USF Libraries
USF Digital Collections

An evaluation of movement patterns and effects of habitat patch size on the demography of the Florida mouse (Podomys flo...

MISSING IMAGE

Material Information

Title:
An evaluation of movement patterns and effects of habitat patch size on the demography of the Florida mouse (Podomys floridanus)
Physical Description:
Book
Language:
English
Creator:
Lukanik, Irmgard
Publisher:
University of South Florida
Place of Publication:
Tampa, Fla.
Publication Date:

Subjects

Subjects / Keywords:
Habitat fragmentation
Mark-recapture
Program MARK
Metapopulation
Genetic analyses
Dissertations, Academic -- Biology -- Masters -- USF   ( lcsh )
Genre:
bibliography   ( marcgt )
theses   ( marcgt )
non-fiction   ( marcgt )

Notes

Abstract:
ABSTRACT: Habitat degradation by humans has been the main reason for the decline in numbers of P. floridanus, the only mammal indigenous to the state of Florida, in the past century. The mouse inhabits what remains of scrub and sandhill associations, which are characterized by patches of sandy soils within a more mesic landscape. It has long been accepted that small populations are more prone to decline and extinction than are larger ones as a result of environmental fluctuations. I hypothesized that the demography of a population of P. floridanus would be affected by a restriction in numbers through habitat patch size in a deterministic way, even without any environmental effects. I also examined dispersal and looked for evidence of metapopulation dynamics. Mark-recapture data were collected from ten scrub fragments in Lake Wales Ridge State Forest, Polk County, FL, ranging in size from 0.5 to 170 ha. Program MARK was used to model survival, recruitment and population growth rate of P. floridanus as a function of habitat patch size and to evaluate temporary migration patterns. Recruitment was positively associated with patch size, but contrary to expectations survival and population growth were negatively associated with patch size. Results suggested that survival was negatively affected by ear tagging, although this effect was temporary. Evidence of migration was found, but would probably have been greater if trapping had been continued until after peak reproduction, when juveniles tend to disperse in search of resources. The degree of interbreeding among patches can only be determined with the help of genetic analyses. Microsatellites have become useful in analyses at the population level because of their high degree of variability. Future research including genetic analyses is recommended to evaluate the importance of gene flow among subgroups to demography and the viability of the study population.
Thesis:
Thesis (M.S.)--University of South Florida, 2007.
Bibliography:
Includes bibliographical references.
System Details:
System requirements: World Wide Web browser and PDF reader.
System Details:
Mode of access: World Wide Web.
Statement of Responsibility:
by Irmgard Lukanik.
General Note:
Title from PDF of title page.
General Note:
Document formatted into pages; contains 70 pages.

Record Information

Source Institution:
University of South Florida Library
Holding Location:
University of South Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 001920218
oclc - 187943131
usfldc doi - E14-SFE0002099
usfldc handle - e14.2099
System ID:
SFS0026417:00001


This item is only available as the following downloads:


Full Text
xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam Ka
controlfield tag 001 001920218
003 fts
005 20080107114346.0
006 m||||e|||d||||||||
007 cr mnu|||uuuuu
008 080107s2007 flu sbm 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0002099
040
FHM
c FHM
043
n-us-fl
035
(OCoLC)187943131
049
FHMM
090
QH307.2 (ONLINE)
1 100
Lukanik, Irmgard.
3 245
An evaluation of movement patterns and effects of habitat patch size on the demography of the Florida mouse (Podomys floridanus)
h [electronic resource] /
by Irmgard Lukanik.
260
[Tampa, Fla.] :
b University of South Florida,
2007.
520
ABSTRACT: Habitat degradation by humans has been the main reason for the decline in numbers of P. floridanus, the only mammal indigenous to the state of Florida, in the past century. The mouse inhabits what remains of scrub and sandhill associations, which are characterized by patches of sandy soils within a more mesic landscape. It has long been accepted that small populations are more prone to decline and extinction than are larger ones as a result of environmental fluctuations. I hypothesized that the demography of a population of P. floridanus would be affected by a restriction in numbers through habitat patch size in a deterministic way, even without any environmental effects. I also examined dispersal and looked for evidence of metapopulation dynamics. Mark-recapture data were collected from ten scrub fragments in Lake Wales Ridge State Forest, Polk County, FL, ranging in size from 0.5 to 170 ha. Program MARK was used to model survival, recruitment and population growth rate of P. floridanus as a function of habitat patch size and to evaluate temporary migration patterns. Recruitment was positively associated with patch size, but contrary to expectations survival and population growth were negatively associated with patch size. Results suggested that survival was negatively affected by ear tagging, although this effect was temporary. Evidence of migration was found, but would probably have been greater if trapping had been continued until after peak reproduction, when juveniles tend to disperse in search of resources. The degree of interbreeding among patches can only be determined with the help of genetic analyses. Microsatellites have become useful in analyses at the population level because of their high degree of variability. Future research including genetic analyses is recommended to evaluate the importance of gene flow among subgroups to demography and the viability of the study population.
502
Thesis (M.S.)--University of South Florida, 2007.
504
Includes bibliographical references.
516
Text (Electronic thesis) in PDF format.
538
System requirements: World Wide Web browser and PDF reader.
Mode of access: World Wide Web.
500
Title from PDF of title page.
Document formatted into pages; contains 70 pages.
590
Advisor: Henry Mushinsky, Ph.D.
653
Habitat fragmentation.
Mark-recapture.
Program MARK.
Metapopulation.
Genetic analyses.
0 690
Dissertations, Academic
z USF
x Biology
Masters.
773
t USF Electronic Theses and Dissertations.
4 856
u http://digital.lib.usf.edu/?e14.2099



PAGE 1

An Evaluation of Movement Patterns and Effects of Habitat Patch Size on the Demography of the Florida Mouse ( Podomys floridanus ) by Irmgard Lukanik A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science Department of Biology College of Arts and Sciences University of South Florida Co-Major Professor: Henry Mushinsky, Ph.D. Co-Major Professor: Earl McCoy, Ph.D. James Garey, Ph.D. Date of Approval: July 18, 2007 Keywords: habitat fragmentation, mark -recapture, program MARK, metapopulation, genetic analyses Copyright 2007, Irmgard Lukanik

PAGE 2

Dedication This thesis is dedicated to my husband, Michael Lukanik, without whose constant emotional and financial support I may never have embarked on a college career, much less finished with a Masters degree. His s acrifices did not go una ppreciated, especially the many evenings he sat alone while I worked. It was often his pride in my accomplishments that kept me going.

PAGE 3

Acknowledgements I would like to thank my majo r professors, Drs. Mushin sky and McCoy, for their guidance and for supporting me through several ch anges in direction with regard to my research. They never told me that I couldn’t do something I envisione d, but left it to me to discover what was and wasn’t possible. My thanks also go to Drs. Garey and Pierce for kindly allowing me the use of their labor atories and expertise when my study took a turn into the field of genetics. Stefi Depovic, Terry Campbell, Haydn Rubelmann, Julie Schwartz and Anna Bass patie ntly taught me what I needed to know about PCR and genetic theory. Brian Halstead was my teacher and guide in the field. He instructed me in mark-recapture techniques, awakened my in terest in snakes and lizards (much to my mother’s dismay) and stayed in the field with me until I stopped getting lost. He also provided me with his mark-recapture data. Fi nally, my thanks go to Anne Malatesta of the Lakeland District Division of Forestry fo r her permission to conduct my study at Lake Wales Ridge State Forest.

PAGE 4

i TABLE OF CONTENTS LIST OF TABLES iii LIST OF FIGURES v ABSTRACT vi INTRODUCTION 1 Taxonomy 3 Physical Description 4 Demography 4 Habitat Requirements 5 Microsatellites 7 METHODS AND MATERIALS 9 Study Site 9 Data Collection 9 Data Analyses 13 Mark-Recapture Data 13 DNA Analysis 14 RESULTS 17 Dispersal 17 Mark-Recapture Analyses 17 Goodness-of-Fit of Global Models 17 Results for 2004 18 Conventional Robust Design 18 Pradel Robust Design with Individual Covariates 19 Pradel Robust Design with Groups 20 Link Barker with Individual Covariates 20 Link Barker with Groups 21 Summary of Parameter Estimates for 2004 Data 21 Results for 2005 Data 25 Conventional Robust Design 25 Pradel Robust Design with Individual Covariates 25 Pradel Robust Design with Groups 26 Link Barker with Individual Covariates 26

PAGE 5

ii Link Barker with Groups 27 Summary of Parameter Estimates for 2005 Data 28 Effect of Limiting Number of Capture Occasions 31 Modeli ng Heterogeneity in Ca pture Probabilities 32 Molecular Analyses 32 DISCUSSION 34 Zero Model Deviances 34 Temporary Effect of Marking on Survival 34 Migration 35 Parameter Estimates 38 Future Direction 43 REFERENCES 45 APPENDICES 53 Appendix A: Program MARK 54 Appendix B: Additional Figures 64

PAGE 6

iii LIST OF TABLES Table 1 Primers Used to Amplify P. floridanus Microsatellites 16 Table 2 2004 Top Models Us ing Conventional Robust Design ranked by AIC 19 Table 3 2004 Top Models Using Li nk Barker with Individual Covariates 20 Table 4 2004 Top Models Using Link Barker with Groups 21 Table 5 Parameter Estimates for 2004 Data 22 Table 6 2005 Top Models Using Conventional R obust Design 25 Table 7 2005 Top Models Using Pradel Robus t with Individual Covariates 26 Table 8 2005 Top Models Using Pradel Robust with Groups 26 Table 9 2005 Top Models Using Li nk Barker with Individual Covariates 27 Table 10 2005 Top Models Using Link Barker with Groups 27 Table 11 Parameter Estimates for 2005 Data 28 Table 12 June 2005 Models for Six Day Sampling Period 32 Table 13 Primer-Specific Annealing Temperatures Used in PCR Amplifications 33 Appendix A Table 14 Actual and Adju sted Habitat Patch Sizes 60

PAGE 7

iv Appendix B Table 15 2004 CJS Model for Herp Array Data 64 Table 16 2005 CJS Model for Herp Array Data 65 Table 17 2004 Models Using Conventional Robust Design Ranked by AIC 66 Table 18 2004 Parameter Estimates for the Conventional Robust Design 66 Table 19 2004 Models Using Pradel Robust Design with Individual Covariates 67 Table 20 2004 Models Using Pr adel Robust Design with Groups 67 Table 21 2004 Models Using Link Ba rker with Individual Covariates 68 Table 22 2004 Models Usi ng Link Barker with Groups 68 Table 23 2005 Models Usi ng Conventional Robust Design 68 Table 24 2005 Models Using Prad el Robust Design with Individual Covariates 69 Table 25 2005 Models Using Pr adel Robust Design with Groups 69 Table 26 2005 Models Using Link Ba rker with Individual Covariates 69 Table 27 2005 Models Usi ng Link Barker with Groups 70

PAGE 8

v LIST OF FIGURES Figure 1. Location of Trap Arrays at LWRSF Study Site 10 Figure 2. Arrangement of Trap Arrays 11 Figure 3. Abundance Estimates for March September, 2004 23 Figure 4. Survival Estimates fo r March-September 2004 Using Different Model Types 24 Figure 5. Recruitment Rates Using Link Barker Models 24 Figure 6. Abundance Estimates for April August, 2005 30 Figure 7. Survival Estimates for April September 2005 Using Different Model Types 30 Figure 8. Estimates for Population Growth Rate in 2005 31 Figure 9. Cumulative First Captures Over A Six-Day Sampling Period 31 Figure 10. Bands of PCR Products Obtained with Primers PO-9 and BW-3 33 Figure 11. Seasonal Variation in Pregnancies in Alachua County 37 Figure 12. Estimates of Recruitment Rates in Large vs. Small Patches 41 Figure 13. Estimates of Population Grow th Rates in Large vs. Small Patches 41 Figure 14. Estimates of Survival Rates in Large vs. Small Patches 42 Appendix A Figure 15. Basic Structure of the Robust Design Model 57

PAGE 9

vi An Evaluation of Movement Patterns and Effects of Habitat Patch Size on the Demography of the Florida Mouse ( Podomys floridanus ) Irmgard Lukanik ABSTRACT Habitat degradation by humans has been the main reason for the decline in numbers of P. floridanus the only mammal indigenous to the state of Florida, in the past century. The mouse inhabits what remains of scrub and sandhill associations, which are characterized by patches of sandy soils within a more mesic landscape. It has long been accepted that small populations are more prone to decline and extinc tion than are larger ones as a result of environmental fluctuations I hypothesized that the demography of a population of P. floridanus would be affected by a restriction in numbers through habitat patch size in a deterministic way, even without any environmental eff ects. I also examined dispersal and looked for evidence of metapopulation dynamics. Mark-recapture data were collected from ten scrub fragments in Lake Wales Ridge State Forest, Polk County, FL, ranging in size from 0.5 to 170 ha. Pr ogram MARK was used to model survival, recruitment and population growth rate of P. floridanus as a function of habitat patch size and to evaluate temporary migration pattern s. Recruitment was positively associated with patch size, but contrary to expectations survival and population growth were negatively associated with patch size. Results s uggested that survival was negatively affected by ear tagging, although this effect was tempor ary. Evidence of migration was found, but

PAGE 10

vii would probably have been greater if trapping had been continued unt il after peak reproduction, when juveniles tend to disperse in sear ch of resources. The degree of interbreeding among patches can only be determined with the help of genetic analyses. Microsatellites have become useful in analyses at the population level because of their high degree of variability. Future research including gene tic analyses is recommended to evaluate the importance of gene flow among subgroups to demography and the viab ility of the study population.

