<%BANNER%>

Effects of mammalian predator exclusion, supplemental feeding, and prescribed fire on small mammal populations in a long...

Permanent Link: http://ufdc.ufl.edu/UFE0042234/00001

Material Information

Title: Effects of mammalian predator exclusion, supplemental feeding, and prescribed fire on small mammal populations in a longleaf pine ecosystem
Physical Description: 1 online resource (101 p.)
Language: english
Creator: Morris, Gail
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2010

Subjects

Subjects / Keywords: adaptive, behavior, consumptive, cotton, effects, fire, food, gossypinus, hispidus, mouse, non, oldfield, peromyscus, polionotus, predation, prescribed, rat, sigmodon, supplementation
Wildlife Ecology and Conservation -- Dissertations, Academic -- UF
Genre: Wildlife Ecology and Conservation thesis, M.S.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Food resources and predation play important roles in determining small mammal population dynamics. Each of these factors may affect population parameters such as abundance, survival, and reproduction. These factors may also interact as individuals under predation pressure make trade offs between exposure to predators and access to food resources. Fires consume food sources and reduce cover, which increases exposure to predators. For species that occur in areas with frequent fires, it is beneficial to consider how all of these factors affect populations of interest. This study used a mark-recapture framework to experimentally examine how supplemental feeding, mammalian predator exclusion, and prescribed fire affected survival, abundance, and reproduction of cotton rats (Sigmodon hispidus), cotton mice (Peromyscus gossypinus), and oldfield mice (P. polionotus). Radio telemetry was used to assess home range size and overlap of cotton rats. Among cotton rats, prescribed fire events had the greatest effect, causing large drops in survival, abundance, and reproduction. Food supplementation increased survival, rates of transitions to reproductive states, and abundances but was not sufficient to prevent post fire declines in any of these parameters. This indicates that predation and emigration are responsible for fire related declines in survival and abundance. Population level effects of predator exclusion effects were small in magnitude. Predator exclusion was, however, associated with increased male home range size, indicating a response to perceived predation risk. Among cotton mice, survival was affected (increased) only by an interaction between burning and predator exclusion. Rates of transitions to reproductive states decreased in burn years but increased with the combination of feeding and predator exclusion. Feeding increased abundances. Among oldfield mice, survival and abundance were greater in predator exclusion areas than controls. Feeding and the interaction of feeding and predator exclusion also increased abundances. Rates of transitions to reproductive states declined during peak breeding seasons during which burning occurred such that breeding transitions in these seasons were lower than in non peak seasons. Some of these effects can only be understood by assuming individuals make behavioral responses to predation risk to limit mortality.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Gail Morris.
Thesis: Thesis (M.S.)--University of Florida, 2010.
Local: Adviser: Oli, Madan K.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2010
System ID: UFE0042234:00001

Permanent Link: http://ufdc.ufl.edu/UFE0042234/00001

Material Information

Title: Effects of mammalian predator exclusion, supplemental feeding, and prescribed fire on small mammal populations in a longleaf pine ecosystem
Physical Description: 1 online resource (101 p.)
Language: english
Creator: Morris, Gail
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2010

Subjects

Subjects / Keywords: adaptive, behavior, consumptive, cotton, effects, fire, food, gossypinus, hispidus, mouse, non, oldfield, peromyscus, polionotus, predation, prescribed, rat, sigmodon, supplementation
Wildlife Ecology and Conservation -- Dissertations, Academic -- UF
Genre: Wildlife Ecology and Conservation thesis, M.S.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Food resources and predation play important roles in determining small mammal population dynamics. Each of these factors may affect population parameters such as abundance, survival, and reproduction. These factors may also interact as individuals under predation pressure make trade offs between exposure to predators and access to food resources. Fires consume food sources and reduce cover, which increases exposure to predators. For species that occur in areas with frequent fires, it is beneficial to consider how all of these factors affect populations of interest. This study used a mark-recapture framework to experimentally examine how supplemental feeding, mammalian predator exclusion, and prescribed fire affected survival, abundance, and reproduction of cotton rats (Sigmodon hispidus), cotton mice (Peromyscus gossypinus), and oldfield mice (P. polionotus). Radio telemetry was used to assess home range size and overlap of cotton rats. Among cotton rats, prescribed fire events had the greatest effect, causing large drops in survival, abundance, and reproduction. Food supplementation increased survival, rates of transitions to reproductive states, and abundances but was not sufficient to prevent post fire declines in any of these parameters. This indicates that predation and emigration are responsible for fire related declines in survival and abundance. Population level effects of predator exclusion effects were small in magnitude. Predator exclusion was, however, associated with increased male home range size, indicating a response to perceived predation risk. Among cotton mice, survival was affected (increased) only by an interaction between burning and predator exclusion. Rates of transitions to reproductive states decreased in burn years but increased with the combination of feeding and predator exclusion. Feeding increased abundances. Among oldfield mice, survival and abundance were greater in predator exclusion areas than controls. Feeding and the interaction of feeding and predator exclusion also increased abundances. Rates of transitions to reproductive states declined during peak breeding seasons during which burning occurred such that breeding transitions in these seasons were lower than in non peak seasons. Some of these effects can only be understood by assuming individuals make behavioral responses to predation risk to limit mortality.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Gail Morris.
Thesis: Thesis (M.S.)--University of Florida, 2010.
Local: Adviser: Oli, Madan K.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2010
System ID: UFE0042234:00001


This item has the following downloads:


Full Text





EFFECTS OF MAMMALIAN PREDATOR EXCLUSION, SUPPLEMENTAL FEEDING,
AND PRESECRIBED FIRE ON SMALL MAMMAL POPULATIONS IN A LONGLEAF
PINE ECOSYSTEM


















By

GAIL MORRIS


A THESIS PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
MASTER OF SCIENCE

UNIVERSITY OF FLORIDA

2010

































2010 Gail Morris
































"[T]hinking about rats, as low-down as it seems, can easily lead to thoughts about larger
topics, such as life and death and the nature of man."





Robert Sullivan, Rats: Observations on the History & Habitat of the City's Most
Unwanted Inhabitants









ACKNOWLEDGMENTS

I received much support during the course of this project. The University of Florida

and the Joseph W. Jones Ecological Research Center provided funding, equipment,

and manpower. My co-chairs Madan Oli and Mike Conner and committee member Mel

Sunquist gave a great deal of assistance, especially with experimental design, analysis,

and editorial suggestions for this thesis. I am also grateful for the help of numerous

technicians who assisted with data collection, entry, and management. Cat Eddins and

Jen Eells got the project off to a good start and helped work out early difficulties.

Bonnie Fairbanks, Amanda Goldberg, and Jamie Utz held down the fort, catching and

chasing many rats while I was away from the field for a semester. James Miller, Megan

Munroe, and Jud Swart, braved the summer heat and gnats in pursuit of ridiculous

number of rats. Evan Hill, Amy Kryzton-Presson, and Erica Rigsby braved heat, gnats,

mosquitoes, and the absence of the lab's fearless leader in pursuit of even more rats

and made tracking around the fire a success. Hayden Martin and Cliff Rushton brought

the field work to a close and assisted with proofing tens of thousands of trapping

records. Jimmy Atkinson, Scott Smith, Cat Eddins, and Briant Williamson carried out

the prescribed burns. The Herpetology lab, especially Jen Linehan and Kelly McKean,

caught many rat-eating snakes and cared for them until the snakes passed the radio

collars. Anna Derrick showed me the smammaling ropes with such great enthusiasm

my own almost seems normal. I am grateful for her advice which I am sure saved me

from many mistakes and frustration. Brent Howze was helpful with many random

aspects of fieldwork, especially with those relating to radio telemetry. Jean Brock,

Michael Simmons, and John Merritt provided assistance with GIS, data management,

and IT support respectively. Liz Cox acquired even the most obscure references with









great speed. I am also grateful for the support of the administrative staff at both the

Jones Center and the Department of Wildlife Ecology and Conservation at UF.

Two individuals deserve special thanks. Jessica Rutledge provided assistance

with field work, data management, equipment, and pretty much everything else

imaginable. Jeff Hostetler provided much patient assistance with data analysis and also

made editorial suggestions. I don't care to think about how much more difficult this

would have been without the help of these two individuals.

I also thank my family and friends for their support and for not saying things like

"Ew, rats!" all the times I talk about rats. I thank my personal aggravation of rat, and her

less long-lived partner, for being highly entertaining despite a prickly nature, and for

showing me many things about the ratty nature.









TABLE OF CONTENTS

page

ACKNOWLEDGMENTS .................................... ............... 4

LIST O F TA BLES ......... ...... ...................................................................... 8

LIST OF FIGURES.................................. ......... 11

A B S T R A C T .............. ..... ............ ................. .................................................. 1 2

CHAPTERS

1 INTRODUCTION .............................................. 14

2 EFFECTS OF PRESCRIBED FIRE, SUPPLEMENTAL FEEDING, AND
MAMMALIAN PREDATOR EXCLUSION ON COTTON RAT SPACE USE AND
POPULATION DYNAMICS IN A LONGLEAF PINE ECOSYSTEM ........... 18

Introduction .................... ............. ............... 18
M methods ....................................... .. ..................... ................ ......... 20
Study Site and Species .................... ....................... 20
Experimental Design .................... ......... ................ 21
Field Methods .......................... ......... ........... 21
Statistical Methods .......................................... 23
Analysis of capture-mark-recapture (CMR) data .................... ............... 23
Survival analysis using radio telemetry data ....... ............. .... ....... 28
Hom e range analysis ..................................... .... ............... ......... 28
R e s u lts .......................................... ................................................. 3 0
CMR Analyses................................ ............. 30
Analyses of Radio Telemetry Data ........................ ...... ....... 32
Collared rat survival ............................................... 32
Effects on home range size................... ..... .................... 33
Effects on home range exclusivity....................................... 34
Effects of fire on radio collared rats.................... ............ 34
D iscussio n ........................................................................................... .............. 36
Fire Effects .................. ... ............................ ..... ....... ............... 36
Predation and Supplemental Feeding Effects ....................... ......... 37
Home Range Exclusivity ................... .......... ...................... 40
C o nclusio ns ....................... ................. ............................................ .. 4 2

3 EFFECTS OF SUPPLEMENTAL FEEDING, MAMMALIAN PREDATOR
EXCLUSION, AND PRESCRIBED FIRE ON COTTON AND OLDFIELD
MOUSE POPULATIONS IN A LONGLEAF PINE ECOSYSTEM ............. 56

Introduction ................. ...... ..... ............ ....................... 56
M e th o d s ....................... ................. ......................................................... 5 7









Study Site and Species .............................................................. .......... 57
F ie ld M e tho d s ...................................................... 5 8
S tatistica l M ethods ............................................... ....................... ........ 59
R e s u lts ......................................................................................... 6 6
Cotton M ice ...................................................................... ......... 66
Oldfield Mice............................................ 68
Discussion ......................................... ................ ......... ............... 70
Treatment Effects on Cotton Mice .......................................................... 71
Treatm ent Effects on O ldfield M ice ................................................................ 74
C conclusions .................................. ........................... ....................... 76

4 CO NCLUSIO NS ......................................................... ................... 92

LIST O F REFER ENC ES ....................................................... ....... ......... 94

B IO G R A P H IC A L S K E T C H ........................................................... ............... 10 1









LIST OF TABLES


Table page

2-1 Model comparison table for multistate capture-mark-recapture analysis
assessing occurrence of peak breeding seasons of cotton rats in general
(Model set 1) and with respect to winter prescribed burns (Model set 2) in
Southwestern Georgia between 2005 and 2009........................ ............... 43

2-2 Model comparison table for multistate capture-mark-recapture analysis
assessing sex, reproductive condition, breeding season and time effects on
capture probability (p), survival (S), and rates of transitions between
reproductive states (P) in cotton rats in Southwestern Georgia between 2005
and 2009. ............................................................. .......... ......... 44

2-3 Model comparison table for multistate capture-mark-recapture analysis
assessing the term of effect of winter prescribed burns on survival (S) of
cotton rats in Southwestern Georgia between 2005 and 2009........................ 45

2-4 Model comparison table for multistate capture-mark-recapture analysis
assessing potential for paired site effects on survival (S) and rate of
transitions to reproductive states (P) of cotton rats in Southwestern Georgia
betw ee n 2 005 a nd 2 009 ....................................... ................... .......................... 46

2-5 Model comparison table for robust design capture-mark-recapture analysis
assessing demographic and time effects on abundance (N), capture
probability (p), and recapture probability (c) in cotton rats in Southwestern
Georgia between 2005 and 2009.. ......................................... ............... 47

2-6 Model comparison table for multistate capture-mark-recapture analysis
examining the effect of predation, supplemental feeding, and fire treatments
on survival (S) and transition probabilities (Y,between reproductive and
non-reproductive states) of cotton rats in Southwestern Georgia between
2 0 0 5 a n d 2 0 0 9 ................ .......................................................... 4 8

2-7 Factors influencing survival of radio collared cotton rats in sites treated with
supplemental feeding, winter prescribed fires, and mammalian predator
exclusion in southwestern Georgia from June 2007 August 2009 ................. 49

2-8 Factors influencing home range size of cotton rats in southwestern Georgia
using 95% minimum convex polygon (MCP) and 95% fixed kernel (Kernel)
estimates. ............ ............ ..... .......................................... 50

2-9 Home range exclusivities for cotton rats in sites treated with supplemental
feeding and mammalian predator exclusion in southwestern Georgia from
June 2007 to August 2009 ...................................................................... 51









2-10 Factors influencing home range exclusivities of cotton rats in sites treated
with supplemental feeding and mammalian predator exclusion in
southwestern Georgia from June 2007 to August 2009................................... 52

3-1 Model comparison table for multistate capture-mark-recapture analysis
assessing occurrence of peak breeding seasons of cotton mice in general
(Model set 1) and with respect to winter prescribed burns (Model set 2) in
Southwestern Georgia between 2005 and 2009................ ..... ................. 77

3-2 Model comparison table for multistate capture-mark-recapture analysis
assessing occurrence of peak breeding seasons of oldfied mice in general
(Model set 1) and with respect to winter prescribed burns (Model set 2) in
Southwestern Georgia between 2005 and 2009................ ..... ................. 78

3-3 Model comparison table for multistate capture-mark-recapture analysis
assessing sex, reproductive condition, breeding season, and time effects on
capture probability (p), survival (S), and transitions rates between
reproductive states (P) in cotton mice in Southwestern Georgia between
2005 and 2009. See Table 3-1 for column definitions. ..................................... 79

3-4 Model comparison table for multistate capture-mark-recapture analysis
assessing sex, reproductive condition, breeding season, and time effects on
capture probability (p), survival (S), and transitions rates between
reproductive states (P) in oldfield mice in Southwestern Georgia between
2005 and 2009 .. ....................................... ...... ................ .............. 80

3-5 Model comparison table for multistate capture-mark-recapture analysis
assessing term of effect of winter prescribed burns on survival of cotton and
oldfield mice in Southwestern Georgia between 2005 and 2009........................ 81

3-6 Model comparison table for multistate capture-mark-recapture analysis
assessing potential for site effects on survival (S) and rate of transitions to
reproductive states (P) of cotton and oldfield mice in Southwestern Georgia
between 2005 and 2009. ................................................ .. ........ 82

3-7 Model comparison table for robust design capture-mark-recapture analysis
assessing demographic and time effects on abundance (N), capture
probability (p), and recapture probability (c) in cotton mice in Southwestern
Georgia between 2005 and 2009. ........... .... ...... .................. ............. 83

3-8 Model comparison table for POPAN capture-mark-recapture analysis
assessing site and time effects on abundance (N), capture probability (p),
and entry probability (pent) in oldfield mice in Southwestern Georgia between
2005 and 2009 .. ....................................... ...... ................ .............. 84









3-9 Model comparison table for multistate capture-mark-recapture analysis
examining the effect of predation, feeding, and fire treatments on survival (S)
and transition probabilities (P, between reproductive and non-reproductive
states) of cotton mice in southwestern Georgia, 2005-2009............................... 86

3-10 Model comparison table for multistate capture-mark-recapture analysis
examining the effect of predation, feeding, and fire treatments on survival (S)
and transition probabilities (P, between reproductive and non-reproductive
states) of oldfield mice in southwestern Georgia, between 2005 and 2009........ 87









LIST OF FIGURES


Figure page

2-1 Model averaged estimates of survival of cotton rats in southwestern Georgia
between 2005 and 2009 in response to prescribed fire, supplemental
feeding, and predator control treatments .................. ................................. 53

2-2 Model averaged estimates of the rates of transitions to reproductive states
for male and female cotton rats during peak breeding seasons (spring and
summer), non-peak seasons during which burning did not occur, and
non-peak seasons during which burning did occur, in southwestern Georiga
betw een 2005 and 2009. ............................ ........... ............. ............... 54

2- 3 Survival estimates ( standard error) of radio collared cotton rats in
southwestern Georgia between 2007 and 2009, generated using Cox
proportional hazard models. ...................................... ............... 55

3-1 Model averaged estimates of survival of cotton mice in southwestern Georgia
between 2005 and 2009 in response to prescribed fire, supplemental
feeding, and predator control treatments .... ...... ............. ......... ........ 88

3-2 Model averaged estimates of breeding transitions for male and female cotton
mice in southwestern Georgia between 2005 and 2009 during peak breeding
seasons (fall and early winter), non-peak seasons during which burning did
not occur and non-peak seasons during which burning did occur. ....... ........ 89

3-3 Model averaged survival estimates in southwestern Georgia between 2005
and 2009 in response to prescribed fire, supplemental feeding, and predator
control treatm ents .......... ......... ......... ........... ................ .............. 90

3-4 Model averaged estimates of breeding transitions for oldfield mice in
southwestern Georgia between 2005 and 2009 during peak breeding
seasons (winter and summer) in burn and non-burn years and non-peak
breeding seasons. ..................... ................. ................ ............... 91









Abstract of Thesis Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Master of Science

EFFECTS OF MAMMALIAN PREDATOR EXCLUSION, SUPPLEMENTAL FEEDING,
AND PRESECRIBED FIRE ON SMALL MAMMAL POPULATIONS IN A LONGLEAF
PINE ECOSYSTEM

By

Gail Morris

August 2010

Chair: Madan K. Oli
Major: Wildlife Ecology and Conservation

Food resources and predation play important roles in determining small mammal

population dynamics. Each of these factors may affect population parameters such as

abundance, survival, and reproduction. These factors may also interact as individuals

under predation pressure make trade-offs between exposure to predators and access to

food resources. Fires consume food sources and reduce cover, which increases

exposure to predators. For species that occur in areas with frequent fires, it is beneficial

to consider how all of these factors affect populations of interest. This study used a

capture-mark-recapture framework to experimentally examine how supplemental

feeding, mammalian predator exclusion, and prescribed fire affected survival,

abundance, and reproduction of cotton rats (Sigmodon hispidus), cotton mice

(Peromyscus gossypinus), and oldfield mice (P. polionotus). Radio telemetry was used

to assess home range size and overlap of cotton rats. Among cotton rats, prescribed

fire events had the greatest effect, causing large drops in survival, abundance, and

reproduction. Food supplementation increased survival, rates of transitions to

reproductive states, and abundances but was not sufficient to prevent post-fire declines









in any of these parameters. This indicates that predation and emigration are

responsible for fire-related declines in survival and abundance. Population-level effects

of predator exclusion effects were small in magnitude. Predator exclusion was,

however, associated with increased male home range size, indicating a response to

perceived predation risk. Among cotton mice, survival was affected (increased) only by

an interaction between burning and predator exclusion. Rates of transitions to

reproductive states decreased in burn years but increased with the combination of

feeding and predator exclusion. Feeding increased abundances. Among oldfield mice,

survival and abundance were greater in predator exclusion areas than controls.

Feeding and the interaction of feeding and predator exclusion also increased

abundances. Rates of transitions to reproductive states declined during peak breeding

seasons during which burning occurred such that breeding transitions in these seasons

were lower than in non-peak seasons. Some of these effects can only be understood

by assuming individuals make behavioral responses to predation risk to limit mortality.









CHAPTER 1
INTRODUCTION

Predation and access to food resources can have large and direct effects on small

mammal populations. Field studies have shown food addition to be associated with

increases in abundance and survival, changes in rates of immigration and emigration,

earlier reproduction, and greater numbers of young produced per reproductive event

(Taitt and Krebs 1983, Boutin 1990, Campbell and Slade 1995, Krebs et al. 1995,

Hubbs and Boonstra 1997, Perrin and Johnson 1999). Experimental removal or

exclusion of predators has been associated with increased densities (Schnell 1968,

Weigert 1972, Meserve et al. 1993, Yunger 2004), increased survival (Meserve et al.

1993, Oli and Dobson 2003), earlier breeding (Arthur et al. 2004), and increased

immigration (Tait and Krebs 1983, Perrin and Johnson 1999).

Indirect effects related to these factors may also be common and often result from

behavioral responses associated with changes in predation or food resources. For

example, food addition may cause changes in intraspecific aggression, and decreases

in home range sizes (Boutin 1990, Desy et al. 1990, Hubbs and Boonstra 1998).

Increased levels of predation risk have been associated with decreases in home range

size and alterations in habitat use (Desy et al. 1990, Dickman 1992, Arthur et al. 2004,

Yunger 2004). Behavioral changes may in turn affect vital rates at the population level.

For example, reductions in space use associated with predation risk may limit access to

reproductive opportunities. Indirect effects may be as large as or larger than direct

effects, and may either take the form and direction of direct effects, or cause

unexpected effects (Pressier et al. 2005).









Although food and predation may individually have large impacts on a population,

there is also a great deal of theoretical and empirical support for interactions between

these factors (Taitt and Krebs 1983, Abrams 1983, 1984, 1991, 1992a, 1992b, 1993,

McNamara and Houston 1987, Desy and Batzli 1989, Hubbs and Boonstra 1997, Clark

and Mangel 2000). In a large-scale experiment examining the effect of food

supplementation and predator exclusion on snowshoe hare (Lepus americanus)

populations, Krebs et al. (1995) reported that food addition tripled population densities,

predator exclusion doubled population densities, and the combination of treatments

resulted in an eleven-fold increase in population densities. Similarly, Hubbs and

Boonstra (1997) found that while feeding and predator exclusion treatments each

individually increased survival of arctic ground squirrels (Spermophilus parryii), survival

in a combined treatment plot was significantly greater than either treatment alone.

Similar trends were found with respect to increased growth rates (Hubbs and Boonstra

1997) and decreasing home range size (Hubbs and Boonstra 1998), indicating a

significant interactive effect on several aspects of species ecology and behavior.

These interactions may stem from the fact that foraging behaviors are likely to

put individuals at greater risk of predation. An individual must make trade-offs between

the need to acquire food and the need to stay safe from predators. Optimally, an

individual should maximize fitness by minimizing risk of death by predation while

maximizing food intake. Rates of food intake are important to avoid starvation and

because reproduction is often associated with the quantity and quality of food available.

A large body of theoretical research suggests that such trade-offs are common

(reviewed in Lima and Dill 1989), may be complex, and may cause counterintuitive









effects (Abrams 1991, 1992a, 1992b, 1993). This is especially likely to be true in

systems with complex food chains and competitive interactions between multiple

predator and prey species. Effects of trade-offs may also depend on a species' life

history and may change with breeding/non-breeding seasons (Abrams 1991).

Theoretical investigations are crucial to understand and predict predator-prey

interactions; however, empirical field data are also necessary to test predictions made in

theoretical studies. Unfortunately, relatively few studies have adequately examined

these factors. Many studies that attempt to do so are aquatic in nature, focus on small

temporal and spatial scales, and/or lack replication. Long term, replicated experimental

studies that consider large spatial scales and examine predator-prey interactions

between terrestrial vertebrates are rare because of the difficulties and expense required

to adequately carry out such experimental studies (but see Krebs et al. 1995).

The primary objective of this study was to experimentally examine how predation

and food resources affect small mammal survival, reproduction, and abundances in a

longleaf pine ecosystem. This study was carried out over four and half years and over a

large spatial scale. We focused on three target species common to the longleaf pine

ecosystem: cotton rats (Sigmodon hispidus), cotton mice (Peromyscus gossypinus),

and oldfield mice (P. polonious). Capture-mark-recapture methodologies were used to

estimate population parameters. Space use, in the form of home range size and

overlap, was additionally considered for adult cotton rats.

The roles of food resources and predation may take on new aspects for small

mammal species following fires, which occur frequently (historically over intervals of one

to five years) in longleaf pine ecosystems. Fires consume both herbaceous food









resources and vegetative cover which increases predation risk. Different small mammal

species exhibit different short-term reactions to fires. Cotton rat populations decline

dramatically (Layne 1974, Bock and Bock 1978) while cotton mice exhibit either a

neutral or short-term positive response to fire (Shadowen 1963, Hatchell 1964, Layne

1974, Suazo et al. 2009). Oldfield mice do not appear to respond strongly to fire over

the short term (Arata 1959, Odum et al. 1973, Suazo et al. 2009). All three of these

species benefit from frequent fire application over the long term (Masters et al. 2002,

2007, Suazo et al. 2009). While the long term benefits of fire are likely related to the

maintenance of ideal habitat structure (Masters et al. 2002), the driving factors behind

population changes over the short term remain poorly understood, although

explanations relating to changes in food resources or cover, and by extension predation

risk, are generally hypothesized. Therefore, a secondary objective of this study was to

experimentally examine the roles of predation and food resources in small mammal

population changes following prescribed fires.









CHAPTER 2
EFFECTS OF PRESCRIBED FIRE, SUPPLEMENTAL FEEDING, AND MAMMALIAN
PREDATOR EXCLUSION ON COTTON RAT SPACE USE AND POPULATION
DYNAMICS IN A LONGLEAF PINE ECOSYSTEM

Introduction

Food availability and predation can have dramatic effects on small mammal

populations. Abundant food resources have been associated with increases in

reproduction (earlier and/or longer reproductive seasons, more young produced per

reproductive event), survival, abundance, and immigration (Boutin 1990, Campbell and

Slade 1995, Webb et al. 2005).

Predation may also have large effects. At its most basic level, predation

removes individuals from a population causing an immediate decrease in abundance.

Predation may also have sub-lethal effects relating to perceived risk of predation.

These include changes in behavior such as reduced activity, changes in habitat use,

decreased home range size, and delay in age of first reproduction or in the onset of a

reproductive season (Lima and Dill 1989). Non-lethal effects that may result from these

behaviors include decreased growth rates, poorer body condition, and decreased

overall reproductive output (Hik 1995, Peckarsky et al. 2008). Non-lethal effects on

prey demographics may be as important as or more important than direct consumption

(Werner and Peacor 2003, Preisser et al. 2005).

Clearly, changes in population parameters caused by changes in food resources

and/or predation can have significant effects on population growth and abundance. For

rodent species which mature and reproduce rapidly ("fast" species), changes in

reproductive rates can have especially large impacts at the population level (Oli and

Dobson 2003).









Effects relating to predation and food resources do not occur in isolation. In the

presence of predation pressure, individuals must make choices that balance predation

risk with the need to engage in activities that put them at risk of predation such as

searching for food (McNamara and Houston 1987, Abrams 1991). Individuals avoid

foraging in areas where they are vulnerable to predators even if good forage occurs

there, unless they are food stressed such that the risk of death by starvation outweighs

the risk of predation. Similar tradeoffs may occur with respect to other behaviors, such

as seeking out reproductive opportunities (Lima and Dill 1989, Clark and Mangel 2000).

For species such as the cotton rat (Sigmodon hispidus), the occurrence of fires

causes a crisis of both food and predation. Fire consumes herbaceous vegetation that

cotton rats require for cover from predators and which they also use as a primary food

source (Whitaker and Hamilton 1998). Multiple authors have noted precipitous declines

in cotton rat populations following fires (Arata 1959, Bock and Bock 1978, Rehmeier et

al. 2005). We know of no studies that have attempted to experimentally determine the

driving forces behind these declines, although hypotheses relating to food resources

and/or predation are generally proposed.

Many studies have examined how predation affects individuals sub-lethally by

influencing behavioral decisions; fewer have been able to connect these decisions to

population-level effects. Fewer still have examined such effects on large spatial and

temporal scales. Experimental manipulation of predators and food resources allows a

better understanding of how predation and food resources influence both individual

behaviors and population dynamics. We examined the effects of predation and food

resource availability on demography and space use of cotton rats over the course of









four and a half years and through three prescribed fire cycles by experimentally

manipulating presence of mammalian predators and food resources. Using

capture-mark-recapture (CMR) and radio telemetry data, we determined the impact of

food resources and predation on cotton rat space use and population dynamics in

general and following prescribed fire events.

Methods

Study Site and Species

This research was conducted at the Joseph W. Jones Ecological Research Center

at Ichauway in Baker County, Georgia. Ichauway is a 12,000 ha property consisting

primarily of longleaf pine (Pinus palustris) and wiregrass (Aristida beyrichiana)

ecosystem. Longleaf pine ecosystems are characterized by a low-density longleaf pine

over-story, a diverse, herbaceous groundcover, and an open, park-like mid-story (Van

Lear et al. 2005). Hardwood tree species occur at limited levels. Frequent, low

intensity fires are key ecological processes. Consequently, application of prescribed fire

is a primary management tool throughout Ichauway; most sites are burned on a two

year rotation (Atkinson et al. 1996).

Cotton rats are solitary, crepuscular rodents found abundantly across the

southeastern United States. They occur in many habitats, but require thick cover,

particularly in the form of dense grasses and shrubs (Goertz 1964). Cover is essential

for protection from a wide range of avian, mammalian, and snake predators.

Herbaceous material is also consumed as a primary food source and used in nest

construction. Predation is the most common cause of death among cotton rats (Weigert

1972, Derrick 2007), and predation pressure is so strong that cotton rat populations

experience near complete turnover in as little as five to eight months (Goertz 1964).









Experimental Design

In 2002, the Jones Center constructed four mammalian predator exclosures, each

paired with a nearby control with similar habitat. Plots range in size from 35.94 to

49.09 ha. Exclosures are surrounded by 1.2 m tall woven wire fences which carry

electrified lines along the top, middle, and bottom to discourage mammals from climbing

over or digging under (the weave is large enough to allow small mammals and snakes

to pass through). Although mammalian predators occasionally enter exclosures, regular

monitoring by track counts and thermal camera surveys indicate significantly fewer

mammalian predators in exclosures than controls (Conner at al. 2010).

In June of 2007, two exclosure and two control grids were randomly selected to

receive a supplemental feeding treatment consisting of placing 113 g (4 oz) of rabbit

chow in cans at every other station trapping grids in each control and exclosure (see

below). Food was replaced every other week. Empty cans were placed on non-feeding

grids. This treatment continued through August 2009. Images from trail cameras

demonstrated that cotton rats, cotton mice (Peromyscus gossypinus), oldfield mice (P.

polionotus), house mice (Mus musculus), woodrats (Neotoma floridana), flying squirrels

(Glaucomys volans), and eastern cottontails (Sylvilagus floridanus) regularly used

feeding stations. We found no evidence that cans were defended by individuals of any

species.

In February of 2005, 2007, and 2009, all plots were burned according to

Ichauway's burn plan which has these study areas on a two year burn rotation.

Field Methods

Each control and exclosure contains a 12x12 small mammal trapping grid with

15 m spacing between stations. Pairs of grids were trapped four times per year (once









each season) from January 2005 through June 2007 and eight times per year (twice per

season) from July 2007 through the June 2009 using Sherman live traps (H.B. Sherman

Traps, Tallahassee, Florida, USA). A small amount of a granular insecticide was

sprinkled around each trap to prevent deaths due to fire ants. New captures were

marked individually with metal ear tags. Data recorded for all captures included

location, species, sex, weight, age (adult or juvenile, based on weight), reproductive

condition (for males, testes descended or not, for females, if pregnant and/or lactating

or not), and hind foot measurement.

In four of the eight study plots (one fed predator exclosure, one unfed predator

exclosure, one fed predator control, and one unfed predator control plot) cotton rats

weighing 90 g or more (so that collar weight was not > 5% of the rats' mass) were

anesthetized with Isoflurane and fitted with radio collars (Advanced Telemetry Systems

Isanti, Minnesota, USA; Sirtrack Wildlife Tracking Solutions, Havelock North, New

Zealand; and Telenax, Playa del Carmen, Mexico). Following recovery, rats were

released at their capture site. Collared rats were located by triangulation or homing a

minimum of three times per week and located visually once a week to confirm status as

alive or dead. Rats were located using TRXC-2000S (Wildlife Materials Murpheysboro,

Illinois, USA), R-1000 (Communication Specialists, Inc. Orange, California, USA), or

R-2000 receivers (Advanced Telemetry Systems Isanti, Minnesota, USA). When rats

were found dead, the location was searched for sign to classify the event as a slip or

death due to avian, mammalian, snake, or unknown predation, handling, or unknown

causes. If rats slipped or chewed off collars or if signals were lost, attempts were made

to retrap and recollar rats. Searches for missing signals were undertaken if rats could









not be retrapped. Collaring and tracking began in July 2007 and continued through

August 2009.

Collared rats were tracked intensively around the 2009 burn. Rats were homed

on to verify status as alive one to two days prior to burning. Signals were monitored

during the burn and rats were homed on again immediately after the fire. For seven

days following the fire, rats were tracked twice a day. Rats surviving past seven days

were tracked daily for an additional seven days. Regular monitoring was restored

thereafter. Missing rats were searched for immediately to detect emigration.

Trapping, handling, and tracking methods followed recommendations of the

American Society of Mammalogists (Gannon et al. 2007) and were approved by the

University of Florida Institutional Animal Care and Use Committee.

Statistical Methods

Analysis of capture-mark-recapture (CMR) data

Capture-mark-recapture (CMR) data considered for this analysis included 26

sessions from January 2005 through June 2009. Analyses were carried out using the R

2.9.1 (R Development Core Team) package RMark 1.9.6 (Laake and Rexstad 2008) to

build models for program MARK 6.0 (White and Burnham 1999). Capture probabilities

were fixed to 1 for radio collared rats.

Multistate models were used to estimate and model state specific survival (S),

capture probability (p), and transitions between reproductive states (P). States used for

S and L were based on reproductive condition. Males were considered to be in

reproductive condition if testes were descended, females if pregnant and/or lactating.

Therefore, models evaluating effects on S which include a reproductive condition term

estimate and model survival separately for reproductive and non-reproductive









individuals, and models evaluating effects on Y which include a reproductive condition

term estimate and model probabilities of individuals moving between reproductive states

(i.e., rates of non-reproductive individuals entering reproductive states, rates of

reproductive individuals remaining reproductive).

Preliminary investigation considered the potential influence of trapping session,

season, year, and the additive effect of season and year on p. Influence of reproductive

condition and sex was assessed for S and Y. Breeding season was also considered for

Yu. Assessment of effects on p, S and Y was carried out in a sequential fashion. First,

effects on p were considered while modeling S and Y using the most general models

described above for each. Effects on S and Y were then considered in a similar

fashion.

Assessment of goodness-of-fit was carried out using the median c approach in

program MARK (White and Burnham 1999). The median c test indicated a mild

overdispersion (6 = 1.262). Models in each parameter's set were compared using

Akaike's information criterion corrected for small sample size (AICc), after

quasi-likelihood adjustments (QAICc) made using c = 1.262. Models were considered

well supported if they had a AQAICc of less than two. The best supported model within

each parameter's set was selected a base for modeling that parameter in further

analyses.

These analyses indicated that reproductive condition and sex modeled in an

interactive fashion (S(reproductive condition*sex)), were important for describing S

(Table 2-2, model 5). Capture probability was best described as varying by session.

