Winter climate and its effects on Taylor's Checkerspot Butterflies (Euphydryas editha taylori) in the Puget Sound

Item

Identifier

Thesis_MES_2019_LagerquistW

Title

Winter climate and its effects on Taylor's Checkerspot Butterflies (Euphydryas editha taylori) in the Puget Sound

Date

June 2019

Creator

Lagerquist, Wendy

extracted text

WINTER CLIMATE AND ITS EFFECTS ON TAYLOR’S CHECKERSPOT
BUTTERFLIES (EUPHYDRYAS EDITHA TAYLORI) IN THE PUGET SOUND

by
Wendy Lagerquist

A Thesis Submitted
in partial fulfillment
of the requirements for the degree
Master of Environmental Studies
The Evergreen State College
June 2019

© 2019 by Wendy Lagerquist. All rights reserved.

This Thesis for the Master of Environmental Studies Degree
By
Wendy Lagerquist

has been approved for
The Evergreen State College by

________________________
John C. Withey, Ph. D
Member of the Faculty

_______________________
June 2019

ABSTRACT
Winter Climate and its Effects on Taylor’s Checkerspot Butterflies (Euphydryas editha
taylori) in the Puget Sound
The effects of three elements of winter climate on abundance of the Taylor’s Checkerspot
Butterfly in the Puget Sound were examined. Winter is the time the butterflies are
hibernating in larval state diapause. The climatic variables examined were winter
precipitation, humidity, and temperature. Butterflies on three sites of Joint Base Lewis
McChord (JBLM) were examined. In one site (Range 76) there were no significant
correlations of log-transformed estimates of peak abundance over time or with climate
variables. In the other two sites, the variable with the most explanatory power on logtransformed estimates of peak abundance was not climate variables but the year. In one
site (Scatter Creek South) abundance increased by 47%/year on average, while in the
other (Range 50) abundance increased by 75%/year on average. There was some
evidence of an association of climatic variables with estimates of peak abundance in
these two sites, but those associations had much less evidence than the increase over
time. These abundance increases could be an effect of the reintroduction of Taylor’s
Checkerspots to these sites, so the potential for climate to influence annual abundance
should not be discounted.

TABLE OF CONTENTS

vi

LIST OF FIGURES.................................................................................................

vii

LIST OF TABLES………………………………………………………………...

viii

ACKNOWLEDGEMENTS.....................................................................................
INTRODUCTION AND LITERATURE REVIEW……………………………...
Introduction..............................................................................................................
Phenological asynchrony.........................................................................................
Habitat work with Taylor’s Checkerspots in the Puget Sound……........................
Biology of Taylor’s Checkerspots and components of their survival……..............
Available data and statistical models from similar research....................................
What we already know about the weather and its effects from past research.........
The need for weather modeling for the Puget Sound re-introductory effort……...
METHODOLOGY..................................................................................................
Data collection for population counts of E. editha adults.………….......…………
Analysis for butterfly count data……………………………………………….....
Data collection for weather elements ……………………………….……………
Methodology for the statistical analysis………………………………………….
RESULTS AND DISCUSSION..............................................................................
R76 Analysis results...…………………………………………………….……....
R50 Analysis results…………………………………….…………………….......
Scatter Creek South analysis results………………………………….…………...
Discussion……………............................................................................................
CONCLUSION……………………………………………………………………
BIBLIOGRAPHY....................................................................................................
APPENDICES.........................................................................................................

ix
1
1
3
4
5
6
10
10
15
15
20
21
23
27
27
29
32
36
39
40
52

LIST OF FIGURES
Figure 1 – Study sites of overall reintroduction project............................................ 17
Figure 2 – Project sites in Fort Lewis with Gray Army Airfield…………............... 18
Figure 3 – Project site Scatter Creek South………………....................................... 19
Figure 4 – Display of 30-year average values from PRISM for location site R76...
Figure 5 – GAM abundance with precipitation values at R76..................................

23
26

RESULTS AND DISCUSSION
Figure 6 – Plot log Peak (GAM) vs Year for Range 76………………………........
Figure 7 – Plot of year vs. log10 peak (GAM) at site R50..………………………...
Figure 8 – Plot of max vapor pressure deficit vs. log10 peak (GAM) at site R50......
Figure 9 – Plot of temperature vs. log10 peak (GAM) at site R50………………….
Figure 10 – Plot of precipitation vs. log10 peak (GAM) at site R50………………..
Figure 11 – Plot of year vs. log10 peak (GAM) at Scatter Creek South....................
Figure 12 – Plot of temperature vs. log10 (peak GAM) at Scatter Creek South........
Figure 13 – Plot of precipitation vs. log10 (peak GAM) at Scatter Creek South…...
Figure 14 – Plot of max vapor pressure deficit vs. log10 (peak GAM) at SCS……..

28
30
30
31
32
34
35
35
36

vii

LIST OF TABLES
Table 1 – Correlation Matrix for Range76……………............................................
Table 2 –Model Selection Results for Range 50…………………………...............
Table 3 – Model Selection Results for Scatter Creek South.....................................

viii

28
29
33

LIST OF APPENDICES
Appendix A – Data used for the linear regressions.….............................................. 52
Appendix B –Results from raw data without transformation…………………….... 71

ix

ACKNOWLEDGEMENTS
The butterfly field count data for this thesis was collected, compiled and analyzed by the
biologists and data scientists at the Washington State Department of Wildlife (WDFW),
notably Mary Linders, Gail Olsen, Ann Potter, Josh Cook, Emily Phillips, and Lisa
Thompson-Randalf. Their work takes great skill from many years of practice. Mary
Linders helped guide me on my research question and Lisa Thompson-Randalf kept me
informed on the news of butterfly conservation. I also thank my co-students at The
Evergreen State College (TESC). I will never forget Paris McClusky and his comradery
as we suffered together in the CAL for hours and hours. Diane Nelson, Sarah Hieber,
Tracy Scalici, and Keegan Curry gave excellent feedback on my work. I thank my kids,
Sammy and Alex Tassoni for putting up with being without me and my husband Peter
Tassoni for covering childcare duties and lending an ear when I needed empathy. He also
aided me with formatting the thesis. I thank my coworkers, Dustin Wiersma, Brian
Cosentino, Ian Trewella and John Talmadge, in the science division of the wildlife
program at WDFW and Professor Mike Ruth for their expertise and feedback on some of
the programming, data source and GIS-related questions. Dustin helped me a great deal
with Python programming and GIS work needed for the presentation. I appreciate my
supervisor, Shelly Snyder for her patience with my repeated requests for time off during
the crunch times. The professors of TESC deserve mention because they taught me
speaking, data analysis and writing skills, particularly Professors Kevin Francis, Erin
Martin, Ted Whitesell, Michael Ruth, Kathleen Saul, Miranda Mellis and Shawn
Hazbourn. Lastly, I owe a great debt of gratitude to my advisor and reader, Professor
John Withey for his patience and excellent feedback on my dissertation. Dr. Withey’s
expertise in ecological science and statistics was pivotal as he provided guidance on this
project’s methods design and statistical analysis.

x

INTRODUCTION AND LITERATURE REVIEW
Introduction
For those who are paying attention, it is no surprise that there is a general trend
toward a global loss of biodiversity and heavy ecological collapse everywhere due to
anthropogenic climate change and habitat fragmentation (Burlew 2010; DeRosa n.d.;
Ehrlich and Ehrlich 2009; Jarvis 2018; Sugarbaker 2017). We are embarking on an age of
human-caused apocalypse—particularly among the insect populations of the world. This
apocalypse is caused by human over population (Ehrlich and Ehrlich 2009).
In the last 20 years, the Monarch butterflies (Danaus plexippus) have had a 90
percent drop in population, a disappearance of 900 million individuals. The Rusty
Patched Bumble Bee (Bombus affinis) had an 87 percent drop in the last 20 years (Jarvis
2018; “Rusty Patched Bumble Bee,” n.d.; Szymanski et al. 2016).
In the Puget Sound region of the Pacific Northwest, there are many species of
concern. The species that are getting the major attention these days is the Southern
Resident Killer Whale (Orcinas orca) (DeRosa n.d.) and the Northern Spotted Owl (Strix
occidentalis caurina), which has generated huge controversies with respect to
conservation vs. the natural resource extraction interests (Glenn et al. 2011b).
The species of concern that will be the subject of this thesis gets less attention in
the press. It too is being threatened by fragmentation of habitat and global climate
change. It is the Taylor’s Checkerspot Butterfly (Euphydryas editha taylori), a species of
butterfly endemic to the Pacific Northwest. They are a listed species of concern by the
US Fish and Wildlife Service (USFWS). There is a concerted effort by local and federal
agencies as well as non-governmental organizations (NGO)s to save this species from
1

extinction (Andrusyszyn 2013; Grosboll 2011; Linders, Lewis, and Curry 2016; Linders,
Lewis, and Dorman 2016).
Conservation organizations are becoming over-stretched by the need to actively
manage populations and the only way to fix this problem is to bridge fragmented habitats
and repair the damage to renew vigor to the ecosystems that support life (Parmesan et al.
2015). This sentiment was echoed by the keynote speaker for The State of Washington’s
GIS Day 2017. Larry Sugarbaker, the keynote speaker discussed the shifting of attitudes
that have happened during his career in public lands that had started in 1979 when the
prevailing attitude about disturbances to the earth at the time in public lands management
as, “The earth will heal by itself and it is OK to do damage” to one of “conservation of
the environment” during his tenure at the Department of Natural Resources. Larry
Sugarbaker spoke of his own evolution in thinking as an individual, as well as the DNR’s
natural resource professionals toward “conservation” and away from “The earth will heal
itself” sentiment. His main point was that working to save individual species from
extinction is the same as working to prevent whole ecological systems from collapse.
This literature review will cover the past research on the effects of climate on the
E. editha and its subspecies and the differences in climate tolerance between the Bay
Checkerspots of California and Taylor’s Checkerspot Butterflies of the Pacific
Northwest. It will touch on phenological asynchrony in general and the scholarship in
such phenomena. It will discuss the statistical modeling of California’s Bay Checkerspot
populations, as well as similar research on other organisms such as the local populations
of the Northern Spotted Owl and how the climate and its various components affect that
listed species. It will touch on the lack of knowledge there is of climate elements and how
2