PAGE 11

1 INTRODUCTION Habitat fragmentation and destruct ion have been the main causes for the decline in numbers of the Florida mouse ( Podomys floridanus ) (Layne 1992), as has been the case for many species of flora and fauna worldw ide (Burkey 1994, reviewed in Campbell and Reece 2002 and Ricklefs and Miller 1999). P. floridanus is the only mammal endemic to the state of Florida (Layne 1992) Its range extends mainly over the northern two thirds of the state’s peninsula, but within this ar ea its distribution is patchy and mostly limited to what remains of natural scrub and sandhi ll associations (Myers 1990). During the Wisconsinan glacial period (approx. 70,000 to 10,000 years ago), when the Florida landmass was much larger and drier, P. floridanus may have had a more extensive and continuous distribution; however, as sea level ro se during the Holocene and much of the remaining landmass became wetlands, its distribut ion became more restricted because of its dependence on xeric habitats (Layne 1992). Si nce the early 1900s, hum an activities such as phosphate mining, agricultural use and real estate development have further fragmented and reduced upland areas by about 70% (McCoy and Mushinsky 1992, Humphrey 1992). As a result, populations of P. floridanus have declined further, and the mouse is currently listed as a species of special c oncern by the Florida Game and Fresh Water Commission (Layne 1992) and as vuln erable on the IUCN Red List. Previous studies have shed some light on microhabitat requirements and natural history of this species (Layne 1966, Layne and Jackson 1994, Jones unpublished, Jones

PAGE 12

2 and Franz 1990, Schmutz unpublished). Howeve r, although it is generally recognized that, as a result of stochastic events, small populations are more prone to extinction than are larger ones (MacArthur and Wilson 1967, Burkey 1995, Frankham 1997), little is known about the effects of habitat fragmenta tion on demographic parameters and extincttion risk of the Florida mouse. Hokit and Branch (2003) have shown that for populations of the scrub lizard ( Sceloporus woodi) which are isolated from each other because of restriction of movement betw een habitat patches (Clark et al. 1999, Hokit et al. 1999), patch size was positively associated with abunda nce, survivorship and recruitment. Thus, the size of a fragment not only constrained popul ation size, thereby in creasing the risk of extinction as a result of chance environmen tal fluctuations, but may also have had a direct deterministic effect on population dem ography. Contrary to the case of groups living in isolation, other studi es have shown that some rodents exist in metapopulations, in which individuals migrate between habitat patches by way of landscape corridors. This dispersal provides increased genetic heteroge neity and thus viability of subpopulations (Merriam and Lanoue 1990, Bennett 1990). In metapopulations, therefore, immigration and emigration of individuals among groups would have effects on demographic parameters in addition to those that may resu lt from patch size (D iffendorfer et al. 1995, Fahrig and Merriam 1992). “Source” populatio ns may provide colonists for groups in other fragments, allowing them to persist when otherwise they might become extinct (Pulliam 1988). Genetic flow between pa tches increases genetic heterogeneity and reduces negative inbreeding e ffects and thus may contribut e to the viability of populations. In the case of a species that exists in metapopulati ons, conservation efforts would have to encompass all subpopulations and thei r habitats in order to have a maximum

PAGE 13

3 effect. The current study was conducted in or der to determine whet her habitat patch size has a deterministic effect on th e demography of a population of P. floridanus and whether evidence of a metapopul ation structure exists. Taxonomy – From 1909 until 1980, P. floridanus was described as a subgenus of the genus Peromyscus (Osgood 1909, Hooper 1968), which is one of the most well represented genera of North American mammals and cons idered the ecological counterpart of Old World Apodemus (Kirkland and Layne 1989). In 1980, Carleton reclassified many groups within Peromyscus elevating Podomys to a generic level. The reclassification was based mainly on studies of the male re productive tract, which indicated a phyletic separation of P. floridanus from others in the genus. Carleton found consistent links among Podomys, Habromys and Neotomodon the latter two of which are found in southern Mexico and Guatemala. Until the 1970s, taxonomic classificat ion had been based largely on morphological features such as dentition, cranial size and shape and the structure of reproductive organs but shortly thereafter el ectrophoretic and karyologic analyses became widely available. Studi es based on chromosomal inversions and rearrangements (Yates et al. 1979, Robbins and Baker 1981) confirmed the relationship between Podomys and Neotomodon, but did not include an examination of Habromys Later, however, H. lepturus was proposed as a sister-group to Neotomodon Johnson and Layne (1961) found corroborating evidence for the close relationship among the three genera based on similarities between their most common ect oparasites. According to King (1968), P. floridanus is thought to have descended fr om stock that was at one time

PAGE 14

4 widely distributed in the s outhern U.S. and possibly in Middle America, lending further support to the idea that it ma y be closely related to form s currently living in Mexico. Physical Description – Podomys floridanus is a relatively large mouse, with adults ranging in total length between 179 and 197 mm and in mass between 25 and 49 g. Mean values vary widely among populations (Layne 1992). Eyes, ears an d hind feet are proportionally large. Adult pelage is brownish or tan dorsally, fading to orange alo ng the sides and white ventrally. Juveniles are grayis h in overall color with a white venter. One of the most distinguishing features of the Florida mouse is the presence of only five plantar tubercles on the soles of the hind feet as opposed to the six found on other mice within its range. In addition to these characteristics, it has a distin ct, skunk-like odor. Demography – Means of abundances measured in individuals/100 trapnights va ry widely with habitat type and trapping method [3.5-18.1 for La yne and Griffo (1961) and 0.26–7.1 for Humphrey et al. (1985)], a nd Layne (1990) reported estimates of population densities with a mean and maximum of 5/ha and 28/ha respectively. In general, scrub systems have been found to support higher densities than have been found in sandhill, presumably because of larger food availability in the form of acorn mast in scrub ecosystems (Layne 1992). Information on home range size for P. floridanus is scarce; however, Jones (unpublished) reported home ranges fo r 20 adults (males and females) ranging between 300 and 1850 m2. Relative estimates of home range sizes based on mean distances between successive captu res of individuals indicated overall larger home ranges in sandhill than in scrub (Layne 1992). Mean survival times were found to be higher for

PAGE 15

5 adults than for younger age classes, with adul ts surviving 4.2 months and 2.0 months in sandhill and scrub, respectively. Five percen t of individuals in a sandhill population survived for more than a year, and one individual survived for mo re than 7 years in captivity (Layne 1992). Predation by snakes, owls and various mammals is thought to be the main source of mortality. Females become sexually mature by the age of 5 weeks, while males take somewhat longer (Layne 1966). Breeding ta kes place primarily in late summer and fall with a lesser peak in late winter. Litter sizes can range from 1 to 5, but most often there are two to three young, a nd females usually produce no more than two litters per season (Layne 1992). Habitat Requirements – The Florida mouse occurs mainly in two types of habitats : scrub systems, including sand pine scrub and scrubby flatwoods, and sand hill (Layne 1992). These are xeric, firedependent plant communities located on we ll drained and nutrie nt-poor, sandy upland soils. Sandhill usually consists of three layers of vegetation. Longleaf pine ( Pinus palustris ) or slash pine ( P. elliottii ) and xeric oaks (mostly turkey oak, Quercus laevis ) form a scattered overstory, the understory consists of Serenoa repens, Diospyrus virginiana and other woody shrubs and the dive rse herbaceous ground layer includes Aristada spp., Asimina angustifolia, As clepias humistrata, A. tuberosa and Chrysopsis scabrella (Taylor 1998, Hartman 1992). Scr ub consists of a sand pine ( P. clausa ) overstory and a shrub layer made up of several scrubby oaks (i.e. Q. myrtifolia, Q. geminata and Q. chapmanii ) and shrubs such as Ceratiola ericoides, S. repens, Lyonia ferruginea, Carya floridana and Persia humilis Herbaceous ground cover is sparse and interspersed with open, sandy ar eas. Scrubby flatwoods occu r on relatively dry ridges in

PAGE 16

6 typical flatwoods; sand pine is usually re placed by longleaf or slash pine, and Lyonia lucida and Ilex glabra are common shrubs. Scrub oak species provide a much more abundant acorn mast than that of the turk ey oaks in sandhill communities, making scrub the preferred habitat for the Fl orida mouse (Layne 1992). Othe r than acorns, the diet of the mouse consists of insects, seeds, fruits, nuts and fungi (Layne 1978, Jones unpublished). In addition to other insects, Florida mice have been observed to feed on engorged soft ticks, Ornithodoros turicata americanus which are known to parasitize the gopher frog ( Rana capito ) and the gopher tortoise ( Gopherus polyphemus ) (Jones unpublished). Predation on this pa rasite and other insects found in gopher tortoise burrows is presumably one reason why P. floridanus is often found living commensally with the tortoise. The mouse is exclusively burrowdwelling, and although there is some debate as to whether or not it constructs burrows, it is most often found i nhabiting tortoise burrows and, to a lesser degree, those of the nine-banded armadillo ( Dasypus novemcinctus ), oldfield mouse ( Peromyscus polionotus ) and cotton rat ( Sigmodon hispidus ) (Layne and Jackson 1994). Excavated gopher tortoise bu rrows have revealed narrow side tunnels, chimneys and nest chambers constructed by Florida mice (Jones and Franz 1990). Burrows serve as a refuge from extreme temper atures and fire. Burrow temperatures are relatively constant compared with those aboveground, which I r ecorded to be as high as 44 degrees Celsius during midsummer at my study site in Lake Wales Ridge State Forest. Florida mice are active at night and pr esumably retreat underground during the day, thereby avoiding high daytime temperatures. Torpor, a physiologically regulated lowering of the body temperature, may also play a role in avoiding heat stress and water loss. Although the phenomenon has not been documented for P. floridanus it has been shown

PAGE 17

7 that torpor may be initiated in P. polionotus which inhabits Florida scrub systems, by a lowering of ambient oxygen concentration such as occurs in burrows. In Peromyscus eremicus and Peromyscus truei, both of which exist in xeric habitats, torpor seems to be induced by negative water balance (Hill 1983). Microsatellites The degree of relatedness of organi sms is reflected in the degree of similarity in their DNA; two individuals of the same speci es or population have more similar genomes than do individuals of different species or populations. Nevertheless, high degrees of variability can be found in nuc lear as well as organellar DNA in the form of heterogeneity in alleles (Parker et al 1998) even between closely re lated individuals. Usually, highly variable regions occur in non-coding DNA because mutations are not subject to the same selective pressures that affect codi ng sections of the genome. The origin of replication in mitochondrial DNA, called the Displacement Loop, is highly variable in most animal species and is therefore useful in studies at the population level. A major drawback, however, is the fact that mito chondrial DNA is inherited in a uniparental fashion, most often as part of the egg’s cyto plasm, and thus studies of this DNA type will only reveal matrilines (Parker et al 1998, re viewed in Klug and Cu mmings 2002). Microsatellites are regions of non-coding nuclea r DNA consisting of tandemly repeating units of a core nucleotide sequence of two to six base pairs. Unlike most alleles, those of microsatellites do not vary in their sequences of base pairs, but ra ther in the number of repeats of the core unit. This variation in number of repeats is common, making microsatellites one of the most variable types of genetic markers in the eukaryotic genome, and because they are found in nuclear rather th an mitochondrial DNA they are able to trace

PAGE 18

8 both parental lineages (Parker et al 1998). Once microsatelli tes have been processed and sequenced, specialized comput er software (e.g. Arlequin, POPGENE) may be used to calculate indices of diversit y, population structure, geneti c distance and clustering in order to determine the relationships of i ndividuals within and among groups (Labate 2000).