However, standard errors around parameter estimates were large for this model, so it









was excluded from the analysis. The next best supported model, with an additive effect

of year and season, p(year+season), was selected to model p (Table 2-2, model 1).

To model Y, we investigated the best approach to modeling a breeding season

effect on Yu. Breeding seasons vary over the cotton rat's geographic range (Cameron

and Spencer 1981, Whitaker and Hamilton 1998). In the southern part of the range,

breeding occurs year round but with peaks at certain times of the year. Because a

literature review did not give a clear indication of when peak seasons occur in

southwestern Georgia, we created a model set, based on the literature (Cameron and

Spencer 1981, Whitaker and Hamilton 1998) and personal observations, to identify

peak and non-peak breeding seasons. This analysis indicated reproductive peaks in

spring and summer (Table 2-1, model 1).

Because of potential confounding effects of prescribed burning treatments and

breeding seasons occurring as occasion-dependant effects, we then assessed whether

there was support for further dividing the non-breeding seasons (winter and fall) into

whether a burn occurred during these seasons or not. Division of non-breeding

seasons into two classes based on burning was well supported (Table 2-2, model 7).

Using this breeding season model (Table 2-2, model 7), we continued the

sequential variable selection for L as described above for p and S. Reproductive

condition, sex, and breeding season, modeled in an additive fashion ( (reproductive

condition+sex+breeding season)), were important for describing P (Table 2-2,

model 10).

Although the prescribed fires occurred at specific times, fire-caused changes in

cover and food resources may last for weeks or months. To determine the best effect









window for the fire treatments, a set of models considering fire effects on survival over

multiple time intervals was considered. Survival was constrained to be similar between

all trapping periods except those following fires. Post-fire survival was allowed to vary

for several different intervals, from including only the interval during which the fire

occurred (interval length of ten weeks following the 2005 and 2007 fires and of five

weeks following the 2009 fire), to including intervals through the summer season

(30weeks), by which time vegetation is typically recovered. This analysis indicated a

short term fire effect on survival with declines occurring only in the interval during which

the fire occurred (Table 2-3, model 1).

Exclosure and control sites were initially selected as pairs based on similar

habitats between pairs. As part of a post-hoc examination of non-treatment effects on S

and LP, we examined the potential for paired site effects on these parameters. Using

the base models indicated by the analyses described above, we ran a second set of

models considering paired site effects on S and LP. This analysis indicated paired site

effects were important for modeling both S and L (Table 2-4, model 1).

Treatment effects were added to the best base multistate model (S(reproductive

condition*sex+site)p(year+season)l(breeding season+reproductive

condition+sex+site)) as additive and interactive effects (two-way only). Due to

confounding effects relating to fire and breeding season occurring as

occasion-dependent effects, only food and predation treatments were considered with

respect to L while food, predation, and fire effects were considered with respect to S.

Model averaging was employed to generate parameter estimates for S and LP.









Pollock's robust design (Pollock 1982) was used to generate abundance estimates

(N). Robust design models estimate probabilities for survival (S), capture (p), recapture

(c), emigration (y"), and staying away after emigrating (y'). The model selection

approach used for the robust design was similar to that used for the multistate analysis.

Preliminary investigation considered potential for time effects on p and c. Paired site

effects were considered for N. S was modeled using the best supported S model from

the multistate analysis (without reproductive condition, S(sex+site+food+fire)). y terms

were modeled using a random emigration effect (y"(.)=y'(.)). Preliminary analyses

indicated that c and p varied by trapping session (Table 2-5 model 3) and that a paired

site effect, as described above, was important for modeling N (Table 2-5, model 1).

Because of the difficulties associated with modeling treatment effects on

abundance directly (White 2002), the best supported robust design model indicated by

the preliminary analysis described above (S(site+sex+food+fire)y"(.)=y'(.)p(session)

c(session)N(site)) was used to generate derived abundance estimates by site and

session, but not to determine treatment effects. Treatment effects on abundance were

evaluated using a repeated measures ANOVA (Schabenberger and Pierce 2002)

implemented using the PROC MIXED procedure in SAS (SAS Institute Inc. 2004). The

model for this ANOVA included food, fire, and predation treatments and their

interactions (two-way interactions only). Paired sites were included as a random effect.

Multiple covariance structures were investigated and the best (variance components

structure, which allows a different variance for each random effect) was selected based

on AICc value (Miller et al. 2004). Treatment effects were considered significant at the

a = 0.05 level.









Survival analysis using radio telemetry data

Survival of collared rats was estimated using the Cox proportional hazard model

(Cox 1972) implemented using the PROC PHREG procedure in SAS (Allison 1995). An

a priori model set was constructed to examine the effects of season, treatment, sex, and

interactions of these factors on survival. Rats that lived less than one week following

collaring or that died due to handling-related causes were not considered for this

analysis. To prevent an upward bias associated the censoring of rats in the latter

group, the first week of all rats was censored as well. Rats whose signals were lost and

rats who lost collars were right-censored. Seasons were pooled except for winters

which were separated into winter 2008 (no burning) and winter 2009 (burning occurred)

to test for a fire effect on survival. Models were assessed using an AIC framework.

Home range analysis

Home ranges were estimated using 95% minimum convex polygons (MCP) and,

for purpose of comparison to other studies, 95% fixed kernel methods. To avoid short

sampling intervals and small sample sizes which may contribute to inaccurate home

range estimates (Swihart and Slade 1985a, Swihart and Slade 1985b, Spencer et al.

1990), we followed the recommendations of Cameron and Spencer (1985) and Swihart

and Slade (1985b) and estimated home ranges only for rats that had a minimum of 15

locations with least 4.5 hours between locations. MCP estimates were generated using

the program CALHOME (Kie et al. 1994). Kernel estimates were generated in ArcGIS 9

(ESRI 2005) using the Hawth's tools extension (Beyer 2004). Kernel bandwidth was

specified by least squares cross validation (LSCV, Seaman and Powell 1996) for all

rats, with the exception of nine rats that were repeatedly found at the same locations)

(usually a nest or burrow). For these rats, the bandwidth was user specified to prevent









bias such behavior can create in home range estimates using LSCV to determine

bandwidth (Seaman and Powell 1996). Home range estimates were annual, not

seasonal (although "annual" is perhaps misleading as no rats lived to be collared for

more than a year).

A t-test indicated that male rats had significantly larger home ranges than

females; therefore, males and females were considered separately in further analysis.

Effects of feeding, predation and the interaction of these treatments on home range size

were examined using a two-way ANOVA (SAS procedure PROC GLM, SAS Institute

Inc. 2004, Schabenberger and Pierce 2002). Fire was not included as a treatment in

this analysis due to small post-fire sample sizes.

Home range exclusivity was estimated by identifying all pairs of rats that lived

during the same period and which had adjacent or overlapping MCP home ranges.

Distances between pairs of such individuals located by radio telemetry within thirty

minutes of each other were measured in ArcGIS 9. Distances between randomly

selected locations for each pair were also measured. Averaged distances were

subtracted (average real average random distance) for each pair to generate an

estimate of exclusivity. Positive differences are interpreted to indicate a pair's

avoidance of each other, while negative differences indicate an affinity. Effects of

treatments on this measure were examined using a two-way ANOVA implemented in

PROC GLM in SAS. Independent variables included in these models were type of

pairing (male/male, female/female, male/female), feeding treatment, predation

treatment, and interactions of type of pairing with feeding and predation treatments









(two-way interactions only). Fire was not included as a treatment due to small post-fire

sample sizes.

Results

CM R Analyses

Over 26 trapping sessions in eight trapping plots, 2557 individual cotton rats (6815

total captures) were trapped. The multistate analysis results show six models with a

AQAICc < 2, none with particularly strong support over the others (Table 2-6). It is

clear from the top ranked models that fire and supplemental feeding effects were

important factors affecting survival. Fire effects appear in the top 30 models and these

hold 100% of the model weight of the set. Food effects appear in the top 14 models

and these carry 92.5% of the model set's weight. There is no evidence that

supplemental food or predator exclusion treatments influenced Y. The lack of clear

support for any particular model indicates model selection uncertainty; therefore, model

averaging was employed for parameter estimation.

Overall model averaged survival estimates showed that males had lower survival

than females and that reproductive individuals had lower survival than non-reproductive

individuals (Figure 2-1). Model averaged estimates show large post-fire declines in

survival for both sexes and strata in both the predator exclosures and controls

(Figure 2-1). Post-fire survival was not greatly impacted by the addition of food and still

approached zero. During non-fire periods, food supplementation increased survival for

both sexes and reproductive conditions (Figure 2-1). Survival was greater in predator

exclosures compared to controls, but this difference was marginal. Similarly, predator

exclusion conveyed some benefit to survival post-fire, but this was small in magnitude

(Figure 2-1).









Model averaged estimates of L show that a greater proportion of males made

transitions into reproductive states than females in all seasons (Figure 2-2).

Additionally, most reproductive individuals that achieved a reproductive state stayed in a

reproductive state; this trend was slightly greater for males than females (Figure 2-2).

Initial investigation indicated a strong fire effect on transitions between

reproductive states: models that included three classes of breeding seasons (peaks in

spring and summer, non-peak breeding in falls and winters without burns, and a second

non-peak in falls and winters with burns; hereafter peak, non-peak/non-burn, and

non-peak/burn, respectively) were clearly better supported than models with only two

classes of breeding seasons (peak and non-peak with no distinguishing between burn

and non-burn years, Table 2-1). Two-season breeding season models had no support

(weight = 0.0) compared to three-season models (Table 2-1, model 7).

Model averaged parameter estimates indicated that transitions to reproductive

states were at their highest during peak breeding seasons but that there was only a

small drop in transitions to reproductive states during non-peak/non-fire seasons

(Figure 2-2). However, transitions to reproductive states dropped considerably during

non-peak/fire seasons (Figure 2-2). Addition of food increased transitions to

reproductive states while predator exclusion had a minimal effect on this parameter

(Figure 2-2).

The repeated measures ANOVA examining treatment effects on abundance

indicated significant effects of feeding and fire treatments and their interaction

(P = 0.001, P < 0.001, and P = 0.045 respectively). Examination of least square means

showed that supplemental feeding increased abundances by 1.9 times and burning









caused a 3 fold decline in abundance. Although the interaction of feeding and burning

was marginally significant, burning in feeding areas still produced large declines in

abundance (by 2.9 times) indicating that this interaction is not biologically significant.

Analyses of Radio Telemetry Data

A total of 279 cotton rats were collared during this study. Of these, 145 had

sufficient locations for home range and exclusivity analysis; 204 met criteria for survival

analysis. The average number of locations per home range ( SE) was 29.83 1.29

(range 15-92). Male home ranges were significantly larger than female home ranges

(P < 0.001). Average home range size was 2948 m2 (range 685-9814 m2) for female

rats (N = 72) using MCP and 5983 m2 (range 888-20917 m2) using fixed kernel

methods. For males (N = 73) the average home ranges were 7891.5 m2 (range

150-30590 m2) and 15845.31 m2 (range 511-84248 m2) using MCP and kernel methods.

Although kernel estimates were substantially larger than MCP estimates, results from

the two methods did not differ qualitatively. Overall, we believe the 95% MCP estimates

provide a more accurate picture of actual space use, and include kernel estimates here

for comparison purposes; therefore, the discussion will primarily focus on results based

on the MCP home range estimates.

Collared rat survival

Of the 204 rats that met criteria for survival analysis, 29 were censored for at least

some time during which they were not tracked (slipped/chewed off collar, or

experienced transmitter failure), but were recollared and reentered into the analysis at

some later point. Sixty-two rats were censored and never reentered into analysis.

Causes for censoring included slipping/chewing off collars (N = 21), emigrating from the

control/exclosure plot where collared (N = 13), transmitter failure (N = 2), and









unexplained loss of the signal (N = 26). The latter could be attributed to transmitter

failure, emigration, or carrying off of the rat and/or collar by a far-ranging predator.

There was strong evidence for seasonal and fire effects on survival of collared rats

(Table 2-7). Models including only spring, summer, and fall effects (models 8, 9, and 11,

Table 2-7) show that spring, summer, and fall survivals were similar (had AICc values

within a range of 2) while survivals during winters where burning occurred and during

winters where burning did not occur differed from each other and from other seasons

(AICc values >2 from other seasonal models). Parameter estimates from the top

ranked model (Table 2-7, model 1; survival varying by season), show that winter

survival in non-burn years was greater than other seasons, while survival in winters of

burn years was quite low (Figure 2-3).

There was poor support for supplemental feeding and predation treatment effects

on collared rat survival; the highest ranked model with a treatment term has moderate

support (AAICc = 4.14, Table 2-7, model 3)). This model indicated an interactive effect

of season with the predator exclusion treatment. Parameter estimates from this model

show that survival was similar between controls and exclosures for all seasons except

winters of burn years, during which time survival was lower in predator controls than

exclosures (Figure 3-3).

Effects on home range size

The only significant treatment effect on home ranges estimated by MCP was for

male rats in the predator treatment (Table 2-8). Examination of least square means

showed that males in the predator exclosures had larger home ranges than males in the

predator controls (P = 0.001). Neither feeding nor interactions of feeding and predator









treatments had significant effects on home range size. Similar results were found when

examining kernel home range estimates.

There was no evidence that feeding or predator exclusion treatments, or their

interaction, had significant effects on female home ranges estimated by MCP

(Table 2-8). Similar tests performed using kernel estimates showed a marginally

significant predation effect (again, larger home ranges in exclosures, P = 0.048), while

feeding and the interaction of feeding and predation treatments showed no significant

effects.

Effects on home range exclusivity

Three hundred and fifty-six pairs of rats had adjacent or overlapping home ranges

during the same time period. Analysis of home range exclusivity suggested no effect of

type of pair or treatment on spacing between rats (Tables 2-9 and 2-10).

Effects of fire on radio collared rats

Thirty-three collared rats were collared and tracked during the February 2009

burns. All rats survived the fire itself by sheltering in holes within their home ranges or

nearby areas that did not burn completely. Forty-one percent of the rats died due to

predation, 34% emigrated to small, unburned patches within the larger burn area and

19% emigrated to unburned areas completely outside of the burn unit. All but one rat

either died or emigrated within seven days of the fire (this rat died twelve days post-fire).

Predation and feeding treatments did not significantly affect response to fire (P > 0.05).

Of the remaining two rats, one chewed its collar off in a hole in the burn area; the

other stayed within the burn area (an unfed predator exclosure) and apparently died due

to starvation. This rat was found near the entrance of a hole within its home range

seven days post-fire. A necropsy revealed that the rat had lost 19% of her body weight









since she was last trapped (34 days prior). There were no signs of trauma commonly

apparent following mammalian, avian, or snake predation/attempted predation. The

stomach contained primarily ash and dirt. Two rats killed and cached by mammalian

predators in a fed/predator access grid during the same period were also necropsied to

reveal stomachs filled with rabbit chow, indicating the individuals in fed grids used

supplemental food following the fire.

Every rat that moved to an unburned patch within the larger burn area moved

less than 50 m. These unburned patches overlapped or were adjacent to the rats'

pre-fire home ranges; it is likely the rats were already familiar with the unburned patches

they invaded. Six of these rats survived and were captured during the next trapping

period. Five had lost weight during this interval. The mean percent weight loss was

-0.083 % over 33 days. To determine if this was an artifact of the winter season itself,

weight changes of collared rats over the 2008 winter were also calculated. Of 9 collared

rats which were captured in both winter 2008 sessions (35 days between sessions), the

mean weight change was a gain of 0.082 %. Change in weight differed (P = 0.003)

between winter of 2008 and 2009.

Of rats that successfully emigrated to areas outside of the burn unit, most did so in

a single night and moved distances of 50 to 700 m. None appeared to have moved the

shortest distance to the burn edge and/or stopped immediately upon reaching an

unburned area. Most of these rats (all but two) died or went missing within two weeks

of emigrating.









Discussion


Fire Effects

Of the three treatments experimentally applied to cotton rat populations in this

study (mammalian predator exclusion, supplemental feeding, and prescribed fire), fire

had the largest impact on cotton rat populations. Prescribed burning caused precipitous

declines in survival, abundance, and transitions to reproductive states regardless of the

presence of supplemental food or absence of mammalian predators. These results

support the hypothesis that cotton rat declines following fires are due primarily to

predation, secondarily to emigration, and not due to changes in food resource

availability. However, one radio collared rat apparently died of starvation following a

burn and other rats that remained in small, unburned patches in the overall burn unit

lost a significant amount of weight. This suggests that the loss of herbaceous food

sources by burning was indeed a problem for this species, but that the crisis of food

resources was overwhelmed by increased exposure to predators due to loss of cover.

This is not surprising given the cotton rat's heavy cover requirements and general

susceptibility to predation; cotton rats support a wide variety of predators including

snake, mammalian, and avian predators. Predation is by far the most common cause of

death (82% of deaths, Derrick 2007, and here, 76% of collared rat deaths overall) and

cotton rat populations can experience near complete turnover in as little as five to eight

months (Goertz 1964).

These results suggest that in ecosystems where fires are frequent, such as

longleaf pine, cotton rat populations are heavily influenced by fire events. Similar sharp

post-fire declines in cotton rat abundances have been observed in southern pine

forests, native tallgrass prairies, and sacaton grasslands (Arata 1959, Layne 1974, Bock









and Bock 1978, Rehmeier et al. 2005). Although cotton rats experience a short-term

negative fire effect, over the long term cotton rats have a positive association with fire.

Populations in tallgrass prairies in Kansas peaked in autumns of the first two years

following spring burns but declined in autumns thereafter if burns were not repeated

(Rehmeier et al. 2005). Rehmeier et al. (2005) hypothesize that this occurs because

fires enhance growth of plants that serve as food resources and reduce litter that may

inhibit movement through vegetation.

Predation and Supplemental Feeding Effects

Given that population dynamics of species such as cotton rats with rapid

maturation and turnover are more sensitive to changes in reproductive parameters than

to changes in survival (Heppell et al. 2000, Oli and Dobson 2003), it is somewhat

surprising that we found no strong evidence of predator or feeding treatment effects on

reproductive transitions but did observe increased survival in supplemental feeding

plots. This may be explained by canalization of vital rates which have great proportional

impact on population growth rate; such rates tend to have little variation due to heavy

selective pressure on those parameters (Pfister 1998) and are unlikely to be greatly

affected by environmental changes.

Cotton rats are extraordinary reproducers even when compared with other rat

species. Cotton rats become reproductive within one to two months, require 27 days for

gestation, and a female may become pregnant again within 24 hours of giving birth

(Whitaker and Hamilton 1998). Young open eyes within 24 hours, wean at five to six

days, and achieve independence soon after. By contrast, Norway rats (Rattus

norvegicus) open eyes at 14 to 17 days, wean at three weeks and become independent

at four weeks (Whitaker and Hamilton 1998). Given the cotton rat's already accelerated









schedule of development and reproduction, it is difficult to imagine there is much room

for improvement. A similar study on effects of supplemental feeding on cotton rats

found that feeding increased the number and weight of young born in feeding plots but

did not affect juvenile survival or recruitment (Campbell and Slade 1995).

The lack of predation effects observed here is also surprising given the

enormous role predation plays in cotton rat mortality. It is possible that mammalian

predator exclusion alone was insufficient to elicit a response in the parameters

examined. Previous studies examining predator exclusion or removal on cotton rats

(Weigert 1972, Guthery and Beasom 1977) suggest that the effects of such treatments

vary according to the predators excluded or removed. Guthery and Beasom (1977)

removed only mammalian predators from areas where cotton rats occurred, and

detected no change in survival or density. Weigert (1972) excluded all predators from

some study areas and only mammalian predators from others and determined that

avian predators had a greater impact on cotton rat populations than mammalian

predators. These results are consistent with our own which suggest that predation by

raptors and snakes make up for losses when mammalian predators are excluded.

Alternatively, it is possible that predation does not regulate cotton rat populations in

areas or periods where cover is sufficient. This hypothesis is supported by the positive

supplemental feeding effect (increased survival, abundance, and transitions to

reproductive states) observed during non-fire periods.

Although we did not observe strong predation effects on survival, abundance, or

reproductive transitions, predator exclusion was associated with increased male home

range sizes. This indicates a sub-lethal predation effect. The best studied examples of









sub-lethal effects of predation deal with the interplay of food acquisition and predation

risk. Foraging increases exposure to predators, causing individuals to make trade-offs

between the need to eat and the need to minimize predation risk (McNamara and

Houston 1987, Abrams 1991). The lack of feeding effects on home range size suggests

that cotton rats do not make such trade-offs, at least with respect to mammalian

predation. Instead, we found that male rats made trade-offs associated with predator

exclusion alone. How can this be explained outside of a food context? Given the

previously stated importance of reproduction over survival in species that have rapid

turnover and maturation (Oli and Dobson 2003), it follows that male rats, who are

promiscuous, range widely, and have no involvement in raising young would make

predation risk decisions based on maximizing reproductive opportunities rather than

food acquisition. That the same trend was not observed with female rats can be

explained by the likelihood that a female rat will be bred regardless of whether she

encounters a single male or several. While female rats are less likely to influence

reproduction by changing space use, male rats ought to increase fitness by mating with

as many females as possible. Maintenance of larger home ranges should increase the

chances of encountering females and the chances of breeding. However, when

predation pressure is high, maintaining a large home range may increase predation risk

causing male rats to restrict movements.

Although we did not detect a strong predation treatment effect on survival of

trapped rats, we found that radio collared rats had greater survival in controls than

exclosures in winters during which burning occurred (compared to winters where

burning did not occur, during which survival was similar between predator controls and









exclosures). This may indicate additional support for sub-lethal effects of predation, but

we believe it to be due instead to site effects. Specifically, one section of the unfed

predator control which contained collared rats did not burn completely. Six collared rats

had home ranges adjacent to this area and, by moving into the area, which provided

cover, were less vulnerable to predation. Given that these six rats made up a significant

portion of rats in predator control treatments during the burn (N = 18), it is possible that

this skewed the results relating to post-fire predation treatment effects on collared rat

survival. If it were not for this, we believe these results would have been similar to

those observed from the CMR analyses: fire caused a decline in survival and that

decline was not affected by the predation treatment. However, these results indicate

that the negative short-term fire effect can be mitigated if unburned refugia remain

(although rats seemed unable to reliably find such refugia if it occurred at distances

greater than 50 to 70 m from their home ranges).

Home Range Exclusivity

Although our analysis failed to detect a treatment effect on cotton rat home range

exclusivity, it is interesting that previous studies have found that female cotton rats have

more exclusive home ranges than males and that heavier males have more exclusive

ranges than smaller males (Fleharty and Mares 1973, Cameron and Spencer 1985).

We found no significant differences in exclusivities between male and female rats or

between rats in any treatment. The inconsistency in these results may be due to

different methodologies used to quantify home range exclusivity. Cameron and

Spencer (1985) estimated overlap for co-occurring rats tracked by radio telemetry

around sun rise and sunset by calculating percent overlap of MCP home ranges.

Fleharty and Mares (1973) also examined overlap and distances between centers of









activity of individuals, although these estimates were based on home ranges generated

with as few as six trapping locations.

We used a different method to evaluate the tolerance of cotton rats for each

other. Locations were taken by radio telemetry at all times between dawn and dusk,

and exclusivity was evaluated by comparing distances between co-occurring individuals

tracked at the same time to random distances between those individuals. Given the

social system of cotton rats, this may provide a more accurate means of evaluating

interactions between rats. Liu (1971) found in a large-scale laboratory study that cotton

rat populations consist of dominant and subordinate individuals (defined by whether an

individual wins or loses fights submissive behavior was not observed). Subordinate

individuals lived within home ranges of dominant individuals but minimized encounters

by foraging at less desirable times; dominant individuals foraged around dawn and dusk

while subordinates foraged during day or night hours. Subordinate rats defended only

areas immediately around their nests. Dominant rats tended not to have overlapping

home ranges with other dominant rats as encounters between two dominants generally

ended with the death of one rat or the other. Mating pairs shared nests at times,

although they did not forage together, and females moved to new nests which were

defended even from her mate shortly before giving birth and while nursing young.

Since cotton rats use extensively overlapping areas, measuring home range

overlap provides limited insight into territoriality, especially for home ranges generated

from points collected at times when only a subset of the population is likely to have

been active. The exclusivity measure used here allows an indirect examination of









tolerance of cotton rats for each other in a system where individuals use the same

space but are solitary and agonistic towards one another.

Other studies have found that rodents decrease agonistic behavior when provided

supplemental food, but not with respect to predation pressure (Desy et al. 1990). We

were unable to detect changes in tolerance of rats towards each other with predation or

feeding treatments. It is possible that the food provided was insufficient or too greatly

dispersed to allow a decrease in aggression and that the presence of avian and snake

predators prevented behavioral responses at this level to mammalian predator

exclusion.

Conclusions

Our results suggest that cotton rat population dynamics in our study site are

primarily driven by fire events. Population responses following fires appear to be

strongly influenced by fire-caused loss of cover and associated increases in predation.

Direct effects relating to mammalian predation do not appear to be strong, but there is

evidence that male cotton rats respond adaptively to predation risk, by decreasing home

range size in areas where mammalian predators have access compared to areas where

they do not. Food is also important to cotton rats, and caused increases in all of the

demographic parameters considered here. This is likely to be true in both fire and

non-fire periods, but food effects were overwhelmed by predation effects following fires.

We detected no treatment effects on cotton rat behavior associated with spacing

between individuals.









Table 2-1. Model comparison table for multistate capture-mark-recapture analysis
assessing occurrence of peak breeding seasons of cotton rats in general
(Model set 1) and with respect to winter prescribed burns (Model set 2) in
Southwestern Georgia between 2005 and 2009. All models had survival set
as S(reproductive condition*sex) and capture probability set as
p(year+season). Table includes number of parameters (K), model weights
(relative likelihood of models in the set), and difference in Akaike's information
criterion corrected for small sample size after quasilikelihood adjustment
(AQAICc). Quasilikelihood adjustments were made using an estimated c of
1.262.
Model Model K AQAICc Model
no. weight
Model 1 W(spring and summer peaks) 15 0.00 1.00
set 1* 2 W(spring peaks) 15 80.92 0.00
3 W(summer and fall peaks) 15 103.07 0.00
4 W(summer peaks) 15 110.02 0.00
5 W(dot) 14 148.42 0.00
6 W(spring, fall, and summer peaks) 15 150.39 0.00

Model 7 W(spring and summer peaks; winter and fall 16 0.00 1.00
set 2** non-peaks divided by years and non-burn
years)
8 W(spring and summer peaks; winter and fall 15 74.08 0.00
non-peaks)
* Model set 1 considers only two classes of breeding season: peak (listed) and
non-peak (all seasons not listed for a given model).
** Model set 2 compares the best model from set 1 (model number 1) with a similar
model that considers three classes of breeding seasons by breaking down non-peak
seasons into those that occurred during burn years (winters and falls of 2005, 2007, and
2009; burns occurred in winters of these years) and those that occurred during non-burn
years (winters and falls of 2006 and 2008).









Table 2-2. Model comparison table for multistate capture-mark-recapture analysis
assessing sex, reproductive condition, breeding season and time effects on
capture probability (p), survival (S), and rates of transitions between
reproductive states (P) in cotton rats in Southwestern Georgia between 2005
and 2009. See Table 2-1 for column definitions.
Model Model K AQAICc Model
no. weight
Effects on capture probability (p) *
1 p(year+season) 18 0.00 1.00
2 p(season) 14 20.79 0.00
3 p(year) 14 83.26 0.00
4 p(constant) 10 124.66 0.00

Effects on survival (S)**
5 S(reproductive condition*sex) 34 0.00 1.00
6 S(sex) 32 11.18 0.00
7 S(reproductive condition+sex) 33 13.10 0.00
8 S(constant) 31 21.63 0.00
9 S(reproductive condition) 32 23.00 0.00

Effects on rates of transitions to reproductive states***
10 Y(breeding season+reproductive condition+sex) 34 0.00 1.00
11 Y(breeding season+reproductive condition) 33 16.72 0.00
12 Y(breeding season) 32 95.93 0.00
13 Y(breeding season+sex) 33 102.04 0.00
14 Y(constant) 30 324.08 0.00
* Additional parameters modeled as S(reproductive condition*sex) l(breeding season+
reproductive condition+sex).
** Additional parameters modeled as p(session) l(breeding season+reproductive
condition+ sex).
***Additional parameters modeled as S(reproductive condtion*sex)p(session).









Table 2-3. Model comparison table for multistate capture-mark-recapture analysis
assessing the term of effect of winter prescribed burns on survival (S) of
cotton rats in Southwestern Georgia between 2005 and 2009. See Table 2-1
for column definitions. All models had additional parameters of capture
probability (p) modeled as p(year+season) and rate of transition between
reproductive states (P) modeled as Y(reproductive condition+sex+breeding
season).
Model Model K AQAICc Model
no. weight


1 S(fire effect over 5 (2009) to 10 weeks (2005 and 16 0.00 0.98
2007))*
2 S(fire effect over 30 weeks) 16 7.85 0.02
3 S(fire effect over 10 weeks) 16 15.72 0.00
4 S(fire effect over 20 weeks) 16 21.32 0.00
* Model 1 shows a range of intervals because the interval between trapping sessions
changed from 10 weeks (2005 and 2007) to 5 weeks (2009) before the 2009 prescribed
burn.









Table 2-4. Model comparison table for multistate capture-mark-recapture analysis
assessing potential for paired site effects on survival (S) and rate of
transitions to reproductive states (P) of cotton rats in Southwestern Georgia
between 2005 and 2009. See Table 2-1 for column definitions. All models
had capture probability (p) modeled as p(year+season).
Model Model K AQAICc Model
no. weight
1 S(reproductive condition*sex+site) 24 0.00 0.84
Y(breeding season+reproductive condition+
sex+site)

2 S(reproductive condition*sex+site) 21 3.26 0.16
Y(breeding season+reproductive condition+sex)

3 S(reproductive condition*sex) Y(breeding season+ 21 21.70 0.00
reproductive condition+sex+site)

4 S(reproductive condition*sex) Y(breeding season+ 18 25.08 0.00
reproductive condition+sex)









Table 2-5. Model comparison table for robust design capture-mark-recapture analysis
assessing demographic and time effects on abundance (N), capture
probability (p), and recapture probability (c) in cotton rats in Southwestern
Georgia between 2005 and 2009. See Table 2-1 for column definitions.
Survival (S) was modeled as (S(site+sex+food+fire) for all models.
Emigration terms (y" and y') were modeled with a random emigration effect
for all models: y"(.)=y'(.).
Model Model K AAICc Model
no. weight
Effects on abundance (N)*
1 N(site) 64 0.00 1.00
2 N(.) 61 119.80 0.00

Effects on capture (p) and recapture (c)
probabilities**
3 p(session)c(session) 64 0.00 1.00
4 p(session)c(p+c')*** 39 35.02 0.00
5 p(.)c(p+c')**** 14 538941.84 0.00
*Additional parameters were modeled as p(session)c(session). **Additional parameter
was modeled as N(site).
***Indicates capture probability varies by session with a constant trap happy response
recapture response (c').
****Indicates a constant capture probability with a constant trap-happy recapture
response (c').









Table 2-6. Model comparison table for multistate capture-mark-recapture analysis examining the effect of predation,
supplemental feeding, and fire treatments on survival (S) and transition probabilities (Y,between reproductive
and non-reproductive states) of cotton rats in Southwestern Georgia between 2005 and 2009. Capture
probability modeled was modeled as p(year+season) for all models. See Table 2-1 for column definitions.
Bolded text indicates treatment effects (all other effects are similar between models throughout the set). Only
models with a AQAICc < 4 are shown here (the top ranked 12 models of 55 in the overall set).
Model Model K AQAICc Model


no. weight


1 S(reproductive condition*sex+site+food*fire) Y(breeding season+reproductive
condition+site+sex)
2 S(reproductive condition*sex+site+food*fire)
Y(breeding season+reproductive condition+site+sex+food)
3 S(reproductive condition*sex+site+food+predation+fire)
Y(breeding season+reproductive condition+site+sex)
4 S(reproductive condition*sex+site+food+predation+fire)
Y(breeding season+reproductive condition+site+sex+food)
5 S(reproductive condition*sex+site+food+fire) Y(breeding season+reproductive
condition+site+sex)
6 S(reproductive condition*sex+site+food+fire)
Y(breeding season+reproductive condition+site+sex+food)
7 S(reproductive condition*sex+site+food*fire)
Y(breeding season+reproductive condition+site+sex+predation)
8 S(reproductive condition*sex+site+food*fire)
Y(breeding season+reproductive condition+site+sex+food+predation)
9 S(reproductive condition*sex+site+food+predation+fire)
Y(breeding season+reproductive condition+site+sex+predation)
10 S(reproductive condition*sex+site+food+predation+fire)
Y(breeding season+reproductive condition+site+sex+food+predation)
11 S(reproductive condition*sex+site+food+fire)
Y(breeding season+reproductive condition+site+sex+predation)
12 S(reproductive condition*sex+site+food+fire)
U(breeding season+reproductive condition+site+sex+food+predation)


27 0.00


0.11

0.93

1.05

1.79

1.90

2.03

2.13

2.96

3.06

3.82

3.91


0.16


0.16

0.10

0.10

0.07

0.06

0.06

0.06

0.04

0.04

0.03

0.02









Table 2-7. Factors influencing survival of radio collared cotton rats in sites treated with
supplemental feeding, winter prescribed fires, and mammalian predator
exclusion in southwestern Georgia from June 2007 August 2009. Models
and associated AICc rankings and weights for collared rat survival were
estimated using Cox proportional hazard models. See Table 2-1 for column
definitions. "Winter (non-burn)" refers to the winter of 2008 during which no
sites were burned. "I" notation indicates that all additive and interaction
combinations of variables are included in the model. "I" notation indicates
that all additive and interaction combinations of variables are included in the
model.
Model Model K AAICc Model weight
no.
1 Season 5 0.00 0.57
2 Winter (burn)* 2 1.21 0.31
3 Predationlseason 12 4.14 0.07
4 Foodlseason 12 6.67 0.02
5 Winter (non-burn) 2 7.88 0.01
6 Sexlseason 12 9.56 0.01
7 Constant survival 1 13.65 0.00
8 Spring 2 13.72 0.00
9 Summer 2 14.74 0.00
10 Predation 2 14.84 0.00
11 Fall 2 14.92 0.00
12 Food 2 15.24 0.00
13 Sex 2 15.58 0.00
14 Foodlsex 4 17.42 0.00
15 Predationlsex 4 17.84 0.00
16 Predationlfood 4 18.23 0.00
17 Global model 27 26.24 0.00
* Winter (burn) refers to the winter of 2009 during which all sites were treated with
prescribed fire.
** Winter (non-burn) refers to the winter of 2008 during which no sites were burned.









Table 2-8. Factors influencing home range size of cotton rats in southwestern Georgia
using 95% minimum convex polygon (MCP) and 95% fixed kernel (Kernel)
estimates. General linear model results are given with interactive effects.
Degrees of freedom (d.f.), mean square (MS), F-statistic values (F), and
significance level (P) are given for each effect. Home ranges were log
transformed for this analysis.


