BLOOD IN THE WATER: WETLANDS, WET SPELLS, AND DROUGHT REGULATE
MOSQUITO-BORNE DISEASE TRANSMISSION AT MULTIPLE SCALES
By
Nicholas Karam Skaff

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
Fisheries and Wildlife—Doctor of Philosophy
Ecology Evolutionary Biology and Behavior—Dual Major
2018

	

ABSTRACT
BLOOD IN THE WATER: WETLANDS, WET SPELLS, AND DROUGHT REGULATE
MOSQUITO-BORNE DISEASE TRANSMISSION AT MULTIPLE SCALES
By
Nicholas Karam Skaff
Outbreaks of mosquito-borne diseases pose a significant global public health challenge
and are difficult to predict over space and time. In the United States, two of the most important
mosquito-borne pathogens, West Nile virus (WNV) and Eastern equine encephalitis virus
(EEEV), have irregular outbreak patterns that are potentially linked to variation in climate and
land cover. In particular, spatial heterogeneity in WNV and EEEV transmission may be
associated with wetlands because they provide suitable habitat for mosquito vectors and avian
hosts. Additionally, drought and wet spells may affect multiple aspects of these disease systems,
including their arthropod vectors and pathogen hosts. However, wetlands are not homogeneous
and the effects of wetlands on WNV and EEEV transmission may vary due to vegetation,
connectivity, size, and inundation characteristics. Also, drought and wet spells can lead to
increases in disease transmission in some contexts and decreases in others. Finally, wetland
characteristics and climate conditions likely interact in complex ways resulting in both spatial
and temporal heterogeneity in disease transmission.
In this dissertation, I address the influence of wetlands and climate on WNV and EEEV
transmission at a range of temporal and spatial scales. In Chapter 1, I examined the effects of
drought conditions and several wetland characteristics on county-level human WNV incidence in
the northeastern and midwestern USA. I found that drought and wetland characteristics had
regionally variable effects: counties west of the Mississippi River with small wetlands and
counties undergoing drought with high levels of semi-permanent wetlands had 100% and 300%

	

higher human WNV incidence, respectively, and counties east of the Mississippi River with high
levels of connected wetlands had 50% higher human WNV incidence than counties without these
wetland characteristics. In Chapter 2, I investigated the influence of wetland characteristics and
drought and wet spells on EEEV vector infection rates and abundance in Connecticut, USA. I
found that evergreen and deciduous forested wetlands were associated with high EEEV vector
abundance and that emergent and shrub wetlands were associated with low vector abundance,
but that the effects of wetlands on EEEV vector infection rates were weak. Wet conditions
during the transmission season and during the fall/winter preceding the transmission season were
also favorable for EEEV transmission. In Chapter 3, I examined the influence of drought and wet
spells on WNV vector infection rates and abundance in Chicago, Illinois; Ft. Collins, Colorado;
and Coachella Valley, California. I detected significant regional differences in the influence of
drought and wet spells, likely due to variation in regional aridity and WNV ecology. I also
detected local-scale dissimilarities in the influence of drought and wet spells, which were likely
caused by surrounding cover of wetlands, impervious surfaces, croplands and forest. These
findings demonstrate that the effects of wetland and climate on WNV and EEEV transmission
are context dependent, and likely mediated by regional aridity, vector natural history, and
wetland characteristics. This underscores the importance of avoiding sweeping generalizations
about the influence of wetlands or climate on mosquito-borne disease transmission in the United
States.

	

	

This dissertation is dedicated to my mom, Anna Drallios, for her inspiring writing and political
activism while I myopically focused on my research; to my dad, Jon Skaff, for raising two kids
while writing a dissertation, which I find incomprehensible; to my sister, Sophia Skaff, for her
inspiring wanderlust; and to my girlfriend, Mary Flegler, for not being complacent in the search
for a meaningful job and life.

	

iv

	
ACKNOWLEDGEMENTS

A doctoral program can be a difficult time in a person’s life, but I had an intellectually
stimulating, fulfilling and fun experience. This was largely due to the people I met and the help I
received along the way. I’m hugely indebted to my advisor Dr. Kendra Cheruvelil for her
ceaseless enthusiasm and flexibility. Dr. Cheruvelil entrusted me with total freedom to pursue
my interests, which made research feel less like a job and more like a hobby. She was also a
great mentor and role model; my dissertation research and more importantly, my approach to
science, improved immensely with her guidance. I also greatly appreciate the support of
committee member Dr. Patricia Soranno, who was always willing to take the time to assist with
research or life and acted like an informal second advisor. Committee members Dr. Michael
Kaufman and Dr. Jennifer Owen also deserve immense thanks for indoctrinating me in the world
of disease ecology by inviting me to their preferred conferences and connecting me with their
friends and colleagues. Connections made at these conferences led to a research collaboration
and a postdoctoral position. I also want to thank committee member Dr. Gary Roloff for
improving my understanding of landscape ecology, and his expert pitching and unbelievable
batting hot-streak that led to a Lansing-area lower-division softball championship.
Several collaborators and mentors including Dr. Zofia Taranu, Dr. Theodore Andreadis,
Dr. Philip Armstrong, and Dr. Jean-François Lapierre, also greatly helped in developing the
ideas and approaches described in this dissertation. Additionally, I want to thank the long list of
labmates and coworkers I’ve had over the past five years, including Dr. Emi Fergus, Dr. Sarah
Collins, Dr. Emily Norton-Henry, Dr. Carolyn Wynn, Dr. Caren Scott, Dr. Jeff White, Angela
DePalma-Dow, Alex Rafalski, Katelyn King, Joe Stachelek, Autumn Poisson, and Nicole Smith

	

v

	
for teaching me about everything from ‘tidyverse’ data-wrangling to pasta-making. Finally, I
would not have succeeded without generous support from multiple funding sources, which I have
described in detail in each chapter’s acknowledgements section.

	

vi

	
PREFACE

The three chapters in this dissertation have been written as stand-alone manuscripts with
coauthors. Chapter 1 was written with Dr. Kendra Cheruvelil and published in Landscape
Ecology in February 2016; Chapter 2 was written with Dr. Kendra Cheruvelil, Dr. Theodore
Andreadis, and Dr. Philip Armstrong and published in Parasites and Vectors in October 2017;
Chapter 3 was written with Dr. Kendra Cheruvelil and Dr. Zofia Taranu and will be submitted
for peer review in 2018. Below are the associated citations:

Skaff, N. K., and K. S. Cheruvelil. 2016. Fine-scale wetland features mediate vector and climatedependent macroscale patterns in human West Nile virus incidence. Landscape Ecology
31:1615-1628.

Skaff, N. K., P. M. Armstrong, T. G. Andreadis, and K. S. Cheruvelil. 2017. Wetland
characteristics linked to broad-scale patterns in Culiseta melanura abundance and eastern equine
encephalitis virus infection. Parasites & vectors 10:1-16.

Skaff NK, Taranu Z, Cheruvelil KS. Regional climate and local land cover alter the effects of
drought and wet spells on West Nile virus transmission. In prep.

	

vii

	
TABLE OF CONTENTS

LIST OF TABLES ........................................................................................................................ ix
LIST OF FIGURES ...................................................................................................................... x
INTRODUCTION ........................................................................................................................ 1
CHAPTER 1: Fine-scale wetland features mediate vector and climate-dependent macroscale
patterns in human West Nile virus incidence ............................................................................... 5
CHAPTER 2: Wetland characteristics linked to broad-scale patterns in Culiseta melanura
abundance and Eastern equine encephalitis virus infection .......................................................... 6
CHAPTER 3: Regional climate and local land cover alter the effects of drought and wet spells on
West Nile virus transmission ........................................................................................................ 7
Abstract ............................................................................................................................. 7
Introduction ....................................................................................................................... 8
Methods............................................................................................................................. 12
Study locations, mosquito sampling, and data sources ............................................... 12
Land cover .................................................................................................................. 15
Analytical methods ..................................................................................................... 16
Results ............................................................................................................................... 19
Question 1: Regional differences in the effects of drought and wet spells ................. 19
Question 2: The importance of drought and wet spell timing .................................... 21
Question 3: Local land cover regulation of drought and wet spell effects ................. 25
Discussion ......................................................................................................................... 29
Regional climate and biota alter the effects of drought and wet spells ...................... 29
Local land cover regulates the effects of drought and wet spells ............................... 31
Conclusion ........................................................................................................................ 33
Acknowledgements ........................................................................................................... 35
CONCLUSION ............................................................................................................................. 36
APPENDIX ................................................................................................................................... 38
LITERATURE CITED ................................................................................................................. 40

	

viii

	
LIST OF TABLES

Table 1. Mean climate and land cover characteristics (within 500m of sampling sites) of
Chicago, Ft. Collins, and Coachella Valley .................................................................................. 34
Table 2. Summary of vector surveillance data included in analysis for Chicago, Ft. Collins, and
Coachella Valley. Values for MIR (minimum infection rate) and abundance represent monthly
mean across all sampling years ..................................................................................................... 34