they affect E. editha in the Pacific Northwest and the importance of research in this area,
finalizing on the ecological imperatives for conservation in general, tying the efforts to
save one species with the overall goal of preventing whole-scale ecological collapse.
Phenological Asynchrony
The relationship between many species of butterflies and their hosts had evolved
to be a precisely timed event, where the host plant and butterfly match their development
to the timing of each other’s’ life stages (Abarca and Lill 2015; M. C. Singer and
Parmesan 2010).
Unfortunately, global climate change has lengthened the “green” season,
prolonging summer and shortening winter. Spring in the temperate zones arrives earlier
and major weather events such as sudden cold snaps and late storms are more frequent
(Miller-Rushing and Primack 2008; Abarca and Lill 2015).
Change in the climate may have caused E. editha to be one of the many species to
fall subject to phenological asynchrony (the animal needing plant food when it is not
available). Most of the stress is nutritional and hydrological-the plant will age before the
caterpillars are ready from hydrological conditions and degradation of their food source,
thereby starving the larvae. (Michael C. Singer and Parmesan 2010; Weiss and Weiss
1998; McInnis 1997; Raloff 1996; Cohen 1996; “Edith’s Checkerspot” 2007; Parmesan
et al. 2015; Parsons 1995; Bonebrake et al. 2010; Liu et al. 2012; Olson 2017).
Synchrony between the host plant timing and many species of a moth or butterfly
is a honed, precise evolutionary trick programed in the genetics of these creatures (Raloff
1996; Ehrlich and Hanski 2004a; Hanski et al. 2004). For example, in the case of Eastern
Tent Caterpillars (Malacosoma americanum)it has been shown that asynchrony has been
3

triggered by warmer temperatures early in the season (Abarca and Lill 2015). Global
climate change has brought about the prevalence of a longer green season, particularly
earlier springs and later winters (Miller-Rushing and Primack 2008). The reason E. editha
is so susceptible to global climate change is that many species have been forced to
migrate north and to higher elevations to survive. E. editha have a complex relationship
with their host plants and are also known to be stationary in their habitat (non-migratory)
due to this complexity. Several studies done on the Bay Checkerspot butterfly (E. editha
bayensis) state that the stress posed by climate change is forcing Bay Checkerspots of
California to move north and to higher elevations where the climate is damper and cooler.
(Cohen 1996; “Edith’s Checkerspot” 2007; Parmesan et al. 2015; Lacy et al. 2017;
Parsons 1995; McInnis 1997; Michael C. Singer and Parmesan 2010).
Habitat work with the Puget Sound Taylor’s Checkerspots
Here in the Pacific Northwest, it has been suggested that microclimates need to be
studied to be able to assess the habitat needs of the Taylor’s Checkerspot Butterfly (Olsen
2017). Due to urban interference and climate change, plans may be made to move the
species to new locations based on predicted weather-pattern changes, depending on
findings with respect to the climatic requirements of the local E. editha populations.
Habitat degradation generally causes increases in migration rates (Raloff 1996; Linders,
Lewis, and Dorman 2016). Up to this point, the choice of habitat locations have been
selected using a rapid habitat assessment using the multitude of variables that affect the E
editha populations (Linders, Lewis, and Dorman 2016). Microclimate research with
respect to populations of E. editha will help in future habitat assessments.

4

Biology of Checkerspot Butterflies and Components of their Survival
Taylor’s Checkerspot Butterfly populations are affected by a variety of factors
including genetics, pollution, pesticides, behavior, predation dynamics, habitat location,
reproduction and site management. E. editha have predators and parasitoids that prey
upon it. Their natural predators include ants, birds, wasps, and a few other members of
the Hemiptera order (true bugs) such as assassin bugs (Ehrlich and Hanski 2004b).
Humans have used a parasitoid wasp known as Cotesia to control Cabbage Butterflies,
and E editha have been collateral damage. E. editha lose about 67% of their larvae to 1-3
parasitoid species (James, Nunnallee, and Pyle 2011). E. editha caterpillars are brightly
colored, warning birds that they are unpalatable due to their iridoid glycosides they get
from their host plants. E. editha larvae eat when they sense host plants that provide them
with a defensive chemical called, iridoid glycosides that help them to defend against
predation by birds. Such plants include Castilleja (Indian Paintbrush), Plantago
(Plantain), and Collinsia (Blue-Eyed Mary) plants. Castilleja plants are used by the very
young caterpillars in the summer, and the Plantagos and Collinsias are used later by
older, half-grown caterpillars in the spring after they have passed through the winter
months (James, Nunnallee, and Pyle 2011).
An understanding of Checkerspot reproductive biology will be critical to their
conservation. Reproductive biology of Checkerspot butterflies is a model system that has
numerous directions. Though the attempt to understand the decay of E. editha’s genetic
variability by quantitative genetics analysis has not been successful, there are
technological advances allowing researchers to work out the small evolutionary forces

5

that affect their phylogeny with their problems of small populations (Ehrlich and Hanski
2004a).
Behaviors of E. editha are important to consider because of their consequences
for population change. They include feeding behaviors such as what larval host plants are
used, adult nectar feeding, how the animal responds to environmental disturbances, and
how they interact with their natural predators (Ehrlich and Hanski 2004a; Grosboll 2011;
Husby 2012; Michael C. Singer and Parmesan 2010; Aubrey 2013; James, Nunnallee,
and Pyle 2011; Weiss and Weiss 1998).
Available Data and Statistical Models from Similar Research
Available and needed data
Long-term population monitoring with respect to microclimate studies are
imperative because E. editha are so weather dependent. Understanding of weather
patterns and meteorology with respect to global climate change and the different climatic
factors such as rainfall, temperature, and sunlight will be vital to E editha behaviors such
as feeding, migration, mating, the structure and dynamics of E editha population models
and different stages of development of E editha (Parmesan et al. 2015).
Since rainfall is known to delay eclosion (emergence from pupae) of adults,
amount and timing of precipitation should be considered. Minimum and maximum
temperature, as well as the amount of insolation (sunlight) should be measured with
respect to geographic location. The local re-introductory effort needs brood data to its
link to bad years versus good years.
At this point, there is an abundance of data available about climate and adult
butterfly counts. We have field butterfly counts dating back to 2007. WDFW has
6

collected larval and adult distributions through distance sampling by sub-dividing study
area samples into grids and counting the numbers in quadrats (one-square-meter squares)
every 9-25 meters (Murphy and Weiss 1988; Linders, Lewis, and Dorman 2016; Grosboll
2011). Furthermore, WDFW has data on temperatures dating back to 2006. Mary
Linders, a WDFW biologist working with the Taylor’s Checkerspots noted the 2006 July
temperature fluctuations were large, varying from around 20-40 degrees Celsius in areas
she put out the heat sensor data collectors. She stated the cases were plastic and should
not be getting hotter than the surroundings. This is significant because the Bay
Checkerspots are losing the night time cooling effects during the summer, which could
have implications for E. editha (Arndt 2015). Given the large amount of data WDFW has
with day and night time fluctuations, we may be able to answer the question about nighttime warming as mentioned by Arndt. This could be the focus of future research.
Digital Elevation Data
Due to the fact that the climate data is dependent on elevation in that the greater
the elevation, the higher the precipitation and the lower the temperature, it was important
to look at the elevation data for our Joint Base Lewis-McCord sites available on the
University of Washington’s School of Oceanography website (Finlayson et al., n.d.) with
respect to the butterfly count locations and the weather station locations. Elevation data
has been considered with the climate data used in the analysis. The data used for the final
analysis incorporated elevation and other factors such as slope and aspect (Daly et al.
2008).
Climate Data

7

Ideally, climate data used for research similar to this thesis should be accurate to
the distance sampling plots, and precise to the count data. Due to the fact these counts are
daily numbers, the weather data needs to be daily too. A good source of weather data is
Wunderground, the website that contains the daily weather values in the Joint Base Lewis
McChord area and is based off the closest weather station (Gray Army Airfield). The
primary source of data this thesis will use is from a model used for a project at Oregon
State University that has assembled spatial climate datasets for short- and long-term
climate patterns known as Parameter-elevation Relationships on Independent Slopes
Model (PRISM). PRISM uses a linear regression model to estimate temperature and
precipitation as a function of elevation (Daly et al. 2008). A review of the weather
stations used by PRISM revealed that the Gray Army Airfield is included in the PRISM
model, thereby making this model a more reliable data source for the data analysis.
Statistical Model Ideas based on Past Research
Variance components analysis is what is currently being considered by the Puget
Sound wildlife professionals in the same way they have analyzed the local effect of
climate on the Northern Spotted Owls. Variance components analysis is a way to
compare populations of E.editha to these different co-variates. The use of variance
components analysis elucidates if climate is a significant influence on the local E. editha
populations (Olsen 2017). Hypotheses for this thesis will be chosen based on the
components of climate and the choice of statistical models will be formulated to the
hypotheses. The process of comparing populations to the different co-variates of climate
to other co-variates, such as habitat and site management gives wildlife managers an idea
if climate is a significant influence on the local E. editha populations (Olsen 2017).
8

There is also another approach to climate modeling on the Taylor’s Checkerspots in the
Pacific Northwest. Climate patterns have been evaluated through the use of species
distribution models with respect to location to determine range shifts depicted on
geographic maps (Parmesan et al. 2015). There are linear regression models outlined in
the publication where Parmesan et al. (2015) and Weiss and Weiss (1998) calculated
range-shifts of Checkerspots through meta-analysis of different datasets to identify
patterns evident in climate data and current Puget Sound population models for the
Taylor’s Checkerspot Butterflies (Parmesan et al. 2015; Weiss and Weiss 1998).
Past Research on how climate affects populations in general and how it relates to this
project
Statistical models that have been used to analyze the variance of climate
components on the Northern Spotted Owl reproduction and populations are being
considered by the Puget Sound wildlife professionals involved with the management of
the Taylor’s Checkerspot Butterfly Glenn et al. (2011a, 2011b). Researchers have
examined the relationship of survival rates of populations of Northern Spotted Owl to
local weather and regional climate variables. Perhaps these studies could be a direction
for the efforts to understand the local effects of weather on the Taylor’s Checkerspots of
the Puget Sound region (Glenn et al. 2011a).
Since rainfall is known to delay eclosion of adults, precipitation should be
considered. Minimum and maximum temperature, as well as the amount of insolation
(sunlight) should be measured with respect to geographic location. Weiss and Weiss
(1998) measured these weather variables as a function of location, taking data points
every seven days (Weiss and Weiss 1998).
9

In addition to the work done by Glenn et al. (2011a, 2011b), climate and weather
work done by Parmesan et al. (2015) and Weiss and Weiss (1998) provide viable
templates for calculating the range-shifts of Checkerspots through meta-analysis of
different datasets to identify patterns evident in climate data and current Puget Sound
population models for the Taylor’s Checkerspot Butterflies. Climate patterns will need to
be evaluated through the use of species distribution models with respect to location to
determine range shifts depicted on geographic map. There are linear regression models
outlined in the Parmesan et al. (2015) publication.
What we already know about the climate and how it affects E. editha from past
research
There has been comprehensive research on future habitat models with regard to
the Bay Checkerspot Butterflies, a different sub-species of Checkerspot butterfly in
Southern California (Parmesan et al. 2015). Climate change brings different climate
situations to different places. California’s funding situation has allowed for studies to be
done on the E. editha in that area, namely the Bay’s Checkerspot, E. editha bayensis.
Research has shown that in California, drought has played a major role in climate
perturbations to the species who are struggling in this bug apocalypse. For the Bay
Checkerspots in California, drought has nasty implications such as lack of time for the
larvae to develop before their host plants senesce. This phenomena has caused the natural
resource managers of this area responsible for the California Bay Checkerspots to
actively relocate this non-migratory species to higher elevations and northward (Raloff
1996; McInnis 1997)
The Need for Climate Modeling for the Puget Sound Re-Introductory Effort
10

Much like California Bay Checkerspots, one of the major problems facing E.
editha taylori has been global climate change. The local weather patterns and regional
climate oscillations affect E. editha and the food plants on which they depend.
Understanding of weather patterns and meteorology with respect to global climate change
and the different weather factors such as rainfall, temperature, and sunlight will be vital
to understanding E editha behaviors such as feeding, migration, and mating. It will also
be important for the understanding of the structure and dynamics of E editha population
models and different stages of development of E editha (Parmesan et al. 2015).
Due to the differences between the heavily studied Bay Checkerspots in
California and the Taylor’s Checkerspots here in the Northwest, wildlife planners in the
Puget Sound need to have local information on how our regional climate patterns affect
this endangered species. The dynamics and differences of the phylogenic asynchrony
may be different for Taylor’s Checkerspot as opposed to the Bay Checkerspot of
California.
Here in the Pacific Northwest, the effects of global climate change may be much
different. The Northwest is a cool, wet place with variable climate such as droughts and
wet years. E. editha requires basking in the sun to survive. Climate change here in the
Pacific Northwest will bring hotter, dryer summers and warmer, wetter winters (Glenn et
al. 2011a; Parsons 1995). Voltinism (number of broods this insect has in a year) is
indicative of climate, as well as the number of instars it has (larval growth stages). Here
in the Puget Sound, E. editha vary in both. This genus uses diapause (dormancy such as
hibernation) to overcome stresses related to weather (James, Nunnallee, and Pyle 2011).

