EPA/600/R-10/061
Assessing Potential Impacts Associated With
Contamination Events in Water Distribution Systems:
A Sensitivity Analysis
United States Environmental Protection Agency
Cincinnati, Ohio 45268
November 2010
-------
11
-------
Contents
Disclaimer v
Foreword vii
List of Figures ix
List of Tables xix
Acronyms and Abbreviations xxi
Acknowledgments xxiii
Summary xxv
1 Introduction 1
1.1 Definitions 1
1.2 Navigating This Report 2
2 Methodology 3
2.1 Water Distribution Systems Examined 3
2.2 Analysis 8
3 Nature of Impacts 15
4 Sensitivity of Impacts to Injection Duration 25
5 Sensitivity of Impacts to Injection Time 45
6 Sensitivity of Impacts to Injection Mass 91
7 Sensitivity of Impacts to Population Distribution 115
8 Sensitivity of Impacts to Ingestion Model 135
9 Sensitivity of Impacts to Combinations of Factors 155
10 Sensitivities Examined 181
iii
-------
iv CONTENTS
11 Discussion 195
11.1 Methodological Issues 195
11.2 Computational Issues 197
11.3 Application of Results 198
11.4 Additional Work 198
12 Conclusions 205
References 207
Appendix: Quality Assurance 209
-------
Disclaimer
The U.S. Environmental Protection Agency (EPA) through its Office of Research and Develop-
ment funded, managed, and collaborated in the research described herein under an interagency
agreement with the U.S. Department of Energy through Contract DE-AC02-06CH11357 with Ar-
gonne National Laboratory. It has been reviewed by the Agency but does not necessarily reflect the
Agency's views. No official endorsement should be inferred. EPA does not endorse the purchase
or sale of any commercial products or services.
Because of the confidentiality of the information, the identity of the water distribution systems
used in this report, the associated network models, and any information that could be used to
identify the systems cannot be disclosed.
-------
vi Disclaimer
-------
Foreword
Following the events of September 11, 2001, the U.S. EPA's mission was expanded to address
critical needs related to homeland security. Presidential directives identified EPA as the primary
federal agency responsible for the country's water supplies and for decontamination following a
chemical, biological, and/or radiological attack. To provide scientific and technical support in
meeting this expanded mission, EPA's National Homeland Security Research Center (NHSRC)
was established. NHSRC is focused on conducting research and delivering products that improve
the capability of the Agency to carry out its homeland security responsibilities.
As a part of this mission, NHSRC conducts research and provides tools and methodologies
to support America's drinking water utilities so that they can improve their security preparedness,
response, and recovery. Over the last several years, NHSRC has been developing new methods
to help design, implement, and evaluate drinking water contamination warning systems. These
new systems integrate a variety of monitoring technologies to rapidly detect contamination. Water
distribution system networks are composed of hundreds to thousands of miles of pipe and the con-
tamination warning system must economically safeguard the largest number of people reasonably
possible. Designers of contamination warning systems need to understand how key factors associ-
ated with contamination events affect public health outcomes. For example, how does the location
of the contaminant release, the toxicity of the contaminant, or the length of time over which the
contaminant is introduced into the water system influence the number of people affected? This
study examines the adverse effects of contamination events in water distribution systems using
models for 12 actual systems that serve populations ranging from about 104 to over 106 persons.
It extends previous, related work and provides an improved understanding of the nature of the
adverse effects that could be associated with contamination events. The results presented support
water utilities, their consultants, and researchers in conducting contaminant vulnerability analyses
and designing and implementing contamination warning systems.
This report results from collaborative research between the NHSRC and Argonne National
Laboratory. It is published and made available by EPA's Office of Research and Development in
order to enable the water community to improve the security of our nation's drinking water.
Jonathan Herrmann, Director
National Homeland Security Research Center
vn
-------
viii Foreword
-------
List of Figures
2.1 Distribution of nodal populations for the networks below 250 persons per node. . . 5
2.2 Network 4 6
2.3 Network 6 7
2.4 Overlap of high-percentile injection nodes 11
2.5 Comparison of a violin plot and a box plot 12
3.1 Impacts for Networks 1 to 4 as function of dose level 17
3.2 Impacts for Networks 5 to 8 as function of dose level 18
3.3 Impacts Networks 9 to 12 as function of dose level 19
3.4 Network 4 showing locations of injection nodes associated with 90th and 95th
percentile or higher impacts at a dose level of 0.0001 mg 20
3.5 Network 4 showing locations of injection nodes associated with 85th and 95th
percentile or higher impacts at a dose level of 0.0001 mg 21
3.6 Network 4 showing locations of injection nodes associated with 80th and 95th
percentile or higher impacts at a dose level of 0.0001 mg 22
3.7 Network 6 showing receptor nodes affected following injections at the 90th per-
centile nodes for dose levels of 0.0001 and 1.0 mg 23
4.1 Impacts associated with 1-h and 24-h injections for Network 12 as a function of
dose level, for the 90th, 95th, 99th, and 100th percentile injection nodes 28
4.2 Impacts associated with 1-h and 24-h injections for Network 12 as a function of
dose level, for the 50th, 60th, 70th, and 80th percentile injection nodes 29
4.3 Impact for a 24-h injection relative to the impact for a 1-h injection, as a function
of dose level, for the 90th, 95th, 99th, and 100th percentile injection nodes 30
4.4 Impact for a 24-h injection relative to the impact for a 1-h injection, as a function
of dose level, for the 50th, 60th, 70th, and 80th percentile injection nodes 31
4.5 Distributions of impacts for Network 12 across all injection nodes for 1-h and
24-h injections for dose levels of 0.0001,0.001,0.01, and O.lmg 32
4.6 Distributions of impacts for Network 12 across all injection nodes for 1-h and
24-h injections for dose levels of 1, 10, and 100 mg 33
4.7 Impacts for Network 12 for 1-h and 24-h injections for dose levels of 0.0001,
0.001,0.01, and O.lmg 34
4.8 Impacts for Network 12 for 1-h and 24-h injections for dose levels of 1, 10, and
100 mg 35
4.9 Impacts for Network 12 that are at or above the 80th percentile level for 1-h and/or
24-h injections for dose levels of 0.0001,0.001,0.01, and O.lmg 36
ix
-------
LIST OF FIGURES
4.10 Impacts for Network 12 that are at or above the 80th percentile level for 1-h and/or
24-h injections for dose levels of 1, 10, and 100 mg 37
4.11 Overlap of high percentile injection nodes for Network 12 for 1-h and 24-h injec-
tions for different dose levels 38
4.12 Overlap of high percentile injection nodes for the networks for 1-h and 24-h in-
jections for dose levels of 0.0001, 0.001, 0.01, and 0.1 mg 39
4.13 Overlap of high percentile injection nodes for the networks for 1-h and 24-h in-
jections for dose levels of 1, 10, and 100 mg 40
4.14 Network 4 showing locations of injection nodes associated with 95th percentile or
higher impacts at a dose level of 0.0001 mg for injection durations of 1 h and 24 h. 41
4.15 Network 4 showing locations of injection nodes associated with 95th percentile or
higher impacts at a dose level of 1.0 mg for injection durations of 1 h and 24 h. . . 42
4.16 Network 6 showing locations of injection nodes associated with 95th percentile or
higher impacts at a dose level of 0.0001 mg for injection durations of 1 h and 24 h. 43
4.17 Network 6 showing locations of injection nodes associated with 95th percentile or
higher impacts at a dose level of 1.0 mg for injection durations of 1 h and 24 h. . . 44
5.1 Impacts associated with different injection times for Network 12 as a function of
dose level, for the 90th, 95th, 99th, and 100th percentile injection nodes 47
5.2 Impacts associated with different injection times for Network 12 as a function of
dose level, for the 50th, 60th, 70th, and 80th percentile injection nodes 48
5.3 Maximum ratio of impacts for the networks for four different injection times as a
function of dose level, for the 90th, 95th, 99th, and 100th percentile injection nodes. 49
5.4 Maximum ratio of impacts for the networks for four different injection times as a
function of dose level, for the 50th, 60th, 70th, and 80th percentile injection nodes. . 50
5.5 Relative impacts for the 12 networks for different injection times, for a dose level
of 0.0001 mg and for injection nodes associated with different percentile impacts. 51
5.6 Relative impacts for the 12 networks for different injection times, for a dose level
of 0.001 mg and for injection nodes associated with different percentile impacts. . 52
5.7 Relative impacts for the 12 networks for different injection times, for a dose level
of 0.01 mg and for injection nodes associated with different percentile impacts. . . 53
5.8 Relative impacts for the 12 networks for different injection times, for a dose level
of 0.1 mg and for injection nodes associated with different percentile impacts. . . 54
5.9 Relative impacts for the 12 networks for different injection times, for a dose level
of 1 mg and for injection nodes associated with different percentile impacts. ... 55
5.10 Distributions of impacts for Network 12 across all injection nodes for different
injection times, for dose levels of 0.0001, 0.001, 0.01, and 0.1 mg 56
5.11 Distributions of impacts for Network 12 across all injection nodes for different
injection times, for dose levels of 1, 10, and 100 mg 57
5.12 Impacts for Network 12 for injection times of 0:00 and 6:00, for dose levels of
0.0001, 0.001, 0.01, and 0.1 mg 58
5.13 Impacts for Network 12 for injection times of 0:00 and 6:00, for dose levels of 1,
10, and 100 mg 59
5.14 Impacts for Network 12 for injection times of 0:00 and 12:00, for dose levels of
0.0001, 0.001, 0.01, and 0.1 mg 60
-------
LIST OF FIGURES xi
5.15 Impacts for Network 12 for injection times of 0:00 and 12:00, for dose levels of
1, 10, and 100 mg 61
5.16 Impacts for Network 12 for injection times of 0:00 and 18:00, for dose levels of
0.0001, 0.001, 0.01, and 0.1 mg 62
5.17 Impacts for Network 12 for injection times of 0:00 and 18:00, for dose levels of
1, 10, and 100 mg 63
5.18 Impacts for Network 12 that are at or above the 80th percentile level for an injec-
tion time of 6:00 and/or an injection time of 0:00, for dose levels of 0.0001, 0.001,
0.01, and 0.1 mg 64
5.19 Impacts for Network 12 that are at or above the 80th percentile level for an injec-
tion time of 6:00 and/or an injection time of 0:00, for dose levels of 1,10, and 100
mg 65
5.20 Impacts for Network 12 that are at or above the 80th percentile level for an in-
jection time of 12:00 and/or an injection time of 0:00, for dose levels of 0.0001,
0.001,0.01, and 0.1 mg 66
5.21 Impacts for Network 12 that are at or above the 80th percentile level for an injec-
tion time of 12:00 and/or an injection time of 0:00, for dose levels of 1, 10, and
100 mg 67
5.22 Impacts for Network 12 that are at or above the 80th percentile level for an in-
jection time of 18:00 and/or an injection time of 0:00, for dose levels of 0.0001,
0.001,0.01, and 0.1 mg 68
5.23 Impacts for Network 12 that are at or above the 80th percentile level for an injec-
tion time of 18:00 and/or an injection time of 0:00, for dose levels of 1, 10, and
100 mg 69
5.24 Overlap of high percentile injection nodes for Network 12 for an injection time of
6:00 with those for an injection time of 0:00 for different dose levels 70
5.25 Overlap of high percentile injection nodes for Network 12 for an injection time of
12:00 with those for an injection time of 0:00 for different dose levels 71
5.26 Overlap of high percentile injection nodes for Network 12 for an injection time of
18:00 with those for an injection time of 0:00 for different dose levels 72
5.27 Overlap of high percentile injection nodes for the networks for an injection time
of 6:00 with those for an injection time of 0:00, for dose levels of 0.0001, 0.001,
0.01, and 0.1 mg 73
5.28 Overlap of high percentile injection nodes for the networks for an injection time
of 6:00 with those for an injection time of 0:00, for dose levels of 1, 10, and 100 mg. 74
5.29 Overlap of high percentile injection nodes for the networks for an injection time
of 12:00 with those for an injection time of 0:00, for dose levels of 0.0001, 0.001,
0.01, and 0.1 mg 75
5.30 Overlap of high percentile injection nodes for the networks for an injection time
of 12:00 with those for an injection time of 0:00, for dose levels of 1, 10, and 100
mg 76
5.31 Overlap of high percentile injection nodes for the networks for an injection time
of 18:00 with those for an injection time of 0:00, for dose levels of 0.0001, 0.001,
0.01, and 0.1 mg 77
-------
xii LIST OF FIGURES
5.32 Overlap of high percentile injection nodes for the networks for an injection time
of 18:00 with those for an injection time of 0:00, for dose levels of 1, 10, and 100
mg 78
5.33 Network 4 showing locations of injection nodes associated with 95th percentile or
higher impacts at a dose level of 0.0001 mg for injection times of 0:00 and 6:00. . 79
5.34 Network 4 showing locations of injection nodes associated with 95th percentile or
higher impacts at a dose level of 0.0001 mg for injection times of 0:00 and 12:00. . 80
5.35 Network 4 showing locations of injection nodes associated with 95th percentile or
higher impacts at a dose level of 0.0001 mg for injection times of 0:00 and 18:00. . 81
5.36 Network 4 showing locations of injection nodes associated with 95th percentile or
higher impacts at a dose level of 1.0 mg for injection times of 0:00 and 6:00. ... 82
5.37 Network 4 showing locations of injection nodes associated with 95th percentile or
higher impacts at a dose level of 1.0 mg for injection times of 0:00 and 12:00. ... 83
5.38 Network 4 showing locations of injection nodes associated with 95th percentile or
higher impacts at a dose level of 1.0 mg for injection times of 0:00 and 18:00. ... 84
5.39 Network 6 showing locations of injection nodes associated with 95th percentile or
higher impacts at a dose level of 0.0001 mg for injection times of 0:00 and 6:00. . 85
5.40 Network 6 showing locations of injection nodes associated with 95th percentile or
higher impacts at a dose level of 0.0001 mg for injection times of 0:00 and 12:00. . 86
5.41 Network 6 showing locations of injection nodes associated with 95th percentile or
higher impacts at a dose level of 0.0001 mg for injection times of 0:00 and 18:00. . 87
5.42 Network 6 showing locations of injection nodes associated with 95th percentile or
higher impacts at a dose level of 1.0 mg for injection times of 0:00 and 6:00. ... 88
5.43 Network 6 showing locations of injection nodes associated with 95th percentile or
higher impacts at a dose level of 1.0 mg for injection times of 0:00 and 12:00. ... 89
5.44 Network 6 showing locations of injection nodes associated with 95th percentile or
higher impacts at a dose level of 1.0 mg for injection times of 0:00 and 18:00. ... 90
6.1 Impacts associated with injection masses of IX and 10X for Network 12 as a
function of dose level, for the 90th, 95th, 99th, and 100th percentile injection nodes. 93
6.2 Impacts associated with injection masses of IX and 10X for Network 12 as a
function of dose level, for the 50th, 60th, 70th, and 80th percentile injection nodes. . 94
6.3 Impact for a 10X injection mass relative to the impact for a IX injection mass, as
a function of dose level, for the 90th, 95th, 99th, and 100th percentile injection nodes. 95
6.4 Impact for a 10X injection mass relative to the impact for a IX injection mass, as
a function of dose level, for the 50th, 60th, 70th, and 80th percentile injection nodes. 96
6.5 Variation in impacts as a function of injection mass for a dose level of 0.0001 mg. 97
6.6 Variation in impacts as a function of injection mass for a dose level of 0.001 mg. . 98
6.7 Variation in impacts as a function of injection mass for a dose level of 0.01 mg. . . 99
6.8 Variation in impacts as a function of injection mass for a dose level of 0.1 mg. . . 100
6.9 Variation in impacts as a function of injection mass for a dose level of 1 mg. ... 101
6.10 Distributions of impacts for Network 12 across all injection nodes for injection
masses of IX and 10X, for dose levels of 0.0001, 0.001, 0.01, and 0.1 mg 102
6.11 Distributions of impacts for Network 12 across all injection nodes for injection
masses of IX and 10X, for dose levels of 1, 10, and 100 mg 103
-------
LIST OF FIGURES xiii
6.12 Impacts for Network 12 for injection masses of IX and 10X, for dose levels of
0.0001, 0.001, 0.01, and 0.1 mg 104
6.13 Impacts for Network 12 for injection masses of IX and 10X, for dose levels of 1,
10, and 100 mg 105
6.14 Impacts for Network 12 that are at or above the 80th percentile level for IX and/or
1 OX injection masses, for dose levels of 0.0001,0.001,0.01, and 0.1 mg 106
6.15 Impacts for Network 12 that are at or above the 80th percentile level for IX and/or
1 OX injection masses, for dose levels of 1, 10, and 100 mg 107
6.16 Overlap of high percentile injection nodes for Network 12 for injection masses of
IX and 10X for different dose levels 108
6.17 Overlap of high percentile injection nodes for the networks for injection masses
of IX and 1 OX, for dose levels of 0.0001,0.001,0.01, and 0.1 mg 109
6.18 Overlap of high percentile injection nodes for the networks for injection masses
of IX and 1 OX, for dose levels of 1, 10, and 100 mg 110
6.19 Network 4 showing locations of injection nodes associated with 95th percentile or
higher impacts at a dose level of 0.0001 mg for injection masses of IX and 10X. .111
6.20 Network 4 showing locations of injection nodes associated with 95th percentile or
higher impacts at a dose level of 1.0 mg for injection masses of IX and 10X. ... 112
6.21 Network 6 showing locations of injection nodes associated with 95th percentile or
higher impacts at a dose level of 0.0001 mg for injection masses of IX and 10X. .113
6.22 Network 6 showing locations of injection nodes associated with 95th percentile or
higher impacts at a dose level of 1.0 mg for injection masses of IX and 10X. ... 114
7.1 Impacts associated with different population models for Network 12 as a function
of dose level, for the 90th, 95th, 99th, and 100th percentile injection nodes 117
7.2 Impacts associated with different population models for Network 12 as a function
of dose level, for the 50th, 60th, 70th, and 80th percentile injection nodes 118
7.3 Relative magnitudes of the impacts for the networks obtained with different pop-
ulation models as a function of dose level, for the 90th, 95th, 99th, and 100th per-
centile injection nodes 119
7.4 Relative magnitudes of the impacts for the networks obtained with different popu-
lation models as a function of dose level, for the 50th, 60th, 70th, and 80th percentile
injection nodes 120
7.5 Distributions of impacts for Network 12 across all injection nodes for different
population models, for dose levels of 0.0001,0.001,0.01, and 0.1 mg 121
7.6 Distributions of impacts for Network 12 across all injection nodes for different
population models, for dose levels of 1, 10, and 100 mg 122
7.7 Impacts for Network 12 obtained with different population models, for dose levels
of 0.0001, 0.001, 0.01, and 0.1 mg 123
7.8 Impacts for Network 12 obtained with different population models, for dose levels
of 1, 10, and 100 mg 124
7.9 Impacts for Network 12 that are at or above the 80th percentile level obtained with
different population models, for dose levels of 0.0001, 0.001, 0.01, and 0.1 mg. . . 125
7.10 Impacts for Network 12 that are at or above the 80th percentile level obtained with
different population models, for dose levels of 1, 10, and 100 mg 126
-------
xiv LIST OF FIGURES
7.11 Overlap of high percentile injection nodes for Network 12 for the two population
models for different dose levels 127
7.12 Overlap of high percentile injection nodes for the networks for the two population
models, for dose levels of 0.0001,0.001,0.01, and 0.1 mg 128
7.13 Overlap of high percentile injection nodes for the networks for the two population
models, for dose levels of 1, 10, and 100 mg 129
7.14 Network 4 showing locations of injection nodes associated with 95th percentile or
higher impacts at a dose level of 0.0001 mg for the two population models 130
7.15 Network 4 showing locations of injection nodes associated with 95th percentile or
higher impacts at a dose level of 1.0 mg for the two population models 131
7.16 Network 6 showing locations of injection nodes associated with 95th percentile or
higher impacts at a dose level of 0.0001 mg for the two population models 132
7.17 Network 6 showing locations of injection nodes associated with 95th percentile or
higher impacts at a dose level of 1.0 mg for the two population models 133
8.1 Impacts associated with different ingestion models for Network 12 as a function
of dose level, for the 90th, 95th, 99th, and 100th percentile injection nodes 137
8.2 Impacts associated with different ingestion models for Network 12 as a function
of dose level, for 50th, 60th, 70th, and 80th percentile injection nodes 138
8.3 Relative magnitudes of the impacts for the networks obtained with different inges-
tion models as a function of dose level, for the 90th, 95th, 99th, and 100th percentile
injection nodes 139
8.4 Relative magnitudes of the impacts for the networks obtained with different in-
gestion models as a function of dose level, for 50th, 60th, 70th, and 80th percentile
injection nodes 140
8.5 Distributions of impacts for Network 12 across all injection nodes for different
ingestion models, for dose levels of 0.0001,0.001,0.01, and 0.1 mg 141
8.6 Distributions of impacts for Network 12 across all injection nodes for different
ingestion models, for dose levels of 1, 10, and 100 mg 142
8.7 Impacts for Network 12 obtained with different ingestion models, for dose levels
of 0.0001, 0.001, 0.01, and 0.1 mg 143
8.8 Impacts for Network 12 obtained with different ingestion models, for dose levels
of 1, 10, and 100 mg 144
8.9 Impacts for Network 12 that are at or above the 80th percentile level obtained with
different ingestion models, for dose levels of 0.0001, 0.001, 0.01, and 0.1 mg. . . . 145
8.10 Impacts for Network 12 that are at or above the 80th percentile level obtained with
different ingestion models, for dose levels of 1, 10, and 100 mg 146
8.11 Overlap of high percentile injection nodes for Network 12 for different ingestion
models for different dose levels 147
8.12 Overlap of high percentile injection nodes for the networks for the two ingestion
models, for dose levels of 0.0001,0.001,0.01, and 0.1 mg 148
8.13 Overlap of high percentile injection nodes for the networks for the two ingestion
models, for dose levels of 1, 10, and 100 mg 149
8.14 Network 4 showing locations of injection nodes associated with 95th percentile or
higher impacts at a dose level of 0.0001 mg for the two ingestion models 150
-------
LIST OF FIGURES xv
8.15 Network 4 showing locations of injection nodes associated with 95th percentile or
higher impacts at a dose level of 1.0 mg for the two ingestion models 151
8.16 Network 6 showing locations of injection nodes associated with 95th percentile or
higher impacts at a dose level of 0.0001 mg for the two ingestion models 152
8.17 Network 6 showing locations of injection nodes associated with 95th percentile or
higher impacts at a dose level of 1.0 mg for the two ingestion models 153
9.1 Histograms of impacts for the 32 combinations of factors for Network 2 for a dose
level of 0.0001 mg and different impact levels 157
9.2 Histograms of impacts for the 32 combinations of factors for Network 2 for a dose
level of 0.01 mg and different impact levels 158
9.3 Histograms of impacts for the 32 combinations of factors for Network 2 for a dose
level of 1 mg and different impact levels 159
9.4 Histograms of impacts for the 32 combinations of factors for Network 2 for a dose
level of 100 mg and different impact levels 160
9.5 Histograms of impacts for the 32 combinations of factors for Network 2 for a
dose level of 0.0001 mg and different impact levels with cases color coded to
show associated injection time 161
9.6 Histograms of impacts for the 32 combinations of factors for Network 2 for a
dose level of 0.01 mg and different impact levels with cases color coded to show
associated injection time 162
9.7 Histograms of impacts for the 32 combinations of factors for Network 2 for a dose
level of 1 mg and different impact levels with cases color coded to show associated
injection time 163
9.8 Histograms of impacts for the 32 combinations of factors for Network 2 for a
dose level of 100 mg and different impact levels with cases color coded to show
associated injection time 164
9.9 Histograms of impacts for the 32 combinations of factors for Network 2 for a
dose level of 0.0001 mg and different impact levels with cases color coded to
show associated injection duration 165
9.10 Histograms of impacts for the 32 combinations of factors for Network 2 for a
dose level of 0.01 mg and different impact levels with cases color coded to show
associated injection duration 166
9.11 Histograms of impacts for the 32 combinations of factors for Network 2 for a dose
level of 1 mg and different impact levels with cases color coded to show associated
injection duration 167
9.12 Histograms of impacts for the 32 combinations of factors for Network 2 for a
dose level of 100 mg and different impact levels with cases color coded to show
associated injection duration 168
9.13 Histograms of impacts for the 32 combinations of factors for Network 2 for a
dose level of 0.0001 mg and different impact levels with cases color coded to
show associated injection mass 169
9.14 Histograms of impacts for the 32 combinations of factors for Network 2 for a
dose level of 0.01 mg and different impact levels with cases color coded to show
associated injection mass 170
-------
xvi LIST OF FIGURES
9.15 Histograms of impacts for the 32 combinations of factors for Network 2 for a dose
level of 1 mg and different impact levels with cases color coded to show associated
injection mass 171
9.16 Histograms of impacts for the 32 combinations of factors for Network 2 for a
dose level of 100 mg and different impact levels with cases color coded to show
associated injection mass 172
9.17 Histograms of impacts for the 32 combinations of factors for Network 2 for a
dose level of 0.0001 mg and different impact levels with cases color coded to
show which population model was used 173
9.18 Histograms of impacts for the 32 combinations of factors for Network 2 for a dose
level of 0.01 mg and different impact levels with cases color coded to show which
population model was used 174
9.19 Histograms of impacts for the 32 combinations of factors for Network 2 for a dose
level of 1 mg and different impact levels with cases color coded to show which
population model was used 175
9.20 Histograms of impacts for the 32 combinations of factors for Network 2 for a dose
level of 100 mg and different impact levels with cases color coded to show which
population model was used 176
9.21 Overlap of injection nodes for the 32 combinations of factors for Network 2 for
different dose levels for 75th percentile and higher impacts 177
9.22 Overlap of injection nodes for the 32 combinations of factors for Network 2 for
different dose levels for 95th percentile and higher impacts 178
9.23 Overlap of injection nodes for the 32 combinations of factors for Network 2 for
different dose levels for 99th percentile and higher impacts 179
10.1 Summary of the ratio of 95th percentile impacts associated with 1- and 24-h injec-
tions for the networks for two dose levels 182
10.2 Summary of the maximum ratio of 95th percentile impacts associated with differ-
ent injection times (0:00, 6:00, 12:00, and 18:00) for the networks for two dose
levels 183
10.3 Summary of the ratio of 95th percentile impacts associated with different injection
masses (IX and 1 OX) for the networks for two dose levels 184
10.4 Summary of the ratio of 95th percentile impacts associated with the two population
models for the networks for two dose levels 185
10.5 Summary of the ratio of 95th percentile impacts associated with the two ingestion
models for the networks for two dose levels 186
10.6 Summary of the overlap of 95th percentile and higher injection nodes associated
with 1-h and 24-h injections for the networks 187
10.7 Summary of the overlap of 95th percentile and higher injection nodes associated
with injection times of 0:00 and 6:00 for the networks 188
10.8 Summary of the overlap of 95th percentile and higher injection nodes associated
with injection times of 0:00 and 12:00 for the networks 189
10.9 Summary of the overlap of 95th percentile and higher injection nodes associated
with injection times of 0:00 and 18:00 for the networks 190
-------
LIST OF FIGURES xvii
10.10 Summary of the overlap of 95th percentile and higher injection nodes associated
with IX and 1 OX injection masses for the networks 191
10.11 Summary of the overlap of 95th percentile and higher injection nodes for the net-
works for the two population models 192
10.12 Summary of the overlap of 95th percentile and higher injection nodes for the net-
works for the two ingestion models 193
11.1 Network 4 showing locations of injection nodes associated with 95th percentile or
higher impacts at dose levels of 0.0001 mg and 0.001 mg 200
11.2 Network 4 showing locations of injection nodes associated with 95th percentile or
higher impacts at dose levels of 0.0001 mg and 0.01 mg 201
11.3 Network 4 showing locations of injection nodes associated with 95th percentile or
higher impacts at dose levels of 0.0001 mg and 0.1 mg 202
11.4 Network 4 showing locations of injection nodes associated with 95th percentile or
higher impacts at dose levels of 0.0001 mg and 1 mg 203
-------
xviii LIST OF FIGURES
-------
List of Tables
2.1 System Descriptions 4
2.2 Ingestion Models 9
2.3 Factors Considered in the Sensitivity Analysis 9
xix
-------
xx LIST OF TABLES
-------
Acronyms and Abbreviations
Acronym Definition
CWS Contamination Warning System
EPA Environmental Protection Agency
EPANET name of software
NZD non-zero demand (applied to nodes)
QA quality assurance
QC quality control
TEVA-SPOT Threat Ensemble Vulnerability Assessment and Sensor Placement
Optimization Toolkit (name of software)
xxi
-------
xxii Acronyms and Abbreviations
-------
Acknowledgments
Authors
The authors of this report are:
Michael J. Davis
Environmental Science Division
Argonne National Laboratory
Argonne, Illinois
Robert Janke
National Homeland Security Research Center
U.S. Environmental Protection Agency
Cincinnati, Ohio
Thomas N. Taxon
Decision and Information Sciences Division
Argonne National Laboratory
Argonne, Illinois
Reviewers
The following individuals reviewed a draft version of this report:
Argonne National Laboratory, Environmental Science Division, Argonne, Illinois
Kurt Picel
U.S. Army Corps of Engineers, Engineering Research Development Center, Construction Engi-
neering Research Laboratory, Champaign, Illinois
Jennie Feliciano
Mark Ginsberg
Kathryn Guy
Page Martin
XXlll
-------
xxiv Acknowledgments
U.S. Environmental Protection Agency, National Homeland Security Research Center, Cincinnati,
Ohio
Terra Haxton
Steve Klosterman
Matthew Magnuson
Jeff Szabo
University of Cincinnati, Department of Civil and Environmental Engineering, Ohio
Ernesto Arandia
-------
Summary
This study examines the adverse effects of contamination events in water distribution systems using
models for 12 actual systems that serve populations ranging from about 104 to over 106 persons. It
extends previous work (Davis and Janke 2010) and provides an improved understanding of the na-
ture of the adverse effects that could be associated with contamination events. The results presented
support water utilities, their consultants, and researchers in conducting contaminant vulnerability
analyses and designing and implementing contamination warning systems.
