The Evergreen State College Institutional Repository

OREGON SPOTTED FROG (Rana pretiosa) ABSENCE IN THE CHEHALIS RIVER BASIN: AN ANALYSIS OF HABITAT, CLIMATE AND INVASIVE SPECIES

Item

Title

Eng OREGON SPOTTED FROG (Rana pretiosa) ABSENCE IN THE CHEHALIS RIVER BASIN: AN ANALYSIS OF HABITAT, CLIMATE AND INVASIVE SPECIES

Date

Eng 2021

Creator

Eng Murtagh, Patrick

Identifier

Eng Thesis_MES_2021_MurtaghP

extracted text

OREGON SPOTTED FROG (Rana pretiosa) ABSENCE IN THE CHEHALIS RIVER BASIN:
AN ANALYSIS OF HABITAT, CLIMATE AND INVASIVE SPECIES

by
Patrick Murtagh

A Thesis
Submitted in partial fulfillment
Of the requirements for the degree
Master of Environmental Studies
The Evergreen State College
June 2021

© 2021 by Patrick Murtagh. All rights reserved.

This Thesis for the Master of Environmental Studies Degree
by
Patrick Murtagh

has been approved for
The Evergreen State College
by

_______________________________
Erin Martin, Ph.D.
Member of Faculty

_______________________________
Date

ABSTRACT
Oregon Spotted Frog (Rana pretiosa) Absence in the Chehalis River Basin: An Analysis of
Habitat, Climate, and Invasive Species
Patrick Murtagh
The Oregon spotted frog (Rana pretiosa) is a federally threatened and Washington state
endangered species now extirpated from close to 80% of its historical range (from British
Columbia to Northern California). This loss is attributed to habitat loss, invasive species, disease
and climate change. In Washington state, there are six drainages currently known to have
populations of frogs and only one in the South Puget Sound: the Black River. The Black River is
a tributary of the Chehalis River, which despite being aquatically linked to the Black River, has
no records of Oregon spotted frog occurrence. One of the recovery objectives for the Oregon
spotted frog is to locate new populations, potentially in the Chehalis River Basin. Prior to this
study, amphibian surveys had been conducted in the Chehalis River mainstem floodplain, all of
which failed to detect Oregon spotted frog.
To determine the extent of suitable habitat in the Chehalis River Basin, habitat was
modelled using the Maxent species distribution model and the environmental variables and
presence of invasive species (centrarchid fishes and American Bullfrog) were compared between
the watersheds. The Maxent model located suitable habitat throughout the Chehalis Basin by
finding locations with similar habitat to the Black River watershed. Surveyed locations in the
Chehalis River floodplain, that were deemed as suitable habitat, were compared to occupied
locations in the Black River by their habitat structure, climate and abundance of invasive species.
According to this comparison, the habitat structure of the Chehalis River is not a limiting factor
for Oregon spotted frog. The climate and abundance of invasive species were different between
the rivers, yet there is stronger evidence to suggest that the presence of American Bullfrogs and
centrarchid fishes (invasives) are having a greater effect towards the presumed absence of
Oregon spotted frog in the Chehalis River Basin outside of the Black River.

TABLE OF CONTENTS

LIST OF FIGURES ...................................................................................................................................... v
LIST OF TABLES ....................................................................................................................................... vi
LIST OF APPENDICES ............................................................................................................................. vii
ACKNOWLEDMENTS ............................................................................................................................ viii
INTRODUCTION ........................................................................................................................................ 1
LITERATURE REVIEW ............................................................................................................................. 4
The Oregon Spotted Frog ......................................................................................................................... 4
Life History .......................................................................................................................................... 6
Threats to the Oregon Spotted Frog ................................................................................................... 10
Oregon Spotted Frog in the Chehalis River Basin ............................................................................. 16
Habitat Modelling ................................................................................................................................... 18
Species Distribution Models .............................................................................................................. 18
Maxent................................................................................................................................................ 21
METHODS ................................................................................................................................................. 25
Maxent .................................................................................................................................................... 25
Presence Point Selection .................................................................................................................... 26
Background Point Selection ............................................................................................................... 29
Environmental Variables .................................................................................................................... 30
Modelling ........................................................................................................................................... 39
Model Evaluation ............................................................................................................................... 39
Watershed Comparison........................................................................................................................... 41
Chehalis Survey Point Selection and Comparison Criteria ................................................................ 41
Statistical Analysis ............................................................................................................................. 44
RESULTS ................................................................................................................................................... 46
Maxent .................................................................................................................................................... 46
Model Evaluation ............................................................................................................................... 46
Predicted Distribution ........................................................................................................................ 46
Environmental Variable Importance .................................................................................................. 50
Watershed Comparison........................................................................................................................... 58
DISCUSSION ............................................................................................................................................. 64
LITERATURE CITED ............................................................................................................................... 76
APPENDIX ................................................................................................................................................. 83

iv

LIST OF FIGURES
Figure 1. The watersheds in Washington state currently known to have populations of Oregon
spotted frog. ............................................................................................................................. 5
Figure 2. Presence points in the Black River watershed as a result of being spatially rarefied at
500 meters. ............................................................................................................................. 29
Figure 3. Location of Chehalis survey points and Black River presence points used in the
comparison analysis. Survey points were selected based on their predicted probability of
presence according to the Maxent model (> 0.7). ................................................................. 43
Figure 4. Suitable breeding habitat for the Oregon spotted frog in the Black River watershed
according to the Maxent model. The warmer colors represent an increased probability of
presence based on the estimate habitat suitability. ................................................................ 48
Figure 5. Suitable breeding habitat for the Oregon spotted frog in the Upper Chehalis River
Basin. The warmer colors represent an increased probability of presence based on the
estimated habitat suitability. .................................................................................................. 49
Figure 6. Suitable breeding habitat for the Oregon spotted frog in the Lower Chehalis River
Basin. The warmer colors represent an increased probability of presence based on the
estimate habitat suitability. .................................................................................................... 50
Figure 7. Jackknife analysis of the Maxent variables. The light blue bars measure the training
gain of the model when that variable is excluded. The dark blue bars measure the training
gain when that variable is run in isolation. ............................................................................ 53
Figure 8. Response curves plotting the probability of presence based on the variation in the
structural variables used in the Maxent model. The variables include (a) distance to cover
class, (b) landcover, (c) slope, (d) hydric soil status, (e) herbaceous vegetation height. ...... 55
Figure 9. The response curves plotting the probability of presence based on the variation in the
climate variables used in the Maxent model. Variables include (a) annual mean temperature,
(b) mean temperature coldest quarter, (c) precipitation driest quarter, (d) precipitation
wettest month, (e) temperature annual range. ........................................................................ 57
Figure 10. The frequency of abundance ranks of Centrarchid fishes at the Black River presence
points and Chehalis River survey points. .............................................................................. 62
Figure 11. The frequency of abundance ranks of bullfrogs in the Black River presence points
and Chehalis River survey points. ......................................................................................... 63

v

LIST OF TABLES
Table 1. Structure variables used in the Maxent model obtained from publicly available sources
and their original resolution. .................................................................................................. 32
Table 2. Climate variables used in the Maxent model obtained from Worldclim and their original
resolution. .............................................................................................................................. 35
Table 3. The percent contribution and permutation importance of all 21 climate variables when
run in a Maxent model. .......................................................................................................... 36
Table 4. Final structure and climate variables representing Oregon spotted frog breeding habitat
used in the Maxent model. ..................................................................................................... 38
Table 5. The success rate, omission rate, and associated p-values of the Maxent model
developed for the Chehalis Basin. Success was evaluated at two thresholds, the minimum
training presence (MTP) and the 10-percentile training presence (10TP). ........................... 46
Table 6. The permutation importance and percent contribution importance of the environmental
variables in the Maxent model. .............................................................................................. 52
Table 7. Results of the comparison analysis of environmental and exotic species variables
between the Black River presence points and the Chehalis River floodplain survey points.
Significant differences are in bold (p < 0.001). ..................................................................... 59
Table 8. Summary of environmental and exotic species variables used in the comparison
analysis between the Black River presence points and the Chehalis River floodplain survey
points. Statistical differences are in bold. .............................................................................. 60
Table 9. The magnitude of change between statistically different variables between the Back
River presence points (B) and the Chehalis River floodplain survey points (C). .................. 61
Table 10. Contingency tables for (a) landcover and (b) hydric soils, the categorical variables
used in the Maxent model. ..................................................................................................... 61

vi

LIST OF APPENDICES
Appendix 1. Predicted suitable habitat for the Oregon spotted frog in the Chehalis Basin
according to the structure variables. ...................................................................................... 83
Appendix 2. Predicted suitable habitat for the Oregon spotted frog in the Chehalis Basin
according to the climate variables. ........................................................................................ 84
Appendix 3. The success rate, omission rate, and associated p-values of the structure-only and
climate-only Maxent models developed for the Chehalis Basin. Success was evaluated at
two thresholds, the minimum training presence (MTP) and the 10-percentile training
presence (10TP). .................................................................................................................... 84
Appendix 4. The percent contribution and permutation importance of the variables in the
structure-only model. ............................................................................................................. 84
Appendix 5. The percent contribution and permutation importance of the variables in the
climate-only model. ............................................................................................................... 85
Appendix 6. Survey points selected according to the structure-only and climate-only Maxent
outputs. Points were selected according to their probability of presence in the Maxent
outputs (> 0.7)........................................................................................................................ 85
Appendix 7. The comparison results of the environmental variables and abundance of exotic
species between the Chehalis River floodplain survey points selected from the structure-only
or climate-only models and the Black River presence points................................................ 86
Appendix 8. Summary statistics for the environmental variables used in the Maxent models and
the abundance of exotic species for the presence points in the Black River and the survey
points selected from the structure-only model and the climate-only model. ......................... 87
Appendix 9. Contingency tables for (a) landcover and (b) hydric soil, the categorical variables
used in the structure-only model............................................................................................ 87
Appendix 10. The magnitude of change between statistically different variables in the structureonly and climate-only models. ............................................................................................... 88
Appendix 11. The abundance ranks of centrarchid fishes at the Black River presence points and
the Chehalis River floodplain survey points selected in the structure-only model and the
climate-only model. ............................................................................................................... 88
Appendix 12. The abundance ranks of bullfrogs at the Black River presence points and The
Chehalis River floodplain survey points selected in the structure-only model and the
climate-only model. ............................................................................................................... 89
Appendix 13. Selected climate variables for the Maxent model. ................................................ 90

vii

ACKNOWLEDMENTS
There were a number of people who made this study possible. I would first like to thank Lisa
Hallock and Julie Tyson of the Washington Department of Fish and Wildlife for providing me
with the data necessary to carry out this research. For training and assistance in GIS, I would like
to thank Mike Ruth and Tyler Cowdrey. Thank you to Marc Hayes, for all of the hours spent
discussing frog habitat and how best to model it. Also, thank you to my reader, Erin Martin for
helping me refine every part of this thesis. Your insights and guidance made this project all the
better. Thank you to the rest of my MES cohort, it was great to be able to commiserate and bond
over this achievement. Finally, I want to thank my family, especially my wife, Jenny. Thank you
for all of the patience, support, and understanding you provided as we navigated a newborn, a
pandemic and a thesis.

viii

INTRODUCTION
The Oregon spotted frog (Rana pretiosa) is a ranid frog endemic to the Pacific
Northwest. Historically, they occupied territory from British Columbia, to Northern California,
however, this range has been reduced by up to 80% (Adams et al., 2014; Hayes, 1997). This
large reduction in range has led to a threatened listing under the Endangered Species Act and an
endangered listing in Washington state. They are wetland specialists with a complex life cycle;
occupying multiple aquatic habitats including seasonally flooded shallows for oviposition,
permanent water during the dry season, and springs, beaver dams or flowing water for overwintering (Pearl & Hayes, 2004). These habitats and the frog are at risk from altered hydrology,
invasive species, disease and climate change (Hallock, 2013; Holgerson et al., 2019; Watson et
al., 2003).
Of particular concern, are the challenges imposed on the Oregon spotted frog by the
introduction of invasive species and the alteration of the natural hydrology of many of
Washington’s wetlands. When waterways are channelized to reduce flooding or convert wetlands
to agriculture, they become static and the seasonal hydrological fluctuations necessary for the
Oregon spotted frog are removed (McAllister & Leonard, 1997). The invasive species of concern
for the Oregon spotted frog include aquatic fauna, specifically the American bullfrog (Lithobates
catesbeianus) and centrarchid fishes (Holgerson et al., 2019). Centrarchids are warmwater fish of
the sunfish family introduced into many of Washington’s rivers for sport fishing (Hallock, 2013).
Invasive species increase predation pressure, are able to outcompete for resources and create
barriers to migration, thus isolating populations (Bradford & Tabatabai, 1993; Pearl & Hayes,
2004). Many native amphibians in the Pacific Northwest are vulnerable to these species but
especially the Oregon spotted frog. Due to their aquatic nature and inability to escape into
1

terrestrial habitat like other native amphibians, frogs are forced to share a larger proportion of
their habitat and thus increase their interactions with these invasive species (Holgerson et al.,
2019; Rowe et al., 2021).
In Washington state, there are six drainages that are currently known to have populations
of Oregon spotted frog, of which, the Black River is the only watershed in the South Puget
Sound (Hallock, 2013). The Black River is a tributary of the Chehalis River, and due the aquatic
connectivity and size of the larger Chehalis Basin, frogs may have occurred in other reaches of
this watershed. However, there has never been a recorded occurrence of Oregon spotted frog
elsewhere in the Chehalis Basin (M. Hayes, personal communication). The Black River may be
unique among the Chehalis tributaries as it has one of the largest intact freshwater wetland
systems remaining in the Puget Sound region, making it ideal habitat for the Oregon spotted frog
(Species Restoration Plan Steering Committee, 2019).
Successful conservation efforts of rare or sensitive species depend on knowledge of the
habitat requirements and the current distribution of the species of interest (Bohannon et al.,
2016). As such, one of the Oregon spotted frog recovery objectives proposed by the Washington
Department of Fish and Wildlife is to locate new populations for this reason (Hallock, 2013).
Despite the lack of occurrences, there is potential for discovery of previously undocumented
population in Chehalis Basin due to the connection to the Black River and the wide array of
habitat found in such a large watershed.
In this study, suitable Oregon spotted frog breeding habitat was modeled in the Chehalis
Basin using the species distribution model, Maxent. The Maxent model used the habitat
conditions of locations in the Black River with known Oregon spotted frog presence and
modelled areas of similar conditions elsewhere in the Chehalis Basin. Suitable habitat was

2

predicted in the Chehalis River floodplain itself, as well as among many of its tributaries.
Further, using the habitat model as a guide, prior surveyed locations in the Chehalis River
floodplain that were deemed as suitable habitat were compared to known breeding locations in
the Black River. These locations were compared by their habitat structure, climate, and
abundance of invasive bullfrogs and centrarchids. The purpose of this comparison was to detect
any difference between the two rivers to better explain the presumed absence of Oregon spotted
frog outside of the Black River.

3

LITERATURE REVIEW
The Oregon Spotted Frog
The Oregon spotted frog (Rana pretiosa) is a ranid frog endemic to the Pacific
Northwest. Historically, they occupied territory from British Columbia to Northern California,
however, they are now believed to be missing from close to 80% of this range (Adams et al.,
2014; Groff et al., 2014; Hayes, 1997). Because of this large range reduction, the Oregon
spotted frog is listed as threatened under the US Endangered Species Act. At the state level,
Oregon spotted frog is designated as endangered in Washington and Canada, sensitive in Oregon
and believed extirpated from California. The loss of Oregon spotted frog from this territory can
be attributed to habitat loss or conversion, the introduction of invasive species, and disease; all of
which have the potential to be exacerbated by climate change (Hallock, 2013; Holgerson et al.,
2019; McAllister & Leonard, 1997).
In Washington state, Oregon spotted frogs historically occupied at least fifteen different
watersheds. Currently, they are known to exist in just four watersheds in the Puget Sound: the
Samish River, Black Slough, Sumas River, and the Black River, and two in the Eastern
Cascades: Trout Lake Creek and Outlet Creek (Figure 1). Conboy Lake National Wildlife
Refuge, part of the Outlet Creek drainage, supports one of the largest populations in the entire
range (Bohannon et al., 2016; Hallock, 2013; McAllister & Leonard, 1997). While the Black
River is a tributary of the Chehalis River, there are no records of Oregon spotted frogs ever
occurring anywhere else in the Chehalis River Basin (M. Hayes, Personal communication).
Outside of Washington state, Oregon spotted frogs currently occupy sites in the Fraser Valley in
British Columbia and in the Oregon Cascades (Hallock, 2013; McKibbin et al., 2008).

