&EPA
United States
Environmental Protection
Agency
Office of
Drinking Water (WH-550)
Washington DC 20460
September 1982
EPA/570/9-82-001
Sampling Frequency-
Microbiological
Drinking Water
Regulations
Final Report
-------
SAMPLING FREQUENCY - MICROBIOLOGICAL
DRINKING WATER REGULATIONS
Final Report
by
Wesley 0. Pipes and Robert R. Christian
Department of Biological Sciences
Drexel University
Philadelphia, Pennsylvania
EPA R-805-637
Project Officers
Paul Berger Edwin E. Geldreich
Criteria & Standards Division Water Supply Research Division
Office of Drinking Water Municipal Environmental Research Lab.
Washington, D.C. 20460 Cincinnati, Ohio 45268
OFFICE OF DRINKING WATER
U.S. ENVIRONMENTAL PROTECTION AGENCY
WASHINGTON, D. C. 20460
-------
DISCLAIMER
This report has- been reviewed by the Office of Drinking Water and the
Municipal Environmental Research Laboratory, U.S. Environmental Protection
Agency and approved for publication. Approval does not signify that the
contents necessarily reflect the views and policies of the U.S. Environ-
mental Protection Agency, nor does mention of trade names of commercial
products constitute endorsement or recommendation for use.
ii
-------
ABSTRACT
The primary goal of this project was to develop a sampling model which
can be used to specify the sampling frequency needed to determine compliance
with the microbiological maximum contaminant levels of the National Interim
Primary Drinking Water Regulations. Two approaches to model development were
used. The emperical approach was based on fitting coliform data to frequency
distributions. The mechanistic approach was an attempt to analyze the phys-
ical elements of the water system in order to determine if samples from a
particular location are representative of water quality over an area and if
it is more likely that coliforms will be found in certain types of sampling
locations.
The emperical approach resulted in the finding that coliform counts in
samples from a water distribution system do not fit either Poisson or Poisson
plus added zeros distributions. The counts do fit negative binomial dis-
tributions and truncated lognormal distributions. It was proposed to use
the truncated lognormal distribution because it is familiar to personnel in
the water works field and computations are somewhat simpler. Truncation in
this case means that coliform densities 80/100ml cannot be
measured because of limitations of the laboratory methodology for coliform
detection and enumeration. The probability of violation of one of the
microbiological maximum contaminant levels is a function not only of the
densities of coliforms in the water and the number of samples collected
but also of the parameters of the lognormal distribution. Higher values of
the geometric standard deviation (that is, a higher degree of aggregation
of coliform bacteria in the system) lead to larger numbers of samples being
needed for detection of contamination.
The mechanistic approach resulted in the finding that water distribution
systems can be divided into hydraulically isolated sections. Coliform occur-
rence can vary significantly among the isolated sections of a water distribu-
tion system which leads to the conclusion that all sections of the system
must be included in the microbiological monitoring program in order to get
an accurate measure of coliform occurrence. No significant difference in
the occurrence of coliforms between peripheral and nonperipheral sampling
locations was found. Also, no increase in coliform occurrence with distance
from the water source into the system was found. Thus, sampling locations
for a microbiological monitoring program should be selected by a randomiza-
tion procedure which gives equal probability of selection to all sampling
locations.
ill
-------
CONTENTS
Foreword ill
Abstract iv
Figures viii
Tables x
List of Abbreviations and Symbols xiv
Acknowledgments xvi
1. Introduction 1
1.1 Objectives of the Project 1
1.2 Approaches 1
1.3 Objectives of Monitoring 2
1.3.1 Safety 2
1.3.2 Reliability 3
2. Conclusions and Recommendations 6
3. Methods and Procedures 8
3.1 Introduction 8
3.2 Planning a Sampling Program 8
3.2.1 Hierarchy of Sampling Units 8
3.2.2 Random Sampling 12
3.2.3 Stratified Random Sampling 15
3.2.4 Other Sampling Patterns 15
3.2.5 Application to Monitoring Programs 16
3.3 Collection and Analysis of Samples 16
3.3.1 Collection of Samples 17
3.3.2 Analysis of Samples 17
3.3.3 Special Quality Assurance Procedures 19
3.3.4 Data Management 19
4. Sample Validity 22
4.1 Introduction 22
4.2 Subsequent Contamination 23
4.2.1 Sterility of Sample Bottles 23
4.2.2 Tap Location 23
4.2.3 Sampler Error 27
4.2.4 Contamination During Transit in Ice 28
4.2.5 Carry Over During Filtration 30
4.2.6 Summary on Subsequent Contamination 35
4.3 Changes in Coliform Densities in Water Samples 36
4.3.1 Preservation Techniques 36
4.3.2 Second Day Filtration 37
4.4 Observations on Interferences 40
4.5 Verifications of Presumptive Coliform and Other Colonies .. 43
-------
5. Maximum Contaminant Levels for Coliform Bacteria and Their
Relation to Frequency Distributions 54
5.1 Introduction 54
5.2 Characteristics of MCL's Independent of Specific
Frequency Distributions 54
5.3 Previous Work on Frequency Distributions of Bacteria .... 56
5.3.1 Distributions Considered 56
5.3.2 Poisson Distribution 56
5.3.3 Poisson Plus Added Zeroes 60
5.3.4 Assumptions Underlying the Simplistic Models 60
5.3.5 Lognormal Distribution 63
5.3.6 Negative Binomial Distribution 67
5.4 Data Fitting 67
5.4.1 Data Sets 70
5.4.2 Poisson Distribution 70
5.4.3 Poisson Plus Added Zeroes Distribution 70
5.4.4 Lognormal Distribution 74
5.4.5 Negative Binomial Distribution 83
5.5 Relationship of the Lognormal Distribution to the National
Interim Primary Drinking Water Regulations 85
5.5.1 General 85
5.5.2 Probability of Exceeding the Average Rule 86
5.5.3 Probability of Exceeding the Maximum Rule 91
5.5.4 Probability of Obtaining Samples above Particular
Densities 94
5.5.5 The Probability of Exceeding Regulations if the
Number of Samples is Reduced 99
5.5.6 Interpreation of Field Results When All Samples
are Negative 99
5.6 Replication of Samples and Larger Volume Samples 100
5.6.1 Introduction 100
5.6.2 Replicate Samples Collected 101
5.6.3 Results from Duplicate Samples 101
5.6.4 Negative Sampling Results Ill
5.6.5 Positive Sampling Results 112
5.6.6 Larger Volume Samples 112
6. Mechanistic Approach 125
6.1 Introduction 125
6.2 Configuration of Sections 125
6.2.1 Definitions of Distribution System Elements 125
6.2.2 Isolated Sections 126
6.2.3 Configuration of Systems Sampled 128
6.2.4 Relationship of Configuration to Sampling 131
6.3 Sampling by Sections 134
6.3.1 Differences Among Sections 134
6.3.2 Differences Within Sections 143
6.4 Hypotheses About Contamination 145
6.4.1 Upstream and Downstream Samples 150
6.4.2 Peripheral Locations 152
6.4.3 Distance from Treatment Plant 153
6.5 Summary and Implications 156
vi
-------
Appendices (Appendices are not included in this publication, but are avail-
able from the Office of Drinking Water.)
A. Coatesville, Pennsylvania
B. Woodbury Heights, New Jersey
C. Long Beach Water Co.
D. Downingtown, Pennsylvania
E. Brooklawn, New Jersey
F. Mt Idy Mobile Home Park
G. Spring Run Homes
H. Marshallton Woods Homes
I. Bradford Glen Homes
J. Effect of Filter Backwashing on Bacteriological Quality of
Drinking Water - (included in this publication)
vii
-------
FIGURES
Number Page
3.1 Woodbury Heights Sampling Location Map 10
3.2 Portion of Location File from Woodbury Heights, N.J. 20
3.3 Portion of Bacteriological File from Woodbury Heights, N.J. 21
5.1 Probability of Exceeding Average Rule for Random Dispersion
of Coliforms at Different Actual Mean Densities 58
5.2 Probability of Exceeding Maximum Rule for Random Dispersion
of Coliforms in a Distribution System 59
5.3 Probability of Obtaining No Positive Samples for Actual Mean
Density of 1 per 100ml 61
5.4 Distribution of computed arithmetic .mean (y) from the geometric
means and geometric standard deviations in lognormal distribu-
tions. Isobars represent different arithmetic means. 64
5.5 The percentage of samples with >^ coliforms/lOOml as a function of
the geometric mean (Gm) and geometric standard deviation. 65
5.6 Likelihood of exceeding the average and maximum rules for
lognormal distributions when a large number of samples are
taken per month. 66
5.7 The Probability of Obtaining a Sample Mean of 1 for a True Mean
Density of 1 as a Function of Number (n) of Samples and
Variability of the System, (after Muenz, 1978). 68
5.8 Probability of obtaining a particular count (X) when y=l and K
is varied from 0.1 to 10 in a negative binomial distribution.
X=0, 1,2,3^ 4. 69
5.9 Log-Probability Plots of Coatesville Coliform Data 75
5.10 Extrapolation of Log-Probability Plot of Coatesville Data for
Fitting a Truncated Lognormal Distribution. 76
5.11 Probability of Exceeding the Average Rule when y=l as a Function
of Number of Samples 88
5.12 Probability of Exceeding the Average Rule when «=1 as a Function
of Number of Samples 90
5.13 Probability of Exceeding the Maximum Rule when «=1 as a Function
of Number of Samples 93
viii
-------
Number Page
5.14 Probability of Obtaining at Least 1 Positive Sample as a Function
of Number of Samples when « = 1. 96
5.15 Probability of Obtaining at Least 1 Sample with >_ 4 Coliforms
per 100ml when «=1 as a Function of Number of Samples 97
5.16 Probability of Obtaining at Least One TNTC Sample when ==1 as
a Function of Number of Samples 98
6.1 Representative Examples of Four Isolated Section Configurations 127
6.2 Relationship of average cross sectional area to volume of water
distribution system sections. 132
6.3 Flow Diagrams for the Isolated Sections of System BL. 151
ix
-------
TABLES
Number Page
3.1
4.1
4.2
4.3
4.4
4.5
4.6
4.7
4.8
4.9
4.10
4.11
4.12
4.13
4.14
4.15
4.16
4.17
4.18
5.1
Sections of Woodbury Heights, N.J.
Comparison of Positive Coliform Results in Source Water
and Distribution System Samples for Seven Systems
Comparison of Coliform Results from Different Sampling Sites
Comparison of Paired Samples from Commercial Establishments
Tests for Subsequent Contamination from Fingers of Samplers
Contamination of Sample Bottles Immersed in Ice Water Containing
Different Densities of Coliforms
Positive Samples from Different Types of Bottles Immersed in
Ice Water for Various Periods of Time
Relationships of Coliform Density, Type of Filter Funnel and
Rinsing Protocol as Determinants of Carry Over
Potential Carry Over Analysis of Selected Samples Collected in
the Field
Effect of Preservation Techniques on Standard Plate Count
Comparison of Preservation Techniques
Delayed Filtration Experiments
Verification Results on Systems Sampled
Summary of Verifications and API Identifications
Summary of Reactions of Isolates Identified as Enterobacter
Agglomerans (IND+)
Summary of Reactions of Isolates Identified as Enterobacter
Agglomerans (IND-)
Summary of Isolates Identified as E. Coli
Comparison of Test Results Between Escherichia Coli and
Enterobacter Agglomeran Strains which are IMViC Type -H —
Summary of API Identification of Non-Coliform Colonies
Examples of Various Violations of the MCL's when Sample Size is
11
24
25
26
28
29
31
33
34
38
39
41
45
46
49
50
51
52
53
10 55
5.2 Summary of Water Systems Sampled 71
-------
Number Page
5.3 Applicability of the Poisson Distribution to the Frequency
Distribution of Confirmed Total Coliforms in Water
Distribution Systems 72
5.4 Applicability of the Poisson Plus Added Zeroes Distribution
to the Frequency Distribution of Confirmed Total
Coliforms in Water Distribution Systems 73
5.5 Example of Computation of Factors to be Used in the Assessment
of the Lognormal Distribution Based on Data from WH I 78
5.6 Parameters of the Lognormal Distribution of Coliforms in
Woodbury Heights, N.J. as Computed by the Graphical Approach
using all Locations Having Equivalent Volumes Filtered 79
5.7 Parameters of the Lognormal Distribution of Coliforms in Water
Distribution Systems as Computed by Least Squares Regression 80
5.8 Relationship of Sample Arithmetic Mean (x) and Computed Arith-
metic Means from the Regression Analysis (ar) and Graphical
Analysis (^g) for the Lognormal Distribution
5.9 Parameters of the Negative Binomial Distribution of Confirmed
Total Coliforms in Water Distribution Systems 84
5.10 Probability of Exceeding the Average Rule as a Function of
Number of Samples when the Geometric Mean is Held Constant 86
5.11 Computed Arithmetic Means (=) When y = 1 as a Function of
Number of Samples 87
5.12 The Average Computed Arithmetic Mean («) When p = 0 as a
Function of Number of Samples 89
5.13 Probability of Exceeding the Maximum Rule as a Function of
Number of Samples When Geometric Mean is Held Constant 92
5.14 Occurrence of Violations in the Average and Maximum Rules
in Water Distribution Systems Studied 94
5.15 Sequences of Coliform Counts in Replicate Samples-Coatesville 102
5.16 Sequences of Coliform Counts in Replicate Samples-Woodbury
Heights 103
5.17 Sequences of Coliform Counts in Replicate Samples-Long Beach
Water Co. 104
5.18 Sequences of Coliform Counts in Replicate Samples-Downingtown 105
5.19 Sequences of Coliform Counts in Replicate Samples-Brooklawn 106
5.20 Sequences of Coliform Counts in Replicate Samples-Mt. Idy
Mobile Home Park 107
5.21 Sequences of Coliform Counts in Replicate Samples-Spring Run 108
5.22 Sequences of Coliform Counts in Replicate Samples-
Marshallton Woods 109
xi
-------
Number Page
5.23 Sequences of Coliform Counts in Replicate Samples-
Bradford Glen 110
5.24 Analysis of Data from Duplicate 100ml Samples 113
5.25 Effect of Second Replicate 100ml Sample on Fraction of Sites
Positive and Percent of Sites with 4 or more Coliforms
per 100ml 114
5.26 Comparison of 100ml and 200ml Samples 115
5.27 Expected Results of Sampling from Lognormal Distributions 118
5.28 Expected Numbers of Counts in Samples from Middle Range
Coliform Density Sites 119
5.29 Expected Results of Sampling from Lognormal Distributions 120
5.30 Effect of Sample Volume on Fraction of Sites with Coliforms 122
5.31 Effect of Sample Volume on Fraction of Sites with Coliforms 123
6.1 Structural Analysis rf Systems Sampled 129
6.2 Summary of Structural Parameters by Section Configuration 130
6.3 Dimensions of Distribution System Pipes by Section 133
6.4 Analysis of System CV Bacteriological Water Quality Parameters
by Section 135
6.5 Analysis of System WH Bacteriological Water Quality Parameters
by Section 136
6.6 Analysis of System LB Bacteriological Water Quality Parameters
by Section 137
6.7 Analysis of System DT Bacteriological Water Quality Parameters
by Section 138
6.8 Analysis of System BL Bacteriological Water Quality Parameters
by Section 139
6.9 Analysis of System SR Bacteriological Water Quality Parameters
by Section 139
6.10 Contingency Table Analyses for Differences among Sections 141
6.11 Analysis of System LB Data for June 1979 142
6.12 Analysis of System WH Data for May-June 1981 144
6.13 Comparison of Sampling Results (100ml Samples) from the Two
Locations of System MW 146
6.14 Comparison of Sampling Results (100ml Samples) from 68
Sampling Locations of System BL 147
6.15 Comparison of Sampling Results (100ml Samples) from 18
Sampling Locations of System SR 149
xii
-------
Number Page
6.16 Locations of Positive Coliform Samples Collected from System
BL - June 1981 152
6.17 Peripheral Versus Other Locations 153
6.18 Analysis of Coliform Occurrence as a Function of Distance from
the Treatment Plant System BL - June 1981 Data ISA
6.19 Analysis of Coliform Occurrence as a Function of Distance from
the Treatment Plant System WH - May-June 1981 155
6.20 Analysis of Coliform Occurrence as a Function of Distance from
the Treatment Plant System LB 155
xiii
-------
LIST OF ABBREVIATIONS AND SYMBOLS
ABBREVIATIONS
BB — Brant Beach service area of the Long Beach Water Co. distribution
system
BG — Bradford Glen water system
BL — Brooklawn water system
CFU — colony forming units
CV — Coatesville water system
DT — Downingtown water system
gpd — gallons per day
LB — Long Beach Water Co. water system
MCL — maximum contaminant level
MGD — million gallons per day
MI — Mount Idy water system
MW — Marshallton Woods water system
SR — Spring Run water system
TNTC — too numerous to count
TP — Terrace Plant service area of the Long Beach Water Co. distribution
system
WH — Woodbury Heights water system
SYMBOLS
E>max Statistic used for Kolmogorov-Smirnov test
D^! Fisher's index of dispersion
e Base of natural logarithms
F Statistic used for regression analysis
GM Geometric mean
GSD Geometric standard deviation
GSE Geometric standard error
k Parameter of negative binomial distribution
n Number of samples
P2 Probability
R Coefficient of Determination
S- Standard deviation
S Variance
S Geometric standard deviation
X° 'Coliform count of a single sample
3t Mean coliform count of a set of samples
X Geometric mean coliform count
Of
Z Statistic used for normal distribution
xiv
-------
SYMBOLS (con't)
X
a
y
Arithmetic mean calculated from parameters of lognormal distribution
Arithmetic mean calculated from parameters of lognormal distribution
estimated by graphical technique
Arithmetic mean calculated from parameters of lognormal distribution
estimated by regression
Fraction of water not contaminated with coliform bacteria
Statistic used for goodness-of-fit and contingency table analyses
Variance
Mean
xv
-------
ACKNOWLEDGMENTS
The basic questions to which this project was directed were formulated
by Joseph A. Cotruvo and Charles W. Hendricks, Office of Drinking Water,
U.S. Environmental Protection Agency. Dr. Hendricks provided guidance
as Project Officer during the first year of the project and maintained his
scientific interest in the research after his move to the Office of Research
and Development. Paul Berger, Office of Drinking Water and E. E. Geldreich,
Municipal Environmental Research Laboratory served as Project Officers
during the last two years of the project and made significant intellectual
contributions.
Several individuals in the Pennsylvania Department of Environmental
Resources and the New Jersey Department of Environmental Protection provided
information which was essential for the identification and selection of
systems for sampling. We would also like to express appreciation to the
City of Coatesville, Pennsylvania, Woodbury Heights Township, New Jersey,
The Long Beach Water Company, Downingtown Borough, and Brooklawn Township,
New Jersey for their cooperation in the sampling programs.
Harvey A. Minnigh supervised the sampling programs and laboratory
analyses for the last two years of the project. Edward M. Podgorski was
responsible for the laboratory studies during the first year of the project.
Marc Goshko, Eric D. Becker, James Sioma, Gary Burlingame, Marlene Troy
and others participated in the data collection and analysis. Mrs. Sandra L.
Sees typed and retyped numerous reports and manuscripts and was an unfailing
source of cheerful encouragement.
xvi
-------
SECTION 1
INTRODUCTION
1.1 OBJECTIVES OF THE PROJECT
The overall objective of the project was to obtain data and information
for use in developing a scientifically verifiable rationale for specifying
monitoring programs for water systems of different sizes and types which can
be used to determine compliance with the maximum microbiological contaminant
level (MCL) of the National Interim Primary Drinking Water Regulations.
Among the items specified in the Regulations are 1) the sampling frequency in
terms of number of samples per month as a function of population served, 2)
the standard sample, and 3) a statement that the samples shall be taken at
points which are representative of conditions in the distribution system.
The primary goal of the project was to develop a sampling model which
could be used for revision of the sampling frequency table. The data base
for the sampling model is the results from short term sampling of nine small,
community water systems. The collection of field sampling data for the data
base was limited by physical considerations to small communities within
about 75 miles of Philadelphia. There may be special water quality conditions
which occur in this region and which are not of any great significance in
other regions or there may be phenomena occurring in other regions which are
not pertinent here. Thus, the sampling model needs to be verified for other
regions of the country.
The data for model development were collected using the membrane filter
method for determining coliform densities of water samples. No information
on the multiple tube technique are presented in this report.
1.2 APPROACHES
Two approaches to the construction of a sampling model were used. One,
which we call the empirical approach, is based on fitting various data sets
to frequency distributions. The frequency distributions are then used to
predict the probabilities of obtaining various results with smaller numbers
of samples. This approach is explored in Section 5.
The other, which is called the mechanistic approach, depends upon postu-
lating mechanisms by which the coliform bacteria are despersed in the distri-
bution system. If the mechanism of contamination and the details of flow and
mixing in the distribution system are known, it should be possible to predict
where the coliforms would be found and possibly the frequency distribution
for the coliform counts. This approach is explored in Section 6.
-------
The use of either approach to the construction of a sampling model re-
quires accurate sampling data for calibration. It is necessary to avoid
erroneous results; that is, coliform bacteria which were not actually in the
distribution system showing up in the laboratory tests or coliform bacteria
in the distribution system not showing up as coliform bacteria in the labora-
tory analyses. The special procedures which were used for sample collection
and handling are described in Section 3 and questions about sample validity
are addressed in Section 4.
1.3 OBJECTIVES OF MONITORING
The overall objective of microbiological monitoring of water distribu-
tion systems is related to the protection of public health, especially to
the prevention of the spread of waterborne diseases. Historically, particu-
larly in the nineteenth century, there were many major epidemics of diseases
transmitted by water contaminated by human wastes. Prevention of waterborne
disease depends upon multiple barriers; i.e., selection of the waters of the
highest quality available for public supply, protection of the water source,
treatment to remove turbidity particles, disinfection to kill pathogens,
maintaining a residual concentration of disinfectant in the distribution
system, and protection of the physical integrity of the distribution system
against cross connections and leaks into the system.
The most difficult barrier to maintain, and the one whose penetration
poses the greatest threat of waterborne disease, is the physical integrity of
the distribution system. All water distribution systems leak and potential
cross connections are not uncommon. Any water which leaks into the system
may contain pathogenic microorganisms. Positive pressure in the distribution
system should assure that the leakage is always out of the system but pressure
variations due to differences in elevation, very high rates of water use for
fighting fires, excessively high demand during hot weather, and accidental
breaks can result in significantly reduced water pressure and enhance the risk
of contaminated water leaking into the system.
Outbreaks of waterborne disease are infrequent in any one system but
twenty to thirty such outbreaks are reported in the United States every year.
Thus, it appears that the multiple barrier approach as presently practiced
is effective but not absolute. Microbiological monitoring of the product
delivered to the consumer is one method of directing attention to the neces-
sity of constant attention and maintenance of the barriers.
Harrison (1978) made the distinction between the safety of an individ-
ual portion of the water and the reliability of the water system. We contend
that the routine microbiological monitoring of water distribution systems as
required by the National Interim Primary Drinking Water Regulations is direct-
ed to demonstrating the reliability of the overall system and that the safety
of individual portions of the water is not an issue.
1.3.1 Safety.
Although the total coliform group is not an infallible indicator, the
demonstration of the absence of coliform bacteria from some portion of water
-------
from a distribution system does provide some reasonable assurance of the
safety of that water (in terms of a low probability of transmission of water-
borne infectious disease). .Unfortunately, it is impractical to test any
significant fraction of the water which will be consumed by humans to achieve
this reasonable assurance of safety. It is quicker, cheaper, and more effec-
tive to boil and cool water before consumption to achieve the assurance of
safety than to to monitor for microorganisms and that is exactly what is
done for particularly susceptable individuals such as young infants and
those with chronic, debilitating diseases.
As an example of the safety issue, consider a city block, 400 feet long
with a 6" diameter main under the street. At any time, there are about 588
gallons or 2224 liters of water in the main. Suppose that on a given day
samples are collected from two residences at either end of the block and that
the samples each have a volume of 250ml. In the laboratory the samples are
well shaken and 100ml of each is filtered. If all goes well, there will be a
demonstration 22 hours later that no coliforms were present in either sample.
(The possibility that 1 or 2 coliforms might have been present in the 250ml
of the sample but were not included in the 100ml which were filtered has to
be admitted but is of little consequence.) This does not provide proof that
there were no coliforms in the other 2223.5 liters of water in the main
between the two sampling points and, even if it did, by the time the coliform
results are available a different 2224 liters of water is in the main. The
water in the samples tested is never consumed and the water which is consumed
is never tested; thus, the safety of the water consumed is never demonstrated.
The sampling frequency assumed in this example is extreme. System WH
(Appendix B) served a population of 3600 distributed over 115 blocks from
which samples could be obtained. Four samples per month were examined for
routine monitoring. If the sites to be sampled were selected randomly, each
block would be sampled once every 29 months on the average rather than the
two samples in one day of the example. Actually, some blocks are never
sampled and some blocks are sampled more frequently than once every 29 months.
In any case, it is clear that confidence can be placed in the results of one
sample from a given block every 2 or 3 years only if the barriers against
microbial contamination are known to be reliable.
1.3.2 Reliability.
The primary objective of a microbiological monitoring program is to
demonstrate the reliability of multiple barriers against transmission of
waterborne disease. This reliability is not absolute. It must be admitted
that for any water supply there will always be conditions under which the
barriers will be penetrated. The reliability of the system needs to be
measured quantitatively in order to determine the extent of protection of
public health.
The techniques available for testing for microorganisms require that
discrete volumes of water be used. The standard sample for the membrane fil-
ter techniques is 100ml at present. Thus, the potential number of samples is
all 100ml volumes of water which pass through the distribution system. For
a one million gallon per day (MGD) system, this would amount to 37,800,000
-------
potential samples per day or 1.134x10^ samples per month. The potential
number of samples which could be tested is, for all practical purposes, in-
finite in relation to the actual number of samples tested. The question to
be considered is what inferences can be drawn about the reliability of the
system from the few samples which are collected and tested.
Reliability is a characteristic which is related to persistence in time
and over the area served by the distribution system. A distribution system
which is free of coliforms one day but has many the next would not be consid-
ered to be reliably protected. Likewise, a system free of coliforms in some
areas but with many of them in other areas would not be considered to be
reliably protected. Thus, in monitoring to demonstrate reliability of protec-
tion the sampling must extend over time and over the entire area covered by
the distribution system. The need for the monitoring program to be represen-
tative of the entire area of the distribution system is obvious. The question
of time persistence of reliability needs additional consideration.
As an example of the reliability question, consider a system supplying
about 450,000 gallons per day (gpd) to a population of 4500. Under present
regulations, 5 samples per month would be examined for microbiological
contamination. Assume that the water system has certain characteristics of
water source, treatment, and the nature of the distribution system such that,
over some relatively long period of time, some fraction of the 100ml samples
will have coliforms present. For any month, if all 5 samples have no coli-
forms, it can be said that we are 95% confident that the fraction of the
100ml samples with coliforms is less than one half (0.5) for that month (see
section 5.3.2). However, if all of the 60 samples collected in a year's
time have no coliforms, it can be said that we are 95% confident that the
fraction of 100ml samples with coliforms is less than one twentieth (0.05)
for that year, and if no coliforms are found in the 300 samples collected
over a five year period, it can be said that we are 95% confident that the
fraction of 100ml samples with coliforms is less than one hundreth (0.01)
over the five year period.
The assumptions used for this example need to be examined carefully.
The accumulation of negative coliform results over a period of time developes
more and more confidence that the barriers to microbial contamination of the
water are reliable. However, it is not clear how long a period of time the
characteristics of the system remain the same so that data may be accumulated
to demonstrate reliability of protection. The coliform density in a water
distribution system can change from one day to the next and almost certainly
will change over the seasons of a year. These changes can be considered to
be normal variability resulting from the characteristics of the water system.
The characteristics of the system may change when there are alterations in
the water source, treatment processes, or distribution system. Examples of
these alterations would include changes in water treatment processes, placing
new water mains in service, failure and replacement of existing water mains,
storage tank painting and repairs, and abnormal variations in water pressure.
These are identifiable occurrences and can be used to separate groups of data
used to demonstrate reliability of the system.
Even in the absence of identifiable changes in the water system, it may
4
-------
be expected that there will be slow deterioration, particularly of the dis-
tribution system, which would change the reliability of protection against
microbial contamination. In the preceding example, time periods of one month,
one year, and five years were used. It seems likely that the characteristics
of a water system would remain consistent for more than one month but prob-
ably not for as long as five years. It seems reasonable to evaluate the
reliability of protection of a water supply against microbiological contami-
nation on the basis of data collected over one or two year periods in the
absence of identifiable occurrences which could change the characteristics
of the system.
There is no hard evidence to show that the reliability of protection of
a water system should be evaluated using data collected over a one year or
two year or five year period. In this project, water systems were sampled
over one or two month periods and the data collected shed no light on this
question. There is, however, a precedent for evaluating the reliability of
protection over a one year period under the previous drinking water standards
of the U.S. Public Health Service (1962). Water supplies had "approval" if
they met the coliform standards all 12 months of a year, "provisional approv-
al" if the standards were violated only once in 12 months, and were "use
prohibited" if the standards were violated any two months in a year's time.
Clearly, averaging coliform results over a month's time is a matter of
convenience for reporting and has no particular scientific basis. However,
there is no evidence that hourly, daily, weekly or quarterly averaging and
reporting would give a superior measure of coliform occurrence and densities
in the system. Since the monitoring and reporting system is run by people
for the protection of people, it is appropriate to select the reporting
period as a matter of convenience in the absence of scientific evidence that
one reporting period would be superior to any other. The same argument can
be made in respect to the period of time over which the reliability of the
system should be evaluated.
-------
SECTION 2
CONCLUSIONS AND RECOMMENDATIONS
1. The reliability of any water system in providing properly treated and
disinfected water is never absolute. The objective of microbiological
monitoring of a water distribution system is to provide a quantitative
measure of this reliability which can only be expressed as a probability
function.
2. Contamination of a water sample with coliforms during or subsequent to
collection can be prevented or recognized by a proper quality assurance
program.
3. When coliforms are present in a water sample collected from a distribu-
tion system, their density can change within twenty-four hours if the
sample is held that long before examination. Thus, there may be a
transit time problem using currently accepted procedures.
4. Coliforms were found in all nine community water systems studied. The
most common verified species were Enterobacter cloacae, Klebsiella pneu-
moniae, and Klebsiella oxytoca. Escherichia coli was found only rarely
and only in two sys terns.
5. Most of the presumptive coliform colonies observed on membrane filters
which did not verify were strains of Enterobacter agglomerans. However,
a sucrose positive strain of this species which did produce gas in
lauryl tryptose and brilliant green bile broths was encountered. This
strain has the same IMViC type as Escherichia coli.
6.- Coliform counts from all systems studied could be fitted to either the
truncated lognormal or the negative binomial distribution. This demon-
strates that coliforms are not randomly dispersed in water distribution
systems. The variance of the counts was much greater than the mean.
7. We recommend the truncated lognormal distribution for modeling coliform
counts in samples from a distribution system because some water works
personnel are already familiar with it and the estimation of parameters
is mathematically simpler.
8. Even when the mean coliform density is greater than 1 per 100ml, the
probability that a 100ml sample will have no coliforms present is very
high. This is a prediction from the truncated lognormal distribution
which was very well supported by the data collected.
-------
9. Most of the systems studied had .coliform counts with geometric means in
the range 10~1 to 10~4 per 100ml and geometric standard deviations be-
tween 10 and 100. The average arithmetic coliform counts in the samples
collected from a system ranged from 0.1 to > 9.4 per 100ml.
10. The number of samples per month required for detection of coliforms de-
pends more on the geometric standard deviation than on the geometric
mean density; more samples are required as the geometric standard
deviation increases.
11. We recommend a minimum of six samples per month for microbiological
monitoring of a water distribution system based on the following
assumptions:
a. A true arithmetic mean coliform density of ^ 1 per 100ml
indicates adequate reliability of protection.
b. If a water system is found to violate the microbiological MCL
more frequently than once a year, action will be taken to im-
prove disinfection and/or other protection of the system.
c. If a water system is found to violate the microbiological MCL
once a year or less, no remedial action will be taken.
d. The geometric standard deviation of the coliform distribution
will be 100 or less.
12. Collection of two replicate 100 ml samples per site sampled increased
the fraction of the sites where coliforms were found by about 50%.
This suggests that increasing the volume of the standard sample to
200 ml would increase the detection of coliforms when they are present
in densities of 5 to 10 per liter.
13. If the true arithmetic mean coliform density is >_ 1 per 100ml and the
geometric standard deviation is _<_ 100, violation of the microbiological
MCL will occur twice a year if 6 samples per month are taken.
14. A water distribution system can be divided into sections which are
hydraulically isolated; that is, coliforms may be present in one section
and not be detected by samples collected from other sections. Signifi-
cance differences in the frequency of coliform occurrence may sometimes
be found among the isolated sections of one distribution system.
15, No significant differences were found in the frequency of coliform
occurrence between peripheral and nonperipheral sampling locations.
The frequency of coliform occurrence was not found to increase with
distance from the water source. Thus, it is recommended that the sam-
pling sites for a microbiological monitoring program be selected by a
randomization procedure and that all parts of the distribution system
be included.
-------
SECTION 3
METHODS AND PROCEDURES
3.1 INTRODUCTION
In this section we define the variables accounted for within our sam-
pling program and describe our protocols for sample collection, transit,
analysis, quality assurance and data management. Information on the Woodbury
Heights, New Jersey water distribution system and some of the data collected
from that system are used to illustrate certain points. Information on and
summaries of the data collected from the other systems are provided elsewhere
including Sections 5 and 6 and the Appendices.
The Regulations require (U.S.E.P.A. 1976) that samples be collected at
points representative of the conditions in the system. The term "represen-
tative" may be interpreted in several different ways, each of which has some
validity depending upon the objectives of the monitoring program. One pur-
pose of this project has been to define what "representative" should mean
for water distribution systems in general, and to provide examples of how
sampling programs can be designed to meet specific obectives. The sampling
conducted for this project has had two major objectives: 1) to obtain random
samples to provide data for fitting frequency distributions and 2) to test
certain hypotheses about the sources, persistence, and transportation of
bacteriological contamination of a water distribution system. These are not
necessarily the objectives of a monitoring program, but the methodology used
in the development of these programs can be applied to the development of
monitoring programs once the objectives of such programs have been specifi-
cally defined. The objectives of a monitoring program to meet federal or
state regulations are discussed in Section 1.
3.2 PLANNING A SAMPLING PROGRAM
We found it necessary to devise a method of designating exactly where
a sample was collected and describing the real relationship among several
samples in order to achieve the objectives of this project. This was
accomplished by subdividing a distribution system into sections, locations,
services and taps to form a hierarchy as described immediately hereafter.
This method may have some utility for planning monitoring programs.
3.2.1 Hierarchy of Sampling Units
The physical units of sampling may be considered in a hierarchy from
largest to smallest as: water system, distribution system, section, location,
service, and tap. A water system as considered here meets the definition in
8
-------
Section 141.2 (e) of the Regulations (U.S.E.P.A. 1976) and as such it in-
cludes "... any collection, treatment, storage, and distribution facilities
...." That part of the water system, after collection, treatment, and storage
of water, that is used to. distribute water is considered the distribution
system.
The definition of sections of a distribution system is more subjective.
A section is defined as a portion in which the pattern of water flow is
isolated to some degree from that of other sections or in which the inflow
and outflow of water to and from the section are limited. The terms "to some
degree" and "limited" are subjective. The desired result of delineating a
section is to have a portion of the distribution system in which recognition
of contamination within the section has potential importance to more than one
location but less than the distribution system as a whole. A section is
delineated from the patterns of mains and streets. We have assigned
configuration types to sections, described in Section 6.
The pipes which make up a distribution system may be designated as
transmission mains, distribution mains, and street laterals. Transmission
mains convey water from the treatment facility to the sections of the
distribution system and typically are the largest pipes of the system. A
distribution main delivers water to street laterals within a section and to
other sections. Street laterals deliver water to service connections and
sometimes to other street laterals.
In this project, a sampling location was defined as a pipe between two
connections supplying water to one or more service laterals. A location
could be either a distribution main or a street lateral. Usually, a location
corresponded to a city block with the main or lateral located under the
street, between the connections under street intersections, and services on
either side of the street receiving water from the same location. A joint in
a distribution system is where three or more street laterals or distribution
mains are joined and indicates where locations are separated. Occasionally,
it was found that a street lateral served two or more blocks so that a sam-
pling location was more than one block or that two street laterals served
the opposite sides of the street in one block making that block two sampling
locations.
A service is any building receiving water from the distribution system
within a location. A tap is a faucet within a service from which water may
be drawn. Fire hydrants represent another way of obtaining samples, but
cannot be described simply as a service or tap.
The public water system of Woodbury Heights, New Jersey is used as an
example of this hierarchial subdivision. Figure 3.1 is a map of the town.
The numbers denote locations. Raw water is obtained and treated at a well at
location 101 and stored in a standpipe between locations 111 and 112. Five
sections have been delineated each containing between 16 and 31 locations.
The sections are shown in Table 3.1. Each location contains at least 1
service, and a service may contain several taps. Thus, a location may be
sampled at several discrete points within it.
-------
o.
§
•H
4-1
fl)
O
60
C
to
CO
4J
60
w
I
,n
T3
O
I
oo
0)
1-1
10
-------
This hierarchy of sampling units is helpful for defining representative
samples. Samples taken within the treatment facility cannot be considered
representative of the water distribution system. Results from these samples
only provide information on the bacteriological quality of the source water
and the efficacy of treatment. This is but one possible source of contamina-
tion to the system as a whole.
TABLE 3.1
SECTIONS OF WOODBURY HEIGHTS, N.J.
Name Locations included No. of locations
East
East central
West central
Southwest
North
101-119
120-132, 218
133-162
163-188
189-217
22
16
31
26
30
Contamination may arise from any point in the distribution system and
may occur in one section without occurring in the others. "Representative
samples" should represent the different sections. If contamination is found
in a section, follow up monitoring may focus on the section from which the
contamination was found. The configurations of sections differ as described
in Section 6. A particular configuration may be more prone to contamination
than another. Information of this sort may be useful in not only designing
a monitoring program, but also designing the distribution system itself.
The basic unit of sampling used for our study was the location. There
are several advantages to this. The number of locations within a system is
often large enough to be considered for application of random sampling, and
yet small to allow for adequate data management. For very small systems, too
few locations may be available for random sampling. Sampling of residential
areas requires that someone be home and give permission for a sample to be
collected. As a location often contains several services there is a better
probability of obtaining a sample from a particular location than from a
particular service. Also, multiple services may be sampled from a location
if information on reproducibility is desired. Consideration must be given
for check sampling, however. The service and tap sampled within a location
must be noted if check sampling is to be conducted according to the current
regulations (Section 141.21(d)(3 and 4)).
A service may have several taps from which samples may be obtained.
The location of the tap may be important in consideration of subsequent
contamination (Section 4). Also, only certain taps may be made accessible
to the sampler. We have found it easier to gain permission to sample outside
taps than those inside. From the point of view of monitoring the water in
the distribution system per se, the tap location within a service should have
no bearing aside from that of possible subsequent contamination. However,
11
-------
the tap location may be of importance if intraservice cross connections are
being considered.
The designation of a section, location, or service is often based on
the visible layout of the municipality rather than the pipe layout. Blue-
prints of pipe layout may be available, but their availability is sometimes
limited. Maps of old distribution systems may be lost; new distribution and
recently modified systems may lack adequate blueprints. One can use street
layout as a first approximation of pipe layout. Exceptions exist, however,
and there is no true substitute for accurate blueprints in designating the
units of sampling.
Another consideration in sampling is time which can be considered in a
hierarchial sense. The unit of time designated in the Regulations (U.S.E.P.
A. 1976) is the month. In evaluating the sustained bacteriological quality
of a system, however, time periods of many months may need to be considered.
Seasonal changes in bacteriological quality may occur or the ability of a
system to meet the coliform MCL's when contamination is continuous but at low
densities or when sample size is small may best be evaluated over several
months (Pipes and Christian 1978).
In theory sampling should occur randomly through time including night
and day since water is supplied continuously and contimination may occur at
any time. In practice this is not feasible. Systems requiring one or a few
samples per month may be sampled only one or at most a few days. The times
of the day or days of the week at which one has access to services or to
personnel is limited. If a limited number of samples is taken and the
samples are distributed over several days, this time may detract from efforts
to obtain adequate sampling of different sections. The sampling pattern
must be a compromise between evaluation of space and time differences in
bacteriological water quality.
3.2.2 Random Sampling
At any time there is a very large (for all practical purposes, infinite)
number of 100ml samples of water which could be collected from a water dis-
tribution system. An "ideal" experiment would be to drain the system instan-
taneously, examine each of the 100ml portions of water before the bacterial
densities could change, and count each coliform organism in each sample.
The results of this experiment would give the actual frequency distribution
of coliform counts per 100ml. However, the water distribution system would
be empty and the frequency distribution found would not necessarily describe
the dispersion of coliform bacteria in the system once it were filled again.
The "ideal" experiment is clearly impossible; the frequency distribution
describing the dispersion of coliforms in a water distribution system must be
estimated from sampling data. The most important characteristic which the
data used for fitting frequency distributions should have is that any 100ml
portion of the water should have the same probability of being collected as
any other 100ml portion of the water. This is the characteristic of a random
sample but it is not possible, for practical reasons, to collect such a set
of samples from a water distribution system. It is only possible to approxi-
12
-------
mate random sampling and to understand the extent of deviation from random-
ness. The two variables considered in the randomization process are location
and time.
At any given time, not all of the water in a distribution system is
equally accessible. Samples can only be obtained from taps on the system.
The sampling protocol calls for allowing the tap to run for at least 3 min-
utes to flush out the service lateral before the sample is taken. The sam-
pling procedure is completed in less than 10 minutes. Thus, the part of the
water which is accessible at any given time is the water which is within 3
to 10 minutes of time-of-flow from a tap.