11

Eclosion (emergence) depends on rainfall and the first flight season of butterflies
determines the oviposition preferences (distribution of eggs).
Excessive rainfall will retard post-diapause (hibernation) larval development,
delaying the time the butterflies can have a chance to fly before their plants die, leaving
their larvae to starve (Parsons 1995). Mary Linders, the biologist in charge of the
Washington State Department of Fish and Wildlife’s E. editha conservation work stated
that there is a concern here in the Pacific Northwest that the excessive moisture may be
causing the pupae to rot in their cocoons before they have a chance to eclose. This is a
theory that needs to be researched, because it is a completely different effect that global
climate change has brought to the local E. editha here in the Pacific Northwest than what
has been in California. Adding to the problems facing these creatures, E. editha’s
reproductive behaviors are being impacted by climate change causing a “lag effect”,
where the butterflies are effected by the previous year’s climate (Parmesan et al. 2015;
Weiss and Weiss 1998). Herein lies the challenge of reintroduction and habitat planning
for the biologists and conservationists involved in the Pacific Northwest.
The 2017 final annual report to the US Fish and Wildlife Service, Joint Lewis-McChord,
and the ACUB Technical Review Committee compiled by the Washington State
Department of Fish and Wildlife (WDFW), the Oregon Zoo, and The Evergreen State
College (TESC) dedicated much of its discussion at the end of the report calling for
research on climate-related effects on Taylor’s checkerspot populations in the Puget
Sound to help answer questions they have about habitat quality such as patch size and
connectivity. (Linders, Lewis, and Curry 2016).

12

The methodology for climate research outlined by biologists are as follows (M. J.
Linders, 2013):
•

Combine the climate components data

•

Find gaps in the data (in number of days)

•

Compile start and end flight season dates per year per site

•

Look for patterns of checkerspot abundance with respect to climate elements.

•

Look for anomalies in the climate data: find average range of temperature and
precipitation on a monthly basis.

•

Generate a summary of butterfly abundance and phenology with respect to the
climate elements.

The data sets come from a variety of different sources, most notably, Wunderground,
PRISM, and NOAA.
Study of the local populations with respect to the local weather patterns should
assist the wildlife managers in finding suitability in habitat assessments. Because of the
lack of knowledge of the local Puget Sound Taylor’s Checkerspots (Euphydryas editha
taylori) with respect to the provincial weather patterns and regional climate oscillations
due to the lack of available funding to conduct such studies. Long-term population
monitoring with respect to microclimate studies are imperative because E. editha are so
weather dependent. WDFW biologists have stressed that adult eclosion (emergence)
depends on rainfall and the first flight season of butterflies determines the oviposition
preferences (distribution of eggs).
Conservation of Taylor’s Checkerspot

13

E. editha are in danger of going extinct. They were listed in 2000 as endangered
and there are only about 100 left in Canada (“Edith’s Checkerspot” 2007). We need to
look at how this species is being affected by global climate change locally and compare
these effects to other taxonomic groups (Ehrlich and Hanski 2004a). Phenological
asynchrony is not just hurting E. editha. It is a problem for the prairie plants who depend
on the timing of their pollinators (Husby 2012). Ecological conservation is not just about
one species, it is about systems and the collapse of entire web networks of organisms that
depend on each-other. The plight of the local populations of E. editha is just a small
component of a breath-taking problem of ecological collapse.
Research such as this brings to light the local problems caused by global climate
change and highlights the importance of habitat conservation. It helps in shifting public
attitudes on global climate change through demonstration of the local effects of weather
and climate change through time and the effects of these changes on specific species.
Studies such as these show that it is imperative habitats are not only repaired but
expanded to prevent the biological apocalypse from happening. This study on climate and
how it affects E. editha is necessary because conservation managers such as WDFW and
the Oregon Zoo must make informed management decisions such as when and where to
place captive bred Taylor’s Checkerspot Butterflies for maximum probability of
reintroduction success knowing that climate and weather will be a component of survival.

14

METHODOLOGY
Data Collection for population counts of E. editha adults
The Washington State Department of Fish and Wildlife (WDFW) biologists used
distance sampling to obtain the population count data. The advantage of distance
sampling is that it is a statistically based methodology that accounts for imperfect and
variable detectability brought about by climate, vegetation, observer, time of day, or
number of individuals. Detectability declines with distance (Olson, 2017), and this can be
modeled to account for variability in data collection.
Distance sampling is a technique where observers walk along transect lines that
are established with fixed coordinates and marked with colored flagging. The observer
walks along the lines looking for the butterflies and records the perpendicular distance
from the line to the butterfly. This is a difficult skill to learn and takes extensive training
or sophisticated measurement equipment (Olson, 2011). They used a survey technique
called “line transect sampling”, which accounts for differences in detectability of the
butterflies in some sites vs. others (Brown & Boyce, 1998). Transects are imaginary 700meter lines drawn through a sampling area. The spacing between the lines are 100 meters
with the segment length altered to be 25 meters (M. J. Linders & Olson, 2014a; M.
Linders, Lewis, & Curry, 2016). The definition of a segment is a unit within the 700meter transect where butterflies are counted. Essentially, each segment is the “container”
containing each butterfly count. The counts were performed between 1000 and 1630
hours (M. J. Linders & Olson, 2014). The WDFW biologists have collected data in the
Pacific Northwest on adult presence, distribution, and relative abundance. Surveys and
transects were conducted only if the ambient temperature was ³11.7° C, there was
15

sufficient sunshine to cast a soft or distinct shadow, or if no shadow temperature was
>15.5°C.
Project sites
There are six project sites in the Puget Sound chosen for the rehabilitation activities.
•

The Scatter Creek Wildlife Area-South Unit (SCS)

•

Range 50, Joint Base Lewis-McCord (JBLM)

•

The Pacemaker Airstrip, JBLM

•

The Glacial Heritage Preserve

•

Training Area 7 South, JBLM

•

Range 76, JBLM

Because of data availability, only three of the six sites will be used for this particular
study: Range 76 (R76) and Range 50 (R50) at JBLM, and Scatter Creek Wildlife AreaSouth Unit (SCS).

16

Figure 1 Taylor’s checkerspot rehabilitation sites (Linders, Lewis, & Curry, 2016)

17

Figure 2: Location of R50 and R76 shown with Gray Army Airfield weather station location (included in PRISM network)

18

Figure 3: Location of Scatter Creek South site

19

Adult butterfly counts
The biologists conducted surveys of adults up to three times during the flight
season, which ranges from between early April and late May. They chose the locations of
the survey sites to be the sites of reintroduction with a 200-foot buffer to include those
butterflies that wandered off the reintroduction sites.
Analysis for butterfly count data
Observer differences, especially in 2016 where there were a few observers who
were in training who broke the guidelines that had been set by Linders and Olson in 2014
(M. J. Linders & Olson, 2014) created confounding factors. Due to this problem, the
biologists had to perform data manipulation to fit curves to the data and increase the 95 %
confidence intervals.
WDFW biologists analyzed the data with the program called, “Distance, Version
6.2”(Thomas et al., 2010). They first generated summary statistics in SAS statistical
software. Models of analysis were developed to account for the observer differences.
Akaike’s Information Criterion (AIC), goodness of fit tests, and additional criteria were
used to detect the variability (error) in encounter rates at transect lines. Variance
estimates of density were calculated using the method of Fewster et al. (2009). Otherwise
a non-parametric bootstrap method was used (Marques, Thomas, Fancy, & Buckland,
2007). This method makes sense in that count numbers to the expert eye should be in
higher numbers near the transect with a smooth drop off in sightings the farther the
distance from the observer. The program, “Distance” takes these curves in the data and
generates density estimates, which are calculated to account for total butterfly abundance

20

for each study site. The abundance calculations were then used for the general additive
models (GAM) that estimate the peak abundance values for each study year.
Data collection for climate elements
Overall, the climate elements biologists at WDFW find pertinent are:
•

Insolation

•

Degree days

•

Rainfall (precipitation)

•

Overwinter moisture, total and delivery in heavy flood-prone events

•

Number of days below freezing.

This thesis used weather data available on the PRISM website maintained and managed
by the University of Washington’s School of Oceanography. PRISM offers a datasets
from an interpolation of a network of weather stations that account for slope, distance
from the coast, elevation, and aspect (Daly et al., 2008). The data from PRISM is
available in the form of a geographic raster DEM format where the climate elements can
be converted to Z-values or can be simply downloaded from a point source as a direct
download in comma separated values into a spreadsheet. The most popular data from
PRISM are the 30-year average datasets, can be used to discuss the findings for long term
implications, or be used to examine climate anomalies (Figure 4).
This thesis used the point source data downloads option to obtain the climate
measurements closest to the study sites. Before downloading the data, it was confirmed
the weather stations included in the network are close to the study sites. The Gray Army
Airfield is the weather station located closest to the sites in Joint Base Lewis-McCord.

21

Range 76 is 4.0 miles and Range 50 is 5.9 miles away. The Centralia weather station is
the closest weather station to Scatter Creek at a distance of 6.5 miles.
Only precipitation, humidity, and temperature were used as the independent
variables for this analysis and only the winter climate elements were examined. Winter
values are for the months of December, January, and February preceding a given year’s
butterfly abundance estimates. Temperature (temp) was calculated as the average, in
degrees Fahrenheit, for the winter months, humidity was the winter average max vapor
pressure deficit (VPD, a measure of dryness), and precipitation was the total winter sum
of precipitation in millimeters (mm). Humidity and temperature values were measured
value readings, and precipitation was a measure summed value from PRISM.
The analysis included the average winter temperature (average of the mean values
returned from PRISM). The overwinter precipitation values was the total sum of
precipitation for the winter months (PRISM measure converted to millimeters and added
over the entire winter), and the over winter humidity was an average of the maximum
dryness for each winter month as opposed to the average of minimum dryness. Taking
the maximum dryness values in vapor pressure deficit was a judgement call to measure
how maximum dryness is affecting the butterfly populations.

