Exploring the effects of environmental characteristics and anthropogenic activities on mosquito
populations: an experimental and spatially explicit model-based approach
by
Daniel Eugene Dawson, BSc, MSc
A Dissertation
In
Environmental Toxicology
Submitted to the Graduate Faculty
of Texas Tech University in
Partial Fulfillment of
the Requirements for
the Degree of
DOCTOR OF PHILOSOPHY
Approved
Christopher Salice, Ph.D.
Chair of Committee
Steven Presley, Ph.D.
Todd Anderson, Ph.D.
Blake Grisham, Ph.D
Mark Sheridan
Dean of the Graduate School
August, 2016
© 2016, Daniel Dawson
Texas Tech University, Daniel Dawson, August 2016
ACKNOWLEDGMENTS
Thanks to my advisor Dr. Christopher Salice for offering me this opportunity, and
for providing feedback and support on research questions and approaches all along the way. I
also would like to thank Dr. Steven Presley, who accepted me into his lab group and served as
my official advisor when Chris moved to Towson University. Thanks also to my other committee
members, Dr. Todd Anderson and Dr. Blake Grisham, for providing help on chemical analyses
and modeling questions, respectively. In addition to my committee, thanks to the wonderful staff
at The Institute of Environmental and Human Health (TIEHH) that made my time as a graduate
student easier and more enjoyable. In particular, thanks to Stephanie White for administrating
the clerical aspects of the EPA STAR Fellowship, Allyson Smith for cheerfully handling
numerous graduate school-related administrative tasks, and to Lori Gibler and Brad Thomas for
helping me with computer needs.
I would like to acknowledge the persons and entities that collaborated with me on this
research. First, thanks to Tarrant County Public Health (TCPH), and cities of Arlington, Haltom
City, Burleson, Colleyville, Southlake, and North Richland Hills in Tarrant County, Texas, for
providing mosquito control records for use in this dissertation. I thank Nina Dacko of TCPH in
particular for acquiring and compiling these records for my use. Next, thanks to Lubbock County
Vector Control for providing insight into mosquito control, and for providing me some of the
surveillance records used in this dissertation. Lastly, thanks to Lucas Heintzman for serving as
my assistant during the summer of 2015, for providing me the landcover dataset used in this
dissertation, and for generally being a source of helpful suggestions and feedback.
Thanks to my lab-mates Scott Weir, Thomas Bilbo, Evelyn Reátegui-Zirena, Adric Olson
and Bridget Fidder for helping with research, as well as being friends and comrades-in-arms. An
extra thanks to Thomas Bilbo, who was a fellow mosquito researcher while at TIEHH, and who
ii
Texas Tech University, Daniel Dawson, August 2016
graciously helped me in the field during the summer of 2014. In addition, thanks to the many
friends at TIEHH and in Lubbock that have made the last 4-5 years an enjoyable time. At
TIEHH, thanks in particular to Nick Dunham, Jordan Hunter, and Francis Loko for your
friendship, and the many interesting conversations we’ve had. Outside TIEHH, thanks especially
to Beth Watson, Neil Knauth, Tiffany Hettrick, and Chris Hopper for providing my wife and I a
like-minded and supportive social nucleus while here in Lubbock; you will be dearly missed.
Thanks also to my family for always encouraging me in the pursuit of my PhD. Importantly,
thanks to my wonderful and brilliant wife Elizabeth Farley-Dawson, who provided me a constant
sounding board for research ideas and concerns, and whose love and support made the worst
times here bearable, and the good times great.
Lastly, I would like to thank the several funding sources that made my research possible.
First, thanks to TIEHH for providing initial funding and research facilities over the course of my
studies. Thanks also to the CH Foundation and the ARCS Foundation for providing additional
funding. Lastly, I acknowledge EPA STAR Fellowship Assistance Agreement no. 91765301-0
for financial support in producing this research. Note that this publication has not been formally
reviewed by the EPA. The views expressed in this publication are solely those of the authors,
and EPA does not endorse any products or commercial services mentioned in this publication.
iii
Texas Tech University, Daniel Dawson, August 2016
TABLE OF CONTENTS
ACKNOWLEDGMENTS .......................................................................................................................... II
ABSTRACT ................................................................................................................................................ V
LIST OF TABLES..................................................................................................................................... VI
LIST OF FIGURES ................................................................................................................................. VII
I. INTRODUCTION .................................................................................................................................. 1
II. POPULATION EFFECTS OF MALATHION EXPOSURE AT DIFFERENT
TEMPERATURES AND AGES OF EXPOSURE IN AEDES AEGYPTI (DIPTERA: CULICIDAE)
..................................................................................................................................................................... 5
III. A MODEL OF MOSQUITO ABUNDANCE CONSTRUCTED USING ROUTINE
SURVEILLANCE AND TREATMENT DATA IN TARRANT COUNTY, TEXAS ......................... 25
IV. THE INFLUENCE OF WATER QUALITY AND PREDATION PRESENCE ON THE
RESPONSE OF CULEX TARSALIS LARVAE TO BACILLUS THURINGIENSIS ISRAELENSIS
(BTI) LARVICIDE .................................................................................................................................. 60
V. MODELING SPATIALLY EXPLICIT MOSQUITO POPULATION DYNAMICS WITH THE
RNETLOGO PACKAGE ......................................................................................................................... 96
VI. SUMMARY AND CONCLUSIONS .............................................................................................. 144
REFERENCES....................................................................................................................................... 149
iv
Texas Tech University, Daniel Dawson, August 2016
ABSTRACT
Mosquitoes pose risks to humans, both as nuisances and as vectors of disease. The
control of mosquito populations typically involves the use of chemical pesticides targeting both
the aquatic larvae (i.e, larvicides) and the terrestrial adults (i.e., adulticides). Both extrinsic (e.g.,
temperature) and intrinsic (e.g., species) factors influence the efficacy of these chemicals, and
their effects on life history characteristics and population dynamics. In the field of mosquito
control, primary goals are to assess mosquito populations in space, and to predict how they will
respond to pesticide applications given conditions. However, because many factors interact to
influence population dynamics, including pesticide exposure, this can be a challenging task. To
assist in this effort, spatially-explicit mathematical population models are promising tools that
can help mosquito control authorities by (1) providing insight into mosquito population dynamics;
(2) predicting mosquito populations in space; and (3) potentially providing quantitative estimates
of risks posed by mosquitoes.
This research had two main goals, including (1) to investigate how some common
environmental factors influence the effects that larvicides have on larval and adult life history
characteristics of two medically important mosquitoes, and (2) to develop spatially-explicit
mosquito population models with the potential of being utilized as operational tools. My research
revealed that (1) larvicides can interact with age, temperature, and water quality to affect the life
history characteristics of both larvae and adults; (2) these affects can potentially influence
population dynamics; (3) real-world larvicide and adulticide applications have significant but
local effects on population dynamics in the landscape; and 4) spatially explicit population
models have the potential to be useful tools for mosquito control, with the recommended model
structure depending upon the needs and capability of the mosquito control authority.
v
Texas Tech University, Daniel Dawson, August 2016
LIST OF TABLES
Table 2.1 Vital rates and traits for the first instar and fourth instar exposure scenarios. ............23
Table 2.2 Thirty-day population projections, intrinsic growth rates (λ), and control
normalized projections for both first and fourth instar exposure scenarios. ................................24
Table 3.1 All variables considered during the modeling process.. .............................................56
Table 3.2 All models evaluated using each modeling approach ................................................57
Table 3.3 Model selection weights based on AICc for each best selected model in each
category ....................................................................................................................................58
Table 3.4 Parameters of average-weighted GLMM model ........................................................59
Table 4.1 Values of water characteristics measured in playa and wastewater media ................92
Table 4.2 All vital rates (averages) measured in both experiments............................................93
Table 4.3 Results from all statistical analyses run for experiment 1. ..........................................94
Table 4.4 Results from all statistical analyses run for experiment 2...........................................95
Table 5.1 Model Parameters used in matrix and NetLogo models...........................................142
Table 5.2 Models and AICc model weights for models considered for the selection of
oviposition wetlands ................................................................................................................143
Table 5.3 Parameter estimates, standard errors, and p-values of model parameters of
negative binomial model developed for sensitivity analysis .....................................................143
vi
Texas Tech University, Daniel Dawson, August 2016
LIST OF FIGURES
Figure 2.1 Average emergence rates for experimentally-exposed Ae. aegypti. .........................19
Figure 2.2 Average times of emergence for experimentally-exposed Ae. aegypti......................20
Figure 2.3 Average wing lengths (mm) of experimentally-exposed female Ae. aegypti. ............21
Figure 2.4 Malathion concentrations (points) in water solutions over 7 days
(day 0 to day 6) at all temperatures studied during the experiment. ..........................................22
Figure 3.1 Collaborating municipalities and traps in unincorporated Tarrant County, TX,
operated by Tarrant County Public Health (TCPH) ....................................................................50
Figure 3.2 Standardized residuals of average weighted General Linear Mixed Model
versus fitted values ...................................................................................................................51
Figure 3.3 Plot of observed counts (log+1 scale) at all included traps (triangles) versus
counts predicted (crosses) by the model ...................................................................................52
Figure 3.4 Plots of observed counts versus counts predicted by the GLMM model ...................53
Figure 3.5 GLMM standardized model residuals against scaled population density (Pop)
and NDVI ..................................................................................................................................54
Figure 3.6 Observed count values and treatment events at example locations in
Burleson and Arlington ..............................................................................................................55
Figure 4.1 Experimental containers used in experiments 1 and 2 .............................................86
Figure 4.2 Average daily temperature and light during experiment 1 .........................................87
Figure 4.3 Average emergence rate and average time to female emergence in
experiment 1 .............................................................................................................................88
Figure 4.4 Average wing size (mm) of females in experiment 1 ................................................89
Figure 4.5 Average emergence rate in experiment 2 .................................................................90
vii
Texas Tech University, Daniel Dawson, August 2016
Figure 4.6 Relationship between female wing length (mm) and fecundity (i.e, number
of eggs laid) in experiment 1 .....................................................................................................91
Figure 5.1 Map of position of Lubbock county in the Southern High Plains of Texas and a
landcover map of Lubbock county ..........................................................................................134
Figure 5.2 Decision tree describing daily behavior of females in NetLogo model ....................135
Figure 5.3 Location of surveillance traps in Lubbock County, TX, operated by Texas Tech
University ................................................................................................................................136
Figure 5.4 Interaction plots of significant interactions identified in the sensitivity analysis
between the oviposition movement parameter and adult daily survival, temperature, and
dispersal behavior ..................................................................................................................137
Figure 5.5 Prediction distribution and 95% intervals for values at 5 traps used in model
evaluation for 100 model simulations ......................................................................................138
Figure 5.6 Average predicted adult population of treatment scenario simulations over
60 days ...................................................................................................................................139
Figure 5.7 Difference rasters at three time points in the simulations of treatment and nontreatment conditions ................................................................................................................140
Figure 5.8 Example of clustering risk cells in landscape in relation to wetlands.......................141
viii
CHAPTER I
INTRODUCTION
Mosquitoes are a diverse group of insects (order Diptera, family Culicidae) comprising
approximately 3500 species that occur in a wide variety of habitats and ecosystems around the
world (A.N. Clements, 1992a). Due to the need for the females of many mosquito species to
take blood-meals from hosts to reproduce (A.N. Clements, 1992b), their role as a nuisance is
well understood by those that have encountered them. This particular aspect of their biology
also means that they pose a threat to public health because of their potential to vector diseases
(Smith et al., 2014). Mosquitoes vector both parasitic diseases like filiariasis and malaria as well
as viral pathogens such as West Nile virus and dengue fever virus (Smith et al., 2014). Because
of the severity and broad distribution of these diseases, a detailed understanding of their
ecology, including the factors that drive population dynamics, is critical to mitigating the risks
that mosquitoes pose.
Mosquitoes are holometabolous insects, meaning their life cycle consists of 4 distinct
stages (egg, larva, pupa, and adult) (A.N. Clements, 1992a). Functionally, this life history can be
grouped into an aquatic phase (larvae and pupae), and a terrestrial adult phase (A.N. Clements,
1992a). The egg phase may be aquatic or terrestrial depending upon the species, but always
requiring water to hatch (A.N. Clements, 1992a). An important part of the response to the health
threat posed by mosquitoes is the use of mosquito control techniques, or just “mosquito control,”
with the aim toward reducing mosquito populations, often via chemical pesticides (Kroeger et
al., 2013). The larvae and adult stages are most commonly targeted for control, with “larvicides”
and “adulticides” being key components of mosquito control programs. Although aquatic larvae
have a distinct ecology from that of terrestrial adults, influences on juvenile development can
1
Texas Tech University, Daniel Dawson, August 2016
have direct implications on the fitness of eventual adults (Akoh et al., 1992; Briegel and
Timmermann, 2001), and therefore population dynamics (Carrington et al., 2013; Walsh et al.,
2011). Naturally, an important influence on larval life history directly related to mosquito control
is exposure to lethal and sub-lethal concentrations of larvicides.
Larvicides are a wide variety of chemical and biological agents, and include neurotoxicants (e.g. organophosphates, pyrethroids), hormone-regulating compounds (e.g.,
methoprene), biologically derived toxins (e.g. Bacillus thuringiensis israelensis (Bti)-based
compounds), mono-molecular oils, and larvivorous predators (e.g. Gambusia affinis) (Connelly
and Carlson, 2009). In addition to application rate, a number of factors have been identified that
influence the effectiveness of larvicides. For example, Bti larvicides are popular because their
mode of action makes them relatively non-toxic to non-target organisms (Lacey, 2007).
However, their effectiveness is known to be reduced in aquatic habitats with a high organic
matter content because of their potential sorption to particles (Lacey, 2007), and they are rapidly
degraded by UV light (Knight et al., 2003). In addition to reducing the effectiveness of
pesticides, biotic and abiotic factors also interact with sub-lethal concentrations to influence
larval life history characteristics that ultimately influence adults and potentially adult population
dynamics. Temperature, for instance, is a highly influential factor in larval ecology, with higher
temperatures leading to faster development rates and smaller adult size (A.N. Clements,
1992c). Temperature can also influence the effect of sublethal exposures to larvicides, with
Muturi (2013) reporting that first instar Aedes aegypti (Lay) larvae exposed to sub-lethal doses
of malathion had larger wings and were more fecund as adults when raised at cooler
temperatures. In another example, the introduction of larvicidal agents can reduce intra-specific
(Muturi et al., 2011a) and interspecific (Blaustein and Karban, 1990) competition and predation
(Bence and Murdoch, 1983), releasing resources to surviving larvae, and benefiting eventual
adult fitness.
2
Texas Tech University, Daniel Dawson, August 2016
Understanding and quantifying adult population dynamics are important to mosquito
control, particularly for the purposes of mitigating disease risk, since adult female mosquitoes
must make physical contact with a person to transmit a pathogen. One way to capture the
effects that pesticide exposures and other extrinsic factors have on adult population dynamics,
especially impacts occurring during juvenile stages, is through mathematical population
modeling. The opportunities for modeling in the field of mosquito control are numerous, with
models ranging widely in scope and complexity. For example, Carrington and others (2013)
constructed a relatively simple, deterministic matrix model describing how population growth of
Ae. aegypti was impacted by fluctuating versus constant larval rearing temperature. In contrast,
Skeeter Buster is a highly complex, spatially explicit mechanistic model developed to predict
population dynamics in Ae. aegypti (Legros et al., 2011; Magori et al., 2009). The nature and
extent of the model construct depends upon its intended use, so although these example
models vary widely in form, they are designed to address specific questions. For questions
regarding how specific stressors, like temperature or pesticide exposures, affect population
dynamics in general, relatively simple, non-spatial models are adequate (Carrington et al.,
2013). To capture population dynamics in real-world systems, more complex models that
account for aspects like spatial heterogeneity are likely required. Spatially explicit models in
particular have the potential to be useful tools for mosquito control, as predictions of mosquito
population dynamics, or the risk posed by them, in space can be used to assess landscape
drivers of populations (Schurich et al., 2014), assess the success of treatment strategies
(Pawelek et al., 2014), and potentially guide where mosquito control treatment should be
allocated. Lastly, spatially-explicit population models can be used as the basis for efforts to
quantify mosquito-borne disease in space (Smith et al., 2004; Tachiiri et al., 2006).
For this dissertation I investigate the influence of larvicide exposure to the larval ecology
of two medically important mosquitoes, and develop spatially explicit models of mosquito
3
Texas Tech University, Daniel Dawson, August 2016
population dynamics in two real-world systems. In Chapter 2, I report an experiment in which
Ae. aegypti mosquito larvae were exposed to malathion at different temperatures, and then
modeled how differences in larval responses based on the age of exposure in larvae potentially
affect mosquito population dynamics. In Chapter 3, I used surveillance and treatment data
collected from collaborating municipalities in Tarrant County, Texas, to develop a linear
regression-based, spatially explicit population model of the mosquito Culex quinquefasciatus. In
Chapter 4, I report two experiments investigating the effects of water quality, predator presence,
and a commonly applied larvicide on life history characteristics of larval Culex tarsalis
mosquitoes. In Chapter 5 I report on the development, analysis, evaluation, and demonstration
of an application of a spatially explicit population model for Cx. tarsalis in Lubbock County, TX. I
conclude with a summary and synthesis of research themes covered by this dissertation.
4
Texas Tech University, Daniel Dawson, August 2016
CHAPTER II
POPULATION EFFECTS OF MALATHION EXPOSURE AT DIFFERENT
TEMPERATURES AND AGES OF EXPOSURE IN AEDES AEGYPTI
(DIPTERA: CULICIDAE)
Abstract
Exposures to sub-lethal toxicant concentrations during development can have far-reaching
effects on life history traits that can potentially manifest at the population level. In the case of the
yellow-fever mosquito (Aedes aegypti (L.)), previous research has shown that when young
larvae (first instar) are exposed to malathion at cooler temperatures, emerging females had
greater emergence rates, larger wings, and increased fecundity than those exposed at higher
temperatures. However, it is uncertain how temperature and malathion concentration interact
with older larvae due to differences in energy allocation and metabolic ability between older and
younger larvae. To address this, an experiment was conducted in which fourth instar larvae
were exposed to malathion at different temperatures. It was found that emergence rates
decreased only in response to increasing malathion concentration, that development rate
increased only in response to increased temperature, and that female wing size decreased in
response to increasing temperature and malathion concentration. These experimental results,
along with published results from a similar experiment with first instar larvae, were extended into
a stage-based population model. Modeling results showed that sub-lethal malathion exposure
was related positively to population growth following first instar exposure, and related negatively
following fourth instar larvae exposure. This research demonstrates that exposure to stressors
during development can have unexpected effects that may manifest in higher levels of biological
organization.
5
Texas Tech University, Daniel Dawson, August 2016
2.1. Introduction
A common goal in ecotoxicology and risk assessment is to predict toxicant effects on
natural systems. Characterizing the dose or exposure concentration of toxicants in relation to an
organism’s response is often of critical importance. For pest organisms targeted with pesticides,
however, factors other than exposure can significantly influence the effectiveness of the
pesticide in controlling populations and, in turn, the quantity of pesticide needed to effectively
control nuisance species. Two such factors considered here are environmental temperature and
age of exposure.
Ambient temperature has far-reaching influence over physiology and toxicant sensitivity
(Hallman and Denlinger 1998), especially for poikilothermic organisms like insects. Mosquitoes
are insects with four distinct life phases, including aquatic egg, larvae, and pupae stages, and a
terrestrial adult stage. During the aquatic phase, temperature is a critical factor influencing larval
and pupal life history characteristics that can translate to the population level. For example,
higher temperatures in the aquatic phase tend to induce faster larval and pupal development
rates that result in smaller adult sizes and lower fecundities (van den Heuvel, 1963). In the adult
phase temperature is also critically important because as temperature increases, the rate of the
gonadotrophic cycle also increases (Carrington et al., 2013), allowing for more frequent egg
laying. For larval mosquitoes, temperature is also an important modulator of pesticide effects on
life history characteristics (Muturi et al., 2011; Muturi, 2013; Muturi et al., 2011c). Some
toxicants, including organophosphorus pesticides, can be potentiated by higher temperatures
(Muturi, 2013; Nareshkumar et al., 2012; Polson et al., 2012). On the other hand, malathion, like
other organophosphorus compounds, is more rapidly hydrolyzed as temperature increases
(Ragnarsdottir, 2000), lessening the exposure duration of larvae.
An organism’s age also significantly influences how it is affected by toxicants. In general,
younger animals are more sensitive than older animals. This phenomena has been
demonstrated in numerous taxa, including vertebrates (Timchalk et al., 2007) and various
6
Texas Tech University, Daniel Dawson, August 2016
invertebrates (Bouvier et al., 2002; Palmquist et al., 2008; Stuijfzand et al., 2000), including
mosquitoes (Nareshkumar et al., 2012). For example, different-age instars of various mosquito
species exposed to the crystalline protein toxins produced by the bacterium Bacillus sphaericus
showed increased toxin tolerance as larvae matured (Lotfy et al., 1992; Nareshkumar et al.,
2012). One reason for this general pattern is that older animals tend to have a greater ability to
metabolically detoxify toxicants than younger animals. In mosquito larvae this is in part due to
older larvae having more developed fat bodies. The fat body is an essential organ present in all
insects (Arrese and Soulages, 2010) that stores energy reserves in the form of lipids (mainly
triglycerides) and glycogen. It also serves important endocrine and immune functions by
producing and directing the activity of hormones and metabolizing enzymes (Arrese and
Soulages, 2010). Although metabolic responses may prevent toxicant-induced mortality, they
are energetically expensive and likely reduce energy reserves (Naylor et al., 1989; Palmquist et
al., 2008). Because mosquito pupae do not eat, newly emerged adult mosquitoes depend upon
energy reserves acquired during the larval stage. Total body energy content positively
correlates with body size and fecundity in newly eclosed adult females (Briegel, 1990; Muturi,
2013), and since resource allocation strategies differ by larval stage, age of exposure to
pesticides can be important. For example, the yellow fever mosquito Ae. aegypti, accumulates
the bulk of its larval energy during the fourth instar (Timmermann and Briegel, 1999). Chemical
insults to fourth instar larvae may lead to smaller and potentially less fecund adults than
unexposed individuals because resources normally allocated to adult development have to be
utilized for the metabolism of toxicants (Vlahović et al., 2009). Alternatively, if young larvae are
exposed to a toxicant that causes only partial mortality, the surviving individuals may experience
a competitive release of resources. Surviving-but-exposed females may eventually become
larger and more fecund than unexposed individuals (Antonio et al., 2009; Muturi, 2013).
Lastly, temperature, pesticides and age have the potential to interact to influence
mosquito life history characteristics, particularly under non-lethal exposures. For instance,
7
Texas Tech University, Daniel Dawson, August 2016
younger larvae exposed to pesticides and developing at high temperatures would have reduced
energy reserves in comparison to those developing at lower temperatures, potentially reducing
the capacity for metabolic detoxification. For older larvae reared at higher temperatures and
exposed to a toxicant, already lower energetic resources (compared to larvae reared at lower
temperatures) would be taxed further by metabolic responses, potentially leading to even
smaller and less fecund adults. Interactions such as these increase the difficulty in predicting
effects of chemical exposures on mosquito populations. Fortunately, insight in this regard can
be gathered by extending experimentally derived estimates of vital rates into a population
modeling framework (Caswell, 2001), as done recently with Ae. aegypti by (Carrington et al.,
2013a; Carrington et al., 2013) . For disease vectors such as Ae. aegypti, predictions from such
models have important management implications and provide insight into how interactions of
multiple factors in the larval environment may translate to population-level responses.
In this work, experimentation and modeling are used to investigate how larval rearing
temperature and malathion exposure affect life history characteristics and population dynamics
differently depending on whether larvae are exposed as first instars (as demonstrated by Muturi
2013) or fourth instars. First, fourth instar larvae are experimentally exposed to malathion at
different temperatures and survival and development rates are evaluated. It was hypothesized
that the combined effects of malathion and temperature on emergence and development rates
would be similar to those of exposed first instar larvae. However, due to differences in metabolic
detoxifying activity and energy allocation between first and fourth instars, the wing sizes of
emerging females should be smaller relative to control at higher concentrations of malathion.
Second, differences in results between first instar exposures and fourth instar exposures
translate to population level effects by extending experimental results to a population model
framework. Ultimately, the goal of this research was to determine how variable intrinsic (age
structure) and extrinsic (concentration, temperature) factors can influence the effect of mosquito
control on mosquito populations.
8
Texas Tech University, Daniel Dawson, August 2016
2.2. Methods and Materials
2.2.1. Mosquitoes
Mosquitoes in this study were acquired from the Liverpool Strain (LVP) of Aedes aegypti
maintained since 1936 by the Liverpool School of Tropical Medicine (Vectorbase, 2015). The
experiment was conducted using individuals produced from a breeding colony established in the
insectary at Texas Tech University from the original stock. Our colony was maintained at
approximately 25° C and 75% relative humidity in cubical cages 0.61 m to a side. Eggs were
produced by females fed bovine blood in 50% Alsever’s solution using a membrane feeding
system (Mishra et al., 2005).
2.2.2. Experimental setup
The experiment was conducted using artificial freshwater consisting of CaSO4 (3 g),
MgSO4 (3 g), KCl (0.2 g), and NaHCO3 (4.9 g) dissolved in 50 L deionized (DI) water, and
hereafter referred to as ‘‘lab water.’’ Eggs were hatched at 30°C in lab water that had been
infused with grass clippings and amended with a pinch of ground brewer’s yeast. Upon
hatching, larvae were transferred at a density of 10 ml per larva into 1 L tubs of lab water, where
they were reared to fourth instar at the three experimental temperatures (20°C, 25°C, and 30°C)
in incubators with 12/12 L:D cycle. When >50% of larvae in each temperature group had
reached the fourth instar, larvae were transferred in groups of 10 into 100 ml of lab water at the
three temperatures listed above, and exposed to malathion. Malathion exposure concentrations
included an acetone carrier control, the LC05 (0.045 mg/l), and LC50 (0.084 mg/l) values
determined from trial experiments, hereafter referred to as “control,” “low,” and “high.” These
concentrations were reasonably similar to the mortality elicited by the "high" and "low"
exposures reported by Muturi (2013), but were slightly higher, as expected due to reduced
sensitivity of older larvae. There were six replicates for each factorial combination of
temperature and malathion concentration. Larvae were reared to pupation and monitored every
9
Texas Tech University, Daniel Dawson, August 2016
24 hrs for mortality and pupal development. They were allowed ad libitum ground TetraFin®
goldfish flakes, initially starting with 0.12 g per experimental container and adding more as
necessary. Pupae were transferred to glass vials topped with cotton balls and reared at the
same temperature to emergence, at which point they were frozen. Lastly, wing lengths of all
females was measured as an indicator of potential fecundity (Armistead et al., 2008; Muturi,
2013)
2.2.3. Chemical analyses
Initial malathion concentrations and degradation rates between temperature treatments
were determined using High Performance Liquid Chromatography (HPLC). A calibration curve
(R2=0.995) was generated for malathion using a malathion standard on an Agilent 1100 High
Performance Liquid Chromatograph. Initial exposure concentration was calculated following the
determination of the concentration of the malathion stock solution. Degradation rates were
characterized using two replicates of a 1:10 dilution of the stock (20.5 ppm) for each
temperature. To simulate the interaction of malathion with food particles in experimental
containers, each stock solution sample was initially amended with a similar amount of food.
Each day for seven days, including the initial day, a sample was drawn from each stock solution
replicate (kept at the appropriate temperature), passed through a 0.2 μm polytetrafluoroethylene
(PTFE) filter into a tube, and frozen at -80 °C. After determination of concentrations via HPLC,
degradation rates were modeled at each temperature using a first-order rate equation.
2.2.4. Statistical analyses
Vital rates calculated for each treatment included average emergence rate and average
time to emergence. Average emergence rate was defined as the average proportion of larvae to
survive to emergence as adults. Average time to emergence was defined as the average day,
counting from hatch day (day 0), on which adults emerged. The overall effects of malathion
concentration and temperature on emergence rate and time was analyzed via multivariate
10
Texas Tech University, Daniel Dawson, August 2016
analysis of variance (MANOVA) using the MVN package (Korkmaz et al., 2014). When
significant effects were obtained, univariate analysis of variance (ANOVA) followed by Tukey’s
Honestly Significant Different (HSD) was used to compare pairwise differences in treatment
means (Zar, Jerrold, 1998). Time to emergence was rank-transformed (Quinn and Keough,
2002) to meet the assumption of multivariate normality, which was assessed with a Mardia test
using the MVN package (Korkmaz et al., 2014). Two-way analysis of variance was used to
determine the effects of concentration and temperature on adult female wing size. A Box-Cox
power transformation was applied to wing-sizes to meet the homoscedasticity assumptions
(Quinn and Keough, 2002). Because sample sizes were uneven due to fewer numbers surviving
in some groups than others, a Type II sum of squares approach was used in these analyses to
assess main effects without significant interactions (p>0.05) (Langsrud and Matforsk, 2003).
Pairwise differences between treatment means were compared using a Tukey’s HSD test.
2.2.5. Population modeling
2.2.5.1. Model structure and vital rates
To evaluate how exposure to concentrations of malathion at different developmental
stages and temperatures would affect population growth and structure, stage-based matrix
population models were constructed for both first and fourth instar exposure scenarios. Models
assumed a daily time step, were deterministic, and considered only females with a postbreeding census. Aquatic stages included an egg and a combined larvae-pupae stage, while
adult stages included a pre-host seeking period (after emergence) stage, and a stage for each
individual gonadotrophic cycle. Experimental data from Muturi (2013) were used to
parameterize the larvae-pupae stage of the first instar exposure model, and included average
emergence rate and female time to emergence. Data from our experiment was used to
parameterize the larvae-pupae stage of the fourth instar exposure model, and included average
emergence rate and overall average time to emergence (Table 2.1). Egg stage vital rates, adult
11
Texas Tech University, Daniel Dawson, August 2016
stage survival and duration per stage were based on temperature and were taken from
Carrington and others (2013). Maximum lifespan was based on the number of gonadotrophic
cycles expected from females before death (4), and was also taken from Carrington and others
(2013).
Daily development rates (r) for a stage were calculated by taking the inverse of the mean
time until event [e.g., daily time to emergence: DTE = 1/mean time to emergence]. Daily stage
survival rates (d) were calculated by raising average stage- based survival rates (a) to the p
power, with p equaling the daily development rate [d = ap; e.g., daily emergence rate: DER =
average emergence rateDTE]. The daily probability of surviving and staying within a stage (stage
s) was calculated by multiplying d*(1-p). The probability of surviving and maturing to another
stage (stage (t)) was calculated by multiplying d*p. Fecundity rates (F) in the model were
derived from a linear regression model constructed using data presented in Muturi (2013) that
predicted the number of eggs females lay as a function of wing size:
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑒𝑔𝑔𝑠 = 58.66 × 𝑤𝑖𝑛𝑔 𝑙𝑒𝑛𝑔𝑡ℎ (𝑚𝑚) − 96.755 .
As suggested by Carrington and others (2013), the number of eggs laid per gonadotrophic cycle
was decreased by 5% with each subsequent cycle over the course of an adult female’s life, and
daily survival decreased by 15% with each gonadotropic cycle. Lastly, because the model only
produces numbers of female offspring, clutch size was multiplied by the sex ratio, which was
assumed to be 1:1.
All vital rates were incorporated into the following general transition matrix structure (A):
A
=
Aquatic Stage
Egg(s)
0
Egg(t) Aquatic(s)
0
Aquatic(t)
0
0
0
0
0
0
0
0
0
0
0
0
Pre-blood(s)
Pre-blood(t)
0
0
0
0
0
0
0
GC1(s)
GC1(t)
0
0
0
12
Adult Stage
F1
0
0
0
GC2(s)
GC2(t)
0
0
F2
0
0
0
0
GC3(s)
GC3(t)
0
F3
0
0
0
0
0
GC4(s)
GC4(t)
F4
0
0
0
0
0
0
0
Texas Tech University, Daniel Dawson, August 2016
Models simulations assumed only a single exposure to the first cohort (Nstart) using the
treatment transition matrix (At). Subsequent cohorts (N), including offspring from the first cohort,
experience control conditions of the corresponding temperature via the control transition matrix
(Ac). Model calculations were structured such that initial cohort individuals overlapped with
subsequent cohorts, and contributed to overall population dynamics, but were subject to the
vital rates associated with their corresponding treatment:
𝑁𝑠𝑡𝑎𝑟𝑡𝑡+1 = 𝐴𝑡 × 𝑁𝑠𝑡𝑎𝑟𝑡𝑡1 ;
𝑁𝑡+1 = (𝐴𝑐 × 𝑁𝑡 ) + 𝑁𝑠𝑡𝑎𝑟𝑡𝑡 .
2.2.5.2. Model simulations
All simulations were initiated with a population of ten female larvae (Carrington et al.,
2013). Simulations of each treatment scenario for both experimental exposures were run for 150
days to assure that all projections converged on a stable finite growth rate (λ). A stable finite
growth rate was defined as the day on which λ (determined numerically) did not change (up to 4
significant digits) with subsequent simulations. Lastly, estimates of adult mosquito abundance of
all projections were compared at 30 days as a benchmark of population growth. Actual
projected abundances between the first and fourth instar exposure scenarios were not directly
comparable because the first instar model used female time to emergence as reported by
Muturi (2013) instead of overall time to emergence. However, a relative comparison can be
made on the effects of early versus late instar exposure by normalizing the projected
populations of treatment groups at each temperature relative to the projections of their
corresponding control groups. To this end, adult abundances of each treatment projection were
converted to percentages of the corresponding control projections at the 30-day benchmark
[e.g. (treatment projection)/(control projection) *100].
13
Texas Tech University, Daniel Dawson, August 2016
2.3. Results
2.3.1. Experimental results
Temperature (Pillai’s trace = 0.9245; df = 4, 90; p<0.0001) and malathion concentration
(Pillai’s trace = 0.8135, df = 4, 90; p<0.0001) had a significant effect on adult mosquito
emergence rate and time to emergence, but there was no significant interaction (Pillai’s trace =
0.268, df = 4,90, p = 0.1). Post-hoc analysis showed that emergence rate was not impacted by
temperature, but that the high malathion concentration groups had significantly lower
emergence than the control or low groups (Fig. 2.1). Post-hoc analysis for time to emergence
showed that development rate significantly increased with each increase in temperature (Fig.
2.2), but was not different between concentrations within temperature groups. For all
experimentally produced vital rates, see Table 2.1.
Wing length of females exposed as fourth instar larvae decreased with both increasing
temperature (p = <0.0001) and malathion concentration (p = <0.0001), with no significant
interaction (Fig. 2.3). The main effect of temperature was driven by females having smaller
wings on average at each increase in temperature (p =<0.0001). The main effect of
concentration was driven by significantly larger wings on average in the control concentrations
compared to both the low (p =<0.002) and high concentrations (p =<0.00001), particularly in the
25°C group (Fig. 2.3).
2.3.2. Chemical analysis
Analysis of malathion concentrations in the daily degradation samples showed that the
average concentration of sample replicates (13.02 ppm) on the initial day was significantly lower
than the targeted concentration (20.5 ppm), but that the averages within temperature groups
(30°C: 12.67 ppm, 25°C: 13.71 ppm; 20°C: 12.69 ppm) were relatively consistent initially and
across the sampled days. As malathion is relatively lipophilic and readily adsorbs to organic
material in aqueous solutions (Bell and Tsezos, 1987), these results suggest that some of the
14
Texas Tech University, Daniel Dawson, August 2016
chemical adsorbed to food particles, which were then intercepted by the filter prior to sampling
from containers. If this was the case, it would suggest that the concentrations of malathion freely
available in the water of the experimental containers may have been lower than the initial
concentrations determined via HPLC.