The measure of adverse effects that we use is the number of people who are exposed to a
contaminant above some dose level due to ingestion of contaminated tap water. For this study the
number of people who receive a dose above a particular level defines the impact associated with
an event. Impact, therefore, refers to the number of people exposed above some level.
We consider a wide range of dose levels in order to accommodate a wide range of potential
contaminants. For a particular contaminant, dose level can be related to a health effect level. For
example, a dose level could correspond to the median lethal dose, i.e., the dose that would be fatal
to 50% of the exposed population. The dose level required to reach a common endpoint can vary
by orders of magnitude, depending on the toxicity of the contaminant. Highly toxic contaminants
may be associated with a particular response at a very low dose level, whereas contaminants with
low toxicity may only be associated with the same response at a much higher dose level.
This report examines how impacts depend on five factors that either define the nature of a con-
tamination event or involve assumptions that are used in assessing exposure to the contaminant:
(1) duration of contaminant injection, (2) time of contaminant injection, (3) quantity or mass of
contaminant injected, (4) population distribution in the water distribution system, and (5) the in-
gestion pattern of the potentially exposed population. For each of these factors, the sensitivities of
impacts to injection location and contaminant toxicity are also examined.
The sensitivity of impacts to the various factors studied is determined by comparing the impacts
associated with different cases of a factor, e.g., 1-h versus 24-h injection durations. Impacts are
estimated for injections at all non-zero demand nodes, the pipe junctions where water is consumed.
For the networks considered in this report, the comparisons involve simulation of injections at
thousands of nodes. (Generally, the demand nodes examined in this report represent groups of
service connections.) In order to facilitate comparisons, we identify locations of contaminant
injection (network model nodes) in terms of the ranking of the associated impact. The nth percentile
injection node is the node associated with the nth percentile impact. We examine two types of
sensitivities to the various factors: sensitivity that results in variations in the magnitude of the nth
percentile impact and sensitivity that results in changes in the injection locations that are associated
with the nth percentile and higher impacts.
Impacts are sensitive to all the factors examined. The degree of sensitivity is dependent on the
particular water system, the location of contaminant injection, and the dose level. Sensitivity of
XXV
-------
xxvi Summary
impacts to all the factors considered tends to increase with dose level (i.e., decreasing toxicity) of
the contaminant, with considerable inter-network variability. With the exception of the population
distribution (factor 4 above), sensitivity to the various factors tends to be highest at lower impact
levels (e.g., impacts below the 80th percentile). Conversely, for the population distribution factor,
sensitivity is lowest at the lower impact levels. For injection duration, impacts generally are higher
for longer duration injections. Definite patterns are present in the sensitivity of impacts to injection
time, but these vary substantially across the networks. As would be expected, impacts are larger
for larger mass injections, but the sensitivity can vary dramatically depending on dose level and the
network. Estimated impacts can be sensitive to assumptions about how population is distributed
in a network, particularly at high impact levels and high dose levels, again with considerable vari-
ability across networks. Finally, impacts can be sensitive to assumptions about ingestion patterns
in the potentially exposed population, with sensitivities varying across networks and tending to be
highest for high dose levels.
When considered in combination with the other factors (but not including the ingestion model
used), impacts at low dose levels (levels at which the effects of highly toxic contaminants can be
significant) are most sensitive to injection duration. Similarly, when considered in combination,
impacts at higher dose levels (levels required for significant effects from contaminants with low
toxicity) are most sensitive to injection mass. At low dose levels, for a likely range in injection
masses, impacts are not particularly sensitive to injection mass.
The influence of the various factors on the location of high percentile injection locations can be
as important or more important than their influence on the magnitudes of impacts. In addition, the
choice of contaminant has a major influence on which nodes are high impact injection locations.
The sharing (overlap) of the same high-percentile injection nodes for different values of a factor
can vary substantially by contaminant and impact level (percentile of impact). Overlap tends to
decrease with decreasing toxicity of the contaminant and increasing impact level for all the factors
considered, with considerable variability among the networks.
Our results demonstrate for a wide range of water systems the great variability in (1) the impacts
that can result from a contamination event and (2) the sensitivity of impacts to the various factors
examined. Although definite patterns exist in the nature and magnitude of impacts for the diverse
set of water systems examined, these results also show that substantial inter-network variability
limits the ability to predict or extrapolate these results to other systems. Therefore, although water
systems do exhibit some similarities in the magnitude and pattern of impacts during contamination
events, each individual water system should be treated as a unique entity when determining its
vulnerabilities to such events.
-------
Section 1
Introduction
Contamination warning systems (CWSs) for water distribution systems need to be designed with a
clear understanding of two issues. First, how are the adverse effects associated with contamination
events influenced by the major factors that define the contamination event? Second, how might
these effects vary for a wide range of distribution systems and contaminants? These issues are
addressed in this report.
Various studies have examined potential adverse effects associated with contamination events,
generally for small water distribution systems (Khanal 2005; Nilsson et al. 2005; Khanal et al.
2006; Davis and Janke 2008). Using a simple network with less than 90 nodes (pipe junctions),
Khanal (2005) and Khanal et al. (2006) examined the sensitivity to selected network variables of
the potential for exposure to contaminants. They determined a potential for exposure but consid-
ered only the presence of a contaminant at network nodes, not the estimated actual exposures of
consumers of contaminated water. No study has examined the sensitivity of potential exposures
to contaminants to major factors used in the simulation considering the actual exposure process
and systems of significant size (i.e., thousands of nodes). Separately, we have analyzed potential
exposures associated with contamination events using 12 models of actual water systems serving
populations ranging from about 104 to over 106 persons, while considering the exposure process
(Davis and Janke 2010). Using these same models and extending that work, this report examines
the sensitivity of potential exposures to (1) a number of major factors that define the nature of a
contamination event and (2) assumptions that are used in assessing exposure to the contaminant.
The results presented here should be useful to water utilities, their consultants, and researchers
interested in conducting contaminant vulnerability analyses and designing and implementating
CWSs in water distribution systems.
1.1 Definitions
The following terms are used throughout this report with the meanings given here.
Dose. A dose is the quantity of a contaminant ingested by consumption of tap water. Inhalation
and dermal contact may also be exposure pathways, but in this study we restrict our analysis to
ingestion. Ingestion doses usually are expressed as a mass of contaminant per unit of body mass
(i.e., in units of mg/kg). In this report doses are expressed as contaminant mass ingested (mg).
Assuming a typical body mass of 70 kg, doses in units of mg/kg can be obtained from the doses
reported here by dividing results by 70 kg.
-------
2 Introduction
Dose Level. A dose level is simply some level of dose. For a particular contaminant, it can
be related to a health effect level. For example, a dose level could correspond to the median lethal
dose. The dose level required to reach a common endpoint can vary by orders of magnitude,
depending on the contaminant. Therefore, we consider a wide range of dose levels in order to
accommodate the potentially wide variability in responses to different contaminants.
Impact. Impact is the size of the population exposed to a contaminant above some dose level
by ingestion of contaminated tap water. It is the measure of adverse effects associated with a
contamination event that we use in this report. Impact could be defined as the number of fatalities
or illnesses that result from exposure to contaminated water. However, to reduce the complexity of
the analysis, the focus is on contaminant doses and the major exposure pathway.
1.2 Navigating This Report
The methodology used in this report is presented in Section 2. Section 3 discusses the nature of
the impacts considered. Results are presented in Sections 4 through 9 and discussed in Sections
10 and 11. Major conclusions are summarized in Section 12. Finally, the appendix describes the
approach to quality assurance used in this study.
The electronic version of this report contains various links that can be used to move from one
location to another. All references to a figure, table, or document have links to the actual figure,
table, or document description. All entries in the Table of Contents and lists of figures and tables
contain links. In Acrobat Reader®, using the "Previous View" button will return the user to the
original reference. If the button is not installed, it can be installed by going to Tools on the Reader
menu bar and then going to Customize Toolbars.
-------
Section 2
Methodology
We simulated impacts to consumers of contaminated tap water during a contamination event for
a number of water distribution systems and examined the sensitivity of the estimated impacts to
major factors characterizing the injection of contaminant, the distribution of population in the sys-
tem, and the ingestion of contaminated water. By simulating diverse distribution systems, it is
possible to determine how sensitivities vary across systems and identify the factors for which esti-
mated impacts are sensitive to the network used and those for which the sensitivitiess are generally
independent of the network.
2.1 Water Distribution Systems Examined
The systems and associated network models used are described in Table 2.1. The systems modeled
are all real systems, but not all of the network models used represent complete systems. (The
networks in some cases may represent incomplete systems but are always complete with respect
to the area being represented.) For security reasons, we do not identify the systems or provide
their general location. However, some of the network models have been used previously. Watson
et al. (2009) used our Networks 1 and 4 (their Networks 1 and 2) and Ostfeld et al. (2008) used
our Network 6 (their Network 2) for sensor monitoring research and evaluation. Isovitsch and
VanBriesen (2008) used our Network 6 in a study of optimal sensor placement. Murray et al.
(2009) examined the influence of district metering areas on a number of factors using case studies,
including two networks that are the same as our Networks 4 and 6.
The populations served by the systems range in size from less than 104 to over 106 and the pop-
ulation density varies from about 102 to over 104 persons/km2. Figure 2.1 shows the distribution of
population at network nodes that have non-zero demands (i.e., nodes where water is consumed) for
nodes with up to 250 people per node, assuming that population is proportional to nodal demand.
(Generally, the nodes examined in this report represent groups of service connections.) Maximum
nodal populations are given in Table 2.1, again assuming population is proportional to demand.
Some portion of the differences in population distribution is due to differences in the degree of
skeletonization of the networks. Four of the network models (Networks 2, 3, 6, and 10) are con-
sidered 'all pipe' models because of the pipe detail per service connection included in the models.
These networks have the smallest mean populations per non-zero demand (NZD) node (Table 2.1).
Average daily water demand for the networks ranges from about 60 to over 9000 L/s (1-200 mil-
lion gal/d). Analysis with EPANET (Rossman 2000), software for simulating the hydraulic and
3
-------
Methodology
Table 2.1. System Descriptionst
Pop.
Network (103)
1
2
3
4
5
6
7
8
9
10
11
12
6
75
130
150
190
250
590
790
840
1,500
1,500
1,800
Mean pop. Mean
Area density water
(km2) (no./km2) age (h)
40
78
130
500
260
490
21
2,700
1,600
1,200
1,400
3,700
160
1,000
1,000
310
740
520
28,000
290
520
1,200
1,100
480
68
17
45
50
24
26
7
30
38
160
110
79
Non-zero demand nodes
No. of
nodes
410
3,200
6,800
3,400
3,000
13,000
4,400
7,400
8,100
43,000
14,000
3,100
No.
110
3,000
6,700
1,600
1,800
11,000
2,200
5,900
6,800
28,000
8,700
1,400
Mean Median
pop. pop.
61
25
20
94
100
24
260
130
120
51
170
1,200
15
17
13
76
58
16
110
83
75
12
53
790
Max.
pop.
1,400
1,000
4,300
1,400
9,000
3,200
18,000
2,000
60,000
73,000
110,000
16,000
The systems are listed by population (pop.) size. All numbers (no.) are rounded independently to two significant
figures.
water quality behavior of networks of pressurized pipes, yields mean water ages (i.e., residence
times) ranging from 7 to 160 h. Water age was simulated separately for each system using a 500-h
simulation duration. Mean water age for each system was determined by averaging over the last
24 h of the 500-h extended period simulation for each node and then averaging again across all
nodes.
We consider a diverse set of distribution systems, as summarized in Table 2.1. In addition,
some of the systems are supplied primarily by surface water, while a few depend almost entirely on
ground-water sources. Tanks dominate some systems as a result of significant elevation obstacles.
Other systems are in areas with little elevation change. For the systems relying on ground water, the
supply wells are widely distributed, potentially making these systems less vulnerable to widespread
contamination due to the continuous, spatially diverse supply of new water being introduced into
the distribution system. The spatially diverse water sources result in areas of the network being
somewhat isolated hydraulically from one another.
Maps of the network models for Networks 4 and 6 are shown in Figures 2.2 and 2.3, respec-
tively. The figures show distorted versions of the actual networks. For security reasons we do not
provide maps for any additional networks.
The systems used in this study were selected because of the availability of complete network
models and a desire to include a wide range of system sizes. The systems are not a random sample
of water distribution systems. However, a broad range of networks is included and therefore our
conclusions should be relevant to a wide range of applications.
-------
Methodology
o.o
50
100 150
Nodal Population
200
250
Figure 2.1. Distribution of nodal populations for the networks below 250 persons per node.
-------
Methodology
B
*
.!
™™
\
WTP Water Treatment Plant
§ Tank
(8) Valve
Pipe Diameter, in
4.0-8.0
8.1 -16.0
16.1 -24.0
48.1 -99.0
Figure 2.2. Network 4.
-------
Methodology
•7
Water Treatment Plant
Production Well
Reservoirs/Tanks
Pipe Diameter, in
0.8-5.0
5.1 -8.0
8.1 - 14.0
14.1 -24.
24.1 -48.0
Figure 2.3. Network 6.
-------
8 Methodology
2.2 Analysis
To examine the sensitivity of impacts to various factors, we defined a number of cases involving
these factors. For each network and case considered, we simulated the following: (1) the inde-
pendent injection of a contaminant at each of the NZD nodes in the network, (2) transport of the
contaminant in the network following injection, and (3) exposure to the contaminant by ingestion
of tap water from the network. As an example, for a particular case involving Network 11 we did
8,700 independent simulations, one for each NZD node. Each of these simulations represents one
contamination event for the network, resulting from an injection at a particular node.
NZD nodes generally represent a number of service connections. We used them in our analysis
because they represent locations where injections are most likely to occur. Zero demand nodes
could also be potential injection locations. However, such locations are more likely to be inacces-
sible than NZD nodes because they often represent pipe connections within water facilities, e.g.,
treatment plants, pump stations, reservoirs, and tanks. Zero demand nodes could also, however,
represent fire hydrants. Although fire hydrants are likely accessible, they also represent locations
that are not easily concealed.
The factors that define a contaminant injection are the time of day that the injection begins, the
duration of the injection, and the mass of contaminant injected. Our baseline case is an injection
beginning at 0:00 hours local time and lasting for one hour. We use a realistic injection mass that
results in significant contaminant concentrations in the networks but we do not specify the mass in
order to prevent the results in this report from being used for design purposes. In our sensitivity
analysis we also consider injections with ten times the mass used for the baseline case, injections
that begin at 06:00, 12:00, and 18:00 hrs, and injections that last for 24 h.
We simulated contaminant transport in the water distribution systems with TEVA-SPOT sensor
placement optimization software (U. S. EPA 2010a), which uses EPANET (version 2.00.12) as its
primary simulation engine. Time steps for hydraulics were 1 h or less; time steps for water quality
were 1 min. Water quality tolerances were specified in the models obtained and were generally
the EPANET default of 0.01 mg/L. Contaminants were assumed to behave as conservative tracers
following injection (i.e., they were not retarded or degraded in the system).
There are numerous parameters that describe a distribution system that must be specified for
EPANET. We used the paramenters in the network models supplied by various utilities. However,
although populations served by the systems are known, their distribution to NZD nodes is not.
Therefore, to account for the sensitivity of results to how population is distributed in a network,
we considered (1) population distributed in proportion to nodal demand and (2) equal population
at all nodes. The total population for a system is the same for both cases. The baseline case is
demand-based population.
We estimated ingestion doses for consumers of tap water using two different models. Ingestion
of tapwater is likely to be the major exposure pathway for most contaminants. Ingestion dose
depends on contaminant concentration in the water ingested and the volume of water ingested.
Because contaminant concentration in tap water varies with time during a contamination event,
ingestion times are needed in order to determine contaminant concentrations in ingested water. The
ingestion models used are summarized in Table 2.2 and have been discussed previously (Davis and
Janke 2008). Both of the ingestion models have a component that addresses ingestion timing and a
component that addresses the volume of tap water consumed. Model 1 is our baseline case. It uses
a probabilistic approach to determining both ingestion times and volume. Model 2 is a completely
-------
Methodology
Table 2.2. Ingestion Models
Model
Timing Component
Volume Component
1
(probabilistic)
2
(deterministic)
Ingestion based, probabilistic (5
events per day), based on time-use
studies
Demand based, every hour (24
events per day), used in some
studies
Variable daily volume (probabilistic),
based on U. S. EPA (2000a)
1 L per day, based on mean volume
of tap water ingested per day
Table 2.3. Factors Considered in the Sensitivity Analysis
Factor
Injection duration
Injection time
Injection mass
Population distribution
Cases Considered
1 and 24 h
0:00, 6:00, 12:00, and 18:00
Realistic mass (IX) and
ten times larger mass (10X)
Demand-based at each node and
Baseline Case
Ih
0:00
IX
Demand-based
at each node
Ingestion model
average value at each node
Models 1 and 2
Model 1
deterministic model that uses an average ingestion volume but assumes ingestion throughout the
day in amounts proportional to nodal demand.
Repeated simulations with Model 1 produce different results. It would be preferable to carry
out multiple simultations and obtain mean results. However, use of such an approach has excessive
computational requirements. Variability in impacts for different simulations with the model is rel-
atively small; the relative standard error in impacts obtained for multiple simulations for Network
4 is about 2% or less (Davis and Janke 2010). Consequently, a single simulation with the model
provides results that are acceptable for this study.
We do not include children in our analysis. Also, we examine only the sensitivity of results to
various factors that do not include the ages of exposed individuals.
Table 2.3 summarizes the factors examined in the sensitivity analysis and the variations con-
sidered in the factors. We also considered combinations of factors to examine interactions. The
first case shown for each factor in Table 2.3 is the baseline case. Although estimated impacts may
be sensitive to uncertainties in the network models used, we did not examine sensitivity to any
parameters or details of the network models.
In our analysis we compare impacts that result from the injection of a contaminant at a partic-
ular node for various cases to determine the sensitivity to different factors, such as injection time.
To allow such comparisons to be made, we need a method to identify nodes by the ranking of the
associated impact, in addition to simply using the name of a particular node. Therefore, we define
-------
10 Methodology
the nth percentile injection node as the injection node associated with the n^ percentile impact.
Nodes corresponding to a particular percentile impact generally vary with the dose level used to
define the impact. Therefore, when identifying a node, we specify both its rank and the dose level
used to define the impact. For example, we might refer to a node as the 90th percentile node for a
dose level of 10 mg.
As indicated in the previous paragraph, when factors change, the impact associated with an
injection at a particular node can change. These changes can influence the importance of the
impact of an injection at a particular node relative to impacts of injections at all other NZD nodes
for the network. To quantify the consistency of the importance of nodes as injection sites, we define
an overlap between injection nodes that are associated with impacts at or above some percentile for
one case of a factor (Case 1) and the injection nodes associated with impacts at or above the same
percentile for another case of a factor (Case 2). Overlap is simply the fraction of nodes possibly
having a particular percentile rank or higher rank that actually have that rank or a higher rank for
both cases. Our interest is in the overlap of high-percentile injection nodes for different values of
some factor.
The concept of overlap of high-percentile injection nodes is illustrated in Figure 2.4, which
provides scatter plots comparing impacts for two hypothetical cases of some factor for a network
with 20 nodes. The plots show examples of overlaps equal to 1, 0.5, and 0 for 80th percentile and
higher impacts for some dose level. Each of the 20 points in the plots corresponds to an injection
node. The red dashed lines give the 80th percentile impacts. For each case there are four nodes
that are associated with impacts at or above the 80th percentile. For an overlap of 1, the same four
nodes are associated with 80th percentile or higher impacts for both cases. (Note that although the
percentile is the same for both cases, the associated impact is not the same for the two cases.) For
an overlap of zero, no node is associated with 80th percentile or greater impacts for both cases.
For an overlap of 0.5, two injection nodes are associated with impacts that have a percentile rank
greater than 80 for both cases.
Overlap can be expressed as:
/
overlap =
^
V1 100
where
N = number of injection nodes (i.e., NZD nodes),
P = the percentile of the impact of interest, and
NP = number of injection nodes associated with impacts > Pth percentile impacts, both cases.
When overlap = 1, all injection nodes for Case 1 that are associated with impacts at or above the
given percentile are also associated with impacts at or above that percentile for Case 2. If N = 20
and P = 80%, then NP would be 4 for overlap = 1, as illustrated in Figure 2.4. When overlap
= 0, there are no injection nodes that are associated with impacts at or above the given percentile
for both cases. When a particular percentile impact is zero, all injection nodes are associated with
impacts that are at or above that impact. In such a case the overlap calculated will be 1. Therefore,
when the impact associated with the percentile of interest for one or both of the cases is zero, no
overlap is determined.
The analysis of overlap shows that the locations of high-percentile injection nodes varies when
the factors studied are changed. However, the analysis does not provide any information on
changes in the actual spatial distribution of the nodes as the factors change. To show the ac-
-------
Methodology
11
Overlap = 1
Overlap = 0.5
Overlap = 0
OM
-------
12
Methodology
o
co
o
CM
Outside values
Upper adjacent value
Third quartile
Median
First quartile
Lower adjacent value
O
Violin Plot
Box Plot
Figure 2.5. Comparison of a violin plot and a box plot.
-------
Methodology 13
doing so would double the number of combinations. Using the set of 32 combinations of factors
for one network, we performed one simulation for each combination for each NZD node for the
network. We then examined the patterns in the estimated impacts to determine which factors, when
considered together, have a major influence on the impacts, as will be discussed in Section 9.
Sections 4 through 8 consider the sensitivity of impacts to a number of factors: injection dura-
tion, injection time, injection mass, the distribution of the population in the network, and the model
used to estimate ingestion dose. Each section considers one of the factors at a time. Combinations
of factors are then considered together in Section 9. In the sections considering the individual
factors, results are presented in a series of seven types of figures that are repeated for each factor.
In addition, maps are used to illustrate how impacts vary spatially as the factors are varied. The
various figures and maps are discussed in Section 4 as each one is introduced.
-------
14 Methodology
-------
Section 3
Nature of Impacts
As already noted, in this study impact is defined as the size of the population exposed above some
dose level due to ingestion of contaminated tap water. This section presents some basic information
about impacts.
Figures 3.1, 3.2, and 3.3 show how impacts vary with dose level for the 50th, 75th, 85th, 95th,
99th, and 100th percentile injection nodes for the 12 networks1. (Note the logarithmic scales on the
horizontal and vertical axes.) The behavior of impacts is generally consistent for all the networks.
First, impact decreases with dose level. This decrease occurs because fewer people receive high
doses than low doses (due to the effects of dilution as the contaminant is transported away from
the injection node) and impact is defined as the number of people exposed above a particular level.
Second, for any dose level, impact increases with the increasing percentile of the injection node.
However, for some networks, such as Network 5, the impacts vary much less between the median
and maximum percentile nodes than for others, such as Network 11. Because significant impacts,
those similar in magnitude to worst-case impacts, are associated only with the high percentile
cases, we focus our attention here on injections at high percentile nodes.
The results in Figures 3.1, 3.2, and 3.3 (and other figures) are not specific to any contaminant
but can be applied to specific contaminants. For example, the upper end of the range of dose levels
in Figures 3.1, 3.2, and 3.3 is 100 mg, or about 1 mg/kg for a 70-kg person, a value similar to the
median lethal dose for phorate, a pesticide. The use of a wide range in dose levels allows us to
consider the equivalent of a wide range in potential contaminants.
The shapes of the curves plotted in Figures 3.1, 3.2, and 3.3 are determined by the characteris-
tics of the various network models. They are independent of the magnitude of the injection mass
used to determine the impacts. However, the values of the dose levels on the horizontal axis are
a function of the injection mass used. If the injection mass were 100 times larger than that used
to obtain the results in the figures, the plots would be identical except that all values of dose level
would be multiplied by 100. Therefore, what is called a low dose level (e.g., 0.0001 mg in this
report) or a high dose level (e.g., 100 mg in this report) depends on the injection mass. These labels
are relative ones. A dose level can be related to some level of health effects for a particular contam-
inant. Low dose levels can be related to a particular health effect level for more toxic contaminants
and high dose levels can be related to the same health effect level for less toxic contaminants.
According to the standard definition of percentile, the rf1 percentile impact is the impact below which n percent
of the impacts lie. Therefore, there can be no 100th percentile impact. However, to simplify graphical presentations in
this report, we use 100th percentile to mean maximum value.
15
-------
16 Nature of Impacts
The location on the horizontal axis of a dose level corresponding to some health effect level for a
particular contaminant will depend on the injection mass. For a small injection mass a particular
health effect level for a contaminant could correspond to a dose level on the right side of the plots,
while for a much larger injection mass the same effect level for the contaminant could correspond
to a dose level on the left side. Equivalently, if sufficient mass is injected, a given percentile impact
for a low toxicity contaminant will be as large as that for a high toxicity contaminant and a much
smaller injection mass.
For higher dose levels, impacts may be zero for some cases. In such cases, no points are shown
in Figures 3.1, 3.2, and 3.3 or similar figures. For example, for Network 5 in Figure 3.2, no 50th
percentile impacts are shown for dose levels of 10 and 100 mg because they are zero.
As already noted, the injection node that is the n^ percentile injection node for a particular
dose level generally is not the nth percentile injection node for a different dose level. Figures such
as Figures 3.1, 3.2, and 3.3 that show impacts for a particular impact level as a function of dose
level are showing impacts associated with generally different injection nodes.