	

ix

	
LIST OF FIGURES

Figure 1. Pearson correlations between monthly Palmer Drought Severity Index (PDSI; June
previous year through December focal year) and annual mean minimum infection rate (MIR; left
panels; a,c,e); and monthly PDSI and total vector abundance (right panels; b,d,f) at each
sampling site. Plots depict correlations in Chicago (top two panels with blue color scheme; a,b);
Ft. Collins (middle two panels with green color scheme; c,d); and Coachella Valley (bottom two
panels with red color scheme; e,f). Sampling sites with statistically significant correlations
(p<0.1) are depicted with colored (dark) points; sites with non-statistically significant
correlations are grey (light). Lowess curves are fitted through the points with statistically
significant correlations to reflect overall patterns in the effects of drought and wet spells across
months. Months during the focal year when mosquito sampling occurred are labeled with a
bolded and colored font. ............................................................................................................... 20
Figure 2. The proportion of sampling sites in Chicago (left, blue), Ft. Collins (middle, green),
and Coachella Valley (right, red) with statistically significant Pearson correlations (p<0.1),
statistically significant positive Pearson correlations, statistically significant negative Pearson
correlations, and both positive and negative significant correlations between Palmer Drought
Severity Index (PDSI) and minimum infection rate (MIR) (a), and PDSI and vector abundance
(b). ................................................................................................................................................. 23
Figure 3. Comparing the Bray-Curtis similarity indices for the sampling sites with significant
Pearson correlations (p<0.1) between Palmer Drought Severity Index (PDSI) and minimum
infection rate (MIR; left panels, a,c,e); and PDSI and vector abundance (right panels, b,d,f) for
each month in Chicago (top panels; a,b), Ft. Collins (middle panels; c,d), and Coachella Valley
(bottom panels; e,f). Higher similarity values (red) indicate that the months compared have
similar sampling sites with significant correlations; low similarity values (blue) indicate that the
months compared have different sampling sites with significant correlations. If both months
compared have no sampling sites with significant correlations, then no similarity value is
calculated. Months labeled with a negative sign represent the previous year (e.g. previous June),
rather than the focal year (e.g. current June) ................................................................................ 24
Figure 4. Land cover variables identified in classification trees that increase or decrease the
likelihood that a sampling site has a statistically significant positive or negative correlation
between Palmer Drought Severity Index (PDSI) and minimum infection rate (MIR; left panels;
a,c,e); and PDSI and vector abundance (right panels; b,d,f) in each month in Chicago (top panels;
a,b), Ft. Collins (middle panels; c,d), and Coachella Valley (bottom panels; e,f). The y-axis
represents the percent of sites that the land cover variable helps to correctly classify and can be
interpreted similarly to the proportion of variance explained by the variable (R2). Land cover
variables associated with significant negative correlations are represented by bars pointing down,
and those associated with significant positive correlations are represented by upwards pointing
bars. Variables that reduce the likelihood a site has a significant correlation (rather than increase
the likelihood) are represented by bars outlined with a dashed line. Lowess curves from Figure 1
are overlaid to show patterns of significant correlations across months. Months during the focal

	

x

	
year when mosquito sampling occurred are labeled with a bolded and colored font. Wetland
refers to the total area of all wetland types around a sampling site; wetland vegetation refers to
the area of either emergent, forest, or shrub wetland; wetland system refers to the area of
wetlands with either a palustrine, riverine, pond or farmed classification; wetland regime refers
to the area of wetlands with either a temporarily flooded, seasonally flooded, semi-permanently
flooded, or permanently flooded inundation pattern..................................................................... 27
Figure S1. Frequency distribution of Palmer Drought Severity Index values for each month for
Chicago (a), Ft. Collins (b), and Coachella Valley (c) ................................................................. 39

	

xi

	
INTRODUCTION
	
Climate and land cover changes in the coming century are expected to trigger profound
shifts in the distribution and severity of disease (Gage et al. 2008, Patz et al. 2008, Lafferty
2009). Mosquito-borne pathogens may be exceptionally susceptible to these changes because
mosquito vectors and hosts are highly responsive to environmental perturbations (Reiter 2001,
Reiter 2008). In particular, drought and wet spells can influence several aspects of mosquitoborne disease transmission, and are expected to change in frequency and severity in the future
(Strzepek et al. 2010, Dai 2013, Brown et al. 2014). Also, wetland land cover can facilitate
transmission because it supplies aquatic habitat for larval mosquitoes and a suitable environment
for multiple aspects of the host life cycle (Carpenter and La Casse 1965, Tiner Jr 1984,
Laderman et al. 1989). By further examining how these factors influence transmission, it is
possible to more effectively predict outbreaks of these pathogens.
Two important mosquito-borne pathogens in the United States, West Nile virus (WNV;
Flaviviridae; Flavivirus) and Eastern equine encephalitis virus (EEEV; Togaviridae,
Alphavirus), have transmission cycles that may be susceptible to the effects of drought and wet
spells and wetland land cover. WNV is geographically pervasive and the most prevalent
mosquito-borne pathogen in the US (Reimann et al. 2008). The primary WNV vector species
differ across US regions. Culex pipiens and Culex restuans proliferate in northern latitudes in
container habitats or other stagnant pools; Culex quinquefasciatus is primarily found in southern
states with mild winter temperatures; and Culex tarsalis is often found in wetlands west of the
Mississippi River (Pratt et al. 1963, Goddard et al. 2002, Kilpatrick et al. 2005). The primary
EEEV vector is Culiseta melanura, which overwinters in the larval stage and prefers cool, acidic
larval habitats (Weaver 2005, Armstrong and Andreadis 2010). Outbreaks of EEEV have been

	

1

	
expanding regionally in the US since 2000, although the reasons for the expansion are not clear
(Armstrong and Andreadis 2013). WNV and EEEV circulate among passerine birds and
ornithophilic vectors and may cause severe illness in humans that are incidentally infected (Scott
and Weaver 1989, Reimann et al. 2008, Reisen 2013). Transmission of both pathogens is usually
most intense during the summer months, but environmental conditions throughout the year can
influence transmission (Weaver 2005, Reisen 2013).
Previous studies suggest that WNV and EEEV transmission may be closely associated
with the occurrence of drought and wet spells. These climate conditions can alter either vector
abundance or vector infection rates, both of which affect the risks of outbreak in humans
(Bolling et al. 2009, Kilpatrick and Pape 2013). Drought can reduce vector abundance by
limiting available aquatic larval habitat (Reisen et al. 1992), or may increase vector abundance
by increasing organic content in larval habitats (Andreadis et al. 2004, Johnson and Sukhdeo
2013). In contrast, wet conditions can increase vector abundance by creating larval habitats
(Reisen et al. 1992) or reduce vector abundance by decreasing larval habitat stability and organic
content due to frequent flushing (Shaman et al. 2002, Andreadis et al. 2004, Koenraadt and
Harrington 2008). Drought can also increase WNV rates in vectors by restricting aquatic habitat
availability, thus facilitating contact between vectors and susceptible hosts in remaining refugia
(Shaman et al. 2005, Johnson and Sukhdeo 2013). In contrast, wet spells can increase infection
prevalence by supporting larger populations of susceptible passerine species, including the hatch
year birds (Rotenberry and Wiens 1989, Sæther et al. 2004) that are important for WNV
amplification (Hamer et al. 2008b). However, it is not clear whether the effects of drought and
wet spells differ across regions with different ecological characteristics, at local scales due to
land cover differences, or if the timing of climate events plays a role. This information will help

	

2

	
to determine when, where, and how the effects of climate conditions manifest in WNV and
EEEV transmission.
Wetlands may also play a key role in transmission, because they can provide habitat for
larval mosquitoes and attract susceptible hosts (Carpenter and La Casse 1965, Tiner Jr 1984,
Laderman et al. 1989). Certain wetland characteristics have been shown to be particularly
important. For example, wetland vegetation imposes important constraints on the suitability of
wetland habitats for larval development and on the locations where contact between vectors and
susceptible avian hosts occurs (Rey et al. 2012). Connectivity of wetlands to other aquatic
habitats may influence mosquito abundance via changes in mosquito predator dispersal, while
wetland size can influence the composition of host communities (Chase and Shulman 2009,
Johnson et al. 2012). Climate conditions, including drought and wet spells, can interact with
specific wetland characteristics to influence transmission. Wetlands with semi-permanent
inundation regimes can be highly favorable for vectors after drought periods (Chase and Knight
2003), and drought can eliminate temporary wetlands forcing vectors and avian hosts onto a few
remaining permanent refuges leading to vector-host contact (Shaman et al. 2005). Finally,
climate-driven increases in groundwater levels can inundate some wetlands resulting in larger
vector populations (Howard et al. 1988). Despite the apparent importance of wetland
characteristics, studies at broad spatial scales have typically failed to account for how differences
among wetland habitats can influence disease transmission (e.g. Gibbs et al. 2006, Ezenwa et al.
2007, Bowden et al. 2011, DeGroote and Sugumaran 2012). Therefore, research is needed to
determine whether specific wetland characteristics have an epidemiologically-relevant influence
on EEEV and WNV transmission over broad areas.