Method
Male rats
MCP



Kernel



Female rats
MCP


Kernel


Source
Food
Predation
Food*Predation

Food
Predation
Food*Predation


Food
Predation
Food*Predation

Food
Predation
Food*Predation


MS
0.055
1.768
0.008

0.023
2.284
0.003


0.023
0.147
0.111

0.036
0.329
0.161


F
0.40
11.020
0.050

0.17
16.85
0.02


0.29
1.85
1.39


0.44
4.04
1.98


P
0.559
0.001
0.822

0.685
<0.001
0.879


0.592
0.178
0.242

0.509
0.048
0.164









Table 2-9. Home range exclusivities for cotton rats in sites treated with supplemental
feeding and mammalian predator exclusion in southwestern Georgia from
June 2007 to August 2009.
Pair type Treatment N Mean difference (m)*
Female/Female Feeding 9 -1.802.08
Non-feeding 21 -4.222.23
Predator access 16 -5.142.69
Predator exclosure 14 -1.621.84
Male/Male Feeding 32 1.453.30
Non-feeding 25 -2.242.46
Predator access 27 1.002.79
Predator exclosure 30 -1.213.23
Male/Female Feeding 42 -2.501.59
Non-feeding 49 2.051.85
Predator access 49 -0.831.86
Predator exclosure 42 0.871.65
* Exclusivity was estimated by finding actual distances between rats with adjacent
minimum convex polygon home ranges that lived during the same time on days that
they were located within 30 minutes of each other by radio telemetry. Random
distances between each pair were also found. Mean difference is the mean of actual
distances random distances (SE).









Table 2-10. Factors influencing home range exclusivities of cotton rats in sites treated
with supplemental feeding and mammalian predator exclusion in
southwestern Georgia from June 2007 to August 2009. General linear model
results are given with interactive effects. Pair type refers to whether pairs
were female/female, male/male, or male/female. See Table 2-4 for column
definitions.
Source d.f. MS F P
Pair type 2 82.190 0.48 0.620
Food 1 9.289 0.05 0.816
Predation 1 17.267 0.10 0.751
Type*Food 2 347.416 2.03 0.135
Type*Predation 2 120.604 0.70 0.496









o ,- Non-reproductive males


C.nnl Cnn3 C.nn9 CnnA FY1 FY3 FY9 FYA
Reproductive females


B u nn 1 nn
D Cnn1 Cnn?!


Cnn C.nnd


Fy1 FY?


Conl Con3 Con2 Con4 Exl Ex3 Ex2 Ex4


Reproductive males


Con1 Con3 Con2 Con4 Exl Ex3 Ex2 Ex4
E[ No fire/No food Food/No fire = Food/No fire E Food/Fire


Figure 2-1. Model averaged estimates of survival of cotton rats in southwestern Georgia between 2005 and 2009 in
response to prescribed fire, supplemental feeding, and predator control treatments. Estimates are given for
non-reproductive (A and B) and reproductive (C and D) male and female rats. Survival is estimated over 10
week intervals. Estimates are given by site: "Ex" sites refer to areas treated with mammalian predator
exclusion while "Con" sites refer to areas where mammalian predators were allowed access. Supplemental
feeding treatments were added to Con and Ex sites 2 and 4 from summer 2007 through 2009. All sites were
burned during the winters of 2005, 2007 and 2009.


&i


a r


ti8


~I












0.8


00 __ T 1 ] oTI i I H '':I'


02 0.2



A Con1 Con3 Con2 Con4 Ex Ex3 Ex2 E4 E B Con1 Con3 Con2 Con4 Ex1 Ex3 Ex2 Ex4
1.0 -
Males: R to R
10-
Females: R to R I
04 0.8 -
008






S0.40
04


02 .- Con1 Con3 Con2 Con4 Exl Ex3 Ex2 Ex4
Con Con3 Con2 Con4 ExPeak/No foodEZ:I Non-peak/No fire/No food Non-peak/Fire/No food
C Con Con3 Con2 Con4 Ex Ex3 Ex2 Ex4 D Peak/Food = Non-peak/No fire/Food EB Non-peak/Fire/Food


Figure 2-2. Model averaged estimates of the rates of transitions to reproductive states for male and female cotton rats
during peak breeding seasons (spring and summer), non-peak seasons during which burning did not occur, and
non-peak seasons during which burning did occur, in southwestern Georiga between 2005 and 2009.
Transitions include movement of individuals from non-reproductive to reproductive states (N to R; A and B), and
reproductive individuals staying in a reproductive state (R to R; C and D). Transitions occurred over 10 week
intervals. Estimates are given by site: "Ex" sites refer to areas treated with mammalian predator exclusion and
"Con" refers to mammalian predator access areas sites. Supplemental food was added to Con and Ex sites 2
and 4 from summer 2007 through 2009. All sites were burned during the winters of 2005, 2007 and 2009.


08










Survival by season


0.6
0.5
0.4
0.3
0.2
0.1
0-


Spring


Summer


Fall Winter (non-burn) Winter (burn)


A

Survival by season and predation treatment


\C


CI


B


Figure 2- 3. Survival estimates ( standard error) of radio collared cotton rats in
southwestern Georgia between 2007 and 2009, generated using Cox
proportional hazard models. Estimates for part A were generated using a
model where survival varied by season (Table 7, model 1). Estimates for part
B were generated using a model where survival varied by season and
predation treatment (Table 7, model 3). Estimates are given by season and
site: "Ex" sites refer to areas treated with mammalian predator exclusion
while "Con" sites refer to areas where mammalian predators were allowed
access. "Winter (burn)" refers to the winter of 2009 during which all sites
underwent a prescribed burn. "Winter (non-burn)" refers to the winter 2008
during which prescribed burns were not carried out.









CHAPTER 3
EFFECTS OF SUPPLEMENTAL FEEDING, MAMMALIAN PREDATOR EXCLUSION,
AND PRESCRIBED FIRE ON COTTON AND OLDFIELD MOUSE POPULATIONS IN A
LONGLEAF PINE ECOSYSTEM

Introduction

Factors which limit populations are commonly of interest to ecologists. Access to

food resources and predation are common causes of limitation. Effects relating to food

and predation are often examined separately, but there is a great deal of evidence from

theoretical and field studies that these factors interact and should be considered

simultaneously (Abrams 1983, McNamara and Houston 1987, Krebs et al. 1995, Hubbs

and Boonstra 1997, 1998). Access to food resources is important for reproduction and

to avoid starvation, but foraging behaviors often increase predation risk. Individuals

should seek a balance that minimizes predation risk while maximizing food intake.

Such trade-offs have obvious implications for survival and abundance and may impact

other vital rates, such as reproduction as well (Lima and Dill 1989). Studies seeking to

understand dynamics of prey species will be benefited by investigating the roles of food

resources and predation individually and in combination on multiple vital rates.

For the target species in this study, cotton and oldfield mice (Peromyscus

gossypinus and P. polionotus respectively), a third factor may interact with food and

predation: prescribed fire. Prescribed fire is a common management tool in longleaf

pine, southern pine, and Florida scrub ecosystems in which both species occur

(Whitaker and Hamilton 1998). Over the short term, burning simultaneously consumes

food resources and reduces cover which increases exposure to predators. Over the

long term, burning maintains open habitat, reduces occurrence of hardwood trees and

shrubs, and improves vegetative growth (Brockway and Lewis 1997). Prescribed fire









benefits these species over the long term, although the exact form of the response may

depend on frequency of fire application (Masters et al. 2002, 2007, Suazo et al. 2009).

Most previous studies examining fire effects on small mammals focused on short term

(less than one year) effects on abundance. These have shown cotton mice to respond

to fire either neutrally or with immediate but temporary population spikes in burned

areas (Shadowen 1963, Hatchell 1964, Layne 1974, Suazo et al. 2009). Oldfield mice

do not appear to have a strong short-term fire response (Arata 1959, Odum et al. 1973,

Suazo et al. 2009). We know of no studies that have attempted to experimentally

determine which fire-related changes (loss of food resources or loss of cover) are

responsible for the observed population-level effects, and few that have examined

effects on a broader range of population parameters such as survival and reproduction.

The objective of this study was to experimentally examine the effects of

supplemental feeding and mammalian predation on cotton and oldfield mouse

populations. We were secondarily interested in determining the roles of these factors in

changes in cover and food availability caused by prescribed burning. This was

accomplished by establishing a large scale factorial experiment with mammalian

predator exclusion and supplemental feeding treatments conducted over four and half

years. Plots were burned three times over the course of the experiment.

Methods

Study Site and Species

This research was conducted at the Joseph W. Jones Ecological Research Center

at Ichauway in Baker County, Georgia. Ichauway is a 12,000 ha property consisting

primarily of longleaf pine (Pinus palustris) and wiregrass (Aristida beyrichiana)

ecosystem. Longleaf pine ecosystems are characterized by a low-density longleaf pine









over-story, a diverse, herbaceous groundcover, and an open, park-like mid-story (Van

Lear et al. 2005). Hardwood tree species occur at limited levels. Frequent, low

intensity fires are key ecological processes. Consequently, frequent application of

prescribed fire is a primary management tool throughout Ichauway; most sites are

burned on a two year rotation (Atkinson et al. 1996).

Cotton and oldfield mice are common across the Southeastern US. The

semi-arboreal cotton mouse prefers bottomland hardwood forests, but the species is a

habitat generalist (Whitaker and Hamilton 1998). Downed woody debris is an important

microhabitat component for this species (McCay 2000). Oldfield mice prefer dry, open

fields with loose soils and beaches. This species is noted for its monogamous breeding

habits (Whitaker and Hamilton 1998).

Field Methods

In 2002, the Jones Center constructed four mammalian predator exclosures, each

paired with a nearby control with similar habitat. Plots range in size from 35.94 to

49.09 ha. Exclosures are surrounded by 1.2 m tall woven wire fences which carry

electrified lines along the top, middle, and bottom to discourage mammals from climbing

over or digging under (the weave is large enough to allow small mammals and snakes

to pass through). Although mammalian predators occasionally enter exclosures, regular

monitoring by track counts and thermal camera surveys indicate significantly fewer

mammalian predators in exclosures than controls (Conner et al. 2010).

Each control and exclosure contained a 12x12 small mammal trapping grid with

15 m spacing between stations. Twenty-four elevated trapping stations were also

interspersed throughout each grid, attached to trees at heights of about 1.5 to 2 m.

Pairs of grids were trapped four times per year (once each season) from January 2005









through June 2007 and eight times per year (twice per season) from July 2007 through

the June 2009 using Sherman live traps (H.B. Sherman Traps, Tallahassee, Florida,

USA). A small amount of a granular insecticide was sprinkled around each trap to

prevent deaths due to fire ants. New captures were marked individually with metal ear

tags. Data recorded for all captures included location, species, sex, weight, age (adult

or juvenile, based on weight), reproductive condition (for males, testes descended or

not, for females, if pregnant and/or lactating), and hind foot measurement.

In June of 2007, two exclosure and two control grids were randomly selected to

receive a supplemental feeding treatment consisting of placing 113 g (4 oz) of

commercial rabbit chow in cans at every other station on the trapping grids. Food was

replaced every other week. Empty cans were also placed in the non-feeding grids.

This treatment continued through August 2009. Images from trail cameras

demonstrated that cotton mice, oldfield mice, cotton rats (Sigmodon hispidus), house

mice (Mus musculus), woodrats (Neotoma floridana), flying squirrels (Glaucomys

volans), and eastern cottontails (Sylvilagus floridanus) regularly used feeding stations.

We found no evidence that cans were defended by individuals of any species.

In February of 2005, 2007, and 2009, all plots were burned according to

Ichauway's burn plan which has these study areas on a two year burn rotation.

Trapping methods followed recommendations of the American Society of Mammalogists

(Gannon et al. 2007) and were approved by the University of Florida Institutional Animal

Care and Use Committee.

Statistical Methods

Data considered for this analysis includes capture-mark-recapture (CMR) data for

cotton and oldfield mice trapped between January 2005 and June 2009 (26 sessions).









Analyses were carried out using the R 2.9.1 (R Development Core Team) package

RMark (Laake and Rexstad 2008) to build models for program MARK (White and

Burnham 1999).

Multistate CMR models were used to estimate and model state specific survival

(S), capture probability (p), and transitions between reproductive states ( L). States

used for S and Y were based on reproductive condition. Males were considered to be

in reproductive condition if testes were descended, females if pregnant and/or lactating.

Therefore, models evaluating effects on S which include a reproductive condition term

estimate and model survival separately for reproductive and non-reproductive

individuals, and models evaluating effects on Y which include a reproductive condition

term estimate and model probabilities of individuals moving between reproductive states

(i.e., rates of non-reproductive individuals entering reproductive states, rates of

reproductive individuals remaining reproductive).

Preliminary analyses considered the potential influence of session, season, and

year on p. Influence of reproductive condition and sex was assessed for S and Y.

Breeding season was also considered for Y. Assessment of effects on p, S and Y was

carried out in a sequential fashion. First, effects on p were considered while modeling S

and Y using the most general models for each described above. Effects on S and Y

were then considered in a similar fashion.

Assessment of goodness-of-fit was carried out using the median c approach in

program MARK (White and Burnham 1999). The median c test indicated a mild

overdispersion (6 = 1.321 for cotton mice and c = 1.339 for oldfield mice). Models in

each parameter's set were compared using Akaike's information criterion corrected for









small sample size (AICc), after quasi-likelihood adjustments (QAICc) made using

c = 1.321 for cotton mice and 1.339 for oldfield mice. Models were considered well

supported if they had a AQAICc of less than two. The best supported model within

each parameter's set was selected a base for modeling that parameter in further

analyses.

These analyses indicated that reproductive condition and sex, modeled in an

additive fashion, were important factors for describing S in cotton mice (S(reproductive

condition+sex), Table 3-3 model 5), while reproductive condition was important for

oldfield mice (S(reproductive condition), Table 3-4, model 5). For both species, capture

probability was best described as fully time varying (p(session)), Table 3-3, model 1

(cotton mice); Table 3-4, model 1 (oldfield mice)).

To model LP, we investigated approach to modeling a breeding season effect on

L. Breeding seasons vary over the geographic range of cotton and oldfield mice (Wolfe

and Linzey 1977, Whitaker and Hamilton 1998). Throughout most of the range,

breeding may occur year round but with peaks at certain times of the year. Because a

literature review did not give a clear indication of when peak breeding seasons for these

species occur in southwestern Georgia, we created model set, based on the literature

(Wolfe and Lindzey 1977, Whitaker and Hamilton 1998, Suazo et al. 2009) and personal

observations, to identify when peak breeding seasons occurred. This analysis

indicated, for cotton mice, breeding peaks in fall and early winter (Table 3-1, model 1),

and, for oldfield mice, peaks in winter and summer (Table 3-2, model 1).

Because of potential confounding effects due to prescribed burning treatments and

breeding seasons occurring as occasion-dependant effects, we then assessed whether









there was evidence to support further division of cotton mouse non-breeding seasons

into whether a burn occurred during these seasons or not. Since the burns occurred

during the oldfield mouse peak breeding season, a similar division was made, but with

respect to peak breeding seasons rather than non-peak seasons. Further division of

breeding seasons based on burning was well supported (Table 3-1, model 7 for cotton

mice, and Table 3-2, model 5 for oldfield mice).

Using these breeding season models, we continued the sequential variable

selection for L as described above for p and S. Reproductive condition, sex, and

breeding season, modeled in an additive fashion, were important for modeling P in

cotton mice (Y(reproductive condition+sex+breeding seasons), Table 3-3, model 10),

while reproductive condition and breeding season, also modeled in an additive fashion,

were important for oldfield mice (Y(reproductive condition+breeding season), Table 3-4,

model 10).

Although the prescribed fires occurred at specific times, fire-caused changes in

cover and food resources may last for weeks or months. To determine the best effect

window for the fire treatments, a set of models considering fire effects on survival over

multiple time intervals was considered. Survival was constrained to be similar between

all trapping periods except those following fires. Post-fire survival was allowed to vary

for several different intervals, from including only the interval during which the fire

occurred (interval length of ten weeks following the 2005 and 2007 fires and of five

weeks following the 2009 fire), to including intervals through the summer season (30

weeks), by which time vegetation is typically recovered. For both species no single

model had overwhelming support over the others (Table 3-5). The top ranked model









was chosen to represent the fire effect on survival in subsequent analysis. For cotton

mice, the highest ranking model indicated a short term fire effect on survival with

declines occurring over a period of ten weeks (Table 3-5, model 1). For oldfield mice,

the best supported model indicated an effect lasting thirty weeks (Table 3-5, model 5).

Exclosure and control sites were initially selected as pairs based on similar

habitats between pairs. As part of a post-hoc examination of non-treatment effects on S

and L, we examined the potential for paired site effects on these parameters. Using

the best models indicated by the analyses described above, we ran a second set of

models considering paired site effects on S and L. This analysis indicated paired site

effects were important in modeling both S and L for cotton and oldfield mice (Table 3-6,

model 1 (cotton mice) and model 5 (oldfield mice)).

Treatment effects were added to the best base model (for cotton mice:

S(reproductive condition+sex+site)p(session) l(breeding season+reproductive

condition+sex+ site); for oldfield mice: S(reproductive condition+site)p(session)

Y(breeding season+reproductive condition+site)) as additive and interactive effects

(two-way only). Due to confounding effects relating to both fire and breeding season

occurring as occasion-dependant effects, only supplemental feeding and predation

treatments were considered with respect to LP, while feeding, predation, and fire effects

were considered with respect to S. Model averaging was employed to generate

parameter estimates for S and L.

Abundance estimates (N) were also generated for both species. Pollock's robust

design (Pollock 1982) was used for cotton mice. Due to computational difficulties, it









was not possible to use the robust design for oldfield mice; the POPAN model was used

instead (Schwarz and Arnason 1996).

Robust design models estimate probabilities for survival (S), capture (p), recapture

(c), emigration (y"), and staying away after emigration (y'). The variable selection

approach used for the robust design was similar to that used for the multistate analysis.

Preliminary investigation considered potential for time and sex effects on p and c.

Paired site effects were considered for N. S was modeled using the best S model from

the multistate analysis (without reproductive condition; S(sex+site+ fire*predation)). y

terms were modeled using a random emigration effect (y"(.)=y'(.)).

Preliminary investigations indicated that cotton mouse capture/recapture models

were best supported when modeled with a capture probability that varied by session

and allowed a constant "trap happy" response. Sex was also important for modeling

capture/recapture probabilities (p(session+sex)c(p+c') where c' is the constant trap

happy response, Table 3-7, model 3). Paired site effects were important for N (Table

3-7, model 1).

Because of the difficulties associated with modeling treatment effects on

abundance directly (White 2002), the best robust design model indicated by the

preliminary analysis described above, S(sex+site+fire*predation)y"(.)=y'(.)

p(session+sex)c(p+c')N(site), was used to generate derived abundance estimates by

site and session, but not to assess treatment effects. Treatment effects on abundance

were evaluated using a repeated measures ANOVA (Schabenberger and Pierce 2002)

implemented using the PROC MIXED procedure in SAS (SAS Institute Inc. 2004). The

variables considered in this ANOVA included food, fire, and predation treatments and









their interactions (two-way interactions only). Paired sites were included as a random

effect. Multiple covariance structures were investigated and the best (variance

components structure, which allows a different variance for each random effect) was

selected based on AICc value (Miller et al. 2004). Treatment effects were considered

significant at the a = 0.05 level.

The POPAN models used for oldfield mice estimate apparent survival (0),

capture probability (p), entry probability (pent), and population size (N). Due to

constraints associated with the POPAN model, the data set was divided by paired sites.

For each of these paired sites, 0 was modeled using the best survival model from the

multistate analysis (minus the reproductive condition term; Q(predation)). The

sequential variable selection process for N, p, and pent followed as described

previously. A site effect was considered with respect to N. Effects of year, burn year,

and season were considered with respect to p and pent. To pick the best common

model for all site pairs, each candidate model in each parameter's candidate model set

was ranked by AAICc score. Ranks were summed across sites for each parameter and

the model with the lowest score was selected as the best. This investigation

determined that p and pent were best modeled across sites using an additive effect

between year and season while N was best modeled varying by site (Table 3-8).

The resulting model Q)(predation)p(year+season)pent(year+season)N(site) was

run for all site pairs to generate derived abundance estimates for each site by trapping

session. Treatment effects on abundance were investigated using a repeated

measures ANOVA in PROC MIXED in SAS as described above for cotton mice.









Results


Cotton Mice

A total of 2108 individual cotton mice (8428 total captures) was trapped over 26

trapping sessions in eight trapping plots. The best supported multistate model

suggested an interactive effect of predation and fire treatments on survival with an

interactive effect of feeding and predation on L (Table 3-9, model 1). Although this

model was the best supported (the second ranked model has a AQAICc > 2), it did not

carry a great deal of weight (0.393). However, it was clear from the top ranked models

that the feeding interacted with predation to affect Y. This interaction appeared in the

top ten models, and models with this interaction held 91.1% of the weight of the overall

model set. Support for treatment effects on survival was less clear. The second best

supported model (model 2, Table 3-9) included no treatment effect on survival,

indicating poor support for treatment effects other than the interactive predation*fire

effect on survival.

The lack of substantial support for any particular model indicates model selection

uncertainty; therefore, model averaging was employed for parameter estimation.

Overall model averaged survival estimates showed that males had lower survival than

females and that reproductive individuals had higher survival than non-reproductive

individuals (Figure 3-1). Model averaged estimates indicated that, in predator access

grids, burning had essentially no effect on survival. In predator exclosures, however,

survival increased dramatically following fires (Figure 3-1). During non-fire periods,

survival was slightly greater in predator access grids than in the exclosures, but this

trend was reversed following burns (Figure 3-1).









The addition of food had minimal impact on survival regardless of whether

predators had access or not, or whether an area had been recently burned or not

(Figure 3-1).

Model averaged parameter estimates for Y showed that a greater proportion of

males made transitions to reproductive states than females in all seasons (Figure 3-2).

Additionally, most reproductive individuals that achieved a reproductive state stayed in a

reproductive state; this trend was slightly greater for males than females (Figure 3-2).

Initial investigation indicated a strong fire effect on transitions between

reproductive states: models that included three classes of breeding seasons (peak

breeding in fall and early winter, non-peak breeding in springs, summers, and late

winters without burns, and a second non-peak in springs, summers, and late winters of

burn years; hereafter, peak, non-peak/non-burn, and non-peak/burn respectively) had

greater support than models with only two breeding seasons (peak and non-peak with

no distinguishing between burn and non-burn years). Two-season models had no

support (weight = 0.0) compared to three-season models (Table 3-1, model 7).

Model averaged parameter estimates indicated that transitions to reproductive

states were at their highest during peak breeding seasons and that there was a small

drop in transitions to reproductive states during non-peak/non-fire seasons (Figure 3-2).

Transitions to reproductive states dropped considerably more during non-breeding/fire

seasons (Figure 3-2). Predator exclusion and feeding alone caused small decreases in

reproductive transitions and the combination of these treatments was associated with

an increase in transitions to reproductive states (Figure 3-2).









The repeated measures ANOVA examining treatment effects on abundance

indicated a significant effect of feeding on abundance (P < 0.001). Examination of least

square means indicated that feeding plots contained 1.8x the number of cotton mice as

unfed plots. No other treatments or their interactions significantly affected abundance

(P > 0.05).

Oldfield Mice

A total of 1203 individual oldfield mice (4828 total captures) was trapped over 26

trapping sessions in eight trapping plots. There was no clear top multistate model for

oldfield mice as there were six models with a AQAICc < 2 and none of these carried

much weight (Table 3-10). However, it was clear from the top ranked models that

predation was an important factor affecting survival. Predation effects appeared in the

top twenty models and these models collectively held a weight of 82.8%. Models

including feeding also had decent support in the model set, holding 67% of the weight of

the overall set. The model set showed limited support for treatment effects on Y. The

lack of clear support for any particular model indicated model selection uncertainty;

therefore, model averaging was employed for parameter estimation.

Model averaged survival estimates indicated that non-reproductive individuals had

lower survival than reproductive individuals (Figure 3-3). Model averaged survival

estimates also showed increased survival in predator exclusion plots compared to

predator access plots. This was true in both pre- and post-fire periods. Following

prescribed fires, survival decreased slightly in predator access grids and increased (by

a slightly greater magnitude) in predator exclosure treatments (Figure 3-3). Addition of

food was associated with declines in survival in both predator access and exclosure

grids, and in both pre- and post-fire periods. The magnitude of the decline was greater









in exclosures than in controls, but the magnitude of the change was not great in either

case.

Initial investigation found strong fire effects on transitions between reproductive

states: models that included three classes of breeding seasons bimodall peak breeding

in winters and summers, distinguishing between burn years and non-burn years, and

non-peak breeding in falls and springs) had better support than models that contained

only two breeding seasons (peak and non peak, with no distinguishing between burn

and non-burn peak seasons). Two season models had no support when compared to

three season models (weight = 0.0, Table 3-2, model 5).

Model averaged parameter estimates for Y indicate that a greater proportion of

reproductive individuals that achieved reproductive states stayed in reproductive states

(Figure 3-4). Transitions to reproductive states were the greatest during peak breeding

seasons of non-burn years. Transitions into breeding states dropped during non-peak

seasons. However, during winters and summers of burn years, transitions to breeding

states dropped dramatically such that transitions during these seasons were below even

that of non-breeding seasons, indicating a strong fire effect on reproduction in oldfield

mice (Figure 3-4).

Predator exclusion was associated with smaller proportions of individuals entering

breeding states, whether food was present or not, although this difference was minimal.

Supplemental feeding was associated with a larger increase in transitions to

reproductive states in both predator access and exclusion areas (Figure 3-4).

The repeated measures ANOVA examining treatment effects on abundance

indicated significant effects of predation and feeding treatments and the interaction of









these treatments on abundance (P = 0.001, P < 0.001, and P = 0.001 respectively).

Examination of least square means showed that feeding increased abundances by 2.7x,

predator exclusion increased abundances by 2.7x and the application of both

treatments simultaneously increased abundances by 7.6x.

Discussion

Although cotton and oldfield mice are closely related species which occur in many

of the same habitats, they were affected in different ways by the application of

mammalian predator exclusion, supplemental feeding, and prescribed fire treatments.

However, both species showed a surprising trend of increased survival among

reproductive individuals compared to non-reproductive individuals. We hypothesize that

this may be because non-reproductive adults are likely to be younger individuals.

Although there is evidence that reproduction exacts a survival cost among small

mammals (Koivula et al. 2003), it may be that among these mice, the cost of being

young is greater. Juveniles and young adults tend to be transient while seeking to

establish home ranges (Bigler and Jenkins 1975) and dispersal behavior is associated

with reduced survival (Van Vuren and Armitage 1994). However, this behavior may be

adaptive in general if dispersed individuals improve reproduction by doing so (Van

Vuren and Armitage 1994).

Among cotton mice, initial analyses identified sex as an important factor with

respect to both survival and reproductive transitions, but among oldfield mice sex was

irrelevant. It is possible that this difference occurs because oldfield mice are

monogamous and form long-term pair bonds while cotton mice are promiscuous (Blair

1951).









Treatment Effects on Cotton Mice

Cotton mice showed different treatment effects on different vital rates, some in

interactive and unexpected ways. It is difficult to tie these effects into a clear picture of

how predation, food resources, and fire affect cotton mouse populations overall, but

interpretation is aided by consideration of ecological theory concerning adaptive prey

behavior in response to predation risk and food availability. This theory has been

largely developed by Abrams (1983, 1984, 1991, 1992a, 1992b, 1993) and is based on

the observation that while access to sufficient food is required for reproduction and to

avoid starvation, foraging increases predation risk. Individuals must make trade-offs

between foraging and predation risk in such a way as to maximize food intake (and by

extension, reproduction) while minimizing mortality. A large body of theoretical and

experimental evidence suggests such trade-offs are common (reviewed in Lima and Dill

1989) and may be stronger than direct consumptive effects of predation (Pressier et al.

2005). Although a simple concept at the core, the effects of these adaptive behaviors

may be complex and counterintuitive (Abrams 1991, 1992a, 1992b, 1993). This is

especially likely in systems with complex food webs and competitive interactions

between multiple predator and prey species. Effects of trade-offs may also depend on

a species' life history and may change with breeding/non-breeding seasons (Abrams

1991).

The only strong treatment effect on cotton mouse survival was an increase in

survival with the combination of predator exclusion and fire. There was no strong fire

effect on survival in predator controls, and survival was similar between controls and

exclosures during non-fire periods. This suggests fire conveys a benefit that is not

realized when predators have access to the burned area. This in turn suggests an









awareness of increased predation risk with loss of cover, a behavioral response of

cotton mice in the predator controls, and a choice to balance predation risk against

taking advantage of this benefit. Differences in predation and associated trade-offs may

contribute to the mixed short-term fire responses (neutral or beneficial) observed with

cotton mice in previous studies (Shadowen 1963, Hatchell 1964, Layne 1974, Suazo et

al. 2009).

The general lack of a supplemental feeding effect on survival suggests that food

resources are not important in this fire response, unless the food supplementation was

insufficient or access to food cans was considered risky. Other mechanisms that may

have caused to the combined benefit of fire and predator exclusion remain unclear. It is

possible that change in abundances of other species in the burned area may have

contributed to this response. For example, cotton rats, usually among the most

abundant species in the study areas, declined to near 0 following burns.

The combination of a lack of a fire effect on abundance and a decline in post-fire

breeding supports a behavioral response to increased predation risk associated with

post-fire conditions. Adaptive anti-predatory behavior may mitigate negative fire effects

on survival and abundance, but a trade-off appears to occur with a decline in

reproduction. Because fires occurred during the non-breeding season, it is reasonable

that cotton mice would make a trade off favoring survival at the cost of reproduction.

Population growth rates of species such as rodents which mature rapidly and reproduce

often are most sensitive to changes in reproduction (Heppell et al. 2000, Oli and

Dobson 2003), and in most cases ought to favor strategies that maximize reproduction

rather than survival. However, during the non-breeding season these species should









maximize fitness by behaving in a similar manner as semelparous species (Abrams

1991). This entails minimizing predation risk by reducing foraging. That feeding had no

apparent effect on post-fire survival and reproductive transitions may be explained by

the observation that for semelparous species, or iteroparous species in the

non-breeding season, increases in food resources should, somewhat counterintuitively,

be associated with a decrease in foraging effort to maximize survival and overall fitness

(Abrams 1991).

Increased predation should cause prey species to make trade-offs between growth

rate (positively associated with food intake) and predation risk (negatively associated

with food intake) when prey species (like cotton mice) achieve reproductive maturity at a

specific size rather than by reaching a specific age (Abrams and Rowe 1996). This

trade-off should cause an increase in predation risk to be associated with an increase in

the age at maturity due to decreased growth rates (Abrams and Rowe 1996). This

relationship may have contributed to the decline in reproduction following fires; if

juvenile development was delayed, a decline in reproductive transitions should be

observed. However, increases in food resources may counteract the direct effect of

predation on growth rate (Abrams and Rowe 1996). This may explain why breeding

transitions were positively affected by the interaction of predator exclusion and food

supplementation.

Food addition was associated with a nearly two-fold increase in abundance. It is

not surprising that food addition caused increases in abundance, but given the

relationship between food resources and the ability to achieve reproductive status in

small mammals (Cameron and Eshelman 1996), it is interesting that feeding was not









also associated with increases in transitions to reproductive states (except in predator

exclusion plots). This suggests that the increased abundances were due to either

increases in juvenile survival, in the number of young produced per reproductive event

or immigration to feeding plots. Such effects have been observed to occur in response

to supplemental feeding in previous studies with small mammals (Boutin 1990, Hubbs

and Boonstra 1997); unfortunately, we were unable to address these factors in the

current study.

Treatment Effects on Oldfield Mice

The strongest treatment effect on oldfield mice was the predator exclusion

treatment; this was associated with increased survival and abundance. Feeding and

the interaction of feeding and predator exclusion were also associated with increased

abundances. On the surface, these results seem more intuitive than the cotton mouse

results and could be interpreted to suggest oldfield mice are more influenced by direct

consumptive effects of predation. Pressier and Bolnick (2008) observed in a review of

non-consumptive effects of predation that non-consumptive effects seemed to dominate

some predator-prey relationships while appearing to be weak or non-existent in others.

However, Peckarsky et al. (2008) note that it is impossible to conclude that changes in

prey survival and abundance are due solely to direct effects of predation as stresses

and trade-offs associated with predation risk can cause declines in survival and

abundance even when prey are not actually at risk (as by experimental manipulation of

a predator's ability to kill). Such indirect effects may be overlooked when they take the

predicted form and direction of direct consumptive effects, and they may at times cause

effects as strong, if not stronger, than direct consumptive effects (Pressier et al. 2005).

Behavioral data can help to distinguish between direct and indirect effects, and indeed









may be necessary to do so (Abrams 1995), but our study design did not include such

observations.

Similar to the results observed with cotton mice, feeding was associated with

increases in abundance but with only minimal increases in reproductive transitions,

indicating the feeding effect on abundance may be due to increases in immigration,

juvenile survival, or the number of young produced per reproductive event.

Predation had a minimal effect on reproductive transitions, but fire caused

declines in transitions to reproductive states. Given the importance of predation on

other population parameters, it might be expected that there would be an interaction of

predation and fire treatments since burning removes cover and should increase risk of

predation. It is possible that increased predation due to loss of cover was not a problem

following fires because this species already prefers open areas (Whitaker and Hamilton

1998). This of course fails to explain why burning caused declines in reproductive

transitions, and why addition of food was insufficient to prevent such declines. Although

the decline in reproductive transitions following fires is similar to that of cotton mice, an

important difference exists between the responses in these species. Cotton mice were

in a non-peak breeding season when the burns occurred while oldfield mice were in a

peak season. In this situation, oldfield mice should be expected to maximize

reproduction during this season, even at the expense of survival (Abrams 1991). That

they were unable to do so (assuming reproductive output is associated with transitioning

to a reproductive state) suggests that either the theory is incorrect or that oldfield mice

were limited in some way despite food addition. Given necessity of sufficient food

quality and quantity for small mammals to achieve reproductive status (Cameron and









Eshelman 1996), and the oldfield mouse's general dietary preference for insects and

seeds over herbaceous material, it is possible that the provided food was insufficient to

allow these mice to maintain normal breeding following a winter burn. Although

supplemental food was available following fires, other preferred food sources may have

become limiting. For example, Odum et al. (1973) observed declines in arthropod

abundances following a winter burn in Georgia for several months.

Conclusions

Although cotton and oldfield mice are closely related species that occur in similar

habitats, feeding, fire, and predation treatments affected these species differently.

Cotton mice appear to make trade-offs with respect to predation risk. These appear to

be especially important following fire events, implying cotton mice are, over the short

term, negatively affected by the loss of cover associated with burning. An exception

seems to occur when mammalian predators are excluded. Oldfield mice also

experience significant effects of predation but it is less clear if these effects are related

directly or indirectly to consumption itself, or merely the risk of being consumed. Fire

effects are less apparent for oldfield mice although reproduction in oldfield mice is

negatively affected by fires.

The retroductive conclusions relating to behavioral responses are in many ways

speculative as we lack behavioral data that would help to confirm these effects. Studies

incorporating behavioral components such as home range size, microhabitat use, giving

up densities for foraging animals, etc., could better address these issues.









Table 3-1. Model comparison table for multistate capture-mark-recapture analysis
assessing occurrence of peak breeding seasons of cotton mice in general
(Model set 1) and with respect to winter prescribed burns (Model set 2) in
Southwestern Georgia between 2005 and 2009. All models had survival set
as S(reproductive condition+sex) and capture probability set as p(session).
Table includes number of parameters (K), model weights (relative likelihood
of models in the set), and difference in Akaike's information criterion corrected
for small sample size after quasilikelihood adjustment (AQAICc).
Quasilikelihood adjustments were made using an estimated c of 1.339.
Model Model K AQAICc Model
no. weight
Model 1 Y(fall and early winter peaks) 30 0.00 1.00
set 1* 2 Y(winter, spring and fall peaks) 30 24.82 0.00
3 Y(fall peaks) 30 32.17 0.00
4 Y(winter peaks) 30 64.67 0.00
5 Y(fall and winter peaks) 30 72.68 0.00
6 Y(constant) 29 75.55 0.00

Model 7 Y(fall and early winter peaks; late winter, 31 0.00 1.00
set 2** spring, summer non-peaks divided by
burn years and non-burn years )
8 Y(fall and early winter peaks; late winter, 30 18.70 0.00
spring, summer non-peaks)
* Model set 1 considers only two classes of breeding season: peak (listed) and
non-peak (all seasons not listed for a given model).
** Model set 2 compares the best model from set 1 (model number 1) with a similar
model that considers three classes of breeding seasons by breaking down non-peak
seasons into those that occurred during burn years (late winters, springs, and summers
of 2005, 2007, and 2009; burns occurred in mid-winters of these years) and those that
occurred during non-burn years (late winters, springs, and summers of 2006 and 2008).