The spatial extent of the population of high percentile injection nodes can expand substantially
as the percentile of interest decreases. Figures 3.4, 3.5, and 3.6 illustrate this behavior for Net-
work 4 for a dose level of 0.0001 mg and the 95th, 90th, 85th, and 80th percentiles. The maps in
the figures show the locations of injection nodes associated with impacts at or above the percentile
given.
The areas affected by injection at the nth percentile node for different dose levels can vary sig-
nificantly depending on the dose levels of interest. Figure 3.7 provides an example for Network 6
showing the areas (receptor nodes) affected following injection at the 90th percentile nodes for dose
levels of 0.0001 and 1.0 mg. For the lower dose level the affected area is substantial and occupies
a large portion of the central part of the network. For a dose level of 1.0 mg, the area affected in-
volves only a relatively small number of nodes in the southwestern portion of the network. Figure
3.7 also illustrates that the location of a given percentile node can vary greatly with dose level.
-------
Nature of Impacts
17
io
4-l
10J
CD
t/5
o
Q
s_
CD
.c
O)
±
5
c
o
10-
1 -
Network 1
10~4 10"3 10"2
Q. 108H
o
CO
Q.
10J
,2-
10
10-
1 -
Network 3
10~4 10~3 10"2
1 10 100
10B-
10
5.
10
,4-
3.
10
10"-
10-
1 -
Network 2
10~4 10~3 10~2 10~1 1 10 100
10'
,6-
10
10J
,4-
,2-
10
10-
1 -
Network 4
10 100 10~4
Dose Level (mg)
1 10 100
Figure 3.1. Impacts for Networks 1 to 4 as function of dose level, for various percentile
injection nodes.
-------
18
Nature of Impacts
o>
(/5
O
Q
s_
CD
.c
O)
±
O
"CD
Q.
O
CO
Q.
10D-
104-
10a-
102-
10-
1 -
ioe-l
io5-l
104-l
10a-
102-
10-
1 -
Network 5
Network 7
10 100
10°-
104-
10a-
102-
10-
1 -
ioe-l
io5-l
104-l
10a-
102-
10-
1 -
Network 6
10 100
Network 8
10"4 10~3 10"2
10 100 10~4
Dose Level (mg)
10 100
Figure 3.2. Impacts for Networks 5 to 8 as function of dose level, for various percentile
injection nodes.
-------
Nature of Impacts
19
io
4-l
10J
CD
t/5
o
Q
s_
CD
.c
O)
±
5
c
o
10-
1 -
Network 9
10"4 10"3 10"2
Q. 108H
o
CO
Q.
10J
,2-
10
10-
1 -
Network 11
10~4 10"3 10"2
1 10 100
10B-
10
5.
10
,4-
3.
10
10"-
10-
1 -
Network 10
10"4 10"3 10~2
10'
,6-
10
5.
10
,4-
10
3.
,2-
10
10-
1 -
Network 12
10 100 10~4
Dose Level (mg)
10 100
1 10 100
Figure 3.3. Impacts for Networks 9 to 12 as function of dose level, for various percentile
injection nodes.
-------
20
Nature of Impacts
Injection Nodes
o 0.0001 mg, 90th
• 0.0001 mg, 95th
Figure 3.4. Network 4 showing locations of injection nodes associated with 90th and 95th
percentile or higher impacts at a dose level of 0.0001 mg. Additional features on the map are
identified in Figure 2.2.
-------
Nature of Impacts
21
Injection Nodes
o 0.0001 mg, 85th
• 0.0001 mg, 95th
Figure 3.5. Network 4 showing locations of injection nodes associated with 85th and 95th
percentile or higher impacts at a dose level of 0.0001 mg.
-------
22
Nature of Impacts
Injection Nodes
o 0.0001 mg, 80th
• 0.0001 mg, 95th
Figure 3.6. Network 4 showing locations of injection nodes associated with 80th and 95th
percentile or higher impacts at a dose level of 0.0001 mg.
-------
Nature of Impacts
23
SOURCE SOURCE
G 0.0001 mg • 1.0 mg
Receptors Receptors
• 1-36
o 37-134
• 135-676
o
o
1 -9
10-28
29-81
Figure 3.7. Network 6 showing receptor nodes affected following injections at the 90th per-
centile nodes for dose levels of 0.0001 and 1.0 mg. The number of individuals affected at each
receptor node is indicated. Additional features on the map are identified in Figure 2.3.
-------
24 Nature of Impacts
-------
Section 4
Sensitivity of Impacts to Injection Duration
Sensitivity to the various factors is considered both in terms of the statistics of the impacts across
the ensemble of injection nodes and in terms of impacts associated with individual injection nodes.
Figures 4.1 to 4.6 provide information on how impact statistics are affected by changes in the in-
jection duration (1 h and 24 h). Figures 4.7 to 4.13 allow an evaluation to be made of how impacts
at individual injection nodes are affected. Figures 4.14 to 4.17 provide maps showing how the
locations of high percentile injections nodes are influenced by changes in injection duration. Sen-
sitivity is examined first by using Network 12 as an example and then using all networks together.
This approach provides a detailed understanding of one network and an overall understanding of
variability in sensitivities across all the networks.
Figures 4.1 and 4.2 provide a comparison for Network 12 of how high percentile impacts
associated with 1-h and 24-h injections vary with dose level. Impacts at a given percentile decrease
with increasing dose level at a rate that depends on the level of impact. The rate of decrease also
varies for different networks. The figure shows that impacts associated with 24-h injections are
generally larger at all dose levels and the relative difference tends to increase for lower percentile
impacts. The curves in Figures 4.1 and 4.2 for the 1-h injection are the base case that is used when
similar comparisons are made for the other factors.
Considering all 12 networks, Figures 4.3 and 4.4 provide plots of the relative magnitudes of
the impacts for the 24-h and 1-h injections versus dose level for different percentile impacts. (For
comparison, absolute values for impacts for the networks for 1-h injections are provided in Figures
3.1 to 3.3.) Generally, the impacts for the 24-h injections are larger than those for the 1-h injection
because there is an increased opportunity for exposure. The ratio tends to increase with dose level,
especially as the percentile of the impact decreases. However, there are considerable differences
among the networks. Results are not shown if impacts for either 1-h or 24-h injections are zero.
Returning to Network 12, Figures 4.5 and 4.6 compare the distribution of impacts across all
injection nodes for 1-h and 24-h injections for different dose levels using violin plots (note the
change in vertical scale factor for dose levels greater than 0.1 mg). The distributions of impacts
for the 1-h and 24-h cases are very similar for the lowest dose level (0.0001 mg) but differ greatly
for the higher dose level of 0.1 mg. Median values are similar for the two cases at the lowest dose
levels and at the highest dose levels (where they are both zero), with the relative difference being
largest for intermediate dose levels. The distribution for the 1-h injection is also the base case for
similar comparisons for other factors.
To show how impacts associated with individual injection nodes vary as injection duration
25
-------
26 Sensitivity of Impacts to Injection Duration
changes, Figures 4.7 and 4.8 compare impacts for Network 12 for the two injection durations
using scatter plots. Impacts for the 1-h and the 24-h injections are compared node-by-node for all
1400 injection nodes. Impacts for the 24-h injection are generally larger (most points lie above the
red lines, which have a slope of 1 and represent equal impacts for the two injection times). The
correlation between the two sets of impacts decreases as the dose level increases. Note the change
in scale factors for dose levels greater than 0.1 mg.
Focusing only on higher percentile nodes, the impacts compared in the scatter plots in Figures
4.9 and 4.10 are all at or above the 80th percentile for the 1-h injection (green points), for the
24-h injection (blue points), or both (red points). The red points identify the most critical injection
locations; they are associated with impacts at the 80th percentile and above for either 1-h or 24-h
injections. The dashed lines show the 80th percentile values. If the same injection nodes had
impacts at or above the 80th percentile for both 1-h and 24-h injection durations, then there would
be 280 red points1. If there were no overlap, there would be no red points, but there would be 280
blue points and 280 green points. Overlap decreases as the dose level increases up to a level of
1 mg. At higher dose levels, the 80th percentile impacts are zero for both cases and, therefore, all
nodes are associated with impacts above the 80th percentile (are red). (Actual values of overlap for
the various dose levels can be obtained from Figure 4.11.) Again, note the change in scale factors
for dose levels greater than 0.1 mg.
Figure 4.11 provides a plot for Network 12 of the overlap of high percentile nodes for 1-h
and 24-h injections as a function of impact percentile. For example, for the 80th percentile and a
dose level of 0.1 mg, about 40% of the possible 280 nodes are 80th percentile or higher injection
nodes for both injection durations. (This quantifies the overlap that can be seen in the scatterplot
in Figure 4.9 for a dose level of 0.1 mg.) The overlap of high percentile nodes for the two injection
durations tends to decrease as the percentile increases and as the dose level increases. However,
for the lowest dose levels (0.0001 and 0.001 mg) the overlap is approximately constant until the
percentile is higher than 90%. The overlap of high percentile nodes for high dose levels is low.
In other words, for contaminants with relatively high thresholds for adverse effects, the injection
locations associated with the highest impacts are quite sensitive to injection duration.
Considering all the networks, Figures 4.12 and 4.13 show the overlap of high-percentile nodes
for different dose levels as a function of impact percentile. There is considerable variability by
network, but, as in the case for Network 12, overlap tends to decrease as the percentile increases
and as the dose level increases.
Using the map for Network 4 (see Figure 2.2), Figure 4.14 illustrates the overlap of injection
nodes associated with impacts at or above the 95th percentile for injection durations of 1 h and 24 h
for a dose level of 0.0001 mg. Consistent with the results in Figure 4.12, there is high overlap of
the injection nodes. Figure 4.15 provides a similar comparison for a dose level of 1.0 mg. At this
dose level, little overlap of the injection nodes occurs. Figures 4.16 and 4.17 show comparisons
for Network 6 for the two dose levels. Again there is high overlap of injection nodes at a dose level
of 0.0001 mg and poor overlap at 1.0 mg.
In summary, when compared on the basis of the ranking of the impact (e.g., the 90th percentile),
the impacts associated with 1-h and 24-h injections often are similar, but the degree of similarity
varies by network and tends to decrease at higher dose levels and for lower impact levels (e.g.,
compare the 99th and 50th percentile impacts in Figures 4.3 and 4.4, respectively). When compared
lrThe network has 1400 injection nodes. Twenty per cent of these (0.2x1400 = 280) are at or above the 80th
percentile.
-------
Sensitivity of Impacts to Injection Duration 27
on the basis of individual injection nodes, the impacts associated with 1-h and 24-h injections
can vary substantially (Figures 4.7 to 4.10). The overlap of the high percentile nodes for the two
injection durations can be low for higher dose levels with higher impact.
The series of figures used to examine the sensitivity of impacts to injection duration are re-
peated in the examination of the other factors.
-------
28
Sensitivity of Impacts to Injection Duration
10'
,6-
10
,5-
,4-
10'
103-|
CD
(/>
O
Q
i_
CD
.C
D)
X
10-
10"4 10"3 10~2
O
jo 10b1
Q.
5-1
CO
E
10
1031
10"
percentile = 100
1 10 100
percentile = 95
10'
,6-
10
,5-
,4-
10'
1021
10-
1 -
10"4 10~3 10"2
,4-
10'
1031
102H
10-
1 -
1 h
24 h
10 100 10~4 10
Dose Level (mg)
"3
percentile = 99
0"1 1 10 100
percentile = 90
10 100
Figure 4.1. Impacts associated with 1-h and 24-h injections for Network 12 as a function of
dose level, for the 90th, 95th, 99th, and 100th percentile injection nodes.
-------
Sensitivity of Impacts to Injection Duration
29
10'
,6-
10
,5-
,4-
10'
103-|
CD
(/>
O
Q
i_
CD
.C
D)
X
10-
10"4 10"3 10~2
O
jo 10b1
Q.
5-1
CO
E
10
1031
10"
percentile = 80
1 10 100
percentile = 60
10'
,6-
10
,5-
,4-
10'
1021
10-
1 -
10"4 10~3 10"2
,4-
10'
1031
102H
10-
1 -
10 100 10
Dose Level (mg)
"4 ~3
10~ 10"
percentile = 70
0"1 1 10 100
percentile = 50
10 100
Figure 4.2. Impacts associated with 1-h and 24-h injections for Network 12 as a function of
dose level, for the 50th, 60th, 70th, and 80th percentile injection nodes.
-------
30
Sensitivity of Impacts to Injection Duration
•s
CO
Q.
E
CNJ,
(/5
•s
CD
Q.
E
12-
ID-
S'
6-
4-
2'
0-
12-
10-
8-
6-
4-
2-
0-
10 100
? f 1 Z=ft3=S^
Percentile = 95
10"4 10"3 10"2
12-
ID-
S'
6-
4-
2'
Network
1
2 -
3 -
4
5 -
6 -
9
10
11
12
0- Percentile = 99
10~3 10~:
10 100
12-
10-
8-
6-
4-
2-
0-
Percentile = 90
1 10 100 10~4
Dose Level (mg)
1 10 100
Figure 4.3. Impact for a 24-h injection relative to the impact for a 1-h injection, as a function
of dose level, for the 90th, 95th, 99th, and 100th percentile injection nodes.
-------
Sensitivity of Impacts to Injection Duration
31
CO
Q.
E
CN,
CD
Q.
E
12-
ID-
S'
6-
4-
2-
0-
12-
10-
8-
6-
4-
2-
0-
Percentile = 80
10~4 10"3 10"2 10~1 1
Percentile = 60
10~4 10"3 10"2
10 100
12-
10-
8-
6-
4-
2-
0-
12-
10-
8-
6-
4-
2-
0-
Network
- e
Percentile = 70
10 100
Percentile = 50
10 100 10~4
Dose Level (mg)
1 10 100
Figure 4.4. Impact for a 24-h injection relative to the impact for a 1-h injection, as a function
of dose level, for the 50th, 60th, 70th, and 80th percentile injection nodes.
-------
32
Sensitivity of Impacts to Injection Duration
8J
o
o-
OM
CO
O.
E
o
COH
o
H
O_
dose level = 0.0001 mg
1 h
24 h
dose level = 0.01 mg
1 h
o
31
o
o-
OM
8-
dose level = 0.001 mg
i
1 h
24 h
dose level = 0.1 mg
24 h 1 h
Injection Durations
24 h
Figure 4.5. Distributions of impacts for Network 12 across all injection nodes for 1-h and
24-h injections for dose levels of 0.0001, 0.001, 0.01, and 0.1 mg. The red dot represents the
median value.
-------
Sensitivity of Impacts to Injection Duration
33
8.
in
o
o-
CO
o
o-
OM
dose level = 1.0 mg
CO
Q.
E
00
o'
to
o'
C\|
o'
q
o'
s-
a-
0_
dose level = 10 mg
1 h
24 h
1 h
24 h
dose level = 100 mg
1 h
24 h
Injection Durations
Figure 4.6. Distributions of impacts for Network 12 across all injection nodes for 1-h and
24-h injections for dose levels of 1,10, and 100 mg.
-------
34
Sensitivity of Impacts to Injection Duration
8j
o
o
§1
8j
CD
O
i1
dose level = 0.0001 mg
o
a>
CO
P o
t co
o _
C\l
O _
dose level = 0.01 mg
200 400 600 800
10 15 20 25 30
o
o
CO
o
o
OM
dose level = 0.001 mg
0 50 100 150 200 250
dose level = 0.1 mg
a*o ^°o
o o
Impact, 1-h Injection (thousands)
Figure 4.7. Impacts for Network 12 for 1-h and 24-h injections for dose levels of 0.0001,
0.001, 0.01, and 0.1 mg. The red lines have a slope of 1. Each point corresponds to one
injection node.
-------
Sensitivity of Impacts to Injection Duration
35
o
81
o
81
o
a
o
o -
o
'•s
o>
i
•*
CM
CO
Q. O
E -
00
o
CD
O
CN
ci
q
ci
dose level = 1 mg
dose level = 100 mg
100 200 300 400
o
CO
o _
dose level = 10 mg
10 15 20 25
30
0.0 0.2 0.4 0.6 0.8 1.0
Impact, 1-h Injection
Figure 4.8. Impacts for Network 12 for 1-h and 24-h injections for dose levels of 1,10, and
100 mg. The red lines have a slope of 1. Each point corresponds to one injection node.
-------
36
Sensitivity of Impacts to Injection Duration
•a
to
o
o>
CD
a.
8
o
8
to
o
°
dose level = 0.0001 mg
>80th, 1 h
O > 80th, 24 h
O > 80th, both
200
400
600
800
o
CO
o
OM
o _
dose level = 0.01 mg
: o
0 5 10 15 20 25 30
o
o
CO
o
o
OM
dose level = 0.001 mg
50 100 150 200 250
dose level = 0.1 mg
QU §b O
$0 O
Impact, 1-h Injection (thousands)
Figure 4.9. Impacts for Network 12 that are at or above the 80th percentile level for 1-h
and/or 24-h injections for dose levels of 0.0001, 0.001, 0.01, and 0.1 mg. The dotted lines
indicate 80th percentile impacts. The black lines have a slope of 1. Each point corresponds to
one injection node.
-------
Sensitivity of Impacts to Injection Duration
37
o
s
o
o -
OM
o>
I
•°° o
o
H
o
CO
o _
dose level = 10 mg
100
200
300
400
dose level = 100 mg
10 15 20 25
30
0.0 0.2 0.4 0.6 0.8 1.0
Impact, 1-h Injection
Figure 4.10. Impacts for Network 12 that are at or above the 80th percentile level for 1-h
and/or 24-h injections for dose levels of 1, 10, and 100 mg. The dotted lines indicate 80th
percentile impacts. The black lines have a slope of 1. Each point corresponds to one injection
node.
-------
38
Sensitivity of Impacts to Injection Duration
1.0-
0.8-
o
2
= 0.6-
Q
c
o
O>
Q. 0.4-
03
0.2-
0.0-
I
50
I
60
I
70
Percentile
i
80
I
90
100
Figure 4.11. Overlap of high percentile injection nodes for Network 12 for 1-h and 24-h
injections for different dose levels.
-------
Sensitivity of Impacts to Injection Duration
39
1.0-
0.8-
0.6-
0.4-
.0 0.2-
° o.o H
50 60 70 80 90 100
o>
Q- 1.0-
O 0.8-
0.6-
0.4-
0.2-
0.0-
dose level = 0.01 mg
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
dose level = 0.001 mg
50 60 70 80 90 100
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
dose level = 0.1 mg
50 60 70 80 90 100 50 60
Percentile
70 80 90 100
Figure 4.12. Overlap of high percentile injection nodes for the networks for 1-h and 24-h
injections for dose levels of 0.0001,0.001,0.01, and 0.1 mg.
-------
40
Sensitivity of Impacts to Injection Duration
1.0-
0.8-
0.6-
0.4-
.o 0.2-
0.0-
dose level = 1.0 mg
o
O>
'^
co" 1'°"
O 0.8-
0.6-
0.4-
0.2-
0.0-
dose level = 100 mg
Network
1
2 -
3 -
4
5 -
6 -
50 60 70 80
90
50 60 70 80 90 100
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
dose level = 10 mg
100
Percentile
50 60 70 80 90 100
Figure 4.13. Overlap of high percentile injection nodes for the networks for 1-h and 24-h
injections for dose levels of 1,10, and 100 mg.
-------
Sensitivity of Impacts to Injection Duration
41
Injection Nodes
o 0.0001 mg, 24 h
• 0.0001 mg, 1 h
Figure 4.14. Network 4 showing locations of injection nodes associated with 95th percentile
or higher impacts at a dose level of 0.0001 mg for injection durations of 1 h and 24 h.
-------
42
Sensitivity of Impacts to Injection Duration
o o
Injection Nodes
o 1.0 mg, 24 h
• 1.0 mg, 1 h
Figure 4.15. Network 4 showing locations of injection nodes associated with 95th percentile
or higher impacts at a dose level of 1.0 mg for injection durations of 1 h and 24 h.
-------
Sensitivity of Impacts to Injection Duration
43
-T
Injection Nodes
o 0.0001 mg, 24 h
O 0.0001 mg, 1 h
Figure 4.16. Network 6 showing locations of injection nodes associated with 95th percentile
or higher impacts at a dose level of 0.0001 mg for injection durations of 1 h and 24 h.
-------
44
Sensitivity of Impacts to Injection Duration
pea A
*^f^a *
Injection Nodes
o 1.0 mg, 24 h
• 1.0 mg, 1 h
Figure 4.17. Network 6 showing locations of injection nodes associated with 95th percentile
or higher impacts at a dose level of 1.0 mg for injection durations of 1 h and 24 h.
-------
Section 5
Sensitivity of Impacts to Injection Time
This section addresses the sensitivity of impacts to changes in injection time. The presentation is
generally similar to that used for injection duration in the previous section.
We considered four injection times, 0:00, 6:00, 12:00, and 18:00. Figures 5.1 and 5.2 show how
ranked impacts vary for Network 12 for the different injection times as a function of dose level.
Generally, the impacts associated with injections at 0:00 are the smallest, with the injection time
associated with the largest impact varying somewhat by percentile. At low dose levels, the impacts
vary little with injection time, while at higher dose levels there can be considerable difference in
impacts for different injection times.
Considering all the networks, Figures 5.3 and 5.4 provide ratios of impacts for the injection
time with the largest impact to impacts for the injection time with the smallest impacts, as a func-
tion of dose level. (For comparison, absolute values for impacts for the networks for injections at
0:00 are provided in Figures 3.1 to 3.3.) Results are provided for different percentile impacts. As
is the case with injection duration, the ratio tends to increase with dose level, particularly as the
percentile of the impact decreases. There is considerable variability among the networks and the
variability increases with dose level.
Again, considering all the networks, impacts associated with the various injection times are
shown in Figures 5.5 through 5.9 relative to the impact associated with an injection at 0:00. Rel-
ative impacts are provided for dose levels from 0.0001 mg to 1.0 mg for injection nodes with
percentiles of 100%, 95%, 75%, and 50%. The diurnal pattern in relative impacts varies with net-
work, dose level, and percentile. At the higher dose levels, the larger networks tend to have a larger
range in impacts for different injection times than do the smaller networks. No figures similar to
5.5 through 5.9 are provided for the analysis of the sensitivity of impacts to injection duration,
population model, or ingestion model because in those cases only two values of the factor were
considered. However, similar figures are provided for the analysis involving injection mass.
The distributions of impacts across injection nodes are compared in Figures 5.10 and 5.11 for
different injection times for Network 12. There can be considerable differences in the distributions
for the different injection times, depending on dose level. For example, compare the distributions
for a dose level of 0.1 mg.
The scatter plots in Figures 5.12 through 5.17 compare impacts by injection node for injection
times of 6:00, 12:00, and 18:00 with the impacts for the same nodes for an injection time of
0:00, for Network 12. Impacts associated with injections at 0:00 are generally smaller than those
associated with the other injection times (most points lie above the red line). (However, as Figures
45
-------
46 Sensitivity of Impacts to Injection Time
5.5 through 5.9 show, this behavior is not always true for all networks.) The degree of scatter
increases with dose level, up to a dose level of about 1 mg.
The scatter plots in Figures 5.18 through 5.23 include only those injection nodes for Network
12 for which impacts are at or above the 80th percentile for at least one of the injection times. The
red points in the plots are associated with injection nodes whose impacts are at or above the 80th
percentile for both injection times used in a plot. The overlap of nodes that are at or above the 80th
or other high percentile for both cases decreases considerably as the dose level increases, at least
up to a dose level of about 0.1 mg, as is shown explicitly in Figures 5.24, 5.25, and 5.26. Even
for the lower dose levels, the degree of overlap can decrease considerably for higher percentile
impacts.
Considering all the networks, the overlap of the high percentile injection nodes is shown in the
plots in Figures 5.27 through 5.32 for injection times of 6:00, 12:00, and 18:00, compared with an
injection time of 0:00. The plots show how overlap varies with dose level for different percentile
impacts. The overlap varies considerably among the networks, but tends to decrease with the rank
of the impact and the dose level.
Figures 5.33 through 5.38 provide maps of Network 4 comparing the locations of 95th per-
centile and higher injection nodes for different injection times (0:00, 6:00, 12:00, and 18:00) for
dose levels of 0.0001 mg (Figures 5.33 to 5.35 ) and 1.0 mg (Figures 5.36 to 5.38). Figures 5.39
through 5.44 provide similar maps for Network 6. For both networks, high overlap of injection
nodes occurs for the different injection times for a dose level of 0.0001 mg. For a dose level of
1.0 mg, the overlap of injection nodes for both networks is much lower. These results are consistent
with the overlap plots for the networks given in Figures 5.27 through 5.32.
Overall, the sensitivity of impacts to injection time is similar to the sensitivity to injection
duration. However, the variability in sensitivity of ranked impacts for the different networks is
generally higher for injection times than for injection duration (cf. Figures 4.3 and 5.3).
-------
Sensitivity of Impacts to Injection Time
47
10B-
103-
10
,4-
10
3.
o>
t/5
O
Q
s_
CD
.c
O)
±
10-
1 -
percentile = 100
1 10 100
ro 10'
^
Q.
O
Q.
,6-
CO
Q.
io4-l
10
10
10-
1 -
r,2-
percentile = 95
10~4 10"3 10"2 10~1 1
10B-
103-
10
,4-
10
102
10
3.
1 -
I I
,6-
10'
105
io4-l
10
10
10-
1 -
r.2_
00:00
06:00
12:00
18:00
10 100
Dose Level (mg)
10~4 10~3
percentile = 99
1 10 100
percentile = 90
1 10 100
Figure 5.1. Impacts associated with different injection times for Network 12 as a function of
dose level, for the 90th, 95th, 99th, and 100th percentile injection nodes.
-------
48
Sensitivity of Impacts to Injection Time
10B-
103-
10
,4-
10
3.
o>
t/5
O
Q
s_
CD
.c
O)
±
10-
1 -
percentile = 80
1 10 100
ro 10'
^
Q.
O
Q.
,6-
CO
Q.
io4-l
10
10
10-
1 -
r,2-
percentile = 60
10~4 10"3 10"2
10B-
103-
10
,4-
10
102
10
3.
1 -
I I
,6-
10'
105
io4-l
10
10
10-
1 -
r.2_
10 100 10~4
Dose Level (mg)
percentile = 70
10 100
percentile = 50
10 100
Figure 5.2. Impacts associated with different injection times for Network 12 as a function of
dose level, for the 50th, 60th, 70th, and 80th percentile injection nodes.
-------
Sensitivity of Impacts to Injection Time
49
CD
E
o
o>
CD
I
b
CO
•s
CO
Q.
E
M—
O
o
'-4—'
CD
o:
16-
14-
12-
10-
8-
6-
4-
2-
0-
Percentile = 100
10
,-4
10
,-3
10 100
16-
14-
12-
10-
8-
6-
4-
2-
0-
Percentile = 95
10
-4
10
-3
10"
10"
10 100
16-
14-
12-
10-
8-
6-
4-
2-
0-
Network
Percentile = 99
10
,-4
10
,-3
10"
10"
10 100
16-
14-
12-
10-
8-
6-
4-
2-
0-
Percentile = 90
10
-4
10
-3
10"
10"
10 100
Dose Level (mg)
Figure 5.3. Maximum ratio of impacts for the networks for four different injection times as
a function of dose level, for the 90th, 95th, 99th, and 100th percentile injection nodes.
-------
50
Sensitivity of Impacts to Injection Time
CD
E
o
o>
CD
I
b
CO
•s
CO
Q.
E
M—
O
o
'-4—'
CD
o:
16-
14-
12-
10-
8-
6-
4-
2-
0-
Percentile = 80
10
,-4
10
,-3
10
100
16-
14-
12-
10-
8-
6-
4-
2-
0-
6**~
Percentile = 60
10
-4
10
-3
10"
10"
10 100
16-
14-
12-
10-
8-
6-
4-
2-
0-
Network
Percentile = 70
10
,-4
10
,-3
10"
10"
10
16-
14-
12-
10-
8-
6-
4-
2-
0-
Percentile = 50
10
-4
10
-3
10"
10"
Dose Level (mg)
100
10 100
Figure 5.4. Maximum ratio of impacts for the networks for four different injection times as
a function of dose level, for the 50th, 60th, 70th, and 80th percentile injection nodes.