Malathion degraded in a non-linear manner, with degradation rate increasing as
temperature increased. Semi-log functions without intercepts were created to model
degradation at each temperature (Fig. 2.4). Functions predict that on the first day the structure
of the model allows for newly hatched larvae (day 6; see section 2.2.5.1), concentrations of
malathion would be 0.026, 0.0074, and 0.00052 ppm at 20°C, 25°C, and 30°C, respectively.
These values are below the “low” concentration used for first instar exposures in Muturi (2013),
meaning that a single dose during the larval phase would likely degrade to well below those
causing toxicity by the time new eggs are laid.
2.3.3. Population modeling
Modeling results showed that population growth in both the first and fourth instar
exposure projections was driven most strongly by temperature, with projections in the 30°C and
20°C temperature groups differing from each other by orders of magnitude after 30 days (Table
2.2). In both the first and fourth instar exposure projections, however, the abundances in the
25°C and 30°C groups tended to be similar. This was due to similar developmental times
between these groups, and was reflected by similar λ’s (Table 2). Malathion concentration and
age of exposure appeared to be secondary but also significant in influencing relative abundance
between concentrations. In the first instar exposure scenario, the high concentration exposure
groups had the highest projected abundances with temperature groups, followed by low and
control concentrations (Table 2.2). In contrast, in the fourth instar exposure scenario, the high
concentration exposure groups had the lowest projected abundances within temperature
groups, followed by low and then control concentrations (Table 2.2).
15
Texas Tech University, Daniel Dawson, August 2016
2.4. Discussion
As expected, this study found that development rate increased with rising temperatures,
and was consistent with other work in finding that female mosquitoes had larger wing sizes at
lower temperatures (Dodson et al., 2012; Padmanabha et al., 2011). However, we found that
females had progressively smaller wing sizes as malathion concentration increased. This
contrasts with previous studies reporting the opposite effect of malathion following exposure to
first instar larvae (Muturi et al, 2011; Muturi, 2013; Muturi et al., 2011a, 2011b). In addition,
previous studies with first instar mosquitoes have shown effects of malathion to interact with
effects of high temperatures to further reduce emergence rates compared to exposures at lower
temperatures (Muturi, 2013; Muturi et al., 2011c), and for development times to decrease with
pesticide concentration (Muturi, 2013; Muturi et al., 2011b). These contrasting results may be
due to differences in the ability of younger and older larvae to metabolically detoxify pesticides,
and differences in how they allocate energy in their remaining larval period.
Because fourth instar larvae are larger with more developed fat bodies, they have a
greater capacity for enzymatic metabolism of toxicants than first instar larvae (Bouvier et al.,
2002). This may enable fourth instar larvae to better tolerate chemical insults compared to first
instar larvae, preventing differences in emergence rates at temperatures known to potentiate
malathion toxicity. However, the use of reserves from the fat body to reduce effects of sub-lethal
malathion concentrations would detract from energy available for development, leading to
reduced adult size. The difference in time remaining to develop may also contribute to
differences in the effect of malathion on wing size between first and fourth instar exposures.
While surviving-exposed first instar larvae have the remainder of their larval stage (3 more
instars) to accumulate resources for adulthood, surviving-exposed fourth instar larvae have a
comparatively short amount of time to do the same thing. The shorter relative time to
accumulate resources after exposure may also explain why time to emergence did not decrease
with concentration, as shown following first instar exposures (Muturi, 2013; Muturi et al., 2011b).
16
Texas Tech University, Daniel Dawson, August 2016
In these exposures, faster stadial development rates may be the result of competitive release of
resources by surviving individuals or due to the selection of phenotypic disposition in survivors
that tends toward faster development. With less time remaining before pupation, the influences
of these two factors may have been reduced. Lastly, the main effect of malathion concentration
on wing size in our experiment was driven by significantly larger wings in the control groups
versus both the low and high concentration groups, suggesting that even a small chemical insult
(e.g., the low concentration) can influence wing size. When wings are used as predictors of
fecundity, changes in wing size can translate to population-level effects.
Population model projections suggested that within temperature groups, impacts to
potential fecundity have the greatest effect on projected abundances. This is demonstrated by
relative population projections within temperature groups (Table 2.2) being directly related to
malathion concentration with first instar exposures (i.e., higher concentrations = lower projected
abundance relative to control), and inversely related to malathion concentration for fourth instar
exposures (i.e., higher concentrations = lower projected abundance relative to control. These
results suggest that there may be unintended population consequences to mosquito control due
to some individuals actually benefiting from treatment, depending on the age of exposure,
temperature, and concentrations of pesticide employed (Antonio et al., 2009; Muturi, 2013).
Although we only considered one correlate of wing size here (fecundity), greater female mass in
mosquitoes (a strong correlate of wing size), is a factor shown to associate with a number of life
history characteristics with implications for vector-borne disease risk. For example, larger
female mosquitoes have greater longevity (Hawley, 1985), and a better ability to obtain blood
meals (Nasci, 1986), while also dispersing shorter distances (Maciel-De-Freitas et al., 2007)
and having lower competence for disseminating dengue virus (Alto et al., 2012b). Therefore, a
complexity of factors can arise as a result of influencing the life history characteristics of adults,
of which fecundity is only one part.
17
Texas Tech University, Daniel Dawson, August 2016
It is important to note that like all models, these results incorporate numerous simplifying
assumptions, and in reality, multiple aspects influence the ultimate number of viable offspring
produced by a female. For example, Antonio and others (2009) found in an experiment that
although exposure to sub-lethal levels of spinosad increased female size and fecundity of Ae.
aegypti, egg hatching rate was reduced in exposed females. Therefore, predictions of higher
fecundity in exposed females may not be valid in all situations because of unaccounted for
factors in the model. In addition, the model presented here is completely deterministic, assumes
density independence, equal effects of pesticides by sex, a 1:1 sex ratio, and that no external
forces are acting, all assumptions that evidence suggest are ultimately invalid. Therefore, it is
largely theoretical and inferences gained from such model projections require qualification.
Despite these apparent limitations, however, these results further demonstrate the potential for
effects on the larval stage to extend to higher levels of organization, and of the utility of
modeling approaches to provide valuable insights.
2.4.1. Management implications
Potential management implications of these findings are that, if possible, it may be more
effective to target older larvae than younger larvae. This would seem to have two main initial
benefits, including that 1) concentrations used to elicit mortality in older larvae will be sufficient
to kill most or all younger larvae, and 2) that surviving older larvae may be less fecund as adults
compared to those not treated. One negative consequence to this approach is that as
concentrations of chemicals degrade in water, they soon fall into the range of a sub-lethal dose
to first instar larvae. And, if eggs are laid and hatched during this period, surviving larvae may
experience a benefit in the form of higher fecundity as eventual adults.
18
Texas Tech University, Daniel Dawson, August 2016
Acknowledgments
I acknowledge Luciano Cosme from the Texas A&M Vector Biology Research Group for
initially supplying me with Aedes aegypti eggs, and providing technical assistance in
establishing the colony used in this work.
Figure 2.1. Average emergence rates for experimentally-exposed Ae. aegypti. Error bars
indicate standard errors around each mean. Different letters indicate significant pairwise
differences in treatment means.
19
Texas Tech University, Daniel Dawson, August 2016
Figure 2.2. Average times of emergence for experimentally-exposed Ae. aegypti. Error bars
indicate standard errors around each mean. Different letters indicate significant pairwise
differences in treatment means.
20
Texas Tech University, Daniel Dawson, August 2016
Figure 2.3. Average wing lengths (mm) of experimentally-exposed female Ae. aegypti. Error
bars indicate standard errors around each mean. Different letters indicate significant pairwise
differences in treatment means.
21
1.0
Texas Tech University, Daniel Dawson, August 2016
20°:Conc=exp( -0.196 X Day), SE= 0.007
25°:Conc=exp( -0.404 X Day), SE= 0.014
0.6
30°:Conc=exp( -0.848 X Day), SE= 0.065
2
0.99
0.4
R
2
R
0.99
0.2
Percent(%) Starting Concentration
0.8
20°
25°
30°
2
0.93
0.0
R
0
1
2
3
4
5
6
Day
Figure 2.4. Malathion concentrations (points) in water solutions over 7 days (day 0 to day 6) at
all temperatures studied during the experiment. Values based on HPLC determination. Lines
shown are predicated values from semi-log models of degradation rate at each temperature.
Model equations with standard errors estimates are shown, along with R2 values for each
model.
22
Texas Tech University, Daniel Dawson, August 2016
Table 2.1. Vital rates and traits for the first instar and fourth instar exposure scenarios. Values
are listed by malathion concentration [Control (C), Low (L), and High (H), within each
temperature group (20°C, 25°C, 30°C)]. Standard errors (SE) are shown for fourth instar vital
rates experimentally derived from this study. Fecundity estimates were generated from a linear
regression equation derived from data presented in Muturi (2013). Non-exposure stage vital
rates were derived from literature values and were constant within temperature groups.
Exposure Stage
1st Instar Larvae
4th
Instar Larvae
Non-exposure
Stage
Adults
Eggs
Vital Rate
Emergence Rate
Time to Emergence
(Days)
Wing Length (mm)
Fecundity Estimate
(eggs/female)
Emergence Rate
SE
Time to Emergence
(Days)
SE
Wing Length (mm)
SE
Fecundity Estimate
(eggs/female)
C
0.51
20° C
L
0.58
H
0.37
C
0.79
25° C
L
0.70
H
0.37
C
0.61
30° C
L
0.60
H
0.24
15.50
15.00
14.50
14.00
13.00
11.00
11.50
11.60
8.50
2.30
2.27
2.75
2.20
2.35
2.74
2.15
2.16
2.50
38.17
36.41
64.57
32.30
41.10
63.98
29.37
29.96
49.90
0.93
0.03
0.80
0.05
0.45
0.04
0.87
0.06
0.83
0.04
0.48
0.06
0.87
0.03
0.88
0.05
0.30
0.06
14.36
14.50
13.94
9.74
9.44
9.28
6.96
6.84
6.89
0.21
3.09
0.03
0.19
3.05
0.02
0.21
2.99
0.05
0.08
2.96
0.03
0.09
2.85
0.03
0.09
2.73
0.08
0.05
2.72
0.02
0.12
2.62
0.03
0.31
2.57
0.05
84.72
82.27
78.39
76.83
70.38
63.59
62.93
56.70
54.22
Vital Rate
20° C
25° C
30° C
Daily survival
Pre-Blood
length(Days)
Gonadotrophic Cycle
Length (Days)
Daily survival
Time to Hatch (Days)
0.90
0.90
0.90
2.50
1.25
1.40
7.50
3.00
2.50
0.99
5.00
0.99
5.00
0.99
5.00
23
Texas Tech University, Daniel Dawson, August 2016
Table 2.2. Thirty-day population projections, intrinsic growth rates (λ), and control normalized
projections for both first and fourth instar exposure scenarios. All exposure scenarios reverted to
control conditions within temperature groups after initial exposure, so there is only one λ value
per temperature group. Control normalized projections indicate the ratio of each projection to
the control projection within temperature groups.
Instar
Temperature
Exposed
(°C)
1st
4th
Lambda
(λ)
20
1.251
25
1.376
30
1.382
20
1.164
25
1.232
30
1.232
Malathion
Concentration
C
L
H
C
L
H
C
L
H
C
L
H
C
L
H
C
L
H
24
Adult Female Population
Projection
347.76
321.84
273.21
5514.22
5115.51
4010.67
8639.79
7933.18
5270.57
50.95
51.44
81.59
255.11
330.29
474.58
277.91
280.56
414.28
Control
Normalized
1.00
0.93
0.79
1.00
0.93
0.73
1.00
0.92
0.61
1.00
1.01
1.60
1.00
1.29
1.86
1.00
1.01
1.49
Texas Tech University, Daniel Dawson, August 2016
CHAPTER III
A MODEL OF MOSQUITO ABUNDANCE CONSTRUCTED USING
ROUTINE SURVEILLANCE AND TREATMENT DATA IN TARRANT
COUNTY, TEXAS
Abstract
Mosquito population dynamics are spatially and temporally variable, making the
establishment of quantitative relationships between mosquito populations and their drivers a
challenging task. While mosquito surveillance data is used by vector control organizations to
assess the response of mosquito populations to climatic and pesticide treatment applications in
a general way, these data have been infrequently used to explicitly quantify drivers of mosquito
population dynamics, particularly in an operational context. Mathematical modeling is a valuable
tool that can help accomplish this task. I used general linear mixed modeling to model
population dynamics of the mosquito Culex quinquefasciatus, a West Nile virus vector, in
Tarrant County, Texas. Pesticide treatment, habitat, and weather data were used as predictor
variables to model mosquito surveillance data collected by six municipalities and in
unincorporated neighborhoods in Tarrant County during the 2014 mosquito season. General
linear mixed modeling (GLMM) using log + 1 transformed data was used to model surveillance
data. An Akaike Information Criterion (AIC) -based model selection and multi-model averaging
techniques were applied to determine the best average model for mosquito counts across the
landscape. The model revealed that counts were driven mainly by seasonally-fluctuating
temperature, precipitation, and treatment. Interestingly, the impacts of habitat factors in driving
mosquito populations across the study system was found to be insignificant. The model was
variable in its predictive ability depending upon trap location, with predicted log counts better
approximating observed log counts at locations with less variability from week to week. Future
25
Texas Tech University, Daniel Dawson, August 2016
efforts should be focused on assessing how other location-specific habitat factors and model
structure may influence model predictive ability, particularly at sites with highly stochastic
counts.
3.1. Introduction
Mosquitoes pose a significant threat to human health and well-being throughout the
world, both as disease vectors and nuisances. In the United States, considerable efforts are
made to control mosquito populations and their associated disease risk by an array of publicallyfunded mosquito control organizations (e.g., county and city health departments, mosquito
control abatement districts, and vector control departments), and by for-profit mosquito control
companies. These operations vary greatly depending upon their location, their level of funding,
the relative health risks that their service areas face, and the demographics of the populations
they serve. However, there are a few consistent aspects of modern mosquito control programs.
First, the reduction of mosquito populations is carried out by targeting larval sources and adults
through water management and pesticide applications (Connelly and Carlson, 2009). Second,
mosquito control programs survey mosquito populations using adult traps and larval sampling in
order to decide where and when to apply mosquito control treatments (Connelly and Carlson,
2009). These two aspects are important because they generate data that inform management
decisions. On the surveillance side, mosquito trapping data indicate how populations fluctuate
over time as a factor of weather and treatment factors. On the treatment side, records provide
spatiotemporal information relative to what kind and how much pesticide is applied in the
environment. When combined, these two data sources are used by managers to make mosquito
control decisions. In general, “data use” often translates to managers examining surveillance
data from the latest sampling period and deciding where they should allocate available
resources based on their previous experience and the priorities of the mosquito control program
(personal communication-mosquito control personnel). Although current practices represent an
26
Texas Tech University, Daniel Dawson, August 2016
important utilization of available information, opportunities exist through mathematical modeling
to enable greater use of this data.
The modeling approach that can most readily utilize the information generated from
mosquito control programs are multiple regression models in which adult mosquito population
dynamics are modeled using count data collected at mosquito traps as part of a surveillance
program. Indeed, a number of efforts have been made in this regard, including landscape coverbased models (Diuk-Wasser et al., 2006; Schurich et al., 2014), temporally autoregressive
models (Brown et al., 2011), and complex generalized linear models that account for temporal
and spatial correlation using Bayesian estimation methods (Yoo, 2014). In addition, surveillance
data has been used in modeling risk factors for the incidence of mosquito borne disease (Pepin
et al., 2015). Interestingly, with some exceptions (Pawelek et al., 2014), few models of mosquito
population dynamics using surveillance data have included treatment as a predicting factor. One
potential reason for this is that for some mosquito control activities, like the application of nonregulated pesticides, records may not be kept (Tim Segura, former manager of City of Lubbock
Vector Control; personal communication). Another potential reason is that even when records
are kept, they are not collected consistently, and they are made in hardcopy paper format that is
never digitized (personal observation). Lastly, treatment records that are collected may not
include a detailed spatiotemporal record. That is, the record of application may exist, but they
are not associated with a specific time and location (personal observation). These issues make
data analysis difficult, and compound if collaboration across multiple mosquito control
authorities is desired. Fortunately, there is an increasing use within the mosquito control field of
Global Positioning Satellite (GPS)-enabled technology that automatically collects spatiotemporal
information of treatment activity, particularly the application of adulticides dispersed from truckborn ultra-low volume (ULV) spray devices. Coupled with handheld GPS devices for larviciding,
there is an expanding opportunity to incorporate treatment information into mathematical models
of mosquito population dynamics.
27
Texas Tech University, Daniel Dawson, August 2016
Tarrant County, Texas, is located in north-central Texas, USA. It is an urban county, with
a human population of 1.8 million (Tarrant County, 2016), and the county seat is Fort Worth.
Several mosquito species of concern are present, including Culex quinquefasciatus and Cx.
tarsalis, both vectors of West Nile virus. Cx. quinquefasciatus are of particular concern in
Tarrant County as they are urban mosquitoes, breeding in small quantities of stagnant water, as
well as water sources like wastewater lagoons (Zequi et al., 2014). A number of municipalities
operate independent mosquito control programs in the county (including the city of Fort Worth),
while Tarrant County Public Health (TCPH) administers a mosquito control program that covers
Unincorporated Tarrant County. These entities all maintain networks of adult mosquito traps,
maintain treatment records, and coordinate with the TCPH to manage vector-borne disease
risks. As far as the author is aware, no previous attempt to date has been made to characterize
mosquito population dynamics in Tarrant County using a mathematical modeling approach.
Such a model, however, could provide insight into population drivers at the landscape scale,
and particularly if model predictions are spatially specific, could lead to the development of
operational tools. With these goals in mind, we developed a mixed linear model of surveillance
data, collected by a subset of these municipalities and TCPH, as a factor of treatment, weather,
and habitat variables.
3.2. Methods and materials
3.2.1. Data sources
Mosquito abundance and treatment data were obtained from TCPH following the 2014
mosquito season (April-October). The study area included a subset of participating
municipalities within the county that administers mosquito control programs and unincorporated
Tarrant County, in which the mosquito control program is operated by TCPH (Fig. 3.1).
Collaborating municipalities included Arlington, Burleson, Colleyville, North Richland Hills,
Southlake, and Haltom City.
28
Texas Tech University, Daniel Dawson, August 2016
3.2.2. Response variable
The response variable of interest in this study was the abundance of female Cx.
quinquefasciatus (commonly called “quinqs”) mosquitoes. Counts of quinqs was collected by
mosquito traps known as “gravid traps” by both individual municipalities and TCPH and
compiled by TCPH. Gravid traps use containers of stagnant water to attract gravid female
mosquitoes into fan-operated traps. In addition to being useful for surveilling viral infection rates
in blood-fed females in general, gravid traps are known to be highly effective at attracting quinqs
(DiMenna et al., 2006; White et al., 2009). Traps throughout the study area were operated on a
weekly basis throughout the mosquito season, with some on a permanent basis, and some on a
temporary basis. For the purposes of comparability, only data from permanent traps were used
in this analysis.
3.2.3. Predictor variables
Based on mosquito biology and previous efforts to model mosquito surveillance data
(Schurich et al., 2014; Yoo, 2014), it was hypothesized that mosquito counts could be best
modeled as dependent on four exogenous factors (or a subset of them), including weather,
mosquito control treatments, habitat quality, and temporal factors that allow for seasonality.
Weather data, including precipitation and temperature records, were obtained from the National
Climatic Data Center via online download (NCEI, 2016), and consisted of data from eight
weather stations distributed around Tarrant County. Mosquito control data included larviciding
and adulticiding records collected and maintained by individual municipalities and the TCPH.
These data were compiled at the request of TCPH and released to the authors for the purposes
of this study. Treatment data was highly variable in form and detail depending upon the source.
Data ranged from hand-written descriptions of treatments with general descriptions of locations,
and chemicals and quantities used (particularly for larvicide records), to high-resolution, GISgenerated maps of adulticide applications. For modeling purpose, habitat quality was accounted
29
Texas Tech University, Daniel Dawson, August 2016
for with two variables including the Normalized Difference Vegetation Index (NDVI) and human
population density. NDVI is an indicator of the vegetative vigor, and therefore, potential water
availability. NDVI was chosen because it is relatively easy to obtain and is significantly
associated with abundance (Yoo, 2014) or distribution (Diuk-Wasser et al., 2006) of mosquitoes.
NDVI was calculated from spectral imagery (described below) downloaded from the USGS
Global Visualization Viewer (USGS, 2016). Human population density was selected as a factor
because the quinq mosquito is often thought of as an anthropophilic mosquito (Murty et al.,
2002) that utilizes small pools of standing water around human settlements (ditches, French
drains, flower pots). Human population density has also been significantly associated with
mosquito counts in other species with affinity for humans (Yoo, 2014). Human population data
was downloaded from the US Census Bureau (USCB, 2016).
3.2.4. Data processing
3.2.4.1. Count data.
Mosquito abundance data consisted of raw counts of quinq females identified by TCPH
technicians. The spatial locations of all traps, including geographic coordinates were entered
into ArcGIS 10.3 and plotted. The processing of predictor variable data varied, but two aspects
were common to the extraction of spatially/temporally specific information. First, although
distances between traps throughout the landscape varied, a number of traps in urban areas
were in close proximity to each other (i.e., less than 400 m apart). In order to maintain
independence of data between spatially-specific predictor variables, a 100 m buffer was
extended around trap locations to extract all spatial data. This buffer distance was within the
spatial range of other studies examining the association of spatial variables and quinqs in urban
environments (Landau and Leeuwen, 2012; Leisnham et al., 2014). Second, because it was
uncertain at which temporal scales count data would be predicted by weather and treatment
30
Texas Tech University, Daniel Dawson, August 2016
information, the data for these variables were aggregated into 1–4 week time intervals after
extraction, and each time scale was evaluated during the model selection process.
3.2.4.2. Treatment data.
All treatment data was digitized and migrated into ArcGIS 10.3. Larviciding information
collected by the participating municipalities was highly heterogeneous in format, and was highly
variable in the specificity of spatial location, with some records including GPS coordinates,
some including addresses, and others were simply marked by hand on a map. Records with
point locations were entered into a corresponding point shapefile, while those associated with
areas were hand digitized into a polygon shapefile. Because some locations were treated over
multiple dates, a new point or polygon was added to the treatment layers for each treatment
event on each date. Due to the heterogeneity of auxiliary information collected with treatment
records, application date was the only auxiliary information consistently associated with
larviciding records in ArcGIS. Because larviciding information consisted of both point and
polygon features, a single larviciding polygon was made by creating 10 m polygon buffers
around all larviciding points, and then joining the buffer file to the previous larvicide polygon file.
The resulting shapefile contained all larvicide events as separate polygons.
Adulticiding information was gathered largely in map form, often being generated from
adulticiding tracking software associated with GPS-equipped spray trucks. These records were
hand-digitized into ArcGIS into an overall adulticiding polygon shapefile. Spray areas were
usually in neighborhoods with highly complex shapes, so polygons were created for each spray
event that encapsulated the entire spray area in a single polygon. While hand digitizing, buffers
were included around spray area margins (as indicated on drawn or produced maps) of
approximately of 10–20 m. Like larviciding records, new polygons were created for specific
spray events so that each event was represented in the shapefile as its own polygon.
Because both surveillance data and treatment records included date information, the
spatial location of treatment events could be associated with surveillance data via their date of
31
Texas Tech University, Daniel Dawson, August 2016
occurrence and their proximity to mosquito trap stations. So, we first calculated intersection
areas between the 100 m buffer of the surveillance data, and the adulticide and larvicide layers.
Then, records were exported into program R, where surveillance data and treatment data were
combined into a single file in which treatment records were associated with particular
surveillance records based on whether they fell into time intervals (1-4 weeks) prior to the
surveillance record. Lastly, measures of treatment activity were calculated for each time interval
for each surveillance record. For larviciding, treatment activity was the total number of
larviciding events within a time interval (e.g., total events in week 3) and the sum total of
larviciding events to occur over the entire period (e.g., sum total of events since and including
week 3). Adulticiding activity was calculated similarly to larviciding, and included the total areas
treated within a particular time interval (e.g., total area in week 3), and the sum total areas over
an entire period (e.g., total areas treated since and including week 3).
3.2.4.3. Weather data
Weather data downloaded from the NCDC included precipitation (average daily rainfall)
and temperature (min, max, and time of observation) records from weather stations in around
Tarrant County with a period of record of March 1, 2014 – November 1, 2014. Because weather
data was limited to eight stations, each mosquito trap was assigned the data from the nearest
weather station using ArcGIS. A preliminary investigation using temperature data loggers at 20
trap locations around the county demonstrated that temperature data collected at weather
stations closely aligned with temperatures at traps. Unfortunately, no localized precipitation
information was available for comparison with weather station data.
Aggregated precipitation and temperature variables were generated for record (i.e.
date) in each station corresponding to the 1-4 week time intervals utilized in the treatment data.
For precipitation these variables included daily average sum per week (e.g. daily average of
week 3), and daily average sum over the entire period (e.g. daily average rainfall from and
including week 3). For temperatures, these variables included average daily temperature
32
Texas Tech University, Daniel Dawson, August 2016
(calculated as the average of high and low per day) per week, and average daily temperature
over the entire period.
3.2.4.4. Census data
Human population data from the 2010 census were downloaded by census block
polygon for Tarrant County. This polygon was imported into ArcGIS and clipped to the extent of
the traps. Then, we calculated the population density of humans within the same 100 m buffer
from mosquito trap locations as the treatment data. We first used the intersect tool to determine
the area of intersect between the census polygon areas and buffer circle around each trap.
Next, the proportion of each census block within buffer circles was multiplied by the original
polygon fragment yielding proportional abundances per fragment. Lastly, the abundances of all
fragments were added, and then divided by the area of the buffer circle to obtain density. These
density values were then assigned to each mosquito trap.
3.2.4.5. NDVI data.
NDVI data were calculated using Landsat 8 multi-spectral imagery, downloaded on
January 16 2016 from USGS Global Visualization Viewer. Downloaded data were from two
dates, July 1 and October 5, 2014, as these had less than 10% cloud cover, and were
representative of warmer and cooler times of the study period. Data were imported into ArcGIS,
where NDVI was calculated for each date using the standard formula [(NIR-Red)/(Red+NIR);
NIR=Near Infra-Red]. Then both NDVI layers were averaged using the raster calculator tool.
Next, 100 m buffers around the mosquito trap locations were used as masks to extract the NDVI
data surrounding each trap. Lastly, summary statistics, including mean and standard deviation,
were gathered from each buffer using the zonal statistics tool.
Upon spatial extraction in ArcGIS, all data was exported and prepared for analysis using
program R. Non-temporal data, including human population density and NDVI, were assigned to
individual mosquito surveillance records based on proximity to trap location. Temporal-
33
Texas Tech University, Daniel Dawson, August 2016
dependent data, including weather and treatment, were associated with each surveillance
record via its location and date using a temporally- and spatially- specific identifier.
3.2.5. Statistical analysis
Prior to statistical modeling, a data exploration was undertaken to elucidate underlying
correlations and potential analysis issues. It was determined that the raw mosquito count data
were distributed according to a negative binomial distribution, and was heavily zero-inflated
(37%). The majority of zero records occurred during the beginning of the mosquito season,
likely due to lower temperatures. To reduce the influence of these temperature-driven zeroes
and because the intention of the model is to predict counts of quinqs when they are physically
present, data was truncated to only include records from the beginning of June through
November 1. This reduced the percentage of zero records to 8% of the dataset. To improve
comparability of longitudinal data between traps, data from traps were only included if traps 1)
were categorized as “static” traps (as opposed to temporary), and 2) if the total number of data
points (i.e., sample weeks) for the sample period was at least 50% of the trap with the most data
points (25). Lastly, we excluded some trapping records where trap effectiveness was
questionable, including trap malfunctions, high wind, or precipitation during trap setting. This
resulted in a total of 53 traps and 1068 observations included in model construction.
3.2.5.1. Predictor variables
Some predictor variables were temporally specific, namely treatment and weather variables.
Uncertainty regarding which temporal scale would best predict counts was accounted for in the
model selection process, as described below. For all temporal variables, we considered either
the cumulative average (temperature) or sum (precipitation, adulticide area, larvicide events)
over an entire interval, or the average or sum within an interval, as discussed above. Nontemporally specific variables included average NDVI, NDVI standard deviation, and human
population density.
34
Texas Tech University, Daniel Dawson, August 2016
During the data exploration phase, a seasonal pattern in quinq counts was noted at
several stations, with counts starting low in June, reaching a peak sometime between July and
September, and declining to November. Therefore, four temporal variables, including Julian
Date (Date), Julian Date2 (Date2), sin (sin((2*pi/365) * Julian day)), and cos (cos(2*pi/365) *
Julian day)), were included in the model construction and selection process. Week number was
also investigated as a substitute temporal variable for Date, since mosquito data were collected
at a weekly scale. However, models with Date were better supported using the method
described below than those using Week number. Lastly, it was observed that quinq counts were
highly heterogeneous through time as a factor of both the trap and the city in which traps were
located, suggesting that spatial and temporal dependencies should be considered as potential
random factors. All variables considered for model construction are shown in Table 3.1.
3.2.5.2. Model building and selection.
Model building was carried out using Generalized Linear Mixed Modeling (GLMM) with
the lme4 package in program R. Data were log +1 transformed (to include 0 observations) prior
to inclusion in the modeling process. Although a Poisson or Negative Binomial regression
approach would ostensibly be more appropriate as the response variable (mosquito counts) was
discrete (Caputo et al., 2015; Yoo, 2014), preliminary analyses using these methods produced
poor model fit due to the large spread of count values (e.g. 0–2000). Previous modeling efforts
of mosquito counts have also log transformed counts to improve model fit (Brown et al., 2011).
Mixed modelling was used here to account for spatial dependencies at trap locations, which
were considered as random factors. Prior to modeling, all predictor variables were centered and
standardized by dividing variables by 2 times their standard deviation to facilitate numerical
parameter estimation and to aid in model interpretation (Gelman, 2008). The modeling process
proceeded in several steps, and used the Akaike Information Criterion for small sample sizes
(AICc) (Anderson, 2008; Ganser and Wisely, 2013):
35
Texas Tech University, Daniel Dawson, August 2016
𝐴𝐼𝐶𝑐 = 𝐴𝐼𝐶 +
2𝑘(𝑘 + 1)
𝑁−𝑘−1
with k equal to the number of parameters and N equal the number observations, to evaluate the
relative fit of models at each step. Though the sample size is reasonably large in this study
(1068 observations), AICc has been shown to be generally superior to AIC (Anderson, 2008;
Ganser and Wisely, 2013). The overall modelling process included 1) fitting the best random
structure, 2) assessing various fixed model structures, 3) assessing models in the final model
set with AICc, and employing multi-modal averaging if the top model doesn’t garner
overwhelming support (>90%), and 4) validating the assumptions of the average model, namely
assessing the distribution of the residuals for lacking of fit, homogeneity of variance, and
normality using graphical techniques.
To evaluate the best fixed structure, the overall hypothesis stated above (see Section
2.3: Predictor Variables) was parsed into 6 sub-hypothesis categories and investigated
hierarchically, with the best sub-model in each category determined by AICc model selection.
Model categories started with the most basic subset of the overall hypothesis and then built to
the full hypothesis. Because mosquitoes are poikilothermic organisms that require standing
water to breed, the most basic plausible hypothesis is one containing a linear combination of
temporal and weather factors, represented by hypothesis 1. The most complex hypothesis
considered includes all four predictor types (i.e., temporal, weather, habitat, treatment), but in
which weather and habitat factors interact (hypothesis 6). The categories investigated, including
the number of variables per factor (in parentheses) are shown below:
1. Temporal (4) + Weather (16)
2. Temporal (2) + Weather (2) + Habitat (3)
3. Temporal (2) + Weather (2) * Habitat (3)
4. Temporal (2) + Weather (2) + Treatment (16)
5. Temporal (2) + Weather (2) + Habitat (1) + Treatment (2)
36
Texas Tech University, Daniel Dawson, August 2016
6. Temporal (2) + Weather (2) * Habitat (1) + Treatment (2).
These hypotheses were nested, with hypothesis category 1 included in all subsequent models.
The best supported model to represent hypothesis 1 was assembled in two steps, first by
selecting the best combination of temporal variables, and then adding the best supported
weather variables. Temporal variables were selected by assessing all “Temporal Only” models
listed in Table 3.2. The best temporal aggregations for weather variables at each time interval
were assessed in independent models (i.e., one temperature or precipitation variable at a time)
using the best supported combination of temporal variables. Lastly, the models included under
“Temporal + Weather” in Table 3.2 were evaluated, with the best selected model used in all
subsequent model comparisons. Likewise, with hypotheses 4–6, all temporal aggregations for
treatment variables were evaluated, with the best models evaluated with AICc. The result of
this was a final model set of 6 models representing the best supported model from each
category, indicated by AICc (Table 3.3). The relative support for each model in the final set was
assessed by calculating the AICc model weights (Anderson, 2008). To address model selection
uncertainty, models and their parameter estimates of the final set were averaged by the “natural
average” method described by Anderson (2008) to produce a final average weighted model.
We assessed model assumptions of homogeneity of variance and normality of residuals
by plotting the average model residuals against fitted values, and a normal quantile plot. The
statistical significance of variables in the averaged model was assessed by computing 95%
confidence intervals based on their unconditional variances. Lastly, because variables were
centered and standardized, the size of the beta coefficients was used to assess relative model
contribution.
37
Texas Tech University, Daniel Dawson, August 2016
3.3. Results
3.3.1. Overall
The best supported model of the 6 general hypotheses evaluated accounted for 46.7%
of AICc weight, and represented the hypothesis that counts were driven by a combination of
weather, treatment, and temporal factors (hypothesis 4) (Table 3.3). The next best-supported
model accounted for 29.2% of AICc weight, and represented the hypothesis that counts were
driven by only weather and temporal factors (hypothesis 1). The bottom four models accounted
for the remaining 24% of AICc weight, and contained all models including habitat factors.
Because no one model had overwhelming support (>90%), models were averaged to produce a
final average model. The final average model included Date and Date2 as temporal factors,
cumulative average temperature over the 3 weeks prior to surveys (Temp), the cumulative
precipitation over the two weeks prior to surveys (Precip), human population density (Pop), the
interaction between Precip and Pop, total larviciding events 2 weeks prior (Lcide), and the total
area adulticided 1 week prior (Acide). The estimated parameters for the final averaged model,
including standard errors and 95% CIs for each variable are listed in Table 3.4. The model also
included a random structure in which the intercept and the effect of date were allowed to
randomly fluctuate as a function of trap location, nested within collaborating municipality.
Random effects significantly improved model fit, indicating that spatial dependencies existed at
the trap and collaborator level.
Of the variables included in the average model, only Date, Date2, Temp, Precip, Acide,
and Lcide were found to be statistically significant (α=0.05) based on their 95% CI’s (Table 3.4).