Table 3-2. Model comparison table for multistate capture-mark-recapture analysis
assessing occurrence of peak breeding seasons of oldfied mice in general
(Model set 1) and with respect to winter prescribed burns (Model set 2) in
Southwestern Georgia between 2005 and 2009. All models had survival set
as S(reproductive condition) and capture probability set as p(session). See
Table 3-1 for column definitions.
Model Model K AQAICc Model
no. weight
Model 1 Y(winter and summer winter peaks) 29 0.00 0.59
set 1* 2 Y(spring peaks) 29 2.56 0.16
3 Y(no peaks) 28 2.73 0.15
4 Y(fall peaks) 29 3.49 0.10

Model 5 Y(winter and summer peaks with peak 30 0.00 1.00
set 2** seasons divided by burn years and non-
burn years )
6 Y(winter and summer peaks) 29 41.07 0.00
* Model set 1 considers only two classes of breeding season: peak (listed) and
non-peak (all seasons not listed for a given model).
** Model set 2 compares the best model from set 1 (model number 1) with a similar
model that considers three classes of breeding seasons by breaking down the peak
seasons into those that occurred during burn years (winters and summers of 2005,
2007, and 2009; burns occurred in winters of these years) and those that occurred
during non-burn years (winters and summers of 2006 and 2008).









Table 3-3. Model comparison table for multistate capture-mark-recapture analysis
assessing sex, reproductive condition, breeding season, and time effects on
capture probability (p), survival (S), and transitions rates between
reproductive states (P) in cotton mice in Southwestern Georgia between
2005 and 2009. See Table 3-1 for column definitions.
Model K AQAICc Model
no. weight
Effects on capture probability (p) *
1 p(session) 34 0.00 0.71
2 p(season) 14 1.84 0.29
3 p(constant) 10 12.84 0.00
4 p(years) 14 13.65 0.00

Effects on survival (S)**
5 S(reproductive condition*sex) 34 0.00 0.48
6 S(reproductive condition+sex) 33 0.47 0.38
7 S(sex) 32 3.83 0.07
8 S(reproductive condition) 32 4.14 0.06
9 S(constant) 31 6.44 0.02

Effects on rates of reproductive transitions***
10 Y(breeding season+reproductive condition+sex) 33 0.00 0.61
11 Y(breeding season+reproductive condition) 32 0.93 0.39
12 Y(breeding season+sex) 32 245.32 0.00
13 Y(breeding season) 31 254.10 0.00
14 Y(constant) 29 348.35 0.00
*Additional parameters modeled as S(reproductive condition*sex) l(breeding season+
reproductive condition+sex).
**Additional parameters modeled as p(session) l(breeding season+reproductive
condition+sex).
***Additional parameters modeled as S(reproductive condtion*sex)p(session).









Table 3-4. Model comparison table for multistate capture-mark-recapture analysis
assessing sex, reproductive condition, breeding season, and time effects on
capture probability (p), survival (S), and transitions rates between
reproductive states (P) in oldfield mice in Southwestern Georgia between
2005 and 2009. See Table 3-1 for column definitions.
Model Model K AQAICc Model
no. weight
Effects on capture probability (p) *
1 p(session) 34 0.00 0.84
2 p(season) 14 3.31 0.16
3 p(years) 14 22.05 0.00
4 p(constant) 10 23.60 0.00

Effects on survival (S)**
5 S(reproductive condition) 32 0.00 0.43
6 S(reproductive condition+sex) 33 0.18 0.40
7 S(reproductive condition*sex) 34 1.98 0.16
8 S(sex) 32 8.32 0.01
9 S(constant) 31 8.55 0.01

Effects on rates of reproductive transitions (u)***
10 Y(breeding season+reproductive condition) 33 0.00 0.54
11 Y(breeding season+reproductive condition+sex) 34 0.36 0.46
12 Y(breeding season+sex) 33 51.21 0.00
13 Y(breeding season) 32 51.34 0.00
14 Y(constant) 30 93.94 0.00
*Additional parameters modeled as S(reproductive condition*sex) l(breeding season+
reproductive condition+sex).
**Additional parameters modeled as p(session) l(breeding season+reproductive
condition+sex).
***Additional parameters modeled as S(reproductive condtion*sex)p(session).









Table 3-5. Model comparison table for multistate capture-mark-recapture analysis
assessing term of effect of winter prescribed burns on survival of cotton and
oldfield mice in Southwestern Georgia between 2005 and 2009. See Table
3-1 for column definitions. All models had additional parameters of capture
probability (p) modeled as p(session) and rate of transition between
reproductive states (P) modeled as Y(reproductive condition+sex+breeding
season) (for cotton mice) or Y(reproductive condition+breeding season) (for
oldfield mice).
Species Model Model K AQAICc Model
no. weight
Cotton 1 S(fire effect over 10 weeks) 32 0.00 0.32
mice 2 S(fire effect over 20 weeks) 32 0.35 0.27
3 S(fire effect over 5 (2009) to 10 32 0.81 0.21
weeks (2005 and 2007))*
4 S(fire effect over 30 weeks) 32 0.85 0.21

Oldfield 5 S(fire effect over 20 weeks) 32 0.00 0.28
mice 6 S(fire effect over 30 weeks) 32 0.25 0.24
7 S(fire effect over 10 weeks) 32 0.26 0.24
8 S(fire effect over 5 (2009) to 10 32 0.31 0.24
weeks (2005 and 2007))*
* Models 3 and 8 show a range of intervals because the interval between trapping
sessions changed from 10 weeks (2005 and 2007) to 5 weeks (2009) before the 2009
prescribed burn.









Table 3-6. Model comparison table for multistate capture-mark-recapture analysis assessing potential for site effects on
survival (S) and rate of transitions to reproductive states (P) of cotton and oldfield mice in Southwestern
Georgia between 2005 and 2009. See Table 3-1 for column definitions. All models had capture probability (p)
modeled as p(session).
Model Species Model K AQAICc Model
no. weight
1 Cotton S(reproductive condition+sex+site*) 39 0.00 0.68
Mice Y(breeding season+reproductive condition+sex+site)
2 S(reproductive condition+sex+site) 36 1.49 0.32
Y(breeding season+reproductive condition+ sex)
3 S(reproductive condition+sex) 36 19.99 0.00
Y(breeding season+reproductive condition+sex+site)
4 S(reproductive condition+sex) 33 21.63 0.00
Y(breeding season+reproductive condition+sex)

5 Oldfield S(reproductive condition+site) 38 0.00 0.68
Mice Y(breeding season+reproductive condition+site)
6 S(reproductive condition) (breeding season+reproductive condition+site) 35 1.53 0.32
7 S(reproductive condition+site) (breeding season+reproductive condition) 35 16.43 0.00
8 S(reproductive condition) (breeding season+reproductive condition) 32 17.80 0.00
*Site effects are paired site effects. This effect pairs each site treated with mammalian predator exclusion with a predator
access site with similar habitat.









Table 3-7. Model comparison table for robust design capture-mark-recapture analysis
assessing demographic and time effects on abundance (N), capture
probability (p), and recapture probability (c) in cotton mice in Southwestern
Georgia between 2005 and 2009. See Table 3-1 for column definitions.
Survival (S) was modeled as S(site+sex+fire*predation) for all models.
Emigration terms (y" and y') were modeled with a random emigration effect
for all models: y"(.)=y'(.).
Model Model K AQAICc Model
no. weight
Effects on abundance (N)*
1 N(site) 41 0.00 1.00
2 N(.) 38 21.69 0.00


Effects on capture (p) and recapture (c)**
3 p(session+sex)c(p+c')t 41 0.00 1.0(
4 p(session)c(p+c')tt 40 29.03 0.0(
5 p(sex)c(p+c') 16 229.00 0.0(
6 p(.)c(p+c')4 15 246.21 0.0(
7 p(.)c(p) 14 362.08 0.0(
* Additional parameters modeled as p(session+sex)c(p+c').
**Additional parameter modeled as N(site).
t Indicates capture probability varies by session, with an additive effect of sex, and with
a constant trap happy response recapture response (c').
tt Indicates capture probability varies by session with a constant trap happy response
recapture response (c').
Indicates capture probability varies by sex with a constant trap happy response
recapture response (c').
$: Indicates a constant capture probability with a constant trap-happy recapture
response (c').
Indicates constant capture and recapture rates (shared).


0
0
0
0
0
J









Table 3-8. Model comparison table for POPAN capture-mark-recapture analysis
assessing site and time effects on abundance (N), capture probability (p), and
entry probability (pent) in oldfield mice in Southwestern Georgia between
2005 and 2009. Effects are given separately for four paired mammalian
predator exclusion and control sites. See Table 3-1 for column definitions.
Survival (0) was modeled as S(predation) for all models. Bolded models
indicate those selected as the best common model for all sites (based on
ranking each model by AAICc score and summing ranks across sites).
Model Model K AAICc Model
no. weight


Effects on abundance (N)*
Control/Exclosure 1
N(site)
N(constant)
Control/Exclosure 2
N(site)
N(constant)
Control/Exclosure 3
N(site)
N(constant)
Control/Exclosure 4
N(constant)
N(site)


0.00
4.08


0.88
0.12


0.00 1.00
14.42 0.00


0.00
1.71


0.70
0.30


0.00 0.71
1.81 0.29


Effects on capture probability (p)**
Control/Exclosure 1
p(year+season)
p(burn year+season)
p(season)
p(burn year)
p(constant)
p(year)
Control/Exclosure 2
p(year+season)
p(year)
p(season)
p(burn year+season)
p(constant)
p(burn year)
Control/Exclosure 3
p(year+season)
p(year)
p(burn year+season)
p(constant)
p(burn year)
p(season)


0.00
30.74
43.33
59.00
60.36
60.39

0.00
9.29
116.85
118.94
132.72
134.30

0.00
10.00
11.73
11.93
14.11
18.09


1.00
0.00
0.00
0.00
0.00
0.00

0.99
0.01
0.00
0.00
0.00
0.00

0.99
0.01
0.00
0.00
0.00
0.00









Table 3-8. Continued
Model Model K AAICc Model
no. weight
Control/Exclosure 4
27 p(year) 15 0.00 0.37
28 p(burn year) 12 0.13 0.35
29 p(constant) 11 1.01 0.22
30 p(year+season) 19 4.63 0.04
31 p(season) 15 6.96 0.01
32 p(burn year+season) 16 7.27 0.01

Effects on entry probability (pent)
Control/Exclosure 1
33 pent(year+season) 19 0.00 0.93
34 pent(year) 15 6.52 0.04
35 pent(burn year+season) 16 6.86 0.03
36 pent(season) 15 14.34 0.00
37 pent(burn year) 12 20.30 0.00
38 pent(constant) 11 28.42 0.00
Control/Exclosure 2
39 pent(year+season) 19 0.00 0.99
40 pent(year) 15 9.93 0.01
41 pent(burn year+season) 16 130.58 0.00
42 pent(burn year) 12 136.50 0.00
43 pent(season) 15 161.11 0.00
44 pent(constant) 11 182.66 0.00
Control/Exclosure 3
45 pent(season) 15 0.00 0.60
46 pent(burn year+season) 16 1.06 0.35
47 pent(constant) 11 6.45 0.02
48 pent(burn year) 12 7.57 0.01
49 pent(year+season) 19 9.49 0.01
50 pent(year) 15 15.81 0.00
Control/Exclosure 4
51 pent(year+season) 19 0.00 1.00
52 pent(burn year+season) 16 21.86 0.00
53 pent(season) 15 30.39 0.00
54 pent(year) 15 106.40 0.00
55 pent(constant) 11 138.22 0.00
56 pent(burn year) 12 139.90 0.00
* Additional parameters modeled as p(burn year+season) and pent(burn year+season).
** Additional parameters modeled as pent(burn year+season) and N(site).
*** Additional parameters modeled as N(site) and p(burn year+season).









Table 3-9. Model comparison table for multistate capture-mark-recapture analysis examining the effect of predation,
feeding, and fire treatments on survival (S) and transition probabilities (P, between reproductive and
non-reproductive states) of cotton mice in southwestern Georgia, 2005-2009. All models had capture
probability set at p(session). See Table 3-1 for column definitions. Bolded text indicates treatment effects (all
other effects are similar between models throughout the set). Table values were adjusted for an estimated
c-hat of 1.321. Only models with a model weight > 0.03 are shown here (the top 8 models of 55 in the overall
set).
Model Model K AQAICc Model
no. weight
1 S(reproductive condition+sex+site+predation*fire) 45 0.00 0.39
Y(breeding season+ reproductive condition+sex+site+food*predation)
2 S(reproductive condition +sex+site) 42 2.08 0.14
Y(breeding season+ reproductive condition +sex+site+food*predation)
3 S(reproductive condition +sex+site+fire) 43 2.73 0.10
Y(breeding season+ reproductive condition+sex+site+food*predation)
4 S(reproductive condition +sex+site+predation) 43 3.81 0.06
Y(breeding season+ reproductive condition+sex+site+food*predation)
5 S(reproductive condition +sex+site+food) 43 4.07 0.05
Y(breeding season+ reproductive condition+sex+site+food*predation)
6 S(reproductive condition +sex+site+food*fire) 45 4.22 0.05
Y(breeding season+ reproductive condition+sex+site+food*predation)
7 S(reproductive condition+sex+site+predation+fire) 44 4.39 0.04
Y(breeding season+ reproductive condition+sex+site+food*predation)
8 S(reproductive condition +sex+site+food+fire) 44 4.66 0.04
YU(breeding season+ reproductive condition+sex+site+food*predation)









Table 3-10. Model comparison table for multistate capture-mark-recapture analysis
examining the effect of predation, feeding, and fire treatments on survival (S)
and transition probabilities (P, between reproductive and non-reproductive
states) of oldfield mice in southwestern Georgia, between 2005 and 2009. All
models had capture probability set at p(session). See Table 3-1 for column
definitions. Bolded text indicates treatment effects (all other effects are
similar between models throughout the set). Only models with a model


weight > 0.03 are shown here (the top 9 models of 55 in
Model K


1 S(reproductive condition+site+food*predation)
Y(breeding season+reproductive condition +site)
2 S(reproductive condition+site+food+predation)
Y(breeding season+reproductive condition+site)
3 S(reproductive condition+site+food*predation)
Y(breeding season+reproductive
condition+site+predation)
4 S(reproductive condition+site+food*predation)
Y(breeding season+reproductive
condition+site+food)
5 S(reproductive condition+site+predation*fire)
Y(breeding season+reproductive condition+site)
6 S(reproductive condition+site+predation)
Y(breeding season+reproductive condition+site)
7 S(reproductive condition +site+food+predation)
Y(breeding season+reproductive
condition+site+predation)
8 S(reproductive condition+site+food+predation)
Y(breeding season+reproductive
condition+site+predation)
9 S(reproductive
condition+site+food+predation+fire)
U(breedinq season+reproductive condition +site)


the overall set).
AQAICc Model
weight
0.00 0.14


39 0.81

41 1.29


41 1.69


40 1.90

38 1.96

40 2.10


40 2.50


40 2.66


0.09

0.07


0.06


0.05

0.05

0.05


0.04


0.04


Model
no.


40














06
Non-reproductive females
0.5-



0.4-

04

S0.3



A Con1 Con3 Con2 Con4 Exl Ex3 Ex2 Ex4 B Con1 Con3 Con2 Con4 Ex1 Ex3 Ex2 Ex4
0 65
Reproductive females Reproductive males
0 60 .6-



0.5-
050 -

0 45- 04.






C Con1 Con3 Con2 Con4 Ex Ex3 Ex2 Ex4 D No firFire/No Food Io FFood/No Fire Food/Fire



Figure 3-1. Model averaged estimates of survival of cotton mice in southwestern Georgia between 2005 and 2009 in
response to prescribed fire, supplemental feeding, and predator control treatments. Estimates are given for
non-reproductive (A and B) and reproductive (C and D) male and female mice. Survival is estimated over 10
week intervals. Estimates are given by site: "Ex" sites refer to areas treated with mammalian predator
exclusion and "Con" refers to mammalian predator access areas sites. Supplemental food was added to Con
and Ex sites 2 and 4 from summer 2007 through 2009. All sites were burned during the winters of 2005, 2007
and 2009.














0 7-
Females N to R
0.6
0 6
0.5
05





02


A Con1 Con3 Con2 Con4 Exl Ex3 Ex Ex4 Con1 Con3 Con2 Con4 Ex Ex3 Ex2 Ex4
1.0 -
Males: R to R
1.0
Females: R to R
0.9 -




0.7 0.7 -




0.6 0.6
Con1 Con3 Con2 Con4 Exl Ex3 Ex2 Ex4
Con1 Con3 Con2 Con4 Exl Ex3 Ex2 Ex4 Peak/No food Non-peak/No fire/No foo I Non-peak/Fire/No food
C 4 Peak/Food = Non-peak/No fire/Food E Non-peak/Fire/Food



Figure 3-2. Model averaged estimates of breeding transitions for male and female cotton mice in southwestern Georgia
between 2005 and 2009 during peak breeding seasons (fall and early winter), non-peak seasons during which
burning did not occur and non-peak seasons during which burning did occur. Transitions include transition of
non-reproductive individuals to reproductive states (N to R, A and B) and reproductive individuals staying in a
reproductive state (R to R; C and D). Transitions occurred over 10 week intervals. Estimates are given by site:
"Ex" refers to areas treated with mammalian predator exclusion; "Con" refers to mammalian predator access
sites. Supplemental food was added to Con and Ex sites 2 and 4 from summer 2007 through 2009. All sites
were burned during the winters of 2005, 2007 and 2009.












Non-reproductive


0.5



0.4 -

CE

0.3



0.2


Con1 Con3 Con2 Con4 Exl Ex3 Ex2 Ex4
A
0.7
Reproductive


0.6



0.5




0.4



0.3
Con1 Con3 Con2 Con4 Exl Ex3 Ex2 Ex4
B No fire/No food EZ3 Fire/No food = Food/No fire E Food/Fire



Figure 3-3. Model averaged survival estimates in southwestern Georgia between 2005
and 2009 in response to prescribed fire, supplemental feeding, and predator
control treatments. Estimates are given for non-reproductive (A) and
reproductive (B) individuals. Survival is estimated over 10 week intervals.
Estimates are given by site: "Ex" sites refer to areas treated with mammalian
predator exclusion while "Con" sites refer to areas where mammalian
predators were allowed access. Supplemental feeding treatments were
added to Con and Ex sites 2 and 4 from summer 2007 through 2009. All sites
were burned during the winters of 2005, 2007 and 2009.












0.8


0.6




S 0.4-




0.2


Con1 Con3 Con2 Con4 Exl Ex3 Ex2 Ex4
A


09 R to R

0.8

07 -

06
06
0.5

0.4 -

0.3
Con1 Con3 Con2 Con4 Exl Ex3 Ex2 Ex4
m Peak/No fire/No food I\> Peak/Fire/No food I I Non-peak/No food
i Peak/No fire/Food I Peak/fire/Food I Non-peak/Food
B

Figure 3-4. Model averaged estimates of breeding transitions for oldfield mice in
southwestern Georgia between 2005 and 2009 during peak breeding seasons
(winter and summer) in burn and non-burn years and non-peak breeding
seasons. Transitions include non-reproductive individuals transitioning to
reproductive states (N to R; A) and reproductive individuals staying in a
reproductive state (R to R; B). Transitions are estimated over 10 week
intervals. Estimates are given by site: "Ex" sites refer to areas treated with
mammalian predator exclusion while "Con" sites refer to areas where
mammalian predators were allowed access. Supplemental feeding
treatments were added to Con and Ex sites 2 and 4 from summer 2007
through 2009. All sites were burned during the winters of 2005, 2007 and
2009.









CHAPTER 4
CONCLUSIONS

Although the three target species in this study, cotton rats, cotton mice, and

oldfield mice occupy similar habitats, each showed different responses to supplemental

feeding, predation, and prescribed fire treatments. Cotton rat populations were primarily

driven by fire events, although food effects were apparent in non-fire periods, and likely

would have become important in post-fire periods if predation and emigration had not

caused rapid population declines. Although cotton rat mortality and turnover is primarily

driven by predation, there was little evidence of direct, negative effects of mammalian

predation on cotton rat survival and reproduction. However, we detected evidence that

cotton rats behave adaptively in response to predation pressure as male rats had

smaller home ranges in predator access grids compared to exclosures.

Of the three treatments applied here, oldfield mice were most strongly affected by

predation. Predator access grids were associated with smaller abundances and lower

survival. The observed effects occurred in the expected form and direction, indicating a

negative effect from mammalian predation. However, lacking behavioral data, it is

unclear whether these effects were consumptive or non-consumptive in nature.

Although previous studies have demonstrated that oldfield mice generally exhibit a

short-term neutral response to fire, probably due to their preference for areas that

already have little cover, we found that prescribed burning had a negative impact on

transitions to reproductive states for this species. We believe this may be due to

changes in availability of preferred food sources such as insects.

Cotton mouse populations saw different treatment effects on the various

population parameters. Food addition increased abundances. Survival increased in









predator exclusion grids following fires. Reproduction was positively influenced by the

interaction of predator exclusion and food supplementation and negatively influenced by

burning. These complex responses may result from behavioral responses to predation

pressure. For example, following fires, when predation risk is high, cotton mice appear

to make trade-offs in favor of survival at the expense of reproduction. However, as with

oldfield mice, the understanding of this system would be improved if behavioral data

were available to supplement the trapping data considered here.

There is a need for long-term, large-scale, replicated experiments to further

address questions relating to interactions between predation and food resources,

particularly in terrestrial systems. Such experiments would benefit from the

incorporation of behavioral data with data regarding vital rates.









LIST OF REFERENCES


Abrams, P. A. 1983. Life history strategies of optimal foragers. Theoretical Population
Biology 24:22 38.

Abrams, P. A. 1984. Foraging time optimization and interactions in food webs. The
American Naturalist 124:80-96.

Abrams, P. A. 1991. Life history and the relationship between food availability and
foraging effort. Ecology 72:1242-1252.

Abrams, P. A. 1992a. Predators that benefit prey and prey that harm predators:
Unusual effects of interacting foraging adaptations. American Naturalist
140:573-600.

Abrams, P. A. 1992b. Why don't predators have positive effects on prey populations?
Evolutionary Ecology 6:449-457.

Abrams, P. A. 1993. Why predation rates should not be proportional to predator
density. Ecology 74:726-733.

Abrams, P. A., and L. Rowe. 1996. The effects of predation on the age and size of
maturity of prey. Evolution 50:1052-1061.

Allison, P. D. 1995. Survival analysis using SAS: A practical guide. SAS Press, Cary,
North Carolina, USA.

Arata, A. A. 1959. Effects of burning on vegetation and rodent populations in a
longleaf pine turkey oak association in north central Florida. Quarterly Journal of
the Florida Academy of Sciences 22:94-104.

Arthur, A. D., R. P. Pech, and C. R. Dickman. 2004. Habitat structure mediates the
non-lethal effects of predation on enclosed populations of house mice. Journal of
Animal Ecology 73:867-877.

Atkinson, J. B., J. Brock, and R. Smith. 1996. Operational longleaf pine management at
Ichauway. The Longleaf Pine Alliance Annual Meeting, Mobile, Alabama, USA.

Beyer, H. L. 2004. Hawth's Analysis Tools for ArcGIS. Available at
http://www.spatialecology.com/htools.

Bigler, W. J., and J. H. Jenkins. 1975. Population characteristics of Peromyscus
gossypinus and Sigmodon hispidus in tropical hammocks of south Florida. Journal
of Mammalogy 56:633-643.









Blair, W. F. 1951. Population structure, social behavior, and environmental relations in a
natural population of the beach mouse (Peromyscus polionotus leucocephalus).
Contributions of the Laboratory of Vertebrate Biology, University of Michigan
48:1-47.

Bock, C. E., and J. H. Bock. 1978. Response of birds, small mammals, and vegetation
to burning sacaton grasslands in southeastern Arizona. Journal of Range
Management 31:296-300.

Boutin, S. 1990. Food supplementation experiments with terrestrial vertebrates:
Patterns, problems, and the future. Canadian Journal of Zoology 68:203-220.

Brockway, D. G., and C. E. Lewis. 1997. Long term effects of dormant-season
prescribed fire on plant community diversity, structure, and productivity in a
longleaf pine wiregrass ecosystem. Forest Ecology and Management 96:167-183.

Byrom, A. E., T. J. Karels, C. J. Krebs, and R. Boonstra. 2000. Experimental
manipulation of predation and food supply of arctic ground squirrels in the boreal
forest. Canadian Journal of Zoology 78:1309-1319.

Cameron, G., and B. D. Eshelman. 1996. Growth and reproduction of hispid cotton rats
(Sigmodon hispidus) in response to naturally occurring levels of dietary protein.
Journal of Mammalogy 77(1):220-231.

Cameron, G. N., and S. R. Spencer. 1981. Sigmodon hispidus. Mammalian Species
158:1-9.

Cameron, G. N., and S. R. Spencer. 1985. Assessment of space-use patterns in the
hispid cotton rat (Sigmodon hispidus). Oecologia 68:133-139.

Campbell, M. T., and N. A. Slade. 1995. The effect of maternal mass on litter size and
offspring survival in the hispid cotton rat (Sigmodon hispidus). Canadian Journal of
Zoology 73:133-140.

Clark, C. W., and M. Mangel. 2000. Dynamic state variable models in ecology. Oxford
University Press, New York, New York, USA.

Conner, L. M., J. R. Rutledge, and L. L. Smith. 2010. Effects of mesopredators on nest
survival of shrub-nesting birds. Journal of Wildlife Management 74:73-80.

Cox, D. R. 1972. Regression models and life tables. Journal of the Royal Statistical
Society, B34:187-220.

Derrick, A. M. 2007. Small mammal responses to prescribed fire and simulated mammal
removal in a longleaf pine ecosystem. M.S. Thesis, University of Georgia, Athens,
Georgia, USA.










Desy, E. A., and G. O. Batzli. 1989. Effects of food availability and predation on prairie
vole demography: A field experiment. Ecology 70:411-421.

Desy, E. A., G. O. Batzli, and J. Liu. 1990. Effects of food and predation on behavior of
prairie voles: A field experiment. Oikos 58:159-168.

Dickman, C. R. 1992. Predation and habitat shift in the house mouse, Mus domesticus.
Ecology 73:313-322.
th
ESRI. 2005. ArcGIS. 9 ed. Environmental Systems Research Institute, Inc., Redlands,
California, USA.

Fleharty, E. D., and M. A. Mares. 1973. Habitat preference and spatial relations of
Sigmodon hispidus on a remnant prairie in west-central Kansas. The
Southwestern Naturalist 18:21-29.

Gannon, W. L., R. S. Sikes, and the Animal Care and Use Committee of the American
Society of Mammalogists. 2007. Guidelines of the American Society of
Mammalogists for the use of wild mammals in research. Journal of Mammalogy
88:809-823.

Goertz, J. W. 1964. The influence of habitat quality upon density of cotton rat
populations. Ecological Monographs 34:359-381.

Guthery, F. S., and S. L. Beasom. 1977. Responses of game and non-game wildlife to
predator control in south Texas. Journal of Range Management 30:404-409.

Hatchell, G. E. 1964. Small mammal species and populations in the longleaf pine forest
type of Louisiana. United States Department of Agriculture Forest Service
Research Paper SO-10.

Heppell, S. S., H. Caswell, and L. B. Crowder. 2000. Life histories and elasticity
patterns: Perturbation analysis for species with minimal demographic data.
Ecology 81:654-665.

Hik, D. S. 1995. Does risk of predation influence population dynamics? Evidence from
the cyclic decline of snowshoe hares. Wildlife Research 22:115-129.

Hubbs, A. H., and R. Boonstra. 1997. Population limitation in Arctic ground squirrels:
Effects of food and predation. Journal of Animal Ecology 66:527-541.

Hubbs, A. H., and R. Boonstra. 1998. Effects of food and predators on the home-range
sizes of arctic ground squirrels (Spermophilus parryii). Canadian Journal of
Zoology 76:592-596.









Kie, J. G., J. A. Baldwin, and C. J. Evans. 1994. CALHOME: Home range analysis
program electronic user's manual. United States Forest Service Pacific Northwest
Research Station, Gekeler Lane, La Grande, Oregon, USA.

Koivula, M., E. Koskela, T. Mappes, and T. A. Oksanen. 2003. Cost of reproduction in
the wild: Manipulation of reproductive effort in the bank vole. Ecology
84:398-405.

Krebs, C. J., S. Boutin, R. Boonstra, A. R. E. Sinclair, J. N. M. Smith, M. R. T. Dale, K.
Martin, and R. Turkington. 1995. Impact of food and predation on the
snowshoe hare cycle. Science 269:1112-1115.

Laake, J., and E. Rexstad. 2008. RMark an alternative approach to building linear
models in MARK. Program MARK: 'A Gentle Introduction', 7th ed. (eds. E. Cooch
and G.C. White), pp. C-1-C-115. Available at:
http://www.phidot.org/software/mark/docs/book, accessed March 13, 2010.

Layne, J. N. 1974. Ecology of small mammals in a flatwoods habitat in north central
Florida with emphasis on the cotton rat (Sigmodon hispidus). American Museum
Novitates 2544:1-48.

Lima, S. L. 1998. Nonlethal effects in the ecology of predator-prey interactions.
BioScience 48:25-34.

Lima, S. L., and L. M. Dill. 1989. Behavioral decisions made under the risk of predation:
A review and prospectus. Canadian Journal of Zoology 68:619-640.

Liu, E. S. 1971. The ethology of the cotton rat, Sigmodon hispidus. M.S. Thesis.
University of Florida, Gainesville, Florida, USA.

Masters, R. E., C. W. Wilson, D. S. Cram, G. A. Bukenhofer, and R. L. Lochmiller. 2002.
Influence of ecosystem restoration for red-cockaded woodpeckers on breeding
bird and small mammal communities. In: Ford, M. W., K. R. Russel, C. E.
Moorman, eds. Proceedings: The role of fire for non-game wildlife management
and community restoration: Traditional uses and new directions. General
Technical Report NE-288. Newton Square, PA. US Department of Agriculture,
Forest Service, Northeastern Research Station: 73-90.

Masters, R. E. 2007. The importance of short-leaf pine for wildlife and diversity in mixed
oak-pine forests and in pine-grassland woodlands. In: Kabrick, J. M., D. C. Dey,
D. Gwaze, eds. 2006. Shortleaf pine restoration and ecology in the Ozarks:
Proceedings of a symposium. General Technical Report NRS-P-15. Newton
Square, PA: US Department of Agriculture, Forest Service, Northern Research
Station: 35-46.









McCay, T. S. 2000. Use of woody debris by cotton mice (Peromyscus gossypinus) in a
southeastern pine forest. Journal of Mammalogy 81:527-535.

McNamara, J. M., and A. I. Houston. 1987. Starvation and predation as factors limiting
population size. Ecology 68:1515-1519.

Miller, D. A., R. E. Thill, M. A. Melchiors, T. B. Wigley, and P. A. Tappe. 2004. Small
mammal communities of streamside management zones in intensively managed
pine forests of Arkansas. Forest Ecology and Management 203:381-393.

Meserve, P. L., J. R. Gutierrez, and F. M. Jaksic. 1993. Effects of vertebrate predation
on a caviomorph rodent, the degu (Octodon degus) in a semiarid thorn scrub
community in Chile. Oecologia 94:153-158.

Odum, E. P., S. E. Pomeroy, J. C. Dickinson, III, and K. Hutcheson. 1973. The effects of
late winter litter burn on the composition, productivity and diversity of a 4-year old
fallow field in Georgia. Proceedings of the Annual Tall Timbers Fire Ecology
Conference, Tallahassee,
Florida, USA, 13:399-427.

Oli, M. K. 2003. Population cycles of small rodents are caused by specialist predators:
Or are they? Trends in Ecology and Evolution 18:105-107.

Oli, M. K., and F. S. Dobson. 2003. The relative importance of life-history variables to
population growth rate in mammals: Cole's predation revisited. The American
Naturalist 161:422-440.

Peckarsky, B. L., P. A. Abrams, D. I. Bolnick, L. M. Dill, J. H. Grabowski, B. Luttbeg, J.
L. Orrock, S. D. Peacor, E. L. Preisser, O. J. Schmitz, and G. C. Trussell. 2008.
Revisiting the classics: Considering nonconsumptive effects in textbook examples
of predator-prey interactions. Ecology 89:2416-2425.

Perrin, M. R., and S. J. Johnson. 1999. The effect of supplemental food and cover
availability on a population of the striped mouse. South African Journal of Wildlife
Research 29:15-18.

Pfister, C. A. 1998. Patterns of variance in stage-structured populations: Evolutionary
predictions and ecological implications. Proceedings of the National Academy of
Sciences of the United States of America 95:213-218.

Pollock, K. H. 1982. A capture-recapture design robust to unequal probability of capture.
Journal of Wildlife Management 46:757-760.

Pressier, E. L., D. I. Bolnick, and M. F. Benard. 2005. Scared to death? The effects of
intimidation and consumption in predator-prey interactions. Ecology 86:501-509.









Pressier, E. L., and D. I. Bolnick. 2008. The many faces of fear: Comparing the
pathways and impacts of nonconsumptive predator effects on prey populations.
PLoS ONE 3:e2465.

R Development Core Team. 2009. R: A language and environment for statistical
computing. R Foundation for Statistical Computing,Vienna, Austria. Available at:
http://www.R-project.org.

Rehmeier, R. L., G. A. Kaufman, D. W. Kaufman, and B. R. McMillan. 2005. Long-term
study of abundance of the hispid cotton rat in native tallgrass prairie. Journal of
Mammalogy 86:670-676.

SAS Institute Inc. 2004. SAS/STAT user's guide, version 9.1. SAS Institute Inc., Cary,
North Carolina.

Schabenerger, O., and F. J. Pierce. 2002. Contemporary statistical models for the plant
and soil sciences. CRC Press, New York, New York, USA.

Schnell, J. H. 1968. The limiting effects of natural predation on experimental cotton rat
populations. Journal of Wildlife Management 32:698-711.

Schwarz, C. J., and A. N. Arnason. 1996. A general methodology for the analysis of
open-model capture-recapture experiments. Biometrics 52:860-873.

Seaman, D. E., and R. A. Powell. 1996. An evaluation of the accuracy of kernel
estimates for home range analysis. Ecology 77:2075-2085.

Shadowen, H. E. 1963. A live-trap study of small mammals in Louisiana. Journal of
Mammalogy 44:103-108.

Spencer S. R., G. N. Cameron, and R. K. Swihart. 1990. Operationally defining home
range: Temporal dependence exhibited by hispid cotton rats. Ecology
71:1817-1822.

Suazo, A. A., J. E. Fauth, J. D. Roth, C. L. Parkinson, and I. J. Stout. 2009. Responses
of small rodents to habitat restoration and management for the imperiled Florida
scrub-jay. Biological Conservation 142:2322-2328.