PAGE 19

9 METHODS AND MATERIALS Study Site – The study was conducted at the Ar buckle Tract in Lake Wa les Ridge State Forest (LWRSF), located five miles southeast of the to wn of Frostproof in Polk County, Florida. This area is part of the Florida Central Ridge, a chain of sand ridges and ancient dunes running north-south from Clay and Putnam C ounties to Highlands County (Myers 1990). The range of formerly extensive scrub ecosystems was likely reduced about 5,000 – 7,000 years ago, when climate changes resulted in rising water levels. High water levels in turn led to the present-day mosaic patte rn of isolated scrub islands surrounded by lowlying, more mesic habitat. LWRSF represents one of the last remnants of a scrub system which continues to be reduced by Flor ida’s ever-increasing human population. Data Collection – Fifteen arrays of 15 Sherman liv e traps each were installed in ten scrub fragments, which ranged in size from approx. 1.5 to 170 ha. (Figure1). My study proceeded in conjunction with ongoing research conducted by Brian Halstead, a PhD student at the University of South Florida, who was colle cting data on prey species of the Eastern coachwhip snake ( Masticophis flagellum flagellum ). Sherman live traps were arranged around trap arrays already in stalled by Brian for the purpos e of capturing reptiles and amphibians. Each of these arrays consisted of four sections of metal drift fence approximately 8 m in length extending in the four cardinal di rections with a large, square trap in

PAGE 20

10 the center and bucket and funnel traps along bot h sides of each section. My small mammal traps were arranged such that a single ro w of three traps exte nded outward from the end of each of the four sections of drift fe nce, with 10 m intervals between traps (Figure 2). A two-by-four post was also installed at the north arm of each array, to which a rain Figure 1. Location of Trap Arrays at LWRSF Study Site

PAGE 21

11 gauge and a max/min thermometer were attached Thermometers were oriented to face north so that they would not be directly inundated with s unlight. Each morning when traps were checked, amount of rainfall a nd maximum and minimum temperatures were recorded and the instruments were reset. Figure 2. Arrangement of Trap Arrays Trapping was conducted for thr ee consecutive nights per month from March to October of 2004 and 2005. Sampling periods of at least five consecutive nights are common for small mammals (Wilson et al 1996, Swilling and Wooton 2002, Rave and Holler 1992); however, because of the number of researchers already working in LWRSF Sherman Live tra p Central trap Bucket trap Funnel tra p

PAGE 22

12 and in the interest of limiting human traffi c through conservation land, I was only given permission to trap for three nights per month by park management. As a result, the number of models that could be used for analys es was reduced, because models incorporating individual heterogeneity in capture probability (a relaxation of one of the assumptions for closed population capture-recaptu re models) cannot be used w ith only three capture occasions per sampling period (Gary C. White, Co lorado State University, personal communication). In order to test for heterogeneity in capture probabilities I trapped for six consecutive nights at eleven of fifteen sites during the month of June 2005 and compared the results with those from other months (see Data Analyses). Traps were opened at dusk and baited with a small handful of sunflower seeds, then checked and closed after daybreak each morning. Rolled oats with and without peanut butter are also commonly used as bait fo r small mammal traps; however, anecdotal evidence suggests that fire an ts, which are known to prey on live-trapped animals in the southern U.S. and were also observed at the study site, may be less attracted by sunflower seeds (B. Halstead, University of South Florida, personal communication). During colder months, mice caught in traps can develop hypothermia (Jones unpublished). Therefore, traps were insulated with a handful of excelsi or (wood shavings used as packing material) when temperatures were forecast to dip below 10 degrees C. Excelsior is superior to Spanish moss (Jones unpublished) and cotton (B. Halstead, personal communication) as an insulation material because it does not absorb moisture. Captured individuals were iden tified to species, weighed, sexed, examined with respect to reproductive status and first-time captures were marked with metal ear tags for individual identificati on using a 1005-1S Monel applicator (Hasco Tag Co.). Initially, I

PAGE 23

13 attempted to mark animals w ith an ear punch code using a stainless steel ear punch [Fine Science Tools (USA), Inc., 2 mm punch diameter], but because the ears of P. floridanus were so fragile, this method re sulted in large tears in the ears and I abandoned the technique. Loss of ear tags became a problem in a small percentage of mice. Because site fidelity was high and there were relatively few occasions on which I captured a mouse in more than one array, I felt comfortable in assigning an individua l which had obviously lost its tag a number that had belonged to an animal of the same sex and had been recorded in the same array in previous months, but not since then. Small tissue samples were taken from the ears of 86 mice with surgical scissors for DNA anal yses. The scissors were swabbed with 70% isopropyl alcohol and flamed with a cigare tte lighter between uses to avoid infection. Tissue samples we re placed in 1.5 ml Eppendorf tubes containing a salt saturated dimethyl sulfoxide (DMSO) buffer, Ph 7.5, and later frozen to –20C. Data from P. floridanus captured incidentally in B. Halstead’s reptile traps during the same time periods were made available to me. I planned to analyze these data as well and compare the results with those obtained from the Sherman Live trap (SLT) arrays. Because recapture rates were very low, how ever, initial analyses revealed that many parameters would be inestimab le and/or estimates would ha ve huge confidence intervals, making these analyses pointless (see Results). Data Analyses – Mark-Recapture Data – Mark-recapture data from tw o years (2004 and 2005) were analyzed using the program MARK. Initially, seve ral model types were compar ed using the 2004 data set.

PAGE 24

14 Those that seemed most appropriate were th en selected and also used on the 2005 data. An explanation of MARK and model se lection is given in Appendix A. A factor that could have cont ributed to an underestima tion of population sizes was the low number of secondary sampling occasions to which I had been limited (see Data Collection). I tested for this possibility using the SLT data from June 2005 (the month during which I had sampled for six consecutive nights instead of th ree) by plotting the cumulative number of first-time captures over ti me. If the slope of the line reached a horizontal asymptote by the third night, it would indicate that all available animals would have been captured by then and the additi onal sampling occasions would not have been necessary. Alternatively, a slope that did no t level off until after the third sampling occasion would indicate that individu als that had been available for capture in other months had been missed as a result of a lack of sampling, and because population size is estimated individually for each primary sampli ng period, these estimates would be biased low. Using the same June data, I also ran models with and without heterogeneity in capture probabilities to estimat e how much more, if any, vari ability could have been explained using models with heterogeneity. As previously explained, these types of models cannot be used with as few as three secondary sampling occasions. DNA Analysis A search of GenBank (an open access, annotated collecti on of publicly available nucleotide sequences produced at the Nati onal Center for Biot echnology Information) failed to produce sequences of any known microsatellites for P. floridanus or either of the genera thought to be most closely related, Habromys or Neotomodon Therefore, primers developed for microsatellites isolated in the oldfield mouse, Peromyscus polionotus and

PAGE 25

15 the deer mouse, P. maniculatus (listed below in Table 1) we re used. Primers PO-9, PO26 and PO-68 were obtained from Prince et al (2002) while the others were among those listed in Mullen et al. (2006). All had prev iously been used to amplify microsatellite DNA across different species of Peromyscus DNA was extracted from ear cli ppings using UltraClean Ti ssue DNA Isolation kits (Mo Bio Laboratories, Inc.; Catalog No. 12334S). Touchdown PCR amplifications were performed in a 25L volume. For all primers, 25ng of template DNA, 1.25 U Taq DNA polymerase (ID Labs Biotechnology, Inc.), 2.5L of 10x ID Proof buffer with 20mM MgCl2 (ID Labs Biotechnology, Inc.), 1.0L each of 10M forward and reverse primers (Integrated DNA Technologies, Inc.; see Ta ble 4) and 5.0L of 1.25mM dNTPs were used. An initial denaturing st ep at 94 C for 2 minutes was followed by 10 cycles of 94 C for 30 s, annealing for 30 s, 72 C for 45 s, 94 C for 30 s, annealing for 30 s and 72 C for 45 s. Initial ann ealing temperature was set at 62 C or lower (depending on melting temperatures of primers) and reduced by 1.0 C each cycle. This procedure was followed by 20 cycles of 94 C for 30 s, 30 s at 10 C below the initial annealing temperature and 72 C for 45 s. The final extension occurred at 72 C for 90 s. If amplification was unsuccessful, initial anneal ing temperature was lowered by 1 C and repeated until DNA amplification could be confirme d using agarose gel electrophoresis.

PAGE 26

16 Table 1. Primers Used to Amplify P. floridanus Microsatellites Primer Repeat Motif in Allele Sequence 5’-3’ PO 9 (AC)20N14(AC)8 N16(AC)20 F: TTTCAGAGGACCAGAGTAGG R: AACTCTGGGTCTTAATACTTT PO 26 (AG)13(ACAGAG)4 F: GCTTCAGTGTTGATGTCTGAT R: GCCTCTCTGTCTCTGTCTAT PO3 68 (TG)22+ F: GTAGTCTGAGAAGCGAAAGG R: TTTATTTGGGTCAGCTCGAC PO 31 (GA)26 F: TTTCAGTGGCTCTCATGGTTA R: AGCTTTCTTCTTCCCAACTA PO 71 (AC)10(AG)32 F:CAGCCAGAACAAAATAGCACT R: AGCTTCATGCCTCCTATATTC BW4 28 (TCTA)15 F: TAATCCAGGTGTATCTAATCT R: CCCAGTATTGCTAGTCT BW4 45 (CTTT)19(CT)19 (CCTT)4 F: ATGGCCTGCCTACCTCA R: AGGGGAAGTGAAAAGCTACA BW4 93 (CCTT)10(TTT)23 F: GACATTTAAAAAGGACTG R: CCCTCTTGATTCCACAC BW4 112 (AGAT)13 F: GGCAGTGCATTCATGGTAA R: TGAGTCCCCAGTTGTATGTA BW4 137 (ATAG)9 (GATA)16 F: GGCTTGGTGGATTAATG R: ATGCCAGAGCTGTTATAC BW4 178 (ATAG)13 F: CCGTTTTTCTTACTCA R: CAAAACAGTGGGTCAA BW4 200 (ATCT)5(GTCT)6 F: GCACATTTCTCCTTCTAAGC R: GACCACCTGATGAGCATAGAT BW4 234 (TGAA)6TAAC (AAAT)4 F: ATTCCAACTCAGCAGGTAGA R: GCCCAGAGTGTGTCATGTAG

PAGE 27

17 RESULTS In total, 419 Florida mice were captured in Sherman Live traps (SLT) during two seasons (207 in 2004 and 212 in 2005) using 7,9 48 trapnights. Other species encountered were mainly Peromyscus polionotus and occasionally P. gossypinus and Sigmodon hispidus Densities of P. floridanus calculated as number of individuals per 100 trapnights, were 0.63 for 2004 and 0.45 for 2005. Only two individuals were captured in both years. A male was captured fifteen times and a female was captured five times, both over ten-month periods. P. floridanus trapped incidentally in herp arrays numbered 112 in 2004 (9,672 trapnights) and 99 in 2005 (8,736 trapnights). Dispersal – Of the 419 Florida mice captured in SLT arrays, 18 were recaptured in a different habitat patch from the one in which they were originally trapped. For thirteen of those individuals only one-way movement was reco rded, while the other five were found to have moved off their original patch and then back again. The longest migration distance recorded was between arrays G1 and B4 (Figure 1), approximately 1.7 km. Mark-Recapture Analyses Goodness-of-Fit of Global Models – Bootstrap tests run on global CJS models [Phi (group*t) p (group*t)] indicated adequate fit for the SLT data sets (p = 0.22 for 2004 and p = 0.15 for 2005 with 100 simulations each), and no adjustments to were necessary (see Appendix A, Goodness-of-Fit

PAGE 28

18 and Model Selection for an explanation of ). Even the least parameterized models that could have served as global starting models for the herp array data, based on the outcome of the SLT data analyses, did not provide meaningful results in the Bootstrap tests. I interpreted the combination of non-estimab le parameters and/or estimates with huge confidence intervals (Appendix B, Tables 15 and 16) and zero values for the models as well as many of the Bootstrap simulations as meaning that the data were too sparse and excluded the herp array data from the analyses. Results for 2004 – Models with two variables conn ected by a “*”, e.g. phi (t patchsize), denote effects of both variables and an interaction between the two. Variables connected by a “+”, e.g. phi (t + patchsize) have additive effects, but there is no interac tion term. The notation (all .) in robust design models means th at a parameter was held constant between secondary as well as primary sampling occas ions (see Appendix A for a description of the robust design model). The notation (.) means the parameter was constant between primary sampling occasions only. Figures showi ng models with at least 10 % of the total weight based on AIC (Akaike’s Information Cr iterion; see Appendix A) are included in this section, while figures incl uding all models used in th e analyses are in Appendix B. Conventional Robust Design – Five models all had model weights greater than 10 % (Table 2). Among them, there is significant support fo r a “time-since-marking” effect (symbolized by “a1”), but only weak (17 %) support for a patch size eff ect (symbolized “PS”) on survival. A timesince-marking effect means that survival was lower in the month following initial capture and marking than in all later months. Thr ee of the top five mode ls incorporated time

PAGE 29

19 since marking, but only the model ranked fourth supported an effect of patch size. All five models included either Markovian migra tion (where the probability of an individual being available for capture depended on its av ailability during the previous session) or random migration, with Markovian movement finding greater support overall. “No migration” models found zero support (see Appendi x B, Table 17). Capture and recapture rates in the top models were constant over primary as we ll as secondary sampling periods. An explanation for the zero deviances in these models is provided in the Discussion section. Although and parameters for the last two sampling intervals had been constrained to be equal in Markovian models, which is recommended to improve estimability (Kendall 2007), one of five "s and two of four 's in these models were inestimable. Estimates for the top model are shown in Appendix B, Table 18. Table 2 2004 Top Models Using Conventional Robust Design ranked by AIC Delta AICc Model Model AICc AICc Weight Likelihood #Par D eviance -------------------------------------------------------------------------------------------------------------------------------{Phi (a1) p,c (all .) N (.) Markov migr} -351.812 0.00 0.29647 1.0000 11.000 0.000 {Phi (.) p,c (all .) N (.) random migr} -350.949 0.86 0.19253 0.6494 9.000 0.000 {Phi (a1) p,c (all .) N (.) random migr} -350.862 0.95 0.18429 0.6216 10.000 0.000 {Phi (a1+PS) p,c (all .) N (.) Markov migr} -350.717 1.10 0.17147 0.5784 13.000 0.000 {Phi (.) p,c (all .) N (.) Markov migr} -350.306 1.51 0.13959 0.4708 11.000 0.000 -------------------------------------------------------------------------------------------------------------------------------Pradel Robust Design with Individual Covariates – A single model [Phi, lambda (p atchsize*t) p,c (.)] had all the support based on a weight of 0.999 (Appendix B, Table 19). Bo th survival and population growth rate changed with patch size and over time, and ther e was an interaction between the two variables. Capture and recapture rates were c onstant over secondary sampling occasions, but