	

3

	
The following chapters examine how climate conditions and wetland characteristics
influence human WNV incidence, and WNV and EEEV vector abundance and infection rates.
Chapter 1 focuses on the effects of drought conditions and three fine-scale wetland
characteristics—wetland size, connectivity, and inundation regime—on broad-scale human
WNV incidence in the northeastern and midwestern United States. Chapter 2 examines the
effects of several wetland characteristics and drought and wet spells on EEEV transmission at
multiple spatial scales and temporal periods in Connecticut, USA. Finally, Chapter 3 addresses
the influence of drought and wet spells on WNV vector abundance and infection rates in
Chicago, Illinois; Ft. Collins, Colorado; and Coachella Valley, California. By determining how
climate, wetlands and interactions between the two influence EEEV and WNV transmission it is
possible to better predict the location and timing of mosquito-borne disease outbreaks. With this
information, future climate and land cover changes will have easier to anticipate effects on the
severity and geographical distribution of mosquito-borne pathogens.

	

4

	
CHAPTER 1: Fine-scale wetland features mediate vector and climate-dependent
macroscale patterns in human West Nile virus incidence

We take a spatially explicit GIS-based approach to determine whether fine-scale wetland
characteristics influence annual county-level human West Nile virus incidence across the
northeastern and midwestern US. Our results indicate that the effects of wetland characteristics
are regionally dependent and may be mediated by other factors such as vector species-specific
traits and climate. For a full text of this work go to:
https://www.researchgate.net/publication/295085677_Finescale_wetland_features_mediate_vector_and_climatedependent_macroscale_patterns_in_human_West_Nile_virus_incidence

	

5

	
CHAPTER 2: Wetland characteristics linked to broad-scale patterns in Culiseta melanura
abundance and Eastern equine encephalitis virus infection

We examined the effects of wetland characteristics and climate conditions on Cs. melanura
abundance and infection with Eastern equine encephalitis virus (EEEV) at multiple spatial scales
in Connecticut, USA. Our results suggest that wetland vegetation is an important determinant of
Cs. melanura abundance and that wet conditions during the summer and winter are linked to
increased risks of EEEV transmission. For a full text of this work go to:
https://parasitesandvectors.biomedcentral.com/articles/10.1186/s13071-017-2482-0

	

6

	
CHAPTER 3: Regional climate and local land cover alter the effects of drought and wet
spells on West Nile virus transmission

Abstract
Climate change is expected to cause regional shifts in the severity and frequency of
drought and wet spells, both of which influence components of mosquito-borne disease
transmission. However, responses to these climate conditions can be complex and may increase
disease transmission in some contexts, but reduce it in others. This is particularly true for West
Nile virus (WNV), the most prevalent mosquito-borne pathogen in the United States, because its
continental range spans a huge variety of ecosystems, resulting in regional variation in vectors,
hosts, and their environmental preferences. Therefore, we evaluated the influence of drought and
wet spells on WNV vector abundance and vector infection rates in Chicago, Illinois; Ft. Collins,
Colorado; and Coachella Valley, California. These locations span major gradients of climate,
land cover, and vector species, offering insight into WNV transmission across a large part of the
WNV range. We found that the influence of drought and wet spells was vastly different among
the three locations, and that local-scale land cover, particularly wetlands, mediated effects on
vector abundance and infection rates. These results point to the importance of multi-scale studies
of disease transmission and underscore the complex and sometimes contradictory responses of
pathogens to climate patterns.

	

7

	
Introduction
Changes in the severity and frequency of drought and wet spells are predicted in the
coming century. Severe drought is projected to increase substantially over a wide area of the
globe, including most regions of the United States (US; Karl 2009, Strzepek et al. 2010, Dai
2013). However, some US regions may experience fewer months of drought and wetter
conditions (Karl 2009, Strzepek et al. 2010). These changes are likely to have important,
regionally-specific implications for disease transmission, particularly transmission of mosquitoborne pathogens. Drought and wet spells have been shown to affect multiple components of
mosquito-borne disease systems, including vectors, pathogens, and pathogen hosts (Githeko et
al. 2000, Brown et al. 2014). However, responses to these climate conditions can be complex,
leading to increases in disease transmission in some contexts, but decreases in others
(Landesman et al. 2007, Brown et al. 2014, Skaff and Cheruvelil 2016). Therefore, it is essential
to understand the spatial and temporal factors that determine specific responses.
West Nile virus (WNV; Flaviviridae; Flavivirusis) one of the most significant vectorborne threats to human health in the US (Reimann et al. 2008). This pathogen circulates
primarily among passerine birds and ornithophilic Culex mosquitoes, with human infection
resulting from spillover transmission (Reimann et al. 2008, Reisen 2013). In the US, the primary
vector species differ across geographic gradients. Culex pipiens and Culex restuans proliferate in
northern latitudes in container habitats or small pools; Culex quinquefasciatus has similar habitat
preferences, but is not cold tolerant and is primarily found in southern states with mild winters;
and Culex tarsalis is often found in wetlands and agricultural settings west of the Mississippi
River (Pratt et al. 1963). Viral transmission can occur between the spring and fall, although it is
most intense in the mid to late summer (Reisen 2013).

	

8

	
Previous studies indicate that WNV transmission is closely associated with drought and
wet spells. Summer Palmer Drought Severity Index (PDSI) was found to be significantly
associated with annual human WNV cases in half the states in the continental US (Paull et al.
2017). Drought and wet spells proceeding the transmission season were also associated with
human WNV incidence, with drought linked to higher WNV incidence in the western USA and
wet conditions linked with higher WNV incidence in the eastern US (Landesman et al. 2007). In
contrast, above average rainfall had a positive effect on human WNV infection in the western US
and a negative effect in the eastern US (Hahn et al. 2015). The effects of climate appear to be
geographically pervasive and may act on multiple aspects of WNV transmission.
Drought and wet spells can influence WNV transmission via changes in either vector
abundance or vector infection prevalence, both of which increase the likelihood that WNVinfected mosquitoes blood-feed on humans (Bolling et al. 2009, Kilpatrick and Pape 2013).
Drought can reduce vector abundance by limiting available aquatic larval habitat (Reisen et al.
1992), or may increase vector abundance by increasing organic content in larval habitats
(Andreadis et al. 2004, Johnson and Sukhdeo 2013), or by eradicating mosquito predators (Chase
and Knight 2003). In contrast, wet conditions can increase vector abundance by creating larval
habitats (Reisen et al. 1992) or reduce vector abundance by decreasing larval habitat stability and
organic content due to frequent flushing (Shaman et al. 2002, Andreadis et al. 2004, Koenraadt
and Harrington 2008). Drought can increase WNV prevalence in vectors by reducing available
aquatic habitat, thus facilitating contact between vectors and susceptible hosts in remaining
refugia (Shaman et al. 2005, Johnson and Sukhdeo 2013). In contrast, wet spells can increase
WNV infection prevalence by supporting larger populations of susceptible passerine species,

	

9

	
including the hatch year birds (Rotenberry and Wiens 1989, Sæther et al. 2004) that are
important for WNV amplification (Hamer et al. 2008b).
Given the wide variation in the effects of drought and wet spells on WNV transmission, it
is important to understand the spatial context determining when and where these divergent
relationships are observed. These relationships may differ based on regional WNV ecology,
multi-scaled landscape cover and use, or the timing of the drought or wet spell. For instance,
drought may have different effects depending on the regional vector species (Landesman et al.
2007, Skaff and Cheruvelil 2016), wetlands and croplands can mediate relationships between
drought and mosquito abundance or infection (Shaman et al. 2005, Shaman et al. 2010, Skaff et
al. 2017), and the timing of drought and wet spells (i.e., prior to or during the transmission
season) can determine how vector populations respond (Chase and Knight 2003, Shaman et al.
2005, Skaff et al. 2017). However, there are few studies comparing the effects of these climate
conditions on vector abundance and WNV infection rates across regions and through time, nor
are there systematic analyses of how these effects may differ depending on local-scale land
cover.
We evaluated how drought and wet spells influence vector abundance and vector
minimum infection rate (MIR) at multiple scales using mosquito surveillance datasets from
Chicago, IL; Ft. Collins, CO; and Coachella Valley, CA. These locations have dissimilar
climates, land cover patterns, and vector species, which provided an opportunity to better
understand links between climate and WNV transmission in different parts of the WNV range.
We addressed three main questions: 1.) How do the effects of drought and wet spells on
abundance and MIR differ among these three locations?; 2.) How does the timing of drought and
wet spells mediate these relationships?; 3.) What local-scale land cover characteristics are

	

10

	
associated with differences in the effects of drought and wet spells on vector abundance and
MIR?

