The Potential for Using the CANARY Event Detection Software to Enhance Security and
Improve Water Quality
Jennifer Hagar1, Regan Murray2, Terra Haxton2, John Hall2, and Sean McKenna3
i OR1SE Fellow National Homeland Security Research Center, U.S. EPA, NG-16, 26 W. Martin Luther King Dr.,
Cincinnati, OH 45268
2Research Scientist, National Homeland Security Research Center, U.S. EPA, NG-16, 26 W. Martin Luther King
Dr., Cincinnati, OH45268
3 Research Manager, Smart Cities Technology Centre, IBM Research, Dublin, Ireland
Abstract
Public health protection and the provision of potable water are main priorities for all drinking
water utilities. Over the last ten years, the threat of contamination of the nation's drinking water
infrastructure has heightened utility awareness of distribution system security. Drinking water
systems face enormous challenges in meeting these goals, and in particular, small water systems,
which comprise 94% of the nation's public water systems, face additional economic and
operational hurdles. The CANARY event detection software was developed to enhance the
detection of contamination in water distribution systems. Working in conjunction with a network
of water quality sensors placed strategically throughout the distribution system, CANARY
increases the likelihood and speed of detection by interpreting sensor data in near real-time,
identifying anomalies, and alerting the operator when a contaminant might be present. Through
pilot studies, CANARY has demonstrated its ability to detect unexpected "normal" events, such
as sensor malfunctions or pipe breaks. This study systematically investigates how CANARY
could be used to provide multiple benefits to water utilities by improving water system
operations, treatment, and security. The probable causes of standard distribution system water
quality issues (e.g., maintenance of an acceptable disinfectant residual, biofilm control, customer
concerns) are presented, as well as the associated changes in water quality parameters that
could be detected by CANARY. The range of conditions under which CANARY could detect each
of these water quality issues is evaluated. For water utilities of all sizes, the value of using this
type of detection software to enhance detection and response to a wide range of water quality
events is discussed.
1. Introduction
Water quality sensors placed optimally throughout a water distribution system as part of a
Contamination Warning System (CWS) in combination with other monitoring approaches, such
as public health surveillance systems and customer complaint monitoring, can detect incidents of
abnormal water quality. Event Detection Systems (EDSs) are a necessary component of a CWS;
they evaluate the considerable amount of data generated from online water quality monitors and
distinguish abnormal from normal water quality patterns. Such systems have been developed to
improve the security of water distribution systems; however, they also act to improve every day
water quality operations. In this paper, the potential for using CWSs and EDSs to detect a wide
1

-------
variety of distribution system issues is investigated including; pipe breaks, cross-connections,
nitrification, water quality decay, intrusion via pressure transients, and caustic overfeeds.
Sensors capable of detecting all possible contaminants are not commercially available; therefore,
the majority of detection methods use standard water quality parameters, such as free chlorine,
pH, and specific conductivity, as surrogate parameters to indirectly monitor for contamination in
the water distribution network (EPA 2005a). Hall et al. (2007) found that several water quality
parameters, including conductivity, total organic carbon (TOC), total/free chlorine, and
oxidation-reduction potential (ORP), exhibit consistent and significant changes in the presence of
a broad spectrum of potentially hazardous substances (see Table 1). Affordable, multi-sensor
probes are commercially available that include the types of sensors evaluated in the study by
Hall (2007).
Table 1. Standard water quality parameters evaluated for use in detecting abnormal changes in water
quality due to contamination \adapted from (Hall 2009) Table 2.2],	
Parameter
Parameter Applicability to Water Quality
Free chlorine
Parameter that indicates the concentration of
disinfectant available to inactivate bacteria and
viruses and protect against re-growth.
Total chlorine
Sum of combined chlorine and available free
chlorine remaining after a given contact time.
Important parameter for utilities that practice
chloramination.
pH
Indicator of acidity or alkalinity of water.
Turbidity
Indicator of suspended matter and microscopic
organisms. Readings may be variable and
system specific.
Conductivity
Capacity of water to carry an electrical current.
Strong indicator of dissolved salts, but can be
variable.
Oxidation reduction potential (ORP)
Gauge of dissolved oxidizing and reducing
agents (metal salts, chlorine, sulfite ion).
Indicates antimicrobial potential of the water.
Total organic carbon (TOC)
Reasonably stable measure of dissolved and
particulate organic compounds. Useful in DBP
control and disinfection optimization.
Temperature
Biological and chemical activities are heavily
influenced by water temperature. Readings
might vary seasonally and be dependent on the
source.
Dissolved oxygen (DO)
Measures concentration of oxygen dissolved in
water. Indicates chemical and biochemical
activity.
Ammonia
Operational control parameter, critical to
chloramine decay and nitrification processes.