PAGE 30

20 varied between primary sampling occasions. One lambda parameter could not be estimated. Unlike the results for the conventiona l robust design type, abundance varied over time in this top model. Pradel Robust Design with Groups – As for the previous models usin g individual covariates, one model [Phi, lambda (gr*t) p,c,N (gr)] had exclus ive support with a model wei ght of 0.997, and it is again the model incorporating effects of patch size (h ere a group effect base d on individuals from large patches versus small patches). Surv ival and population growth varied between groups and over time with a group-time inter action, and group effects were detected for capture and recapture rates and for abundance N (Appendix B, Table 20). Estimates were higher in small patches than la rge ones, which was contrary to expectations, at least for survival and population growth. Two la mbda parameters were inestimable. Link-Barker with Individual Covariates – Two models carried more than 10 % of the total weight (Table 3). Time variation and patch size effect with interaction were i ndicated for recruitment in both models. Survival varied over time but not with patch size. Recapture rate varied over time in the first model (43 % weight), but first and last recaptu re parameters were inestimable. In the model ranked second (33 % weight), recapture rate was constant. Table 3. 2004 Top Models Using Link Barker with Individual Covariates Delta AICc Model Model AICc AICc Weight Likelihood #Par Devia nce -------------------------------------------------------------------------------------------------------------------------------{Phi (t) p (t) f (patchsize*t)} 1346.410 0.00 0.43380 1.0000 23.000 1297.458 {Phi (t) p (.) f (patchsize*t)} 1346.979 0.57 0.32629 0.7522 19.000 1306.969 ------------------------------------------------------------------------------------------------------

PAGE 31

21 Link Barker with Groups – Similar to the Link Barker models with covariates, variation through time and group effect were strongly supported with regard to recruitment. Th e top model, carrying a weight of 39 %, indicated a group effect only on survival, while models ranked second and third incorporated varia tion over time but no group effect Together the second and third models had one third of the model weight (Table 4). Capture ra te again varied over time in the top two models but showed no group effect as in the Pradel robust models. Estimates for recruitment were lower in small patches than in large ones, while survival estimates were slightly higher in sm aller patches than larger ones. Table 4. 2004 Top Models Using Link Barker with Groups Delta AICc Model Model AICc AICc Weight Likelihood #Par Deviance --------------------------------------------------------------------------------------------------------------------{Phi (gr) p (t) f (gr*t)} 1351.883 0.00 0.38923 1.0000 19.000 169.329 {Phi (t) p (t) f (gr*t)} 1353.494 1.61 0.17392 0.4468 22.000 164.252 {Phi (t) p (.) f (gr*t)} 1353.805 1.92 0.14891 0.3826 18.000 173.456 --------------------------------------------------------------------------------------------------------------------Summary of Parameter Estimates for 2004 Data – Table 5 below shows parameter estimates and standard errors obtained from the five model types. Abundance N is defined in all cases as the total number of animals in the population exposed to sampling efforts (Ams trup et al. 2005). Where best-fit models allowed parameters to vary over time, the estimates represent average values. The two values shown for in the conventional Robust Design are estimates for the first month after initial capture and marki ng (top) and all later sampling in tervals combined (bottom).

PAGE 32

22 In all other cells containing two estimates, the upper and lower values stand for large and small patches, respectively. Table 5. Parameter Estimates for 2004 Data Model Type p c N f Conventional Robust Design 0.64; 0.05 0.69; 0.04 0.60; 0.03 0.69; 0.02 72; 1 N/A N/A 0.22; 0.09 not estimated Pradel Robust with Individ. Covariates 0.74; 0.06 0.53; 0.07 0.68; 0.05 68; 6 0.98; 0.15 N/A N/A N/A Pradel Robust with Two Groups 0.60; 0.08 0.84; 0.04 0.52; 0.04 0.75; 0.03 0.68; 0.03 0.69; 0.03 32; 1 40; 0 0.96; 0.15 1.33; 0.18 N/A N/A N/A Link Barker with Individ. Covariates 0.67; 0.08 0.79; 0.08 N/A N/A N/A 0.20; 0.08 N/A N/A Link Barker with Two Groups 0.69; 0.08 0.70; 0.08 0.80; 0.07 N/A N/A N/A 0.29; 0.12 0.15; 0.08 N/A N/A Several trends became evident in models where parameters were time-dependent. Abundance varied with time in the Pradel r obust design model with c ovariates. Estimates peaked at 87 individuals in April, the second month of tr apping, and decreased continually to 36 in September (Figure 3). Surviv al was time-dependent in top models of all types except the conventional robust design, wh ere it showed a time-since-marking effect instead. Similar patterns of an overall in crease from April through July or August followed by a sharp decline to September were evid ent in all time-variant models (Figure 4). An exception in the last sampling interval was the estimate for small patches using the Pradel robust design; it was higher than the estimate for the previous interval. In the Pradel robust design model with covariates, su rvival rate peaked at 0.94 in July and fell

PAGE 33

23 to 0.44 in September. In the Pradel robust design with two groups, survival for large patches peaked at 0.77 in July and decreased to 0.31, while for small patches it was about 1 until July and decreased to 0.61 by September. Link Barker models showed August peaks of 0.82 with individual covariates and 0.75 and 0.76 for large and small groups, respectively, followed by September lows of 0.40 with covariates and 0.57 and 0.58 for large and small groups, respectively. Trends in population growth rate were somewhat difficult to assess because several parameters were inestimable, but estimated values of fluctuated slightly between March and Augus t and fell sharply in September. Recruitment parameters for the first and last interv als were not estimable as a result of confounding with other parameters in fully time-var iant models. For the intervening months, values were at a maximum in the April-May interval, followed by lows in the next two intervals and a second peak in July-August (Figure 5). Figure 3. Abundance Estimates for March September, 2004 0 20 40 60 80 100 120 March April May June July Aug Sept Sampling periodAbundance

PAGE 34

24 Figure 4. Survival Estimates for March-September 2004 Using Different Model Types 0 0.2 0.4 0.6 0.8 1 1.2 MarchApr AprMay MayJune JuneJuly JulyAug AugSept Sampling periodSurvival rate Pradel Robust w/covariates Pradel Robust large patches Pradel Robust small patches Link Barker w/covariates Link Barker large patches Link Barker small patches Figure 5. Recruitment Rates Using Link Barker Models 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 April-May May-June June-July July-Aug Sampling intervalRecruitment rate Link Barker w/Covariates Link Barker w/Groups-large Link Barker w/Groups-small

PAGE 35

25 Results for 2005 Data Conventional Robust Design – Similar to the results for the previous year, all best-f it models included a timesince-marking effect on survival (Table 6). Support for a patch size effect was marginal (Appendix B, Table 23). Markovian migrati on was again strongly s upported in the top models (69 % model weight), but a “no migr ation” model also received some support, while random movement had zero support in this year. Capture and recapture parameters were constant over secondary sampling periods but varied between primary periods, and abundance was constant over time. Two out of three parameters were inestimable in the top model. Table 6. 2005 Top Models Using Conventional Robust Design Delta AICc Model Model AICc AICc Weight Likelihood #Par D eviance -------------------------------------------------------------------------------------------------------------------------------{Phi (a1) p,c (.) N (.) Markov migr} -1 69.681 0.00 0.52935 1.0000 20.000 0.000 {Phi (a1) gammas (.) p,c (.) N (.) Markov migr} -167.341 2.34 0.16436 0.3105 17.000 0.000 {Phi (a1) p,c (.) N (.) no migr} -166.620 3.06 0.11457 0.2164 15.000 0.000 -------------------------------------------------------------------------------------------------------------------------------Pradel Robust Design with Individual Covariates – Models ranked first and second showed only time variation in survival and population growth rate Only the model ranked third (10 % model weight) supported a patch size effect on (Table 7). Capture and recapture rates varied between sampling periods but were constant within them. The top m odel held N constant, but the second and third models (41 % combined model weight) showed variation over time in abundance.

PAGE 36

26 Table 7. 2005 Top Models Using Pradel Robus t with Individual Covariates Delta AICc Model Model AICc AICc Weight Likelihood #Par Deviance -------------------------------------------------------------------------------------------------------------------------------{Phi (t) lambda (t) p,c (.) N (.)} 263.740 0.00 0.50239 1.0000 23.000 215.600 {Phi (t) lambda (t) p,c (.) N (t)} 264.720 0.98 0.30768 0.6124 27.000 207.767 {Phi (t) lambda (PS+t) p,c (.) N (t)} 266.875 3.14 0.10477 0.2085 28.000 207.697 --------------------------------------------------------------------------------------------------------------------Pradel Robust Design with Groups – The top two models had almost equal weights (Table 8). Both showed variation over time in survival and population growth, but there was no support for a group effect. The top model showed time variation in abunda nce while the second one held N constant. Table 8. 2005 Top Models Using Pradel Robust with Groups Delta AICc Model Model AICc AICc Weight Likelihood #Par Devia nce -------------------------------------------------------------------------------------------------------------------------------{Phi (t) lambda (t) p,c (.) N (t)} 670.382 0. 00 0.46722 1.0000 27.000 613.429 {Phi (t) lambda (t) p,c (.) N (.)} 670.440 0. 06 0.45405 0.9718 23.000 622.300 -------------------------------------------------------------------------------------------------------------------------------Link-Barker with Individual Covariates – Only weak support (16 %) existe d for a patch size effect on recruitment in best-fit models of this type, although the two top models held parameters constant. No patch size effect was indicated for survival, but time va riation had support. Recapture rate was constant over time (Table 9).

PAGE 37

27 Table 9. 2005 Top Models Using Link Barker with Individual Covariates Delta AICc Model Model AICc AICc Weight Likelihood #Par Deviance --------------------------------------------------------------------------------------------------------------------{Phi (t) p (.) f (.)} 918.824 0.00 0.46775 1.0000 7.000 904.453 {Phi (.) p (.) f (.)} 920.705 1.88 0.18267 0.3905 3.000 914.626 {Phi (t) p (.) f (PS)} 920.920 2.10 0.16401 0.3506 8.000 904 .442 --------------------------------------------------------------------------------------------------------------------Link Barker with Groups – The top five models all had sup port, with the first one being more than twice as likely as the second and the remaining four all having similar weights (Table 10). The top model did not support a group effect in eith er survival or recruitment, but models ranked second through fourth showed 31 % suppor t for a group effect on survival and 31 % support for a group effect on recruitment. Recapture rates we re constant in all models and survival varied over time in the fi rst and third models. Similar to results in 2004, models indicating group effects showed higher estimates for survival in small versus large patches and lower estimates for recruitment in small versus large patches. Table 10. 2005 Top Models Using Link Barker with Groups Delta AICc Model Model AICc AICc Weight Likelihood #Par Deviance --------------------------------------------------------------------------------------------------------------------{Phi (t) p (.) f (.)} 918.824 0. 00 0.36181 1.0000 7.000 94.339 {Phi (gr) p (.) f (.)} 920.431 1.61 0.16199 0.4477 4.000 102.186 {Phi (t) p (.) f (gr)} 920.515 1.69 0.15540 0.4295 8.000 93.922 {Phi (gr) p (.) f (gr)} 920.636 1.81 0.14626 0.4042 5.000 100.324 {Phi (.) p (.) f (.)} 920.705 1.88 0.14130 0.3905 3.000 104.512 -----------------------------------------------------------------------------------------------------

PAGE 38

28 Summary of Parameter Estimates for 2005 Data – Table 11 shows parameter estimates and standard errors for the 2005 data. As for the 2004 data set, abundance is the total numbe r of animals exposed to sampling efforts, estimates are mean values where parameters varied over time, and two values in a cell are estimates either for time-since-marking c ohorts (Conventional Robust Design) or for large vs. small patches (group effect models). Table 11. Parameter Estimates for 2005 Data Abundance varied over time in seve ral best-fit models of the Pradel robust types, although parameters for September, the last sampling period, were inestimable. In the models using covariates, model-averaged esti mates were highest in April with 77 individuals and fell to 64 by August (Figure 6). Th e decrease in numbers was not as dramatic as in 2004, but it showed a similar trend. Prad el robust models using groups resulted in Model Type p c N f Conventional Robust Design 0.65; 0.12 0.92; 0.17 0.49; 0.06 0.50; 0.06 72; 3 N/A N/A 0.20; 0.17 not estimated Pradel Robust with Individ. Covariates 0.64; 0.07 0.47; 0.07 0.50; 0.06 71; 5 0.84; 0.11 N/A N/A N/A Pradel Robust with Two Groups 0.63; 0.07 0.63; 0.07 0.49; 0.06 0.49; 0.06 0.50; 0.06 0.50; 0.06 40; 2 0.84; 0.11 0.84; 0.11 N/A N/A N/A Link Barker with Individ. Covariates 0.65; 0.08 0.80; 0.04 N/A N/A N/A 0.19; 0.04 N/A N/A Link Barker with Two Groups 0.64; 0.08 0.67; 0.08 0.80; 0.04 0.80; 0.04 N/A N/A N/A 0.20; 0.03 0.18; 0.04 N/A N/A