Consider a residential service on block number 147 (Beech Avenue between
Third Street and Fourth Street) on the sampling location map of Woodbury
Heights, New Jersey (figure 3.1). The water main for that block is 6 inches
in diameter, 700 feet long and contains approximately 1030 gallons of water
or about 1.5 gallons per foot. There are seven houses on the north side of
the street and eight on the south side or 15 potential sampling sites. If
a service lateral is 50' of 3/4" pipe and in the house there are 30' of 1/2"
pipe from the water meter to the tap to be sampled, then there is slightly
more than one gallon of water between the street lateral and the tap. If the
tap is allowed to run for 3 minutes at about 1 gallon per minute the service
lateral and pipes in the building are flushed out three times but the water
in the street lateral moves forward only about 2 feet (assuming no other
water usage). If the tap is run for 10 minutes at one gallon per minute, the
water moves forward in the main about 7 feet. Since there are, on the aver-
age 46.7 feet between service laterals, some fraction between 0.04 and 0.15
of the water in the main is accessible for sampling at a particular service.
From this example it is clear that any tap on a particular service should be
equivalent to any other tap on the same service in terms of the quality of
water obtained for a sample but that each service on the block represents a
different portion of the water in the main for sampling purposes. When this
example is worked out for different pipe sizes, block lengths, service con-
nections per block, etc. the numbers change somewhat, but in the context of
what is considered to be acceptable, modern water supply practice the con-
clusions remain the same.
The best way to obtain a quasi-random sample would be to number each
service connection on the system and select the service connections to be
sampled from a table of random numbers. Unfortunately, this is impractical
as mentioned previously because at the time of sampling there may be no one
at home in a particular residence to provide access to a tap. Even when
someone is at home, entrance to the residence for the purpose of sampling may
be refused. Thus we adopted the street lateral as the location, numbering
each as a potential sampling location and selecting location numbers for
sampling from a table of random numbers. Then any service is considered to
represent a sample from that street lateral. Even this approach does not
always work out because it may not be possible to gain access to any of the
services for sampling purposes.
In most cases (except for dead ends), it is expected that the water in
a main will be flowing rather than still. Thus, a sample from a residence
13
-------
represents extracting a small fraction of the flowing water rather than ex-
traction from a static body of water. Although there is very little longitu-
dinal mixing of water flowing in a pipe, the fact that the water is flowing
means that a larger fraction of the water in the system is potentially acces-
sible for sampling.
Usually, we attempted to obtain samples from a tap in the yard because
this avoided the need to enter a house. Of course, this was not possible in
the winter and some home owners preferred that the samples be collected in-
side. The second most frequently sampled tap was the kitchen sink. Other
taps which have been sampled include the bathroom lavatory, laundry room
sink, and utility sink, (usually in the basement). Because the service
lateral was flushed out several times before the samples were collected,
there should have been no difference in water quality in samples collected
at the different taps.
Since water is flowing in the distribution system, the coliform density
in any part of the system may be expected to be changing continuously. To
collect a random sample from the distribution system it would be necessary to
randomize the time of sample collection as well as the locations of the
samples. However, it would be futile to ring door bells at 3:00 am to ask
permission to obtain water samples. As a practical matter, the time of sam-
ple collection for this project, as it would be for routine monitoring pro-
grams, was even more restricted by the necessity of processing the samples
within a relatively short period of time after collection and the necessity
of providing reasonable working hours for the samplers and laboratory person-
nel. Thus, most samples were collected on weekday mornings.
The bacteriological quality of water in a distribution system can vary
with time. A major question for planning sampling programs is how rapidly
and how often it changes. If the bacteriological quality of water in a dis-
tribution system remains stable for several days, data collected over that
period can be lumped together for fitting frequency distributions. Changes
in bacteriological quality would be recognized by significant changes in
parameters such as fraction of sampling locations yielding coliforms, mean
coliform density, etc. One of the results of this project has been informa-
tion on temporal changes in the bacteriological quality of water in a distri-
bution system ^and the reasonableness of calculating statistical parameters
from data collected on several different days.
Data from random samples are needed for fitting to frequency distribu-
tions. It is impossible, in an absolute sense, to randomize collection of
samples from a water distribution system because 1) not all of the water in
the system is equally accessible at any time and 2) it is impractical to
collect samples at certain times of the day. The best that can be done is
to avoid known sources of bias and to test for changes in the parameters
from day to day. This is more of a conceptual than a real problem. The
objective of the project was to develop a model which can be used to specify
the parameters of a routine monitoring program. As long as the data used
for model development are comparable with data obtained from routine monitor-
ing programs, the model should be valid.
14
-------
3.2.3 Stratified Random Sampling
"Representative" sampling usually means selecting certain groups of
samples to represent characteristics of the population being sampled. The
division of the possible samples into groups is called "stratification".
The water distribution systems sampled have been stratified by dividing
them into sections. Sections were selected insofar as possible to be isolat-
ed from other sections, but some of the systems sampled have had central por-
tions which were not isolated, and the system LB had no isolated sections.
The division of a large interconnected grid into sections has to be somewhat
arbitrary and based on convenience of sampling.
The number of sections should be small in comparison with the number of
samples collected during any time period. If the number of sections were
equal to the number of samples collected, then there would be one sample from
each section. This would be regular rather than random sampling. There have
to be several samples per section for the requirements of a stratified random
sampling pattern to be met.
The purpose of stratified random sampling in this project was to deter-
mine if there were differences in bacteriological quality among the different
parts of a water distribution system. If such differences were found, a
monitoring program would have to be designed somewhat differently from a
program for a homogeneous system.
3.2.4 Other Sampling Patterns
One hypothesis which was proposed before the project started (Pipes and
Christian, 1978) was that if there were a point source of bacteriological
contamination such as occurs in the case of a cross connection, samples
downstream of a positive sample would also be positive and all samples up-
stream of a negative sample would also be negative. Testing such a hypoth-
esis required a linear sampling pattern along particular mains and the labo-
ratory physical model.
Another hypothesis proposed before the project started (Pipes and
Christian, 1978) is that the probability of finding coliform bacteria in a
sample increases with distance from the water source and the greatest proba-
bility occurs at the periphery of the system. Testing the first part of the
hypothesis required a sample stratified according to distance from the water
source. Testing the second part of the hypothesis required a peripheral
sampling pattern; that is, a set of samples collected at peripheral locations
and a comparable set of samples collected at other locations. For the pur-
pose of this project a peripheral location was defined as a street lateral
which supplied water only to the service connections on that block. A dead
end street lateral clearly meets this definition. A loop consists of two
dead ends connected together; that is, two street laterals which, if not
connected, would not influence the flow in any other parts of the system.
Once these definitions are understood, it is clear that peripheral location
can occur anywhere in the system, not just at the edges, and they can be
close to the water source as well as distant from it.
15
-------
3.2.5 Application to Monitoring Programs
It is unlikely that any program for routine monitoring of a water dis-
tribution system would be set up for random sampling. The randomization of
sampling locations which was used for acquisition of some of our data sets
approximated random sampling as closely as is possible given the physical
limitations of a typical water distribution system. There is no reason to
believe that these physical limitations seriously biased any data sets so
they have been used for fitting frequency distributions. The frequency
distributions selected have been used for evaluation of various types of
sampling programs which might be used by water departments.
The term "representative locations" in the Regulations infers a strati-
fied random sampling program. It is clear both from the Regulations and
from our sampling experience that there should not be areas of a water dis-
tribution system which are never sampled. Recognition of sections will aid
in preventing this. Most of the positive samples collected from system CV
were from areas of the system which are never sampled by the Water Department
samplers. The sampling locations should be changed from month to month so
that all subsections of the system are covered over a period of time. Other-
wise a serious problem in one part of the system could go undetected for
long periods of time.
In order to cover all sections of a water distribution system it is
necessary to obtain samples from private residences as well as from public
and commercial buildings. Some water departments avoid sampling private
residences. Our experience is that sampling private residences is slower than
sampling public and commercial buildings because in many cases there is no
one home and in some cases entrance to the residence is refused. However, it
is possible to obtain large numbers of samples from private residences if
adequate effort is made. Peripheral locations as well as other parts of the
system should be sampled.
The goal of a bacteriological monitoring program should be to achieve
the highest probability of detecting any bacterial contamination of the
distribution system possible under the resource constraints of the program.
This section has provided qualitative information on the design of a sampling
program. Data and the results of tests of hypotheses about quantitative
aspects are provided in Section 6.
3.3 COLLECTION AND ANALYSIS OF SAMPLES
For the development of procedures and protocols for this project, the
primary references used were two EPA Manuals (Geldreich 1976 and Bordner, et
al. 1978) and Standard Methods, 14th Ed. (APHA 1975). The methods of
sampling and sample analysis were not major subjects of investigation. Well
accepted procedures which have been developed and tested by others over a
period of years were used. It would be a waste of space to summarize those
procedures from the reference documents. Only a brief description of our
specific adaptions of the accepted procedures which have been made is given
here.
16
-------
A Quality Assurance Manual has been written for the project giving de-
tailed descriptions of the procedures being used for collection, transporta-
tion, handling, and analysis of samples. Copies of that Manual are available
upon request.
3.3.1 Collection of Samples
Selection of location, service and tap have been described. Sampling
involved recording information as to the address of the service, location of
tap, time of sampling and sampler. The tap water was allowed to run for no
less than three minutes at which time the temperature of the water and chlo-
rine residual were taken. Chlorine measurements included free and total
chlorine determined by the use of DPD kits (Hach Chemical Co.). Samples for
bacteriological analyses were collected in sterile bottles containing thio-
sulfate for dechlorination. The number of bottles and volume of water per
bottle varied according to desired replication and filtering volume. Volume
ranged from 125m£ to 3.8£, and replicate bottle numbers ranged from 1
to 12 for normal sampling.
The temperature during transit to the laboratory should be < 10°C.
Samples were stored in ice chests with frozen, sealed ice packs whenever
necessary. The time from sampling to processing was less than 8 hr with the
average time between 3 and 4 hr.
3.3.2 Analysis of Samples
The membrane filter procedure was used for coliform analysis of 100ml or
200ml on most samples. Each analyst filtered about 50ml of sterile buffered
rinse water before and after each series of samples using the graduated
cylinder, funnel assembly, and rinse bottle used for processing samples. The
membrane from these rinse water controls were incubated on m-Endo medium in
the normal manner and serve as negative controls. Serial dilutions of cul-
tures of Escherichia coli (ATCC8739) and Enterobacter cloacae (ATCC13047)
were also filtered as positive controls to demonstrate that the Endo medium
produced the typical coliform colonies.
After portions from the samples and negative controls were filtered for
coliform tests, one ml portions were removed for standard plate counts. A
standard plate count was also made on the Endo medium and rinse water from
all rinse bottles used that day to test for contamination. The Endo medium
and the tryptone-glucose-extract agar was made up fresh for each sampling
day. The pH and turbidity determinations were made on each sample after the
portions had been removed for the coliform tests and standard plate counts.
In some instances a very turbid sample left a noticeable deposit of
particulate matter on a membrane filter. Even though it is well known that
such particulate matter can interfere with the development of coliform
colonies, the filter was placed on Endo medium and incubated for 22-24 hours.
If no coliform colonies developed, the filter was placed in lauryl tryptose
broth and incubated for another 48 hrs. If gas was produced in the broth,
this provided presumptive evidence of coliform bacteria. This, of course,
does not give a coliform count for that sample, but the number of samples
17
-------
with coliforms present for any sampling trip is an important parameter to be
evaluated accurately.
For parts of a distribution system where samples routinely contained
large amounts of particulate matter, to preclude this interference from
causing negative errors in the data generated, samples were filtered in
several smaller aliquots to give a total of 100ml. If necessary, these
filters were subjected to the non-quantitative procedure outlined above.
Alternatively, larger volume samples were collected and a portion of the
sample analyzed by the multiple tube fermentation test for comparison with
the membrane filter results. The objective here was to assure that coliforms
were detected if present and to obtain the best quantitative information
possible, not to make a comparison of the efficiency of the two techniques.
Typical coliform colonies on Endo medium are dark with a green metallic
sheen. Some colonies on Endo medium have dark centers with no sheen, and in
some instances it was difficult to judge the presence or absence of the
sheen. In those cases in which the presence or absence of the sheen was a
difficult judgement, the colony was counted as a presumptive coliform and
transferred to lauryl tryptose broth and brilliant green bile broth for ver-
ification. In this way, any doubt about the recognition of coliform colonies
on Endo medium was resolved by testing rather than by opinion. A differen-
tial count of colonies with similar morphology was made, and coliform counts
were adjusted according to the verification results. In addition all typical
coliform colonies on a filter were picked for verification if there were
five or less. If there were more than five coliform colonies on a filter,
the number picked for verification depended on the total number of coliform
colonies on all filters that required verification that day; however, the
number was never less than five. Verification involved growth and gas pro-
duction in lauryl tryptose broth (LTB) and brilliant green bile broth (BGLB),
growth on EMB agar, growth and reactions on triple sugar iron agar (TSI),
and Gram stain.
Dark colonies without a sheen were counted as atypical colonies. Also,
red, and clear colonies were counted. A small number of these colonies were
run through the verification procedure. One or more of the colonies from the
filter with the ATCC organisms were also run through the verification proce-
dure as a positive control.
Some fraction of the isolated verified coliforms and other isolated
colonies from each system were identified using the API 20E system. Supple-
mental macrotube tests were employed when questions arose about identifica-
tion. When few colonies were isolated in a system virtually all colonies
were identified. When many colonies were collected the fraction was reduced
to as low as one half to two thirds the isolates.
Water from each sample was used for standard plate count (SPC) deter-
minations. Aliquots of 1 ml were used for pour plates with Trypticase Glucose
Extract Agar (TGEA). Negative controls at the beginning and end of each
flask of TGEA poured and triplicate samples every tenth sample were prepared
for quality assurance. Plates were incubated at 35°C for 48 hour prior to
counting.
18
-------
The turbidity of each sample was determined with a Hach Model 2100A
turbidimeter. Water was taken periodically for other chemical analyses.
Chemical parameters included pH, alkalinity, hardness, and dissolved ortho-
phosphate, ammonia, oxygen, nitrate, and nitrite concentrations. All proce-
dures were in accordance with Standard Methods (APHA 1975) or the Manual of
Methods for Chemical Analysis of Water and Wastes (U.S.E.P.A. 1974).
3.3.3 Special Quality Assurance Procedures
All laboratory personnel on this project kept individual logs in which
all activities, such as preparation of a batch of medium or collection of
samples and all data collected were recorded. Separate log books were
maintained for the autoclave, hot air oven, pH meter, all incubators, media
preparation, media controls, media performance, and laboratory water quality.
Records of samples collected were kept on field sheets, and the data for each
sampling site were recorded on permanent data sheets in a separate notebook.
Each sample was identified by the sampling location number and replicate
number. On the field sheets, the street address of the building sampled and
the exact location on the tap used for sampling were recorded along with the
water temperature, chlorine residual, and time of sampling. Thus, it was
possible to resample the same tap if needed.
When the samples were being filtered, each analyst recorded, in
chronological order, the identification numbers of the samples filtered. It
was then possible to ascertain which samples were filtered before and after
any given sample. When several positive samples were obtained on a sampling
day, the records were checked to make sure that the positive samples occurred
at random in the sampling order and among the analysts.
Records on all project activities were maintained at least in duplicate
and sometimes in triplicate. It was possible to reconstruct the history of
any sample, batch of medium, etc. from these records if the need arose. In
this way, reliance upon the memories of project personnel was minimized and
all data used for modeling was documented and could be subjected to critical
scrutiny if necessary.
During the grant period we applied for certification equivalence
through Region 3 of the U.S.E.P.A. After site visitation this equivalence
was given.
3.3.4 Data Management
Beyond the maintenance of records described above, much of the informa-
tion was stored in computer files. Two primary files used were the "location
file", identifying the position of each location within the system and its
pipeline characteristics, and the "bacteriological file", containing results
from each sample taken. A representative "location file" and "bacteriologi-
cal file" are shown in Figures 3.2 and 3.3, respectively. From these files
most of our data analysis was done. We employed both the SAS programming
facilities and Fortran IV for these analyses on the IBM 370 computer operated
by UniColl.
19
-------
v U
V cr
a «
1/1 I
3 a
C_JCJ_JQO OOO O Q O OO
a. ULJi/iQ oo uu'UJU.u.iuu^rrz cri-QZa u;»-
-< z >c>>>>>>ocu.c:a:zioff-ix3 a.z:
— z a < c < — .
o »-> oonoo^-k- cooooc1
a: LU eto. QcctatT'^roda. or at: a a. a oe of
Z CJ •- > OO OUlOCJOOCJU-OU-U-C'CUOO CO
bi CO > a£ t- t- 33 CD W1ZZC Q3Z>ifl^ wiirtook-^-ii X trtcoviXt/l «/; Vl
jt i/»ujtuzuJZ2'Zt/itn'JJ*/ii.'ii/ii/it/)^r*^">— »— H-i/it/it/1. K"-oiu*/)t/i
-
§ ,
tu a jiuj>-K-i-i->-ozi/>i/!
I o uj— = 3 3 3 3 z o o 35 > i-
. _ -_
« t- LU <«t<_i<<«r«t<-J3:ixizoQi<<_j-iH^'OQO^*4&r*tiOaocoO4>f^ 4J
5 < s
*-• o
o oooooooooooooooooooincjoooomoooo .1
I OOOOOOOOO'OOOOOO.OO'^ON-OOOOOr-O'^OO '"'
O. fM«nj*O>O^O-CO1O^<»<«o o m uiin iAin in JMO m « mm t -t < m N mm o o m -^ in o IA m o")
-------
O O Z |0 O OUJ
O O Z IO O _l « OOOOO«-«OC'--|-OOOC(T|OGOOOO
i ) v* <-4 •* I
in m in ir>
cdcd •••• «$"»$• *^ »}•
• ' i Q O UJUJ »->— Kf- + + + +
f~f-^«l _J_J u^t/)l/)V) IUUJUJUJ
•^ 3 to to o.a_i_JUjujUJUj ix ct a. a.
ZH SEZ-_j_iaiai aa: aocoooo
wi«^<<_j_)XjC»-t-n«i')iov)r3oxaEXx
^< ^< O Z3 "< ro «r m»i<-o>Oinin-4-H>rsroo ij
••••.••••. .......... ^
OOOOOOOOOOOOOOOCOOOO "^jj
: >H
, ffi
O. _l «I l~ LU lOO OOC'^OOOOmOOOOmOOOOOO ^>
..•••' .....•...< ..... M
OOO-^r-tOOo^moOO—iO-^inOOO 3
r- m ir> o —< ,0
^ »-i IQ
i ' ; . i o
z ! S
O -I of IOOQUJ . »- • ; **
@
O O_JlU4a: lOO OOOOOOOOOOOO»-I«^HOOOOO g
Q£ • >4-l
Q. Z Z Z
rvi
it aeaia 100 oo-—
«* Z ' ,
Q < l~ > IOO/3UJ . t- !
->-o.'—o«i_j oocooooao «ooo»*-oooooo
I ! . i
O.OiUJ|OOOUJ ! ' ,
Q. a: UJ IOO_J—> OOOOO--O-J |U-i--Jt~ OOOOOOO-'JOOOOOOOOOOOO
I > u -J
-
acujo. |l/> Jj
rtlClOO'-<-<-<--OLUJ oc at O O aC, oC O O oc cC O O O O at'cC. o O O O
21
-------
SECTION 4
SAMPLE VALIDITY
4.1 INTRODUCTION
For the purposes of this project, a valid result from a sample is one in
which the coliform count on the membrane filter accurately reflects the
coliform density in the main or street lateral supplying the service sampled.
If it is assumed that the coliforms suspended in the water in the main or
street lateral are carried through the service pipes and out the tap, there
are four different occurrences which can invalidate the results.
Coliforms may be introduced into a sample by non-sterile sample bottles,
washings from the surface of the tap sampled, transfer from the hands of the
sampler, leakage of contaminated water into the sample bottle during trans-
port, or from the glassware used for measuring 100ml of the sample and con-
taining it during filtration. We use the term "subsequent contamination" to
describe all these cases in which coliforras could be introduced into the
water subsequent to its discharge from the service pipes. Our studies of
these occurrences and the steps we took to prevent them are described in
Section 4.2.
Coliform bacteria are living organisms. As such they may die or they
may reproduce. Either death or reproduction of coliforms in a water sample
can occur between the time the sample is collected and when the sample is
filtered. Changes in coliform density due either to death or reproduction
would produce inaccurate results. Our studies of these changes are described
in Section 4.3.
Certain types of particulate matter in the samples or large numbers of
non-coliform bacteria may prevent the growth of coliform bacteria on the
surface of membrane filters to form typical coliform colonies on M-Endo
medium. This project did not include an investigation of these types of
interferences as such. However, evidence of interference due to particulate
matter in a few samples was observed. This is described in Section 4.4 along
with the procedures used to identify its occurrence.
Non-coliform bacteria in the water sample may grow on M-Endo medium and
produce colonies with a green sheen or coliform bacteria may produce colonies
without a sheen. These two occurrences are known respectively as "false
positive results" and "false negative results." False positive results are
eliminated by the confirmation procedure. False negative results are more
difficult to deal with in a routine monitoring program. Our experiences with
both false positives and false negatives are described in Section 4.5.
22
-------
4.2 SUBSEQUENT CONTAMINATION
When faced with a set of coliform data consisting of many zeros and a
few counts such as 31, 2, 17, 1 etc., the person responsible for the system
is often inclined to believe the zeros and to blame the positive results, at
least the high positive results, on bad samples. Excuses such as "the lab
sent over some bad sample bottles," "the tap was dirty," and "the sampler
stuck his finger in the bottle" are frequently heard, particularly if check
samples were negative. This type of thinking can lead to indications of
failure of the disinfection process being ignored; however, it must be admit-
ted that such mistakes can occur and steps should be taken to avoid them.
4.2.1. Sterility of Sample Bottles
Several steps were taken to assure the sterility of sample bottles used
in this project. A log of the temperature and time of autoclaving for each
use of the autoclave was maintained. Some sample bottles in each batch
autoclaved were labeled with indicator tape on which the word "STERILE"
appeared when there was adequate exposure to sterilizing temperature. Every
three months the effectiveness of the autoclave was checked by autoclaving
a strip of Bacillus spores and then incubating it in trypticase soy broth to
obtain evidence that all of the spores were killed. With each batch of sam-
ple bottles autoclaved, 10ml of trypticase soy broth made up with distilled
water containing at least 300 viable bacteria per ml was autoclaved in a
culture tube in a 500ml sample bottle. The sample bottle was then incubated
for three days and observed for growth.
In seven of the community systems sampled, the finished water from the
well or treatment plant had many fewer positive coliform samples and lower
coliform densities than the water collected from the distribution system.
Well water samples were not obtained for systems MW and BG. These data are
presented in Table 4.1 and are evidence that the sample bottles used in this
project were properly sterilized. The data in Table 4.1 include all samples
and the volumes filtered varied depending upon the system and sampling
location.
4.2.2 Tap Location
The Regulations state that samples for the determination of microbio-
logical water quality "shall be taken at points which are representative of
the conditions within the distribution system." (U.S.E.P.A. 1976). This
may be interpreted in different ways (Geldreich 1971). Samples may be taken
from different sampling sites including private residences, public buildings
such as schools and fire houses, commercial establishments, and fire hy-
drants. Also, samples may be taken from bathrooms, kitchens, utility rooms,
outside hydrants, or drinking fountains. Interviews of personnel of water
departments indicated that some communities sample solely from residences,
others from fire hydrants, others from public buildings and commercial es-
tablishments, and others from combinations of location types. Public rest-
rooms by their very nature are used by a diversity of people with an equally
diverse assortment of personal cleanliness habits, and taps in public rest-
rooms may be subjected to contamination not common to other faucets. Also,
23
-------
w
.J
EH
> CO
Distribution System Samples |
Finished Water Samples
CO
01
rn
P
n
CO
d
Density a |
Fraction
Number
Number (Fraction (Density
Density3
Fraction
Number
o
01
p.
Positive
o
rH
01
-5
Positive
rH
<§
O
rH
r- CO OO
r^ co r**
i-l VO CM
O rH rH
A A
in oo oo
r~. in o
O CM i-H
o o o
r-» 00 CM
CM ON sr
CM CM CO
CO
o
o d o
rH
CO
O
o d o
ON in
m o
CM i-H
O 0
d d o
rH O
H i-l
rH O
d d o
OO rH ON
•-H O CM
rH
P*» rH
ON OO
i-H ON
rH
nj CT\ o)
X r-. a
1 ON S
H rH ^
•rl 1
^1 >N >N
rH
sr in
sr CM
O ON
A
H vo
VO CO
0 rH
0 0
CO OO
rH ON
CM H
0 0
o o
rH VO
CM rH
ON
CM
O
d o
ON
CM
O
d o
-* rH
CO rH
ON ON
ON ON
rH rH
0) >-.
43
pq
M
ON CO
CM in
co m
A A
vo ro
f» co
CM CM
0 0
CM ON
ON O
H CO
O O
0 0
00 O
i-H rH
0 O
O O
00 O
CM CO
ON ON
i-H rH
01 X
a I-H
a s
di-aij
00 ON
CM CM
O ST
O O
A
in CM
CM CO
o o
o o
CM CM
SI- CM
CM CM
O O
0 O
ON in
sr m
CO O
ON O
vo ON
A
§O
0
0 0
rH rH
m CM
i-H rH
OO 00
ON ON
i-l rH
14 01
o a
O "-J
H
O
ON ON CM
rH CO rH
o m f-
o o o
A
ON sr sr
rH CO ON
o o o
o o o
O ON ON
vo i-H ON
CM co CM
CO
CO
o
o i d
CO
CO
o
O 1 O
m o
CO O CO
i-H
ON
CM
O
d o
CM
CM
0
d o
m m
CO O CM
o o
00 00 rH
ON ON 00
rH i-H ON
i-H
43 rH
O -rl 0!
rl rl a
j
VO
ON
CO
d
A
i-H
CM
H
d
CM
rH
O
CM
in
O
00
ON
rH
§
7
1
,0.
M
ON
rH
rH
O
rH
d
o
r-.
CM
CM
rH
O
d
CM
i-H
O
O
00
CO
ON
rH
O
01
a
4J
O
o
CO
co
01
24
-------
CM
sr
CO
H
g
O
CO
H
1
lishments
ings
.0 T3
CO rH
4J -H
CO 3
*3 BQ
rH rH
•rl O
O iH
VJ 0
0> iH
a c
J*
tn
j_i
4J
10
CD
(A
•H
^Q
£
co
0)
ej
01
•O
•rH
ce
01
CA
41
«
•H
t-t
PH
t>0
(3
•rl "rj
H O
fX-rl
a vi
cd 0)
CO CUi
g
4-1
CO
to
a
c«
0)
ction]
i^
f»t
MH
o
•
O
9
£
o
•H
4J
O
£
<4H
O
•
O
&
s
cd
<1)
ss
g
jj
o
CD
£
*w
o
•
o
; Density
per 100ml)
i
z
4J
•H
CO
O
P«
CO
0>
rH
1
CO
•u o
iH rH
C >-<
o a
>
4-1
•H
co
O
a
a)
rH
IX
ed
w
/-N
rH
£•1
•H rH
CO
Ol 4)
>
TJ
•H
co
O
Pi
CO
01
rH
a.
1
0 0
0 O
00 OO
CM
^
••J
0 O
rH
*
O 0
CM vo
en
CM CM
o m
en O
d d
Ox Ox
CM rH
0 0
d d
CM 00
rH CM
OX
Ox Ox
r^ I-H
Ox
H £.
• O
•£> ^>
01 cd
Pu
U
0
o
I-l
m
xO
d
m
o
o
o
CM
0
r-
o
^
A
O
r-
0
d
^j-
rH
rH
Ox
r-
rH
s**»
rt
T
rH
•rl
n.
33
3
d
VO
rH
O
rH
O
rH
d
O
i-H
O
O
H
r-
in
rH
en
xO
CM
d
«^r
r—
CM
ox
f^.
O^
rH
OX Ol
^^ C-i
ox 3
rH r,
1
II
o
o
CM
O
O
CM
CM
m
0
rH
CO
O
O
r-
Ox
rH
OX
Ox
rH
1
«
ca
PC
i-J
CM
OX
CM
en
en
en
0
1
O I*1*
in en
CM 00
d 1-5
A
en CM
00 OO
O 0
o o
CM Ox
rH sr
O O
0 0
sr oo
sr oo
f- Ox
r- r~-
CM vO
A A
CM O
Ox en
CM CM
0 O
vO CM
CM m
CM CM
Ox Ox
OX Ox
rH rH
01 >x
O rH
3 3
p
PQ
>J
O O
o o
OO vO
i-H rH
OX in
in CM
O rH
d d
Ox CM
in vo
o o
0 O
r^ xo
rH rH
sr oo
en CM
O r»
o o
A
sr vo
CM en
0 0
o o
f^ xO
O en
CM rH
O 0
oo co
Ox Ox
rH H
• *
rt iQ
ft 0)
r-J fj^
1 1
o a
O) ed
Q >-J
H
Q T
0 0
m oo
O rH
d d
o o
m CM
0 rH
o o
0 O
CM m
0
o
d o
o
3
o o
in
CM CO
Ox Ox
O en
o oo
o o
A
OX rH
o en
0 0
0 O
in rH
H vO
CM CM
O O
co co
OX Ox
i-H rH
,fl rH
0 iH
I-l I-
:§•§
1-3
.. *
i—
oo
Ox
!—
0
'*.
25
-------
contamination of faucets can result from aerosols produced by toilet flushing
(Gerba et al. 1975). Thus it was postulated that samples from public rest-
rooms may provide a larger fraction of positive coliform results than those
from other taps due to subsequent contamination.
Two approaches were taken to examine the influence of faucet type on
subsequent contamination. Samples were obtained from public restrooms and
from other taps in commercial establishments in six of the systems. These
data were analyzed with respect to the probability of positive results from
each of the faucet types. In Table 4.2 the results of sampling of private
residences, public restroom faucets, and taps in other commercial or munici-
pal buildings for six water distribution systems are shown. Systems MI, SR,
MW, and BG were small residential communities with no municipal or commercial
buildings available for sampling. The data show no differences between
samples collected in public restrooms and those collected in private resi-
dences or those collected from other taps in municipal and commercial build-
ings.
The second approach was used to eliminate any space dependent variations
in coliform density. For this approach two faucets from either a commercial
establishment or public building were sampled within 15 minutes of one
another. A restroom faucet was sampled as well as another faucet within the
same building. The results of this study are shown in Table 4.3. One posi-
tive sample was obtained from the 17 sites sampled. The positive sample
was from a restroom faucet. These data suggest the possibility of subsequent
contamination from restroom faucets, although they do not prove that it
occurred.
TABLE 4.3
COMPARISON OF PAIRED SAMPLES FROM COMMERCIAL ESTABLISHMENTS
System Number Vol. Public Restroom Other Faucet
of sites filtered Number Coliform Number
sampled Positive Count Positive
cv
8
100ml
800ml
100ml
0
0
0
0
0
0
WH 9 100ml 0 0
100ml 1 (13) 0
Totals 17 10
Fraction 0.059 0.0
Positive
95% Confidence 0.003 to 0.254 0.0 to 0.167
Interval
26
-------
4.2.3 Sampler Error
Water flowing into the bottle may accidentally come in contact with the
sampler's finger, or the sampler's finger may brush against the lip of the
sample bottle. The appropriate course of action should be to eliminate that
sample immediately. However, often samplers have a set number of bottles for
an equal number of samples to be taken on the day or may not be aware of the
incident. Also, during the processing of samples, contamination of the sam-
ple by the finger of the analyst may occur. If no coliforms are present
upon testing, these errors are likely to be forgotten. If, however, a posi-
tive coliform result is obtained, these errors may be used as an explanation
by the sampler. This situation poses a problem in interpreting the validity
of positive coliform results.
Three questions were asked in this study. 1. If a finger were placed
into a water sample, what is the probability of obtaining a positive coliform
result? 2. If a positive coliform result is obtained, is the number of
coliforms always large enough as to suggest this route of subsequent
contamination? 3. When positive coliform results are obtained, are the
species of organisms different from those normally found in distribution
systems?
One hundred ml of dechlorinated tap water were placed in 125ml sample
bottles, sterilized by autoclaving at 121°C for 30 rain., and cooled to room
temperature. Test subjects were asked to place one finger into the mouth of
the bottle and shake the bottle vigorously for approximately 5 sec. The
bottle was then capped, returned to the laboratory, and processed by membrane
filtration techniques. Selected colonies from positive plates were isolated,
verified, and identified by the normal procedures used for this project.
Test subjects were divided into three categories: 1. Water samplers
from our laboratory, 2. Water samplers and water department employees from
four different municipalities, and 3. Drexel University staff and students.
In categories 1. and 3., a portion of the samples were taken from the same
subjects on different days. In category 2. each sample represented a dif-
ferent subject.
The procedure used to determine subsequent contamination from fingers was
designed to maximize the probability of contamination. Thus the results pre-
sented may be interpreted as a worst case situation. Despite this approach
only 5.4% of the samples contained coliforms (Table 4.4). Positive coliform
results were not found from any municipal water department personnel. The
positive results from Drexel University water samplers were more numerous
than for the other groups. These water samplers were also responsible for
transfering coliform cultures and bacterial identifications. It appears that
a positive coliform result may occur from the hands of the sampler, but such
results are not highly probable.
Initially it was postulated that if a positive result were obtained, the
potential for this route of contamination may be recognized. A water sample
contaminated from a finger may have an unusually high bacterial density or a
species of bacteria not normally found in distribution systems. However
27
-------
TABLE 4.4
TESTS FOR SUBSEQUENT CONTAMINATION FROM FINGERS OF SAMPLERS
Subjects
No. of
Samples
Taken
No. with
Coliforms
Fraction
Contaminated
(95% confidence interval)
Project staff 12
Water Department 21
Employees
Drexel Staff 78
and Students
Overall 111
0
0.083
(0.000-0.236)
0
(0.000-0.087)
0.064
(0.010-0.119)
0.054
(0.010-0.096)
when positive samples were obtained the number of coliforms or non-coliforms
ranged from 1 to too numerous to count (TNTC = >80/100ml). Thus no indica-
tion of subsequent contamination could be found based merely on the number of
colonies present. Also, the species of organisms found were similar to those
we have found in water distribution systems. Coliforms found included
Klebsiella pneumoniae, Enterobacter cloacae, and Enterobacter agglomerans as
determined using the API system. The possibility of this type of subsequent
contamination could not be determined by species identification.
4.2.4 Contamination During Transit In Ice
It is recommended that microbiological samples be stored at less than
4°C during transit to the laboratory (Bordner et al 1978). This may be done
by keeping samples in a chest filled with ice. As the ice melts the sample
bottles may have their tops immersed in the ice water. This immersion may be
a route of subsequent contamination. As the bottles cool, differences in the
coefficients of expansion of the materials used for the bottles and the caps
.may cause the caps to loosen. As cooling occurs the air within the bottle
may contract producing a negative pressure. A combination of these two
events may allow the intrusion of the surrounding ice water into the bottle
and thus subsequent contamination. Previously, Geldreich (1975) has cau-
tioned about this route of contamination but did not provide quantitation.
A series of experiments to determine the likelihood of occurrence of this
route of subsequent contamination was conducted.
Sample bottles were filled with phosphate buffer made up with distilled
deionized water and autoclaved at 121°C for 30 min. Caps were tightly se-
cured after cooling of the water. The bottles were then placed into ice
water for 4 hours. The time in which the bottle tops were immersed in the
28
-------
ice water was controlled. In some cases the tops of the bottles were im-
mersed for the entire 4 hours. Also, tops were immersed for 0.5 hr and
0.08 hr after cooling for 3.5 hr and 3.92 hr, respectively. One hundred ml
of water from each bottle was then membrane filtered, and the filter was
incubated on Endo medium for 22 hr at 35°C. Both coliform and non-coliform
colonies were counted. Also, the ice water was sampled for coliforms by the
same membrane filter technique.
Four different types of bottles were employed: 1. 125 ml glass dilution
bottles with plastic caps, 2. 1 qt. (945 ml) glass bottles with plastic caps,
3. 125 ml polypropylene bottles with polypropylene caps, 4. 1 liter polypro-
pylene bottles with polypropylene caps. In one experiment surface water was
added to the ice water to increase the coliform density.
Bottles were immersed in ice water containing different densities of
coliforms ranging from 1 CFU/lOOml to 105 CFU/lOOml. The results in Table
4.5 represent a composite of all bottle types and all times of immersion.
Subsequent contamination occurred at all coliform densities tested. At low
densities (< 10 CFU/lOOml) as many as 12.5% of the bottles were positive
for coliforms. At a high coliform density (105 CFU/lOOml due to the intro-
duction of creek water) all bottles were positive for coliforms. Unless one
is assured that ice used in transit is completely devoid of coliforms, the
potential for contamination is too great to allow the transit of sample
bottles in ice where there is a possibility of bottle tops being immersed in
the ice water.
TABLE 4.5
CONTAMINATION OF SAMPLE BOTTLES IMMERSED IN ICE
WATER CONTAINING DIFFERENT DENSITIES OF COLIFORMS
Density of Coliforms
in Ice Water
(per 100 ml)
< 1
rv 7
~ 10
~ 105
No. of
Bottles
Under
Ice Water
17
24
10
48
No. of
Bottles
With
Coliforms
1
3
1
48
Fraction
Positive
0.050
0.125
0.100
1.000
(95% Confidence
Interval)
(0.003-0.254)
(0.035-0.308)
(0.005-0.397)
(0.925-1.000)
Ice may contain bacteria which are not coliforms but which may grow on
Endo medium. Such organisms may also contaminate samples and may be respon-
sible for interference with coliform counting (Geldreich et al 1978). We
were able to use this population of organisms as a tracer in our subsequent
contamination study. The ice we used could be divided into two states with
29
-------
respect to non-coliform organism densities. "Uncontaminated" ice water
possessed < 1 CFU/100ml of these organisms, and "contaminated" ice water
contained > 200 CFU/lOOml. Experiments were divided according to these cri-
teria. In Table 4.6 the results of experiments with different periods of
immersion of bottle tops are shown. In "contaminated" ice water positive
results were obtained in all but one case (65 of 66 bottles were positive).
Both glass and polypropylene bottles could be contaminated, and contamination
occurred even when bottles were immersed for as little as 5 rain. In "uncon-
taminated" ice water the fraction positive was considerably less (2 of 17
bottles were positive). The positive results in this case were for the larger
sample bottles. The results confirm those for coliforms and extend the obser-
vation that only a short period of immersion is required to obtain contamina-
tion.
Based on these and other findings, it is clear that bottles should not
be stored in ice in such a way as to allow immersion of the bottle top
(Geldreich 1975). However, samples should be maintained at 1-4°C during
transit whenever possible. This problem may be alleviated by using sealed
coolant instead of loose ice.
4.2.5 Carry Over During Filtration
In the membrane filter technique for coliforms the same filtration unit
is used for a number of samples. It is recommended that a filtration unit be
rinsed between each sample to insure that all coliform bacteria in the sample
reach the filter and to avoid the carry over of coliforms from one sample to
the next. If coliforms were found in high density in one sample, inadequate
rinsing could permit coliforms to be found in one or more subsequent samples.
This source of subsequent contamination must be avoided or at least identi-
fied if it were to happen. The prescribed technique of rinsing is to "rinse
the sides of the funnel at least twice with 200 ml of sterile dilution
water". However, a number of variables in procedure exist which are not
addressed in this general protocol. These variables include 1) measuring
sample volume in graduated cylinders or directly in the funnel, 2) rinsing
with a squeeze bottle of dilution water or from a graduated cylinder, and
3) using different brands of filter funnels. These variables were examined
to develop a rinsing procedure to minimize the potential for carry over.
The potential for carry over was investigated by pure culture experi-
ments and observations of filtering order of field samples. Pure cultures
of Escherichia coli or Enterobacter cloacae were grown in nutrient broth
overnight and diluted in sterile buffered dilution water to appropriate
densities. One hundred ml of diluted cultures were measured in graduate
cylinders and then filtered in either glass or plastic polysulfone filter
funnels. The glass funnel was held in place by a spring loaded clamp while
the plastic funnel was held in place by a magnetic ring. Two rinsing proto-
cols were followed. For one series of filters, a 30 ml aliquot of sterile
dilution water was placed in the graduated cylinder used previously, and
this aliquot was used to wash down the sides of the filter funnel while
vacuum was still being applied. This procedure was conducted a total of
three times. The filter was then removed and incubated on Endo medium. A
second series of rinsing was done on replicate samples using sterile dilution
30
-------
W
Si
M
H
b
0
co
a
o
H
a!
w
Q-J
j— j
0
M
53
>
0
b
1
r3"
W
•-•
S3
H
Q
W
CO
Di
§
53
H
CO
W
13
H
O
oa
b
0
co
W
PM
^4
H
H
S3
2
w
b
b
M
Q
as
o
b
CO
W
0,
y
3
co
W
M
H
rH
CO
O
Qj
0)
O
a
CJ U
(U
»•? 4-1
O\ H
**^
a
o
•H
4->
j.
4-1
- i
•rl
CU
CO > CO
>4H CU iH 4J
O rH 4-1 rH
4-1 -H 3
• 4-1 CO CO
O O O 0)
CO
<4-l 01
O rH
4-1
• 4-1
O O
S3 tO
i-4
CU
4-1
C cfl
•H (X
co cu
rJ U
m H
rH
4-> ^
« H
•'^
rH
O
i— I
C^
b
CJ
o
o
CM
\^f
)-l
0)
4-1
cd
jj
0)
o
•H
T)
0)
4-1
n5
C
•H
e
cO
4-1
c
0
CJ
X-s ^v S~^
O O O
o o o
0 O O
1 1 1
co vo m
VO r-4 00
oo oo vo
o o o
V^V-' N_/
m o o
CJ\ O O
*
—I CM
CM ^H OO
CM CM
CM —1 00
00
m o
-* o o
CO
CO
rt
rH
DO
•a
m
C\J
i— I
^^ ^^ ^v ^^s x^ ^^\ ^^
o o o r^ o -H —i
o o o CM o m m
o o o m m r-~ r^.
Ill till
m m m o o co co
00 00 00 O O >-i <-i
vO vO vO C^ CO t^ CO
O O O 0 O O O
^XN_X^^ \-/ **S <*** \^
o o o m m
O O O x-\ CM CN
o
o
-^
g
U
i—4
v^^
W
a;
4J
a)
oooooo |j o o i-i 1-1
cu
o
•H
*^?