4

Figure 1. The watersheds in Washington state currently known to have populations of Oregon
spotted frog.
5

Life History
Non-Breeding Habitat
Oregon spotted frog are strictly aquatic, warmwater wetland specialists occupying a
variety of habitat types depending on lifecycle stage or season (Pearl & Hayes, 2004). Their
habitat varies depending on the breeding, non-breeding, and over-wintering season (Watson et
al., 2003). To meet this diverse need, occupied wetlands tend to be at least four hectares in size
and contain perennial streams and water bodies aquatically connected to areas of seasonal
inundation (Pearl & Hayes, 2004; Watson et al., 2000). A gradual topographic gradient sloping
towards the permanent waterway is ideal to allow for the persistent and required water levels
used during the year (Watson et al., 2000). Wetland systems used by the Oregon spotted frog are
palustrine and lacustrine (Bohannon et al., 2016). Palustrine systems are shallow, permanent or
temporary, vegetated wetlands found on the edges and floodplains of rivers, lakes or ponds.
Marshes, swamps, bogs, prairies and fens are all palustrine systems. The emergent vegetation,
plants that root in the soil of aquatic environments but grow above the water level, in these
systems is different than that of the vegetation in running or permanent, deeper water (Cowardin
et al., 1979). Lacustrine systems are deeper than palustrine and tend to be more permanent. They
are formed by dammed rivers and depressions (Cowardin et al., 1979). Palustrine, emergent
wetlands are the primary system occupied by Oregon spotted frog. Wetland complexes that have
multiple habitat types are more likely to be occupied. (Pearl & Hayes, 2004).
During the non-breeding season, Oregon spotted frogs reside in permanent, usually
deeper water, unlike the seasonally inundated shallow areas used during the breeding season
(Watson et al., 2000, 2003). During the summer months, water temperatures in these permanent
waterways typically exceeds 20 °C (Hayes, 1994). These deeper pools contain sparse to

6

moderate emergent vegetation, often hardhack (Spirea douglasii) dominated, and vegetative
mats; on which they can float and bask (McAllister & Leonard, 1997; Watson et al., 2003) The
vegetation allows for cover and escape from predators, while the mats offer better thermal
regulation as they can move from the warmer, shallower water created by the mats or to deeper
water depending on temperature needs. (McAllister & Leonard, 1997; Popesu et al., 2013;
Watson et al., 2000). Oregon spotted frogs will either swim in the deep water or crawl across the
mats to ambush prey, mainly consisting of aquatic breeding insects (Pearl et al., 2005).
There is flexibility in Oregon spotted frog habitat selection during this season, both in
water depth, such as deep pools or shallow flooded fields, and among what dominant vegetation,
usually sedge or shrub-scrub, but the availability of water remains the limiting factor (Watson et
al., 2000). As seasonal water levels recede, frogs will stay in permanent pools, only venturing out
if wetter conditions provide temporary flooding of surrounding areas and corridors between
habitat types (McAllister & Leonard, 1997; Popesu et al., 2013; Watson et al., 2003). This need
for both aquatic habitat and corridors can constrain frogs to small areas during the summer if
drier conditions prevent adequate water levels for movement (Popesu et al., 2013; Watson et al.,
2003).
Oregon spotted frogs use water temperature as a sign of changing seasons and will move
to overwintering habitat along aquatic corridors as the temperature begins to lower. In the colder
regions of their range, water temperatures approaching five degree Celsius cue the frogs to find
suitable habitat before the more extreme winter conditions occur. (Hayes et al., 2001; Pearl et al.,
2018). Overwintering habitat for the Oregon spotted frog requires adequate levels of dissolved
oxygen and cover from predators (Hayes et al., 2001). Dissolved oxygen levels will be lower in
water that is iced-over, therefore overwintering in springs and beaver dams can allow flowing

7

water to prevent hypoxic conditions and lessen the chance of freezing. However, if under ice,
frogs will move around to find pockets of higher oxygenated water (Hallock, 2013; Hayes et al.,
2001; Pearl et al., 2018). The absence of predators in overwintering sites is important and can
include fish, birds, minks and otters (Chelgren et al., 2008; Hayes et al., 2001; Pearl et al., 2018;
Watson et al., 2000). Cover from predators can be found in beaver dams and bank hollows (Pearl
et al., 2018; Tattersall & Ultsch, 2008).
These different habitat types need to exist within a relatively small distance as frogs do
not disperse over large areas. They need to be connected by aquatic corridors, thus wetland
complexes that contain multiple habitat types are ideal (Pearl & Hayes, 2004). Habitat elements
that exist in isolation will not be available to the frogs, such as a disconnected pool or shallow
flooded areas that are not connected to a perennial water source (Watson et al., 2000). Frogs may
migrate between populations or colonize new areas that are aquatically connected to their home
ranges (Watson et al., 2003).
Breeding Habitat
In late winter and early spring, as water temperatures begin to consistently rise above 5
°C, Oregon spotted frogs will move out to breeding sites to begin oviposition (Bowerman &
Pearl, 2020; Licht, 1974; Mcallister & White, 2001; Pearl & Hayes, 2004). Reliance on
environmental cues, as opposed to calendar dates, give the frogs flexibility for starting migration
and breeding and can therefore avoid large mortality events. These events can occur when
moving away from overwintering sites too early and freezing, too late and leaving eggs stranded
in drier conditions, or exposing themselves to predators while at the breeding site (Bowerman &
Pearl, 2020). Transitioning from overwintering sites, frogs move out into shallow, lentic water to
begin oviposition (McAllister & Leonard, 1997; Pearl & Hayes, 2004; Tattersall & Ultsch, 2008;

8

Watson et al., 2003). Oregon spotted frog are communal breeders and have a high site fidelity,
often returning to the same site each year to lay egg masses in groups of varying sizes (Licht,
1971; Pearl & Hayes, 2004).
Because of Oregon spotted frog’s aquatic nature, areas of temporary shallow water, such
as floodplains and seasonal wetlands, must have a hydrological connection to a permanent
waterway (Watson et al., 2003). Shallow water depth, typically less than 25 cm, is critical for
oviposition, therefore, these sites are often located near the margins of seasonally inundated
floodplains or near the shores of ponds, but small depressions or even tire tracks may be
sufficient (Licht, 1971; McAllister & Leonard, 1997; Mcallister & White, 2001; Pearl, Adams, et
al., 2009; Pearl & Hayes, 2004; Watson et al., 2000, 2003). Temperatures are warmer in shallow
water, which aids in incubating egg masses (Mcallister & White, 2001; Pearl, Adams, et al.,
2009). However, these temporary shallows create potential for large mortality events if water
recedes too quickly and egg masses are left to desiccate or tadpoles and juvenile frogs are unable
to reach permanent water in the drier season due to lack of aquatic connections (Licht, 1974;
Pearl & Hayes, 2004; Watson et al., 2003).
Breeding sites are mostly located in palustrine wetlands among emergent vegetation or
aquatic beds (Bohannon et al., 2016; Pearl & Hayes, 2004). These areas typically have moderate
to low density vegetation with open canopies to prevent shading and allow for more surface
exposure to the sun (Kapust et al., 2012; Watson et al., 2003). The dominant vegetation at
oviposition sites is often a mix a sedges and rushes, however, the structure of the vegetation is
more important than the species and frogs will lay egg masses atop denser vegetated substrates if
the density and height of the surrounding vegetation is low enough to provide adequate sun
exposure (McAllister & Leonard, 1997; Pearl & Hayes, 2004; Watson et al., 2000, 2003).

9

Suitable vegetation structures mimic early successional stages with short heights and low
densities, however, with a lack of natural openings or disturbance, conditions can be found from
light grazing, winter snowpack flattening and compressing the vegetation, or human activities
such as mowing or haying (Hallock, 2013; Kapust et al., 2012; Watson et al., 2003).
Beaver dams have the potential to create and expand Oregon spotted frog habitat. The
resulting hydrology from beaver dams satisfy many of the requirements for the Oregon spotted
frog lifecycle. The increase of permanent water behind the dam, the resulting flooding of
adjoining fields and meadows, and the shelter provided during overwintering are all critical
elements of the lifecycle (Hallock, 2013; Pearl et al., 2018; Romansic et al., 2020).

Threats to the Oregon Spotted Frog
Amphibians are in decline worldwide (Stuart et al., 2004). In the United States, surveys
of species that have a threat designation, such as threatened or endangered, show a continual
decline across their ranges (Adams et al., 2013). Sources of the decline comes from habitat loss,
invasive species, climate change, and disease, specifically the fungal disease Batrachochytrium
dendrobatidis (Bd) (Wake & Vredenburg, 2008). Human interference can be attributed to many
of the threats posed to amphibians, including altering the hydrology and destroying habitat,
introducing invasive species and contributing to climate change (Arkle & Pilliod, 2015; Wake &
Vredenburg, 2008). In the Pacific Northwest, aquatic amphibians, including Oregon spotted frog,
face challenges from disease, habitat loss, altered hydrology from the removal of beaver (Castor
canadensis) and human interference, climate change and the introduction of invasive plant and
animal species (Arkle & Pilliod, 2015; Hallock, 2013; Hayes et al., 2009; Pearl, Adams, et al.,
2009). The species that are most adversely affected by these threats occur in small population

10

sizes in habitat that is located in areas undergoing human encroachment and development (Wake
& Vredenburg, 2008).
The Oregon spotted frog lifecycle naturally lends itself to high mortality events and
population fluctuations due to communal breeding activities. High reproductive mortality can
occur when disturbances affect egg masses concentrated in a small area or when frogs
congregate at the breeding site and are exposed to high predation events (Chelgren et al., 2008;
Mcallister & White, 2001). These natural fluctuations combined with isolated populations and
novel interferences are contributing to the decline of the Oregon spotted frog (Blouin et al.,
2010; Pearl & Hayes, 2004).
Habitat Conversion and Hydrology Changes
One of the major causes of Oregon spotted frog decline is the loss of suitable habitat from
altered hydrology and disturbance regimes. The lowland floodplains and wetlands inhabited by
Oregon spotted frog are readily converted to agricultural and livestock fields (McAllister &
Leonard, 1997). These conversions are accomplished by channelizing, draining and dredging
wetlands and marshes, and the resulting hydrology can no longer support Oregon spotted frog
(Hallock, 2013; McAllister & Leonard, 1997). The channelized waterways prevent seasonal
flooding and diminish areas of sustained water, thus eliminating aquatic corridors between
different populations and local habitat types (Cushman, Kathleen & Pearl, Christopher, 2007;
Watson et al., 2000, 2003). Isolated populations with no means of reaching other populations
will experience genetic diversity loss (Blouin et al., 2010; Cushman, Kathleen & Pearl,
Christopher, 2007; McKibbin et al., 2008). Human control of water flows can lead to drastic
changes in the water levels of Oregon spotted frog habitat and can strand egg masses or tadpoles
if levels recede too rapidly (Hallock, 2013; McAllister & Leonard, 1997).

11

The increase of human presence in Oregon spotted frog habitat has led to a decline in
disturbance responsible for resetting wetlands to early successional stages. Fire has been
removed from many natural systems following European settlement and flood events have
decreased due to the channelization of many waterways and removal of beavers (Hallock, 2013).
The absence of these disturbances has accelerated the establishment of woody plants and
succession into upland habitat not favored by Oregon spotted frogs (Cushman, Kathleen & Pearl,
Christopher, 2007; Hayes, 1997). Additionally, human restoration actions in many wetlands
involve planting woody species to benefit salmon, thus further reducing habitat for frogs
(Bohannon et al., 2016; Hallock, 2013). The increased cover from woody plants increases the
vegetation height and thus decreases the solar exposure relied upon for egg mass incubation and
warmer water temperatures.
Many sites now occupied by Oregon spotted frog are maintained by human activities
such as mowing or livestock grazing (Hallock, 2013; Kapust et al., 2012). Intensive grazing can
be detrimental to wetlands but moderate to light grazing mimics natural disturbance at many
Oregon spotted frog occupied sites by keeping vegetation heights and densities low and
preventing the establishment of woody species (Bohannon et al., 2016; Hayes, 1997; McAllister
& Leonard, 1997; Watson et al., 2000, 2003). Oregon spotted frogs breed more often in grazed
areas when the natural vegetation structure has been altered from lack of disturbance or the
introduction of invasive species such as reed canary grass (Phalaris arundinacea) (Bohannon et
al., 2016; Watson et al., 2003). The cessation of grazing at many Oregon spotted frog sites has
led to a decline in populations due to the reestablishment of invasive vegetation (Bohannon et al.,
2016; Hallock, 2013).
Invasive Species

12

Reed canary grass is a prolific invader of wetlands in the Pacific Northwest and
aggressively colonizes and degrades Oregon spotted frog habitat. It establishes rapidly after
disturbance and grows under a variety of environmental conditions and growth patterns
(Reinhardt Adams & Galatowitsch, 2005). This flexibility allows it to outcompete many native
species, including the sedges and rushes used by the Oregon spotted frog (Lavergne & Molofsky,
2004; Watson et al., 2003). Reed canary grass does not establish well under a shaded canopy but
is more successful in wetlands at an early stage of succession with no canopy and low vegetation
density (Maurer et al., 2003). Disturbance and the resulting early successional stage wetlands
such as these are ideal habitat for Oregon spotted frog and thus reed canary grass is a major
contributor to the decline in available habitat (Hallock, 2013). As abundance of reed canary grass
increases, the structure of the vegetation community changes with increasing density, height, and
thatch and litter depth (Spyreas et al., 2010). These altered vegetation characteristics run counter
to the breeding site preferences of Oregon spotted frog and they will no longer use these sites if
reed canary grass is prevalent at high densities (Bohannon et al., 2016; Watson et al., 2003). If
sites occupied by Oregon spotted frog are invaded by reed canary grass, frogs are forced to seek
out new areas that are less dense and will not hamper movement (Popesu et al., 2013). Because
the structure of the vegetation is more important than the specific species, oviposition may still
occur among reed canary grass if there are large enough openings. These openings may be
caused by snow compaction, grazing or other human interventions that mimic the short, sparse
vegetation characteristics preferred by the frog (Hallock, 2013; Kapust et al., 2012; Watson et
al., 2003).
Oregon spotted frog is an important food source for many native species, however, the
increased predation threat and habitat overlap from nonnative species amplifies pressure on small

13

populations and can act as a barrier to movement between aquatic habitats (Bradford &
Tabatabai, 1993). The occurrence of exotic fish and bullfrogs has been associated with reduced
presence or absence of native amphibians in the Pacific Northwest (Holgerson et al., 2019; Pearl
et al., 2004). Oregon spotted frog is particularly at risk of these species due to its completely
aquatic lifecycle. Frogs cannot escape predators by retreating to terrestrial habitat during the nonbreeding season as many other native amphibians can (Holgerson et al., 2019).
The introduced American bullfrog (Lithobates catesbeianus) is problematic for many
native Pacific Northwest frogs due to their opportunistic feeding behavior and large sized
juvenile’s ability to outcompete other species (Kiesecker & Blaustein, 1998; Pearl & Hayes,
2004). The American bullfrog shares similar habitat with the Oregon spotted frog and the
resulting interaction can force Oregon spotted frog into suboptimal habitat (Pearl & Hayes, 2004;
Rowe et al., 2021). This can result in reduced development in young frogs due to diminished
water conditions or food sources for tadpoles. They can also be forced to retreat into areas where
they interact with novel predators not normally encountered (Kiesecker & Blaustein, 1998).
Non-native fish can outcompete and feed on Oregon spotted frogs, especially during low
water years and while in overwintering habitat. Further, they can cause Oregon spotted frogs to
be forced out of preferred habitat by other non-native predators (Holgerson et al., 2019; Pearl,
Adams, et al., 2009). An abundance of warmwater centrarchid fishes (i.e., basses and other
sunfish) negatively affect native amphibian occurrence, especially in permanent water bodies
(Hayes, 1997; Holgerson et al., 2019). These are fish that have been introduced into many of
Washington’s rivers for sport fishing (Hallock, 2013). When frogs move into different,
potentially deeper, aquatic habitat they can encounter fish species that they would not normally
interact with (Hallock, 2013; McAllister & Leonard, 1997; Pearl, Adams, et al., 2009).

14

Additionally, predatory fish can be agents of isolation because they do not allow frogs to move
along aquatic corridors to reach new populations (Bradford & Tabatabai, 1993). The increase in
human presence can introduce synanthropic predators such as raccoons and crows (Hallock,
2013).
Disease - Batrachochytrium dendrobatidis in the PNW
A lesser threat to Oregon spotted frog persistence is the fungal disease, Batrachochytrium
dendrobatidis (Bd), which is attributed to global declines of amphibians (Wake & Vredenburg,
2008). Bd is prevalent throughout the Pacific Northwest and Oregon spotted frog has a high rate
of infection in comparison with other amphibians in the region (Hayes et al., 2009; Pearl,
Bowerman, et al., 2009). However, while infected frogs tend to be smaller, there is not a high
rate of mortality (Padgett-Flohr & Hayes, 2011; Pearl, Bowerman, et al., 2009). This may be an
artifact of the fungus being present for a long time in the region and frogs that survived today are
the ones that are resistant to it (Padgett-Flohr & Hayes, 2011).
Climate Change
With a changing climate, wetlands in the Pacific Northwest are expected to experience
increased temperatures and changing precipitation regimes (Hudec et al., 2019). These changes
are expected to cause a shift in the hydrology of wetlands by changing the evapotranspiration
rate, the duration of inundation, the seasonal water levels, and the groundwater recharge or
depletion rate (Hallock, 2013; Hudec et al., 2019). However, many of these changes, and the
effects they have on wetland species, is location dependent, affected by the local geology,
climate, and surrounding land uses (Hudec et al., 2019). For example, a snow melt dependent
system may experience more precipitation as rain, increasing stream inputs in the winter and fall

15

instead of the spring and summer, or if neighboring agriculture is forced to draw more from
groundwater storage during drier summers (Hallock, 2013; Hudec et al., 2019).
The Oregon spotted frog is considered highly vulnerable to climate change (Hohmann &
Wall, 2017). While the effects of climate change are hard to predict, due to the aquatic nature of
its entire life history, changes to the hydrology of wetlands will inevitably have effects on the
frog. Increased temperatures and precipitation falling earlier and occurring as rain instead of
snow pack, will shift historic water levels earlier in the year (Hallock, 2013). Coupled with more
extreme winters, earlier springs can cause Oregon spotted frogs to mis-time their breeding,
resulting in decreased reproductive success (Bowerman & Pearl, 2020). Shallow, ephemeral
wetlands are at risk of not forming or drying quicker, thus stranding egg masses and tadpoles.
Permanent wetlands can shrink in extent which will reduce available habitat for frogs during the
drier months and allow for woody plants and reed canary grass to establish in the newly dried
area (Hallock, 2013; Hudec et al., 2019). Oregon spotted frogs depend on aquatic corridors for
migration, which will either shift in availability due to altered precipitation and drier summers, or
not form at all and isolate populations and reduce gene flow (Robertson et al., 2018). A potential
positive effect of climate change is the increased chance of disturbances, especially fire and
flood events. These disturbances can reset wetlands to an early seral stage, ideal habitat for the
Oregon spotted frog (Hudec et al., 2019).

