer demands, and reaction coefficients). Step 2
updated the probability of the contamination scenarios
based on available measurements following the possible
alert of a contamination incident. Step 3 calculated the
probability that a node was contaminated using the
precomputed simulation results and the contamination
scenario probabilities from Steps 1 and 2, respectively.
Step 4 categorized nodes based upon their probability of
contamination given a specific confidence level. Step 5
assessed the number of nodes that were categorized as
uncertain in terms of contamination, and if the number
was close to zero, the process stopped. Otherwise, Step 6
used an optimization-based approach to determine the
best locations to take additional samples. Step 7 obtained
new sample measurements and returned the process back
to Step 2. A node's probability of contamination was
adjusted as more sample measurements were obtained
and the uncertainty in the contamination source and
extent were reduced. The optimization formulations
presented solved for multiple optimal sampling locations
simultaneously and efficiently, even for large systems
with a large uncertainty space. The efficiency and
effectiveness of the framework was demonstrated in two
case studies.

Figure 2 shows node probability maps after each
sampling cycle of four samples each for one of the case
studies. Approximately 30% of the nodes were highly
unlikely to be contaminated when the initial alarm was
triggered, while the remaining 70% remained uncertain as
to whether they were contaminated. By cycle 2 with a
total of eight measurements, the level of uncertainty in
the extent of contamination had been greatly reduced, and
by cycle 3 with a total of 12 measurements, the extent of
contamination was almost completely characterized. To
identify sampling locations quickly and optimally, a
broad set of contamination scenarios, including hydraulic
patterns to reflect significant decreases in the overall
demand due to public health notices (e.g., do not drink,
do not use), should be precomputed. However, new
contamination scenarios can be included during the
sampling process to account for scenarios generated
based on real-time data or other system knowledge during
an actual incident.

ConnectwithusOnline

EPA/600/S-23/268

You
IfliTil




-------
*>EPA

5

(a)



4

g

\i

(b)







mm ly







UN



Cc)

(d)

mm ln

Figure 2. Nodal probability maps in which LY (red) is likely
contaminated, UN (yellow) is uncertain if contaminated, andLN
(blue) is likely not contaminated for different sampling cycles: (a)
alatwi is triggered (cycle 0 - no measurements); (b) after four
measurements (cycle 1); (c) after eight measurements (cycle 2); and
(d) after twelve measurements (cvcle 3) [modifiedfrom Rodriguez et
a I. (2021)].

Regulatory Sampling for Emergencies

While sampling locations can be determined following
an alert of a possible contamination, drinking water
utilities already have established routine sampling
locations within the distribution system for operational
and regulatory purposes. Since utilities are already used
to taking samples at these locations, they might begin an
investigation of a contamination incident there first.
Haxton et al. (2021) evaluated the effectiveness of
routine, regulatory sampling locations for emergency
response scenarios. They also investigated the
performance of emergency response sampling locations
for regulatory purposes. For the systems assessed in the
paper, the sampling locations identified for one purpose
(regulatory or emergency response) detected less
scenarios when evaluated against the opposite purpose.
The average performance was reduced by 3%-4% when
emergency response locations were used for regulatory
goals, while the performance was reduced by 7%—10%
when regulatory response locations were used for
emergency response. Figure 3 illustrates the sampling
locations and the contribution each sampling location
had on scenario detection for the emergency response
scenarios. This work highlighted that regulatory
sampling locations could provide value in responding to
an emergency for the system evaluated.

0.1 0.2 0.3
Fraction detected

Figure 3. Fraction of detected scenarios using 120 sampling
locations for (a) emergency response optimization and (b) regulatory
optimization, evaluated for emergency response conditions. Each
figure includes the total fraction detected and a histogram showing
the frequency of sampling locations for each fraction detected
[modified from Haxton el al. (2021)].

Drinking water distribution modeling tools can support
water utilities in making decisions with regards to
sampling locations during emergency response situations.
During an emergency, prioritizing the use of limited
personnel and resources is critical to (1) determining the
nature and extent of contamination within a system and
(2) determining the effective mitigation strategies to
address the emergency . While having an up-to-date
emergency response plan with available sampling
locations identified is desired, these sampling locations

Connect with us Online



You
trim

EPA/600/S-23/268


-------
v>EPA

might not be predefined for all systems. However,
systems will likely have designated locations already
established for complying with regulatory requirements.
The approaches discussed here can be applied to help
utilities understand how effective manual samples would
be for their system and to improve the sampling location
selection process. If a system does not have an
emergency response plan, then regulatory sampling
locations could provide a good basis for an initial round
of sampling during a suspected emergency, allowing
utility personnel to be quickly dispatched to get
preliminary data related to contamination extent.

CONTACT

Dr. Terra Haxton

Office of Research & Development
513-569-7810, haxton.terra@epa.gov

Dr. Jonathan Burkhardt

Office of Research & Development

513-569-7466, burkhardt.jonathan@epa.gov

www.epa.gov/ord

REFERENCES

D. Hart, J. S. Rodriguez, J. Burkhardt, B. Borchers, C. Laird, R.
Murray, K. Klise, T. Haxton (2019) "Quantifying hydraulic and
water quality uncertainty to inform sampling of drinking water
distribution systems" Journal of Water Resources Planning and
Management 145(1).

J. S. Rodriguez, M. Bynum, C. Laird, D. B. Hart, K. A. Klise, J.
Burkhardt, T. Haxton (2021) "Optimal sampling locations to reduce
uncertainty in contamination extent in water distribution systems"
Journal of Infrastructure Systems 27(3).

T. Haxton, K. A. Klise, D. Laky, R. Murray, C. D. Laird, J. B.
Burkhardt (2021) "Evaluating manual sampling locations for
regulatory and emergency response" Journal of Water Resources
Planning and Management 147(12).

DISCLAIMER

This document has been reviewed in accordance with U.S.
Environmental Protection Agency, Office of Research and
Development, and approved for publication.

D

Connect with us Online
¥ ••

You
ifirn

EPA/600/S-23/268


-------