CU
4-1
oo oo oo (t) *3" in *^" <±
rj
•H
0
flj
4J
(J
O
o
rj
00 £3
m o
-------
water kept in a squeeze bottle. In this series the graduated cylinder was
rinsed well with a jet of water from the squeeze bottle and poured onto the
filter. Approximately 5 to 10 ml of water were used for this process. The
sides of the filter funnel were washed with a jet of water from the squeeze
bottle (20-30 ml). Again the filter was removed and incubated on Endo me-
dium. The bottom ring of the plastic filter funnel top was then rinsed with
a jet of sterile dilution water (5-10 ml). For both rinse series a new
filter was placed in the filter funnel and 100ml of sterile dilution water
was filtered. These filters were incubated on Endo medium to determine the
extent of carry over.
The results of one series of laboratory experiments using E. cloacae are
shown in Table 4.7. The variables investigated in this study included coli-
form density, filter funnel type, and rinsing protocol. In this study nei-
ther rinsing protocol nor filter funnel type altered the potential for carry
over. However, the initial density of coliforms was a major determinant for
the potential of carry over. Little or no carry over was found at the lower
two densities studied, and carry over was almost unavoidable at the higher
two densities. In other experiments with E. coli and contaminated surface
water, carry over routinely occurred after densities too numerous to count
(>80 coliforms/100 ml) were filtered. During these experiments squeeze
bottle rinsing was found to be the more effective protocol. Also, the im-
portance of rinsing the bottom ring of the plastic filter funnel should be
noted. Apparently the magnetic ring which holds the filter funnel in place
does not produce enough pressure to prevent seepage of water out of the
funnel during filteration. Extra rinsing of this ring to help eliminate
carry over is recommended.
Samples collected in the field were used to demonstrate the reliability
of the squeeze bottle procedure. Duplicate 100 ml samples were taken at all
locations sampled, and occasionally 1 liter samples were taken in addition.
Filtration of samples were carried out 2 to 6 hours later. Samples were
filtered, filter funnels were rinsed by the squeeze bottle protocol, and the
order of filtering recorded. The extent of possible carry over was deter-
mined by examination of sequential positive coliform results. Carry over
was considered possible if two criteria were fulfilled: 1) if the positive
coliform result immediately followed in order of filtration another positive
result with a greater coliform density, and 2) if the positive coliform re-
sult was not supported by a positive result in a replicate sample.
The coliform results from selected field samples are presented in Table
4.8. One hundred three samples gave positive results during this study of
which 4 were considered possible cases of carry over by the two criteria
described above. Two of these were from 1 liter samples in which a count
of >80/100 ml was followed by a count of 14/100 ml and then 7/100ml. The
next filtered sample was not positive for coliforms. One liter samples are
rarely used in routine analyses, but may offer a greater possibility of
carry -over than 100 ml samples. In liter samples the 300 ml filter funnel
is filled to the top and water is in contact with the funnel for a longer
period of time. The remaining two possible carry over cases each gave 1/100
ml after a sample containing >10/100 ml was filtered. As stated, the repli-
cates of these samples were negative. However, at low coliform densities it
32
-------
w
. T
HH
w
#3
5s p^
r^ W
fa ^>
_, o
021
W >H
H p£j
rH ^^
fa O
fa fa
0 0
W CO
fa H
r* 55
H ^
5^
* M
^"* r£j
EH p3
M W
CO H
z w
w p
p
co
s
O HH
fa O
H CJ
rJ O
0 H
°§
fa P-l
0
o
CO !Z
A. i ]
H CO
sc a
CO M
IS Cd
0
M Q
H 5
M
PQ
Ol
rH
•H
fa
*o
d
o
0
0>
CO
Jf*
CJ
cO
w
d
o
4J
d
3
O
CJ
g
o
<4H
•H
rH
o
O
O
g
<^
O
rH
^
01
a
\^s
x"x
P
co
^^
+i
^
4J
•H
co
a
0)
p
g
o
•H
rH
O
CJ
rH
01
d
d
3
fa
^4
0)
4-1
rH
•H
fa
CJ
•H
4-1
CO
cO
rH
fa
rH
ai
d
d
fa
CO
co
CO
rH
O
CO
M
01
4J
rH
•H
fa
•a
d
o
o
01
co
*4H
O
d
0
•H
4J
CJ
cO
^
fa
01
N
cr
CO
T3
01
4J
to
3
13
CO
r-*
g
O
<4H
•H
rH
O
CJ
0>
o
d
01
TJ
MH
d
o
CJ
&•?
in
ON
^
O)
>
iH
4-1
tH
CO
o
fa
0>
rH
4-1
4-1
O
ta
IH
01
"d
•H
rH
^
CJ
0)
rH
4-1
4-1
O
pa
^4
a>
•o
d
•H
rH
CJ
O
4.)
•H
CO
d
0)
P
^
rH
CO
J-*
a)
ij
d
M
^
4J CD
•H CJ
co d
d o)
0) *O
P iH
r7 S d
cfl M O
> 0 CJ
M <4H
01 -H fr«
4-1 rH m
d O ON
H CJ ^
01
CO
d
iH
P-^
cu
CO
d
p^
01
w
iH
p^
CU
CO
d
Prf
01
r*i ^^
H-d o
CO 01 O
M rH
4-1 01
CO 4-> r<
1-1 H 01
•H -H CX
fa fa *~*
o
o
o ^
• o
r- 1 •
1 CO
-» rH
^O 1
r-^ «N
o m
v^ / v^
O
O
O rH
^•H O^
y^S
i-H
•
ro
N^^
CO +1
r*x co
M *
ON vO
^-^
O
ON
**D
i-H +|
A
m o
CSJ •
ft SO
r-v »-H
^^
m
rH
N.^
^O +!
A
in co
A •
oo NO
CN
•
«^-
\^
ON
•> +1
rH
rH (•*•.
co r*^
o
o
o
o
CO
m
vO
oo
•
o
I
CO
CO
CO
o
\^s
m
co
\o
•
O
o
i-H
A
O
£SJ
M
CO
rH
rH
A
CO
rH
ft
O
0
o
o
o
M
CO
/i— \
00
•
f— I
1
o
*~s
i-H
•
i— 1
/••x
f— (
•
CO
v_ /
+1
CO
•
^0
^^
o
i— H
x^ x
_|_|
O
•
CM
•—4
<^
•
i— 1
\^s
+1
O
•
CM
vO
•
O
\*^s
+1
CO
O
vO
^H;
CO
• /^^
O CO
1 •
-* o
0 1
0 O
0 0
CO
oo
O -H
• •
O 0
s~\
o
o
O +1
*»
i-H ^^
A •
0 0
^^
ViO
•
o
rH +|
O CO
o o
^
o
o
O +1
cTo
M •
0 0
o
•
o
•*_/
*t
0 +1
•s
O 0
0 0
m
CN
F-H
o
m
CN
•
o
1
0
0
o
0
\^s
o
o
o o
o o
/-^
o
o
v^ /
O +1
O 0
A •
o o
^- N
o
o
°« +l
o o
A •
o o
^
o
o
O +1
0 O
* •
0 0
o
•
o
X^ '
+1
0 0
0 0
ON
1— 1
33
-------
is to be expected that some replicate samples from a site will be positive
when the other is negative.
TABLE 4.8
POTENTIAL CARRY OVER ANALYSIS OF SELECTED SAMPLES
COLLECTED IN THE FIELD
Initial
Coliforms Coliform Count of Next Sample (No. of Second
Count Samples with the Density Listed)
1 0(30), 1(6), 3(3), 4(2), 5(1), 11(1), 15(1), 35(1), 48(1), TNTC(l)
2 0(9), 2(2), 11(1)
3 0(8), 1(3), 3(1), 9(1)
4 0(3)
5 0(1), 24(1)
6-10 0(7), 1(1)
11-20 0(9), 7(1*)
21-40 0(2), 1(1*,1), 2(1)
41-60 0(2), 1(1)
61-80 1(1*)
> 80 14(1*)
* potential carry over
Based on these findings about subsequent contamination, carry over is
a source of error. Therefore, to avoid carry over or recognize it when it
does occur, the following procedures are recommended:
1. Rinsing of any graduated cylinder used to determine sample volume
must be included in the protocol.
2. Rinsing with a jet of water from a squeeze bottle is at least as
efficient and perhaps more efficient than rinsing from a graduated
cylinder.
3. The bottom ring of the plastic filter funnel top should be rinsed
between each sample.
4. The order of filtration must be recorded and positive results
interpreted within the context of filtration order.
5. Any positive results filtered following a positive result of greater
than 20 coliforms/100ml should be considered suspect of subsequent
contamination by carry over.
34
-------
4.2.6 Summary on Subsequent Contamination
To assess accurately the microbiological water quality of a water
distribution system, procedures must be designed to avoid subsequent contami-
nation or account for it if it does occur. Potential routes of subsequent
contamination include 1) tap location, 2) sampler error by placing his
finger in the bottle, 3) transit of sample bottles in ice, and 4) carry over
during filtration. Of these, transit in ice and carry over during filtration
give the greatest likelihood of subsequent contamination. Sampling public
restroom faucets and finger contamination of samples demonstrated a low
probability of leading to subsequent contaminated positive results. However,
any chance of subsequent contamination should be minimized or eliminated if
possible.
Based on observations from these studies, recommendations for sampling
and sample processing are as follows:
1. Routine sampling should avoid public restroom faucets. If a sample
from a commercial establishment or public building is desired, the sample
should be taken from a faucet from a kitchen, utility room or other non-
restroom location.
2. If a water sample comes in contact with a finger or part of the sam-
pler's hand at any time, the sample should be discarded. Although a negative
result is highly probable, the chance for a positive result exists. However,
the probability of obtaining a positive result is so low that the use of
this as an excuse for a positive result is tenuous at best.
3. Water samples should be cooled to 4°C during transit but the tops of
sample bottles should never be allowed to be immersed in ice or ice water.
Sealed containers of ice or coolant should always be used in preference to
loose ice. This precaution is so simple that no substitute short of portable
refrigeration should be allowed.
4. Carry over of coliforms from one sample to the next during filtration
is of major concern as a route of subsequent contamination. If a sample
contains a large number of coliforms (>80 coliforms/100 ml), subsequent con-
tamination is highly probable. Optimum rinsing procedures must be maintained
at all times. Rinsing by squeeze bottle is at least as effective and prob-
ably more effective than using a graduated cylinder rinse. Any graduated
cylinder used to measure sample volume must be rinsed into the sample filter.
The bottom ring of a plastic filter funnel top, if used, must be rinsed
after the funnel has been disconnected. Even with these precautions carry
over may occur. To account for potential carry over, the order of filtration
must be recorded. Positive samples should then be compared to filtration
order. If a positive result of low coliform density is found immediately
after one of a higher density, subsequent contamination may have occurred.
Repeat samples of both should be taken and a note of explanation of probable
subsequent contamination included in any report.
If these recommendations are followed, the probability of obtaining a
positive coliform result through subsequent contamination is minimized. None
35
-------
of the recommendations are contrary to present regulations or guidelines and
can be instituted with little or no added effort or capital expenditure. The
importance of obtaining accurate results of coliform densities is obvious.
Employment of these recommendations would help insure that accuracy.
4.3 CHANGES IN COLIFORM DENSITIES IN WATER SAMPLES
The accepted procedures for handling samples from water distribution
systems allow for a delay of up to 30 hours between the time of sample col-
lection and when the sample is filtered (Bordner et al. 1978). In this
project samples were collected in the morning and filtered in the afternoon.
The maximum time between sample collection and filtration was eight hours and
most samples were filtered within 2 to 6 hours after collection. However, we
did obtain some information on changes in coliform densities in water samples
as a result of two types of experiments.
4.3.1. Preservation Techniques
As a part of the development of procedures to be used for this project,
before the field sampling was initiated we tried an experiment with sample
preservation techniques. The samples for this experiment were obtained from
system WD, a large water system which was not included in the field studies.
The reason for use of this system was that the laboratory director knew of
a location which consistantly gave samples with a low coliform density.
All samples were obtained from a single fire hydrant. One gallon sample
bottles were used. The preservation techniques used were 1. chilling the
sample bottles in an ice chest (there was no opportunity for immersion of the
bottle tops due to the size of the bottles), 2. adding 2.4ml of 10% thiosul-
fate solution for a 3 liter sample, and 3. adding 3 gm of peptone and 15 gm
boric acid for a 3 liter sample. The third treatment was based on the pre-
servation medium of Brodsky et al (1978) as modified by Lariviere (1978).
Chilled and unchilled samples were used for each of the preservatives and
for samples with no chemical preservatives, giving a total of six different
treatments.
Eighteen three liter samples were collected in one gallon bottles. This
gave three samples for each treatment. The sample bottles were numbered;
bottles 1 through 6 had no preservatives added and 1-3 were held at ambient
temperature while 4-6 were chilled, and so forth. The time sequence of
filling the numbered bottles was obtained from a table of random numbers.
During sample collection the temperature of the water from the hydrant
decreased from 18°C to 16°C. The temperature of the water in the bottles
held at ambient temperature increased to 27°C over the first 24 hours and
held constant for the second 24 hours. The temperature of the water in the
chilled samples decreased to 6°C in the first 24 hrs and then held constant.
Three replicate 100ml aliquots were taken from each sample bottle at 3,
6, 21, 27, and 48 hours after sample collection and filtered through membrane
filters. The filters were incubated at 35°C for 22 hours and coliform counts
were made. The presumptive coliform colonies were not verified in this ex-
periment because our interest was solely in comparing the preservation tech-
36
-------
niques. At the same time the aliquots for coliform counts were taken, three
aliquots were taken for standard plate counts.
The standard plate count results are given in Table 4.9 and the coliform
results in Table 4.10. Analysis of variance was used to detect significant
differences in the rows and columns of these tables . These differences are
exhibited by differences in the 95% confidence intervals for the individual
counts which are given in the tables. The standard plate counts are discuss-
ed first because those results are useful in understanding the coliform
results.
There were variations in the bacterial densities in the water during the
45 minute period that it took to fill the 18 gallon bottles. In spite of the
fact that the order of filling the bottles was randomized to try to avoid
significant differences among the treatments initially, one treatment (pep-
tone borate preservative held at ambient temperature) had a significantly
lower standard plate count to start with. No significant changes in the
standard plate counts between 3 and 6 hours after collection were found.
The standard plate counts in the samples held at ambient temperatures with
no preservative and with thiosulfate increased greatly during the first day
after collecton and apparently continued to increase the second day after
prevented collection of precise data on this point. The samples with no
preservative or with thiosulfate show no significant changes in 48 hours.
The peptone-borate preservative caused a decrease during the first day fol-
lowed by an increase the second day. It may be speculated that borate at
the concentration used was toxic to some of the bacteria but other types of
bacteria which are a part of the standard plate count can grow in the pre-
sence of borate.
The coliform data are more difficult to interpret because the mean
counts were much lower and the variability of the counts was such that sta-
tistically significant differences are rare. It appears that the coliform
density was decreasing in the samples with no preservative and with thiosul-
fate which were held at ambient temperature. However, since the standard
plate counts were increasing greatly for these two treatments, the apparent
decrease may have been the result of interference from the non-coliform
bacteria growing in those samples.
4.3.2 Second Day Filtration
During the sampling of system LB, a question arose as to why we were
finding so many samples with coliforms whereas the twelve samples per week
collected by system personnel never showed any coliforms. They allowed
their samples to sit at ambient temperature, sometimes in direct sunlight,
for up to seven hours while they were being collected and then refrigerated
them for about 20 hrs before they were filtered. Our samples were placed in
an ice chest within an hour after collection and filtered the same day.
To test coliform die-off in the 24 hours after collection, we collected
some 250ml samples and saved the rest of the sample after filtering 100ml the
day of collection. The samples were stored in a refrigerator. On the day
after the samples were collected, 100ml aliquots from those which had coli-
37
-------
ON
H
E-*
55
t3
o
o
W
H
jj
Pk
Q
pq
^3
Q
s
^
H
CO
5Z
O
co
g
cy
M
55
e
W
SZ
0
H
^
(Hi
CO
£
a,
fa
o
H
CJ
W
fa
ptj
U
X"N
rH
CO
M
0)
4-1
d
H
O>
U
0>
•H
d
0
CJ
B-S
m
ON
^^
cu
e
•H
H
TJ
0)
4-1
CO
O
•H
TJ
d
H
4J
cfl
r~H
e
M
a)
a,
4-J
d
3
O
O
o>
JJ
CO
rH
PL,
TJ
M
CO
TJ
d
4-J
CO
4-1
d
e
cO
cu
H
CO
M
3
0
•d
00
CO
M
3
O
rC
t".
CM
CO
M
3
O
rH
CM
CO
rJ
3
0
rd
vO
co
M
3
O
j£
co
oo
•
rH
m
o
4-1
CM
•
CM
r*^ st
st ^^
^-^
rH
•
CM
VO
O
ON
rH
CM CM
^-^
vO
vO
st
O
4-1
**
f*».
CN| CO
•^ *"*-•'
J^
•
o
vO
O
4-1
CO
in
co st
m *-*
^^
VO
ON
in
o
4-1
^-
•
vO
CO st
in \^
01
•H
4J rg
•H 0)
T3 ^H
*X3 i"H
•^J *H
I"
o
o
o
1— 1
/\ 1
s*\
vD
O
vO
0
4-1
r-. oo
CO vO
co ^
/-s
m
•
rH
vO
CM
0
4J
m
oo
ON
•
ON
m co
*^ V— /
S~+i
f"^,
•
00
o
4-1
CO
•
CO
rH
CO
rH 0)
3 rH
CO rH
O -H
•H rC
H
O
O
O
rH
/N. 1
^^
r^
o
CO
rH
O
4-1
in co
r^ sj~
O 00
rH \^s
s~\
m
CM
m
o
4-1
m
vn si-
CO CO
sf ^^
00
•
ON
si-
O
4-1
CM
•
oo
sf CO
sf v-x
/^^
oo
•
CO
o
4-1
CM
•
cu
4J
cO
MH 4-1
rH d
3 0)
CO -H
O rO
/^%
rH
•
oo
CO
o
4J
ON
•
CM m
CM N-X
^^
r*^
•
^.
i— i
o
4-1
CO
•
CM vO
rH N— f
s~^
CO
O
4-1
^
oo •
vO
\^s
^
•
oo
CO
o
CO
•
rH
m -H
CM ^
^*i
r»^
0)
d rH
O rH
4J -H
0) CJ
PM
,^
CO
•
ON
rH
O
4-1
f1^
•
CM sf
rH N^X
^^
CO
•
-*
o
4-J
f-x.
•
CO ^H
%_/
S~*i
r^s
00
0
4->
CO
m •
i— i
^^
ON
•
vO
CM
O
4J
rH
•
p —
CM rH
CM
^^
vO
•
CM
CO
o
4-1
«^-
•
CO
00 CM
CM s-<
0)
a)
M
0
1 4->
^
•
-------
co
W
a
o-
H
3
Q_J
w
H
S
O
M
0 H
* ^>
»
CM
CO
M
3
O
^
rH
CM
CO
3
O
rC
vO
CO
3
0
en
^-^
m
*^
•
en
o
4J
m
o m
o •
• o
CM v-^
x-»\
en
CM
en
0
4J
en
oo en
r^> •
• o
rH ^^
^
ON
en
•
sf
O
4J
oo
ON CM
oo •
• rH
CM -^
/~x
i-H
CM
sf
O
4-1
m
oo en
r^» •
• i — i
CM \^
S^,
rH
en
00
o
4-1
m
en en
en •
• CM
cu
•H
4J
•iH rrj
T3 — '
^
vO
•
en
0
en 4->
en
• O
^
CM
oo
CM
O
4J
CN
r^ m
vO •
• O
i-H s_x
/•~s
o
en
o
4-1
CM
VO CM
m •
• rH
CM v_/
/^
ON
OO
m
o
4-1
in
vO •
• en
Sf v—'
a.
cu e
> cu
--I H
4J
•H 4-1
^3 C3
T*
en f^
en •
• o
CM — '
,_,
CM
m
-/
yr^
CM
in
sf
o
4J
vO
sf en
sf •
• o
CM N-'
P<
0) &
4-1 CU
ed H
<4H
rH 4J
3 C
co cu
0 -H
•H rQ
x-x
ON
o
•
CM
o
4-1
m
r*x. rH
rH •
• 0
/r^
vO
CM
en
o
4-1
o
oo en
r^ *
• o
rH >••*/
/_s
O
m
•^
o
4J
o
o m
o •
• rH
en v-'
s-**
ON
CM
en
O
4J
rH
O 1^
o •
• o
CM v-'
s-*.
oo
ON
m
o
4-1
vO
r^ en
vo •
• rH
en ^
cu
i j
ed
U
O
to
I
CU T3
G CU
O rH
4J -H
CX-H
CU rC
PJ U
/"X
O
oo
•
m
r-H
o
en 4J
en
• o
vO •>— '
^«v
m
oo
rH
O
4-1
m
o -H
o •
• o
x-x
00
CM
•
CM
O
4-1
00
en en
en •
• O
rH v-x
s~^
vO
en
^
o
4-»
O
en en
en •
• O
CM v^
s~\
O
-*
vO
O
4-1
VO
OO rH
r- •
• en
^^ ^^/
cu
^J 0
cd p4
M e
O CU
PQ H
1
01 4-1
c a
o cu
4-1 *r4
OH J^
tf \-s
CO
cu
rH
CX
CO
-H
39
-------
forms were filtered and the filters were incubated on M-Endo medium. The
results from these samples are presented in Table 4.11. All but one of the
samples from system LB showed a decrease in coliform count and in most cases
the coliform count on samples filtered the day after collection was zero.
Later in the project, the question of how large a density could occur
in samples which were counted as too numerous to count (TNTC interpreted as >
80/100ml) arose. The second day filtration experiment was tried with several
samples with coliforms and 1ml and 10ml aliquots were filtered from the
samples which gave TNTC results on the day of collection. These results are
also presented in Table 4.11. Significant numbers of samples from systems
WH and BL were examined in this way. Coliform counts increased in some
samples and decreased in others. On the average there was a slight increase.
It is clear that coliform counts on samples filtered the day after
collection are questionable. It is difficult to predict whether the coliform
count will increase or decrease in samples within the first 24 hours after
collection. This apparently depends upon water quality factors not yet
elucidated.
4.4 OBSERVATIONS ON INTERFERENCE
During the sampling of system CV filtration of one liter samples was
tried as a method of detecting coliforms at densities lower than 1 per 100ml.
As can be seen from Tables A5 and A6 there was essentially no correspondence
between finding coliforms in one liter samples and finding them in 100ml sam-
ples. Later in the project we found that coliform density changes abruptly in
water flowing from a tap and there is no reason to expect such correspondence,
However, at the time, finding coliforms in 100ml samples from three locations
sampled on 3/13/79 and not finding them in one liter samples from the same
locations raised a question. When coliforms were found in 100ml samples
from 2 locations .on 3/15/79 without coliforms in the one liter samples, the
filters from the one liter samples were inspected carefully. They were
found to be covered with a film of particulate matter. The filters from the
one liter samples from the two positive locations were carefully transferred
from M-Endo medium to Lauryl Tryptose broth. Gas was produced from fermenta-
tion of the LTB within 24 hours and the presence of coliforms was confirmed
by transfer to and production of gas in BGLB. It was concluded that coli-
forms were present in those samples and the particulate matter had interfered
with the development of recognizable coliform colonies on the filters when
they were incubated on M-Endo.
The following week, three one gallon samples were obtained from a loca-
tion in the system which was known to give turbid water samples. A multiple
tube fermentation test did not reveal any coliforms in these samples. The
samples were spiked with a strain of Enterobacter cloacae which had been
isolated from a previous sample from the CV system. Aliquots of 100ml,
200ml, 500ml, and 1000ml were filtered from each of the three samples. The
filters were incubated on M-Endo medium. The coliform counts obtained were
proportial to the volumes filtered and the coliform recovery gave the same
density as calculated from the coliforms added. Thus, it appeared that the
particulate matter in the samples, which formed a film on the filter, did
40
-------
TABLE 4.11
DELAYED FILTRATION EXPERIMENTS
System
LB-BB
LB-TP
SR
MW
BG
Sampling
Date
7/10/79
7/17/79
7/18/79
7/10/79
7/17/79
7/18/79
10/31/80
11/14/80
10/31/80
11/12/80
11/14/80
10/31/80
11/20/81
Sample
Code
331-1
178
258-2
334-1
328-1
329-1
331-1
688-2
699-1
699-2
470-2
488-1
566-1
638-1
717-1
475-1
480-1
577-1
591-1
113-1
111-3
22-4
23-3
51-7
52-7
55-3
61-4
61-8
72-12
122A-1
123A-1
123A-4
123B-1
123B-2
123B-5
1646-1
1646-2
MF Coliform
Filtered
3 to 6 Hours
After Sampling
1
10
TNTC
26
9
10
3
6
1
71
1
2
2
1
6
49
3
27
26
1
1
1
26
5
3
14
2
7
1
8
2
8
13
9
1
89
4
Count
Filtered
21 to 24 Hours
After Sampling
0
5
TNTC
18
0
0
0
0
0
25
6
0
0
0
0
0
0
1
0
0
2
7
33
28
2
26
0
6
0
0
0
1
3
23
4
117
8
41
-------
TABLE 4.11
(continued)
System Sampling Sample
Date Code
WH 5/27/81 114-2
172-1
206-1
206-2
5/28/81 114-1
114-2
174-2
216-1
6/2/81 206-1
195-2
185-1
185-2
106A-2
133-2
132-1
143-2
6/4/81 163-1
163-2
198-1
198-1
106C-2
209-2
212-1
212-1
214-1
215-1
BL 6/16/81 122-1
152-1
152-2
154-1
160-1
160-2
163-2
164-2
6/18/81 118-1
127C-2
146-2
146A-1
154-1
154-2
156A-1
168-1
170-2
MF Coliform
Filtered
3 to 6 Hours
After Sampling
6
1
12
5
6
4
TNTC
3
2
47
1
TNTC
3
1
3
2
3
43
1
42
5
TNTC
1
1
15
2
5
2
2
3
78
47
1
1
3
2
6
1
7
20
2
1
1
Count
Filtered
21 to 24 Hours
Sampling
3
0
3
5
15
0
1600
3
9
66
1
330
1
1
0
2
TNTC
21
1
33
TNTC
2700
5
6
2
4
34
2
0
0
290
TNTC
0
2
1
4
18
1
9
36
0
1
0
42
-------
not interfere with the recovery of Enterobacter cloacae freshly grown in a
laboratory culture. It is also possible that the particulate matter which
caused the turbidity in the samples which were spiked with Enterobacter
cloacae was different from the particulate matter which caused the turbidity
in the earlier samples.
One liter samples were used during the first two weeks of sampling
system WH. Of 13 one liter samples of well water collected on 4/26/79, two
produced coliform colonies on membrane filters on M-Endo. The other 11
filters did not produce coliform colonies on M-Endo but did produce gas in
LTB after they were transferred. Of four one liter samples collected at the
well (2 raw water samples and 2 chlorinated water samples) on 5/1/79 none
produced coliform colonies on filters on M-Endo but two of the four filters
produced gas in LTB. This was also seen on two filters from liter samples
from the distribution system collected on 4/26/79. We were unable to find
a way to eliminate this interference from particulate matter in large volume
samples and stopped using one liter samples.
Filters from 100ml samples were inspected carefully for the presence of
a film of particulate matter throughout the project. Occasionally, filters
with noticeable particulate matter and no coliform colonies after incubation
on M-Endo were transferred to LTB. None of these filters caused the produc-
tion of gas in LTB so, we do not have any evidence of particulate matter
interference on 100ml samples.
One liter samples were again used during the sampling of system MI.
This system uses well water with a very low pH and all the pipes in the
system are plastic. The turbidity of the water in the system is very low
and none of the liter samples produced a film of particulate matter on the
membrane filters. No colifonus were found in any of these one liter samples.
Le Chevalier et al. (1980) presented some evidence that some bacteria
which are enumerated by the standard plate count technique are antagonistic
to coliforms and may suppress their growth on membrane filters when present
in large numbers. They also cited references to three previous investiga-
tions which showed interference with coliform detection by some of the
standard plate count organisms. We found evidence of a negative correlation
between standard plate counts and coliform densities only for system CV and
even for that system the correlation was significant at the 5% level (Goshko
et al. 1981). Positive correlations significant at the 1% level between
standard plate counts and coliforms were found for systems WH, LB, and BL
using the Spearman rank correlation test (Goshko et al. 1981). Thus, what
we found was some evidence of an increase in the probability of finding
coliforms with increasing standard plate counts rather than any evidence of
interference. The standard plate count flora is probably different for
different water distribution systems, so it is not necessarily expected that
the relationship between coliform count and standard plate count would be
the same for all systems.
4.5 VERIFICATIONS OF PRESUMPTIVE COLIFORM AND OTHER COLONIES
Tables 6 and 7 in the various appendices give tha results of verifica-
43
-------
tions for individual systems by dates and locations. Both lauryl tryptose
broth (LTB) and brilliant green bile broth (BGLB) were used for verification.
The results from both broths were consistent throughout the project; i.e.,
all presumptive coliforms which produced gas in LTB also produced gas in
BGLB and, of course, those which did not produce gas in LTB did not produce
gas in BG.
The verification protocol described in section 3 was followed with a few
exceptions. The number of presumptive coliforms found in systems WH and LB
sometimes exceeded our capacity to make verifications and fewer than five
colonies were picked from filters which had many colonies. During the sam-
pling of syt,.-~n "1 several of the attempted transfers from the membrane
filters on M-Endo medium resulted In no growth. This occurred occasionally
in other systems. No explanation of this phenomenon was found. The occa-
sional misses on transfers could be due to technician fatigue, but the large
number of missed transfers on 10/23/80 may have been the result of some
material filtered from the water which interacted with a component of the
M-Endo medium to eventually inactivate the coliforms.
The presumptive coliform counts were reduced by the fraction of colonies
verified on a sample by sample basis to give the verified coliform counts.
In the cases when no coliform colony was isolated for verification for an
individual sample the presumptive coliform count was reduced by the overall
fraction of colonies verified for that sampling day. In the case of the LB
samples for 7/18/79 when no colonies were isolated for verification, the
overall fraction of colonies which verified for that week was used to adjust
the counts. Since all of the verifications attempted for all samples from
that system for July 1979 were positive, this amounted to assuming that all
of the presumptive conforms were verified.
Table 4.12 gives the overall fraction of coliforms which verified for
the various systems sampled. The percent verified changes from system to
system and from one sampling period to another for some systems. It appears
that the percent of the presumptive coliforms which verified is greater in
warm weather than in cold weather, but this not entirely consistent for
all systems.
Table 4.13 gives the verification results broken down according to the
API identifications and also according to the system and sampling period.
The data from system DT were further broken down into those which were col-
lected from the water distribution system (DTS) and those which were collect-
ed at the water treatment plant (DW). The coliforms found in the water
treatment plant were mostly from the raw water or the settled water. The
data from the other systems include coliforms from samples collected at the
water treatment plant as well as those from the distribution system.
Overall the most frequently identified coliform was Enterobacter cloacae
(271.isolates) followed by Enterobacter agglomerans (165), Klebsiella oxytoca
(96), Klebsiella pneumoniae (77), Citrobacter (42T7 and Escherichia coli (25).
The Escherichia coli isolates were from the DT raw water and Systems WH and
BL. The E. coli from system BL were from samples collected either at the
water treatment plant or from the elevated storage tank. The E. coli from
44
-------
TABLE 4.12
VERIFICATION RESULTS ON SYSTEMS SAMPLED
System
CV
WH
LB-BB
LB-TP
DT
BL
MI
SR
MR
BG
Sampling Verifi- Positive Negative Percent
Period cations Verifi- Verifi- Verified
Attempted cations cations
Feb. 1979
March 1979
I April-May 1979
II May 1979
III May- June 1981
I June 1979
II July 1979
I June 1979
II July 1979
Jan. -Feb. 1980
Water Plant
Dist. System
I March- April 1980
II June 1980
May- June 1980
Sept .-Dec. 1980
Oct. -Dec. 1980
Sept .-Dec. 1980
21
26
39
140
131
15
38
60
56
143
27
54
97
45
145
104
101
13
25
39
134
120
13
38
49
56
131
22
32
83
33
126
102
91
8
1
4
6
11
3
0
11
0
12
5
22
14
12
19
2
10
61.9
96.1
90.7
95.7
91.6
80.0
100
81.7
100
91.6
59.5
59.3
85.6
73.3
86.9
98.1
90.1
system WH were in samples collected at the treatment plant and in two speci-
fic sites in the community. These results tend to indicate that E. coli
does not survive in these water distributions very long. However, apparently
species of Enterobacter and Klebsiella do survive and probably reproduce in
these systems.
A few of the Enterobacter cloacae and a few of the Klebsiella pneumoniae
failed to verify as coliforms. These may be considered to be false negative
results.
By far the great majority of the presumptive coliforms which failed to
verify were Enterobacter agglomerans. On the other hand over half of the
Enterobacter agglomerans isolates did verify as coliforms. Are the Enterobac-
45
-------
co
O
M
H
CJ
M
fn
H
55
Ed
Q
h- 1
M
CO
*-H Q
-* 5
W CO
iJ 55
PQ O
H H
CJ
M
l_l
w
fe
0
SH
3
§
pj
co
1
CO
H +
Q 0-
f
£5
IH -J-
Q N— '
~
M
I-l
pq ~f-
iJ x-x
^
1
M
hJ ^-^
^
M 1
M
M
g +
1
1-1
f"^
£ +,
1
M
|T^
i
T
" +
N_ ^
^
i-H
•^* co
co
> CM m
CO CM CM
1 — 1
CO *3" ' — i A
«j- ^H r~- r~- co i— i CM
*vf" ^^ CO i— *
O •— I •— I
o CM m «* m >-i
i— I O A
i-H
O\ "^ *^ C^
0
rHOOcd tajscj-H in
bogoti cucxcdco -H
bo 3 4-1 cu a o y-i 3 -,a -H i-H O -H > i-l
Or-IHr-l4JrH CO MH-H <4H
Cdr-li-li-HOCdCdcOCd4JlH4JT)'OlH
(HCOCOCOO-Hl-ia-rH 0) 3 03 M-l >
0) *O ,0 vJd M £H CO O M * T3 CO i— 1 iH
pg^adGSiSlcSSIcS^II
46
-------
w
o
I-l
H
•-'
T
O
pQ >f-
v"
T
§i
^•^
1
toi
x-*\
i
u +
s ^
/_s
1
h-l
H
J +
PQ xlx
/_N
1
h-l
jj
"i
^^ «^- ^5 i-H f"** ^H ^^ CD ^^ *^ ^"^ ^~*
r^
i^ (-H oo ^^ r^ ^D ^^ ^^ ^H ^^ ^D ^if
^^ ^^ ^? ^^ ^^ ^^
CM -H
in m CT» i— * i— i o
m o o\ -H
CM i-l
CO i— 1 i— ' i-H
CO CM /\
i— ( ^H
r>.CM O COOsl-CMCM
CM CM -H Sf in O O
CM i-l i-H
i— ivO i— I i— i ON in ^C ON
-H «* CM ^H
cyx *^ CO /N
CM m m co CM
i-l CTi •* CO i— 1
O —I 1-1 — i /\
— < CM
I-H i— I O CM -3" ^ CO *3"
•— i co cj\ oo ~H
CM
vO •— I CM CO 1-1
CM CM i— I CM
\o m oo -* CM CM
1-1 ^f m co CM
CM
CO
c
•H co
•H M CU CO CO
CO ^ CO CO cO Cd &
CO CO S i-l CO iH CU CU M
ONOCCflCO -H -H -H > O
CO CO iH O O (0 T) J2 O -H 14-1
iH O ^i bO 0 O C3 CU ft CO CO *r4
O CJ CO (0 CU 4J <0 MMiHi-H O
O RMN 01 T3 3 O CO
MMMftOO -H 4J >i CT C t) *O
CO CU CU CU MCUCJ2-HOCUO) CU
•H4J4->4JCOcOcOCO>-H i-HOiH> i-l
43OUUiHiHiH4JH CO 14-11-1 U-l
O CO CO cOHrHi-l O Cfl CO CO CO -U -H *-> T3 T3 -H
•H^lja^CUCUCUcO •HCiHO4-lftCUCOM
MOO O-Hi-liH^QCO C O 4J !Z C 6 W -H CO
COMMMCOCOCQOiH-H0iH CU 3 Cfl U-l >
r£ CO CO CO v^ *A pO M C3 CO O M * *O CO T-^ *r"l
O 4-J 4-1 4-1 CU CU 0) 4J <4-l MM MQ*HCU OM4J
CO (3 C C >H r-j iH -H CO COCO CO • CM COCO O
wiaww!^J^!^;utiJ>-isz
47
-------
ter agglomerans which do not verify false negatives? Are the Enterobacter
agglomerans which do verify false positives? Is there any reason that some
Enterobacter agglomerans should be considered as microbiological contaminants
of drinking water while others are not? There seems to be a lack of corre-
spondence between the functional definition of the coliform group and a
taxonomic category which requires some further consideration.
Information on the API code number and reactions on various media of
the Enterobacter isolates is given in tables 4.14 and 4.15. The API manual
suggests dividing Enterobacter agglomerans into four biotypes based on the
reactions to indole (IND) and sucrose (SAC). The information in tables
4.14 and 4.15 are divided according to the API suggested biotypes with the
INIH- strains in table 4.14. The IMViC series of tests was carried out for
all of the INIH- Enterobacter agglomerans and these results are included in
table 4.15. It can be seen that the great majority of Enterobacter agglomer-
ans isolates which did verify as coliforms had IMViC type -H— which is
usually associated with E. coli.
The information on the E. coli isolates is given in table 14.16 and a
comparison of the test results used to differentiate between E. coli isolates
and Enterobacter agglomerans IMViC type -H— are given in table 14.17. It is
clear that there are distinct biochemical differences between E. coli and
Enterobacter agglomerans although the test differences are few.
Enterobacter agglomerans strains were previously considered to be mem-
bers of the genus Erwinia, but were renamed by Ewing and Fife (1972). Most
of them are plant saprophites but a few are opportunistic pathogens of man
and other animals. Their sanitary significance as indicators of contamina-
tion of drinking water is unclear.
Occasionally non-coliform colonies were picked from the membrane fil-
ters, incubated on M-Endo medium, and processed for verification and identi-
fication. The non-coliform colonies were described as atypical for red
colonies with a dark center but no sheen, red and clear. These results are
given in the appendicies. A summary of the results are given in Table 4.18.
Enterobacter agglomerans also turns up relatively frequently in Table
4.18. The numbers in the parentheses with a following + are the numbers of
the isolates which produced gas in both LTB and BG. Thus even some of the
Enterobacter agglerans which did not pruce sheen colonies on M-Endo media
would be counted as coliforms by the fermentation tube technique. Other
species of Enterobacter also sometimes produce non-sheen colonies on M-Endo
and some of those produce gas in LTB and BG.
The coliform group defined by laboratory methology does not conform
exactly to any taxonomic group. Members of the genera Escherichia, Entero-
bacter, Klebsiella, and Citrobacter usually are coliforms but not in every
case. Hafnia Aeromonas, Serritia, and Proteus usually are not coliforras but
their reactions to the coliform test procedures are sometimes close enough
that they can be confused with coliforms.
The coliform group, as defined by the laboratory test procedures has
48
-------
W
M
v^/
CO
55
% cu cu
•^ ,J3 <-! 4J 4.1 f^ ^
co co oo cd cd co co
co co
—i m
»^- *^
*^ *^
o o
1— 1 I— 1
1
^4
co
5
Ss
M
49
-------
T
M
CO
£2
<
M
S
o
B
o
W
O
W
t_l
W
<2
IT)
rH O
•* M
fa
W M
hJ H
pq fe
< w
H Q
CO
§
1 ^
o
CO
M
Pn
o
CO
2
o
M
EH
0
S
pE^
0
>>
CO
O CO
0)
M 4J
v cd
"e 'o
3 CO
a H
a
o
•iH
4J
cd
i PQ
O W
"o C
u o
o
•a
>i W
a i
0 0
O (3
U 0
CU
r-l M
•H 0)
H 0 S
Pn M S
5j c^ ss
Q)
cu
H 0
< PQ
COCOCOCOCOCOCOCOCOCOCOCOCOCOCO CO
cdcdcdcdcdcdcdcdcdcdcdcdcdcdcd cd
bObObObObObobObObObObObObObObO bo
CO CO
OOOOOOOOOOOOOOOcdOcd
cd
O 'Q *^3 trj t^ nj »rj ^rj »rj »rj trj
r-((i)ci) Q)a)
4J iH r-4 i-( 4J iH
td cd cd cd cd cd
*• O O O ^ O
CS *H C (3 «H *r^ CZ! C fS C3 *H Ca C C CJ CJ
C) (X d) CU O* OH Q) Cl) (1) O O< flJ O CU O O
-------
M
j 1
o
.
w
C/3
•"•
o
w
vD 1— 1
i-H H
co C
o cu
o C
CO CU
. 1 pTj
a"§
O Ed
i-l
o d
cj O
o
a
S-, W
a i
o s
0 C
0 0
01
iH i-i
•H CU
M-l J3
H O S
PM M 3
< CM S
cu
p.
4J
M O
cO ci cO
C cu i-l C cu
CU iH CO CU i-l
cu o co cu o
ȣ- 3 "^ ,^ 3
co C S co C
C3 d d C CH
CU CU CU CU CU
CU CU CU CU CU
ȣ* .^ f^ f^ f*-
CO CO CO CO CO
CM CM CM CM
m m r** in
i— i m m m
•^ *^" ^^ "d"
o o -H o
m co 4-1
CO CO
c a cu cu c
CU > cu a> cu cu
4_l n rt n f*.
CO CO CO CO CO
CM CM CM CM
r^« m r^ r^*
m i— i r- 1 m
-3" "^ -^ ^~
*^ H^ *^
-------
TABLE 4.17
COMPARISON OF TEST RESULTS BETWEEN ESCHERICHIA COLI AND
ENTEROBACTER AGGLOMERAN STRAINS WHICH ARE IMViC TYPE-H—.