Swihart R. K. and N. A. Slade. 1985a. Influence of sampling interval on estimates of
home-range size. Journal of Wildlife Management 49:1019-1025.

Swihart R. K. and N. A. Slade. 1985b. Testing for independence of observations in
animal movements. Ecology 66:1176-1184.

Tiatt, M. J., and C. J. Krebs. 1983. Predation, cover, and food manipulations during a
spring decline of Microtus townsendii. Journal of Animal Ecology 52:837-848.









Van Lear, D. H., W. D. Carroll, P. R. Kapeluck, and R. Johnson. 2005. History and
restoration of the longleaf pine-grassland ecosystem: Implications for species at
risk. Forest Ecology and Management 211:150-165.

Van Vuren, D., and K. B. Armitage. 1994. Survival of dispersing and philopatric yellow-
bellied marmots: What is the cost of dispersal? Oikos 69:179-181.

Webb, R. E., D. M. Leslie, Jr., R. L. Lochmiller, and R. E. Masters. 2005. Impact of food
supplementation and methionine on high densities of cotton rats: Support for the
amino-acid-quality hypothesis? Journal of Mammalogy 86:46-55.

Weigert, R. G. 1972. Avian versus mammalian predation on a population of cotton rats.
Journal of Wildlife Management 36:1322-1327.

Werner, E., and S. Peacor. 2003. A review of trait-mediated indirect interactions in
ecological communities. Ecology 84:1083-1100.

Whitaker, J. O., Jr., W. J. Hamilton, Jr. 1998. Mammals of the eastern United States, 3rd
edition. Cornell University Press, Ithaca, New York, USA.

White, G. C., and K. P. Burnham. 1999. Program MARK: Survival estimation from
populations of marked animals. Bird Study 46(Supplement): 120-138.

White, G. C. 2002. Discussion comments on: The use of auxiliary variables in capture-
recapture modeling. Journal of Applied Statistics 29:103-106.

Wolfe, J. L., and A. V. Linzey. 1977. Peromyscus gossypinus. Mammalian Species
70:1-5

Yunger, J. A. 2004. Movement and spatial organization of small mammals following
vertebrate predator exclusions. Oecologia 139:647-651.


100









BIOGRAPHICAL SKETCH

Gail Morris is originally from Williamsport, Pennsylvania. She attended

Muhlenberg College in Allentown, PA and received a BS in Biology in 2004. Following

a series of seasonal wildlife field jobs, she ended up at the Joseph W. Jones Ecological

Research Center where, among other things, she caught and chased rats for a year.

This lead to the unexpected discovery of a fascination with rats, mice, and other critters

at the bottom of the food chain. She was eventually offered a graduate assistantship

with the University of Florida which enabled the continuation of rat and mouse studies at

the Jones Center.


101





PAGE 1

EFFECTS OF MAMMALIAN PREDATOR EXCLUSION, SUPPLEMENTAL FEEDING, AND PR ESECRIBED FIRE ON SMALL MA MMAL POPULATIONS IN A LONGLEAF PINE ECOSYSTEM By GAIL MORRIS A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE UNIVERSITY OF FLORIDA 2010 1

PAGE 2

2010 Gail Morris 2

PAGE 3

[T]hinking about rats, as low-down as it seems, can easily lead to thoughts about larger topics, such as life and death and the nature of man. Robert Sullivan, Rats: Observations on the History & Habitat of th e Citys Most Unwanted Inhabitants 3

PAGE 4

ACKNOWLEDGMENT S I received much support during the course of this project. The University of Florida and the Joseph W. Jones Ecological Resear ch Center provided funding, equipment, and manpower. My co-chairs Madan Oli and Mike Conner and committee member Mel Sunquist gave a great deal of assistance, es pecially with experimental design, analysis, and editorial suggestions for this thesis. I am also grateful fo r the help of numerous technicians who assisted with data collecti on, entry, and management. Cat Eddins and Jen Eells got the project off to a good star t and helped work out early difficulties. Bonnie Fairbanks, Amanda Goldberg, and Jami e Utz held down the fort, catching and chasing many rats while I was away from th e field for a semester. James Miller, Megan Munroe, and Jud Swart, braved the summer heat and gnats in pursuit of ridiculous number of rats. Evan Hill, Amy Kryzton-Presson, and Erica Rigsby braved heat, gnats, mosquitoes, and the absence of the labs fearless leader in pursuit of even more rats and made tracking around the fire a success. Hayden Martin and Cliff Rushton brought the field work to a close and assisted with proofing tens of t housands of trapping records. Jimmy Atkinson, Scott Smith, Ca t Eddins, and Briant Williamson carried out the prescribed burns. The Herpetology lab, especially Jen Linehan and Kelly McKean, caught many rat-eating snakes and cared fo r them until the snakes passed the radio collars. Anna Derrick showed me the smam maling ropes with such great enthusiasm my own almost seems normal. I am grateful for her advice which I am sure saved me from many mistakes and frus tration. Brent Howze was helpful with many random aspects of fieldwork, especially with those relating to radio telemetry. Jean Brock, Michael Simmons, and John Merritt prov ided assistance with GIS, data management, and IT support respectively. Liz Cox acquir ed even the most obscure references with 4

PAGE 5

great speed. I am also gratef ul for the support of the admini strative staff at both the Jones Center and the Department of Wildlife Ecology and Conserva tion at UF. Two individuals deserve special thanks. Jessica Rutledge provided assistance with field work, data management, equipmen t, and pretty much everything else imaginable. Jeff Hostetler provided much patient assistance with data analysis and also made editorial suggestions. I dont care to think about how much more difficult this would have been without the help of these two individuals. I also thank my family and friends for t heir support and for not saying things like Ew, rats! all the times I ta lk about rats. I th ank my personal aggravation of rat, and her less long-lived partner, for being highly enter taining despite a prickly nature, and for showing me many things about the ratty nature. 5

PAGE 6

TABLE OF CONTENTS page ACKNOWLEDGMENTS ..................................................................................................4LIST OF TABLES ............................................................................................................8LIST OF FIGURES ........................................................................................................11ABSTRACT ...................................................................................................................12 CHAPTERS 1 INTRODUC TION....................................................................................................142 EFFECTS OF PRESCRIBED FIRE SUPPLEMENTAL FEEDING, AND MAMMALIAN PREDATOR EXCLUSION ON COTTON RAT SPACE USE AND POPULATION DYNAMICS IN A LONGLEAF PINE ECOSYST EM........................18Introduction .............................................................................................................18Methods ..................................................................................................................20Study Site and Species ....................................................................................20Experimental Design ........................................................................................21Field Methods ...................................................................................................21Statistical Methods ...........................................................................................23Analysis of capture-mark-recapture (CMR) data ........................................23Survival analysis using radio telemetry data ..............................................28Home range analysis .................................................................................28Results ....................................................................................................................30CMR Analyses ..................................................................................................30Analyses of Radio Telemetry Data ...................................................................32Collared rat survival ...................................................................................32Effects on home range size ........................................................................33Effects on home range exclusivity ..............................................................34Effects of fire on radio collared rats ............................................................34Discussion ..............................................................................................................36Fire Effects .......................................................................................................36Predation and Supplem ental Feeding Effects ..................................................37Home Range Exclusivity ...................................................................................40Conclusions ......................................................................................................423 EFFECTS OF SUPPLEMENTAL F EEDING, MAMMALIAN PREDATOR EXCLUSION, AND PRESCRIBED FIRE ON COTTON AND OLDFIELD MOUSE POPULATIONS IN A LO NGLEAF PINE ECOSYST EM...........................56Introduction .............................................................................................................56Methods ..................................................................................................................57 6

PAGE 7

Study Site and Species ....................................................................................57Field Methods ...................................................................................................58Statistical Methods ...........................................................................................59Results ....................................................................................................................66Cotton Mice ......................................................................................................66Oldfield Mice .....................................................................................................68Discussion ..............................................................................................................70Treatment Effects on Cotton Mice ....................................................................71Treatment Effects on Oldfield Mice ..................................................................74Conclusions ......................................................................................................764 CONCLUS IONS.....................................................................................................92LIST OF REFERENCES ...............................................................................................94BIOGRAPHICAL SKETCH ..........................................................................................101 7

PAGE 8

LIST OF TABLES Table page 2-1 Model comparison table for mult istate capture-mark-recapture analysis assessing occurrence of peak breeding se asons of cotton rats in general (Model set 1) and with respect to winter prescribed burns (Model set 2) in Southwestern Georgia between 2005 and 2009.. ............................................... 43 2-2 Model comparison table for mult istate capture-mark-recapture analysis assessing sex, reproductive condition, breeding season and time effects on capture probability ( p ), survival ( S ), and rates of transitions between reproductive states ( ) in cotton rats in Southwestern Georgia between 2005 and 2009. ........................................................................................................... 44 2-3 Model comparison table for multis tate capture-mark-recapture analysis assessing the term of effect of win ter prescribed burns on survival ( S ) of cotton rats in Southwestern Georgia between 2005 and 2009. ..........................45 2-4 Model comparison table for multis tate capture-mark-recapture analysis assessing potential for paired site effects on survival ( S ) and rate of transitions to reproductive states ( ) of cotton rats in Southwestern Georgia between 2005 and 2009. ....................................................................................46 2-5 Model comparison table for robus t design capture-mark-recapture analysis assessing demographic and ti me effects on abundance ( N ), capture probability ( p ), and recapture probability ( c) in cotton rats in Southwestern Georgia between 2005 and 2009.. .....................................................................47 2-6 Model comparison table for mult istate capture-mark-recapture analysis examining the effect of predation, s upplemental feeding, and fire treatments on survival ( S ) and transition probabilities ( ,between reproductive and non-reproductive states) of cotton rats in Southwestern Georgia between 2005 and 2009. ...................................................................................................48 2-7 Factors influencing surv ival of radio collared cott on rats in sites treated with supplemental feeding, winter prescr ibed fires, and mammalian predator exclusion in southwes tern Georgia from June 2007 August 2009. ..................49 2-8 Factors influencing ho me range size of cotton rats in southwestern Georgia using 95% minimum convex polygon (MC P) and 95% fixed kernel (Kernel) estimates. ...........................................................................................................50 2-9 Home range exclusivities for cotton rats in sites treated with supplemental feeding and mammalian predator exclusi on in southwestern Georgi a from June 2007 to August 2009. .................................................................................51 8

PAGE 9

2-10 Factors influencing ho me range exclusivities of co tton rats in sites treated with supplemental feeding and mammalian predator exclusion in southwestern Georgia from June 2007 to August 2009. .....................................52 3-1 Model comparison table for mult istate capture-mark-recapture analysis assessing occurrence of peak breeding se asons of cotton mice in general (Model set 1) and with respect to winter prescribed burns (Model set 2) in Southwestern Georgia between 2005 and 2009. ................................................77 3-2 Model comparison table for mult istate capture-mark-recapture analysis assessing occurrence of peak breeding se asons of oldfied mice in general (Model set 1) and with respect to winter prescribed burns (Model set 2) in Southwestern Georgia between 2005 and 2009. ................................................78 3-3 Model comparison table for mult istate capture-mark-recapture analysis assessing sex, reproductive condition, breeding season, and time effects on capture probability ( p ), survival ( S ), and transitions rates between reproductive states ( ) in cotton mice in Sout hwestern Georgia between 2005 and 2009. See Table 3-1 for column definitions. ......................................79 3-4 Model comparison table for mult istate capture-mark-recapture analysis assessing sex, reproductive condition, breeding season, and time effects on capture probability ( p ), survival ( S ), and transitions rates between reproductive states ( ) in oldfield mice in So uthwestern Georgia between 2005 and 2009. ...................................................................................................80 3-5 Model comparison table for mult istate capture-mark-recapture analysis assessing term of effect of winter prescribed burns on survival of cotton and oldfield mice in Southwes tern Georgia between 2005 and 2009. ....................... 81 3-6 Model comparison table for multis tate capture-mark-recapture analysis assessing potential for site effects on survival ( S ) and rate of transitions to reproductive states ( ) of cotton and oldfield mice in Southwestern Georgia between 2005 and 2009. ....................................................................................82 3-7 Model comparison table for robus t design capture-mark-recapture analysis assessing demographic and ti me effects on abundance ( N ), capture probability ( p ), and recapture probability ( c ) in cotton mice in Southwestern Georgia between 2005 and 2009. ......................................................................83 3-8 Model comparison table for PO PAN capture-mark-recapture analysis assessing site and time effects on abundance ( N ), capture probability ( p ), and entry probability ( pent ) in oldf ield mice in S outhwestern Georgia between 2005 and 2009. ......................................................................... 84 9

PAGE 10

3-9 Model comparison table for mult istate capture-mark-recapture analysis examining the effect of predation, feedi ng, and fire treatments on survival (S) and transition probabilities ( between reproductive and non-reproductive states) of cotton mice in southwestern Georgia, 2005-2009. ..............................86 3-10 Model comparison table for mult istate capture-mark-recapture analysis examining the effect of predation, feedi ng, and fire treatments on survival (S) and transition probabilities ( between reproductive and non-reproductive states) of oldfield mice in southw estern Georgia, between 2005 and 2009.. ......87 10

PAGE 11

LIST OF FIGURES Figure page 2-1 Model averaged estimates of survival of cotton rats in southwestern Georgia between 2005 and 2009 in response to prescribed fire, supplemental feeding, and predator control treatments. ...........................................................532-2 Model averaged estimates of the rates of transitions to reproductive states for male and female cotton rats during peak breeding seasons (spring and summer), non-peak seasons during wh ich burning did not occur, and non-peak seasons during which burning did occur, in southwestern Georiga between 2005 and 2009. ....................................................................................5423 Survival estimates ( standard error) of radio collared cotton rats in southwestern Georgia between 2 007 and 2009, generated using Cox proportional hazard models. ...............................................................................553-1 Model averaged estimates of survival of cotton mice in southwestern Georgia between 2005 and 2009 in response to prescribed fire, supplemental feeding, and predator control treatments. ...........................................................883-2 Model averaged estimates of breeding transitions for male and female cotton mice in southwestern Georgia between 2005 and 2009 during peak breeding seasons (fall and early winter), non-peak seasons during which burning did not occur and non-peak seasons during which burning did occur. .....................893-3 Model averaged survival estimate s in southwestern Georgia between 2005 and 2009 in response to prescribed fire, supplemental feeding, and predator control treatments. ..............................................................................................903-4 Model averaged estimates of breedi ng transitions for oldfield mice in southwestern Georgia between 2 005 and 2009 during peak breeding seasons (winter and summer) in burn and non-burn years and non-peak breeding seasons. ..............................................................................................91 11

PAGE 12

Abstract of Thesis Pres ented to the Graduate School of the University of Florida in Part ial Fulf illment of the Requirements for the Master of Science EFFECTS OF MAMMALIAN PREDATOR EXCLUSION, SUPPLEMENTAL FEEDING, AND PRESECRIBED FIRE ON SMALL MA MMAL POPULATIONS IN A LONGLEAF PINE ECOSYSTEM By Gail Morris August 2010 Chair: Madan K. Oli Major: Wildlife Ecology and Conservation Food resources and predation play important roles in determining small mammal population dynamics. Each of these factors may affect population paramet ers such as abundance, survival, and reproduction. These fa ctors may also intera ct as individuals under predation pressure make trade-offs bet ween exposure to predators and access to food resources. Fires consume food s ources and reduce cover, which increases exposure to predators. For species that occu r in areas with frequent fi res, it is beneficial to consider how all of these factors affect populations of interest. This study used a capture-mark-recapture framework to ex perimentally examine how supplemental feeding, mammalian predator exclusion, and prescribed fire affected survival, abundance, and reproduction of cotton rats ( Sigmodon hispidus ), cotton mice ( Peromyscus gossypinus ), and oldfield mice ( P. polionotus ). Radio telemetry was used to assess home range size and overlap of co tton rats. Among cotton rats, prescribed fire events had the greatest effect, causi ng large drops in survival, abundance, and reproduction. Food supplementation increased survival, rates of transitions to reproductive states, and abundances but was not sufficient to prevent post-fire declines 12

PAGE 13

13 in any of these parameters. This i ndicates that predation and emigration are responsible for fire-related dec lines in survival and abundance. Population-level effects of predator exclusion effect s were small in magnitude. Predator exclusion was, however, associated with increased male home range size, indicating a response to perceived predation risk. Among cotton mice, survival was affected (increased) only by an interaction between burning and predator exclusion. Rates of transitions to reproductive states decreased in burn year s but increased with th e combination of feeding and predator ex clusion. Feeding increased ab undances. Among oldfield mice, survival and abundance were greater in predator exclusion areas than controls. Feeding and the interaction of feeding and predator exclusion also increased abundances. Rates of transitions to reproduc tive states declined during peak breeding seasons during which burning occurred such t hat breeding transitions in these seasons were lower than in non-peak seasons. Some of these effects can only be understood by assuming individuals make behavioral responses to predation risk to limit mortality.

PAGE 14

CHAPTER 1 INTRODUCTION Predation and access to food resources c an have large and direct effects on small mammal populations. Field studies have shown food addition to be associated with increases in abundance and survival, changes in rates of immigration and emigration, earlier reproduction, and great er numbers of young produced per reproductive event (Taitt and Krebs 1983, Boutin 1990, Campbe ll and Slade 1995, Krebs et al. 1995, Hubbs and Boonstra 1997, Perrin and Johnson 1999). Experimental removal or exclusion of predators has been associated with increased densities (Schnell 1968, Weigert 1972, Meserve et al. 1993, Yunger 2004), increased survival (Meserve et al. 1993, Oli and Dobson 2003), ear lier breeding (Arthur et al. 2004), and increased immigration (Tait and Krebs 1983, Perrin and Johnson 1999). Indirect effects related to these factors may also be common and often result from behavioral responses associated with changes in predation or food resources. For example, food addition may cause changes in intraspecific aggression, and decreases in home range sizes (Boutin 1990, Desy et al. 1990, Hubbs and Boonstra 1998). Increased levels of predation risk have been associated with decreases in home range size and alterations in habitat use (Desy et al. 1990, Dickman 1992, Arthur et al. 2004, Yunger 2004). Behavioral changes may in turn affect vital rates at the population level. For example, reductions in space use asso ciated with predation risk may limit access to reproductive opportunities. Indi rect effects may be as large as or larger than direct effects, and may either take the form and di rection of direct effects, or cause unexpected effects (Pressier et al. 2005). 14

PAGE 15

Although food and predation may individually have large impacts on a population, there is also a great deal of theoretical and empirical support for interactions between these factors (Taitt and Krebs 1983, Abrams 1983, 1984, 1991, 1992 a 1992 b 1993, McNamara and Houston 1987, Desy and Batz li 1989, Hubbs and Boonstra 1997, Clark and Mangel 2000). In a large-scale exper iment examining the effect of food supplementation and predator e xclusion on snowshoe hare ( Lepus americanus) populations, Krebs et al. (1995) reported that food addition tripled p opulation densities, predator exclusion doubled pop ulation densities, and the co mbination of treatments resulted in an eleven-fold increase in popul ation densities. Similarly, Hubbs and Boonstra (1997) found that wh ile feeding and pr edator exclusion treatments each individually increased survival of arctic ground squirrels (Spermophilus parryii ), survival in a combined treatment plot was significantly greater than either treatment alone. Similar trends were found with respect to increased growth rates (Hubbs and Boonstra 1997) and decreasing home range size (Hubbs and Boonstra 1998), indicating a significant interactive effect on several aspects of species ecology and behavior. These interactions may stem from the fa ct that foraging behaviors are likely to put individuals at greater risk of predation. An individual must make trade-offs between the need to acquire food and the need to stay safe from predators. Optimally, an individual should maximize fitness by mi nimizing risk of death by predation while maximizing food intake. Rates of food in take are important to avoid starvation and because reproduction is often associated with t he quantity and quality of food available. A large body of theor etical research suggests that such trade-offs are common (reviewed in Lima and Dill 1989), may be complex, and may cause counterintuitive 15

PAGE 16

effects (Abrams 1991, 1992 a 1992 b 1993). This is especially likely to be true in systems with complex food chains and compet itive interactions between multiple predator and prey species. Effects of tr ade-offs may also depend on a species life history and may change with breeding/nonbreeding seasons (Abrams 1991). Theoretical investigations are crucial to understand and predict predator-prey interactions; however, empirical field data are al so necessary to test predictions made in theoretical studies. Unfortunately, rela tively few studies have adequately examined these factors. Many studies that attempt to do so are aquatic in nature, focus on small temporal and spatial scales, and/or lack replication. Long term, rep licated experimental studies that consider large spatial sca les and examine predator-prey interactions between terrestrial vertebrates are rare becaus e of the difficulties and expense required to adequately carry out such experimental studies (but see Krebs et al. 1995). The primary objective of this study was to experiment ally examine how predation and food resources affect small mammal survival, reproduction, and abundances in a longleaf pine ecosystem. This study was carried out over four and half years and over a large spatial scale. We focused on three ta rget species common to the longleaf pine ecosystem: cotton rats (Sigmodon hispidus), cotton mice (Peromyscus gossypinus), and oldfield mice ( P. polonious). Capture-mark-recapture methodologies were used to estimate population parameters. Space us e, in the form of home range size and overlap, was additionally consi dered for adult cotton rats. The roles of food resources and predation may take on new aspects for small mammal species following fires, which occur frequently (historically over intervals of one to five years) in longleaf pine ecosystems. Fires consume both herbaceous food 16

PAGE 17

resources and vegetative cover which increases predation risk. Different small mammal species exhibit different short-term reacti ons to fires. Cott on rat populations decline dramatically (Layne 1974, Bock and Bock 1978) while cotton mice exhibit either a neutral or short-term positive response to fire (Shadowen 1963, Hatchell 1964, Layne 1974, Suazo et al. 2009). Oldfield mice do not appear to respond strongly to fire over the short term (Arata 1959, Odum et al. 1973, Suazo et al. 2009). All three of these species benefit from frequent fire applicati on over the long term (Masters et al. 2002, 2007, Suazo et al. 2009). While the long term benefits of fire are likely related to the maintenance of ideal habitat structure (Masters et al. 2002), the driving factors behind population changes over the short term remain poorly understood, although explanations relating to changes in food res ources or cover, and by extension predation risk, are generally hypothesized. Therefore, a secondary objective of this study was to experimentally examine t he roles of predation and food resources in small mammal population changes following prescribed fires. 17

PAGE 18

CHAPTER 2 EFFECTS OF PRESCRIBED FIRE, SUPPL EMENTAL FEEDING, AND MAMMALIAN PREDATOR EXCLUSION ON COTTON RAT SPACE USE AND POPULATION DYNAMICS IN A LONGLEAF PINE ECOSYSTEM Introduction Food availability and predation can hav e dramatic effects on small mammal populations. Abundant food resources hav e been associated with increases in reproduction (earlier and/or longer reproduc tive seasons, more young produced per reproductive event), survival, abundance, and immigration (Boutin 1990, Campbell and Slade 1995, Webb et al. 2005). Predation may also have large effects. At its most basic level, predation removes individuals from a population causing an immediate decrease in abundance. Predation may also have sub-lethal effects relating to perceived risk of predation. These include changes in behavior such as reduced activity, changes in habitat use, decreased home range size, and delay in age of first reproduction or in the onset of a reproductive season (Lima and Dill 1989). Non-le thal effects that ma y result from these behaviors include decreased growth rate s, poorer body condition, and decreased overall reproductive output (Hik 1995, Pecka rsky et al. 2008). Non-lethal effects on prey demographics may be as impor tant as or more important than direct consumption (Werner and Peacor 2003, Preisser et al. 2005). Clearly, changes in population parameters caused by changes in food resources and/or predation can have significant effects on population growth and abundance. For rodent species which mature and reproduce rapidly (fast species), changes in reproductive rates can have especially lar ge impacts at the population level (Oli and Dobson 2003). 18

PAGE 19

Effects relating to predation and food resources do not occur in isolation. In the presence of predation pressure, individuals mu st make choices that balance predation risk with the need to engage in activities that put them at risk of predation suc h as searching for food (McNamara and Houston 1987, Abrams 1991). Individuals avoid foraging in areas where they are vulnerable to predator s even if good forage occurs there, unless they are food stressed such t hat the risk of death by starvation outweighs the risk of predation. Similar tradeoffs may occur with res pect to other behaviors, such as seeking out reproductive opportunities (Lima and Dill 1989, Clark and Mangel 2000). For species such as the cotton rat ( Sigmodon hispidus ), the occurrence of fires causes a crisis of both food and predation. Fire consumes herbaceous vegetation that cotton rats require for cover from predators and which they also use as a primary food source (Whitaker and Hamilton 1998). Multiple authors have noted precipitous declines in cotton rat populations following fires (A rata 1959, Bock and Bock 1978, Rehmeier et al. 2005). We know of no studies that have attempted to experimentally determine the driving forces behind these declines, alt hough hypotheses relating to food resources and/or predation are generally proposed. Many studies have examined how predation affects indivi duals sub-lethally by influencing behavioral decisions; fewer have been able to connect these decisions to population-level effects. Fewer still have examined such effects on large spatial and temporal scales. Experimental manipulat ion of predators and food resources allows a better understanding of how predation and food resources influence both individual behaviors and population dynamics. We examined the effe cts of predat ion and food resource availability on demography and space use of cotton rats ov er the course of 19

PAGE 20

four and a half years and through three pre scribed fire cycles by experimentally manipulating presence of mamma lian predators and food resources. Using capture-mark-recapture (CMR) and radio telemetry data, we determined the impact of food resources and predation on cotton rat space use and population dynamic s in general and following prescribed fire events. Methods Study Site and Species This research was conducted at the Joseph W. Jones Ecological Research Center at Ichauway in Baker County, Georgia. Ichauway is a 12,000 ha property consisting primarily of longleaf pine ( Pinus palustris ) and wiregrass ( Aristida beyrichiana ) ecosystem. Longleaf pine ecosystems are char acterized by a low-density longleaf pine over-story, a diverse, herbaceous groundcover, and an open, park-like mid-story (Van Lear et al. 2005). Hardwood tree species occur at limited levels. Frequent, low intensity fires are key ecologic al processes. Consequently, app lication of prescribed fire is a primary management tool throughout Ic hauway; most sites are burned on a two year rotation (Atkinson et al. 1996). Cotton rats are solitary, crepuscu lar rodents found abun dantly across the southeastern United States. They occur in many habitats, but require thick cover, particularly in the form of dense grasses and shrubs (Goertz 1964). Cover is essential for protection from a wide range of avian, mammalian, and snake predators. Herbaceous material is also consumed as a primary food source and used in nest construction. Predation is the most commo n cause of death among cotton rats (Weigert 1972, Derrick 2007), and predation pressure is so strong t hat cotton rat populations experience near complete turnover in as little as five to eight months (Goertz 1964). 20

PAGE 21

Experimental Design In 2002, the Jones Center constructed four mammalian predator exc losures, each paired with a nearby control wit h similar habitat. Plots range in size from 35.94 to 49.09 ha. Exclosures are surrounded by 1.2 m tall woven wire fences which carry electrified lines along the top, middle, and bottom to discourage mammals from climbing over or digging under (the weave is lar ge enough to allow small mammals and snakes to pass through). Although mammalian predator s occasionally enter exclosures, regular monitoring by track counts and thermal camera surveys indicate significantly fewer mammalian predators in exclosures t han controls (Conner at al. 2010). In June of 2007, two exclosure and two control grids were randomly selected to receive a supplemental feeding treatment cons isting of placing 113 g (4 oz) of rabbit chow in cans at every other station trapping grids in each control and exclosure (see below). Food was replaced every other week. Empty cans were placed on non-feeding grids. This treatment continued through August 2009. Images from trail cameras demonstrated that cotton rats, cotton mice ( Peromyscus gossypinus ), oldfield mice ( P. polionotus ), house mice ( Mus musculus ), woodrats (Neotoma floridana ), flying squirrels ( Glaucomys volans ), and eastern cottontails ( Sylvilagus floridanus ) regularly used feeding stations. We found no evidence that cans were defended by individuals of any species. In February of 2005, 2007, and 2009, all plots were burned according to Ichauways burn plan which has these study areas on a two year burn rotation. Field Methods Each control and exclosure contains a 12x12 small mammal trapping grid with 15 m spacing between stations. Pairs of grids were trapped four times per year (once 21

PAGE 22

each season) from January 2005 through June 2007 and eight times per year (twice per season) from July 2007 through the June 2009 using Sherman live traps (H.B. Sherman Traps, Tallahassee, Florida, USA). A sm all amount of a granul ar insecticide was sprinkled around each trap to prevent deaths due to fire ants. New captures were marked individually with meta l ear tags. Data recorded for all captures included location, species, sex, weight, age (adult or juvenile, based on we ight), reproductive condition (for males, testes descended or no t, for females, if pr egnant and/or lactating or not), and hind foot measurement. In four of the eight study plots (one fed predator exclos ure, one unfed predator exclosure, one fed predator control, and one unfed predator control plot) cotton rats weighing 90 g or more (so t hat collar weight was not > 5% of the rats mass) were anesthetized with Isoflurane and fitted with r adio collars (Advanced Telemetry Systems Isanti, Minnesota, USA; Sirtrack Wildlife Tracking Solutions, Havelock North, New Zealand; and Telenax, Playa del Carmen, Me xico). Following recovery, rats were released at their capture site. Collared rats were located by triangulation or homing a minimum of three times per w eek and located visually once a week to confirm status as alive or dead. Rats were located using TRXC-2000S (Wildlife Ma terials Murpheysboro, Illinois, USA), R-1000 (Communication Specia lists, Inc. Orange, California, USA), or R-2000 receivers (Advanced Telemetry Systems Isanti, Minnesota, USA). When rats were found dead, the location was searched for sign to classify the event as a slip or death due to avian, mammalian, snake, or unknown predation, handling, or unknown causes. If rats slipped or chewed off collars or if signals were los t, attempts were made to retrap and recollar rats. Searches for mi ssing signals were undertaken if rats could 22

PAGE 23

not be retrapped. Collaring and tracking beg an in July 2007 and continued through August 2009. Collared rats were tracked intensivel y around the 2009 burn. Rats were homed on to verify status as alive one to two da ys prior to burning. Signals were monitored during the burn and rats were homed on again i mmediately after the fire. For seven days following the fire, rats were tracked twice a day. R ats surviving past seven days were tracked daily for an additional sev en days. Regular monitoring was restored thereafter. Missing rats were searched fo r immediately to detect emigration. Trapping, handling, and tracking met hods followed recommendations of the American Society of Mammalogists (Gannon et al. 2007) and were approved by the University of Florida Instituti onal Animal Care and Use Committee. Statistical Methods Analysis of capture-mark-recapture (CMR) data Capture-mark-recapture (CMR) data considered for this analysis included 26 sessions from January 2005 through June 2009. Analyses were carried out using the R 2.9.1 (R Development Core Team) pack age RMark 1.9.6 (Laake and Rexstad 2008) to build models for program MARK 6.0 (White and Burnham 1999). Capture probabilities were fixed to 1 for r adio collared rats. Multistate models were used to estimate and model state specific survival ( S ), capture probability ( p ), and transitions between reproductive states ( ). States used for S and were based on reproductive condition. Males were considered to be in reproductive condition if testes were descend ed, females if pregnant and/or lactating. Therefore, models ev aluating effects on S which include a reproductive condition term estimate and model survival separate ly for reproductive and non-reproductive 23

PAGE 24

individuals, and models evaluating effects on which include a reproductive condition term estimate and model probabilities of indi viduals moving between reproductive states (i.e., rates of non-reproductive individuals entering repr oductive states, rates of reproductive individuals remaining reproductive). Preliminary investigation considered the potential infl uence of trapping session, season, year, and the additive effect of season and year on p Influence of reproductive condition and sex was assessed for S and Breeding season was also considered for Assessment of effects on p S and was carried out in a sequential fashion. First, effects on p were considered while modeling S and using the most general models described above for each. Effects on S and were then considered in a similar fashion. Assessment of goodness-of-fit wa s carried out using the median approach in program MARK (White and Burnham 1999). The median test indicated a mild overdispersion ( = 1.262). Models in each param eters set were compared using Akaikes information criterion corrected for small sample size (AICc), after quasi-likelihood adjustment s (QAICc) made using = 1.262. Models were considered well supported if they had a QAICc of less than two. T he best supported model within each parameters set was selected a base for modeling that param eter in further analyses. These analyses indicated t hat reproductive condition and sex modeled in an interactive fashion ( S (reproductive condition*sex)), were important for describing S (Table 2-2, model 5). Capt ure probability was best described as varying by session. However, standard errors around parameter estimates were la rge for this model, so it 24

PAGE 25

was excluded from the analysis. The next best supported m odel, with an additive effect of year and season, p (year+season), was selected to model p (Table 2-2, model 1). To model we investigated the best approach to modeling a breeding season effect on Breeding seasons vary over the co tton rats geographic range (Cameron and Spencer 1981, Whitaker and Hamilton 1998). In the s outhern part of the range, breeding occurs year round but with peaks at certain times of the year. Because a literature review did not give a clear i ndication of when peak seasons occur in southwestern Georgia, we created a model set, based on the literature (Cameron and Spencer 1981, Whitaker and Hamilton 1998) and personal observations, to identify peak and non-peak breeding seasons. This analysis indicated reproductive peaks in spring and summer (Table 2-1, model 1). Because of potential confounding effect s of prescribed bur ning treatments and breeding seasons occurring as occasion-dependant effects, we then assessed whether there was support for further dividing the non-breeding seasons (winter and fall) into whether a burn occurred during these s easons or not. Div ision of non-breeding seasons into two classes based on burning was well supported (Table 2-2, model 7). Using this breeding season model (Table 2-2, model 7), we continued the sequential variable selection for as described above for p and S Reproductive condition, sex, and breeding season, modeled in an additive fashion ( (reproductive condition+sex+breeding season)), were important for describing (Table 2-2, model 10). Although the prescribed fires occurred at sp ecific times, fire-caused changes in cover and food resources may last for weeks or months. To determine the best effect 25

PAGE 26

window for the fire treatments, a set of models considering fire effects on survival over multiple time intervals was considered. Su rvival was constrained to be similar between all trapping periods except those following fires. Post-fire survival was allowed to vary for several different intervals, fro m incl uding only the interval during which the fire occurred (interval length of ten weeks fo llowing the 2005 and 200 7 fires and of five weeks following the 2009 fire), to includ ing intervals through the summer season (30weeks), by which time vegetation is typi cally recovered. This analysis indicated a short term fire effect on survival with decli nes occurring only in the interval during which the fire occurred (Table 2-3, model 1). Exclosure and control sites were initia lly selected as pairs based on similar habitats between pairs. As part of a post-hoc examination of non-treatment effects on S and we examined the potential for paired site effects on these parameters. Using the base models indicated by the analyses de scribed above, we ran a second set of models considering paired site effects on S and This analysis indicated paired site effects were import ant for modeling both S and (Table 2-4, model 1). Treatment effects were added to the best base multistate model ( S (reproductive condition*sex+site)p(year+season) (breeding season+reproductive condition+sex+site)) as additive and intera ctive effects (two-way only). Due to confounding effects relating to fi re and breeding season occurring as occasion-dependent effects, only food and predation treatme nts were considered with respect to while food, predation, and fire effe cts were considered with respect to S Model averaging was employed to generate parameter estimates for S and 26