2

-------
Prompt identification of atypical water quality conditions is a critical element of successful event
detection (Byer 2005; Hall 2007; Hart 2010; Murray 2010; Yang 2009). When an EDS indicates
an early warning of a possible contamination, utilities can be proactive in mitigating the potential
impacts. The mitigation measures can be implemented while the more lengthy process of
accurate identification of specific contaminants takes place, which typically involves plate and
culture methods for biological contaminants and advanced analytical instrumentation (e.g., gas
chromatography - mass spectrometry (GC-MS), atomic absorption (AA), or liquid
chromatography - mass spectrometry (LC-MS)) for chemical agents (Hall 2007).
Successful event detection requires that hazards are clearly distinguished from background noise.
Considerable variation in water quality throughout the distribution network adds a layer of
complexity to detecting contamination with water quality sensors. Water quality data is noisy
due to a variety of factors; some of which can be attributed to operational events, while others
are a reflection of daily and seasonal patterns and source mixing (Byer 2005; Kroll 2006;
McKenna 2008). In addition, the general condition of the distribution system infrastructure (e.g.,
the type of material, age) can have a significant affect on water quality (AwwaRF 2007).
The CANARY EDS (Hart 2009; Murray 2010) was developed through collaboration between the
U.S. Environmental Protection Agency (EPA) and the Department of Energy's Sandia National
Laboratories. CANARY offers an advance in detection technology by enabling the use of water
quality sensors to indirectly detect the occurrence of contaminants in conjunction with statistical
tools that analyze the water quality data in real-time and alert the utility operator of a possible
contamination. Multiple event detection algorithms have been developed, tested and included in
the CANARY software. The detection algorithms process data at each time step and the
likelihood of a water quality event is calculated with respect to the recent water quality values.
Algorithms in CANARY constantly adapt to new water quality values and look for significant
variation from the changing background. If enough outliers (observed values that differ from
predicted values) are identified within a specific time period, CANARY signals an alarm to
utility operators indicating an episode of abnormal water quality.
CANARY is a free software tool for drinking water utilities. It is accessible worldwide, and can
be downloaded from EPA's website. EPA offers CANARY webinars and email updates to the
600 users in 15 countries which have accessed the software. In addition, EPA has trained several
water utilities and consultants to use the software. CANARY has been pilot tested at utilities in
five major U.S. cities, is running at utilities in major cities, including Cincinnati and Singapore,
and is being evaluated at several other utilities.
Given the broad scope of water quality problems that can occur within a distribution system,
EPA along with distribution system specialists have begun to assemble literature on the topic of
public health risks that could be directly connected to distribution systems (EPA 2006a). These
EPA reports are focused on nine specific distribution issues: intrusion, cross-connection control,
aging infrastructure and corrosion, permeation and leaching, nitrification, biofilms/growth,
covered storage, decay in water quality over time, and new or repaired mains. Three of these,
cross-connection control, finished water storage facilities, and new or repaired mains, are
identified by the National Academy of Science (NAS) as high risk priorities with respect to
public health threats, system vulnerability, and security (NAS 2006).
3

-------
This study considers the potential for using CANARY to detect not only intentional
contamination incidents but also unanticipated, unintentional normal occurances in water
distribution systems such as pipe breaks, nitrification episodes, cross-connections, or pressure
issues identified by EPA and partners. In this paper, selected distribution system issues are
investigated to determine the potential for CANARY or other EDSs to detect such incidents. The
cause of each distribution system issue is identified, the costs incurred by utilities to manage
each issue are presented, and the surrogate water quality parameters are listed that would be
impacted prior to or during the event. Additional consideration is given to issues that have been
previously detected with an EDS in case studies. Section 2 examines specific distribution system
issues (pipe breaks, corrosion, cross-connections) as well as water quality issues (nitrification,
decay in water quality over time) and operational issues (chemical overfeeds or pressure
transients) that could potentially be detected by CANARY. The final section is a summary and
discussion on the potential for using the CANARY EDS for purposes beyond security and
highlights research needs.
2, Potential Distribution System Issues Detectable by CANARY
The nine distribution system issues have been combined into a smaller set of issues addressed in
the remainder of this paper. Pipe breaks, as a result of aging infrastructure and corrosion, are
discussed first, followed by cross-connections, nitrification, water quality decay, and intrusion
via pressure transients. Though not explicitly part of the nine issues identified in the EPA
whitepapers, chemical overfeeds are also discussed, given that case studies have shown their
potential for detection by an EDS. The potential for the CANARY EDS to detect each of the
distribution system issues is assigned as high, medium, or low. A high detection potential
indicates one or more case studies exist that show another EDS has previously detected this
issue. A medium detection potential signifies that research exists which confirms the capability
of water quality sensors to detect changes in water quality associated with that specific issue. A
low detection potential is assigned to issues that need additional study. A discussion of the
potential for CANARY to detect these problems rapidly enough to mitigate impacts follows.