Oregon Spotted Frog in the Chehalis River Basin
The Black River is one of the few remaining watersheds in Washington state to have
Oregon spotted frogs. This tributary of the Chehalis River is connected aquatically to the greater

16

Chehalis River Basin, yet there are no historical records of frogs occurring outside of this
watershed (M. Hayes, personal communication).
Oregon Spotted Frog Habitat
The Chehalis River Basin is the second largest river basin in Washington State behind the
Columbia River, draining 2,600 square miles. Beginning in southern Lewis county, its runs
north and west for 126 miles before ending in Grays Harbor (Thurston County Public Health and
Social Services Department, 2006). It is primarily a rain driven system (Department of Ecology,
2016). Many of its major tributaries begin in forested headwaters and drain down to lower, more
expansive floodplains. The tributaries nearer to the headwaters and middle of the Chehalis River
contain palustrine wetland and prairie habitats, however the historic extent of these habitats has
been reduced by development and agriculture. These tributaries include Salzer Creek, the
Newakum River, the Skookumchuck River, Scatter Creek, the Black River, and the Satsop River
(Department of Ecology, 2016; Species Restoration Plan Steering Committee, 2019). Wetlands
in the Skookumchuck and Newakum Rivers have been reduced by 90 and 75 percent
respectively. The Black River is an exception to this extensive habitat loss as it contains one of
the largest intact wetland complexes in the Puget Sound (Species Restoration Plan Steering
Committee, 2019).
Flowing out of Black Lake, near Tumwater, WA, The Black River meanders south for 28
miles, draining 136 square miles before joining the main stem of the Chehalis River (Dickes,
1990). The first seven miles south of Black Lake are protected riparian and wetland habitat.
Combined with additional intact wetlands along the main stem and in multiple tributaries, the
Black River watershed contains one of the largest contiguous, freshwater wetland complexes in
the Puget Sound (Species Restoration Plan Steering Committee, 2019; Watson et al., 2000). The

17

watershed was formed by retreating glaciers resulting in many areas of porous, glacial outwash
soils. These porous soils and gentle gradients of many of the river’s tributaries cause extensive
groundwater flooding when the water table rises and they are unconfined in wetlands and
agricultural fields (Thurston County Public Health and Social Services Department, 2006). This
flooding and intact wetland habitat makes for ideal habitat for the Oregon spotted frog, making
the Black River watershed one of the few rivers in Washington state and the only tributary in the
Chehalis River Basin that the frog can be found (Species Restoration Plan Steering Committee,
2019).
Oregon Spotted Frog in the Black River
Oregon spotted frogs were first identified in the Black River during surveys of the Puget
Sound in 1990 (McAllister et al., 1993). They have since been located in many other sites both
along the Black River and in many of its tributaries. These include Allen Creek, Beaver Creek,
Dempsey Creek, Fish Pond Creek, Michelle Creek, Mima Creek and Salmon Creek (Washington
Deptartment of Fish and Wildlife, 2020).

Habitat Modelling
Species Distribution Models
Species distribution models (SDM) use the natural history and ecology of a species and
combines them with statistical methods to explain the current and predicted species’ distribution
(Elith & Leathwick, 2009). They accomplish this by using locations with known species
occurrences and combine them with relevant environmental variables to predict a distribution of
similar environmental conditions across a targeted geographical area (Elith & Leathwick, 2009).
SDMs are also known as ecological niche models as they model a species ecological niche
18

within the studied environmental (Phillips et al., 2006). A species occupies a specific niche in the
environment that contains elements for long term persistence without the need of immigration
from other populations (Pulliam, 2000). The environmental niche is habitat suitable enough for a
species to survive and reproduce in perpetuity. The fundamental niche is the entire collection of
this suitable habitat and represents the geographic range available to this species, while the
realized niche is the proportion of this available habitat that the species actually occupies
(Pulliam, 2000). The difference between the fundamental niche and the realized niche can be
attributed to external influences such as disturbance removing a population, competition from
other species, and loss of connectivity between habitat (Phillips et al., 2006; Pulliam, 2000). A
combination of environmental variables comprise the ecological niche, therefore, it is important
to understand the ecology of the species of interest when choosing the environmental layers used
in the model (Guisan & Thuiller, 2005). Different model approaches have been developed to
incorporate either the realized nice or the fundamental niche when predicting suitable habitat; the
difference being whether the model incorporates absences or just presences for the occurrence
data.
Types of Species Distribution Models
Species distribution models can be classified as either presence-absence or presence-only
depending on the type of occurrence data used in the model (Elith et al., 2011). Presence-absence
models use inputs of both known presence locations and locations that were surveyed but no
species were detected. Presence-only models use inputs of only known locations. There are
advantages and disadvantages to using either model, but presence-only models perform better
when species occurrence data is limited, especially in the case of threatened species where there
are small numbers of populations (Cianfrani et al., 2010). Presence-absence models assume that

19

the species is in equilibrium with the environment and all suitable habitat is occupied, thus using
the fundamental nice as a model input. The assumption is that species are absent from sites
because there is something key missing from that location (Guisan & Thuiller, 2005). Problems
with this assumption occur when species are difficult to detect during surveys, corridors have
been disrupted making suitable habitat unreachable to species with limited dispersal capacity, or
a prior disturbance removed the population and they have been thus far unable to recolonize
(Elith et al., 2006, 2011). SDMs are constrained to modelling in temporal space as well as
geographic space, therefore, if a species has been removed in the past from suitable area, it
should not be considered absent because in time they may recolonize. These factors can lead to
false absences which result in false predictions of suitable habitat (Cianfrani et al., 2010).
Presence-only models remove the assumption of environmental equilibrium and only use
the realized niche. Inputs are collected from the habitat that is currently being used, to predict the
fundamental niche, the habitat that could be used (Phillips et al., 2006). These models have been
shown to create better predictions for rare and threatened species as well as modelling potential
recolonization areas. These potential recolonization area are missing from presence-absence
models because they would have been inputted as absent from the start (Cianfrani et al., 2010;
Elith et al., 2006).
However, there are some disadvantages to presence only modes. Presence-only models
cannot predict the prevalence, the proportion of occurrences that occur at certain sites, of a
species (Elith et al., 2011). Presence-only models are also more affected by sample bias, which
occur when the sampling method favors areas either easily accessible to surveys due to
geographic location, close to roads and towns, or access issues from private property (Boria et
al., 2014; Elith et al., 2011). There is no way to know if areas that do not have occurrences are

20

due to simply being unsampled, therefore the conditions at these sites may still be viable habitat.
This bias can lead to spatial autocorrelation and overfitting of the model, leading predictions to
favor the already known locations as these were the locations used when training the model
(Boria et al., 2014). Models that are overfit will do poorly when new testing data is inputted into
the model and will under-predict additional habitat beyond the original inputs (Radosavljevic &
Anderson, 2014).

Maxent
Maxent is a presence-only model that performs well when compared to other SDMs
(Ortega-Huerta & Peterson, 2008). Maxent is a species distribution model that estimates the
probability of presence based on an index of habitat suitability across a targeted geographic area
(Phillips et al., 2006). The model predicts presence at maximum entropy, or the maximum
dispersal while being bounded by some environmental constraint (Phillips et al., 2006). The
constraints are defined by the conditions of environmental variables, deemed ecologically
relevant, at known locations of species presence. The details of how Maxent predicts habitat is
described in the Methods section. The use of Maxent is widespread and a popular method of
species distribution modelling.
Modelling for amphibians with Maxent
Maxent is a useful tool with a variety of uses in amphibian ecology. Compared to other
modelling algorithms, Maxent is better at capturing the ecology of the species rather than solely
being a result of statistical analysis (Preau et al., 2018). After all, species distribution models are
meant to be an exercise in combining statistics with ecology, rather than a simple mathematical
equation. Given the ability to perform well with small sample sizes, Maxent is useful for
modelling distributions of amphibians with limited occurrence localities (Pearson et al., 2007;

21

Tarrant & Armstrong, 2013). The outputs of these models offer potential locations of
conservation importance for endangered species and guide surveys for locating new populations
of less studied species (Blank & Blaustein, 2012; Tarrant & Armstrong, 2013). Models created
during different seasons or stages of the lifecycle can highlight migration patterns or species
specific behaviors in amphibians (Najibzadeh et al., 2017). Additionally, Maxent works well
with climate data. Like many species, amphibians have climatic restraints to their ranges and
Maxent models can help define these (Cunningham et al., 2016). Amphibians tend to be less
mobile than other taxa and are therefore face greater risks to climate or land use changes.
Modelling with projected climate and land use conditions can define the loss or geographic shift
in available habitat (Gül et al., 2018; Struecker & Milanovich, 2017).
Modelling for Oregon Spotted Frog
Prior modelling for Oregon spotted frog involved using a GIS “screen” to assess site
suitability for the frog. These screens were created by combining data layers determined as
ecologically relevant from literature studies. Layers include wetland type and size, soil type,
elevation, and aquatic connectivity (Bohannon et al., 2016; Germaine & Cosentino, 2004). This
method was applied in the North Puget Sound to successfully identify three new watersheds that
contain previously undocumented populations, the Samish, Nooksack, and Sumas rivers
(Bohannon et al., 2016). Maxent adds a statistical element to this type of ecological modelling
(Na et al., 2018). It was used to predict potential populations in Southern Oregon and Northern
California, where frogs are believed to be extirpated. Subsequent surveys of predicted locations
did not find frogs in California, however, individual frogs were located in a previously
undocumented location in Oregon which was predicted by the Maxent model (Groff et al., 2014).

22

When modelling for Oregon spotted frog with Maxent, the environmental variables need
to reflect the frog’s ecology, including limits on dispersal and habitat selection. Aquatic habitat is
required at all life stages and is usually located among emergent wetlands (Pearl & Hayes, 2004).
This habitat requirement can be represented in Maxent modelling by using land cover class and
soil data. Hydric soils are formed under anaerobic conditions by permanent water or temporary
flooding (Ecology, 1997). The distinction between hydric and non-hydric soils can delineate
aquatic environments. The vegetation structure is another critical variable for breeding habitat
selection, as frogs lay egg masses in areas of short statured vegetation that allows for exposure to
air and solar radiation (Kapust et al., 2012; Watson et al., 2003). Habitat tends to be at lower
elevations with higher water temperatures and gentle slopes for the lentic water used for
oviposition (Pearl & Hayes, 2004). As such, elevation, slope, and climate data are variables that
are important to predicting suitable locations. Variables that can capture these aspects of Oregon
spotted frog will benefit the model’s predictive ability and are included as the key inputs in this
study.
One of the main uses of Maxent is to locate areas of suitable habitat to guide survey
efforts for less documented species (Blank & Blaustein, 2012). Applying Maxent in this way to
model Oregon spotted frog breeding habitat in the entire Chehalis Basin could provide locations
that may have frogs but have not been previously documented. However, many of the wetlands
in the floodplain and off-channel habitats in the main stem of the Chehalis River have been
surveyed, which allows for an additional use for the Maxent model. If survey locations coincide
with the model output, it can be assumed that they are suitable breeding habitat, however, none
of the surveys detected Oregon spotted frog (M. Hayes, personal communication). Even with this
discrepancy, this can be useful as the habitat of the survey sites can be compared with the habitat

23

of known breeding locations to test for differences. A comparison in this manner will also allow
for the inclusion of additional variables such as the abundance of invasive species, specifically
bullfrogs and exotic fish. Because, biotic influences can determine a species dispersal beyond the
abiotic factors, and the Maxent model is only using abiotic variables yet predicting suitable
habitat, the presence of these invasive species may be influencing the presence of Oregon spotted
frog (Elith & Leathwick, 2009).

24

METHODS
Oregon spotted frog were not known to occur in the Chehalis Basin until 1990, when they
were discovered at Dempsey Creek, a tributary of the Black River (McAllister et al., 1993).
Several additional populations have been discovered in the Black River in the years following
this initial find. Frogs may have occurred beyond the Black River watershed in the main stem of
the Chehalis River floodplain or in its other tributaries. However, surveys in the Chehalis River
floodplain have failed to detect Oregon spotted frog. To better understand this apparent absence,
Maxent was used to model Oregon spotted frog breeding habitat in the Chehalis Basin,
specifically to see if suitable habitat exists elsewhere in the Chehalis Watershed that has the
same characteristics of known Oregon spotted frog sites in the Black River. A Maxent species
distribution model was developed from the habitat and climatic data associated with extant
Oregon spotted frog populations in the Black River watershed. The outputs of the Maxent model
revealed that sites surveyed in the Chehalis River mainstem floodplain were predicted to be
suitable habitat. The structure and climatic conditions of these sites were compared to the known
breeding locations for Oregon spotted frog in the Black River. An additional variable set was
evaluated for these locations: the abundance of centrarchid fishes and bullfrogs, exotic species
known to be detrimental to Oregon spotted frog. The purpose of this addition was to assess if
these exotic species could be limiting Oregon spotted frog distribution in the Chehalis River
mainstem floodplain.

Maxent
Maxent is a species distribution model that estimates the probability of presence based on
an index of habitat suitability across a targeted geographic area (Phillips et al., 2006). Habitat
25

suitability is defined by the environmental conditions at locations of known occurrence.
Occurrence locations are inputted as presence points and environmental data are inputted as
raster grid layers that encompass the entire study area, with each cell of the grid being assigned a
value of that variable (Phillips, 2017). For example, each cell of a raster layer for slope at a 30meter resolution would contain a value for the degree of slope found in that 30 m2. The slope
value for a presence point is based off of what cell the point occurs in. Habitat suitability is
determined by a value calculated from a set of mathematical transformations, called features.
Features are performed on and between the set of environmental values at each presence point, as
well as a set of background points randomly selected from the study area (Phillips et al., 2006;
Phillips & Dudík, 2008). How much the means of the features at the background points differ
from the means of the presence points determines the level of habitat suitability. A habitat
suitability index is created that defines the range of values and their level of suitability (Elith et
al., 2011). This process is described as model training , as the model is being trained to identify
suitable habitat based on the presence points. (Phillips et al., 2006; Phillips & Dudík, 2008). The
final Maxent output is the probability of presence of each cell in the raster grid according to the
suitability index. Cells that have habitat suitability values close to those of the presence points
are deemed to have a high probability of presence while cells that differ substantially from the
presence points have a low probability of presence (Elith et al., 2011; Phillips et al., 2017).

Presence Point Selection
To model Oregon spotted frog breeding habitat, egg mass locations were acquired from
the Washington State Department of Fish and Wildlife’s Priority Habitat and Species database
(Washington Deptartment of Fish and Wildlife, 2020). This database includes locations of
sensitive species and their associated habitats across the state of Washington. Data points from

26

the Black River were acquired and filtered to include only locations of egg masses collected with
GPS during annual egg mass surveys. This filter was applied to avoid including points that might
represent dispersal of juveniles, or other conditions and might represent non-representative
habitat that was not regularly occupied. In the dataset, the locations of egg mass points were
representative of breeding habitat, and confidence was high that these points were accessible to
non-breeding habitat. Other points in the dataset were classified as occurrences when individual
frogs were observed, however, the context of these observations was uncertain, therefore they
were excluded to focus the model on breeding habitat. Since the Oregon spotted frog was
discovered in the Black River in 1990, subsequent populations in the watershed have been
discovered at various times in the following years (Hallock, 2013; McAllister et al., 1993).
Therefore, the data obtained from the WDFW database spans the years 1996 to 2019, with some
populations having been sampled multiple times since 1996 and others only recently. Despite
this discrepancy, the entire range of surveys years was included for analysis to maximize
information on breeding habitat usage. Overall. there were 894 breeding points considered as
presence points in the Black River.
Egg masses in the Black River are clustered in distinct groupings of varying sizes. In
order to reduce effects from spatial-autocorrelation, points within these groups were spatially
filtered using the “Spatially Rarefy Occurrence Data for SDMs” tool in SDMToolbox (Brown et
al., 2017). Spatially rarefying points is a process that selects a presence point and then removes
every other point within a specified distance (Brown et al., 2017). The objective was to find a
spatial distance that balances maximizing information on habitat use (retains a maximum number
of points) and keeps the points independent as possible (removes points too close together to be
considered independent) (Boria et al., 2014). The points were spatially rarefied to a scale of 300,

27

400, and 500 meters and the results of each distance were visually compared to gauge the
distribution of the remaining points both within and between the original groupings of egg
masses. After visually comparing the results, the points remaining at the 500-meter rarified
distance were chosen for the model. Because the breeding points occur in distinct groups of
varying sizes, the 500-meter distance reduced each group to less than five points, depending on
the total area encompassed by the original group. While the groups are not completely
independent, as some of the larger groups are represented by multiple points, the remaining
points capture the variability of the habitat used in the larger groups. Too much information on
habitat use may be lost if all but one point is removed. This process resulted in 29 presence
points used for the model (Figure 2). The large reduction in presence points from 894 to 29
avoids overfitting in the final model as the points are concentrated in the Black River, a small
portion of the Chehalis Basin. Overfitting can occur when a model is trained from a large
number of points in one area, many with non-independent habitat characteristics. The model will
overpredict habitat suitability of that small area and fail to generalize predictions to the rest of
the study area. Using the minimum number of points to capture the same information lessens
this possibility (Boria et al., 2014; Radosavljevic & Anderson, 2014). Maxent is still capable of
robust predictions when using a small number of presence points (Pearson et al., 2007). The
points were then projected into the WGS 1984 UTM Zone 10 coordinate system and the
longitude and latitude were calculated for each point.