PAGE 27

Pollocks robust design (Pollock 1982) was used to generate abundance estimates ( N ). Robust design models estimate probabilities for survival ( S ), capture ( p ), recapture ( c), emigration (), and staying away after emigrating ( ). The model selection approach used for the robust design was similar to that used for the multistate analysis. Preliminary investigat ion considered potential for time effects on p and c. P aired site effects were considered for N S was modeled using the best supported S model from the multistate analysis (without reproductive condition, S (sex+site+food+fire)). terms were modeled using a random emigration effect ( (.)= (.)). Preliminary analyses indicated that c and p varied by trapping session (Table 2-5 model 3) and that a paired site effect, as described abov e, was important for modeling N (Table 2-5, model 1). Because of the difficulties associat ed with modeling treatment effects on abundance directly (White 2002), the best su pported robust design model indicated by the preliminary analysis described above ( S (site+sex+food+fire) (.)= (.)p (session) c(session) N (site)) was used to generate derived abundance estimates by site and session, but not to determine treatment effects. Treatm ent effects on abundance were evaluated using a repeated measures ANOVA (Schabenberger and Pierce 2002) implemented using the PROC MIXED procedure in SAS (SAS Institute Inc. 2004). The model for this ANOVA included food, fi re, and predation treatments and their interactions (two-way interactions only). Pa ired sites were included as a random effect. Multiple covariance structures were in vestigated and the best (variance components structure, which allows a different variance for each random effect) was selected based on AICc value (Miller et al. 2004) Treatment effects were co nsidered significant at the = 0.05 level. 27

PAGE 28

Survival analys is using radio telemetry data Survival of collared rats was estimat ed using the Cox proportional hazard model (Cox 1972) implemented using the PROC P HREG procedure in SAS (Allison 1995). An a priori model set was constructed to examine the effects of season, treatment, sex, and interactions of these factors on survival. Rats that lived less than one week following collaring or that died due to handling-related causes were not considered for this analysis. To prevent an upward bias associat ed the censoring of rats in the latter group, the first week of all rats was censored as well. Rats whose signals were lost and rats who lost collars were right-censored. Seasons were pooled except for winters which were separated into winter 2008 (no burning) and winter 2009 (burning occurred) to test for a fire effect on survival. Model s were assessed using an AIC framework. Home range analysis Home ranges were estimated using 95% minimum convex polygons (MCP) and, for purpose of comparison to other studies, 95 % fixed kernel methods. To avoid short sampling intervals and small sample sizes which may contribute to inaccurate home range estimates (Swihart and Slade 1985 a Swihart and Slade 1985 b Spencer et al. 1990), we followed the recommendations of Cameron and Spencer (1985) and Swihart and Slade (1985 b ) and estimated home ranges only for rats that had a minimum of 15 locations with least 4.5 hours between locati ons. MCP estimates were generated using the program CALHOME (Kie et al. 1994). Kernel estimate s were generated in ArcGIS 9 (ESRI 2005) using the Hawths tools ext ension (Beyer 2004). Kernel bandwidth was specified by least squares cross validat ion (LSCV, Seaman and Powell 1996) for all rats, with the exception of nine rats that we re repeatedly found at the same location(s) (usually a nest or burrow). For these rats, the bandwidth was user specified to prevent 28

PAGE 29

bias such behavior can create in home r ange estimates using LSCV to determine bandwidth (Seaman and Powell 1996). Home range estimates were annual, not seasonal (although annual is perhaps misleading as no rats lived to be collared for more than a year). A t -test indicated that male rats had significa ntly larger home ranges than females; therefore, males and females were considered separ ately in further analysis. Effects of feeding, predation and the interact ion of these treatm ents on home range size were examined using a two-way ANOVA (S AS procedure PROC GLM, SAS Institute Inc. 2004, Schabenberger and Pierce 2002). Fire was not included as a treatment in this analysis due to small post-fire sample sizes. Home range exclusivity was estimated by identifying all pairs of rats that lived during the same period and which had adjac ent or overlapping MCP home ranges. Distances between pairs of such individuals located by radio telemetry within thirty minutes of each other were measured in ArcGIS 9. Distances between randomly selected locations for each pair were al so measured. Averaged distances were subtracted (average real av erage random distance) for each pair to generate an estimate of exclusivity. Positive differ ences are interpreted to indicate a pairs avoidance of each other, while negative differ ences indicate an affinity. Effects of treatments on this m easure were examined using a two-way ANOVA implemented in PROC GLM in SAS. Independent variables included in these models were type of pairing (male/male, female/female, male /female), feeding treatment, predation treatment, and interactions of type of pairing with feeding and predation treatments 29

PAGE 30

(two-way interactions only). Fire was not included as a treatment due to small post-fire sample sizes. Results CMR Analyses Over 26 trapping sessions in eight trappi ng plots, 2557 individual cotton rats (6815 total captures) were trapped. The multista te analysis results show six models with a QAICc < 2, none with particularl y strong support over the others (Table 2-6). It is clear from the top ranked m odels that fire and supplem ental feeding effects were important factors affecting survival. Fire effects appear in the top 30 models and these hold 100% of the model weight of the set. Food effects appear in the top 14 models and these carry 92.5% of the model set s weight. There is no evidence that supplemental food or predator exclusion tr eatments influenced The lack of clear support for any particular model indicates model selection uncertainty; therefore, model averaging was employed for parameter estimation. Overall model averaged survival estimate s showed that males had lower survival than females and that reproductive individuals had lower survival than non-reproductive individuals (Figure 2-1). Model averaged esti mates show large post-fire declines in survival for both sexes and strata in bot h the predator exclos ures and controls (Figure 2-1). Post-fire surv ival was not greatly impacted by the addition of food and still approached zero. During non-fire periods, f ood supplementation increased survival for both sexes and reproductive condit ions (Figure 2-1). Survival was greater in predator exclosures compared to controls, but this di fference was marginal. Similarly, predator exclusion conveyed some benefit to survival post-fire, but this was small in magnitude (Figure 2-1). 30

PAGE 31

Model averaged estimates of show that a greater proportion of males made transitions into reproductive states than females in all se asons (Figure 2-2). Additionally, most reproductive individuals that achieved a r eproductive state stayed in a reproductive state; this trend was slightly gr eater for males than females (Figure 2-2). Initial investigation indicated a str ong fire effect on transitions between reproductive states: models that included three classes of breeding seasons (peaks in spring and summer, non-peak breeding in falls and winters without burns, and a second non-peak in falls and winters with burns ; hereafter peak, non-peak/non-burn, and non-peak/burn, respectively) were clearly better supported than models with only two classes of breeding seasons (peak and non-peak with no distinguishing between burn and non-burn years, Table 21). Two-season breeding season models had no support (weight = 0.0) compared to three-s eason models (Table 2-1, model 7). Model averaged parameter estimates indica ted that transitions to reproductive states were at their highest during peak breeding seasons but that there was only a small drop in transitions to reproductive states during nonpeak/non-fire seasons (Figure 2-2). However, transitions to r eproductive states dropped considerably during non-peak/fire seasons (Figure 2-2). Addi tion of food increased transitions to reproductive states while predator exclusi on had a minimal effect on this parameter (Figure 2-2). The repeated measures ANOVA examin ing treatment effects on abundance indicated significant effects of feeding and fire treatments and their interaction ( P = 0.001, P < 0.001, and P = 0.045 respectively). Exam ination of least square means showed that supplemental feeding increased abundances by 1.9 times and burning 31

PAGE 32

caused a 3 fold decline in abundance. Although the interaction of feeding and burning was marginally sign ificant, burning in feeding areas still produced large declines in abundance (by 2.9 times) indicati ng that this interaction is not biologically significant. Analyses of Radio Telemetry Data A total of 279 cotton rats were collar ed during this study. Of these, 145 had sufficient locations for home range and exclusivity analysis; 204 met criteria for survival analysis. The average number of locations per home range ( SE) was 29.83 1.29 (range 15-92). Male home ranges were signifi cantly larger than female home ranges ( P < 0.001). Average home range size was 2948 m2 (range 685-9814 m2) for female rats (N = 72) using MCP and 5983 m2 (range 888-20917 m2) using fixed kernel methods. For males (N = 73) th e average home ranges were 7891.5 m2 (range 150-30590 m2) and 15845.31 m2 (range 511-84248 m2) using MCP and kernel methods. Although kernel estimates were substantiall y larger than MCP estimates, results from the two methods did not differ qu alitatively. Overall, we believe the 95% MCP estimates provide a more accurate picture of actual space use, and include kernel estimates here for comparison purposes; therefore, the discu ssion will primarily focus on results based on the MCP home range estimates. Collared rat survival Of the 204 rats that met criteria for surviv al analysis, 29 were censored for at least some time during which they were not tracked (slipped/chewed off collar, or experienced transmitter failure ), but were recollared and reentered into the analysis at some later point. Sixty-tw o rats were censored and never reentered into analysis. Causes for censoring included slipping/chewing off collars (N = 21), emigrating from the control/exclosure plot where collared (N = 13), transmitter failure (N = 2), and 32

PAGE 33

unexplained loss of the signal (N = 26). The latter coul d be attributed to transmitter failure, emigration, or carrying off of the rat and/or collar by a far-ranging predator. There was strong evidence for seasonal and fi re effects on survival of collared rats (Table 2-7). Models including only spring, summe r, and fall effects (models 8, 9, and 11, Table 2-7) show that spring, summer, and fall survivals were similar (had AICc values within a range of 2) while survivals during winters where burning occurred and during winters where burning did not occur differ ed from each other and from other seasons (AICc values >2 from other seasonal model s). Parameter esti mates from the top ranked model (Table 2-7, model 1; survival varying by season), show that winter survival in non-burn years was greater than other seasons, while survival in winters of burn years was quite low (Figure 2-3). There was poor support for supplemental feeding and predation treatment effects on collared rat survival; the highest ranked model with a treatment term has moderate support ( AICc = 4.14, Table 2-7, model 3)). Th is model indicated an interactive effect of season with the predator excl usion treatment. Parameter estimates from this model show that survival was similar between cont rols and exc losures for all seasons except winters of burn years, during which time survival was lower in predator controls than exclosures (Figure 3-3). Effects on home range size The only significant treatment effect on home ranges estimated by MCP was for male rats in the predator treatment (Table 2-8). Examination of least square means showed that males in the predator exclosures had larger home ranges than males in the predator controls ( P = 0.001). Neither feeding nor inte ractions of feeding and predator 33

PAGE 34

treatments had signif icant effects on home range size. Similar results were found when examining kernel home range estimates. There was no evidence that feeding or predator exclusio n treatments, or their interaction, had significant effects on female home ranges estimated by MCP (Table 2-8). Similar tests performed using kernel estimates showed a marginally signific ant predation effect (again, larger home ranges in exclosures, P = 0.048), while feeding and the interaction of feeding and pr edation treatments showed no significant effects. Effects on home range exclusivity Three hundred and fifty-six pairs of rats had adjacent or overlapping home ranges during the same time period. Analysis of home range exclusivity suggested no effect of type of pair or treatment on spacing between rats (Tables 2-9 and 2-10). Effects of fire on radio collared rats Thirty-three collared rats were colla red and tracked during the February 2009 burns. All rats survived the fire itself by sheltering in holes within their home ranges or nearby areas that did not burn completely. Forty-one percent of t he rats died due to predation, 34% emigrated to small, unburned patches within the larger burn area and 19% emigrated to unburned areas completely outside of the burn unit. All but one rat either died or emigrated within se ven days of the fire (this rat died twelve days post-fire). Predation and feeding treatm ents did not significant ly affect response to fire (P > 0.05). Of the remaining two rats, one chewed its colla r off in a hole in the burn area; the other stayed within the burn area (an unfed predator exclos ure) and apparently died due to starvation. This rat was found near t he entrance of a hole within its home range seven days post-fire. A necropsy revealed that the rat had lost 19% of her body weight 34

PAGE 35

since she was last trapped (34 days prior). There were no signs of trauma commonly apparent following mammalian, avian, or s nake predation/attempted predation. The stomach contained primarily ash and dirt. Two rats killed and cached by mammalian predators in a fed/predator a ccess grid during the same period were also necropsied to reveal stomachs filled with rabbit chow, i ndicating the individuals in fed grids used supplemental food following the fire. Every rat that moved to an unburned pat ch within the larger burn area moved less than 50 m. These unburned patches ov erlapped or were adjacent to the rats pre-fire home ranges; it is likely the rats were already familiar with the unburned patches they invaded. Six of these rats surv ived and were captured during the next trapping period. Five had lost weight during this interval. The mean percent weight loss was 0.083 % over 33 days. To determine if this was an artifact of the winter season itself, weight changes of collared rats over the 2008 winter were also calculated. Of 9 collared rats which were captured in both winter 2008 sessions (35 days between sessions), the mean weight change was a gain of 0.082 %. Change in weight differed ( P = 0.003) between winter of 2008 and 2009. Of rats that successfully emigrated to areas outside of the burn un it, most did so in a single night and moved distanc es of 50 to 700 m. None appeared to have moved the shortest distance to the burn edge and/or stopped immediately upon reaching an unburned area. Most of these rats (all but two) died or went missing within two weeks of emigrating. 35

PAGE 36

Discussion Fire Effects Of the three treatments exper imentally applied to cotton rat populations in this study (mammalian predator exclusion, supplemental feeding, and prescribed fire), fire had the largest impact on cotton rat populations Prescribed burning caused precipitous declines in survival, abundance, and transitions to reproductive stat es regardless of the presence of supplemental food or absence of mammalian predators. These results support the hypothesis that cotton rat decli nes following fires are due primarily to predation, secondarily to emigration, and not due to changes in food resource availability. However, one radio collared rat apparently di ed of starvation following a burn and other rats that rema ined in small, unburned patches in the overall burn unit lost a significant amount of weight. This suggests that the loss of herbaceous food sources by burning was indeed a problem for this species, but that the crisis of food resources was overwhelmed by increased exposur e to predators due to loss of cover. This is not surprising given the cott on rats heavy cover requirements and general susceptibility to predation; cotton rats support a wide vari ety of predators including snake, mammalian, and avian predators. Predat ion is by far the most common cause of death (82% of deaths, Derrick 2007, and here, 76% of collared rat deaths overall) and cotton rat populations can experience near complete turnover in as little as five to eight months (Goertz 1964). These results suggest that in ecosystems where fires are frequent, such as longleaf pine, cotton rat populat ions are heavily influenced by fire events. Similar sharp post-fire declines in cotton rat abundances have been observed in southern pine forests, native tallgrass prairies, and sacat on grasslands (Arata 1959, Layne 1974, Bock 36

PAGE 37

and Bock 1978, Rehmeier et al. 2005). Al though cotton rats experience a short-term negative fire effect, over the long term cotton rats have a posit ive association with fire. Populations in tallgrass prairies in Kansas peaked in autumns of the first two years following s pring burns but declined in autumns thereafter if burns were not repeated (Rehmeier et al. 2005). Rehmeier et al. (2005) hypothesize that this occurs because fires enhance growth of plants that serve as food resources and reduce litter that may inhibit movement through vegetation. Predation and Supplemental Feeding Effects Given that population dynam ics of species such as cotton rats with rapid maturation and turnover are more sensitive to changes in reproductive parameters than to changes in survival (Heppell et al. 2000, Oli and Dobson 2003), it is somewhat surprising that we found no strong evidence of predator or feeding treatment effects on reproductive transitions but did observe increased survival in supplemental feeding plots. This may be explained by canalization of vital rates which have great proportional impact on population growth rate ; such rates tend to have little variation due to heavy selective pressure on those parameters (P fister 1998) and are unl ikely to be greatly affected by environmental changes. Cotton rats are extraordinary reproducers even when compared with other rat species. Cotton rats become reproductive within one to two months, require 27 days for gestation, and a female may become pregnant again within 24 hours of giving birth (Whitaker and Hamilton 1998). Young open eyes within 24 hours, wean at five to six days, and achieve independence soon after. By contrast, Norway rats ( Rattus norvegicus ) open eyes at 14 to 17 days, wean at three weeks and become independent at four weeks (Whitaker and Hamilton 1998). Given the cotton rats already accelerated 37

PAGE 38

schedule of development and reproduction, it is difficult to imagine there is much room for improvement. A similar study on effe cts of supplemental feeding on cotton rats found that feeding increased the number and we ight of young born in feeding plots but did not affect juvenile survival or recruitment (Campbell and Slade 1995). The lack of predation effects observ ed here is also surprising given the enormous role predation plays in cotton rat mortality. It is po ssible that mammalian predator exclusion alone was insufficient to elicit a response in the parameters examined. Previous studies examining pre dator exclusion or removal on cotton rats (Weigert 1972, Guthery and Beas om 1977) suggest that the ef fects of such treatments vary according to the predators excluded or removed. Guthery and Beasom (1977) removed only mammalian predators from areas where cotton ra ts occurred, and detected no change in survival or density. Weigert (1972) excluded all predators from some study areas and only mammalian pre dators from others and determined that avian predators had a greater impact on cotton rat populations than mammalian predators. These results ar e consistent with our own wh ich suggest that predation by raptors and snakes make up for losses when mammalian predators are excluded. Alternatively, it is possibl e that predation does not regul ate cotton rat populations in areas or periods where cover is sufficient. This hypothesis is supported by the positive supplemental feeding effect (increased survival, abundance, and transitions to reproductive states) observed during non-fire periods. Although we did not observe strong predat ion effects on survival, abundance, or reproductive transitions, predator exclusi on was associated with increased male home range sizes. This indicates a sub-lethal pred ation effect. The best studied examples of 38

PAGE 39

sub-lethal effects of predati on deal with the interplay of food acquisition and pr edation risk. Foraging increases exposure to predator s, causing individuals to make trade-offs between the need to eat and the need to mi nimize predation risk (McNamara and Houston 1987, Abrams 1991). The lack of feeding effects on home range size suggests that cotton rats do not make such tradeoffs, at least with respect to mammalian predation. Instead, we found that male rats made trade-o ffs associated with predator exclusion alone. How can this be explained outside of a food context? Given the previously stated importance of reproduction over survival in species that have rapid turnover and maturation (Oli and Dobson 2003) it follows that male rats, who are promiscuous, range widely, and have no involvement in raising young would make predation risk decisions based on maximizing reproductive opportunities rather than food acquisition. That the same trend wa s not observed with female rats can be explained by the likelihood that a female rat will be bred regardless of whether she encounters a single male or several. While female rats are less likely to influence reproduction by changing space use, male ra ts ought to increase fitness by mating with as many females as possible. Maintenance of larger home ranges should increase the chances of encountering females and the chances of breeding. However, when predation pressure is high, maintaining a large home range may increase predation risk causing male rats to restrict movements. Although we did not detect a strong predat ion treatment effect on survival of trapped rats, we found that radio collared rats had greater survival in controls than exclosures in winters during which bur ning occurred (compared to winters where burning did not occur, during which survival was similar between predator controls and 39

PAGE 40

exclosures) This may indicate additional su pport for sub-lethal effects of predation, but we believe it to be due instead to site effect s. Specifically, one section of the unfed predator control which contained collared rats did not burn completely. Six collared rats had home ranges adjacent to this area and, by moving into the area, which provided cover, were less vulnerable to predation. Gi ven that these six rats made up a significant portion of rats in predator cont rol treatments during the burn (N = 18), it is possible that this skewed the results relating to post-fire predation treatment ef fects on collared rat survival. If it were not fo r this, we believe these results would have been similar to those observed from the CMR analyses: fire caused a decline in survival and that decline was not affected by t he predation treatment. Howeve r, these results indicate that the negative short-term fire effect can be mitigat ed if unburned refugia remain (although rats seemed unable to reliably find su ch refugia if it occurred at distances greater than 50 to 70 m from their home ran ges). Home Range Exclusivity Although our analysis failed to detect a tr eatment effect on cotton rat home range exclusivity, it is interesti ng that previous studies have f ound that female cotton rats have more exclusive home ranges than males and that heavier males have more exclusive ranges than smaller males (Fleharty and Ma res 1973, Cameron and Spencer 1985). We found no significant differences in exclus ivities between male and female rats or between rats in any treatment. The incons istency in these results may be due to different methodologies used to quantify home range exclusivit y. Cameron and Spencer (1985) estimated overlap for co-occ urring rats tracked by radio telemetry around sun rise and sunset by calculating percent overlap of MCP home ranges. Fleharty and Mares (1973) also examined ov erlap and distances between centers of 40

PAGE 41

activity of individuals, although these esti mates were based on home ranges generated with as few as six trapping locations. We used a different method to evaluate the tolerance of cotton rats for each other. Locations were taken by radio tele metry at all times between dawn and dusk, and exclusivity was evaluated by comparing distances between co-occurring individuals tracked at the same time to random distanc es between those individuals. Given the social system of cotton rats, this may provide a more accurate means of evaluating interactions between rats. Liu (1971) found in a large-scale laboratory study that cotton rat populations consist of dominant and subordi nate individuals (defined by whether an individual wins or loses fights submissi ve behavior was not observed). Subordinate individuals lived within home ranges of dominant individuals but minimized encounters by foraging at less desirable times; dom inant individuals foraged around dawn and dusk while subordinates foraged during day or ni ght hours. Subordinat e rats defended only areas immediately around their nests. Dominant rats tended not to have overlapping home ranges with other dominant rats as encounters between two dominants generally ended with the death of one rat or the other. Mating pairs shared nests at times, although they did not forage together, and fe males moved to new nests which were defended even from her mate shortly before giving birth and while nursing young. Since cotton rats use extensively ov erlapping areas, measuring home range overlap provides limited insight into territo riality, especially for home ranges generated from points collected at times when only a s ubset of the population is likely to have been active. The exclusivity measure used here allows an indirect examination of 41

PAGE 42

42 tolerance of cotton rats for each other in a system where individuals use the same space but are solitary and agonistic towards one another. Other studies have found that rodents decrease agonistic behavior when provided supplemental food, but not with respect to predation pressure (Desy et al. 1990). We were unable to detect changes in tolerance of rats towards each other with predation or feeding treatments. It is possible that the food provided wa s insufficient or too greatly dispersed to allow a decrease in aggression and that the presence of avian and snake predators prevented behavioral responses at this level to mammalian predator exclusion. Conclusions Our results suggest that cotton rat populat ion dynamics in our study site are primarily driven by fire events. Populat ion responses following fires appear to be strongly influenced by fire-caused loss of cover and associated increases in predation. Direct effects relating to mammalian predatio n do not appear to be strong, but there is evidence that male cotton rats respond adaptiv ely to predation risk, by decreasing home range size in areas where mammalian predato rs have access compared to areas where they do not. Food is also important to co tton rats, and caused increases in all of the demographic parameters considered here. This is likely to be true in both fire and non-fire periods, but food effect s were overwhelmed by predati on effects following fires. We detected no treatment effects on cott on rat behavior associated with spacing between individuals.

PAGE 43

Table 2-1. Model comparison table for multistate capture-mark-recapture analysis assessing occurrence of peak breeding se asons of cotton rats in general (Model set 1) and with respect to winter prescribed burns (Model set 2) in Southwestern Georgia between 2005 and 2 009. All models had s urvival set as S (reproductive condition*sex) and capture probability set as p (year+season). Table includes number of parameters (K), model weights (relative likelihood of models in the set), and difference in Akaikes information criterion corrected for small sample size after quasilikelihood adjustment ( QAICc). Quasilikelihood adjustment s were made using an estimated of 1.262. Model no. Model K QAICc Model weight 1 (spring and summer peaks) 15 0.001.00 Model (spring peaks) 15 80.92 0.00 set 1* 2 (summer and fall peaks) 15 103.07 0.00 3 (summer peaks) 15 110.02 0.00 4 (dot) 14 148.42 0.00 5 6 (spring, fall, and summer peaks) 15 150.39 0.00 7 (spring and summer peaks; winter and fall non-peaks divided by years and non-burn years) 16 0.00 1.00 Model set 2** (spring and summer peaks; winter and fall non-peaks) 15 74.08 0.00 8 Model set 1 considers only two classe s of breeding season: peak (listed) and non-peak (all seasons not listed for a given model). ** Model set 2 compares the best model fr om set 1 (model number 1) with a similar model that considers three classes of breeding seasons by breaking down non-peak seasons into those that occurred during burn years (winters and falls of 2005, 2007, and 2009; burns occurred in winters of these ye ars) and those that occurred during non-burn years (winters and falls of 2006 and 2008). 43

PAGE 44

Table 2-2. Model comparison table for multistate capture-mark-recapture analysis assessing sex, reproductive condition, breeding season and time effects on capture probability ( p ), survival ( S ), and rates of transitions between reproductive states ( ) in cotton rats in Southwestern Georgia between 2005 and 2009. See Table 2-1 for column definitions. Model no. Model K QAICc M odel weight Effects on capture probability ( p ) 1 p (year+season) 18 0.001.00 2 p (season) 14 20.790.00 3 p (year) 14 83.260.00 4 p (constant) 10 124.660.00 Effects on survival (S )** 5 S (reproductive condition* sex) 34 0.001.00 6 S (sex) 32 11.180.00 7 S (reproductive condition+sex) 33 13.100.00 8 S (constant) 31 21.630.00 9 S (reproductive condition) 32 23.000.00 Effects on rates of transiti ons to reproductive states*** 10 (breeding season+reproductive condition+sex) 34 0.00 1.00 11 (breeding season+reproductive condition) 33 16.72 0.00 12 (breeding season) 32 95.93 0.00 13 (breeding season+sex) 33 102.04 0.00 14 (constant) 30 324.08 0.00 Additional parameters modeled as S (reproductive condition*sex) (breeding season+ reproductive condition+sex). ** Additional parameters modeled as p (session) (breeding season+reproductive condition+ sex). ***Additional parameters modeled as S (reproductive condtion*sex) p (session). 44

PAGE 45

Table 2-3. Model comparison table for multistate capture-mark-recapture analysis assessing the term of effect of win ter prescribed burns on survival ( S ) of cotton rats in Southwestern Georgi a between 2005 and 2009. See Table 2-1 for column definitions. All model s had additional parameters of capture probability ( p ) modeled as p (year+season) and rate of transition between reproductive states ( ) modeled as (reproductive condition+sex+breeding season). Model no. Model K QAICc Model weight 1 S (fire effect over 5 (2009) to 10 weeks (2005 and 2007))* 16 0.00 0.98 2 S (fire effect over 30 weeks) 16 7.85 0.02 3 S (fire effect over 10 weeks) 1615.72 0.00 4 S (fire effect over 20 weeks) 1621.32 0.00 Model 1 shows a range of intervals becaus e the interval between trapping sessions changed from 10 weeks (2005 and 2007) to 5 weeks (2009) before the 2009 prescribed burn. 45

PAGE 46

Table 2-4. Model comparison table for multistate capture-mark-recapture analysis assessing potential for paired site effects on survival ( S ) and rate of transitions to reproductive states ( ) of cotton rats in Southwestern Georgia between 2005 and 2009. See Table 2-1 fo r column definitions. All models had capture probability ( p ) modeled as p (year+season). Model no. Model K QAICc Model weight 1 S (reproductive condition*sex+site) (breeding season+reproductive condition+ sex+site) 24 0.000.84 2 S (reproductive condition*sex+site) (breeding season+reproductive condition+sex) 21 3.260.16 3 S (reproductive condition*sex) (breeding season+ reproductive condition+sex+site) 21 21.700.00 4 S (reproductive condition*sex) (breeding season+ reproductive condition+sex) 18 25.080.00 46

PAGE 47

47 Table 2-5. Model comparison table for r obust design capture-mark-recapture analysis assessing demographic and ti me effects on abundance ( N ), capture probability ( p ), and recapture probability ( c) in cotton rats in Southwestern Georgia between 2005 and 2009. See Tabl e 2-1 for column definitions. Survival (S ) was modeled as (S(site+sex+food+fire) for all models. Emigration terms ( and ) were modeled with a random emigration effect for all models: (.)= (.). Model no. Model K AICc Model weight Effects on abundance (N )* 1 N (site) 64 0.001.00 2 N (.) 61 119.800.00 Effects on capture ( p ) and recapture ( c) probabilities** 3 p (session) c(session) 64 0.00 1.00 4 p (session)c( p + c)*** 39 35.02 0.00 5 p (.)c( p + c )**** 14 538941.84 0.00 *Additional parameter s were modeled as p (session) c(session). **Additional parameter was modeled as N (site). ***Indicates capture probability varies by session with a constant trap happy response recapture response ( c). ****Indicates a constant capture probability with a constant trap-happy recapture response ( c).

PAGE 48

Table 2-6. Model comparison table for multistate capture-mark-recapture analysis examining the effect of predation, supplemental feeding, and fire treatments on survival ( S ) and transition probabilities ( ,between reproductive and non-reproductive states) of cotton rats in Southw estern Georgia between 2005 and 2009. Capture probability modeled was modeled as p(year+season) for all models. See Table 2-1 fo r column definitions. Bolded text indicates treatment effect s (all other effects are similar between models throughout the set). Only models with a QAICc < 4 are shown here (the top ranked 12 models of 55 in the overall set). Model no. Model K QAICc Model weight S (reproductive condition*sex+site+ food*fire ) (breeding season+reproductive condition+site+sex) 27 0.00 0.16 1 S (reproductive condition*sex+site+ food*fire ) (breeding season+reproductive condition+site+sex+ food ) 28 0.11 0.16 2 S (reproductive condition*sex+site+ food+predation+fire ) (breeding season+reproductive condition+site+sex) 27 0.93 0.10 3 S (reproductive condition*sex+site+ food+predation+fire ) (breeding season+reproductive condition+site+sex+ food ) 28 1.05 0.10 4 S (reproductive condition*sex+site+ food+fire ) (breeding season+reproductive condition+site+sex) 26 1.79 0.07 5 6 S(reproductive condition*sex+site+ food+fire ) (breeding season+reproductive condition+site+sex+ food ) 27 1.90 0.06 S (reproductive condition*sex+site+ food*fire ) (breeding season+reproductive condition+site+sex+ predation ) 28 2.03 0.06 7 S (reproductive condition*sex+site+ food*fire ) (breeding season+reproductive condition+site+sex+ food+predation ) 29 2.13 0.06 8 S (reproductive condition*sex+site+ food+predation+fire ) (breeding season+reproductive condition+site+sex+ predation ) 28 2.96 0.04 9 S (reproductive condition*sex+site+ food+predation+fire ) (breeding season+reproductive condition+site+sex+ food+predation ) 29 3.06 0.04 10 S (reproductive condition*sex+site+ food+fire ) (breeding season+reproductive condition+site+sex+ predation ) 27 3.82 0.03 11 S (reproductive condition*sex+si te+ food+fire ) (breeding season+reproductive condition+site+sex+ food+predation ) 29 3.91 0.02 12 48

PAGE 49

Table 2-7. Factors influencing survival of ra dio collared cotton rats in sites treated with supplemental feeding, winter prescr ibed fires, and mammalian predator exclusion in southwes tern Georgia from June 2007 August 2009. Models and associated AICc rankings and weights for collared rat survival were estimated using Cox propor tional hazard models. See Table 2-1 for column definitions. Winter (non-bur n) refers to the winter of 2008 during which no sites were burned. | notation indica tes that all additive and interaction combinations of variables are included in the model. | notation indicates that all additive and interaction combinat ions of variables are included in the model. Model no. Model K AICc Model weight 1 Season 5 0.00 0.57 2 Winter (burn)* 2 1.21 0.31 3 Predation|season 12 4.14 0.07 4 Food|season 12 6.67 0.02 5 Winter (non-burn) 2 7.88 0.01 6 Sex|season 12 9.56 0.01 7 Constant survival 1 13.65 0.00 8 Spring 2 13.72 0.00 9 Summer 2 14.74 0.00 10 Predation 2 14.84 0.00 11 Fall 2 14.92 0.00 12 Food 2 15.24 0.00 13 Sex 2 15.58 0.00 14 Food|sex 4 17.42 0.00 15 Predation|sex 4 17.84 0.00 16 Predation|food 4 18.23 0.00 17 Global model 27 26.24 0.00 Winter (burn) refers to the winter of 2009 during which all sites were treated with prescribed fire. ** Winter (non-burn) refers to the winter of 2008 during which no sites were burned. 49

PAGE 50

Table 2-8. Factors influencin g home range size of cotton rats in southwestern Georgia using 95% minimum convex polygon (MC P) and 95% fixed kernel (Kernel) estimates. General linear model results are given with interactive effects. Degrees of freedom ( d.f .), mean square (MS), F -statistic values ( F ), and signific ance level ( P ) are given for each effect. Home ranges were log transformed for this analysis. Method Male rats Source d.f. MS F P 1 0.055 0.40 0.559 MCP Kernel Food Predation Food*Predation Food Predation Food*Predation 1 1.768 11.020 0.001 1 0.008 0.050 0.822 1 0.023 0.17 0.685 1 2.284 16.85 <0.001 1 0.003 0.02 0.879 Female rats MCP Kernel Food Predation Food*Predation Food Predation Food*Predation 1 0.023 0.29 0.592 1 0.147 1.85 0.178 1 0.111 1.39 0.242 1 0.036 0.44 0.509 1 0.329 4.04 0.048 1 0.161 1.98 0.164 50

PAGE 51

Table 2-9. Home range exclusivities for co tton rats in sites treated with supplemental feeding and mammalian predator exclusi on in southwes tern Georgia from June 2007 to August 2009. Pair type Treatment N Mean difference (m)* Female/Female Feeding Non-feeding Predator access Predator exclosure 9 21 16 14 -1.80.08 -4.22.23 -5.14.69 -1.62.84 Male/Male Feeding Non-feeding Predator access Predator exclosure 32 25 27 30 1.45.30 -2.24.46 1.00.79 -1.21.23 Male/Female Feeding Non-feeding Predator access Predator exclosure 42 49 49 42 -2.50.59 2.05.85 -0.83.86 0.87.65 Exclusivity was estimated by finding actual distances between rats with adjacent minimum convex polygon home ranges that lived during the same time on days that they were located within 30 minutes of each other by radio telemetry. Random distances between each pair were also found. Mean difference is the mean of actual distances random distances (SE). 51

PAGE 52

52 Table 2-10. Factors influencing home range excl usivities of cotton ra ts in sites treated with supplemental feeding and mammalian predator exclusion in southwestern Georgia from June 2007 to August 2009. General linear model results are given with interactive effect s. Pair type refers to whether pairs were female/female, male/m ale, or male/female. S ee Table 2-4 for column definitions. Source d.f. MS F P Pair type 2 82.190 0.48 0.620 Food 1 9.289 0.05 0.816 Predation 1 17.267 0.10 0.751 Type*Food 2 347.416 2.03 0.135 Type*Predation 2 120.604 0.70 0.496

PAGE 53

A Non-reproductive females Co n1 Con 3 Con 2 Co n4 Ex 1 Ex 3 Ex 2 Ex 4 Survival 0.0 0.1 0.2 0.3 0.4 0.5 B Non-reproductive males C on1 C on3 C on2 C on4 Ex 1 Ex 3 Ex 2 Ex 4 0.0 0.1 0.2 0.3 0.4 0.5 C Reproductive females Co n1 Co n 3 Co n2 Co n4Ex1Ex 3 Ex2Ex4Survival 0.0 0.1 0.2 0.3 0.4 D Reproductive males Con1Con3Con2Con4Ex1Ex3Ex2Ex4 0.0 0.1 0.2 0.3 0.4 No fire/No food Food/No fire Food/No fire Food/Fire Figure 2-1. Model averaged estimates of survival of cotton rats in southwestern Georgia between 2005 and 2009 in response to prescribed fire, supplem ental feeding, and predator control treatments. Estimates are given for non-reproductive (A and B) and reproductive (C and D) male and female rats. Surviv al is estimated over 10 week intervals. Estimates are given by site: Ex sites refer to areas treat ed with mammalian predator exclusion while Con sites refer to areas where mamm alian predators were allo wed access. Supplemental feeding treatments were added to Con and Ex sites 2 and 4 from summer 2007 through 2009. All sites were burned during the winters of 2005, 2007 and 2009. 53

PAGE 54

54 A Females: N to R Con1Con3Con2Con4Ex1Ex3Ex2Ex4Transition rate 0.0 0.2 0.4 0.6 0.8 B Males: N to R Con1Con3Con2Con4Ex1Ex3Ex2Ex4 0.0 0.2 0.4 0.6 0.8 C Females: R to R Con1Con3Con2Con4Ex1Ex3Ex2Ex4Transition rate 0.2 0.4 0.6 0.8 1.0 D Males: R to R Con1Con3Con2Con4Ex1Ex3Ex2Ex4 0.2 0.4 0.6 0.8 1.0 Peak/No food Non-peak/No fire/No food Non-peak/Fire/No food Peak/Food Non-peak/No fire/Food Non-peak/Fire/Food Figure 2-2. Model averaged estimates of t he rates of transitions to reproductive states for male and female cotton rats during peak breeding seasons (spring and summer), non-peak seasons during which burning did not occur, and non-peak seasons during which burning did occur, in southwestern Georiga between 2005 and 2009. Transitions include movement of individuals from non-reproductive to reproductive states (N to R; A and B), and reproductive individuals staying in a r eproductive state (R to R; C and D). Transitions occurred over 10 week intervals. Estimates are given by si te: Ex sites refer to areas treat ed with mammalian predator exclusion and Con refers to mammalian predator access areas sites. Supplemental f ood was added to Con and Ex sites 2 and 4 from summer 2007 through 2009. All sites were burned during the winters of 2005, 2007 and 2009.