	

11

	
Methods
Study locations, mosquito sampling, and data sources
Chicago: This US city with a population of 2.7 million (2010 US census) and an area of
600 km2 borders the southwestern edge of Lake Michigan in the north central US. It has a
temperate climate, with four distinct seasons including cold snowy winters and warm
humid summers (Peel et al. 2007). The city receives an average of 99.3 cm of
precipitation per year (Midway Airport station; https://www.ncdc.noaa.gov/cdoweb/datatools/normals). The most important WNV vector is Cx. pipiens, which bloodfeeds on both avian and mammalian hosts (Hamer et al. 2008a). Another species, Cx.
restuans, has similar blood-feeding preferences and may also play a role in WNV
transmission in this region (Hamer et al. 2008a). We grouped these two species in our
analysis, because they are morphologically similar and were often not distinguished
during sample collection and processing.
Mosquito surveillance data were acquired from the Chicago Department of Public
Health (CDPH). Gravid traps baited with a grass infusion and BG-sentinel traps
(https://www.bg-sentinel.com) baited with CO2 and a BG-lure were typically operated
twice per week between late May and early October from 2004–2016. All collected Cx.
pipiens and Cx. restuans were pooled in groups of up to 50 and tested for WNV weekly
using RT-PCR. Further details on data collection and virus detection protocols have been
previously reported (Jones et al. 2011). Traps that were operated, but failed to capture
mosquitoes were not recorded in the available dataset and therefore were not included in
our analyses. Sampling sites were located in a range of habitat types and were generally
surrounded by large amounts of impervious cover (Table 1). The exact locations of traps

	

12

	
were not provided by CDPH, but were listed within city blocks (~0.02 km2). We used
ggmap package in R version 3.3.3 to geocode city blocks to approximate geographic
coordinates.

Fort Collins: This US city with a population of 145,000 (2010 US census) and an area of
148 km2 is located in the foothills of the Rocky Mountains in the west central US. It has a
semi-arid climate with four distinct seasons, and minimal precipitation (average of 40.9
cm per year) that typically occurs in the spring and summer
(https://www.ncdc.noaa.gov/cdo-web/datatools/normals; Peel et al. 2007). Cx. tarsalis
and Cx. pipiens transmit WNV in this region, although Cx. tarsalis is the primary
enzootic and epizootic vector (Bolling et al. 2009). Here we focus only on Cx. tarsalis.
Mosquito surveillance data were acquired from Vector Disease Control
International, a company specializing in Integrated Mosquito Management
(http://www.vdci.net). CDC Miniature Light Traps with a CO2 attractant and gravid traps
baited with an alfalfa and MicrobeLift (https://www.microbelift.com) infusion were
operated weekly from late May through mid-September 2006–2012. Collected Cx.
tarsalis were combined in pools of up to 50 and tested for WNV virus using RT-PCR.
Traps that were operated, but failed to capture mosquitoes were not recorded in the
available dataset and therefore were not included in our analyses. Sampling sites were
located in a range of wetland, impervious and cropland cover but generally had little
forest (Table 1).

	

13

	
Coachella Valley: The Coachella Valley is located in southern California on the northern
shore of the Salton Sea, an inland waterbody. The valley has an area of ~1,728 km2 and is
bounded by several mountain ranges. It consists of nine cities with a total population of
346,518 and several small, heavily-agricultural communities. The climate is extremely
arid with mild winters and extremely hot summers. Most precipitation occurs during the
winter, with a yearly average of 8.1 cm (Mecca Station, https://www.ncdc.noaa.gov/cdoweb/datatools/normals). The valley is dominated by resorts and residential areas in the
north, transitioning to irrigated cropland moving south toward the Salton Sea. Areas
around the Salton Sea consist of managed wetlands used for bird conservation and duck
hunting (http://ca.audubon.org/conservation/roadmap-protecting-bird-habitat-salton-sea;
Reisen et al. 2008). Cx. tarsalis and Cx. quinquefasciatus are the primary vectors in this
region, although Cx. tarsalis plays a key role in enzootic transmission (Reisen et al.
2008). We focus only on Cx. tarsalis.
Mosquito surveillance data were acquired from the Coachella Valley Mosquito
and Vector Control District. CDC light traps with a CO2 attractant, gravid traps baited
with an alfalfa infusion, and a small number of BG-sentinel traps were operated for one
night every week or two weeks from January through December, 2006-2016. Collected
Cx. tarsalis were combined in pools of up to 50 and tested for WNV virus using RTPCR. We did not include data from traps that failed to capture mosquitoes in order to
remain consistent with the surveillance datasets from Chicago and Ft. Collins. Unlike in
Chicago and Ft. Collins, sampling occurred year-round due to the mild conditions, and
Cx. tarsalis abundance was relatively high during all months. WNV was detected in
mosquitoes in all months, although it was generally highest in July and August. Sampling

	

14

	
sites were located in areas with a range of wetland and cropland cover, but generally had
little forest and impervious cover (Table 1).

Land Cover
We quantified the land cover characteristics within buffers of 100 m, 500 m, 1000 m, and
5000 m around each sampling site in Chicago, Ft. Collins, and Coachella Valley. The
characteristics included were: 1) the percent of impervious surfaces, 2) the percent of cultivated
land cover, 3) the area of irrigated cropland, 4) the percent of forest cover, 5) the area of all
wetlands, and 6) the area of wetlands with specific vegetative, system and inundation
classifications (details below). Impervious surface data were acquired from the 2006 National
Land Cover Database Percent Developed Imperviousness layer
(https://www.mrlc.gov/nlcd06_data.php) and the percent impervious surface was calculated by
determining the mean pixel value within each buffer. Cultivated land cover data were also
acquired from the 2006 National Land Cover Database (https://www.mrlc.gov/nlcd06_data.php)
and the percent of cultivated land was determined by calculating the percent of pixels within
each buffer with a value of 82 (NLCD code for cultivated land). In order to also assess the
influence of irrigated croplands, we included data from the Moderate Resolution Imaging
Spectroradiometer (MODIS) Irrigated Agriculture Dataset for the United States (MIrAD-US;
https://earlywarning.usgs.gov/USirrigation). We calculated the percent cover of irrigated
cropland by calculating the proportion of pixels within each buffer that contained irrigated
cropland. Forest cover was acquired from the United States Geological Survey (USGS) 2010
Global Tree Canopy Cover dataset

	

15

	
(https://landcover.usgs.gov/glc/TreeCoverDescriptionAndDownloads.php). The mean value of
each pixel representing the percent of tree canopy cover was calculated within each buffer.
Finally, we used the National Wetlands Inventory to determine the total wetland area and area of
wetlands with three vegetation types, four different system classifications and four different
water inundation regimes (https://www.fws.gov/wetlands/). Wetland vegetation types included:
1) emergent, 2) forest, and 3) scrub/shrub wetlands; wetland system classifications included: 1)
palustrine, 2) riverine, 3) pond, and 4) farmed wetlands; wetland inundation regimes included 1)
temporarily flooded, 2) seasonally flooded, 3) semi-permanently flooded and 4) permanently
flooded wetlands. We calculated the total area of polygons of each wetland type within buffer
areas. We used the raster, rgeos and sp packages in R version 3.3.3 to calculate land cover
around sampling sites.

Analytical methods
Our analyses consisted of four steps: 1) we calculated response variables (vector
abundance and MIR) annually for each individual sampling site; 2) we calculated Pearson
correlations between each response variable and monthly PDSI for each sampling site; 3) we
determined whether similar sampling sites had significant correlations with monthly PDSI across
months; and 4) we identified the land cover variables associated with significantly and nonsignificantly correlated sampling sites during each month. Step 2 allowed us to determine how
the effects of drought and wet spells on abundance and MIR differed between regions; steps 2
and 3 allowed us to determine how the timing of drought and wet spells mediated these regional
relationships; and steps 2 and 4 determined whether local-scale land cover characteristics were
associated with differences in the effects of drought and wet spells at sampling sites.

	

16

	

1) Quantifying vector abundance and MIR: We calculated mean annual MIR and mean
annual vector abundance per trap night for each sampling site. MIR was determined by
calculating the annual average weekly MIR ((# positive pools in week/total weekly
capture)*1000). Annual abundance per trap night was calculated by determining the total
number of mosquitoes captured at each site per year, then dividing by the total number of
trap nights. In order to ensure an adequate sample size to detect correlations between
PDSI and MIR/vector abundance, sampling sites were only included in our analyses if
they had at least four years with detections of WNV for MIR analysis, and at least four
years of sampling for vector abundance analyses.

2) Correlation Analysis: In order to assess the impacts of drought and wet spells at each
sampling site, we calculated Pearson correlations between monthly PDSI and a) annual
vector abundance per trap night and b) mean annual MIR. For each site, we calculated
mean PDSI for each month from the June of the previous year through the last month of
sampling in the focal year (year mosquito surveillance data were collected). This
information was derived using rasterized data from the Wild West Drought Tracker
(https://wrcc.dri.edu/wwdt/about.php) and was calculated by determining the mean pixel
value within a 500 m buffer of each site (Abatzoglou et al. 2017). Pearson correlations
were considered significant at p < 0.1 in order to include sites with obvious correlations
that had a small sample size. We were willing to accept an elevated probability of falsely
rejecting the null hypothesis (type I error), because findings hinged on overall monthly
and regional patterns across multiple sites rather than the correlations at individual sites.