22

Figure 4: 30-year average from 1981-2010 of precipitation values in millimeters on and near site R76

Methodology for the statistical analysis
The purpose of this thesis is to assist land managers in the task of predicting the
phenology of the Taylor’s Checkerspot Butterfly in the areas of reintroduction. The
question this thesis aims to answer is, “Do winter climate variables in the South Puget
Sound predict the population counts of Taylor’s Checkerspot Butterflies?” We already
know that post-diapause larvae are dependent on insolation to develop at the correct
physiological time (Weiss, Murphy, Ehrlich, & Metzler, 1993; Weiss & Weiss, 1998).

23

The statistical analysis
The statistical analysis chosen for this thesis included regression models
performed in R: A Language and Environment for Statistical Computing (R Core Team
2019). Each climate element was inspected with respect to the peak general additive
model (GAM) values for abundance. The choice to use a generalized linear model was
affirmed by an examination of all the plots of the residuals, which indicated reasonable
model fit.
For the three sites with enough data for analysis (R76, R50 and SCS), the
assumption was made that the count of butterflies has its peak when winter climate
conditions were ideal – i.e. the species needs a certain range of temperatures, humidity,
and precipitation. The null hypothesis for each model was H0: “The winter climate
element does not affect the abundance of butterflies for the following spring”. This null
hypothesis was tested in a series of steps:
1) Plot the peak GAM point of butterfly abundance against each climate
variable. Consider a log transformation of peak GAM values.
2) Perform a correlation analysis on the data with respect to each climate
variable to assess whether regression analyses are merited.
3) If called for, run a simple linear regression on each climate variable and
year, with either raw peak GAM or log-transformed peak GAM as the
response.
4) Use a model selection approach to assess multiple models (both simple
and multiple linear regression models, as deemed reasonable).

24

If the scatterplots and examination of the plots of the residuals show the climatic
variables have a linear relationship to the peak GAM abundance values, the linear
regression would give this equation:
𝑌~𝛽$ + 𝛽' 𝑥)' + 𝛽* 𝑥)* + 𝛽+ 𝑥)+ + 𝜀)
Here is the multiple linear regression model for this quadratic relationship, if there is one:
𝑌~𝛽$ + 𝛽' 𝑥)' + 𝛽* 𝑥)' * + 𝛽+ 𝑥)* + 𝜀)
Where
𝑌 = 𝐵𝑢𝑡𝑡𝑒𝑟𝑓𝑙𝑦 𝑐𝑜𝑢𝑛𝑡(𝑝𝑒𝑎𝑘 𝑎𝑏𝑢𝑛𝑑𝑎𝑛𝑐𝑒 𝑜𝑟 𝑎𝑟𝑒𝑎 𝑢𝑛𝑑𝑒𝑟 𝑡ℎ𝑒 𝑐𝑢𝑟𝑣𝑒 𝑓𝑜𝑟 𝑒𝑎𝑐ℎ 𝑠𝑒𝑎𝑠𝑜𝑛)
𝑥' = 𝑃𝑟𝑒𝑐𝑖𝑝𝑖𝑡𝑎𝑡𝑖𝑜𝑛 in mm
𝑥* = 𝑇𝑒𝑚𝑝𝑒𝑟𝑎𝑡𝑢𝑟𝑒 in Fahrenheit
𝑥+ = 𝐻𝑢𝑚𝑖𝑑𝑖𝑡𝑦 𝑖𝑛 𝑉𝑃𝐷(𝑎 𝑚𝑒𝑎𝑠𝑢𝑟𝑒 𝑜𝑓 𝑑𝑟𝑦𝑛𝑒𝑠𝑠)
𝑖 = 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑜𝑏𝑠𝑒𝑟𝑣𝑎𝑡𝑖𝑜𝑛𝑠
Other climate variables (x) could be added to the model, such as wind, number of days
below freezing, insolation, heavy flood events, etc., but for the scope of this thesis, the
multiple linear regression will be limited to temperature, humidity and precipitation.
Furthermore, it will be assumed that the climatic variables used as independent variables
in this analysis will be independent enough from each other if the correlation coefficient
is less than r = 0.8.

25

Modeling the abundance values
The values for abundance data from the three sites that had been compiled by
WDFW statisticians and biologists as output from the program “Distance” were used as
inputs for the GAM model (Hastie & Tibsirani, 1987) for the purpose of extracting peak
abundance values for each site per year. The data for these models needed to have the
degrees of freedom calculated due to a low number of observations (from 3-12 in a given
year).

Figure 5: Plot of abundance vs. Julian date for the year 2014 in R76 as a general additive model (GAM). All of these
plots are available for inspection in the Appendices.

As an example, the GAM model shown for the Range 76 site in 2014 (Figure 5) yields a
peak GAM value of 2919 butterflies (see Appendix A for all individual GAM plots).
These yearly maximum values from the GAM models were used in linear regression and
26

correlation models to determine if the winter climatic variables affect the populations of
butterflies.
Three climatic variables were tested against the peak GAM values for the three
study sites: precipitation in millimeters, temperature in Fahrenheit, and humidity in
average maximum vapor pressure deficit (VPD), which is a measurement of dryness
(Prenger & Ling, n.d.). An analysis with the peak GAM values as the response was run
first with no significant results (See Appendix B).
Regression models were created with log10-transformed peak GAM values as the
response variable. All candidate models were evaluated using Akaike’s information
criterion corrected for small sample sizes (AICc, recommended for cases where n/K < 40,
Burnham and Anderson 2002) to determine empirical support for the different candidate
models. This analysis differed from the raw data analysis in that it also included the year
as an independent variable.

RESULTS AND DISCUSSION
R76 data analysis of log-transformed peak GAM data
Correlation of log-transformed peak GAM data from site R76 with winter climate
variables was low (Table 1), and there did not appear to be any trend across the years of
the data (Figure 6). Therefore, regression analyses were not performed for the R76 peak
GAM data.

27

Table 1: Correlation Matrix for site R76.
PeakGAM

Log
Peak(GAM)

Precipitation

Temperature

Vapor Pressure
Deficit

Year

1

NA

.102

.180

-.357

.074

NA

1

.122

.245

-.254

.135

Precipitation

.102

.122

1

.431

-.186

-.056

Temperature

.180

.245

.431

1

-.297

.253

-.357

-.254

-.186

-.297

1

-.647

.074

.135

-.056

.253

-.647

1

Variable
PeakGAM
Log(peakGAM)

Vapor Pressure deficit
Year

Figure 6: Plot of the log Peak (GAM) vs. year at site R76

28

R50 data analysis of log-transformed data
Using a model selection approach, a simple linear regression with ‘year’ as the only
independent variable had the best support and year was included in the top four supported
models (Table 2, Figure 7). In the best-supported model, year had a coefficient of 0.1684
for an annual average increase of 47% (Figure 7). For illustration, bivariate plots of logtransformed peak GAM from R50 with each climatic variable are also included here
(Figures 7-10), but these variables were only part of models with relatively little support
(Akaike weights < 0.05). In comparing models using the evidence ratio, the simple
regression model with the variable, “Year” is 23.8 times more likely to be a better model
than the simple linear regression model with vapor pressure deficit, based on the Akaike
weights.

Table 2: Model selection results for site R50 with log-transformed peak GAM values as the response variable. Only
models with ΔAICc < 10 and Akaike weights > 0.01 are shown.
Model: log10 (peakGAM) ~
Year

AICc

ΔAICc

Akaike
weight

Adjusted R2

P.Value

F-Statistic

8.20

0.00

0.882

0.78

<0.001

30.19 on 1 and 7 df

vapor pressure deficit + year

14.56

6.36

0.037

0.77

<0.001

14.48 on 2 and 6 df

temperature + year

14.74

6.54

0.033

0.77

0.005

14.15 on 2 and 6 df

precipitation + year

15.39

7.19

0.024

0.75

0.007

12.94 on 2 and 6 df

vapor pressure deficit

16.30

8.10

0.015

0.47

0.025

8.11 on 1 and 7 df

29

Figure7: Plot of year vs. log peak (GAM) at site R50. Regression equation is log(peak GAM) = -336 +17 * year.

Figure 8: Plot of max vapor pressure deficit vs. log peak (GAM) at site R50. Regression equation is log(peak GAM) = 5.54-.77*vpd

30

Figure 9: Plot of Temperature vs. log (peak GAM) at site R50. Regression equation is log peak (GAM) = 4.89-.04*T

31

Figure 10: Plot of Precipitation vs log peak (GAM) at site R50. Equation is log (peak (GAM)=2.25 + .003*ppt

SCS data analysis of log-transformed peak GAM data
Using a model selection approach, a simple linear regression with year as the only
independent variable had the best support, and year was included in all supported models
(i.e. with Akaike weight > 0.01, Table 3, Figure 11). In the best-supported model, year
had a coefficient of 0.2436, for an annual average increase of 75% (Figure 11). For
illustration, bivariate plots of log-transformed peak GAM from SCS with each climatic
variable are also included here (Figures 11-14) but these variables were only part of
32

models with relatively little support (Akaike weights < 0.10). In comparing models using
the evidence ratio, the simple regression model with the variable, “Year” is 13.9 times
more likely to be a better model than the multiple linear regression model with
temperature + year, based on the Akaike weights.