-------
Sensitivity of Impacts to Injection Time
51
o
o
o
o>
E
o 0.4-
o
o>
0:00 6:00 12:00 18:00 24:00
0:00 6:00 12:00 18:00 24:00
o
to
Q.
2.5 H
~ 2.0 H
CD
Q.
E
M—
O
o
'-4—'
CD
o:
1.5-
1.0-
Percentile = 75
2.0-
1.5-
1.0-
0.5-
0:00 6:00 12:00 18:00
24:00 0:00
Injection Time
6:00 12:00 18:00 24:00
Figure 5.5. Relative impacts for the 12 networks for different injection times, for a dose level
of 0.0001 mg and for injection nodes associated with different percentile impacts. Note that
the vertical scale is logarithmic.
-------
52
Sensitivity of Impacts to Injection Time
24:00
0:00 6:00 12:00
18:00 24:00 0:00
Injection Time
6:00 12:00 18:00 24:00
Figure 5.6. Relative impacts for the 12 networks for different injection times, for a dose level
of 0.001 mg and for injection nodes associated with different percentile impacts. Note that
the vertical scale is logarithmic.
-------
Sensitivity of Impacts to Injection Time
53
Dose Level = 0.01 mg
10.0-
1.0-
o
o
o
CD
E
o 0.01
"8
Percentile = 100
0:00
6:00
12:00 18:00 24:00
o
CD
Q- 3.0-
— 2.5-
2.0-
CD
Q.
E
M—
o
o
'-4—'
CD
a:
1.5-
1.0-
0.5-
Percentile = 75
1.8-
1.6-
1.4-
1.2-
1.0-
0.8-
0.6-
Percentile = 95
0:00
6:00
12:00 18:00 24:00
2.0-
1.0-
0.5-
Percentile = 50
0:00 6:00 12:00 18:00 24:00 0:00
Injection Time
6:00 12:00 18:00 24:00
Figure 5.7. Relative impacts for the 12 networks for different injection times, for a dose level
of 0.01 mg and for injection nodes associated with different percentile impacts. Note that the
vertical scale is logarithmic.
-------
54
Sensitivity of Impacts to Injection Time
Dose Level = 0.1 mg
10.0-
1.0-
o
o
o
o> 0.1
E
g 0.01
Percentile = 100
0:00
6:00
12:00 18:00 24:00
o
to
Q.
2 5.0-
CD
Q.
E
CD
or
2.0-
1.0-
0.5-
Percentile = 75
o- ^
0:00
6:00
12:00 18:00
2.0-
1.0-
0.5-
Percentile = 95
0:00 6:00
12:00 18:00
24:00
10.0-
5.0-
2.0-
1.0-
0.5-
Percentile = 50
24:00 0:00
Injection Time
6:00
12:00 18:00
24:00
Figure 5.8. Relative impacts for the 12 networks for different injection times, for a dose level
of 0.1 mg and for injection nodes associated with different percentile impacts. Note that the
vertical scale is logarithmic.
-------
Sensitivity of Impacts to Injection Time
55
Dose Level = 1.0 mg
100.0H
10.0-
o
o
O>
E
1.0-
0.1-
0.01 -
o
o>
o
CD
Q.
O
5
Percentile = 100
5.0-
2.0-
1.0-
0.5-
Percentile = 95
0:00
6:00
12:00 18:00
20.0-
10.0-
E
"o 5.0-
o
'-4—'
CD
2.0-
1.0-
0.5-
Percentile = 75
24:00 0:00
50.0-
6:00 12:00 18:00 24:00
20.0'
10.0-
5.0-
2.0-
1.0-
0.5-
Percentile = 50
0:00 6:00 12:00 18:00
24:00 0:00
Injection Time
6:00
12:00 18:00 24:00
Figure 5.9. Relative impacts for the 12 networks for different injection times, for a dose level
of 1 mg and for injection nodes associated with different percentile impacts. Note that the
vertical scale is logarithmic.
-------
56
Sensitivity of Impacts to Injection Time
o
o-
00
o
CD-
CO
o
a-
CO
•a
to
CO
I
dose level = 0.0001 mg
§•
00:00 06:00 12:00 18:00
dose level = 0.01 mg
o
81
o
o-
dose level = 0.001 mg
00:00 06:00 12:00 18:00
dose level = 0.1 mg
00:00 06:00 12:00 18:00 00:00 06:00 12:00 18:00
Injection Times
Figure 5.10. Distributions of impacts for Network 12 across all injection nodes for different
injection times, for dose levels of 0.0001,0.001,0.01, and 0.1 mg.
-------
Sensitivity of Impacts to Injection Time
57
o
o-
co
8j
CN
dose level = 1.0 mg
CO
Q.
E
00:00 06:00 12:00 18:00
o_
dose level = 100 mg
dose level = 10 mg
00:00 06:00 12:00 18:00
00:00 06:00 12:00 18:00
Injection Times
Figure 5.11. Distributions of impacts for Network 12 across all injection nodes for different
injection times, for dose levels of 1,10, and 100 mg.
-------
58
Sensitivity of Impacts to Injection Time
o
o
00
-S? 8
-a ^
c
CD
3
O °
.c
o
o
CD
I 8
a.
E o
— in
o _
dose level = 0.0001 mg
S-
o
OM
O _
dose level = 0.01 mg
0 200 400 600 800
o
s
o
o -
dose level = 0.001 mg
O
,o „ o
\ I
0 50 100 150 200 250
dose level = 0.1 mg
0 5 10 15 20 25 30 01
Impact, 0:00 (thousands)
Figure 5.12. Impacts for Network 12 for injection times of 0:00 and 6:00, for dose levels of
0.0001, 0.001, 0.01, and 0.1 mg. The red lines have a slope of 1. Each point corresponds to
one injection node.
-------
Sensitivity of Impacts to Injection Time
59
8
o
o
in
o
8
o
o
OM
8-
o
<~>.
CD
•S
CO
dose level = 1 mg
100
200
300 400
dose level = 100 mg
0.0 0.2 0.4 0.6 0.8
o _
o
to
o
in
o
CO
o
OM
O _
dose level = 10 mg
1.0
Impact, 0:00
0 5 10 15 20 25 30
Figure 5.13. Impacts for injection times of 0:00 and 6:00, for dose levels of 1,10, and 100 mg.
The red lines have a slope of 1. Each point corresponds to one injection node.
-------
60
Sensitivity of Impacts to Injection Time
o
o -
00
dose level = 0.0001 mg
O O
o
o
csi o _|
T— 1^
•g"
CD §
Q.
8-
O _
CO
O _
C\l
o _
dose level = 0.01 mg
O
O
O O
200 400 600 800
0 5 10 15 20 25 30
o
o
CO
8
C\l
O
o
dose level = 0.001 mg
00
0 50 100 150 200 250
dose level = 0.1 mg
Impact, 0:00 (thousands)
Figure 5.14. Impacts for Network 12 for injection times of 0:00 and 12:00, for dose levels of
0.0001, 0.001, 0.01, and 0.1 mg. The red lines have a slope of 1. Each point corresponds to
one injection node.
-------
Sensitivity of Impacts to Injection Time
61
o
o
to
o
o -
OM
O
o
•s
CO
Q.
dose level = 1 mg
O
100
200 300
400
dose level = 100 mg
0.0 0.2 0.4 0.6 0.8
o
in
o
CO
o
OM
O _
dose level = 10 mg
1.0
Impact, 0:00
10 15 20 25 30
Figure 5.15. Impacts for Network 12 for injection times of 0:00 and 12:00, for dose levels of
1,10, and 100 mg. The red lines have a slope of 1. Each point corresponds to one injection
node.
-------
62
Sensitivity of Impacts to Injection Time
8j
00
8
to
O
(/5 O
"O CN
•a
CD
O o -
dose level = 0.0001 mg
°00
00
o
o
06
> o
CD 03
Q.
dose level = 0.01 mg
O
o
0 200 400 600 800
O
O -
dose level = 0.001 mg
O
O
0 50 100 150 200 250
dose level = 0.1 mg
O O
00°
0 5 10 15 20 25 30 01
Impact, 0:00 (thousands)
Figure 5.16. Impacts for Network 12 for injection times of 0:00 and 18:00, for dose levels of
0.0001, 0.001, 0.01, and 0.1 mg. The red lines have a slope of 1. Each point corresponds to
one injection node.
-------
Sensitivity of Impacts to Injection Time
63
o
o
CO
CO
Q.
o
o
o ^
o
o -
to
dose level = 1 mg
o _
0.0
100
200
300
400
dose level = 100 mg
0.2 0.4 0.6 0.8
o
OM -
O _
00
O _
1.0
Impact, 0:00
dose level = 10 mg
10 15 20 25 30
Figure 5.17. Impacts for Network 12 for injection times of 0:00 and 18:00, for dose levels of
1,10, and 100 mg. The red lines have a slope of 1. Each point corresponds to one injection
node.
-------
64
Sensitivity of Impacts to Injection Time
o
o
00
dose level = 0.0001 mg
o
o
-S? 8
-a ^
c
CD
3
O °
.c
o
o
CD
I 8
a.
E o
— in
o _
>80th, 0:00
O p... O >80th, 6:00
O > 80th, both
200 400 600 800
s-
o _
dose level = 0.01 mg
o
S
o
a
dose level = 0.001 mg
O
0 50 100 150 200 250
dose level = 0.1 mg
0 5 10 15 20 25 30 01
Impact, 0:00 (thousands)
Figure 5.18. Impacts for Network 12 that are at or above the 80th percentile level for an
injection time of 6:00 and/or an injection time of 0:00, for dose levels of 0.0001, 0.001, 0.01,
and 0.1 mg. The dotted lines indicate 80th percentile impacts. The black lines have a slope of
1. Each point corresponds to one injection node.
-------
Sensitivity of Impacts to Injection Time
65
o
o
CO
o
o
in
o
o
CO
o
o
CN
O
O
o
q o
CD
CO
I -
dose level = 1 mg
0 100 200 300 400
dose level = 100 mg
s-
o
CO
o
CN
o _
dose level = 10 mg
0.0 0.2 0.4 0.6 0.8 1.0
Impact, 0:00
0 5 10 15 20 25 30
Figure 5.19. Impacts for Network 12 that are at or above the 80th percentile level for an
injection time of 6:00 and/or an injection time of 0:00, for dose levels of 1, 10, and 100 mg.
The dotted lines indicate 80th percentile impacts. The black lines have a slope of 1. Each
point corresponds to one injection node.
-------
66
Sensitivity of Impacts to Injection Time
o
o
00
o
o
CO
CO
C/5
O o -
O >80th, 0:00
O > 80th, 12:00
O >80th, both
O
O
Csi o
I
E
200 400 600 800
8-
i-
o
CO
o
CN
O _
dose level = 0.01 mg
0
Q
o o
o
o
CO
o
o
CN
o
o
dose level = 0.001 mg
00
0 50 100 150 200 250
dose level = 0.1 mg
.o :
8
<§>
o o
10 15 20 25 30 0
Impact, 0:00 (thousands)
Figure 5.20. Impacts for Network 12 that are at or above the 80th percentile level for an
injection time of 12:00 and/or an injection time of 0:00, for dose levels of 0.0001, 0.001,0.01,
and 0.1 mg. The dotted lines indicate 80th percentile impacts. The black lines have a slope of
1. Each point corresponds to one injection node.
-------
Sensitivity of Impacts to Injection Time
67
o
o
CO
o
o
to
o
o -
OM
dose level = 1 mg
O
: o
100 200 300 400
dose level = 100 mg
o
in
o
CO
o _
dose level = 10 mg
0.0 0.2 0.4 0.6 0.8 1.0
Impact, 0:00
10 15 20 25
30
Figure 5.21. Impacts for Network 12 that are at or above the 80th percentile level for an
injection time of 12:00 and/or an injection time of 0:00, for dose levels of 1,10, and 100 mg.
The dotted lines indicate 80th percentile impacts. The black lines have a slope of 1. Each
point corresponds to one injection node.
-------
68
Sensitivity of Impacts to Injection Time
8j
00
o
o
to
.O
(/5 O
"O CN
•a
CD
O o -
dose level = 0.0001 mg
oc
O >80th, 0:00
O > 80th, 18:00
O >80th, both
o
o
06
O O
CD 03
Q.
200
400
600
800
dose level = 0.01 mg
o:
'• • -o- • • •
o
o -
dose level = 0.001 mg
0 50 100 150 200 250
dose level = 0.1 mg
O O
OcP
0 5 10 15 20 25 30 01
Impact, 0:00 (thousands)
Figure 5.22. Impacts for Network 12 that are at or above the 80th percentile level for an
injection time of 18:00 and/or an injection time of 0:00, for dose levels of 0.0001, 0.001,0.01,
and 0.1 mg. The dotted lines indicate 80th percentile impacts. The black lines have a slope of
1. Each point corresponds to one injection node.
-------
Sensitivity of Impacts to Injection Time
69
8-1
o
o
CO
CO
Q.
o
o
00
o
o
to
o _
dose level = 1 mg
o
OM -
O _
00
O _
dose level = 10 mg
0 100
200 300 400
dose level = 100 mg
0.0 0.2 0.4 0.6 0.8 1.0
Impact, 0:00
10 15 20 25 30
Figure 5.23. Impacts for Network 12 that are at or above the 80th percentile level for an
injection time of 18:00 and/or an injection time of 0:00, for dose levels of 1,10, and 100 mg.
The dotted lines indicate 80th percentile impacts. The black lines have a slope of 1. Each
point corresponds to one injection node.
-------
70
Sensitivity of Impacts to Injection Time
1.0-
-0.8H
o
CD
•a
c
CO
o
o
o, 0.6-
CO
CD
CD
0.4-
O.
_CO
0.2-
0.0-
Dose level (mg)
0.0001
0.001
0.01
0.1
1.0
- 10
I
50
I
60
I
70
Percentile
i
80
I
90
100
Figure 5.24. Overlap of high percentile injection nodes for Network 12 for an injection time
of 6:00 with those for an injection time of 0:00 for different dose levels.
-------
Sensitivity of Impacts to Injection Time
71
1.0-
o
o
CO
o
o
o,
CO
CD
O
'•s
CD
Q.
_CO
0.8-
0.6-
0.4-
O 0.2-
0.0-
Dose level (mg)
0.0001
0.001
0.01
0.1
1.0
- 10
I
50
I
60
I
70
Percentile
i
80
I
90
100
Figure 5.25. Overlap of high percentile injection nodes for Network 12 for an injection time
of 12:00 with those for an injection time of 0:00 for different dose levels.
-------
72
Sensitivity of Impacts to Injection Time
1.0-
o
o
CO
CO
o
o
o,
t/5
CD
O
'•s
CD
Q.
_CO
0.8-
0.6-
0.4-
O 0.2-
0.0-
I
50
I
60
I
70
Percentile
i
80
I
90
100
Figure 5.26. Overlap of high percentile injection nodes for Network 12 for an injection time
of 18:00 with those for an injection time of 0:00 for different dose levels.
-------
Sensitivity of Impacts to Injection Time
73
CD
•a
c
ro
o
o
o
^
O>
o
"8
a>
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
dose level = 0.0001 mg
50 60 70 80 90 100
1.0-
0.6-
0.4-
0.2-
0.0-
dose level = 0.01 mg
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
dose level = 0.001 mg
50
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
60 70
80
90 100
dose level = 0.1 mg
50 60 70 80 90 100 50 60
Percentile
70 80 90 100
Figure 5.27. Overlap of high percentile injection nodes for the networks for an injection time
of 6:00 with those for an injection time of 0:00, for dose levels of 0.0001, 0.001, 0.01, and 0.1
mg.
-------
74
Sensitivity of Impacts to Injection Time
1.0-
0.8-
0.6-
o
° 0.4-
CD
•a
o
o
o
^
E
0.2 -|
0.0 H
dose level = 1.0 mg
50
60 70 80 90 100
o
"8 1.0-
a>
0.8 -
O 0.6-
0.4-
0.2-
0.0-
dose level = 100 mg
Network
1
2 -
3 -
4
5 -
6 -
50 60 70 80
90
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
dose level = 10 mg
100
Percentile
50 60 70 80 90 100
Figure 5.28. Overlap of high percentile injection nodes for the networks for an injection time
of 6:00 with those for an injection time of 0:00, for dose levels of 1,10, and 100 mg.
-------
Sensitivity of Impacts to Injection Time
75
o
o
1.0-
0.8-
0.6-
0.4-
ro 0.2-
o
o
o
"c/r °-°"
CD
dose level = 0.0001 mg
50 60 70 80 90 100
'•g 1.0-
cp^
'^
d. 0.8-
O 0.6-
0.4-
0.2-
0.0-
dose level = 0.01 mg
50 60 70 80
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
dose level = 0.001 mg
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
dose level = 0.1 mg
90 100 50 60 70 80
Percentile
50 60 70 80 90 100
90 100
Figure 5.29. Overlap of high percentile injection nodes for the networks for an injection time
of 12:00 with those for an injection time of 0:00, for dose levels of 0.0001,0.001,0.01, and 0.1
mg.
-------
76
Sensitivity of Impacts to Injection Time
1.0-
60 70 80 90
60 70 80 90 100
'•g 1.0-
g^
'^
d. 0.8-
O 0.6-
0.4-
0.2-
0.0-
dose level = 100 mg
Network
1
2 -
3 -
4
5 -
6 -
50 60 70 80
90
100
Percentile
Figure 5.30. Overlap of high percentile injection nodes for the networks for an injection time
of 12:00 with those for an injection time of 0:00, for dose levels of 1,10, and 100 mg.
-------
Sensitivity of Impacts to Injection Time
77
o
o
06
1.0-
0.8-
0.6-
0.4-
ro 0.2-
o
o
o
"c/r °-°"
CD
dose level = 0.0001 mg
50 60 70 80 90 100
'•g 1.0-
cp^
'^
d. 0.8-
O 0.6-
0.4-
0.2-
0.0-
dose level = 0.01 mg
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
dose level = 0.001 mg
50
60 70 80 90 100
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
dose level = 0.1 mg
50 60 70 80 90 100 50 60 70 80 90 100
Percentile
Figure 5.31. Overlap of high percentile injection nodes for the networks for an injection time
of 18:00 with those for an injection time of 0:00, for dose levels of 0.0001,0.001,0.01, and 0.1
mg.
-------
78
Sensitivity of Impacts to Injection Time
o
o
CO
1.0-
0.8-
0.6-
0.4-
ro 0.2-
o
o
o
"c/r °-°"
CD
dose level = 1.0 mg
50
'•g 1.0-
cp^
'^
d. 0.8-
O 0.6-
0.4-
0.2-
0.0-
dose level = 100 mg
Network
1
2 -
3 -
4
5 -
6 -
50 60 70 80
90
60 70 80 90 100
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
dose level = 10 mg
100
Percentile
50 60 70 80 90 100
Figure 5.32. Overlap of high percentile injection nodes for the networks for an injection time
of 18:00 with those for an injection time of 0:00, for dose levels of 1,10, and 100 mg.
-------
Sensitivity of Impacts to Iiijection Time
79
Injection Nodes
o 0.0001 mg, 6:00
• 0.0001 mg, 0:00
Figure 5.33. Network 4 showing locations of injection nodes associated with 95th percentile
or higher impacts at a dose level of 0.0001 mg for injection times of 0:00 and 6:00.
-------
80
Sensitivity of Impacts to Injection Time
Injection Nodes
o 0.0001 mg, 12:00
O 0.0001 mg, 0:00
Figure 5.34. Network 4 showing locations of injection nodes associated with 95th percentile
or higher impacts at a dose level of 0.0001 mg for injection times of 0:00 and 12:00.
-------
Sensitivity of Impacts to Injection Time
81
Injection Nodes
o 0.0001 mg, 18:00
o 0.0001 mg, 0:00
Figure 5.35. Network 4 showing locations of injection nodes associated with 95th percentile
or higher impacts at a dose level of 0.0001 mg for injection times of 0:00 and 18:00.
-------
82
Sensitivity of Impacts to Injection Time
•
Injection Nodes
o 1.0 mg, 6:00
• 1.0 mg, 0:00
Figure 5.36. Network 4 showing locations of injection nodes associated with 95th percentile
or higher impacts at a dose level of 1.0 mg for injection times of 0:00 and 6:00.
-------
Sensitivity of Impacts to Injection Time
83
o
m
Injection Nodes
0 1.0 mg, 12:00
• 1.0 mg, 0:00
Figure 5.37. Network 4 showing locations of injection nodes associated with 95th percentile
or higher impacts at a dose level of 1.0 mg for injection times of 0:00 and 12:00.
-------
84
Sensitivity of Impacts to Injection Time
\iv
°9 •
a
«oj^ ^ «
0 § «o0 m °0 • o ^^
XI ^S c= ==a ^ ^?1
#0 ^e B
s ^g ^ ^c/
^
*••* V /^'tJlL;*""
^S^^W'
, rr V 1 rn TXrv^ VN
M
J
© e *•
^
/D o . '
w Oo /A^^^
a» aAiPw /Kx ^^^^.
(pGBOgCr ^^
o o5
cP
^^^
o
a/i csg -
•f • ^
V
/? r°o
#'. *•
Si ^°
\v
\ V
H VA^-,
* 0
3
.
Injection Nodes
o 1.0 mg, 18:00
• 1.0 mg, 0:00
Figure 5.38. Network 4 showing locations of injection nodes associated with 95th percentile
or higher impacts at a dose level of 1.0 mg for injection times of 0:00 and 18:00.
-------
Sensitivity of Impacts to Injection Time
85
B**"lIP
//
tut
Injection Nodes
o 0.0001 mg, 6:00
• 0.0001 mg, 0:00
Figure 5.39. Network 6 showing locations of injection nodes associated with 95th percentile
or higher impacts at a dose level of 0.0001 mg for injection times of 0:00 and 6:00.
-------
86
Sensitivity of Impacts to Injection Time
B**"lIP
//
tut
Injection Nodes
o 0.0001 mg, 12:00
• 0.0001 mg, 0:00
Figure 5.40. Network 6 showing locations of injection nodes associated with 95th percentile
or higher impacts at a dose level of 0.0001 mg for injection times of 0:00 and 12:00.
-------
Sensitivity of Impacts to Injection Time
87
Injection Nodes
o 0.0001 mg, 18:00
• 0.0001 mg, 0:00
Figure 5.41. Network 6 showing locations of injection nodes associated with 95th percentile
or higher impacts at a dose level of 0.0001 mg for injection times of 0:00 and 18:00.
-------
88
Sensitivity of Impacts to Injection Time
I
Injection Nodes
o 1.0 mg, 6:00
• 1.0 mg, 0:00
Figure 5.42. Network 6 showing locations of injection nodes associated with 95th percentile
or higher impacts at a dose level of 1.0 mg for injection times of 0:00 and 6:00.
-------
Sensitivity of Impacts to Injection Time
89
I
-f
ILJS
Injection Nodes
o 1.0 mg, 12:00
• 1.0mg,0:00
Figure 5.43. Network 6 showing locations of injection nodes associated with 95th percentile
or higher impacts at a dose level of 1.0 mg for injection times of 0:00 and 12:00.
-------
90
Sensitivity of Impacts to Injection Time
-f
Injection Nodes
o 1.0 mg, 18:00
• 1.0mg,0:00
Figure 5.44. Network 6 showing locations of injection nodes associated with 95th percentile
or higher impacts at a dose level of 1.0 mg for injection times of 0:00 and 18:00.
-------
Section 6
Sensitivity of Impacts to Injection Mass
To assess the sensitivity of impacts to injection mass, we consider two values for injection mass.
The base case uses a realistic injection mass, which we designate IX. The second case uses ten
times that mass and is labeled 10X. Figures 6.1 and 6.2 show that increasing injection mass by
a factor of ten can have a significant influence on impacts for Network 12, except at the lowest
dose levels. The pattern is similar for percentile impacts ranging from 50 to 100%, except that as
the percentile of the impact decreases below about 90%, the range of dose levels over which the
influence is small increases. For example, for the 90th percentile, the impacts for the two different
injection masses diverge above a dose level of about 0.0001 mg, while for the 50th percentile, the
impacts for the two cases diverge only for dose levels above about 0.001 mg. The impacts at the
lowest dose levels do not increase as the injected mass increases because the number of people
exposed does not increase as the injected mass increases. Doses will increase but the number of
individuals with doses above the lowest dose levels does not; the contaminated area in the network
does not continue to expand as more contaminant is injected. The area does not expand because
all areas connected to the injection node have already been reached by the contaminant.
Figures 6.3 and 6.4 provide the relative magnitudes of impacts for IX and 10X injection masses
versus dose level for all the networks. The effect of increasing injection mass is qualitatively sim-
ilar for all the networks considered. Relative impacts tend to increase as the dose level increases,
but with considerable quantitative variability among the networks. Overall behavior across the
networks becomes more consistent for percentiles below about 90%. However, there is still con-
siderable variability from network to network in their relative response to increased injection mass.
For example, for 60th percentile impacts and a dose level of 0.1 mg, the relative increase in impact
as the injection mass is increased by a factor of ten is about two for Network 6 and about 15 for
Network 7. (For comparison, absolute values for impacts for the networks for IX injections are
provided in Figures 3.1 to 3.3.)
The relative influence of changes in injection mass on impact for the different networks is
shown quantitatively in Figures 6.5 through 6.9. The plots illustrate how varying injection mass
over five orders of magnitude affects impacts relative to the impact for the baseline case of a IX
mass injection, for dose levels ranging from 0.0001 mg to 1.0 mg. Results are provided for four
impact levels (percentiles equal to 100, 95, 75, and 50). At the lower dose levels, impact can be
quite insensitive to injection mass over a considerable range in injection mass, but sensitivity varies
considerably among the networks. For example, for a dose level of 0.0001 mg (Figure 6.5), the
50th percentile impact for Network 6 shows minimal change as injection mass is varied from about
91
-------
92 Sensitivity of Impacts to Injection Mass
0.01X to 10X, a change in injected mass of three orders of magnitude. However, for Network 11
the change is about 10, substantially larger, but still small relative to the change in injection mass.
At higher dose levels, impacts increase considerably as injection mass increases, but again with
large variability among the networks.
Using violin plots, the distribution of impacts associated with the IX and 10X injection masses
is provided in Figures 6.10 and 6.11 for Network 12. Consistent with the results in Figures 6.1
and 6.2, the distributions for the two cases can be considerably different, except at the lowest dose
level.
The scatter plots in Figures 6.12 and 6.13 show for Network 12 that on a node-by-node basis
the impacts associated with the 10X injection mass are usually larger than those associated with the
IX injection mass, but, in general, there is no linear relationship between the impacts for the two
cases. Impacts are often much larger for the 10X case, as would be expected. However, the overlap
of high percentile nodes for the two cases can be small at and above the 80th percentile impact level
for dose levels of 0.1 mg and higher (Figures 6.14 and 6.15). Increasing the injection mass can
result in a substantial shifting of the locations of the high-percentile injection nodes. Figure 6.16
shows the low overlap of the highest percentile nodes for Network 12, particularly for the higher
dose levels.
Considering all the networks, Figures 6.17 and 6.18 shows the overlap of injection nodes for the
two injection masses. The overlap is high for a dose level of 0.0001 mg for most of the networks
for all levels of impact. However, as the dose level increases, the degree of overlap decreases
considerably, although there is considerable variability across the networks.
The locations of injection nodes in Network 4 associated with impacts at and above the 95th per-
centile for injection masses of IX and 10X are shown in Figure 6.19 for a dose level of 0.0001 mg.