28

Figure 2. Presence points in the Black River watershed as a result of being spatially rarefied at
500 meters.

Background Point Selection
During the model training process, random background points are selected to compare to
the values of the presence points to compute habitat suitability and create a suitability index
(Phillips et al., 2006). When selecting background points, Maxent will randomly select points
across the study area, therefore the entire range of values from the environmental variables have
the potential to be used for training the model. However, if the species is only expected to occur
in habitat that is bounded within a limited range of environmental values, a random background
point selection can select many points outside of the range of what would be suitable habitat
(Merow et al., 2013). The values of many of these points will be far from the values of the

29

presence points, thus biasing the suitability index to be very close to the values of the presence
points. The final output will be overfit and predict little probability of presence in cells that are
not close in value to the presence points. To increase the predictive ability of the model, it can be
beneficial to limit the selection of background points to a spatial range that the species would be
expected to disperse, thus limiting the model training to environmental values that are relevant to
the species (Merow et al., 2013).
In the case of the Oregon spotted frog, which only occur in aquatic habitat and are limited
in their dispersal ability, background points heavily selected from upland habitat would reduce
the predictive ability of the model. Therefore, background point selection was limited to a 1.5kilometer buffer around each of the presence points. This distance was chosen to constrain points
to only encompass the breeding site and areas accessible to dispersing Oregon spotted frogs (M.
Hayes, personal communication). A “bias file” was created for the modelling software using the
“Sample by Buffered Local Adaptive Convex-Hull” tool in SDMToolbox (Brown et al., 2017).
The Maxent default of 10,000 background points were used to train the models.

Environmental Variables
Models become more meaningful when the environmental variables are refined and
focused on the ecology of the species of interest (Elith & Leathwick, 2009; Guisan & Thuiller,
2005). If too many variables exist in the model, it is difficult to find meaningful relationships
between them. In contrast, if too few variables exist, the prediction may be too broad to be useful
(Merow et al., 2013). A discerning approach to variable selection in ecological modelling should
be adopted and be limited to variables with potential to describe the distribution of the studied
species. These can include resource availability, disturbance, and climate. Variables affect
species distribution at different scales with resource availability and disturbance describing

30

species distribution at a fine scale, while climate, typically at the resolution measured, tends to be
more coarse (Guisan & Thuiller, 2005).
The environmental variables used to model Oregon spotted frog breeding habitat were a
collection of raster grids obtained from publicly available sources in a variety of formats and
resolutions. Maxent requires variables to be in the same coordinate system and match in
resolution and geographic extent. ArcGIS Pro Version 2.7 was used to process variables to meet
these requirements and derive new variables from the public data. The variables used in the
model were projected into the WGS 1984 UTM Zone 10 coordinate system, resampled to a
resolution of 30-meter grid cells if necessary, and clipped to the extent of the Chehalis Basin.
Two sets of variables, structure and climate, were used to create the model for Oregon
spotted frog breeding habitat. The structure variables describe the physical environment at a finer
scale, while the climate variables generally capture a broader, but local regional scale (Guisan &
Thuiller, 2005).
Structure Variables
The structure variables were included based on specific characteristics of Oregon spotted
frog breeding habitat. In order to decrease complexity in the model, variables were chosen that
are directly related to the ecology, and potentially describe the distribution of the frog. Inclusion
of each variable was based on the literature and expert opinions (M. Hayes, personal
communication). Table 1 lists the variables collected from public sources and their original
resolution.

31

Table 1. Structure variables used in the Maxent model obtained from publicly available sources
and their original resolution.
Variable
Digital Elevation Model

Landcover
Open Water
Developed, Open Space
Developed, Low Intensity
Developed, Medium Intensity
Developed, High Intensity
Barren Land
Deciduous Forest
Evergreen Forest
Mixed Forest
Shrub/Scrub
Herbaceous
Hay/Pasture
Cultivated Crops
Woody Wetlands
Emergent Herbaceous Wetlands
Vegetation Height Class
Non-Vegetated Classes
Developed
Barren
Quarries/Mines
Agriculture
Sparse Vegetation
Tree Height
Shrub Height
Herbaceous Height
Hydric Soil
Non-hydric
Hydric

Unit
Meters

Original Resolution
1 arc-sec (~30m²)

Source
National Elevation Dataset
(USGS, 2020)

Categorical

30 m²

Multi-Resolution Land
Characteristics Consortium
(Yang et al., 2018)

Meters

30 m²

LANDIRE Existing
Vegetation
(LANDFIRE, 2016b)

Categorical

Vector

WA DNR Geospatial Open
Data Portal

A digital elevation model was obtained from the United States Geological Service
(USGS). Elevation was not directly used as a variable in the final model, but instead was used to
create a mask for the structure variables, removing areas from consideration in the model. Cells
above 634 meters and below 2.8 meters were removed using the mask. Oregon spotted frog has
not been found above 634 meters in Washington state and the 2.8-meter cutoff removes cells
below the high tide level in Grays Harbor, removing any marine influence. A slope variable was
32

derived from the digital elevation model using the “Slope” tool in ArcGIS Pro. Oregon spotted
frog breed in shallow stable water, found in conjunction with gentle slopes.
The land cover variable defines the dominant cover class of each cell using 20 unique
classifications, 15 of which occur within the Chehalis Basin (Yang et al., 2018). Of the excluded
classifications, four are unique to Alaska while the only classification of the lower 48 states not
found in the Chehalis Basin was perennial ice/snow. The 15 cover classes were reduced to three
classes relevant to Oregon spotted frog breeding habitat using the “Reclassify” tool. Reducing
the classes to ones relevant to Oregon spotted frog breeding habitat reduces the complexity of the
model and removes variables that are not predictive for breeding habitat usage. The new classes
were non-habitat, agriculture/pasture, and emergent herbaceous wetland. The classifications were
chosen based on the literature, expert opinions and creating a Maxent model only using the
landcover variable. The classes in this landcover-only model that were influential in predicting
habitat use were kept, while the remaining classes were reclassified into a single “non-habitat”
class. An additional variable was derived from the reclassified landcover variable, described as
follows: Using the “Euclidian Distance” tool, the proximity of each cell to the nearest emergent
wetland or agriculture/pasture cell was calculated. Oregon spotted frog breed in seasonally
flooded areas, many of which may extend beyond the boundaries of what was designated as
emergent wetland or agriculture/pasture during the image classification process when the cover
class variable was created. The proximity variable was meant to capture the movement away
from the predictive cover classes if seasonal water levels carried beyond the boundary of the
classified cells.
Vegetation height is classified by the dominant vegetation type of each cell (LANDFIRE,
2016a). Similar to landcover, vegetation height was reclassified for relevance to Oregon spotted

33

frog breeding habitat, which consists of short, sparse vegetation. Non-herbaceous vegetation was
re-classified as one class and herbaceous heights were classified into 4 categories of 0.2-meter
intervals from 0 meters to 0.8 meters. Due to the nature of this classification, this is an ordinal
variable, yet it is based on a discrete scale and so run as a continuous variable in the final model
(Phillips & Dudík, 2008).
Because of the aquatic nature of Oregon spotted frog, the hydric status of the soil was
included. This vector data was obtained from the Washington Department of Natural Resources
soil database and converted to raster format using the “Polygon to Raster” tool. The five
reconfigured structure variables were included in the final model (Table 4).
Climate Variables
Climate data was obtained from Worldclim as a collection of bioclimatic variables (Table
2). Bioclimatic variables are variations in monthly and annual temperatures and precipitation
amounts representing meaningful trends or limiting factors in ecology (Fick & Hijmans, 2017).
Trends were calculated based on data collected in the 30-year period between 1970 and 2000
(Fick & Hijmans, 2017). Average solar radiation was obtained for February and March, the
period of Oregon spotted frog breeding represented by the presence points.

34

Table 2. Climate variables used in the Maxent model obtained from Worldclim and their original
resolution.
Variable
Bioclimatic Variables
Annual Mean Temperature
Mean Diurnal Range
Isothermality
Temperature Seasonality
Max Temp of Warmest Month
Min Temp of Coldest Month
Temperature Annual Range
Mean Temp of Wettest Quarter
Mean Temp of Driest Quarter
Mean Temp of Warmest Quarter
Mean Temp of Coldest Quarter
Annual Precipitation
Precipitation of Wettest Month
Precipitation of Driest Month
Precipitation Seasonality
Precipitation of Wettest Quarter
Precipitation of Driest Quarter
Precipitation of Warmest Quarter
Precipitation of Coldest Quarter
Solar radiation February
Solar Radiation March

Climatic Variables
Unit
Original Resolution
30 arc-sec (~1km²)

Source
WorldClim
(Fick & Hijmans, 2017)

°C

mm

kJ m-2 day-1

All climate variables were obtained in an original resolution of 30 arc-sec, covering about
1 square kilometer. To match the resolution of the structure variables, each variable was rescaled
to a 30-meter resolution using the “Resample” tool in ArcGIS Pro. The finer resolution (30meter) was chosen over the coarser (1 kilometer) in order to retain the level of detail in the
structure variables. To select the final climate variables, an intermediate model was created
containing all 21 climate variables (Table 3). It is sometimes advised to remove correlated
variables if they are known to be irrelevant; however, Maxent will remain stable when using
correlated variables (Elith et al., 2011). The climatic conditions for Oregon spotted frog are less
understood than the structural requirements, therefore all of the variables were included and then
chosen for the final model based on their importance to this intermediate model. Maxent grades
35

variables on two metrics of their importance to the model, the percent contribution and the
permutation importance. Out of the 21 variables in the intermediate model, those with a
permutation importance above 5% were chosen for the final model.
Table 3. The percent contribution and permutation importance of all 21 climate variables when
run in a Maxent model.
Variable
Precipitation of Driest Quarter
Precipitation of Wettest Month
Solar radiation February
Mean Temp of Coldest Quarter
Temperature Annual Range
Annual Mean Temperature
Mean Temp of Wettest Quarter
Precipitation of Warmest Quarter
Min Temp of Coldest Month
Precipitation Seasonality
Mean Diurnal Range
Precipitation of Wettest Quarter
Solar Radiation March
Precipitation of Driest Month
Isothermality
Temperature Seasonality
Precipitation of Coldest Quarter
Mean Temp of Warmest Quarter
Mean Temp of Driest Quarter
Max Temp of Warmest Month
Annual Precipitation

Percent Contribution
22.3
18.3
14.1
12.5
8.8
8.4
3.6
2.9
2.7
2.7
1.3
1.2
0.6
0.3
0.2
0
0
0
0
0
0

Permutation Importance
20
14.9
1.9
10.6
26
19.9
2.6
0
1.4
1.1
0.2
1.1
0.2
0
0.3
0
0
0
0
0
0

The percent contribution of each variable is calculated by Maxent as it is creating the
model (Phillips, 2017). As each variable is included in the algorithm, it increases the training
gain. The training gain is a measure of the model’s ability to differentiate background points
from a presence point (Merow et al., 2013). The model starts as a uniform distribution or
maximum entropy, of suitable habitat. As variables are included, they add constraints, which are
defined by the values at the presence points, so the distribution beings to shrink. The gain is how
much this distribution shrinks around the presence points. As the gain increases with the addition
of variables, the model becomes more defined according to the presence points. The percent

36

contribution of each variable is a measure of how much they increase the gain when the model is
being created (Phillips, 2017; Phillips et al., 2006).
The permutation importance is defined by the variable’s influence on the final model.
The area under the receiving operator curve (AUC) is a measure of the model’s ability to
differentiate between suitable and unsuitable habitat (Merow et al., 2013). To measure the
permutation importance of each variable, the values of that variable at each training point are
randomly permuted. When these values change, the resulting AUC will change as well (Songer
et al., 2012). A decrease in AUC indicates a decrease in the model’s predictive ability. The
decrease is converted to a percentage and is reported as the permutation importance. A large
decrease in AUC indicates that the model is relying heavily on that variable to differentiate
suitable and unsuitable habitat (Phillips, 2017). Permutation importance is a better indicator of
variable importance to the final model compared to the percent contribution (Songer et al.,
2012). There were five climatic variables with a permutation importance above five percent:
precipitation of the driest quarter, precipitation of the wettest month, mean temperature of the
coldest quarter, temperature annual range, and annual mean temperature. These variables were
selected to be used as the climate variables in the Maxent model (Table 4).

37

Table 4. Final structure and climate variables representing Oregon spotted frog breeding habitat
used in the Maxent model.
Structure Variables
Variable
Slope
Landcover
Non-habitat
Agriculture/Pasture
Emergent Herbaceous Wetland
Proximity to Cover Class
Herbaceous Vegetation Height
Non-herbaceous
0 - 0.2 meters
0.2 - 0.4 meters
0.4 - 0.6 meters
0.6 - 0.8 meters
Hydric Soil
Non-hydric
Hydric

Unit
Degrees
Categorical

Data Type
Continuous
Categorical

Meters
Meters

Continuous
Continuous

Categorical

Categorical

Climate Variables
Variable
Precipitation of Driest Quarter
Precipitation of Wettest Month
Mean Temperature of Coldest Quarter
Temperature Annual Range
Annual Mean Temperature

Unit
Mm
Mm
°C
°C
°C

Data Type
Continuous
Continuous
Continuous
Continuous
Continuous

When running the model in Maxent, the options to create response curves and variable
jackknives were selected. Response curves chart the probability of presence with the values
within each environmental variable, thus displaying the bounds of suitable habitat for each
variable (Phillips, 2017). The variable jackknife is an additional indicator of importance of each
variable to the model. A model is created with each variable in isolation and the increase in
training gain is calculated. A large increase in training gain indicates that the variable has
information useful for training the model. An additional model is created by removing each
variable and calculating the decrease in training gain. A large decrease means the variable has
information that is not found in the other variables (Phillips, 2017).

38

Modelling
Using Maxent, version 3.4.4 (Phillips et al., 2020), Oregon spotted frog breeding habitat
was modelled for the Chehalis Basin. The 29 spatially rarefied points in the Black River, and the
1.5-kilometer buffer around each point were used for the presence points and background point
selection bias file. The environmental variables were the five structure variables and five climate
variables detailed in Table 4. The default settings in Maxent were used which include 10,000
background points, linear, quadratic, product and hinge features, and a regularization multiplier
of 1 (Phillips & Dudík, 2008). The default output format, complimentary log-log (cloglog), was
chosen for the final model output (Phillips et al., 2017). This output format displays the
probability of presence of each cell based on the estimate of suitable habitat (Phillips et al.,
2017).

Model Evaluation
Model evaluation is a critical step to determine the predictive ability of the model when
presented with data other than the training data. Commonly, when there are sufficient presence
points, a subset of these points is removed to be used as testing data. The training points (the
ones left) are used to build the model and the testing points (the ones removed) are used to test
the model’s ability at determining if they are suitable habitat. The AUC is a common metric of
evaluating the model’s success at differentiating test points from background points (Merow et
al., 2013; Pearson et al., 2007). However, when sample sizes are small (about 25 or less) this
method is less robust and a different evaluation approach was developed (Pearson et al., 2007;
Shcheglovitova & Anderson, 2013).
When the sample size of the presence points is small, removing a portion of the points for
testing may limit the model’s predictive ability due to the loss of information from those points.
39

Therefore, a jackknife, or leave-one-out, approach has been developed for small samples
(Pearson et al., 2007). This approach to model evaluation retains all of the presence points as
training points, save one. The remaining point is used as the test point. The model is trained with
the remaining presence points and then determines the habitat suitability of the testing point.
Suitability is determined by a threshold based on the values of the training points. A value
greater than the threshold is considered a success and an omission if it is less. The test point is
replaced and a new test point is removed from the presence points. This process is repeated for
every point (Groff et al., 2014; Pearson et al., 2007). The results of each iteration are compiled
and the success and omission rates are calculated for the model. A p-value is computed using
pvalueCompute, a software developed by Pearson et al 2007. A significant p-value indicates that
the success rate of the model using the testing points is better than randomly using a set of
background points (Pearson et al., 2007).
Due to the limited number of samples (n=29) in the Black River, the Maxent model was
evaluated using the leave-one-out approach. The model was run 29 times and the success of each
iteration at predicting the testing point was evaluated at two different thresholds, the minimum
training presence and the 10-percentile training presence. While all are considered suitable
habitat, training points are ranked based on their values according to the habitat suitability index
created by the model. The thresholds are based on these rankings. The minimum training
presence threshold is set at the lowest valued training point (Pearson et al., 2007). If the testing
point has a suitability value less than the lowest ranked training value, it is an omission. The
threshold for the 10-percentile training presence is set at the lowest ten percent of the training
points (Pearson et al., 2007). The significance of the success rate at each threshold was evaluated
using the pvalueCompute software developed by Pearson et al 2007.