Escherichia coli Enterobacter agglomerans
API 20E Tests
Similarities
Nitrate reduction + +
Urease - -
Motility ± ±
Gelatin liquification
g-Galactosidase + +
Tryptophane deaminase - -
Arginine dehydrolase
Glucose + +
Arabinose + +
Meliobiose ± ±
Sucrose ± ±
Mannitol + +
Rhamnose ± ±
Insitol ± ±
Sorbitol ± ±
Indole + +
Citrate
Differences
Lysine decarboxylase ±
Ornithine decarboxylase ±
Amygdalin - +
Other Tests
Malonate utilization - ±
Tartrate utilization +
KCN Growth - ±
Pigment formation (25°C) - ±
Growth at 44.5°C ±
been used for monitoring potable water systems for many decades. It has long
been known that, although coliform bacteria are present in large numbers in
sewage, there are non-sewage sources of some of the coliform organisms. The
lack of exact correspondence between the coliform group and varis taxonomic
categories has not been found to be a major impediment to the beneficial use
of the coliform group as a monitoring tool.
52
-------
en
w
55
O
O
o
fa
w
o
0
1
§
55
co fa
^H O
•
~* O
W H
hJ H
a "
H H
fa
H
55
O
M
M
O
SH
Pq
1
en
O
-
0
£
PQ
H
PQ
.J
W
£>
t^
HP
i-H
VO
CM
i-H
i-H i-H
p-H LO
^
^
1-1
^^
^.
CM
^
+ ^>
Ci i
CM ^
i— i i— i
?
^H
t— 1
CO
cu
CU (3
cd o>
u bo
cd o
•H O M
i-l tH CU
o u cd
u
M M
cd cu cu
•H 4-1 4J
J3 0 CJ
u cd cd
•H ,0 ,0
MOO
CU I-l M
^3 Q) 0)
0 4J 4J
CD C C
^x
rH
•
cu cu CM M cu si
4-> 4J Cd CU >
U U (X i-l 4-1 i-H CO
cd cd s H cj cd rt
,Q f O CU Cd (3
O O l-i -H ,O CO O
M M O CO O -H 0
CU CU ,Q M C O
4-1 4J H CU 4J >4-l M
C (3 P.J r-j -H cd 01
fy*j NJ ^c ^^ ^^ Q^ ^(
cp
^
m
~
1— 1
H
CS|
^^
i
^-^
•— 1 r-H
^
CO 00 CM ^H
i— 1 ^H
co CT\ in
CM
CO
a cd
CU i-H
CJ -H
CO fi
CU ft
CO MO
•H O 4J
iH 3 i-H
•H iH Cd
& M
-------
SECTION 5
MAXIMUM CONTAMINANT LEVELS FOR COLIFORM BACTERIA
AND THEIR RELATION TO FREQUENCY DISTRIBUTIONS
5.1 INTRODUCTION
The Maximum Contaminant Levels (MCL) for coliform bacteria in the
National Interim Primary Drinking Water Regulations (U.S.E.P.A. 1976) are
based on two rules. The rules considered here are those which apply when
the MF technique is used. The first, "the average rule," states that the
MCL is 1 coliform per 100ml "as the arithmetic mean of all samples examined
per month pursuant to § 141.21 (b) or (c)". The second or "maximum rule"
states that the MCL is exceeded when more than 4 coliforms per 100ml are
found in more than 1 sample when less than 20 samples are examined per month
or in more than 5% of the samples when 20 or more samples are taken per
month. Thus the MCL's are based on two factors of the frequency distribution
of coliform bacteria: the mean and the probability of obtaining counts >4
per 100ml. However these two factors are not independent of one another.
Such dependence is obvious in some instances but more subtle in others. In
this section we describe the theoretical dependencies of these two factors
and the empirical relationships based on our sampling program. Specifically
we address: 1) MCL relationships independent of frequency distribution, 2)
frequency distributions considered, 3) results of field sampling, 4) rela-
tionship between frequency distributions, sample size, and MCL's, 5) replica-
tion of samples, and 6) sample volume.
5.2 CHARACTERISTICS OF MCL'S INDEPENDENT OF SPECIFIC FREQUENCY DISTRIBUTIONS.
Either one or both of the two MCL's may be exceeded depending upon con-
ditions in the distribution system and the monitoring program. Table 5.1
presents hypothetical sampling results for each of these possibilities using
a sample size of 10. Results recorded as "too numerous to count" (TNTC) are
considered to be >80 per 100ml for calculating the means and variances.
For the results presented in rows a through e the average rule is exceeded,
but the maximum rule is not. The results in rows a through d may be consid-
ered as the minimum conditions needed to exceed the average rule, those in
row e represent the maximum variance result, and those in row f represent
the maximum average count which exceeds the average rule but not the maximum
rule. Overall, sampling results which exceed the average rule but not the
maximum would generally be characterized by: 1) a mean in the range of 1 to
3 with a variance approximately equal to or less than the mean; 2) a mean
>1 due largely to 1 or a few «5%) very high counts. In the latter case
the variance would be large.
54
-------
H
*
lO
w
pj
2
H
O
rH
CO
H
W
tvl
CO
w
i
CO
25
i
CO
••
^J
g
u
h*
P
[14
O
CO
53
O
M
H
rH
O
M
>
CO
O
H
P=j
>
Ptj
O
CO
fe
§
s
W
0)
u
a
"2 3
H TO
g%
0)
u
a
CO
M
CO
>
fi
2
s
4-1
c
3
o
o
TJ
a)
4J *
co o
u oo
•H A
TJ
C tTi
H f>.
J3 rH
4J rH
> O
rH
ca
a) a\
p, oo
cd r^
CO
vO
rH
S m
o <«•
CO
iw
O CM
T-\
O
S3 0
-a
-------
The results presented in rows g through i are examples in which the max-
imum rule but not the average rule is exceeded. In all cases most samples
contain zero coliforms per 100ml, and two samples contained >4 coliforms
per 100ml. With a sample size of 10, the number of possible results which
satisfy these criteria is more limited than those which exceed the average
rule but not the maximum rule. This may not hold if the number of samples is
greater than 20. In general, the sampling results which exceed the maximum
rule and not the average rule may be characterized as: 1) when a preponder-
ance of samples will have no coliforms; 2) when less than 20 samples are
examined no more than 4 will have >4 coliforms per 100ml; 3) when 20 or
more samples are examined more than 5%, but not more than 20% will have >4
per 100ml; 4) when the percentage of samples with >4 per 100ml is large,
the counts of most if not all will be equal to or slightly greater than 4
per 100ml; 5) when the percentage is only slightly greater than 5%, the
counts per 100ml may be much greater than 4 for at least one or a few sam-
ples. If one sample is recorded as 80 per 100ml, and the maximum rule is
exceeded, a minimum of 96 samples are required such that the average rule
may not be exceeded.
Both rules may be exceeded in a large number of cases. The results
presented in rows j through m slightly exceed the rules and row n represents
the opposite extreme. The variety of results by which both rules are exceed-
ed is too great to summarize adequately in a brief table.
Some of the results shown in Table 5.1 are quite improbable (e.g., those
in rows a, e, and n). A major purpose of the research described in this re-
port has been to determine the type of frequency distribution of coliforms
most commonly found in water distribution systems and to relate this to the
MCL's and sampling protocol.
5.3 PREVIOUS WORK ON FREQUENCY DISTRIBUTIONS OF BACTERIA.
5.3.1 Distributions Considered
Five frequency distributions are considered for fitting with sampling
results. They are 1) Poisson, 2) Poisson plus added zeroes, 3) Lognormal, 4)
Delta, and 5) Negative binomial. Previously, the Gamma-Poisson distribution
has also been used (Muenz 1978) but it is essentially the same as the nega-
tive binomial. The following is a discussion of the attributes of each.
5.3.2 Poisson Distribution
Pipes and Christian (1978) elaborated on Muenz's approach to evaluating
MCL's using simpler frequency distributions: the Poisson and the Poisson plus
added zeroes. These simulations were extended somewhat during the first year
of the project.
The simplest case occurs when the coliform bacteria are randomly dis-
persed in the distribution system and the counts on some number of samples
would fit a Poisson distribution. This may not be a realistic case, but it
forms a lower limit on the number of samples needed to have any given prob-
ability of detecting coliforms in the system because it is assumed that the
56
-------
variance is equal to the mean.
If it is assumed that a mean density of one coliform organism per 100ml
is a realistic and significant goal, then questions about the relationship
between the sample mean and the actual mean density in the distribution
system become important ones. Figure 5.1 presents plots of the probability
of the sample mean being greater than 1 per 100ml versus number of samples
per month for various true mean densities in the system ("x is the sample
mean and p is the actual mean density). For an actual mean density of 1 per
100ml, the probability that the sample mean will exceed 1 per 100ml is low
for a small number of samples but increases to almost 50% as the number of
samples increases. A 50% probability of the sample mean being greater than
1 per 100ml means that the water distribution system would fail to meet the
regulation 6 out of 12 months on the average. Although the water supply
would be judged safe some months, the failure rate over a period of time
certainly would not be accepted and adjustments would have to be made to
provide better treatment. The failure rate at one sample per month is 26.4%
or approximately one month out of four. Even this failure rate would prob-
ably not be accepted, so in the long run one sample per month would be ade-
quate to detect a mean coliform density of 1 per 100ml if the bacteria were
randomly dispersed in the system. From the two upper curves on figure 5.1,
it is clear that a mean density of greater than 1 per 100ml would not be
accepted even if only a few samples per month are taken.
The other side of the issue is the possibility of judging a water system
with an actual mean coliform density of less than 1 per 100ml as unaccept-
able. From figure 5.1, it can be seen that if the actual mean density is
0.5 per 100ml and only 1 sample per month is taken, the probability of that
sample having more than one coliform per 100ml is about 9% which means that
such a system would fail to meet the regulation about once a year. As the
number of samples per month is increased the chances of judging a water
system with an actual mean density of less than 1 per 100ml as unacceptable
decreases rapidly.
Figure 5.2 gives the probability of the samples from a water distribu-
tion system failing to meet the maximum rule for coliform bacteria randomly
dispersed in the system with actual mean densities of 1, 1.5, and 2 per
100ml. The functions are "saw toothed" because of the integral nature of
the samples; that is, 2 samples with 4 or more coliforms is more than 5% of
39 samples per month but it takes 3 samples to be more than 5% or 40 samples
per month. Comparison of figures 5.2 and 5.3 shows that the average rule is
a more stringent criterion than the maximum rule if the coliform bacteria
are randomly dispersed in the distribution system.
Another way of considering the question is to ask what is the probabili-
ty of obtaining some number of negative samples (no coliforms present) if the
actual mean density is 1 per 100ml or less. The left most curve on figure
5.3 gives the probability of obtaining no positive samples for various num-
bers of samples for an actual coliform density of 1 per 100ml if the coli-
forms are randomly dispersed in the system. For such a condition the chances
of obtaining all samples as negative when more than 2 samples are taken per
month is slightly less than 5%.
57
-------
40
60 80
NUMBER
100 120 140
OF SAMPLES
160 180 200 220
Figure 5.1 Probability of Exceeding Average Rule for Random Dispersion
of Coliforms at Different Actual Mean Densities
58
-------
a
o
•H
co
M
0)
fX
CO
s
o
T3
a
PS
n
o
K-l
-------
The overall conclusion from this very simple case is that if coliform
bacteria were randomly dispersed in a water distribution system, they would
be relatively easy to detect with only a few samples. It would make no
difference where the samples were collected and the size of the distribution
system would not influence the number of samples required. Unless the actual
mean coliform density in the system were considerably less than 1 per 100ml
the sample mean would fail to meet the Regulations frequently enough that
the operator would be forced to improve the disinfection practices.
5.3.3 Poisson Plus Added Zeroes
The next simplest assumption which can be made about the dispersion of
coliform bacteria in water is that some of the water is clean, only a part is
contaminated (coliforms present) and the coliforms in the contaminated part
are randomly dispersed. The mathematical equivalent of this assumption is
the compound binomial-Poisson distribution also called the Poisson plus
added zeroes. This is a two parameter distribution. The parameters are the
fraction of water contaminated and the mean coliform density of the contam-
inated fraction. In the earlier paper (Pipes and Christian, 1978) the frac-
tion of water contaminated was represented by 1 - 5 (the Greek letter zeta
representing the fraction of water not contaminated) and the mean density
of the contaminated fraction was \. The overall mean density of the water in
the system is y = A(l-£).
Figure 5.3 gives the probability of obtaining no samples with coliforms
versus numbers of samples collected for various values of £ with y = 1 per
100ml. To keep y = 1 per 100ml it is necessary to vary X with ^; for instance
when r, = 0.5, X = 2 per 100ml, when 5 = 0.9, X = 10 per 100ml, etc. The left
most curve for 5 = 0 and \ - y = 1 per 100ml is the random dispersion of
coliforms throughout the distribution system as previously discussed. From
Figure 5.3, it may be seen that with 50% of the water contaminated it is
necessary to collect at least 6 samples to have a 95% probability that at
least one sample will have coliforms, with 10% of the water contaminated 30
samples are needed, with 5% of the water contaminated 60 samples are needed,
and with 1% of the water contaminated 300 samples are needed. From this it
is clear that it is never possible to take enough samples to be absolutely
sure that there are no coliforms in the system. It is possible to discuss
the probability that some small fraction (such as 5% or 1%) of the water is
contaminated even for a large number of samples with no coliforms.
This model behaves the same in relation to the average rule as does the
random dispersion model for y = 1, it just requires more and more samples for
the probability of the sample mean to approach 50% as 5 gets larger and larg-
er (Pipes and Christian 1978). Also if y>l the probability of the sample
mean being greater than 1 approaches 100% for large numbers of samples.
5.3.4 Assumptions Underlying the Simplistic Models
In these simplistic models bacteria are treated as if they are single,
inert beads floating around individually in the system. Bacteria are not
inert. If a chlorine residual is present, they will die-off in the system.
Some are able to multiply in the system. They may occur as clumps of two,
60
-------
td
4-1
u
t-l
o
CO
0)
en
0)
to
o
00
C
•H
OJ O
M
(1)
CO
O C
V4 (U
fi< p
co
CO N _ (£> IT) ^-.
dodddo
S31dlNVS 3AlllSOd ON JO
TO
o
(\J
d
o
61
-------
three, four or more cells adhering together. If the bacteria are growing
slowly, it is more likely that the bacteria will occur as clumps rather than
as single cells. Also in the presence of chlorine, it is likely that bac-
teria in clumps would survive longer than individual cells. If the bacteria
are in clumps widely dispersed in the distribution system and it is assumed
that the shaking of samples required before analysis breaks up the clumps,
it would be expected that most of the samples would have no coliforms and
that the samples with coliforms would have several.
The second assumption underlying these simplistic models is that one
frequency distribution can describe the dispersion of coliform bacteria
throughout the system. There is no particular reason to believe, a priori,
that this should be true. It has been considered as a hypothesis to be
tested. If the hypothesis were confirmed that the coliform counts from an
entire system did fit a single random frequency distribution, this would
suggest a single mechanism of contamination. However, distribution systems
with different pipe materials in different areas, or different ages of pipe
or differences in pressure probably are subject to different mechanisms of
contamination. A more complex frequency distribution may be observed if
this is the case.
The same argument can be made with respect to the period of time over
which the samples are collected. Time is not a variable in the simplistic
models. No differentiation is made among samples collected over an hour's
time or a day or a week or a month. The dispersion of coliform bacteria in
a water distribution system may change with time. One of the aims of this
project has been to determine the periods of time over which the microblal
quality of water in a distribution system is likely to remain constant and
attempt to identify significant occurrences in the operation of the distribu-
tion system which could cause changes in the level of contamination and the
dispersion of the coliforms in the system.
The simplistic models provide some insight into the general relationship
between the actual mean density of coliforms in a water distribution system
and the sample mean as a function of the number of samples collected. If
the actual mean density in the distribution system is 1 per 100ml and enough
samples are taken, the sample mean density will be greater than 1 frequently
enough that the disinfection practices will have to be improved. "Enough
samples" may be very few (3 or 4) if the coliforms are randomly dispersed
throughout the system or several hundred if only a small fraction of the
water is contaminated. In these simplistic models, the number of samples
needed is independent of the size of the distribution system. However, if
the microbial quality of the water varies from one part of the system to
another, a dependence between the number of samples needed and the size of
the system might be shown. A point which has been investigated is how large
a portion of a distribution system will show consistent water quality and
what features of the system could be used to identify areas of different
water quality. It is also clear that the time period over which the micro-
bial quality of the water will remain consistent and the types of changes in
the operation of the system that might signal possible changes in microbial
quality of the water are important in designing a sampling program.
62
-------
5.3.5 Lognormal Distribution
The lognormal or variations of the lognormal distribution were investi-
gated during our study. In the lognormal distribution the arithmetic mean
is dependent on both the geometric mean and geometric standard deviation.
This dependence is exponential rather than linear such that relatively small
changes in the geometric standard deviation may produce large changes in the
arithmetic mean. This may be seen in Figure 5.4. In this figure the arith-
metic mean is plotted within the plane formed by the log geometric mean and
log geometric standard deviation. In this case the line designated as "1"
is the arithmetic mean density of 1 per 100ml. From the slopes of the
lines it can be seen that the geometric standard deviation plays a larger
role in establishing the arithmetic mean than does the geometric mean. Thus
with a lognormal distribution the variability between samples is important
in determining whether the average rule is exceeded. In this distribution
samples with large densities are to be expected to occur with a probability
not much less than that for low densities. Thus the probability of exceeding
the maximum rule increases with an increase in the geometric standard devia-
tion. This is shown in Figure 5.5. While the geometric mean remains con-
stant the probability of obtaining counts >4 increases as the geometric
standard deviation increases. The discussion above is theoretical and based
on an infinite number of samples. The relationships as functions of number
of samples examined are discussed later.
For the lognormal distribution the "average rule" is more stringent than
the "maximum rule" when a large number of samples are taken. That is the
probability of exceeding the "average rule" is always greater than that of
exceeding the "maximum rule." This is shown graphically in Figure 5.6. This
may not occur with small sample sizes, as shown later but does indicate
redundancy of the rules.
The lognormal distribution assumes no limits as to the upper and lower
boundaries of coliform densities. Obviously, however, these limits exist
and may be recognized at two levels: first, the actual limits of coliform
densities and, second, the sampling limits. The upper limits of coliform
densities are the maximum densities which could occur in the water. If coli-
forms are growing in the water, this could be considered the carrying capac-
ity of the population dependent on the nutrient availability in the water,
the growth yield of the bacteria, and physical and time constraints on growth
such as temperature and residence time of the water in the system. If coli-
forms recovered in the water have grown elsewhere (e.g., pipe walls or ground
water) the carrying capacity of the growth location and entry mechanism must
be taken into account. The actual lower limit is zero for the entire volume
of water in the system.
The detection limits depend on the volume of samples. If a standard
volume of 100ml is used, the lowest positive count is 1 per 100ml. The
upper limit is 80 per 100ml (Bordner et al 1978). To assess the lognormal
distribution the data are considered truncated above and below these limits.
Samples with 0 per 100ml are considered as <1 per 100ml, and TNTC samples
are considered as >80 per 100ml. No distinctions are made as to how much
lower or higher than these limits the densities may be. These limits may be
63
-------
O
o
o
rO
,0
CD
«O
o
o
o
o
o
g
x) co
§41
14
CO CU
a *
co
O ,0
•H O
M CO
4J H
O •
a) co
M a
o
CU -H
O
t-
CO
•H
LU ^,H
c, Q 2
<0 H
^ § I
tZ o o
CO ^H r-l
O
•H
CO
tr -H «
i_ £ O ;
I Cd iH CO
— xi cd cd
CU -rl 4)
jj > g
- 5
•o o
O
UJ
O T3 -U
O 1-1 CU
O
o
•H
O *J
-H C
fc CU CU
•H a) in
Q 60X1
•H
64
-------
50r
I 10 100 1,000
GEOMETRIC STANDARD DEVIATION
Figure 5.5 The percentage of samples with >_ 4 coliforms/lOOml as a function
of the geometric mean (Gm) and geometric standard deviation.
65
-------
CM
cn
CM
E .
^
•i «
x — i
0 2
E
\
Nk^/'
^/
/
(
\
'\
& ^
c 00<^
^- c o,
CM — CO
^ CM e
+ _^ —
CTI .
IX C^
|X CT> \
C |X ~
"«, ^ Al
ii , . CP
^ Al ix /
Al e - /
_J /
U /
S /
/
0) /
Qi /
E /
^_ / ^ -
o / ^-^^"^
C .' ^^^ '
(/) / ^.^^^
i-
o
_l
O
2
(/)
^—
O LJ
— Q
Q
o:
^•<
Q
t-
0 ^
O
VL
1-
LJ
^
O
UJ
o
_
o o o
0 P
o
f\
H
cd
O 4J
0 0
00 O
5 B
1— 1
n
^( Q)
o a
0
to a)
flj AJ
in rt
5 w
0)
I M
•H 0)
y Q)
S a
T3 tfl
M-l
0) O
00
cd t-i
M (U
« 3
0
0)
,0 (U
^i
60 nJ
0 H
•H
-a a)
(U 0
o a)
X ,C
(U S
M— i CO
0 0
o
-a -H
o -u
o s
•H -H
rH M
tu -u
^«J to
t-^l 13
•
LO
(U
M
60
•H
X) NV3W Omi3IAI039
66
-------
extended by multiple and larger volume sampling as will be discussed later.
5.3.6 Negative Binomial Distribution
Muenz (1978) assumed that the dispersion of coliform bacteria in a water
distribution system could be described by a compound gamma-Poisson frequency
distribution which is mathematically equivalent to using a negative binomial
distribution. Thomas (1952) suggested the use of the gamma distribution for
averaging coliform data from water distribution system samples and later
(Thomas, 1955) showed that the negative binomial distribution was equivalent.
Pipes et al (1977) found that coliform counts from a raw water supply would
fit either a negative binomial or a lognormal distribution. Recently, El-
Shaarawi et al (1981) used the negative binomial distribution to describe
bacteriological data from Lake Erie. The negative binomial distribution is
similar to the Poisson distribution but has a variance much greater than the
mean. Muenz (1978) interpreted a high variance as indicating greater hetero-
geneity of the water distribution system with respect to microbial water
quality. Part of his results are reproduced in Figure 5.7. In his terms,
"probability of acceptance" means the probability of obtaining a sample mean
of less than 1 per 100ml for n samples and the "coefficient of variation" is
the ratio of the variance to the mean. If the true mean coliform density of
the water is 1 per 100ml, the sample mean will be less than or equal to 1
only about half the time if 20 samples per month are collected. However, if
fewer samples are collected the chances of getting a sample mean of less
than or equal to 1 per 100ml are greater especially if the variance is high.
The negative binomial distribution is commonly used to describe clumped
or aggregated patterns of dispersion. The negative binomial may be derived
as a compound distribution. It models a random or Poisson distribution of
clumps or clusters in which the pattern of coliforms in each clump is log-
rithmic (Pielou 1969). The negative binomial has three parameters but,
since the three parameters are interdependent only 2 are needed to describe
the distribution completely. These parameters are the mean, variance, and a
measure of aggregation (k). The true mean and variance are estimated well
by the simple calculation of arithmetic mean and variance. The measure of
aggregation (k) is derived from the mean and variance such that as the degree
of aggregation increases, k decreases. As k becomes large the negative bi-
nomial distribution merges with the Poisson distribution. Figure 5.8 shows
the relationship of the probability of obtaining samples with various den-
sities (0,1,2,3,>4 coliforms/lOOml) as the log k is varied for a true mean
of 1 per 100ml. When k is small most samples have 0 coliforms and the prob-
ability of obtaining counts >4 is comparable to that of obtaining a count
of 1. This is similar to what was found for the lognormal distribution. As
k increases, the probability of obtaining a positive sample increases, but
most positive samples would have a low density (e.g. 1 or 2/100ml). This is
similar to the Poisson distribution. Thus, the flexibility of the negative
binomial is demonstrated here.
5.4 DATA FITTING
The following frequency distributions were examined: Poisson, Poisson
plus added zeroes, lognormal, and negative binomial. The following is a
67
-------
LU
o
I
Q-
LJ
O
o
o
t
-J
CO
CD
O
cr
Q.
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.5
0.2
O.I
0 O.I 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
COEFFICIENT OF VARIATION FOR PRIOR
DISTRIBUTION OF THE MEAN
Figure 5.7 The Probability of Obtaining a Sample Mean of 1 for a True
Mean Density of 1 as a Function of Number (n) of Samples and
Variability of the System, (after Muenz, 1978).
68
-------
I.Or
-I 0 0.3 I
LOG K
Figure 5.8 Probability of obtaining a particular count (X) when p=l and
K is varied from 0.1 to 10 in a negative binomial distribution.
X=0, 1, 2, 3 > 4.
69
-------
discussion of the applicability of each.
5.4.1 Data Sets
Nine water systems were sampled during the project's three years.
Information on sampling each system is summarized in Table 5.2. More infor-
mation on each is presented in the Appendices. Total coliform results from
each system were used to assess the applicability of the various frequency
distributions. In some cases the sampling results were divided into shorter
periods of time for analysis of time dependent changes in water quality.
System LB was divided into two major areas on the basis of separate well
water sources. These areas are denoted as Brant Beach (BB) and Terrace (TP).
5*4.2 Poisson Distribution
The simplest assumption as to the frequency distribution of coliforms is
that they occur randomly within the system. This is a Poisson distribution
in which the mean coliform density is equal to the variance. A test of the
applicability of this distribution is the testing of the null hypothesis of
equality of mean and variance. This is accomplished by a ^ test using
Fisher's Index of Dispersion (D ) where D = (n-l)S /~x. The term n is the
number of samples, S^ is the sample variance, and TT is the sample mean.
Another method of testing the hypothesis that the coliforms are randomly
distributed is the KolmogorovSmirnov test for goodness of fit (Sokal and
Rohlf 1969). In this test the observed cumulative frequency distribution is
compared with the expected Poisson cumulative frequency distribution based on
the sample mean. The maximum difference between the two curves divided by
the number of samples (Dmax) is the test statistic. Each of these procedures
were used to test the applicability of the Poisson distribution. The number
of samples used for these analyses was less than the total number taken.
Only locations in which a constant number of replicates could be analyzed
were used. A minimum of duplicate 100ml samples per location was used to
restrict unreplicated locations. In systems SR, MW and BG the replication
was 2-100ml and 2-200ml samples. Each replicated location was used as a
sample for computation.
The results of the analyses for the Poisson distribution are shown in
Table 5.3. Mean coliform densities ranged from 0.11 to >9.38 per 100ml
including 9 systems in compliance and 6 in violation of the average rule.
Both methods of testing failed to accept the hypothesis of a Poisson distri-
bution of confirmed total coliforms in any system. Thus, calculations as to
the effectiveness of sampling in determining bacteriological water quality
should not be based on the assumption of randomness.
5.4.3 Poisson Plus Added Zeroes Distribution
The second frequency distribution considered assumes that some portion
of the water is devoid of coliforms and that in portions of water containing
coliforms the bacteria are dispersed randomly. To test this hypothesis the
Fisher's Index of Dispersion was used on the means and variances of locations
which gave positive coliform results. The results of these analyses are shown
in Table 5.4. The fractions of uncontaminated water ranged from 0.585 to
70
-------
w
r^
•3
CO
g
CM H
• CO
m {H
co
w
rJ &
% B
H g
tv.
HH
o
M
s
g
CO
co
?-H
0
f*.
I
CO
T3
CO -H
C rH
O CO CO
MH iH > 0)
O -U rH
cd j3 cx
. O 4J 0
O O -H cd
55 rJ 15 CO
00
c
•H
rH CO
ft a)
§ ti
cd cd
co O
CO
00 C
a o
•H -H
rH CO
O *i-4
l_u m
CO Q
G
a>
4J
CO
ts
CO
> 5
u 5
m vo
CM vO
CM
CT\ CTi
PS. PS.
m ro
rH ^s.
•^s, m
n i
1 -st
H CM
•s-s» *Ss^
CM -st
M
CO
4J
jd
00
iH
0) a)
rH W
rH
iH >>
> rl
CO 3
0) .0.
W TJ
cd o
0 0
u &
H
H M
HI 1 LJ . |
• " r^ ^
M
S w pq pq
3 3 pq pq
-* O CM a,
in P^ CTI CTI
rH rH
CT<
CJ> rH Oi ps.
ps. co r*. ^s.
^s, ^s, -^ o>
P- sf rH rH
rH — CM —
^s, VO >s^ fs.
m i vo i
1 CM 1 O
00 CM PS. rH
Ss^ s^ s^ s^^
m m vo p~-
M
H H
o
cd
cu
pq
4J
d
H cd
H M M
M H pq
T3
d
cd
rH
CO
H
X
U
cd
vOO 1 ~-s m 00 ^s. rH
rHrHCTis3-CM-sv.CM«s^
CM ~-s. P-. CM UlrHCM
^s. ps. ^s, -V. VO 1 rH
vO 1 I-* -st 1 vO VO 1
1 O CM 1 vO 1 rH 00
f-s.rH~-s.-strHvO-ss.~-,
— . --s. CM ~s, — s~-s.o O
vOps-HrOvOinrHrH
H
1— 1 M
0)
O
Cd
rl
(_|
CU M
H H H
CO
T3
Q
O
JS
d d
9 a o
O d 3 W
*J ? Pi rH
oo cd >> I-H
C rH 13 00 Cd
•H ^ M d J2
do IH co
? O M »-i
O rl 4J ft Cd
Q pq g en g
u
pq
vO
«st
O
00
CM
CM
rH
in
CM
CTi
i
d
cu
i— i
o
T3
M
O
M-l
T)
cd
rl
pq
71
-------
TABLE 5.3
APPLICABILITY OF THE POISSON DISTRIBUTION TO THE
FREQUENCY DISTRIBUTION OF CONFIRMED TOTAL
COLIFORMS IN WATER DISTRIBUTION SYSTEMS
System Number of Mean Density3 Variance3
Locations per 100ml
Sampled
Dmaxc
cv
WH I
II
III
BB I
II
TP I
II
DT
BL I
II
MI
SR
MW
BG
225
66
154
170
92
99
122
146
174
238
169
207
144
99
46
0.11
0.40
> 1.52
> 1.76
0.51
> 9.38
> 2.58
> 6.16
> 0.30
> 0.35
> 0.50
> 0.40
> 0.93
0.41
> 2.15
0.67
3.62
> 25. 213
> 72.95
15.41
> 439
> 59.06
> 274.9
> 9.40
> 26.89
> 23.24
> 30.92
> 25.72
3.41
> 69.73
1364
588
> 2538
> 7004
2750
> 4587
> 2770
> 6471
> 5421
> 18208
> 7809
> 15924
> 3955
809
> 1459
0.956
0.909
0.677
0.869
0.880
0.789
0.766
0.801
0.954
0.967
0.893
0.986
0.840
0.818
0.813
3The > symbol has been used when at least one sample was "too numerous
to count." The density for such a sample was considered >80 per
100ml.
systems had a probability of having a Poisson distribution of
p<0.005 when Fisher's Index of Dispersion (D2) was analyzed.
CA11 systems had a probability of having a Poisson distribution of
p<0.01 when the Kolmogorov-Smirnov goodness of fit test (Dmax) was
analyzed .
72
-------
TABLE 5.4
APPLICABILITY OF THE POISSON PLUS ADDED ZEROES
DISTRIBUTION TO THE FREQUENCY DISTRIBUTION OF
CONFIRMED TOTAL COLIFORMS IN WATER DISTRIBUTION SYSTEMS
System
CV
WH I
II
III
BB I
II
TP I
II
DT
BL I
II
MI
MI
SR
MW
BG
Number of
Locations
(Number
with
Coliforms)
225(10)
66(6)
154(61)
170(26)
92(11)
99(41)
122(41)
146(53)
174(8)
236(8)
169(18)
207(7)
207(7)
144(36)
99(19)
46(10)
Fraction of Mean Density3 Variance D^b
Locations Not per 100ml of of Contam-
Contaminated Contaminated inated
Water Water
0.956
0.909
0.604
0.847
0.880
0.585
0.664
0.637
0.954
0.966
0.893
0.966
0.966
0.750
0.808
0.783
2.52
4.42
> 3.84*
> 11.49
4.27
> 22.65
> 7.67
> 16.99
> 6.56
> 10.44
> 4.67
> 11.89
> 11.89
> 3.73
2.15
> 9.90
10.03
25.84
> 55.26
> 337.05
122.52
> 767.52
> 138.62
> 579.42
> 185.39
> 790.03
> 209.03
> 902.08
> 902.08
> 94.35
14.59
> 263.43
40
29
> 863
> 820
287
> 1355
> 723
> 1773
> 198
> 530
> 761
> 455
> 455
> 885
122
239
aThe > symbol has been used when at least one sample was "too numerous to
count." The density for such a sample was considered >80 per 100ml.
systems had a probability of having a Poisson plus added zeroes dis-
tribution of p<0.005 when Fisher's Index of Dispersion (D^) was analyzed.
73
-------
0.966. None of the frequencies of distribution of coliforms could be satis-
fied by the Poisson plus added zeros distribution. The added assumption of
a portion of water devoid of coliforms was not sufficient to explain the
dispersion of bacteria. The coliforms in contaminated water were not random-
ly dispersed. The fact that the D^'s in these analysis were larger than
could be explained by randomness indicates that the bacteria are aggregated
in some fashion. The probability of obtaining high counts (>4 per 100ml)
in samples is greater than would be predicted by the two distributions invok-
ing randomness. Again the calculations based on a simple model are inade-
quate to describe the effectiveness of sampling.
5.4.4 Lognormal Distribution
The lognormal distribution is useful in fitting data which are aggre-
gated and which have samples containing both small and large densities.
This frequency distribution has been used previously for the analysis of
coliform data (Velz 1951, Pipes et al 1977). The graphical approach using
log-probability paper provides a quick method of checking for a fit. Ait-
chison and Brown (1957) discuss the rationale underlying this method in
some detail. Figure 5.9 gives two log-probability plots of the Coatesville
coliform data based on individual samples rather than locations. There are
two approaches to analysis of these data. One is to assume that some frac-
tion of the water is not contaminated with coliform bacteria (in this case
0.9781 of the water) and fit the positive counts to a lognormal distribution.
This is tantamount to assuming a lognormal plus added zeroes or delta distri-
bution. This approach is represented by the line labeled "percent of 10
positive samples." The mean log for the 10 positive counts is 0.457 giving
a geometric mean of 2.86 per 100ml and the log of the standard deviation is
0.522 giving a geometric standard deviation of 3.32. The leftmost line of
Figure 5.9 passes through the geometric mean at the 50 percent point and has
a slope corresponding to the geometric standard deviation of 3.32. This ap-
proach requires three parameters to describe the frequency distribution of
coliform counts: namely, the fraction of the water contaminated, the geo-
metric mean, and the geometric standard deviation for the contaminated frac-
tion.
The other approach is to assume that all of the water in the distribu-
tion system is contaminated and the coliform counts fit a lognormal distribu-
tion, but counts of less than 1 per 100ml are not observed. This approach
is represented by the line on Figure 5.9 labeled percent of all 457 samples.
This line is replotted on a different scale on Figure 5.10 so that it can be
extrapolated to the 50th percentile which represents the geometric mean (3.8
x 10~° in this case). For this type of plot the ratio of the 84.1 percentile
value to the 50 percentile value is the geometric standard deviation (in this
case 1.45 x 10~3/3.8 x 10~6 = 382). This approach describes the frequency
distribution of coliform counts using only two parameters.
The lognormal distribution represents the frequency distribution of a
continuous variable. Bacterial counts are discrete rather than continuous,
but bacterial density is actually a continuous variable. A bacterium may
occur in 150ml of water, or in 386ml of water, or in 6.84 liters of water,
etc. These occurrences would give bacterial densities of 0.667, 0.259, and
74
-------
30
20
1 '5
o: 10
UJ
^8.0
o
o
e
6.0
4.0
3.0
2.0
1.5
1.0
PERCENT OF 10
POSITIVE SAMPLES
^PERCENT
OF ALL
457 SAMPLES
GEOMETRIC
MEAN = 2.9
10 20 30 40 50 60 70 80 90 95 98 99 99.5
PERCENT OF COUNTS EQUAL TO OR LESS THAN
99.9
Figure 5.9 Log-Probability Plots of Coatesville Coliform Data
75
-------
10
o 10"'
cr
LU
a.
^ 10-2
8 |0-3
a:
o
^ IO'4
o
o
10
" 5
/*^-84.l PERCENTILE 11.5 X to"3
GEOMETRIC
MEAN
3.8 X 10
_6
10 20 3040506070 80 90 95 98 99 99.8
PERCENT OF COUNTS EQUAL TO OR LESS THAN
Figure 5.10
Extrapolation of Log-Probability Plot of Coatesville Data for
Fitting a Truncated Lognormal Distribution.
76
-------
0.0147 per 100ml respectively. If a 100ml sample happens to include a coli-
form, a count of 1 per 100ml is produced, but there is no information about
how much more water from that sampling location would need to be examined
before another coliform would be found. The practical realities of the
sampling and sample processing procedures have imposed an artificial dis-
creteness on the data. Densities of less than 1 per 100ml are not observed
but they almost certainly exist. When certain values of a lognormal variable
cannot be observed the frequency distribution is said to be "truncated"
(Aitchison and Brown, 1957). Truncation may also occur at densities above
the range studied (i.e., >80 per 100ml).
It can be seen from Figures 5.9 and 5.10 that either approach gives a
reasonable representation of the data. Under most circumstances the three
parameter approach gives a better fit, but the two parameter approach would
be preferred if it gives a reasonable fit because fewer data are needed to
estimate the parameters. For this problem there is an additional implica-
tion. If it is assumed that some fraction of the water is not contaminated,
this implies that in certain parts of the system taking larger volume samples
or taking more samples would not increase the number of coliforms found.
If, on the other hand, it is assumed that all the water is contaminated but
the variability of the countj gives some 100ml samples with no coliforms,
this implies that taking larger volume samples would give a greater probabi-
lity of finding coliforms and of getting higher coliform counts (if the
problem of interference could be overcome).
While the graphic approaches described here are useful in providing a
quick estimate of parameters, they are subjective. They rely on an "eyeball"
fit of the data by the investigator for the construction of the best fit line
and hence the geometric mean and geometric standard deviation. Also, no
estimation of variability around the line is easily obtained. A more objec-
tive fit may be obtained by regression analysis. This is achieved by the
following steps:
1) The coliform densities are ordered from least to most. As multiple
samples were taken per site and these replicates may not be consi-
dered as independent of one another, we will consider the replicate
samples from a site to give one density measurement. To determine
density per site, the total volume filtered in 100ml units is
divided into the total number of coliforms for each replicate.
The mean of the replicates is then determined. Therefore all
densities are given as coliforms per 100ml. This technique can
provide fractional densities of coliforms. For example, if 2
replicate 100ml sample have counts of 0 and 3, the total density
is 3 per 200ml or 1.50 per 100ml.
Sites with no coliforms are not considered zero but rather
some density below the minimum observable density. Following our
previous example 1 per 200ml or 0.50 per 100ml would be the mini-
mum observable density. Thus, sites with no coliforms are consi-
dered to have a density of <0.50 per 100ml. Truncation is said to
occur at 0.50 per 100ml. All sites considered must have the same
volume filtered.
77
-------
If a replicate sample from a site contains >80 per 100ml it
is too numerous to count. Truncation must occur at this high
density because of the uncertainty of the true density of any such
replicates. The density at which such truncation occurs is depen-
dent on the data set. The lowest density in our example at which
this can occur is >80 per 200ml or >40 per 100ml. However, a
TNTC is often associated with other positive counts (e.g., 5, >80).
This would make truncation at >85 per 200ml or >42.50 per iOOml.
2) Once the densities per site have been ordered, the percent of sam-
ples < each density is computed based on n + 1 as 100%. An example
of this calculation is shown in Table 5.5. When a density between
truncations is represented more than once the percent is computed
for each (not shown).
TABLE 5.5
EXAMPLE OF COMPUTATION OF FACTORS TO BE USED IN THE
ASSESSMENT OF THE LOGNORMAL DISTRIBUTION BASED ON DATA FROM WH I
Number of
Sites
60
2
1
1
1
1
Density per 100ml
<0.5
0.5
1.0
4.5
6.5
13.5
% < density
90.0
92.5
94.0
95.5
97.0
98.5
a of %
1.35
1.37
1.55
1.64
1.88
2.13
In density
<-0.69
-0.69
0.00
1.50
1.87
2.60
66 locations sampled; n + 1 = 67
Example computation for % < 4.5/100 ml: (64/67) x 100 = 95.59%.
3) The percent < each density is converted to standard deviation (0)
units to linearize the coordinate. Fifty percent is Oa and each
percent above and below that can be represented by some value in
standard deviations. For example, 1 standard deviation is 84.1%
total or 34.1% above 50%. The o units are shown in Table 5.5 for
the WH I example.
4) The densities per 100ml are transformed to In or log densities,
78
-------
again to linearize the coordinate.
5) Least squares regression is performed on the a values between the
lower and upper truncation densities. The antilog of the intercept
(50%) is the geometric mean (GM), and the antilog of the slope is
the geometric standard deviation (GSD). A correlation coefficient
may be derived indicating goodness of fit, and variances and confi-
dence limits about the GM and GSD may be computed.
In Table 5.6 we show the parameters of the lognormal distribution for WH
I, II, and III as calculated by the graphical approach using all locations
having equivalent volumes filtered. The geometric means (GM) were lower than
the sample arithmetic means (x) in every case. Arithmetic means (<=) were
calculated from the lognormal distribution parameters by the equation_ln « =
(In GM + 1/2 (In GSD)2). In all « exceeded the sample means. The «:x was
greatest when the geometric standard deviation was greatest. The lognormal
distribution predicted that a small but finite portion of the water contained
coliform densities in excess of 80 per 100ml (TNTC). These high densities
are recorded as 80 per 100ml in sampling; thus the sample mean is biased low
when TNTC's are found.
TABLE 5.6.
PARAMETERS OF THE LOGNORMAL DISTRIBUTION OF
COLIFORMS IN WOODBURY HEIGHTS, N.J. AS COMPUTED
BY THE GRAPHICAL APPROACH USING ALL LOCATIONS HAVING
EQUIVALENT VOLUMES FILTERED
Time
I
II
III
Geometric
Mean
per 100ml
1.3 x 10~3
0.17
5.5 x 10~4
Geometric
Standard
Deviation
82
12
336
Sample
Mean
per 100ml
0.40
> 1.52
> 1.76
Computed3
Arithmetic
Mean (
-------
f^
•
m
w
,J
S
H
Pi
H 55
H 0
•< M
JS M
to
<3 W
H o
s 2
O to
Pn W
0 P
u o-
to
O H
^ ^J
O PH
M rJ
t^> ^-«
PP PQ
M
(V^ Q
H W
CO H
M P
^1
t-J O
s! u
S CO
0 > Q)
•^ 6
rH O
•H *O
cd cd ^-N
^ >-i ex
o ^
M MH
PH 0
1
4J rj
C -H
cu e
•H ft
CJ CU
•H -U
MH cd -u
(U T3 td
e e -H
o cd >
0) 4J
CTi CT> 00
000
/-x
p^.