The overlap of the nodes is high. Figure 6.20 compares the location of 95th percentile and higher
injection nodes for Network 4 for injection masses of IX and 10X for a dose level of 1.0 mg. In
this case, the overlap is low. Figures 6.21 and 6.22 provide a similar comparison for Network 6
and demonstrate similar results, all of which are consistent with the overlap plots shown in Figures
6.17 and 6.18.
Impacts can show considerable sensitivity to large increases in injected mass, both in terms of
ranked impacts and overlap of high percentile nodes. However, at low dose levels the sensitivity
even to large increases in injected mass is generally small.
-------
Sensitivity of Impacts to Injection Mass
93
10
,4-
10J
o>
t/5
O
Q
s_
CD
.c
O)
±
5
c
O
10-
1 -
,6-
ro 10'
^
Q.
O
Q.
CO
§• 104-l
10J
,2-
10
10-
1 -
10~4 10~3 10"2
percentile = 100
10 100
percentile = 95
10B-
103-
10
,4-
3.
10
10-
1 -
10'
,6-
10
10J
,4-
,2-
10
10-
1 -
1X mass
10X mass
percentile = 99
1 10 100
percentile = 90
10 100 10~4
Dose Level (mg)
10"
1 10 100
Figure 6.1. Impacts associated with injection masses of IX and 10X for Network 12 as a
function of dose level, for the 90th, 95th, 99th, and 100th percentile injection nodes.
-------
94
Sensitivity of Impacts to Injection Mass
10
,4-
10J
o>
t/5
O
Q
s_
CD
.c
O)
±
5
c
O
10-
1 -
,6-
ro 10'
^
Q.
O
Q.
CO
§• 104-l
10J
,2-
10
10-
1 -
10~4 10~3 10"2
percentile = 80
10 100
percentile = 60
10B-
103-
10
,4-
3.
10
10-
1 -
10'
,6-
10
10J
,4-
,2-
10
10-
1 -
1X mass
10X mass
percentile = 70
1 10 100
percentile = 50
10 100 10~4
Dose Level (mg)
10"
1 10 100
Figure 6.2. Impacts associated with injection masses of IX and 10X for Network 12 as a
function of dose level, for the 50th, 60th, 70th, and 80th percentile injection nodes.
-------
Sensitivity of Impacts to Injection Mass
95
x
•s
CO
Q.
E
"
CD
Q.
E
40-
35'
30-
25-
20-
is-
le-
s'
0-
40-
35-
30-
25-
20-
15-
10-
5-
0-
Percentile = 100
10 100
Percentile = 95
10"4 10~3 10"2
10 100
40-
35-
30-
25-
20-
15-
10-
5-
0-
40-
35-
30-
25-
20-
15-
10-
5-
0-
Network
10"
10"
10"
10"
Percentile = 99
10 100
^ O
Percentile = 90
10"
10"
10"
10 100
Dose Level (mg)
Figure 6.3. Impact for a 10X injection mass relative to the impact for a IX injection mass, as
a function of dose level, for the 90th, 95th, 99th, and 100th percentile injection nodes.
-------
96
Sensitivity of Impacts to Injection Mass
x
•s
CO
Q.
E
40-
35-
30-
25-
20-
15-
10-
5-
0-
C- 40-
"G -3C_
CD M>
Q.
I 30-
25-
20-
15-
10-
5-
0-
Percentile = 80
10"
10"
10"
10
100
Percentile = 60
10"
10"
10"
10 100
40-
35-
30-
25-
20-
15-
10-
5-
0-
40-
35-
30-
25-
20-
15-
10-
5-
0-
Network
10"
10"
Percentile = 70
10"
10"
10"
10
100
Percentile = 50
10"
10"
10"
10 100
Dose Level (mg)
Figure 6.4. Impact for a 10X injection mass relative to the impact for a IX injection mass, as
a function of dose level, for the 50th, 60th, 70th, and 80th percentile injection nodes.
-------
Sensitivity of Impacts to Injection Mass
97
Dose Level = 0.0001 mg
o>
(/5
to
O
CD
t/5
CO
CO
CO
Q.
E
1.0-
0.5-
0.0-
Percentile = 100
10
2.0-
o
CO
Q.
E
M—
O
O
'-4—'
CO
a:
1.5-
1.0-
0.5-
0.0-
Percentile = 75
10"
10"3 10"2
10"
10
3.0-
2.5-
2.0-
1.5-
1.0-
0.5-
0.0-
Percentile = 95
Network
10"
10"
1.5-
1.0-
0.5-
0.0-
Percentile = 50
10"
10"3 10"2
10"
10
10
Injection Mass Relative to Value for Base Case
Figure 6.5. Variation in impacts as a function of injection mass for a dose level of 0.0001 mg,
for the 100th, 95th, 75th, and 50th percentile injection nodes. All impacts are relative to those
for an injection mass of IX.
-------
98
Sensitivity of Impacts to Injection Mass
Dose Level = 0.001 mg
o>
(/5
to
O
CD
t/5
CO
CO
3.01
2.5-
2.0-
1.5-
1.0-
0.5-
CO
Q.
E
o
CO
Q.
E
M—
o
o
'-4—'
CO
or
0.0-
Percentile = 100
10"
10"
10
2.0-
1.5-
1.0-
0.5-
0.0-
Percentile = 75
10"
10
3.5-
3.0-
2.5-
2.0-
1.5-
1.0-
0.5-
0.0-
Percentile = 95
10"
10"
10"
10"
10
1.5-
1.0-
0.5-
0.0-
Percentile = 50
10"
10"
10
Injection Mass Relative to Value for Base Case
Figure 6.6. Variation in impacts as a function of injection mass for a dose level of 0.001 mg,
for the 100th, 95th, 75th, and 50th percentile injection nodes. All impacts are relative to those
for an injection mass of IX.
-------
Sensitivity of Impacts to Injection Mass
99
Dose Level = 0.01 mg
o>
(/5
to
O
CD
t/5
CO
CO
CO
Q.
E
o
CO
Q.
E
M—
o
o
'-4—'
CO
o:
6-
4-
2-
0-
Percentile = 100
3-
2-
1-
0-
Percentile = 75
10
8-
6-
4-
2-
0-
Percentile = 95
4-
3-
2-
1 -
0-
Percentile = 50
10"1 1 10 10"4 10^ 1Q~2
Injection Mass Relative to Value for Base Case
10
10
Figure 6.7. Variation in impacts as a function of injection mass for a dose level of 0.01 mg,
for the 100th, 95th, 75th, and 50th percentile injection nodes. All impacts are relative to those
for an injection mass of IX.
-------
100
Sensitivity of Impacts to Injection Mass
Dose Level = 0.1 mg
o>
(/5
to
O
CD
t/5
CO
CO
CO
Q.
E
o
CO
Q.
E
M—
o
o
'-4—'
CO
o:
8-
6-
4-
2-
0-
Percentile = 100
15-
10-
5-
0-
Percentile = 75
10
10-
6-
4-
2-
0-
Percentile = 95
Network
1
2 -
3 -
4
5 -
6 -
7
8
9
10
11
12
20-
15-
10-
5-
0-
Percentile = 50
10
10"1 1 10 10"4 10^ 1Q~2
Injection Mass Relative to Value for Base Case
10
Figure 6.8. Variation in impacts as a function of injection mass for a dose level of 0.1 mg, for
the 100th, 95th, 75th, and 50th percentile injection nodes. All impacts are relative to those for
an injection mass of IX.
-------
Sensitivity of Impacts to Injection Mass
101
Dose Level = 1.0 mg
o>
(/5
to
O
CD
t/5
CO
CO
CO
Q.
E
o
CO
Q.
E
CO
o:
15H
10-
5-
0-
Percentile = 100
40-
20-
10-
0-
Percentile = 75
10"
10
35-
30-
25-
20-
15-
10-
5-
n-
Percentile = 95
Network
1
2
3
4
5
6
7
8
9
10
11
12
jt
10"
10"
10"
10"
50-
40-
30-
20-
10-
0-
Percentile = 50
10 1 10 101010 10
Injection Mass Relative to Value for Base Case
10
10
Figure 6.9. Variation in impacts as a function of injection mass for a dose level of 1 mg, for
the 100th, 95th, 75th, and 50th percentile injection nodes. All impacts are relative to those for
an injection mass of IX.
-------
102
Sensitivity of Impacts to Injection Mass
•s
CD
O.
O
8-
8.
00
8.
OM
o
o-
OM
o
in-
dose level = 0.0001 mg
1X mass
10X mass
dose level = 0.01 mg
1X mass
o_
dose level = 0.001 mg
1X mass
10X mass
dose level = 0.1 mg
10X mass IXmass
Injection Masses
10X mass
Figure 6.10. Distributions of impacts for Network 12 across all injection nodes for injection
masses of IX and 10X, for dose levels of 0.0001,0.001,0.01, and 0.1 mg.
-------
Sensitivity of Impacts to Injection Mass
103
8.
o
8-
OM
s-
CO
Q.
i-
0_
dose level = 1.0 mg
1X mass
10X mass
dose level = 100 mg
1X mass
10X mass
dose level = 10 mg
1X mass
10X mass
Injection Masses
Figure 6.11. Distributions of impacts for Network 12 across all injection nodes for injection
masses of IX and 10X, for dose levels of 1,10, and 100 mg.
-------
104
Sensitivity of Impacts to Injection Mass
•a
c
CD
CD
E
X
o
CD
O.
E
o
o
OM
8
o
o
o
00
8
to
o
o
OM
dose level = 0.0001 mg
200
400
600
800
o
o
to
0 50 100 150 200 250
10 15 20 25 30 01
Impact, 1X mass (thousands)
Figure 6.12. Impacts for Network 12 for injection masses of IX and 10X, for dose levels of
0.0001, 0.001, 0.01, and 0.1 mg. The red lines have a slope of 1. Each point corresponds to
one injection node.
-------
Sensitivity of Impacts to Injection Mass
105
100 200 300 400
b
CO
Q.
E
i-
o
CO
0 .
OM
O .
0 -
dose level = 100 mg
O
O
O
8
-1 °"
10 15 20 25 30
0.0 0.2 0.4 0.6 0.8 1.0
Impact, 1X mass
Figure 6.13. Impacts for Network 12 for injection masses of IX and 10X, for dose levels of
1,10, and 100 mg. The red lines have a slope of 1. Each point corresponds to one injection
node.
-------
106
Sensitivity of Impacts to Injection Mass
•a
c
CD
CD
X
o
CD
O.
E
8
o
8
to
O
O
OM
dose level = 0.0001 mg
>80th, 1X
.. O >80th, 10X
O > 80th, both
200
400
600
800
10 15
dose level = 0.001 mg
(BD
0 50 100 150 200 250
20 25 30 01
Impact, 1X mass (thousands)
Figure 6.14. Impacts for Network 12 that are at or above the 80th percentile level for IX
and/or 10X injection masses, for dose levels of 0.0001, 0.001, 0.01, and 0.1 mg. The dotted
lines indicate 80th percentile impacts. The black lines have a slope of 1. Each point corre-
sponds to one injection node.
-------
Sensitivity of Impacts to Injection Mass
107
0 .
O .
100 200 300 400
dose level = 100 mg
0.0 0.2 0.4 0.6 0.8 1.0
Impact, 1X mass
10 15 20 25 30
Figure 6.15. Impacts for Network 12 that are at or above the 80th percentile level for IX
and/or 10X injection masses, for dose levels of 1,10, and 100 mg. The dotted lines indicate
80th percentile impacts. The black lines have a slope of 1. Each point corresponds to one
injection node.
-------
108
Sensitivity of Impacts to Injection Mass
1.0-
0.8-
o>
1 0.6 H
c
o
o>
ro" 0.4-
0.2-
0.0-
I
50
I
60
I
70
Percentile
i
80
I
90
100
Figure 6.16. Overlap of high percentile injection nodes for Network 12 for injection masses
of IX and 10X for different dose levels.
-------
Sensitivity of Impacts to Injection Mass
109
1.0-
0.8-
0.6-
0.4-
t/5
t/5
CO
c 0.0-
o
'•Q
CD
dose level = 0.0001 mg
50 60 70 80 90 100
co" 1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
dose level = 0.01 mg
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
dose level = 0.001 mg
50
60 70 80 90 100
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
dose level = 0.1 mg
**.
50 60 70 80 90 100 50 60 70 80
Percentile
90 100
Figure 6.17. Overlap of high percentile injection nodes for the networks for injection masses
of IX and 10X, for dose levels of 0.0001,0.001,0.01, and 0.1 mg.
-------
110
Sensitivity of Impacts to Injection Mass
1.0-
0.8-
0.6-
0.4-
t/5
£ 0.2-1
o
CD
0.0-
dose level = 1.0 mg
50 60 70 80 90 100
CD" 1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
dose level = 100 mg
Network
1
2 -
3 -
4
5 -
6 -
50 60 70 80
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
dose level = 10 mg
90 100
Percentile
50 60 70 80 90 100
Figure 6.18. Overlap of high percentile injection nodes for the networks for injection masses
of IX and 10X, for dose levels of 1,10, and 100 mg.
-------
Sensitivity of Impacts to Injection Mass
111
Injection Nodes
o 0.0001 mg, 10X
• 0.0001 mg, 1X
Figure 6.19. Network 4 showing locations of injection nodes associated with 95th percentile
or higher impacts at a dose level of 0.0001 mg for injection masses of IX and 10X.
-------
112
Sensitivity of Impacts to Injection Mass
Injection Nodes
o 1.0 mg, 10X
O 1 mg, 1X
Figure 6.20. Network 4 showing locations of injection nodes associated with 95th percentile
or higher impacts at a dose level of 1.0 mg for injection masses of IX and 10X.
-------
Sensitivity of Impacts to Injection Mass
113
Injection Nodes
o 0.0001 mg, 10X
• 0.0001 mg, 1X
Figure 6.21. Network 6 showing locations of injection nodes associated with 95th percentile
or higher impacts at a dose level of 0.0001 mg for injection masses of IX and 10X.
-------
114
Sensitivity of Impacts to Injection Mass
•7
Injection Nodes
o 1.0 mg, 10X
• 1.0 mg, 1X
Figure 6.22. Network 6 showing locations of injection nodes associated with 95th percentile
or higher impacts at a dose level of 1.0 mg for injection masses of IX and 10X.
-------
Section 7
Sensitivity of Impacts to Population
Distribution
Two models for population distribution were considered. The first, the base case, uses a demand-
based population at each network node. The second uses the same (network average) population
at each NZD node. The total population for a network is the same for both cases. Figures 7.1
and 7.2 illustrate how the choice of population distribution influences estimated ranked impacts
for Network 12. Only for the highest percentile impacts is there any significant difference between
impacts for the two distributions and then only for higher dose levels.
The plots in Figures 7.3 and 7.4 provide results for all the networks, many of which show a
sensitivity to population distribution similar to that for Network 12. With only a few exceptions,
the impacts are not sensitive to the population model for impact percentiles below about 90%.
However, for the highest percentiles and higher dose levels, impacts for some of the networks show
considerable sensitivity to the population model. This sensitivity occurs because in these cases the
injection affects only a small number of nodes and the population assigned to those nodes can have
a significant effect on the impact. (For comparison, absolute values for impacts for the networks
obtained using the demand-based population model are provided in Figures 3.1 to 3.3.)
The distribution of impacts for Network 12 (Figures 7.5 and 7.6) illustrates that for the lowest
dose levels little difference exists between the impacts associated with the two population models.
However, as the dose level increases, there is considerable difference between the two cases for the
upper tail of the distribution of impacts, consistent with the plots in Figures 7.1 and 7.2.
The scatter plots in Figures 7.7 and 7.8 compare impacts node-by-node for Network 12 for the
two different population models. Scatter increases as the dose level increases. The overlap of high
percentile injection nodes generally decreases as dose level increases (Figures 7.9 and 7.10) and
generally is low for all dose levels for the highest percentile impacts (Figure 7.11).
Overlap of high percentile nodes for the two population models is shown in Figures 7.12 and
7.13 as a function of impact level for all the networks for different dose levels. For most networks,
the overlap decreases at the highest percentile impacts and can be quite low, especially for higher
dose levels.
The influence of the population model on the locations determined to be high percentile injec-
tion nodes is shown in Figures 7.14 through 7.17 for Networks 4 and 6 for dose levels of 0.0001
and 1.0 mg. For Network 4 (Figures 7.14 and 7.15), the choice of population model has limited
influence on the locations of nodes at or above the 95th percentile level for either dose level, con-
115
-------
116 Sensitivity of Impacts to Population Distribution
sistent with the overlaps shown in Figures 7.12 and 7.13 for the network (near 100% for 0.0001 mg
and about 65% for 1.0 mg). For Network 6 (Figures 7.16 and 7.17), the high percentile injection
nodes determined using the two population models are much the same for the 0.0001 mg dose
level, but differ considerably for a dose level of 1.0 mg. From Figures 7.12 and 7.13, the overlaps
for the two population models for the network are about 80% and 30% at 0.0001 mg and 1.0 mg,
respectively, which is consistent with the maps.
Impacts can have considerable sensitivity to population distribution at high dose levels, if the
impacts have very high rankings (greater than about the 95th percentile), with considerable vari-
ability among the networks. Overlap of high percentile injection nodes tends to decrease as dose
level increases and can be low for the two different population models, especially at high dose
levels. The model used for the population distribution matters most for estimating impacts at high
dose levels; it is less important at low dose levels.
-------
Sensitivity of Impacts to Population Distribution
117
10
,4-
10J
o>
t/5
O
Q
s_
CD
.c
O)
±
5
c
O
10-
1 -
,6-
ro 10'
^
Q.
O
Q.
CO
§• 104-l
10J
,2-
10
10-
1 -
10~4 10~3 10"2
percentile = 100
10 100
percentile = 95
10B-
103-
10
,4-
3.
10
10-
1 -
10'
,6-
10
10J
,4-
,2-
10
10-
1 -
percentile = 99
1 10 100
percentile = 90
Demand Based
Fixed
10 100 10~4
Dose Level (mg)
10~3 10"2 10"1 1 10 100
Figure 7.1. Impacts associated with different population models for Network 12 as a function
of dose level, for the 90th, 95th, 99th, and 100th percentile injection nodes.
-------
118
Sensitivity of Impacts to Population Distribution
10
,4-
10J
o>
t/5
O
Q
s_
CD
.c
O)
±
5
c
O
10-
1 -
,6-
ro 10'
^
Q.
O
Q.
CO
§• 104-l
10J
,2-
10
10-
1 -
10~4 10~3 10"2
percentile = 80
10 100
percentile = 60
10B-
103-
10
,4-
3.
10
10-
1 -
10'
,6-
10
10J
,4-
,2-
10
10-
1 -
percentile = 70
1 10 100
percentile = 50
Demand Based
Fixed
10 100 10~4
Dose Level (mg)
10~3 10"2 10"1 1 10 100
Figure 7.2. Impacts associated with different population models for Network 12 as a function
of dose level, for the 50th, 60th, 70th, and 80th percentile injection nodes.
-------
Sensitivity of Impacts to Population Distribution
119
10-
6-
CD 4-
co
m
CO
E
CD
Q,
(/5
"8
CO
Q.
E
2-
0- Percentile = 100
10-
1 10 100
CO
I 6H
4-
2-
0-
Percentile = 95
10~4 10~3 10"2
Network
6-
4-
2-
0- Percentile = 99
10~4 10~3 10~2 10"1 1 10 100
10-
6-
4-
2-
0-
.e?
«.
-------
120
Sensitivity of Impacts to Population Distribution
10-
6-
CD 4-
CD
00
CD
E
CD
O
2-
0-
9. „
Percentile = 80
g 10~4 10~3 10~2
Q.
.
1°H
CD
I
4-
2-
0-
9- -
Percentile = 60
10~4 10~3 10"2
10-
1 10 100
6-
4-
2-
0-
Network
1
2 -
3 -
4
5 -
6 -
9
10
11
12
Percentile = 70
10"4 10"3 10"
1 10 100
10-
6-
4-
2-
0-
Percentile = 50
1 10 100
Dose Level (mg)
10~4 10~3 10"2 10~1 1 10 100
Figure 7.4. Relative magnitudes of the impacts for the networks obtained with different
population models as a function of dose level, for the 50th, 60th, 70th, and 80th percentile
injection nodes.
-------
Sensitivity of Impacts to Population Distribution
121
o
o-
00
o
CD-
CO
o
a-
CO
•a
to
CO
•s
CO
O.
E
8-
o_
dose level = 0.0001 mg
Demand Based
Fixed
dose level = 0.01 mg
Demand Based
CM.
o_
dose level = 0.001 mg
Demand Based
Fixed
dose level = 0.1 mg
Fixed Demand Based
Population Model
Fixed
Figure 7.5. Distributions of impacts for Network 12 across all injection nodes for different
population models, for dose levels of 0.0001,0.001,0.01, and 0.1 mg.
-------
122
Sensitivity of Impacts to Population Distribution
o
o
a
sj
in
o
8H
_
I CO.
dose level = 1.0 mg
Demand Based
Fixed
dose level = 100 mg
Demand Based
Fixed
o
a-
8-I
8-
dose level = 10 mg
Demand Based
Population Model
Fixed
Figure 7.6. Distributions of impacts for Network 12 across all injection nodes for different
population models, for dose levels of 1,10, and 100 mg.
-------
Sensitivity of Impacts to Population Distribution
123
o
o
00
o
o
to
O O
'-4— '
co
o
H_T in
O
to
I i
o
CO
o
OM
O _
dose level = 0.0001 mg
200 400 600
dose level = 0.01 mg
800
0 5 10 15 20 25 30
o
in
c\i
o
o
c\i
o
in
8-
o _
dose level = 0.001 mg
50 100 150 200 250
dose level = 0.1 mg
Impact, Demand-Based Population (thousands)
Figure 7.7. Impacts for Network 12 obtained with different population models, for dose levels
of 0.0001,0.001,0.01, and 0.1 mg. The red lines have a slope of 1. Each point corresponds to
one injection node.
-------
124
Sensitivity of Impacts to Population Distribution
o
'-4—'
_CD
Q.
£
o
to
Q.
E
8.
o
o
in
8
o
8
in
dose level = 1 mg
O
8
100 200 300 400
dose level = 100 mg
o
s
o
in
o
in
dose level = 10 mg
O
O
O
8
10 15 20 25 30
0.0 0.2 0.4 0.6 0.8 1.0
Impact, Demand-Based Population
Figure 7.8. Impacts for Network 12 obtained with different population models, for dose levels
of 1,10, and 100 mg. The red lines have a slope of 1. Each point corresponds to one injection
node.
-------
Sensitivity of Impacts to Population Distribution
125
O
O
00
O
O
to
to
O O
'-4— '
CO
O
H_T in
O
CD
I i
o
CO
o
OM
O _
O > 80th, demand
O > 80th, fixed
O > 80th, both
200
400
600
800
dose level = 0.01 mg
10 15 20 25 30
o
in
c\i
o
o
c\i
o
in
8-
o _
dose level = 0.001 mg
50
100 150 200 250
dose level = 0.1 mg
Impact, Demand-Based Population (thousands)
Figure 7.9. Impacts for Network 12 that are at or above the 80th percentile level obtained
with different population models, for dose levels of 0.0001, 0.001, 0.01, and 0.1 mg. The
dotted lines indicate 80th percentile impacts. The black lines have a slope of 1. Each point
corresponds to one injection node.
-------
126
Sensitivity of Impacts to Population Distribution
o
'-4—'
_CD
Q.
£
o
to
Q.
E
8.
o
o
in
8
o
8
in
dose level = 1 mg
O
100 200 300 400
dose level = 100 mg
o
S
o
in -
o
a
dose level = 10 mg
O
O
O
8
10 15 20 25
30
O > 80th, demand
O > 80th, fixed
O > 80th, both
0.0 0.2 0.4 0.6 0.8 1.0
Impact, Demand-Based Population
Figure 7.10. Impacts for Network 12 that are at or above the 80th percentile level obtained
with different population models, for dose levels of 1, 10, and 100 mg. The dotted lines
indicate 80th percentile impacts. The black lines have a slope of 1. Each point corresponds to
one injection node.
-------
Sensitivity of Impacts to Population Distribution
127
1.0-
0.8-
o>
•a
o
0.6 H
a.
o
Q.
0.2-
0.0-
Dose level (mg)
0.0001
0.001
0.01
0.1
1.0
- 10
I
50
I
60
I
70
Percentile
i
80
I
90
100
Figure 7.11. Overlap of high percentile injection nodes for Network 12 for the two population
models for different dose levels.
-------
128
Sensitivity of Impacts to Population Distribution
1.0-
0.8-
0.6-
0.4-
0.2-
o 0.0-
dose level = 0.0001 mg
50
60 70 80 90 100
Q.
£
f
O 0.8-
0.6-
0.4-
0.2-
0.0-
dose level = 0.01 mg
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
dose level = 0.001 mg
50
60 70 80 90 100
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
dose level = 0.1 mg
50 60 70 80 90 100 50
Percentile
60 70 80 90 100
Figure 7.12. Overlap of high percentile injection nodes for the networks for the two popula-
tion models, for dose levels of 0.0001,0.001,0.01, and 0.1 mg.
-------
Sensitivity of Impacts to Population Distribution
129
1.0-
0.8-
0.6-
0.4-
3 0.2-|
o
o 0.0-
'-4—'
_ro
^
Q.
£
-------
130
Sensitivity of Impacts to Population Distribution
Injection Nodes
o 0.0001 mg, Fixed
• 0.0001 mg, Demand
Figure 7.14. Network 4 showing locations of injection nodes associated with 95th percentile
or higher impacts at a dose level of 0.0001 mg for the two population models.
-------
Sensitivity of Impacts to Population Distribution
131
.
•
Injection Nodes
o 1.0 mg, Fixed
• 1.0 mg, Demand
Figure 7.15. Network 4 showing locations of injection nodes associated with 95th percentile
or higher impacts at a dose level of 1.0 mg for the two population models.
-------
132
Sensitivity of Impacts to Population Distribution
Injection Nodes
o 0.0001 mg, Fixed
• 0.0001 mg, Demand
Figure 7.16. Network 6 showing locations of injection nodes associated with 95th percentile
or higher impacts at a dose level of 0.0001 mg for the two population models.
-------
Sensitivity of Impacts to Population Distribution
133
•7
Injection Nodes
° 1.0 mg, Fixed
• 1.0 mg, Demand
Figure 7.17. Network 6 showing locations of injection nodes associated with 95th percentile
or higher impacts at a dose level of 1.0 mg for the two population models.
-------
134 Sensitivity of Impacts to Population Distribution
-------
Section 8
Sensitivity of Impacts to Ingestion Model
In this section we present our analysis of the sensitivity of impacts to the ingestion model used
in estimating the impacts. The presentation follows that used for the other factors in the previous
sections.
Figures 8.1 and 8.2 compare ranked impacts for Network 12 estimated using the two ingestion
models (Models 1 and 2, described in Table 2.1). The deterministic model (Model 2) consistently
yields the larger impacts. The ratio of impacts for this network tends to increase with dose level.
The relative impacts are approximately four or less.
The sensitivity of ranked impacts to the ingestion model varies by network (Figures 8.3 and
8.4). For some networks sensitivity tends to increase with dose level except for the worst-case
injection nodes (100th percentile). For the majority of the networks the ratios of impacts are about
four or less. For some networks the differences in results can be larger. (For comparison, absolute
values for impacts for the networks obtained using Model 1 are provided in Figures 3.1 to 3.3.)
The distributions of impacts for Network 12 are most similar for the two ingestion models for
the lowest dose level (Figures 8.5 and 8.6). There can be considerable differences in the upper tails
of the distributions for higher dose levels.
When compared injection node by injection node, there can be a considerable difference in
the impacts estimated for Network 12 using the two ingestion models, as illustrated in the plots in
Figures 8.7 and 8.8. The scatter in the plots increases with dose level up to a dose level of 0.1 mg.