	

17

	

3) Similarity of sampling sites: We determined whether similar sampling sites had
significant correlations between PDSI and a) vector abundance and b) mean annual MIR
in different months. Bray-Curtis similarity indices were generated using the vegan
package in R version 3.3.3.

4) Land cover drivers of correlations: Classification trees (rpart package in R version
3.3.3) were used to identify potential land cover drivers of significant correlations
between monthly PDSI and a) annual vector abundance per trap night, and b) mean
annual MIR. Land cover variables associated with a higher or lower likelihood of a
significant correlation at a site during each month (previous June – focal October) were
selected using the Gini index as the splitting criteria. We limited overfitting of trees by
setting the minimum bucket size to half the sample size of the minority binary outcome,
setting complexity parameter to 0.01, and by only growing trees for months where at least
15% of sites had a significant correlation. Due to low sample sizes, there was instability
in 10-fold cross validation outcomes, so these results were not used to limit tree size.
However, trees generally had fewer than three splits due to other efforts to limit
overfitting.
All figures were generated using the ggplot2 package in R version 3.3.3.

	

18

	

Results
Question 1: Regional differences in the effects of drought and wet spells
We detected differences in the influence of drought and wet spells among our three study
locations. In Chicago, PDSI was consistently and negatively correlated with both MIR and
vector abundance, while it changed through time and differed by response variable in the other
locations (Figure 1a,b; Figure 2a,b). In Fort Collins, negative correlations between MIR and
PDSI dominated from May to September of the focal transmission season (Figure 1c). During
other time periods, few sampling sites had significant correlations between MIR and PDSI
(Figure 1c). During June through November of the previous year, significant correlations
between PDSI and vector abundance were exclusively negative (Figure 1d). There was a
subsequent period of transition from December to April when there were few to no sites with
significant correlations, then from May to September, significant correlations between PDSI and
abundance were exclusively positive (Figure 1d).
In Coachella Valley, few sites had correlations between PDSI and MIR from June
through December of the previous year, but from January through December of the focal year,
positive correlations were common and were observed in approximately a quarter of the
sampling sites (Figure 1e; Figure 2a). Vector abundance was more tightly linked with PDSI than
was MIR. During an extended period from January to December of the focal year, vector
abundance was consistently and positively correlated with PDSI in Coachella Valley (Figure 1f;
Figure 2b). Overall, our results show that dry conditions increased MIR and abundance in
Chicago, dry and wet conditions increased MIR and abundance in Ft. Collins depending on the

	

19

	
timing and response variable, and wet conditions consistently increased MIR and abundance in
Coachella Valley.

Figure 1. Pearson correlations between monthly Palmer Drought Severity Index (PDSI; June
previous year through December focal year) and annual mean minimum infection rate (MIR; left
panels; a,c,e); and monthly PDSI and total vector abundance (right panels; b,d,f) at each
sampling site. Plots depict correlations in Chicago (top two panels with blue color scheme; a,b);
Ft. Collins (middle two panels with green color scheme; c,d); and Coachella Valley (bottom two
panels with red color scheme; e,f). Sampling sites with statistically significant correlations
(p<0.1) are depicted with colored (dark) points; sites with non-statistically significant
correlations are grey (light). Lowess curves are fitted through the points with statistically
significant correlations to reflect overall patterns in the effects of drought and wet spells across

	

20

	
months. Months during the focal year when mosquito sampling occurred are labeled with a
bolded and colored font.
Question 2: The importance of drought and wet spell timing
We found that the timing of drought and wet spells (i.e. the month of PDSI measurement)
had two distinct consequences for observed correlations with MIR and vector abundance. First,
timing determined whether significant correlations were observed and the direction of observed
correlations at a site. Henceforth, we refer to this as a ‘directional effect’. Second, timing
determined which sampling sites were significantly correlated with PDSI – we refer to these as
‘site-level effects’. Site-level effects were identified by determining the similarity of the
sampling sites with significant correlations in different months. In Chicago, the directional
effects of timing were minimal given that significant correlations were overwhelmingly negative
during all months for both MIR and vector abundance (Figure 1a,b; Figure 2a,b). However, sitelevel effects of timing were prevalent. We found that the sampling sites where MIR and vector
abundance were significantly correlated with PDSI differed depending on the month, even
though the directional effects were consistently negative through time (Figure 3a, b). For MIR,
the sampling sites negatively correlated with PDSI between June and October of the focal year
were very different from the sampling sites with significant negative correlations between the
previous August and the previous December (<11% similarity; Figure 3a). A similar but weaker
pattern was observed for vector abundance in Chicago (Figure 3b).
Both directional and site-level effects of the timing of drought and wet spells were
identified in Ft. Collins and Coachella Valley. In these locations, the directional effects tended to
change between the transmission season and the months preceding the transmission season
(Figure 1c-f). Also, our analysis of site-level effects indicates that sampling sites in Ft. Collins

	

21

	
with significant positive correlations between PDSI and vector abundance from May to
September of the focal year were very different from the negatively correlated sites during July
through December of the preceding year (0% similarity; Figure 3d). In Coachella Valley,
sampling sites with significant positive correlations between PDSI and vector abundance from
January through December were markedly different from the sites where significant correlations
were observed from June to December of the previous year (<25% similarity; Figure 3e,f). In Ft.
Collins and Coachella Valley, there were very few significant correlations between PDSI and
MIR prior to the transmission season, so site-level effects of timing could not be evaluated
(Figure 3c,e). These results indicate that the month when drought and wet spells occur heavily
influences the site-level effects in Chicago, and both the directional and site-level effects in Ft.
Collins and Coachella Valley.

	

22

	

Figure 2. The proportion of sampling sites in Chicago (left, blue), Ft. Collins (middle, green),
and Coachella Valley (right, red) with statistically significant Pearson correlations (p<0.1),
statistically significant positive Pearson correlations, statistically significant negative Pearson
correlations, and both positive and negative significant correlations between Palmer Drought
Severity Index (PDSI) and minimum infection rate(MIR) (a), and PDSI and vector abundance (b)

	

23

	

Figure 3. Comparing the Bray-Curtis similarity indices for the sampling sites with significant
Pearson correlations (p<0.1) between Palmer Drought Severity Index (PDSI) and minimum
infection rate (MIR; left panels, a,c,e) and PDSI and vector abundance (right panels, b,d,f) for
each month in Chicago (top panels; a,b), Ft. Collins (middle panels; c,d), and Coachella Valley
(bottom panels; e,f). Higher similarity values (red) indicate that the months compared have
similar sampling sites with significant correlations; low similarity values (blue) indicate that the
months compared have different sampling sites with significant correlations. If both months
compared have no sampling sites with significant correlations, than no similarity value is
calculated. Months labeled with a negative sign represent the previous year (e.g. previous June),
rather than the focal year (e.g. current June)
	

24

	
Question 3: Local land cover regulation of drought and wet spell effects
Although we observed distinct regional patterns, many sampling sites within each region
did not conform to the dominant paradigm. Therefore, we examined whether observed
differences among sampling sites in the same region were related to a site’s surrounding land
cover context.
We found that the local land cover variables associated with significant correlations
differed among locations and by response variable. In Chicago, forest cover and impervious
surfaces increased the likelihood that a site had a significant negative correlation with MIR
(Figure 4a), but wetlands and impervious surfaces were most important in determining the
likelihood of a significant correlation with vector abundance (Figure 4b). In Ft. Collins, wetland
area consistently affected the likelihood of a significant correlation for both MIR and vector
abundance (Figure 4c,d). In Coachella Valley, the total wetland area was important in
determining the likelihood of significant correlations for MIR, but wetlands, forests, and
croplands were all important for vector abundance. (Figure 4 e,f). In all, we found that a wide
variety of land cover variables were important and that multiple land cover variables could
concurrently mediate the effects drought and wet spells on the response variables.
We also found that local land cover variables could either increase or decrease the
likelihood of detecting a significant correlation at a sampling site. In Chicago, land cover
variables generally increased the likelihood of detecting a significant negative correlation (Figure
4a,b), but in Ft. Collins land cover could increase or decrease the likelihood (Figure 4c,d). For
example, wetlands increased the likelihood of detecting a significant negative correlation, but
decreased the likelihood of detecting a significant positive correlation at sampling sites in Ft.
Collins (Figure 4c,d). In Coachella Valley, wetlands increased the likelihood of a significant

	

25

	
positive correlation with MIR, but decreased the likelihood of a significant correlation with
vector abundance. Croplands also increased the likelihood of detecting a significant positive
correlation with vector abundance (Figure 4e,f). Overall, local land cover had a pervasive
influence on the importance of drought and wet spells that could either increase or decrease the
likelihood of a significant correlation at a sampling site.