X^N • S~\
rH O
s-\ s~\
CO CO
1 1
O O
rH rH
CO CO
• •
* CO
O r7o
rH CM i-H
X • X
00 O O
• 1 •
CO ON CN
^~s rH >ta-^
co «co
1 O 1
o ^ o
rH rH
CO CN in
rH O CO
H
M I-H
HUM
§
CM rH
O O
O O
O O
O O
V
VO rH
vO CM
CO vO
l-~
CO O
O vO
00 ITi
O O
/""\ X~N
I-H m
I i
i-H 00
CN O"\
O O
s~* x"x
rH P^
CM !>•
I 1
p^ CT»
rH m
\~S N^
H vo
/^^
CM
1
O
rH
m
•
CT\
1
Co" I
rH O
• rH
O X
1 •*
CO •
rH r--
OCN
0
rH
in X
i — i —^
r-.
0
• o
O CM
X-N
CM
1
O
i-H
m
•
rH
1
CO
1
o
rH
-*
•
^-\ S3"
\w/
1 co
1
v-'O
rH
o x
m H
o oo
H
M M
£
oo
VO
rH
O
O
H
00
CO
•*
o\
O"N
CTi
O
s~*
oo
rH
m
i
m
r^
CO
^-^
rH
-------
This would not occur randomly, but did not give a slope reasonable for the
lognormal distribution. One TNTC location was obtained from this system and
one location had 6 presumptive coliforms per 100ml, but these did not con-
firm. The regression analysis does not include these samples. If they had
been included a lognormal distribution would be reasonable. With the excep-
tion of BL I all coefficients of determination were greater than 80%, and
all but three greater than 90%. Thus a large portion of the variance asso-
ciated with the In density of coliforms was explained by the sigma unit (and
hence % < density) coordinate. Based on these analyses, it is quite reason-
able to assume that the dispersion of coliforms in water distribution systems
is often compatible with the lognormal distribution.
The geometric means of confirmed total coliforms for all systems except
BL I ranged from 1.4 x 10~6 per 100ml for MI to 0.21 per 100ml for BB II.
The variation around the geometric mean was computed as ± 1 standard error.
In the case of WH II the total range of this variation (0.19 - 0.21) was 10%
of the mean. However, in other cases the variation exceeded 100% (e.g., DT
and MI). As the number of positive samples increased, the variation around
the geometric mean decreased.
The geometric standard deviations of confirmed total coliforms for all
systems except BL I ranged from 10 per 100ml for WH II to 441 per 100ml for
MI. Again an estimate of variation is given as ± 1 standard error. The
percent of the range of the standard deviations were from 11 for WH II to
211 for DT. Again as the number of positive samples used in the analysis
increased, the variation as a percentage decreased. Thus the precision of
the geometric means and geometric standard deviations increases as the
number of positive samples obtained increases.
The arithmetic means (°0 were computed for the various systems from
the parameters derived by regression analyses. These are shown in Table 5.8.
They ranged from 0.55 to_693 colifojrms per 100ml. In all case « exceeded the
sample arithmetic mean (x). The «:x ranged from <1.44 to <394. Thus the
sample arithmetic mean appears to underestimate the computed arithmetic mean.
It has been stated that the graphical procedure is useful in obtaining
estimates of parameters of the lognormal distribution. If the regression
analysis is a more rigorous form of estimation, a comparison of the proce-
dures is important. The geometric means and standard deviations for the 3
WH sample periods may be compared using Tables 5.6 and 5.7. Table 5.8 com-
pares the computed arithmetic means of each. The values of GM's and GSD's
estimated graphically may deviate from those estimated by regression. When
large deviations occurred the graphical procedure tended to give lower GM's
and larger GSD's than the regression procedure. This is because of the
tendency to place subjectively large weight on large densities when best
fitting a line by eye. When many samples are positive and more evenly dis-
persed along the line the discrepancies are minimized.
The computed arithmetic means are more dependent on the GSD than the GM.
The
-------
TABLE 5.8
RELATIONSHIP OF SAMPLE ARITHMETIC MEAN (x) AND COMPUTED ARITHMETIC MEANS
FROM THE REGRESSION ANALYSIS («r) AND GRAPHICAL ANALYSIS («g)
FOR THE LOGNORMAL DISTRIBUTION
System
CV
WH I
II
III
BB I
II
TP I
II
DT
BL I
II
MI
SR
MW
BG
X
0.11
0.40
>1.52
>1.76
0.51
>_9.38
>2.58
>6.16
0.30
>0.35
j>0.50
^0.40
>0.93
0.41
>2.15
ocr
0.55
29.31
2.83
96.39
3.04
693.03
11.45
79.85
1.83
0.50
0.72
157.46
5.06
4.83
420.37
•r
X «g
5.01
73.27 21.42
< 1.86 3.73,
£54.75 1.2xl04
5.96
£73.88
< 4.44
£94.13
6.09
<1.43
^1.44
£393.66
£5.44
11.79
<195.52
82
-------
The following conclusions may be drawn from the evaluation of the log-
normal distribution.
1) Most water distribution systems studied possessed a dispersion of
confirmed total coliforms compatible with the lognormal distribu-
tion.
2) If the lognormal distribution is assumed, the probability of obtain-
ing samples with densities >80 per 100 ml is appreciable even if
most samples show no growth of coliforms. All but 3 systems studied
had TNTC samples.
3) Estimation of parameters of the lognormal distribution by regression
analysis is superior to the graphical technique.
4) The sample arithmetic mean may deviate considerably from the arith-
metic mean computed from the parameters of the lognormal distribu-
tion.
5) Precision of parameter estimates is very important to estimating the
geometric mean or the computed arithmetic mean.
6) Precision is enhanced by increasing the number of positive samples
analyzed.
7) Most systems studied had geometric means of the order of 10~1 to
10~4 coliforms per 100ml.
8) Most systems studied had geometric standard deviations of the order
to 10 to 100 coliforms per 100ml.
9) Systems studied had computed arithmetic means of 0.5 to 700 coli-
forms per 100ml.
The consequences of these findings on evaluation of bacteriological
water quality and the MCL's will be discussed in Section 5.5.
5.4.5 Negative Binomial Distribution
The negative binomial distribution is quite versatile in fitting many
dispersion patterns. As stated in Section 5.3.6, the negative binomial has
been used to represent coliform dispersion patterns previously (Thomas 1955,
Pipes et al. 1977, Muenz 1978, El-Shaarawi et al. 1981). This distribution
requires three parameters: mean, variance and k. The parameter k is a coef-
ficient of aggregation and is estimated from the mean and variance. We
estimated k using the maximum likelyhood method.
The results of fitting coliform data from the water distribution systems
to the negative binomial are shown in Table 5.9. The k's ranged from 0.006
to 0.16. As k decreases the degree of aggregation increases. The fact that
all k values were <1 is indicative of a clumped distribution. The analysis
for goodness of fit was performed by the Kolmogorov-Smirnov test in which
83
-------
the test statistic is Dmax. Dmax represents the maximum deviation between
expected and observed cumulative frequency distributions divided by the
sample size. The larger the Dmax the less the probability of expected re-
sults fitting the negative binomial. All data fit the negative binomial by
this test.
TABLE 5.9
PARAMETERS OF THE NEGATIVE BINOMIAL DISTRIBUTION OF CONFIRMED
TOTAL COLIFORMS IN WATER DISTRIBUTION SYSTEMS
System
CV
WH I
II
III
BB I
II
TP I
II
DT
BL I
II
MI
SR
MW
BG
aThe probability of
negative binomial
Mean
CPU/ 100ml
0.11
0.40
> 1.52
> 1.76
0.51
> 9.38
> 2.58
> 6.16
> 0.30
> 0.50
> 0.40
> 0.93
0.41
> 2.15
rejecting the
distribution in
Variance
0.67
3.62
> 25.21
> 72.91
15.41
> 439.0
> 59.06
> 274.9
> 9.40
> 23.24
> 30.92
> 25.72
3.41
> 69.73
hypothesis that
all systems was
k
0.017
0.028
0.160
0.036
0.037
0.103
0.101
0.090
0.012
0.032
0.006
0.072
0.064
0.056
the coliforms
p>0.2.
Dmaxa
0.005
0.019
0.056
0.025
0.047
0.082
0.041
0.045
0.009
0.032
0.016
0.069
0.037
0.041
have a
Thus the dispersion of coliforms in all systems examined could be de-
scribed by the negative binomial distribution. El-Shaarawi et al. (1981)
have discussed the advantages of applying the negative binomial to bacterio-
logical data. However, the estimation of k is laborious and the data are
equally well described by the lognormal distribution. As the lognormal
distribution is used in other areas of water quality evaluation, we have
84
-------
reasoned that the latter may be a more suitable model upon which to evaluate
the MCL's and sampling frequency. Further discussions are restricted to the
lognormal distribution.
5.5 RELATIONSHIP OF THE LOGNORMAL DISTRIBUTION TO THE NATIONAL INTERIM
PRIMARY DRINKING WATER REGULATIONS
5.5.1 General
There are several consequences of the fact that coliforms are lognormal-
ly distributed in water systems to the National Interim Primary Drinking
Water Regulations. In this section we discuss these consequences with respect
to the following as functions of the number of samples collected:
1) The probability of exceeding the average rule.
2) The probability of exceeding the maximum rule.
3) The probability of obtaining positive samples.
4) The probability of obtaining samples with high densities of
coliforms.
5) The probability of exceeding the MCL if the number of samples is
reduced.
6) The interpretation of field results when all samples are negative.
The lognormal distribution has several characteristics that should be
explained prior to these discussions. Most microbiologists are accustomed to
analyses of samples in which the sample mean ("x) is a good representation of
the true mean (p). Such would be the case if coliforms were randomly dis-
persed. In the lognormal distribution this may not be the case. The sample
mean ("x) is associated with the geometric mean (GM), the geometric standard
deviation (GSD), and the number of samples (n). The GM and GSD may be used
to compute an arithmetic mean [«) (In a = In GM + 1/2 (In GSD)2] which,
as has been shown in section 5.4.4, may deviate greatly from TT. Also, IT is
invariably greater than the GM. If coliforms were randomly dispersed, and
if a x" > 1 were unacceptable; then a p > 1 is unacceptable. If coliforms
were lognormally dispersed, and if an x~ > 1 were unacceptable; then there
are several options as to what "true" parameter is used for acceptability.
One may use p, GM, GSD, or «. Another way of stating this is that with
a lognormal distribution, various combinations of parameters may be obtained
for any "x".
In the following discussions we restrict our examples of parameter com-
binations. We consider primarily several initial distributions. Two distri-
butions had an initial « equal to 1. The GSD's were 10 and 100 which were
values bracketing most of the observed GSD's (Table 5.7). The respective
initial GM's were 0.07 and 2.48xlO~5 coliforms per 100ml. From these two
distributions calculations for various numbers of samples were made holding
either « constant at 1 or holding the GM's constant at either 0.07 or 2.48
85
-------
x 10~5, respectively. The remaining distributions had a u of 1 (GM=2.72)
or 0(GM=1) and a GSD of either 10 or 100.
5.5.2 Probability of Exceeding the Average Rule
The cc for a lognormal distribution is the computed arithmetic mean.
If the GM is held constant, as the number of samples used to compute « in-
creases, 1) is low under all conditions and ap-
proaches 0 with increased number of samples. It would be highly unlikely
that the sample mean would exceed 1 under either of these conditions, and
the greatest likelihood would be with very small numbers of samples.
TABLE 5.10
PROBABILITY OF EXCEEDING THE AVERAGE RULE AS A FUNCTION OF NUMBER OF
SAMPLES WHEN THE GEOMETRIC MEAN IS HELD CONSTANT
Geometric Standard
Deviation 10
Geometric Mean (colif./100ml) 0.07
100
2.48x10-5
Number of
Samples
j>rob. of
x > 1
Prob. of
x > 1
1
2
5
10
10
20
30
40
50
60
70
80
1
0.264
0.120
0.091
0.091
0.080
0.076
0.075
0.074
0.073
0.073
0.072
0.125
0.051
0.005
<0.001
<0.001
<0.001
<0.001
<0.001
<0.001
<0.001
<0.001
<0.001
5.00x10-3
2.08xlO-4
7.19x10-5
7.19x10-5
4.23x10-5
3.55x10-5
3.25x10-5
3.55x10-5
2.97x10-5
2.90x10-5
2.84x10-5
0.011
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
<0.0001
86
-------
If the p equals 1, the GM would be 2.72 collforms per 100ml (y = In GM).
The 0.99 when n is greater than 30.
TABLE 5.11
COMPUTED ARITHMETIC MEANS («) WHEN y = 1 AS A
FUNCTION OF NUMBER OF SAMPLES
Geometric Standard
Deviation
Geometric Mean (colif./ 100ml)
Number of
Samples
1
2
5
10
20
20
30
40
50
60
70
80
10
2.72
~x
38.51
10.23
4.62
3.54
3.10
3.10
2.97
2.90
2.87
2.84
2.82
2.81
100
2.72
oT
l.lxlO5
545.5
22.65
7.85
4.62
4.62
3.87
3.54
3.36
3.24
3.16
3.10
The relationship of_number of samples to probability of exceeding the
Average rule when y or GM is held constant is highly dependent on the y and
GM chosen. In Table 5.10 the GM's were less than 1, and the probability of
exceeding_the rule decreased with increased number of samples. In Figure
5.11 the GM was greater than 1, and the probability of exceeding the rule
increased with increased number of samples. At a GM equal to 1 (hence y =
0) the probability is always 0.50 and is independent of the number of sam-
ples. The « values for two such distributions (y = 0, GSD = 10 and 100)
are shown in Table 5.12. Again, «" approaches GM as the number of samples
is increased, and the «"' s for GSD = 10 decreased more rapidly than those
for GSD=100.
87
-------
0.99
cr
CP
o
i_
CD
cu
0.90
0.70
0.50
en
| 0.30
CD
CD
O
LU 0 10
JD
O
JQ
O
0.01
0.0001
GSD = IO
10 20 30 40 50 60 70 80
Samples per month
Figure 5.11 Probability of Exceeding the Average Rule when y
a Function of Number of Samples
1 as
88
-------
TABLE 5.12
THE AVERAGE COMPUTED ARITHMETIC MEAN («)
WHEN y = 0 AS A FUNCTION OF NUMBER OF SAMPLES
Geometric Standard
Deviation
Geometric Mean (colif ./100ml)
Number of
Samples
1
2
5
10
20
30
40
50
60
70
80
10
1
"ae
14.17
3.76
1.70
1.30
1.14
1.09
1.07
1.05
1.05
1.04
1.03
100
1
oT
4.03x10^
200.7
8.34
2.89
1.70
1.42
1.30
1.24
1.19
1.16
1.14
In all of the examples above, \i was held constant. If a is held con-
stant, the GM increases to «" with increased number of samples. The prob-
abilities of exceeding the average rule when «" = 1 are shown in Figure 5.12.
The probability of exceeding the average rule increases asymptotically with
increased number of samples. The probabilities remained below 0.5 for n <
80, and were always higher for GSD = 10 than for GSD = 100.
It^should be evident from all of these calculations that the choice of
parameter sets is very important to determining the relationship between sam-
ples per month and the probability of exceeding the average rule. To summa-
rize, we have examined 4 general cases:
1) After establishment of
-------
0.99
tr
S'0.90
D
i_
CD
< 0.70
0)
en
0.50
^030
o>
0)
O
u 0.10
-g 0.01
_Q
O
V-
Q_
0.0001
GSD = IO
GSD= 100
10 20 30 40 50 60 70 80
Samples per month
Figure 5.12 Probability of Exceeding the Average Rule when
a Function of Number of Samples
1 as
90
-------
probability increased with increased numbers of samples approaching 40-42% at
80 samples per month. In all cases where GM was held constant, ^ decreased
with increased number of samples, approaching the GM. A sample mean (IT) of
>1 may easily be obtained without GM or p being equal to 1. Thus if it were
desirable to establish the basis for regulations on a GM or y of 1, the
amount of bacterial contamination allowable would be greater than current
conditions.
In case 1, the GM's were far less than 1 and ^r equaled 1 initially. If
these distributions were considered as a basis for the average rule, the
probability of "x > 1 would be extremely insensitive to rather large changes
in the GM or GSD. For example, if 20 samples were taken per month with a GSD
of 10, and the GM rose from 0.07 to 0.54 (a 770% increase) the probability of
exceeding the average rule would be only 10%. Thus the average rule is a
poor indicator of changes in the geometric mean or median and hence the true
mean.
In case 4 the computed arithmetic mean ( 1. When n < 10, the probability of exceeding the average rule in-
creases substantially with increased number of samples. However under the
current regulations, if less than 4 samples per month are taken, the results
are averaged over a three month period, such that the minimum number of
samples for calculation of the average is 3. The probabilities of violation
of this minimum would generally range from 10 to 25%. In other words if «
equaled 1 and 1 sample per month was averaged over three months the system
would be in violation of the average rule once in every 4 to 10 quarters.
But if 3 samples were taken per month and averaged per quarter, violation
would be very likely to occur at least once per year. When n = 6 per month,
the rule would be exceeded about twice per year, enough to insure that reme-
dies would be taken to improve bacteriological water quality. For systems
taking 10 or more samples per month the probability of violation is at least
0.25 or at least once in every 4 months.
5.5.3 Probability of Exceeding the Maximum Rule
The probability of exceeding the maximum rule is a function of the num-
ber of samples and the parameter sets chosen for the lognormal distribution.
In this section we consider two cases where: 1) the initial « = 1 and the
geometric means are held constant at 0.07 for GSD = 10 and 2.48xlO~5 for
GSD=100, and 2) the « is held constant at 1 but the initial conditions are
set as above. The probability of exceeding the maximum rule <0.0001 in the
first case (Table 5.13). Violation would occur about 1 month every 833 years.
Clearly, the rule is insensitive to such distributions when the GM is far
less than 1 coliform per 100ml.
The probability of violation for the second case is shown in Figure 5.13.
The probabilities of violation for both distributions increase with increased
numbers of samples from 2 to 39 samples for which 2 samples with >4 coliforms
91
-------
per 100ml exceeds the rule. Thus as the number of samples increases the
probability of obtaining 2 such samples increases. Above 39 samples 1 more
sample of >4 per 100ml is required for each 20 samples added. Thus the
probability of obtaining 3 such samples in 40 is less than obtaining 2 in 39.
The sawtooth effect seen here is similar to that seen for the Poisson distri-
bution (Figure 5.2).
TABLE 5.13
PROBABILITY OF EXCEEDING THE MAXIMUM RULE AS A FUNCTION OF
NUMBER OF SAMPLES WHEN GEOMETRIC MEAN IS HELD CONSTANT.
GSD
GM (colif. /100ml)
Number of
Samples
2
5
10
10
0.07
Probability of
Violation
<0.0001
<0.0001
<0.0001
100
2.48x10-5
Probability of
Violation
<0.0001
<0.0001
<0.0001
The maximum probability of violation occurs at 39 samples for both dis-
tributions . The peaks of the sawtooths decrease with increased number of
samples. The maximum probability of violation for GSD=10 is 0.53 and for GSD=
100 is 0.013. In other words with these distributions violation would occur
no more than 1 month in approximately 2 for GSD=10 and 1 month in approxi-
mately 6.4 years for GSD=100. As the GSD reflects the degree of aggregation
of coliforms, the probability of exceeding the maximum rule decreases as
aggregation or patchiness increases. This rule is very insensitive to very
patchy distributions of coliforms.
The maximum rule is generally less sensitive to distributions studied
than is the average rule. The probability of exceeding the average rule was
greater than that of exceeding the maximum rule with few exceptions. They
all occurred when « was held constant for a GSD= 10 at n = 36 to 39 and n =
58 to 59. The differences between the two rules at these exceptions were
not substantial. Thus, the maximum rule is redundant to the average rule,
and in almost all circumstances violation of the maximum rule would be accom-
panied by violation of the average rule.
This concept was tested with field data shown in Table 5.14. The mean
density per location was used for evaluation. Of the 15 analyses 6 exceeded
the average rule, and 7 exceeded the maximum rule. In no case the average
rule but not the maximum rule was violated, and in 1 case was the maximum
rule violated without also violating the average rule. These results are
from large numbers of samples. Can the same be said for small numbers of
samples? Nine sampling dates from WH are analyzed independently for proba-
bility of exceeding either rule (See appendix B). The number of samples
92
-------
0.99
jCD
^
cr
E 0.90
X
o
0.70
CD
£ 0.50
en
% 0.30
CD
CD
LJ 0.10
15 0.01
D
_Q
O
v_
Q_
0.0001
GSD = IOO
10 20 30 40 50 60 70 80
Samples per month
Figure 5.13 Probability of Exceeding the Maximum Rule when «
a Function ef Number of Samples
1 as
93
-------
ranged from 6 to 50 per day. On 5 days the average rule was exceeded, and
the maximum rule was violated on 4 of these days. At no time was the maximum
rule violated without violation of the average rule. Thus the theoretical
relationship between the maximum and average rules is generally confirmed
by our field data. It should be noted that these analyses were based on
locations from which volumes greater than 100ml were filtered.
TABLE 5.14
OCCURRENCE OF VIOLATIONS IN THE AVERAGE AND MAXIMUM
RULES IN WATER DISTRIBUTION SYSTEMS STUDIED
System
Sample
Mean
Violation of
Average Rule
% >4
colif.per 100ml
Violation of
Maximum Rule
cv
WH I
II
III
BB I
II
TP I
II
DT
BL I
II
MI
SR
BG
0.11
0.40
> 1.52
> 1.76
0.51
> 9.38
> 2.58
> 6.16
> 0.30
> 0.35
> 0.50
> 0.40
> 0.41
0.41
> 2.15
No
No
Yes
Yes
No
Yes
Yes
Yes
No
No
No
No
No
No
Yes
1.3
4.5
10.4
5.9
2.2
23.2
11.5
18.5
1.1
0.4
0.6
1.4
4.9
5.1
10.9
No
No
Yes
Yes
No
Yes
Yes
Yes
No
No
No
No
No
Yes
Yes
5.5.4. Probability of Obtaining Samples above Particular Densities
In this section we discuss the likelihood of obtaining positive samples
and samples of particular density ranges in water distribution systems. From
our field data and retrospective studies, two points are clear. First, most
samples from water distribution systems under normal conditions will be de-
void of coliforms. In Table 5.4 we show the fraction of locations for each
system studied that produced samples without coliforms. In all systems but
two that met the average rule greater than 90% of the samples (locations) had
0 coliforms per 100ml. Even when the average rule was violated over one half
of the locations were without coliform growth. Second, when positive loca-
94
-------
tions are obtained the likelihood of obtaining a high density count is rela-
tively large and the likelihood of obtaining a sample that is too numerous
to count (TNTC = 80 per 100ml) is finite. TNTC samples were obtained in 11
system analyses. Of these the average rule was violated only 6 times. What
then are the consequences of the lognormal distribution on obtaining positive
samples and high density samples?
The two cases of theoretical parameter sets discussed in the previous
section will be used here. When GM was held constant at 0.07 (GSD=10, <* = 1)
the probability of obtaining at least 1 positive sample was 0.125 for 1
sample, 0.099 for 2 samples, 0.025 for 5 samples and <0.001 for any number
above 10 samples. When the GM was held constant at 2.48xlO~5 (GSD 100, « =
1), the probability was 0.010 for 1 sample and <0.0002 for all higher sam-
ples. Thus the probability of obtaining a positive sample is very small for
low GM's when GSD ranges from 10 to 100.
When a is held constant at 1 and GSD's are 10 and 100, the probability
of obtaining positive samples increases with increased number of samples
(Figure 5.14). The probability of obtaining a positive sample increases to
greater than 99% for n > 40 when GSD=10. The probability is always less when
GSD=100 than for GSD=10 and reaches 50% by 63 samples. If GSD=10, at least
one positive sample would be found on average <2 months if 6 or more sam-
ples are taken per month. If GSD=100, at least 63 samples would be required
for the same result. Thus as the coliforms become more clumped in their
distribution, the likelihood of obtaining positive samples decreases.
If the likelihood of obtaining positive samples is very small when GM's
are held constant and are small, the probability of samples with high coli-
form densities is even smaller. The discussion of samples with high coliform
densities is restricted to the cases where « is held constant at 1. The
probabilities of obtaining at least 1 sample with 4 or more coliforms per
100ml are shown in Figure 5.15. Again probabilities increase with increased
number of samples and are always greater for GSD=10 than GSD=100. These
probabilities are similar to but not exactly those of check sampling. The
density of samples for which check sampling is required is actually >4 coli-
forms per 100ml. However, this depiction is a reasonable guide. If GSD=10
and oc = 1, a check sampling program would occur at least once every two
months when 15 or more samples are taken and at least once every four months
when >6 samples are taken per month. If GSD=100 and « = 1, check sampling
would occur for less often (for example once every 14 months when 15 samples
are taken per month).
The occurrence of a TNTC sample is often taken as an error in sampling
or processing by water system personnel. A TNTC count is particularly dam-
aging to the compliance to the average rule for small water systems. It is
often considered as an 80 coliforms per 100ml density; and thus when one
TNTC is found, 79 negative samples must be obtained if the average rule is
to be met. What then is the probability of obtaining a TNTC? Is the proba-
bility high enough such that obtaining a TNTC may not be regarded solely as
an error? The probabilities for obtaining at least 1 TNTC, when « = 1 and
GSD's=10 and 100, are shown in Figure 5.16. The probabilities for neither
distribution exceed 10% for n < 80 per month. However, when the number of
95
-------
0.0001
10 20 30 40 50 60 70 80
Samples per month
Figure 5.14 Probability of Obtaining at Least 1 Positive Sample as
a Function of Number of Samples when « = 1.
96
-------
0.99 r
0.90
jf 0.70
Q.
§ 0.50
oo
^ 0.30
o
0)
O
0.10
o 0.01
o
Ql
0.0001
= 100
10 .20 30 40 50 60 70 80
Samples per month
Figure 5.15 Probability of Obtaining at Least 1 Sample with >_4 Coliforms
per 100ml when a = 1 as a Function of Number of Samples
97
-------
o
0.90
0.70
•5 0.50
CP
- 0.30
0.10
_a
o
jQ
o
Ql
0.01
0.0001
GSD=IO
10 20 30 40 50 60 70 80
Samples per Month
Figure 5.16 Probability of Obtaining at Least One TNTC Sample when
<* = 1 as a Function of Number of Samples
98
-------
samples is greater than 9 for GSD=10 and 19 when GSD=100 the probability is
greater than 1%. For most systems having an « =1 and sampling 20 times per
month, a TNTC may be expected to be found at least every 8 to 10 years.
With this low probability it can be seen why most operators consider such
samples as errors. However, another way of looking at this aspect is to
consider that if 100 systems sample 20 times per month with « = 1 for all,
at least 1 system would probably have a TNTC. The state or federal official
examining the records of many systems on a monthly basis must be prepared
for the eventuality of occasional TNTC samples.
5.5.5. The Probability of Exceeding Regulations if the Number of Samples
is Reduced
The National Interim Primary Drinking Water Regulations (U.S.E.P.A.
1976) state that as many as 75% of the microbiological samples required per
month may be substituted by free chlorine measurements with 4 chlorine mea-
surements required per coliform determination (Section 141.21h). As the
likelihood of violating the average and maximum rules are dependent on sample
size, one must consider the consequences of such a rule. Our discussion is
restricted to the cases where « is held constant at 1 coliform per 100ml
and GSD=10 and 100.
The probability of exceeding the average rule would largely be unaffect-
ed if the number of microbiological samples examined is >20 per month. When
sampling falls below 20, then the probability of exceeding the average rule
is diminished significantly as fewer samples are taken. Clearly, this rule
should be given special scrutiny for small water systems.
The sawtooth nature of the probability of exceeding the maximum rule
should be recognized in establishing the degree of substitution. The proba-
bility of exceeding the maximum rule can change by more than 0.2 by a differ-
ence of 1 sample (e.g., 39 to 40 samples in Figure 5.13). Substitutions
may be made such that the remaining number of samples examined for coliforms
maximizes the sensitivity of sampling within each 20 sample unit. That is
the retained microbiological samples could be set at 39, 59, 79, etc. If
the retained microbiological sample size falls below 39, the probability of
exceeding the maximum rule decreases significantly for fewere samples. Thus
the cholerine substitution rule can be used by small systems to avoid viola-
tion of the MCL even when microbiological contamination is high.
5.5.6. Interpretation of Field Results When All Samples are Negative
What can be said about the degree of bacterial contamination of a water
system if all samples in a month are negative? From the previous discussions
it can be seen that systems may be contaminated with that contamination
rarely being detected. This was certainly the case when the GM's were held
constant at 0.07 and 2.48xlO~5 coliforms per 100ml and GSD's were 10 and
100 respectively. Also, in spite of the fact that we found coliforms in
every system we studied, the historical records of those systems rarely
noted coliforms. In fact in system LB no coliforms were reported for several
years prior to our study and were not obtained by required monitoring during
our study. The possibility existed that the procedures used by those at LB
99
-------
were not satisfactory for proper recovery of coliforms (see Section 4). But
the fact remains that coliforms may occur in relatively high densities in a
system and go undetected.
The probability of « being 1 and all samples during a month being nega-
tive (0 coliforms per 100ml) may be derived from Figure 5.14. As one minus
the probability of obtaining at least one positive sample. For example if
20 samples are taken in one month and all are negative, the probability of
this occurring when « = 1 and GSD=100 is 0.8. In other words 80% of the
months that 20 samples are taken, detection of coliforms would not occur
even though they would be present with the above distribution. When <* =1
and GSD=10, contamination should be detectible in the form of at least 1
positive sample at least half the time when greater than 5 samples are
taken per month. If only one sample is taken per month 7 out of 8 months
would be negative. The probability of detecting coliforms decreases as
the GSD increases. If only one sample is taken per month when « =1 and
GSD=100, 90 out of 91 months would have no positive samples. Even when 40
samples are taken per month approximately two out of three months may not
give positive samples. Thus, small numbers of samples, when coliform distri-
butions are highly aggregated, are quite insensitive to the detection of
contamination.
5.6 REPLICATION OF SAMPLES AND LARGER VOLUME SAMPLES
5.6.1 Introduction
During the sampling programs it was found that the time required per
site sampled varied from about 10 minutes to 30 minutes because of the prob-
ability of finding someone at home to provide access to a tap, and the number
of refusals to permit access to the residence. The amount of time needed to
measure water temperature and chlorine residual and obtain the sample was
small in comparison with the time required to gain access to the tap. Thus,
the number of samples collected on any given day could be readily increased
with little additional time and effort by obtaining replicate samples at
each site sampled.
Replicate samples at a site are not independent observations on the
quality of water in the system and thus replication of samples is not equiv-
alent to obtaining additional samples at other sites. However, it is clear
that obtaining replicate samples at each site provides more information than
single samples at each site. How much more information is obtained from the
second and additional replicates is the subject of this Section.
One method of interpreting the data from replicate samples is to consid-
er them as equivalent single, larger volume samples. Two replicate 100ml
samples could be considered to be the same as a single 200ml sample or two
replicate 200ml samples as a single 400ml sample and so forth. Larger volume
samples, mostly 1000ml and 200ml samples, were collected from some communi-
ties on some sampling days to test this interpretation.
The purpose of this Section is to explore the questions associated with
replication of samples from individual sites and with the use of larger vol-
100
-------
ume samples. The questions are concerned with the most efficient method for
monitoring a water distribution system. The objective of bacteriological
monitoring is to find coliform organisms if they are present or, stated from
the opposite viewpoint, to demonstrate that the disinfection of the water
during treatment and the integrity of the distribution system are maintained
so that the probability of bacterial contamination of the water is very
small. Replication of samples and/or use of larger volume samples are evalu-
ated in terms of how they might expedite the achievement of these objectives.
5.6.2 Replicate Samples Collected
The number of sites sampled and replicate samples collected for the var-
ious communities included in this project are given in Tables 5.15-5.23 along
with the confirmed coliform counts in samples where they were present. The
numbers assigned to the individual replicates represents in most instances,
the order of collection.
One liter samples were collected during the sampling of systems CV and
WH. It was found that if large amounts of particulate matter were present
in the samples, filtration of 1000ml required an excessive amount of time
(30 to 60 minutes), and the particulate matter on the surface of the filter
interfered with development of typical coliform colonies. The use of one
liter samples was discontinued about half way through the WH system sampling.
During the second half of the CV system sampling and thereafter a mini-
mum of two 100ml replicates were collected at each site sampled. In DT and
BL some extra 100ml samples were collected at a few sites and a few 200ml
samples were also collected. In MI, SR, MW and BG duplicate 200ml samples
were frequently collected in addition to the two 100ml replicates. One
liter replicates were tried again on two sampling days in MI because the
water there was exceptionally free of particulate matter, but no coliforms
were found in any of those samples.
5.6.3 Results from Duplicate Samples
The largest mass of data on replicate samples which we have collected
consists of counts on duplicate 100ml samples. These data are summarized in
Table 5.24. All sites where less than 2 replicate 100ml samples were col-
lected are not included. For sites where more than 2 samples were collected
data from only the first two 100ml samples are used.
One or more samples with more than eighty coliforms per 100ml were col-
lected from each of the systems sampled except CT, WH, and MW. When one or
more coliform results are recorded as TNTC the calculation of a mean coliform
density is uncertain. To obtain the figures in Table 5.24, TNTC results
were taken as more than eighty coliforms per 100ml. This gives a minimum
value for the mean and variance of the density.
From the confidence intervals in Table 5.24 it can be readily seen that
there are no significant differences in the fraction of samples with coli-
forms or in the mean density between the first and second replicates. This
is as expected and clearly shows that the chance of finding a coliform is
101
-------
w
w
PH
§
H 53 CO
i-J 5 W
PQ OH
< o <;
H O
S cj
O
Pn
M
. •]
0
CJ
PK
o
en
W
CJ
§
1 >
o-
w
C/D
CO
cu
4J
S -H
rl CO
0
MH 0)
•H >
rH TH
O 4J
CJ -H
co
M-l O
O PH
CO 4-1
cu id
CJ
Pi CO
CU 4-1
3 0
cr 3
CU 0
CO CJ
4-1
3
o
,C CO
•H C
> o
MH
W iH
CU rH
4-1 O
•H CJ
CO
p_l
C3
o
o co
o cu
i-H 4J
cd
<*H U
O -H
rH
• CX
co cu
O Pn
55
r-J
0
O CO
o cu
rH 4J
cd
MH cj
O- 1
•rl
i-H
• CX
CO CU
O f*H
55
T)
CU
CO i-H
CU CX
4J E
•H cd
co co
M
c
•H
rH
CX CU
3 4-J
cd cd
CO Q
o
i
o
1
i-H
*
O
1
O
1
O O rH
1 1
CTi O «
1 O
O 1
O O 1 O
II CN 1
rH ro co H
VO CN in O O i-H
rH CN rH rH CN CO
cd ,0
CN CN CN rH <• rH
*^" CO
M A
CO CO CN CO
H H rH CN rH CN
vO "^ **O C^ rH ^*
i-H CN rH H CN CO
CTi O>
^; ^ ^ d C^ ^
"^. ""^^ ^""s. "^ LO "**^»
rH VO 0> H rH VO
CN CN CN CN CN CO
O
1
O
1
f-N.
O
1
o
1
rH
A
o
1
CN
1
O
«v
o
1
o
1
CO
rH
vO
CN
rH
CO
CN
O
CO
Oi
"^.
oo
CO
fc^
I
o
1
rH
0
•> 1
O CTv
1 1
CN CJ
1 C"
o
*
" 0
rH 1
1 rH
0 1
1 U
f^ ^«
O CTi
CO CO
rH rH
CO co
CN CN
CO rH
CO «3"
ON CTi
^ ^
co m
rH rH
CO CO
CU
U
cd
t-H
CX
co
cd
cu
rH
S-
g
cd
CO
rl
CU
•H
H
rH
*
cu a
• ,C 0
rH 4-1 -H
0 0 cd
r^ S "3
>4-l U
*rj rj
C rl -H
cd cu
** J~* M-l
31— -j Q
•H
MH co
O M
CN CU 3
J2 O
" 4-1 J3
^ «•*
CD 6 CN
• O 3
H H -H (3
S t) 4H
CD I4H 0)
O O 0 CO
o cd
rH CO O M)
C -T3
a o c -a
Cd -H W CU
r^3 4J 1 CJ
•»-• rl S 3
O 13
M CX d 0
CU O rl
rC CU CX
4-1 4J CO
cd cd H 13
rl rl 3 C
cd o cd
0 CU J3
CO Cfl CN 4-1
CO CN O
00 (3 rl
•H rl J3
co o
cd 'O MH cu
Is cu co
rl C O
"O CU O 4->
CU 4J -H p.
rl I-H 4-> >>
CU -H Cd rl
4J M-l ^5 4J
rH 3
•H rl CJ rH
>4H CU (3 >%
4J -H rl
CU -rl 3
0 rH H Cd
3 CU rH
H CU 4J
O C 4-1 C
> 0 <3 -H
• * *
Cd rO O
102
-------
VO
I-l
m
W
rJ
g
CO
w
^
CO
63
d
M
i-J
Cki
W CO
« H
55 ^2
M H
M
co W
Z (M
ie
u en
§8
O S
Pb
H
r3
0
o
<3
CO
u
§
D
o-
CO
09
01
£«
•H >
rH iH
CJ £
u
I«H O
O P-
0) 4-1
41 cd
o
C CO
4) U
9 a
V 9
41 O
CO U
4J
9
O
43 01
^ S
•H H
? O
MH
O -rl
41 i-l
ScS
CO
3
0 09
0 41
H W
cd
>rl U
O iH
H
• a
0) 41
O (A
85
.
g
o u
0 4)
CO
U-l U
O -H
rH
• o.
Is
•s
« rH
fl) pi
*•* §
•rl CO
CO CO
60
a
•H
( — 1
a 4i
B ^-*
CO CO
CO O
f^
jf
c- ov
1
• O
vO 1
H
i-l «
H O O
id o o
1 -^ "^
CO M m
iH O rH
444
vO CM O
vO -» O 00
rH CM rH
rH H H i-l
CO CO CO CO
•» * *» M
CM CM CM CM
vO f- CM rH
rH CM CM
Ov Ov
st VO — •-»
CM CM rH CO
*if
rH 1-H
m m
in
rH •
1 O
O 1
OV
M
VO •
sr o
i i
00 H
• i-T
O 1
1 O
rH
•t
rH*?
1 CM CO
O CM 1
O
•t •>
00 st •
rH CM O
1 1 1
O CO rH
CM?"?"
rH i-H rH
M « •>
CO rH O
inirH
a
cd
*
4)
H
0)
rl •
41 0
U O
iH tH
i-H U
M 01 rl
rl 4)
rl 9 4J
4) O rH
•rl -J-
\
O I-l O
1 1 . j J 1^
a H co
0X9
•H rl 0
*J 9 rl
cd rt 4)
,0 rH S
9 9
CJ (3 13
fl -H
•H O
rl 4) -U
U CJ
•u cd I
<4H rH
<
-------
Ed
Id
<
CO
S
u
I-J
E o
2 o
r*
rH S5 PS
M W
m H
W H ^
9 O 0
H U ^
gg
33
8
o
CO
U
»
s
CO
n
is
& 3
(0
UH
»l
O U
O vt
C a
3 o) iw
CO 0
rL
P? 0
H ft
CO U
£*
5 4)
•rl
H to
1 'g
W 0) i-l
H u a.
to co
a
w
§8
8Z
CO
IM
0 0
O U
•< 0 iH
W (5 «
2 a o
< 3 P«
vT*
W - ^
rH CM
9V 9t
9t r^ ^
•v. CM •»
r» H i-H
vO vo vO
2
V
0
M
1-t
i
rH
0
s
?
CM
.
•t M rH *•
° ? cn f
CM rH « O
• «CM •
CO CM 1 rH
H CM
\Q \O
H'IH
i 7
r- •
H 00
7?
en rH
• "O
"? Yin
o o
?
00
vo in
rrf
vO vO
cn
H"
H
•t
EH
A
sr
rH
d,
•s
s
H
vo vo
vO O
CM
9i 9t
fs» |*s»
rH CM
i-l rH
fNi FN.
•> M
- H -a-
m i i
I H in
~oo
rH in 1
A«=o"
ss/.
H EH 7 7
1 1 -tr O
0 H rH •
« » »cn
0 rH EH 0
EH
O
in H
a
rH U-1
-fv"
EH_O
00 CM
ca"
cn
o
i
CM
cT
1
in o
rH 1
-VO
"If
rH i-l
.3.
rH
-._
CM
CM
•I
t
O
cn o
M ft
0 H
rH 1
1 vo
O CM
•t ft
rH rH
O O
M •
O EH
A H
m
^
rH
ej\
t-
^
rH
fs.
vO
CM
O
1
r-
CM
rH
|
cn
f
CM H
to*
9. 1
•* CM
in
rH
CM
CM
rH
rH
f.
0
rH
H
9>
en
1
O
•t
O rH
1 1
CM en
00
^^
rH
9>
r—
00
rH
r-
rH
*
rH
1
O
CM*
1
O
*
rH
1
H
r^.
rH
en
CM
?
cn
o
t
vO
*
en
1
cn
A
m
A
-*
Ov
^
i-l
9i
r-
97
rH
r^
104
-------
oo
rH
in
TABLE
CO
a
CO
H
§
h4
PM
Pi
S3 S3
H S
O
CO H
P H
O S3
U 3
COLIFOI
fa
o
CO
W
53
H
1
CO
CO
4J
PI
O
CJ
a
M
O
M-l
•H CO
rH CU
O 4J
CJ -H
CO
M-l
O CU
CO vH
CJ -H
a co
CU O
cr
CU 4J
co a)
4-1
g
O
.e
4-1 CO
o
co MH
01 -^
w m
4-1 l-H
•H O
CO CJ
O CO
O CU
CM 4-1
a)
M-4 O
0 tH
rH
• P.