PAGE 39

29 much lower estimates than those obtained using covariates (see Table 11 and Figure 6). The values decreased from 44 individuals in April to 35 in August and are probably faulty. Models that inco rporated no group effect on a bundance obtained the greatest support, but apparently program MARK cal culated abundance for one group for these models rather than for the en tire population. Two models in the set that incorporated a group effect on abundance but received no s upport based on AIC values showed estimates of 42 and 40 for large groups and 30 and 26 for small groups, totaling 72 and 69, respectively, and approximating the estimates fr om models using covariates. As in 2004, time variation in survival wa s found in all models except th e conventional robust design (Figure 7). Only one set of survival estimat es was obtained for the Pradel robust design using groups because only models without a gr oup effect on survival received support. Estimates peaked in the MayJune interval, whereas in th e previous year maximum values were found one to two months later. Ov erall, however, similar trends of increasing survival from spring into summer followed by a decline into early fall could be observed in both years. Survival estimates were sli ghtly higher in small pa tches than in large patches in the Link Barker models, but ther e was considerable ove rlap of confidence intervals. Estimates for population growth ra te were virtually iden tical in all Pradel robust models. They peaked at 0.97 in the May-June interval and decreased to 0.57 and 0.56 in models with covariates and models with groups, respectively (Figure 8). Recruitment did not vary with time in best-fit models for 2005, so that no trends could be observed in this parameter.

PAGE 40

30 Figure 6. Abundance Estimates for April August, 2005 0 10 20 30 40 50 60 70 80 90 Apr May June July Aug Sampling periodAbundance Pradel Robust w/Covariates Pradel Robust w/Groups Figure 7. Survival Estimates for April Sept ember 2005 Using Different Model Types 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 April-May May-June June-July July-Aug Aug-Sept Sampling intervalSurvival rate Pradel Robust w/Covariates Pradel Robust w/Groups Link Barker w/Covariates Link Barker w/Groupslarge Link Barker w/Groupssmall

PAGE 41

31 Figure 8. Estimates for Population Growth Rate in 2005 0.2 0.4 0.6 0.8 1.0 1.2 Apr-May May-June June-July July-Aug Aug-Sept Sampling intervalPopulation growth rate Pradel Robust w/Covariates Effect of Limiting Number of Capture Occasions – Figure 9 shows a plot of the numb er of cumulative first-time captures over the only six-day trapping session conducte d in June of 2005. The capture rate clearly did not level off by the third sampling occasion, lending strong support to the idea that not all available individuals were trapped in other months when only three consecutive sampling occasions were used. As a result, abundance estimates are most likely biased low. Figure 9. Cumulative First Captures Ov er A Six-Day Sampling Period 0 2 4 6 8 10 12 14 16 18123456 Sampling occasionCumulative # of first captures

PAGE 42

32 Modeling Heterogeneity in Capture Probabilities Using the same six days of data from June 2005, a comparison between models using “closed captures” and “full closed captu res with heterogeneity” data types revealed that incorporating heterogeneity in capture probabilities into a model resulted in significantly better fit than the m odel without heterogeneity (Table 12). The model ranked second is the best supported one for the clos ed captures data type but the same model using the closed captures with heterogeneity data type carri es more than 1500 times as much weight. This model type relaxes the assumption of equal catchability among individuals, but it requires more than three sec ondary sampling occasions and could not be used in this study. Table 12. June 2005 Models for Six Day Sampling Period Delta AICc Model Model AICc AICc Weight Likelihood #Par Deviance --------------------------------------------------------------------------------------------------------------------{P,c (t) pt=ct full closed caps w/het} 88.794 0.00 0.99920 1.0000 20.000 52.291 {P,c (t) pt=ct closed caps} 103.4 68 14.67 0.00065 0.0007 11.000 86.654 {P,c (t) p(t)=p(t-1) closed caps} 106.441 17.65 0.00015 0.0000 11.000 89.627 {P,c (.) closed caps} 117.129 28.33 0.00000 0.0000 3.000 116.994 --------------------------------------------------------------------------------------------------------------------Molecular Analyses – Twenty-four of the 86 tissue sa mples collected from Florida mice were used in PCR amplifications. Of those, 16 samples we re successfully amplified at least once and some as many as five times. With the excep tion of PO-26, all prim er pairs successfully amplified P. floridanus DNA at least once. Table 13 show s the initial annealing temperatures (Ta) used in successful trials. PCR produc ts were run on agarose gels along with a

PAGE 43

33 positive control (DNA from Peromyscus polionotus obtained from the Peromyscus Genetic Stock Center, University of South Caro lina). This procedure confirmed that the fragments that had been obtaine d were in the correct size range (Figure 10), but could not identify them as the desired microsatellites A sequencing step would have been necessary to determine nucleotide base sequences and thereby allow identification of the fragments, but because of time constraints this step was never reached. After several months of running PCR under various conditions wit hout repeatable results DNA analyses were abandoned. Table 13. Primer-Specific Annealing Temperat ures Used in PCR Amplifications Primer Initial Ta (C) PO 9 62 PO 26 none found to work PO3 68 62 PO 31 58 PO 71 60 BW4 28 57 BW4 45 60 BW4 93 55 BW4 112 60 BW4 137 57 BW4 178 50 BW4 200 61 BW4 234 62 Figure 10. Bands of PCR Products Obtained with Primers PO-9 and BW4-28

PAGE 44

34 DISCUSSION Zero Model Deviances – Deviances for all models run unde r the conventional robust de sign data type were reported as zero (Tables 2 and 6). Accord ing to a response posted by G. White on the Analysis Forum, an online discussion forum for the MARK program, this is because a saturated model likelihood has not yet been computed for the robust design model and a negative constant is left out of the likeli hood to speed up computation. Leaving out the constant leads to positive likelihoods, which in turn result in negative deviances. The lack of a saturated model combined with negative deviances causes deviances to be reported as zero. The AIC values, however, do scale appropriately and can be used to assess the fit of models. Temporary Effect of Marking on Survival Models run under the robust de sign with a time-since-marking effect (“a1”) on survival received strong support for 2004 a nd sole support for 2005 (Tables 2 and 6). Probabilities of survival were 0.64 (+/0.05) for the month immediately following initial capture and tagging versus 0.69 (+/0.04) for all later months combined in 2004 (Table 5) and 0.65 (+/0.12) and 0.92 (+/0.17), respectively, in 2005 (Table 11). These differences could mean that the process of ear tagging caused a temporary reduction in survival, but there could also be another explana tion. Because survival in juveniles is often lower than in adults (Schwarz and Seber 1999, Dinsmore et al. 2003, Layne 1992,

PAGE 45

35 Gardali et al. 2003), I examined the possibility that first-time captures may have been concentrated at a time of year when juven iles may have been abundant relative to other times and the difference in survival could therefore be between age classes and not a result of marking. The majority of first-ti me captures took place in March and April of both years, when trapping was initiated. Resu lts of the 2004 data analyses showed a peak in abundance in April (Figure 3) and a peak in recruitment in the April-May interval (Figure 4), indicating that lowe r juvenile survival may have been a factor during that time period. In 2005, however, recruitment was cons tant in best-fit models and abundance did not vary as much as in the previous year (F igure 5), but the difference in survival estimates between time-since-marking cohorts in conventional robust design models was greater than in 2004 (Tables 5 and 11). This result indicates that th e process of ear tagging temporarily affected survival in 2005, but the cause for the difference in survival in 2004 cannot be determined. Infections were observed in a few individuals over the course of the study, and possibly mice could be distracted by the tags and could therefore be more susceptible to predators. I chose th is method of identification because it seemed less invasive than toe clipping and my initia l attempts at using an ear punching code resulted in severe damage to the ears of mi ce (albeit my inexperience with the technique may have played a role). Migration Results of conventional robust design models showed strong support for temporary migration of individuals into and out of sampling areas. Best-fit models in 2004 incorporated both Markovian and random movement while in 2005 only Markovian migration was supported. The difference in interpreta tion between the two is that Markovian

PAGE 46

36 models assume the probability of being avai lable for capture depends on an individual’s availability in the previous session, while in random mode ls it does not (Kendall 2007). In other words, the Markovian model assumes that an animal “remembers” whether or not it was in the trapping ar ea in the previous session. When modeling Markovian migration, no constraints are placed on migration parameters ( and ") if survival is timeinvariant. For random migration, 's are set equal to "s, the interpretation being that the probability of being out of the study area is the same whether an animal was in or out of the study area during the previ ous occasion. The greater pa rameterization of Markovian models (probably combined with sparseness of data) led to inestimability of several migration parameters and unusually large conf idence intervals. The probability of temporary emigration ( ") calculated for the 2004 data from five out of six parameters was 0.22 (+/0.09), but four out of five immigration parameters we re inestimable. Deriving the probability of immigration (1') from a single was deemed unreliable and therefore no value was reported. For 2005, was estimated as 0.20 (+/0.17), and two out of three 's were again inestimable. Although obt aining values for migration parameters was problematic, AIC ranking strongly indicated that temporary migration existed in this population. Whether it can be interpreted as m ovement out of habitat patches or simply as movement out of the trapping area to anothe r part of the same patch is not clear, but the previously reported dispersal of 18 individuals between patches (see Results) strengthens the argument that metapopulation dynamics may exist. Although 18 individuals is only a small portion of the animal s encountered, I suspect that a higher number of dispersing mice would have been obser ved had trapping been continued into the winter. This suspicion is ba sed on the fact that juveniles and subadults usually disperse

PAGE 47

37 from their natal areas in search of mate s and other resources (Swilling and Wooton 2002, Zug et al. 2001). According to Layne (1966), the vast majority of pregnancies in a population of P. floridanus he studied in Alachua County occurred during September and October (see Figure 11). Assu ming reproduction also occurs mainly during those months in the population studied here, offspring would not be weaned until late fall or early winter, at which time they would begin to di sperse. Because trapping was only conducted from March through September, I was not able to observe whether or not this occurred. However, even a small number of dispersi ng animals may be sufficient to augment the gene pools of subpopulations and increase fitnes s. Dispersal itself does not provide insight into whether or not the individuals i nvolved are interbreed ing with other subpopulations. Genetic analyses, on the other hand, can determine the degree of gene flow among subpopulations and the relatedness of individuals from neighboring patches. Figure 11. Seasonal Variation in Pre gnancies in Alachua County 0 10 20 30 40 50 60 70 JanFebMarAprMayJuneJulyAugSeptOctNovDec MonthPercent pregnant females ( Ada p ted from La y ne 1966 )

PAGE 48

38 Parameter Estimates – With the exception of abunda nce estimates obtained for 2005 using the Pradel robust design with groups, which must be vi ewed with caution (see Results), estimates from all model types and for both years indi cated that approximately 70 individuals on average were exposed to sampling efforts in each month (Tables 5 and 11). In markrecapture studies, traps are usually arrange d on rectangular grids and the sampling area can be calculated relatively easily, but b ecause of the unusual arrangement of traps around herp arrays in the current study, the ar ea sampled was more difficult to determine. A rough estimate of the area per trap array is 200 m2, and with 15 arrays the total sampling area would have been approximately 3000 m2. Overall, abundance estimates are likely biased low as a result of the limited nu mber of secondary sampling occasions (see Results). In addition, low estimates of a bundances as well as recruitment and population growth rates are likely a resu lt of not having included the peak reproductive period in the overall sampling period. Finally, it has been shown that heterogeneity in capture probabilities is strongly indicated for this population, and hete rogeneity has been known to produce unreliable abundance estimates (Po llock and Alpizar-Jara 2005, Conn 2006). In both 2004 and 2005, time-variant models indicate d that the number of mice was high in April and decreased into the fall (Figures 3 a nd 6), which probably reflects a minor peak in reproduction around Februa ry (see Figure 11). Average monthly survival estimates were between 0.67 (+/0.08) and 0.74 (+/0.06) in 2004 and between 0.63 (+/0.07) and 0.66 (+/0.08) in 2005 (see Tables 5 and 11). In conventional robust design models, av erages of survival per cohort were 0.67 (+/0.05) in 2004 and 0.79 (+/0.15) in 2005. The outlier in 2005 was the value for the