The coefficients for Date and Date2 were very similar and of opposite sign. This means that the
combined contribution of Date (positive) and Date2 (negative) was heavily influenced by the
contribution from the random slope (estimated SE=0.62, Table 3.4). Slopes for the remaining
significant variables were all negative. In addition to being in all models, an examination of the
38
Texas Tech University, Daniel Dawson, August 2016
beta coefficients clearly indicated the most important variables driving log counts were Date,
Date2, Temp, and Precip. Treatment parameters (Lcide and Acide) were included in 3 out 6
models, with those models accounting for 61% of AICc weight (Table 3.3). Due to the relative
rarity of treatment, mean treatment values were very low (e.g., 0.067 larvicide events per two
week period prior to trapping events; 0.0016 km2 adulticided per week prior to trapping events),
and treatment parameters appearing to play a small role in population dynamics on the
landscape scale. This is reflected their low beta coefficients (Table 3.4). However, at the trapscale, treatment impacts on trap counts were shown to be quite pronounced. For example,
when holding other variables at their mean values, the model predicted that a single larvicide
application event (3.22 SD from average) reduced counts at a trap by approximately 20%
relative to no larviciding. In addition, each 0.05 km2 sprayed with adulticide (5.8 SD’s from
average) was predicted to reduce counts at a trap by 26% relative to no adulticiding. Pop and
the interaction of Pop and Precip were not significant based on their CI’s (Table 3.4). However,
the models Pop was included in accounted for 24% of AICc weight, suggesting some support
(Table 3.2). The models containing the interaction term accounted for less than 4% AICc weight.
3.3.2. Model validation
A plot of standardized residuals versus fitted values shows low heteroscedasticity about
zero (Fig. 3.2), with the exception of observed values near zero as would be expected with
transformed count data. However, there is an obvious pattern in the residuals indicating some
lack of fit. A wide variability in model predictive ability is also suggested by the estimated
residual standard error (1.164, Table 3.4) (mean predicted log count= 3.22). This is born out
when predicted values are plotted against observed values on a location-specific basis (Fig.
3.3), where it can be seen that although the model is able to reproduce general seasonal
patterns of counts at locations, it is inconsistent in its ability to reproduce observed counts at
locations. Although in some trap locations the model performed quite poorly to predict observed
39
Texas Tech University, Daniel Dawson, August 2016
counts (Fig. 3.4A), at other locations modeled and observed counts matched more closely (Fig.
3.4B). This suggests site-specific drivers, potentially habitat-based, of count stochasticity.
3.4. Discussion
3.4.1. Main model drivers
As reflected in the highest AICc support for the Temporal + Weather hypothesis, the
main drivers of mosquito counts were climatic in nature, and included fluctuations in cumulative
temperature and precipitation, as well as a seasonal pattern in counts due to variables Date and
Date2 in which counts generally increased with time to an inflection point and then decreased in
an inverse parabolic pattern. Although the effects of Date and Date2 largely cancelled each
other out due to their similar size and reversed sign, the shape of the seasonal pattern in counts
varied between sites was captured by the random factor for Date. Traps with negative random
contributions to Date had inflection points for the effect of season on counts that occurred earlier
than for random effects with positive contributions. For example, using only the coefficients from
the GLMM model for considering Date and Date2, one location with a negative slope
(unincorporated: -97.2066, random slope = -1.39) reached its asymptote for the effect of Date
10 days earlier (August 3) than another location with a positive slope (August 30)
(unincorporated -97.4594, random slope = 1.12). Indeed, an examination of the observed points
for these locations (Fig. 3.3) showed that the peaks in observed mosquito counts at these
locations occurred earlier and later in the year, respectively.
The physical driver of the parabolic pattern created by Date and Date2 is largely due to
Temp, as indicated by high correlations (>0.9 Pearson’s rho) between the fixed effects of Temp
and both Date and Date2. This indicates that Temp may be a redundant variable to Date and
Date2. To address this, the top model set was refit and averaged without Temp, and the model
performance of between the two average models was compared using the Root Mean Square
40
Texas Tech University, Daniel Dawson, August 2016
Error (RMSE). RMSE is an overall indicator of model precision and performance (Brown et al.,
2011), calculated as
Σ(𝑜𝑏𝑠𝑒𝑟𝑣𝑒𝑑−𝑝𝑟𝑒𝑑𝑖𝑐𝑡𝑒𝑑)2
𝑛
𝑅𝑀𝑆𝐸 = √
,
with smaller values indicating better overall predictive ability. The RMSE for the model without
Temp was 1.81, while the RMSE for the model with Temp was 1.79, indicating only a slightly
improved predictive ability with the addition of Temp. Although models with Temp were
supported via AICc more than those without, the small change in RMSE suggest that Temp
might be removed without much of loss predictive ability. This may because the data was
truncated to include June-November. Since during this period quinqs are both present and
active, the influence of Temp may be largely captured by the seasonal pattern replicated by
Date and Date2. One exception to this appears to be at the very end of the season in which
observed counts at almost all locations fall to near zero, likely due to decreases in temperature,
a pattern poorly captured by the model (Fig. 3.3). This suggests that the current model would be
likely unable to be extended beyond the temperature range with which it was parameterized.
Although a negative sign for the Precip coefficient initially seems counter-intuitive,
mosquito counts can be negatively associated with precipitation in the short to intermediate term
(i.e., weeks to months) due to the dilution of nutrients in stagnant water in which mosquitoes
seek to lay eggs (Jian et al., 2014). This is reflected in the fact that the infusion media used to
attract gravid mosquitoes is generally aged for at least seven days before it used (Burkett et al.,
2004). Quinqs have a temperature-dependent development rate, with a time of emergence that
ranges from approximately 7–14 days in a temperature range of 30°C to 20°C, respectively
(Rueda et al., 1990). The cumulative 3-week average (Temp predictor used in the model)
ranged from approximately 20°C to 32°C over the season. Although other studies have found
support for a one week lag in precipitation prior to surveys as a weather predictor of mosquito
counts (Ganser and Wisely 2013, Yoo 2014), a two-week lag here appears to better align with
41
Texas Tech University, Daniel Dawson, August 2016
quinq developmental biology, and therefore likely represents an underlying driver of population
dynamics. Quinqs are known to use a variety of standing water to oviposit, ranging from aboveground containers like flower pots in urban areas and cemeteries to natural and anthropogenic
catch basins (Leisnham et al., 2014), including underground French drains (Nina Dacko,
personal communication). Such anthropogenic basins may be more consistently available, and
indeed, the negative coefficient in the model is suggestive of consistent breeding habitat that is
reduced in quality after rains, not new habitat that is created after rains. This would also suggest
that habitat factors and interactions with Precip would be important model drivers. However, the
coefficients for both Pop and the interaction were relatively small compared to Precip, and both
were statistically insignificant (Table 3.4).
Despite the lack of a significant habitat factor in the averaged model, and all hypotheses
containing habitat factors accounting for <25% of AICc weight, a comparison of fitted values at
trap locations shows fit of model predictions to be highly variable depending on their location
(Fig. 3.3). This suggests that underlying site-specific differences may potentially be uncaptured
habitat characteristics. Previous studies incorporating similar habitat predictors to the current
one (Yoo, 2014), as well as land-cover and other habitat features (Ganser and Wisely, 2013;
Leisnham et al., 2014; Schurich et al., 2014) found strong impacts of these factors. This
suggests that the habitat variables included in the modeling process were insufficient,
particularly in more rural locations. This is reflected in plots of model residuals against predictor
variables that showed model residuals to be slightly larger at locations with lower human
population and higher NDVI (Fig. 3.5). As an example, the poorly predicted location in Fig. 3.4A
was located in a rural subdivision in Unincorporated Tarrant County, where the well predicted
location (Fig. 3.4B) was located at an urban location in the in the City of Burleson. For quinqs in
general, rural versus urban prevalence appears to be variable, with some authors reporting
significantly higher abundances of quinqs in rural areas compared to urban ones (Murty et al.,
2002), while others report quinqs abundance to be associated both developed and non42
Texas Tech University, Daniel Dawson, August 2016
developed areas (Leisnham et al., 2014). In both these cases, the availability of oviposition sites
was indicated as the primary driver of quinq counts. This suggests that the availability of
oviposition habitat throughout the mosquito season has the potential to greatly influence local
population dynamics. In urban locations in Tarrant County, the availability of breeding habitat
may be more consistent throughout the season, and site-specific effects of temperature and
precipitation may be adequate to describe population dynamics. In contrast, oviposition site
availability may change frequently in more rural areas, leading to more stochastic population
dynamics, and driving poor model performance. One habitat measure not explicitly included in
the modeling process was the spatial extent of wetlands near trap locations. The association of
wetland land cover was negatively associated with quinqs in urban areas in Florida, likely due to
the availability of human-supplied oviposition locations (Leisnham et al., 2014). However, in
rural areas this may be less available, making it important to consider incorporating wetland
extent into future modeling efforts in Tarrant County. Potentially complicating matters, a
temporal component has also been found to exist in which quinqs shifted their habitat use
throughout the year, apparently due to competition from Aedes species (Leisnham et al., 2014).
To this end, characterizing oviposition habitat, and factors influencing its use and availability
may be important to predicting population dynamics of quinqs in rural Tarrant County.
Another related explanation for site-level heterogeneity in model performance is that
precipitation and temperature information was collected at an insufficient scale to predict sitelevel heterogeneity of counts. Because weather data was collected from weather stations, it only
approximated weather conditions at trap locations. Small-scale temperature fluctuations due to
habitat influences, and differences in actual rainfall between weather stations and trap locations
were likely. Precipitation in particular is prone to being highly variable across landscapes, with
precipitation data of insufficient spatial resolution and accuracy cited as an important source of
error in spatially-explicit hydrological modelling (Tetzlaff et al., 2005). These issues could
certainly lead to poorer model fit than if data were available otherwise, and therefore poorer
43
Texas Tech University, Daniel Dawson, August 2016
predictive performance. One potential solution for this is to install rain gages at trap site
locations that could be read by technicians as part of weekly operations. Fortunately, some
mosquito control authorities are already instituting this as part of their surveillance programs
(personal observation), with some spatial database management software for mosquito control
including rain measurements (e.g. MapVision, Leading Edge Associates).
Lastly, an important point to consider is that in addition to representing underlying
population levels, mosquito counts also represent patterns of activity among the fraction of adult
mosquitoes attracted to traps (Jian et al., 2014). Movements by this “active fraction” (Jian et al.,
2014) are likely influenced by microclimate conditions not collected at the weather station scale,
including localized relative humidity, wind velocity, and tree cover shade (Verdonschot and
Besse-Lototskaya, 2014). Also, the attraction of mosquitoes to particular habitats, either due to
presence of hosts or for resting, may influence the activity of mosquitoes around traps. For
example, quinq counts were highly associated with the presence of medium-height trees in
urban environments in Tucson, Arizona (Landau and Leeuwen, 2012). Therefore in Tarrant
County, differences in variability between trap sites may be due to factors contributing to the
activity of mosquitoes as well as their underlying population dynamics.
3.4.2. Treatment effects
The sign and temporal scales of the treatment effects included in the average model are
consistent with what we expected, given the biology of quinqs, and the mechanisms of
adulticide and larvicide pesticides. Adulticides used in Tarrant County during 2014 consisted of
largely pyrethroids sprayed out of truck-mounted ULV (ultra-low volume) sprayers. Sprayed
adulticides are designed to impinge upon the bodies of flying mosquitoes, with no additional
expected mortality to mosquitoes after the droplets have settled (Bonds, 2012). Therefore, that
the best supported adulticide variable had 1 week lag time makes intuitive sense. In addition,
given the expected development rate of quinqs (see above), a two week lag time on the effect of
44
Texas Tech University, Daniel Dawson, August 2016
larvicide is reasonable. Larvicides can have multiple mechanisms, including causing direct
toxicity to larvae (e.g. Bacillus thuringiensis israelensis (Bti) toxin-based products), mimicking
hormones that prevent larvae from developing (e.g. Methoprene), and smothering larvae and
pupae on the surface of water (e.g. larvicidal oils) (Connelly and Carlson, 2009). The larvicides
used in Tarrant County were variable, with each of the above examples represented.
An interesting result is the significant, but highly localized effects of treatment on counts
in the model. For example, a plot of log observed counts versus adulticiding events at a trap
location in the City of Burleson shows notable reductions after treatment events occur (Fig.
3.6A). Likewise, a plot of log counts versus the number of larviciding events within 2 weeks of
the trapping survey at a site in the City of Arlington show a similar, though less clear pattern for
larviciding activity (Fig. 3.6B). However, when treatment is considered on a landscape scale,
treatment events occurring in proximity to traps become rather rare. For example, at the spatial
scale considered for treatment effects (100 m from trap center), the average number of total
adulticiding events and larviciding events occurring over the previous 4 weeks within the
proximity of any given trap was 0.23 and 0.25, respectively. In addition, no adulticiding or
larviciding events occurred within the proximity of 26 (49%) and 37 (70%) traps, respectively
over the course of the season. Therefore, the significant local effects of treatment in the model
were likely masked by the effects of non-treatment occurring most of the time in the landscape,
resulting in small beta coefficients relative to other significant factors.
One potential solution to better characterize the efficacy of treatment is to simply widen
the allowable area around traps to be considered for treatment effects, as the 100 m buffer used
in this analysis may be insufficient in size. To that end, spatial data for treatment were extracted
at both 500 and 1000 m buffers from around trap locations, and the model selection process
was repeated with this data using the GLMM approach described above. However, the models
containing treatment variables at 500 m and 1000 m distances from traps were less supported
by AICc than those with treatment data at 100 m. This suggests that at least when treatment
45
Texas Tech University, Daniel Dawson, August 2016
effects were quantified as they are in the current model, 100 m is the best distance for
predicting impacts on counts.
3.4.3. Summary and future considerations
3.4.3.1. Model performance
This research represents a first step in the utilization of surveillance data and treatment
data collected by mosquito control authorities in Tarrant County, Texas, towards two goals,
including 1) providing inference into mosquito population dynamics and 2) providing an
operational tool. The model clearly indicates that quinq counts in Tarrant County from the period
of June-November appear driven in large part by the influence of seasonal temperature patterns
and precipitation, and by the effects of treatment. Overall, quinq counts tended to increase over
course of the season and then decrease after a certain point with temperature, with the specific
pattern driven by the influence of date that was particular to the city and trapping location. The
negative coefficient for precipitation is likely to due to the dilution of nutrients in larval habitat by
rain, thereby making those environments less productive. The effect of treatment factors are
important but local, due to the many incidences of no treatment within proximity of traps
throughout the season. In contrast, habitat variables included in the average model were not
statistically significant, and did account for much AICc weight (<25%).
If a model is to be used as operational tool, it must adequately capture fluctuations of
population dynamics. The current model is inconsistent in its predictive ability depending upon
location. Counts at more urban locations with lower stochasticity were predicted better, whereas
counts in more rural locations with higher stochasticity were predicted more poorly. Inconsistent
prediction may be related to location-specific habitat factors, particularly those in rural areas, or
activity by the active fraction of mosquitoes (Jian et al., 2014).
One general approach to address this issue is to model surveillance separately by
collaborator, or groups of collaborators. As suggested above, specific habitat features driving
46
Texas Tech University, Daniel Dawson, August 2016
population dynamics likely varies depending upon location (e.g., municipality), or type of
location (urban vs. rural). In addition, patterns in abundance throughout as season may be due
to seasonal shifts in habitat availability (e.g., due to competition; Leisnham et al., 2014) that are
specific to a general location. Information available to describe such patterns in a model may be
more specific and at a finer spatial scale if data are considered on per municipality or group
basis. A second reason behind this approach is to account for the quality of the data collected
by mosquito control authorities. On the surveillance side, a need to make data as consistent as
possible for the purposes of modeling prompted the removal of temporary traps or traps with
few surveys, as well as surveys of questionable quality (e.g., during rain events). This could be
at least partially alleviated if all points within an analysis came from the same collaborating
organization. On the treatment side, treatment data was highly variable in terms of quality and
content depending upon municipality. Because the current model was constructed to
encompass the data from all collaborating organizations, many of the attributes accompanying
the original treatment dataset were dropped due to a lack of correspondence in data collection
detail between collaborators. For example, because information regarding dosing rate, area
applied, weather conditions, habitat conditions and chemical used was inconsistently collected
by collaborators, the treatment dataset utilized for model construction consisted only of records
of where and when larvicide and adulticide treatments occurred. Modeling focused on a
collaborator basis would allow for this ancillary data to be incorporated for the municipalities that
consistently collect it. Lastly, from an operational point of view, models developed on a per
municipality basis may be desirable, as specific predictors and consistent ancillary treatment
data may increase the predictive ability, and flexibility of models such that they can be more
readily used to support management decisions. The downside to such an approach is a general
reduction in data available for model development. This is of particular concern for collaborators
with relatively small vector control infrastructure. Therefore, careful consideration of the data
47
Texas Tech University, Daniel Dawson, August 2016
available and the goals of the modeling effort should be made when splitting or grouping
collaborators.
3.4.3.2. Treatment effects
Utilizing surveillance data to quantify the effects of treatment on population dynamics is
a reasonable goal because it utilizes data already collected during mosquito control operations,
and doesn’t come with the challenge and expense of experimentation. In addition to this,
however, is the goal of assessing treatment efficacy, which would also require the incorporation
of detailed treatment information such as application rate and chemical types. One potential
approach, similar to that of Caputo and others (2015), would be to compare differences in
counts at traps prior to and after events. By modeling these differences with a GLMM, perhaps
by employing random intercepts for trap and date, the influences of other factors like habitat,
weather, and application rates could also be gauged relative to that of treatment. This would
have the benefit of being able to easily compare the relative effects of different temporal lags
and spatial scales on model performance via AICc model selection criteria, as done in this
study. Another solution may be utilize surveillance data in conjunction with a non-regression
modeling technique. For example, Pawalek and others (2014) constructed a stage-based
population model for Cx. pipiens using a system of Ordinary Differential Equations (ODE) that
included the impacts of both larviciding and adulticiding. This model was parameterized using a
combination of surveillance data and previous studies and the authors were able to replicate the
impacts of treatments on trap counts. One drawback of this sort of approach is that each trap
requires its own population model, and all or a majority of larval sources contributing to trap
counts need to be identified. In general, however, the collection of such fine scale data may be
worth the effort if better estimates of treatment impacts on mosquito population dynamics, and
potentially treatment efficacy can be made.
48
Texas Tech University, Daniel Dawson, August 2016
Acknowledgments
I would like to thank the vector control departments of the collaborating entities, including
the cities of Arlington, Burleson, North Richland Hills, Colleyville, Haltom City, and Southlake,
and the Tarrant County Department of Public Health (TCDPH) for providing the surveillance and
treatment data used in this study. In addition, I thank the TCDPH and Nina Dacko specifically for
compiling the data provided by collaborators prior to use in this study.
49
Texas Tech University, Daniel Dawson, August 2016
Figure 3.1. Collaborating municipalities and traps in unincorporated Tarrant County, TX,
operated by Tarrant County Public Health (TCPH). Some municipalities, including Burleson and
Arlington overlap into adjacent counties.
50
Texas Tech University, Daniel Dawson, August 2016
Figure 3.2. Standardized residuals of average weighted General Linear Mixed Model versus
fitted values.
51
Texas Tech University, Daniel Dawson, August 2016
Figure 3.3. Plot of observed counts (log+1 scale) at all included traps (triangles) versus counts
predicted (crosses) by the model. Labels for each trap include the combination of its longitude
and municipality.
52
Texas Tech University, Daniel Dawson, August 2016
Figure 3.4. Plots of observed counts versus counts predicted by the GLMM model. (A) Located
in rural subdivision in unincorporated Tarrant County, counts at this trap were relatively poorly
predicted by the model; (B) located in an urban setting in Burleson, counts at this trap were
relatively well predicted by the model.
53
Texas Tech University, Daniel Dawson, August 2016
Figure 3.5. GLMM standardized model residuals against scaled population density (Pop) and
NDVI. Slight heterogeneity toward low values of Pop and higher values of NDVI suggest model
fit is lacking in rural areas, leading to poor predictive performance in those areas.
54
Texas Tech University, Daniel Dawson, August 2016
Figure 3.6. Observed count values (circles) and treatment events(squares), including (A) the
number of adulticide events 1 week prior to trapping surveys at a location in Burleson and (B)
the number larvicide events 2 weeks prior to survey at a location in Arlington.
55
Texas Tech University, Daniel Dawson, August 2016
Table 3.1. All variables considered during the modeling process. All spatially specific variables
collected via 100 m buffers from trap locations except Temp and Precip, which were derived
from the nearest weather station from trap locations. Note that multiple temporal aggregations
were considered for weather variables and treatment variables, with the best supported
aggregation selected via AICc.
Variable Category
Temporal
Weather
Habitat
Treatment
Abbreviation
Date
Julian date
Model Variable
Date2
sin
cos
Julian date 2
sin((2*pi/365) * Julian date)
cos((2*pi/365)*Julian date)
Temp
Temperature
Precip
Pop
NDVI
NDVISD
Lcide
Acide
Precipitation
Human population density
Normalized difference vegetation Index
NDVI Standard deviation
Larvicide events
Area adulticided
56
Temporal aggragations
considered
NA
NA
NA
NA
1-4 weeks; cumulative average
and within-interval average
1-4 weeks; cumulative sum and
within-interval sum
NA
NA
NA
1-4 weeks; cumulative average
and within-interval average
Texas Tech University, Daniel Dawson, August 2016
Table 3.2. All models evaluated using each modeling approach. The best supported
combination of temporal and weather variables (via AICc), representing hypothesis category 1,
was used throughout all subsequent modeling categories. The best supported model in each
category is shown in bold.
Model
Structure
Hypothesis category
Random
NA
Temporal Only
Temporal + Weather
Temporal + Weather + Habitat
Fixed
Temporal + Weather *Habitat
Temporal + Weather + Treatment
Temporal + Weather + Habitat + Treatment
Temporal + Weather + Habitat * Treatment
Evaluated models
1|City
1|Trap
1|City/Trap
Date|City
Date|Trap
Date|City/Trap
Sin+Cos+ Date+ Date2
Date+ Date2
Sin+Cos
Date
Date + Date2 + Temp + Precip
Date + Date2 + Temp * Precip
Date + Date2 + Temp
Date + Date2 + Precip
Temp + Precip + Pop
Temp + Precip +NDVI
Temp + Precip + NDVISD
Temp + Precip + Pop + NDVI
Temp + Precip + Pop + NDVISD
Temp + Precip + NDVI + NDVISD
Temp + Precip + Pop + NDVI + NDVISD
Temp + Precip + Pop * NDVI
Temp + Precip + Pop * NDVISD
Temp + Precip + Pop * NDVI + NDVISID
Temp + Precip + Pop * NDVISD+ NDVI
Temp + Precip + NDVISD * NDVI
Temp + Precip + Pop + NDVISD * NDVI
Temp + Precip + Pop*(NDVISD + NDVI)
Temp + Precip + NDVI*(Pop + NDVISD)
Temp + Precip + NDVISD*(Pop + NDVI)
Temp + Precip + Pop*NDVISD * NDVI
Temp + Precip * Pop
Temp + Precip * Pop + NDVI
Temp + Precip * Pop + NDVISD
Temp + Precip * Pop + NDVISD + NDVI
Temp + Precip * NDVI
Temp + Precip * NDVI + NDVISD
Temp + Precip * NDVI + NDVISD + Pop
Temp + Precip * NDVISD
Temp + Precip * NDVISD + NDVIS
Temp + Precip * NDVISD + NDVI + Pop
Temp + Precip * (Pop + NDVI)
Temp + Precip * (Pop + NDVISD)
Temp + Precip * (NDVI+ NDVISD)
Temp + Precip * (Pop + NDVI) + NDVIDSD
Temp + Precip * (Pop + NDVISD) + NDVI
Temp + Precip * (NDVI + NDVISD) + Pop
Temp + Precip * (NDVI + NDVISD + Pop)
Temp + Precip + Lcide + Acide
Temp + Precip + Pop Lcide + Acide
Temp + Precip * Pop + Lcide + Acide
57
Texas Tech University, Daniel Dawson, August 2016
Table 3.3. Model selection weights based on AICc for each best selected model in each
category. AICc=Second order Akaike Information Criterion. dAICc = Delta AICc; W = AICc
relative model weight. In addition to the fixed parameters shown, there was also an intercept,
and 7 estimated parameters included in each model due to the random structure, including 1)
the residual variance, the intercept variances for 2) municipality and 3) trap location nested
within municipality; the slope variances on the Date parameter for 4) municipalities and 5) trap
location nested within municipality; and the correlation between the slope and intercept
variances for 6) municipality and 7) trap location nested within municipality. For models with
interactions, there was also an interaction parameter in addition to the fixed parameters shown.
Model Category
Weather + Treatment
Weather
Weather + Habitat + Treatment
Weather + Habitat
Weather * Habitat + Treatment
Weather * Habitat
Model
Date + Date2 + Temp + Precip + Lcide +Acide
Date + Date2 + Temp + Precip
Date + Date2 + Temp + Precip + Pop + Lcide + Acide
Date + Date2 + Temp + Precip + Pop
Date + Date2 + Temp + Precip * Pop + Lcide + Acide
Date + Date2 + Temp + Precip*Pop
58
Estimated
Parameters
14
12
15
13
16
14
AICc
3488.26
3489.20
3490.90
3491.78
3494.21
3495.37
dAICc
0.00
0.94
2.64
3.52
5.95
7.11
W
0.47
0.29
0.12
0.08
0.02
0.01
Texas Tech University, Daniel Dawson, August 2016
Table 3.4. Parameters of average-weighted GLMM model. The top table includes all fixed
parameter estimates with their associated standard errors, and the 95% CI for each indicating
statistical significance. The bottom part of the table includes standard errors for the estimated
random effects for the intercept and the effect for date for each trap, nested within collaborator
city.
Variable
Estimate
Intercept
3.12
Date
21.15
Date2
-21.16
Temp
-0.90
Precip
-0.44
Lcide
-0.13
Acide
-0.10
Pop
0.05
Precip*Pop
0.00
Random Effects: Variance Estimates
Group Level
Intercept
Trap, nested within
Collaborator
0.61
Collaborator
0.66
Residual Variance
1.16
SE
0.27
2.12
2.16
0.20
0.07
0.05
0.05
0.04
0.01
Date
0.53
0.92
59
Confidence Interval
5%
95%
2.60
3.65
16.99
25.31
-25.41
-16.92
-1.30
-0.50
-0.59
-0.30
-0.23
-0.04
-0.19
-0.01
-0.03
0.12
-0.01
0.02
Texas Tech University, Daniel Dawson, August 2016
CHAPTER IV
THE INFLUENCE OF WATER QUALITY AND PREDATION PRESENCE
ON THE RESPONSE OF CULEX TARSALIS LARVAE TO BACILLUS
THURINGIENSIS ISRAELENSIS (BTI) LARVICIDE
Abstract
The combination of larvicides and other exogenous stressors can have significant
impacts on life history characteristics of mosquito larvae. Water quality and predation are two
factors that are likely to interact with the activity of larvicides such as Bacillus thuringiensis
israelensis (Bti). Because of higher rates of sedimentation and toxin degradation, increasing
organic pollution in water would be expected to reduce Bti effectiveness. In contrast, by
prompting energetically costly predator avoidance strategies, the presence of predators would
be expected to enhance Bti effectiveness. However, because responses to factors may be
situation and species-specific, these effects are difficult to predict. I conducted a series of
experiments in an attempt to disentangle the effects of these different factors. In the first
experiment, we exposed larvae of the mosquito Culex tarsalis Coquillet to varying levels of Bti in
either water collected from playa wetlands or water collected from an urban wastewater pond. In
the second experiment, we exposed Cx. tarsalis larvae to Bti, as well as the presence or
absence of Odonate predators. In the first experiment, I found that wastewater caused
significantly lower emergence rates at higher application rates of Bti, that emerging females
developed faster in wastewater and higher rates of Bti, and that they had shorter wings in
wastewater, overall. In contrast, in the second experiment we found that although emergence
rate decreased with Bti exposure, there was no effect of predator presence. Although
unexpected, results from experiment 1 may be due to higher salinity and turbidity, and the
presence of contaminants in wastewater. In addition, experiment 2 may suggest that indirect
60
Texas Tech University, Daniel Dawson, August 2016
predator effects may be less important than direct consumptive effects for Cx. tarsalis. Overall,
results suggest that water quality can have important but counter-intuitive interactions with Bti,
and predation effects, both direct and indirect, may vary by species and environmental
conditions.
4.1. Introduction
The aquatic larvae of mosquitoes have a distinct ecology from that of terrestrial adults,
but influences on juvenile development have direct implications on the fitness of eventual adults.
One important influence on the development of mosquito larvae is the application of chemical
larvicides as a part of mosquito control activities. Although application rate is an important
predictor of the effect of larvicides, their efficacy is also influenced by natural stressors that
themselves play important roles in larval ecology. For example, the effectiveness of several
pesticides is influenced by temperature, with some (e.g. organophosphates) increasing in
toxicity with temperature (Muturi, 2013), and others (e.g., DDT, pyrethroids) decreasing in
toxicity with temperature (Davies et al., 2007). Temperature alone is also highly influential in
larval ecology, with higher temperatures leading to faster development rates and smaller adult
size (A.N. Clements, 1992c). Because such exogenous factors vary widely under natural
conditions, understanding their effects are important to predicting the impacts of pesticide
applications to target populations.
Bacillus thuringiensis israelensis (Bti)-based larvicides, referred to hereafter as Bti, are
pesticides based on parasporal crystalline proteins generated by the bacteria Bti that are toxic
when ingested by mosquitoes. Toxicity is exerted by these proteins by causing a loss of
osmoregulation and subsequent cell death in the midgut of affected organisms (Lacey, 2007).
Because Bti is considered to be highly specific in its activity to mosquitoes and a few other
dipteran groups, they are popular larvicides in the field of mosquito control (Lacey, 2007). Like
other pesticides, the efficacy of Bti is influenced by several exogenous factors. One is water
61
Texas Tech University, Daniel Dawson, August 2016
quality, particularly organic pollution, as Bti proteins are rapidly degraded by bacterial activity or
adsorbed on to organic matter (Lacey, 2007). This is reflected in the dosing instructions for
Aquabac XT (Clark Pharmaceuticals), a Bti larvicide, which specifies that the maximum
application rate (2 pts/acre) be employed in habitats like wastewater retention ponds. Further, a
study of mosquito control in meat processing wastewater lagoons with Aquabac XT found that
the maximum rate (2 L/hectare) with weekly application is necessary for complete control (Zequi
et al., 2014). In addition to its effects on Bti, mosquitoes developing in water with elevated
nutrient contents, and therefore higher bacterial activity, had increased development and growth
rates (Peck and Walton, 2005). This suggests that not only would mosquitoes reared in such
environments have better protection from Bti-induced toxicity, but they would potentially be
more fit adults upon emergence. Therefore, water quality of aquatic habitat, especially in
relation to its biological activity, has important implications for the success of mosquito control
efforts with Bti.
Aquatic phase predation is another exogenous factor that plays an important role in
mosquito ecology. Mosquito larvae can make up significant biomass within aquatic ecosystems
(Fang, 2010), and thus have myriad predators (Mogi, 2007), including invertebrates like odonate
nymphs and predaceous aquatic beetles (e.g. Dytiscidae, Hydrophilidae), and vertebrates like
mosquito fish (Gambusia affinis). Using general terms suggested by Preisser and others (2007)
(Preisser et al., 2005), the effects of predation fall into two general categories, including densitymediated influences (DMI) (direct mortality from consumption), and trait-mediated influences
(TMI), which include phenotypic responses of prey to avoid or escape predation. DMI of
predators can be substantial for mosquito larvae, with various predator taxa, including
mosquitofish, tadpole shrimp (Triops longicaudatus), odonate larvae, cyclopoid copepods, and
even larvivorous mosquito larvae evaluated as potential mosquito control agents (Kumar and
62
Texas Tech University, Daniel Dawson, August 2016
Hwang, 2006). Some species, particularly mosquitofish, have been commonly employed as part
of current mosquito control programs for over 80 years (Knight et al., 2003).
Despite the importance of DMI in directly regulating populations, TMI’s of predation can
be just as important in influencing prey fitness (Preisser et al., 2005). When mosquito larvae and
pupae are exposed to predators, they display both escape behaviors (e.g., diving), as well as
predator avoidance strategies like reduced activity and movement into more densely vegetated
habitat (A. N. Clements, 1999a). Although appropriate anti-predator behaviors can reduce
predation rates while minimizing fitness losses (Sih, 1986), reduced foraging time or energy
expended to escape predators can still exact significant costs. For example, Aedes notoscriptus
larvae exposed to chemical cues from predator fish experienced slower growth rates and
reduced size at emergence than those not exposed to predator cues (van Uitregt et al., 2012).
Because higher fecundity and longer adult survival is positively associated with larger energy
reserves at adult emergence (Alto et al., 2012a; Briegel and Timmermann, 2001), such impacts
can potentially have effects at higher levels of biological organization.
The effects of combined predator pressure and pesticide exposure, particularly at sublethal concentrations, is a likely scenario under natural conditions. Alone, sub-lethal pesticide
exposures may result in larger surviving individuals, either through the selection of individuals of
higher fitness or through the competitive release of resources (Muturi, 2013). Similarly,
consumptive predation alone may benefit survivors by increasing resource availability (Alto et
al., 2012a). These two stressors combined, however, can serve to both reduce larvae
populations directly, and to influence larval and eventual adult fitness. In examples of the
former, the combination of Bti and Gambusia affinis were found to provide better control of
mosquito larvae than Bti alone in rice fields (Kramer et al., 1988; Stewart et al., 1983). In an
example of the latter, Cx. pipiens displayed a reduction in alarm response to beetle predators
63
Texas Tech University, Daniel Dawson, August 2016
after larvae were exposed to fenvalerate, and thus were more susceptible to predation
(Reynaldi et al., 2011).
In this study I use two experiments to explore how water quality and predator presence
interact with mosquito control-relevant concentrations of Bti to 1) influence Bti’s efficacy, and 2)
influence the fitness of surviving individuals in the mosquito Cx. tarsalis. Culex tarsalis is a wideranging species of western North America. In the Southern Great Plains, it frequently inhabits
the vegetated edges of ephemeral wetlands known as playa wetlands (Richardson et al., 1972).
However, it can also utilize polluted water sources like wastewater lagoons (Peck and Walton,
2005). Bti is commonly used as a larvicide against Cx. tarsalis throughout its range. In the first
experiment, Cx. tarsalis larvae (third/fourth instar) were exposed to varying Bti concentrations in
two different aquatic media, including water representing typical “playa” habitat, and water
collected from a wastewater lagoon. In the second experiment, larvae were exposed to varying
Bti concentrations while also being exposed to visual and chemical predation signals or not. In
both experiments, larvae were reared through the adult stage, and life history characteristics
were measured and compared.
4.2. Methods and materials
4.2.1. Mosquitoes
Mosquitoes in these studies were obtained from BEI resources as eggs (BEI resources,
NIAID, NIH: Cx. tarsalis YOLO, NR-43026). Experiment 1 described below used eggs directly
received from BEI resources (generation F34). The second study described below used eggs
produced from a colony maintained by the authors (generation F41), and initially started with
eggs from BEI resources (generation F31). Our colony was maintained at approximately 25° C
and 65% relative humidity in cubic cages 0.61 m to a side, with a 14hr-10hr light dark cycle,
including 1 hour dawn/dusk periods of reduced light. Eggs were produced by females blood-fed
bovine blood in 50% Alsever’s solution using a membrane feeding system (Mishra et al., 2005),
64
Texas Tech University, Daniel Dawson, August 2016
augmented by the placement of small piece of dry ice atop the cage during feeding events. One
day after eggs were laid (or the day received), eggs were transferred to rearing trays containing
an artificial freshwater mixture of moderately hard water (“mod hard”), consisting of CaSO (3 g),
MgSO4 (3 g), KCl (0.2 g), and NaHCO3 (4.9 g) dissolved in 50 L deionized (DI) water to hatch.