40

Watershed Comparison
Chehalis River Survey Point Selection and Comparison Criteria
The purpose of the watershed comparison is to assess the habitat suitability of surveyed
sites in the Chehalis River mainstem floodplain based on the habitat conditions of Oregon
spotted frog breeding habitat in the Black River. No historical records of Oregon spotted frogs
exist in the Chehalis Basin outside of the Black River; survey efforts in the Chehalis River
floodplain have not been successful at locating new populations. Using Maxent to model suitable
habitat across the Chehalis Basin will allow for the comparison of the habitat characteristics of
the Chehalis survey points to those of known Oregon spotted frog breeding locations in the
Black River. Differences in these characteristics may provide insight into explaining the
presumed absence of the frog in the Chehalis River.
The locations of survey sites in the Chehalis River floodplain were acquired from
Washington Department of Fish and Wildlife biologists. Surveys for pond breeding amphibians
were conducted from 2013 to 2017 (Holgerson et al., 2019). The survey data was a GIS
containing polygons delineating each wetland that was surveyed. The “Feature to Point” tool in
ArcGIS was used to generate a point within each survey polygon to represent each survey site.
There were 189 survey points.
The Maxent model output predicted the probability of presence of Oregon spotted frog
across the Chehalis Basin based on an index of habitat suitability derived from the environmental
characteristics of the presence points in the Black River. Survey points selected for comparison
were chosen according to a threshold of probability of presence above 0.7. Because the
probability of presence is associated with habitat suitability, points selected at a threshold of 0.7
are expected to have moderate to high habitat suitability. The points selected at the 0.7 threshold
41

were then filtered by size and form. Wetlands smaller than 0.4 hectares or classified as creeks
were removed. Oregon spotted frog breed in lentic water, therefore survey points in creeks are
not suitable as breeding habitat. Points classified as ponds or oxbows were retained for use in the
analysis. Frogs also tend to occupy large wetland complexes, therefore survey sites smaller than
0.4 hectares were removed. While still not very large in isolation, many of the sites 0.4 hectares
and larger are connected aquatically to others, resulting in larger wetland complexes. Under this
assumption, the 0.4-hectare threshold was used to avoid being too restrictive in the selection of
survey points. There were 47 survey points that met the filtering criteria, 29 of which were
randomly chosen to be compared to the 29 presence points in the Black River (Figure 3).

42

Figure 3. Location of Chehalis River mainstem floodplain survey points and Black River
presence points used in the comparison analysis. Survey points were selected based on their
predicted probability of presence according to the Maxent model (> 0.7).
At each Chehalis River survey and Black River presence point, the values of the 10
environmental variables used in the Maxent model were obtained using the “Extract Multi
Values to Points” tool in ArcGIS Pro. This tool extracts the values of a raster dataset at the
specified point. Three additional variables describing the abundance and presence of exotic
species (bullfrogs and centrarchid fish species) were included for each survey and presence
point. These exotic species are especially detrimental to Oregon spotted frog persistence, and
their presence may represent another basis for the absence of frogs outside of the Black River
(Holgerson et al., 2019; Pearl et al., 2004). The abundance ranks of centrarchids and bullfrogs
43

were scored as 0, 1, or 2 according to the number of observed individuals at each presence or
survey point. The ranks were classified as absent (0, no observations), rare (1, less than 10
observations), and abundant (2, 10 or more observations). The frequency of presence of these
species was also included in this comparison. If each point had either a fish or bullfrog
occurrence the point was considered occupied by an exotic species.
Following the methods described above, two additional models were created solely for
the comparison analysis: a model using only the five structure variables and one using the five
climate variables (Appendix 1). The purpose of the variable specific models was to select survey
locations for comparison from the Chehalis River mainstem floodplain according to habitat
suitability based only on structure or climate. Survey points selected from the structure-only
model were compared to the Black River presence points by the five structure variables and the
points in the climate-only model were compared by the five climate variables. Both exotic
species abundance variables were included for the variable-specific models.

Statistical Analysis
The Black River presence points and Chehalis River survey points were compared by the
ten environmental variables (five structure and five climate) used in the Maxent model and two
variables describing the abundance of exotic species (centrarchids and bullfrogs). For the
continuous and ordinal variables, each was tested for a normal distribution with a Shapiro-Wilk
test. If they were normally distributed a two-sample t-test was conducted if the variances were
similar and a Welch two sample t-test was conducted if the variances were not. Three variables
met the assumption of a normal distribution: annual mean temperature, precipitation of the
wettest month and precipitation of the driest quarter, but only the variances of the annual mean
temperature were similar. Square root and log-transformations were attempted if the variables
44

were not normally distributed. however, no transformations were successful in creating a normal
distribution for any of the remaining continuous variables in either model. As a result, a MannWhitney U test was performed on the following variables: mean temperature of the coldest
quarter, the temperature annual range, the distance to cover class, herbaceous vegetation height,
slope, the abundance of centrarchids, and the abundance of bullfrogs. A Chi-squared test was
conducted on the categorical variables: landcover and hydric soil. Significance was established
when the p-values was < 0.05. All statistical analysis was conducted using the computing
software, R (R Core Team, 2020).

45

RESULTS
Maxent
A model was created using Maxent to predict the probability of presence of Oregon
spotted frog in the greater Chehalis Basin based on habitat suitability defined by the conditions
of presence points in the Black River. The habitat of the presence points was defined by ten
environmental variables, five of which describe the structure of the habitat and five describe the
climate.

Model Evaluation
The Maxent model created for the Chehalis Basin was evaluated using the leave-one-out
method. The model had a high predictive ability at both thresholds and was statistically
significant compared to a random selection of background points (p ≤ 0.001) (Table 5). The
success rate at the minimum training presence threshold was 93% with an omission rate of 7%.
The success rate at the 10-percentile training presence threshold was 83% with an omission rate
of 17%.
Table 5. The success rate, omission rate, and associated p-values of the Maxent model
developed for the Chehalis Basin. Success was evaluated at two thresholds, the minimum
training presence (MTP) and the 10-percentile training presence (10TP).
Maxent Model Evaluation
Threshold

Success Rate

Omission Rate

p-value

MTP

0.931034

0.068966

0.000

10TP

0.827586

0.172414

0.000

Predicted Distribution
The Maxent model predicted the distribution of suitable breeding habitat for the Oregon
spotted frog not only in the Black River, where frogs currently persist, but in the greater Chehalis
46

Basin as well. Because no records of Oregon spotted frog in the Chehalis Basin exist outside of
the Black River watershed, this distribution may potentially be mapping the historic presence of
the frog if they ever occurred outside of the Black River. Analysis of the environmental
conditions at these locations may reveal factors for their current, presumed absence. The cloglog
output format predicts the probability of presence based on habitat suitability. The higher the
probability, means the greater the estimate of suitable habitat.
In the Black River watershed, the largest concentration of highly suitable habitat was
located along Dempsey Creek, Blooms Ditch, Allen Creek and along the Back River between
Blooms Ditch and Waddell Creek (Figure 4). Moderate to high, but more dispersed, habitat was
predicted along Salmon Creek, Beaver Creek and the southern stretches of the river between
Shaner Creek and Mima Creek. Additional moderate habitat was predicted at the southern point
of Black Lake, the source of the Black River and west of Offutt Lake on the eastern edge of the
watershed. Much of the predicted habitat occurred around the presence points and was not
distributed into many areas not already known to have frogs.

47

Figure 4. Suitable breeding habitat for the Oregon spotted frog in the Black River watershed
according to the Maxent model. The warmer colors represent an increased probability of
presence based on the estimate habitat suitability.
The highest concentration and highest probability of presence in the entire Chehalis Basin
occurred in the tributaries flowing into the Chehalis River in the upper reaches of the watershed
(Figure 5). The Skookumchuck River, Hanaford Creek, Salzer creek and the Newakum River
were all predicted to contain highly suitable habitat. Much of the habitat along these tributaries
moving eastwards from the Chehalis River have a probability of presence above 0.8. Additional
moderate suitable habitat was found along Lincoln and Bunker Creeks, tributaries on the west
side of the Chehalis River. In the upper Chehalis Basin, little suitable habitat was predicted along

48

the Chehalis River itself, other than at the confluences of the eastern tributaries. However, this
trend changes moving down river into the lower reaches of the watershed.

Figure 5. Suitable breeding habitat for the Oregon spotted frog in the Upper Chehalis River
Basin. The warmer colors represent an increased probability of presence based on the estimated
habitat suitability.
Suitable habitat in the lower Chehalis Basin was confined to the floodplain of the
Chehalis River (Figure 6). Most of the highest suitable habitat along this stretch was located
among the sloughs, specifically, Wenzel Slough at the confluence of the Satsop River, Metcalf
Slough, and the assemblage of sloughs near the mouth of the Chehalis River at Grays Harbor,
including Elliot, Mox Chuck and Blue sloughs. Additionally, there was scattered habitat along
the Wynoochee and Humptulips Rivers but these were not predicted to have a high probability of
presence. Oregon spotted frog prefers large wetland complexes with gentle gradients. In this
49

stretch of the Chehalis River, the tributaries are flowing out of the Olympic Mountains at much
higher elevations and steeper slopes, thus diminishing the available habitat.

Figure 6. Suitable breeding habitat for the Oregon spotted frog in the Lower Chehalis River
Basin. The warmer colors represent an increased probability of presence based on the estimate
habitat suitability.

Environmental Variable Importance
Maxent evaluates the importance of variables to the model in multiple ways: the percent
contribution of each variable as the model is being trained (percent contribution), the importance
of each variable to the final model’s predictive ability (permutation importance), and how the
variability within each variable defines habitat suitability and the resulting probability of

50

presence (response curves). The evaluation of variables can describe the influence they have on
the distribution and ecology of the species, yet they should be considered with caution. The
importance of the variable may reflect the Maxent algorithm and not necessarily the preferences
of the species.
The distance to cover class variable contributed the most to the Maxent model with a
permutation importance of 65% and a percent contribution of 35.8%. This variable contributed
the most to the training gain when creating the model and is highly relied upon for the final
model’s predictive ability. The hydric soil variable contributes substantially less to the final
model’s predictive ability with a permutation importance of 14.9%; however, it had a high
contribution to the training gain with a 27.5 percent contribution. Landcover had a similar
percent contribution to hydric soil with 27.9%, however it only has a permutation importance of
3.4%. Slope has the third highest permutation importance, but it is only marginal at 5.4% and
only a 2.4 percent contribution. The herbaceous vegetation height and all of the climate variables
have a permutation importance and percent contribution value less than 5%. These include: mean
temperature of the coldest month, precipitation of the wettest month, annual mean temperature,
temperature annual range, and precipitation of the driest quarter. Precipitation of the driest
quarter has a permutation importance and percent contribution of 0 percent (Table 6).

51

Table 6. The permutation importance and percent contribution importance of the environmental
variables in the Maxent model.
Permutation Importance and Percent Contribution of Variables to Maxent Model
Variable
Permutation Importance
Percent Contribution
Distance to Cover Class
65
35.8
Hydric Soil
14.9
27.5
Slope
5.4
2.4
Mean Temperature of Coldest Quarter
3.5
1.2
Landcover
3.4
27.9
Precipitation of Wettest Month
3.3
0.8
Annual Mean Temperature
1.8
0.2
Temperature Annual Range
1.6
0.6
Herbaceous Vegetation Height
1
3.5
Precipitation of Driest Quarter
0
0

The importance of the variables to the model are further highlighted in the jackknife
analysis (Figure 7). Maxent runs a model with either the variable in isolation (dark blue) or
excluded (light blue) and calculates the resulting training gain. The training gain is the measure
of how the model defines a presence point against a background point. The higher the training
gain, the more defined the model is by the presence points and is thus better able to distinguish
the habitat characteristics at those points (Merow et al., 2013). When excluded, the distance to
cover class variable decreased the training gain the most (light blue bar), indicating that this
variable provided the most information on the habitat characteristics of the presence points that is
not found in the other variables. Hydric soil decreased the training gain second to distance to
cover class when it was excluded from the model. These two variables have information
pertaining to the presence points that was not found in the other variables. Their high
permutation importance further supports this.

52

When each variable was run in isolation, landcover increased the training gain the most
(dark blue bar). By itself, landcover was better able to define the habitat characteristics of the
presence points compared to the other variables. When run in isolation, the set of structure
variables generally were better able to define the presence points compared the climate variables,
which perform poorly in terms of training gain when they were run by themselves.

Figure 7. Jackknife analysis of the Maxent variables. The light blue bars measure the training
gain of the model when that variable is excluded. The dark blue bars measure the training gain
when that variable is run in isolation.
The variable response curves identified the range of values within the variables and how
they contributed to the probability of presence in the Maxent model output. The probability of
Orgon spotted frog presence favored aquatic conditions and gentle gradients (Figure 8).
Probability increases as the distance to the preferred cover class decreases and becomes likely
(>50% probability of presence) at less than 15 meters (Figure 8a). The probability of presence
was highest among herbaceous vegetation taller than 0.2 meters, and probability increased until
the maximum vegetation height of 0.8 meters was reached, which was the tallest height in the
data (Figure 8e). While there was not a clear trend between the height classes, the probability of
53

presence was higher in every class compared to the non-herbaceous vegetation class. Although
the probability of presence was 40% in the agriculture cover class, probability increased to over
90% when located in the emergent wetland cover class (Figure 8b). Finding Oregon spotted
frogs at high slopes was highly improbable but started becoming increasingly probable once
slope dropped below 3˚, with a much higher probability at slopes less than 1° (Figure 8c). Soil
type was important as well, with an 80% probability of presence within hydric soils (Figure 8d).

54

Structure Variable Response Curves
(a)

(b)

PROBABILITY OF PRESENCE

PROBABILITY OF PRESENCE

Distance to Cover Class
1
0.8
0.6
0.4
0.2
0
0.89

1225.01

Landcover
1.0
0.8
0.6
0.4
0.2
0.0
Non-Habitat

Emergent
Wetland

Cover Class

DISTANCE (METERS)

(d)

PROBABILITY OF PRESENCE

Slope
1
0.8
0.6
0.4
0.2
0
0.02

27.89
DEGREES

PROBABILITY OF PRESENCE

(c)

Agriculture

Hydric Soil
1.0
0.8
0.6
0.4
0.2
0.0
Non-Hydric

Hydric

HYDRIC STATUS

(e)

PROBABILITY OF PRESENCE

Herbaceous Vegetation Height
1.0
0.8

0.6
0.4
0.2
0.0
Non-Herbaceous

0 - 0.2

0.2 - 0.4

0.4 - 0.6

0.6 - 0.8

HEIGHT (METERS)

Figure 8. Response curves plotting the probability of presence based on the variation in the
structural variables used in the Maxent model. The variables include (a) distance to cover class,
(b) landcover, (c) slope, (d) hydric soil status, (e) herbaceous vegetation height.
55

The probability of presence, according to the climate variables, trended towards
increasing probability with warmer and drier conditions (Figure 9). Presence became more likely
when the annual mean temperature was greater than 10.15 °C (Figure 9a). Probability increased
until the mean temperature of the coldest quarter reached 4.28 °C and then leveled off at 65%
(Figure 9b). Probability began to increase as the temperature annual range began to widen
beyond 24.47 °C (Figure 9e). The probability of presence increased with drier conditions.
Probability began to decrease as the precipitation of the driest quarter exceeded 91 millimeters
and the precipitation of the wettest month exceeded 200 millimeters (Figure 9 c and d).

56

Climate Variable Response Curves
(b)

Annual Mean Temperature

Mean Temp Coldest Quarter

1
0.8
0.6
0.4

0.2
0
9.79

10.57

PROBABILITY OF PRESENCE

PROBABILITY OF PRESENCE

(a)

1
0.8
0.6
0.4
0.2
0
4.10

TEMPERATURE (°C)

4.66
TEMPERATURE (°C)

(c)

(d)

Precipitation Wettest Month

1
0.8
0.6
0.4
0.2
0
90.18

114.31
PRECIPITATION (MILLIMETERS)

PROBABILITY OF PRESENCE

PROBABILITY OF PRESENCE

Precipitation Driest Quarter

1
0.8
0.6

0.4
0.2
0
195.64

PRECIPITATION (MILLIMETERS)

250.07

(e)

PROBABILITY OF PRESENCE

Temperature Annual Range
1
0.8
0.6
0.4
0.2
0
23.36

25.73
TEMPERATURE (°C)

Figure 9. The response curves plotting the probability of presence based on the variation in the
climate variables used in the Maxent model. Variables include (a) annual mean temperature, (b)
mean temperature coldest quarter, (c) precipitation driest quarter, (d) precipitation wettest month,
(e) temperature annual range.