The overlap of high percentile nodes for Network 12, illustrated by plots in Figures 8.9, 8.10, and
8.11, decreases considerably with dose level, at least up to a dose level of 0.1 mg. At the 95th
percentile, the overlap at a dose level of 0.1 mg is only about 0.1 (Figure 8.11).
The extent of overlap of high percentile injection nodes varies substantially across the networks
(Figures 8.12 and 8.13). At the lowest dose level, the overlap is high for all impact levels, except
for Network 7. Network 7 tends to have the lowest overlap of injection nodes for dose levels of
1.0 mg or lower, but other networks can also have low overlap of high percentile nodes for the
two different models. At higher dose levels, the overlap of high percentile nodes is low for all the
networks.
Maps illustrating the overlap of high percentile injection nodes obtained using the two ingestion
models are shown in Figures 8.14 through 8.17. Figures 8.14 and 8.15 provide results for Network
4 for dose levels of 0.0001 mg and 1.0 mg, respectively and Figures 8.16 and 8.17 provide similar
results for Network 6. As indicated in Figures 8.12 and 8.13, the overlaps are high at the lower
dose level but low at the higher dose level.
135
-------
136 Sensitivity of Impacts to Ingestion Model
The magnitudes of ranked impacts can be sensitive to the ingestion model for some networks,
particularly at higher dose levels. However, the more important sensitivity to these models involves
the impacts associated with individual injection nodes. The overlap of the injection nodes associ-
ated with the highest impacts can be low for some networks for the two ingestion models even for
relatively low dose levels. There is considerable variability from network to network.
-------
Sensitivity of Impacts to Ingestion Model
137
10
,4-
10J
o>
t/5
O
Q
s_
CD
.c
O)
±
5
c
O
10-
1 -
,6-
ro 10'
^
Q.
O
Q.
CO
§• 104-l
10J
,2-
10
10-
1 -
10~4 10~3 10"2
percentile = 100
10 100
percentile = 95
10B-
103-
10
,4-
3.
10
10-
1 -
10'
,6-
10
10J
,4-
,2-
10
10-
1 -
Model 1
Model 2
percentile = 99
1 10 100
percentile = 90
10 100 10~4
Dose Level (mg)
10"
1 10 100
Figure 8.1. Impacts associated with different ingestion models for Network 12 as a function
of dose level, for the 90th, 95th, 99th, and 100th percentile injection nodes.
-------
138
Sensitivity of Impacts to Ingestion Model
10
,4-
10J
o>
t/5
O
Q
s_
CD
.c
O)
±
5
c
O
10-
1 -
,6-
ro 10'
^
Q.
O
Q.
CO
§• 104-l
10J
,2-
10
10-
1 -
10~4 10~3 10"2
percentile = 80
10 100
percentile = 60
10B-
103-
10
,4-
3.
10
10-
1 -
10'
,6-
10
10J
,4-
,2-
10
10-
1 -
Model 1
Model 2
percentile = 70
1 10 100
percentile = 50
10 100 10~4
Dose Level (mg)
10"
1 10 100
Figure 8.2. Impacts associated with different ingestion models for Network 12 as a function
of dose level, for 50th, 60th, 70th, and 80th percentile injection nodes.
-------
Sensitivity of Impacts to Ingestion Model
139
12-
10-
O>
•a
o
CO
a.
E
CD
•a
o
"
CD
O.
E
Percentile = 100
10 100
12-
10-
8-
6-
4-
2-
0-
Percentile = 95
10~4 10"3 10"2
12-
10-
8-
6-
4-
2-
0-
Network
1
2 -
3 -
4
5 -
6 -
9
10
11
12
Percentile = 99
10 100
12-
10-
8-
6-
4-
2-
0-
Percentile = 90
10 100 10"4
Dose Level (mg)
10 100
Figure 8.3. Relative magnitudes of the impacts for the networks obtained with different in-
gestion models as a function of dose level, for the 90th, 95th, 99th, and 100th percentile injection
nodes.
-------
140
Sensitivity of Impacts to Ingestion Model
12-
10-
O>
•a
o
CO
a.
E
CD
•a
o
Percentile = 80
12-
o
§_ 10-|
0-
Percentile = 60
10~4 10"3 10"2
10 100
12-
10-
8-
6-
4-
2-
Network
1
2 -
3 -
4
5 -
6 -
9
10
11
12
0- Percentile = 70
10"4 10
,-3
10"
12-
10-
8-
6-
4-
2-
0-
Percentile = 50
10 100 10~4
Dose Level (mg)
10 100
10 100
Figure 8.4. Relative magnitudes of the impacts for the networks obtained with different
ingestion models as a function of dose level, for 50th, 60th, 70th, and 80th percentile injection
nodes.
-------
Sensitivity of Impacts to Ingestion Model
141
§•
o
O'
00
8.
OM
CD
a.
E
dose level = 0.0001 mg
Model 1
Model 2
dose level = 0.01 mg
Model 1
o_
dose level = 0.001 mg
Model 1
Model 2
dose level = 0.1 mg
Model 2 Model 1
Ingestion Model
Model 2
Figure 8.5. Distributions of impacts for Network 12 across all injection nodes for different
ingestion models, for dose levels of 0.0001,0.001,0.01, and 0.1 mg.
-------
142
Sensitivity of Impacts to Ingestion Model
8-
o
o-
00
o
o-
CO
o
CD-
CM
CO
Q.
E
oq
CD"
CD
CD"
q
CD"
dose level = 1.0 mg
dose level = 10 mg
Model 1
Model 2
Model 1
Model 2
dose level = 100 mg
Model 1
Model 2
Ingestion Model
Figure 8.6. Distributions of impacts for Network 12 across all injection nodes for different
ingestion models, for dose levels of 1,10, and 100 mg.
-------
Sensitivity of Impacts to Ingestion Model
143
o
8
8
to
= §
8 °
CM
"CD
•a
o
o
to
§
o
OM
dose level = 0.0001 mg
dose level = 0.01 mg
200 400 600 800
O
o
-00
o
o -
in
o
o -
OM
8-
o _
dose level = 0.001 mg
50 100 150 200 250
dose level = 0.1 mg
5 10 15 20 25 30 01
Impact, Model 1 (thousands)
Figure 8.7. Impacts for Network 12 obtained with different ingestion models, for dose levels
of 0.0001,0.001,0.01, and 0.1 mg. The red lines have a slope of 1. Each point corresponds to
one injection node.
-------
144
Sensitivity of Impacts to Ingestion Model
CM
"CD
•a
o
CO
O.
E
o
o
OM
O
O
O
O
O
00
8-
8
OM
dose level = 1 mg
00
o
CD
o
CN
o
q
o
QDOD) SEOCO O
100 200 300 400
dose level = 100 mg
o
o -
g-
o
CO H
o
• H
O O
00
OO
OO
0.0 0.2 0.4 0.6 0.8 1.0
Impact, Model 1
dose level = 10 mg
OOO OO
10 15 20 25
30
Figure 8.8. Impacts for Network 12 obtained with different ingestion models, for dose levels
of 1,10, and 100 mg. The red lines have a slope of 1. Each point corresponds to one injection
node.
-------
Sensitivity of Impacts to Ingestion Model
145
IS)
IS)
3
o
CM
"CD
•a
o
o
CO
o
o
OM
O
8
o
o
to
8
o _
o
OM
dose level = 0.0001 mg
O > 80th, Model!
O > 80th, Model 2
O > 80th, both
200
400
600
800
dose level = 0.01 mg
O
0
J30
o
o
in
o
o
OM
8-
o _
dose level = 0.001 mg
0 50 100 150 200 250
dose level = 0.1 mg
10 15 20 25 30 01
Impact, Model 1 (thousands)
Figure 8.9. Impacts for Network 12 that are at or above the 80th percentile level obtained with
different ingestion models, for dose levels of 0.0001,0.001,0.01, and 0.1 mg. The dotted lines
indicate 80th percentile impacts. The black lines have a slope of 1. Each point corresponds to
one injection node.
-------
146
Sensitivity of Impacts to Ingestion Model
CM
"CD
•a
o
CO
O.
o
o
CN
O
O
00
dose level = 1 mg
00
o
CD
o
CN
o
q
o
ODOBSEGOOO •
o
OM -
100
200
300
400
dose level = 100 mg
o
o -
g-
o
CO H
O
• H
O O
00
OO
OO
0.0 0.2 0.4 0.6 0.8 1.0
Impact, Model 1
dose level = 10 mg
•ooo -oo
o—o- •
•O--
10 15 20 25
30
O > 80th, Model!
O > 80th, Model 2
O > 80th, both
Figure 8.10. Impacts for Network 12 that are at or above the 80th percentile level obtained
with different ingestion models, for dose levels of 1,10, and 100 mg. The dotted lines indicate
80th percentile impacts. The black lines have a slope of 1. Each point corresponds to one
injection node.
-------
Sensitivity of Impacts to Ingestion Model
147
1.0-
0.8-
O>
•a
o
0.6 H
a>
O)
0.4-
0.2-
0.0-
I
50
I
60
I
70
Percentile
i
80
I
90
100
Figure 8.11. Overlap of high percentile injection nodes for Network 12 for different ingestion
models for different dose levels.
-------
148
Sensitivity of Impacts to Ingestion Model
O>
•a
o
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
dose level = 0.0001 mg
O>
O)
50 60 70 80 90 100
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
dose level = 0.01 mg
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
dose level = 0.001 mg
50 60 70 80 90 100
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
dose level = 0.1 mg
50 60 70 80 90 100 50
Percentile
60 70 80 90 100
Figure 8.12. Overlap of high percentile injection nodes for the networks for the two ingestion
models, for dose levels of 0.0001,0.001,0.01, and 0.1 mg.
-------
Sensitivity of Impacts to Ingestion Model
149
o
1.0-
0.8-
0.6-
0.4-
0.2 H
0.0-
dose level = 1.0 mg
O)
50 60 70 80 90 100
0.8 H
0.6-
0.4-
0.2-
0.0-
dose level = 100 mg
Network
^^—
^^^
1
2
3
4
5
6
7
- - 8
9
10 S^\
— 11 *5i
12 ""'X-G
50 60 70 80
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
dose level = 10 mg
90 100
Percentile
50 60 70 80 90 100
Figure 8.13. Overlap of high percentile injection nodes for the networks for the two ingestion
models, for dose levels of 1,10, and 100 mg.
-------
150
Sensitivity of Impacts to Ingestion Model
.
Injection Nodes
o 0.0001 mg, M2
• 0.0001 mg, M1
Figure 8.14. Network 4 showing locations of injection nodes associated with 95th percentile
or higher impacts at a dose level of 0.0001 mg for the two ingestion models. Ml is Model 1
and M2 is Model 2.
-------
Sensitivity of Impacts to Ingestion Model
151
•
°o m o tf*> % 0
o CPq) o © m
° * $P ^
n /"' 0%^ ° £? ^
B . ;. ' *° o. ^
ft ff
S &,'
&s^£#°
©
s ^
81 gs
0^
".
Injection Nodes
o 1.0mg,M2
• 1.0 mg, M1
Figure 8.15. Network 4 showing locations of injection nodes associated with 95th percentile
or higher impacts at a dose level of 1.0 mg for the two ingestion models. Ml is Model 1 and
M2 is Model 2.
-------
152
Sensitivity of Impacts to Ingestion Model
Injection Nodes
o 0.0001 mg, M2
• 0.0001 mg, M1
Figure 8.16. Network 6 showing locations of injection nodes associated with 95th percentile
or higher impacts at a dose level of 0.0001 mg for the two ingestion models. Ml is Model 1
and M2 is Model 2.
-------
Sensitivity of Impacts to Ingestion Model
153
Injection Nodes
o 1.0 mg, M2
o 1.0 mg, M1
Figure 8.17. Network 6 showing locations of injection nodes associated with 95th percentile
or higher impacts at a dose level of 1.0 mg for the two ingestion models. Ml is Model 1 and
M2 is Model 2.
-------
154 Sensitivity of Impacts to Ingestion Model
-------
Section 9
Sensitivity of Impacts to Combinations of
Factors
The preceding analysis examined the influence of changes in one factor while all others were held
constant. However, in reality, the various factors do not affect estimated impacts in isolation. It
is therefore necessary to examine the importance of changes in the factors when variation in all
factors is assumed.
As discussed in Section 2, we examined the sensitivity of impacts to combinations of factors.
We consider four injection times (0:00, 6:00, 12:00, and 18:00), two injection durations (0 and
24 h), two injection masses (IX and 10X), and two population distributions (average population
value at each node and demand-based population at each node), yielding a total of 32 combinations
of factors. Considering the ingestion model in addition to these four factors would have doubled
the number of required simulations and therefore was not pursued.
Figures 9.1 through 9.4 provide histograms of the impacts obtained for the 32 cases (combi-
nations of factors) for Network 2. Results are shown for four dose levels (0.0001, 0.01, 1, and
100 mg) and for four impact levels (100th, 90th, 70th, and 50th percentiles). Note that impacts in
Figures 9.1 and 9.3 are in units of thousands. Using the 100th percentile plot for a dose level of
0.0001 mg in Figure 9.1 as an example, there are six cases that have impacts that lie in the bin
centered on 75,000 impacts. The bin has a width of 2,000 impacts. This means that six out of the
32 cases have their largest impacts in the range from 74,000 to 76,000. Similarly, one case out
of the 32 has its 70th percentile impacts for a dose level of 0.0001 mg in the range from 3,000 to
3,200.
At the lowest dose level, the magnitudes of the impacts for the various cases are similar (there
is only a small change in their relative values), particularly at the higher percentiles. However,
at the higher dose levels, the relative magnitudes of the impacts can vary substantially across the
various cases. Changes in the various factors have a much larger influence on impacts at the higher
dose levels than at the lower dose levels. The histograms in Figures 9.1 through 9.4 demonstrate
that changes in the factors can significantly influence impacts, but they do not tell us which factors
have the most influence when all factors are considered together. That information can be obtained
from Figures 9.5 through 9.20, which show how impacts vary with each of the individual factors.
Figures 9.5 through 9.20 provide the same histograms as shown in Figures 9.1 through 9.4.
However, they are now color coded to show the relationship of impacts to injection time (Fig-
ures 9.5 through 9.8), injection duration (Figures 9.9 through 9.12), injection mass (Figures 9.13
155
-------
156 Sensitivity of Impacts to Combinations of Factors
through 9.16), and population distribution (Figures 9.17 through 9.20). In the histograms, a partic-
ular color corresponds to a particular value of a factor. For example, in Figure 9.5 blue corresponds
to an injection time of 12:00. If colors are distributed more or less uniformly across a histogram,
then variations in the factor are not having much influence on the magnitudes of the impacts. If
patterns are present in the distribution of the colors then impacts are being influenced differently
by different values of the factor. For example, in Figure 9.9, at a dose level of 0.0001 mg, there is
strong division between the distribution of impacts associated with an injection duration of 1 h and
that of an injection duration of 24 h. The impacts associated with the latter are larger than those
associated with the former, independent of all the other factors. However, at a dose level of 100 mg
(Figure 9.12) the pattern disappears and no correlation between injection duration and impact is
evident. If there is no correlation with duration, then other factors are more important and duration
is not a good predictor of impacts. Figures 9.13 throught 9.16 show similar results for injection
mass, except that now the simple relationship between the factor and impacts occurs for higher
dose levels (high injection mass results in high impacts) but disappears at the lowest dose level. As
would be expected, at higher dose levels the higher impacts generally are associated with the 10X
injection mass. Increasing the injection mass by a factor of ten generally results in increased im-
pacts, independent of any other factor. However, at the lowest dose level the relationship between
injection mass and impacts is less clear: impacts are not strongly related to injection mass at this
dose level. No simple, consistent patterns are apparent for injection time (Figures 9.5 through 9.8)
or the population model (Figure 9.17 through 9.20).
Overall, when all the factors are considered together on the basis of ranked impacts, impacts
are most sensitive to injection mass and injection duration, but the degree of sensitivity depends
on dose level. No significant relationship exists between impact level and injection time (Figures
9.5 through 9.8). Finally, impacts are not strongly related to the distribution used for population
(Figures 9.17 through 9.20).
When considered on the basis of individual injection nodes, the degree of consistency with
which particular nodes remain high impact injection nodes across the combinations of factors
depends on the dose level and impact level. The overlap of nodes across the 32 combinations of
factors is shown in Figures 9.21 to 9.23 for Network 2 for different dose levels for 75th, 95th, and
99th and higher percentile impacts. The horizontal axis in the plots gives a listing of individual
injection nodes. The nodes are labeled simply 1, 2, 3 etc., after sorting the nodes by the number
of times the same node is a particular percentile or higher node for the 32 cases. If the same
nodes were the high percentile injection nodes for all combinations of factors, the plots would
show a constant count of 32 and a number of nodes equal to about 750, 150, and 30 for the case
of 75th, 95th, and 99th and higher percentiles, respectively. (The number of nodes would be equal
to about 3000 times the fraction of nodes at or above the given percentile, e.g., 0.25 for the 75th
percentile.) The dashed red lines show what the plots would look like if the same nodes were the
high percentile nodes for all cases. The figures show that overlap is highest for the low dose levels
and lower impact levels. For example, for a dose level of 0.0001 mg, the same 600-plus nodes
(out of a possible 750 or so) are 75th percentile and higher impact nodes for all 32 combinations
of factors (Figure 9.21). However, for a dose level of 100 mg and the 99th and higher percentiles
(Figure 9.23), at most only about 10 combinations of factors share a common high impact node,
out of a possible 32. In summary, the plots show that the consistency with which particular nodes
remain high impact nodes is low for high dose levels, but can be high for low dose levels, as is
shown for the 75th percentile case in Figure 9.21.
-------
Sensitivity of Impacts to Combinations of Factors
157
t/5
CD
CO
CO
O
CD
.Q
E
10-
6-
4-
2-
0-
0.0001 mg, percentile = 100
6-
4-
2-
0-
60
65
70
0.0001 mg, percentile = 70
n
75
10-
6-
4-
2-
0-
0.0001 mg, percentile = 90
10.0 10.5 11.0 11.5 12.0 12.5 13.0 13.5
0.0001 mg, percentile = 50
6-
4-
2-
0-
3.0 3.5 4.0 4.5 5.0 0.4 0.6 0.8 1.0 1.2 1.4
Impact (thousands)
Figure 9.1. Histograms of impacts for the 32 combinations of factors for Network 2 for a
dose level of 0.0001 mg and different impact levels. The number of cases on the vertical axis
is the number of combinations of factors, out of a total of 32, that are in a particular bin in a
histogram.
-------
158
Sensitivity of Impacts to Combinations of Factors
0.01 mg, percentile = 100
0.01 mg, percentile = 90
8-
6-
A. -
(/I
CO n
O U"
° 20
CD
.Q
E
1 1 1
30 40 50
7-
6-
5-
4-
3-
1 -
0-
1 1
60 70 6 7 £
I 1
I 1
1
i i i i
! 9 10 11 12
0.01 mg, percentile = 70
0.01 mg, percentile = 50
10-
8-
6-
4-
0 _
r
—
1
—
-n
1 1
1
10-
8-
6-
4-
2-
1
n
1 1 1 1 1
1.5 2.0 2.5 3.0 3.5 4.0 0.4 0.6
Impact (thousands)
0.8
1.0
1.2
Figure 9.2. Histograms of impacts for the 32 combinations of factors for Network 2 for a
dose level of 0.01 mg and different impact levels. The number of cases on the vertical axis is
the number of combinations of factors, out of a total of 32, that are in a particular bin in a
histogram.
-------
Sensitivity of Impacts to Combinations of Factors
159
1 mg, percentile = 100
14-
12-
10-
8-
6-
4-
to 2-
CD
to
CO n
O °"
I — |
^ i i i i i i i i
° 01234567
CD
.Q
E
10-
6-
4-
2-
0-
1 mg, percentile = 70
0.0 0.5
1.0
1.5
15-
10-
5-
0-
1 mg, percentile = 90
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
1 mg, percentile = 50
10-
6-
4-
2-
0-
I I I I I
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Impact (thousands)
Figure 9.3. Histograms of impacts for the 32 combinations of factors for Network 2 for a
dose level of 1 mg and different impact levels. The number of cases on the vertical axis is
the number of combinations of factors, out of a total of 32, that are in a particular bin in a
histogram.
-------
160
Sensitivity of Impacts to Combinations of Factors
t/5
CD
CO
CO
O
CD
.Q
E
6-
4-
2-
0-
100 mg, percentile = 100
15-
10-
5-
0-
100 mg, percentile = 90
50 100 150
100 mg, percentile = 70
200
0-
8-
6-
4-
2-
n-
"
-,
i 1
n
0 10 20 30 40 50
100 mg, percentile = 50
15-
10-
5-
0-
0 5 10 15 20 25 0
Impact
i i i
10
15
Figure 9.4. Histograms of impacts for the 32 combinations of factors for Network 2 for a
dose level of 100 mg and different impact levels. The number of cases on the vertical axis is
the number of combinations of factors, out of a total of 32, that are in a particular bin in a
histogram.
-------
Sensitivity of Impacts to Combinations of Factors
161
0.0001 mg, percentile = 100
0.0001 mg, percentile = 90
10-
CD 8-
CD
o
6-
CD 4-
_Q
2-
0-
time
60 65 70 75
Impact (thousands)
10-
W r,
CD 8-
CD
O
6-
CD 4-
_Q
2H
o-
10.0 10.5 11.0 11.5 12.0 12.5 13.0 13.5
Impact (thousands)
0.0001 mg, percentile = 70
0.0001 mg, percentile = 50
10-
CD
w
CD
O
CD
6-
4-
15 2-
0-
Jll
3.0 3.5 4.0 4.5
Impact (thousands)
5.0
10-
CD
w
CD
O
CD
6-
4-
15 2-
0-
0.4 0.6 0.8 1.0 1.2 1.4
Impact (thousands)
Figure 9.5. Histograms of impacts for the 32 combinations of factors for Network 2 for a
dose level of 0.0001 mg and different impact levels with cases color coded to show associated
injection time. The number of cases on the vertical axis is the number of combinations of
factors, out of a total of 32, that are in a particular bin in a histogram.
-------
162
Sensitivity of Impacts to Combinations of Factors
0.01 mg, percentile = 100
0.01 mg, percentile = 90
10-
CD
CD
O
"5
CD
6-
4-
2-
0-
20 30 40 50 60 70
Impact (thousands)
10-
to
CD
w
CD
O
—
CD
6-
4-
2-
0-
6 7 8 9 10 11 12
Impact (thousands)
0.01 mg, percentile = 70
0.01 mg, percentile = 50
10-
CD 8-
CD
O
6-
O
CD 4.
E
z 2-
0-
time
2.0 2.5 3.0 3.5 4.0
Impact (thousands)
10-
CD
w
CD
O
CD
6-
4-
15 2-
0-
0.4 0.6 0.8 1.0 1.2
Impact (thousands)
Figure 9.6. Histograms of impacts for the 32 combinations of factors for Network 2 for a
dose level of 0.01 mg and different impact levels with cases color coded to show associated
injection time. The number of cases on the vertical axis is the number of combinations of
factors, out of a total of 32, that are in a particular bin in a histogram.
-------
Sensitivity of Impacts to Combinations of Factors
163
1 mg, percentile = 100
1 mg, percentile = 90
14-
12-
03
O 8H
CD
4H
2H
0-
time
00
06
12
18
0 1
Impact (thousands)
15-
CD
oioH
CD
-Q 5j
E 5H
0-
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
Impact (thousands)
1 mg, percentile = 70
1 mg, percentile = 50
10-
„
CD 8-
w
CD
O
6-
CD 4-
2-
0-
0.0 0.5 1.0 1.5
Impact (thousands)
10-
n
CD 8-
w
CD
O
6-
CD 4-
2-
0-
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Impact (thousands)
Figure 9.7. Histograms of impacts for the 32 combinations of factors for Network 2 for a dose
level of 1 mg and different impact levels with cases color coded to show associated injection
time. The number of cases on the vertical axis is the number of combinations of factors, out
of a total of 32, that are in a particular bin in a histogram.
-------
164
Sensitivity of Impacts to Combinations of Factors
10-
to
CD
to
CD
o
CD
6-
4-
15 2-
0-
100 mg, percentile =100
0 50 100
Impact
150 200
10-
to
CD
to
CD
O
CD
6-
4-
15 2-
0-
100 mg, percentile = 90
0 10 20 30 40 50
Impact
100 mg, percentile = 70
100 mg, percentile = 50
14-
to12'
CD
eg 10.
—
CD
.Q
E
6-
2-
0'
L
0 5 10 15 20 25
Impact
20-
CD
to
CD
O
10H
CD
0-
time
5 10 15
Impact
Figure 9.8. Histograms of impacts for the 32 combinations of factors for Network 2 for a dose
level of 100 mg and different impact levels with cases color coded to show associated injection
time. The number of cases on the vertical axis is the number of combinations of factors, out
of a total of 32, that are in a particular bin in a histogram.
-------
Sensitivity of Impacts to Combinations of Factors
165
0.0001 mg, percentile = 100
0.0001 mg, percentile = 90
10-
CD 8-
CD
o
6-
CD 4-
_Q
2-
0-
60 65 70 75
Impact (thousands)
10-
to
CD
w
CD
O
6-
CD 4-
_Q
2H
o-
10.0 10.5 11.0 11.5 12.0 12.5 13.0 13.5
Impact (thousands)
0.0001 mg, percentile = 70
0.0001 mg, percentile = 50
10-
CD
CD
O
"5
CD
£1
E
6-
4-
2-
0-
3.0 3.5 4.0 4.5
Impact (thousands)
5.0
10-
CD
w
CD
O
CD
6-
4-
15 2-
0-
0.4 0.6 0.8 1.0 1.2 1.4
Impact (thousands)
Figure 9.9. Histograms of impacts for the 32 combinations of factors for Network 2 for a
dose level of 0.0001 mg and different impact levels with cases color coded to show associated
injection duration. The number of cases on the vertical axis is the number of combinations
of factors, out of a total of 32, that are in a particular bin in a histogram.
-------
166
Sensitivity of Impacts to Combinations of Factors
0.01 mg, percentile = 100
0.01 mg, percentile = 90
10-
CD
"3 fij
o 6H
"5
JD l
D 2-
Z
0-
20 30 40 50 60
Impact (thousands)
70
10-
to
CD
w
CD
O
—
CD
6-
4-
2-
0-
6 7 8 9 10 11 12
Impact (thousands)
0.01 mg, percentile = 70
0.01 mg, percentile = 50
10-
CD 8-
CD
O
6-
O
CD 4.
E
z 2-
0-
duration
2.0 2.5 3.0 3.5 4.0
Impact (thousands)
10-
CD
w
CD
O
—
CD
6-
4-
15 2-
0-
0.4 0.6 0.8 1.0 1.2
Impact (thousands)
Figure 9.10. Histograms of impacts for the 32 combinations of factors for Network 2 for a
dose level of 0.01 mg and different impact levels with cases color coded to show associated
injection duration. The number of cases on the vertical axis is the number of combinations
of factors, out of a total of 32, that are in a particular bin in a histogram.
-------
Sensitivity of Impacts to Combinations of Factors
167
14-
12-
w
CD
O 8H
*6H
CD
|4H
Z 2H
0-
1 mg, percentile = 100
1 mg, percentile = 90
duration
01
Impact (thousands)
15-
to
CD
w
CD
.Q
E
D
0-
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
Impact (thousands)
1 mg, percentile = 70
1 mg, percentile = 50
10-
„
CD 8-
w
CD
O
6-
CD 4-
2-
0-
0.0 0.5 1.0 1.5
Impact (thousands)
10-
n
CD 8-
w
CD
O
6-
CD 4-
2-
0-
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Impact (thousands)
Figure 9.11. Histograms of impacts for the 32 combinations of factors for Network 2 for
a dose level of 1 mg and different impact levels with cases color coded to show associated
injection duration. The number of cases on the vertical axis is the number of combinations
of factors, out of a total of 32, that are in a particular bin in a histogram.