Mod hard water was augmented with a 1:1 mixture of finely ground Tetrafin™ and Bovine liver
power, along with a whole grass plant (blades, stems, and roots) to ensure adequate food
content for developing larvae. Larvae were then reared to the instar stages utilized in
experiments, as described below.
4.2.2. Experimental setup
4.2.2.1.1. Experiment 1: water quality and Bti
The influence of differing aquatic media on the sensitivity of Cx. tarsalis larvae to Bti
were assessed using a fully factorial design. Treatments consisted of two aquatic mediums of
distinctly different water quality, and varying levels of Bti (Control, Low, Medium, and High).
Each treatment level (eight) was replicated 4–5 times, totaling eight treatments and 35 total
replicates. In the playa water treatments, the control and high exposure groups had five
replicates, and the low and medium exposure groups had four. In the wastewater exposure
groups, the low exposure group had five replicates, and the control, medium, and high exposure
groups had four. Aquatic media in this study was contained within glass jars (2.625 cm radius),
the bottom of which were covered with approximately 0.5 to 1 cm of sediment, 16 grams per
container. Sediment was collected from the edge of an inundated playa lake in Lubbock County.
To reduce the effects of volatile organic compounds and aquatic organisms present in the
sediment, sediment was autoclaved and oven-dried (60°C). To ensure homogenization and aid
in the equal allocation of sediment between replicates, autoclaved and dried sediment was
crushed through a 2 mm sieve and mixed prior to being scooped into sample containers.
65
Texas Tech University, Daniel Dawson, August 2016
Aquatic media types included water collected from representative playa wetland (“playa
water”), and water collected from a wastewater lagoon (“wastewater”). Playa media consisted of
water collected at four representative playa lakes within the city limits of Lubbock, TX, screened
through a 500 μm sieve (to remove macroinvertebrates), and pooled in equal proportions.
Wastewater consisted of water collected from a wastewater lagoon in Shallowater, Texas, also
screened through a 500 μm sieve. Water quality differences between pooled playa and
wastewater was assessed at the beginning of the experiment, including measures of nitrates,
pH, specific conductance and turbidity. Nitrates were assessed using an AQI aquarium testing
unit. Specific conductance (μS/cm) and pH were assessed using a YSI Professional Plus Multiprobe meter. Turbidity (nephelometric turbidity units, ntu) was assessed using a Lovibond TB
205 WL portable turbidimeter. In addition, samples of both medias were taken from water
collected for experiments, and stored in HDPE bottles at approximately -24°C for additional
analytical determination (see section 4.2.2.1.2 Chemical Analyses).
The aquatic phase of the experiment was conducted within a shaded outdoor (Fig. 4.1A)
enclosure from October 9th to October 3first, 2015 in Lubbock, TX. The experiment was
conducted outdoors in order to incorporate the effects of natural light and temperature
fluctuations into larval responses. Large temperature fluctuations have been shown to influence
life history traits of larval mosquitoes (Carrington et al., 2013a), and are more reflective of field
conditions than static temperatures. Within the enclosure, sample containers were kept within
water-filled boxes (Fig. 4.1B) that served to modulate drastic temperature fluctuations due to
changes in air temperature throughout the day. Prior to starting the experiment, 256 ml of
aquatic media and 16 grams of sediment (approximately 0.5–1 cm in depth) was added to the
sample containers and the mixture was allowed to settle for 24 hrs. This volume of aquatic
media was selected because it resulted in an approximate water depth of 5 cm, a depth shown
to be naturally inhabited by Cx. tarsalis within natural wetlands (Ward, 1968), and 2) is the same
depth as the experimental containers used in experiment 2 (below). To assess temperature and
66
Texas Tech University, Daniel Dawson, August 2016
light variation over the course of the experiment, 4 HOBO pendant water data-loggers (Onset
Corporation) were added to representative blank containers, including 2 in each aquatic media
types. Lastly, replicate containers were loosely topped with plastic lids during the experiment.
The larvicide used was Aquabac XT (Adapco Company), a viscous liquid Bti larvicide
that contains a mixture of Bti toxins and bacteria components but no live bacterial cells.
Concentrations used were based on area-based (pts/Acre) exposure rates on the product’s
label scaled to the approximate surface area of the sample containers (21.637 cm2).
Concentrations included a control (0), Low (0.0625 pts/acre (0.073 L/ha)), Medium (0.125
pts/acre (0.15 L/ha)), and High (0.25 pts/acre (0.29 L/ha)) exposures. Application rate was
adapted to the sample containers by determining the average weight of 1 ml of AquabacXT,
measuring and determining the approximate surface area of the sample containers, and then
determining the mass of product needed per application rate. Lastly, an exposure concentration
was calculated, based on the volume of aquatic media. Exposure concentrations selected
represent a range leading up the minimum label application rate of 0.25 pts/acre (High), with low
and medium exposure concentrations equal to ¼ and ½ this rate. Preliminary experiments
showed 100% mortality at 0.5 pts/acre.
To begin the experiment, 15 second instar Cx. tarsalis larvae were transferred to each
container. Larvae were monitored for mortality and development on a daily basis, and fed a
prepared suspension of the TetraFin/Liver powder mixture described above. Feeding rate was
intended to be at a level not causing nutritional stress in Cx. tarsalis larvae (Dodson et al.,
2011), and varied on a daily basis depending upon the number and stage of surviving
individuals. On the day in which the majority of larvae within a container reached the fourth
instar stage (which occurred simultaneously), containers were spiked with a Bti solution (see
below). As wetlands in the field are unlikely to be treated again within the time span of a single
mosquito’s aquatic phase, sample containers were only exposed once during the experiment. It
was noted that turbidity in the wastewater treatments was initially much higher than the playa
67
Texas Tech University, Daniel Dawson, August 2016
water, but the difference decreased with time, probably due to larval feeding and/or settling in
the containers. Because it was suspected that differences in turbidity between wastewater and
playa water may contribute to differences in larval responses to Bti, larvae were transferred to
containers with new aquatic media and sediment one day prior to being exposed.
After exposure, mosquito survival and development were monitored daily until pupation.
At pupation, pupae were moved to separate containers containing the same aquatic media
where they were allowed to emerge as adults. Upon emergence, they were identified by sex
and transferred to cubical cages that were pooled by treatment. During daily surveys, larvae
were not counted as surviving to pupation unless they were found alive as pupae the day they
were discovered as pupae. Pupae were not counted as emerging as adults unless they
managed to leave their pupal exuvae.
Caged adults were maintained in the insectory conditions mentioned above, and given
continuous access to a 10% sucrose solution. When seven days had elapsed following the
emergence of a treatment group’s first female, two blood-feeding attempts were made for each
treatment group using the procedure listed above for the colony. Blood-fed females were
captured via aspirator and placed within individual oviposition jars with both sugar-water and a
small cup of water for oviposition. Oviposition jars were monitored on a daily basis for
oviposition, and after eggs were laid, for hatching. The day after hatching was initially noted in a
jar, females were knocked down with dry ice and collected, and the number of eggs and larvae
per jar were counted.
4.2.2.1.2. Chemical analyses
Previous studies suggested that ionic differences, particularly sulfate (Mian, 2006), as
well as the presence of higher concentrations of organic contaminants in wastewater, including
painkillers and anti-biotics (Pennington et al., 2015), may negatively effect larval life history
characteristics. To account for this, concentrations of common cations and anions, as well as
68
Texas Tech University, Daniel Dawson, August 2016
concentrations of acetaminophen and three antibiotics (Ciprofloxacin, Lincomycin, and
Oxytetracycline) were determined from collected samples using Ion chromatography (IC) and
Liquid Chromatography-Mass Spectrometry/Mass Spectrometry (LCMS/MS), respectively. Prior
to analytical work, both playa and wastewater samples were centrifuged, and passed through a
0.2 μm nylon filter.
Ion Chromatography was carried out for anions (F-, Cl-, SO4-2) and cations (Na+, NH4+,
K+, Mg+2, Ca+2) using a Dionex IC25 Ion Chromatograph. Sample concentrations were
determined using calibration curves constructed using multi-anion and multi-cation standards
(Sigma Aldrich). This procedure was used to determine ion concentrations in 5 pseudoreplicates taken from each of the two media, with values averaged to produce a representative
concentration for each media.
Prior to LC-MS/MS, 5 pseudo-replicate samples apiece from each of the two media, as
well as matrix spike (100 ppb) and lab control spike (100 ppb) replicates, were prepared with
Oasis HLB Solid Phase Extraction cartridges. Sample preparation and LC-MS/MS optimization
was followed previously published methods (Ferrer and Thurman, 2008; Vanderford et al.,
2003). LC-MS/MS determination was carried out using an Accela™ LC system with a TSQ
Quantum™ Access MAX triple quadrapole MS. Following LC-MS/MS, calibration curves were
constructed from standards for each compound (Oxytetracycline hydrochloride, Sigma, Israel;
Lincomycin hydrochloride, Sigma-Aldrich, Germany; Ciprofloxacin, Fluka, Switzerland,
Acetaminophen, Sigma-Aldrich, USA). Initial sample concentrations were determined by
applying the calibration curve determined values. Final sample concentrations were calculated
by applying a correction factor using the matrix spike samples in each media to compensate for
imperfect detection in the matrix. Lastly, an average concentration was calculated over the 5
pseudo-replicates for each aquatic media.
69
Texas Tech University, Daniel Dawson, August 2016
4.2.2.2. Experiment 2: Bti and predation cues
The influence of chemical and physical predator cues on sensitivity to Bti in Cx. tarsalis
larvae was assessed, also using a factorial design. Treatments included the presence or
absence of an aquatic predator (a dragonfly nymph, Libellula sp.) and varying levels of Bti
concentration (Control, Low and High). Each treatment (six) was replicated four times, with 24
total replicates. This study was conducted in playa water contained within aquaria made from
plastic shoeboxes (31.8 cm x 18.5 cm), the bottom of which were covered with approximately
0.5 to 1 cm of the sediment described above (about 120 grams per container). Aquaria were
fitted with removable screen-tops attached by Velcro®, allowing for light exposure and
ventilation (Fig. 1 C). Playa water was newly collected and pooled from the same playas used in
experiment 1 above. Experimental larvae were confined within smaller, predator exclusion
containers inside the plastic shoeboxes. Predator exclusion containers consisted of round
plastic cups (5.3 cm radius) with screened sides that allowed for the movement of water through
the containers, but prevented experimental mosquito larvae from escaping. In addition, it
prevented mosquito larvae from being consumed by nymphs, while allowing mosquito larvae
inside containers to be subjected by harassment from nymphs. Third instar larvae were selected
as the starting stage in this experiment (as opposed to second instar in experiment 1), because
they were large enough to ensure that they could not move through the screening material.
Lastly, the experiment was run within two incubators set at 25°C and a 14/10 L:D cycle.
Prior to starting the experiment, 3 L of playa water and sediment was added to the
sample containers and the mixture was allowed to settle for 24 hrs., resulting in approximately 5
cm of depth (same as experiment 1). The maximum water level was marked on each container,
and water volume was maintained at this level via the addition to DI water on a daily basis for
the duration of the experiment. After the sediment settled, a single dragonfly nymph was added
to each predator presence replicate. Nymphs were allowed to swim in sample replicate
containers for 5 days prior to adding experimental larvae to ensure adequate chemical presence
70
Texas Tech University, Daniel Dawson, August 2016
of the predators. Throughout the experiment, nymphs were monitored on a daily basis to ensure
they were alive, with dead nymphs being replaced upon discovery. In addition, with the
exception of the last six days, nymphs were fed two to three second–fourth instar Cx. tarsalis
larvae on a daily basis. Due to a lack of availability of Cx. tarsalis during the last six days of the
experiment, nymphs were fed second–fourth instar Ae. aegypti larvae.
Due to third instar larvae availability (and differential development rates of individual
larvae), the experiment was started in three groups, one day apart from each other. Treatments
assigned to replicates in each starting group were randomized. In each group, 25 third instar
Cx. tarsalis larvae were transferred to each replicate container. This number of larvae were
selected so that the density would be similar to those employed in experiment 1 (≈17 ml/larvae).
Larvae were monitored for mortality and development on a daily basis, and fed a prepared
suspension of the TetraFin/Liver powder mixture described above. Feeding rate was same as
described for experiment 1 above. On the day in which the majority of larvae within a container
reached the fourth instar stage, containers were spiked with Aquabac XT. Like experiment 1 in
this chapter, concentrations were based on area-based (pts/Acre) dosing on the product’s label,
scaled to the approximate surface area of the sample containers (588.3 square cm). The low
(0.0625 pts/acre) and high (0.25 pts/acre) exposure concentrations were the same as the
corresponding concentrations used in experiment 1. In addition, the surface area to volume ratio
was maintained between the two experiments, so the concentrations between the two
experiments were the same.
After exposure, mosquito survival and development were monitored daily until
emergence. No mortality was noted in dragonfly nymphs immediately following dosing. At
emergence, mosquitoes were mouth aspirated out of aquaria, identified by sex and transferred
to cubical cages (0.61 m to a side) that were pooled by treatment.
71
Texas Tech University, Daniel Dawson, August 2016
Caged adults were kept at the insectory conditions mentioned above. Females were
blood-fed, and fecundity and egg viability was assessed using the methods described earlier for
experiment 1 of this chapter.
4.2.3. Statistical analysis
The same statistical analyses were used to analyze comparable vital rates measured in
both experiments whenever possible. All analyses were carried out using program R (R Core
Team, 2015). Emergence rate was calculated as the proportion of larvae surviving to emerge as
adults. Average time to female emergence was calculated as the average day from hatching on
which pupated larvae emerged as female adults. Fecundity was defined as the number of eggs
laid by a female. Some blood-fed females did not lay eggs, but since this might have been a
reflection of insufficient insemination (which was not controlled for), fecundity was only
calculated for females that laid at least 1 egg. Egg viability rate was calculated as the proportion
of eggs that hatched two days after being laid. In some instances, eggs were laid on the ground
instead of inside oviposition cups, or were knocked out of oviposition cups accidentally, making
hatching highly unlikely. Therefore, hatch rate was calculated for only those females that laid
eggs inside oviposition cups. Lastly, average female wing size was calculated by measuring
both left and right wings of females (blood fed and non-blood fed), and averaging them. Length
was measured by detaching and photographing wings using a Leica MZ95 microscope-camera,
and then using the program ImageJ (Schneider et al., 2012) to make measurements in pixels,
which were then converted to mm using a photographed scale bar. Wing length was measured
as the distance from the axillary incision to the furthest distance on the wing margin, excluding
fringe scales (Nasci, 1986). In experiment 1, light and temperature data collected by HOBO
water data-loggers was aggregated by unit and described, and water quality differences
between wastewater and playa water are described. All vital rates measured in both
experiments are described in Table 4.2.
72
Texas Tech University, Daniel Dawson, August 2016
4.2.3.1. Emergence rate and time to emergence
In experiment 1, differences in emergence rate and female time to emergence between
treatment groups (water quality, Bti) were analyzed via 2-way analysis of variance (ANOVA).
Tukey’s Honestly Significant Difference (HSD) were used to make post-hoc comparisons of
significantly different groups. Multivariate analysis of variance (MANOVA) were not appropriate
in this situation because some replicates had 100% mortality and therefore, 0 days to
emergence. In experiment 2, two additional nuisance variables were included due to
experimental logistics. These factors included starting group and incubator. The starting group
factor reflected that the experiment was started in three groups (over three sequential days) due
to the availability of third instar larvae. The incubator factor reflected that two different
incubators were used in the experiment. To include these factors in the consideration of
emergence rate and female to emergence, these vital rates were separately analyzed via
general linear mixed modeling (GLMM). Models included predator presence, Bti concentration,
and their interaction as fixed effects, and start group and incubator as random intercept effects.
If main affects or interactions were found to be significant, Tukey’s HSD was used to determine
significantly different groups.
4.2.3.2. Fecundity, egg viability, and wing size
Because emerged adults were pooled into a single cage per treatment group for
breeding purposes, fecundity and egg viability rate in both experiments were analyzed via
MANOVA. In experiment 1, many adults in the wastewater group, but particularly males, were
found drowned in the pupation media after initially emerging from their pupal exuvae. Although
potentially due to experimental effects, the experiment wasn’t optimized to compare postemergent adult survival. In addition, other factors, including non-optimal emergence containers
and unusually cold weather at the time many wastewater adults began to emerge, may have
contributed to this effect. However, the end result was that very few wastewater females blood-
73
Texas Tech University, Daniel Dawson, August 2016
fed, and in fact, none laid eggs. For this reason, the MANOVA for fecundity and egg viability
was one-way, with Bti as the treatment variable. In experiment 2, more equitable numbers of
females blood-fed between the predator presence and predator absence treatment groups, so
the MANOVA for fecundity and egg viability was two-way (Bti and predator presence/absence).
However, because no eggs were laid by few females in the H concentration, predator presence
group, the MANOVA only considered C and L groups. Differences in average female wing sizes
between treatment groups were analyzed via 2-way ANOVA in both experiments, and included
both blood-fed and non-blood fed females.
4.2.2.3. Model assumptions
For ANOVA and GLMM analyses, assumptions of normality of model residuals and
homoscedasticity of variance were checked via visual assessment of normal quantile plots (i.e.
“qqplot”) of residuals, and plots of model residuals versus fitted values. For all conducted
MANOVA’s, the assumption of multivariate normality was assessed using a Mardia test using
the MVN package (Korkmaz et al., 2014). For all post-hoc ANOVA’s, assumptions of normality
were tested visually via qqplot. Lastly, in both experiments, fecundity was modeled as a function
of average wing size via general linear modeling.
4.3. Results
4.3.1. Experiment 1: water quality and Bti
In experiment 1, the average daily temperature fluctuated widely (Fig. 4.2A) from a
maximum of 25.3°C to a minimum of 15.2°C, with an average of 19.6°C. Average daily light
intensity readings also fluctuated widely (Fig. 4.2B), from a maximum of 1611 lumens/ft2, to a
low of 0 occurring at night, with an average of 95 lumens/ft2. To avoid mortality due to the
effects of temperature extremes, forecasted freezing overnight temperatures prompted the
moving of all experimental replicates indoors for the last 6 days of the experiment. However,
only three experimental replicates remained with larvae at this point. Water quality
74
Texas Tech University, Daniel Dawson, August 2016
measurements taken at the beginning of the experiment showed that of the characteristics
measured, turbidity and specific conductance in wastewater were higher than in playa water,
whereas pH and nitrates were similar between the two media. Analytical determinations made
via ion chromatography showed that all anions (F-, Cl-, and SO4-2) and cations (Na+, NH4+, K+,
Mg+2, and Ca+2) able to be determined were orders of magnitude higher in wastewater than
playa water, with the concentrations of all but Ca+2 and K+ in wastewater >10 times the
concentrations in playa water. Of the organic constituents determined via LCMS, only
Ciprofloxacin and Lincomycin were detected. Lincomycin was only detected in trace amounts
(i.e. below quantification limits) in wastewater. Ciprofloxacin was detected in low amounts in
both (<2 ppb), but the concentration in wastewater was approximately 1.5 times that found in
playa water. Determined values of all water quality characteristics assessed can be found in
Table 4.1.
A two-way ANOVA found a significant interaction between Bti concentration and aquatic
medium (Table 4.3), obscuring the interpretation of main affects. Tukey’s HSD comparisons
(Fig. 4.2A) showed that the significant interaction was largely driven by differences between the
Bti treatment groups, particularly between the High (H) and Medium (M) treatment groups
between the two aquatic media. Although aquatic media had no effect on emergence rate at
control and Low (L) Bti levels (25% of minimum field application rate), emergence rate was
significantly lower in the wastewater group at M (50% of minimum field application rate) to H
(100% of minimum field application rate) levels
For female time to emergence. ANOVAs showed that development rates were
significantly higher for females reared in wastewater and females exposed to higher Bti
concentration (Table 4.3). Post hoc tests (Fig. 4.2B) showed that these effects were largely
driven by difference between the H wastewater group and other treatment combinations.
75
Texas Tech University, Daniel Dawson, August 2016
The one-way MANOVA comparing the effects of Bti concentration on fecundity and egg
viability was insignificant (Table 3). A two-way ANOVA also showed that Bti had no effect on
wing size, but playa water reared individuals had significantly larger wings than those reared in
wastewater (Table 4.3). Post hoc comparisons (Fig. 4.3) showed this effect was driven by the
control, low and high wastewater groups. Lastly, a linear model of fecundity as a function of
wing size was not statistically significant (p = 0.065, df = 25, R2 = 0.13).
4.3.2 Experiment 2: predator presence and Bti
The general linear mixed model (GLMM) of emergence rate as a factor of predator
presence and Bti concentration showed that emergence rate was only significantly impacted by
the main effects of increasing Bti concentration (Table 4.4). Tukey’s HSD post hoc comparisons
(Fig. 4.4) showed that emergence rate significantly decreased with each increase in
concentration. The GLMM of female time to emergence showed no significant effect of either
predator presence, Bti, or the interaction of the two (Table 4.4). Both the two-way MANOVA on
the effects of predator presence and Bti concentration on fecundity and egg viability, and the
two-way ANOVA on the effects of predator presence and Bti on female wing length were found
to be insignificant (Table 4.4). Lastly, a linear model of fecundity as a function wing size was
found to be significant (p<0.01, df = 37, R2 = 0.22), with wing size positively associated with
fecundity.
4.4. Discussion
I examined how the efficacy of a commonly used Bti-based larvicide was impacted by 1)
aquatic media of differing water quality and 2) predator presence. In experiment 1 (aquatic
media and Bti), I found that several vital rates, including emergence rate, female time to
emergence, and wing size, were significantly influenced by Bti concentration, aquatic medium,
or the interaction between the two. In experiment 2 (predator presence and Bti), I found that Bti
concentration was responsible for impacts to only one vital rate (emergence rate), and that
76
Texas Tech University, Daniel Dawson, August 2016
surprisingly predator presence had no statistically significant effect on any measured vital rates.
The results of these experiments demonstrated that some, but not all exogenous factors, have
the potential to interact with Bti to influence the life history characteristics of Cx. tarsalis larvae.
4.4.1. Experiment 1
4.4.1.1. Impacts to emergence rate
Emergence rate was determined by an interaction between Bti and aquatic media such
that the effect of the pesticide was significantly greater in the wastewater group than in the playa
group at the M and H groups, but not the C and L groups. That wastewater would increase the
efficacy of Bti is an unexpected result, as polluted water is thought to reduce the efficacy of Bti
pesticides due the higher presence of suspended materials that can adsorb to Bti proteins,
making them inaccessible for ingestion (Boisvert and Boisvert, 2000). Turbidity comparisons
between wastewater and playa water did indeed show higher turbidity in the wastewater (114)
compared to the playa water (23). Also, chlorine is a common additive to wastewater that has
also been found to reduce the efficacy of Bti by degrading the toxin (Rydzanicz et al., 2010;
Sinegre et al., 1981), and chloride levels, likely derived from treatment with free-chlorine, was
found to be >11 times higher in the wastewater compared to the playa water.
Wastewater is a milieu of inorganic and organic compounds, some of which could
undoubtedly negatively effect the survival of aquatic organisms (Mian, 2006; Pennington et al.,
2016, 2015). Despite this, the control and L concentration groups in both the playa water and
wastewater groups had similar emergence rates, suggesting that one or more aspects of the
wastewater acted synergistically with Bti to enhance its toxicity, but only after a certain threshold
was reached. Four potential factors to consider following the analysis of water quality
characteristics are turbidity, specific conductance, ionic concentrations, and organic wastewater
contaminants. First, although increased turbidity is a potential reducer of Bti efficacy, the
opposite effect may have occurred as well. For example, turbidity in water in microcosms
77
Texas Tech University, Daniel Dawson, August 2016
caused by movements of Triops longicaudatus was found to increase the effectiveness of Bti
against Cx. quinquefasciatus larvae (Fry-O’Brien and Mulla, 1996). The authors attributed this to
adsorption of Bti toxins to suspended material, which may have resulted in greater ingestion by
the larvae. In combination with a formulation of Bti designed to stay suspended longer
(Biocontrol Network, 2006), this could certainly have been the case in the current experiment.
Specific conductance was found to be much higher in wastewater compared to the
pooled playa water. Specific conductance is a measure of the electrical conductivity of water at
25°C, and is strongly related to salinity and chlorides and sulfates (Patiño et al., 2014). In
contrast to this study, previous studies have shown that increases in salinity tend to reduce
effectiveness of Bti pesticides in mosquitoes (Jude et al., 2012; Osborn et al., 2007). However,
in research with Daphnia pulex, greater reductions in densities of experimental populations
occurred in higher salinity water as exposures to Bti increased (Duchet et al., 2010). The
authors posited that higher salinity may interfere with the active transport of ions, which may
combine with the stress caused by the pesticide. Cx. tarsalis has been characterized as a
euryhaline species that varies in its response to changing salinity. In relatively low saline
conditions (<40% seawater) it osmoregulates it’s ion balance by actively excreting ions (A. N.
Clements, 1992), while at higher salinities it osmoconforms to its surrounding environment via
the accumulation of the polysaccharide trehalose and the amino acid proline in its hemolymph
(Patrick and Bradley, 2000). Although the wastewater treatment was higher in salinity than the
playa water treatment, it was still relatively low (<1% seawater), meaning Cx. tarsalis likely
osmoregulated rather than osmoconformed is this experiment. As osmoregulation involves
actively transporting ions against a gradient, it entails an energetic cost which may combine with
the stress of the pesticide. In addition, stresses that increase energetic costs may necessitate
increased feeding to compensate (Clark et al., 2004), which may in turn lead to higher rates of
ingestion and therefore higher Bti exposure. Lastly, some mosquitoes will increase rates of
drinking aquatic media to prevent water loss in more saline environments (A. N. Clements,
78
Texas Tech University, Daniel Dawson, August 2016
1992), potentially serving as a route of additional exposure to Bti. Although drinking rates in Cx.
tarsalis to reduce water loss after acute salinity increases have been shown to become
indistinguishable from lower salinity groups after a few hours (Patrick and Bradley, 2000),
chronic drinking rates between water of higher and lower salinities have not been compared.
Of the ions determined via IC, those relating in particular to salinity (Na+, Cl- and SO4-2;
Dawson et al., 2015), hardness (Mg+2), and metabolic waste (NH4+) were particularly high (e.g.,
>10 times) in comparison to playa water. Of these, NH4+ and SO4-2 have been implicated as
potentially causing toxicity to Cx. tarsalis. In one example for NH4+, larvae reared in exclusion
cages in polluted wetlands mosquitoes had lower survivor rates than those reared in less
polluted wetlands with lower NH4+ (Reisen et al., 1989). In another example, Peck and Walton
(2005) attributed lower emergence rates in larvae reared in undiluted dairy water compared to
those reared in diluted dairy water to “putative toxic compounds” including NH4+. In both cases,
it was suggested to be reflective of the typical preference of less polluted water for oviposition
by Cx. tarsalis. For sulfate, in a study by Mian (2006), significantly higher SO4-2 was found in
dairy wastewater ponds in which Cx. tarsalis wasn’t found to breed, suggesting it may influence
their survival. In the current study, despite high levels of both ions in wastewater compared to
playa water, emergence was only impacted at medium and high concentrations of Bti exposure.
This suggests that while these ions may cause stress to larvae (as indicated by wing size;
discussed below), it is insufficient to cause mortality until another stressor, such as Bti
exposure, combines with it in sufficient combination.
Organic contaminants found in human wastewater may have acted synergistically to
enhance the toxicity of Bti (Pennington et al., 2015). Pennington and others (2015) showed that
mortality rates of Cx. quinquefasciatus exposed to Bti in combination with acetaminophen or a
mixture of antibiotics (Lincomycin, Oxytetracycline, and Ciprofloxacin) was higher than when
exposed to the same levels of Bti in control water. In addition, mortality was variable between
the compounds, indicating that different chemicals interact with Bti in specific ways. In the
79
Texas Tech University, Daniel Dawson, August 2016
current study, only Lincomycin and Ciprofloxacin were detected, with Lincomycin found only at
trace levels in wastewater, and similar levels of Ciprofloxacin found in both media. In addition,
concentrations of Ciprofloxacin detected (<2 ppb) were far less the 31,000 ppb used to produce
an effect in Pennington and others (2015). This preliminarily suggests that at least for the
compounds selected for testing, organic contaminants may not be significant drivers in
increasing sensitivity to Bti in wastewater. It should be noted, however, that samples were held
for five months prior to determination, much longer than the seven day extraction limit
suggested for analysis of organic contaminants in environmental samples (Dean, 2013). So,
concentrations may have degraded below detection limits after collection but prior to
determination. On the other hand, because aquatic media were taken from open, standing water
environments, tested-for compounds could also have been in low concentrations prior to
collection due to microbial (Wu et al., 2012) or photolytic degradation pathways (Xuan et al.,
2010). Therefore, the potential effects of organic contaminants is ultimately inconclusive, and
requires further study.
4.4.1.2. Development rate and wing size
Although there was ostensibly more food available in wastewater compared to playa
water due to higher bacterial and algal activity, smaller average wing sizes in the wastewater
group (Fig. 4.4) reflect lower food availability or higher energetic costs (Dodson et al., 2011).
One such cost, as discussed above, is that of dealing with higher salinity in wastewater relative
to playa water (Clark et al., 2004). Another is the impact of dealing with potentially toxic
constituents in wastewater. In an example of the latter, smaller lengths and weights of Cx.
tarsalis larvae and pupae were observed in larvae reared in diluted water from dairy wastewater
lagoons in which mosquitoes were found to not contain breeding mosquitoes as compared to
those that do (Mian, 2006). In another example, third and fourth instar Cx. tarsalis larvae were
smaller when reared in water from ponds treated with wastewater effluent than water from
80
Texas Tech University, Daniel Dawson, August 2016
control ponds (Mian et al., 2009). In both cases, it was postulated that higher rates of sulfate
(SO4-2) in the non-breeding (400–500%) or effluent-treated water (16%), respectively, could
have contributed to these larval growth differences. In the present study, sulfate levels in the
wastewater were >13 times greater than that in the playa water, and therefore consistent with
these previous studies. It is unclear, however, if the presence of high levels of sulfate are toxic
themselves, or whether sulfate simply contributes to the impacts of salinity. Together with
previous work, the results shown here thus may justify specific research into this area.
From the above discussion, it seems that wastewater may have been a stressful
environment that reduced fitness relative to playa water. Although longer wing length is
generally associated with female fitness, particularly greater fecundity (Blackmore and Lord,
2000; Briegel, 1990; Muturi et al., 2011; Muturi, 2013; Styer et al., 2007), no eggs were laid by
blood-fed females in the wastewater group (see above), preventing a direct comparison. In
addition, the linear regression for fecundity (using only females from playa water treatments)
was only marginally significant (p = 0.055, df = 1,35), with wing size only explaining 11% of the
variance. Therefore, it’s possible that although female size was smaller in the wastewater group,
this difference may not have translated to meaningful difference in fecundity. However, it may
translate to other differences, such as reductions in adult female survival (Reisen et al., 1984).
The driver of faster development in the wastewater group (Fig. 4.3B) is unclear, as faster
growth in Cx. tarsalis suggests greater food availability or higher temperature (Reisen et al.,
1989). In addition, Cx. tarsalis larvae reared in diluted dairy wastewater were found to develop
faster than those reared in wetland water, with the direct implication by the authors that
increased food availability increased development rate (Peck and Walton, 2005). However, as
discussed above, smaller wings in females reared in wastewater suggests otherwise, and
temperature data loggers suggested very similar temperatures across containers. One potential
explanation is the specific conductance differences noted between wastewater and playa water.
Though the mechanism is unclear, previous studies have shown that some mosquito species
81
Texas Tech University, Daniel Dawson, August 2016
tend to develop faster as salinities increase to moderate levels (Clark et al., 2004; Mottram et
al., 1994). Therefore, higher salinity in the wastewater group may have driven faster
development rates compared to the playa water group.
The positive effect of Bti concentration on development rate, largely driven by the
influence of the H groups, is consistent with previous studies of pesticides on times to
emergence (Muturi, 2013; see Chapter 2). Two potential hypotheses to explain this are 1) the
random killing of some larvae by pesticide exposure leads to the competitive release of
resources, benefiting surviving individuals, and 2) pesticide exposure selects for more fit
individuals, skewing the development rate down (Antonio et al., 2009; Muturi, 2013). Although
these hypotheses can’t be disentangled here, the latter hypothesis is better supported, as larger
wings, indicative of greater nutrient accumulation, were not found in the M or H wastewater
groups. Regardless of the mechanisms at play, there appears to be a tradeoff for larvae reared
in wastewater rather than playa water, and those exposed to Bti. Wastewater reared-individuals
development faster, thereby reducing the chances of predation and being exposed to
pesticides. Surviving individuals exposed to Bti in wastewater may have an even greater
development-rate advantage. However, those emerging from wastewater may be less fit overall,
and if exposed to Bti, very few survivors will develop faster.
4.4.2. Experiment 2
Interestingly, predator presence was determined to have no significant effect on any vital
rate, with only Bti concentration impacting emergence rate. In addition, in contrast to experiment
1, there was no significant effect of Bti concentration on time to emergence, once start group
and incubator effects were accounted for. Overall these results are surprising, as exposure to
predator presence in some species has been shown to be highly impactful on larval vital rates.
For example, Ae. notoscriptus exposed to the presence of predatory fish were smaller, and had
slower growth (van Uitregt et al., 2012). One possibility for the results here is that inadequate
82
Texas Tech University, Daniel Dawson, August 2016
predator chemical cues reached larvae within the enclosures. However, given that dragonfly
nymphs were in continuous contact with aquatic media throughout the experiment, including
several days prior, and were fed Cx. tarsalis larvae for the majority of this time, this is unlikely.
Another possibility is that responses to predators vary by species. For example, the vital rates of
some mosquito species (Cx. quinquefasciatus, Culiseta longareolata) were not impacted by the
non-consumptive presence of dragonflies and damselflies, while the development rate and size
of another (Cx. sinaiticus) was reduced (Roberts, 2012). Although prior research has
demonstrated that consumptive predation (i.e., density-mediated influences, DMI) can be a
dominant factor in the regulation of Cx. tarsalis larval populations (Bence and Murdoch, 1983;
Walton et al., 1990), non-consumptive predation effects (i.e., Trait-mediated influences, TMI)
may be less important. In contrast, indirect predator consumption (i.e., consumption of
competitors and predators), although not assessed here, has a potentially significant effect on
Cx. tarsalis ecology. For example, Cx. tarsalis larvae reared in the presence of mosquitofish
(Gambusia affinis) but excluded from consumptive predation had higher survival and emerged
faster than larvae not in the presence of fish, due to the reduction of mosquito competitors
(Blaustein and Karban, 1990) and insect predators (Bence and Murdoch, 1983) by the fish.
Because similar sized competitors and aquatic predators were screened out of playa water
when it was collected, and food was added on a per-larvae basis, a non-effect of predator
presence here is consistent with these previous studies.