Table 3: Model selection results for site SCS with log-transformed peak GAM values as the response variable. Only
models with ΔAICc < 10 and Akaike weights > 0.01 are shown.
Model: log10 (peakGAM) ~

AICc

ΔAICc

Akaike
weight

Adjusted R2

Year

13.34

0.00

0.849

temperature + year

18.62

5.28

precipitation + year

19.13

vapor pressure deficit + year

19.26

P.Value

F-Statistic

0.84

0.000122

47.92 on 1 and 8 df

0.061

0.83

0.000861

22.79 on 2 and 7 df

5.79

0.047

0.82

0.00103

21.48 on 2 and 7 df

5.92

0.044

0.82

0.00108

21.14 on 2 and 7 df

33

Figure 11: Scatter Creek South plot of log peak (GAM) vs. Year Regression Equation is log peak(GAM)=-488.19 + .24 *
Year

34

Figure 12: Scatter Creek South plot of log peak (GAM) vs. Temperature. Regression equation is log peak (GAM) = 4.86 .06 * T

Figure 73: Plot of Scatter Creek South log peak(GAM) vs. precipitation. Regression equation is log peak(GAM) = 0.74 +
.003 * ppt

35

Figure 14: Plot of Scatter Creek South log peak(GAM) vs. vapor pressure deficit. Regression equation is log peak(GAM)
= 4.78 - .82 * vpd

Discussion
There is an active effort to reintroduce butterflies to several locations, including
Scatter Creek South and R50. As these prairies are being improved for habitat and
butterflies are reestablishing themselves, the populations may just be naturally increasing
due to habitat filling where the butterflies are being returned to their historic ranges,

36

resulting in a significant increase over the past ten years, and reflected in the model
selection results for R50 and SCS
Perhaps the population numbers are expanding due to having been introduced to
high-quality habitat and could still level off and/or become more subject to variations due
to climatic conditions. Abundance of this species is a complex question that involves
many different elements besides climate. Additionally, the question of climate is further
complicated by the possibility of a lag effect, where the previous year’s climatic variables
are probably affecting the numbers of the butterflies for the current year. Many studies on
the climate should be done to answer the question of climate and its effect on E. editha
taylori here in the Pacific Northwest.
This thesis was a small-scale examination of the winter climatic variables, but
there is the possibility that summer climatic variables are just as, or more important to
consider and should be the topic of future studies. WDFW biologists working with the
Taylor’s Checkerspots noted the 2006 July temperature fluctuations were large, varying
from around 20-40 degrees. Bay Checkerspots of California are losing the night time
cooling effects during the summer, which could have implications for E. editha (Arndt
2015). Given the large amount of summer data WDFW has with day and night time
fluctuations, we may be able to answer the question about night-time warming as
mentioned by Arndt.
Peak GAM points
One important thing to note is that the GAM models needed to have the degrees
of freedom calculated and added to the models due to the low number of observations.

37

This calculation may have added some error to the process. Site R76 needed to have year
2008 excluded due to insufficient data.
These findings in the context of climate change
The choice of following the data for winter climatic variables as a possible effect
on the local Taylor’s Checkerspot butterflies is due to the phenomenon of climate change.
Unfortunately, global climate change has lengthened the “green” season. This could be
the reason the humidity values are so important in this data analysis. If the humidity is
such that winters are too dry, we may be witnessing the deleterious effects of the
unseasonal dryness during the winter. (Miller-Rushing and Primack 2008; Abarca and
Lill 2015).
Indeed, studies have shown that the Bay Checkerspot butterflies (E. editha
bayensis) in California have move north and to higher elevations where the climate is
damper and cooler due to the deleterious effects of the dryness brought on by climate
change. (Cohen 1996; “Edith’s Checkerspot” 2007; Parmesan et al. 2015; Lacy et al.
2017; Parsons 1995; McInnis 1997; Michael C. Singer and Parmesan 2010). Perhaps the
increases in butterfly abundance at these sites is due to the fact that the effects of Global
Climate change here in the Pacific Northwest may be helpful to the butterfly during the
winter months. The Northwest is a cool, wet place, though droughts do occur here.
Climate change here in the Pacific Northwest will bring hotter, dryer summers and
warmer, wetter winters (Glenn et al. 2011a; Parsons 1995). The worry here in the Pacific
Northwest is that the wetter winters will cause the pupae to rot. Pupation does not occur
till spring, and the butterflies are in larval stage diapause during the winter, so the
humidity values over winter are not the whole picture.
38

In the context of climate change, the work to save the E. editha taylori is not just
about that species, it is about the repair of whole ecological systems.

CONCLUSION
This research revealed evidence of winter climate affecting the estimates of peak
abundance in the three sites examined. (See tables 2 and 3). The main finding of this
thesis was that the two reintroduction sites increased with time, where Scatter Creek
South increased an average of 47% over time and Range 50 increased in abundance an
average of 75% over time. This phenomenon could simply be a reintroduction effect and
not related to climate at all.
Maximum dryness of the air during the winter is possibly to be linked to lower
abundances of butterflies the following spring in Range 50 and Scatter Creek South.
Range 76 showed no significant correlations between the climate variables and the
abundance estimates. Possible links of high vapor pressure deficit to lower abundance
estimates hint similar findings from studies on the Bay checkerspot, which show excess
dryness in the air to be a detriment.
Funding is being sought here in the Pacific Northwest to explore the effects of
climate on the abundance of this endangered species. This work is a small part of a giant
project where future researchers can and should explore many angles to help solve the
question of what the effects of climate are on E editha of the Puget Sound.

39

BIBLIOGRAPHY
10 Terrific Facts About Trilobites. (2015, September 23). Retrieved November 5, 2017,
from http://mentalfloss.com/article/68881/10-terrific-facts-about-trilobites
Abarca, M., & Lill, J. T. (2015). Warming affects hatching time and early season survival
of eastern tent caterpillars. Oecologia, 179(3), 901–912.
https://doi.org/10.1007/s00442-015-3371-x
Allen, E., Dodero, M., Longcore, T., Murphy, D., Parmesan, C., Pratt, G., … Anderson,
A. (2003). Recovery Plan for the Quino Checkerspot Butterfly (p. 133) [Recovery
Plan]. U.S. Fish and Wildlife Service.
Andrusyszyn, R. (2013). Using Fire for Butterflies: Soil Characteristics Across a Burn
Gradient in Western Washington Prairies (Master of Environmental Study,
Evergreen State College).
Arndt, D. (2015, November 24). Climate change rule of thumb: cold “things” warming
faster than warm things. [Blog]. Retrieved October 4, 2017, from National Oceanic
and Atmospheric Administration web site website: https://www.climate.gov/newsfeatures/blogs/beyond-data/climate-change-rule-thumb-cold-things-warming-fasterwarm-things
Aubrey, D. (2013). Oviposition Preference in Taylor’s Checkerspot Butterflies
(Euphydryas editha taylori): Collaborative Research and Conservation with
Incarcerated Women (Master of Environmental Study, Evergreen State College).
Bennett, V. J., Smith, W. P., & Betts, M. G. (2012). Evidence for Mate Guarding
Behavior in the Taylor’s Checkerspot Butterfly. Journal of Insect Behavior, 25(2),
183–196. https://doi.org/10.1007/s10905-011-9289-1
40

Boersma, D. (1998). Butter fly Territoriality and Dispersal at Fort Lewis.
Bonebrake, T. C., Ponisio, L. C., Boggs, C. L., & Ehrlich, P. R. (2010). More than just
indicators: A review of tropical butterfly ecology and conservation. Biological
Conservation, 143(8), 1831–1841. https://doi.org/10.1016/j.biocon.2010.04.044
Bornad, C. N., Kery, M., Bueche, L., & Fischer, M. (2014). Hide-and-seek in vegetation:
time-to-detection is an efficient design for estimating detectability and occurrence.
Methods in Ecology and Evolution, 5, 433–442. https://doi.org/10.1111/2041210X.12171
Boughton, D. A. (1999). Empirical Evidence for Complex Source-Sink Dynamics with
Alternative States in a Butterfly Metapopulation. Ecology, 80(8), 2727–2739.
Breuker, C. J., Brakefield, P. M., & Gibbs, M. (n.d.). The association between Wing
Morphology and Dispersal is Sex-Specific in the Glanville fritillary Butterfly
Meliatea cinxia (Lepidoptera: Nymphalidae). European Journal of Entomology,
104(3), 445–452.
Brown, J., & Boyce, M. S. (1998). Line transect sampling of Karner blue butterflies
(Lycaedes melissa samuelis). Environmental and Ecological Statistics, 5, 81–91.
Burlew, D. A. (2010). The Effects of Pesticide-Contaminated Pollen on Larval
Development of the Honey Bee, Apis mellifera (Master of Environmental Study,
Evergreen State College).
Burnham, K. P., Anderson, D. R., & Burnham, K. P. (2002). Model selection and
multimodel inference: a practical information-theoretic approach (2nd ed). New
York: Springer.

41

Center, N. N. C. D. D. (n.d.). Metadata Standards | National Centers for Environmental
Information (NCEI) [Document]. Retrieved October 15, 2017, from
https://www.ncddc.noaa.gov/metadata-standards/
Cohen, P. (1996). Edith’s butterfly flees north. New Scientist, 151(2045), 9–9.
Daly, C., Halbleib, M., Smith, J. I., Gibson, W. P., Doggett, M. K., Taylor, G. H., …
Pasteris, P. P. (2008). Physiographically sensitive mapping of climatological
temperature and precipitation across the conterminous United States. International
Journal of Climatology, 28(15), 2031–2064. https://doi.org/10.1002/joc.1688
Dempster, J. P., & Hall, M. L. (1980). An attempt at re-establishing the swallowtail
butterfly. Ecological Entomology, 5, 327–334.
DeRosa, K. (n.d.). Southern resident killer whales at risk of extinction: study. Retrieved
November 1, 2017, from Times Colonist website:
http://www.timescolonist.com/news/local/southern-resident-killer-whales-at-risk-ofextinction-study-1.23078363
Dobkin, D. ., Olivieri, I., & Ehrlich, P. R. (1987). Rainfall and the interaction of
microclimate with larval resources in the population dynamics of checkerspot
butterflies inhabiting serpentine grassia. Oecologia, 71, 161–166. Retrieved from
Berlin.
Dodge, J., & Writer. (n.d.). “Just a bug, but a very beautiful bug.” Retrieved October 14,
2017, from the Olympian website:
http://www.theolympian.com/news/local/article25254070.html
Dunn, C. D., & Napier, J. A. (1975). An evaluation of factors affecting the in vitro
bioassay for erythropoietin. Experimental Hematology, 3(6), 362–374.
42

Dzurisin, J., Schultz, C. B., & Sheperdson, D. (2004). Captive rearing of Fender’s and
Puget blue butterflies: Protocols and conservation consequences [Report to
American Zoo and Aquarium Association Fore project funded by AZA’s
Conservation Endowment Fund]. Washington State University Vancouver, 14204
NE Salmon Creek Ave, Vancouver, WA 98686 and Oregon Zoo, 4001 SW Canyon
Rd. Portland, OR 97221.
Edith’s Checkerspot. (2007). Canadian Wildlife, 13(2), 6.
Ehrlich, P. R., & Davidson, S. E. (1960). Techniques for Capture-Recapture Studies of
Lepidoptera Populations. Journal of the Lepidopterists’ Society, 14(4), 227–229.
Ehrlich, P. R., & Ehrlich, A., H. (2009). The Population Bomb Revisited. The Electronic
Journal of Sustainable Development, 1(3), 63–71.
Ehrlich, P. R., & Hanski, I. (2004a). Checkerspot Research. In On the Wings of
Checkerspots a Model System for Population Biology (pp. 3–16). Oxford : New
York: Oxford University Press.
Ehrlich, P. R., & Hanski, I. (Eds.). (2004b). On the wings of checkerspots: a model
system for population biology. Oxford : New York: Oxford University Press.
Fewster, R. M., Buckland, S. T., Burnham, K. P., Borchers, D. L., Jupp, P. E., Laake, J.
L., & Thomas, L. (2009). Estimating the Encounter Rate Variance in Distance
Sampling. Biometrics, 65(1), 225–236. https://doi.org/10.1111/j.15410420.2008.01018.x
Finlayson et al. (n.d.). Data Archives for Puget Sound. Retrieved from
https://www.ocean.washington.edu/data/pugetsound/