	

26

	

Figure 4. Land cover variables identified in classification trees that increase or decrease the
likelihood that a sampling site has a statistically significant positive or negative correlation
between Palmer Drought Severity Index (PDSI) and minimum infection rate (MIR; left panels;
a,c,e) and PDSI and vector abundance (right panels; b,d,f) in each month in Chicago (top panels;
a,b), Ft. Collins (middle panels; c,d), and Coachella Valley (bottom panels; e,f). The y-axis
represents the percent of sites that the land cover variable helps to correctly classify and can be
interpreted similarly to the proportion of variance explained by the variable (R2). Land cover
variables associated with significant negative correlations are represented by bars pointing down,
and those associated with significant positive correlations are represented by upwards pointing

	

27

	
Figure 4 (cont’d)
bars. Variables that reduce the likelihood a site has a significant correlation (rather than increase
the likelihood) are represented by bars outlined with a dashed line. Lowess curves from Figure 1
are overlaid to show patterns of significant correlations across months. Months during the focal
year when mosquito sampling occurred are labeled with a bolded and colored font. Wetland
refers to the total area of all wetland types around a sampling site; wetland vegetation refers to
the area of either emergent, forest, or shrub wetland; wetland system refers to the area of
wetlands with either a palustrine, riverine, pond or farmed classification; wetland regime refers
to the area of wetlands with either a temporarily flooded, seasonally flooded, semi-permanently
flooded, or permanently flooded inundation pattern

	

28

	
Discussion
We found that WNV vector abundance and infection rates had complex and contradictory
responses to drought and wet spells – conditions that were favorable for disease transmission in
some circumstances were unfavorable or had no effect in others. Our findings help to identify the
specific context and spatial scales at which these differing outcomes occur. We found that the
influence of drought and wet spells varied significantly across regions, that the timing of drought
and wet spells had a strong influence on corresponding changes in vector abundance and
infection rates, and that differences in local-scale land cover altered these relationships.

Regional climate and biota alter the effects of drought and wet spells
We detected regional differences in relationships between PDSI and our response
variables that appear to be mediated by a combination of climate differences and vector habitat
preferences. In Chicago, which receives plentiful annual rainfall and has poorly drained soils, we
found that drier conditions during any part of the previous or focal transmission season were
associated with both higher vector abundance and MIR. This suggests that dry weather in this
area, at nearly any time of year, could lead to increased risks of transmission to humans.
This result is consistent with the current understanding of WNV transmission in Chicago
and other areas of the Northeast US. The primary vectors, Culex pipiens and Culex restuans,
generally prefer stagnant water sources with high organic content for larval development (Pratt et
al. 1963, Gardner et al. 2012). Dry conditions promote these characteristics because they reduce
habitat flushing and support the accumulation of organic matter (Munstermann and Craig Jr
1976, Gardner et al. 2012). Catch-basins, built in Chicago to ameliorate backyard and basement
flooding, may exacerbate the issue by creating plentiful habitat for local vectors during dry

	

29

	
periods (Ruiz et al. 2004). Rainfall events of just 3.5 cm were found to flush larva from catchbasins in Chicago (Gardner et al. 2012), so even moderate droughts that reduce the likelihood of
heavy precipitation may lead to substantial increases in vector populations. Associations between
drought and high vector infection rates are likely tied to drought-driven changes in vector-host
contact due to increases in vector abundance, abundance of key hosts, or greater overlap in the
spatial distribution of vectors and hosts (Kilpatrick et al. 2005, Shaman et al. 2005, Hamer et al.
2009, Johnson and Sukhdeo 2013).
Locations in drier climates responded differently; wet conditions were more likely to be
associated with high vector abundance and high MIR. The driest study location, Coachella
Valley, receives ~8.1 cm of precipitation per year. We found that wet conditions in this location
were associated with greater vector abundance and MIR, particularly around the transmission
season. This result indicates that in more arid climates, the availability of larval habitats may
constrain abundance and that increases in water availability may be an important driver of WNV
transmission. Previous research in Coachella Valley supports this idea; aquatic larval habitats
suitable for the primary vector, Culex tarsalis, are generally in short supply and dry, hot periods
can reduce water sources even in wetlands and agricultural areas (Reisen et al. 1995). Available
water in Coachella Valley, especially associated with wetlands in the nearby Salton Sea, may
also increase the attractiveness of the location to bird populations (Patten et al. 2003), which
could increase the pool of susceptible hosts. Therefore, wet spells may have important
implications for both WNV vector abundance and infection rates in other similar arid
environments.
Responses to drought and wet spells in Ft. Collins had characteristics reminiscent of the
other two locations. This is not surprising since Ft. Collins has an intermediate climate; it is

	

30

	
semi-arid and receives less than half the precipitation of Chicago but ~30 cm more precipitation
per year than Coachella Valley. Wet spells may fill ephemeral larval habitats in Ft. Collins, like
depressions and retention ponds (Schurich et al. 2014), leading to increases in vector abundance.
However, as in Chicago, dry conditions may reduce predator populations and enable vector
proliferation (Chase and Knight 2003) or restrict vectors and hosts to persistent aquatic habitats
(particularly riparian and agricultural zones), facilitating vector-host contact and viral
amplification (Shaman et al. 2005, Shaman et al. 2010).
Regionally divergent responses may also be a reflection of biological differences between
the primary vector species in each location, rather than purely a reflection of variation in
environmental aridity. In Ft. Collins and Coachella Valley, Cx. tarsalis utilizes clean, natural
larval habitats in vegetated margins around waterbodies (Zou et al. 2006) and thus may benefit
from water-induced flooding rather than the drought-related stagnation favored by Cx.
pipiens/Cx. restuans in Chicago. The overall influence of drought and wet spells is likely a
complex reflection of biology and climate that ultimately determines transmission dynamics and
human infection risks.

Local land cover regulates the effects of drought and wet spells
Although an influence of drought and wet spells was detected in all three locations, the
effects were not uniform across sampling sites. Only ~60% of sampling sites had a significant
correlation with PDSI (Figure 2). In any individual month, a third or fewer sampling sites had
statistically significant correlations. We also observed a large degree of dissimilarity in the sites
with significant correlations, depending on the timing of the drought and wet spell. These
differences were associated with local-scale land cover characteristics and the important land

	

31

	
cover variables reflected the regional WNV transmission ecology. For example, forest cover and
impervious surfaces increased the likelihood of a negative correlations between PDSI and MIR
in Chicago, likely because damp and cool urban green spaces provide a favorable setting for
vector-host contact. In Ft. Collins and Coachella Valley, wetlands decreased the likelihood of
significant correlations between PDSI and vector abundance potentially because, in more arid
regions, wetlands act as supplemental sources of water, thus limiting the sensitivity of vector
population to fluctuations in PDSI. This effect may be especially prominent in Coachella Valley,
because many of the wetland habitats are artificially inundated
(https://www.fws.gov/refuge/Sonny_Bono_Salton_Sea/wildlife_and_habitat/index.html) and
therefore less dependent on rainfall or ground water for inundation.
Land cover either increased or decreased the likelihood of significant correlations
between PDSI and the response variables, meaning that it coupled or decoupled relationships
between WNV transmission and climate conditions. This has important implications for how the
risks of WNV transmission are distributed across the landscape. Local land cover can create
pockets that have persistently high or low risks regardless of climate conditions, or with risks
that fluctuate significantly with the occurrence of drought and wet spells. Although we were
unable to acquire sufficiently detailed data on vector control efforts, these management activities
likely decouple relationships between climate and WNV transmission. Management efforts in the
three locations varied across sampling sites and through time depending on perceived risks of
transmission by local specialists, perhaps creating pockets of reduced risk on the landscape.
Future studies should attempt to account for these activities in order reduce unexplained spatial
variation in local responses to drought and wet spells.

	

32

	
Conclusion
The observed regional differences in WNV transmission underscore the complex and
sometimes contradictory responses of pathogens to climate patterns. Our findings warn against
over-generalizing disease responses to climate change, as the effects may differ depending on
context. For instance, based on our findings, an increased frequency of drought could increase
WNV transmission in Chicago, decrease it in Coachella Valley and have uncertain effects in Ft.
Collins. Regional biotic differences, such as vector species habitat preferences, and climate
differences, like environmental aridity, may help to explain divergent regional responses.
However, the relative importance of climate and biology are difficult to disentangle, because
regional biology is a reflection of evolutionary adaptation to climate conditions. Future studies
should assess the influence of drought and wetness among different vectors in the same location,
to help distinguish these effects.
These findings also highlight the multi-scale nature of disease risks. Although our results
indicate that drought and wet spells had consistent effects at broad scales, we did not detect these
effects at individual sampling sites in many land cover contexts. Although local-scale processes
are commonly obscured by regional aggregations of environmental heterogeneity (Huston 1999),
this idea has been under-emphasized in a disease context, particularly in studies of WNV
transmission. Future studies should consider the heterogeneous impacts of climate conditions at
local scales when developing models of disease risk.