4.4.3. Overall comparison
A comparison of the two experiments shows some similarities and some distinct
differences. Both experiments found that Bti concentration had no effect on fecundity or egg
viability. However, the MANOVA considering fecundity and egg viability in experiment 1 was
limited to only the playa water group, and the MANOVA in experiment 2 did not include H
concentrations. Therefore, the ability to detect differences between all treatment groups for
83
Texas Tech University, Daniel Dawson, August 2016
these variables was limited. Another similarity is the relative lack of relationship between wing
size and fecundity. Wing sizes overall were larger (Fig. 4.6, Table 4.2) and development rates
were slower (Table 4.2) in experiment 1. This was likely due to lower temperatures during
experiment 1 (Fig. 4.2), and the well documented effects of lower temperature of these
characteristics (A.N. Clements, 1992c). Despite this, average overall fecundity between the two
experiments were similar (Exp1:99.36 ± 3.17 (SE), Exp2: 90.91 ± 5 .67(SE)). In addition,
although the linear model of fecundity as a function of wing size was statistically significant in
experiment 2, its predictive ability was still very low (R2 = 0.22). Together, this suggests that
wing size in Cx. tarsalis is not an effective predictor of fecundity. Fecundity in mosquitoes can
be highly variable (Mahmood et al., 2004; Reisen et al., 1984; Styer et al., 2007), and depends
on multiple factors including body and wing size (Blackmore and Lord, 2000; Briegel, 1990; Zhu
et al., 2014), blood meal size (Zhu et al., 2014), temperature (Ciota et al., 2014), and age of the
female (Akoh et al., 1992; Mahmood et al., 2004; Reisen et al., 1984). In these experiments, all
females were fed to repletion as far as could be assessed, were kept at the same temperature,
were in their first gonotrophic cycle, and controlled for size. Therefore, other factors associated
with fecundity, such as more comprehensive wing morphometrics or body weight, may need to
be explored to better explain fecundity variability.
A difference of note between the experiments is that the emergence rates of the playa
group in experiment 1 (e.g., H: 0.25) differed noticeably from those in the non-predator
exposure group in experiment 2 (e.g., H: 0.07), even though Bti concentrations were the same,
and conditions (e.g., source water, density, feeding ratios) were similar. However, experiment 1
was conducted in an outside enclosure under a varying temperature regime that averaged
approximately 20°C (Fig. 4.1A), while experiment 2 was conducted inside incubators with a set
25°C temperature. Because Bti toxicity is known to increase with increasing temperature
(Boisvert and Boisvert, 2000; Sinegre et al., 1981), higher emergence rates at the same
84
Texas Tech University, Daniel Dawson, August 2016
concentration in experiment 1 may have been due to the temperature difference with
experiment 2. To assess this possibility, a follow-up experiment was conducted in which second
instar Cx. tarsalis (10 per replicate) larvae were exposed to varying concentrations of Bti (0,
0.03125, 0.0625, 0.125 , 0.25 pts/acre) in 100 ml mod hard laboratory media at either 20°C or
30°C (5 replicates each) for 48 hrs., at which time larval survival was assessed and LC50 values
were calculated using the “Mass” package (Venables and Ripley, 2002). This ad hoc experiment
found that the LC50 for larvae reared at 20°C was significantly higher (LC50: 0.051 ± 0.002) than
those raised at 30°C (0.034 ± 0.003), supporting the supposition that the difference in
emergence rates was due to temperature. It should be noted that experiment 2 was conducted
towards the end of the “mosquito season” (October), and that average temperatures at or
exceeding that used in experiment 2 (25°C) would be present throughout the height of the
season in Lubbock, TX (June-August). Therefore, Bti sensitivity in experiment 2 is likely to be
better reflective of field-sensitivity to Bti in playa water for more of the mosquito season in the
Lubbock region.
4.4.4. Conclusions
Together, the results of the two experiments suggest a few main points. First, the
aquatic media that Cx. tarsalis is reared in can have significant impacts on its sensitivity to Bti
pesticides. Secondly, as demonstrated by lower emergence rates at M and H Bti concentration
in wastewater compared to playa water, these impacts can be counter-intuitive, and likely
situation specific. In situations in which Cx. tarsalis are reared in water with higher organic
activity but without high ionic constituents and potential organic contaminants, the opposite, and
indeed more expected effect of aquatic medium on Bti concentration may be anticipated. Next,
unlike the prominent impacts of direct and indirect consumptive pressure noted in other studies,
predator presence alone does not appear to affect Cx. tarsalis vital rates. Lastly, temperature is
an important consideration when controlling mosquitoes with Bti, with lower temperature
85
Texas Tech University, Daniel Dawson, August 2016
increasing the tolerance of Cx. tarsalis to Bti. However, even though lower temperature also
result in larger adults, this may not translate to significantly higher fecundity.
Acknowledgments
I acknowledge Lucas Heintzman for his efforts in helping to prepare for this work.
Figure 4.1. Outdoor enclosure used in experiment 1 (A); Arrangement of glass containers within
water-filled boxes within outdoor enclosure (B); Aquaria with internal predator exclusion cage,
and Velcro top used in experiment 2 (C).
86
Texas Tech University, Daniel Dawson, August 2016
Figure 4.2. Average daily temperature (A) and light (B) range (minimum, average, maximum)
during experiment 1. Date range includes two days prior to the introduction of larvae to the last
day larvae were in aquatic phase. The large reduction in temperature and light variation after
October 25 is due to containers being brought inside to protect against forecasted outside
freezing temperatures. Temperature is expressed in °C, and light is expressed in lumens/ft2.
87
Texas Tech University, Daniel Dawson, August 2016
Figure 4.3. Average emergence rate (A) and average time to female emergence (B) by
treatment combination, including Bti concentration (Control, Low (L, 0.0625 pts/acre), Medium
(M, 0.125 pts/acre), and High (H, 0.25 pts/acre)), and aquatic medium (Playa or Waste) in
experiment 1. Letters denote significant differences (p<0.05) based on Tukey’s HSD.
88
Texas Tech University, Daniel Dawson, August 2016
Figure 4.4. Average wing size (mm) of females by treatment combination, including Bti
concentration (Control, Low (L, 0.0625 pts/acre), Medium (M, 0.125 pts/acre), and High (H, 0.25
pts/acre)), and aquatic medium (Playa or Waste) in experiment 1. Letters denote significant
differences (p<0.05) based on Tukey’s HSD.
89
Texas Tech University, Daniel Dawson, August 2016
Figure 4.5. Average emergence rate by treatment combination, including Bti concentration
(Control, Low (L, 0.0625 pts/acre), Medium (M, 0.125 pts/acre), and High (H, 0.25 pts/acre)),
and predator presence (No Predators, Predators). Letters denote significant differences
(p<0.05) based on Tukey’s HSD.
90
Texas Tech University, Daniel Dawson, August 2016
Figure 4.6. Relationship between female wing length (mm) and fecundity (i.e, number of eggs
laid) in experiment 1 (open squares) and experiment 2 (filled squares). Shown are the linear
models constructed for both experiments, the R2, and the line of best fit.
91
Texas Tech University, Daniel Dawson, August 2016
Table 4.1. Values of water characteristics measured by analytical determination of frozen
sample (Anions, Cations, Organic Contaminants), and at the beginning of the experiment
(Turbidity, pH, Nitrates, Specific conductance) in playa and wastewater media. In addition, the
ratio of the playa values compared to the wastewater value are shown for each characteristic.
Water Quality Meausure
Anions (ppm)
Analyte
+
Ratio (Playa/Wastewater)
7.41
147.34
19.89
NH4+
BQL
42.57
85.14
2.38
19.58
8.23
+
+2
Mg
2.24
58.32
26.01
Ca+2
40.70
156.00
3.83
2.88
25.54
8.87
14.09
167.52
11.89
6.05
1.36
ND
ND
ND
83.87
1.99
BQL
ND
ND
13.86
1.46
F
-
Cl
Organic Contaminants (ppb)
Wastewater
Na
K
Cations (ppm)
Playa
-
SO4-2
Ciprofloxacin
Lincomycin
Oxytetracyclin
Acetaminophen
Turbidity (ntu)
23.79
114.00
4.79
pH
8.05
7.87
0.98
Nitrates (ppm)
<0.25
<0.25
NA
Specific Conductance (μS/cm)
194
800
4.12
BQL = Quantification Limit; QL for Anions = 0.5 ppm; QL for Organic Contaminants = 0.2 ppb; ND = Not detected
92
Texas Tech University, Daniel Dawson, August 2016
Table 4.2. All vital rates (averages) measured in both experiments. Bti concentrations include
Control, Low (L, 0.0625 pts/acre), Medium (M, 0.125 pts/acre), and High (H, 0.25 pts/acre).
Missing values in experiment 1 are due to a lack of females laying eggs in wastewater group.
Missing values in experiment 2 are due to no M group being included, and no females (2) in the
H, no predator presence group laying eggs.
Aquatic Media
Playa Water
Experiment 1:
Water Quality
& Bti
Waste water
Predation Presence
Predator absent
Experiment 2:
Predator
Presence & Bti
Predator Present
C
Vital Rate
Emergence Rate (% larvae emerge)
Time to Female Emergence (Days)
Fecundity (eggs laid)
Hatch Viability (% eggs hatch)
Wing Size (mm)
Emergence Rate (% larvae emerge)
Time to Female Emergence (Days)
Fecundity (eggs laid)
Hatch Viability (% eggs hatch)
Wing Size (mm)
0.78
19.89
105.33
0.83
3.75
0.85
18.41
3.68
Vital Rate
Emergence Rate (% larvae emerge)
Time to Female Emergence (Days)
Fecundity (eggs laid)
Hatch Viability (% eggs hatch)
Wing Size (mm)
Emergence Rate (% larvae emerge)
Time to Female Emergence (Days)
Fecundity (eggs laid)
Hatch Viability (% eggs hatch)
Wing Size (mm)
0.66
13.90
90.50
0.63
3.45
0.77
13.10
83.60
0.52
3.46
93
C
SE
0.03
0.29
4.88
0.06
0.02
0.04
0.15
0.03
L
0.68
19.76
93.39
0.71
3.81
0.74
18.80
3.67
SE
0.05
0.19
4.86
0.07
0.03
0.06
0.14
0.04
SE
L
0.05
0.36
0.79 14.29
10.37 96.80
0.11
0.75
0.04
3.51
0.02
0.25
0.72 12.50
8.27 101.58
0.08
0.69
0.02
3.46
SE
0.07
0.99
14.16
0.07
0.05
0.08
0.54
16.34
0.13
0.04
M
0.76
19.56
98.78
0.69
3.79
0.52
18.92
3.75
M
-
H
SE
0.05
0.50
8.64
0.10
0.03
0.05
0.15
0.02
0.32
18.33
106.40
0.88
3.84
0.05
16.50
3.62
SE
-
0.02
13.50
3.47
0.05
11.25
118.25
0.79
3.49
H
SE
0.05
1.01
6.93
0.05
0.06
0.02
0.22
0.15
SE
0.01
2.50
0.15
0.04
1.25
3.65
0.05
0.06
Texas Tech University, Daniel Dawson, August 2016
Table 4.3. Results from all statistical analyses run for experiment 1. Table headings pertaining
to all analyses include the vital rate of interest, the statistical analysis used, the variables
included in the analysis, the number of observations (N). For ANOVA, degrees of freedom (df),
sum of squares of the mean (SSM), F statistic and P-value are shown. For MANOVA, degrees
of freedom, Pillai’s trace, the approximate F value (Approx. F), and P-values are shown.
Vital Rate
Statistical Analysis
Variable
Bti
WQ
Emergence Rate
2-way ANOVA
Bti * WQ
Residuals
Bti
WQ
Female time to Emergence
2-way ANOVA
Bti * WQ
Residuals
Bti
WQ
Wing Size
2-way ANOVA
Bti * WQ
Residuals
Vital Rate
Statistical Analysis Variable
Fecundity, Egg viability
1-way MANOVA Bti
94
N
35
33
124
N
27
df
SSM
3
1.9568
1
0.0696
3
0.2212
27
0.2514
3
15.1
1
11.07
3
1.49
25
34.73
3
0.0489
1
0.3175
3
0.0736
116
2.0755
df Pillai's Trace
6, 46
0.154
F
70.045
7.472
7.919
P-value
< 0.001
< 0.05
< 0.001
3.623
7.967
0.356
< 0.03
< 0.01
0.7849
0.91
17.746
1.371
0.438
<0.001
0.255
Approx. F P-value
0.638
0.699
Texas Tech University, Daniel Dawson, August 2016
Table 4.4. Results from all statistical analyses run for experiment 2. Table headings pertaining
to all analyses include the vital rate of interest, the statistical analysis used, the variables
included in the analysis, the number of observations (N). For the GLMM’s of emergence rate
and female time to emergence, the beta coefficients (Beta), standard errors (SE), test statistics
(t-value), and P-values of all variables are shown. For ANOVA analyzes, degrees of freedom
(df), sum of squares of the mean (SSM), F statistic and P-value are shown. For MANOVA,
degrees of freedom, Pillai’s trace, the approximate F value (Approx. F), and P-values are
shown.
Vital Rate
Emergence rate
Female time to
Emergence
Vital Rate
Wing Size
Vital Rate
Fecundity, Egg viability
Statistical Analyses
Variable
Intercept
Concentration L
Concentration H
GLMM
Predation
Concentration L * Predation
Concentration H * Predation
Intercept
Concentration L
Concentration H
GLMM
Predator Presence
Concentration L * Predator Presence
Concentration H * Predator Presence
Statistical Analyses
Variable
Predator Presence
Bti
2-way ANOVA
Bti * Predator Presence
Residuals
Statistical Analyses
Variable
Bti
2-way MANOVA Predator Presence
Bti * Predator Presence
95
N
24
20
N
92
N
39
Beta
0.656
-0.300
-0.620
0.081
-0.179
-0.071
14.165
0.389
-0.102
-0.166
-0.767
-0.967
Df
1
2
2
86
Pillai's Trace
0.058
0.027
0.005
SE
0.064
0.061
0.063
0.062
0.088
0.088
0.834
0.519
0.755
0.542
0.772
0.967
SSM
0.0013
0.0185
0.022
df
2, 34
2,34
2, 34
t-value
10.197
-4.933
-9.875
1.312
-2.033
-0.810
16.991
0.750
-0.135
-0.307
-0.993
-0.999
F
0.05
0.359
0.428
P-value
<0.001
<0.001
<0.001
0.211
0.061
0.431
<0.001
0.471
0.895
0.765
0.344
0.341
P-value
0.823
0.699
0.653
Approx. F P-value
1.050
0.615
0.479
0.351
0.080
0.923
Texas Tech University, Daniel Dawson, August 2016
CHAPTER V
MODELING SPATIALLY EXPLICIT MOSQUITO POPULATION
DYNAMICS WITH THE RNETLOGO PACKAGE
Abstract
Mosquitoes pose risks to humans, both as nuisances and vectors of disease. They have
a complex ecology comprised of an aquatic larval phase and a highly mobile terrestrial adult
phase, and numerous factors influence their population dynamics at a landscape scale. Thus,
control of mosquitoes and the management of risk posed by can be challenging. Spatially
explicit population models can help by providing predictions of mosquitoes, and quantifications
of risk, given conditions. In this chapter, I describe the development of a spatially explicit
population model for the western encephalitis mosquito, Culex tarsalis, in the representative
landscape of Lubbock County, TX. The model works by simultaneously using a matrix model
approach in program R to model the aquatic phase, and an individual-based model approach in
NetLogo using the package RNetLogo. After development, I carry out a sensitivity analysis and
a model evaluation, and demonstrate a potential model application. Model sensitivity analysis
showed that the function governing oviposition movement, adult survival and starting conditions
are the most important parameters influencing overall population dynamics. The model
evaluation showed that the model was able to reasonably replicate field-collected surveillance
data. Lastly, the model application scenario found that a risk-based treatment strategy can
disproportionately reduce mosquito contact risk compared to overall mosquito population. In
regards to mosquito control, this research demonstrates that this model has potential to be a
useful management tool.
96
Texas Tech University, Daniel Dawson, August 2016
5.1. Introduction
Mosquitoes pose risks to humans, both as nuisances and as vectors of disease. As a
group of organisms, they are highly diverse in their use of both aquatic breeding locations and
terrestrial habitats. The result is that mosquitoes occur in a variety of environments, from
tropical to artic ecosystems (Darsie and Ward, 2005), and the aquatic systems in which they lay
their eggs can range from small containers to a large diversity of natural wetlands (Rey et al.,
2012). The threats posed by mosquitoes to humans are not uniform, since different species 1)
vary in their terrestrial and aquatic habitat use, 2) have greater or lesser tendencies to use
humans as hosts, and 3) differ in their capacities to vector particular pathogens (A.N. Clements,
1992a). Coupled with the fact that many mosquito species are highly mobile (sometimes flying
long distances; Verdonschot and Besse-Lototskaya, 2014) can make managing the risk posed
by mosquitoes a challenging task.
In the western United States, a particular mosquito of concern is the “Western
Encephalitis mosquito” Cx. tarsalis, a wide-ranging species that is known to vector several
pathogenic viruses, including Western Equine Encephalitis virus (Mahmood et al., 2004) and
West Nile Virus (Clements, 2012). Cx. tarsalis is able to disperse relatively long distances
(>8km in a night; Bailey et al., 1965), inhabits both anthropogenic (e.g., croplands, developed
areas) and natural habitats (Reisen et al., 2003), and utilizes a variety of wetlands depending
upon availability in the landscape. In the Southern High Plains of Texas, a dominant wetland
type are playa wetlands, which are shallow, depressional, ephemeral wetlands widely
distributed across the region (Haukos and Smith, 1994). Culex species in general (Ward, 1964,
1968), and Cx. tarsalis in particular (personal observation), is known to use the vegetated edges
of playa wetlands as oviposition habitat, making these habitats important drivers of mosquito
populations in the region.
97
Texas Tech University, Daniel Dawson, August 2016
Lubbock county is a semi-urban county located in the Southern High Plains of Texas
(Fig. 5.1) and is a representative example of a semi-urban landscape in the region where
mosquito-borne disease risk has historically be recognized as a significant threat (Harmston et
al., 1956). As of this writing, mosquito populations are actively managed by a county-wide
mosquito control program known as Lubbock County Vector Control. This organization utilizes
methods common to mosquito control in general, and includes targeted management of adults
and larvae, and conducting surveillance of their populations (personal observation). First, adults
are controlled via the application of aerosolized insecticide (e.g. adulticides) via the use of ultralow volume (ULV) spray devices, mounted on trucks. To reduce larval populations, aquatic
habitats are surveyed for larvae by technicians, and then treated with chemical (larvicides) or
biological means (e.g. introducing predators) if larvae are found. Surveillance of adults is
conducted using a network of traps distributed throughout the county and operated on a weekly
basis during the mosquito season. Information collected about the distribution of larvae and
adult mosquitoes from surveillance efforts are used by vector control management to make
decisions about where to direct treatment resources in the immediate future. This is a semiquantitative process in which the experience of a manager is combined with newly collected
information to make such decisions and is likely reflective of how many mosquito control
departments throughout the United States operate (personal observation; based on
conservations with vector control managers at 2015–2016 American Mosquito Control
Associated Annual Meetings).
Although the control strategy described is generally thought to be successful at reducing
mosquito populations, it is less clear how this strategy actually mitigates risk posed by
mosquitos. This is because risk, the chance of an adverse event occurring, is a complex
concept often comprised of different components. Two measures of risk related to mosquitoes
considered here are “contact risk” and “mosquito-borne disease risk.” Contact risk can be
98
Texas Tech University, Daniel Dawson, August 2016
defined as simply the relative risk of a human coming into contact with a questing mosquito, with
the implication that the mosquito can take a blood-meal from a human. In contrast, defining
disease risk is much more complicated, and can be in influenced by a variety of biological and
epidemiological aspects in addition to mosquito and human presence (Eisen and Eisen, 2008).
However, both measures depend upon mosquitoes making physical contact with humans,
requiring the spatially explicit estimation of mosquito populations.
To that end, a number of spatially explicit mosquito population models have been
developed to provide insight into population drivers and patterns of spatial heterogeneity at a
landscape scale (Schurich et al., 2014; Yoo, 2014), with some incorporating mosquito
population predictions into estimates of risk (Pawelek et al., 2014; Tachiiri et al., 2006; Winters
et al., 2008). In addition, a few models have integrated mosquito control impacts as drivers of
mosquito population dynamics (Magori et al., 2009; Pawelek et al., 2014). Although weatherdriven, spatially implicit population models for mosquitoes have built for the Southern High
Plains of Texas (Erickson et al., 2010a, 2010b), to my knowledge no spatially explicit models for
mosquitoes have been developed to date. Such a model could be helpful in mosquito control
management decisions, particularly if the explicit management of risk was a goal.
In this study, I describe the development, analysis, and application of a spatially explicit
population model of the mosquito, Cx. tarsalis, in Lubbock County, Texas. This mechanistic
model captures larval population dynamics in individual oviposition sites over time, thus allowing
for temporally and spatially specific impacts of larvicide treatment to be explicitly account for at
the landscape scale. In addition, after simulation of mosquito populations, mosquito-abundance
projections are combined with human density information to translate mosquito population
predictions into spatially explicit estimations of contact risk. Following the description of model
development, I describe an analysis of model sensitivity using a Monte Carlo approach, a model
evaluation using surveillance data, and present a contact risk-based treatment scenario.
99
Texas Tech University, Daniel Dawson, August 2016
5.2. Methods and materials
The overall modeling approach simulates the aquatic and terrestrial phases by separate
means, but simultaneously, such that the ecologies of both phases are best accounted for.
Specifically, a matrix model approach is used to model the aquatic phase, while an individualbased model (IBM) is used to model the adult phase. This was accomplished by coupling
program R (R Core Team, 2015) and NetLogo (Wilensky, 1999) using the R package RNetLogo
(Thiele, 2014; Thiele et al., 2012). Although conceptually similar to that of the previously
published model SkeeterBuster (Legros et al., 2011; Magori et al., 2009), it differs in that it is
designed to capture population dynamics at a landscape scale, and gives spatially specific
predictions of mosquito population densities. It also leverages the power of the R statistical
language, which includes a rich analytical toolset and flexible graphical outputs, to translate
mosquito population predictions into measures of risk.
5.2.1. Program R: aquatic phase matrix model
5.2.1.1. Model structure
The aquatic phase of the mosquito life cycle is accounted for via a stage-based matrix
model approach (Carrington et al., 2013; Erickson et al., 2010b) using program R, in which the
aquatic population within each breeding wetland within a landscape is individually modeled. This
capitalizes on the fact that all the mosquito larvae within a particular wetland are more likely to
experience the same or similar conditions than larvae in other wetlands.
An underlying assertion of this modeling approach is that heterogeneity in conditions at
individual oviposition wetlands influences overall population dynamics. Therefore, the conditions
at each wetland were modeled separately, which was accomplished by allocating separate
transition matrices for each wetland included in the model landscape. The dimensions of each
transition matrix (A) was 3 x 3, and included vital rates for an egg stage, a combined larvaepupae stage, and an adult stage, as shown below.
100
Texas Tech University, Daniel Dawson, August 2016
A
=
Egg(s)
Egg(t)
0
0
Aquatic(s)
Aquatic(t)
0
0
Adult(1)
Each matrix was accompanied by a projection matrix (N) of dimension 3 x n, where n = the
number of time steps in the simulation. For each stage except adults, daily development rates
(r) were calculated by taking the inverse of the mean time until event (e.g., daily time to
emergence (DTE = 1/mean time to emergence). Daily stage survival rates (d) were calculated
by raising average stage-based survival rates (a) to the p power, with p equaling the daily
development rate (d = ap; e.g, daily emergence rate (DER) = average emergence rateDTE). The
daily probability of surviving and staying within a stage (stage, s was calculated by multiplying
d*(1-p). The probability of surviving and transitioning to another stage (stage (t)) was calculated
by multiplying d*p. In the case of adults, calculated numbers of individuals produced each day in
the projection matrix were passed to the IBM in NetLogo via the RNetLogo package. After this
occurred, the adult stage in the projection matrix was set to zero. To introduce new eggs into a
matrix model, eggs had to be laid into oviposition wetlands in the NetLogo model (as described
below) corresponding to the correct projection matrix in program R, after which they were
allocated to the first column of the projection matrix. Daily model projections were calculated as:
𝑁𝑡+1 = 𝐴 × 𝑁𝑡 .
Lastly, in order to transition the continuous population projections produced by the matrix model
to the discrete population dynamics of the IBM, matrix projections were rounded down to the
nearest integer on a daily basis.
5.2.1.2. Matrix model parameterization
Vital rates for the matrix model were largely derived from two experiments with Cx.
tarsalis under both outdoor (fluctuating temperature) and indoor (constant temperature)
conditions (see Chapter 3) in which larvae were reared under various exposure conditions.
101
Texas Tech University, Daniel Dawson, August 2016
Common to both experiments were treatments in which larvae were reared under non-exposure
(i.e. “control”) conditions in water collected from four playa wetlands within the city of limits of
Lubbock. Only groups and individuals in these control treatments were used as vital rates in the
model. Parameters solely derived from experiments included aquatic phase survival and egg
hatching rate. Time to hatching was assumed to be two days, which is consistent with egg
hatching in insectory conditions (personal observation). Aquatic phase development was driven
by temperature, and experiments were conducted over a relatively narrow temperature window
(19°C and 25°C). Therefore, experimental data was combined with published data (Milby and
Meyer, 1986; Reisen et al., 1989) to develop a simple linear model (R2=0.42, df=16, p=0.001) to
predict aquatic phase development as a function of temperature of the form:
𝐷𝑒𝑣𝑒𝑙𝑜𝑝𝑚𝑒𝑛𝑡 𝑟𝑎𝑡𝑒 (𝑑𝑎𝑦𝑠) = −0.4693 ∗ 𝑇𝑒𝑚𝑝𝑒𝑟𝑎𝑡𝑢𝑟𝑒(°𝐶) + 25.2098.
All model parameters used in the matrix models are listed in Table 5.1.
5.2.2. Adult phase: NetLogo IBM
5.2.2.1. Model structure
The adult phase was modeled via an IBM, which accounts for the activities of every
individual in a simulation, separately. IBM’s are useful techniques when system properties are
thought to emerge from the behaviors of individuals (Grimm and Railsbeck, 2005). In the case
of adults, an individual is expected to contribute to population dynamics and risk conditions
based on its location and its individual life history. In a simple example, adults in a part of a
landscape that is adulticided may have a lower chance of survival and lower reproductive
potential than adults in non-treated parts. By capturing individual variability of adults in the
system, an IBM can reproduce the spatially heterogeneous population dynamics expected to
emerge. Recent studies have taken advantage of spatially explicit IBM’s to model mosquito
population dynamics to understand temporal fluctuations in weather-driven population dynamics
102
Texas Tech University, Daniel Dawson, August 2016
(Jian et al., 2014) and to predict mosquito population dynamics to inform disease-vector control
(Magori et al., 2009). In addition, other IBM’s of mosquito populations have been used to model
mosquito-borne disease transmission (Karl et al., 2014), and to understand how the spatial
locations of bird roosts and oviposition habitat influence viral amplification (Shaman, 2007).
Therefore, IBM’s are a powerful approach to incorporating the contributions of individual adult
life history in understanding mosquito population dynamics.
In natural systems, the life cycle of female mosquitoes is governed by reproductive
activity patterns. After emergence, females pass through a pre-blood feeding phase lasting
approximately 2 days, during which oogenesis proceeds to the pre-vitellogenic phase in
preparation for the acquisition of a blood meal (A.N. Clements, 1992d). During this time,
females seek out nectar-bearing plants to replenish their energy stores and they may mate,
although it is uncertain whether females mate before or after taking a blood-meal (Clements,
1999). After this time, females begin seeking a blood-meal host. Upon taking a blood-meal, they
enter their first gonotrophic cycle (A.N. Clements, 1992d), a period lasting several days during
which their oocytes develop into mature follicles (A.N. Clements, 1992d). Lastly, females seek
out an appropriate oviposition water source, oviposit, and the GC starts over again. The length
of GC’s are inversely proportional to temperature, with females continuing to repeat the process
until they die (A.N. Clements, 1992d).
To approximate this basic ecology of female Cx. tarsalis mosquitoes, the NetLogo IBM
employs relatively simple rules to govern behavior. An overview of these rules is shown as a
decision tree in Fig. 5.2. Overall, individual life histories in the model are dictated by their stage
(pre-blood feeding, questing-egg development, gravid), and operate on a daily time step. After
emerging, a mosquito ages one GC day, and a probabilistic function (similar to NetLogo Simple
Birth Rates model; Wilensky, 1997) uses the specified adult daily survival probability to
determine whether a female lives or dies. No maximum life span is specified in the model, so
103
Texas Tech University, Daniel Dawson, August 2016
the length of any female’s life is a function of the survival rate. Next, its GC age is checked
against the GC length required to lay eggs, depending upon the specified temperature. Females
in their first GC (GC1) must wait a longer time compared to subsequent GC’s (GCsub) before
they can lay eggs, as GC1 incorporates the pre-blood period discussed above. If a female’s GC
age is sufficient to lay eggs, then their movement behavior is dictated by an oviposition
movement function. If it less than the GC length, the mosquito moves according to dispersal
functions. Both oviposition and dispersal movement function parameterizations are discussed
below. If individuals are moving according to the oviposition movement function and move to a
suitable oviposition wetland habitat, they oviposit a number of eggs dictated by a fecundity
parameter. The number of eggs laid is then multiplied by an assumed sex ratio (generally 0.5)
so that only female eggs are laid. Eggs laid in wetlands are passed to the corresponding matrix
model assigned to each wetland via RNetLogo. After ovipositing, females reset their GC ages to
zero, and repeat the cycle until they die.
5.2.2.1.1. Dispersal functions
If dispersal is selected instead of oviposition movement, a probabilistic function based on
a long distance dispersal probability parameter (operating similar to live/die function described
above) first determines whether the mosquito moves a short or long distance in the next day.
This reflects the fact that although long-distance dispersal is frequently observed with Cx.
tarsalis, mark-recapture studies have shown most recaptures (75-84%) tend to occur relatively
close (e.g., <1 km) to release points (Reisen and Lothrop, 1995; Reisen et al., 1992). If a long
distance dispersal is selected, then the mosquito selects a distance (specified by the long
distance dispersal distance parameter) and a random direction, and moves. If a short distance
is selected, then one of two parameterizations, including a random-movement (“random”)
function or a landcover-based (LC-based) probabilistic dispersal function, is used to determine
both the distance and direction of travel of a female mosquito. This reflects two hypotheses as
104
Texas Tech University, Daniel Dawson, August 2016
to how mosquitoes choose to move in the landscape. In the random parameterization,
mosquitoes essentially employ a random walk in their movements. Spatial simulation models
with Aedes aegypti have previously assumed such dispersal behavior for short-distance
behavior (Oléron Evans and Bishop, 2014; Otero et al., 2008), and assume that habitat does not
influence where mosquitoes move, and implicitly, where they find suitable hosts. The LC-based
movement parameterization reflects that host-seeking behavior and capture patterns of Cx.
tarsalis are influenced by landscape composition (Lothrop and Reisen, 2001; Thiemann et al.,
2011). With LC-based movement, the land-cover class (as discussed below) of the cells
surrounding a mosquito’s location are used to determine direction of travel. First, a circular
buffer at the selected dispersal distance (via average daily dispersal parameter) is extended
from the mosquito’s location, and the cells at the dispersal distance are selected and assessed
for their land-cover type. The land-cover values of each cell in the selection are passed to a
probabilistic function (described below), which assigns them a specific probability. These
probabilities are then converted to relative probabilities based on the other cells being
considered. These relative probabilities are then used as weights in a weighted random
selection process (RND extension, Nicolas Payette), in which a particular cell is selected. Lastly,
the mosquito moves the specified dispersal distance in the direction towards the direction of the
selected cell’s center, plus or minus random variation within 10 degrees.
5.2.2.1.2. Oviposition movement functions
As discussed in section 5.2.2.3: Oviposition Wetlands, a number of factors apparently
influence the selection of oviposition habitat, including chemical emissions from various sources
(e.g. bacteria, larvae, eggs, and predators, among others), visual appearance, water movement,
and elevation (A. N. Clements, 1999b). However, it is unknown at what distance Cx. tarsalis can
detect potential oviposition sites in the first place. To reflect this uncertainty, two possible
parameters govern the ability of females to find oviposition wetlands. In the “wetland search”
105
Texas Tech University, Daniel Dawson, August 2016
parameterization, it is assumed that mosquitoes can detect wetlands only within a distance of
their typical dispersal distance of a few hundred meters. In the “nearest wetland
parameterization”, it is assumed that mosquitoes can detect available wetlands from an
unlimited distance away. These two parameterizations represent opposite hypotheses regarding
the ability of mosquitoes in general, and Cx. tarsalis in particular, to detect oviposition habitats.
5.2.2.2. NetLogo IBM parameterization
5.2.2.2.1. Individual-based parameters
Parameters for the individual-based model were selected from both experimental and
literature sources. Average daily survival (Nelson et al., 1978; Nelson and Milby, 1980; Reisen
and Lothrop, 1995; Reisen et al., 1992), daily average and long distance dispersal distance
(Bailey et al., 1965; Dow et al., 1965; Lothrop and Reisen, 2001; Nelson and Milby, 1980;
Reisen et al., 1995, 1992), and long distance dispersal probability (Reisen and Lothrop, 1995;
Reisen et al., 1992) were derived from published mark-recapture studies. GC lengths were
temperature dependent, and were based on a simple linear model (R2=0.8, df=11, p=0.001)
constructed from published GC lengths (Nelson et al., 1978; Reisen and Lothrop, 1995; Reisen
et al., 1992, 1991), and was of the form:
𝐺𝑜𝑛𝑜𝑡𝑟𝑜𝑝ℎ𝑖𝑐 𝑐𝑦𝑐𝑙𝑒 𝑙𝑒𝑛𝑔𝑡ℎ (𝑑𝑎𝑦𝑠) = −0.13847 ∗ 𝑇𝑒𝑚𝑝𝑒𝑟𝑎𝑡𝑢𝑟𝑒 (°𝐶) + 8.44.
Fecundity was the only adult parameter collected from experimental data. All adult parameters
used in the model are listed in Table 5.1.
5.2.2.2.2. Development of probabilistic landcover-based dispersal function
Landscape composition impacts habitat use by Cx. tarsalis adults, with different habitats
varying with regard to oviposition suitability and host availability (Lothrop and Reisen, 2001;
Schurich et al., 2014). Thus, Cx. tarsalis may be more or less willing to disperse through
different types of cover. Therefore, dispersal potential through a habitat may be inferred by
106
Texas Tech University, Daniel Dawson, August 2016
relating the proportion of time a mosquito is found at a location (i.e., a probability of occurrence),
with the proportions of different surrounding habitat types. To that end, a land-cover
classification in Lubbock County, TX, for 2014 was combined with Cx. tarsalis occurrence
information at mosquito traps positioned around Lubbock County to construct a function that
influences the directionality of mosquito movement. The land-cover classification was initially
based on a USDA NASS CropScape classification from 2014. This classification was
reclassified via the incorporation of wetlands known to contain standing water during 2014 using
previously published data (Collins et al., 2014; Starr et al., 2016), resulting in a landcover
surface with 11 cover types. This was subsequently reduced to five main cover types, including
row-crop agriculture, grassland/pasture, developed habitat (ranging from residential to industrial
development), shrubland, and standing water/wetlands. Lastly, this surface was reduced in
resolution from the native 30 x 30 m resolution of Landsat 8 imagery to 120 x 120 m due to
NetLogo computational limitations. As can be seen from Fig. 5.1, row-crop agriculture is the
dominant non-developed land-cover type, followed by grassland/pasture. Shrubland is largely
represented in the eastern side of the county along a wide riparian corridor. Wetland land-cover
consists of playa wetlands, as well as riparian areas.