2.1 Pipe Breaks
Roughly 880,000 miles of installed water main pipe are buried underground in the U.S. (Sadiq
2007). A number of variables, including surrounding soil type, pipe material, diameter, and date
of installation, can affect pressurized mains sometimes causing a break (EPA 2002d). A water
main break is typically defined as the structural failure of a pipe (EPA 2005b). Main breaks are a
frequent occurrence, totaling 240,000 yearly breaks, or roughly 650 breaks per day across the
U.S. (EPA 2007b). Repair operations provide an opportunity for contaminant intrusion when the
interior of the pipes and fittings come into contact with contaminated soil or water nearby
(Besner 2002). A study by Haas et al. (1998) emphasized the health risks after examining
bacterial loading of repair trenches, and finding that the largest bacterial densities were localized
alongside existing pipe. Additionally, of the 619 waterborne disease outbreaks reported in water
systems from 1971-1998, two of the largest outbreaks were linked to contamination of a broken
main, one of which resulted in 1,272 cases of Giardia (Craun 2001).
4

-------
Utilities could incur significant costs associated with main breaks, especially if repair costs
continue to rise. The average cost to a large mid-western utility to repair a main break was over
$5,300 in 2011 (Klopfer 2012). Indirect costs associated with main breaks, including water loss,
and lost revenue from service disruption, can add 20% to 40% more (EPA 2002b). Additional
adverse effects involve public safety issues (traffic disruptions, electrical shock hazards, loss of
fire flow), resource depletion (both water and energy), environmental degredation (chlorinated
water release), and public health problems (waterborne disease outbreaks, sewer overflows from
flooding).
During a pipe break, multiple water quality parameters would likely be impacted. Sudden
increases in turbidity with a corresponding fluctuation in disinfectant residual generally indicate
a system disturbance such as a main break (Kirmeyer 2002a). In a 2009 case study, Kroll
reported that 24 hours prior to a major main rupture there was an increase in turbidity. When the
main ruptured the next day, there was a spike in turbidity (over a roughly 4 hour period).
Generally, an increase in turbidity is associated with an increased disinfectant demand, resulting
in a sudden decrease in the residual (Kirmeyer 2002b). Loss or reduction in pressure might also
occur system-wide in connection with a main break. CANARY would detect abrupt and
sustained changes in these parameters.
A crucial factor in the likelihood of a CANARY alert is the distance between the break location
and the nearest sensor station. If the break occurs in an area of the distribution system with no
sensor coverage, or it occurs just below a sensor station, the probability of detection may
decrease, depending on the type and concentration of the contaminant.
Prior to a pipe break, distribution system problems could manifest as aesthetic problems (e.g.,
"red" or "rusty" water) that are not directly related to any public health effects (Lytle 2005;
Sadiq 2007), yet might be indicative of a larger problem (e.g., corrosion) that could eventually
lead to a pipe break. Variability in the mechanisms that induce corrosion makes it difficult to
detect using an EDS. A variety of water quality parameters are associated with corrosion (e.g.,
pH, DO, disinfectant dose). The time scale over which the change in water quality parameters
would occur is also variable and likely system specific. For example, pitting, a localized
acceleration of corrosion that thins the pipe wall (Lytle 2005), can cause pipe failure in one
system as little as a few weeks (Schock 2011a), while in a different system the same type of
corrosion could take significantly longer to cause a failure. The changes in water quality
parameters associated with corrosion (e.g., pH or TDS fluctuation) would need to display sudden
and persistent changes in average value over multiple time steps to be detected by CANARY,
however, the continuous time period over which corrosion usually occurs would not be
conducive to CANARY detection.
5

-------
Table 2. Summary of the water quality parameters that would be affected by pipe breaks and corrosion,
Incident/Issue
Water Quality
Parameter
Time Scale
EDS Detection
Potential
Pipe break
Turbidity (+)
Free chlorine (-)
Pressure (-)
Hours
High, depending
on the proximity
to water quality
sensors
Corrosion
pH(-)
Free chlorine (-)
DO(+)
TDS (+)
Weeks to years
Low, due to long
time scale
2.2 Cross-connection Control/Backflow
A cross-connection occurs when a nonpotable substance comes in contact with the potable
drinking water supply, such as when drinking water lines are connected to chemical feeds for
industrial processes or for firefighting (EPA 2001). Cross-connections create a potential pathway
for backflow of nonpotable water into the distribution system. Backflow is a result of reduced
pressure in the distribution system (also known as backsiphonage), or conversely, it is due to
increased pressure from a nonpotable source (backpressure). Unexpected events such as pump
failures and main breaks can be caused by backsiphonage. Backsiphonage can draw in nearby
water through cracks in mains, or leaky joints, during negative pressure events (Kirmeyer
2002a). Corrosion can be induced when backflows introduce certain contaminants (e.g., acids
and carbon dioxide) (EPA 2001).