57

Watershed Comparison
To assess the reason for the apparent absence of Oregon spotted frog in the Chehalis
Basin, outside of the Black River, the values of the ten environmental variables used in the
Maxent model and the presence and abundance of two exotic species were compared between
sites surveyed in the Chehalis River mainstem floodplain to populations in the Black River. The
Chehalis survey sites were selected based on moderate to high habitat suitability predicted by the
Maxent model.
There was no statistical difference (p < 0.05) in the structural variables between points in
the Chehalis and points in the Black (Table 7). These variables describe the physical habitat
structure and include: the distance to cover class, the herbaceous vegetation height, landcover,
slope, and hydric soils (Table 8).
The five climate variables were significantly different (p < 0.05) between the rivers
(Table 7). The Chehalis points were warmer, wetter and narrower in temperature range compared
to the Black (Tables 8). The annual mean temperature was 0.09 °C warmer and the mean
temperature of the coldest quarter was 0.53 °C warmer in the Chehalis compared to the Black.
There was 50.5 millimeters more precipitation during the wettest month and 24.45 millimeters
more precipitation in the driest quarter in the Chehalis compared to the Black. The annual
temperature range was 1.42 °C narrower in the Chehalis compared to the Black (Table 9).
The abundance and presence of centrarchids and bullfrogs, were significantly different
(p< 0.05) between the watersheds (Table 7). The abundance of centrarchids was higher in the
Chehalis River floodplain with an average abundance rank of 0.79 compared to 0.10 in the
Black. The abundance of bullfrogs was higher in the Chehalis as well with an average abundance

58

rank of 1.59 compared to 0.45 in the Black (Table 8). Exotic species were present at 90 percent
of the Chehalis River points compared to 45 percent of the Black River points (Table 8).
Table 7. Results of the comparison analysis of environmental and exotic species variables
between the Black River presence points and the Chehalis River floodplain survey points.
Significant differences are in bold (p < 0.001).
Black River Presence Points and Chehalis River Survey Points Comparison Results
Variable
Test
df
Statistic
Structure Variables
Distance to Cover Class
Mann-Whitney U test
W = 420.5
Herbaceous Vegetation Height
Mann-Whitney U test
W = 401.5
Landcover
Chi-squared test
2
X² = 4.2125
Slope
Mann-Whitney U test
W = 444
Hydric Soil
Chi-squared test
1
X² = 0
Climate Variables
Annual Mean Temperature
Two Sample t-test
56
t = -3.7316
Mean Temperature of Coldest Quarter
Mann-Whitney U test
W= 0
Precipitation of Wettest Month
Welch Two Sample t-test
29.097
t = -6.8479
Precipitation of Driest Quarter
Welch Two Sample t-test
29.433
t = -9.1964
Temperature Annual Range
Mann-Whitney U test
W = 799
Exotic Species Abundance Variables
Centrarchid Abundance
Mann-Whitney U test
W = 247.5
Bullfrog Abundance
Mann-Whitney U test
W = 102
Exotic Presence
Mann-Whitney U test
W=609

p-value
1.000
0.754
0.122
0.710
1.000
0.000
0.000
0.000
0.000
0.000
0.001
0.000
0.000

59

Table 8. Summary of environmental and exotic species variables used in the comparison
analysis between the Black River presence points and the Chehalis River floodplain survey
points. Statistical differences are in bold.
Summary Statistics of Black River Presence Points and Chehalis River Survey Points
Black
Variable

Chehalis

Mean

St Dev

Mean

St Dev

Distance to Cover Class (Meters)

6.21

12.37

6.21

12.37

Herbaceous Vegetation Height (Height Rank)

1.24

1.12

1.38

1.42

Structure Variables

Landcover
Slope (Degrees)

Categorical
0.54

Hydric Soil

0.57

0.51

0.59

Categorical

Climate Variables
Annual Mean Temperature (°C)

10.31

0.09

10.41

0.10

Mean Temperature of Coldest Quarter (°C)

4.32

0.09

4.84

0.15

Precipitation of Wettest Month (Millimeters)

212.15

5.52

262.74

39.40

Precipitation of Driest Quarter (mm)

96.07

2.26

120.52

14.14

Temperature Annual Range (°C)

24.69

0.40

23.26

0.87

Centrarchid Abundance (Abundance Rank)

0.10

0.31

0.79

0.90

Bullfrog Abundance (Abundance Rank)

0.45

0.51

1.59

0.68

45

0.51

90

0.31

Exotic Species Abundance Variables

Exotic Presence (Percent Occurrence)

60

Table 9. The magnitude of change between statistically different variables between the Back
River presence points (B) and the Chehalis River floodplain survey points (C).
Magnitude of Change in Statistically Different Variables
Variables

Difference

Direction

Annual Mean Temperature (°C)

0.09 °C

C warmer than B

Mean Temperature of Coldest Quarter (°C)

0.53 °C

C warmer than B

Precipitation of Wettest Month (Millimeters)

50.50 mm

C wetter than B

Precipitation of Driest Quarter (Millimeters)

24.45 mm

C wetter than B

1.42 °C

C narrower than B

Centrarchid Abundance (Abundance Rank)

0.69

C more abundant than B

Bullfrog Abundance (Abundance Rank)

1.14

C more abundant than B

Exotic Presence (Percent Occurrence)

45%

C more occurrences than B

Climate Variables

Temperature Annual Range (°C)
Exotic Species Abundance Variables

The categorical variables, landcover class and soil hydric status, were independent of the
watershed the points occurred in. The frequency of points classified as each landcover class was
not statistically different between the Black and Chehalis (Table 10). The emergent wetland
landcover class had the highest frequency of points with 17 in the Black and 22 in the Chehalis.
There were six points in the Black River classified as agriculture but only one in the Chehalis.
However, there were six points in both the Black River and the Chehalis River floodplain
classified as non-habitat (Table 10a). The frequency of non-hydric and hydric soils at the Black
River and Chehalis points was the same (Table 10b).
Table 10. Contingency tables for (a) landcover and (b) hydric soils, the categorical variables
used in the Maxent model.
(a)
Landcover Class Frequency
Black
Chehalis
Non-habitat
6
6
Agriculture
6
1
Emergent Wetland
17
22

(b)
Hydric Soil Frequency
Black
Chehalis
Non-Hydric

6

6

Hydric

23

23

61

Exotic species were more prevalent and abundant in the Chehalis watershed. The
abundance of centrarchids at the Chehalis River floodplain survey points were ranked as rare and
abundant more often than the Black River presence points, which were more often ranked as
absent (Figure 10). The abundance ranks of bullfrogs at the Chehalis River points were more
often ranked as abundant, while the Black River points were more often ranked as absent or rare
(Figure 11).

Centrarchid Abundance Ranks
30

# of Points

25

20
15
10
5

0
Absent

Rare

Abundant

Abundance Rank
Black

Chehalis

Figure 10. The frequency of abundance ranks of centrarchid fishes at the Black River presence
points and Chehalis River floodplain survey points.

62

Bullfrogs Abundance Ranks
30

# of Points

25
20
15
10
5
0

Absent

Rare

Abundant

Abundance Ranks
Black

Chehalis

Figure 11. The frequency of abundance ranks of bullfrogs in the Black River presence points
and Chehalis River floodplain survey points.
When survey sites were selected from the structure-only and climate-only models, and
their respective variables were compared, the results were the same. There was no statistical
difference between the structure variables at the structure-only points and the Black River points.
There was a significant difference between the climate variables at the climate-only selected
survey points and the Black River presence points. The abundance of exotic species was
statistically different between the Chehalis River points in both models when compared to the
Black River points (Appendix).

63

DISCUSSION
The Oregon spotted frog, is a federally threatened, and Washington state endangered
species due to a large reduction in historical range. As such, one of the recovery objectives is to
locate additional populations to increase the current knowledge of the frog’s distribution
(Hallock, 2013). There are six watersheds in Washington that are currently known to have
populations of Oregon spotted frog, one of which, the Black River, is a tributary of the Chehalis
River. Due to the aquatic connectivity of the Black and Chehalis rivers, it could be likely that the
frogs occupy additional watersheds in the Chehalis Basin, however no records of occurrences
outside of the Black River have been documented. To determine potential causes for the absence
of Oregon spotted frog from the Chehalis Basin, a Maxent model was created to predict the
distribution of suitable habit. That model was based off the habitat of known frog breeding
locations in the Black River, and therefore similar conditions were modelled in the rest of the
basin. Using this habitat distribution as a selection guide, previous survey sites in the Chehalis
River mainstem floodplain were chosen that met a moderate to high level of suitability according
to the habitat conditions of known breeding locations. Although these sites were predicted to
have habitat characteristics similar to the occupied locations, all surveys in the Chehalis River
floodplain have thus far been unsuccessful at locating Oregon spotted frog. The apparent absence
of frogs at these sites is consistent with the historical records, as no frogs have been reported in
the Chehalis Basin, except in the Black River watershed, which were not discovered until 1990.
It can be hypothesized that frogs did once occur in parts of the greater Chehalis Basin, yet by the
time they had been discovered in the Black River in 1990, conditions in the remainder of the
basin may have been altered to their detriment and eventual extirpation. Therefore, after using

64

the Maxent model to identify locations that are predicted to have similar habitat in the Chehalis
Basin, a comparison between the survey points in the Chehalis River floodplain found within
habitat considered to be suitable and presence points in the Black River may offer insights as to
how these conditions may differ between the Black and Chehalis Rivers. The locations were
compared by three criteria: the habitat structure, the climate, and the abundance of exotic
species. According to this comparison, habitat structure does not appear to be a limiting factor,
yet the climate and abundance of exotic species differ between the two rivers.
The Maxent distribution of suitable habitat extended well beyond the boundary of the
Black River watershed, despite no records of Oregon spotted frog occurring elsewhere in the
Chehalis Basin. This output may therefore serve as a historical distribution of frogs in the
Chehalis Basin, if they ever occurred outside of the Black River. According to the model
evaluation, the Maxent model had a high predictive ability at the minimum training presence
threshold and a moderately high ability at the 10-percentile training presence threshold with a
success rate of 93% and 83% respectively. When determining the importance of the
environmental variables, the structure variables contributed the most to the model in both
permutation importance and percent contribution. Therefore, the model output is being driven
mainly by the conditions of the habitat structure of the Black River presence points and predicted
sites across the rest of the Chehalis Basin can be expected to have similar structure based on the
variables provided. Indeed, the structure variables at the Chehalis River floodplain survey points
and the Black River presence points were not statistically different according to the comparison
analysis. This holds true in the complete model as well as the survey points selected for the
structure-only model (Appendix).

65

Although there was not a statistical difference between the structural variables across the
basin, it is worth discussing whether the desirable habitat structure is consistent with that
presented in the literature for Oregon spotted frog. Probability was highest when points were
located within or near emergent wetlands, the preferred habitat types for Oregon spotted frog
(Watson et al., 2003). Emergent wetlands provide the appropriate hydrology and vegetation
structure needed for the diverse, seasonal habitat requirement of the frog (Watson et al., 2003).
While habitat requirements for the frog may differ between season and life cycle stage, all must
be aquatic and the Maxent model predicted a higher probability of presence within hydric soils.
Hydric soils are formed under anaerobic conditions due to permanent water or seasonal flooding
and therefore an indicator of aquatic conditions (Ecology, 1997). Frogs breed in seasonally
flooded, shallow areas and reside in permanent water in the non-breeding season; therefore, this
prediction is consistent with the frog’s lifecycle. The seasonal flooding of Oregon spotted frog
breeding habitat may extend beyond the boundary of what is classified as emergent wetland in
the data, however, the probability of presence decreases when the distance increases away from
the preferred cover classes. Presence is most likely when within the preferred cover class or vey
near it.
The vegetation height variable, while less clear in determining the preference of the frog,
still makes an important distinction between vegetation type. Oregon spotted frog breed among
short-statured and sparse herbaceous vegetation and avoid wooded areas or areas with dense
cover (Watson et al., 2003). The model predicted a high likelihood of presence at most of the
herbaceous height classes and was less likely when located in the non-herbaceous vegetation
class. The preference for frogs is to breed among short vegetation, yet the model predicted
suitable habitat at every height class up to one meter, the highest height class in the data. This

66

discrepancy could be a result of the data itself. The vegetation height data describes the dominant
vegetation type and height of each 30 square meter cell of the raster layer, yet there is
opportunity for variation in heights of each cell (LANDFIRE, 2016a). For example, a cell
designated as being dominated by 0.8-meter-tall herbaceous vegetation may still have sufficient
areas with variation in height to provide plenty of microsites for breeding habitat. What is
significant about the prediction within this variable is that cells designated as non-herbaceous
vegetation had a low probability of presence compared to cells classified as herbaceous
vegetation. While the exact vegetation height may be lost in the data, the vegetation type is
consistent with the preference found in the literature.
Confidence should be high that the model is predicting suitable habitat structure for
Oregon spotted frog in the Chehalis Basin. According to the permutation importance of the
variables and the model evaluation, the model relied on structure variables the most for its
prediction and had a moderately high success rate at detecting test points as suitable habitat.
Additionally, the probability of presence increased within the values of each structure variable to
mimic the habitat requirements recorded in the literature. Finally, there was no statistical
difference between the Chehalis and Black River structure variables. However, despite the
predicted availability of structural habitat, there are no occurrences of frogs in the Chehalis Basin
and consideration of the structure variables is therefore needed.
Oregon spotted frog have very specific requirements for suitable breeding habitat and
while the structure variables were chosen for the model in an attempt to emulate these ecological
needs, there are aspects difficult to capture in a way that can be incorporated into the Maxent
model, specifically the hydrology. The need for seasonally inundated shallow, lentic water at
breeding sites is a critical determinant for Oregon spotted frog (Watson et al., 2003). However,

67

modelling the variability of seasonal inundation on a site-by-site basis, as well as a year-by-year
basis can be a challenge for reliable predictions. Besides the variability of the natural hydrology,
the level of land use alteration, including the channeling of waterways and habitat conversion to
agriculture and urban areas may be markedly different between the Chehalis and the Black River
watersheds. The Black River contains one of the largest intact emergent wetland system in the
Puget Sound, while the wetlands of the Skookumchuck and Newakum Rivers, Chehalis River
tributaries predicted to be highly suitable habitat, have been reduced by up to 75 percent (Species
Restoration Plan Steering Committee, 2019). Increased development and agriculture can result in
the channeling of waterways for wetland conversion and flood mitigation, which occurs
frequently in stretches of the Chehalis River (Ecology, 2016; McAllister & Leonard, 1997).
Additionally, the invasion of reed canary grass and the subsequent alteration of the vegetation
structure of many wetlands, creates a situation where Oregon spotted frog is reliant on microsites
found in opportune openings in the grass (Hallock, 2013; Kapust et al., 2012). The landcover and
vegetation data used in the model was classified by vegetation systems and not individual
species. Therefore, areas classified as emergent freshwater wetlands, the highest predictive land
cover class, may be infested with reed canary grass across the entire basin and the number of
microsites available to the frogs can be highly variable between locations. While the Maxent
model was highly predictive using the variables provided for the model, individual site visits will
be necessary to validate the suitability for Oregon spotted frog. Despite the discrepancies
between the data and what may actually be available for Oregon spotted frog, the model
predicted suitable habitat at wetlands that were independently surveyed by Washington
Department of Fish and Wildlife biologists and determined to be suitable habitat. Even with the

68

field validation of these sites by biologists, Oregon spotted frog was not observed in these
wetlands and their absence may be explained by factors other than habitat structure.
According to the comparison analysis, the climate of the Chehalis River floodplain
survey points is statistically different than the Black River presence points. The Chehalis River
points were wetter and warmer than the Black, yet the Maxent model predicts a higher
probability of presence when the climate is warmer and drier. The annual mean temperature of
the Chehalis Basin follows an elevational gradient and decreases as elevation increases
(Appendix 13a). The higher temperatures are located along the Chehalis River valley and lower
elevations of the tributaries matching much of the distribution of predicted suitable habitat from
the Maxent model. While they are statistically different, the variation in the average annual mean
temperature between the Chehalis and Black River points is only 0.09 degrees C. According to
the response curves, the probability of presence is highest when the temperature is above 10
degrees C and both the Chehalis and Black River points have a mean temperature higher than
this value. In contrast to the annual mean temperature, the temperature annual range follows an
east-west gradient and narrows as the Chehalis River flows westwards to the coast, resulting in
the mean temperature of the coldest quarter becoming warmer along the coast than the eastern
portion of the Basin, where the Black River is located (Appendix 13b and c). The narrower
temperature range provides a more moderated climate condition therefore, a warmer winter on
the coast than inland. Similar to the annual mean temperature, while the survey and presence
points are statistically different, the means of the temperature annual range and mean
temperature of the coldest quarter only vary by 1.42 and 0.53 °C. The temperature variables are
statistically different but they are not necessarily inhospitable to frogs. Based on the range of
Oregon spotted frog, this variation does not seem likely to be responsible for the absence of frogs

69

in the Chehalis River floodplain as frogs survive in more extreme climates within their range.
For example, the winter temperatures at Conboy Lake, the largest remaining population of
Oregon spotted frog, regularly reaches temperatures below freezing (Hayes et al., 2001). On the
other end of the temperature spectrum, frogs prefer warmwater wetlands, occupying summer
water temperatures above 20 °C. If the temperature is more moderate closer to the coast, perhaps
the water temperature does not get warm enough for frogs (Hayes, 1994). As long as there is
permanent water available throughout the dry season, frogs may be able to persist in warmer
climates.
The precipitation in the Chehalis Basin follows an east-west gradient with less rainfall
during both the wettest month and the driest quarter occurring in the eastern portion of the basin
(Appendix 13d and e). Oregon spotted frog is completely aquatic, yet the Maxent model predicts
the highest amount of suitable habitat in the driest tributaries. Granted, this prediction is most
likely based off of the habitat structure of these tributaries as the precipitation of the driest
quarter had zero percent contribution and permutation importance and the precipitation of the
wettest month had a permutation importance of only 3.3 and a percent contribution of 0.8.
Where climate may have an influence, but is missing from the model, is the effect on
water temperatures. The elevation of the headwaters of the tributaries may have an effect on
water temperatures, which could be a limiting factor for Oregon spotted frog, which use
warmwater habitat. Typically, frogs inhabit wetlands with water temperatures that exceed 20 °C
during the summer months (Hayes, 1994). The Black River originates from Black Lake at a low
elevation and remains so for its entire length before meeting the Chehalis River. The river is
exposed to the warmer, low elevation temperatures for its entirety. In contrast, the tributaries in
the north of the basin, such as the Wynoochee or Satsop rivers, originate in the Olympic

70

mountains, where temperatures are colder than the Black River and water temperatures may be
influenced by snowmelt as opposed to just rain. Additionally, in the northern tributaries, the
mean temperature cools off rapidly when moving north and increasing elevation from the
mainstem of the Chehalis River, which exposes greater lengths of the rivers to cooler
temperatures. In contrast, the annual mean temperature of the eastern tributaries, such as the
Skookumchuck, Newakum, and Black rivers, remains warmer for a greater distance when
moving upriver from their confluences at the Chehalis River. Additional data would be needed to
evaluate this hypothesis.
It is important to interpret the results of the Maxent model with caution, especially when
the variables responsible for the distribution of OSF are not entirely understood. This is
especially true for the climate favored by the Oregon spotted frog. Unlike habitat structure, the
climate of the Oregon spotted frog is less understood, therefore, variable selection was not
ecologically focused, but instead relied on the mathematical side of the relationship between
ecology and statistics in species distribution modelling. Predicting the climate of Oregon spotted
frog habitat in the Chehalis Basin may be limited by using a presence-only model. Because the
model only uses information from known locations and formulates the prediction off of the
conditions of these locations, when all of the occurrence samples are located in one end of a
spectrum, such as the precipitation gradient in the Chehalis Basin, it is only capable of predicting
habitat that falls in the conditions of that end of the spectrum. As a result, greater predictive
weight may be placed on the drier conditions of the eastern basin, as that is the conditions of the
presence points that the model was trained with. It cannot be said definitively if climate is a
limiting factor for Oregon spotted fogs in the Chehalis Basin, however based on this study, the
climate outside of the Black River watershed is statistically different and warrants further study.