PAGE 55

A B Figure 23. Survival estimates ( standar d error) of radio collared cotton rats in southwestern Georgia between 2 007 and 2009, generated using Cox proportional hazard models. Estimate s for part A were generated using a model where survival varied by season (T able 7, model 1). Estimates for part B were generated using a model where survival varied by season and predation treatment (Table 7, model 3) Estimates are given by season and site: Ex sites refer to areas tr eated with mammalian predator exclusion while Con sites refer to areas wher e mammalian predators were allowed access. Winter (burn) refers to the winter of 2009 during which all sites underwent a prescribed burn. Winter (non-burn) refers to the winter 2008 during which prescribed burns were not carried out. 55

PAGE 56

CHAPTER 3 EFFECTS OF SUPPLEMENTAL FEEDING, MAMMALIAN PREDATOR EXCLUSION, AND PRESCRIBED FIRE ON COTTON AND OLDFIELD MOUSE POPULATIONS IN A LONGLEAF PINE ECOSYSTEM Introduction Factors which limit populations are commonly of interest to ecol ogists. Access to food resources and predation are common causes of limitation. Effects relating to food and predation are often examined separately, but there is a great deal of evidence from theoretical and field studies that these factors interact and should be considered simultaneously (Abrams 1983, McNamara and Houston 1987, Krebs et al. 1995, Hubbs and Boonstra 1997, 1998). Access to food res ources is important for reproduction and to avoid starvation, but foraging behaviors often increase predation risk. Individuals should seek a balance that minimizes predat ion risk while maximizing food intake. Such trade-offs have obvious implications for survival and abundance and may impact other vital rates, such as reproduction as we ll (Lima and Dill 1989). Studies seeking to understand dynamics of prey sp ecies will be benefited by invest igating the roles of food resources and predation individually and in combination on multiple vital rates. For the target species in this study, cotton and oldfield mice ( Peromyscus gossypinus and P. polionotus respectively), a third fact or may interact with food and predation: prescribed fire. Prescribed fi re is a common management tool in longleaf pine, southern pine, and Fl orida scrub ecosystems in which both species occur (Whitaker and Hamilton 1998). Over the short term, burning simultaneously consumes food resources and reduces cover which increas es exposure to predators. Over the long term, burning maintains open habitat, r educes occurrence of hardwood trees and shrubs, and improves vegetative growth (Brockway and Lewis 1997). Prescribed fire 56

PAGE 57

benefits these species over the long term, although the exact form of the response may depend on frequency of fire application (Masters et al. 2002, 2007, Suazo et al. 2009). Most previ ous studies examining fire e ffects on small mammals focused on short term (less than one year) effects on abundance. These have shown cotton mice to respond to fire either neutrally or with immediat e but temporary population spikes in burned areas (Shadowen 1963, Hatchell 1964, Layne 1974, Suazo et al. 2009). Oldfield mice do not appear to have a strong shor t-term fire response (Arata 1959, Odum et al. 1973, Suazo et al. 2009). We know of no studies that have attempted to experimentally determine which fire-related changes (loss of food resources or loss of cover) are responsible for the observ ed population-level effects, and few that have examined effects on a broader range of populat ion parameters such as survival and reproduction. The objective of this study was to ex perimentally examine the effects of supplemental feeding and mammalian pr edation on cotton and oldfield mouse populations. We were secondarily interested in determining the roles of these factors in changes in cover and food availability caused by prescribed burning. This was accomplished by establishing a large scale factorial experiment with mammalian predator exclusion and supplemental feeding treatments conducted over four and half years. Plots were burned three times over the course of the experiment. Methods Study Site and Species This research was conducted at the Joseph W. Jones Ecological Research Center at Ichauway in Baker County, Georgia. Ichauway is a 12,000 ha property consisting primarily of longleaf pine ( Pinus palustris ) and wiregrass ( Aristida beyrichiana ) ecosystem. Longleaf pine ecosystems are char acterized by a low-density longleaf pine 57

PAGE 58

over-story, a diverse, herbaceous groundcover, and an open, park-like mid-story (Van Lear et al. 2005). Hardwood tree species occur at limited levels. Frequent, low intensity fires are key ecological processes. Conseque ntly, frequent application of prescribed fire is a primary management tool throughout Ichauway; most sites are burned on a two year rotation (Atkinson et al. 1996). Cotton and oldfield mice are common across the Southeastern US. The semi-arboreal cotton mouse prefers bottoml and hardwood forests, but the species is a habitat generalist (Whitaker and Hamilton 1998). Downed woody debris is an important microhabitat component for this species (McCay 2000). Oldfield mice prefer dry, open fields with loose soils and beaches. This species is noted for its monogamous breeding habits (Whitaker and Hamilton 1998). Field Methods In 2002, the Jones Center constructed four mammalian predator exclosures, each paired with a nearby control wit h similar habitat. Plots range in size from 35.94 to 49.09 ha. Exclosures are surrounded by 1.2 m tall woven wire fences which carry electrified lines along the top, middle, and bottom to discourage mammals from climbing over or digging under (the weave is lar ge enough to allow small mammals and snakes to pass through). Although mammalian predator s occasionally enter exclosures, regular monitoring by track counts and thermal camera surveys indicate significantly fewer mammalian predators in exclosures t han controls (Conner et al. 2010). Each control and exclosure contained a 12x12 small mammal trapping grid with 15 m spacing between stations. Twenty-f our elevated trapping stations were also interspersed throughout each grid, attached to trees at heights of about 1.5 to 2 m. Pairs of grids were trapped four times per year (once each season) from January 2005 58

PAGE 59

through June 2007 and eight times per year (twic e per season) from July 2007 through the June 2009 using Sherman live traps (H.B Sherman Traps, Tallahassee, Florida, USA). A small amount of a granular insecticide was sprinkled ar ound each trap to prevent deaths due to fire ants. New captures were marked individually with metal ear tags. Data recorded for all captures incl uded location, species, sex, weight, age (adult or juvenile, based on weight), reproductive condition (for males, testes descended or not, for females, if pregnant and/or la ctating), and hind foot measurement. In June of 2007, two exclosure and two control grids were randomly selected to receive a supplemental feeding treatment consisting of placing 113 g (4 oz) of commercial rabbit chow in cans at every ot her station on the trapping grids. Food was replaced every other week. Empty cans were also placed in the non-feeding grids. This treatment continued through August 2009. Images from trail cameras demonstrated that cotton mice, ol dfield mice, cotton rats ( Sigmodon hispidus ), house mice (Mus musculus ), woodrats ( Neotoma floridana ), flying squirrels (Glaucomys volans ), and eastern cottontails ( Sylvilagus floridanus ) regularly used feeding stations. We found no evidence that cans were def ended by individuals of any species. In February of 2005, 2007, and 2009, all plots were burned according to Ichauways burn plan which has these study areas on a two year burn rotation. Trapping methods followed recommendations of the American Societ y of Mammalogists (Gannon et al. 2007) and were approved by the Univ ersity of Florida Institutional Animal Care and Use Committee. Statistical Methods Data considered for this analysis includes capture-mark-recapt ure (CMR) data for cotton and oldfield mice trapped between January 2005 and June 2009 (26 sessions). 59

PAGE 60

Analyses were carried out using the R 2.9. 1 (R Development Core Team) package RMark (Laake and Rexstad 2008) to build models for program MARK (White and Burnham 1999). Multistate CMR models were used to esti mate and model state specific survival ( S ), capture probability ( p ), and transitions between reproductive states ( ). States used for S and were based on reproductive condition. Males were considered to be in reproductive condition if testes were desce nded, females if pregnant and/or lactating. Therefore, models ev aluating effects on S which include a reproductive condition term estimate and model survival separate ly for reproductive and non-reproductive individuals, and models evaluating effects on which include a reproductive condition term estimate and model probabilities of indi viduals moving between reproductive states (i.e., rates of non-reproductive individuals entering repr oductive states, rates of reproductive individuals remaining reproductive). Preliminary analyses consider ed the potential influence of session, season, and year on p Influence of reproductive condition and sex was assessed for S and Breeding season was also considered for Assessment of effects on p S and was carried out in a sequential fa shion. First, effects on p were considered while modeling S and using the most general models for each described above. Effects on S and were then considered in a similar fashion. Assessment of goodness-of-fit wa s carried out using the median approach in program MARK (White and Burnham 1999). The median test indicated a mild overdispersion ( = 1.321 for cotton mice and = 1.339 for oldfield mice). Models in each parameters set were compared using Ak aikes information cr iterion corrected for 60

PAGE 61

small sample size (AICc), after quasi-lik elihood adjustments (QAICc) made using = 1.321 for cotton mice and 1.339 for oldfie ld mice. Models wer e considered well supported if they had a QAICc of less than two. T he best supported model within each parameters set was selected a base for modeling that param eter in further analyses. These analyses indicated t hat reproductive condition and sex, modeled in an additive fashion, were impor tant factors for describing S in cotton mice ( S (reproductive condition+sex), Table 3-3 model 5), while reproductive condition was important for oldfield mice ( S (reproductive condition), Table 3-4, model 5). For both species, capture probability was best described as fully time varying ( p (session)), Table 3-3, model 1 (cotton mice); Table 3-4, model 1 (oldfield mice)). To model we investigated approach to mode ling a breeding season effect on Breeding seasons vary over the geographic range of cotton and oldfield mice (Wolfe and Linzey 1977, Whitaker and Hamilton 19 98). Throughout most of the range, breeding may occur year round but with peaks at certain times of the year. Because a literature review did not give a clear indica tion of when peak breeding seasons for these species occur in southwestern Georgia, we created model set, based on the literature (Wolfe and Lindzey 1977, Whitaker and Ham ilton 1998, Suazo et al. 2009) and personal observations, to identify when peak breeding seasons occurred. This analysis indicated, for cotton mice, breeding peaks in fall and early win ter (Table 3-1, model 1), and, for oldfield mice, peaks in wint er and summer (Table 3-2, model 1). Because of potential confounding effects due to prescribed burning treatments and breeding seasons occurring as occasion-dependant effects, we then assessed whether 61

PAGE 62

there was evidence to support further divisi on of cotton mouse non-breeding seasons into whether a burn occurred during these s easons or not. Since the burns occurred during the oldfield mouse peak breeding season, a similar di vis ion was made, but with respect to peak breeding seasons rather than non-peak season s. Further division of breeding seasons based on burning was well supported (Table 3-1, model 7 for cotton mice, and Table 3-2, model 5 for oldfield mice). Using these breeding season models, we continued the sequential variable selection for as described above for p and S Reproductive condition, sex, and breeding season, modeled in an additive fa shion, were important for modeling in cotton mice ( (reproductive condition+ sex+breeding seasons), Table 3-3, model 10), while reproductive condition and breeding season, also modeled in an additive fashion, were important for oldfield mice ( (reproductive condition+breedi ng season), Table 3-4, model 10). Although the prescribed fires occurred at sp ecific times, fire-caused changes in cover and food resources may last for weeks or months. To determine the best effect window for the fire treatments, a set of models considering fire effects on survival over multiple time intervals was considered. Su rvival was constrained to be similar between all trapping periods except those following fires. Post-fire survival was allowed to vary for several different intervals, from incl uding only the interval during which the fire occurred (interval length of ten weeks fo llowing the 2005 and 200 7 fires and of five weeks following the 2009 fire), to including intervals through the summer season (30 weeks), by which time vegetation is typica lly recovered. For both species no single model had overwhelming support over the ot hers (Table 3-5). The top ranked model 62

PAGE 63

was chosen to represent the fire effect on survival in subsequent analys is. For cotton mice, the highest ranking model indicated a s hort term fire effect on survival with declines occurring over a period of ten weeks (Table 3-5, model 1). For oldfield mice, the best supported model indicate d an effect lasting thirty weeks (Table 3-5, model 5). Exclosure and control sites were initia lly selected as pairs based on similar habitats between pairs. As part of a post-hoc examination of non-treatment effects on S and we examined the potential for paired site effects on these parameters. Using the best models indicated by the analyses described above, we ran a second set of models considering paired site effects on S and This analysis indicated paired site effects were important in modeling both S and for cotton and oldfield mice (Table 3-6, model 1 (cotton mice) and model 5 (oldfield mice)). Treatment effects were added to the best base model (for cotton mice: S (reproductive condition+sex+site) p (session) (breeding season+reproductive condition+sex+ site); for oldfield mice: S (reproductive condition+site) p (session) (breeding season+reproductive condition+sit e)) as additive and interactive effects (two-way only). Due to confounding effect s relating to both fire and breeding season occurring as occasion-dependant effects, only supplemental feeding and predation treatments were consi dered with respect to while feeding, predation, and fire effects were considered with respect to S Model averaging was employed to generate parameter estimates for S and Abundance estimates (N ) were also generated for both species. Pollocks robust design (Pollock 1982) was used for cotton mice. Due to computational difficulties, it 63

PAGE 64

was not possible to use the robust design fo r oldfield mice; the POPAN model was used instead (Schwarz and Arnason 1996). Robust design models estimate probabilities fo r survival ( S ), capture ( p ), recapture ( c), emigration (), and staying away after emigration ( ). The variable selection approach used for the robust design was similar to that used for the multistate analysis. Preliminary investigati on considered potential for time and sex effects on p and c. Paired site effects were considered for N S was modeled using the best S model from the multistate analysis (without reproductive condition; S (sex+site+ fire*predation)). terms were modeled using a random emigration effect ( (.)= (.)). Preliminary investigations indicated that cotton mouse capture/recapture models were best supported when model ed with a capture probability that varied by session and allowed a constant trap happy response. Sex was also important for modeling capture/recapture probabilities ( p (session+sex) c( p + c) where c is the constant trap happy response, Table 3-7, model 3). Pa ired site effects were important for N (Table 3-7, model 1). Because of the difficulties associ ated with modeling treatment effects on abundance directly (White 2002), the best robust design model indicated by the preliminary analysis described above, S (sex+site+fire*predation) (.)= (.) p (session+sex) c( p + c) N (site), was used to generate derived abundance estimates by site and session, but not to assess treatment effects. Treatment effects on abundance were evaluated using a repeated measur es ANOVA (Schabenberger and Pierce 2002) implemented using the PROC MIXED procedure in SAS (SAS Institute Inc. 2004). The variables considered in this ANOVA incl uded food, fire, and pr edation treatments and 64

PAGE 65

their interactions (two-way interactions onl y). Paired sites were included as a random effect. Multiple covariance structures were investigated and the best (variance components structure, which allows a different variance for each random effect) was selected ba sed on AICc value (Miller et al. 20 04). Treatment effects were considered significant at the = 0.05 level. The POPAN models used for oldfield mice estimate apparent survival ( ), capture probability ( p ), entry probability ( pent ), and population size (N ). Due to constraints associated with the POPAN model, the data set was divided by paired sites. For each of these paired sites, was modeled using the best survival model from the multistate analysis (minus the reproductive condition term; (predation)). The sequential variable selection process for N p and pent followed as described previously. A site effect was considered with respect to N Effects of year, burn year, and season were considered with respect to p and pent To pick the best common model for all site pairs, each candidate model in each parameters candidate model set was ranked by AICc score. Ranks were summed ac ross sites for each parameter and the model with the lowest score was sele cted as the best. This investigation determined that p and pent were best modeled across sites using an additive effect between year and season while N was best modeled varying by site (Table 3-8). The resulting model (predation) p (year+season) pent (year+season) N (site) was run for all site pairs to generate derived abu ndance estimates for each site by trapping session. Treatment effects on abundance were investigated using a repeated measures ANOVA in PROC MIXED in SA S as described above for cotton mice. 65

PAGE 66

Results Cotton Mice A total of 2108 individual cotton mice (8428 total captures) was trapped over 26 trapping sessions in eight trapping plots. The best supported multistate model suggested an interactive effect of predation and fire treatments on survival with an interactive effect of feeding and predation on (Table 3-9, model 1). Although this model was the best supported (the second ranked model has a QAICc > 2) it did not carry a great deal of weight (0.393). However, it was clear from the top ranked models that the feeding interacted with predation to affect This interaction appeared in the top ten models, and models with this interacti on held 91.1% of the we ight of the overall model set. Support for treatm ent effects on survival was less clear. The second best supported model (model 2, Table 3-9) in cluded no treatment effect on survival, indicating poor support for treatment effect s other than the interactive predation*fire effect on survival. The lack of substantial support for any par ticular model indicates model selection uncertainty; therefore, model averaging was employed for parameter estimation. Overall model averaged survival estimates showed that males had lower survival than females and that reproductive individuals had higher survival than non-reproductive individuals (Figure 3-1). Model averaged estimates indicated that, in predator access grids, burning had essentially no effect on su rvival. In predator exclosures, however, survival increased dramatically following fi res (Figure 3-1). During non-fire periods, survival was slightly greater in predator a ccess grids than in the exclosures, but this trend was reversed following bur ns (Figure 3-1). 66

PAGE 67

The addition of food had minimal impac t on survival regardless of whether predators had access or not, or whether an area had been recently burned or not (Figure 3-1). Model averaged parameter estimates for showed that a greater proportion of males made transitions to reproductive states than females in all seasons (Figure 3-2). Additionally, most reproductive individuals that achieved a r eproductive state stayed in a reproductive state; this trend was slightly gr eater for males than females (Figure 3-2). Initial investigation indicated a st rong fire effect on transitions between reproductive states: models that included three classes of breeding seasons (peak breeding in fall and early winter, non-peak breeding in springs, summers, and late winters without burns, and a second non-peak in springs, summers, and late winters of burn years; hereafter, peak, non-peak/non-bu rn, and non-peak/burn respectively) had greater support than models with only tw o breeding seasons (peak and non-peak with no distinguishing between bur n and non-burn years). Tw o-season models had no support (weight = 0.0) compared to threeseason models (Table 3-1, model 7). Model averaged parameter estimates indica ted that transitions to reproductive states were at their highest during peak br eeding seasons and that there was a small drop in transitions to reproductive states during non-peak/non-fire s easons (Figure 3-2). Transitions to reproductive states dropped considerably more during non-breeding/fire seasons (Figure 3-2). Predat or exclusion and feeding alone caused small decreases in reproductive transitions and the combination of these treatments was associated with an increase in transitions to reproductive states (Figure 3-2). 67

PAGE 68

The repeated measures ANOVA examin ing treatment effects on abundance indica ted a significant effect of feeding on abundance ( P < 0.001). Examination of least square means indicated that feeding plots contained 1.8x the number of cotton mice as unfed plots. No other treatm ents or their interactions significantly affected abundance ( P > 0.05). Oldfield Mice A total of 1203 individual oldfield mice (4828 total captures) was trapped over 26 trapping sessions in eight trapping plots. Th ere was no clear top multistate model for oldfield mice as there were six models with a QAICc < 2 and none of these carried much weight (Table 3-10). However, it was clear from the t op ranked models that predation was an important fact or affecting survival. Predation effects appeared in the top twenty models and these models collective ly held a weight of 82.8%. Models including feeding also had decent support in the model set, holding 67% of the weight of the overall set. The model set showed li mited support for treatment effects on The lack of clear support for any particular model indicated model selection uncertainty; therefore, model aver aging was employed for parameter estimation. Model averaged survival estimates indicated that non-reproductive individuals had lower survival than reproductive individual s (Figure 3-3). Model averaged survival estimates also showed increased survival in predator exclusio n plots compared to predator access plots. This was true in both preand post-fire periods. Following prescribed fires, survival decreased slightly in predator access grids and increased (by a slightly greater magnitude) in predator exclosure treatments (Figure 3-3). Addition of food was associated with declines in survival in both predator access and exclosure grids, and in both preand postfire periods. The magnitude of the decline was greater 68

PAGE 69

in exclosur es than in controls, but the m agnitude of the change was not great in either case. Initial investigation found strong fire effects on transitions between reproductive states: models that included three classes of breeding seasons (bimodal peak breeding in winters and summers, distinguishing between burn years and non-burn years, and non-peak breeding in falls and springs) had be tter support than models that contained only two breeding seasons (peak and non peak with no distinguishing between burn and non-burn peak seasons). Two season m odels had no support when compared to three season models (weight = 0.0, Table 3-2, model 5). Model averaged parameter estimates for indicate that a greater proportion of reproductive individuals that achieved reproduc tive states stayed in reproductive states (Figure 3-4). Transitions to reproductive states were the greates t during peak breeding seasons of non-burn years. Transitions into breeding states dropped during non-peak seasons. However, during winters and summers of burn years, transitions to breeding states dropped dramatically su ch that transitions during t hese seasons were below even that of non-breeding seasons, indicating a strong fire effect on reproduction in oldfield mice (Figure 3-4). Predator exclusion was associated with smaller proportions of individuals entering breeding states, whether food was present or not although this difference was minimal. Supplemental feeding was associated with a larger increase in transitions to reproductive states in bot h predator access and exclusio n areas (Figure 3-4). The repeated measures AN OVA examining treatment effects on abundance indicated significant effects of predation and feeding treatments and the interaction of 69

PAGE 70

these treatments on abundance ( P = 0.001, P < 0.001, and P = 0.001 respectively). Examination of least square means showed that feeding increased abundances by 2.7x, predator exclusion increased abundances by 2.7x and the application of both treatments simultaneous ly increased abundances by 7.6x. Discussio n Although cotton and oldfield mice are closely related species which occur in many of the same habitats, they were affected in different ways by the application of mammalian predator exclusion, supplemental feeding, and prescribed fire treatments. However, both species showed a surprising trend of increased survival among reproductive individuals compared to non-reproduc tive individuals. We hypothesize that this may be because non-reproductive adults are likely to be younger individuals. Although there is evidence that reproduc tion exacts a survival cost among small mammals (Koivula et al. 2003), it may be that among these mice, the cost of being young is greater. Juveniles and young adults tend to be transient while seeking to establish home ranges (Bigler and Jenkins 1975) and dispersal behavior is associated with reduced survival (Van Vuren and Armi tage 1994). However, this behavior may be adaptive in general if disper sed individuals improve reproduction by doing so (Van Vuren and Armitage 1994). Among cotton mice, initial analyses ident ified sex as an impor tant factor with respect to both survival and reproductive trans itions, but among oldfield mice sex was irrelevant. It is possible that this di fference occurs because oldfield mice are monogamous and form long-term pair bonds whil e cotton mice are promiscuous (Blair 1951). 70

PAGE 71

Treatment Effects on Cotton Mice Cotton mice showed different treatment effe cts on different vital rates, some in interactive and unexpected ways. It is difficult to tie these e ffects into a clear picture of how predation, food resources, and fire affe ct cotton mouse populations overall, but interpretation is aided by consideration of ec ological theory concerning adaptive prey behavior in response to predation risk and fo od availability. This theory has been largely developed by Abrams (1983, 1984, 1991, 1992 a 1992 b 1993) and is based on the observation that while access to sufficient food is required for reproduction and to avoid starvation, foraging incr eases predation risk. Individua ls must make trade-offs between foraging and predation risk in such a wa y as to maximize food intake (and by extension, reproduction) while minimizing mo rtality. A large body of theoretical and experimental evidence suggests such tradeoffs are common (reviewed in Lima and Dill 1989) and may be stronger than direct consumptiv e effects of predation (Pressier et al. 2005). Although a simple concept at the core the effects of thes e adaptive behaviors may be complex and counterin tuitive (Abrams 1991, 1992a 1992 b 1993). This is especially likely in systems with complex food webs and competitive interactions between multiple predator and prey species. Effects of trade-offs may also depend on a species life history and may change with breeding/non-breedi ng seasons (Abrams 1991). The only strong treatment effect on cotton mouse survival was an increase in survival with the combination of predator ex clusion and fire. There was no strong fire effect on survival in predator controls, and survival was similar between controls and exclosures during non-fire periods. This s uggests fire conveys a benefit that is not realized when predators have access to the burned area. This in turn suggests an 71

PAGE 72

awareness of increased predation risk with loss of cover, a behavioral respons e of cotton mice in the predator controls, and a choice to balance predation risk against taking advantage of this benefit. Differences in predation and asso ciated trade-offs may contribute to the mixed shortterm fire responses (neutral or beneficial) observed with cotton mice in previous studies (Shadowen 1963, Hatchell 1964, Layne 1974, Suazo et al. 2009). The general lack of a supplemental feeding effe ct on survival suggests that food resources are not important in this fire response, unless the food supplementation was insufficient or access to food cans was c onsidered risky. Other mechanisms that may have caused to the combined benefit of fire and predator exclusion remain unclear. It is possible that change in abundances of othe r species in the burned area may have contributed to this response. For example, cotton rats, usually among the most abundant species in the study areas, dec lined to near 0 following burns. The combination of a lack of a fire effect on abundance and a decline in post-fire breeding supports a behavioral response to increased predation risk associated with post-fire conditions. Adaptive anti-predatory behavior may mitigate negative fire effects on survival and abundance, but a trade-of f appears to occur with a decline in reproduction. Because fires occurred during the non-breeding season, it is reasonable that cotton mice would make a trade off favoring survival at the cost of reproduction. Population growth rates of species such as rodents which mature rapidly and reproduce often are most sensitive to changes in reproduction (Heppell et al. 2000, Oli and Dobson 2003), and in most cases ought to favor strategies that maxi mize reproduction rather than survival. However, during the non-breeding season t hese species should 72

PAGE 73

maximize fitness by behaving in a similar manner as semelparous species (Abrams 1991). This entails minimizing predation risk by reducing foraging. That feeding had no apparent effect on post-fire survival and r eproductive transitions may be explained by the observation that for semelparous s pecies, or iteroparous species in the non-breeding season, increases in food resource s should, somewhat counterintuitively, be associated with a decrease in foraging effort to maximize survival and overall fitness (Abrams 1991). Increased predation should cause prey spec ies to make trade-offs between growth rate (positively associated with food inta ke) and predation risk (negatively associated with food intake) when prey species (like cotton mice) achieve reproductive maturity at a specific size rather than by reaching a specific age (Abrams and Rowe 1996). This trade-off should cause an increase in predation risk to be associated with an increase in the age at maturity due to decreased grow th rates (Abrams and Rowe 1996). This relationship may have contributed to the dec line in reproduction following fires; if juvenile development was delayed, a decline in reproductive transitions should be observed. However, increases in food res ources may counteract the direct effect of predation on growth rate (Abrams and Rowe 1996). This may explain why breeding transitions were positively affected by the interaction of predator exclusion and food supplementation. Food addition was associated with a nearly two-fold increase in abundance. It is not surprising that food addition caused increases in abundance, but given the relationship between food resources and the abili ty to achieve reproductive status in small mammals (Cameron and Eshelman 1996), it is interesting that feeding was not 73

PAGE 74

also associated with increases in transitions to reproductive states (except in predator exclusion plots). This suggests that t he increased abundances wer e due to either increases in juvenile survival, in the number of young produced per reproductive event or immigration to feeding plots. Such e ffects have been observed to occur in response to supplemental feeding in previous studies with small mammals (Boutin 1990, Hubbs and Boonstra 1997); unfortunately, we were unable to address these factors in the current study. Treatment Effect s on Oldfield Mice The strongest treatment effect on oldfie ld mice was the predator exclusion treatment; this was associated with in creased survival and abundance. Feeding and the interaction of feeding and predator exclusion were also associated with increased abundances. On the surface, these results seem more intuitive than the cotton mouse results and could be interpreted to suggest oldf ield mice are more influenced by direct consumptive effects of predation. Pressier and Bolnick (2008) observed in a review of non-consumptive effects of predation that non-consumptive effects seemed to dominate some predator-prey relationships while appearing to be weak or non-existent in others. However, Peckarsky et al. (2008) note that it is impossible to conclude that changes in prey survival and abundance are due solely to direct effects of predation as stresses and trade-offs associated with predation risk can cause declines in survival and abundance even when prey are not actually at risk (as by experiment al manipulation of a predators ability to kill). Such indirect effects may be overlooked when they take the predicted form and direction of direct consumpt ive effects, and they may at times cause effects as strong, if not str onger, than direct consumptive effe cts (Pressier et al. 2005). Behavioral data can help to distinguish betw een direct and indirect effects, and indeed 74

PAGE 75

may be necessary to do so (Abrams 1995), but our study design did not include such observations. Similar to the results observed with cotton mice, feeding was associated with increases in abundance but with only minim al increases in reproductive transitions, indicating the feeding effect on abundance ma y be due to increases in immigration, juvenile survival, or the number of young produced per reproductive event. Predation had a minimal effect on r eproductive transitions, but fire caused declines in transitions to reproductive stat es. Given the import ance of predation on other population parameters, it might be expected that there would be an interaction of predation and fire treatments since burning removes cover and should increase risk of predation. It is possible that increased pr edation due to loss of cover was not a problem following fires because this species already prefers open areas (Whitaker and Hamilton 1998). This of course fails to explain wh y burning caused declines in reproductive transitions, and why addition of food was insuffici ent to prevent such declines. Although the decline in reproductive transit ions following fires is similar to that of cotton mice, an important difference exists between the responses in these s pecies. Cotton mice were in a non-peak breeding season when the burns occurred while oldfield mice were in a peak season. In this situation, oldfie ld mice should be expected to maximize reproduction during this season, even at the expense of survival (Abrams 1991). That they were unable to do so (assuming reproducti ve output is associated with transitioning to a reproductive state) suggests that either the theory is inco rrect or that oldfield mice were limited in some way despite food additi on. Given necessity of sufficient food quality and quantity for small mammals to achi eve reproductive status (Cameron and 75

PAGE 76

Eshelman 1996), and the oldfield mouses general dietary preference for insects and seeds over herbaceous material, it is possible that the provided food was insufficient to allow these mice to maintain normal br eeding following a winter burn. Although supplemental food was available following fire s, other preferred food sources may have become limiting. For exampl e, Odum et al. (1973) obser ved declines in arthropod abundances following a winter burn in Georgia for several months. Conclusions Although cotton and oldfield mice are closely related species that occur in similar habitats, feeding, fire, and pr edation treatments affected th ese species differently. Cotton mice appear to make trade-offs with res pect to predation risk. These appear to be especially important following fire events, implying cotton mice are, over the short term, negatively affected by the loss of cove r associated with burning. An exception seems to occur when mammalian predators ar e excluded. Oldfield mice also experience significant effects of predation but it is less clear if these effects are related directly or indirectly to consumption itsel f, or merely the risk of being consumed. Fire effects are less apparent for oldfield mice although reproduction in oldfield mice is negatively affected by fires. The retroductive conclusions relating to behavioral responses are in many ways speculative as we lack behavioral data that would help to confirm these effects. Studies incorporating behavioral components such as home range size, microhabitat use, giving up densities for foraging animals, etc. could better address these issues. 76

PAGE 77

Table 3-1. Model comparison table for multistate capture-mark-recapture analysis assessing occurrence of peak breeding seasons of cotton mice in general (Model set 1) and with respect to winter prescribed burns (Model set 2) in Southwestern Georgia between 2005 and 2 009. All models had survival set as S(reproductive condition+sex) and c apture probability set as p(session). Table includes number of parameters (K ), model weights (relative likelihood of models in the set), and difference in Akaikes information criterion corrected for small sample size after quasilikelihood adjustment ( QAICc). Quasilikelihood adjustments were made using an estimated of 1.339. Model no. Model K QAICc Model weight 1 (fall and early winter peaks) 30 0.00 1.00 Model set 1* 2 (winter, spring and fall peaks) 30 24.82 0.00 3 (fall peak s) 30 32.17 0.00 4 (winter peaks) 30 64.67 0.00 5 (fall and winter peaks) 30 72.68 0.00 6 (constant) 29 75.55 0.00 7 (fall and early winter peaks; late winter, spring, summer non-peaks divided by burn years and non-burn years ) 31 0.00 1.00 Model set 2** 8 (fall and early winter peaks; late winter, spring, summer non-peaks) 30 18.70 0.00 Model set 1 considers only two classes of breeding season: peak (listed) and non-peak (all seasons not listed for a given model). ** Model set 2 compares the best model from set 1 (model number 1) with a similar model that considers three classes of breeding seasons by breaking down non-peak seasons into those that occurred during burn years (late winters, springs, and summers of 2005, 2007, and 2009; burns occurred in midwinters of these y ears) and those that occurred during non-burn years (late winters, springs, and summers of 2006 and 2008). 77

PAGE 78

Table 3-2. Model comparison table for multistate capture-mark-recapture analysis assessing occurrence of peak breeding se asons of oldfied mice in general (Model set 1) and with respect to winter prescribed burns (Model set 2) in Southwestern Georgia between 2005 and 2 009. All models had s urvival set as S (reproductive condition) and capture probability set as p (session). See Table 3-1 for column definitions. Model no. Model K QAICc Model weight 1 (winter and summer winter peaks) 29 0.00 0.59 Model set 1* 2 (spring peaks) 29 2.56 0.16 3 (no peaks) 28 2.73 0.15 4 (fall peaks) 29 3.49 0.10 5 (winter and summer peaks with peak seasons divided by burn years and nonburn years ) 30 0.00 1.00 Model set 2** 6 (winter and summer peaks) 29 41.07 0.00 Model set 1 considers only two classe s of breeding season: peak (listed) and non-peak (all seasons not listed for a given model). ** Model set 2 compares the best model fr om set 1 (model number 1) with a similar model that considers thr ee classes of breeding seasons by breaking down the peak seasons into those that occurred during burn years (winters and summers of 2005, 2007, and 2009; burns occurred in winters of these years) and those that occurred during non-burn years (winters and summers of 2006 and 2008). 78

PAGE 79

Table 3-3. Model comparison table for multistate capture-mark-recapture analysis assessing sex, reproductive condition, breeding season, and time effects on capture probability ( p ), survival ( S ), and transitions rates between reproductive states ( ) in cotton mice in Sout hwestern Georgia between 2005 and 2009. See Table 3-1 for column definitions. K QAICc Model weight Model no. Effects on capture probability ( p ) 1 p (session) 340.00 0.71 2 p (season) 141.84 0.29 3 p (constant) 1012.84 0.00 4 p (years) 1413.65 0.00 Effects on survival (S )** 5 S (reproductive conditi on*sex) 34 0.00 0.48 6 S (reproductive conditi on+sex) 33 0.47 0.38 7 S (sex) 32 3.83 0.07 8 S (reproductive condition) 32 4.14 0.06 9 S (constant) 31 6.44 0.02 Effects on rates of reproductive transitions*** 10 (breeding season+reproductive condition+sex) 33 0.00 0.61 11 (breeding season+reproductive condition) 32 0.93 0.39 12 (breeding season+sex) 32 245.32 0.00 13 (breeding season) 31 254.10 0.00 14 (constant) 29 348.35 0.00 *Additional param eters modeled as S (reproductive condition*sex) (breeding season+ reproductive condition+sex). **Additional parameters modeled as p (session) (breeding season+reproductive condition+sex). ***Additional parameters modeled as S (reproductive condtion*sex) p (session). 79

PAGE 80

Table 3-4. Model comparison table for multistate capture-mark-recapture analysis assessing sex, reproductive condition, breeding season, and time effects on capture probability ( p ), survival ( S ), and transitions rates between reproductive states ( ) in oldfield mice in So uthwestern Georgia between 2005 and 2009. See Table 3-1 for column definitions. Model K QAICc Model weig ht Model no. Effects on capture probability ( p ) 1 p (session) 34 0.00 0.84 2 p (season) 14 3.31 0.16 3 p (years) 14 22.05 0.00 4 p (constant) 10 23.60 0.00 Effects on survival (S )** 5 S (reproductive condition) 32 0.00 0.43 6 S (reproductive conditi on+sex) 33 0.18 0.40 7 S (reproductive conditi on*sex) 34 1.98 0.16 8 S (sex) 32 8.32 0.01 9 S (constant) 31 8.55 0.01 Effects on rates of reproductive transitions ( )*** 10 (breeding season+reproductive condition) 33 0.00 0.54 11 (breeding season+reproductive condition+sex) 34 0.36 0.46 12 (breeding season+sex) 33 51.21 0.00 13 (breeding season) 32 51.34 0.00 14 (constant) 30 93.94 0.00 *Additional param eters modeled as S (reproductive condition*sex) (breeding season+ reproductive condition+sex). **Additional parameters modeled as p (session) (breeding season+reproductive condition+sex). ***Additional parameters modeled as S (reproductive condtion*sex) p (session). 80

PAGE 81

81 Table 3-5. Model comparison table for multistate capture-mark-recapture analysis assessing term of effect of winter prescribed burns on survival of cotton and oldfield mice in Southwestern G eorgia between 2005 and 2009. See Table 3-1 for column definitions. All models had additional parameters of capture probability ( p ) modeled as p (session) and rate of transition between reproductive states ( ) modeled as (reproductive condition+sex+breeding season) (for cotton mice) or (reproductive condition+breeding season) (for oldfield mice). Species Model no. Model K QAICc Model weight 1 S (fire effect over 10 weeks) 32 0.00 0.32 2 S (fire effect over 20 weeks) 32 0.35 0.27 3 S (fire effect over 5 (2009) to 10 weeks (2005 and 2007))* 32 0.81 0.21 Cotton mice 4 S (fire effect over 30 weeks) 32 0.85 0.21 5 S (fire effect over 20 weeks) 32 0.00 0.28 6 S (fire effect over 30 weeks) 32 0.25 0.24 7 S (fire effect over 10 weeks) 32 0.26 0.24 Oldfield mice 8 S (fire effect over 5 (2009) to 10 weeks (2005 and 2007))* 32 0.31 0.24 Models 3 and 8 show a range of intervals because the interval between trapping sessions changed from 10 weeks (2005 and 2007) to 5 weeks (2009) before the 2009 prescribed burn.