-------
168
Sensitivity of Impacts to Combinations of Factors
10-
to
CD
to
CD
o
6-
—
CD
£1
=5 2H
0-
100 mg, percentile =100
50 100
Impact
150 200
10-
to
CD
to
CD
O
CD
6-
4-
15 2-
0-
100 mg, percentile = 90
0 10 20 30 40 50
Impact
100 mg, percentile = 70
100 mg, percentile = 50
14-
to12'
CD
eg 10.
—
CD
£1
E
6-
2-
0'
L
0 5 10 15 20 25
Impact
20-
CD lo
to
CD
O
t5 10'
CD
E 5
0-
duration
1
24
aura
.
5 10
Impact
15
Figure 9.12. Histograms of impacts for the 32 combinations of factors for Network 2 for a
dose level of 100 mg and different impact levels with cases color coded to show associated
injection duration. The number of cases on the vertical axis is the number of combinations
of factors, out of a total of 32, that are in a particular bin in a histogram.
-------
Sensitivity of Impacts to Combinations of Factors
169
0.0001 mg, percentile = 100
0.0001 mg, percentile = 90
10-
CD 8-
CD
o
6-
CD 4-
_Q
2-
0-
60 65 70 75
Impact (thousands)
10-
W r,
CD 8-
CD
O
6-
CD 4-
_Q
2H
o-
10.0 10.5 11.0 11.5 12.0 12.5 13.0 13.5
Impact (thousands)
0.0001 mg, percentile = 70
0.0001 mg, percentile = 50
10-
CD
w
CD
O
CD
6-
4-
15 2-
0-
3.0 3.5 4.0 4.5
Impact (thousands)
5.0
10-
CD
w
CD
O
CD
6-
4-
15 2-
0-
0.4 0.6 0.8 1.0 1.2 1.4
Impact (thousands)
Figure 9.13. Histograms of impacts for the 32 combinations of factors for Network 2 for a
dose level of 0.0001 mg and different impact levels with cases color coded to show associated
injection mass. The number of cases on the vertical axis is the number of combinations of
factors, out of a total of 32, that are in a particular bin in a histogram.
-------
170
Sensitivity of Impacts to Combinations of Factors
0.01 mg, percentile = 100
0.01 mg, percentile = 90
10-
CD
"3 fij
o 6H
"5
JD l
D 2-
Z
0-
20 30 40 50 60
Impact (thousands)
70
10-
to
CD
w
CD
O
—
CD
6-
4-
2-
0-
6 7 8 9 10 11 12
Impact (thousands)
0.01 mg, percentile = 70
0.01 mg, percentile = 50
10-
„
CD 8-
w
CD
O
6-
CD 4-
2-
0-
2.0 2.5 3.0 3.5 4.0
Impact (thousands)
10-
CD
w
CD
O
—
CD
6-
4-
D 2-
0-
0.4 0.6 0.8 1.0 1.2
Impact (thousands)
Figure 9.14. Histograms of impacts for the 32 combinations of factors for Network 2 for a
dose level of 0.01 mg and different impact levels with cases color coded to show associated
injection mass. The number of cases on the vertical axis is the number of combinations of
factors, out of a total of 32, that are in a particular bin in a histogram.
-------
Sensitivity of Impacts to Combinations of Factors
171
14-
12-
w
CD
O 8H
*6H
CD
|4H
Z 2H
0-
1 mg, percentile = 100
1 mg, percentile = 90
mass
1
1 OX
1X
01
Impact (thousands)
15-
CD
w
CD
.Q
E
D
0-
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
Impact (thousands)
1 mg, percentile = 70
1 mg, percentile = 50
10-
„
CD 8-
w
CD
O
6-
CD 4-
2-
0-
0.0 0.5 1.0 1.5
Impact (thousands)
10-
n
CD 8-
w
CD
O
6-
CD 4-
2-
0-
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Impact (thousands)
Figure 9.15. Histograms of impacts for the 32 combinations of factors for Network 2 for
a dose level of 1 mg and different impact levels with cases color coded to show associated
injection mass. The number of cases on the vertical axis is the number of combinations of
factors, out of a total of 32, that are in a particular bin in a histogram.
-------
172
Sensitivity of Impacts to Combinations of Factors
100 mg, percentile =100
100 mg, percentile = 90
10-
to
CD
to
CD
o
CD
6-
4-
15 2-
0-
0 50 100 150 200
Impact
10-
to
CD
to
CD
O
CD
6-
4-
15 2-
0-
0 10 20 30 40 50
Impact
100 mg, percentile = 70
100 mg, percentile = 50
14-
to12'
CD
eg 10.
—
CD
.Q
E
6-
2-
0'
L
0 5 10 15 20 25
Impact
20-
CD K
CD
O
t5 10~l
CD
E 5^
0-
mass
10X
1X
mas'.
m
5 10
Impact
15
Figure 9.16. Histograms of impacts for the 32 combinations of factors for Network 2 for a
dose level of 100 mg and different impact levels with cases color coded to show associated
injection mass. The number of cases on the vertical axis is the number of combinations of
factors, out of a total of 32, that are in a particular bin in a histogram.
-------
Sensitivity of Impacts to Combinations of Factors
173
0.0001 mg, percentile = 100
0.0001 mg, percentile = 90
10-
CD 8-
CD
" 6H
CD 4-
_Q
2-
0-
pop
60 65 70 75
Impact (thousands)
10-
w „
CD 8-
CD
" 6H
CD 4-
_Q
2H
o-
10.0 10.5 11.0 11.5 12.0 12.5 13.0 13.5
Impact (thousands)
0.0001 mg, percentile = 70
0.0001 mg, percentile = 50
10-
CD
w
CD
O
CD
E
6-
4-
2-
0-
3.0 3.5 4.0 4.5
Impact (thousands)
5.0
10-
CD
w
CD
O
CD
6-
4-
2-
0-
0.4 0.6 0.8 1.0 1.2 1.4
Impact (thousands)
Figure 9.17. Histograms of impacts for the 32 combinations of factors for Network 2 for
a dose level of 0.0001 mg and different impact levels with cases color coded to show which
population model was used. D is the demand-based population model and F is the population
model with an average, fixed population at each node. The number of cases on the vertical
axis is the number of combinations of factors, out of a total of 32, that are in a particular bin
in a histogram.
-------
174
Sensitivity of Impacts to Combinations of Factors
0.01 mg, percentile = 100
0.01 mg, percentile = 90
10-
CD
CD
O
"5
CD
6-
4-
2-
0-
20 30 40 50 60 70
Impact (thousands)
10-
to
CD
w
CD
O
—
CD
6-
4-
2-
0-
6 7 8 9 10 11 12
Impact (thousands)
0.01 mg, percentile = 70
0.01 mg, percentile = 50
10-
CD 8-
CD
*6
CD 4.
E
z 2-
0-
pop
2.0 2.5 3.0 3.5 4.0
Impact (thousands)
10-
CD
w
CD
O
—
CD
6-
4-
2-
0-
0.4 0.6 0.8 1.0 1.2
Impact (thousands)
Figure 9.18. Histograms of impacts for the 32 combinations of factors for Network 2 for
a dose level of 0.01 mg and different impact levels with cases color coded to show which
population model was used. D is the demand-based population model and F is the population
model with an average, fixed population at each node. The number of cases on the vertical
axis is the number of combinations of factors, out of a total of 32, that are in a particular bin
in a histogram.
-------
Sensitivity of Impacts to Combinations of Factors
175
1 mg, percentile = 100
1 mg, percentile = 90
14-
12-
w
CD
O 8H
*6H
CD
|4H
Z 2H
0-
pop
01
15-
CD
w
010H
CD
JD
E
D
0-
Impact (thousands)
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
Impact (thousands)
1 mg, percentile = 70
1 mg, percentile = 50
10-
„
CD 8-
w
CD
O
6-
CD 4-
2-
0-
0.0 0.5 1.0 1.5
Impact (thousands)
10-
n
CD 8-
w
CD
O
6-
CD 4-
2-
0-
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Impact (thousands)
Figure 9.19. Histograms of impacts for the 32 combinations of factors for Network 2 for a
dose level of 1 mg and different impact levels with cases color coded to show which population
model was used. D is the demand-based population model and F is the population model
with an average, fixed population at each node. The number of cases on the vertical axis is
the number of combinations of factors, out of a total of 32, that are in a particular bin in a
histogram.
-------
176
Sensitivity of Impacts to Combinations of Factors
100 mg, percentile =100
100 mg, percentile = 90
10-
to
CD
to
CD
o
CD
6-
4-
2-
0-
10-
0 50 100 150 200
Impact
to
CD
to
CD
O
CD
6-
4-
2-
0-
0 10 20 30 40 50
Impact
100 mg, percentile = 70
100 mg, percentile = 50
14
12'
CO
CD
«
i— R •
CD b
.Q
E 4.
D
Z 2.
0
L
0 5 10 15 20 25
Impact
20-
CD
to
CD
O
CD
0-
pop
5 10 15
Impact
Figure 9.20. Histograms of impacts for the 32 combinations of factors for Network 2 for
a dose level of 100 mg and different impact levels with cases color coded to show which
population model was used. D is the demand-based population model and F is the population
model with an average, fixed population at each node. The number of cases on the vertical
axis is the number of combinations of factors, out of a total of 32, that are in a particular bin
in a histogram.
-------
Sensitivity of Impacts to Combinations of Factors
177
35-
30-
25-
20-
15-
10-
5-
0-
CD
o
o
o>
o
E
o
O
o>
0.0001 mg
35-
30-
25-
20-
15-
10-
5-
0-
35-
30-
25-
20-
15-
10-
5-
0-
200 400 600
Injection Nodes
800
0.1 mg
200 600 1000
Injection Nodes
100 mg
500 1500
Injection Nodes
35-
30-
25-
20-
15-
10-
5-
0-
0.001 mg
35-
30-
25-
20-
15-
10-
5-
0-
200 400 600 800
Injection Nodes
1.0 mg
500 1000 1500
Injection Nodes
35-
30-
25-
20-
15-
10-
5-
0-
0.01 mg
35-
30-
25-
20-
15-
10-
5-
0-
200 400 600 800
Injection Nodes
10 mg
500 1500 2500
Injection Nodes
Network 2
percentile = 75
and greater
Figure 9.21. Overlap of injection nodes for the 32 combinations of factors for Network 2 for
different dose levels for 75th percentile and higher impacts. The dashed red line illustrates
results for perfect overlap of the injection nodes for all cases. The parameter in the plots is
the dose level.
-------
178
Sensitivity of Impacts to Combinations of Factors
35-
30-
25-
20-
15-
10-
5-
0-
CD
o
o
o>
o
E
o
O
o>
0.0001 mg
35-
30-
25-
20-
15-
10-
5-
0-
35-
30-
25-
20-
15-
10-
5-
0-
50 100 150 200
Injection Nodes
0.1 mg
100 200 300 400
Injection Nodes
100 mg
500 1000
Injection Nodes
1500
35-
30-
25-
20-
15-
ID-
S'
0-
0.001 mg
35-
30-
25-
20-
15-
10-
5-
0-
50 100 150 200
Injection Nodes
1.0 mg
200 400 600 800
Injection Nodes
35-
30-
25-
20-
15-
10-
5-
0-
0.01 mg
35-
30-
25-
20-
15-
10-
5-
0-
50 150 250
Injection Nodes
10 mg
400 800 1200
Injection Nodes
Network 2
percentile = 95
and greater
Figure 9.22. Overlap of injection nodes for the 32 combinations of factors for Network 2 for
different dose levels for 95th percentile and higher impacts. The dashed red line illustrates
results for perfect overlap of the injection nodes for all cases. The parameter in the plots is
the dose level.
-------
Sensitivity of Impacts to Combinations of Factors
179
35-
30-
25-
20-
15-
10-
5-
0-
0.0001 mg
0 10 20 30 40
Injection Nodes
03
O
O
O
'•Q
O>
'E*
c
Commoi
M—
O
s_
O>
E
^
z
35-
30-
25-
20-
15-
10-
5-
0-
35-
30-
25-
20-
15-
10-
5-
0-
0.1 mg
— t
^^^_
0 50 100 150 20
Injection Nodes
100 mg
\_,
0 100 300 500
Injection Nodes
35-
30-
25-
20-
15-
10-
5-
0-
0.001 mg
0 10 20 30 40
Injection Nodes
35-
30-
25-
20-
15-
10-
5-
0-
1.0 mg
L
>1— ^
100 200 300
Injection Nodes
400
35-
30-
25-
20-
15-
10-
5-
0-
0.01 mg
0 10 20 30 40 50
Injection Nodes
35-
30-
25-
20-
15-
10-
5-
0-
10 mg
- 1
i
i
i
i
i
I'
v
V^^
100 200 300 400
Injection Nodes
Network 2
percentile = 99
and greater
Figure 9.23. Overlap of injection nodes for the 32 combinations of factors for Network 2 for
different dose levels for 99th percentile and higher impacts. The dashed red line illustrates
results for perfect overlap of the injection nodes for all cases. The parameter in the plots is
the dose level.
-------
180 Sensitivity of Impacts to Combinations of Factors
-------
Section 10
Sensitivities Examined
We examined two types of sensitivities to the various factors: sensitivity that results in variations
in the magnitudes of the nih percentile impact and sensitivity that results in changes in the injection
nodes that are associated with the nih percentile and higher impacts. Sensitivity involving injec-
tion nodes was examined quantitatively using the overlap of high percentile injection nodes and
qualitatively using maps of injection nodes for two of the networks.
Figures 10.1 through 10.5 summarize the sensitivity of impacts to the various individual factors
for all of the networks. Results are shown for the 95th percentile and two dose levels (0.0001 and
1.0 mg). The ratios of impacts are generally about 5 or less except for injection mass, for which
the ratios can exceed 10 for a dose level of 1.0 mg.
Figures 10.6 through 10.12 summarize the overlaps of injection nodes for the various individual
factors. Again, results are shown for the 95th percentile and two dose levels (0.0001 and 1.0 mg).
As can be seen from the figures, the sensitivity of impacts and degree of overlap vary from
network to network and depend on the dose level. The figures also provide plots showing how
the results are related to the areas, populations, and average population densities of the networks.
There is no obvious relationship between any of these quantities and the sensitivity to the various
factors or the degree of overlap of injection nodes exhibited by the networks. The inter-network
variability does not appear to be related to these quantities.
The results presented in Sections 4 through 9 show that not only do the magnitudes of impacts
vary with dose level and impact level (e.g., 95th versus 99th percentile), but the sensitivity of impacts
to the various factors considered also varies with those same quantities. Impacts are sensitive to all
the factors and the sensitivity depends on the network. For all factors, sensitivity to the factor tends
to increase with dose level. With the exception of the population model, sensitivity to the various
factors tends to be high at low impact levels. For the population model, sensitivity is lowest at the
lower impact levels.
The overlap of high percentile injection nodes for different values of the same factor (e.g., 1-h
and 24-h injection durations or IX and 10X mass injections) can vary substantially with dose level
and impact level, as shown in Sections 4 through 9. Overlap tends to decrease with both increasing
dose level and increasing impact level. The sensitivity of overlap to all the factors varies with the
network.
181
-------
182
Sensitivities Examined
Percentile = 95
15-
10-
c 5-
o
1
Q
r—
^t"
CN
•a
A o A 0 A
123456789 10 12
15-
10-
5-
0-
0 A o A A
£ 4 A A A A^* 0
1.5 2.0 2.5 3.0 3.5
C 2
m Network Iog10(area, km )
i
£ 15-
CD
Q.
E
"o 10-
o
'-4— '
CD
o:
5-
0-
A oA o^
£ A 40AA A ^AQ
15-
10-
5-
0-
Dose Level (mg)
0 0.0001
A 1.0
A
A
£ o (A A4&
4.0 4.5 5.0 5.5 6.0
Iog10(population)
2.5
3.0
3.5
4.0
4.5
Iog10(pop. density, persons/km)
Figure 10.1. Summary of the ratio of 95th percentile impacts associated with 1- and 24-h
injections for the networks for two dose levels. Summaries are provided by network, by
network area, network population, and average population density for each network.
-------
Sensitivities Examined
183
CD
E
[5
"
CO
Q.
E
CO
a:
15-
10-
,-
5-
o
'•s
.2. OH
oo
1 23456789 10
Network
15-
10-
5-
0-
.
A A
0 CP° 0
Percentile = 95
15-
10-
5-
0-
12
0 CP0
4.0 4.5 5.0 5.5
Iog10(population)
6.0
^ *
0 0 0 ~ 0 CPO
15-
10-
5-
0-
o
1.5 2.0 2.5 3.0 3.5
Iog10(area, km2)
Dose Level (mg)
O 0.0001
A 1.0
2.5 3.0 3.5 4.0
Iog10(pop. density, persons/km2)
4.5
Figure 10.2. Summary of the maximum ratio of 95th percentile impacts associated with dif-
ferent injection times (0:00, 6:00, 12:00, and 18:00) for the networks for two dose levels.
Summaries are provided by network, by network area, network population, and average
population density for each network.
-------
184
Sensitivities Examined
Percentile = 95
15-
10-
CD
(/5 5-
CD
^
O
'•s
CD
'B 0-
A A A
A A A
A
A A A
A
0 0
oooO°o° o0 o
£ 123456789 10 12
15-
10-
5-
0-
A A A
A A A
A
A AA
A
o o
°ooo°0 Oo o
1 1 1 1 1
1.5 2.0 2.5 3.0 3.5
"^ Network Iog10(area, km2)
CD
X
o
H 15"
£
CD
Q.
I 10-
M—
o
o
'-4— '
CD
o:
5-
0-
A A A
Al A
A
A A A
A
0 0
o ocP°o °o0o
15-
10-
5-
0-
Dose Level (mg)
A o 0.0001
A
4 A
A
0 0
Q O (© O QJJQ
4.0 4.5 5.0 5.5 6.0
Iog10(population)
2.5 3.0 3.5 4.0 4.5
Iog10(pop. density, persons/km2)
Figure 10.3. Summary of the ratio of 95th percentile impacts associated with different in-
jection masses (IX and 10X) for the networks for two dose levels. Summaries are provided
by network, by network area, network population, and average population density for each
network. No point is shown in the plots for Network 7 for a dose level of 1.0 mg because the
ratio of 21.3 is off the vertical scale.
-------
Sensitivities Examined
185
Percentile = 95
15-
10-
o
0
o 5-
CL
•a
CD
CO
CO
m
i
•a 0-
A
A A A
A
A A o
fflAAAoAOoO OO
15-
10-
5-
0-
A
A A A
A
A o A
°fflAAo£ OOQO
co ' ' ' ' '
E 1 2 3 4 5 6 7 8 9 10 12 1.5 2.0 2.5 3.0 3.5
a Network Iog10(area, km2)
•a
en
•a 15-
CD
X
L_
£
CO
In 10"
O.
E
M—
O
o
'•S3 5-
a:
0-
A.
A AA
A
(B A A^oA O <§) OO
15-
10-
5-
0-
Dose Level (mg)
0 0.0001
A 1.0
^
A A A
A
A A o 0
4.0 4.5 5.0 5.5
Iog10(population)
6.0
2.5
3.0
3.5
4.0
4.5
Iog10(pop. density, persons/km)
Figure 10.4. Summary of the ratio of 95th percentile impacts associated with the two popu-
lation models for the networks for two dose levels. Summaries are provided by network, by
network area, network population, and average population density for each network.
-------
186
Sensitivities Examined
o>
•a
o
n
15-
10-
5-
o-
Percentile = 95
15-
10-
A
a> 123456789 10 12
E Network
•Q 15-1
to
O.
E
o
o
'-4—'
CD
a:
1°-
5-
0-
4.0 4.5 5.0 5.5
Iog10(population)
6.0
5-
0-
1.5
15-
10-
5-
0-
A
A
0 A
QOO 0 0
2.0 2.5 3.0 3.5
Iog10(area, km2)
Dose Level (mg)
O 0.0001
A 1.0
2.5
3.0 3.5 4.0 4.5
Iog10(pop. density, persons/km)
Figure 10.5. Summary of the ratio of 95th percentile impacts associated with the two ingestion
models for the networks for two dose levels. Summaries are provided by network, by network
area, network population, and average population density for each network.
-------
Sensitivities Examined
187
Percentile = 95
1.0-
0.8-
0.6-
0.4-
^-
S 0.2-
E
Q
s= 0.0-
CN
•a
c
CD
1
d. 1.0-
_CD
t_
O 0.8-
c
o
"G
CD 0.6-
LJ_
0.4-
0.2-
0.0-
0
0 ° °
O
0
o
0 0
0
0 A
A A A A
A A
A A A
A
1 1 1 1 1 1 1 1 1 1 1 1
123456789 10 12
Network
o
O 0.0001
0 A 1.0
0
0 0
o
0 A
AA A
A * A
4.0 4.5 5.0 5.5
Iog10(population)
6.0
2.5 3.0 3.5 4.0
Iog10(pop. density, persons/km2)
4.5
Figure 10.6. Summary of the overlap of 95th percentile and higher injection nodes associated
with 1-h and 24-h injections for the networks for two dose levels. Summaries are provided
by network, by network area, network population, and average population density for each
network.
-------
188
Sensitivities Examined
Percentile = 95
1.0-
0.8-
0.6-
o
o 0.4-
cb
•a
ro 0.2-
o
o
o
"o 0.0-
O>
P
o
| 1.0-
c*
CL
-g 0.8-
o
§ 0.6-
O °
0 °
o
0 °
0 0
0
A 0 A A
A
A A A A
A A
1 1 1 1 1 1 1 1 1 1 1 1
123456789 10 12
Network
o °
cP
o
0 °
° o
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
1.0-
0.8-
0.6-
O °
0 °
o
0 0
0 0
0
0 A A
A A
* A
A
1 1 1 1 1
1.5 2.0 2.5 3.0 3.5
Iog10(area, km2)
Dose Level (mg)
0 ° o 0.0001
0 ° A 1-0
0 °
O °
0.4-
0.2-
0.0-
0.4-
0.2-
0.0-
4.0 4.5 5.0 5.5 6.0
Iog10(population)
2.5
3.0
3.5
4.0
4.5
Iog10(pop. density, persons/km2)
Figure 10.7. Summary of the overlap of 95th percentile and higher injection nodes associated
with injection times of 0:00 and 6:00 for the networks for two dose levels. Summaries are
provided by network, by network area, network population, and average population density
for each network.
-------
Sensitivities Examined
189
Percentile = 9!
1.0-
0.8-
0.6-
o
* °'4-
•a
ro 0.2-
o
o
o
° 0 °
o
o
o
0 0
o
A
A A A A
A A A A
A
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
123456789 10 12
Network
0
O
0 0
O
o
o
4
A A
0
O
o
o
o
A
1.5 2.0 2.5 3.0 3.5
Iog10(area, km )
o
o>
o
CO
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
O
A
° 0
*
0
0 0
0
\ * A A
- A A "
4.0 4.5 5.0 5.5 6.0
Iog10(population)
1.0-
0.8-
0.6-
0.4-
0.2-
0
o
o
0
A
A A A
A
Dose Level (mg)
0 0.0001
A 1.0
0
A
2.5 3.0 3.5 4.0
Iog10(pop. density, persons/km2)
4.5
Figure 10.8. Summary of the overlap of 95th percentile and higher injection nodes associated
with injection times of 0:00 and 12:00 for the networks for two dose levels. Summaries are
provided by network, by network area, network population, and average population density
for each network.
-------
190
Sensitivities Examined
Percentile = 95
1.0-
0.8-
0.6-
o
•a
co 0.2-
o
o
o
"o °-°"
CD
E
i-
o
.8 1.0-
CL
ro 0.8-
O
c 0.6-
o
"8
CO
LL 0.4-
0.2-
0.0-
O
0
0 0
0
o
o
A °
A A
A A
A A A
A
1 1 1 1 1 1 1 1 1 1 1 1
123456789 10 12
Network
o
o
0°
0 0
0 °
0
0
A°
A A
AA
A A A
A A
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
O
0
0
0 0
0
o
o
A °
A A
A A
A A
A
A
1 1 1 1 1
1.5 2.0 2.5 3.0 3.5
Iog10(area, km2)
O Dose Level (mg)
0 0 0.0001
° 0 A 1.0
0 0
0
0
A A
A A
A1 A
A A A
4.0 4.5 5.0 5.5 6.0
Iog10(population)
2.5 3.0 3.5 4.0 4.5
Iog10(pop. density, persons/km2)
Figure 10.9. Summary of the overlap of 95th percentile and higher injection nodes associated
with injection times of 0:00 and 18:00 for the networks for two dose levels. Summaries are
provided by network, by network area, network population, and average population density
for each network.
-------
Sensitivities Examined
191
Percentile = 95
1.0-
0.8-
0.6-
0.4-
0)
(/5 0.2-
CD
2
X o.O-
•a
c
CD
X
_§" 1.0-
i_
| 0.8-
o
'•s
Ul 0.6-
0.4-
0.2-
0.0-
O
o o o o
00° 0
0 0
A
O
0 A
A *
A
AAA* AA A
1 1 1 1 1 1 1 1 1 1 1 1
123456789 10 12
Network
0
OD ° Q
0 0
A
0
0 A
A A
A ^AA A AA
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
O
o 8 0° °
0 0
A
O
A 0
A A
A A A ^ AA A
1 1 1 1 1
1.5 2.0 2.5 3.0 3.5
Iog10(area, km2)
O Dose Level (mg)
% O 0.0001 o
o° o °°
A
0
0 A
A A
Tk A^A
4.0 4.5 5.0 5.5 6.0
Iog10(population)
2.5 3.0 3.5 4.0 4.5
Iog10(pop. density, persons/km2)
Figure 10.10. Summary of the overlap of 95th percentile and higher injection nodes associ-
ated with IX and 10X injection masses for the networks for two dose levels. Summaries are
provided by network, by network area, network population, and average population density
for each network.
-------
192
Sensitivities Examined
Percentile = 95
1.0-
0.8-
0.6-
0.4-
(/5
CD
E 0.2-
^
c
£ 0.0-
CO
Q_
£
o
d. 1.0-
~
O 0.8-
o
co 0.6-
LL
0.4-
0.2-
0.0-
0
O
0 00 ° 0 °
o
0
0 A A
0 A *
A A A
A A
A
A
A
1 1 1 1 1 1 1 1 1 1 1 1
123456789 10 12
Network
o
000 ° 0Q°
0
o
0 A A
A
0 AA
AAA
A
A
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
0
O
0 0 0 0Q °
O
0
A 0 A
AO A
A A
A
A
A
1 1 1 1 1
1.5 2.0 2.5 3.0 3.5
Iog10(area, km2)
O Dose Level (mg)
0 o 0 0.0001
%°?P
O
0 A A
A
4) A
A A A
A
4.0 4.5 5.0 5.5
Iog10(population)
6.0
2.5
3.0
3.5
4.0
4.5
Iog10(pop. density, persons/km)
Figure 10.11. Summary of the overlap of 95th percentile and higher injection nodes for the
networks for the two population models for two dose levels. Summaries are provided by
network, by network area, network population, and average population density for each
network.
-------
Sensitivities Examined
193
Percentile = 95
1.0-
0.8-
0.6-
0.4-
"CD
•a 0.2-
o
to
CD
O)
O
°- 1 0-
co
0 0.8-
o
"8
2. 0.6-
LJ_
0.4-
0.2-
0.0-
O
0°° °00°0
o o
o
A
A
* * *
1 1 1 1 1 1 1 1 1 1 1 1
123456789 10 12
Network
0
o°°0 °° °°°
o °
0
A
A
'^ A
A
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
1.0-
0.8-
0.6-
0.4-
0.2-
0.0-
O
o o o _
o o
o
A
A
A A * A
1 1 1 1 1
1.5 2.0 2.5 3.0 3.5
Iog10(area, km2)
0 ig, 0 Dose Level (mg)
0 0*° 0 0.0001
A 1.0
0
A
A
A +
A ^^ A
4.0 4.5 5.0 5.5
Iog10(population)
6.0
2.5
3.0
3.5
4.0
4.5
Iog10(pop. density, persons/km2)
Figure 10.12. Summary of the overlap of 95th percentile and higher injection nodes for the
networks for the two ingestion models for two dose levels. Summaries are provided by net-
work, by network area, network population, and average population density for each net-
work.