71

A climate model that incorporates the climatic conditions of the entire range of Oregon spotted
frog may allow for the inclusion of more inputs and better predict the conditions favored by the
frog.
Besides climate being a potential limiting factor, the presence of exotic species is another
difference between the Black and Chehalis rivers. The greater abundance of exotic fish and
bullfrogs in the Chehalis River floodplain may be a direct cause of Oregon spotted frog absence
as the presence of these exotic species is associated with a negative abundance of native
amphibians (Holgerson et al., 2019). In the Chehalis River floodplain, 90% of the survey points
had occurrences of fish, bullfrogs or both compared to the Black River where only 45% of the
presence points had occurrences of these species.
The exotic fish of greatest concern are in the Centrarchid family and include species of
bass and crappie. These are warmwater fish introduced into many of Washington’s rivers for
sport fishing (Hallock, 2013). They increase competition and predation on Oregon spotted frog.
In the Black River, centrarchids were absent at most of the sites with only a few sites having a
rare occurrence. In the Chehalis River floodplain, half of the sites did not have exotic fish but
there were many sites with an abundance of occurrences. The difference in abundance of
centrarchids between the two rivers was statistically different. As such, presence of the frog may
be limited in the Chehalis by the presence of centrarchids.
Almost all of the Chehalis River floodplain sites had some degree of bullfrog occurrence,
many of which were abundant. In the Black River, half of the sites did not have bullfrogs and
half were rare. In both cases, the abundance of these species was statistically different between
the rivers and greater in the Chehalis River.

72

Centrarchids and bull frogs have been shown to be detrimental to Oregon spotted
persistence and if left unmanaged, can cause local extirpations. At Conboy Lake National
Wildlife Refuge, when bullfrog management was halted, the populations of Oregon spotted frog
began to decline (M. Hayes, personal communication). Bullfrogs share similar habitat with
Oregon spotted frog and may be outcompeting for resources and increasing predation (Pearl &
Hayes, 2004; Rowe et al., 2021). Exotic fish increase predation pressure on Oregon spotted frog
and when they are present in aquatic connectors between populations, especially permanent ones,
they can increase isolation and sever genetic exchange (Bradford & Tabatabai, 1993). The
abundance of exotic fish is negatively associated with many amphibians native to the Pacific
Northwest (Holgerson et al., 2019). Oregon spotted frogs are particularly vulnerable to both of
these species due to their aquatic nature as they cannot escape into terrestrial habitat during the
non-breeding season (Hayes, 1994).
The exotic abundance data for the Chehalis River floodplain sites was collected at
breeding sites as well as non-breeding habitat, while the data from the Black River was collected
only during breeding surveys. Surveys of non-breeding habitat in the Black River could inform
of their presence in overwintering habitat or in the permanent water habitat used in the warmer
months. However, it could be assumed that if frogs are using breeding habitat, the exotics are not
overly abundant in non-breeding habitat as they are connected aquatically and would be
detrimental to frogs during the remainder of the year. Following this assumption, the nonbreeding sites are still likely to have fewer exotics than the Chehalis River or the breeding
capability of frogs in the Black River would be greatly diminished. Never the less, locating and
documenting the abundance of exotic species in all habitat types can guide conservation and
management efforts before the balance shifts too far in favor of exotic species. From what is

73

currently known, the Black River is the only bastion for Oregon spotted frog in the Chehalis
Basin, therefore, protecting it from invasive species should be a top conservation goal.
This study was an initial attempt at using a species distribution model to locate potential
habitat and asses site differences between occupied and unoccupied sites for the Oregon spotted
frog. However, additional variables and a “finer tuned” modelling approach may be necessary to
fully understand the absence of the frog in the Chehalis Basin. Because of the aquatic nature of
Oregon spotted frog, including variables on the seasonal hydrology, water temperature and water
quality may capture barriers to dispersal missing from this study. The Maxent model itself may
need adjustment as well. Statistical packages have been developed that guide the user to tune
different settings to aid with issues such as correlated variables and over fitting. In certain cases,
these methods have been shown to create a better prediction beyond the default settings within
Maxent.
The strongest evidence for the presumed absence of Oregon spotted frog in the Chehalis
Basin, based on the variables considered in this study, is the presence and abundance of exotic
species. According to the model, the structural habitat found in the Black River is available in
the Chehalis River and many of its tributaries. While the climate is statistically different between
the points in the two rivers, it cannot be said with certainty that there is enough of a difference to
explain their absence. However, the abundance of exotics, has a clear distinction between the
two rivers and both species have a higher instance of occurrence and abundance in the Chehalis
River floodplain compared to the Black River. As these species have been associated with a
decrease in Oregon spotted frog presence, this points to a potential factor for their absence in the
Chehalis River floodplain. The Black River remains one of the only known rivers in Washington
to have populations of Oregon spotted frog because the habitat structure is intact and the

74

presence of invasive species appears to be low enough to allow for their continued persistence.
Conserving this remaining habitat and preventing invasive species from establishing to harmful
levels needs to be a high priority to preserve this already threatened species.

75

LITERATURE CITED
Adams, M. J., Miller, D. A. W., Muths, E., Corn, P. S., Grant, E. H. C., Bailey, L. L., Fellers, G.
M., Fisher, R. N., Sadinski, W. J., Waddle, H., & Walls, S. C. (2013). Trends in Amphibian
Occupancy in the United States. PLoS ONE, 8(5), 1–5.
https://doi.org/10.1371/journal.pone.0064347
Adams, M. J., Pearl, C. A., McCreaery, B., & Galvan, S. K. (2014). Short-Term Occupancy and
Abundance Dynamics of the Oregon Spotted Frog ( Rana pretiosa ) Across Its Core Range.
U.S. Geologic Survey.
Arkle, R. S., & Pilliod, D. S. (2015). Persistence at distributional edges: Columbia spotted frog
habitat in the arid Great Basin, USA. Ecology and Evolution, 5(17), 3704–3724.
https://doi.org/10.1002/ece3.1627
Blank, L., & Blaustein, L. (2012). Using ecological niche modelling to predict the distributions
of two endangered amphibian species in aquatic breeding sites. Hydrobiologia, 693(1),
157–167. https://doi.org/10.1007/s10750-012-1101-5
Blouin, M. S., Phillipsen, I. C., & Monsen, K. J. (2010). Population structure and conservation
genetics of the Oregon spotted frog, Rana pretiosa. Conservation Genetics, 11(6), 2179–
2194. https://doi.org/10.1007/s10592-010-0104-x
Bohannon, J. S., Gay, D. R., Hayes, M. P., Danilson, C. D., & Warheit, K. I. (2016). Discovery
of the Oregon spotted frog in the Northern Puget Sound Basin, Washington state.
Northwestern Naturalist, 97(2), 82–97. https://doi.org/10.1898/nwn15-19.1
Boria, R. A., Olson, L. E., Goodman, S. M., & Anderson, R. P. (2014). Spatial filtering to reduce
sampling bias can improve the performance of ecological niche models. Ecological
Modelling, 275, 73–77. https://doi.org/10.1016/j.ecolmodel.2013.12.012
Bowerman, J., & Pearl, C. A. (2020). Oregon Spotted Frog (Rana pretiosa) Migration from an
Aquatic Overwintering Site: Timing, Duration, and Potential Environmental Cues.
American Midland Naturalist, 184(1), 87–97. https://doi.org/10.1637/0003-0031-184.1.87
Bradford, D. F., & Tabatabai, F. (1993). Isolation of remaining populations of the native frog,
Rana mucosa, by introduced fishes in sequioa and kings canyon national parks,
California.pdf. Conservation Biology, 7(7), 882–888.
Brown, J. L., Bennett, J. R., & French, C. M. (2017). SDMtoolbox 2.0: The next generation
Python-based GIS toolkit for landscape genetic, biogeographic and species distribution
model analyses. PeerJ, 5(e4095), 1–12. https://doi.org/10.7717/peerj.4095
Chelgren, N. D., Pearl, C. A., Adams, M. J., & Bowerman, J. (2008). Demography and
movement in a relocated population of Oregon Spotted Frogs (Rana pretiosa): Influence of
season and gender. Copeia, 2008(4), 742–751. https://doi.org/10.1643/CH-07-142
Cianfrani, C., Le Lay, G., Hirzel, A. H., & Loy, A. (2010). Do habitat suitability models reliably
predict the recovery areas of threatened species? Journal of Applied Ecology, 47, 421–430.
76

https://doi.org/10.1111/j.1365-2664.2010.01781.x
Cowardin, L. M., Carter, V., Golet, F. C., & Laroe, E. T. (1979). Classification of Wetlands and
Deepwater Habitats of the United States. In FWS/OBS-79/31 (Issue December 1979).
https://doi.org/10.1002/047147844x.sw2162
Cunningham, H. R., Rissler, L. J., Buckley, L. B., & Urban, M. C. (2016). Abiotic and biotic
constraints across reptile and amphibian ranges. Ecography, 39(1), 1–8.
https://doi.org/10.1111/ecog.01369
Cushman, Kathleen, A., & Pearl, Christopher, A. (2007). A Conservation Assement for the
Oreogon Spotted Frog (Rana Pretiosa).USDA Forest Service.
Department of Ecology (1997). Washington State Wetlands Identification and Delineation
Manual (Issue March).
Department of Ecology. (2016). Chehalis Basin Strategy Draft Programmatic EIS Reducing
Flood Damage and Restoring Aquatic Species Habitat.
Department of Ecology. (2016). Chehalis Basin Strategy Draft Programmatic EIS.
Dickes, B. (1990). Black River Water Quality, Winter, 1989/1990. Washington State Department
of Ecology
Elith, J., Graham, C. H., Anderson, R. P., Dudik, M., Ferrier, S., Guisan, A., Hijmans, R. J.,
Huettmann, F., Leathwick, J. L., Lehmann, A., Li, J., Lohmann, L. G., Loisell, B. A.,
Manion, G., Moritz, C., Nakamura, M., Nakazawa, Y., Overton, J. M., Peterson, A. T., …
Zimmermann, N. E. (2006). Novel methods to improve prediction of species’ distributions
from occurence data. Ecography, 29, 129–151.
Elith, J., & Leathwick, J. R. (2009). Species distribution models: Ecological explanation and
prediction across space and time. Annual Review of Ecology, Evolution, and Systematics,
40, 677–697. https://doi.org/10.1146/annurev.ecolsys.110308.120159
Elith, J., Phillips, S. J., Hastie, T., Dudík, M., Chee, Y. E., & Yates, C. J. (2011). A statistical
explanation of MaxEnt for ecologists. Diversity and Distributions, 17(1), 43–57.
https://doi.org/10.1111/j.1472-4642.2010.00725.x
Fick, S. E., & Hijmans, R. J. (2017). WorldClim 2: new 1km spatial resolution climate surfaces
for global land areas. International Journal of Climatology, 37(2), 4302–4315.
Germaine, S. S., & Cosentino, B. L. (2004). Screening Model for Determining Likelihood of Site
Occupancy by Oregon Spotted Frogs ( Rana pretiosa ) in Washington State. Washington
Departmetn of Fish and Wildlife.
Groff, L. A., Marks, S. B., & Hayes, M. P. (2014). Using ecological niche models to direct rare
amphibian surveys: A case study using the oregon spotted frog (Rana pretiosa).
Herpetological Conservation and Biology, 9(2), 354–368.
Guisan, A., & Thuiller, W. (2005). Predicting species distribution: Offering more than simple
habitat models. Ecology Letters, 8(9), 993–1009. https://doi.org/10.1111/j.14610248.2005.00792.x
77

Gül, S., Kumlutaş, Y., & Ilgaz, Ç. (2018). Potential distribution under different climatic
scenarios of climate change of the vulnerable Caucasian salamander (Mertensiella
caucasica): A case study of the Caucasus Hotspot. Biologia (Poland), 73(2), 175–184.
https://doi.org/10.2478/s11756-018-0020-y
Hallock, L. A. (2013). DRAFT State of Washington Oregon Spotted Frog Recovery Plan.
Washington Deprtment of Fish and Wildlife.
Hayes, M. P. (1994). The spotted frog in Western Oregon. Oregon Department of Fish and
Wildlife.
Hayes, M. P. (1997). Status of the Oregon spotted frog (Rana pretiosa senu stricto) in the
Deschutes Basin and other selected systems in Oregon and northeastern California with a
rangewide synopsis of the species’ status. US Fish and Wildlife Service.
http://www.ncbi.nlm.nih.gov/pubmed/20892930%0Ahttp://www.pubmedcentral.nih.gov/art
iclerender.fcgi?artid=PMC2624385
Hayes, M. P., Engler, J. D., Van Leuren, S., Friesz, D. C., Quinn, T., & Pierce, D. J. (2001).
Overwintering of the Oregon Spotted Frog (Rana pretiosa) at Conboy Lake National
Wildlife Refuge, Klickitat County, Washington, 2000-2001. Washington Department of
Transportation.
http://wdfw.wa.gov/publications/00852/wdfw00852.pdf%5Cnpapers2://publication/uuid/7E
DD92BA-C328-4112-B5D7-3AB8F6180C88
Hayes, M. P., Rombough, C. J., Padgett-Flohr, G. E., Hallock, L. A., Johnson, J. E., Wagner, R.
S., & Engler, J. D. (2009). Amphibian Chytridiomycosis in the Oregon Spotted Frog (Rana
pretiosa) in Washington state, USA. Northwestern Naturalist, 90(2), 148–151.
https://doi.org/10.1898/nwn08-28.1
Hohmann, M. G., & Wall, W. A. (2017). Multiscale Assessment of Listed and At-Risk Species ’
Climate Change Vulnerabilities. US Army Corps of Engineer Research and Development
Center.
Holgerson, M. A., Duarte, A., Hayes, M. P., Adams, M. J., Tyson, J. A., Douville, K. A., &
Strecker, A. L. (2019). Floodplains provide important amphibian habitat despite multiple
ecological threats. Ecosphere, 10(9). https://doi.org/10.1002/ecs2.2853
Hudec, J. L., Halofsky, J. E., Halsey, S. M., Halofsky, J. S., & Donatoa, D. C. (2019). Climate
Change Vulnerability and Adaptation in Southwest Washington. Ch 5: Effects of Climate
Change on Special Hbitats in Southwest Washington. USDA Forest Service.
Kapust, H. Q. W., McAllister, K. R., & Hayes, M. P. (2012). Oregon spotted frog (Rana
pretiosa) response to enhancement of oviposition habitat degraded by invasive reed canary
grass (Phalaris arundinacea). Herpetological Conservation and Biology, 7(3), 358–366.
Kiesecker, J. M., & Blaustein, A. R. (1998). Effects of introduced bullfrogs and smallmouth bass
on microhabitat use, growth, and survival of native red-legged frogs (Rana aurora).
Conservation Biology, 12(4), 776–787. https://doi.org/10.1111/j.1523-1739.1998.97125.x
LANDFIRE. (2016a). LANDFIRE Remap 2016 Existing Vegetation Height (EVH) CONUS. U.S.
Geological Suvey. https://www.landfire.gov/viewer/
78