Cross-connections are recognized by the National Academy of Sciences as a high-priority issue
pertaining to public health risks (Wilkes 2008). A 2001 report on waterborne disease outbreaks
spanning from 1971-1998 states that of the outbreaks resulting from distribution system failures
approximately 51% were due to cross-connection and backflow (Craun 2001). In addition to
public health impacts, cross connections can cause denial of service to customers and loss of
revenue for the utility. Results from a 1999 American Backflow Prevention Association (ABPA
1999) survey of 92 utilities show that an average of 494 hours of operator time are used per
backflow event, at an average cost of $14,800 per event.
Utilities face a considerable challenge with detecting cross-connections, since there are few
immediate indicators and such events tend to be localized. A significant disinfectant demand
could indicate the existence of cross-connections or a backflow event depending on the
substance that enters into the distribution system (AWWA/PWNS 1995; EPA 2001; Haas 1998;
Schneider 2010). Schneider et al. (2010) suggested a change in ORP as possible indicator of a
cross-connection. Turbid water can also be a product of a cross-connection (Deb 2000). Rapid
and sustained changes in one or more of these parameters would trigger an alarm from
CANARY, indicating a potential contamination event. Additionally a sudden change in water
temperature could signal a problem with distribution system integrity, since water of a different
temperature might enter the system during a backflow event. An increase in water temperature
6

-------
will also increase the rate of decay for chlorine and chloramine (EPA 2007a). Utilities with an
understanding of the water temperature ranges in their system can employ CANARY to detect
any changes in temperature, while filtering out recurring changes in operation that can affect
water quality parameters. Pressure monitoring in the distribution system can identify changes in
pressure that may leave a system vulnerable to contaminant entry into the distribution.
LeChevallier et al. (2003) described negative pressure events that were short-lived, spanning
seconds or minutes. CANARY is capable of sampling at short intervals, although a typical
sampling interval for water quality data analyzed by CANARY is two minutes (Murray 2010).
Table 3. Summary of the water quality parameters that would be affected by cross-connections, the time
Incident/Issue
Water Quality
Parameter
Time Scale
EDS Detection
Potential
Cross-connection
Turbidity (+)
Free chlorine (-)
Pressure (-)
Temperature (+/-)
Hours
Medium,
depending on the
magnitude and
type of the event,
and its proximity
to water quality
sensors
2,3 Nitrification
Nitrification episodes present a managment challenge for utilities that use monochloramine as a
residual disinfectant. Monochloramines can break down into free ammonia in drinking water
which can be oxidized sequentially into nitrite and nitrate during the nitrification process (EPA
2002c). Typical water quality problems that result from nitrification include: a decrease in
disinfectant residual, an increase in heterotrophic bacteria, an increase in nitrate and nitrite, a
decrease in dissolved oxygen (resulting in corrosivity and taste and odor complaints), and a
decrease in pH (Deb 2000). The reduction of nitrate to nitrite can be an issue in a distribution
system with low pH as potentially carcinogenic compounds could form as a result (De Roos
2003). Nitrification problems typically occur in the distribution system during warmer times of
the year and lower water usage. Probable causes of nitrification include: the presence of
ammonia in source water, the growth of nitrifying bacteria due to increasing temperatures, and
increased hydraulic residence time (Deb 2000; EPA 2002a).
An investigation by Cook et al. (2012) noted that a number of water quality parameters are
indicative of nitrification in the distribution system including pH, DO, and total chlorine residual.
Cook's study tracked the progress of a mid-sized utility installing several online monitoring
stations throughout their distribution system to improve water quality. A total of ten months of
chlorine residual data was collected at 15 minute intervals from three distinct areas of the
distribution system in which all received water from the same treatment plant. The data showed
that two of the three monitored sites generally exhibited the same trends. The dissimilarities in
the third site were believed to be a result of nitrification issues upstream. The researchers
concluded that conditions recognized as contributing to nitrification (i.e., low chlorine residual)
can be identified using online water quality sensors and event detection systems.
7

-------
In addition to a decrease in chlorine residual, a reduction in pH and DO are also indicative of a
nitrification event (Deb 2000; EPA 2002c). An increase in temperature can cause nitrification,
because when temperatures increase due to seasonal changes, the potential for the growth of
nitrifying bacteria increases (Wilczak 2006). Given that the time scale associated with seasonal
fluctuation in distribution water temperature, this would occur at a gradual rate that could
prohibit CANARY from detecting changes, however sudden changes in pH, DO, and
temperature from baseline levels would trigger an alarm in CANARY.