	

33

	
Table 1. Mean climate and land cover characteristics (within 500 m of sampling sites) of
Chicago, Ft. Collins, and Coachella Valley.

Table 2. Summary of vector surveillance data included in analysis for Chicago, Ft. Collins, and
Coachella Valley. Values for MIR (minimum infection rate) and abundance represent monthly
mean across all sampling years.

	

34

	
Acknowledgements
We would like to thank the individuals and organizations that made this research possible by
providing mosquito surveillance data collected at great cost and effort, including Dr. Cortland
Lohff with the Chicago Department of Public Health, Dr. Broox Boze with Vector Disease
Control International, and Jennifer Henke with the Coachella Valley Mosquito Control District.
This research was developed under STAR Fellowship Assistance Agreement no. FP-91778301-0
awarded by the U.S. Environmental Protection Agency (EPA). It has not been formally reviewed
by the EPA and the views expressed in this publication are solely those of the authors.
Assistance was also provided by the GLEON Fellowship Program, the Ambrose Pattullo Fund
for Environmental Issues Graduate Fellowship in Literary Work, the Kellogg Biological Station
Graduate School Fellowship, the MSU ESPP program, and the NSF MacroSystems Biology
Program in the Emerging Frontiers Division of the Biological Sciences Directorate (EF1065786).

	

35

	
CONCLUSION
Unpredictability in the location and timing of mosquito-borne disease outbreaks is a
significant public health issue and a major challenge for researchers (Reiter 1988, Brown et al.
2008). My dissertation addresses this issue by identifying where and when outbreaks of WNV
and EEEV are likely to occur as a result of favorable climate conditions and land cover
characteristics. Given expectations of extreme changes in climate and land cover in the coming
century (Pielke 2005, Karl 2009), these results may have important implications for predicting
the future geographical distribution and timing of mosquito-borne disease outbreaks.
We established that land cover characteristics, particularly wetlands with specific
vegetative and inundation features, were likely important drivers of WNV and EEEV outbreaks.
Further, we found that these land cover types could either increase or decrease the likelihood of
pathogen transmission, and that their effects influence areas several kilometers away. These
findings address a significant gap in the literature, because they show that land cover features,
including wetlands with very specific characteristics, can have far-reaching implications on
landscape-scale disease transmission. This knowledge will allow health and mosquito control
specialists to more efficiently identify and target habitats that broadly increase disease risks to
humans. Further research should be conducted to identify the potential mechanisms linking
wetlands with vector infection rates and abundance.
I also determined that the occurrence of drought and wet spells may influence the timing
of WNV and EEEV outbreaks. Wet periods were associated with increased risks of EEEV
transmission and both drought and wet spells were associated with increased risks of WNV
transmission depending on geographical region and timing of the climate conditions. We also
found that the influence of these climate conditions could vary depending on local-scale wetland

	

36

	
characteristics and other land cover features. Therefore, our findings warn against overgeneralizing pathogen responses to climate change, as the effects may differ depending on the
regional and local context. Although previous work has determined that climate conditions play a
role in WNV and EEEV transmission, we expand on this by specifying how the effects of
drought and wet spells are modified by local land cover and by the timing of these climate
conditions. Future work is needed to determine the mechanisms by which drought and wet spells
affect vector and host populations, and more specifically how these conditions alter vector
abundance, and contact between vectors and hosts.
This research identifies specific environmental characteristics that favor WNV and EEEV
transmission and thus helps to reduce the uncertainty in when and where outbreaks of these
pathogens occur. As a result, interventions and resources aimed at reducing transmission, like
educational outreach and mosquito control, can be allocated more effectively at times and in
locations with the greatest need. Decision-making informed by this research should help to
lessen the serious human health burden of these pathogens.

	

	

37

	

APPENDIX

	

38

	

Figure S1. Frequency distribution of PDSI values for each month for a) Chicago, b) Ft. Collins,
and c) Coachella Valley.

	

39

	

LITERATURE CITED

	

40

	
LITERATURE CITED

Abatzoglou, J. T., D. J. McEvoy, and K. T. Redmond. 2017. The West Wide Drought Tracker:
Drought Monitoring at Fine Spatial Scales. Bulletin of the American Meteorological
Society.
Andreadis, T. G., J. F. Anderson, C. R. Vossbrinck, and A. J. Main. 2004. Epidemiology of West
Nile virus in Connecticut: a five-year analysis of mosquito data 1999–2003. VectorBorne & Zoonotic Diseases 4:360-378.
Armstrong, P. M., and T. G. Andreadis. 2010. Eastern equine encephalitis virus in mosquitoes
and their role as bridge vectors. Emerging infectious diseases 16:1869-1874.
Armstrong, P. M., and T. G. Andreadis. 2013. Eastern equine encephalitis virus–old enemy, new
threat. N Engl J Med 368:1670-1673.
Bolling, B. G., C. M. Barker, C. G. Moore, W. J. Pape, and L. Eisen. 2009. Seasonal patterns for
entomological measures of risk for exposure to Culex vectors and West Nile virus in
relation to human disease cases in northeastern Colorado. Journal of medical entomology
46:1519-1531.
Bowden, S. E., K. Magori, and J. M. Drake. 2011. Regional differences in the association
between land cover and West Nile virus disease incidence in humans in the United States.
The American journal of tropical medicine and hygiene 84:234-238.
Brown, H. E., J. E. Childs, M. A. Diuk-Wasser, and D. Fish. 2008. Ecologic factors associated
with West Nile virus transmission, northeastern United States. Emerging infectious
diseases 14:1539-1545.
Brown, L., J. Medlock, and V. Murray. 2014. Impact of drought on vector-borne diseases–how
does one manage the risk? Public health 128:29-37.
Carpenter, S. J., and W. J. La Casse. 1965. Mosquitoes of North America. Univ of California
Press, Berkeley.
Chase, J. M., and T. M. Knight. 2003. Drought-induced mosquito outbreaks in wetlands.
Ecology Letters 6:1017-1024.
Chase, J. M., and R. S. Shulman. 2009. Wetland isolation facilitates larval mosquito density
through the reduction of predators. Ecological entomology 34:741-747.
Dai, A. 2013. Increasing drought under global warming in observations and models. Nature
Climate Change 3:52-58.

	

41

	
DeGroote, J. P., and R. Sugumaran. 2012. National and regional associations between human
West Nile virus incidence and demographic, landscape, and land use conditions in the
coterminous United States. Vector-Borne and Zoonotic Diseases 12:657-665.
Ezenwa, V. O., L. E. Milheim, M. F. Coffey, M. S. Godsey, R. J. King, and S. C. Guptill. 2007.
Land cover variation and West Nile virus prevalence: patterns, processes, and
implications for disease control. Vector-Borne and Zoonotic Diseases 7:173-180.
Gage, K. L., T. R. Burkot, R. J. Eisen, and E. B. Hayes. 2008. Climate and vectorborne diseases.
American journal of preventive medicine 35:436-450.
Gardner, A. M., G. L. Hamer, A. M. Hines, C. M. Newman, E. D. Walker, and M. O. Ruiz.
2012. Weather variability affects abundance of larval Culex (Diptera: Culicidae) in storm
water catch basins in suburban Chicago. Journal of medical entomology 49:270-276.
Gibbs, S. E., M. C. Wimberly, M. Madden, J. Masour, M. J. Yabsley, and D. E. Stallknecht.
2006. Factors affecting the geographic distribution of West Nile virus in Georgia, USA:
2002-2004. Vector-Borne & Zoonotic Diseases 6:73-82.
Githeko, A. K., S. W. Lindsay, U. E. Confalonieri, and J. A. Patz. 2000. Climate change and
vector-borne diseases: a regional analysis. Bulletin of the World Health Organization
78:1136-1147.
Goddard, L. B., A. E. Roth, W. K. Reisen, and T. W. Scott. 2002. Vector competence of
California mosquitoes for West Nile virus. Emerging infectious diseases 8:1385-1391.
Hahn, M. B., A. J. Monaghan, M. H. Hayden, R. J. Eisen, M. J. Delorey, N. P. Lindsey, R. S.
Nasci, and M. Fischer. 2015. Meteorological conditions associated with increased
incidence of West Nile virus disease in the United States, 2004–2012. The American
journal of tropical medicine and hygiene 92:1013-1022.
Hamer, G. L., U. D. Kitron, J. D. Brawn, S. R. Loss, M. O. Ruiz, T. L. Goldberg, and E. D.
Walker. 2008a. Culex pipiens (Diptera: Culicidae): a bridge vector of West Nile virus to
humans. Journal of medical entomology 45:125-128.
Hamer, G. L., U. D. Kitron, T. L. Goldberg, J. D. Brawn, S. R. Loss, M. O. Ruiz, D. B. Hayes,
and E. D. Walker. 2009. Host selection by Culex pipiens mosquitoes and West Nile virus
amplification. American Journal of Tropical Medicine and Hygiene 80:268-278.
Hamer, G. L., E. D. Walker, J. D. Brawn, S. R. Loss, M. O. Ruiz, T. L. Goldberg, A. M.
Schotthoefer, W. M. Brown, E. Wheeler, and U. D. Kitron. 2008b. Rapid amplification of
West Nile virus: the role of hatch-year birds. Vector-Borne and Zoonotic Diseases 8:5768.
Howard, J., C. Morris, D. Emord, and M. Grayson. 1988. Epizootiology of Eastern equine
encephalitis virus in upstate New York, USA. VII. Virus surveillance 1978–85,