LANDFIRE. (2016b). LANDFIRE Remap 2016 Existing Vegetation Type (EVT) CONUS. U.S.
Geological Suvey. https://www.landfire.gov/viewer/
Lavergne, S., & Molofsky, J. (2004). Reed canary grass (Phalaris arundinacea) as a biological
model in the study of plant invasions. Critical Reviews in Plant Sciences, 23(5), 415–429.
https://doi.org/10.1080/07352680490505934
Licht, L. E. (1971). Breeding Habits and Embryonic Thermal Requirements of the Frogs , Rana
Aurora Aurora and Rana Pretiosa Pretiosa , in the Pacific Northwest. Ecology, 52(1), 116–
124.
Licht, L. E. (1974). Survival of embryos, tadpoles, and adults of the frogs Rana aurora aurora
and Rana pretiosa pretiosa sympatric in southwestern British Columbia. Canadian Journal
of Zoology, 52(5), 613–627. https://doi.org/10.1139/z74-079
Maurer, D. A., Lindig-Cisneros, R., Werner, K. J., Kercher, S., Miller, R., & Zedler, J. B. (2003).
The replacement of wetland vegetation by reed canary grass (Phalaris arundinacea).
Ecological Restoration, 21(2), 116–119.
McAllister, K. R., & Leonard, W. P. (1997). Washington State status report for the Oregon
spotted frog. Washington Department of Fish and Wildlife.
McAllister, K. R., Leonard, W. P., & Storm, R. M. (1993). Spotted Frog (Rana pretiosa) Surveys
in the Puget trough of Washington, 1989-1991. Northwestern Naturalist, 74(1), 10.
https://doi.org/10.2307/3536575
Mcallister, K. R., & White, H. Q. (2001). Oviposition Ecology of the Oregon Spotted Frog at
Beaver Creek , Washington. Washington Department of Fish and Wildlife.
McKibbin, R., Dushenko, W. T., vanAggelen, G., & Bishop, C. A. (2008). The influence of
water quality on the embryonic survivorship of the Oregon spotted frog (Rana pretiosa) in
British Columbia, Canada. Science of the Total Environment, 395(1), 28–40.
https://doi.org/10.1016/j.scitotenv.2008.01.050
Merow, C., Smith, M. J., & Silander, J. A. (2013). A practical guide to MaxEnt for modelling
species’ distributions: What it does, and why inputs and settings matter. Ecography, 36(10),
1058–1069. https://doi.org/10.1111/j.1600-0587.2013.07872.x
Na, X., Zhou, H., Zang, S., Wu, C., Li, W., & Li, M. (2018). Maximum Entropy modelling for
habitat suitability assessment of Red-crowned crane. Ecological Indicators, 91(December
2017), 439–446. https://doi.org/10.1016/j.ecolind.2018.04.013
Najibzadeh, M., Gharzi, A., Rastegar-Pouyani, N., Rastegar-Pouyani, E., & Pesarakloo, A.
(2017). Habitat suitability and patterns of sex-biased migration of the Iranian long-legged
wood frog, Rana pseudodalmatina (Anura: Ranidae). Biologia (Poland), 72(6), 686–693.
https://doi.org/10.1515/biolog-2017-0074
Ortega-Huerta, M. A., & Peterson, A. T. (2008). Modelling ecological niches and predicting
geographic distributions: A test of six presence-only methods. Revista Mexicana de
Biodiversidad, 79(1), 205–216. https://doi.org/10.22201/ib.20078706e.2008.001.522
Padgett-Flohr, G. E., & Hayes, M. P. (2011). Assessment of the vulnerability of the oregn
79

spotted frog (Rana pretiosa) to the amphibian chytrid fungus (Batrachochytrium
dendrobatidis). Herpetological Conservation and Biology, 6(2), 99–106.
Pearl, C. A., Adams, M. J., Bury, R. B., & Mccreary, B. (2004). Asymmetrical Effects of
Introduced Bullfrogs (Rana catesbeiana) on Native Ranid Frogs in Oregon. Copeia,
2004(1), 11–20.
Pearl, C. A., Adams, M. J., & Leuthold, N. (2009). Breeding habitat and local population size of
the Oregon spotted frog (Rana pretiosa). Northwestern Naturalist, 90(2), 136–147.
Pearl, C. A., Bowerman, J., Adams, M. J., & Chelgren, N. D. (2009). Widespread occurrence of
the chytrid fungus batrachochytrium dendrobatidis on oregon spotted frogs (Rana
pretiosa). EcoHealth, 6(2), 209–218. https://doi.org/10.1007/s10393-009-0237-x
Pearl, C. A., Bowerman, J., & Knight, D. (2005). Feeding Behavior and Aquatic Habitat Use by
Oregon Spotted Frogs (Rana pretiosa) in Central Oregon. Northwestern Naturalist, 86(1),
36–38.
Pearl, C. A., & Hayes, M. P. (2004). Habitat associations of the Oregon spotted frog (Rana
pretiosa): a literature review. Washington Department of Fish and Wildlife.
Pearl, C. A., McCreary, B., Rowe, J. C., & Adams, M. J. (2018). Late-season movement and
habitat use by Oregon spotted frog (Rana pretiosa) in Oregon, USA. Copeia, 106(3), 539–
549. https://doi.org/10.1643/ch-18-031
Pearson, R. G., Raxworthy, C. J., Nakamura, M., & Townsend Peterson, A. (2007). Predicting
species distributions from small numbers of occurrence records: A test case using cryptic
geckos in Madagascar. Journal of Biogeography, 34(1), 102–117.
https://doi.org/10.1111/j.1365-2699.2006.01594.x
Phillips, S. J. (2017). A Brief Tutorial on Maxent. https://doi.org/10.18388/abp.2002_3772
Phillips, S. J., Anderson, R. P., Dudík, M., Schapire, R. E., & Blair, M. E. (2017). Opening the
black box: an open-source release of Maxent. Ecography, 40(7), 887–893.
https://doi.org/10.1111/ecog.03049
Phillips, S. J., Anderson, R. P., & Schapire, R. E. (2006). Maximum entropy modelling of
species geographic distributions. Ecological Modelling, 190(3–4), 231–259.
https://doi.org/10.1016/j.ecolmodel.2005.03.026
Phillips, S. J., & Dudík, M. (2008). Modelling of species distributions with Maxent: New
extensions and a comprehensive evaluation. Ecography, 31(2), 161–175.
https://doi.org/10.1111/j.0906-7590.2008.5203.x
Phillips, S. J., Dudík, M., & Schapire, R. E. (2020). Maxent software for modelling species
niches and distributions (Version 3.4.4).
http://biodiversityinformatics.amnh.org/open_source/maxent/
Popesu, V. D., Kissel, A. M., Pearson, M., Palen, W. J., Govindarajulu, P., & Bishop, C. A.
(2013). Defining conservation-relevant habitat selection by the highly imperiled Oregon
spotted frog, Rana pretiosa. Herpetological Conservation and Biology, 8(3), 688–706.
80

Preau, C., Trochet, A., Bertrand, R., & Isselin-Nondeau, F. (2018). Modelling potential
distributions of three european amphibian species comapring ENFA and MaxEnt.
Herpetological Conservation and Biology, 13(1), 91–104.
Pulliam, H. R. (2000). On the relationship between niche and distribution. Ecology Letters, 3(4),
349–361. https://doi.org/10.1046/j.1461-0248.2000.00143.x
R Core Team. (2020). R: A language and environment for statistical computing (4.0.3). R
Foundation for Statistical Computing. https://www.r-project.org/
Radosavljevic, A., & Anderson, R. P. (2014). Making better Maxent models of species
distributions: Complexity, overfitting and evaluation. Journal of Biogeography, 41(4), 629–
643. https://doi.org/10.1111/jbi.12227
Reinhardt Adams, C., & Galatowitsch, S. M. (2005). Phalaris arundinacea (reed canary grass):
Rapid growth and growth pattern in conditions approximating newly restored wetlands.
Ecoscience, 12(4), 569–573. https://doi.org/10.2980/i1195-6860-12-4-569.1
Robertson, J. M., Murphy, M. A., Pearl, C. A., Adams, M. J., Páez-Vacas, M. I., Haig, S. M.,
Pilliod, D. S., Storfer, A., & Funk, W. C. (2018). Regional variation in drivers of
connectivity for two frog species (Rana pretiosa and R. luteiventris) from the U.S. Pacific
Northwest. Molecular Ecology, 27(16), 3242–3256. https://doi.org/10.1111/mec.14798
Romansic, J. M., Nelson, N. L., Moffett, K. B., & Piovia-Scott, J. (2020). Beaver dams are
associated with enhanced amphibian diversity via lengthened hydroperiods and increased
representation of slow-developing species. Freshwater Biology, August, 1–14.
https://doi.org/10.1111/fwb.13654
Rowe, J. C., Wilson-Romine, L., & Romine, J. (2021). Active season and pre-overwitering
microhabitat use by Oregon spotted frogs (Rana pretiosa) and American bullfrogs
(Lithobates Catesbeianus) at Conboy Lake National Wildlife Refuge, Washington..
Northwest Naturalist, 102, 55–75.
Shcheglovitova, M., & Anderson, R. P. (2013). Estimating optimal complexity for ecological
niche models: A jackknife approach for species with small sample sizes. Ecological
Modelling, 269, 9–17. https://doi.org/10.1016/j.ecolmodel.2013.08.011
Songer, M., Delion, M., Biggs, A., & Huang, Q. (2012). Modelling impacts of climate change on
giant panda habitat. International Journal of Ecology, 2012.
https://doi.org/10.1155/2012/108752
Aquatic Species Restoration Plan Steering Committee. (2019). Chehalis Basin Strategy Aquatic
Species Restoration Plan Chehalis Basin Strategy.
https://www.chehalisbasinstrategy.com/wp-content/uploads/2019/11/ASRP_Phase-1.pdf
Spyreas, G., Wilm, B. W., Plocher, A. E., Ketzner, D. M., Matthews, J. W., Ellis, J. L., & Heske,
E. J. (2010). Biological consequences of invasion by reed canary grass (Phalaris
arundinacea). Biological Invasions, 12(5), 1253–1267. https://doi.org/10.1007/s10530-0099544-y
Struecker, B. P., & Milanovich, J. (2017). Predicted suitable habitat declines for midwestern
United States amphibians under future climate change and land-use change scenarios.
81

Herpetological Conservation and Biology, 12(3), 635–654.
Stuart, S. N., Chanson, J. S., Cox, N. A., Young, B. E., Rodrigues, A. S. L., Fischman, D. L., &
Waller, R. W. (2004). Status and Trends of Amphibian Declines and Extinctions
Worldwide. Science New Series, 306(5702), 1783–1786.
https://doi.org/10.1177/002205741508100313
Tarrant, J., & Armstrong, A. J. (2013). Using predictive modelling to guide the conservation of a
critically endangered coastal wetland amphibian. Journal for Nature Conservation, 21(5),
369–381. https://doi.org/10.1016/j.jnc.2013.03.006
Tattersall, G. J., & Ultsch, G. R. (2008). Physiological ecology of aquatic overwintering in ranid
frogs. Biological Reviews, 83(2), 119–140. https://doi.org/10.1111/j.1469185X.2008.00035.x
Thurston County Public Health and Social Services Department. (2006). Water Resources
Monitoring Report 2003-2004 and 2004-2005 Water Years.
https://co.thurston.wa.us/health/ehrp/pdf/AR03-05/AR03-05.pdf
US Geologic Survey (2020). USGS NED 1 arc-second. U.S. Geological Suvey.
Wake, D. B., & Vredenburg, V. T. (2008). Are we in the midst of the sixth mass extinction? A
view from the world of amphibians. Proceedings of the National Academy of Sciences of
the United States of America, 105(SUPPL. 1), 11466–11473.
https://doi.org/10.1073/pnas.0801921105
Washington Deptartment of Fish and Wildlife. (2020). Priority Habitats and Species: Maps |
Washington Department of Fish & Wildlife. https://wdfw.wa.gov/species-habitats/atrisk/phs/maps
Watson, J. W., Mcallister, K. R., & Pierce, D. J. (2003). Home ranges, movements, and habitat
selection of Oregon spotted frogs (Rana pretiosa). Journal of Herpetology, 37(2), 292–300.
Watson, J. W., McAllister, K. R., Pierce, D. J., & Alvarado, A. (2000). Ecology of a remnant
population of Oregon spotted frogs (Rana pretiosa) in Thurston County, Washington.
Washingotn Department of Fish and Wildlife.
Yang, L., Jin, S., Danielson, P., Homer, C., Gass, L., Bender, S. M., Case, A., Costello, C.,
Dewitz, J., Fry, J., Funk, M., Granneman, B., Liknes, G. C., Rigge, M., & Xian, G. (2018).
A new generation of the United States National Land Cover Database: Requirements,
research priorities, design, and implementation strategies. ISPRS Journal of
Photogrammetry and Remote Sensing, 146, 108–123.

82

APPENDIX

Appendix 1. Predicted suitable habitat for the Oregon spotted frog in the Chehalis Basin
according to the structure variables.

83

Appendix 2. Predicted suitable habitat for the Oregon spotted frog in the Chehalis Basin
according to the climate variables.
Appendix 3. The success rate, omission rate, and associated p-values of the structure-only and
climate-only Maxent models developed for the Chehalis Basin. Success was evaluated at two
thresholds, the minimum training presence (MTP) and the 10-percentile training presence
(10TP).
Threshold
Model
Success Rate
Omission Rate
p-value
MTP
Structure
0.965517
0.034483
0
Climate
0.862069
0.137931
0.071904
10TP
Structure
0.896552
0.103448
0
Climate
0.793103
0.206897
0.022506
Appendix 4. The percent contribution and permutation importance of the variables in the
structure-only model.
Variable
Proximity to Cover Class
Landcover
Hydric Soil
Herbaceous Vegetation Height
Slope

Percent Contribution
35.9
29.1
28.4
4.4
2.2

Permutation Importance
69.5
8.7
5.8
2
14.1

84

Appendix 5. The percent contribution and permutation importance of the variables in the
climate-only model.
Variable
Precipitation of Wettest Month
Mean Temperature of Coldest Quarter
Precipitation of Driest Quarter
Temperature Annual Range
Annual Mean Temperature

Percent Contribution
26.5
24.4
24.4
15.6
9

Permutation Importance
34.2
14.9
17.6
18.4
14.9

Appendix 6. Survey points selected according to the structure-only and climate-only Maxent
outputs. Points were selected according to their probability of presence in the Maxent outputs (>
0.7).

85

Appendix 7. The comparison results of the environmental variables and abundance of exotic
species between the Chehalis River floodplain survey points selected from the structure-only or
climate-only models and the Black River presence points.
Comparison Results of Structure-only and Climate-only Variables
Structure Only Model
Variable
Test
df
Statistic
p-value
Environmental Variables
Distance to Habitat
Mann-Whitney U test
W= 420.5
1.0000
Herbaceous Vegetation Height
Mann-Whitney U test
W= 393.5
0.6509
Landcover
Chi-squared test
2
X²= 2.4211 0.2980
Slope
Mann-Whitney U test
W= 444
0.7098
Hydric Soil
Chi-squared test
1
X²= 0
1.0000
Exotic Species Abundance
Variables
Centrarchid Abundance
Mann-Whitney U test
W= 218.5
0.0002
Bullfrog Abundance
Mann-Whitney U test
W= 100.5
0.0000
Climate Only Model
Variable
Test
df
Statistic
p-value
Environmental Variables
Welch Two Sample tAnnual Mean Temperature
test
51.555 t= -8.6819
0.0000
Mean Temperature of Coldest
Quarter
Mann-Whitney U test
W= 0
0.0000
Precipitation of Wettest Month
Mann-Whitney U test
W= 731
0.0000
Welch Two Sample tPrecipitation of Driest Quarter
test
49.754 t= -10.85
0.0000
Temperature Annual Range
Mann-Whitney U test
W= 681
0.0000
Exotic Species Abundance
Variables
Centrarchid Abundance
Mann-Whitney U test
W= 122.5
0.0000
Bullfrog Abundance
Mann-Whitney U test
W= 79.5
0.0000

86

Appendix 8. Summary statistics for the environmental variables used in the Maxent models and
the abundance of exotic species for the presence points in the Black River and the survey points
selected from the structure-only model and the climate-only model.
Summary Statistics of Structure-only and Climate-only Variables
Black
Structure-Only
Variable
Mean
St Dev
Mean
St Dev
Environmental Variables
Distance to Habitat (Meters)
6.21
12.37
6.21
12.37
Herbaceous Vegetation Height (Height
1.24
1.12
1.41
1.43
Rank)
Landcover
Categorical
Slope (Degrees)
0.54
0.57
0.45
0.47
Hydric Soil
Categorical
Exotic Species Abundance Variables
Centrarchid Abundance (Abundance Rank)
0.10
0.31
0.86
0.88
Bullfrog Abundance (Abundance Rank)
0.45
0.51
1.55
0.63
Black
Climate-Only
Variable
Mean
St Dev
Mean
St Dev
Environmental Variables
Annual Mean Temperature (°C)
10.31
0.09
10.56
0.12
Mean Temperature of Coldest Quarter (°C)
4.32
0.09
4.75
0.11
Precipitation of Wettest Month
212.15
5.52
204.04
4.69
(Millimeters)
Precipitation of Driest Quarter (Millimeters)
96.07
2.26
101.61
1.56
Temperature Annual Range (°C)
24.69
0.40
24.27
0.26
Exotic Species Abundance Variables
Centrarchid Abundance (Abundance Rank)
0.10
0.31
1.28
0.84
Bullfrog Abundance (Abundance Rank)
0.45
0.51
1.62
0.56
Appendix 9. Contingency tables for (a) landcover and (b) hydric soil, the categorical variables
used in the structure-only model.
The Frequency of Points Occurring in Land Cover Class and Hydric Soils
(a)
(b)
Landcover Class Frequency
Hydric Soil Frequency
Black
Chehalis
Black Chehalis
Non-habitat
6
6
Non-Hydric
6
7
Agriculture
6
2
Hydric
23
22
Emergent Wetland
17
21

87

Appendix 10. The magnitude of change between statistically different variables in the structureonly and climate-only models.
Magnitude of Change in Statistically Different Variables
Variable
Difference
Direction
Climate-Only Variables
Annual Mean Temperature (°C)
0.25 °C
C warmer than B
Mean Temperature of Coldest Quarter (°C)
0.43 °C
C warmer than B
Precipitation of Wettest Month (Millimeters)
8.11 mm
C drier than B
Precipitation of Driest Quarter (Millimeters)
5.54 mm
C wetter than B
Temperature Annual Range (°C)
0.41 °C
C narrower than B
Climate-only Exotic Species Abundance Variables
Centrarchid Abundance (Abundance Rank)
1.17
C more abundant than B
Bullfrog Abundance (Abundance Rank)
1.17
C more abundant than B
Structure-only Exotic Species Abundance Variables
Centrarchid Abundance (Abundance Rank)
0.76
C more abundant than B
Bullfrog Abundance (Abundance Rank)
1.10
C more abundant than B

Centrarchid Abundance Ranks
30

# of Points

25
20
15
10
5
0
Absent

Rare

Abundant

Abundance Ranks
Black

Structure

Climate

Appendix 11. The abundance ranks of centrarchid fishes at the Black River presence points and
the Chehalis River floodplain survey points selected in the structure-only model and the climateonly model.

88

Bullfrog Abundance Ranks
30

# of Points

25

20
15
10
5
0
Absent

Rare

Abundant

Abundance Ranks
Black

Structure

Climate

Appendix 12. The abundance ranks of bullfrogs at the Black River presence points and The
Chehalis River floodplain survey points selected in the structure-only model and the climate-only
model.

89

(a)

(b)

(c)

(d)

(e)

Appendix 13. Selected climate variables for the Maxent model.
90

Item sets

The Evergreen State College Masters Theses: Environmental Studies

Media

Thesis_MES_2021_MurtaghP.pdf