Table 4. Summary of the water quality parameters that would be affected by nitrification, the time frame
Incident/Issue
Water Quality
Parameter
Time Scale
EDS Detection
Potential
Nitrification
Total chlorine (-)
pH(-)
DO (-)
Temperature (+)
Ammonia (+)
Weeks
Medium,
depending on
proximity to
water quality
sensors
2,4 Water Quality Decay
Maintenance of an acceptable disinfectant residual in finished water is a primary concern for
drinking water utilities. The Surface Water Treatment Rule (SWTR) sets regulations on
minimum and maximum residual levels that can reach a customer's tap. Under the SWTR,
finished water entering the distribution system cannot fall below the 0.2 mg/L disinfectant
residual benchmark for more than a four hour period without falling out of compliance. The
residual concentration within the distribution system must be maintained at a detectable
concentration, meaning that 95% of the samples taken must detect the presence of a residual.
Low disinfectant residuals in the distribution system increase the chances of biological re-growth
(i.e., biofilms) and can result in customer complaints regarding taste and odor. Earthy, musty,
and moldy taste and odor complaints are associated with a loss of disinfectant (EPA 2007a).
High concentrations of disinfectant, on the other hand, can result in the formation of potentially
harmful disinfection by-products (DBPs). In 2002, the Stage 1 Disinfectants and Disinfection
By-Product Rule instituted a maximum disinfectant residual level of 4.0 mg/L for chlorine and
chloramines.
Depletion of the disinfectant residual can be attributed to distribution system conditions (e.g.,
topography of the network, excessive corrosion and corrosion by-products, presence of biofilm,
other organic or inorganic material, backflows, high water age, or other contaminant intrusions)
(AWWA 2011; Clark 1998; Deb 2000; EPA 2002e; Kirmeyer 2002a; Schock 2011b), or the
chemical or biological characteristics of treated water entering the distribution system (e.g., pH,
temperature, inadequate organic removal from source water).
The most direct parameter that would indicate an issue with residual maintenance in a system
that chlorinates would be a decrease in free or total chlorine; however, there are additional
8

-------
parameters that would confirm this problem. For example, if the chlorine (oxidant) is being
reduced, ORP readings would show a corresponding decrease (EPA 2008a). Temperature and
TOC increases generally increase the rate of disinfectant decay (EPA 2007a). Based on the
water chemistry associated with residual decay, CANARY could theoretically detect any rapid
changes in ORP, free chlorine, TOC, and/or temperature.
Table 5. Summary of the water quality parameters that would be affected by water quality decay, the time
Incident/Issue
Water Quality
Parameter
Time Scale
EDS Detection
Potential
Decay in Water
Quality
Free chlorine (-)
TOC (-)
ORP (-)
Temperature (+)
Days to Months
Medium,
depending on the
time scale and
proximity to
water quality
sensors
2,5 Intrusion via Pressure Transients
Pressure problems in the distribution system, either an excess or deficiency, can be caused by
main breaks, rapid demand changes, sudden stopping or starting of a pump, power failure, fire
flow, or air valve slam (EPA 2001). When velocity in a pipe is suddenly stopped, there is a
corresponding water pressure increase of 50-60 psi per 1 ft/sec of velocity (Kirmeyer 2002b).
These changes in pressure can occur in a matter of seconds, and are known as transients.
Transients involve a sudden change in velocity and pressure; however, it is the after- effects of
pressure issues that are concerning to utilities. High pressure events can cause device failures
and pipe breaks, while low pressure events are typically associated with pipe collapse or
contaminant intrusion (Collins 2012). Damage from a transient may not be immediately noticed,
as in the case of small cracks. Intensified corrosion in the damaged areas may cause future
problems (presented in Table. 2). In addition, these events may produce intense velocities that
could potentially resuspend settled particles or detach biofilm.
Pipes positioned below the water table are subject to pressure from the exterior water (depending
on the height of the water table above the pipe). Given this, water outside of the pipe could
intrude into the pipe under low or negative pressure conditions within the pipe. It is assumed
that all distribution systems have some pipe below the water table at least some time of the year.
Any contaminant exterior to the distribution system could enter potable water supplies during a
negative pressure event. Chemical contaminants in the groundwater might include pesticides,
petroleum products, fertilizers, solvents, detergents, pharmaceuticals, and other compounds.
Other studies have detected insect repellants, fire retardants, and other industrial chemicals
(Koplin 2002). If chemical compounds intrude in sufficient concentration or volume, they might
result in acute toxicity, or long term exposure to low levels could result in cancers (EPA 2009).
Microbial contaminants are a concern because even with dilution some microbes (e.g., viruses)
could cause an infection with a single organism.