-------
194 Sensitivities Examined
-------
Section 11
Discussion
11.1 Methodological Issues
Importance of Dose Level. We have considered dose level as a parameter in our analysis of
sensitivity. Instead of using dose level as a parameter, we could use median lethal dose as a factor
with several levels, corresponding to different contaminants. Impact, the number of people exposed
above some dose level, is very sensitive to the dose level and, consequently, is also very sensitive
to the value used for median lethal dose. As can be seen in Figures 3.1, 3.2, and 3.3, for 95th
percentile impacts, varying dose level from 0.0001 to 10 mg results in impacts changing by over
two to over five orders of magnitude, depending on the network. Figures 11.1 through 11.4 provide
maps of Network 4 comparing the locations of high percentile injection nodes for dose levels of
0.001, 0.01, 0.1, and 1.0 mg with locations for a dose level of 0.0001 mg, using our baseline
cases for factors. The locations of the injection nodes change greatly as the dose level increases.
No overlap of the injection nodes occurs for dose levels of 0.0001 mg and either 0.1 or 1.0 mg.
What are considered important nodes for high toxicity contaminants are generally not important
nodes for low toxicity contaminants and any strategy for network protection needs to consider this
difference.
Use of NZD Nodes as Injection Nodes. As noted above, we used NZD nodes as injection nodes
because we believe that they represent the most likely injection locations. However, to assess the
significance of our choice of injection nodes, we estimated impacts for a number of the networks
using all nodes as potential injection nodes, in addition to considering only NZD nodes. We ex-
amined Networks 1, 4, 5, 7, and 12, which are the networks for which NZD nodes constitute the
smallest fraction of all network nodes. We found that the results for impacts versus dose level (as
illustrated by Figures 3.1, 3.2, and 3.3) obtained using all the nodes did not differ significantly from
those obtained using only NZD nodes. (A quantitative comparison is provided in the supplemental
material for Davis and Janke (2010).)
Injection at Terminal Nodes. Distribution system models consist of nodes with a demand and
those with no demand. For the distribution system models examined here, demand nodes generally
represent groups of service connections. Depending on the level of model detail, the number
of service connections per node ranges from approximately 5 to 200. EPANET, and therefore
also TEVA-SPOT, model contaminant injection as mass injected per unit time (mg/min), with the
195
-------
196 Discussion
assumption that any volume changes are insignificant. TEVA-SPOT does not have the ability to
model volume additions to pressurized pipe flow. If a model has sufficient detail that all NZD
nodes in a model represent terminal nodes in the network or individual service connections, people
located at these nodes will only receive water from the distribution system but never release water
back to the network. As a result, simulating an injection of a contaminant at these nodes in TEVA-
SPOT will result in only people at the injection node being affected.
For example, if an injection of contaminant occurs at a terminal node at hour 0, the entire mass
of the injected contaminant is removed from the system within the hour (assuming a demand is
specified that results in the removal of water) and unless ingestion occurs within the hour that the
contaminant is introduced no one is affected. For the hourly, demand-based timing model (used
in Model 2), the number of people affected will equal the number of people at the node. For the
probabilistic timing model (used in Model 1), however, only a fraction of the people at the node
will be affected since not all individuals will ingest water during the hour that the contaminant was
introduced.
An example of the issue involving terminal nodes occurs for the synthetic model named Mi-
cropolis (Brumbelow et al. 2007), but will occur with any model having nodes that only receive
water. It is particularly evident with Micropolis because all the NZD nodes represent individual
service connections or terminal nodes in the network, which only receive water. We modeled both
the non-zero and zero demand nodes in Micropolis and found that, unlike the case for the 12 real
systems, simulation of contamination events at the NZD nodes produced very small or near zero
impacts for the various dose levels using ingestion Model 1. However, our simulation of contam-
ination events at the zero demand nodes resulted in impacts generally consistent with the impacts
obtained for the 12 real systems.
As a result, TEVA-SPOT simulations of contaminant injections at terminal nodes or nodes
that only receive water will likely underpredict impacts for two reasons. First, not everyone at
such nodes who would likely ingest contaminated water may be determined to be exposed due to
removal of all contaminated water via demand at the node before ingestion occurs. Second, in
the models, contaminants injected at terminal nodes have no means for conveyance back into the
distribution system. Any ingestion of the contaminant occurs at the injection node.
The inability to properly determine impacts associated with injections at terminal nodes results
in some uncertainty in our results, which will be network specific. The portion of NZD nodes in
our 12 networks that are terminal nodes ranges from about 1% for Network 7 to about 27% for
Network 3. The median value is 12.8%. Impacts associated with injections at terminal nodes tend
to be underestimated and, therefore, impacts at some percentile levels for the networks also will
tend to be underestimated. However, unless these nodes are located such that associated impacts
are larger than would be expected for randomly located nodes, the magnitudes of high percentile
impacts should not be much affected. The actual significance of any underestimate likely depends
on the network, both because of the variation in the fraction of such nodes in the networks and
their location. Quantifying the significance of any underestimate requires modifying the network
models.
-------
Discussion 197
11.2 Computational Issues
Computational Effort. The analyses needed for this report required a large number of simu-
lations and a large amount of computational time. Additionally, a considerable amount of data
storage was needed for the processing of the simulations and analysis of results. The cases an-
alyzed for the 12 networks required the simulation of over 1.6 million contamination events and
occupied approximately 1.3 terabytes of storage. Archiving capabilities available within TEVA-
SPOT significantly reduced the storage requirements.
For a computer with a single processor (with approximately 3 gigahertz processor speed) and
with 2 gigabytes of memory, TEVA-SPOT impact-analysis runtimes per simulation vary consider-
ably for the 12 networks, from approximately 30 seconds (Network 1) to approximately 6 minutes
(Networks 10 and 11). The more than 1.6 million simulations for the 12 networks would have
taken approximately 100,000 hours (11 years) to complete on a single computer with one pro-
cessor. Given the distributed computing capabilities utilized for this work (76 processors), total
runtime was reduced considerably, to approximately 1,300 hours (54 days).
Expanding our analysis of the interaction of factors, particularly to use a Monte Carlo approach
for all of the 12 networks, encounters computational difficulties. For Network 2, 32 combinations
of factors were considered, requiring 32 separate simulations for each of the 3,000 NZD nodes in
the network. For the other 11 networks, only 12 such sets of simulations were done. These simu-
lations required about 60% of the processor time for this study, requiring a total runtime of about
30 days. Expanding these simulations to include all 32 combinations of factors for all 12 networks
would therefore require about another 50 days of runtime. Addition of another factor doubles the
required runtimes. To include a Monte Carlo approach for all of the networks could increase the
total number of simulations by a factor of 1,000 or more, resulting in total runtimes of months or
longer on all but very large distributed computing systems. Even for Network 2 alone, for which
about 11 hours of runtime were required to examine the 32 combinations of factors considered, use
of a Monte Carlo approach would require large runtimes on available computational resources.
Water Quality Tolerance Parameter. The EPANET water quality tolerance parameter specifies
the smallest change in water quality that will result in a new parcel of water being created in a pipe.
The default setting in EPANET is 0.01 mg/L for chemicals (Rossman 2000). The water quality
tolerance determines whether the quality of one parcel of water is the same as that of another
parcel. For a chemical contaminant the value could be the detection limit of the analytical method
used to measure the chemical contaminant, adjusted by a suitable safety factor (Rossman 2000).
Generally, lowering the water quality tolerance will provide increased accuracy of the water quality
modeling results.
We examined the influence of the water quality tolerance value on estimated impacts for six
networks (1, 2, 4, 5, 7, and 12). Mean impacts and 50th to 100th percentile impacts were determined
for each network using the default water quality tolerance value (0.01 mg/L) and using values of
10"6 and zero. For the lowest dose level, differences in impacts ranged from about 1% to 12%,
depending on the network and the percentile of the scenario. For higher dose levels, differences
were generally 1% or less. For a few scenarios for high dose levels of 1 mg or greater, we ob-
served larger differences (20% to 40%). A component of the differences observed is a result of the
randomness in the impacts estimated using the probabilistic exposure model. Given the relative
nature of the sensitivity results, these differences do not affect the conclusions reported.
-------
198 Discussion
The differences between the impacts obtained using water quality tolerance values of 10~6 and
zero are generally much smaller than the differences between the default value and zero, indicating
that better results can be obtained without using the more computationally burdensome value of
zero for water quality tolerance. The water quality tolerance value selected should be aligned with
the chemical contaminant being modeled and its associated mass release rate.
11.3 Application of Results
Ultimately, the value of the sensitivity analysis presented here relates to its significance for the de-
sign of contamination warning systems and network protection. CWSs are designed to minimize
the adverse effects associated with a contamination event. Therefore, estimating such adverse ef-
fects is an important part of any CWS design. In general, CWSs have not been designed with
attention given to the importance of the various factors involved in estimating adverse effects. The
sensitivity analysis presented here should help designers of CWSs to also consider (1) the various
factors involved in defining a contamination event, (2) the possible uncertainties associated with
establishing the distribution of the population within a network and estimating ingestion doses,
(3) the dose level (or contaminant) used in determining impacts, and (4) network-to-network vari-
ability.
Although it is tempting to want to use results obtained with a single network model (or multiple
models) as the basis for general conclusions about the response of water distribution systems to
the intentional injection of contaminants, this study demonstrates that great variability in responses
can occur for different systems. Although definite patterns exist in responses that extend across
networks, as is shown here and in Davis and Janke (2010), substantial inter-network variability lim-
its the current ability to generalize quantitative results from a limited sample to new applications.
Individual water distribution systems should be treated as unique systems with some similarities
in the magnitudes and patterns of impacts during contamination events.
11.4 Additional Work
The work presented here could be expanded by including additional water distribution systems,
other factors affecting the estimated impacts, and a Monte Carlo analysis of impacts for the com-
bination of factors. The current analysis already includes a diverse group of systems. The major
value of analyzing multiple systems is to demonstrate the importance of inter-network variability
in the sensitivity to the factors studied. This variability is demonstrated by the group of 12 systems
examined in this study. Unless additional systems having specific characteristics of interest are
identified or the goal is to examine statistically how sensitivity depends on some system character-
istic (e.g., use of skeletonization in the network model), the value of including additional systems
is limited. Other factors are involved in the simulations used to obtain impacts that could be in-
cluded in a sensitivity analysis. In particular, we did not consider explicitly any factors related to
the network model or to the approach used to determine flows and water quality in the networks.
(We did check for possible effects of network population, area, and population density on our re-
sults and found no apparent influence.) The emphasis here is on the nature of the injection and
the exposure process. As noted above, use of a Monte Carlo approach to examining interactions
-------
Discussion 199
among factors encounters computational difficulties for the large networks included in this study.
However, as computational abilities improve, use of such an approach may become practical.
Two related areas where additional work is needed are the expansion of the analysis of impacts
to include the effects of contaminant decay or removal in the network and the sensitivity of the
design of CWSs to the various factors influencing estimated impacts. Decreases in contaminant
concentrations due to processes other than dilution could have a significant effect on estimated
impacts and the sensitivity of estimated impacts to such processes should be considered. CWS
designs could be significantly affected by variations in the factors considered here and by the
nature of the contaminant, as well as by any nonconservative behavior of the contaminant in the
water distribution system. CWSs may perform differently if challenged by contaminants with low
or high median lethal doses, for example. We are currently pursuing additional work in these two
areas.
-------
200
Discussion
e
Injection Nodes
• 0.001 mg
• 0.0001 mg
Figure 11.1. Network 4 showing locations of injection nodes associated with 95th percentile
or higher impacts at dose levels of 0.0001 mg and 0.001 mg. The baseline case was used for
all factors.
-------
Discussion
201
%
* A
1' o +
B
Injection Nodes
o 0.01 mg
• 0.0001 mg
Figure 11.2. Network 4 showing locations of injection nodes associated with 95th percentile
or higher impacts at dose levels of 0.0001 mg and 0.01 mg. The baseline case was used for all
factors.
-------
202
Discussion
Orv
O
Q O
/
As
O
-------
Discussion
203
00
_
0 OB
,
Injection Nodes
o 1.0 mg
O 0.0001 mg
Figure 11.4. Network 4 showing locations of injection nodes associated with 95th percentile
or higher impacts at dose levels of 0.0001 mg and 1 mg. The baseline case was used for all
factors.
-------
204 Discussion
-------
Section 12
Conclusions
This analysis considered five important factors related to assessing potential impacts associated
with contamination events in water distribution systems: injection duration, injection time, in-
jection mass, the distribution of population in the network, and the ingestion model. Estimated
impacts are sensitive to all these factors, in varying amounts that depend on the network. When the
factors are considered together (not including the ingestion model), depending on the dose level,
the magnitudes of impacts are most sensitive to injection mass or duration:
• At low dose levels, impacts are most sensitive to injection duration, although the relative
changes in impacts due to changes in duration may not be large for high percentile impacts.
Impacts are larger for longer duration injections and the increases tend to be more important
for lower percentile impacts.
• At low dose levels, impacts are not particularly sensitive to injection mass, given a likely
range in injection masses.
• At higher dose levels, impacts are most sensitive to injection mass, with impacts increasing
for larger injection mass.
The network-to-network variability in the sensitivity to the various factors considered here can
be large. We did not find a correlation between any network-related feature and the sensitivities
observed for the various networks. However, we examined only a small number of such features.
The influence of the various factors on the location of high percentile injection nodes can be as
important or more important than their influences on the magnitudes of impacts. In addition, dose
level has a major influence on the location of high percentile injection nodes.
The sensitivity of the location of high percentile injection nodes to the various factors is greatest
at high dose levels.
Injection time, duration, and mass, as well as uncertainties in the distribution of population in
a network and in the model used to estimate ingestion dose can all have a significant influence on
estimated impacts for a contamination event. Designers of contamination warning systems need to
consider the possible influences of all of these factors on their results, as well as the importance of
network-to-network variability and the significance of the dose level or contaminant being used.
205
-------
206 Conclusions
-------
References
Brumbelow, K., Torres, J., Guikema, S., Bristow, E., and Kanta, L. (2007). "Virtual cities for
water distribution and infrastructure system research." Proceedings of the World Environmental
and Water Resources Congress 2007, American Society of Civil Engineers. Tampa, Florida.
DOI:10.1061/40927(243)469.
Davis, M. J., and Janke, R. (2008). "Importance of exposure model in estimating impacts when a
water distribution system is contaminated." Journal of Water Resources Planning and Manage-
ment, 134(5), 449-456.
Davis, M. J., and Janke, R. (2010). "Patterns in potential impacts associated with contamination
events in water distribution systems." Journal of Water Resources Planning and Management,
(16 March 2010), 10.1061/(ASCE)WR.1943-5452.0000084
Hintze, J. L., and Nelson, R. D. (1998). "Violin plots: a box plot-density trace synergism." The
American Statistician, 52(2), 181-184, May.
Isovitsch, S.E., and VanBriesen, J.M. (2008). "Sensor placement and optimization criteria depen-
dencies in a water distribution system." Journal of Water Resources Planning and Management,
134(2), 186-196.
Khanal, N. (2005). Generalized sensitivity analysis of water distribution system vulnerability to
deliberate intrusions, Master of Science thesis, University of Cincinnati, Cincinnati, Ohio.
Khanal, N., Buchberger, S. G., and McKenna, S. A. (2006). "Distribution system contamination
events: exposure, influence, and sensitivity." Journal of Water Resources Planning and Manage-
ment, 132(4), 283-292.
LaTeX project (2010). "LaTeX - a document preparation system." http://www.
latex-project.org/
Morley, K., Janke, R., Murray, R., and Fox, K. (2007). "Drinking water contamination warning
systems: water utilities driving water security research." Journal of the American Water Works
Association, 99(6), 40-46.
Murray, R.E., Grayman, W.M., Savic, D.A. and Farmani, R. (2009). "Effects of DMA redesign on
water distribution system performance." Integrating Water Systems, Boxall, J. and Maksimovic,
C. (eds.), Taylor and Francis, London, pp. 645-650.
207
-------
208 REFERENCES
Nilsson, K. A., Buchberger, S. G., and Clark, R. M. (2005). "Simulating exposures to deliberate
intrusions into water distribution systems." Journal of Water Resources Planning and Manage-
ment, 131(3), 228-236.
Ostfeld, A., et al. (2008). "The Battle of the Water Sensor Networks (BWSN): A design challenge
for engineers and algorithms." Journal of Water Resources Planning and Management, 134(6),
556-568.
R Development Core Team (2010). "R: A language and environment for statistical computing."
R Foundation for Statistical Computing, Vienna, Austria, http : / /www. R-pro ject. org
(Feb. 21, 2010).
Rossman, L. A. (2000). "EPANET 2 users manual," EPA/600/R-00/057, U.S. Environmental Pro-
tection Agency, National Risk Management Research Laboratory, Office of Research and De-
velopment, Cincinnati, Ohio.
U. S. EPA (U. S. Environmental Protection Agency). (2000a). "Estimated per capita water inges-
tion in the United States," EPA-822-00-008, Office of Water, Washington, D.C.
U. S. EPA. (2000b). "Policy and program requirements for the mandatory agency-wide qual-
ity system." EPA Order CIO 2105.0. Available at http://www.epa.gov/irmpoli8/
policies/21050.pdf
U. S. EPA. (2000c). "EPA quality manual for environmental programs." CIO 2105-P-01-0. Avail-
able at http://www.epa.gov/irmpoli8/policies/2105PO10.pdf
U. S. EPA. (2010a). "Threat ensemble vulnerability assessment research program." http://
www. epa . gov/nhsrc/water/teva . html (April 15, 2010).
U. S. EPA. (2010b). "Policy and Guidance: Guide to Requirements for Research Reports Submit-
ted to the National Homeland Security Research Center." http : / /www. epa . gov/nhsrc/
guidance . html (April 26, 2010).
U. S. EPA. (2010c). 'Threat Ensemble Vulnerability Analysis - Sensor Placement Optimization
Tool (TEVA-SPOT) Graphical User Interface Users Manual (2010)," version 2.2.0 BETA, EPA
600/R-08/147, U.S. Environmental Protection Agency, Office of Research and Development,
National Homeland Security Research Center, Water Infrastructure Protection Division, Cincin-
nati, Ohio. Available at http : //www. epa . gov/nhsrc/pubs/600r08147 .pdf
Watson, J-P, Murray, R., and Hart, W. E. (2009). "Formulation and optimization of robust sensor
placement problems for drinking water contamination warning systems." Journal of Infrastruc-
ture Systems, 15(4), 330-339.
-------
Appendix: Quality Assurance
The primary objective of the United States Environmental Protection Agency (EPA) Quality Assur-
ance requirements is to ensure that all environmental data assembled, models and methodologies
developed, and analytical results obtained are of sufficient quantity and quality to support the con-
clusions and observations produced. The EPA Quality System is defined by EPA Order CIO 2105.0
(U. S. EPA 2000b) and is described in the EPA Quality Manual for Environmental Programs (U. S.
EPA 2000c). Quality assurance (QA) is the planned and systematic processes implemented to pro-
vide confidence in the results produced and the suitability of those results. Quality control (QC)
is defined by the procedures implemented to ensure that the results and conclusions adhere to the
defined set of quality criteria.
We implemented quality assurance and quality control methodologies and procedures at each
step of our research project to ensure that the results produced are based on state-of-the art science
and engineering principles, are reproducible, and suitable for estimating adverse public health
effects associated with water distributions systems given an intentional contamination event.
All software development, research, model and algorithm development, data collection and
analyses, and report writing were performed by the authors. QA/QC practices were implemented
at each of the following phases of the research and report preparation:
• Research study (purpose and goals)
• Algorithmic and methodological model development
• Software engineering
• Model and software verification
• Computational modeling and simulation
• Data analysis and report preparation
In following paragraphs, we discuss the QA/QC practices implemented in each of these phases.
Research Study (purpose and goals.) The proper design of CWSs for water distribution sys-
tems depends on a thorough understanding of how adverse public health effects associated with
contamination events are influenced by the major factors that define a contamination event. We
conducted a literature review and found no information available for quantifying the sensitivity of
estimated adverse health effects to the major factors defining intentional contamination events for
a range of real water distribution systems. We summarize our findings of the literature study in
the introduction. Given the lack of such information, we formulated a research plan, the technical
209
-------
210 Appendix: Quality Assurance
portion of which is documented in this report. This plan was used to investigate the sensitivity of
adverse effects associated with an intentional contamination event to the major factors defining the
event and the major assumptions made in assessing exposure to the contaminant. In developing
the research plan, we fully described the purpose and goals of the study. Separately, we identified
potential customers and stakeholders (water utility partners, water industry consultants, American
Water Works Association, EPA Office of Water, Department of Homeland Security, and Depart-
ment of Defense (military installations) who would likely find the research results and findings
useful (Morley et al. 2007). Our approach to the study relied on established software to conduct
simulations and established science and engineering principles.
Algorithmic and Methodological Model Development. We relied on the established EPANET
(Rossman 2000) and TEVA-SPOT (U. S. EPA 2010a) software programs. We used only the most
up-to-date versions of the programs for the results included in this report. Our set of real and oper-
ationally valid water distribution system models were obtained from collaborating water utilities.
Our approach for assessing adverse effects was previously developed and described (Davis and
Janke 2008, 2010). Our exposure models for the timing of ingestion of contaminated tap water
were previously developed and described ((Davis and Janke 2008). Model input parameters were
specified in each water system's network model. Although network model parameters were not in-
dependently verified, each model represents a real, operationally valid system. Three of the water
system models utilized have previously been reported on in peer-reviewed publications. Our data
analysis, including all statistical analyses, relied on established methods and procedures, with all
such procedures fully described and referenced in this report.
Software Engineering. TEVA-SPOT is a stand-alone program developed to study contamination
vulnerability assessment and sensor network design for water distribution systems. TEVA-SPOT
was developed based on EPANET and is maintained to be consistent with the latest version of
EPANET. EPANET (version 2.00.12) is integrated into TEVA-SPOT through the EPANET Pro-
grammer's Toolbox application. Several important differences exist between the execution of
EPANET within TEVA-SPOT as opposed to the execution of EPANET through its stand-alone
graphical user interface. These differences, important to the assessment of exposure and risk, in-
clude: (1) TEVA-SPOT's ability to average contaminant water quality concentration data over the
reporting interval and (2) TEVA-SPOT's use of the STEP Q function instead of NEXT Q function.
Details on these EPANET functions are discussed in the EPANET users manual (Rossman 2000).
A mass balance feature was incorporated into TEVA-SPOT to allow the tracking of contam-
inant mass, i.e., contaminant mass injected, mass removed, mass in pipes and mass in tanks, in
order to help verify the estimated impacts for reasonableness. The mass balance tracking feature
supported the selection of a 1-minute water quality time step. Early simulations using a 5- minute
water quality time step resulted in mass balance discrepancies.
Updates to TEVA-SPOT are independently verified with hand or spreadsheet calculations. All
of our simulations were implemented on a distributed high performance computing system that
includes 38 dual processor compute nodes for a total of 76 processors. TEVA-SPOT is openly
distributed by request through the TEVA Research Program web site (U. S. EPA 2010a).
We maintain a subversion repository for all TEVA-SPOT and EPANET computer code. Our
subversion repository tracks all changes made to the software by person, date, and time and allows
for regression to earlier versions if needed.
-------
Appendix: Quality Assurance 211
Model and Software Verification. Our water utility models were provided by partnering water
utilities. No independent verification of model input parameters was performed. Models were run
in TEVA-SPOT using the input parameters specified in the models. Network distribution system
models were run for water-age analysis to help determine appropriate simulation runtimes. Models
obtained from partnering water utilities met the requirements specified in the TEVA-SPOT Users
Manual, Appendix A, "Water Utility Requirements for Using EPA's TEVA-SPOT Software" (U. S.
EPA 2010c). Input parameters for TEVA-SPOT were developed consistent with the approach
described in Section 2. Contamination scenarios for the 12 networks were defined consistent with
Section 2 and implemented in TEVA-SPOT on a network-by-network basis. Collections of similar
scenarios were designated as an ensemble and run in TEVA-SPOT. Each ensemble was uniquely
named for easy recognition and verification of input parameters.
Using the ensemble naming structure, we routinely verified input parameters to ensure consis-
tency with the network model (e.g., Network 1 versus Network 2) and required input parameters,
e.g., population and ingestion model. Each executed TEVA-SPOT ensemble produced a uniquely
named output file of dose-over-threshold results.
Our maps were created by exporting health impacts assessment results as Database IV (dBase®
IV) data files from TEVA-SPOT and then merging and mapping the results in ArcGIS® 9.3.
We regularly backed up output consistent with EPA information technology and information
management requirements. Additionally, we regularly archived dose-over-threshold data files to a
database. We used these data files for verification of the consistency of dose-over-threshold results
following updates to TEVA-SPOT.
Computational Modeling and Simulation. Given the large number of TEVA-SPOT ensembles
executed, verification of input parameters for all simulations was a significant requirement and
was routinely performed. When input errors or problems were found, the ensemble(s) was re-run
and the analyses were updated with the new results. Results from early runs were compared with
results from later runs to ensure consistency of results following any modifications of software.
The structure of the TEVA-SPOT software facilitated best management QA/QC practices:
• The database structure of TEVA-SPOT output provides organization and management of
ensembles for easy data retrieval and analysis.
• TEVA-SPOT provides management of ensembles by providing date status information on
each ensemble, e.g., (1) date the ensemble was last saved, (2) date ensemble was executed
last, (3) date that health impacts assessment was executed last, and (4) size of the ensemble.
• Each dose-over-threshold output file is uniquely and automatically named by TEVA-SPOT
to link to its parent ensemble name.
• Each ensemble executed in TEVA-SPOT is documented with an output of the network model
used in the simulation, which was then used for QA/QC purposes to verify that the ap-
propriate model was used. The network model and its associated input parameters appear
in the directory structure specific to the ensemble, e.g., TEVA-SPOT-Database/Collection
Name/BaseEnsemble.
-------
212 Appendix: Quality Assurance
• The TEVA-SPOT archival feature provides the ability to archive ensemble data files while
preserving the ensemble definition and description along with the results. All archived data
are fully retrievable, if needed.
Data Analysis and Report Preparation TEVA-SPOT provides dose-over-threshold results in
a text file with column number 1 corresponding to the node identification name and columns 4
through N corresponding to the impacts for the dose thresholds specified in TEVA-SPOT. The
thresholds or dose levels appear as a header in row number 1.
TEVA-SPOT provides some statistical analyses. For example, the identification of the 10th,
25th, 50th, 75th, 90th, and 100th percentile scenarios (identification of injection node) for each en-
semble and the associated receptors affected are provided by TEVA-SPOT. Additional analyses
were developed using Microsoft Excel® and R (R Development Core Team 2010). All plots in
this report, e.g., scatter and histogram, were developed separately from TEVA-SPOT using R. Se-
lected results and statistical analyses were independently verified using hand calculations. Maps
were created by exporting health impacts assessment results as dBase® IV data files from TEVA-
SPOT and then merging and mapping the results in ArcGIS® 9.3. Report writing was version
controlled with a revision number and date. Reviewer comments were addressed and tracked by
revision number. The report was prepared using ETgX 2£ (LaTeX Project 2010). Report formatting
is consistent with EPA guidelines (U. S. EPA 201 Ob).
------- |