PAGE 49

39 second cohort (0.92 +/0.17), but estimates for both cohor ts in this year showed very large confidence intervals because best-fit models incorporated Markovian movement with numerous parameters. Values were si milar when comparing survival estimates between models using covariates and those us ing groups. In terms of trends over time, survival rate estimates in both years were low between Ap ril and May, increased during the summer and declined substa ntially by September (Figures 4 and 7). Low survival in the April-May interval is probably the result of a relatively high ratio of juveniles born around March (see minor peak in February pregnancies in Figure 11) and weaned by April. The drop in survival rates in September might be explained by females retreating underground in preparation for the peak re productive period and being unavailable for capture. Recruitment rates averaged 0.21 (+/0.09) in 2004 and 0.19 (+/0.04) in 2005. All Link Barker models for 2004 showed peaks in the April-May interval (Figure 5), correlating with the peak in abundance in April of th at year. The trend in recruitment rates for 2004 approximates the trend in pregnancies in the Alachua County population shown in Figure 11, but with a one-to-two month lag. Gestation time for P. floridanus is thought to be 23-24 days (Layne 1968b), followed by another three weeks before newborns would be weaned and begin to be encountered in traps. Top models for 2005 held recruitment constant over time, which is consis tent with the smaller decrease in abundances over time for this year compared with the pr evious one (Figure 6). Only models ranked fourth (with 7 % model weight) and lower show ed a decrease in recruitment similar to the trend seen in 2004. Link Barker models w ith covariates resulted in somewhat lower survival estimates than those using groups in 2004 (Table 5), but in 2005 estimates were

PAGE 50

40 very similar (Table 11). Population growth rates were lower in models using covariates versus groups for 2004, but were identical between model types for 2005. Estimates indicated that the population was increasing on a whole in 2004, but decreasing in 2005. However, because population growth is partly a function of recruitment and recruitment rates obtained during the st udy did not include peak re production, yearly population growth rates are likely higher. The greatest differences observe d among model types were in estimates of recapture probabilities be tween Link Barker and robust desi gn models. Link Barker models estimate recapture probabilities although the parameter is sy mbolized as p (which in robust design models is the probability of fi rst-time capture) and the resulting estimates should therefore be compared with recapture probability c in robust design models (Tables 5 and 11). Link Barker estimates are much higher th an those obtained with the robust design type (0.80 in both years fo r Link Barker vs. 0.69 in 2004 and 0.50 in 2005 using robust design) because estimates are de rived from pooled data in the case of Link Barker models. Link Barker estimates the probability of being captured at least once during several consecutive sampling occasions (three in the current study) as opposed to the robust design, which estimates recapture probabilities between two consecutive occasions. Using models with group effects proved valuable in unders tanding the direction of observed differences. Based on Hokit and Bran ch (2003), it was expe cted that survival, population growth rate and recruitment would al l be lower in smaller habitat patches than larger ones, but in both years only recruitmen t followed this trend while the opposite was true where differences in survival and populatio n growth were indicate d (Figures 12-14).

PAGE 51

41 Figure 12. Estimates of Recruitment Rates in Large vs. Small Patches 0.0 0.1 0.2 0.3 0.4 0.5 20042005 YearRecruitment rate large patches small patches Figure 13. Estimates of Population Growth Rates in Large vs. Small Patches 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 20042005 YearPopulation growth rate large patches small patches

PAGE 52

42 Figure 14. Estimates of Survival Rates in Large vs. Small Patches 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 20042005 YearSurvival rate Link Barker large patches Link Barker small patches Pradel Robust large patches Pradel Robust small patches An explanation for these unexpected findings mi ght be that for some reason predation is higher in large scrub patches than in small ones. It is conceivable th at predators that can easily move between patches may spend more time in large ones, where there is greater selection and/or overall abundan ce of prey. Coachwhip snakes, for example, tended to be found more often in larg er patches than smaller ones (B Halstead, personal communication). If this theory is true, recruitment ra te at the study site might not have been noticeably affected because it was already quite low (~ 0.2) duri ng the months when the study took place. Survival rate, how ever, was about 0.7 and diffe rences between groups would have been more easily discernable. Because population growth rate is the sum of survival and recruitment rates and recruitment was relatively low during the entire study, population growth would have been high where survival was high and low where survival was low.

PAGE 53

43 Future Direction The mark-recapture analyses disc ussed here have shown th at habitat patch size was associated either positively or negatively with different demographic parameters of P. floridanus They also indicated a temporary eff ect of ear tagging on su rvival and showed support for migration between habitat patches. However, the analyses fell short of determining whether or not a metapopulation struct ure exists in the study population. I had originally planned to address this question us ing genetic analyses of microsatellites, but because of time constraints and the need to use non-specific primers that proved to be of limited utility, I was forced to abandon this portion of my study. PCR annealing temperatures given in the original literature from which the primers were obtained did not always lead to successful or consistent amplif ication, so that numerous attempts were necessary in order to find th e conditions under which the st udy organism’s DNA could be amplified. Eventually, however, all primers ex cept one were successfully used in PCR, and primer PO-26 may possibly have work ed with continued persistence. The analyses of mark-recapture data showed that habitat area was associated with demographic parameters of the study populati on. However, had no associations been evident, it would not have been clear whether th ey did not in fact ex ist or whether sourcesink dynamics (Pulliam 1988) might be counter acting the relationships. Even with the current results, metapopulation dynamics may s till be at play and may lessen or increase effects of habitat patch size. The rate and direction of movement of individuals among subpopulations can strongly affect variati on in abundances and population viability (Diffendorfer et. al. 1995). For example, one scenario might be that large patches serve as sources for small patches, animals migrat e to the small patches and because predation

PAGE 54

44 is lower in small patches survival is higher th ere, but without new arrivals from the large patches, the smaller ones could not persist. Future research employing genetic analyses should be conducted on the study population in order to determine the degree and kinds of interactions among subgroups. The primer s discussed here may serve as a starting point from which customized primers could be designed that would ensure more consistent results in PCR amplification.

PAGE 55

45 REFERENCES Analysis Forums. http://www.phidot.org/forum/index.php Accessed on May 14, 2007. Bennett, A. F. 1990. Habitat corridors and the conservation of small mammals in a fragmented forest environment. Landscape Ecology 4:109-122. Burkey, T. V. 1995. Extinction rates in archip elagoes: implications for populations in fragmented habitats. Conservation Biology 9(3):527-541. Carlton, M. D. 1980. Phylogene tic relationships in neot omine-peromyscine rodents (Muroidea) and a reappraisal of the di chotomy within New World Cricetinae. Miscellaneous Publications Museum of Zool ogy, University of Mi chigan 157:1-146. Campbell, N. A. and J. B. Reece 2002. Conservation Biology. Pages 1224-1245 in Biology 6th Edition. Benjamin Cummings, San Francisco, California. Clark, A. M., B. W. Bowen and L. C. Bran ch 1999. Effects of natural habitat fragmentation on an endemic scrub lizard ( Sceloporus woodi ): an historical perspective based on a mitochondrial DNA gene genealogy. Molecular Ecology, 8(7):1093-1104. Conn, P. B., A. D. Arthur, L. L. Bailey and G. R. Singleton 2006. Estimating the abundance of mouse populations of known size: pr omises and pitfalls of new methods. Ecological Applications, 16(2):829-837 Cooch, E. and G. C. White 2007. A Gentle Introduction, 6th Edition. www.phidot.org/software/mark/docs/book/ Accessed January 15, 2007.

PAGE 56

46 Diffendorfer, J. E., M. S. Gaines and R. D. Holt 1995. Habitat fragmentation and movements of three small mammals ( Sigmodon, Microtus and Peromyscus ). Ecology 76(3):827-839. Dinsmore, S. J., G. C. White and F. L. Knopf 2003. Annual surviv al and population estimates of Mountain Plovers in S outhern Phillips County, Montana. Ecological Applications 13(4): 1013-1026. Fahrig, L. and G. Merriam 1994. Cons ervation of fragmented populations. Conservation Biology 8(1):50-59. Frankham, R. 1998. Inbreeding and extinction: island populations. Conservation Biology 12(3): 665-67. Gardali, T., D. C. Barton, J. D. White and G. R. Geupel 2003. Juvenile and adult survival of Swainson’s Thrush ( Catharus ustulatus ) in coastal California: annual estimates using capture-recapture analyses. The Auk 120(4):1188-1194. Gaines M. S. and L. R. McClenaghan, Jr. 1980. Dispersal in small mammal populations. Annual Review of Ecology and Systematics 11:163-196. GenBank. National Center for Biotechnology Information. http://www.ncbi.nlm.nih.gov/ Accessed on December 12, 2005. Hartman, B. 1992. Major terrestrial a nd wetland habitats. Pages xvii-xxviii in S. R. Humphrey, editor, Rare and Endangered Biota of Florida. Vol. 1: Mammals University Presses of Florida, Gainesville, Florida. Hill R. W. 1983. Thermal physiology and energetics of Peromyscus ; ontogeny, body temperature, metabolism, in sulation, and microclimatology. Journal of Mammalogy 64(1):19-37.

PAGE 57

47 Hokit, D. G., B. M. Stith and L. C. Branch 1999. Effects of landscape structure in Florida scrub: a population perspective. Ecological Applications 9:124-134. Hokit, D. G. and L. C. Branch 2003. Habita t patch size affects demographics of the Florida scrub lizard ( Sceloporus woodi ). Journal of Herpetology 37 (2):257-265. Hooper, E. T. 1968. Classification. Pages 27-74 in J. A. King editor, Biology of Peromyscus (Rodentia) Spec. Publ., American Soci ety of Mammalogists, 2:1-593. Huggins, R. M. 1989. On the statistical analysis of capture experiments. Biometrika 76: 133-140. Humphrey, S. R., J. F. Eisenberg and R. Fr anz 1985. Possibilities for restoring wildlife of a longleaf pine savanna in an abandoned citrus grove Wildife Society Bulletin, 13: 487-496. IUCN 2006. 2006 IUCN Red List of Threatened Species. www.iucnredlist.org Accessed on July 8, 2007. Jones, C. A. 1990. Microhabitat use by Podomys flori danus in the high pine lands of Putnam County, Florida. Ph. D. Thesis, University of Florida, Gainesville, Florida. Jones, C. A. and R. Franz 1990. Use of gopher tortoise burrows by Florida mice ( Podomys floridanus ) in Putnam County, Florida. Florida Field Naturalist 18:45-51. Kendall, W. L. 2007. The Robust Design. Pages 475-513 in E. Cooch and G. C. White, editors, A Gentle Introduction, 6th Edition. www.phidot.org/software/mark/docs/ book/ Accessed January 15, 2007. Kendall, W. L., K. H. Pollock and C. Br ownie 1995. A likelihood-based approach to capture-recapture estimation of demogra phic parameters under the robust design. Biometrics 51:293-308.

PAGE 58

48 King, J. A. 1968 Biology of Peromyscus (Rodentia). Spec. Publ., American Society of Mammalogists, 2:1-593. Kirland, Jr., G. and J. N. Layne 1989. Advances in the study of Peromyscus (Rodentia). Texas Tech University Press, Lubbock, Texas. Klug, W. S. and M. R. Cummings 2002. Chromosome Structure and DNA Sequence Organization. Pages 348-364 in Essentials of Genetics, 4th Edition. Prentice Hall, Upper Saddle River, New Jersey. Labate, J. A. 2000. Software for population gene tic analyses of molecular marker data. Crop Science 40:1521-1528. Layne, J. N.1966. Postnatal development and growth of Peromyscus floridanus. Growth, 30:23-45. Layne, J. N. 1968b. Ontogeny. Pages 148-253 in Biology of Peromyscus (Rodentia) (J.A. King, ed.). Spec. Publ., The American Society of Mammalogists, 2:1-593. Layne, J. N. 1978. Florida mouse. Pages 21-22 in J. N. Jayne, editor, Rare and Endangered Biota of Flori da. Volume 1: Mammals. University Presses of Florida, Gainesville, Florida. Layne, J. N. 1990. The Florida mouse. Pages 1-21 in C. K. Dodd, Jr., R. E. Ashton, Jr., R. Franz and E. Wester, editors, Burrow Associates of the Gopher Tortoise. Proc. 8th Ann. Meeting of the Gopher Tortoise Council, Florida Museum Natl. Hist., Gainesville, Florida. Layne, J. N. 1992. Florida mouse. Pages 250264 in S. R. Humphrey, editor, Rare and Endangered Biota of Florida. Vol. 1: Mammals. University Presses of Florida, Gainesville, Florida.