	

42

	
description of 1983 outbreak, and series conclusions. Journal of medical entomology
25:501-514.
Huston, M. A. 1999. Local processes and regional patterns: appropriate scales for understanding
variation in the diversity of plants and animals. Oikos 86:393-401.
Johnson, B., and M. Sukhdeo. 2013. Drought-induced amplification of local and regional West
Nile virus infection rates in New Jersey. Journal of medical entomology 50:195-204.
Johnson, B. J., K. Munafo, L. Shappell, N. Tsipoura, M. Robson, J. Ehrenfeld, and M. V.
Sukhdeo. 2012. The roles of mosquito and bird communities on the prevalence of West
Nile virus in urban wetland and residential habitats. Urban Ecosystems 15:513-531.
Jones, R. C., K. N. Weaver, S. Smith, C. Blanco, C. Flores, K. Gibbs, D. Markowski, and J.-P.
Mutebi. 2011. Use of the vector index and geographic information system to
prospectively inform West Nile virus interventions. Journal of the American Mosquito
Control Association 27:315-319.
Karl, T. R. 2009. Global climate change impacts in the United States. Cambridge University
Press.
Kilpatrick, A. M., L. D. Kramer, S. R. Campbell, E. O. Alleyne, A. P. Dobson, and P. Daszak.
2005. West Nile virus risk assessment and the bridge vector paradigm. Emerging
infectious diseases 11:425-429.
Kilpatrick, A. M., and W. J. Pape. 2013. Predicting human West Nile virus infections with
mosquito surveillance data. American journal of epidemiology 178:829-835.
Koenraadt, C. M., and L. Harrington. 2008. Flushing effect of rain on container-inhabiting
mosquitoes Aedes aegypti and Culex pipiens (Diptera: Culicidae). Journal of medical
entomology 45:28-35.
Laderman, A. D., M. Brody, and E. Pendleton. 1989. The ecology of Atlantic white cedar
wetlands: a community profile. Biological Report 85:1-125.
Lafferty, K. D. 2009. The ecology of climate change and infectious diseases. Ecology 90:888900.
Landesman, W. J., B. F. Allan, R. B. Langerhans, T. M. Knight, and J. M. Chase. 2007. Interannual associations between precipitation and human incidence of West Nile virus in the
United States. Vector-Borne and Zoonotic Diseases 7:337-343.
Munstermann, L. E., and G. B. Craig Jr. 1976. Culex mosquito populations in the catch basins of
northern St. Joseph County, Indiana. Proceedings of the Indiana Academy of Science
86:246-252.

	

43

	
Patten, M. A., G. McCaskie, and P. Unitt. 2003. Birds of the Salton Sea: status, biogeography,
and ecology. Univ of California Press.
Patz, J. A., S. H. Olson, C. K. Uejio, and H. K. Gibbs. 2008. Disease emergence from global
climate and land use change. Medical Clinics of North America 92:1473-1491.
Paull, S. H., D. E. Horton, M. Ashfaq, D. Rastogi, L. D. Kramer, N. S. Diffenbaugh, and A. M.
Kilpatrick. 2017. Drought and immunity determine the intensity of West Nile virus
epidemics and climate change impacts. Proc. R. Soc. B 284:1-10.
Peel, M. C., B. L. Finlayson, and T. A. McMahon. 2007. Updated world map of the KöppenGeiger climate classification. Hydrology and earth system sciences discussions 4:439473.
Pielke, R. A. 2005. Land use and climate change. Science 310:1625-1626.
Pratt, H. D., R. C. Barnes, K. S. Littig, and C. D. Center. 1963. Mosquitoes of public health
importance and their control. US Department of Health, Education, and Welfare, Public
Health Service, Communicable Disease Center, Atlanta.
Reimann, C. A., E. B. Hayes, C. DiGuiseppi, R. Hoffman, J. A. Lehman, N. P. Lindsey, G. L.
Campbell, and M. Fischer. 2008. Epidemiology of neuroinvasive arboviral disease in the
United States, 1999–2007. The American journal of tropical medicine and hygiene
79:974-979.
Reisen, W., H. Lothrop, and J. Hardy. 1995. Bionomics of Culex tarsalis (Diptera: Culicidae) in
relation to arbovirus transmission in southeastern California. Journal of medical
entomology 32:316-327.
Reisen, W. K. 2013. Ecology of West Nile Virus in North America. Viruses 5:2079-2105.
Reisen, W. K., H. D. Lothrop, S. S. Wheeler, M. Kennsington, A. Gutierrez, Y. Fang, S. Garcia,
and B. Lothrop. 2008. Persistent West Nile virus transmission and the apparent
displacement St. Louis encephalitis virus in southeastern California, 2003-2006. Journal
of medical entomology 45:494-508.
Reisen, W. K., M. M. Milby, and R. P. Meyer. 1992. Population dynamics of adult Culex
mosquitoes (diptera: culicidae) along the Kern River, Kern County, California, in 1990.
Journal of medical entomology 29:531-543.
Reiter, P. 1988. Weather, vector biology, and arboviral recrudescence. Pages 245-255 in T.
Monath, editor. The arboviruses: epidemiology and ecology. CRC Press, Boca Raton.
Reiter, P. 2001. Climate change and mosquito-borne disease. Environmental health perspectives
109:141-161.

	

44

	
Reiter, P. 2008. Climate change and mosquito-borne disease: knowing the horse before hitching
the cart. Rev Sci Tech 27:383-398.
Rey, J. R., W. E. Walton, R. J. Wolfe, C. R. Connelly, S. M. O'Connell, J. Berg, G. E. SakolskyHoopes, and A. D. Laderman. 2012. North American wetlands and mosquito control.
International journal of environmental research and public health 9:4537-4605.
Rotenberry, J. T., and J. A. Wiens. 1989. Reproductive biology of shrubsteppe passerine birds:
geographical and temporal variation in clutch size, brood size, and fledging success.
Condor 91:1-14.
Ruiz, M. O., C. Tedesco, T. J. McTighe, C. Austin, and U. Kitron. 2004. Environmental and
social determinants of human risk during a West Nile virus outbreak in the greater
Chicago area, 2002. International Journal of Health Geographics 3:1-11.
Sæther, B.-E., W. J. Sutherland, and S. Engen. 2004. Climate influences on avian population
dynamics. Advances in Ecological Research 35:185-209.
Schurich, J. A., S. Kumar, L. Eisen, and C. G. Moore. 2014. Modeling Culex tarsalis abundance
on the northern Colorado front range using a landscape-level approach. Journal of the
American Mosquito Control Association 30:7-20.
Scott, T. W., and S. C. Weaver. 1989. Eastern equine encephalomyelitis virus: epidemiology and
evolution of mosquito transmission. Advances in virus research 37:277-328.
Shaman, J., J. F. Day, and N. Komar. 2010. Hydrologic conditions describe West Nile virus risk
in Colorado. International journal of environmental research and public health 7:494-508.
Shaman, J., J. F. Day, and M. Stieglitz. 2005. Drought-induced amplification and epidemic
transmission of West Nile virus in southern Florida. Journal of medical entomology
42:134-141.
Shaman, J., M. Stieglitz, C. Stark, S. Le Blancq, and M. Cane. 2002. Using a dynamic hydrology
model to predict mosquito abundances in flood and swamp water. Emerging infectious
diseases 8:6-13.
Skaff, N. K., P. M. Armstrong, T. G. Andreadis, and K. S. Cheruvelil. 2017. Wetland
characteristics linked to broad-scale patterns in Culiseta melanura abundance and eastern
equine encephalitis virus infection. Parasites & vectors 10:1-16.
Skaff, N. K., and K. S. Cheruvelil. 2016. Fine-scale wetland features mediate vector and climatedependent macroscale patterns in human West Nile virus incidence. Landscape Ecology
31:1615-1628.
Strzepek, K., G. Yohe, J. Neumann, and B. Boehlert. 2010. Characterizing changes in drought
risk for the United States from climate change. Environmental Research Letters 5:1-9.

	

45

	

Tiner Jr, R. W. 1984. Wetlands of the United States: current status and recent trends. United
States Fish and Wildlife Service, Washington DC.
Weaver, S. 2005. Host range, amplification and arboviral disease emergence. Pages 33-44
Infectious Diseases from Nature: Mechanisms of Viral Emergence and Persistence.
Zou, L., S. N. Miller, and E. T. Schmidtmann. 2006. Mosquito larval habitat mapping using
remote sensing and GIS: implications of coalbed methane development and West Nile
virus. Journal of medical entomology 43:1034-1041.

	

46