43

Ghazoul, J. (2005). Buzziness as usual? Questioning the global pollination crisis. Trends
in Ecology & Evolution, 20(7), 367–373. https://doi.org/10.1016/j.tree.2005.04.026
Glenn, E. M., Anthony, R. G., Forsman, E. D., & Olson, G. S. (2011a). Local Weather,
Regional Climate, and Annual Survival of The Northern Spotted Owl. The Condor,
113(1), 150–158. https://doi.org/10.1525/cond.2011.100118
Glenn, E. M., Anthony, R. G., Forsman, E. D., & Olson, G. S. (2011b). Reproduction of
northern spotted owls: The role of local weather and regional climate. The Journal
of Wildlife Management, 75(6), 1279–1294. https://doi.org/10.1002/jwmg.177
Grosboll, D. (2011). Taylor’s Checkerspot (Euphydryas Editha Taylori) Oviposition
Habitat Selection and Larval Hostplant Use in Washington State (Master of
Environmental Study, Evergreen State College).
Gross, K., Kalendra, E. J., Hudgens, B. R., & Haddad, N. M. (2007). Robustness and
uncertainty in estimates of butterfly abundance from transect counts. Population
Ecology, 49(3), 191–200. https://doi.org/10.1007/s10144-007-0034-8
Grüss, A., Chagaris, D. D., Babcock, E. A., & Tarnecki, J. H. (2018). Assisting
Ecosystem-Based Fisheries Management Efforts Using a Comprehensive Survey
Database, a Large Environmental Database, and Generalized Additive Models.
Marine and Coastal Fisheries, 10(1), 40–70. https://doi.org/10.1002/mcf2.10002
Hanski, I., Ehrlich, P. R., Nieminen, M., Murphy, D., D., Hellmann, J., J., Boggs, C., L.,
& McLaughlin, J. F. (2004). Checkerspots and Conservation Biology. In On the
Wings of Checkerspots: A Model System for Population Biology (pp. 264–287).
Oxford : New York: Oxford University Press.

44

Hastie, T., & Tibsirani, R. (1987). General Additive Models: Some Applications. Journal
of the American Statistical Association, 82(398), 371–386.
Hellmann, J., J. (2005). Distribution, abundance, and adaptation of butterflies at their
northern range limit 2005 Research Summary Report-- Washington DFW (p. 7)
[Research Summary Report]. Washington State: Washington State Department of
Fish and Wildlife.
Hetrick, W. ., Rich, P. ., Barnes, F. ., & Weiss, S. B. (n.d.). GIS-Based Solar Radiation
Flux Models American Society for Photogrammetry and Remote Sensing Technical
Papers. GIS. Photogrammetry and Modeling, 3, 132–143.
Husby, J. F. (2012). Pollinators May Not Limit Native Seed Viability for Puget Lowland
Prairie Restoration (Master of Environmental Study, Evergreen State College).
James, D. G., Nunnallee, D., & Pyle, R. M. (2011). Life histories of Cascadia butterflies.
Corvallis, Or: Oregon State University Press.
Jarvis, B. (2018, November 27). The Insect Apocalypse is Here. The New York Times, p.
41. Retrieved from https://www.nytimes.com/2018/11/27/magazine/insectapocalypse.html
Koh, L. P., Sodhi, N. S., Hugh Tiang Wah Tan, & Kelvin S.-H. Peh. (2002). Factors
Affecting the Distribution of Vascular Plants, Springtails, Butterflies and Birds on
Small Tropical Islands. Journal of Biogeography, 29(1), 93–108.
Konvicka, M., Hula, V., & Fric, Z. (2003). Habitat of pre-hibernating larvae of the
endangered butterfly Eupydryas aurinia (Lepidoptera: Nymphailidae): What can be
learned from vegetation composition and architecture? European Journal of
Entomology, 100, 313–322.
45

Kuefler, D., & Haddad, N. M. (2006). Local versus landscape determinants of butterfly
movement behaviors. Echography, 29(4), 549–560.
Lacy, R. C., Williams, R., Ashe, E., Iii, K. C. B., Brent, L. J. N., Clark, C. W., … Paquet,
P. C. (2017). Evaluating anthropogenic threats to endangered killer whales to inform
effective recovery plans. Scientific Reports, 7(1), 14119.
https://doi.org/10.1038/s41598-017-14471-0
Linders, M. J. (2013, February 25). WeatherDataDocumentation_25Feb2013.
Linders, M. J., & Olson, G. S. (2014, January). Taylor’s checkerspot monitoring South Puget
Sound, Washington, methodology of the Washington State Department of Fish and
Wildlife. Unpublished.
Linders, M., Lewis, K., & Curry, K. (2016). Taylor’s checkerspot (Euphydryas editha taylori)
Captive Rearing and Reintroduction: South Puget Sound, Washington, 2016-2017 (p. 49)
[2017 Final Annual Report to US Fish and Wildlife Service, Joint Base Lewis-McChord,
ACUB Technical Review Committee]. Washington State Department of Fish and
Wildlife, Oregon Zoo, and The Evergreen State College.
Linders, M., Lewis, K., & Dorman, S. (2016). Taylor’s checkerspot (Euphydryas editha
taylori) Captive Rearing and Reintroduction: South Puget Sound, Washington, 20152016 (p. 61) [2016 Annual Report to US Fish and Wildlife Service, Joint Base LewisMcChord, ACUB Technical Review Committee]. Washington State Department of Fish
and Wildlife, Oregon Zoo, and The Evergreen State College.
Liu, H., Feng, C.-L., Chen, B.-S., Wang, Z.-S., Xie, X.-Q., Deng, Z.-H., … Luo, Y.-B. (2012).
Overcoming extreme weather challenges: Successful but variable assisted colonization of

46

wild orchids in southwestern China. Biological Conservation, 150(1), 68–75.
https://doi.org/10.1016/j.biocon.2012.02.018
Marques, T. A., Thomas, L., Fancy, S. G., & Buckland, S. T. (2007). Improving Estimates of
Bird Density using Multiple-Covariate Distance Sampling. The Auk, 124(4), 1229–1243.
McInnis, D. (1997). Go north, young checkerspot. Earth (1056148X), 6(1), 16.
McIntyer, J. (2010). Habitat Variables, Mammal Interactions and Recovery Approaches
Important to a Rare, New Mexican Butterfly, Euphydrias anicia cloudcrofti
(Lepidoptera: Nyphalidae) (Doctor of Philosophy Biology). University of New
Mexico, Albuquerque, New Mexico.
Miller-Rushing, A. ., & Primack, R. (2008). Global warming and flowering times in
Thoreau’s Concord. Ecology, 89, 332–341.
Morrison, M. L., Smallwood, S., & Hall, L. S. (2003). Creating Habitat through Plant
Relocation. Ecological Restoration, 21(2), 95–100.
Murphy, D. D., & Weiss, S. B. (1988). A Long-Term Monitoring Plan for a Threatened
Butterfly. Conservation Biology, 2(4), 367–374.
Olson, G. S. (2011, Spring). Distance Sampling Field Techniques.
Olson, G. S. (2017a). DISTANCE Sampling Training 2017. PowerPoint presented at the
WDFW. WDFW.
Olson, G. S. (2017b). Modeling the Effects of Weather on Taylor’s Checkerspot
Abundance. Population Estimate, 10. Oregon Zoo: Washington State Department of
Fish and Wildlife.
Ovaskainen, O., Skorokhodava, S., Yakovleva, M., Sukhov, A., Kutenkov, A.,
Kutenkova, N., … Delgado, M. del M. (2013). Community-level phenological
47

response to climate change. Proceedings of the National Academy of Sciences of the
United States of America, 110(33), 13434–13439.
Parmesan, C., Williams-Anderson, A., Moskwik, M., Mikheyev, A. S., & Singer, M. C.
(2015). Endangered Quino checkerspot butterfly and climate change: Short-term
success but long-term vulnerability? Journal of Insect Conservation, 19(2), 185–
204. https://doi.org/10.1007/s10841-014-9743-4
Parsons, P. A. (1995). Evolutionary Response to Drought Stress: Conservation Implications.
Biological Conservation, 74, 21–27.
Penuelas, J., Rutishauser, T., & Filella, I. (2009). Phenology Feedbacks on Climate Change.
Science, 324(5929), 887–888. https://doi.org/10.1126/science.1173004
Pietras, M., & Netzel, P. (2012). The method of assessment of solar potential for selected area
with use Geographical Information Systems. EPJ Web of Conferences, 33, 1010.
https://doi.org/10.1051/epjconf/20123301010
Pollard, E. (1988). Temperature, Rainfall and Butterfly Numbers. Journal of Applied Ecology,
25(3), 819–828. https://doi.org/10.2307/2403748
Prenger, J. J., & Ling, P., P. (n.d.). Greenhouse Condensation Control: Understanding and
Using Vapor Pressure Deficit (VPD) AEX-804-01. Retrieved from
http://www.ecaa.ntu.edu.tw/weifang/classcea/Greenhouse%20Condensation%20Control%20VPD,%20AEX-804-01.htm
Pyle, R. M. (2002). The butterflies of Cascadia: a field guide to all the species of Washington,
Oregon, and surrounding territories. Seattle: Seattle Audubon Society.
Raloff, J. (1996). Butterfly Displaced by Climate Change? Science News, 150(9), 135–135.

48

Rausher, M. (1982). Population Differentiation in Euphydryas Editha Butterflies: Larval
Adaptation to Different Hosts. Evolution, 36(3), 581–590.
Rusty Patched Bumble Bee [US Fish and Wildlife Service]. (n.d.). Retrieved from
https://www.fws.gov/midwest/endangered/insects/rpbb/
Schoonmaker, P. K., & Foster, D. R. (1991). Some Implications of Paleoecology for
Contemporary Ecology. Botanical Review, 57(3), 204–245.
Severns, P. M., & Breed, G. A. (2014). Behavioral consequences of exotic host plant
adoption and the differing roles of male harassment on female movement in two
checkerspot butterflies. Behavioral Ecology and Sociobiology, 68(5), 805–814.
https://doi.org/10.1007/s00265-014-1693-z
Singer, M. C., & Lee, J., R. (2000). Discrimination within and between host species by a
butterfly: implications for design of preference experiments (No. 3; pp. 101–105).
Section of Integrative Biology, University of Texas, Austin Texas 78712, USA.
Singer, M. C., & Parmesan, C. (2010). Phenological asynchrony between herbivorous
insects and their hosts: signal of climate change or pre-existing adaptive strategy?
Philosophical Transactions: Biological Sciences, 365(1555), 3161–3176.
Singer, M. C., & Parmesan, C. (2010). Phenological asynchrony between herbivorous
insects and their hosts: signal of climate change or pre-existing adaptive strategy?
Philosophical Transactions of the Royal Society B: Biological Sciences, 365(1555),
3161–3176. https://doi.org/10.1098/rstb.2010.0144
Singer, M. C., & Wee, B. (2005). Spatial pattern in checkerspot butterfly-host plant
association at local, metapopulation and regional scales. Ann. Zool. Fennici, 42,
347–361.
49