PAGE 59

49 Layne, J. N. and R. J. Jackson 1994. Burrow use by the Florida mouse ( Podomys floridanus ) in south-central Florida. American Midland Naturalist 131:17-23. Layne, J. N. and J. V. Griffo, Jr. 1961. Incidence of Capillaria hepatica in populations of the Florida deer mouse, Peromyscus floridanus J. Parasit 47:31-37. Lebreton, J. D., K. P. Burnham, J. Clobert and D. R. Anderson 1992. Modeling survival and testing biological hypothe ses using marked animals: a unified approach with case studies. Ecological Monographs 62 (1):67-118. Lukacs, P. 2007. Closed Population Capt ure-recapture Models. Pages 455-474 in E. Cooch and G. C. White, editors, A Gentle Introduction, 6th Edition. www.phidot.org/software/mark/docs/ book/ Accessed January 15, 2007. MacArthur, R. H. and E. O. Wilson 1967 The Theory of Island Biogeography Princeton University Press. Princeton, New Jersey. McCoy, E. D. and H. R. Mushinsky 1992. Rarity of organisms in sand pine scrub habitat of Florida. Conservation Biology 6:537-548. Merriam, G. and A. Lanoue 1990. Corridor us e by small mammals: field measurement for three experimental types of Peromyscus leucopus Landscape Ecology 4:123131. Mullen, L. M., R. J. Hirschmann, K. L. Prin ce, T. C. Glenn, M. J. Dewey and H. E. Hoekstra 2006. Sixty polymorphic microsat ellite markers for the oldfield mouse developed in Peromyscus polionotus and Peromyscus maniculatus Molecular Ecology Notes 6:36-40. Myers, R. L. 1990. Scrub and high pine. Pages 150-193 in R.L. Myers and J. J. Ewel, editors, Ecosystems of Florida. University of Central Florida Press, Orlando, Florida.

PAGE 60

50 Nichols, J. D. 2005. Modern Open-populati on Capture-Recapture Models. Pages 88-123 in S. C. Amstrup, T. L. McDona ld and B. F. J. Manly, editors, Handbook of CaptureRecapture Analysis. Princeton University Press, Princeton, New Jersey. Osgood, W. H. 1909. Revision of the mice of the American genus Peromyscus. N. Amer. Fauna 28:1-285. Otis, D.L., K.P. Burnham, G. C. White a nd D. R. Anderson. 1978. Statistical inference from capture data on closed animal populations. Wildlife Monographs 62. Parker, P. G., A. A. Snow, M. D. Schug, G. C. Booton and P. A. Fuerst 1998. What molecules can tell us about populations: choosing and using a molecular marker. Ecology 79(2):361-382. Pollock, K. H. and R. Alpizar-Jara 2005. Cl assical Open-populati on Capture-Recapture Models. Pages 36-57 in S. C. Amstrup, T. L. McDonald and B. F. J. Manly, editors, Handbook of Capture-Recapture Analysis. Princeton University Press, Princeton, New Jersey. Prince, K. L., T. C. Glenn and M. J Dewey 2002. Cross-species amplification among peromyscines of new microsatellite DNA loci from the oldfield mouse ( Peromuscus polionotus subgriseus ). Molecular Ecology Notes, 2:133-136. Pulliam, H. R. 1988. Sources, sinks and population regulation. American Naturalist 132:652-661. Rave, E. H. and N. R. Holler 1992. P opulation dynamics of Beach mice ( Peromyscus polionotus ammobates ) in Southern Alabama. Journal of Mammalogy 73(2):347355.

PAGE 61

51 Ricklefs, R. E. and G. L. Miller 1999. Ecology, 4th Edition. W. H. Freeman and Company, New York, New York. Robbins, L. W. and R. J. Baker 1981. An assess ment of the nature of rearrangements in eighteen species of Peromyscus (Rodentia: Cricetidae ). Cytogenetics and Cell Genetics, 31:194-202. Schmutz, D. D. 1997. Translocation and microhabitat distri bution of Podomys floridanus on native upland and reclaimed mined sites M. S. Thesis, University of South Florida, Tampa, Florida. Schwarz, C. J. and A. N. Arnason 2007. Jolly-Seber Models in MARK. Pages 402-454 in E. Cooch and G. C. White, editors, A Gentle Introduction, 6th Edition. www.phidot.org/software/mark/docs/ book/ Accessed January 15, 2007. Schwarz, C. J. and G. A. F. Seber 1999. Estimating animal abundance: Review III. Statistical Science 14(4):427-456. Swilling, Jr. W. R. and M. C. Wooten 2002. Subadult dispersal in a monogamous species: the Alabama Beach mouse ( Peromyscus polionotus ammobates ). Journal of Mammalogy 83(1):252-259. Taylor, W. K. 1998. Florida Wildflowers in their Natural Communities University Press of Florida, Gainesville, Florida. Wilson, D. E., F. R. Cole, J. D. Nichols, R. Rudran and Mercedes S. Foster, editors 1996. Measuring and Monitoring Biological Diversity: Standard Methods for Mammals Smithsonian Institution Press, Washington, D.C. Yates, T. L., R. J. Baker and R. K. Barn ett 1979. Phylogenetic analys is of karyological variation in three genera of peromyscine rodents. Systematic Zoology 28:40-48.

PAGE 62

52 Zug, G. R., L. J. Vitt and J. P. Caldwell 2001. Spacing, Movements and Orientation. Pages 199-220 in Herpetology, An Introductory Biology of Amphibians and Reptiles, 2nd Edition. Academic Press, San Diego, California.

PAGE 63

53 APPENDICES

PAGE 64

54 Appendix A: Program MARK Program MARK is based on ma ximum likelihood estimation of the probabilities defining the occurrence of one or more ev ents (Cooch and White 2007). A likelihood function containing the parameter(s) in questi on is constructed and the value that maximizes the likelihood function, given the set of da ta, is then chosen as the best estimator for the parameter. The input data consist of a set of encounter histories, one for each individual, in the form of a row of dummy variables (0 and 1). Each dummy variable represents a sampling occasion, whereby a 0 indicates that the individual was not captured at that occasion and a 1 indicates th at it was captured. For example, based on a three-year study during which animals are mark ed and released on the first occasion and sampling is conducted once per year after that an encounter history of 101 would mean that the individual associated with that enc ounter history was captured and marked in the first year, not seen in the second year and recap tured in the third year. In all, four encounter histories are possible for the individuals in this hypothetical study: 111, 110, 101 and 100. Each of these encounter histories is associated with a certa in probability of occurrence, which in turn is based on two parameters: phit ( t, the probability of an individual surviving from occasion t to occasion t + 1) and pt (catchability, i.e. the probability that, if alive and in the sample at time t, an individual will be captur ed). In practice, the death of an individual cannot be distinguish ed from permanent emigration, so that is more correctly defined as “apparent survival”. In our example, the encounter history 101 would be defined by the probability 1 (1-p2) 2 p3, meaning that the individual survived to the second occasion (we know this because it was seen alive at the third occasion), was not captured at the second occasion (with probability 1 – p2), survived to the third occa-

PAGE 65

55 Appendix A: (Continued) sion and was seen alive at the third occas ion. Depending on how ma ny individuals were found to have a particular encounter histor y, each history would occur with a certain frequency. The problem addre ssed in MARK, then, is to es timate for which values of and p the probability of finding this set of en counter histories with the given frequencies would be maximized. Many models can be run using the MARK program, and different models derive estimates for different parameters. Cormack-Jolly-Seber (CJS) models estimate survival and catchability as in the above example. Jolly Seber (JS) models also estimate and p, but make the assumption of equal survival a nd catchability for all animals in the population, whether marked or unmarked, while CJS models make no assumptions about the unmarked population and assume these paramete rs to be equal only for marked individuals (Schwarz and Arnason 2007). The main difference between these models, then, is that estimates pertain only to the marked popul ation in CJS models, while they apply to the entire population in JS models. The overa ll assumption of equal catchability in the JS models allows for the estimation of additional parameters such as recruitment and population growth and size. The assumptions of equal catchability and survival among individuals are necessary for the estimation of parameters, but are often unrealistic, and methods have been developed that can relax one or both of these assumptions (Lebreton et al. 1992, Schwarz and Seber 1999). JS and CJS models are referred to as open population models because they allow changes in popula tion size over time in the form of births, deaths, immigration and emigration of indivi duals. Closed population models, on the other hand, assume a constant population size throughout the study. They estimate p (the

PAGE 66

56 Appendix A: (Continued) probability that an animal in the population will be captured for the first time), c (the probability of recapture given that an animal was captured at least once before) and population size or abundance N (Lukacs 2007). Because of the closure assumption it is appropriate to use these models only for data sets that were collected in a short time period, during which it can be assumed that there was virtually no change in population size. MARK supports a number of closed population data types. They can be grouped broadly into those of Otis et al. (1978), which have the a bundance estimates in the likelihood function, and those of Huggins (1989), which are conditioned on the number of animals captured, and N must be estimated as a derived parameter. Within these broad groups, models become increasingly complex as the equal catchability assumption is relaxed and/or uncertainty in identification (u sually a result of genotyping errors) is incorporated. MARK allows parameters to vary over time for all model types and between cohorts for some types (a cohort can refer to an age class or to a gr oup of individuals first captured and marked at a particular occasion) In the case of fully time-dependent models, where all parameters may change over time, several parameters are usually confounded (especially for first and last sampling occasions) and constraints must be placed on some of them (Cooch and White 2007). Gr oup effects (effects resulting from individuals being categorized in some way, e.g. by size or sex) and individu al covariates can be incorporated into some models, while others can handle data from multiple sources or strata (discrete locat ions or conditions, e.g. breedi ng vs. non-breeding, in which the marked individual may potentially be encountered).

PAGE 67

57 Appendix A: (Continued) The current study was designed to use models of the robust design type, which is a combination of open and closed population models (Kendall et. al. 1995, Kendall 2007). The main difference between an open model a nd the robust design model is that, instead of only one capture occasion between sampling intervals, there are multiple (k) occasions sufficiently close in time so th at the population can be assumed to be closed for that time period (Figure 15). These c onsecutive capture occasions a llow estimation of population size for each primary sampling period, while su rvival is estimated between sampling periods when the population is assumed to be open to births, deaths, immigration and emigration. In addition, the clas sical robust design models allow estimation of the probabilities of temporary emigration of individuals from the trappi ng area and immigration of marked animals back to the trapping area (K endall 2007). These calculations are possible because a distinction is made in the robust design model between the capture probabilities estimated in closed and open populat ion models. Capture probability p was previously defined as the probability that, if alive and in the sample at time t, an individual will be captured. However, just as the “apparent survival” estimated in open popuFigure 15. Basic Structure of the Robust Design Model -----------------------Time ---------------------------Primary sampling periods 1 2 3 Secondary sampling 1,2,… k1 1,2,… k2 1,2, … k3 periods (Adapted from Cooch and White 2007)

PAGE 68

58 Appendix A: (Continued) lation models is actually a product of true su rvival and the probability of not permanently emigrating, capture probability in these models is also a product of two other probabilities: that of being captured, conditional on be ing alive and in the sample, and that of being available for capture. Animals may be alive and in the sample, but may not be available for capture and will thus not be captured (e.g. if sampling of birds were conducted at a nesting site, only breeding bird s would be available for encounter and nonbreeding individuals would not be observed). In closed population models, on the other hand, p is the true probability of capture b ecause by definition N is constant in these models and individuals can neither enter nor exit the sample. With estimates for both “apparent” and “true” capture probabilities available from th e combined use of open and closed models in the robust design, gamma ( ), the probability of being available for capture during a particular primary sampling pe riod, can also be obtained. In fact, two gamma parameters are used in assessing temporary migration: and ". is the probability of being off the study area during a par ticular primary sampling period t given that the individual was also off the study area du ring the sampling period t – 1. It follows then that 1' is the probability that an individual enters the study ar ea between t – 1 and t, given that it was off the st udy area at time t – 1, which is a measure of temporary immigration. is the probability that an individual is off the st udy area and unavailable for capture during sampling period t given that it was in the study area at sampling period t – 1, a measure of temporary emigration. Temporary migration can be modeled to be random or Markovian. In the case of random movement, the probability of an individual being available for capture during a primary sampling period does not depend on whether

PAGE 69

59 Appendix A: (Continued) or not it was available in the previous sampling period, whereas for Markovian movement it does. A comparison between robust desi gn models using regular closed captures (after Otis et. al.) and Huggins clos ed captures data types showed similar results in terms of selection of the top model a nd parameter estimates, althoug h standard errors for population size estimates were slightly larger fo r the Huggins models. I chose to use regular closed capture models in my analyses. Ne xt, I compared models with a group effect (comparing data obtained from mice captured in large habitat patches versus in small patches) to models that used patch size as an individual c ovariate. In several cases, I adjusted the patch size covariate to reflect where the trap array was located within the patch. For example, arrays B1, B2, B3 and B4 were all located in a patch approximately 170 ha. in size (Figure 1), but B1 lay on a fingerlike projectio n of scrub away from the main portion of the patch and surrounded by fl atwoods. This relative isolation presumably reduced access to resources and c onspecifics (for mating purposes) for P. floridanus located in B1 relative to mice in other parts of the patch. B2 was located closer to the main portion, but separated from it by Tram Road (a sand road approximately 9 m in width), while arrays B3 and B4 lay nearer to the center of the patch. The adjustments I made were subjective, but I kept them small in order to err on the side of caution (see Table 14). Patch sizes were ad justed down for arrays locate d at the edge of a habitat patch and up when patches lay close to one a nother, effectively enlarging the sizes of