9

-------
Collins (2012) investigated depressurization events including the effect of the transient on the
distribution system. In this work, negative pressures were identified at durations up to twelve
seconds. Similar durations of negative pressures resulting from operations have been reported by
LeChevallier (2003). This study noted negative pressure durations lasting 24 seconds during a
pump shutdown, and zero pressure for 51 seconds resulting from a power outage. Given that
these types of events occur so rapidly, it is unknown at this time whether CANARY or another
EDS could detect changes in pressure.
Table 6. Summary of the water quality parameters that would be affected by pressure transients, the time
Incident/Issue
Water Quality
Parameter
Time Scale
EDS Detection
Potential
Pressure
Transients
Pressure (-)
Seconds
Minutes
Low due to rapid
time scale
2.6 Chemical Overfeed
Although most treatment chemicals added in the distribution system are intended to provide
protection against waterborne disease, to mitigate other potential contaminants, or to control
chemical stability, utilities must ensure that they are not fed into the system in excess. The
activation of either dry chemical feeders or solution pumps should occur instantaneously with the
start-up of the water supply pumps, because the feeding of chemicals into the distribution system
when the water supply pumps are inactive could result in detrimental overfeed to the customers
(EPD 2000).
An overabundance of disinfectant in the system increases the chances that disinfection
byproducts such as trihalomethanes or haloacetic acids could form (EPA 2008b). Overfeed of
disinfectant or other typically used chemicals can be recognized by taste, odor, and color
complaints, but it can also be detected by changes in water quality parameters.
To control pH and reduce corrosivity, operators might use a caustic feed (e.g., orthophosphate)
into the system. An overload of this chemical could occur as a result of human error. Kroll
(2009) noted that this type of overfeed can affect pH and the conductivity of the water. In the
Kroll study, an EDS alarm was triggered when these parameters fluctuated over a four hour
period. In a similar event, a possible chlorine overfeed was suspected when a sudden increase in
pH and turbidity occurred along with a decrease in chlorine and conductivity over a span of a
few hours. CANARY would detect rapid changes in any of the parameters (i.e., pH,
conductivity, turbidity, free chlorine) that might change in response to a chemical feed issue.
CANARY would alarm if these parameters exceeded set point values, or changed significantly
within the parameter limits set by the utility.
10

-------
Table 7. Summary of the water quality parameters that would be affected by chemical over feeds, the
Incident/Issue
Water Quality
Parameter
Time Scale
EDS Detection
Potential
Caustic
Chemical
Overfeed
pH(-)
Conductivity (+)
Chlorine (+)
Hours
High, depending on
the time scale, and
proximity to water
quality sensors
Disinfectant
Overfeed
pH (+)
Turbidity (+)
Chlorine (+)
Hours
High, depending on
the time scale and
proximity to water
quality sensors
3.0 Summary and Conclusions
Online monitoring of common water quality parameters in conjunction with the CANARY EDS
can detect abnormal changes in water quality values and alert operators of a possible water
security event. Recent literature suggests that this technology can also be used to identify other
more common distribution system issues experienced by utilities (Cook 2012; Hart 2010; Kroll
2009; Kroll 2006; Murray 2010). Advanced, continuously adapting algorithms used in
CANARY could be configured to alarm on distribution system conditions that have been shown
to be indicative of a specific issue in order to detect a broader set of water quality problems.
Based on previous case studies and general water chemistry knowledge, this paper demonstrates
the potential for CANARY to detect unexpected events such as pipe breaks and cross-
connections. These events would likely occur over a relatively fast time scale, making them
more likely to be detected by an EDS. Conversely, while issues like water quality decay can
have significant impacts on the distribution system, detection by CANARY might not be as
likely, mostly due to the gradual time scale over which decay occurs. Likewise, it is unknown at
this time whether events that occur very rapidly, over the span of seconds, such as pressure
transients would be detected by CANARY. However, if this type of data was collected on a
smaller time scale, an EDS would, in theory, be capable of detecting an event.
Additional research needs to be performed to verify that the issues discussed in this paper could
be detected with an EDS. Data sets that would be useful for future research would include
information on multiple parameters from a diverse selection of distribution systems with distinct
water quality conditions and operations. Ideally the data would be representative of a range of
geographic locations and size. The data sets would need to include water quality incidents
mentioned in this study (i.e., pipe breaks, nitrification, chemical overfeed, pressure transients).
Simulation studies can be performed to further inform this research. Future studies could include
pipe break simulations to predict network performance (i.e., pressure deficiencies directly caused
by a break) under pipe failure conditions. Useful data would include knowledge of the extent of
an area that is typically affected by these situations.