CO CU
0*
,_,
M en
O CU
rH 4-1
(4
M-l CJ
O iH
• P.
CO CU
O Pi
S3
13
CU
CO i-l
cu a
i i g
iH CO
CO CO
00
a
•H
rH
P. CU
cd cd
CO Q
O
1
O
1
0
1
K
o
1
rH
1
O
1
O
*
o
1
CM
1
O
1
O
o
1
H
1
rH O
1 1
O O
in oo •* -* co
CM
1 1 co" i i
«* •*
CO CO
* A
CM CM CM CM CM
in cy\ oo «*^ co
CM
o>
r^ o o o
— O 00 00 00
r^ oo **•*. "^ ^^*
CM ^- m vo r-
^. 00 rH rH rH
CM "^ *v. ^v. — .
rH rH rH rH i-H
O
t^
M
rH
rH
1
O
o"
1
M
o rH m
I i i
rH i-H O
CMCM •* co oo mm vom H
CM H rH -*
CM
1 1 1 1 II 1 r-T 1
"*
CO
CM CMCMCMCMCMCM«*CM
CMCM •* co oo o-\m r»-in CM
CM i-H iH •*
O O O O O O
oo oo oo oo oo o oo
— . «^ -^. ^> -^ oo «^
CM co -a- oo cy> — CN
CM CM CM CM CM m rH
^^, ^^. ^^, *^^. ^**. "^. ^^.
rH rH rH rH rH CM CM
rl
CU
4J
i-H
•H
MH
CU
a
CU
1
H
O
CJ
o
oo
1
4->
CU
o
0
II
a
H
o
o
4-1
CO
3
O
)4
a)
g
Jj
C
o
o
4-1
II
H
9
cd
105
-------
I
CO
w
1
co
W
H
u
^
hJ
A<
fVJ
M g
*5
CO <
2 2
§8
U A
M
o
M
1-5
o
u
p^
o
CO
H
0
0-
H
CO
n
a
§
u
E
0
•H n
rH 4)
cSS
CO
*W
O 4)
>
in -H
4) 4-1
0 -H
a on
01 O
3 PW
tr
01 4J
co a)
4J
3
O
£
is C
o
CO M-i
4) -rl
AJ ^H
CO U
rH
o
m *J
CO CO
M
C
•H
rH (0
p 4)
£ 4-)
re n
co p
o
o o
rH O
O
rH 1
rH
- 1
rH O rH h* H rH
1 II 1
-------
o
CN
10
W
rQ
vj
H
H
rJ
g
S
w
w
3
H ^i^
h-1 J-<
p 1 f|J
W Pk
a
M 0
pr{
C/2
|H a)
si
s
^t ^^
Is
PTH H
H
rJ •
0 4-1
U S
O
W
U
2;
H
?v
ty
CO
a
a
a
o
u
g
o
t-l
,' J M
•rt w
.-1 4)
O 4J
O iH
CO
0-1
O 4)
CO iH
4) 4J
O -rl
a co
41 O
3 PH
cr
4) 4J
CO CD
4J
9
O
A CO
4n €
S o
CO iH
41 rH
•U O
•H O
rj
(§
o n
O 4)
^ j ( ,1
cd
^M O
O iH
H
• CU
CO 41
0 Pd
H ^**» ***» ^^ ^**» ^»* ^»«fc **^
ininm u-iioinioir,voxo
41
•o
1-1
o
4)
JS
w
4J
9
CO
•o
I
CO
rH
1
iH
>
•H
•O
5 ^
a 4i
O 4J
H
TJ • Tl
4) co m
CD O 41
^ ^W (X
•H
CO rH CO
41 O 41
U U *H
(4 fl
0 43 0
•rl *J H
i-l iH O
a. » o
4)
rl CO 0
41 CO
i-H i-l
0 (It 43
CM CO AJ
"fi "S M
CO O
CO B
p S t
c5 9
O i-H 4J
U-l O
O 41 O
O U
iH CO
4J 41 CO
O 4J 3
41 Pi O
rH 53 tJ
rH 01 (U
O 41 g
O M 3
0. (3
I4_l (J
O VJ O
o
»J T3 4J
41 41
•own
I-l iH
O 0 H
( ,
CO ft
107
-------
.
CM
tn
W
eq
H
to
Cd
fe
5j
to
W
H
U
M
P-i
§
ss
to §
fs o
5 55
O M
O Pi
CU
S to
p5
g
H
O
PM
O
to
H
!=>
cr
M
CO
to
e
3
o
o
g
o
M-J
•H CO
rH 01
ta
O 41
10 -rl
01 U
O -H
C 10
01 O
3 Pi
o-
4) 4J
to n)
CO
S
4J O
3 -H
rH
10 O
01 U
U
•H O
W (3
rH
o to
O 0)
CM 4J
at
lw O
0 -H
rH
• CX
10 01
0 PS
rH
<§ to
O 01
cd
<4-l U
O -H
• CX
10 01
O PS
S5
13
41
0) rH
41 P.
U gj
•H Cd
to to
60
C
•H
rH
CX Q)
e w
m cd
to o
o
1
rH
O
O
rH
•« |
O 0
rH O
1 1
O 0
0 •
o
« 1
O rH
| |
O O
1 1 •
rH CM
1 rH
O -
-o"
0 1
1 O
rH 1
1 VO
O 1
1 CO
0
rH • «
1 O O
0 1 1
0 rH
1 1
rH CM O
1 1 1
0 00
VO in rH vO
CO CO rH
1
CM sf
* 1 1
rH CO
1
CM CM CM
M ft ft
rH rH rH
vO r~ rH in
CO CO CM
v|3 rH
m 00 rH CM
CM -^ -^ -^
•^ o o o
O\ rH rH rH
•i
rH •
1 O
rH 1
1 CO
O 1
O iH
- 1
rH -CM
1 CM rH
O 1 1
1 O r-.
0 1 1
1 O 0
O 1
o -
« CM
rH - 1
1 O VO
rH 1 1
1 O O>
i-l 1 1
rH O 10
1 1 -?
0 1-1
- -oT
O rH 1
1 1 O
O O 1
1 1 O
o o i
1 1 0
H 0
ft
- -00
0 O 1
1 1 0
rH O 1
1 1 u-l
0 O rH
1 1 1
O rH \O
•^
^
CO
CM
•>
rH
oo
rH
»O
CM
^^
O
rH
00
CO
m
'i
H
O
M
o
1
o
1
co
i-H
„
EH
H
1
vo
r*.
rH
1
O
1
0
o
00
CO
CM
«
rH
CM
rH
00
CM
•^^
O
rH
0
rH
1
°
O
CO
CO
CM
M
rH
O\
O
CO
^^
o
rH
r-Te^
^3 c**
1 i
??
0 0
*• «t
o,a
o7
c!*?
1 0
O I
o
CM -
1 O
rH 1
| |
0 °
>O H
» •
O O
O rH
1 1
rH O
?? s ? ?
O O 1 rH O
1 1 rH 1 1
o CM I o r*
110 1 1
rH O i-l O O
rH *3" 00 CO ^ in
H
v£>
in
CO CO CO CO rH rH
00
f«s.
*
CM CM CM CM -*
0 -H CS
O *O
0) fj
f~^ Q «r4
CO
C -o cd
,5 w M
u S "S
01 U
rl C 3
O O T3
g 0
CO rl
II Vl CX
3
4J O •«
§* a
O CM
0 CM JS
4J
0 «H O
4J O Vl
rQ
to c
3 O 41
0 -H CO
^ 4J O
0) Cd 4J
e -° o-
3 3 P^
a o ^
fl *J
O 1^
O rH
4J rl rS
01 rl
1 4J 3
"4-1 Cd
H < rH
« *
CO rO
108
-------
CM
CM
m
r
3
CO
CO
4-J
a
Q
U
E
o
U-l
•H CO
rH 01
O 4-1
U iH
M-l
O 01
CO iH
0) 4-1
O iH
C CO
0) O
3 Pi
cr1
0) 4-1
CO CO
4-1
3
O
42 CO
1 i g
Js o
M-l
CO -H
0) rH
4J O
•H C_>
CO
O CO
o a)
CM 4J
CO
M-< O
O iH
rH
• CX
CO CD
o f*:
25
O CO
O 01
(0
MH O
0 nH
• CX
CO Q)
O Pd
S3
•ti
01
CO rH
O> CX
4J &
•H Cg
CO CO
bC
(3
•H
rH
CX 0)
§ a)
CO O
O
1
o
1
rH
0
M
rH
1
t-H
1
O
1
rH
-3- CM
1
>*
CO
CM CM
rH rH
-3* *3*
O
O 00
00 ---
•^- CO
00 CM
O O
rH rH
i-H
1
0
1
O
1
o
*
o
1
0
1
o
CM
*
0
1
o
1
o
1
CM
f.
0
1
o
1
rH
1
O
CT*
•*
CO
CM
rH
CO
rH
O
00
^^
o
CO
o
rH
O
1
vO
1
O
1
o
*
o
1
o
1
o
1
oo
f*
o
1
ON
i-H
1
O
1
o
vO
rH
«*
CO
CM
rH
ON
rH
O
00
*s^
vO
i-H
rH
O
1
VO
CM
1
O
O
*,
o
1
o
1
o
1
i-H
CO
rH
Sf
CO
CM
rH
in
rH
0
00
•^^
rH
rH
rH
i-H
f\ f*
00 "
1 1 0
CO O 1
1 1 O
CO O 1
1 1 0
O O 1
1 1 O
O O 1
1 1 0
0 0 1
1 1 O
O O 1
1 1 0
O •<]• 1
t-H rH
*
rH * » O
1 CM r- I
rH 1 1 O
1 O O 1
olio
1 Q rH 1 O
m T i o i
1 O O 1 O
CM 1 1 O 1
1 O CM 1 rH
O 1 1 0 1
1 O O 1 rH
O 1 1 CM 1
| 0 0 rH 0
CO 1 1 1 1
vo O H rH H
VO O
00 vO
* A
i*» m
""I
CM -vf
rH ro
CO
CO rH
rH
O
00
^•^
CO
i-H
rH
i—l
O
o
o
o
I
rH
O
O
O
O
I
O
o
o
I
CO
o
o
o
CO
*
CM
vO
CM
00 CO
d- st
A ft
ro co CO
vo
0 0
oo oo
CM
rH O
O rH
O
0 O
O
•H JS
4-1 4-1
S-g
i-H
rH 03
O 01
O rH
CX
HH 0
O CO
CO
M
Q) 4-1
T) O
CO
01
CD 01
CO ^
01 „
4J M
ct) _
o o
o> .,
u o
-------
CO
CM
m
rJ
pq
o
CO
C
O
U
fi
0
<4-t
•H CO
rH CU
O 4->
U iH
CO
14H
O
CJ iH
C CO
CU O
3 PH
CU 4J
co cd
CO
fi ft
4-1 0
•H M-H
ES -H
CO O
cu o
4-1
iH O
co 2;
<§ M
o cu
CM 4J
03
>4H U
O iH
rH
• P.
CO CU
o &
rH
o co
O CU
rH 4J
(!)
14H U
0 iH
rH
• a
co cu
O Pd
55
TJ
01
CO rH
CU PL,
4J 0
•H ctj
oo
£3
•H
rH
a cu
S 4 11
Oj 03
co n
H
i
oo
I
H
O
ft
O
1
H
1
rH
1
^.
•V
rH VD
1 0 1
O 1 Oi
o in
• 1 1
o o o
1 cfll 1
>* H CM
CO *t
rH rH rH rH
in *3* ro co ro
^j
r— 1
O 0 O
o o oo oo oo
oo oo --- ^ ^^
^~ \O rH OO
in oo I-H CM CM
CM "^^ ^-^ "^-^ ^^
•^- o o o o
ON rH rH rH rH
00
1
o
1
o
1
CM
A
o
1
o
1
rH
1
00
CM
«^-
A
CO
CM
»,
i-H
-*
O
00
O
CO
o
rH
O
1
O
1
o
1
rH
1
CM
1
o
1
1
CO
rH
ft
o
1
o
1
o
1
o
1
O
1
O
1
^d"
i
H
m r--
vO
A
m
**
CM -tf
* *
rH CO
00
ft
f>»
ft
-------
the same for either replicate. The only physical difference between the two
replicates is the length of time the water was allowed to run from the tap
before the sample was collected, and this is not expected to influence the
chance of finding a coliform.
The important question about the use of replicate 100ml samples is con-
cerned with how much is gained by the collection of replicate samples at one
site as compared with the collection of single samples at other sites.
Empirical attempts to develop an answer to this question were made during
the sampling of the systems LB, DT, BL, SR, MW and BG.
5.6.4. Negative Sampling Results
On June 14, 1979 twenty-nine replicate 100ml samples were collected at
a single residence in the BB service area. Two of the samples had presump-
tive coliforms but they did not verify. In the rest of the BB service area
on that day, 7 out of 24 sites sampled had coliforms present in one or both
of the samples collected, and all colonies picked verified as coliforms.
On July 19, 1979 nineteen replicate samples were collected in a motel room
in the TP service area but none of them had coliforms present. In the rest
of the TP service area on that day 6 out of 22 sites sampled had verified
coliform organisms present. The conclusion from this is that even when
coliform densities are relatively high in a good portion of a distribution
system there can be parts of the system where coliforms are absent or, at
most, present in very, very low densities.
During the sampling of DT 4 replicates were collected at 4 sites on
1/16/80, at 3 sites on 1/17/80, at 5 sites on 1/29/80 and at 5 sites on
2/5/80. None of these samples had coliforms present. During the sampling
of BL 8 replicate samples were collected at 6 sites on 4/24/80 but no coli-
forms were found. In MI 4 replicates were collected from each site sampled
and no coliforms were found on six different sampling days on which a total
of 115 sites were sampled. In SR no coliforms were found in 4 replicates
from 8 sites on 11/11/80 and in 8 replicates from 4 sites on 11/19/80. In
MW no coliforms were found in eight replicates from 14 sites on 11/19/80 and
in BG no coliforms were found in four replicates from 7 sites on 12/2/80.
Some of the replicate samples referred to in this paragraph were 200ml
samples but since coliforms were not found, it seems reasonable to assume
that coliforms would not have been found if only 100ml of those samples had
been filtered. The results of all these sampling efforts tend to confirm
the conclusion that several replicate samples per site may not reveal any
coliforms.
One sample with no coliforms from a particular site is usually inter-
preted as a coliform density of less than I/100ml at that site. Do several
replicates from a particular site provide any more information? This ques-
tion may be treated statistically using the binomial distribution. One
sample provides only about 63% confidence that the coliform density at that
site is actually
-------
al information, but the amount of information gained with each additional
replicate decreases as the number of replicates increases.
5.6.5 Positive Sampling Results
It was already shown in Table 5.24 that no significant differences in
the average coliform counts were found between the first and second 100ml
replicates. Thus, although collecting and analyzing a second 100ml replicate
may change the average coliform count, the change is not significant within
the range of variability of the counts. Indeed, it would be very strange if
the results did show a significant difference.
Does collecting second 100ml replicates increase the number of sites
where coliforms are found? Comparison of columns 2 and 5 of Table 5.25 shows
clearly that it does. The increase in the fraction of sites positive gained
by taking the second replicate sample varies between 0.002 and 0.096. How-
ever, as might be expected, the increase in fraction positive is greatest
for the systems where many coliforms were found and least where coliforms
were rare. Thus, the gain in information about the overall level of contam-
ination in the system is small. The values in column 3 of Table 5.25 are
the binomial probabilities, P=l-(l-p)2, for having either one or both of
two replicates positive where p is the probability of a single sample having
coliforms (column 2). The fact that the actual values in column 4 are less
than the calculated values in column 3 merely shows that the two replicates
are not independent measurements on the system.
Does collecting the second replicate help in locating sites where coli-
form densities are high (>4 per 100ml)? The last two columns of Table 5.25
shows that it sometimes helps, but not a great deal. The systems for which
the greatest increase was found were those with very high levels of coliforms
which were frequently found with only one sample.
All-in-all, collecting and analyzing a second 100ml replicate does not
add a great deal of information. Systems with low levels of contaminations
and systems with high levels of contamination show the same result with
either one or two replicates per site if enough sites are sampled.
5.6.6 Larger Volume Samples
Interference with the detection of coliforms in 1000ml samples due to
suspended matter was observed on several occasions. Thus, samples of that
large a volume using the MF technique must be considered to be unreliable
unless the interference problem can be overcome. This could be accomplished
by filtering the sample in several portions. Actually, the same type of
interference was observed in some 200ml samples collected in SR on 11/5/81.
However, those were samples collected soon after the system had been flush by
opening fire hydrants and were very turbid. Such interference is usually not
seen in 200ml samples.
A comparison of sampling results obtained from both 100ml samples and
200ml is made in table 5.26. Data from days on which no coliforms were found
were eliminated. Of course, only data from those days on which both 100ml
112
-------
CO
a
co
rH
d
0
rH
w
H
-a- <
N. r^
10 £
W |=)
s 2
*** r*H
E-* O
oi
H
Q
FK
O
CO
M
CO
S
<
5
4J O
C 1H CM
gg co 03
« 41 0 IH
U X 41 41
« o a
O >-'
•H
i-l IW
O. O 41
ad • 01 -H
4J 4) 4J
,C O 4-1 iH
4J CO iH m
O IH CO O
•IS
tj j (y
fr"5 ti 4-1
m o c
cn O rH
rS O
4-» O
C -H rH
41 (0 01
4-1 41 C IH
(0 £ 41 41
0 0 ^
i-H
a
41 o • >
m c 41
•O CT> O 4J
c o a
0 H
0 •
4) 4-*
CO O •
CO 01
rl O
•
^
* 1
m a 4J
cn o ti
O H
H
>i<5
4) 4J O
4J fj iH r-H
CO CO 01
u 4i c: ri
•H 2 4) 41
rH 0 0,
O. ^
cS
• r*
4J &•£ U-4 j^
tn in c 4)
•H u a
Ci-i • h~l
4J
y •
cd w
M 0
U-l
o
(0
• Q)
01 4J
O -rl
a w
•o
e 41
4) rH
4J p.
0] g
^» cd
CO IO
CM
•
o
00
CO
o
•
o
o
+1
CM
O
O
m
rH
o
+ 1
rH
i-l
O
in
CO
d
+ 1
> 01
rl 4-1
S 43
•Q ciO
13 "H
O 41
co
•
oo
A
CT>
cn
CM
•
oo
O
co
A
+ 1
r-.
CO
VO
A
rH
VO
O
+ 1
O
CM
CM
VO
•*
CM
A
+ 1
hx
vO
-a-
A
Ix.
in
o
+ 1
vO
oo
rx.
r>
rH
f*
4J O
ti to
CO 41
^ M
»
CO
•
VO
A
O
CO
co
•
CO
CM
A
+ 1
00
vo
A
-a-
m
o
+ |
cn
in
CM
rH
VO
H
A
+ 1
•-I
CM
co
A
CM
m
o
•fl
cn
CO
CM
m
m
CM
41
U 4-1
CO C
M CO
h i-H
41 P-.
H
O
vo
•
O
A
VO
•a-
o
•
rH
O
A-
+ 1
0
rH
O
A
CM
CM
O
+ 1
CO
CM
O
0
cn
O
A
tl
O
in
o
A
m
CM
o
+1
cn
CM
o
-a-
fx.
•H
g
§
O
M
C
•H
O
P
rH
•O
rH
•
JJ
X
CO
•
rH
A
.4.
CM
rH
•
00
o
A
+ 1
O
00
d
A
0
-3-
O
+ 1
vO
00
O
CO
cn
O
A
+ 1
CO
cn
d
A
m
co
o
+ 1
m
vO
o
vO
00
rH
jj
3
CO
a
•rl
IH
P.
rH
^
i-H
CO
rH
*
vO
H ti
•o a)
<0 •-!
M O
113
-------
TABLE 5.25
EFFECT OF SECOND REPLICATE 100ml SAMPLE ON FRACTION OF SITES
POSITIVE AND PERCENT OF SITES WITH 4 OR MORE COLIFORMS PER 100ml
System
Sampled
(1)
Coatesville
Woodbury
Heights
Brant
Beach
Terrace
Plant
Downingtown
Brooklawn
Mt. Idy
Spring Run
Marshallton
Woods
Bradford
Glen
Fraction
of 100ml
Samples
Positive
(2)
0.019
0.184
0.203
0.249
0.026
0.021
0.013
0.0755
0.081
0.098
Fraction of Sites
Positive from 2-
100ml Samples
Calculated Actual
(3) (4)
0.038
0.0333
0.365
0.436
0.051
0.042
0.025
0.145
0.155
0.186
0.038
0.239
0.299
0.380
0.046
0.033
0.015
0.124
0.131
0.152
Gain in Percent of Sites
Fraction of with >4 per 100ml
Sites First Both
Positive Replicate Replicate
(5) (6) (7)
0.019
0.055
0.096
0.131
0.020
0.011
0.002
0.048
0.050
0.054
1.08
5.58
10.73
10.58
1.15
0.84
0.49
3.76
2.04
8.70
1.08
7.44
14.12
15.29
1.72
0.84
0.49
4.30
3.06
8.70
114
-------
TABLE 5.26
COMPARISON OF 100ml AND 200ml SAMPLES
System
Sampled
(1)
Downing town
Mt. Idy
Spring Run
Marshallton
Woods
Bradford
Glen
Fraction
of 100ml
Samples
Positive
(2)
0.018
0.029
0.140
0.099
0.265
Fraction of
Sites Positive
from 2-100ml
Samples
(3)
0.036
0.034
0.220
0.185
0.412
Fraction
Positive
of Sites
from 1
200ml Sample
First
Replicate
(4)
0.107
0.057
0.170
0.072
0.294
Second
Replicate
(5)
0
0.046
0.170
0.074
0.294
Percent
with >
of Samp.
4
Coliforms/lOOml
100ml
Samples
(6)
1.1
1.2
8.2
2.4
11.8
200ml
Samples
(7)
0,6
1.2
8.2
1.9
23.5
Note: Data only from sampling days on which at least 2-100ml and 2-200ml
replicates were taken at some sites and coliforms were found in
all samples.
115
-------
and 200ml replicates were taken could be used. Data from only the first two
100ml replicates and the first two 200ml replicates were used because there
were only a few days on which additional replicate samples were collected.
The fraction of 200ml samples in which coliforms were found was usually
greater than the fraction of 100ml samples in which coliforms were found
(columns 4 and 5 versus column 2) but not much greater. The data from SR,
MW, and BG suggest that two 100ml samples give a better chance of detecting
coliforms at a sampling site than one 200ml sample (columns 4 and 5 versus
column 3) but those differences are not statistically significant. In four
of the five systems sampled the percent of samples with >4 coliforms per
100ml in 200ml samples was equal to or less than the percent for 100ml sam-
ples. Only in BG where only 17 sets of samples were used for the data in
Table 5.26 were the 200ml samples better for detecting high coliform densi-
ties.
During the sampling of MW several replicate 100ml and 200ml samples were
collected at each site sampled on three different days. It was expected that
these sets of samples might show that 200ml samples were significantly better
for detecting coliforms or high densities of coliforms than 100ml samples.
However, the data do not demonstrate this because the fraction of samples
positive and the coliform densities were actually greater for 100ml samples
than for 200ml samples. This puzzling result is probably caused by the high-
ly non-uniform dispersion of coliforms in these systems and should not be
extrapolated.
The data used for comparison of 100ml and 200ml samples demonstrates
mainly that coliform counts on samples from water distribution systems are
quite variable. There is no empirical evidence to show that 200ml samples
give a significantly greater probability of detecting coliforms or high
densities (>/100ml) of coliforms than 100ml samples. Intuitively, it seems
that 200ml samples should be better than 100ml samples. However, this intui-
tion is explored from a theoretical approach in the next part of this section.
The following is intended to demonstrate that there is at least one
theoretical model of coliform dispersion in water distribution systems which
is consistent with the conclusions about 200ml sampling versus 100ml samples;
that is, that they are not significantly different for detecting coliforms
or high densities of coliforms.
Assume that the dispersion of coliform bacteria in some water distribu-
tion systems follows a lognormal distribution. Consider six such systems
designated A, B, C, D, E, and F with geometric means and geometric standard
deviations of 10"1 and 10, 10~3 and 100, 10~5 and 1000, 10~2 and 10, 10~4 and
100, and 10~6 and 1000 respectively. The values of the parameters of the
distributions have been selected to conform to the ranges of these parameters
estimated from sampling data. Systems A, B, and C have arithmetic mean coli-
form densities of more than one per 100ml and should fail to meet the micro-
biological MCL of the Regulations. Systems D, E, and F have arithmetic mean
coliform densities of less than 1 per 100ml and should meet the microbiologi-
cal MCL of the Regulations.
116
-------
Assume that in some time period in which the bacteriological quality of
the water does not change, 1000 samples are obtained from each of these six
systems. The expected results from such samplings are summarized in Tables
5.27 and 5.28. The values of the fraction of sites sampled in each density
range were obtained from a table of the standardized normal distribution
using the logarithms of the densities to calculate the standardized normal
variate, Z = (X)/S, from a statistical table. The numbers of samples expect-
ed from each density range were calculated by multiplying the fraction of
sites by the assumed number of samples. The values in the last three columns
were obtained by multiplying the number of samples in each density range by
an assumed count in that density range and dividing by 1000. The samples
with more than eighty coliforms per 100ml were averaged in as 80, those in
the range of 10-80/100ml as 30, those in the range of l-10/100ml as 3, those
in the range of O.l-l/100ml as 0.3, and so forth. Thus, the 21 samples
collected from sites in system A where the coliform density is between 10
and 80 are assumed to have an average of 30 coliforms per sample for a total
of 630 coliforms which contribute 0.63 to the arithmetic mean of the coliform
count from 1000 samples. Notice that if enough samples are collected from
sites where the coliform density is low, it is expected that some coliforms
will turn up. For instance, the 191 samples collected from sites in system
B where the coliform density is between 0.001 and 0.01 are expected to turn
up one coliform. However, samples from sites where the coliform density is
less than 10~-VlOOml are not expected to turn up coliforms because it would
require 1000 standard 100ml samples of water with a density of 0.001 per
100ml to turn up one coliform. The significant item to notice from these
tables is that the largest contributions to the arithmetic mean coliform
densities are made by the few samples from sites with coliform densities of
more than 10 per 100ml and for systems B, C, E, and F by the samples from
sites where the density is more than 80 per 100ml.
The calculated values for the arithmetic mean, « = exp [In Xg+l/2(ln
S 2)] for the six distribution systems are: A, « = 1.417; B, a = 40.29;
C, « = 229,949.74; D, « = 0.142; E, a = 4.029; and F, « = 22,994.97. The
large differences between the theoretical arithmetic means and those calcu-
lated for a hypothetical set of 1000 samples is due to the occurrence of sam-
ples with more than 80 coliforms per 100ml. The use of the lognormal distri-
bution to represent the frequencies of various counts forces the conclusion
that some of the TNTC samples contain many more than eighty coliforms per
100ml. One sample with thousands or millions of coliforms would completely
dominate the theoretical arithmetic mean but was averaged in as >80/100ml
for the hypothetical sample sets.
In this analysis, it is assumed that collection of two 100ml samples at
a given site is equivalent to collecting one 200ml sample at the same site.
A consequence of this assumption is that, on the average, 200ml samples will
turn up twice as many coliforms as 100ml samples. This assumption is not
inconsistent with the sampling data presented earlier; although the data do
not demonstrate it.
Table 5.29 gives the expected numbers of counts of 0, 1, 2, and 3 coli-
forms per sample for both 100ml and 200ml samples from sites with densities
in the range of 0.001 to 10 per 100ml. These counts were obtained by assum-
117
-------
g
PO
H B
co o
3 rH
d i!
& a
i g
U 0
^ i
in B IH
3 IS §
M H4 J3
^ k*J JP
b^ &f
£
4J
d 0
Joe
s
B
C m
«S w
rH 0
t5 -g
/- H
4J
^s
j
^5
o
4J
0 M
9 V
o "S
O
00
00
U
0
o
NO
m
m
o
o
NO
o
O v*H
^^
n p*
•< CM
o
rH
O O
0
^
fx
PQ O
O
ON
rH
< §
O
G
2
5
0
00
A
O
NO
c
O
o
00
sr
d
o
VO
o
CM
rH
NO
rH
rH
CM
rH
CM
3
d
P^
m
o
0
ON
o
CM
o
0
o
00
2
m
o
(1
o
CM
CO
rH
d
00
O
sr
0
*f\
C*4
^
"*
NO
CO
1-1
15
CM
0
d
o
I
o
ON
m
CO
rH
o
o
rH
1
rH
CO
rH
O
*
O
00
CM
O
d
CM
O
rH
0
^
^*
CM
ON
rH
CO
CO
I
d
ON
H
ON
O
O
CO
H
•*
CO
0
rH
1
rH
d
CM
O
o
d
m
o
o
d
o
rH
O
O
.r-
O
a
r4
sr
CO
ON
NO
O
d
00
ON
f-f
0
CO
sr
CO
o
rH
'o
rH
CM
1
S
1
rH
O
O
d
i
CO
ON
1-1
ON
rH
NO
CO
rH
1*.
CM
ON
O
d'
m
rH
ON
rH
O
ON
m
CO
rH
O
CM
I
0
rH
CO
1
0
H
O
eo
CO
1 1 •
rH
NO
O
CM
1 1
rH
O
rH
1 1 CO
H
CO
0
£
o
1
1
ON ON -H
rH CM Id
rH NO <
iH ON
ON 0
rH CO
CM rH
CM
CO CO
ON ON
rH CM
rH NO
* •
O O
in i~»
rH l-l
ON sr
rH CM
0 0
in co
H H
CM 0
0 0
0 0
CO
O
rH
tl
0
O rH
rH V
g
0
H
t
60
CO
*
m
i
o
i
iff
BW
U
§
4J
™
-fc
O
o
rH
I
60
CO
M
CO
s
1
iff
m
u
GO
5^
CO
.«
o
rH
60
CO
A
rH
d
I
iff
..
<^J
S
•M
00
to
118
-------
TABLE 5.28
EXPECTED NUMBERS OF COUNTS IN SAMPLES
FROM MIDDLE RANGE COLIFORM DENSITY SITES
(Based on 1000 Total Samples)
System Density No. of 100ml Total
Range Samples with Samples
(per 100ml) Counts of
0123
A 1-10
0.1-1
0.01-0.1
0.001-0.01
B 1-10
0.1-1
0.01-0.1
0.001-0.01
C 1-10
0.01-1
0.001-0.01
0.001-0.01
D 1-10
0.1-1
Ot01-0.1
0.001-0.01
E 1-10
0.01-1
0.01-0.1
0.001-0.01
F 1-10
0.1-1
0.01-0.1
7
252
331
136
2
68
145
190
1
32
65
93
1
100
331
340
1
32
89
150
1
19
43
20
77
10
0
7
20
5
1
4
11
2
0
3
31
10
1
2
11
3
0
2
5
1
30
11
0
0
10
4
0
0
6
1
0
0
5
5
0
0
4
1
0
0
3
1
0
30
1
0
0
10
0
0
0
6
0
0
0
5
0
0
0
4
0
0
0
3
0
0
136
341
341
136
44
92
150
191
25
44
67
93
22
136
341
341
16
44
92
150
13
25
44
Total No. of 200ml
Counts Samples with
Counts of
0123
408
102
10
0
132
28
5
1
75
13.
2
0
66
41
10
1
48
13
3
0
39
7
1
0
187
321
135
0
50
140
190
0
24
63
92
0
74
321
339
0
24
86
149
0
14
42
2
112
20
1
1
30
10
1
0
15
4
1
0
45
20
2
0
15
6
1
0
8
2
6
34
0
0
2
10
0
0
1
4
0
0
1
14
0
0
1
4
0
0
0
3
0
12
8
0
0
4
2
0
0
2
1
0
0
2
3
0
0
1
1
0
0
1
0
0
Total
Samples
136
341
341
136
44
92
150
191
25
44
67
93
22
136
341
341
16
44
92
150
13
25
44
Total
Counts
816
204
20
1
264
56
10
1
150
26
4
1
132
82
20
2
96
26
6
1
78
14
2
119
-------
cr>
CM
m
M
pj
9
H
CO
S5
O
H
B
S
H
M
|
SJ
0
U
o
»J
3S
9
E
o
g
J§
55,
o
CO
H
i
Kj
C_| -
O
^
la
o
H
01
a.
fi
o
•H
O
U
rH
c
p
a
a
-fl
j-
.U
"SI-
o
1
4J
>.
CO
O
•H
4J
01
1
•H
M
<
°
a
o
•H
4-1
3
•H
4-1
a
8
tW
o
M
01
,0
y
*
"S
4J
U
(U
CO
01
•H
CO
<4H
O
a
o
•H
4J
U
BJ
r*4
M-l
O
01
61
fft
CO
01
rH
CO
CO
O
O
O
rH
O
a
CO
£
•8
4J
CO O
U O
•H O
•O rH
rHVH
o
*s
CO O CO
01 01
rH 01 rH
a so o.
OT « W
.g
CO
4J
C fl)
SCO
a
U cd
(2
e^
O 01
MH 4J
^H w
iH 0
O *H
rt
«JS M
4J
•s
^
"d
Q
O
tO rH
4-1
S M
O v-/
0 0
CM 00
en H
EH 0 o
O O
vo in
rH rH
o o
o
a i o
r»t ^> ^Q
W CM m
O 1 H
en vo
fm
0
O
VO vO
rH ST
o o
U O O
• •
en
rH rH
§o
o
V
0
00
0 1
CO O
A H
- vo
•* VO CM 00
-* vo Cft -»
O O O I-.
ov oo in CM
rH en rH en
ON sr ov en
O rH rH -*
• • • •
en en OA oo
rH rH m CM
sr sr en CM
en en H O
rt «M en
1 1 1
O O O
rH rH rH
1 1 1 ST
CM en sr 1
1 1 1 0
O O O rH
rH rH H V
O
1
60
CO
1* "
o
rH
1
I*
£
8
01
4J
eg
^
CO
• *
o
o
rH
1
60
CO
„
-------
ing that the coliforms from samples in a given density range are randomly
distributed among those samples and calculation of the counts from Poisson
probabilities. For instance, the 341 samples from sites in system A where
the coliform density is between 0.1 and 1 per 100ml are expected to yield 102
coliforms for an average of 0.299 per 100ml. The probability of obtaining a
count of zero (negative sample) is e~0«299 or 0.7415, which multiplied by 341
samples gives 252.9 samples. The 252.9 samples would usually round off to
253 but the most likely way to distribute the 102 coliforms is in 77 samples
with one coliform, 11 samples with 2, and 1 sample with 3. This leaves 252
samples with no coliforms. It is assumed that if those samples were 200ml
samples (or two 100ml samples) they would yield 204 coliforms for an average
of 0.598 per 200ml. The probability of obtaining a count of zero in this
case is e~0.598 or 0.550 which when multiplied by 341 samples gives 187 sam-
ples.
This method of distributing the coliforms among samples could be called
a segmented lognormal-Poisson model. It is necessary to change, at some
point, from the lognormal which is a continuous distribution to a discrete
distribution to obtain integral counts per sample. The Poisson is the sim-
plest discrete distribution to use for this purpose and the parameters of
the lognormal distribution are preserved. The division of the density ranges
into factors of 10 is arbitrary. They could be divided into smaller ranges
but finer division of the density ranges would not significantly change the
results of these calculations.
It is expected that all samples collected from sites where the density
is greater than 10 per 100ml will have some coliforms present and that all
samples collected from sites where the density range is less than 0.001 per
100ml will have no coliform present. Thus, the fraction of the sites where
coliforms are found will be changed by collecting two 100ml samples or one
200ml sample per site only where the density is in the range between 0.001
and 10 per 100ml. This is illustrated in Tables 5.30 and 5.31. The fraction
of the sites sampled changes only a small amount when the sample volume is
doubled because only a portion of the sites have a coliform density in the
range of 0.001 to 10 per 100ml.
This result is not intuitively obvious. It can be made clearer by di-
viding the sample sites into three categories. At sites where the coliform
density is greater than 10 per 100ml, there is an extremely high probability
that some coliforms will be found in either 100ml or 200ml samples. At sites
where the density is much less than 1 per 100ml, the probability of finding
a coliform is twice as high for 200ml but still so low that many samples
must be collected before even one coliform is found. (Note that the 1000
samples used in the example is far more than is required for any distribution
system in a months time). The sample volume will make a difference only
where the coliform density is close to 1 per 100ml.
The sample volume should have no effect on the measured coliform density.
This was found, within the sampling variability, for the systems in which
both 100ml and 200ml replicates were collected. Doubling the sample volume
will only slightly increase the number of sites where coliforms are found.
This is also true within the variability of the sampling results for those
121
-------
0
CO
m
S
to
H
CO
B
H
CO
vS
1
E*"*
w
E!
E
Z
u
s
^J
^>
a
1
CO
to
o
("*
M
W
H
1
o
H
M
01
a
g
o
IM
iH
CO rH
J2 O
S o
o
M
rJ fl
8|
pH
fn QJ
H M
JS
S
m
0
w
*J
CO
CO
^^
IM
O
o
2
U
u-i
O
O
*
IM
O
O
Z
m
05 tm
01 O
I-l
S J2J
CO
i °
o o
04 Z
IM
O
•1
O
•1
u
IM
O
s
IM
O
jg
OQ
IM
O
O
CO PS
4)
rH
& _
B
CM CM 00
rH
O i-l
00 O 1
O 1 rH rH
00 O 1 .
A H H 0
CO
VO
•*
0
H
O
i-l
rH
CM
CO
O
04
m
CM
m
i-H
m
f^
CO
CO
o
I-l
o
i
rH
O
o
vo
OV
H
O
H
H
m
co
rH
1-1
CO
Oi
0
o
o\
rH
i-H
vo
CO
o
rH
0
O
H
O
O
O
c/\
,_j
rH
O
i-l
rH
O
CM
04
O
rH
rH
O
rH
O\
O
04
04
O
CO
1
o
rH
1
O
H
O%
CM
VO
0
o
CO
o
rH
O
CTV
CM
vO
6
OV
0
CO
o
rH
O
«»
o
rH
V
JS.
CM
a>
CO
r-
O
co
00
o
CM
rH
vO
vO
vO
•a-
CO
CO
ON
CO
OV
rH
vO
2
m
o\
tv.
rH
m
CM
rH
a
4J
H
CO
O
o
o
04
i-l
co
CO
O
rH
vO
O
•
0
m
O
o
H
m
CM
o
o S*
•rl iH
4J 4J
0 -rl
g) CO
M O
122
-------
rH
CO
m'
TABLE
CO /— s
W H
H J
M O
CO O
i-H
PH
O M
CO
2 Pu
S a
H C
O O
M
t5 CO
PH CO
1-4 CU
^C ,^*1
CO 4->
PH &
O
CO
H 6
U CU
W 4J
PH CO
PH t*.
W CO
**^
4-1 CO
O • CU
• 00 4->
O CU iH
2 & CO
4-1 CO
PH O • CU
• CO 4-1
O O iH
2; PM co
4-4 CO
O • CU
• 00 4J
O CD i-l
2 2 CO
CO
CU 4-1 CO
i-H W O • CU
(X . tO 4J
g O O iH
CO 2 Pk CO
CO
H 4-1 CO
B o • cu
. &0 4->
O O CU iH
0 2 2 CO
4-1 CO
Q O • CU
• CO 4->
O O *H
2 PH CO
4-1 CO
O • CU
• 00 4J
O 0) -H
2 2 co
. CO
O • CU
• CO 4-1
O O iH
2 PH CO
4-1 CO
O • CU
• 00 4-1
O CU tH
co 2 2 co
fll
w
rH 4-4 CO
DM W O • CU
s • w 4-i
CO O O iH
CO 2 PH 00
'jd
4-1 CO
O O • CU
O • 00 4J
rH O CU iH
2 2 co
_
4-1 CO
O • CO
• CO 4J
O O -H
2 Pi CO
^>
4-1
•H CU
co oo
CO rd
Q K
^5 1 ^3 "^
•-I
sf vO CO i-H
i-H i-H
o o o sr
CN
CN m vo o
i-H CN
o o sr
r*^
1 i-H CN CN
CN vO
O O i-H Ov
i-H
m^- vO CM \O
rH
O O i-H CN
co
CN m m CN
rH rH
O i-H O
o
rH
rH rH VO
CN CO
O rH
co o I
O 1 rH rH
oo o i •
A i-H i-H O
CN
rH
i-H i-H
o
m
rr\
i-H O
0 0
m m
0
0
C^ vO
CO CO
CO i-H
CO
ON
00
CN O
f*^ t^
0 O
<-l •— 1
O
r-» co
vO CTi
i-H
°"
o o
0 H
m cr\
rH rH
C^ CTN
CN CN
O
O
CO
VO
0 0
r-. r^
CO CO
o
d
O vO CO
sr co CN
CO rH
i-H
vO
i-H 0
O^ CT^
\O vO
o
•
o
rH
O co
1 C CU
o o o >
1 rH -rl iH
rH 1 ST 4J 4J
o sr i o -H
O 1 O CO CO
O rH J-l O
O rH V PH PH
123
-------
systems where 100ml and 200ml samples were collected. In terms of a monitor-
ing program, the information gained either by taking two 100ml replicates or
by doubling the volume of the standard sample is very small. On the other
hand collecting and analyzing 200ml replicates is no more expensive than
collecting and analyzing 100ml replicates if the MF technique is used for
examination and the samples are transported by vehicle. Sending the larger
volume samples by mail would cause some extra expense. An increase in the
frequency of occurence of interference due to particulate matter on the
membrane filter could be expected with larger volume samples. Using larger
volume samples could have some advantages for larger systems which have their
own microbiological laboratories and can deal with the interference problem.
However, for smaller systems, it could be a disadvantage without equivalent
benefit.
124
-------
SECTION 6
MECHANISTIC APPROACH
6.1 INTRODUCTION
The mechanistic approach is an effort to make use of physical informa-
tion about the system such as the sizes and locations of the pipes, the
layout of the pipes, the amount of water in the system, and the flow of
water through the system as a guide for microbiological monitoring of a
water distribution system. Various components of the mechanistic approach
are explored in this section. A method of describing the configuration or
layout of the distribution system is developed. Then several hypotheses
about mechanisms by which coliform bacteria are dispersed in a water distri-
bution system are examined. The goal of this approach was to develop infor-
mation which could be useful in identifying areas where there is a high
probability of coliform contamination.