PAGE 82

Table 3-6. Model comparison table for multistate capture-mark-rec apture analysis assessing potential for site effects on survival ( S ) and rate of transitions to reproductive states ( ) of cotton and oldfield mice in Southwestern Georgia between 2005 and 2009. See Tabl e 3-1 for column definitions. All models had capture probability ( p ) modeled as p (session). Species Model K QAICc Model weight Model no. 1 S (reproductive condition+sex+site*) (breeding season+reproductive condition+sex+site) 390.000.68 Cotton Mice 2 S (reproductive condition+sex+site) (breeding season+reproductive condition+ sex) 36 1.49 0.32 3 S (reproductive condition+sex) (breeding season+reproductive condition+sex+site) 36 19.99 0.00 4 S (reproductive condition+sex) (breeding season+reproductive condition+sex) 33 21.63 0.00 5 Oldfield Mice S (reproductive condition+site) (breeding season+reproductive condition+site) 38 0.00 0.68 6 S (reproductive condition) (breeding season+reproductive condition+site) 35 1.53 0.32 7 S (reproductive condition+site) (breeding season+reproductive condition) 35 16.43 0.00 8 S (reproductive condition) (breeding season+reproductive condition) 32 17.80 0.00 *Site effects are paired site effects. This effect pairs each site treated with mammalian predator exclusion with a predator access site with similar habitat. 82

PAGE 83

Table 3-7. Model comparison table for r obust design capture-mark-recapture analysis assessing demographic and ti me effects on abundance ( N ), capture probability ( p ), and recapture probability ( c ) in cotton mice in Southwestern Georgia between 2005 and 2009. See Tabl e 3-1 for column definitions. Survival (S ) was modeled as S(site+sex+fire*predation) for all models. Emigration terms ( and ) were modeled with a random emigration effect for all models: (.)= (.). Model K QAICc Model weight Model no. Effects on abundance ( N )* 1 N (site) 41 0.001.00 2 N (.) 38 21.690.00 Effects on capture (p) and recapture (c)** p (session+sex) c( p + c) 41 0.00 1.00 3 p (session) c( p + c) 40 29.03 0.00 4 p (sex) c( p + c) 16 229.00 0.00 5 p (.)c( p + c) 15 246.21 0.00 6 p (.)c( p ) 14 362.08 0.00 7 Additional parameters modeled as p (session+sex) c( p + c). **Additional parameter modeled as N (site). Indicates capture probability varies by sessi on, with an additive effe ct of sex, and with a constant trap happy res ponse recapture response ( c). Indicates capture probability varies by se ssion with a constant trap happy response recapture response ( c). Indicates capture probability varies by sex with a constant trap happy response recapture response ( c). Indicates a constant capture probability with a constant trap-happy recapture response ( c). Indicates constant capture and recapture rates (shared). 83

PAGE 84

Table 3-8. Model comparison table for POPAN capture-mark-recapture analysis assessing site and time effects on abundance ( N ), capture probability ( p ), and entry probability ( pent ) in oldfield mice in Southwestern Georgia between 2005 and 2009. Effects are given separ ately for four paired mammalian predator exclusion and control sites. S ee Table 3-1 for column definitions. Survival ( ) was modeled as S(predation) for all models. Bolded models indicate those selected as the bes t common model for all sites (based on ranking each model by AICc score and summing ranks across sites). Model no. Model K AICc Model weight Effects on abundance (N )* Control/Exclosure 1 1 N (site) 16 0.00 0.88 2 N (constant) 15 4.08 0.12 Control/Exclosure 2 3 N (site) 16 0.00 1.00 4 N (constant) 15 14.42 0.00 Control/Exclosure 3 5 N (site) 16 0.00 0.70 6 N (constant) 15 1.71 0.30 Control/Exclosure 4 7 N (constant) 15 0.00 0.71 8 N (site) 16 1.81 0.29 Effects on capture probability ( p )** Control/Exclosure 1 9 p (year+season) 19 0.00 1.00 10 p (burn year+season) 16 30.74 0.00 11 p (season) 15 43.33 0.00 12 p (burn year) 12 59.00 0.00 13 p (constant) 11 60.36 0.00 14 p (year) 15 60.39 0.00 Control/Exclosure 2 15 p (year+season) 19 0.00 0.99 16 p (year) 15 9.29 0.01 17 p (season) 15 116.85 0.00 18 p (burn year+season) 16 118.94 0.00 19 p (constant) 11 132.72 0.00 20 p (burn year) 12 134.30 0.00 Control/Exclosure 3 21 p (year+season) 19 0.00 0.99 22 p (year) 15 10.00 0.01 23 p (burn year+season) 16 11.73 0.00 24 p (constant) 11 11.93 0.00 25 p (burn year) 12 14.11 0.00 26 p (season) 15 18.09 0.00 84

PAGE 85

85 Table 3-8. Continued Model no. Model K AICc Model weight Control/Exclosure 4 27 p (year) 15 0.00 0.37 28 p (burn year) 12 0.13 0.35 29 p (constant) 11 1.01 0.22 30 p (year+season) 19 4.63 0.04 31 p (season) 15 6.96 0.01 32 p (burn year+season) 16 7.27 0.01 Effects on entry probability (pent ) Control/Exclosure 1 33 pent (year+season) 19 0.00 0.93 34 pent (year) 15 6.52 0.04 35 pent (burn year+season) 16 6.86 0.03 36 pent (season) 15 14.34 0.00 37 pent (burn year) 12 20.30 0.00 38 pent (constant) 11 28.42 0.00 Control/Exclosure 2 39 pent (year+season) 19 0.00 0.99 40 pent (year) 15 9.93 0.01 41 pent (burn year+season) 16 130.58 0.00 42 pent (burn year) 12 136.50 0.00 43 pent (season) 15 161.11 0.00 44 pent (constant) 11 182.66 0.00 Control/Exclosure 3 45 pent (season) 15 0.00 0.60 46 pent (burn year+season) 16 1.06 0.35 47 pent (constant) 11 6.45 0.02 48 pent (burn year) 12 7.57 0.01 49 pent (year+season) 19 9.49 0.01 50 pent (year) 15 15.81 0.00 Control/Ex closure 4 51 pent (year+season) 19 0.00 1.00 52 pent (burn year+season) 16 21.86 0.00 53 pent (season) 15 30.39 0.00 54 pent (year) 15 106.40 0.00 55 pent (constant) 11 138.22 0.00 56 pent (burn year) 12 139.90 0.00 Additional parameters modeled as p (burn year+season) and pent (burn year+season). ** Additional parameters modeled as pent (burn year+season) and N (site). *** Additional parameters modeled as N (site) and p (burn year+season).

PAGE 86

Table 3-9. Model comparison table for multistate capture-mark-recapture analysis examining the effect of predation, feeding, and fire treatments on surviv al (S) and transition probabilities ( between reproductive and non-reproductive states) of cotton mice in southwes tern Georgia, 2005-2009. All models had capture probability set at p(session). See Table 3-1 for column definitions. Bolded text indica tes treatment effects (all other effects are similar between models throughout the set). Table values were adjusted for an estimated c-hat of 1.321. Only models with a model weight > 0. 03 are shown here (the top 8 models of 55 in the overall set). Model K QAICc Model weight Model no. 1 S(reproductive condition+sex+site+predation*fire ) (breeding season+ reproductive condition+sex+site+ food*predation ) 45 0.00 0.39 2 S(reproductive condition +sex+site) (breeding season+ reproductive condition +sex+site+ food*predation ) 42 2.08 0.14 3 S(reproductive condition +sex+site+ fire ) (breeding season+ reproductive condition+sex+site+ food*predation ) 43 2.73 0.10 4 S(reproductive condition +sex+site+ predation ) (breeding season+ reproductive condition+sex+site+ food*predation ) 43 3.81 0.06 5 S(reproductive condition +sex+site+ food ) (breeding season+ reproductive condition+sex+site+ food*predation ) 43 4.07 0.05 6 S(reproductive condition +sex+site+ food*fire ) (breeding season+ reproductive condition+sex+site+ food*predation ) 45 4.22 0.05 7 S(reproductive condition+sex+site+predation+fire ) (breeding season+ reproductive condition+sex+site+ food*predation ) 44 4.39 0.04 8 S(reproductive condition +sex+site+ food+fire ) (breeding season+ reproductive condition+sex+site+ food*predation ) 44 4.66 0.04 86

PAGE 87

Table 3-10. Model comparison table for multistate capture-mark-recapture analysis examining the effect of predation, feedi ng, and fire treatments on survival (S) and transition probabilities ( between reproductive and non-reproductive states) of oldfield mice in southwes tern Georgia, between 2005 and 2009. All models had capture probability set at p( session). See Table 3-1 for column definitions. Bolded text indicates tr eatment effects (all other effects are similar between models throughout the set). Only models with a model weight > 0.03 are shown here (the top 9 models of 55 in the overall set). Model K QAICc Model weight Model no. 1 S(reproductive condition+site+ food*predation ) (breeding season+reproductive condition +site) 400.00 0.14 2 S(reproductive condition+site+ food+predation ) (breeding season+reproductive condition+site) 390.81 0.09 3 S(reproductive condition+site+ food*predation ) (breeding season+reproductive condition+site+ predation ) 411.29 0.07 4 S(reproductive condition+site+ food*predation ) (breeding season+reproductive condition+site+ food ) 411.69 0.06 5 S(reproductive condition+site+ predation*fire ) (breeding season+reproductive condition+site) 401.90 0.05 6 S(reproductive condition+site+ predation ) (breeding season+reproductive condition+site) 381.96 0.05 7 S(reproductive condition +site+ food+predation ) (breeding season+reproductive condition+site+ predation ) 402.10 0.05 8 S(reproductive condition+site+ food+predation ) (breeding season+reproductive condition+site+ predation ) 402.50 0.04 9 S(reproductive condition+site+ food+predation+fire ) (breeding season+reproductive condition +site) 402.66 0.04 87

PAGE 88

A Non-reproductive females Con1Con3Con2Con4Ex1Ex3Ex2Ex4Survival 0.3 0.4 0.5 0.6 B Non-reproductive males Con1Con3Con2Con4Ex1Ex3Ex2Ex4 0.3 0.4 0.5 0.6 C Reproductive females Con1Con3Con2Con4Ex1Ex3Ex2Ex4Survival 0.35 0.40 0.45 0.50 0.55 0.60 0.65 D Reproductive males Con1Con3Con2Con4Ex1Ex3Ex2Ex4 0.3 0.4 0.5 0.6 No fire/No food Fire/No Food Food/No Fire Food/Fire Figure 3-1. Model averaged estimates of survival of cotton mice in southwestern Georgia between 2005 and 2009 in response to prescribed fire, supplem ental feeding, and predator control treatments. Estimates are given for non-reproductive (A and B) and reproductive (C and D) male and female mice. Survival is estimated over 10 week intervals. Estimates are given by site: Ex sites refer to areas treat ed with mammalian predator exclusion and Con refers to mammalian predator access areas sites. Supplemental food was added to Con and Ex sites 2 and 4 from summer 2007 th rough 2009. All sites were burned during the winters of 2005, 2007 and 2009. 88

PAGE 89

89 Figure 3-2. Model averaged estimates of breeding transitions for male and female cotton mice in southwestern Georgia between 2005 and 2009 during p eak breeding seasons (fall and early winter), non-peak seasons during which burning did not occur and non-peak seasons during which burning did occur. Transitions include transition of non-reproductive individuals to reproducti ve states (N to R, A and B) and re productive individuals staying in a reproductive state (R to R; C and D). Tr ansitions occurred over 10 week intervals. Estimates are given by site: Ex refers to areas treated with ma mmalian predator exclusion; Con re fers to mammalian predator access sites. Supplemental food was added to Con and Ex sites 2 and 4 from summer 2007 through 2009. All sites were burned during the winters of 2005, 2007 and 2009. A Females: N to R Con1Con3Con2Con4Ex1Ex3Ex2Ex4Transition rate 0.1 0.2 0.3 0.4 0.5 0.6 0.7 B Males: N to R Con1Con3Con2Con4Ex1Ex3Ex2Ex4 0.2 0.3 0.4 0.5 0.6 0.7 C Females: R to R Con1Con3Con2Con4Ex1Ex3Ex2Ex4Transition rate 0.6 0.7 0.8 0.9 1 0 D Males: R to R Con1Con3Con2Con4Ex1Ex3Ex2Ex4 0.6 0.7 0.8 0.9 1.0 Peak/No food Non-peak/No fire/No food Non-peak/Fire/No food Peak/Food Non-peak/No fire/Food Non-peak/Fire/Food

PAGE 90

A Non-reproductive Con1Con3Con2Con4Ex1Ex3Ex2Ex4Survival 0.2 0.3 0.4 0.5 0.6 B Reproductive Con1Con3Con2Con4Ex1Ex3Ex2Ex4Survival 0.3 0.4 0.5 0.6 0.7 No fire/No food Fire/No food Food/No fire Food/Fire Figure 3-3. Model averaged survival estima tes in southwestern Georgia between 2005 and 2009 in response to presc ribed fire, supplemental feeding, and predator control treatments. Estimates are given for non-reproductive (A) and reproductive (B) individuals. Survival is estimated over 10 week intervals. Estimates are given by site: Ex si tes refer to areas treated with mammalian predator exclusion while Con site s refer to areas where mammalian predators were allowed access. S upplemental feeding treatments were added to Con and Ex sites 2 and 4 from summer 2007 through 2009. All sites were burned during the winters of 2005, 2007 and 2009 90

PAGE 91

91 A N to R Con1Con3Con2Con4Ex1Ex3Ex2Ex4Transition rate 0.2 0.4 0.6 0.8 R to R Con1Con3Con2Con4Ex1Ex3Ex2Ex4Transition rate 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Peak/No fire/No food Peak/Fire/No food Non-peak/No food Peak/No fire/Food Peak/fire/Food Non-peak/Food B Figure 3-4. Model averaged estimates of bre eding transitions for oldfield mice in southwestern Georgia between 2005 and 2009 during peak breeding seasons (winter and summer) in burn and non-burn years and non-peak breeding seasons. Transitions include non-reproductive individuals transitioning to reproductive states (N to R; A) and reproductive individuals staying in a reproductive state (R to R; B). Trans itions are estimated over 10 week intervals. Estimates are given by site: Ex sites refer to areas treated with mammalian predator exclusion while C on sites refer to areas where mammalian predators were allowed a ccess. Supplemental feeding treatments were added to Con and Ex sites 2 and 4 from summer 2007 through 2009. All sites were burned during the winters of 2005, 2007 and 2009.

PAGE 92

CHAPTER 4 CONCLUSIONS Although the three target species in th is study, cotton rats, cotton mice, and oldfield mice occupy similar habitats, each showed different responses to supplemental feeding, predation, and prescri bed fire treatments. Cotton ra t populations were primarily driven by fire events, although food effects were apparent in non-fire periods, and likely would have become important in post-fire periods if predation and emigration had not caused rapid population declines. Although cott on rat mortality and turnover is primarily driven by predation, there was little evidenc e of direct, negative effects of mammalian predation on cotton rat survival and reproduction. However, we detected evidence that cotton rats behave adaptively in response to predation pressure as male rats had smaller home ranges in predator a ccess grids compared to exclosures. Of the three treatments applie d here, oldfield mice were most strongly affected by predation. Predator access grids were associated with smaller abundances and lower survival. The observed effects occurred in th e expected form and direction, indicating a negative effect from mammalian predation. However, lacking behavioral data, it is unclear whether these effects were cons umptive or non-consumptive in nature. Although previous studies hav e demonstrated that oldfie ld mice generally exhibit a short-term neutral response to fire, probabl y due to their preference for areas that already have little cover, we found that prescribed burning had a negative impact on transitions to reproductive states for this species. We believe this may be due to changes in availability of preferred food sources such as insects. Cotton mouse populations saw different treatment effects on the various population parameters. Food addition increased abundances. Survival increased in 92

PAGE 93

93 predator exclusion grids following fires. Reproduction was positively influenced by the interaction of predator exclusion and food supplementation and negatively influenced by burning. These complex responses may resu lt from behavioral responses to predation pressure. For example, fo llowing fires, when predation risk is high, cotton mice appear to make trade-offs in favor of survival at the expense of reproducti on. However, as with oldfield mice, the understanding of this sys tem would be improved if behavioral data were available to supplement the trapping data considered here. There is a need for long-term, large-sca le, replicated experiments to further address questions relating to interactions between predation and food resources, particularly in terrestrial systems. Such experiments would benefit from the incorporation of behavioral data with data regarding vital rates.

PAGE 94

LIST OF REFERENCES Abrams, P. A. 1983. Life history strategies of optimal foragers. Theoretical Population Biology 24:22 38. Abrams, P. A. 1984. Foragi ng time optimization and interactions in food webs. The American Naturalist 124:80-96. Abrams, P. A. 1991. Life history and t he relationship between food availability and foraging effort. Ecology 72:1242-1252. Abrams, P. A. 1992 a Predators that benefit prey and prey that harm predators: Unusual effects of interacting fora ging adaptations. American Naturalist 140:573-600. Abrams, P. A. 1992 b Why dont predators have positiv e effects on prey populations? Evolutionary Ecology 6:449-457. Abrams, P. A. 1993. Why predation rate s should not be proportional to predator density. Ecology 74:726-733. Abrams, P. A., and L. Rowe. 1996. The effects of predation on the age and size of maturity of prey. Evolution 50:1052-1061. Allison, P. D. 1995. Survival ana lysis usi ng SAS: A practical guide. SAS Press, Cary, North Carolina, USA. Arata, A. A. 1959. Ef fects of burning on vegetation and rodent populations in a longleaf pine turkey oak association in nor th central Florida. Quarterly Journal of the Florida Academy of Sciences 22:94-104. Arthur, A. D., R. P. Pech, and C. R. Dickm an. 2004. Habitat st ructure mediates the non-lethal effects of predation on enclosed populations of house mice. Journal of Animal Ecology 73:867-877. Atkinson, J. B., J. Brock, and R. Smith. 1996. Operational longleaf pi ne management at Ichauway. The Longleaf Pine Alliance Annu al Meeting, Mobile, Alabama, USA. Beyer, H. L. 2004. Hawths Analysi s Tools for ArcGIS. Available at http://www.spatialecology.com/htools. Bigler, W. J., and J. H. Jenkins. 1975. Population characteristics of Peromyscus gossypinus and Sigmodon hispidus in tropical hammocks of south Florida. Journal of Mammalogy 56:633-643. 94

PAGE 95

Blair, W. F. 1951. Population structure, soci al behavior, and environmental relations in a natural population of the beach mouse ( Peromyscus polionotus leucocephalus ). Contributions of the Laborat ory of Vertebrate Biology, University of Michigan 48:1-47. Bock, C. E., and J. H. Bock. 1978. Respons e of birds, small mammals, and vegetation to burning sacaton grasslands in southeastern Arizona. Journal of Range Management 31:296-300. Boutin, S. 1990. Food supplementation ex periments with terrestrial vertebrates: Patterns, problems, and the future. Canadian Journal of Zoology 68:203-220. Brockway, D. G., and C. E. Lewis. 1997. Long term effects of dormant-season prescribed fire on plant community diversit y, structure, and productivity in a longleaf pine wiregrass ecosystem. Fo rest Ecology and Management 96:167-183. Byrom, A. E., T. J. Kare ls, C. J. Krebs, and R. Boons tra. 2000. Experimental manipulation of predation and food supply of arctic ground squirrels in the boreal forest. Canadian Journal of Zoology 78:1309-1319. Cameron, G., and B. D. Eshel man. 1996. Growth and reproduc tion of hispid cotton rats ( Sigmodon hispidus ) in response to naturally occurri ng levels of dietary protein. Journal of Mammalogy 77(1):220-231. Cameron, G. N., and S. R. Spencer. 1981. Sigmodon hispidus Mammalian Species 158:1-9. Cameron, G. N., and S. R. Spencer. 1985. Assessment of space-use patterns in the hispid cotton rat (Sigmodon hispidus ). Oecologia 68:133-139. Campbell, M. T., and N. A. Slade 1995. The effect of mate rnal mass on litter size and offspring survival in the hispid cotton rat ( Sigmodon hispidus ). Canadian Journal of Zoology 73:133-140. Clark, C. W., and M. Mangel. 2000. Dynamic state variable models in ecology. Oxford University Press, New York, New York, USA. Conner, L. M., J. R. Rutledge, and L. L. Smith. 2010. E ffects of mesopredators on nest survival of shrub-nesting birds. Journal of Wildlife Management 74:73-80. Cox, D. R. 1972. Regressi on models and life tables. Journal of the Royal Statistical Society, B34:187-220. Derrick, A. M. 2007. Small mammal responses to prescribed fire and simulated mammal removal in a longleaf pine ecosystem. M.S. Thesis, University of Georgia, Athens, Georgia, USA. 95

PAGE 96

Desy, E. A., and G. O. Batzli. 1989. Effects of food availability and predation on prairie vole demography: A field expe riment. Ecology 70:411-421. Desy, E. A., G. O. Batzli, and J. Liu. 1990. Effects of food and predation on behavior of prairie voles: A field experiment. Oikos 58:159-168. Dickman, C. R. 1992. Predation and habi tat shift in the house mouse, Mus domesticus Ecology 73:313-322. ESRI. 2005. ArcGIS. 9th ed. Environmental Systems Resear ch Institute, Inc., Redlands, California, USA. Fleharty, E. D., and M. A. Mares. 1973. Habi tat preference and spatial relations of Sigmodon hispidus on a remnant prairie in west-central Kansas. The Southwestern Naturalist 18:21-29. Gannon, W. L., R. S. Sikes, and the Animal Care and Use Committee of the American Society of Mammalogists. 2007. Guidel ines of the American Society of Mammalogists for the use of wild mammals in research. Journal of Mammalogy 88:809-823. Goertz, J. W. 1964. T he influence of habitat quality upon density of cotton rat populations. Ecological Monographs 34:359-381. Guthery, F. S., and S. L. Beasom. 1977. Responses of game and non-game wildlife to predator control in south Texas. Journal of Range Management 30:404-409. Hatchell, G. E. 1964. Small mammal species and populations in the longleaf pine forest type of Louisiana. United States Department of Agri culture Forest Service Research Paper SO-10. Heppell, S. S., H. Caswell, and L. B. Crow der. 2000. Life histories and elasticity patterns: Perturbation analysis for spec ies with minimal demographic data. Ecology 81:654-665. Hik, D. S. 1995. Does risk of predation in fluence population dynamics? Evidence from the cyclic decline of snowshoe hares. Wildlife Research 22:115-129. Hubbs, A. H., and R. Boonstra. 1997. Population limitati on in Arctic ground squirrels: Effects of food and predation. Jour nal of Animal Ecology 66:527-541. Hubbs, A. H., and R. Boonstra. 1998. Effe cts of food and predator s on the home-range sizes of arctic ground squirrels ( Spermophilus parryii ). Canadian Journal of Zoology 76:592-596. 96

PAGE 97

Kie, J. G., J. A. Baldwin, and C. J. Evans. 1994. CALHOME: Home range analysis program electronic users manual. United States Forest Service Pa cific Northwest Research Station, Gekeler Lane, La Grande, Oregon, USA. Koivula, M., E. Koskela, T. Mappes, and T. A. Oksanen. 2003. Cost of reproduction in the wild: Manipulation of reproductive effort in the bank vole. Ecology 84:398-405. Krebs, C. J., S. Boutin, R. Boonstra, A. R. E. Sinclair, J. N. M. Smit h, M. R. T. Dale, K. Martin, and R. Turkington. 1995. Impact of food and predation on the snowshoe hare cycle. Science 269:1112-1115. Laake, J., and E. Rexstad. 2008. RMark an alternative approach to building linear models in MARK. Program MARK: A Gentle Introduction, 7th ed. (eds. E. Cooch and G.C. White), pp. C-1C -115. Available at: http://www.phidot.org/software/mark/docs/book accessed March 13, 2010. Layne, J. N. 1974. Ecology of small mamma ls in a flatwoods habitat in north central Florida with emphasis on the cotton rat ( Sigmodon hi spidus ). American Museum Novitates 2544:1-48. Lima, S. L. 1998. Nonlethal effects in the ecology of predator-prey interactions. BioScience 48:25-34. Lima, S. L., and L. M. Dill. 1989. Behavioral decisions m ade under the risk of predation: A review and prospectus. Canad ian Journal of Zoology 68:619-640. Liu, E. S. 1971. The ethology of the cotton rat, Sigmodon hispidus M.S. Thesis. University of Florida, Ga inesville, Florida, USA. Masters, R. E., C. W. Wilson, D. S. Cram G. A. Bukenhofer, and R. L. Lochmiller. 2002. Influence of ecosystem restoration for red-cockaded woodpeckers on breeding bird and small mammal communities. In: Ford, M. W., K. R. Russel, C. E. Moorman, eds. Proceedings: The role of fire for non-game wildlife management and community restoration: Traditiona l uses and new directions. General Technical Report NE-288. Newton Square, PA. US Department of Agriculture, Forest Service, Northeastern Research Station: 73-90. Masters, R. E. 2007. The im portance of short-leaf pine fo r wildlife and diversity in mixed oak-pine forests and in pine-gr assland woodlands. In: Kabr ick, J. M., D. C. Dey, D. Gwaze, eds. 2006. Shortleaf pine re storation and ecology in the Ozarks: Proceedings of a symposium. General Technical Report NRS-P-15. Newton Square, PA: US Department of Agricultur e, Forest Service, Northern Research Station: 35-46. 97

PAGE 98

McCay, T. S. 2000. Use of w oody debris by cotton mice ( Peromyscus gossypinus) in a southeastern pine forest. Jour nal of Mammalogy 81:527-535. McNamara, J. M., and A. I. Houston. 1987. Starvation and predation as factors limiting population size. Ecology 68:1515-1519. Miller, D. A., R. E. Thill, M. A. Melchiors, T. B Wigley and P. A. Tappe. 2004. Small mammal communities of streamside manag ement zones in intensively managed pine forests of Arkansas. Forest Ecology and M anagement 203:381-393. Meserve, P. L., J. R. Gutierre z, and F. M. Jaksic. 1993. E ffects of vertebrate predation on a caviomorph rodent, the degu ( Octodon degus ) in a semiarid thorn scrub community in Chile. Oecologia 94:153-158. Odum, E. P., S. E. Pomeroy, J. C. Dickinson, III, and K. Hutcheson. 1973. The effects of late winter litter burn on the composition, productivity and diversity of a 4-year old fallow field in Georgia. Proceedings of the Annual Tall Timbers Fire Ecology Conference, Tallahassee, Florida, USA, 13:399-427. Oli, M. K. 2003. Populati on cycles of small rodents are caused by specialist predators: Or are they? Trends in Ec ology and Evolut ion 18:105-107. Oli, M. K., and F. S. Dobson. 2003. The relative importance of life-history variables to population growth rate in mammals: Coles predation revisited. The American Naturalist 161:422-440. Peckarsky, B. L., P. A. Abrams, D. I. Bolni ck, L. M. Dill, J. H. Grabowski, B. Luttbeg, J. L. Orrock, S. D. Peacor, E. L. Preisser, O. J. Schmit z, and G. C. Trussell. 2008. Revisiting the classics: Considering noncon sumptive effects in textbook examples of predator-prey interactions. Ecology 89:2416-2425. Perrin, M. R., and S. J. Johnson. 1999. Th e effect of supplemental food and cover availability on a population of the striped mouse. South Af rican Journal of Wildlife Research 29:15-18. Pfister, C. A. 1998. Patterns of variance in stage-structured populations: Evolutionary predictions and ecological implications Proceedings of the National Academy of Sciences of the United St ates of America 95:213-218. Pollock, K. H. 1982. A capt ure-recapture design robust to unequal probability of capture. Journal of Wildlife Management 46:757-760. Pressier, E. L., D. I. Bolni ck, and M. F. Benard. 2005. Scared to deat h? The effects of intimidation and consumption in predator-p rey interactions. Ecology 86:501-509. 98

PAGE 99

Pressier, E. L., and D. I. Bolnick. 2008. The many faces of fear: Comparing the pathways and impacts of nonconsumptive pr edator effects on prey populations. PLoS ONE 3:e2465. R Development Core Team. 2009. R: A language and environment for statistical computing. R Foundation for Statistical Com puting,Vienna, Austria. Available at: http://www.R-project.org. Rehmeier, R. L., G. A. Kauf man, D. W. Kaufman, and B. R. McMillan. 2005. Long-term study of abundance of the hispid cotton rat in native tallgrass prai rie. Journal of Mammalogy 86:670-676. SAS Institute Inc. 2004. SAS/STAT users guide, versio n 9.1. SAS Institute Inc., Cary, North Carolina. Schabenerger, O., and F. J. Pier ce. 2002. Contemporary statis tical models for the plant and soil sciences. CRC Press, New York, New York, USA. Schnell, J. H. 1968. The limiting effects of natural pr edation on experimental cotton rat populations. Journal of W ildlife Management 32:698-711. Schwarz, C. J., and A. N. Arnason. 1996. A general me thodology for the analysis of open-model capture-recapture exper iments. Biom etrics 52:860-873. Seaman, D. E., and R. A. Powell. 1996. An evaluation of the accuracy of kernel estimates for home range analysis. Ecology 77:2075-2085. Shadowen, H. E. 1963. A live-trap study of small mammals in Louisiana. Journal of Mammalogy 44:103-108. Spencer S. R., G. N. Camer on, and R. K. Swihart. 1990. Operationally defining home range: Temporal dependence exhibited by hispid cotton rats. Ecology 71:1817-1822. Suazo, A. A., J. E. Fauth, J. D. Roth, C. L. Parkinson, and I. J. Stout. 2009. Responses of small rodents to habitat restorati on and management for the imperiled Florida scrub-jay. Biological Conservation 142:2322-2328. Swihart R. K. and N. A. Slade. 1985 a Influence of sampling interval on estimates of home-range size. Journal of Wildlife Management 49:1019-1025. Swihart R. K. and N. A. Slade. 1985 b Testing for independence of observations in animal movements. Ecology 66:1176-1184. Tiatt, M. J., and C. J. Krebs. 1983. Predati on, cover, and food manipulations during a spring decline of Microtus townsendii Journal of Animal Ecology 52:837-848. 99

PAGE 100

100 Van Lear, D. H., W. D. Carroll, P. R. K apeluck, and R. Johnson. 2005. History and restoration of the longleaf pine-grassland ecosystem: Imp lications for species at risk. Forest Ecology and Management 211:150-165. Van Vuren, D., and K. B. Armitage. 1994. Su rvival of dispersing and philopatric yellowbellied marmots: What is the cost of dispersal? Oikos 69:179-181. Webb, R. E., D. M. Leslie, Jr., R. L. Lochmille r, and R. E. Masters. 2005. Impact of food supplementation and methionine on high dens ities of cotton rats: Support for the amino-acid-quality hypothesis? Journal of Mammalogy 86:46-55. Weigert, R. G. 1972. Avian versus mamma lian predation on a populat ion of cotton rats. Journal of Wildlife Management 36:1322-1327. Werner, E., and S. Peacor. 2003. A review of trait-mediated indirect interactions in ecological communities Ecology 84:1083-1100. Whitaker, J. O., Jr., W. J. Hamilton, Jr. 1998. Mammals of the eastern United States, 3rd edition. Cornell University Pr ess, Ithaca, New York, USA. White, G. C., and K. P. Bur nham. 1999. Program MARK: Survival estimation from populations of marked animals. Bi rd Study 46(Supplement):120-138. White, G. C. 2002. Discussi on comments on: The use of auxiliary variables in capturerecapture modeling. Journal of Applied Statistics 29:103-106. Wolfe, J. L., and A. V. Linzey. 1977. Peromyscus gossypinus. Mammalian Species 70:1-5 Yunger, J. A. 2004. Movem ent and spatial organization of small mammals following vertebrate predator exclusio ns. Oecologia 139:647-651.

PAGE 101

BIOGRAPHICAL SKETCH Gail Morris is originally from Willia msport, Pennsylvan ia. She attended Muhlenberg College in Allentown, PA and receiv ed a BS in Biology in 2004. Following a series of seasonal wildlife field jobs, she ended up at the Joseph W. Jones Ecological Research Center where, among other things, she caught and chased rats for a year. This lead to the unexpected discovery of a fa scination with rats, mice, and other critters at the bottom of the food chai n. She was eventually offered a graduate assistantship with the University of Florida which enabled t he continuation of rat and mouse studies at the Jones Center. 101