11

-------
CANARY supports water utilities in understanding the large volumes of water quality data. An
improved understanding of water quality can lead to enhanced operations (e.g., more consistent
disinfectant residuals, an early warning of nitrification). Because CANARY combines water
quality data from several sensor stations, along with hydraulic information, it might be able to
more accurately and rapidly detect contamination incidents. Rapid detection of contamination
events allows a utility to effectively mitigate public health and economic consequences with less
impact to customers. As a free software tool, CANARY offers significant cost-savings for
utilities. Costs for multi-sensor probes can range from $10,000 to $20,000. Costs associated
with waterborne disease outbreak can reach billions of dollars, and even everyday distribution
system issues (i.e., pipe breaks) can create considerable financial burdens. If these issues are
detected early by CANARY, repairs can be made and more severe impacts can be avoided.
Disclaimer: This project has been subjected to the U.S. Environmental Protection Agency's
review and has been approved for publication. The scientific views expressed are solely those
of the authors and do not necessarily reflect those of the U.S. EPA. Mention of trade names or
commercial products does not endorsement use constitute or recommendation for use.
References
ABPA, A. B. P. A. (1999). "1999 Survey of State and Public Water System Cross-Connection Control
Programs."
AWWA. (2011). Water Quality & Treatment: A Handbook on Drinking Water.
AWWA/PWNS. (1995). Summary of Backflow Incidents.
AwwaRF. (2007). "Advancing the Science of Water: AwwaRF and Distribution System Water Quality."
Besner, M. C., Gauthier, V., Servais, P., and A. Camper. (2002). "Explaining the Occurence of Coliforms in
the Distribution System." J. A WW A, 94(8), 95-109.
Byer, D., and K.H. Carlson. (2005). "Real-time detection of intentional chemical contamination- in the
distribution system "J. A WW A, 97(7), 130-133.
Clark, R. (1998). "Chlorine Demand and TTHM Formation Kinetics: A Second-Order Model." Journal of
Environmental Engineering, 124(1), 16-24.
Collins, R. P., Boxall, B.W., Karney, B.W., Brunone, B., and S. Meniconi. (2012). "How severe can
transients be after a sudden depressurization?" J. AWWA, 104(4), 67-68.
Cook, J. B., Roehl, E.A., Daamen, R.C. (2012). "Control Nitrification with Online Monitoring." Optflow.
Craun, G. F. a. R. L. C. (2001). "Waterborne Disease Outbreaks Caused by Distribution System
Deficiencies." J. AWWA, 93(9), 64-75.
De Roos, A. J., Ward, M.H., Lynch, C.F., and K.P. Cantor. (2003). "Nitrate in public water supplies and the
risk of colon and rectum cancers." Epidemiology, 14(6), 640-649.
Deb, A. K., Momberger, K.A., Hasit, Y.J., and F.M. Grablutz. (2000). "Guidance for Managment of
Distribution Sytem Operation and Maintenance." Denver, CO.
EPA, U. S. (2001). "Potential Contamination Due to Cross-Connections and Backflow and the Associated
Health Risks." Wasington, D.C.
EPA, U. S. (2002a). "Effects of Water Age on Distribution System Water Quality." Washington, D.C.
EPA, U. S. (2002b). "New or Repaired Water Mains." Washington, D.C.
EPA, U. S. (2002c). "Nitrification." Washington, D.C.
EPA, U. S. (2002d). "Deteriorating Buried Infrastructure Management Challenges and Strategies."
12

-------
EPA, U. S. (2002e). "Health Risks from Microbial Growth and Biofilms in Drinking Water Distribution
Systems." Washington, D.C.
EPA, U. S. (2005a). "Overview of Event Detection Systems for WaterSentinel." Washington, D.C.
EPA, U. S. (2005b). "White Paper on Improvement of Structural Integrity Monitoring for Drinking Water
Mains."
EPA, U. S. (2006a). "Total Coliform Rule (TCR) and Distribution System Issue Papers Overview."
EPA, U. S. (2007a). "The Effectivness of Disinfectant Residuals in the Distribution System ", Washington,
D.C.
EPA, U. S. (2007b). "Innovation and Research for Water Infrastructure for the 21st Century."
EPA, U. S. (2008a). "Pilot Scale Tests and Systems Evaluation for the Containment, Treatment and
Decontamination of Selected Materials from T&E Pipe Loop Equipment."
EPA, U. S. (2008b). "Stage 2 Disinfectants and Disinfection Byproducts Rule."
EPA, U. S. (2009). "National Primary Drinking Water Regulations."
EPD, G. (2000). "Guidance Manual for Preparing Public Water Supply System O&M Plans." Atlanta, GA.
Haas, C. N., Chitluru, R.B., Gupta, M., Pipes, W.O., and G.A. Burlingame. (1998). "Development of
Disinfection Guidelinesfor the Installation and Replacement of Water Mains." Denver, CO.