6.2 CONFIGURATION OF SECTIONS
A water distribution system can be divided into isolated sections based
on projected water flow patterns. This division aids in the design of sam-
pling programs. By recognizing sections of a system, samples may be taken
from each section and hence each segment of the water flow. Sections delin-
eated on the basis of pipe layout and hence water flow patterns may have
different probabilities of coliform occurrence. Recognition of this could
aid in understanding why coliforms are or are not found within different
parts of a system.
6.2.1 Definitions of Distribution System Elements
According to the National Interim Primary Drinking Water Regulations
(EPA 1977) a water system means a system for the provision of piped water
for human consumption and includes collection, treatment, storage, and dis-
tribution facilities. We consider the distribution system to include the
pipes used to convey water after collection and treatment (if any) to the
service connections plus the storage facilities connected directly to such
pipes. In some cases, although this did not occur in any of the systems we
sampled, chlorine may be added to the water already in the distribution
system particularly as it is withdrawn from storage.
We have designated the water conveyance elements (pipes) of a water
distribution system as follows: A transmission main is a pipe which carries
water from the source (well or treatment plant) to the distribution system.
Transmission mains are typically the largest pipes of the system. Any bacte-
125
-------
riological quality changes which occur in the transmission mains may affect
water quality in the entire system. A distribution main is a pipe which
carries water into or out of an isolated section of the system. It has a
distinct transmission function but the area served is more limited than that
of the transmission main. Flow reversals in a distribution main occur only
when it is connected with a storage tank.
A street lateral is a pipe which serves to deliver water to customer
services within a single city block and sometimes transmits water to other
street laterals. We call a street lateral which transmits water to at least
two other street laterals a transmission street lateral. Flow reversal in
transmission street laterals may occur frequently due to changes in water
demand. A street lateral which supplies water only for local use is called
a peripheral street lateral, of which there are two types. A street lateral
which delivers water only to customer services within a single city block and
usually ends in either a fire hydrant or a flush out valve is called a dead
end. Flow reversal is impossible and the water is semi-stagnant. A loop
consists of two dead end street laterals connected together.
In addition to the water conveyance elements of a distribution system,
we define pumping stations and storage facilities. Pumping Stations are
located on transmission mains or on distribution mains and are intended to
increase the water pressure on the downstream side. They may be used as
sampling locations for determination of the water quality supplied to a
section or several sections of the distribution system. Distribution system
storage is either elevated or is operated in conjunction with a pumping
station. The purpose is to maintain pressure in parts of the system and to
serve as a reservoir of water to meet extraordinary demand such as fire flow.
Standpipes and storage tanks usually have large volumes of water which mix
and interchange only slowly with water which is actually distributed to
services. Uncovered reservoirs are well known to be sources of bacterial
contamination and have been eliminated from many distribution systems. Even
covered reservoirs, storage tanks, and standpipes may allow access of birds,
rodents, plant materials, dust, etc. to the water in the distribution system
and are likely sites of bacterial contamination of water in the system.
6.2.2 Isolated Sections
An isolated section of a water distribution system is a collection of
street laterals and distribution mains which supply water for local use to
several blocks. There is input to the isolated section from at least one but
no more than two mains. There may be output from the section to other iso-
lated sections. Water quality changes which occur within an isolated section
of a distribution system may not be measured by samples collected from other
parts of the distribution system. Therefore, some samples must be collected
in each isolated section to insure that the samples as a whole are represen-
tative of water quality conditions throughout the entire system.
The four configurations which the pipes that make up an isolated section
can have are linear, loop, dendritic, and grid, as illustrated in figure 6.1.
An isolated section can have a compound configuration; i.e., a combination of
two of the basic patterns; but not a complex configuration; i.e., more than
126
-------
WATER SOURCE
SECTION B LINEAR
1 SAMPLING LOCATION
SECTION C LOOP
4 SAMPLING LOCATIONS
i r
•TRANSMISSION MAIN
1 I I
i i I I
i i
I T
I I
l I
I i
il I i
I I
j i
I I I
I I I
SECTION A GRID
31 SAMPLING LOCATIONS
DISTRIBUTION MAIN
DISTRIBUTION MANIFOLD
i i i i
STREET
LATERAL
SECTION D DENDRITIC
9 SAMPLING LOCATIONS
Figure 6.1 Representative Examples of Four Isolated Section
Configurations.
127
-------
two of the basic patterns. If a mixture of three of the basic patterns is
found, it is divided into two or more sections. Thus, in addition to the
four basic configurations, there are six possible compound configurations
for a total of ten possible configurations overall.
For this project, a block on a street which usually corresponds to a
street lateral or a portion of a distribution main with service connections
is designated as a sampling location. The ends of the pipes corresponding
to sampling locations are joints connecting with two or more other pipes.
Usually, the street lateral is underneath a street and the joints are under-
neath intersections so that the sampling location corresponds to a block on
a street and may be sampled from service connections on either side of the
street. However, this needs to be checked from a blueprint of the distribu-
tion system because there sometimes are two laterals underneath a street, a
street lateral may be two blocks long, or some other unusual situation may
occur. Any service connection in a building, either residential, commercial,
or publically owned is a sample site. Any tap or hydrant in or associated
with the building should be the same as any other for sampling purposes as
long as the water is allowed to run for a few minutes to clear the service
line and obtain water from the street lateral.
6.2.3 Configuration of Systems Sampled
Table 6.1 summarizes the structural parameters of the nine systems sam-
pled. The number of sampling locations is more or less proportional to the
population served but the number of isolated sections is a function of other
factors. The largest system sampled (LB) was a single large grid which was
divided into four sections based on the location of the two treatment plants
which provide the water. It is located on flat terrain and was not crossed
by any highways or railroad tracks. The two systems which have the most
isolated sections (CV and DT) are located in hilly terrain and are divided
by highways and railroads. WH and BL are relatively small systems which are
divided into isolated sections by highways and railroads. The other four
systems are located on hilly terrain but are so small that they have few
isolated sections.
As shown in Table 6.1 the number of sampling locations corresponds very
well with the number of pipes (one per block) and the number of joints (one
for every two blocks). The number of dead ends is more variable and depends
upon the configuration of individual sections.
The last four systems sampled are too small, to contribute much to the
analysis by isolated sections. The first five systems sampled had a total of
twenty eight sections. Eight of the ten possible configurations were found.
The most frequently found configuration was the dendritic grid(10) followed
by the grid(6). The other configurations found were dendritic(4), linear
loop(3), loop grid(2), linear(l), linear grid(l), and dendritic loop(l).
Thus, the grid and grid combinations comprised 68% of the sections delin-
eated. The dendritic pattern and its combinations comprised 54%; the loop
and its combination, 21%; and the linear and its combinations, 18%. The
linear-loop and loop-dendritic were not found in any of these five systems
but were found in the very small systems.
128
-------
Q
£
CO
co
s
w
H
rH CO
VO p4
O
W
j_J CO
M HH
5 CO
H >H
1
<3
H
^
PH
£3
H
0
§
H
CO
a
o
CO 4J
Pt CJ
H 0
cd M
0) 0)
O P.
r4
CD
P. PI
O
CO -H
4-1 4J
Cj Cd
•H CJ
O O
CO
T)
Pi
4-< W
o
• cd
O CU
53 O
(4-1 CO
0 4J
a
• iH
0 0
fe! ^
u-i ca
O CU
• p.
O iH
00 PI
a o
ii i , | , i
•+H *rt *rt
O i-H 4J
p. cd
•00
O CQ O
^ W i-^
T3 CO
0) PI
14-1 4-1 O
O cd iH
M 1 if
• o o
O CO CU
IS M CO
PI
o
4-1
cd -o
rH CU
P. t
O CU
PH CO
0
CU
4-1
co
CO
CN CN rH CTl -* -* CO
rH CN CN O rH in CN
ood ooodo
co O o oo ^* *3" r*^
•^ in in m m vo .OOCN
O\ vO i— | O ^f
H CO rH
CTif^in OcOrHOOcN
-^- CN CO r^ r-N rH rH
CO rH VO rH
coin>* -*«*Hr^cN
in CN ^d" 00 r>> rH rH
-* rH VO rH
CTlin-vJ" 1^-COrHCOrHrH
ooooooinoo
oooooocNmo
Ovooooor--rHinrH
O co co m a\ CM
CN si-
>ffipq HtJHPtfSO
U5rJ QPQSCOgpq
129
-------
Table 6.2 summarizes information on the structural parameters of the
five configurations for which more than one example was found. The size of
an isolated section is measured by the number of sampling locations in that
section. The structural parameters for the section show how well the physi-
cal make up of the distribution system corresponds with the layout of city
blocks for that section. The ratios of pipes and joints per sampling loca-
tion found for the entire systems also held for the individual sections.
The number of dead ends per sampling location did vary with configurations.
Dendritic configurations have an abundance of dead ends because that is basic
to the definition. Grids have the smallest portions of dead ends. This ap-
pears to be due to the large number of joints and pipes which made up a grid
and the fact that if many dead ends are found, the section was designated as
a dendritic grid. The most dead ends per location, joint, or pipe are- in
dendritic configurations. The other common configurations have varying num-
bers of dead ends per location, joint, or pipe. These relative numbers of
dead ends are large for these sections but they are not given a dendritic
designation as the other properties of the system dominated the configuration.
TABLE 6.2
SUMMARY OF STRUCTURAL PARAMETERS BY SECTION CONFIGURATION
Grid Dendritic Loop Dendritic Loop
Configuration Grid Grid Linear
Number of
Examples 6 10 2 4 3
Pipes per Location 0.89 0.93 0.86 1.0 0.93
Joints per Location 0.51 0.48 0.57 0.54 0.65
Dead Ends per
Location 0.07 0.20 0.11 0.36 0.25
Dead Ends per
Pipe 0.08 0.23 0.20 0.31 0.28
Dead Ends per
Joint 0.14 0.46 0.21 0.65 0.38
Pipes per Joint 1.69 1.97 1.57 2.06 1.40
130
-------
These analyses demonstrate that choosing sample locations on the basis
of street patterns often parallels the more complicated procedure of choosing
sample locations on the basis of the layout of the actual distribution mains.
There are descrepancies however. In one section of System CV the number of
lines per location and joints per location are exceptionally low (0.44 and
0.18, respectively). This is because of a dendritic pattern in which street
laterals extend through several blocks without joints to other street later-
als. The question that arises is, given a limited number of samples, might
it be more efficient in such an instance, to make a sampling location one
street lateral rather than one block?
The parameters of greatest variability are those involved with the num-
ber of dead ends. As dead ends are associated with decreased water flow,
sections containing a large number of dead ends were postulated to have
greater probability of coliforms from regrowth than other sections (Pipes
and Christian 1978, 1980).
Other aspects of section structure include the pipe length, volume and
average pipe cross sectional area. These parameters were calculated and are
presented in table 6.3. Pipe diameter and length for each location were
cataloged, the total length of each pipe size was summed for each isolated
section, and volume and average cross section of mains for the various sec-
tions. No one section configuration represented the greatest length or
volume of pipe. The smallest sections by either criterion were linear loop
and linear. Sizes of the other section types were medium to large with no
obvious pattern. If the flow pattern of water does in fact relate to section
characteristics, it is expected that the volume of the section may be posi-
tively correlated with the cross sectional area. Larger sections require
larger mains on average for the maintenance of adequate water supply. A
scatter diagram of the data is presented in figure 6.2. To examine this
hypothesis, a Kendall's rank correlation analysis and a Spearman's rank cor-
relation analysis were conducted between section volume and crosssectional
area. Both demonstrated a significant correlation for 28 sections analyzed.
The Kendall's rank coefficient of correlation was 0.360 providing a ts[HoT=0]
of 2.74 with a probability that the ts arose by chance of 0.0062. The
Spearman's rank coefficient of correlation was 0.44 with a probability that
the correlation occured by chance of less than 0.02. The major contributing
factor to the significance of correlation appears to be the presence of
large transmission mains in the large sections as would be expected.
6.2.4 Relationship of Configuration to Sampling
Clearly, samples should be collected from all of the isolated sections
of a water distribution system because contamination occurring in one section
usually does not affect water quality in other sections. However, it is not
clear that the number of samples collected and the locations used within an
isolated section should depend upon the configuration of that section. The
direction of the flow of water can be established in a rather straightforward
manner for linear and dendritic sections, and individual sampling locations
can be designated as either upstream or downstream of other locations. On
the other hand, the direction of flow in individual lines in a grid or loop
configuration is not obvious and probably changes from time to time with
131
-------
30
20
UJ
cr
< 10
o:
10 20
RANK VOLUME
30
Figure 6.2 Relationship of average cross sectional area to volume
of water distribution system sections.
132
-------
TABLE 6.3
DIMENSIONS OF DISTRIBUTION SYSTEM PIPES
BY SECTION
System Section
CV West
Central
N. Central
E.F.
Cain
S.C.
Foundry
Tooth
North
Configuration
Grid
11
Dendritic
"
Dend . -Grid
Dend . -Loop
Linear-Loop
ii ii
Grid-Loop
Average
Pipe
Area (ft2)
0.399
0.332
0.240
0.195
0.209
0.229
0.280
0.269
0.179
Total
Pipe
Length (ft)
31,609
98,704
8,708
11,220
15,118
21,355
8,220
3,985
6,880
Volume of
-Water
(ft3)
12,614
32,791
2,091
2,199
3,159
4,881
2,300
1,070
1,229
(gals)
94,359
245,298
15,642
16,449
23,609
36,513
17,205
8,004
9,194
WH
LB
DT
BL
Total
W. Central Grid
North Dend.-Grid
Southwest " "
East Dendritic
E. Central "
Total
North Grid
N. Central Dend.-Grid
S. Central " "
South "
Total
Central
North
Southwest
East
West
Northwest
Southeast
Grid
ii
Dend. -Grid
ii ii
Loop -Grid
Loop-Linear
Linear
Total
Central Grid
Southeast Dend.-Grid
Northeast Dend.-Grid
Total
0.307
0.174
0.193
0.196
0.369
0.203
0.235
0.215
0.185
0.198
0.161
0.186
0.350
0.298
0.237
0.295
202,589
15,500
16,750
14,675
18,400
8,100
73,425
33,165
56,636
66,330
65,925
222,056
33,045
15,735
13,765
19,680
32,205
7,300
10,040
131,770
9,965
13,150
9,970
33,085
62,331 466,273
2,698
3,240
2,881
6,784
1,650
17,253
7,123
10,485
13,130
10,627
41,365
10,034
7,230
5,420
6,973
11,690
1,535
5,712
48,595
3,492
3,917
2,364
9,771
20,182
24,232
21,551
50,753
12,345
129,063
60,433
78,533
98,212
79,495
316,673
75,060
54,077
40,544
52,162
87,447
11,483
42,729
363,516
26,122
29,294
17,676
73,092
133
-------
changes in water demand. Thus, samples collected anywhere in a section
within a grid or loop configuration may be considered to representative of
the entire section whereas samples collected in a section with a linear or
dendritic configuration may be used to isolate contamination in a street
lateral or possibly even in a portion of a street lateral. These ideas are
explored further by testing of specific hypotheses.
6.3 SAMPLING BY SECTIONS
Isolated sections of a distribution system are defined so that there is
some hydraulic separation. Microbiological contamination occurring in one
section may not be detected in samples collected from another section of the
same distribution system. Of course, it is possible that contamination could
be general throughout the system and would be detectable in samples collected
in any part of the system. The lack of significant differences among sec-
tions of a distribution system would not prove that the concept of isolated
sections does not hold but finding an instance of significant differences
between two sections of the same system would establish the validity of the
concept.
The other aspect of the question about the validity of the concept of
isolated sections is how large can an isolated section be. In some water
distribution systems, hundreds of city blocks may be supplied water from
mains and street laterals which are part of the same large grid. If there
is no hydraulic separation could contamination entering the distribution
system be detected by samples collected 10 blocks away, or 20, 30, or 40
blocks away?
The data collected during this project are analyzed to determine what
light they shed on these two questions. First, are there significant differ-
ence among isolated sections of the same distribution system? Second, in
the absence of a clear hydraulic separation, do adjacent sampling sites or
adjacent blocks give equivalent results?
6.3.1 Differences Among Sections
The coliform and standard plate count data for the six systems which had
more than one isolated section are summarized in tables 6.4 through 6.9. The
data for systems WH, LB, and BL are further subdivided by sampling period.
These analyses are based on the results in the first two 100ml replicate sam-
ples collected from each site sampled in order to give an equivalence of
data. The coliform counts and standard plate counts for each site sampled
are the average of the counts obtained on the first two replicates.
Differences in the extent of microbiological contamination among the
isolated sections of a distribution systems might be seen as statistically
significant differences in the mean coliform densities, in the mean standard
plate counts, or in the fraction of the samples in which coliform organisms
are found.
Significant differences among the mean coliform counts or among the mean
standard plate counts could be sought using analysis of variance. However,
134
-------
u
w
EH
CO
>-"
CO
Pn
o
CO
CO
§
5
•
•U M
C cd
3"-» >
•O O rH
M U 0
cd
T3 O h
C 4J 0)
CO cd CX C
4J rH v-' Cd
CO Cu CD
e
•
/-\ M
rH Cd
B a >
rl t^O
O 4J O
*"W *H rH
•H (0
rH a rl
O 4) Q) C
u o a cd
N~> CU
B
ca
a B
0 M
•H O
4-J m
CJ ^3 -H
Cd 4J H
rl -H O
fe > U
14-1
o
rrH
^
M 0)
0) W rH
,0. CU Pu
S *J B
3 tH cd
25 co co
a
0
•H
4J
u
(U
CO
o o o
O rH O O O
vo • r^ vo m
rH Si- CO CO rH
00 1^. vO vO 00
oo o r^ r^ sj-
rH m r>»
O O CO
• • 1
rH O rH
ai vo sr
O O rH
• • •
o o o o o
CT* CO vO
m oo co
o o o
• • •
o o o o o
00 st- O CO CTi
VO CM rH 00
rH
cd
1-1
4-1
a
a>
U H
I cd
.C J3 M
4J 4J 4-1 4J £3
03 M VJ d r-l
Z S5 0 U
O vo
cs r>.
. o •
O rH O
o r^
CM O st
O rH O
m
SJ-
1 • 1
rH
vO
CO
•
o o o
rH
4H
> >.
0 rl
H T3
•W rH (3 ,£5
CO Ct) 3 4-1
cd PC< o o
W PL, H
CM
rH
»
o
CO
CM
o
1
o
o
o
rH
CD
r-l
I-l
•H
>
co
0)
,£ 4-1
4J Cfl
3 0
o u
CO
135
-------
5*5
o
H
W
1
s
i
PH
S-j
E_l
H
£3
&
vO H
^3
s *
M J
H CJ
O
O
H
w
H
CJ
e
s
w
E— *
w
CO
fe
O
co
H
CO
s!
3
^
4J M
u cd
Jj kT
P /—N ^
•0 0 rH
M 0 0
CO
TJ 4) M
C 4-1 0)
cd cd ft a
4J rH v-' Cd
CO PL) Cl)
0
"N rl
rH Cd
B ,1 >
M >~> C5
O 4J O
14H iH H
•H CO
rH a r<
0 (U (1) d
CO P ft Cd
v"> i|t
1
0 §
o E
•H 0
•P ,13 <4H
o 4-1 in
«- 1 1
•H rH
M 3 0
Cn O
14-1 CO
o a
O T3
rl «H 0)
Q) 4J rH
0 O 0
^ o cd
23 >4 co
00
a
•H TJ
rH O
ft-H
Cd 1)
CO P-!
ti
o
iH
4J
CJ
0)
CO
in o CN co o
•«* o o co m vo
• 00 CN ...
rH CO i-H vO •* -sf
CO CO CO
H CN O 00 CN OO
vO O CN CJ\ O\ CN
O O vo rH CO ^3-
CN
vo in o o\ -ft
in O\ in rH rH
CN O CN CN O
rH
>* vO CN O CN
r^ CN a» in rH
O O O O O O
o -tf
O - o co
rH "d" *d°
M M
H
4-1
CO
0)
3
,c
4-1
Jj
0
CO
0 0
o o o
r^ a. rH
rH in O-
co -d-
r~ o m
-d- vo oo
CO rH
vO 00
CM O
1 . •
1 " *
rH VO
vo
A
in rH
O O CN
A
VO VO
rH -d"
CO CM
odd
CN OO CO
CM co in
M M
M
irt.
4-1
M
0
S5
/-N/-V/-N
cq co co
t^j i^ t^j
a) (U u
$ $ $
CN CN CN
^X NtaX N,^
O\
P^ rH
-.>>
^ g g
II II II
M H H
H H
H
CO
0
•H
rt
0)
PH
00
a
•H
rH
ft
B
cd
CO
136
-------
VO
•
vO
W
rJ
H
25
0
H
H
0
W
w
PQ
M
H
s
3
CH
U
M-l CO
0 (3
O ^3
M -H (U
CD 4J rH
ja cd a
d o cd
a (J CO
bd
1-1 T3
rH O
§*Tl
Co CU
CO PL|
a
0
4J
y
CO
vO vO vO
oo oo oo oo
VOO OO OrH i-HrH
CT\rH rHCO r~»A AA
A i-H rH rH
••3-O vOrH vtO OO
coco -A COCO CNA AA
CN rH i-H
HCM mcTi VOO COO
*
•A «A • rH 'CO
CO O O\ A rH A
co ^f r^
A A
in oo -3- oo o -j- in
ON r^ rH r^ co o^ co ov
ooo ocTt CNCO cor^
A A A A A
OrH CMOO -~*
CO CO
t*^ t^
cu •>
£3 rH
3 3
>-> ^
II II
H H
M
13
O
•H
rl
(1)
AH
00
c
•H
rH
a
e
cd
CO
137
-------
M
H
U
W
to
M
to
W
H
§
p<
^
p j
SH
H
M
H
Co
o-
r- &
W
vO H
H £3
PQ |_J
^ ^
EH CJ
M
C5
O
1-4
o
M
H
E-i
U
^
PQ
H
P
^
W
H
to
to
O
w
M
to
i
^3
*
fl flJ
3 >
*X3 O ^^^
rl U rH
CO 0
-O cj
(3 4J H
CO cd CU C
4J H P, Cd
W fc^ g
.
^-* s p
I-H co
g J >
o -u o
14-1 -H rH
•rl CO
rH C rl
O 0) 0) C
o p & cd
B
co
§ fi
•rl 0
4_t IM
CJ ^3 -H
Cd 4J rH
rl -H 0
fe ^ U
4-4 CO
0 C
O T3
M -H 4J 4J 4J
CD M 3 B CO It 3
CD o o co cd o o
|2 2; w U W 3 c/)
138
-------
a
0
H
u
w
CO
«
%
W
w
2
5
Pn
^|
|H
H
3
^*
o-
Pi
00 W
H
vo <3
w
pp ^4
« U
E-i H
|_J
O
M
Pi
H
0
2
^J
m
*y1
pq
H
CO
CO
p4
o
CO
CO
£j
1
^
M
4J eg
d >
*O O S**\
co a
T3 CU
d 4J M
cd cd cu d
4-> rH P_ CO
CO P-i ^^ w
— ^_ «J
HCd
O 4J O
14-1 fH i-H
•H CO
rH d M
O CU CU d
O Q D, Cd
X-X fll
s
m
C3 S
O H
•rl O
4-1 «4-l
CO 4J H
1-1 -H O
PH tj C-3
M-l CO
o d
o *o
^1 iH OJ
CU 4-1 rH
S5 >J CO
00
d
rH O
CX-rl
cd o
CO PH
§
»_J
^^
4-1
U
<0
CO
O O O st CN
O VO vo
H H H
co co
CO rH CO
cu cd cu
4-1 4J 4-1
3 d H
O 0) O
co c_> a
£5
0
M
U
w
CO
m
CO
Pi
w
H
2
^
PH
KH
fH
H
S3
o-
Pi
ON W
• H
VO <1
w
j 1 t 1
PQ ^
^ O
H S
8
H?
o o
00 M
ON Pi
i f_-i
r-1 VH
H
i-H U
•H d
cu <
60
g
cfl
^f
4-1 Cd
d >
a
"d o x-x
rl CJ rH
cd S
13 0)
d 4J M
Cd CO CU d
4-1 rH P. Cd
ty2 p^ ^^ CJ
^^_ t.
^^ M
rH Cd
O 4J O
14-1 vH rH
•H CO
rH d M
o cu cu d
U O PH Cd
s
co
O M
•H O
4-> 4-4
Or* i^
ri-4 rl
CO 4-1 rH
>-i -H O
Pn |j cj
<4-t CO
o d
O ^O
H -H OJ
0) 4J i-H
^3 cfl p.
3 O CO
a t— 1 co
oo
d
•H T3
rH O
cd cu
CO p..
§
fcj
*n
4-1
0
cu
CO
0 0 -*
o o o
CO VO •
rH CM 4-1
co to t-i
cd cu o
w 13 2;
139
-------
inspection of tables 6.4 through 6.9 shows that in most cases the variances
are much greater than the mean. The only cases in which the variances are
low are when the samples were collected in the winter or when the number of
samples collected in a section was relatively few. The high values for the
variances indicate that seeking significant differences by analysis of vari-
ance on the untransformed data would be futile. A rational for using a data
transformation to stabilize the variances could be developed. However, since
most of the counts were zeroes the data would have to be coded and then
transformed. This could introduce artifacts into the analysis of variance
and any significant differences found would be questionable. Therefore, it
was decided not to use this approach in searching for significant differences
among sections.
If there is a significant difference in the proportion of the sites
where coliforms are found among several sections, it can reasonably be con-
cluded that there is a different level of microbiological contamination among
the sections. Significant differences in proportions can be sought using a
)(2 contingency table analysis. Such analysis can be applied in a straight-
forward manner for most of the systems and most of the sampling periods.
However, one of the requirements for the analysis is that the expected number
of positive sites for each section, which is obtained by multiplying the
overall fraction positive by the number of sites sampled in the section, must
be greater than 1. This requirement is violated for system CV where the
overall fraction positive was 0.044 which means that 23 sites would have to
be sampled to give an expected number of positive sites of 1.02, for the
first sampling period for system WH where the overall fraction positive was
0.091 and 11 sites would have to be sampled to give an expected number of
positive sites of 1.001, and for system DT where the overall fraction posi-
tive was 0.052 and 20 sites would have to be sampled to give an expected
number of positive sites of 1.04. The only way to meet this requirement for
the )(2 test is to group some sections together. When sections are grouped,
it is necessary to group contiguous sections.
The results of the y^- contingency table analyses are presented in
table 6.10. For system CV the North and North Central sections were grouped
together, as were the Central and Cain sections, and the last four sections.
For the first sampling period for system WH, the East, East Central, and
West Central sections were grouped together. For system DT, the Northwest
and Southwest sections were grouped together as were the North, East, and
Southeast sections. The null hypothesis (Ho) for these analyses is that
there are no differences among the sections of each system for each sampling
period in the fraction of the sites where coliforms were found. The only
instances where the null hypothesis can be rejected at the 5% level are the
third sampling period for system WH and the first sampling period for system
LB.
It is instructive to pursue the differences detected further. The LB
system is one large interconnected grid which is supplied water from two
different sources. The separation into isolated sections is based on the
location of the treatment plants. The North Section is north of the Brant
Beach Plant and the South Section is south of the Terrace Plant. The separa-
tion between the North Central and South Central Sections was made halfway
140
-------
TABLE 6.10
CONTINGENCY TABLE ANALYSES FOR
DIFFERENCES AMONG SECTIONS
System Sampling
Period
CV
WH I
II
III
LB I
II
DT
BL I
II
SR
a = significant at 5%
Degrees of
Freedom
3
2
4
4
3
3
3
2
2
2
level.
x2
0.91
3.84
4.90
9.83
9.22
2.62
1.47
0.92
1.05
0.79
Probability of
Null Hypothesis P(HO)
0.99 < P0 < 0.95
0.10 < P0 < 0.25
0.25 < P0 < 0.5
0.025 < PQ < 0.05a
0.025 < PQ < 0.05a
0.25 < P0 < 0.5
0.50 < P0 < 0.75
0.95 < P0 < 0.975
0.50 < P0 < 0.75
0.95 < P0 < 0.975
between two plants.
2
The X contingency table analysis only determines that significant
differences among the fractions of locations exist. It does not identify
where they occur. Significant differences between two sections can be iden-
tified using the Z statistic. The Z statistic is the number of standard
deviations between two fractions when the normal distribution is used as an
approximation of the binomial distribution. The Z values for two section
comparisons for the samples collected from system LB during June 1979 are
given in table 6.11. It can be seen that there are no significant differ-
ences between the North and North Central sections or between the South and
South Central section. The only significant differences occur between the
sections in the Brant Beach service area and sections in the Terrace Plant
service area and, of course, between the two service areas.
141
-------
TABLE 6.11
ANALYSIS OF SYSTEM LB DATA FOR JUNE 1979
Section
North
North
Central
South
Central
South
Brant
Beach
Service
Area
Terrace
Plant
Service
Area
Date
6/7
6/12
6/14
6/19
6/21
overall
6/7
6/12
6/14
6/19
6/21
overall
6/7
6/12
6/14
6/19
6/21
overall
6/7
6/12
6/14
6/19
6/21
overall
6/7
6/12
6/14
6/19
6/21
overall
6/7
6/12
6/14
6/19
6/21
overall
Number of
Locations
Sampled
1
10
13
10
9
43
0
7
11
11
20
49
2
16
12
14
18
62
0
8
21
16
16
61
1
17
24
21
29
92
2
24
33
30
34
123
Fraction
with
Coliforms
0
0
0.308
0
0.111
0.116
_
0
0.2
0.182
0.050
0.122
0
0
0.538
0.500
0.167
0.274
_
0
0.429
0.500
0.188
0.328
0
0
0.292
0.095
0.069
0.119
0
0
0.485
0.500
0.177
0.301
Z Statistic
North North
Central
0.19
1.42
0.60
0.09
0.19
1.42
0.60
0.09
1.39 1.50
2.66a 1.64
0.36 1.17
1.96a 1.96a
0.71 0.86
2.68a 1.86*>
0.50 1.30
2.49a 2.52a
1.47
3.02a
1.41
7.34a
South
Central
1.39
2.66a
0.36
1.96
1.50
1.64
1.17
1.96a
0.86
0
0.16
0.65
1
3
1
7
South
0.71
2.68a
0.50
2.49a
0.86
1.68b
1.30
2.52a
0.86
0
0.16
0.65
.47
.02a
.41
,34a
a = significant at 5% level, b
significant at 10% level
142
-------
The significant differences are primarily the result of a large fraction
of positive samples collected from the Terrace Plant service area on 6/14/79.
On that day, the high lift pumps at the Terrace Plant were out of service for
three hours during the period when the samples were being collected. The
pressure in the Terrace Plant service area dropped and water from the Brant
Beach Plant flowed into the area causing a reversal of flow in the two large
mains connecting the service areas. It can be postulated that the hydraulic
disturbance in the Terrace Plant service area caused a resuspension of sedi-
ments in the mains or a shearing of bacterial growth from the walls of the
mains. This hydraulic disturbance then would be responsible for the large
number of positive samples collected in the South and South Central Sections
on 6/14/79.
The comparisons between pairs of sections for the third sampling period
for system WH are presented in table 6.12. The significant differences are
between North and West Central Sections and between the Southwest and West
Central Sections and are primarily the result of samples collected on 5/28/81.
The samples collected on 6/4/82 were collected mostly in the Southwest and
North Sections as a measure of the occurrence of coliform bacteria in the
sections of the system most distant from the well. Some fire hydrants were
flushed on that day to increase the flow in those distant sections. Although
no samples were from the West Central section on that last sampling day, the
large numbers of samples collected and the relatively high fractions of the
locations with coliforms in the Southwest and North Sections contributed to
the significant differences found.
It is clear that there can be significant differences in the occurrence
of coliforms among the isolated sections of a distribution system or between
two sections. These differences can occur on a single day and then not occur
on another day of the same week. The possible occurrence of such differences
should be taken into consideration in the planning of a monitoring program.
6.3.2 Differences Within Sections
The value of the concept of an isolated section of a water distribution
system is that it forces recognition that there may be places where coliforms
may be present and not be detected unless samples are collected in the imme-
diate vicinity. One problem with applying the concept to some water distri-
bution systems is that there may not be clear hydraulic separations into
isolated sections. If a distribution system is one large interconnected
grid how can it be divided into sections for sampling? One way of approaching
this problem is to look for significant differences between adjacent sampling
locations or between sampling locations two blocks apart, three blocks apart,
etc.
The smallest system sampled in terms of the number of locations was the
MW system which had two. The two streets were Sherwood Drive and Hall Road.
The two water mains were connected together at each end and the transmission
line from the well fed into the Hall Road Main. Thus, the entire distribu-
tion system was a single loop.
The coliform results from the 100ml samples collected from MW are given
143
-------
TABLE 6.12
ANALYSIS OF SYSTEM WH DATA FOR MAY-JUNE 1981
Section
Date
Number of Fraction
Z Statistic
Locations with East
Sampled Coliforms
East
East
Central
West
Central
South-
west
North
5/22
5/27
5/28
6/2
6/4
Overall
5/22
5/27
5/28
6/2
6/4
Overall
5/22
5/27
5/28
6/2
6/4
Overall
5/22
5/27
5/28
6/2
6/4
Overall
5/22
5/27
5/28
6/2
6/4
Overall
0
8
6
9
0
23
0
5
6
5
1
17
0
13
12
12
0
37
1
11
6
12
13
43
1
11
6
8
27
53
— m
0
0.167
0.111
-
.087
_
0
0
0.4
0
0.118
_
0
0
0.083
-
0.027
0
0.091
0.333
0.167
0.308
0.209
0
0.091
0.167
0.375
0.296
0.245
0
1.04
1.26
0.32
0
0.55
0.22
1.04
0.88
0.67
0.36
1.27
0.88
0
1.28
1.59
East
Central
0
1.04
1.26
0.32
0
0
1.56
1.35
0.70
1.54
1.13
0.83
0.70
1.04
0.09
1.12
West
Central
0
0.55
0.22
1.04
0
0
1.56
1.35
1.11
2.12a
0.62
2.46a
1.11
0.55
1.60
2.81a
South- North
west
0.88
0.67
0.36
1.27
0.7
1.54
1.13
0.83
1.11
2.121
0.62
2.46a
0
0.67
1.05
0.07
0.63
0.88
0
1.28
1.59
0.7
1.04
0.09
1.12
1.11
0.55
1.60
2.81a
0
0.67
1.05
0.07
0.63
a = significant at the 5% level
144
-------
in table 6.13 divided according to location and by sampling date. The Z sta-
tistic for comparison of the fractions of samples with coliforms between the
two sampling locations on a day by day basis are given in the last line of
the table. It is clear that there are no significant differences between
the results obtained from the two locations. The tentative conclusion from
this is that there was no detectable difference in the level of contamination
between the two locations. It should be noted that a great many samples were
collected from each location and the level of contamination was relatively
low.
The sampling results from system BL divided according to the locations
and the sampling periods are presented in table 6.14. Some of the sampling
location numbers are missing from the table because it was not possible to
collect samples from those locations. Most of the locations which could be
sampled were sampled each of the sampling periods and the number of sites
sampled per location varied from 1 to 8. Samples with coliforms present
were found here and there but no consistent differences among the locations
are evident. Adjacent locations sometimes yielded similar results and some-
times different results and the microbiological contamination of the system
could not be localized. The fraction of sites with coliform present is very
small for all sections and for system BL there is no good way to group con-
tiguous locations so that a ^ test for significant differences among loca-
tions can be performed. From these data it is impossible to determine if
there either are or are not differences among the locations within each sec-
tion. It appears that the microbiological contamination is general through-
out the system but at a low level during both sampling periods. It takes a
large number of samples to detect the coliforms but the locations where they
are found appear to be distributed without any detectable pattern.
The sampling results for system SR divided according to the 18 loca-
tions are given in table 6.15. In this system the fraction positive was
0.125, which means that only 8 sites sampled are needed to give an expection
of 1 positive site and it is easier to group contiguous locations for the ^
this system. The locations were grouped as is shown in the table and the
results of the x test are given at the bottom of the table. No significant
differences among the locations were found. This analysis is equivalent to
dividing the East section into four subsections and the North section into
two subsections and it is not at all surprising that the results in table
6.15 are not different from the results of the analysis by section for system
SR as shown in table 6.10.
The net outcome of all this is that there is no basis, in these data,
for limiting the size of an isolated section. Samples collected at one
location are not necessarily similar to those collected at nearby locations
nor necessarily different from those collected at locations some distance
away.
6.4 HYPOTHESES ABOUT CONTAMINATION
The mechanistic approach to modeling coliform dispersion in a water
distribution system is an attempt to find a method of localizing contamina-
tion in the system. If this approach were to prove to be successful, it
145
-------
CO
rH
•
vO
w
pj
9
H
1
s
W
H
CO
F^
CO
fa
O
CO
O
H
0
O
rJ
O
£_j
[r~]
prj
H
a
0
erf
fa
/«-\
CO
w
hJ
PM
^
co
,_(
£j
O
o
^
co
H
,J
O
co
3
o
£3
H
&
^
CO
o
3
O
CO
23
s
o
u
_,
!-H
cfl
F-I
cu
o
00
F— 4
f-H
^^
CM
rH
O
oo
o\
-^
F— 1
F—t
CU
4-1
cfl o
O oo
rO CO
CU rH
4J \
Cfl rH
O rH
T)
a o
M 00
•* — _
C rH
O rH
^^
CO rH
4-1 rH
rH
3
co o
cu oo
vO
00^.
C3 -i
•H rH
ft
6 0
cfl 00
CO ^
o
S CO
E -^
0 0
K-l rH
rH
O 0
o oo
CN
"--^,
o
1 — 1
0
oo
00
o
1 — 1
c
0
•H
4-1
cfl
CJ
o
,-J
CO O
CO CM
CN
o o
1 — 1
•H -H
4-1 4J
0 -H
Cfl CO
i-i O
fa PM
m
m
•
o
CM
O
o
o
vO
m
•
r^
O
o
•
o
m
CM
O
m
CM
•
O
o
/-N
f-t
a
o
4-1 rH
•H
C CO F-I
cfl c! CU
CU CU ft
S 0 ^
OO rH
*^ r^*
O 0
oo >>O
pi CU F-I CU -H -H 4-1 r-H
CO rH CU rH 4-1 4-1 -H
rH CUftrOftO-HCCOlH
rH 4-lS6SCflCOCflCCU
cfl *Hcd3cdF-ioa)(uo4
33 wcopacofapHSO^
CM
F-H
•
1— 1
C7N
i — 1
•
O
O
O^
CM
•
O
i-H
CO
•
i— i
^D
ON
•
o
. — 1
vO
rH
0
CU
•H
4-1
M-l -H
O CO
H O
o co PM
MH C
O CO
CO CO (3
CU i-l O
3 F-I -H
rH Cfl 4J
CO ft O
F> S Cfl
O M
N CJ fa
146
-------
TABLE 6.14
COMPARISON OF SAMPLING RESULTS (100ml SAMPLES)
FROM 68 SAMPLING LOCATIONS OF SYSTEM BL
March- April 1980
Location
Number
102
103
104
105
106
107
108
108
109
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
Total for
Southeast
Section
132
133
134
135
136
137
138
139
140
141
Sites
Sampled
8
3
3
6
8
3
3
3
3
3
3
2
4
0
2
2
2
1
4
3
4
3
3
3
1
3
1
81
2
5
1
2
2
2
2
3
4
3
Fraction Mean
Positive Density
(per 100ml)
0.125 0.06
0
0
0
0
0
0
0
0.333 >26.67
0.333 0.17
0
0
0
-
0
0
0
0
0
0
0
0
0
0
0
0.333 0.17
0
0.049 >1.01
0
0
0
0
0
0
0
0
0
0
June 1981
Sites
Sampled
1
1
4
2
3
4
4
4
1
2
2
3
3
1
1
1
2
2
2
0
3
3
0
1
0
3
1
50
3
2
3
0
1
0
1
1
1
3
Fraction
Positive
0
0
0
0
0.333
0
0
0
0
0
0
0
0
0
0
0
0.500
0
0
-
0
0.333
-
0
-
0.333
0
0.080
0.333
0
0
-
0
-
0
0
0
0
Mean
Density
(per 100ml)
_
-
-
-
0.67
-
-
-
-
-
-
-
-
-
-
-
0.75
-
-
-
-
0.67
-
-
-
0.33
-
0.11
0.17
-
-
-
-
-
-
-
-
-
147
-------
TABLE 6.14 (continued)
March-April 1980
Location
Number
142
143
144
145
146
147
148
149
150
151
151
152
153
154
155
156
Total for
Central
Section
158
159
160
161
162
163
164
165
167
168
169
170
171
172
173
174
Total for
Northeast
Section
Sites
Sampled
4
3
3
4
3
3
5
3
6
4
4
2
2
5
4
7
84
1
1
5
3
3
3
5
3
7
4
4
8
6
6
5
6
70
Fraction
Positive
0
0
0
0
0.333
0
0
0
0
0
0
0
0.500
0
0
0
0.024
0
0
0
0
0
0
0
0
0
0.250
0
0
0
0.167
0
0
0.029
Mean
Density
(per 100ml)
_
-
-
-
0.17
-
-
-
-
-
-
-
0.25
-
-
-
0.01
—
-
-
-
-
-
-
-
-
0.12
-
-
-
0.08
-
-
0.01
Sites
Sampled
0
2
4
5
2
3
5
0
2
2
2
5
1
5
2
3
56
4
3
4
4
4
3
3
7
1
2
6
7
6
5
3
2
64
June 1981
Fraction
Positive
—
0
0
0.200
0.500
0.333
0
-
0
0
0
0.200
0
0.400
0
0.333
0.142
0
0
0.500
0
0
0.333
0.333
0
0
0.500
0
0.142
0
0
0.333
0
0.109
Mean
Density
(per 100ml)
_
-
-
0.60
0.25
0.33
-
-
-
-
-
0.40
-
0.40
-
0.33
0.18
_
-
15.62
-
—
0.17
0.17
-
-
0.25
-
0.07
-
-
1.00
-
1.06
148
-------
TABLE 6.15
COMPARISON OF SAMPLING RESULTS (100ml SAMPLES)
FROM 18 SAMPLING LOCATIONS OF SYSTEM SR
Location Sites Fraction
Number Sampled Positive
105
0.111
No. of Positive Sites (Grouped Locations)
Expected Observed
101
102
103
104
4
5
5
3
0~
o_
0~
0
1.125
1
0
0
1.125
106
107
108
109
110
111
7
15
24
1
6
0.143
0.133_
0.167~
0
0
2.75 3
3.875 4
Total for
East 79
Section
0.101
112
113
4
40
0.500
0.100
5.5
Total for
West 44
Section
0.136
114
115
116
117
118
Total for
North
Section
3
3
3
22
22
53
0
0
0_
0.181
0.181
0.151
1
2
2
.12
.75
.75
0
4
4
=5.11 for 8 df
.50
-------
would be possible to work out a sampling plan which would be superior to ran-
dom selection of sampling sites. That is, a sampling site could be selected
which would represent a main or a part of an isolated section and other
samples from that main or part would not be required. Achieving such a goal
would greatly increase the efficiency of sampling.