The city of Lubbock surveyed for mosquitoes from June-November 2014 at 24 locations
using New Jersey Light Traps, a common type of trap used by mosquito control authorities that
uses light as a lure. Light traps were located throughout the county, and included both urban
and rural locations. Cx. tarsalis was detected at all trapping locations over the course of the
season, indicating that it was present throughout the county in all habitat types. However,
detection rates over the season were highly variable from trap to trap, indicating differential
habitat use by Cx. tarsalis.
To relate landcover to species occurrence, the landcover dataset described above, and
the spatial locations of the 24 light traps were imported into ArcGIS 10.2. Landcover information
107
Texas Tech University, Daniel Dawson, August 2016
around traps was extracted using 500 m buffer masks extended around each trap. The
proportion of each habitat type within the buffer around each trap was calculated by first
converting the masks to polygons using the landcover class as the grouping factor, and
calculating the area of each land cover group. Next, the proportion of each landcover type within
each 500 m buffer was calculated.
The proportion of occurrence of Cx. tarsalis at each trap was calculated over 24 total
weekly sampling events occurring from June 3 to November 10, 2014. This proportion of
occurrence was then regressed against the proportions of the five habitat types listed above
using logistic regression, with each proportion of occurrence weighted by the total surveys
considered. Data exploration revealed that that two habitat types, including standing
water/wetland (“wetlands”) and forest/shrubland (“shrubland”) cover types tended to make up
very small proportions of landcover near light traps and were infrequently represented. In
contrast, the other three habitat types (cropland, grassland/pasture, and developed) were widely
variable and frequently represented. To avoid biasing the probability of occurrence in wetland
and shrubland cover types, they were excluded from the statistical model, and assumed to have
proportional probabilities of occurrence (i.e. 1/5=0.2). The other three cover types were entered
into a logistic regression of the form:
𝑙𝑜𝑔𝑖𝑡(𝐶. 𝑡𝑎𝑟𝑠𝑎𝑙𝑖𝑠 𝑜𝑐𝑐𝑢𝑟𝑟𝑒𝑛𝑐𝑒) = −0.726 + 1.164 ∗ 𝐶𝑟𝑜𝑝𝑙𝑎𝑛𝑑 + 2.419 ∗ 𝐺𝑟𝑎𝑠𝑠𝑙𝑎𝑛𝑑 +
1.142 ∗ 𝐷𝑒𝑣𝑒𝑙𝑜𝑝𝑒𝑑.
Probabilities of occurrence were produced for each habitat type by setting the proportion of the
type of interest to 1.0 (i.e. 100%), setting all other proportions to zero, and solving the model. To
utilize these probabilities to inform dispersal behavior, relative dispersal probabilities were
calculated by summing the individual occurrence probabilities for each habitat type, and then
dividing each individual probability by the sum. The individual probabilities for wetland and
108
Texas Tech University, Daniel Dawson, August 2016
shrubland cover types were scaled so that their relative dispersal probabilities would be 0.2.
Relative dispersal probabilities for cropland, grassland/pasture (“grassland”), and developed
habitat were 0.198, 0.234, and 0.167. These probabilities were assigned to the landscape in
NetLogo, and were used in determining mosquito daily movement, as described above.
5.2.2.3. Oviposition wetlands
5.2.2.3.1. Development of oviposition wetland selection model
In addition to playa wetlands, Cx. tarsalis is known to use a variety of shallow freshwater
wetlands throughout its range, including irrigation ditches (Harmston et al., 1956), floodedagricultural fields (Harmston et al., 1956) and rice-fields (Kramer et al., 1988). It is generally
thought to use more recently created, and less polluted aquatic habitats than other congeners
like Culex quinquefasciatus (Reisen and Meyer, 1990). In addition to the multitude of largely
natural playa wetlands (934 noted in Lubbock County; Haukos and Smith, 1994), a large
number of wetlands have been identified in Lubbock county, and include highly modified playa
wetlands in city parks, drainage ditches, and riparian zones. And while individual wetlands can
be readily assessed for the presence of Cx. tarsalis in the field, assessing oviposition habitat on
a landscape scale necessitates a remote methodology. Therefore, a habitat selection survey
was carried out during the 2014 mosquito season in which 35 inundated wetlands selected from
a US Fish and Wildlife National Wetland Inventory (NWI) dataset in ArcGIS 10.3 were surveyed
for Cx. tarsalis larvae. Wetlands in the survey included relatively equal proportions of different
hydro-periods (seasonally inundated, temporary, semi-permanent, and permanent) and NWIS
wetland classifications (lake, freshwater pond, emergent freshwater wetland, and forested
wetland). Sampled wetlands included municipal park playas, riparian wetlands, Lubbock lakes,
golf-course ponds, and private playas. Temporary (n = 7) wetlands incidentally encountered in
ditches and retention ponds were also surveyed, for a total of 42 sampled aquatic habitats.
Wetlands were sampled in Lubbock County from June-July 2014. Larval surveys were
109
Texas Tech University, Daniel Dawson, August 2016
conducted via a standard pint dipper along the edges of the wetland. Wetlands were sampled
proportionally to their size by dipping for one second for every meter of shoreline, with a
minimum of five minutes per wetland. Sampling activity was concentrated in locations with
emergent vegetation, when present, to maximize the chances of detecting larvae. Captured
larvae were returned to an insectory facility, and were identified as fourth instar larvae to
species, or were allowed to emerge and were identified as adults to species.
After sampled wetlands were denoted as being present or absent for Cx. tarsalis,
Landsat 8 imagery was imported into ArcGIS 10.2 corresponding to the dates closest to the
dates in which wetlands were sampled. Next, using the polygon of each wetland as a mask,
cells in each sampled Landsat 8 band were extracted. Then, the mean values of each were
computed for each band for each mask. These values, along with two categorical variables,
including a NWIS wetland classification and a hydro period variable, were considered as
potential predictor variables in a model to predict the probability of occurrence of Cx. tarsalis
larvae. Initial data exploration revealed that spectral bands 1–4 (aerosol (0.43-0.45 nm), blue
(0.45-0.51 nm), green (0.53-0.59 nm) and red (0.64-0.67 nm)), the normalized difference
vegetation index (NDVI, computed as (Band 4 - Band 6)/(Band 4 + Band 6)), and NWIS wetland
type classification (freshwater emergent, lake, freshwater pond, were associated with Cx.
tarsalis presence. These variables were then included in a logistic regression modeling process
in program R, using the glm package. Because bands 1–4 were highly correlated, each band
was included in its own model, along with NDVI and wetland type. Of these, the best supported
model (Band 4) was selected via the Akaike Information Criterion for small sample sizes (AICc).
Next, all seven model combinations of the Band 4 model were assessed via AICc (Table 5.2).
Calculation of AICc weights (Anderson, 2008) showed that no model had overwhelming support
(90% of weight), so an average model was produced using the “natural average” described in
Anderson (2008) over all models. The final average model was:
110
Texas Tech University, Daniel Dawson, August 2016
𝑙𝑜𝑔𝑖𝑡(𝐶. 𝑡𝑎𝑟𝑠𝑎𝑙𝑖𝑠 𝑙𝑎𝑟𝑣𝑎𝑒 𝑝𝑟𝑒𝑠𝑒𝑛𝑐𝑒) = 6.93 + −0.00061 ∗ 𝐵𝑎𝑛𝑑 4 + 0.479 ∗ 𝑁𝐷𝑉𝐼 +
14.82 ∗ 𝐿𝑎𝑘𝑒 − 2.497 ∗ 𝑃𝑜𝑛𝑑 − 2.5 ∗ 𝐹𝑜𝑟𝑒𝑠𝑡𝑒𝑑𝑊𝑒𝑡𝑙𝑎𝑛𝑑 − 1.658 ∗ 𝐼𝑛𝑐𝑖𝑑𝑒𝑛𝑡𝑎𝑙.
5.2.2.3.2. Application of model to landscape
We applied the habitat selection model described above to determine a representative
distribution of breeding wetlands within Lubbock County in which to conduct the simulationbased analyses and applications described herein. First, Landsat 8 imagery taken on August 7,
2014 was selected, as the imagery had <10% cloud cover, and this also coincided with an
active part of the mosquito season in Lubbock county. Multi-spectral Landsat 8 imagery was
downloaded from the USGS Global Visualization Viewer (GLOVIS) as a Level 1 product.
Imagery was imported into ArcGIS 10.3 and clipped to the boundary of Lubbock County. Cells
containing open water were identified using the LS8 band 6 < LS8 band 4 method, modified
from the TM band 5 < TM band 3 method described for Landsat 7 ETM+ imagery (Collins et al.,
2014; Ruiz et al., 2014). In this method, band 6 (short-wave infrared, 1.57-1.65 μm) is
subtracted from band 4 (Red, 0.64-0.67 μm) using raster algebra, with resulting positive cells
indicating standing water. Wetland polygons were derived from the NWIS wetland inventory
wetland polygon dataset. Raster cells identified with standing water were then used to select the
polygons in the NWI wetland dataset containing standing water at the data of the imagery. Next,
using the wetland polygons as masks, the raster cells from Bands 4 and 6 were selected from
each wetland in the dataset, and the mean values of each band for each wetland were
calculated using zonal statistics. Lastly, NDVI was calculated (using methods previously
described), and wetland type information was associated with each wetland. With this
assembled dataset, the model described above was used to predict the probability of
occurrence of Cx. tarsalis larvae in each wetland. With the data used to construct the habitat
selection model described above, Cx. tarsalis were not found in sampled wetlands with
approximately < 20% predicted probability of occurrence. Therefore, to reduce the chances of
111
Texas Tech University, Daniel Dawson, August 2016
falsely dis-including oviposition habitat as predicted by the model, all wetlands with probability
values over 20% were selected for inclusion in the representative landscape. As a last step to
further reduce the number of potential wetlands, and to aid in the importation of wetland spatial
location into NetLogo, polygons within 50 m of each other were consolidated into single
polygons. This process resulted in a total of 199 wetland polygons to be included within the
representative wetland distribution.
5.2.2.3.3. Importing selected wetlands into the NetLogo world
To import wetland locations into NetLogo, wetland polygons were converted to a raster
surface at the same spatial resolution as the landcover data (120 m). Wetland polygon centers
(i.e. points) were also converted to a raster, with a single cell representing the location of each
one. This allowed for very small, isolated wetlands (< 60 m radius), to be included in the wetland
landscape as a single cell. In the NetLogo model, the polygon-derived wetland raster serves as
the wetland locations at which female mosquitoes lay eggs, and the point-based raster serves
as the spatial reference point from which new adults emerge from a given wetland. In this
arrangement, eggs laid into any cell of a wetland by a female are transferred into the matrix
models associated with that polygon. Then, when the matrix model produces new adults, they
emerge into the NetLogo world from the single cell associated with the wetland by the pointbased raster. In this way, the impacts of wetland size and shape differences on mosquito
population dynamics are implicitly incorporated into the model, as mosquitoes are more likely to
encounter larger, longer wetlands than smaller, shorter ones.
5.2.3. Model analyses and applications
To evaluate model behavior, evaluate its accuracy, and demonstrate its potential, three
groups of simulations were conducted. In the first, a sensitivity analyses was completed to
assess model behavior over a range of conditions. Second, the model was run under restricted
conditions, and model results were compared against field-collected surveillance data to
112
Texas Tech University, Daniel Dawson, August 2016
evaluate model predictions. Lastly, to demonstrate a potential application of the model as
mosquito-control tool, a treatment scenario using a contact risk-based approach was simulated.
5.2.3.1. Sensitivity analysis
Assessing model sensitivity of individual-based models is challenging because of the
relatively large number of parameters, and the inherent stochasticity of model simulations due to
the use of random number generators for some parameters (Shaman, 2007). In addition, though
sensitivity and elasticity of matrix model parameters can be analyzed with established sensitivity
and elasticity methods (Caswell, 2001), their contribution to overall population dynamics is more
difficult to ascertain due to the model structure. One way to assess model sensitivity in this case
is to use Monte-Carlo methods. In such a methodology, the model is run over multiple
simulations with parameters of interest allowed to vary over a distribution between simulations.
Then model outputs over the simulation are regressed against model parameter values using a
multiple regression approach, and relative model sensitivity to parameters can be assessed, as
well as model parameter uncertainty.
Applying this approach here, 10 continuous parameters were allowed to vary between
simulations in factorial combinations (n = 8) of three binary categorical model parameterizations.
Continuous parameters included daily adult survival, average daily dispersal distance, maximum
long distance dispersal distance, the daily probability of long-distance dispersal, first and
subsequent gonotrophic cycle length, fecundity, aquatic phase daily survival, time to
emergence, and hatch rating rate. Ranges of continuous variables were randomly selected from
a uniform distribution over a range constructed from both experimental (aquatic phase survival,
fecundity, hatching rate), literature sources (remaining adult parameters), and a mixture of the
two (larval development rate). For adult daily survival, average daily dispersal distance, and
long-distance dispersal distance, ranges included mean ± 1 SD of literature values. For larval
survival, fecundity, and hatching rate ranges included mean ± 1 SD of values collected from
113
Texas Tech University, Daniel Dawson, August 2016
experiments conducted by the author. The range of the probability of long-distance dispersal
was determined from sources describing shorter distance flight to occur between 16 and 25% of
the time, based on mark-recapture experiments (Reisen and Lothrop, 1995; Reisen et al.,
1992). Because time to emergence and gonotrophic cycle lengths are determined by
temperature via linear models (see sections 5.2.1.2 and 5.2.2.2.1), they were varied by allowing
temperature to vary over the ranges at which data was collected.
Categorical model parameterizations included varying the dispersal behavior
parameterization, the oviposition movement parameterization, and a starting egg parameter
between each of two parameterizations. As described in sections 5.2.2.1.1 and 5.2.2.1.2, the
dispersal behavior parameter included “LC-based” and “random” parameterizations, and the
oviposition movement parameter included “wetland search” and “nearest wetland”
parameterizations. The starting egg parameter governed whether wetlands at the beginning of
the simulation started with either one or two egg rafts, with one egg raft containing 48 female
eggs (1/2 of average egg raft based on experimental data, assuming 50:50 female/male ratio).
This parameter was included to assess how starting conditions impacted model dynamics. All
model parameters, as well as the ranges they were allowed to vary over during sensitivity
analysis simulations, are found in Table 5.1.
Model simulations were run for 30 days using the representative distribution of wetlands
(described above) within Lubbock County. Each factorial combination (eight) of the three binary
categorical variables were represented with 25 simulations, for a total of 200 simulations. All
simulations started with the same number of egg rafts (one or two) in each wetland to ensure an
even starting distribution of mosquitoes across the landscape. The total number of adult
mosquitoes at the end of the simulation across the landscape were counted. Using a negative
binomial generalized linear modeling (package nb.glm), these values were regressed against
parameter values used in each simulation. Temperature was used in the place of gonotrophic
114
Texas Tech University, Daniel Dawson, August 2016
cycle and aquatic phase development rate variables in the model. Prior to model construction,
parameter values were subtracted from their means and divided by their standard deviations
(i.e., centered and standardized). Putting parameters on the same scale (i.e. 1 SD) helps in the
interpretation of model coefficients, and centering reduces collinearity between parameters,
aiding in the interpretation of main effects in the presence of significant interaction terms (Quinn
and Keough, 2002). The regression model was assessed for assumptions of independence and
homogeneity of variance via a visual examination of fitted values versus model residuals.
5.2.3.2. Model evaluation
To evaluate model performance, model forecasts were compared with mosquito
surveillance data collected by the Vector Zoonoses Laboratory (VZL) of the Institute of
Environmental and Human Health (TIEHH) at Texas Tech University. The VZL operates five
EVS (encephalitis virus surveillance) adult mosquito traps (known here as Reece, Quaker,
KRFE, MGGC, and Equine traps) stationed around the city of Lubbock during the “mosquitoseason” in Lubbock county, typically from May–October, on a twice-weekly basis (Fig. 5.3).
These traps use CO2 and light to attract female mosquitoes, particularly questing mosquitoes.
To reduce computation time, only wetlands within 8 km (maximum single-day dispersal
considered) of trap locations were selected for inclusion in a truncated representative wetland
landscape, a total of 71 wetlands. Although starting conditions were influential in model
predictions (see 5.3.1 Results: Sensitivity Analysis), it is not known at any point in the season
how many egg masses are in any particular wetland. However, surveillance data is a way in
which the relative distribution of mosquitoes at the beginning of simulations can be influenced.
In this “landscape seeding” methodology, adult questing mosquitoes (>two days old) were
assumed to be at trap locations (i.e., were “seeded”) at the start of model simulations, with the
number per trap corresponding to the average mosquito counts collected during the week the
model imagery (and therefore, the wetland distribution) was collected (August 7, 2014). In
115
Texas Tech University, Daniel Dawson, August 2016
addition, to bolster starting adult mosquito numbers, some of which were quite low (i.e. three at
one location), each wetland in the simulation also included a single egg mass. Simulations were
run using average parameter values (Table 5.1) for population level variables including aquaticphase survival, egg survival, and the probability of long-distance dispersal. As gonotrophic cycle
lengths and aquatic phase development rate were controlled by temperature, values for these
variables was based on the average temperature for Lubbock in August 2014 (27°C). Some
parameters were allowed to vary (mean value ± 1 SD) by individual mosquito, including daily
short and long dispersal distances and fecundity. Lastly, adult daily survival was set at 0.7, and
model simulations were run using the LC-based and nearest wetland dispersal and oviposition
behavior parameterizations. This combination was selected because preliminary analysis
demonstrated that they resulted in population projections that were relatively stable after an
initial period of fluctuation. In addition to the adult survival value being similar to the mean
literature source value (0.65), it was not known whether populations were increasing or
decreasing, so choosing conditions which prompted relatively stable population dynamics was
appealing.
Using the reduced wetland dataset, 100 model simulations were run for 14 days. Model
predictions within a buffer around each trap of 2 * average daily dispersal distance were
extracted, and 95% intervals of distribution of predictions were calculated. Lastly, the average
observed data at each CO2 trap from the period two weeks after the surveillance data used to
initially seed the model was compared to the constructed intervals.
5.2.3.3. Scenario exploration
Two promising potential applications of this model are the spatially explicit estimation of
risk posed by mosquitoes, and the estimation of the effects of management strategies on that
risk. To demonstrate these applications, I applied a simple metric of contact risk to simulation
results to identify high-risk locations in the landscape, and then ran additional simulations
116
Texas Tech University, Daniel Dawson, August 2016
assuming both non-treatment and larvicidal treatment conditions to determine how treatment
would affect risk relative to non-treatment.
First, relative contact risk posed by mosquitoes in Lubbock, assuming non-treatment
conditions, was calculated by combining spatially explicit mosquito predictions produced by
model simulations with human population information. As described below, risk is quantified on
a cell basis within the NetLogo world, leading to the generation of a risk surface. To generate
mosquito predictions from which to calculate the risk surface, 135 model runs were conducted
over a 60-day period using the same parameter set as in the model evaluation (Section 5.2.4),
and the same representative wetland landscape as in the sensitivity analyses described in
Section 5.2.5.1. A 60-day simulation was selected because preliminary work showed that with
the parameterization used, average adult population dynamics become relatively asymptotic
between 30–60 model days, making impacts to populations by stressors easier to discern.
At the end of model runs, model output was exported from NetLogo and converted into a
raster using the “raster” package in program R for further analysis. First, the number of model
runs a mosquito was predicted to be in each grid cell was calculated. To facilitate the
construction of an index of contact risk, this was partitioned into five categories, including 0, 1–
2, 3–4, 5–6, 7–8, and >9. These categories were assigned the values 0–5, with cells never
containing mosquitoes making up the vast majority (>90%), and the last category (including
cells that were occupied in every simulation with >2 mosquitoes) making up less than 0.01% of
observations. Next, human population data from the 2010 census were downloaded by census
block polygon for Lubbock County. This polygon dataset was imported into ArcGIS 10.3 and
clipped to the extent of the county. The areas of the polygons were used to calculate a density
per km2 per polygon. Then the density polygons were converted to a raster at the same scale as
the landcover dataset (120 m), and also clipped to the extent of the landcover data. Next, the
landcover data were classified into 20% quantiles, with categorical values of 0–5 assigned to
117
Texas Tech University, Daniel Dawson, August 2016
each quantile. A value of “0” was assigned to all values of 0 humans per km2. A simple contact
risk index was calculated by overlaying the classified mosquito abundance and human
population rasters, and adding their values (Salice et al., 2016). Then, the raster of combined
values was reclassified into five ordinal categories (0-5), corresponding to cells with combined
scores of 0–4, 5, 6, 7, 8, and 9–10. In this index, the highest category (5) had both high values
of human density and predicted mosquito density (at least 4 or 5 in either category) and
therefore represented locations with the highest overall risk of contact between humans and
mosquitoes. Conversely, the lowest category (1) represented locations in which relatively few
humans or mosquitoes are predicted to be, and therefore constituted the lowest overall contact
risk.
Next, the constructed risk surface was used to explore how a risk-based treatment
strategy may affect risk in the landscape. To achieve this, wetlands that were likely to contribute
to high risk areas were selected based on their proximity to the highest risk cells (category 5),
as calculated above. Wetlands were included in the selection if they were within the polygon
approximation of 2 * the average daily dispersal distance of a high risk cell. To simulate a
larvicide event in these wetlands, simulations were run (described below) in which larvicidal
treatment were assigned to occur on day 41 of the simulation, with treatment reducing larval
daily survival to zero on that day. This represents a scenario in which a Bacillus thuringiensis
israelensis (Bti)-based larvicide is used at a concentration sufficient to cause 100% mortality in
all larvae within a wetland, and is persistent for a one-day period. Experimental data (see
chapter 3) suggests that dosages well within the limits of the label application rates are sufficient
to cause 100% mortality in mosquito larvae in a very short amount of time, often less than 24
hours. The short duration of effectiveness assumed in this simulation was selected because Btibased pesticides often have a very low persistence in the environment due to sedimentation
and breakdown of toxins by bacteria (Lacey, 2007).
118
Texas Tech University, Daniel Dawson, August 2016
Prior to running simulations, the structure of the aquatic phase of the model was slightly
altered to allow for different larval and pupal vital rates, including survival and times to pupation
and times to emergence, respectively. Parameters for these four “new” parameters were
derived from the control groups of experimental data (see chapter 3), and are shown in Table
5.1. This was done because larvicide pesticides only impact feeding larvae, and not non-feeding
pupae. As the parameterization used in previous model runs included a combined larva-pupae
survival and development time parameter, it was not suitable to examine the effects of larvicideinduced mortality on landscape-level population dynamics. All other parameterizations were
identical to those used in the evaluation analysis. Following this alteration, 100 model runs
assuming non-treatment and treatment conditions were run over 60 days.
Overall adult population in both non-treatment and treatment simulations at each of three
time points were compared. Time points included the day of larvicide treatment (41 days), the
day of the point of largest difference in overall average population between treatment and nontreatment conditions after treatment (46 days), and at the end of the model run (60 days). In
addition, risk surfaces were constructed at each time point for both non-treatment and treatment
landscapes, and differences in the proportion of cells in each risk category were characterized.
Lastly, “difference” rasters were created in which risk surfaces of the average non-treatment
simulation was subtracted from the average treatment simulation to provide a visual guide to the
spatial differences in risk following treatment.
5.3. Results
5.3.1. Sensitivity analysis
The negative binomial model constructed for the sensitivity analysis consisted of 14
parameters, including three categorical variables and three interaction terms. A plot of model
residuals against fitted model values showed a non-linear pattern indicative of nonindependence of residuals (Zuur et al., 2009), which was corrected by the addition of interaction
119
Texas Tech University, Daniel Dawson, August 2016
terms, particularly the interaction between the gravid movement parameterization and adult
survival. All model parameters except the long distance dispersal distance and the main effects
of temperature and the movement function were significant (Table 5.3). The gravid movement
parameter had the largest influence on total mosquito numbers, with the nearest wetland
parameterization producing higher numbers of mosquitoes than the wetland search
parameterization. This was followed by adult survival and starting egg masses. The relatively
large coefficient of the starting egg mass parameter demonstrated that starting conditions were
important to model dynamics. Interestingly, average daily dispersal and the probability of long
distance dispersal had opposite signs, implying that longer “short” dispersal distances increased
mosquito abundance, whereas the increased tendency to make long distance jumps decreased
mosquito abundance. Main effects of movement and temperature were non-significant, with the
influence of these variables contingent upon their interaction with the gravid movement
parameter. To understand the nature of the interaction terms, I investigated the effect of each
term individually. For interactions with continuous variables (adult daily survival, temperature), I
solved the model with both oviposition movement parameterizations (“nearest wetland”,
“wetland search”) allowing values for the continuous variable of interest to range between -1
and 1 SD. Mean values were used for all other continuous variables, and 1 egg mass, and LCdependent dispersal was assumed. For the interaction between the gravid movement and the
dispersal behavior parameter, I solved the model using 1 egg mass, and set all continuous
variables to mean values. Plotting model predictions demonstrated that the nearest wetland
behavior increased the effect of both higher adult daily survival rates and higher temperature on
mosquito abundance, relative to wetland search (Fig. 5.4A, 5.4B). In addition, while the effect of
the two dispersal movement parameterizations (LC-dependent, random) on abundance was
very similar when wetland search is employed, mosquito abundances were significantly higher
with LC-dependent movement than random movement when nearest-wetland oviposition
behavior was used (Fig. 5.4C).
120
Texas Tech University, Daniel Dawson, August 2016
5.3.2. Model evaluation with field data
Average observed counts for the period two weeks following the period from which the
model was initialized all fell within the range of model predictions, and four of the five fell within
the range of 95% of predictions (Fig. 5.5). The location of the observed data point within the
distribution of model predictions was not uniform across sites. For most locations, the observed
value fell at or near the edges of the range of 95% of predictions, with the observed value at the
Reece trap, the Equine trap, and the MGBC trap falling between the bottom or upper 5–10% of
predictions. In addition, the observed value for the KRFE trap (3) was the maximum predicted
value. In contrast, the Quaker trap observation fell between the lower 30–40% of observations.
5.3.3. Contact risk scenario
Model simulations run under non-treatment and treatment conditions were similar in
terms of overall population project until the day of larvicidal treatment (41). At this point,
treatment simulations declined until day 46, and then began increasing until day 60 (Fig. 5.6). At
its lowest point (day 46), the average treatment scenario abundance projections were reduced
by approximately 21% (296.5) compared to those of the non-treatment scenario simulation
(374.3). At the last day of the simulation (day 60), the difference in average abundance
projections between treatment and non-treatment scenario simulations was reduced from its
lowest point, but the average treatment scenario abundance projection was still lower by
approximately 15%.
When the spatial distribution of cells in the landscape classified as high risk (category 5)
was compared between the different time points (Treatment Day (41), Minimum Population Day
(46), Final Run Day (60)), the average proportion of cells categorized as high risk was similar
between the two sets of projections on the day of treatment (non-treatment = 0.059%, treatment
= 0.055%), but was lower by 36% in the treatment scenarios (0.039%) compared to the nontreatment scenarios (0.061%) by day 46. By the end of the simulation (day 60), the difference in
121
Texas Tech University, Daniel Dawson, August 2016
average proportion had reduced to 21% (non-treatment = 0.054%, treatment = 0.043%).
Surprisingly, the proportion of cells classified as either 3 or 4 was reduced in the treatment
landscape by 64% and 61% on day 46, respectively, compared to those in the non-treatment
landscape, despite treatment targeting the highest risk cells.
Difference rasters at the three time points show that on day 41 (Fig. 5.7A), few large
differences exist in the landscape, and on day 46 (Fig. 5.7B), clusters of risk reductions from
highest risk (5) to lowest risk (1) category (e.g. -4) occur in the central part of the county. In
contrast, in the northeast part of the county, there were fewer large reductions in risk, and some
increases in risk score. This reflects the fact that because the human population is generally
lower in the northeast part of the county, fewer wetlands were selected for treatment. The
difference raster at day 60 (Fig. 5.7C) showed few large negative differences, reflecting
increases in population following treatment.
5.4. Discussion
5.4.1. Overview
In this study, I describe the development of a spatially explicit model of mosquito
populations in semi-arid environments, and apply it to populations of the mosquito Cx. tarsalis in
Lubbock County, TX. Two model diagnostic analyses were carried out, including a sensitivity
analysis and a model evaluation, followed by the application of the model in a hypothetical
treatment scenario. In the paragraphs below, I discuss the two diagnostics separately and then
provide a synthesis with a discussion of the scenario application. Lastly, I discuss potential
model extensions and applications.
5.4.2. Sensitivity analysis
The sensitivity analysis regression model demonstrated that both adult and aquaticphase parameters were influential on predictions of adult abundance. Of particular influence
122
Texas Tech University, Daniel Dawson, August 2016
was the oviposition movement parameter. This parameter had the largest beta coefficient in the
model, and interacted significantly with three other adult parameters. Of the two
parameterizations of oviposition movement behavior, “nearest wetland” behavior tended to
produce higher abundances in the model than “wetland search” behavior. This is because unlike
wetland search, nearest wetland guaranteed that a mosquito would find a wetland in which to
oviposit if it simply lived long enough and continued to move. In the case of the interactions with
the two continuous variables (adult daily survival and temperature), nearest wetland behavior
increased the positive impacts that higher survival rates and higher temperature would be
expected to have on population size (Fig. 5.4A, 5.4B). In the case of survival rate, gravid
females were more likely to arrive at a wetland to oviposit before they died with nearest wetland
behavior than they were with wetland search behavior. In the case of temperature, higher
temperatures shorten the time of both the aquatic phase and the length of the gonotrophic
cycle. Because moving towards the nearest wetland to oviposit instead of searching for a
wetland further reduces the time to oviposit, impacts of higher temperatures on population
dynamics were magnified. As demonstrated by other modelling efforts, adult survival and
temperature are important drivers of mosquito population dynamics and mosquito-borne
disease risk in natural systems (Ellis et al., 2011; Erickson et al., 2010a; Tachiiri et al., 2006).
With higher survival, more adults will be present on the landscape, and longer living adults have
a greater chance of both becoming infected with a pathogen and passing it to a host (Smith et
al., 2014). Higher temperature can both increase the numbers of mosquitoes present and
reduce the incubation times of viral pathogens (Hardy et al., 1983). Thus far, the potential
interactions of survival and temperature with the ability of mosquitoes to find oviposition habitat
is a seemingly unexplored avenue of research.
Less clearly defined was the interaction between oviposition behavior and dispersal
movement. With wetland search behavior, landcover (LC)-based and random dispersal
123
Texas Tech University, Daniel Dawson, August 2016
movement produced very similar numbers of mosquitoes (Fig. 5.4C). However, the combination
of nearest wetland behavior and LC-based dispersal produced more mosquitoes on average
than the combination of nearest wetland behavior and random movement (Fig. 5.4C).
Functionally, this indicates that if mosquitoes move according to landcover-based probabilities,
they are more likely to be closer to, and more likely to oviposit in, the nearest wetland. This is
likely due to the relative locations of landcover types and the probabilities assigned to them in
the dispersal function. For example, a number of wetlands were located inside the urban
landscape of the city of Lubbock. Of all the landcover types, developed landcover had the
lowest probability of mosquito dispersal (0.16), based on presence/absence data collected in
adult mosquito traps by the city in 2014. Therefore, mosquitoes emerging on wetland landcover
(probability assumed = 0.2) that was surrounded-by or adjacent-to developed landcover had a
higher probability of staying in or near wetland-landcover when dispersing. The addition of the
nearest wetland behavior would likely intensify this effect.
In natural systems, mosquitoes select oviposition habitat using a variety of cues. For Cx.
tarsalis, the presence of eggs and larvae in water can stimulate oviposition (Hudson and
McLintock, 1967), whereas the presence of predators (Walton et al., 2009) and modifications
that discourage the growth of emergent vegetation (Ward, 1968) can be deterrents. In the case
of chemical influences like larval or egg presence, some evidence suggests that mosquitoes
have to “sample” the water via touch before they make a decision to oviposit (Hudson and
McLintock, 1967). Although little is known about the distance at which mosquitoes can detect
oviposition wetlands, some chemical or visual cue, or perhaps a combination of the two, must
be available at some range. Cx. tarsalis has been described as a “dispersive colonizing species”
that takes advantage of newly created surface water (Reisen et al., 2003). Therefore, the ability
to detect new standing water from relatively far distances would be advantageous, particularly in
semi-arid landscapes in which oviposition wetlands may be scarce (Verdonschot and Besse-
124
Texas Tech University, Daniel Dawson, August 2016
Lototskaya, 2014). In addition, although evidence suggests that site fidelity to oviposition
wetlands may be low in Cx. tarsalis (Beehler and Mulla, 1995), mosquitoes have been shown to
exhibit spatial memory for oviposition sites (McCall and Kelly, 2002), and there is some
evidence that Cx. tarsalis may memorize flight paths (Reisen et al., 2003). Therefore, either by
detecting suitable oviposition sites from a distance or by returning to known wetlands, the
nearest wetland parameterization in this model may better reflect natural actual mosquito
behavior than wetland search. However, further study into oviposition site detection by Cx.
tarsalis is required for clarification.
Starting egg masses was the third most influential model parameter, and indicates the
importance of starting conditions in the model. This was further demonstrated during the
evaluation phase, in which different numbers of adults “seeded” at trap locations at the start of
model runs resulted in spatial heterogeneity of adults 1–2 generations later that were reflective
of the starting distribution. Since short-term predictions (e.g. 1–4 weeks) are generally desired
for management decisions, the importance of starting conditions on the use of the model to
make management decisions is therefore quite high. This issue is further discussed in section
5.4.3. (Model evaluation).
Of all of the model parameters found to be significant, only the probability of longdistance dispersal had a negative coefficient. Conversely, average daily dispersal distance was
positively associated with abundance. One potential reason for this is that making longer short
flights may bring mosquitoes closer to oviposition habitats faster, and therefore reducing time
before ovipositing. In contrast, while long flights may bring mosquitoes to new oviposition
habitats, the converse can also happen. The landscape in which sensitivity analysis simulations
were run was static, with wetlands never drying, but also new wetlands never becoming
available. Therefore, this result may be an artefact of model conditions. Under natural
conditions, wetland availability in semi-arid landscapes fluctuates. Therefore, the ability to make
125
Texas Tech University, Daniel Dawson, August 2016
long flights to find new wetlands may be advantageous (Verdonschot and Besse-Lototskaya,
2014). In addition, while long distance movement (both probability and distance) in the model
was random, in real systems it is likely influenced by several factors, particularly wind. Although
dispersal tends to be low during higher wind events (Reisen et al., 2003), wind can facilitate
dispersing Cx. tarsalis up to 25 km in a night (Bailey et al., 1965), greatly increasing its natural
dispersal ability. Wind was not considered here but could be an important factor in determining
the distribution of Cx. tarsalis on the landscape in future iterations of the model.
5.4.3. Model evaluation
The evaluation analysis demonstrated that the model reasonably replicated the spatial
heterogeneity of mosquito numbers at 5 traps located around the city of Lubbock after two
weeks from particular starting conditions. It is important to note that at the end of the 14-day
model run, the vast majority of starting individuals would have died in the model. Thus, the
mosquitoes “captured” within the buffer distance of trap locations were the sum of the adults
produced from the eggs starting in wetlands at the beginning of model runs (one egg mass per
wetland), as well as the offspring of the starting adults. This means that the influence of the
starting adults in the model translated to the spatially heterogeneous distribution of mosquitoes
“caught” by the model at trap locations two weeks later. This suggests that starting conditions
clearly have a significant influence on transient model behavior (Caswell, 2001). Because the
most useful model predictions in regards to mosquito control would be in the short term (weeks
to a month), seeding the model landscape with adults at trap locations based on surveillance
data could thus be a useful technique to account for heterogeneous start conditions.