Sugarbaker, D. (2017, November). GIS day 2017 Keynote. Keynote presented at the GIS
day, DES Building. Retrieved from http://www.wsdot.wa.gov/mapsdata/gisday.htm
Szymanski, J., Smith, T., Horton, A., Parkin, M., Ragan, L., Masson, G., … Hill, L.
(2016). Rusty Patched Bumble Bee (Bombus affinis) Species Status Assessment
Final Report Version 1 (p. 100). Retrieved from U.S. Fish and Wildlife Service
website:
https://www.fws.gov/midwest/endangered/insects/rpbb/pdf/SSAReportRPBBwAdd.
pdf
Thomas, J. A., Simcox, D. J., & Hovestadt, T. (2011). Evidence based conservation of
butterflies. Journal of Insect Conservation, 15(1–2), 241–258.
https://doi.org/10.1007/s10841-010-9341-z
Thomas, L., Buckland, S. T., Rexstad, E. A., Laake, J. L., Strindberg, S., Hedley, S. L., …
Burnham, K. P. (2010). Distance software: design and analysis of distance sampling
surveys for estimating population size. Journal of Applied Ecology, 47(1), 5–14.
https://doi.org/10.1111/j.1365-2664.2009.01737.x
Trilobites and Cambrian Utah – Utah Geological Survey. (n.d.). Retrieved November 5, 2017,
from https://geology.utah.gov/map-pub/survey-notes/glad-you-asked/trilobites-andcambrian-utah/
Wallis DeVries, M. F., Baxter, W., & Van Vliet, A. J. H. (2011). Beyond climate envelopes:
effects of weather on regional population trends in butterflies. Oecologia, 167(2), 559–
571. https://doi.org/10.1007/s00442-011-2007-z

50

Weiss, S. B., Murphy, D. D., Ehrlich, P. R., & Metzler, C. F. (1993). Adult emergence
phenology in checkerspot butterflies: the effects of macroclimate, topoclimate, and
population history. Oecologia, 96, 261–270.
Weiss, S. B., Murphy, D. D., & White, R. (1988). Sun, Slope, and Butterflies: Topographic
Determinants of Habitat Quality For Euphydryas Editha. Ecology, 69(5), 1486–1496.
Weiss, S. B., & Weiss, A. D. (1998). Landscape-Level Phenology of a Threatened
Butterfly: A GIS-Based Modeling Approach. Ecosystems, 1(3), 299–309.
White, R., & Singer, M. C. (1987). Marking Technique for Larvae. Pan-Pacific
Entomologist, 63(4), 341–345.

51

APPENDIX A
Final Data used for the Linear Regressions:
Butterfly abundance data:
•

PeakGamPoint: Each year butterfly counts were recorded and
analyzed to get density estimates, which was then used to calculate
the total abundance of butterflies in each study site, R76, R50, and
Scatter Creek South. The count data was fitted to a model. The
model was then used to generate an approximation of each day’s
abundance over time for that year, and the maximum
approximation from that model was then used for the regression
analyses. Site R76 had too few data abundance points to model a
peak GAM point for the year 2008.

Weather data: The winter’s precipitation data was downloaded from the point
source data available from the PRISM website maintained and managed by the
University of Washington’s School of Oceanography. Furthermore, only
precipitation, humidity, and temperature will be used as the independent variables
for the analysis.
•

Precip.mm: This represents the sum total of all the precipitation for
each year’s winter values.

•

Temperature Degrees Fahrenheit: This field represents each year’s
average winter temperature.

52

•

Average Max Vapor Pressure Deficit (VPD): This field represents
each year’s average winter humidity, or in the case of vapor
pressure deficit it should be called, dryness.

R76 (excludes year 2008 due to lack of data)
Temperature
Degrees
Year PeakGamPoint Precip.mm Fahrenheit
2006
9825.97
518.41
2007
5736.80
472.44
2009
197.81
320.04
2010
1420.88
300.48
2011
5658.57
363.98
2012
10681.72
311.40
2013
4055.43
315.21
2014
2918.96
304.55
2015
3567.57
327.66
2016
1401.09
621.28
2017
4153.43
383.29
2018
14040.73
398.78

39.57
39.57
37.77
34.40
39.63
38.90
39.23
38.33
43.40
42.03
36.90
39.93

Average Max
Vapor Pressure
Deficit (VPD)
3.10
3.22
3.10
3.89
3.39
3.28
2.62
2.58
3.20
2.75
2.65
1.99

R50
Year PeakGamPoint Precip.mm
2010
135.54
269.75
2011
856.91
339.60
2012
1484.32
282.45
2013
987.59
310.64
2014
924.43
292.35
2015
1710.68
321.56
2016
2520.72
610.87
2017
4382.89
387.86
2018
8948.88
404.37

Average Max
Temperature
Vapor Pressure
Degrees Fahrenheit Deficit (VPD)
41.23
3.98
39.93
3.40
39.23
3.34
39.60
2.66
38.60
2.73
43.73
3.44
42.30
3.04
37.10
2.92
40.17
2.38

53

SCS
Year PeakGamPoint Precip.mm
2009
24.90
422.66
2010
80.17
385.06
2011
29.33
450.60
2012
26.96
409.96
2013
157.20
450.09
2014
126.44
352.55
2015
235.38
452.63
2016
923.91
721.36
2017
2151.40
418.34
2018
4123.85
494.03

Temperature
Average Max
Degrees
Vapor Pressure
Fahrenheit
Deficit (VPD)
45.30
3.21
41.23
3.80
39.73
3.09
39.03
3.23
39.53
2.46
36.30
2.71
43.77
3.56
42.33
2.95
36.97
2.89
40.07
2.59

GAM Plots
The abundance calculations are derived from the density calculations that had
been computed from the “Distance” software by the data scientists at WDFW were used
for the general additive models (GAM) to estimate the peak abundance values for each
study year.

The red lines in the plots below represent estimations of the abundance at each
site throughout each year derived from a general additive model (GAM). The time (julian
date) is the date measured in days for each year. For instance, day 1(julian date =1) is
January first. A GAM could not be modeled for the year 2008 in site R76 due to a lack of
abundance data.
Site R50
R50-2010

54

55

R50-2011

R50-2012

56

R50-2013

R50-2014

57

R50-2015

R50-2016

58

R50-2017

R50-2018

59

Site R76
R76-2006

R76-2007

60

R76-2008: This year had only three data points; not enough for a GAM.
R76-2009

R76: 2010

61

R76-2011

R76-2012

62

R76-2013

R76-2014

63

R76-2015

R76-2016

64

R76-2017

R76-2018

65

Scatter Creek South site
SCS-2009

SCS-2010

66

SCS-2011

SCS-2012

67

SCS-2013

SCS-2014

68

SCS-2015

SCS-2016

69

SCS-2017

SCS-2018

70

APPENDIX B
Results of raw data without transformations
Three climatic variables were tested against the peak GAM values for the three study
sites: precipitation in millimeters, temperature in Fahrenheit, and humidity in average
maximum vapor pressure deficit (VPD), which is a measurement of dryness (Prenger &
Ling, n.d.). The results of the simple and multiple linear regression models in all three
sites show the need for further weather modeling due to the low correlations. An
examination of all the plots of the residuals for the data indicated that the model of choice
is a linear model.

The null hypothesis was H0: “The winter weather element does not affect the populations
of butterflies.” The simple and multiple linear regression model proved the null
hypotheses to be true. However, there was a higher correlation of peak abundance
estimates with respect to humidity.

R76 data analysis:
Because the year 2008 had only three observations, there were not enough data points to
model a peak GAM abundance value. Because of this, the year 2008 was left out of the
analysis for site R76.

71

Precipitation
The analysis used linear regression models where the maximum value of the fitted
general additive model (GAM) of the abundance estimates was the dependent variable
and the sum of the total winter precipitation in millimeters was the independent variable.

Plot of total winter precipitation plotted against abundance at site R76

The correlation coefficient between precipitation and peak GAM values of the abundance
data is 0.10. Because of this, no simple linear regression is needed

Temperature

72

The correlation coefficient of temperature with respect to abundance is r=0.18, too low to
run a simple linear model.

Winter average temperature(˚F) plotted against peak GAM butterfly abundance values from site R76

Humidity
Winter dryness may have a deleterious effect on the populations of butterflies, but not
enough to reject the null hypothesis, which was, H0: “The populations of butterflies in the
site, R76 are not affected by humidity”.

There is a sufficient correlation between vapor pressure deficit and peak GAM abundance
values to run a linear regression model. The correlation coefficient is -0.36, showing that
dryness and populations of butterflies may be negatively correlated.
73

Winter average of max dryness measured in vapor pressure deficit (Vpd) plotted against the peak GAM points for
butterfly abundance at R76

The simple linear regression showed the coefficient of correlation, the adjusted R2 to be
0.04. The F-statistic (with 1 and 10 d.f.) was 1.47 with a p-value of 0.254. These values
show an effect, but not enough to determine that humidity alone is affecting the butterfly
populations. Higher winter humidity is associated with greater spring abundance of
butterflies in R76.

Multivariate analysis in R76
Before running the multiple linear regression on the peak GAM estimates for site R76
and the three climatic variables, a correlation model was run to avoid including the
independent variables that are highly correlated. In R76, the climatic variables
74

precipitation and temperature had a correlation coefficient of 0.42. The multiple linear
regression model did not include these two variables together. Due to low number of data
points for abundance in the year 2008, that year was not included.

A run of the multiple linear regression model of the peak GAM abundance points with
respect to precipitation and humidity returned an adjusted R2 value of -.06. The F-statistic
(with 2 and 9 d.f.) was 0.67 with a p-value of 0.536. This model is not a good fit.

A second model was run looking at the values of temperature and humidity with respect
to the peak GAM values. This model returned an adjusted R2 value of -0.06. The Fstatistic (with 2 and 9 d.f.) was 0.70 with a p-value of 0.523. This model is not a good fit
either.

R50 data analysis
Precipitation
The correlation coefficient between precipitation and peak GAM values is 0.38. This
small amount of correlation indicated the need for a simple linear regression model. The
value for the adjusted R2 is 0.02. The F statistic in this regression model (with 1 and 7
d.f.) is 1.2 and the p-value is 0.310. There may be an effect here, but the p-value is too
large to reject the null hypothesis, further indicating that there may be other weather or
confounding variables affecting the butterfly abundance in this site.

75

Total winter precipitation plotted against peak GAM values at site R50

Temperature
The correlation between the peak GAM abundance values and temperature was negative
showing r=-0.15. There is not much correlation, so no linear model was needed for
abundance values with respect to average winter temperature at this site.

76

Plot of peak GAM abundance values against the average winter temperature for site R50

Humidity
There was a negative correlation between the peak GAM points for abundance in this site
and the humidity values measured in vapor pressure deficit. The correlation coefficient is
r= -0.64, indicating a need for a linear model to be run. The linear model returned an
adjusted R2=0.32. The F-statistic (with 1 and 7 d.f.) was 4.8, with a p-value of 0.065.
Higher winter humidity is somewhat associated with greater abundance of butterflies in
R50 during the following spring.