The hypotheses discussed in this section could be used to formulate sam-
pling plans which would be very efficient for detecting contamination and
for localizing such contamination. Unfortunately, none of the hypotheses
were verified by the data collected during this project and thus it is not
possible to construct a sampling model using the mechanistic approach. Since
all of the systems sampled during this project were small, it is possible
that some of the hypotheses might be true for larger systems.
All of the hypotheses were tested using data from all nine of the sys-
tems sampled. However, since none of these hypotheses were found to be
valid for any of the systems, it seems unnecessary to present all of the
analyses which were made. Data from the sampling of system BL during June
1981 are used as examples of the results obtained. Some data from other
systems are also used to illustrate particular points.
In order to test these hypotheses, it was necessary to construct flow
diagrams for the sampling locations of each system. The flow diagram for the
isolated sections of system BL is presented in figure 6.3. In constructing
these diagrams it was sometimes necessary to make assumptions about the di-
rection of flow in a particular street lateral. The assumptions were based
on the concepts that flow is normally from the street laterals closer to the
treatment plant to the ones further away and from the larger pipes to the
smaller pipes. Using these concepts, the shortest distance of flow from the
treatment plant to any street lateral was selected.
6.4.1 Upstream and Downstream Samples
The first of these hypotheses tested was that samples collected down-
stream from a site from which a positive sample had been obtained would also
have coliforms present. The companion hypothesis is that samples collected
upstream from a site where a negative sample would also be negative. These
hypotheses should be tested using data collected on a single day because
sources of contamination may not presist for more than a short time. It is
clear that if one of these hypotheses were true, the other one would also be
true.
The locations of the samples with coliforms collected from system BL in
June 1981 are listed in table 6.16 along with all downstream locations sam-
pled. The downstream locations are divided into those where coliforms were
found and those where coliforms were not found. From this table it is clear
that negative samples are frequently collected downstream from positive
samples and, conversely, that positive samples are often collected upstream
from negative samples. The same conclusion was reached by examination of
similar data from other systems. Thus, the two hypotheses have to be re-
jected, at least in the general case.
150
-------
SOUTHEAST SECTION
treatment
plant
122
-t-
117 «-
114 «-
126 «-
4-
127
115 «-
4-
116
113 •*•
109 «-
4-
110 -»•
4-
13 L
4-
101
4-
102
4-
103
4
107
4-
108
111
•*• 126
•*• 104
-> 112
D
123 *r 120 -«- 118
4-
124p ^ 121 ^ 119 116 103 -> 104 -^- 105T
125 -*-*• 129
130 •*•*- 128 •«-
CENTRAL SECTION
138
NORTHEAST
134 -»• 157 -*• 158 -w- 159 •*-*• 160 ++ 161 -e-^ 162 ++ 163^-*- 164
1^4 175 Stfd-
D pipe
+ 4-
165
166
Figure 6.3 Flow Diagrams for the Isolated Sections of System BL.
151
-------
TABLE 6.16
6/23/81
6/25/81
LOCATIONS OF POSITIVE COLIFORM SAMPLES COLLECTED FROM
SYSTEM BL - JUNE 1981
Date
6/16/81
6/18/81
Location of
Positive
Sample
122
160
163
164
152B
154
118
127C
145
146
154
156A
168
170
Locations of
Downstream
Positive
Samples
none
163,164
164
none
none
none
none
none
154,156A
154,156A
156A
none
none
none
Locations of
Downstream
Negative
Samples
none
161, 162, 170A, 1708,171, 172, 174
156
156
120A,120B,123
none
148,152,153,155,1568
155,1568
155,1568
1568
169A,169B
none
106
132
147
173
none
147
none
none
none
133,134,136,143,144,145,148,151,
152,155
148
none
A modification of these two hypotheses would be that positive samples
occur more frequently downstream of other positive samples. However, there
were no instances in which enough samples were collected on a single day to
provide a reasonable basis for testing that hypothesis.
6.4.2 Peripheral Locations
Another hypothesis tested is that coliform bacteria are found more fre-
quently in samples from peripheral locations than in samples from other
locations. Table 6.17 gives comparisons of the fractions of peripheral and
other (nonperipheral) locations positive for coliforms for the various sys-
tems and sampling periods. System MI was omitted because there were too few
positive samples for a reasonable test and system MW was omitted because the
entire system consists of two locations connected to for a loop. The Z sta-
152
-------
tistic is based on the use of the normal distribution to approximate the
binomial distribution and is a measure of the number of standard deviations
between the two fractions positive. A Z value of 1.96 would indicate a
difference significant at the 5% level. No significant differences, were
found. Actually, the fraction positive for the peripheral locations was
greater than the fraction positive for the other locations in only 7 of the
11 cases.
When the comparisons were made for the isolated sections and for indi-
vidual sampling days, there were a few significant differences between pe-
ripheral and other locations. For instance, in system SR all the positive
samples collected in the North Section were from peripheral locations and
all the positive samples collected in the East Section were from non-periph-
eral locations. However, when the comparisons were made for the entire
system the results from the individual sections balanced out. There is
clearly no basis for concluding that peripheral locations are more likely to
yield coliforms when the systems are considered as a whole.
TABLE 6.17
PERIPHERAL VERSUS OTHER LOCATIONS
System-
Sampling
Period
CV
WH-I
WH-II
WH-III
LB-I
LB-II
DT
BL-I
BL-II
SR
BG
Peripheral
No. of
Locations
Sampled
70
28
54
55
87
92
57
73
54
71
24
Locations
Fraction
Positive
0.029
0.107
0.463
0.127
0.195
0.369
0.053
0.055
0.167
0.155
0.250
Other Locations
No. of
Locations
Sampled
145
38
100
118
128
140
117
162
116
105
29
Fraction
Positive
0.055
0.079
0.360
0.169
0.250
0.407
0.051
0.025
0.086
0.105
0.172
Z
Statistic
0.85
0.39
0.76
0.65
1.38
0.58
0.06
1.17
0.86
1.05
0.70
6.4.3 Distance from Treatment Plant
Another hypothesis of the proposed mechanistic approach is that samples
with coliform are more likely to be found far from the treatment plant
(source of water to the distribution system) than close to the treatment
plant. The concept behind this hypothesis is that the microbiological
contamination is introduced into the water after it leaves the plant and the
further the water has traveled, the greater the chance it has become contami-
nated.
153
-------
TABLE 6.18
ANALYSIS OF COLIFORM OCCURRENCE AS A FUNCTION
OF DISTANCE FROM THE TREATMENT PLANT
SYSTEM BL - JUNE 1981 DATA
Distance
from
Plant
(blocks)
1
2
3
4
5
6
7
8
9
10
Potential
Sampling
Locations
2
4
8
11
13
11
11
8
4
2
No. of
Sites
Sampled
3
3
12
27
37
26
29
16
11
6
Fraction
Positive
0.333
0
0.083
0.074
0.081
0.115
0.172
0
0.091
0.333
The coliform data from the 1981 sampling of system BL are presented in
table 6.18. It is clear that the fraction of the sites which yielded samples
with coliforms does not increase with distance from the treatment plant.
Indeed, in some instances, locations nearer the treatment plant yielded more
positive samples and more coliform bacteria than more distant locations.
However, system BL has an elevated storage tank which is eleven blocks from
the treatment plant which could be considered as a "source" of water to the
distribution system and which could affect the microbiological water quality
in the northeast and southeast sections of the distribution system. Thus,
system BL may not provide a good test of this hypothesis.
Systems WH and LB both have finished water storage in standpipes close
to the treatment plants. Thus, they should provide adequate tests of the
hypothesis. Data from these two systems are presented in tables 6.19 and 6.20.
Again it is clear that the frequency of occurrence of samples with coliforms
does not increase with distance from the water source. The same type of
analysis was made of the occurance of coliforms in the other systems and no
evidence supporting the hypothesis was found.
154
-------
TABLE 6.19
ANALYSIS OF COLIFORM OCCURRENCE AS A FUNCTION
OF DISTANCE FROM THE TREATMENT PLANT
SYSTEM WH MAY-JUNE 1981
Distance
from
Plant
(blocks)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
Potential
Sampling
Locations
8
11
22
11
11
2
5
7
6
6
8
8
7
2
2
No. of
Sites
Sampled
7
16
25
18
11
2
8
10
12
10
16
18
10
4
6
Fraction
Positive
0
0.125
0.040
0.111
0.182
0.500
0
0.200
0.167
0.300
0.375
0.167
0.200
0.500
0.167
TABLE 6.20
ANALYSIS OF COLIFORM OCCURRENCE AS A FUNCTION
OF DISTANCE FROM THE TREATMENT PLANT
SYSTEM LB
Distance
from
Plant
(blocks)
Potential
Sampling
Locations
No. of
Sites
Sampled
Fraction
Positive
0-5
6-10
11 - 15
16 to 20
21 - 25
26 - 30
>30
120
99
88
108
106
75
48
98
66
43
58
67
79
36
0.357
0.378
0.326
0.345
0.373
0.304
0.278
155
-------
6.5 SUMMARY AND IMPLICATIONS
The concept of dividing a water distribution system into hydraulically
isolated sections for monitoring purposes is important. Coliform bacteria
in a water distribution system travel with the flow of water not against it.
Thus, if they are introduced into a system at only one point, they would be
found downstream of that point and not upstream of it. Clearly, there are
places in a water distribution system where coliforms can be present and
they will not be detected by samples taken from other parts of the system.
The concept of classifying isolated sections of a water distribution
system according to their geometry is also important. Flow reversals are
possible and probably occur in the pipes which make up grid or loop sections
depending upon the location of the water demand; thus, it is not possible to
determine upstream and downstream directions in such sections in any absolute
sense. On the other hand, flow reversals cannot occur in the pipes which
make up linear or dendritic sections unless there is more than one source of
flow into the section. Such factors as direction of flow and reversal of
flow should be considered in the selection of "representative" sampling
locations.
In testing for differences among sections, the small fraction of sites
with coliforms present coupled with a relatively low number of samples spread
over several sections was a problem for the analysis of data from systems CV
and DT and from the first sampling period for system WH. This problem was
circumvented by grouping together data from contiguous sections. This group-
ing of data was necessary but it somewhat weakens the purpose of the test.
Significant differences among the sections were found for the data sets
for the third sampling period for system WH and for the first sampling period
for system CV. This shows that such differences can exist at times and thus
that monitoring results may be biased if all sections of a distribution
system are not included in the monitoring program.
No significant differences were found among the locations within the
sections of a system in the fraction of the sites with coliforms present.
Actually, the ^ contingency table test could not be used for testing for
differences among locations for most systems because there were few samples
per location and a relatively low fraction of samples positive for coliforms.
The contingency table test was used for system SR by grouping results of
some locations. The conclusion from these results is that there is no basis
in any of the data collected from these systems for subdividing hydraulically
isolated sections.
If coliform contamination in a water distribution system was localized
in one part of the system or in one particular type of location, the mecha-
nistic approach could be of value in determining where to sample to find the
coliforms. In this section, we have presented the results of testing the
hypotheses that if coliforms are found in a particular location they should
be found in all downstream locations, that coliforms are more likely to
occur in peripheral than in non-peripheral locations, and that the frequency
of occurrence of coliforms increases with distance of flow from the water
156
-------
source into the system. No evidence to support any of these hypotheses was
found. The conclusion from this is that in planning a monitoring program
all parts of the system and all types of locations should be included. The
best method of selecting locations for sampling is a randomization procedure
which gives each potential sampling location the same probability of being
collected.
In the water works industry it is commonly assumed that if coliforms
are not found in the finished water leaving the treatment plant and are not
found in locations the most distance from the treatment plant, they will not
occur at intermediate locations. Our results demstrate that this is not
true, at least, for the systems we sampled. The implications of this finding
is that all parts of the distribution system have to be included in the
monitoring program in order to have the samples collected at points which
are representative of the conditions with the distribution system.
157
-------
REFERENCES
Aitchison, J. and J. A. C. Brown. 1957. The Lognormal Distribution.
Cambridge University Press, Cambridge.
APHA. 1976. Standard Methods for the Examination of Water and Wastewater
14th Ed., American Public Health Association, Washington, D.C.
Bordner, R., J. Winter, and P. Scarpino. 1978. Microbiological Methods for
Monitoring the Environment - Water and Wastes, EPA-600/8-78-017,
Environmental Monitoring and Support Laboratory, U.S. Environmental
Protection Agency, Cincinnati, Ohio.
Brodsky, M. H., B. W. Ciebin, and D. A. Schiemann. 1978. Simple bacterial
preservation medium and its application to proficiency testing in water
bacteriology. Appl. Environ. Microbiol., 35:487-491.
El-Shaarawi, A. H., S. R. Esterby, and B. J. Dutka. 1981. Bacterial density
in water determined by Poisson or negative binomial distributions.
Appl. Environ. Microbiol., 41:107-116.
Ewing, W. H. and M. A. Fife. 1972. Enterobacter agglomerans (Beijerinck)
Comb. Nov. (the Herbicola-Lathyri Bacteria). Int. Jour. Systm. Bact.,
22:4-11.
Geldreich, E. E.,1971. Application of bacterial data in potable water
surveillance. Jour. Am. Water Works Assoc. 63:4:225-229.
Geldreich, E. E., 1975. Handbook for Evaluating Water Bacteriological Lab-
oratories, 2nd Ed., EPA-670/9-75-006, Municipal Environmental Research
Laboratory, U.S. Environmental Protection Agency, Cincinnati, Ohio.
Geldreich, E. E., M. J. Allen, and R. H. Taylor. 1978. Interferences to
coliform detection, pp. 13-20. In. Evaluation of the Microbiology
Standards for Drinking Water (C. W. Hendricks, ed.), U. S. Environmental
Protection Agency, Washington, D. C.
Gerba, Charles, P., C. Wallis, and J. L. Melnick. 1975. Microbiological
hazards of household toilets: droplet production and the fate of
residual organisms. Appl. Microbiol. 30:2:229-237.
Goshko, M. A., H. A. Minnigh, W. 0. Pipes, and R. R. Christian. 1981. Rela-
tionships between standard plate counts and other parameters in water
distribution systems. Presented at the 1981 Annual Conference.
American Water Works Association, St. Louis, Mo., June 1981.
158
-------
Harrison, J. 1978. Experiences with the coliform standard, pp. 159-168.
In. Evaluation of the Microbiology Standards for Drinking Water
(C. W. Hendricks, ed.), U. S. Environmental Protection Agency,
Washington, D. C.
Lariviere, J. W. 1978. Survival of Escherichia coli in four diluents.
M.S. thesis, Department of Biological Sciences, Drexel University,
Philadelphia, Pa.
LeChevalier, M. W., R. J. Seidler, and T. M. Evans. 1980. Enumeration and
characterization of standard plate count bacteria in chlorinated and
raw water supplies. Appl. Environ. Microbiol. 40;922-929.
Muenz, L. R. 1978. Some statistical considerations in water quality control.
pp. 49-56. In. Evaluation of the Microbiology Standards for Drinking
Water (C. W. Hendricks, ed.), U. S. Environmental Protection Agency,
Washington, D. C.
Pielou, E. C. 1969. An Introduction to Mathematical Ecology, Wiley-Inter-
science, New York.
Pipes, W. 0., P. Ward, and S. H. Ahn, 1977. Frequency distributions for
coliform bacteria in water. Jour. Am. Water Works Assoc., 69:644-647.
Pipes, W. 0. and R. R. Christian. 1978. A sampling model for coliforms in
water distribution systems. Presented at the 1978 Annual Conference,
American Water Works Association, Atlantic City, N.J., June 1978.
Sokal, R. R., and F. J. Rohlf. 1969. Biometry. W. H. Freeman and Co.,
San Francisco, Ca.
Thomas, H. A., Jr. 1952. On averaging results of coliform tests, Jour.
Boston Soc. Civil Engineers, _39:253-264.
Thomas, H. A., Jr. 1955. Statistical analysis of coliform data. Sew. Ind.
Wastes, 27:212-228.
U. S, E. P. A. 1974. Methods for Chemical Analysis of Water and Wastes.
Office of Water Supply, U. S. Environmental Protection Agency,
EPA-625/6-74-003. Washington, D. C.
U. S. E. P. A. 1976. National Interim Primary Drinking Water Regulations.
Office of Water Supply, U. S. Environmental Protection Agency,
EPA-570/9-76-003. Washington, D. C.
U. S. Public Health Service. 1962. Drinking Water Standards. Department of
Health, Education, and Welfare., Washington, D. C.
Velz, C. J. 1951. Graphical approach to statistics. IV. Evaluation of
bacterial density. Water Sew. Works, Feb. 1951, pp. 66-73.
159
-------
APPENDIX J
EFFECT OF FILTER BACKWASHING
ON BACTERIOLOGICAL QUALITY OF
DRINKING WATER
Report of Special Study
-------
APPENDIX J
EFFECT OF FILTER BACKWASHING ON BACTERIOLOGICAL
QUALITY OF DRINKING WATER
J.I OBJECTIVE
To determine if there are differences in the microbial quality of water
as measured by coliform and standard plate counts entering a distribution
system between when the filters are being backwashed and when they are
operating normally.
J.2 HYPOTHESES
The purpose of water filtration is to remove particulate matter, includ-
ing bacteria, from the water before it enters a distribution system. These
solids are separated from the filter by backwashing but some small amounts
may be left in the filter after backwashing is complete and the filter
medium is allowed to settle back into place. During backwashing the total
amount of water being produced by the treatment plant is reduced and finished
water from storage may enter the distribution system.
1. If prechlorination is not used or is not completely effective,
the filter backwash water may contain large numbers of bacteria,
some of which will be coliforms.
2. After backwashing and after the filter medium has been allowed
to settle back into place, the first flush of water through
the filter may contain large numbers of bacteria including
some coliforms.
3. Some coliforms may enter water distribution systems from filtered
water reservoirs, standpipes or elevated storage tanks if the
pressure in the distribution system is allowed to drop during
filter backwashing.
J.3 SYSTEMS STUDIED
The water systems selected for study were Downingtown, Pennsylvania
(sampled January and February 1980) and Brooklawn, New Jersey (sampled
March and April 1980). In addition some samples were collected in July 1979
during filter backwashing at the two treatment plants used by the Long Beach
Water Company. Although these data were collected before the backwash study
was designed they are included as additional evidence related to two of the
hypotheses being tested. Basic descriptive information on the treatment
Jl
-------
plants and their water sources is given in Appendicies D, E, and C.
J.3.1 Long Beach Water Company
The two treatment plants are 3 miles apart and serve different areas of
the distribution system. However, they are essentially identical facilities.
During the summer, when these samples were collected, the filters are back-
washed on Mondays, Wednesdays, and Fridays, those at the Terrace Plant in
the morning and those at the Brant Beach Plant in the afternoon. The back-
wash water is discharged into the sewer system. There are three high lift
pumps at each plant rated at 1000 gpm, 500 gpm, and 500 gpm. During normal
operation the 1000 gpm pump is used to pump water through the filters into
the distribution and is turned on and off by a pressure switch located at
the base of the elevated storage reservoir or standpipe. During the winter
the high lift pump operates about 2.5 hours per day. During backwashing the
1000 gpm pump is used to backwash one filter while the two 500 gpm pumps are
used to pump water through the other two filters into the distribution system.
Since the backwash water is taken from the distribution system, the water
use for backwashing must come from the elevated storage. There is no level
or pressure recorder on either the standpipe at the Terrace Plant or the
elevated storage tank at the Brant Beach Plant and the water pumpage recorder
is turned off during backwashing. Thus, there is no documentation of water
use during backwash nor of the drawdown of water in storage.
J.3.2 Downingtown
The water filtration plant at Downingtown is operated only 16 hours per
day. The filtered water in the 3.75 million gallon open reservoir is pumped
into the distribution system by high lift pumps which are operated indepen-
dently of the filtration plant. Chlorine to provide a disinfectant residual
in the treated water is added at the high lift pumping station. The elevated
storage tanks are on the distribution system at maximum distances from the
treatment plant and contain enough water to supply the system for 2 to 4 days
even if the high lift pump were not operating. There are no loss-of-head
meters on the filters. The filters are backwashed after 30 hours of opera-
tion using water from an elevated backwash water tank. The backwash water
is discharged to a settling pond from which the overflow goes to the raw
water storage.
J.3.3 Brooklawn
The operation of the Brooklawn water treatment plant is similar to that
of the two plants of the Long Beach Water Co. The high lift pump supplies
pressure to force the water through the filters and into the distribution
system. It is turned on and off by a pressure switch on the main which
supplies water to the distribution system. During backwashing, the high
lift pump is turned off and water from the distribution system is used for
backwash. The three filters are backwashed every morning for 10 minutes
each and the drawdown in the elevated storage tank on the other side of town
is only 1 to 2 feet. The backwash water is discharged into a storm sewer
which empties into a tidal flat area of the Delaware estuary.
J2
-------
J.4 RESULTS OF TESTING HYPOTHESES
J.4.1 Bacteria in Backwash Water
J.4.1.1 Long Beach Water Company
Backwash water samples were not collected at either plant because this
sampling was carried out before the backwash study was designed. However,
there is no disinfection of the water before filtration so the hypothesis
could not be tested properly by samples from these plants. Bacteriological
results for raw water samples are given in Table Jl.
The one confirmed coliform found in the raw water at the Brant Beach
Plant on 6/14 was identified as Citrobacter freundii. A presumptive coliform
was found in one of the samples collected on 6/19 but it did not confirm and
was identified as Enterobacter agglomerans. The standard plate counts were
consistently low except for one sample collected on 6/21.
No presumptive coliforms were found in the raw water at the Terrace
Plant. The standard plate counts were more variable than those on samples
from the Brant Beach Plant. However, it is clear that the well water is
not a major source of bacteria.
J.4.1.2 Downingtown Water Treatment Plant
The results pertinent to the first hypothesis are presented in Table J2.
The volumes of raw water filtered for the coliform determinations were 1 ml
and 10 ml. Even so, some filters on samples from 1/15 had more than 80
coliform colonies present. Coliforms identified from the raw water included
Escherichia coli, Citrobacter freundii, Klebsiella oxytoca, and Enterobacter
spp.. The standard plate counts on the raw water were also very high.
It is clear that chlorination, flocculation and settling at this plant
is very effective in reducing the coliform density and standard plate count.
The one coliform found in the treated water samples was Klebsiella pneumoniae.
The multiple tube fermentation tests on the backwash water did not
provide definite evidence of the presence of coliforms. The tests were run
using standard 10 ml portions on the first three sampling days and no pos-
itive tubes were found. On the last two sampling days five 100 ml, five 10
ml, and five 1 ml portions were used. Gas production was found in some of
the 100 ml portions but it was not possible to confirm the tests or isolate
coliforms from the broth. One isolate from one backwash sample collected
on 1/29 was identified as Enterobacter agglomerans but it did not produce
gas in brilliant green bile broth. The standard plate counts on the backwash
water were not significantly different from those on the treated water
samples.
J.4.1.3 Brooklawn Water Treatment Plant
These results are presented in Table J3. No coliforms were found in the
J3
-------
CO
i!
O 4J
O 4J Q)
O CO Q
0)
4-1 x-.
Cd rH
rH 0
PM
• i»J C2
T3 (U Cd
4-1 PH d)
cn "^ g
• •
^ *S 5
OT 6 C/3 Q
6 o
O rH
U-l
•HUC
rH CU Cd
° £ 3
CO
<4H £>
CO CM CM CO rH
• • • • •
O rH O O 1 1 CO 1 CO 1
CM CM CM
m
CM u-| rH u-| CO
oinoooooocoo
H CO rH
i i i i i i i i i i
rH
oo
•
O
rH
m
-------
CM
1-1
W
1
Z
O
H
1
H
1"^
M
PM
W
O
EM
M
co
rJ
p I
2*
3
CO
p^ £5
W §
H g
§ H
B co
O j^i
H rV
CO
[V] A
E"^ £3
G
H o
< H
o o
H Z
O M
° 5
O O
M O
H
O
pq
Fn
o
CO
H
g
CO
W
fx|
Q>
4J
£t
m
i
o
cd
PQ
^
01
4J
cd
S
•o
01
iH
4J
4)
CO
W
0)
4J
si
1
4)
i
41
4-1
cd
J]
•
o
4J
yj
u
>
a.
s
U-J
O
Q
2
0)
4J
*J D
co O
a
cd
O
S
^
rH
J
O
Q
rH
rl
01
a
• •
~3 4)
co a
a
id
s
, .
•o >
en Q
C
cd
S§
•W !>
u a)
co a
c
•a >
4-1 01
CO O
a
cd
33
CM
•
M
O
•
irH
CM
V
m
CO
"^
rH
i-H
CO
•
CM
O
O
CO
vD
rH
CM
CO
CM
•-H
co
CM
co
CM
VO
O
00
03
rH
CM
CM
•
r^
CM
0
•
0>
CM
CM
V
m
CM
in
rH
O
H
CO
^
i-l
0
H
vO
1
O
1
o
EH
H
VO
O
00
in
H
rH
CO
•
r-
vo
•
in
rH
CM
V
m
rH
VO
CM
CO
,_4
CM
0
O
VO
1
CJ
s
H
^.
CM
03
m
co
vO
o
00
CM
CM
rH
m
vO
•
-3-
rH
CM
•
00
CM
CM
in
•*
co
rH
CM
CM
CM
0
O
in
rH
VO
rH
O
0
VO
O\
in
o
m
sr
rH
VO
O
CO
CT*
CM
rH
0
O
•
ao
o\
•
o\
co
CM
m
CO
o
00
-a-
m
0
o
vO
'
CJ
1
f^
rH
CO
CO
in
vO
0
00
•^^
m
CM
VO
CTi
•
CO
rH
CO
00
•
CM
rH
in
CM
r~
^.
CO
in
^
CM
co
O
o
co
o
o
rH
CO
1
O
H
1
CJ
0
CO
rH
rH
Cd
r<
41
J5
-------
u
41
U
IS
*«3
(0
u
£
r*
0)
*J
cd
•a
41
cd
41
M
H
M
0)
td
Jj
.3
41
U
cd
a
u
c
3 • •
O T3 >
J 4J 01
CO O
41
•PC
O 41 Cd
4J CXi 4)
» v-S
Z
a
41 ^
^ ^H *O
H S 0 U
U O -H CO
P.O l-l
0 rH 4J
3 0)
to l-i 6 C
P* ^O 5
tpw
|S5 to
/_s
i-l • •
BS S;
4) M CO Q
JJ 41
tQ (^
• c c
2 3 (d
W 0 ^
y-N • •
t"H *O ^
W S *J 4(
sown
w O
O rH
MH
•H u a
rH 4) Cfl
O Cu 4)
O xx S
IW •
O cd
^ CO
x^ * •
rt j_i flj
0 wo
41 t*
4J 4)
(d Q,
ft, 4J
• B Ci
•d 3 cd
4J O QJ
M o a
X"N • •
rH -0 >
ra B -u 5
a o co a
H C3
O rH
•H VJ C
H 41 Cd
° £ &
VH •
2; co
o\
CO
CM
CN|
•
CO
,,_
sr
CM
CO
ao
d
m
^j.
VO
CM
d
m
CM
d
1
o
0
rH
0V
CM
sr
CM
o\
•*
1
O
VO
0
CO
d
CM
•
o
o
CO
CM
m
CM
CM
sr
d
o
CM
d
i
o
o
f_l
-------
well water or treated water samples. The standard plate count for the
treated water is consistently lower than that of the raw water. The 10 mg/1
chlorine dose added after aeration and before filtration disappears rapidly
in the filters. Free chlorine levels in the treated water and filtered
water samples on 3/4, 3/11, and 3/18 ranged between 0.4 and 0.7 mg/1. On
3/25 the free chlorine concentration in the filtered water was 1 mg/1. It
appears that the variation in the chlorine residual is related to variable
chlorine demand in the well water and a reduction in the standard plate
count is achieved in spite of the chlorine demand.
The multiple tube fermentation test of the backwash water samples con-
sisted of five 100 ml tubes, five 10 ml tubes, and five 1 ml tubes. No gas
production was detected in any of the 1 or 10 ml tubes. Gas production was
detected in all of the 100 ml tubes for all five samples on 3/18 but in none
of the 100 ml tubes on 3/25. Variable numbers of 100 ml tubes showed gas
production in the different samples on 3/4 and 3/11. Transfers from the
LTB tubes with gas to BGLB did not result verification of the presence of
coliforms. An isolate from one of the 100ml tubes from 3/19 was identified
as Klebsiella ozaenae and required 5 days to produce gas in BGLB. No other
isolates were members of the Enterobacteriacae.
J.4.2 Coliforms in Filtered Water
None of the three systems tested filter to waste following backwash.
The treatment plants using well water take backwash water out of the distri-
bution systems and filtering to waste would allow further drawdown of
elevated water storage. Since the filtered water at the Downingtown Water
Treatment Plant passes through an open reservoir and is chlorinated before
it is pumped into the distribution system, there is no particular reason to
filter to waste there.
J.4.2.1 Long Beach Water Company Plants
Samples of the water entering the distribution system during backwash
were collected on 7/18/79. Samples of water entering the distribution sys-
tem while the filters were being operated normally were collected on other
sampling days. A comparison of the bacteriological results of these two
different sets of samples is given in Table J4. No coliforms were found
in the finished water samples and the standard plate counts are lower during
backwashing than at other times.
J.4.2.2 Downingtown Water Treatment Plant
The bacteriological results on filtered water samples are given in Table
J5. The volumes filtered were 1000ml each on 1/8 and 500 ml each on the
other sampling days. The turbidity of all of these samples was very low,
< 0.1 NTU, but the time required for filtration of the 1000 ml samples was
10 to 15 minutes each. The 500 ml samples filtered in about 2 minutes each.
Two presumptive coliforms were found in the filtered water samples after
backwashing on 1/8. One did not produce gas in BGLB and did not grow on EMB
J7
-------
-*
^"3
3
«2
P_J
co
W
t!
CO
H
H
£
g
H
H
1— t K-I
2 &!
O 25
^
CO CJ
E-i
CO OS
ptj pri]
H H
jj
CJ ffi
1 1 PJ
0
CO S 4J CU
B c5 co Q
M 0
O iH
•H I* 0
rH CU Cd
O ft CU
CO
cu
O ft
• s
o ca
& CO
60
•H
CO
ca
o
ca
60
C
•H
1
+-I
fj
g . .
0 TJ >
U 4J CU
CO Q
4-1 /^
ctf i-H
•H S
P-i
• M C
T) cu ca
4J ft CU
co ^-^ g
/^ • •
rH T3 >
CO fi 4-1 CU
s o co n
fi o
0 iH
•rl M C
H CU ca
cSvS^
CO
CU
0 ft
• B
o ca
3 co
4-1
a
ca
r-l
P*
ro r^«
cr» co
• •
CO rH
vO 00
• »
r-H CO
i i
o o
a^ vo
CM eg
H in
•
-------
CO
W
1
CO
PS
E~i
^4
15
Q
m
PQ
W
H
rfS
5|
O £j
^4
CO CO
H Z
1 CO Z
1 H 0
H m
•""! **
CJ 2
H O
O H
O O
. *X3 >
H -U 0)
W J CO Q
s ^?
w ^^
0 H
M-t
•rl H C
rH (1) CO
O O-i Q)
c.3 ^^^ 5r^
«4H •
°t
O Cfl
Z co
j-
(0
Cfl
O
cd
PQ
01
M
O
(U
PQ
4J
C >
3 0)
o n
x^\ *rj
0) r-j 4J
4J S CO
Cfl <5
H O
P4 H
• u a
13 01 cfl
4J 04 CU
CO '-•' g
/*s • •
i™j *o ^
co s -u a)
ggcop
0 rH
14-1
•H rl C
rH 01 CO
o >-' S
M-l 4
o a
• S
O CO
Z co
CD
4J
CO
Q
oo
m
i-H
o
m
o
•
0
CN
CO
o
o
rH
o
O
O
i-H
1^
m
o
vO
vO
•
o
1
o
CO
o
oo
00
i-H
rH CO
CN CN
CN rH
O CO
CO CN
• •
vo -
m in
m CN
CO CN
co a\
• •
oo «*
I i
0 O
m m
rH H
O O
oo oo
•^^ *^^
m CN
rH CN
^^ *^
rH i-H
rH ^^ O^
rH 00 0\
rH CN CN
oo -a- «»
00 vO >*
• • »
O -* CO
m
rH
o
i i
o
O O rH
CN m r~.
rH rH VO
r^ o o
CTi vO O
H rH «Jt
-l
CN m (U
*^*^ ^^. ^
rH CN O
J9
-------
agar and is not recorded as a verified coliform. The other did verify in
BGLB and was identified by the API system as a strain of Enterobacter which
was never found in the raw water. The most likely source for these bacteria
is the backwash water storage tank.
No significant differences in the standard plate count in the filtered
water before and after filtration were found.
J.4.2.3 Brooklawn Water Treatment Plant
The bacteriological results on filtered water samples are presented
in Table J6. The volumes filtered were 500 ml. No coliforms were detected
in the filtered water before or after filtration which is consistent with
the results on the raw water and treated water samples. The standard plate
counts on samples before and after filtration are not significantly differ-
ent from each other nor from the standard plate counts on the treated water
before filtration (Table J3).
J.4.3 Coliforms from Elevated Storage Tanks
J.4.3.1 Long Beach Water Company
There is no record of how much drawdown of water in elevated storage
occurs during backwashing. The water pumpage at each of the two plants on
week days during July 1979 varied between 900,000 and 1,150,000 gallones.
If it is assumed that the water useage is distributed evenly throughout the
day and the water going into the distribution system during backwashing
comes from elevated storage then the amount of water taken out of elevated
storage would be between 112,500 and 143,750 gallons. The standpipe at the
Terrace Plant holds 173,000 gallons and the elevated storage tank at the
Brant Beach Plant holds 150,000 gallons; thus, the drawdown during backwash-
ing would be rather large. However, the two plants pump water into the same
distribution system and the drawdown at one plant could be limited by
additional water pumped from the other plant. The most likely occurrence
during backwashing would be a drawdown of water from elevated storage at
the plant where the filters are being backwashed and a reversal of flow in
the north-south mains in the area of the distribution system between the
two plants. The reversal of flow in a large area of the distribution system
is likely to be a more significant cause of increased numbers of coliforms
in distribution system samples than the drawdown of water from elevated
storage.
J.4.3.2 Downingtown Water Treatment Plant
The pressure in the water distribution system is completely unaffected
by backwashing of filters, so this hypothesis could not be tested on this
system.
J.4.3.3 Brooklawn Water Treatment Plant
The only reasonable test of the hypothesis that elevated storage tanks
J10
-------
CO
w
H^
pn
s
^
CO
p4
1
§
P . ^,
••5 rn
H CO
W
55 ^
vO O
CO Ed
(4 H 55
rJ CO
M W «
< H g
H Es
U S
H O
8O
P4
rJ PP
O
s
w
1
P*
O
CO
t"j
c
CO
2
j-
CO
• •
n g 4J Q)
S O CO O
M O
O r-t
•H M g
H 4) 3
° ^ J
o o.
o §
2 CO
4J • •
§tJ >
4J V
o co n
0)
4J <^
• M C
a ai rt
4J Q< 0)
co ^X
x*^ • •
g o co a
0 H
•HMO
r-f 0) C0
o a. J
o ^^ z
M-l •
o a.
• §
o 3
2 CO
rH •»
rH H
-CM •*
0 0
O O
0 0
1 1
0 0
o m
H t-l
rH CM
H m
CM CO
0 0
CO
O CO
rH rH
0 0
1 1
0 0
o m
rH rH
O
O 00
00 -^
•«^ rH
•sr iH
*>fc ^^
CO CO
CM
00
rH
O
H
1
O
in
i-H
CO
CO
o
CO
o
1
o
in
rH
O
CO
1 — .
00
rH
^^
CO
m
CO
0
m
CM
H
o
1
o
00
CM
H
m
CO
0
,
o
r^
O
00
m
CM
CO
CO
H
H
CO
m
o
1
o
CO
*^
\o
H
O
O
o
f^
«^
rH
ca
H
8
Jll
-------
could be a source of collform bacteria for the distribution system was made
in Brooklawn after the sampling of the filters and backwash was completed.
The separation in time was necessary to avoid overloading our laboratory
capacity to process samples.
Samples from the top of the elevated storage tank were obtained by
having the operator manually override the cut off on the high lift pump and
allow the pump to operate until the tank overflowed. The overflow pipe was
flushed out for 20 minutes before the samples were taken. The samples from
the bottom of the tank were taken from the line which connects the tank to
the distribution system.
The bacteriological results of the samples from the elevated storage
tank are presented in Table J7. Coliforms were found only on one sampling
day on samples from the top of the tank. The standard plate counts were
always higher on the samples from the top of the tank than on the samples
from the bottom of the tank but not significantly different.
No free chlorine residual was found in any of these samples but the
total chlorine residual ranged from a trace to 0.7 mg/1. The lowest
chlorine residuals were found on 4/15.
As shown in Table J8 eight of the nine coliforms from the elevated
storage tank were identified as Escherichia coli. According to the results
of the API test all of these J2. coli isolates were the same biotype. An
j£. coli of the same biotype was found in a distribution system sample col-
lected one block from the elevated storage tank on an earlier sampling
date. This was the only J5. coli isolate from the distribution system during
the entire sampling period.
The most likely source of the coliforms found in the top of the storage
tank is birds. The predominance of IS. coli is suggestive of recent fecal
contamination. It would be expected that E_. coli would die-away rapidly in
the presence of a free chlorine residual.
J.5 SUMMARY AND FINDINGS
Sampling programs carried out at four different water treatment plants
provide some information on the effect filter backwashing may have on the
microbiological quality of drinking water. Since the results were largely
negative and the two treatment plants operated by the Long Beach Water Com-
pany are similar to the treatment plant of the Brooklawn Water Department,
these findings must be considered to be tentative until or unless other
treatment plants with different raw water quality and operating conditions
are examined.
1. Backwash water may contain coliforms and other bacteria removed by
filtration, but the coliform density and standard plate count can be
expected to be very low in most cases.
J12
-------
IS^.
r^
>->
w
nJ
3
Q
H
EH
1
W
,_q
H
a
§
Pn
to
U
rJ
5J w
to to
Pi
2; w
° t*! "^
to 5 g
to H sz
H
H W •>
dp
<1 w ^S
^^ • •
CO 5 ^ 9
gO CO 0
o
O H
>w
•H n a
rH 01 Cd
o a J
O »-' S
0)
MH V
0 rH
o a
Z CO
•a
cd
£-4
MH
O
a
Q
H
8
0
1-1
VH
ca
01
rH
PS
z
V.
•
4J >•
g «
O T3
u
01 CO
4J
cd *-N
rH rH
IX 0
• ^ fi
•O CD Cd
w a ai
co >-- £
x-s • •
rH *O ^
W 0 4J 0)
g^S CA /"»
0 COP
O rH
UH
•rl P fl
rH 0) Cd
O Q. >O 00
rH CM H
1
rH rH i-H
rH VO 0\
• • •
rH rH 0 0
till
1 1 1 1
O O O O
o o o o
rH rH rH CO
«* vO CO •*
00 rH 00 r»-
• • • •
rH CO CM CM
r»
vo co r^ oo
• • • •
rH CO CO CM
r^ in
CO vO
• •
i too
o
O» CO
• •
O O 0 O
o o o o
rH H rH CO
O rH
O O 00 H
OO OO ^- Cd
— — in u
rH 00 rH 01
-* -» sr c5
J13
-------
TABLE J8
IDENTIFICATION OF COLIFORMS FROM STORAGE
TANK SAMPLES
BROOKLAWN, NEW JERSEY
Date Sample Coliform Colonies Verification IMViC API
No. Count Isolated LTB BGLB Identification
4/15 175T-2 1
175T-3 2
175T-4 1
175T-5 1
175T-6 2
175T-7 2
1 + + -H— - Escherichia coli
2 + + ++— Escherichia
coli(2)
1 + + ++-- Escherichia coli
1 + + ++— Escherichia coli
2 + + -f-f Citrobacter
freundii
+ + ++ — Escherichia coli
2 + + -H — Escherichia
coli(2)
J14
-------
a. When coliforms are not found in the raw water and treated
water before filtration, they are not found in the backwash
water (Brooklawn and the two Long Beach plants).
b. Standard plate counts on the backwash water are not signif-
icantly higher than standard plate counts on the treated water
before filtration.
c. The high concentrations of suspended solids in the backwash
water make the interpretation the results of fermentation tube
tests questionable at best. Even when gas production in lauryl
tryptose broth is observed it is difficult either to verify the
presence of coliforms using brilliant green bile broth or to
isolate coliforms from the original tubes.
2. A coliform was found only once in the first flush of water from a
filter following backwash.
a. Clearly coliforms will not be introduced into the filter
by backwashing if there are no coliforms in the water used for
backwashing. This apparently was the case at Brooklawn and the
two Long Beach Plants.
b. If the backwash water is accumulated in an elevated storage
tank, the storage tank may be a source of coliforms which could
be found in the filter immediately following backwashing.
3. Elevated storage tanks on the distribution system may contain high-
er numbers of coliform bacteria than most of the water in the distri-
bution system. If drawdown of water in the elevated storage tank
occurs during backwashing, coliform bacteria may be introduced into the
distribution system.
•u.s. eoTMomiT vKuma omcxi 1982-0-361-082/316
J15
-------
|