An important consideration here is that the time scales and data distributions of the IBM
and the matrix models do not align perfectly, due to the discrete time nature of the IBM, and the
stage-nature of matrix models. For example, in the IBM a gonotrophic cycle specified as six
days would mean that the time required to elapse for an individual between taking a blood-meal
126
Texas Tech University, Daniel Dawson, August 2016
to laying eggs has to be at least six days. In contrast, in the matrix model the mean value of a
stage is spread out over the entire length of time specified for the stage. For example, if the
larvae-pupae aquatic phase was specified to last 12 days (depending on temperature), a
fractional amount of larvae-pupae and therefore adults would still be produced at every time
step. The result is that instead of a pulse of adults being sent to the IBM after the completion of
the aquatic stage, the matrix model produces a continuous number over the duration of the
stage. This time disconnect is ignored, however, with the reasoning that after an initial period of
population establishment, population dynamics in the model stabilize, and approach asymptotic
conditions (Caswell, 2001) as demonstrated by the scenario analysis (Fig. 5.6). In the model
evaluation, model predictions may have reasonably aligned with observed values because the
number of adults seeded at trap locations at the beginning of model simulations was
proportional to adults present in the area, and to the aquatic population already present in
nearby wetlands. In natural systems, the developmental stages of both phases would be
expected to be asynchronous, with new eggs being laid and new adults emerging on a daily
basis. Thus, the model was probably able to reasonably reproduce the spatial heterogeneity in
abundance patterns at traps because trap counts reflect general abundance patterns in space.
However, further analysis is needed to determine how temporal inconsistencies between the
modeling phases influence predictive ability.
One approach to better temporally align the matrix and NetLogo models is to convert the
stage-based matrix model to an age-based matrix based on a daily age structure. This would
come at the cost of requiring a different-size matrix for changes in temperature prompting a
greater than one day change in development rate. However, for prediction over the short term (a
few weeks to a month), using a single matrix (i.e. assuming the same temperature) may be
reasonable. For predictions over a longer period, like an entire season, an alternative may be to
employ three-dimensional matrix model (Lončarić and K. Hackenberger, 2013), which can
127
Texas Tech University, Daniel Dawson, August 2016
incorporate the impacts of minimum and maximum expected temperatures on development
rates. However, such an approach may greatly increase computational requirements of the
model without yielding improvements in output.
5.4.4. Scenario exploration
In the treatment scenario explored, the effect of risk-based larvicidal treatment was
clearly to reduce the proportion of high-risk cells in the landscape compared to non-treatment
conditions. The effect of larviciding on day 41, both on overall population projection and the
spatial distribution of high risk cells, was also shown to persist until the end of the model run
(day 60). The difference in the proportion of high risk cells in the treatment surface compared to
the non-treatment surface reduced from a maximum of 36% on day 46 (five days after
treatment) to about 21% on day 60. This indicates that the treatment surface could eventually
approach non-treatment levels, assuming static conditions, although the model was not
projected beyond 60 days.
It should be noted that this particular treatment scenario does not reflect real-world
conditions in several ways that limit its usefulness in making specific management
recommendations. First, due to limitations on access to private property, it is unlikely that
mosquito control personnel would have ready access to every wetland selected for treatment. In
this case, this is partially compensated for by many of the wetlands in locations of high human
populations being located within public parks which can be accessed by county personnel.
Second, because larvicides used in landscape-scale mosquito control are often applied via
truck-mounted equipment (such as is the case in Lubbock County) only wetlands near roads
can be easily treated. Next, due to the limitations of employee time, it is unlikely that all 57
wetlands selected could be treated on a single day. Lastly, as the sensitivity and evaluation
analyses suggest, starting conditions are important to resulting model dynamics. Scenario
128
Texas Tech University, Daniel Dawson, August 2016
simulations started with one egg mass in each wetland and no adults present, conditions highly
unlikely in the real world.
Despite these limitations, model results were informative in the context of mosquito
control for two main reasons. First, they demonstrated that the effectiveness of a larvicide at
reducing mosquito populations and their associated disease risk can be limited to a single day
(assuming 100% mortality) and still result in a significant influence on population dynamics on
the landscape scale. These results provide justification for the use of Bti-based larvicides with
short windows of efficacy (Kroeger et al., 2013). Next, simulations showed that the targeting of
high risk wetlands for treatment can result in a disproportionate reduction of high risk areas on
the landscape. For example, the treatment surface on day 46 had approximately 35%, 64%,
61%, and 61% fewer cells classified as category 5, 4, 3, and 2, respectively, than the nontreatment landscape, while having only 21% total fewer mosquitoes. This was possible because
cells classified as 0 or 1 (i.e., the lowest risk classifications) were the most common cells in
either landscape and more cells were classified as 0 or 1 in the treatment landscape (98%) than
the non-treatment landscape (95%).
The greater reduction in the proportion of cells in risk categories 3 and 4 than category 5
(which was actually the target of treatment) was unexpected. An examination of the risk surface
reveals that this was likely due the distribution of category 5 cells. In particular, some clusters of
category 5 cells were centered over oviposition wetlands themselves, with category 3 and 4
cells spreading away in somewhat concentric circles (Fig. 5.8A). In other situations, higher risk
cells occurred between wetlands, where dispersing mosquitoes would appear to congregate
(Fig. 5.8B). In both cases, there were more category 3 and 4 cells then category 5 cells
associated with wetlands, and therefore more to reduce when treatment was applied.
129
Texas Tech University, Daniel Dawson, August 2016
The second case of high risk cell distribution (i.e., high risks cells between wetlands) is
interesting, as it demonstrates the importance of selection criteria in determining a risk-based
treatment strategy. In this scenario, wetlands were selected for treatment if high risk cells fell
within a particular distance threshold of wetlands, according to initial non-treatment model
simulations. A problem arises if high risk cells fall outside of the distance threshold used, as
wetlands would not necessarily be selected for treatment. One way to address this problem may
be to simply select the nearest wetland to each high risk cell, thereby avoiding the need for a
distance threshold. However, this approach would result in a greater number of wetlands to
treat, which may not be feasible. Evidence suggests that abundances of Cx. tarsalis in natural
systems change little with distance within 400 m of oviposition wetlands up (Barker et al., 2009).
However, it is not clear how abundances of Cx. tarsalis relate to distances to oviposition sites at
larger scales, or between networks of oviposition sites. Overall, this uncertainty suggests that
hypotheses regarding the spatial distribution of mosquito populations in relation to the
distribution of wetlands should be explored. In regards to mosquito control, this information
could help determine treatment selection criteria that optimizes the balance between
organizational priorities and capabilities.
5.4.5. Model potential
The model of mosquito population dynamics in the Southern High Plains of Texas
developed here provides opportunity to explore a variety of hypothetical and real-world
applications. In addition, because of its generality and flexibility, it can be applied to other
landscapes and organisms. Many model limitations, like the ones mentioned in section 5.4.4,
can be accounted for by either incorporating them into model conditions or by altering the
underlying model structure. In addition, the capabilities of program R and NetLogo make
possible a number of model parameterizations and analyses. Below, I briefly discuss two
examples of potential future model applications.
130
Texas Tech University, Daniel Dawson, August 2016
5.4.5.1. Estimating disease risk
The spatial distribution of mosquito-borne disease risk in a landscape is uneven, and is
influenced by the distribution of larval habitat, landcover, blood-meal hosts, and humans (Smith
et al., 2004). Therefore, an important potential application of this model is the estimation of
spatially heterogeneous disease risk, and the exploration of strategies to help in its mitigation.
Risk indices are potentially effective tools for mosquito control authorities to prioritize treatment,
as demonstrated in section 5.4.4. In the above section, contact risk was defined with a simple
index in which human and mosquito information was combined such that locations in which high
human population density and high mosquito density overlapped were deemed to have high risk
of contact. However a number of possibilities exist for the construction of other, more
informative risk indices. For example, a spatially-explicit disease risk index implemented by
Tachiiri et al. (2006) was constructed by multiplying human population with an estimate of
disease condition risk. In another study, a risk index was built by combining epidemiology
information (cases of disease) with mosquito population predictions (Winters et al., 2008). One
way to construct a disease risk index using this model is to base it on the potential number of
infectious mosquitoes in the landscape. For example, adapting methodologies described by
Pawalek (2014) for modeling West Nile Virus (WNV) disease risk, the mosquito population can
be divided into Susceptible, Exposed, and Infectious groups. Susceptible individuals can
become exposed, and eventually infectious individuals according to an estimate of background
rate of infection. Because WNV is transmitted to mosquitoes via birds (Pawelek et al., 2014),
information available on infection dynamics between avian hosts and mosquitoes, including
biting rates, can be considered. In addition, if the location of communal roosts, or areas of high
WNV incidence area known, these can be explicitly incorporated into the NetLogo landscape.
Lastly, if such information is not available, then hypotheses regarding how these factors might
influence disease risk can be explicitly investigated via simulation.
131
Texas Tech University, Daniel Dawson, August 2016
Once a method for estimating the number of infected mosquitoes on the landscape is
determined, a risk index can be constructed. In addition to the methods mentioned above, a
promising avenue of estimating risk incorporates the movement of humans. Human movement
has been identified as an important consideration in estimating disease risk because of its
influence on biting risk and the spread of new infections (Cosner et al., 2009; Smith et al.,
2014). Human movement has been included in a recent IBM of disease risk (Karl et al., 2014)
and could be modeled in NetLogo in a similar way. Upon the construction of a risk index,
treatment scenarios can then be systematically explored based on the capabilities of the
mosquito control organization to determine how to best mitigate risk in the landscape.
5.4.5.3. Other applications
Although the model described here was developed with the eventual goal of being used
as a tool in the field of mosquito control, its general and flexible structure also give it potential as
a tool for ecological study. For mosquitoes specifically, factors such as nutrient availability,
predation, and competition affect larval survival and development, while factors like disease
infection status influence adult survival and fecundity. Such aspects could be incorporated into
simulations to investigate hypotheses regarding their impacts on population dynamics. In
addition to mosquitoes, the model is suitable for simulating populations of organisms with
amphibious life histories, particularly those with aquatic larval forms and terrestrial adult forms.
Lastly, although the model was developed for Lubbock County, its flexible structure can
accommodate any number of other landscapes, including those at larger or smaller scales.
5.4.6. Summary and conclusions
In this study, I described the development of a spatially explicit simulation model for the
estimation of mosquito populations in semi-arid environments. Then, I applied the model to
simulate populations of the mosquito Cx. tarsalis in Lubbock County, TX. A sensitivity analysis
conducted using a Monte-Carlo and regression-based approach demonstrated that oviposition
132
Texas Tech University, Daniel Dawson, August 2016
behavior, along with its interactions with several parameters governing adult life history
characteristics were important factors governing population dynamics. A model evaluation in
which model projections were compared against mosquito surveillance observations
demonstrated that the model was able to reasonably reproduce observed data over the short
term. In addition, starting conditions were influential on transient model behavior. Lastly, a
contact-risk based treatment scenario was simulated in which non-treatment and treatment
model simulations were run over 60 days, and a selected subset of wetlands were treated in the
treatment set. Simulation results showed that a risk-based approach to prioritizing mosquito
control can disproportionately reduce areas of high contact risk in the landscape compared to
overall mosquito abundance. Overall, the model has a high potential to be eventually applied by
mosquito control authorities to mitigate disease risk and to investigate research hypotheses.
Acknowledgments
I acknowledge Lucas Heintzman for preparing the initial landcover dataset used in this
research.
133
Texas Tech University, Daniel Dawson, August 2016
Figure 5.1. Map of position of Lubbock county in the Southern High Plains of Texas (left), and
landcover map of Lubbock county (right). Landcover categories include cropland agriculture
(red), grasslands/pasture (green), shrublands/forest (yellow), developed (white), and
wetlands/standing water (blue).
134
Texas Tech University, Daniel Dawson, August 2016
Figure 5.2. Decision tree describing daily behavior of females in NetLogo model. Live or die is
determined by random chance, weighted by the average daily adult survival parameter. GC
(gonotrophic cycle) length depends on temperature-based parameters for first and subsequent
GC’s. Short or long distance dispersal is dependent upon random chance, dependent upon long
distance dispersal. Oviposition movement and dispersal functions are described in detail in the
text.
135
Texas Tech University, Daniel Dawson, August 2016
Figure 5.3. Location of surveillance traps in Lubbock County, TX, operated by Texas Tech
University, used in model evaluation.
136
Texas Tech University, Daniel Dawson, August 2016
Figure 5.4. Interaction plots of significant interactions identified in the sensitivity analysis
between the Oviposition Movement parameter and A) adult daily survival, B) Temperature, and
C) Dispersal behavior (Landcover-based (LC) or Random). Values plotted are population values
predicted while allowing parameter of interest to vary, either over 1 Standard Deviation (A&B) or
between binary parameterizations (C) for both oviposition movement behaviors (Nearest
Wetland (“Nearest”), and Wetland Search (“Search”)). All other continuous variables are set to
mean values, and number of starting egg rafts is set to one.
137
Texas Tech University, Daniel Dawson, August 2016
Figure 5.5. Prediction distribution and 95% intervals for values at 5 traps used in the model
evaluation for 100 model simulations. Traps include locations known as Quaker, MBGC, Reece,
KRFE, and Equine. Observed values collected at each trap 14 days after day model initiated are
shown as red lines. 95% prediction intervals are shown as dotted blue lines. For all traps except
KRFE, observed value falls within or on (Equine) the range of 95% of predicted values.
138
Texas Tech University, Daniel Dawson, August 2016
Average Population
400
300
200
0
100
Landscape Population
500
600
Control Conditions
Treatment on Day 41
95% Prediction Interval
0
10
20
30
40
50
60
Days
Figure 5.6. Average predicted adult population of treatment scenario simulations over 60 days.
Shown are the average predicted population of 100 control simulations (black, solid line), 100
treatment simulations (red, solid line), and 95% predictions intervals for both (dotted lines). In
treatment simulations, treatment occurred on day 41.
139
Texas Tech University, Daniel Dawson, August 2016
Figure 5.7. Difference rasters at three time points in the simulation, including A) Treatment day
(41), B) the day of maximum average population difference between control and treatment
populations (Day 46), and C) end of the model run (day 60). Surfaces calculated by subtracting
average control scenario from average treatment scenario. Values range from 4(dark red)
indicating a risk increase in the treatment landscape to -4 (dark blue) a risk decrease. Cluster of
dark blue (negative differences on Day 46 (B) indicates maximum reduction from high (5) to low
(1) contact risk, reflecting treatment of wetlands in locations in proximity to highest human
population on day 41.
140
Texas Tech University, Daniel Dawson, August 2016
Figure 5.8. Example of clustering risk cells (Red=5, Orange=4, 3=Yellow, 2=Green, 1=Black,
0=White) in landscape in relation to wetlands (wetland centers shown as blue circles). In case
A, highest risk tends to be centered on wetland locations; in case B, high risk tends to be
highest between wetlands.
141
Texas Tech University, Daniel Dawson, August 2016
Table 5.1. Model Parameters used in matrix and NetLogo models. Parameters included in the
sensitivity analysis, the model evaluation, and the treatment scenario are shown. For the
sensitivity analysis, average values, plus the variable range of 1 SD are listed. Temperatureassociated variables and long distance dispersal probability parameter varied over the minimum
and maximum values listed. In the evaluation section, if there is variable range listed for a
parameter, it was allowed vary over that range (1 SD). Otherwise it was static for the simulation.
The scenario simulations used the same parameterizations as the evaluation simulations,
except for the parameters listed.
Simulation
Model Phase
Matrix: Aquatic
NetLogo: Adult
Sensitivity Analysis
Both
Matrix: Aquatic
NetLogo: Adult
Simulation
NetLogo: Adult
Model Phase
Matrix: Aquatic
NetLogo: Adult
Evaluation
Both
Matrix: Aquatic
NetLogo: Adult
Simulation
Model Phase
Matrix: Aquatic
Scenario
Both
Matrix: Aquatic
NetLogo: Adult
Parameter
Hatching rate
Time to hatching
Aquatic phase survival
Adult daily survivial
Fecundity
Average daily dispersal distance
Long distance dispersal distance
Temperature-Dependent Parameters
Temperature
Aquatic phase development
Gontrophic cycle 1
Gontrophic cycle subsequent
Min-Max Parameter
Long distance dispersal probability
Parameter
Hatching rate
Time to hatching
Aquatic phase survival
Adult daily survivial
Fecundity
Average daily dispersal distance
Long distance dispersal distance
Long distance dispersal probability
Temperature-Dependent Parameters
Temperature
Aquatic phase development
Gontrophic cycle 1
Gontrophic cycle subsequent
Parameter
Larvae survival
Pupae survival
Temperature-Dependent Parameters
Temperature
Larvae development
Pupae development
Gontrophic cycle 1
Gontrophic cycle subsequent
142
Average Value
0.74
2
0.73
0.65
96
0.39
3.21
Min
18
16.76
7
5
Min
0.16
Mean Value
0.74
2
0.73
0.70
96
0.39
3.21
20.50
Value
27
12.53
7
5
Value
0.85
0.85
Value
25
12.31
2.04
7
5
Range (1SD)
0.23
NA
0.083
0.14
28.48
0.321
1.76
Max
28
12.04
8
6
Max
0.25
Range (1SD)
NA
NA
NA
NA
28.48
0.32
1.76
NA
Range (1SD)
NA
NA
NA
NA
Range (1SD)
NA
NA
Range (1SD)
NA
NA
NA
NA
NA
Units
Probability
Days
Probability
NA
Eggs
km
km
Units
°C
Days
Days
Days
Units
Probability
Units
Probability
Days
Probability
Probability
Eggs
km
km
Probability
Units
°C
Days
Days
Days
Units
Probability
Probability
Units
°C
Days
Days
Days
Days
Texas Tech University, Daniel Dawson, August 2016
Table 5.2. Models and AICc model weights for models considered for the selection of
oviposition wetlands.
Model
Band 4 Mean + Wetland Type
Wetland Type
Band 4 Mean + Wetland Type + NDVI
NDVI + Wetland Type
Band 4 Mean
Band 4 Mean + NDVI
NDVI
AICc
51.3
53.2
54.2
55
56.5
57.7
58.2
dAICc
0
1.8
2.9
3.6
5.1
6.4
6.9
df
6
5
7
6
2
3
2
Likelihood
1.00
0.41
0.23
0.17
0.08
0.04
0.03
AICc Weight
0.51
0.21
0.12
0.08
0.04
0.02
0.02
Table 5.3. Parameter estimates, standard errors, and p-values of model parameters of negative
binomial model developed for sensitivity analysis. Estimates are scaled and centered for
parameter comparison.
Model Parameter
(Intercept)
Gravid movement (Search vs Nearest)
Adult survival
Starting Egg Masses (1 or 2)
Gravid Movement * Movement
Gravid Movement * Adult Survival
Hatch Rate
Gravid Movement * Temperature
Fecundity
Larval survival
Average daily dispersal distance
Long distance dispersal probability
Long distance dispersal distance
Temperature
Movement (LC vs Random)
Estimate Std. Error
2.63
0.09
1.52
0.11
1.06
0.05
0.76
0.07
0.60
0.15
0.55
0.08
0.45
0.04
0.22
0.08
0.19
0.04
0.13
0.04
0.12
0.04
-0.08
0.04
-0.04
0.04
0.01
0.06
-0.09
0.11
143
p-value
p<0.001
p<0.001
p<0.001
p<0.001
p<0.001
p<0.001
p<0.001
p<0.01
p<0.001
p<0.001
p<0.01
p<0.01
NS
NS
NS
Texas Tech University, Daniel Dawson, August 2016
CHAPTER VI
SUMMARY AND CONCLUSIONS
In this dissertation, I investigated how environmental and anthropogenic factors combine
to influence larval ecology and population dynamics of epidemiologically relevant mosquitoes.
Specifically, this research has focused on topics with potential applications in the field of
mosquito ecology and control. In Chapter 2, experimental and modeling results showed that age
of exposure can be an important determinant of how other stressors, like temperature and
pesticides, affect life history characteristics, and can lead to population-level effects in the
yellow-fever mosquito, Aedes aegypti. In Chapter 3, a population model constructed for the
mosquito Culex quinquefasciatus in Tarrant County, TX, using surveillance data demonstrated
that although factors like temperature are important in predicting mosquito population dynamics
at a landscape scale, other factors such as treatment and habitat promote spatial heterogeneity
in populations at local scales. In Chapter 4, experiments demonstrated that water quality and
temperature but not predator presence impacted the effects of Bacillus thuringiensis israelensis
(Bti)-based larvicide exposure on life history characteristics of the mosquito Culex tarsalis.
Lastly, in Chapter 5 a spatially explicit population model of Cx. tarsalis was developed,
analyzed, and evaluated, and a risk-based mosquito-control treatment application was
demonstrated.
Overall, the research findings presented in this dissertation can be categorized into a
few main themes. First, the effects of pesticides to larvae combine with extrinsic and intrinsic
factors to influence the life history characteristics of both larvae and eventual adults. In Chapter
2, experimental results showed that temperature had no significant influence on the effects of
malathion on Ae. aegypti larval survival when larvae were treated as fourth instars. This
144
Texas Tech University, Daniel Dawson, August 2016
contrasted with the findings from a similar, previously published study on first instar (Muturi,
2013) that found larvae treated at higher temperatures had lower survival. However, wing sizes
of surviving individuals treated as fourth instar larvae decreased with concentration (indicating
lowered potential fecundity), the opposite of results found in the published experiment on first
instar larvae (i.e, higher fecundity with malathion concentration). These results may be due to
age differences in exposed individuals, as older larvae have larger energy reserves than
younger larvae with which to deal with toxic insults (Bouvier et al., 2002), but also less time to
replenish reserves. In Chapter 4, results showed that larval Cx. tarsalis were more sensitive to
Bti in wastewater than in water collected from playa wetlands, potentially from higher salinity,
sulfates, and organic contaminants in the wastewater. In addition, adults emerging from
wastewater had smaller wing sizes, indicating likely lower overall fitness. These results were
surprising, since degradation rates of Bti are expected to be higher in wastewater (Lacey, 2007),
and larvae reared in water with high nutrient content, like wastewater, have been found develop
into larger adults (Peck and Walton, 2005). On the other hand, Cx. tarsalis larvae may
potentially be sensitive to high levels of sulfates (Mian, 2006; Mian et al., 2009) and NH4+ (Peck
and Walton, 2005; Reisen et al., 1989), both of which were found to be significantly higher in
wastewater treatments. In contrast to the water quality experiment, no effect of predator
presence was found on either larval or adult life history characteristics, suggesting that chemical
and visual predation cues may be less important for Cx. tarsalis than direct (Walton et al., 1990)
and indirect consumption pressure (Bence and Murdoch, 1983; Blaustein and Karban, 1990) by
predators. Lastly, consistent with known patterns of Bti activity (Boisvert and Boisvert, 2000), Bti
effectiveness between the experiments differed because of temperature, with higher
temperatures increasing larval sensitivity. Together, these results illustrate that pesticide
impacts to larvae and the eventual adults that emerge depend upon a variety of factors. Of the
ones considered here, effects of differential temperature and water quality are likely the most
easily utilized in a mosquito control context. In addition, these results may justify conducting
145
Texas Tech University, Daniel Dawson, August 2016
larvicide bio-assays, preferably using locally captured individuals and local water, to guide
treatment practices in the landscape over the course of the season.
Second, this research demonstrated that influences on both larval and adult life history
characteristics relevant to mosquito control can have significant influences on population
dynamics in both hypothetical space and real world environments. In an example of the former,
experimental results in Chapter 2 extended to a matrix-based population model showed that a
single impact during the larvae phase can potentially influence adult population dynamics. In an
example of the latter, the sensitivity analysis of the model developed for Lubbock County, TX in
Chapter 5 demonstrated that several life history characteristics of adult and larval stages were
influential on adult population dynamics. And although adult survival was found to be relatively
more influential than larval survival to overall population levels, a single larvicide impact to larval
survival in the treatment scenario at wetlands selected as “high risk” produced a reduction in
high contact risk that was disproportionate to the reduction in overall mosquito population.
These results are consistent with previous findings that targeting larvae (Pawelek et al., 2014)
or adults (Elnaiem et al., 2008) with pesticides can significantly affect populations. However,
modeling results in Chapter 2 also found that mosquito control can potentially have unintuitive
results (Antonio et al., 2009; Muturi, 2013), as groups treated as first instar larvae had higher
projected abundances than control groups due to wing-length associated fecundity. Overall,
those results emphasized the need in mosquito control to understand both the underlying
conditions influencing mosquito populations, as well as how the target species reacts to
pesticide exposures.
Importantly, while mosquito control effectiveness over short temporal scales can
significantly affect population dynamics, their effects can be limited to local scales. In Chapter 2,
the model predicted that adult populations emerging from high concentration exposure groups
substantially differed from those emerging from control groups after multiple generations after
146
Texas Tech University, Daniel Dawson, August 2016
only the first cohort was exposed. In addition, the treatment scenario in Chapter 5 showed that a
wide-scale application of a short-lived (1 day,100% effectiveness) larvicide (on day 46), caused
a reduction in adult population that persisted through the end of model simulations (day 60).
However, the model in Chapter 3 demonstrated that although treatment events significantly
affected local population dynamics near traps, the general rareness of treatment events at the
landscape scale made their influence on landscape-level population dynamics rather small. In
addition, the treatment scenario in chapter 4 clearly showed the reduction of risk following
treatment was only found in the vicinity of treated wetlands, not over the entire county. In
contrast to model results, the effectiveness of treatment with pesticides on mosquito populations
in field situations has been shown to be short-lived (Pawelek et al., 2014; Teng et al., 2005),
with frequent retreatment suggested (Teng et al., 2005; Zequi et al., 2014). Together, these
results suggest that while impacts to survival by mosquito control can result in population and
risk reduction, those impacts are likely to be very local, and may be short-termed in nature. And
while they support the continued field use of short-lived mosquito control products (like Btibased larvicide), they further emphasize the need to prioritize where mosquito control resources
are allocated to achieve mosquito control goals.
Lastly, my research further demonstrates that mathematical models, particularly spatially
explicit population models, have potential to be useful tools in the field of mosquito control. In
Chapter 3, the deterministic statistical model was constructed using surveillance and treatment
data collected by mosquito control authorities as part of normal operating procedures. It was fit
using freely available software, and is capable of produce spatially explicit predictions by simply
plugging in relative parameter values and solving the model. This model is similar to other
another regression model of Cx. tarsalis abundance (Schurich et al., 2014), except the one
described here incorporates treatment data. In Chapter 5, the simulation model was constructed
by coupling two-modeling platforms, program R and NetLogo, to simultaneously model both
147
Texas Tech University, Daniel Dawson, August 2016
aquatic and adult life stages on a daily time step at a landscape scale. It also uses freely
available software, but due its inherent complexity and stochastic nature, requires running
multiple simulations to produce reliable estimations of populations in space, a process that can
take considerable computation time. This model is conceptually similar, but far less complex
than the model SkeeterBuster published for Ae. aegypti (Legros et al., 2011; Magori et al.,
2009). Both models described here produce predictions of mosquito population in space, and
both were demonstrated to reproduce field-collected mosquito abundance data. Due to its
relative ease and rapidity of use, the statistical model would be more likely to be utilized in a
management capacity. As an example of its use, population predictions made by the model for
coming weeks could help guide and assess treatment effects. However, this type of model is
dependent upon an adequate network of surveillance traps in the area of interest, and the
collection of good quality surveillance and treatment data. In addition, like the model in Schurich
and others (2014), its generalizability may be limited. Despite its complexity, the mechanistic
nature of the simulation model give it several advantages over the regression model. First, it has
the ability to predict populations in space regardless of available surveillance data, and
generalizable to a number of landscapes. Next, it has the capability to explicitly assess how
different treatment scenarios affect populations, and therefore risk. Lastly, the simulation model
could be potentially adapted to quantify disease risk, a highly useful application prior to or during
disease outbreaks. Regardless of the form, however, spatially-explicit models can provide
quantitative guidance in making on the ground management decisions. Thus, mathematical
models like the ones presented here have potential to be valuable tools to control mosquitoes
and mitigate mosquito-borne disease risks.
148
Texas Tech University, Daniel Dawson, August 2016
REFERENCES
Akoh, J.I., Aigbodion, F.I., Kumbak, D., 1992. Studies on the effect of larval diet, adult body
weight, size of blood-meal and age on the fecundity of Culex quinquefasciatus (Diptera:
Culicidae). Insect Sci. its Appl. 13, 177–181.
Alto, B.W., Malicoate, J., Elliott, S.M., Taylor, J., 2012a. Demographic consequences of
predators on prey: trait and density mediated effects on mosquito larvae in containers.
PLoS One 7, e45785.
Alto, B.W., Reiskin, M.H., Lounibos, L.P., 2012b. Size alters susceptibility of vectors to Dengue
virus infection and dissemination. Am. J. Trop. Med. Hyg. 29, 997–1003.
Anderson, D., 2008. Model based inference in the life sciences: a primer on evidence, 1st ed.
Springer-Verlag, New York.
Antonio, G.E., Sánchez, D., Williams, T., Marina, C.F., 2009. Paradoxical effects of sublethal
exposure to the naturally derived insecticide spinosad in the dengue vector mosquito,
Aedes aegypti. Pest Manag. Sci. 65, 323–326.
Armistead, J.S., Arias, J.R., Nishimura, N., Lounibos, L.P., 2008. Interspecific larval competition
between Aedes albopictus and Aedes japonicus (Diptera: Culicidae) in northern Virginia. J.
Med. Entomol. 45, 629–637.
Arrese, E., Soulages, J.L., 2010. Insect fat body: Energy, metabolism, and regulation. Annu.
Rev. Entomol. 55, 207–225.
Bailey, S.F., Eliason, D.A., Hoffmann, B.L., 1965. Flight and dispersal of the mosquito Culex
tarsalis Coquillett in the Sacramento Valley of California. Hilgardia 37, 1–113.
Barker, C.M., Bolling, B.G., Moore, C.G., Eisen, L., 2009. Relationship between distance from
149
Texas Tech University, Daniel Dawson, August 2016
major larval habitats and abundance of adult mosquitoes in semiarid plains landscapes in
Colorado. J. Med. Entomol. 46, 1290–8.
Beehler, J.W., Mulla, M.S., 1995. Effects of organic enrichment on temporal distribution and
abundance of culicine egg rafts. J. Am. Mosq. Control Assoc. 11, 167–171.
Bell, J.P., Tsezos, M., 1987. Removal of hazardous organic pollutants by biomass adsorption.
Water Sci. Technol. 19, 191–198.
Bence, J., Murdoch, W.W., 1983. Some indirect effects of Gambusia upon Culex tarsalis
survival in rice paddies. Bull. Soc. Vector Ecol. 8.
Biocontrol Network, 2006. Aquabac Technical Bulletin [WWW Document]. URL
http://www.biconet.com/biocontrol/infosheets/AquabacBulletin.pdf
Blackmore, M.S., Lord, C.C., 2000. The relationship between size and fecundity in Aedes
albopictus. J. Vector Ecol. 25, 212–217.
Blaustein, L., Karban, R., 1990. Indirect effects of the mosquitofish Gambusia affinis on the
mosquito Culex tarsalis. Limnol. Oceanogr. 35, 767–771.
Boisvert, M., Boisvert, J., 2000. Effects of Bacillus thuringiensis var. israelensis on target and
nontarget organisms: a review of laboratory and field experiments. Biocontrol Sci. Technol.
10, 517–561.
Bonds, J. a. S., 2012. Ultra-low-volume space sprays in mosquito control: a critical review. Med.
Vet. Entomol. 26, 121–130.
Bouvier, J.-C., Boivin, T., Beslay, D., Sauphanor, B., 2002. Age-dependent response to
insecticides and enzymatic variation in susceptible and resistant codling moth larvae. Arch.
Insect Biochem. Physiol. 51, 55–66.
150
Texas Tech University, Daniel Dawson, August 2016
Briegel, H., 1990. Metabolic relationship between female body size, reserves, and fecundity of
Aedes aegypti. J. Insect Physiol. 36, 165–172.
Briegel, H., Timmermann, S.E., 2001. Aedes albopictus (Diptera: Culicidae): physiological
aspects of development and reproduction. J. Med. Entomol. 38, 566–571.
Brown, H.E., Doyle, M.S., Cox, J., Eisen, R.J., Nasci, R.S., 2011. The effect of spatial and
temporal subsetting on Culex tarsalis abundance models--a design for sensible reduction
of vector surveillance. J Am Mosq Control Assoc 27, 120–128.
Burkett, D., Kelly, R., Porter, C.H., Wirtz, R.A., 2004. Commercial mosquito trap and gravid trap
oviposition media evaluation, Atlanta, Georgia. J. Am. Mosq. Control Assoc. 233–238.
Caputo, B., Ienco, A., Manica, M., Petrarca, V., Rosà, R., della Torre, A., 2015. New adhesive
traps to monitor urban mosquitoes with a case study to assess the efficacy of insecticide
control strategies in temperate areas. Parasit. Vectors 8:134, 1-12.
Carrington, L.B., Armijos, M.V., Lambrechts, L., Barker, C.M., Scott, T.W., 2013. Effects of
fluctuating daily temperatures at critical thermal extremes on Aedes aegypti life-history
traits. PLoS One 8, e58824.
Carrington, L.B., Seifert, S.N., Willits, N.H., Lambrechts, L., Scott, T.W., 2013. Large diurnal
temperature fluctuations negatively influence Aedes aegypti (Diptera: Culicidae) life history
traits. J. Med. Entomol. 50, 43–51.
Caswell, H., 2001. Matrix Population Models: Construction, Analysis, and Interpretation, 2nd ed.
SinauerAssociates, Inc., Sunderland, MA, USA.
Ciota, A.T., Matacchiero, A.C., Kilpatrick, a M., Kramer, L.D., 2014. The effect of temperature
on life history traits of Culex mosquitoes. J. Med. Entomol. 51, 55–62.
151
Texas Tech University, Daniel Dawson, August 2016
Clark, T.M., Flis, B.J., Remold, S.K., 2004. Differences in the effects of salinity on larval growth
and developmental programs of a freshwater and a euryhaline mosquito species (Insecta:
Diptera, Culicidae). J. Exp. Biol. 207, 2289–2295.
Clements, 1999. Mating, in: The Biology of Mosquitoes, Volume 2: Sensory Reception and
Behavior. CABI, Cambridge, UK, pp. 360–400.
Clements, A.N., 2012. Arboviruses-case studies of transmission, in: Cutts, R., Head, T.,
Chippendale, F. (Eds.), The Biology of Mosquitoes, Volume 3: Transmission of Viruses and
Interactions with Bacteria. Cambridge University Press, Cambridge, pp. 174–361.
Clements, A.N., 1999a. Behavior and aspects of the biology of larvae and pupae, in: The
Biology of Mosquitoes, Volume 2: Sensory Reception and Behavior. CABI, Cambridge, UK,
pp. 169–177.
Clements, A.N., 1999b. Egg Laying, in: The Biology of Mosquitoes, Volume 2: Sensory
Reception and Behavior. CABI, Cambridge, UK, pp. 552–636.