Hall, J., Zaffiro, A., Marx, R., Kefauver, P., Krishnan, R., Haught, R., and J. Herrmann. (2007). "On-line
Water Quality Parameters as Indicators of Distribution System Contamination." J. A WWA, 99(1),
66-77.
Hall, J. S., Szabo, J.G., Panguluri, S., and G. Meiners. (2009). "Distribution System Water Quality
Monitoring: Sensor Technology Evaluation Methodology and Results." Cincinnati, OH.
Hart, D. B., McKenna, S.A., Klise, K.A., Wilson, M.P. and Murray, R.M. (2009). ""CANARY Users Manual
Version 4.3." ", US EPA 600/R-08/040A..
Hart, D. B., McKenna, S.A., Murray, R., and T. Haxton. "Combining Water Quality and Operational Data
For Improved Event Detection." in proceeds of: Water Distribution Systems Analysis (WDSA),
Tucson, AZ.
Kirmeyer, G., Friedman, J., Martel, K., Thompson, G., Sandvig, A., Clement, J., and M. Frey. (2002a).
"Guidance Manual for Monitoring Distribution System Water Quality." Denver, CO.
Kirmeyer, G. J., Friedman, M., Matel, K., and G. Thompson. (2002b). "Guidance Manual for Monitoring
Distribution System Water Quality." Denver, CO.
Klopfer, D., Schramuk, J. (2012). "Implementing and managing a large water utility's underground
corrosion control program." J. AWWA, 104(7), 56-63.
Koplin, D. W., Furlong, E.T., Meyer, M.T., Thurman, E.M., Zaugg, S.D., Barber, L.B. and H.T. Buxton.
(2002). "Pharmaceuticals, hormones, and other organic wastewater contaminants in U.S.
streams, 1999-2000: a national reconnaissance." Environ. Sci. Technol., 36(6), 1202-11.
Kroll, D. "Getting More From Your Investment. Using Water Security Monitoring Technology for Everday
Operations." World Environmental and Water Resources Congress, Kansas City, MO.
Kroll, D., King, K. "Laboratory and flow loop validation and testing of the operational effectivness of an
on-line security platform for the water distribution system." Proc., 8th Annual Water
Distribution Systems Analysis Symposium, ASCE, Reston, VA.
LeChevallier, M. W., Gullick, R.W., Karim, M.R., Friedman, M., and J. E. Funk. (2003). "The potential for
health risks from intrusion of contaminants into the distribution system from pressure
transients." Journal of Water and Health, 1(1).
Lytle, D. A. "Copper Pitting Corrosion and Pinhole Leaks: A Case Study." American Water Work
Association Annual Conference, San Francisco, CA, June 12-16.
McKenna, S. A., Klise, K.A., and Wilson, M.P. (2008). "Detecting changes in water quality data." J.
AWWA, 100(1), 74-85.
13

-------
Murray, R., Haxton, T., McKenna, S.A., Hart, D.B., Klise, K., Koch, M., Vugrin, E.D., Martin, S., Wilson, M.,
Cruz, V., and L. Cutler. (2010). "Water Quality Event Dectection Systems for Drinking Water
Contamination Warning Systems Development, Testing, and Application of CANARY."
EPA/600/R-10/036, Cincinnati, OH.
NAS. (2006). Public Water Supply Distribution Systems:Assessing and Reducing Risks.
Sadiq, S., Imran, S.A., and Y. Kleiner. (2007). "Examining the Impact of Water Quality on the Integrity of
Distribution Infrastructure." Denver, CO.
Schneider, O. D., Hughes, D.M., Bukhari, Z., Lechevallier, M., Schwartz, P., Sylvester, P., and J.J. Lee.
(2010). "Determining vulnerability and occurence of residential backflow." J. A WW A, 102(8), 52-
63.
Schock, M. R., and D.A. Lytle. (2011a). "Internal Corrosion and Deposition Control." Water Quality and
Treatment A Handbook on Drinking Water, J. K. Edzwald, ed., American Water Works
Association, Denver, CO.
Schock, M. R., and D.A. Lytle. (2011b). "Internal Corrosion and Deposition Control." Water Quality &
Treatment: A Handbook on Drinking Water, A. W. W. Association, ed., Edzwald, J.K. , Denver,
CO.
Wilczak, A. (2006). "Fundamentals and Control of Nitrification in Chloraminated Drinking Water
Distribution Systems." Manual of Water Supply Practices- M56, AWWA, ed., Denver, CO.
Wilkes, D. (2008). "Drinking Water Quality Optimization: Focus on the Distribution System." Florida
Water Resources Journal.
Yang, J. Y., Haught, R.C., and J.A. Goodrich. (2009). "Real-time contaminant detection and classification
in a drinking water pipe using conventional water quality sensors: techniques and experimental
results." Journal of Environmental Management, 90(8), 2494-2506.
14

-------