77

Plot of winter average max vapor pressure deficit (Vpd) plotted against the peak GAM abundance values at site R50

Multivariate analysis at R50
A correlation analysis of all the weather variables for R50 showed the two variables,
humidity and precipitation to be somewhat correlated, where r=.30. Additionally,
humidity and temperature were also correlated at r= 0.40. Because of these correlations,
humidity was not included in any of the models.

Upon looking at the multiple linear regression model of temperature and precipitation and
how it affects the peak GAM points, an adjusted R2 value was returned as -0.5. The Fstatistic (with 2 and 6 d.f.) was 0.80 with a p-value of 0.493. This model is not a good fit.

78

Scatter Creek South
Precipitation
There is not much effect that the sum of total winter precipitation is having on the
abundance numbers of butterflies at Scatter Creek South. The correlation coefficient is r=
0.24. There seems to be some correlation, indicating a need for a linear regression model.

The adjusted R2 is -.06. The F statistic (with 1 and 8 d.f.) in this regression is 0.51 and
the p-value is 0.496. This indicates that winter precipitation does not affect the
populations of butterflies. There may be an effect here, however the p-value is too large
to reject the null hypothesis, further indicating that there may be other weather or
confounding variables.

Plot of peak GAM abundance values against total winter precipitation values at Scatter Creek South

79

Temperature
The return of a low correlation coefficient of r=-.20 indicates no need to run a linear
regression model of temperature as it affects the values of peak GAM abundance values
at Scatter Creek South.

Plot of average winter temperature against peak GAM values of abundance at the Scatter Creek South study site

Humidity
Though not as strong a correlation as site R50, there is still a negative correlation
between the measure of vapor pressure deficit and the peak GAM values for butterfly
abundance at this site. The correlation coefficient, r = -0.44, is a large enough correlation
to run a linear regression model. The model returns an adjusted R2 of 0.1. There is an Fstatistic (with 1 and 8 d.f.) of 2.0 with a p-value of .198. This shows a pretty strong effect,
but not enough to reject the null hypothesis.
80

Plot of the average max winter vapor pressure deficit (Vpd) against the peak GAM abundance values of butterflies at
Scatter Creek South

Multivariate analysis at SCS
The correlation analysis of the independent variables (winter climatic variables) showed
some correlation between humidity and temperature (r=0.49), as well as correlation
between precipitation and temperature (r=0.35). For this reason, the only reasonable
multiple linear regression model is that of precipitation and humidity that has a
correlation value of r= (-0.18).

This multiple linear regression model returned an adjusted R2 of -0.005. The F-statistic
(with 2 and 7 d.f.) is 1.0 with a p-value of 0.408. This is not a good model fit.

81

Plots of the residuals:
All the plots of the residuals performed on the data show the appropriateness of
the linear model as a model of choice for this data analysis.
The statistics program, R has the function of displaying the plots of the residuals. These
plots help measure how far the data points are off the regression line, or in other words,
how much the regression line misses the data point.
Plots of the residuals explained in detail:
The most common plot is the Residuals vs. Fitted plot. The residuals are plotted
against the fitted value. Lots of scatter and no patterns indicate that the linear models are
appropriate for this data.
The Normal Q-Q plot tells us from the straight line that the errors in this data set
are normally distributed. This data shows that the linear model is appropriate and we do
need to use a non-linear model for the regression.
The Residuals vs Leverage plot lets us know if the variance of all the residuals is
constant. Leverage is the measure of the influence of the points. The higher the influence,
the farther from the mean value (the regression line). If the variance (value of the distance
each point is from the regression line) get larger as the dependent variable
(PeakGamPoint) increases, it will show that our assumption for using the linear
regression that the variance of these residuals are constant is actually false and a different
model other than linear regression should be used.
The Scale-Location plot is a measure of what the predicted values are plotted
against the square root of the residuals for the linear model of the data. We would have to

82

consider a different statistical model if the data showed a pattern of residual increase as
the predicted values increase.
R76 Precipitation Data:
Low correlation for R76 precipitation data and peak GAM points indicated no
linear model was needed.

83

R76 Humidity Data

84

85

R76 Temperature Data:
R76 Temperature data was not modeled due to low correlation

86

R50 Precipitation Data:

87

88

R50 Temperature Data:
R50 Temperature data was not modeled due to low correlation
89

90

R50 Humidity Data:

91

92

93

Scatter Creek South Precipitation Data:

94

95

96

Scatter Creek South Temperature Data:
Scatter Creek South did not get modeled for temperature due to low correlation

SCS Humidity Data:

97

98

Scatter Creek South Temperature was not modeled due to low correlation
99

R76 data analysis of raw data:
Because the year 2008 had only three observations, there were not enough data
points to model a peak GAM abundance value. Because of this, the year 2008 was left
out of the analysis for site R76.
Precipitation
The analysis used linear regression models where the maximum value of the fitted
general additive model (GAM) of the abundance estimates was the dependent variable
and the sum of the total winter precipitation in millimeters was the independent variable.

Plot of total winter precipitation plotted against abundance at site R76

The correlation coefficient between precipitation and peak GAM values of the abundance
data is 0.10. Because of this, no simple linear regression is needed
Temperature
100

The correlation coefficient of temperature with respect to abundance is r=0.18,
too low to run a simple linear model.

Winter average temperature (˚F) plotted against peak GAM butterfly abundance values from site R76

Humidity
Winter dryness may have a deleterious effect on the populations of butterflies, but
not enough to reject the null hypothesis, which was, H0: “The populations of butterflies in
the site, R76 are not affected by humidity”.
There is a sufficient correlation between vapor pressure deficit and peak GAM
abundance values to run a linear regression model. The correlation coefficient is -0.36,
showing that dryness and populations of butterflies may be negatively correlated.

101

Winter average of max dryness measured in vapor pressure deficit (Vpd) plotted against the peak GAM points for
butterfly abundance at R76

The simple linear regression showed the coefficient of correlation, the adjusted R2 to be
0.04. The F-statistic (with 1 and 10 d.f.) was 1.47 with a p-value of 0.254. These values
show an effect, but not enough to determine that humidity alone is affecting the butterfly
populations. Higher winter humidity is associated with greater spring abundance of
butterflies in R76.
Multivariate analysis in R76
Before running the multiple linear regression on the peak GAM estimates for site
R76 and the three climatic variables, a correlation model was run to avoid including the
independent variables that are highly correlated. In R76, the climatic variables
precipitation and temperature had a correlation coefficient of 0.42. The multiple linear
102

regression model did not include these two variables together. Due to low number of data
points for abundance in the year 2008, that year was not included.
A run of the multiple linear regression model of the peak GAM abundance points
with respect to precipitation and humidity returned an adjusted R2 value of -.06. The Fstatistic (with 2 and 9 d.f.) was 0.67 with a p-value of 0.536. This model is not a good fit.
A second model was run looking at the values of temperature and humidity with
respect to the peak GAM values. This model returned an adjusted R2 value of -0.06. The
F-statistic (with 2 and 9 d.f.) was 0.70 with a p-value of 0.523. This model is not a good
fit either.
R50 data analysis of raw data
Precipitation
The correlation coefficient between precipitation and peak GAM values is 0.38.
This small amount of correlation indicated the need for a simple linear regression model.
The value for the adjusted R2 is 0.02. The F statistic in this regression model (with 1 and
7 d.f.) is 1.2 and the p-value is 0.310. There may be an effect here, but the p-value is too
large to reject the null hypothesis, further indicating that there may be other weather or
confounding variables affecting the butterfly abundance in this site.

103

Total winter precipitation plotted against peak GAM values at site R50

Temperature
The correlation between the peak GAM abundance values and temperature was
negative showing r=-0.15. There is not much correlation, so no linear model was needed
for abundance values with respect to average winter temperature at this site.

104

Plot of peak GAM abundance values against the average winter temperature for site R50

Humidity
There was a negative correlation between the peak GAM points for abundance in
this site and the humidity values measured in vapor pressure deficit. The correlation
coefficient is r= -0.64, indicating a need for a linear model to be run. The linear model
returned an adjusted R2=0.32. The F-statistic (with 1 and 7 d.f.) was 4.8, with a p-value
of 0.065. Higher winter humidity is somewhat associated with greater abundance of
butterflies in R50 during the following spring.

105

Plot of winter average max vapor pressure deficit (Vpd) plotted against the peak GAM abundance values at site R50

Multivariate analysis at R50
A correlation analysis of all the weather variables for R50 showed the two
variables, humidity and precipitation to be somewhat correlated, where r=.30.
Additionally, humidity and temperature were also correlated at r= 0.40. Because of these
correlations, humidity was not included in any of the models.
Upon looking at the multiple linear regression model of temperature and
precipitation and how it affects the peak GAM points, an adjusted R2 value was returned
as -0.5. The F-statistic (with 2 and 6 d.f.) was 0.80 with a p-value of 0.493. This model is
not a good fit.
Scatter Creek South analysis of raw data
Precipitation
106

There is not much effect that the sum of total winter precipitation is having on the
abundance numbers of butterflies at Scatter Creek South. The correlation coefficient is r=
0.24. There seems to be some correlation, indicating a need for a linear regression model.
The adjusted R2 is -.06. The F statistic (with 1 and 8 d.f.) in this regression is 0.51
and the p-value is 0.496. This indicates that winter precipitation does not affect the
populations of butterflies. There may be an effect here, however the p-value is too large
to reject the null hypothesis, further indicating that there may be other weather or
confounding variables.

Plot of peak GAM abundance values against total winter precipitation values at Scatter Creek South

107

Temperature
The return of a low correlation coefficient of r=-.20 indicates no need to run a
linear regression model of temperature as it affects the values of peak GAM abundance
values at Scatter Creek South.

Plot of average winter temperature against peak GAM values of abundance at the Scatter Creek South study site

Humidity
Though not as strong a correlation as site R50, there is still a negative correlation
between the measure of vapor pressure deficit and the peak GAM values for butterfly
abundance at this site. The correlation coefficient, r = -0.44, is a large enough correlation
to run a linear regression model. The model returns an adjusted R2 of 0.1. There is an Fstatistic (with 1 and 8 d.f.) of 2.0 with a p-value of .198. This shows a pretty strong effect,
but not enough to reject the null hypothesis.
108

Plot of the average max winter vapor pressure deficit (Vpd) against the peak GAM abundance values of butterflies at
Scatter Creek South

Multivariate analysis at SCS
The correlation analysis of the independent variables (winter climatic variables)
showed some correlation between humidity and temperature (r=0.49), as well as
correlation between precipitation and temperature (r=0.35). For this reason, the only
reasonable multiple linear regression model is that of precipitation and humidity that has
a correlation value of r= (-0.18).
This multiple linear regression model returned an adjusted R2 of -0.005. The Fstatistic (with 2 and 7 d.f.) is 1.0 with a p-value of 0.408. This is not a good model fit.

109

Media

Thesis_MES_2019_LagerquistW.pdf