Clements, A.N., 1992a. Introduction, in: The Biology of Mosquitoes, Volume 1: Development,
Nutrition, and Reproduction. Chapman and Hall, Cambridge, UK, pp. xiv–xxii.
Clements, A.N., 1992b. Nutrition and fertility of anautogenous mosquitoes, in: The Biology of
Mosquitoes, Volume 1: Development, Nutrition, and Reproduction. Chapman and Hall,
Cambridge, UK, pp. 409–423.
Clements, A.N., 1992c. Growth and Development, in: The Biology of Mosquitoes, Volume 1:
Development, Nutrition, and Reproduction. Chapman and Hall, Cambridge, UK, pp. 150–
170.
Clements, A.N., 1992. Osmostic and ionic regulation, in: The Biology of Mosquitoes, Volume 1:
152
Texas Tech University, Daniel Dawson, August 2016
Development, Nutrition, and Reproduction. Chapman and Hall, Cambridge, UK, pp. 124–
143.
Clements, A.N., 1992d. Oogenesis, in: The Biology of Mosquitoes, Volume 1: Development,
Nutrition, and Reproduction. Chapman and Hall, Cambridge, UK, pp. 340–353.
Collins, S.D., Heintzman, L.J., Starr, S.M., Wright, C.K., Henebry, G.M., McIntyre, N.E., 2014.
Hydrological dynamics of temporary wetlands in the southern Great Plains as a function of
surrounding land use. J. Arid Environ. 109, 6–14.
Connelly, C.R., Carlson, D.B., 2009. Florida mosquito control: the state of the mission as
defined by mosquito controllers, regulators, and environmental managers. University of
Florida, Institute of Food and Agricultural Sciences, Florida Medical Entomology
Laboratory, Vero Beach, FL.
Cosner, C., Beier, J.C., Cantrell, R.S., Impoinvil, D., Kapitanski, L., Potts, M.D., Troyo, A., Ruan,
S., 2009. The effects of human movement on the persistence of vector-borne diseases. J.
Theor. Biol. 258, 550–560.
Darsie, R.F., Ward, R.A.W., 2005. Identification and geographical distribution of the mosquitoes
of North America, North of Mexico. University Press of Florida, Gainesville, FL.
Davies, T.G.E., Field, L.M., Usherwood, P.N.R., Williamson, M.S., 2007. DDT, pyrethrins,
pyrethroids and insect sodium channels. IUBMB Life 59, 151–162.
Dawson, D.E., Vanlandeghem, M.M., Asquith, W.H., Patino, R., 2015. Long-term trends in
reservoir water quality and quantity in two major river basins of the southern Great Plains.
Lake Reserv. Manag. 31, 254–279.
Dean, J.R., 2013. Storage of samples for analysis, in: Environmental Trace Analysis:
153
Texas Tech University, Daniel Dawson, August 2016
Techniques and Applications. Wiley and Sons, Inc., pp. 45–50.
DiMenna, M. a, Bueno, R., Parmenter, R.R., Norris, D.E., Sheyka, J.M., Molina, J.L., LaBeau,
E.M., Hatton, E.S., Glass, G.E., 2006. Comparison of mosquito trapping method efficacy
for West Nile virus surveillance in New Mexico. J. Am. Mosq. Control Assoc. 22, 246–53.
Diuk-Wasser, M. a, Brown, H.E., Andreadis, T.G., Fish, D., 2006. Modeling the spatial
distribution of mosquito vectors for West Nile virus in Connecticut, USA. Vector Borne
Zoonotic Dis. 6, 283–295.
Dodson, B.L., Kramer, L.D., Rasgon, J.L., 2012. Effects of larval rearing temperature on
immature development and West Nile virus vector competence of Culex tarsalis. Parasit.
Vectors 5, 199.
Dodson, B.L., Kramer, L.D., Rasgon, J.L., 2011. Larval nutritional stress does not affect vector
competence for West Nile virus (WNV) in Culex tarsalis. Vector-Borne Zoonotic Dis. 11,
1493–1497.
Dow, R.P., Reeves, W.C., Bellamy, R.E., 1965. Dispersal of female Culex tarsalis into a
larvicided area. Am. J. Trop. Med. Hyg. 14, 656–670.
Duchet, C., Caquet, T., Franquet, E., Lagneau, C., Lagadic, L., 2010. Influence of environmental
factors on the response of a natural population of Daphnia magna (Crustacea: Cladocera)
to spinosad and Bacillus thuringiensis israelensis in Mediterranean coastal wetlands.
Environ. Pollut. 158, 1825–1833.
Eisen, R.J., Eisen, L., 2008. Spatial modeling of human risk of exposure to vector-borne
pathogens based on epidemiological versus arthropod vector data. J. Med. Entomol. 45,
181–192.
154
Texas Tech University, Daniel Dawson, August 2016
Ellis, A.M., Garcia, A.J., Focks, D. a., Morrison, A.C., Scott, T.W., 2011. Parameterization and
sensitivity analysis of a complex simulation model for mosquito population dynamics,
dengue transmission, and their control. Am. J. Trop. Med. Hyg. 85, 257–264.
Elnaiem, D.-E. a, Kelley, K., Wright, S., Laffey, R., Yoshimura, G., Reed, M., Goodman, G.,
Thiemann, T., Reimer, L., Reisen, W.K., Brown, D., 2008. Impact of aerial spraying of
pyrethrin insecticide on Culex pipiens and Culex tarsalis (Diptera: Culicidae) abundance
and West Nile virus infection rates in an urban/suburban area of Sacramento County,
California. J. Med. Entomol. 45, 751–757.
Erickson, R. a., Presley, S.M., Allen, L.J.S., Long, K.R., Cox, S.B., 2010a. A dengue model with
a dynamic Aedes albopictus vector population. Ecol. Modell. 221, 2899–2908.
Erickson, R. a., Presley, S.M., Allen, L.J.S., Long, K.R., Cox, S.B., 2010b. A stage-structured,
Aedes albopictus population model. Ecol. Modell. 221, 1273–1282.
Fang, J., 2010. A world without mosquitoes. Nature 466, 432–434.
Ferrer, I., Thurman, M.E., 2008. EPA Method 1694 : Agilent ’ s 6410A LC/MS/MS Solution for
Pharmaceuticals and Personal Care Products in Water , Soil , Sediment , and Biosolids by
HPLC/MS/MS, Agilent Technologies.
Fry-O’Brien, L.L., Mulla, M.S., 1996. Effect of tadpole shrimp, Triops longicaudatus, (Notostraca:
Triopsidae), on the efficacy of the microbial control agent Bacillus thuringiensis var.
israelensis in experimental microcosms. J. Am. Mosq. Control Assoc. 12, 33–38.
Ganser, C., Wisely, S.M., 2013. Patterns of spatio-temporal distribution, abundance, and
diversity in a mosquito community from the eastern Smoky Hills of Kansas. J. Vector Ecol.
38, 229–236.
155
Texas Tech University, Daniel Dawson, August 2016
Gelman, A., 2008. Scaling regression inputs by dividing by two standard deviations. Stat. Med.
27, 2865–2873.
Grimm, V., Railsbeck, S., 2005. Individual-based modeling and Ecology. Princeton University
Press, Princeton, NJ.
Hardy, J.L., Houk, E.J., Kramer, L.D., Reeves, W.C., 1983. Intrinsic factors affecting vector
competence of mosquitoes for arboviruses. Annu. Rev. Entomol. 28, 229–262.
Harmston, F.C., Shultz, G.R., Eads, R.B., Menzies, G.C., 1956. Encephalitis in the Irrigated
Plains of Texas. Public Health Rep. 71, 759–766.
Haukos, D. a., Smith, L.M., 1994. The importance of playa wetlands to biodiversity of the
Southern High Plains. Landsc. Urban Plan. 28, 83–98.
Hawley, W.A., 1985. The effect of larval density on adult longevity of a mosquito, Aedes
sierrensis: epidemiological consequences. J. Anim. Ecol. 54, 955–964.
Hudson, a, McLintock, J., 1967. A chemical factor that stimulates oviposition by Culex tarsalis
Coquillet (Diptera, culicidae). Anim. Behav. 15, 336–341.
Jian, Y., Silvestri, S., Brown, J., Hickman, R., Marani, M., 2014. The temporal spectrum of adult
mosquito population fluctuations: conceptual and modeling implications. PLoS One 9,
e114301.
Jude, P.J., Tharmasegaram, T., Sivasubramaniyam, G., Senthilnanthanan, M., Kannathasan,
S., Raveendran, S., Ramasamy, R., Surendran, S.N., 2012. Salinity-tolerant larvae of
mosquito vectors in the tropical coast of Jaffna, Sri Lanka and the effect of salinity on the
toxicity of Bacillus thuringiensis to Aedes aegypti larvae. Parasit. Vectors 5, 269.
Karl, S., Halder, N., Kelso, J.K., Ritchie, S. a, Milne, G.J., 2014. A spatial simulation model for
156
Texas Tech University, Daniel Dawson, August 2016
dengue virus infection in urban areas. BMC Infect. Dis. 14: 447, 1-22
Knight, R.L., Walton, W.E., O’Meara, G.F., Reisen, W.K., Wass, R., 2003. Strategies for
effective mosquito control in constructed treatment wetlands. Ecol. Eng. 21, 211–232.
Korkmaz, S., Goksuluk, D., Zararsiz, G., 2014. MVN: An R package for assessing multivariate
normality. R J. 6, 151–162.
Kramer, V.L., Garcia, R., Colwell, a E., 1988. An evaluation of Gambusia affinis and Bacillus
thuringiensis var. israelensis as mosquito control agents in California wild rice fields. J. Am.
Mosq. Control Assoc. 4, 470–478.
Kroeger, I., Liess, M., Dziock, F., Duquesne, S., 2013. Sustainable control of mosquito larvae in
the field by the combined actions of the biological insecticide Bti and natural competitors. J.
Vector Ecol. 38, 82–89.
Kumar, R., Hwang, J.-S., 2006. Larvicidal efficiency of aquatic predators: a perspective for
mosquito biocontrol. Zool. Stud. 45, 845–853.
Lacey, L. a, 2007. Bacillus thuringiensis serovariety israelensis and Bacillus sphaericus for
mosquito control. J. Am. Mosq. Control Assoc. 23, 133–163.
Landau, K.I., Leeuwen, W.J.D. Van, 2012. Fine scale spatial urban land cover factors
associated with adult mosquito abundance and risk in Tucson, Arizona. J. Vector Ecol. 37,
407–418.
Langsrud, Ø., Matforsk, O., 2003. ANOVA for unbalanced data: Use Type II instead of Type III
sums of squares. Stat. Comput. 13, 163–167.
Legros, M., Magori, K., Morrison, A.C., Xu, C., Scott, T.W., Lloyd, A.L., Gould, F., 2011.
Evaluation of location-specific predictions by a detailed simulation model of Aedes aegypti
157
Texas Tech University, Daniel Dawson, August 2016
populations. PLoS One 6, e22701.
Leisnham, P.T., LaDeau, S.L., Juliano, S.A., 2014. Spatial and temporal habitat segregation of
mosquitoes in Urban Florida. PLoS One 9.
Lončarić, Ž., K. Hackenberger, B., 2013. Stage and age structured Aedes vexans and Culex
pipiens (Diptera: Culicidae) climate-dependent matrix population model. Theor. Popul. Biol.
83, 82–94.
Lotfy, N.M.., Merdan, A.I.., Abdel Rahman, H.A.., Abdel Razik, N.A., 1992. Factors governing
mosquito susceptibility to Bacillus sphaericus strains. Qatar Univ. Sci. J. 12.
Lothrop, H.D., Reisen, W.K., 2001. Landscape affects the host-seeking patterns of Culex
tarsalis (Diptera: Culicidae) in the Coachella Valley of California. J. Med. Entomol. 38, 325–
332.
Maciel-De-Freitas, R., Codeço, C.T., Lourenço-De-Oliveira, R., 2007. Body size-associated
survival and dispersal rates of Aedes aegypti in Rio de Janeiro. Med. Vet. Entomol. 21,
284–292.
Magori, K., Legros, M., Puente, M.E., Focks, D. a., Scott, T.W., Lloyd, A.L., Gould, F., 2009.
Skeeter Buster: a stochastic, spatially explicit modeling tool for studying Aedes aegypti
population replacement and population suppression strategies. PLoS Negl. Trop. Dis. 3,
e508.
Mahmood, F., Reisen, W.K., Chiles, R.E., Fang, Y., 2004. Western equine encephalomyelitis
virus infection affects the life table characteristics of Culex tarsalis (Diptera: Culicidae). J.
Med. Entomol. 41, 982–986.
McCall, P., Kelly, D., 2002. Learning and memory in disease vectors. Trends Parasitol. 18, 429–
158
Texas Tech University, Daniel Dawson, August 2016
433.
Mian, L.S., 2006. Effects of dairy wastewater on mosquitoes in southern California. J. vector
Ecol. 31, 305–10.
Mian, L.S., Lovett, J., Dhillon, M.S., 2009. Effect of effluent-treated water on mosquito
development in simulated ponds at the Prado Wetlands of southern California. J Am Mosq
Control Assoc 25, 347–355.
Milby, M., Meyer, R., 1986. The influence of constant versus fluctuating water temperatures on
the preimaginal development of Culex tarsalis. J. Am. Mosq. Control Assoc. 2, 7–10.
Mishra, K., Kumar Raj, D., Hazra, R.K., Dash, a P., 2005. A simple, artificial-membrane feeding
method for the radio-isotope labelling of Aedes aegypti polypeptides in vivo. Ann. Trop.
Med. Parasitol. 99, 803–806.
Mogi, M., 2007. Insects and other invertebrate predators. J. Am. Mosq. Control Assoc. 23, 93–
109.
Mottram, P., Kay, B.H., Fanning, I.D., Lee, B., 1994. Development and survival of Culex sitiens
Wiedemann ( Diptera : Culicidae ) in relation to temperature and salinity. J. Aust. Entomol.
Soc. 33, 81–85.
Murty, U.S., Sai, K.S.K., Satya Kumar, D.V.R., Sriram, K., Rao, K.M., Krishna, D., Murty, B.S.N.,
2002. Relative abundance of Culex quinquefasciatus (diptera: culicidae) with reference to
infection and infectivity rate from the rural and urban areas of east and west godavari
districts of Andhra Pradesh, India. Southeast Asian J. Trop. Med. Public Health 33, 702–
710.
Muturi, E.J., 2013. Larval rearing temperature influences the effect of malathion on Aedes
159
Texas Tech University, Daniel Dawson, August 2016
aegypti life history traits and immune responses. Chemosphere 92, 1111–1116.
Muturi, E.J., Alto, B.W., 2011. Larval environmental temperature and insecticide exposure alter
Aedes aegypti competence for arboviruses. Vector Borne Zoonotic Dis. 11, 1157–63.
Muturi, E.J., Costanzo, K., Kesavaraju, B., Alto, B.W., 2011a. Can pesticides and larval
competition alter susceptibility of Aedes mosquitoes (Diptera: Culicidae) to arbovirus
infection? J. Med. Entomol. 48, 429–436.
Muturi, E.J., Kim, C.-H., Alto, B.W., Berenbaum, M.R., Schuler, M. a., 2011b. Larval
environmental stress alters Aedes aegypti competence for Sindbis virus. Trop. Med. Int.
Heal. 16, 955–964.
Muturi, E.J., Lampman, R., Costanzo, K., Alto, B.W., 2011c. Effect of temperature and
insecticide stress on life-history traits of Culex restuans and Aedes albopictus (Diptera:
Culicidae). J. Med. Entomol. 48, 243–250.
Nareshkumar, A., Murugan, K., Baruah, I., Madhiyazhagan, P., Nataraj, T., 2012. Larvicidal
potentiality, longevity and fecundity inhibitory activities of Bacillus sphaericus (Bs G3-IV) on
vector mosquitoes, Aedes aegypti and Culex quinquefasciatus. J. Entomol. Acarol. Res.
44, 15.
Nasci, R.S., 1986. The size of emerging and host-seeking Aedes aegypti and the relation of size
to blood-feeding success in the field. J. Am. Mosq. Control Assoc. 2, 61–62.
Naylor, C., Malby, L., Callow, P., 1989. Scope for growth in Gammarus pulex, a freshwater
detritivore. Hydrobiologia 188, 517–523.
NCEI, 2016. Climate Data Online [WWW Document]. NOAA Natl. Centers Environ. Inf. URL
https://www.ncdc.noaa.gov/cdo-web/ (accessed 1.1.15).
160
Texas Tech University, Daniel Dawson, August 2016
Nelson, R., Milby, M., Reeves, W.C., Fine, P.E., 1978. Estimates of survival, population size,
and emergence of Culex tarsalis at an isolated site. Ann. Entomol. Soc. Am.
Nelson, R.L., Milby, M.M., 1980. Dispersal and survival of field and laboratory strains of Culex
tarsalis (Diptera: Culicidae). J. Mol. Evol. 17, 146–150.
Oléron Evans, T.P., Bishop, S.R., 2014. A spatial model with pulsed releases to compare
strategies for the sterile insect technique applied to the mosquito Aedes aegypti. Math.
Biosci. 254, 6–27.
Osborn, F.R., Herrera, M.J., Gómez, C.J., Salazar, A., 2007. Comparison of two commercial
formulations of Bacillus thuringiensis var. israelensis for the control of Anopheles aquasalis
(Diptera: Culicidae) at three salt concentrations. Mem. Inst. Oswaldo Cruz 102, 69–72.
Otero, M., Schweigmann, N., Solari, H.G., 2008. A stochastic spatial dynamical model for Aedes
aegypti. Bull. Math. Biol. 70, 1297–1325.
Padmanabha, H., Bolker, B., Lord, C.C., Rubio, C., Lounibos, L.P., 2011. Food availability alters
the effects of larval temperature on Aedes aegypti growth. J. Med. Entomol. 48, 974–984.
Palmquist, K.R., Jepson, P.C., Jenkins, J.J., 2008. Impact of aquatic insect life stage and
emergence strategy on sensitivity to esfenvalerate exposure. Environ. Toxicol. Chem. 27,
1728–1734.
Patiño, R., Dawson, D., VanLandeghem, M.M., 2014. Retrospective analysis of associations
between water quality and toxic blooms of golden alga (Prymnesium parvum) in Texas
reservoirs: Implications for understanding dispersal mechanisms and impacts of climate
change. Harmful Algae 33, 1–11.
Patrick, M.L., Bradley, T.J., 2000. The physiology of salinity tolerance in larvae of two species of
161
Texas Tech University, Daniel Dawson, August 2016
Culex mosquitoes: the role of compatible solutes. J. Exp. Biol. 203, 821–830.
Pawelek, K.A., Niehaus, P., Salmeron, C., Hager, E.J., Hunt, G.J., 2014. Modeling dynamics of
Culex pipiens complex populations and assessing abatement strategies for West Nile virus.
PLoS One 9, e108452.
Peck, G.W., Walton, W.E., 2005. Effect of bacterial quality and density on growth and whole
body stoichiometry of Culex quinquefasciatus and Culex tarsalis (Diptera: Culicidae). J.
Med. Entomol. 43, 25–33.
Pennington, M.J., Prager, S.M., Walton, W.E., Trumble, J.T., 2016. Culex quinquefasciatus
larval microbiomes vary with instar and exposure to common wastewater contaminants.
Sci. Rep. 6, 1–9.
Pennington, M.J., Rivas, N.G., Prager, S.M., Walton, W.E., Trumble, J.T., 2015.
Pharmaceuticals and personal care products alter the holobiome and development of a
medically important mosquito. Environ. Pollut. 203, 199–207.
Pepin, K.M., Leach, C.B., Marques-Toledo, C., Laass, K.H., Paixao, K.S., Luis, A.D., Hayman,
D., Johnson, N.G., Buhnerkempe, M.G., Carver, S., Grear, D. a, Tsao, K., Eiras, A.E.,
Webb, C.T., 2015. Utility of mosquito surveillance data for spatial prioritization of vector
control against dengue viruses in three Brazilian cities. Parasit. Vectors 8, 98.
Polson, K., Brogdon, W.G., Rawlins, S.C., Chadee, D.D., 2012. Impact of environmental
temperatures on resistance to organophosphate insecticides in Aedes aegypti from
Trinidad. Rev Panam Salud Publica 32, 1–8.
Preisser, E.L., Bolnick, D.I., Bernard, M.F., 2005. Scared To death ? The effects of intimidation
and consumption in predator - prey interactions. Ecology 86, 501–509.
162
Texas Tech University, Daniel Dawson, August 2016
Quinn, G.P., Keough, M.J., 2002. Experimental Design and Data Analysis for Biologists.
Cambridge University Press, New York, NY.
R Core Team, 2015. R: A language and environment for statistical computing. R Foundation for
Statistical Computing, Vienna, Austria.
Ragnarsdottir, K. V., 2000. Environmental fate and toxicology of organophosphate pesticides. J.
Geol. Soc. London. 157, 859–876.
Reisen, W.K., Lothrop, H.D., 1995. Population ecology and dispersal of Culex tarsalis in the
Coachella valley of California. J. Med. Entomol. 32, 490–502.
Reisen, W.K., Lothrop, H.D., Hardy, J.L., 1995. Bionomics of Culex tarsalis (Diptera: Culicidae)
in relation to arbovirus transmission in southeastern California. J Med Entomol 32, 316–
327.
Reisen, W.K., Lothrop, H.D., Lothrop, B., 2003. Factors influencing the outcome of markrelease-recapture studies with Culex tarsalis (Diptera: Culicidae). J. Med. Entomol. 40,
820–829.
Reisen, W.K., Meyer, R.P., 1990. Attractiveness of selected oviposition substrates for gravid
Culex tarsalis and Culex quinquefasciatus in California. J. Am. Mosq. Control Assoc. 6,
244–250.
Reisen, W.K., Meyer, R.P., Shields, J., Arbolante, C., 1989. Population ecology of preimaginal
Culex tarsalis (Diptera: Culicidae) in Kern County, California. J. Med. Entomol. 26, 10–22.
Reisen, W.K., Milby, M.M., Bock, M.E., 1984. The effects of immature stress on selected events
in the life history of culex tarsalis. Mosq. News.
Reisen, W.K., Milby, M.M., Meyer, R., 1992. Population dynamics of adult Culex mosquitoes
163
Texas Tech University, Daniel Dawson, August 2016
(Diptera: Culicidae) along the Kern River,. J. Med. Entomol. 29, 531–543.
Reisen, W.K., Milby, M.M., Meyer, R.P., Pfuntner, A.R., Spoehel, J., Hazelrigg, J.E., Webb, J.P.,
1991. Mark-release-recapture studies with Culex mosquitoes ( Diptera : Culicidae ) in
southern California. J. Med. Entomol. 28, 357–371.
Rey, J., Walton, W., Wolfe, R., Connelly, C., O’Connell, S., Berg, J., Sakolsky-Hoopes, G.,
Laderman, A., 2012. North American wetlands and mosquito control. Int. J. Environ. Res.
Public Health 9, 4537–4605.
Reynaldi, S., Meiser, M., Liess, M., 2011. Effects of the pyrethroid fenvalerate on the alarm
response and on the vulnerability of the mosquito larva Culex pipiens molestus to the
predator Notonecta glauca. Aquat. Toxicol. 104, 56–60.
Richardson, G., Ward, C.R., Huddleston, E.W., 1972. Aquatic Macroinvertebrates of the Playa.
Logan, UT.
Roberts, D., 2012. Responses of three species of mosquito larvae to the presence of predatory
dragonfly and damselfly larvae. Entomol. Exp. Appl. 145, 23–29.
Rueda, L.M., Patel, K.J., Axtell, R.C., Stinner, R.E., 1990. Temperature-dependent development
and survival rates of Culex quinquefasciatus and Aedes aegypti (Diptera: Culicidae). J.
Med. Entomol. 27, 892–898.
Ruiz, L., Parikh, N., Heintzman, L.J., Collins, S.D., Starr, S.M., Wright, C.K., Henebry, G.M., van
Gestel, N., McIntyre, N.E., 2014. Dynamic connectivity of temporary wetlands in the
southern Great Plains. Landsc. Ecol. 29, 507–516.
Rydzanicz, K., Sobczyński, M., Guz-Regner, K., 2010. Comparison of the activity and
persistence of microbial insecticides based on Bacillus thuringiensis israelensis and
164
Texas Tech University, Daniel Dawson, August 2016
Bacillus sphaericus in organic polluted mosquito- breeding sites. Polish J. Environ. Stud.
19, 1317–1323.
Salice, C.J., Dawson, D.E., Weir, S.M., 2016. Challenges and paths forward in predicting risk of
vector-borne diseases: from mechanistic to rule-based modeling frameworks, in: Kendall,
R.J., Presley, S.M., Ramkumar, S. (Eds.), New Developments in Biological and Chemical
Terrorism Countermeasures. CRC Press, Boca Raton, pp. 22–49.
Schneider, C.A., Rasband, W.S., Eliceiri, K.W., 2012. NIH Image to ImageJ: 25 years of image
analysis. Nat Meth 9, 671–675.
Schurich, J., Kumar, S., Eisen, L., Moore, Chester, G., 2014. Modeling Culex tarsalis
abundance on the northern Colorado front range using a landscape-level approach. J. Am.
Mosq. Control Assoc. 20, 7–20. doi:10.2987/13-6373.1
Shaman, J., 2007. Amplification due to spatial clustering in an individual-based model of
mosquito–avian arbovirus transmission. Trans. R. Soc. Trop. Med. Hyg. 101, 469–483.
doi:10.1016/j.trstmh.2006.11.007
Sih, A., 1986. Antipredator responses and the perception of danger by mosquito larvae. Ecology
67, 434–441.
Sinegre, G., Gaven, B., Vigo, G., 1981. Contribution to standardization of laboratory tests on
experimental and commerical formulations of the serotype H-14 of Bacillus thuringiensis. IIInfluence of temperature, free chlorine, pH, and water depth of biological activity of primary
powder. Cah. ORSTOM, ser. Ent. Parasitol 19, 149–155.
Smith, D.L., Dushoff, J., McKenzie, F.E., 2004. The risk of a mosquito-borne infection a
heterogeneous environment. PLoS Biol. 2, e368.
165
Texas Tech University, Daniel Dawson, August 2016
Smith, D.L., Perkins, T. a., Reiner, R.C., Barker, C.M., Niu, T., Chaves, L.F., Ellis, a. M.,
George, D.B., Le Menach, a., Pulliam, J.R.C., Bisanzio, D., Buckee, C., Chiyaka, C.,
Cummings, D. a. T., Garcia, a. J., Gatton, M.L., Gething, P.W., Hartley, D.M., Johnston,
G., Klein, E.Y., Michael, E., Lloyd, a. L., Pigott, D.M., Reisen, W.K., Ruktanonchai, N.,
Singh, B.K., Stoller, J., Tatem, a. J., Kitron, U., Godfray, H.C.J., Cohen, J.M., Hay, S.I.,
Scott, T.W., 2014. Recasting the theory of mosquito-borne pathogen transmission
dynamics and control. Trans. R. Soc. Trop. Med. Hyg. 108, 185–197.
Starr, S., Heintzman, L., Mulligan, K., Barbato, L., McIntyre, N., 2016. Using remotely sensed
imagery to document how land use drives turbidity of playa waters in Texas. Remote Sens.
8, 192.
Stewart, R.J., Schaeffer, C.H., Miura, T., 1983. Sampling Culex tarsalis (Diptera: Culidae)
immatures on rice fields treated with combinations of mosquitofish and Bacillus
thuringiensis H-14 Toxin. J. Econ. Entomol.
Stuijfzand, S.C., Poort, L., Greve, G.D., Van Der Geest, H.G., Kraak, M.H.S., 2000. Variables
determining the impact of diazinon on aquatic insects: taxon, developmental stage, and
exposure time. Environ. Toxicol. Chem. 19, 582–587.
Styer, L.M., Meola, M. a, Kramer, L.D., 2007. West Nile virus infection decreases fecundity of
Culex tarsalis females. J. Med. Entomol. 44, 1074–1085.
Tachiiri, K., Klinkenberg, B., Mak, S., Kazmi, J., 2006. Predicting outbreaks: a spatial risk
assessment of West Nile virus in British Columbia. Int. J. Health Geogr. 5, 21.
doi:10.1186/1476-072X-5-21
Tarrant County, 2016. Tarrant County, Texas [WWW Document]. URL
http://access.tarrantcounty.com/en/county/about-tarrant.html (accessed 1.1.16).
166
Texas Tech University, Daniel Dawson, August 2016
Teng, H.-J., Lu, L.-C., Wu, Y.-L., Fang, J.-G., 2005. Evaluation of various control agents against
mosquito larvae in rice paddies in Taiwan. J. Vector Ecol. 30, 126–132.
Tetzlaff, D., Uhlenbrook, S., Molnar, P., 2005. Significance of spatial variability in precipitation
for process-oriented modelling: results from two nested catchments using radar and ground
station data. Hydrol. Earth Syst. Sci. 9, 29–41. doi:10.5194/hess-9-29-2005
Thiele, J.C., 2014. R Marries NetLogo: Introduction to the RNetLogo Package. J. Stat. Softw.
58, 1–41.
Thiele, J.C., Kurth, W., Grimm, V., 2012. RNetLogo: an R package for running and exploring
individual-based models implemented in NetLogo. Methods Ecol. Evol. 3, 480–483.
Thiemann, T., Nelms, B., Reisen, W.K., 2011. Bloodmeal host congregation and landscape
structure impact the estimation of female mosquito (Diptera: Culicidae) abundance using
dry ice-baited traps. J. Med. Entomol. 48, 513–517.
Timchalk, C., Kousba, A. a, Poet, T.S., 2007. An age-dependent physiologically based
pharmacokinetic/pharmacodynamic model for the organophosphorus insecticide
chlorpyrifos in the preweanling rat. Toxicol. Sci. 98, 348–65.
Timmermann, S.E., Briegel, H., 1999. Larval growth and biosynthesis of reserves in
mosquitoes. J. Insect Physiol. 45, 461–470..
USCB, 2016. Geography: TIGER Products [WWW Document]. United States Census Bur. URL
https://www.census.gov/geo/maps-data/data/tiger.html (accessed 2.15.15).
USGS, 2016. USGS Global Visualization Viewer [WWW Document]. Earth Resour. Obs. Sci.
Cent. URL http://glovis.usgs.gov/ (accessed 1.1.16).
van den Heuvel, M.J., 1963. The effect of rearing temperature on the wing length, thorax length,
167
Texas Tech University, Daniel Dawson, August 2016
leg length and ovariole number of the adult mosquito, Aedes aegypti (L.). Trans. R.
Entomol. Soc. London 115, 197–216.
van Uitregt, V.O., Hurst, T.P., Wilson, R.S., 2012. Reduced size and starvation resistance in
adult mosquitoes, Aedes notoscriptus, exposed to predation cues as larvae. J. Anim. Ecol.
81, 108–115.
Vanderford, B.J., Pearson, R.A., Rexing, D.J., Snyder, S.A., 2003. Analysis of endocrine
disruptors, pharmaceuticals, and personal care products in water using liquid
chromatography/tandem mass spectrometry. Anal. Chem. 75, 6265–6274.
Vectorbase, 2015. Liverpool [WWW Document]. URL
https://www.vectorbase.org/organisms/aedes-aegypti (accessed 1.1.16).
Venables, W.N., Ripley, B.D., 2002. Modern Applied Statistics with S, 4th ed. Springer, New
York.
Verdonschot, P.F.M., Besse-Lototskaya, A.A., 2014. Flight distance of mosquitoes (Culicidae):
A metadata analysis to support the management of barrier zones around rewetted and
newly constructed wetlands. Limnologica 45, 69–79.
Vlahović, M., Lazarević, J., Perić-Mataruga, V., Ilijin, L., Mrdaković, M., 2009. Plastic responses
of larval mass and alkaline phosphatase to cadmium in the gypsy moth larvae. Ecotoxicol.
Environ. Saf. 72, 1148–1155. doi:10.1016/j.ecoenv.2008.03.012
Walsh, R.K., Facchinelli, L., Ramsey, J.M., Bond, J.G., Gould, F., 2011. Assessing the impact of
density dependence in field populations of Aedes aegypti. J. vector Ecol. 36, 300–307.
Walton, W.E., Tietze, N.S., Mulla, M.S., 1990. Ecology of Culex tarsalis (Diptera: Culicidae):
factors influencing larval abundance in mesocosms in southern California. J Med Entomol
168
Texas Tech University, Daniel Dawson, August 2016
27, 57–67.
Walton, W.E., Van Dam, A.R., Popko, D. a, 2009. Ovipositional responses of two Culex
(Diptera: Culicidae) species to larvivorous fish. J. Med. Entomol. 46, 1338–1343.
Ward, C.R., 1964. Ecological changes in modified playa lakes with species emphasis on
mosquito production. Texas Technological College.
Ward, R.W., 1968. A sampling study of mosquito production in natural and modified playa lakes.
Cornelll University.
White, S.L., Ward, M.P., Budke, C.M., Cyr, T., Bueno, R., 2009. A comparison of gravid and
under-house CO2-baited CDC light traps for mosquito species of public health importance
in Houston, Texas. J. Med. Entomol. 46, 1494–7.
Wilensky, U., 1999. NetLogo. Center for Connected Learning and Computer-Based Modeling,
Northwestern University, Evanston, IL.
Wilensky, U., 1997. NetLogo simple birth rates model. Center for Connected Learning and
Computer-Based Modeling, Northwestern University, Evanston, IL.
Winters, A.M., Bolling, B.G., Beaty, B.J., Blair, C.D., Eisen, R.J., Meyer, A.M., Pape, W.J.,
Moore, C.G., Eisen, L., 2008. Combining mosquito vector and human disease data for
improved assessment of spatial West Nile virus disease risk. Am. J. Trop. Med. Hyg. 78,
654–665.
Wu, S., Zhang, L., Chen, J., 2012. Paracetamol in the environment and its degradation by
microorganisms. Appl. Microbiol. Biotechnol. 96, 875–884.
Xuan, R., Arisi, L., Wang, Q., Yates, S.R., Biswas, K.C., 2010. Hydrolysis and photolysis of
oxytetracycline in aqueous solution. J. Environ. Sci. Health. B. 45, 73–81.
169
Texas Tech University, Daniel Dawson, August 2016
Yoo, E.-H., 2014. Site-specific prediction of West Nile virus mosquito abundance in Greater
Toronto Area using generalized linear mixed models. Int. J. Geogr. Inf. Sci. 28, 296–313.
Zar, Jerrold, H., 1998. Biostatistical Analysis, 4th ed. Prentice Hall, Upper Saddle River, NJ.
Zequi, J.A.C., Dos Santos, F.P., Lopes, J., 2014. Control of Culex quinquefascitus and Cx.
saltanensis (Diptera: Culicidae) with Bacillus thuringiensis israelensis in wastewater
treatment lagoons. Rev. Colomb. Entomol. 40, 98–103.
Zhu, L., Fulcher, A., Hossain, T., Davidson, C., Beier, J.C., Zhu, L.I.N., Fulcher, a L.I., 2014.
Body size, blood feeding activity, and fecundity of Psorophora howardii, Psorophora ciliata,
and Psorophora ferox ( Diptera : Culicidae ). J. Med. Entomol. 51, 382–386.
Zuur, A.F., Ieno, E.N., Walker, N.J., Saveliev, A.A., Smith, G.M., 2009. Mixed effects models
and extensions in ecology with R, 1st ed. Springer, New York, NY.
170