PAGE 70

60 Appendix A: (Continued) Table 14. Actual and Adjusted Habitat Patch Sizes Array Actual size of patch (ha.) Adjusted size of patch (ha.) O1 80 80 O2 80 70 H1 10 10 G1 2 2 B1 170 100 B2 170 150 B3 170 170 B4 170 170 A1 12 20 C1 1.5 1.5 F1 3 3 E1 2 2 D1 1.5 1.5 I1 40 40 I2 40 40 both fragments. A comparison of the two mode l sets again showed that the models with greatest support based on AIC values (see Goodness-of-Fit and Mode l Selection) were virtually identical, but estimates for abundan ce were lower in the two-group models. There were several disadvantages to using two-group models. One was that they required a greater number of parameters and some of them were inestimable because of small sample sizes. Another was that the cut-off be tween large and small patch sizes was rather arbitrary. On the other hand, Cooch and White (2007) point out that individual covariates are difficult to interpret in m odels estimating population growth ( ) and recruitment (f) because t = t + f t, and “while individual covariates c ould apply to survival rates, the recruitment parameter is not tied to any i ndividual – it is a popul ation-based, average

PAGE 71

61 Appendix A: (Continued) recruitment per individual in the population”. I decided to use both model types and compare estimates between them. Next, a comparison between Pradel robust models and simple Pradel models (e.g. Pradel Robust Surv ival and Lambda and Pradel Survival and Lambda) showed that top models were again id entical and estimates were very similar, so that there was no added benefit in also r unning the simpler models. The set of model types I decided on was as follows: The classical robust design model, becau se it allows modeling of temporary migration and age/cohort effects to test for differences in survival The Pradel Robust Survival and Lambda for estimation of population growth (models with group effect and with individua l covariates) Link-Barker models with group effect and individual covariates for estimation of recruitment rate f The Link-Barker model, like the Pradel mode ls, is based on the original JS model and estimates the per capita recruitment rate f. It uses pool ed capture histories, where secondary sampling occasions (see Figure 15) are poole d and an individual is either captured at least once during a primary sampling period or it is not captured. Goodness-of-Fit and Model Selection – Much of data analysis relies on the correct choice of a model; in other words, a model must be chosen which most adequately fits the data at hand (Lebreton et al. 1992, Cooch and White 2007). When using MARK, a set of models that seem biologically reasonable is chosen a priori. The next step that should be performed is to test the most global model of the set (i.e. the most parame terized one) for goodness-of-fit. If the most

PAGE 72

62 Appendix A: (Continued) global model adequately fits the data, it can be assumed that the others in the set do as well. All models can then be run and comp ared as to which one is most parsimonious, meaning which model represents the given data adequately and with the fewest possible parameters. If more than one model has significant support, es timates from the top models can be averaged. Unfortunately, none of the goodness-of-fit tests that have been developed to date (e.g. Bootstrap, Chi Square Release) can be used for robust design models (G. White, personal communication, W. R. Clark, Iowa State University, personal communication). Cooch and White (2007) recommend using Program RELEASE or the Parametric Bootstrap method on the fully pa rameterized CJS model that corresponds to the more complicated model being used, and if the CJS model is sup-ported by the test, one can proceed with the more complex mode l. Program RELEASE could not be used because of insufficient data; therefore I a pplied the Bootstrap method. Bootstrapping estimates the variance inflation factor, which is a measure of the lack of fit of the model to the underlying data. If equals 1, the model fits the data well, but a greater than 1 indicates overdispersion (extra varia tion) in the data. This means that “the arrangement of the data do not meet the e xpectations determined by the assumptions underlying the model” (Cooch and White 2007), most importantly the assumptions of equal catchability and survival If overdispersion of data is indicated, the value of can be adjusted to account for the lack of fit by calculating the ratio of the model deviance and the mean deviance (or the ratio of model to mean ) from the bootstrap simulations. Once a set of models had been run, the most parsimonious model was identified with the help of Akaike’s Information Cr iterion (AIC). AIC is defined as AIC = -2 ln (L) + 2 K,

PAGE 73

63 Appendix A: (Continued) where L is the model likelihood and K is the nu mber of parameters. The most parameterized model in a set will always fit the data best; however, the more parameters a model includes, the lower the precision becomes for th e individual estimates. The AIC strikes a balance between the best possible fit of a model (reflected by a low log likelihood) and the number of parameters, choosing the model that is most parsimonious overall. Results browsers in MARK list models in order of lowest to high est AIC values, and the model with the lowest AIC value has the greatest support. Where there was significant support (which I chose to be an AIC weight of at least 0.1) for more than one model, I used model averaging.

PAGE 74

64 Appendix B: Additional Figures Table 15. 2004 CJS Model for Herp Array Data Real Function Parameters of {Phi (gr*t) p (gr)} 95% Confidence Interval Parameter Estimate St andard Error Lower Upper -------------------------------------------------------------------------------------------------------------------1:Phi 0.9528634 0.4980721 0.7350119E-08 1.0000000 2:Phi 0.2657063 0.2106244 0.0417901 0.7501424 3:Phi 1.0000000 0.2290427E-07 1.0000000 1.0000000 4:Phi 1.0000000 0.3366622E-06 0.9999993 1.0000007 5:Phi 0.8364372 0.2876243 0.0766558 0.9968355 6:Phi 1.0000000 0.6238900E-07 0.9999999 1.0000001 7:Phi 1.0000000 0.5633052E-07 0.9999999 1.0000001 8:Phi 0.9621496 0.5417493 0.3530284E-09 1.0000000 9:Phi 0.2013092E-15 0.1003268E-07 -0.1966405E-07 0.1966405E-07 10:Phi 0.2404760E-15 0.9807671E-08 -0.1922304E-07 0.1922304E-07 11:Phi 0.5737598E-15 0.3387505E-07 -0.6639511E-07 0.6639511E-07 12:Phi 0.3408025 0.6322205E-15 0.3408025 0.3408025 13:Phi 0.2000000 0.1788854 0.0271820 0.6910541 14:Phi 0.2500000 0.2165063 0.0335100 0.7621677 15:Phi 0.1919038E-15 0.7997995E-08 -0.1567607E-07 0.1567607E-07 16:Phi 0.2000000 0.1264911 0.0504114 0.5407151 17:p 0.3041469 0.0996881 0.1479421 0.5238764 18:p 1.0000000 0.4897021E-06 0.9999990 1.0000010 -------------------------------------------------------------------------------------------------------------------

PAGE 75

65 Appendix B: (Continued) Table 16. 2005 CJS Model for Herp Array Data Real Function Parameters of {Phi (gr*t) p (t)} 95% Confidence Interval Parameter Estimate Standard Error Lower Upper --------------------------------------------------------------------------------------------------------------------1:Phi 0.5556494 0.2392547 0.1576577 0.8931003 2:Phi 0.2318002 0.1593927 0.0496135 0.6355863 3:Phi 1.0000000 0.2465402E-06 0.9999995 1.0000005 4:Phi 0.5852174 0.3601328 0.0715127 0.9627499 5:Phi 0.5153360 0.3192536 0.0798827 0.9286854 6:Phi 0.4840061 0.2001161 0.1632128 0.8185449 7:Phi 0.5708425 270.39279 0.1847300E-10 1.0000000 8:Phi 0.4979545 0.1704301 0.2067913 0.7905116 9:Phi 1.0000000 0.3611931E-07 0.9999999 1.0000001 10:Phi 0.6165312 0.3737527 0.0676301 0.9727050 11:Phi 0.7617800 0.5250086 0.0109009 0.9989234 12:Phi 0.7246216 0.4216839 0.0401398 0.9939967 13:Phi 0.6143448 0.2089097 0.2205224 0.8996959 14:Phi 0.8086938 383.05697 0.5870741E-10 1.0000000 15:p 0.7634795 0.1983777 0.2726260 0.9652780 16:p 0.4140473 0.1378648 0.1883095 0.6827665 17:p 0.1396365 0.1036179 0.0290631 0.4680853 18:p 0.3234532 0.1815741 0.0859400 0.7085501 19:p 0.4253910 0.1983320 0.1311289 0.7840884 20:p 1.0000000 0.1249396E-06 0.9999998 1.0000002 21:p 0.6182810 292.86328 0.2249470E-10 1.0000000 --------------------------------------------------------------------------------------------------------------------

PAGE 76

66 Appendix B: (Continued) Table 17. 2004 Models Using Conventiona l Robust Design Ranked by AIC Delta AICc Model Model AICc AICc Weight Likelihood #Par D eviance -------------------------------------------------------------------------------------------------------------------------------{Phi (a1) p,c (all .) N (.) Markov migr} -3 51.812 0.00 0.29647 1.0000 11.000 0.000 {Phi (.) p,c (all .) N (.) random migr} -350.949 0.86 0.19253 0.6494 9.000 0.000 {Phi (a1) p,c (all .) N (.) random migr} -350.862 0.95 0.18429 0.6216 10.000 0.000 {Phi (a1+PS) p,c (all .) N (.) Markov migr} -350.7 17 1.10 0.17147 0.5784 13.000 0.000 {Phi (.) p,c (all .) N (.) Markov migr} -350.306 1.51 0.13959 0.4708 11.000 0.000 {Phi (t) p,c (all .) N (.) Markov migr} -344.750 7.06 0.00868 0.0293 27.000 0.000 {Phi (PS) p,c (all .) Markov migr} -341.799 10.01 0.00198 0.0067 16.000 0.000 {Phi (a1+PS) p,c (.) Markov migr} -341.270 10.54 0.00152 0.0051 32.000 0.000 {Phi (.) p,c (all .) N (t) Markov migr} -341.063 10.75 0.00137 0.0046 16.000 0.000 {Phi (a1) p,c (all .) Markov migr} -340.484 11.33 0.00103 0.0035 17.000 0.000 {Phi (a1) gammas (.) p,c (all .) N (.) Markov migr -338.747 13.07 0.00043 0.0015 7.000 0.000 {Phi (t+PS) p,c (.) Markov migr} -338.625 13.19 0.00041 0.0014 34.000 0.000 {Phi (a1+PS) gammas (.) p,c (all .) N (.) Markov migr} -337.172 14.64 0.00020 0.0007 9.000 0 .000 {Phi (t*PS) p,c (.) Markov migr} -332.841 18.97 0.00002 0.0001 39.000 0.000 {Phi (a1) p,c (all .) N (.) no migr} -325.151 26.66 0.00000 0.0000 5.000 0.000 {Phi (.) p,c (all .) N (.) no migration} -324.109 27.70 0.00000 0.0000 4.000 0.000 -------------------------------------------------------------------------------------------------------------------------------Table 18. 2004 Parameter Estimates for the Conventional Robust Design Real Function Parameters of {Phi (a1) p,c (all .) N (.) Markov migr} 95% Confidence Interval Parameter Estimate Standard Error Lower Upper -----------------------------------------------------------------------------------------------------------------------------1:S 0.6202346 0. 0432286 0.5326612 0.7006210 2:S 0.6970948 0. 0405358 0.6123580 0.7702579 3:Gamma'' 0.2699954 0.0825089 0.1400182 0.4565716 4:Gamma'' 0.3693414 0.1023471 0.1984113 0.5808265 5:Gamma'' 0.0410575 0.0586186 0.0023081 0.4420833 6:Gamma'' 0.5195290E-06 0.3588293E-03 0.7215194E-17 0.9999733 7:Gamma'' 0.2490882 0.0829622 0.1220852 0.4417329 8:Gamma' 0.1361717E-07 0.2937626E-04 0.1891145E-18 0.9989812 9:Gamma' 0.4471174 0.2289154 0.1163638 0.8323919 10:Gamma' 0.5685376E-08 0.0000000 0.5685376E-08 0.5685376E-08 11:Gamma' 0.2073623 0.4470773 0.0012640 0.9818437 12:p Session 1 0.6026984 0.0270800 0.5486117 0.6543871 13:c Session 1 0.6870859 0.0188669 0.6489745 0.7228268 14:N Session 1 72.253391 1.1212247 70.687215 75.273387 ------------------------------------------------------------------------------------------------------------------------------

PAGE 77

67 Appendix B: (Continued) Table 19. 2004 Models Using Pradel Robust De sign with Individual Covariates Delta AICc Model Model AICc AICc Weight Likelihood #Par Devi ance -------------------------------------------------------------------------------------------------------------------------------{Phi, lambda (patchsize*t) p,c (.)} 223.360 0.00 0.99856 1.0000 44.000 130.231 {Phi, lambda (patchsize*t) p,c,N (.)} 236.454 13.09 0.00143 0.0014 39.000 154.438 {Phi, lambda (patchsize*t) p,c,N (all t) pt=ct} 247.306 23.95 0.00001 0.0000 58.000 122.277 {Phi, lambda (patchsize*t) p,c (all t) pt=ct N (.)} 252.292 28.93 0.00000 0.0000 54.000 136.496 {Phi(t) lambda(patchsize*t) p,c(.)} 282.995 59.64 0.00000 0.0000 39.000 200.980 {Phi, lambda (patchsize+t) p,c (.)} 315.683 92.32 0.00000 0.0000 35.000 242.456 {Phi(patchsize*t) lambda(t) p,c(.)} 322.567 99.21 0.00000 0.0000 39.000 240.551 {Phi, lambda(com inter-patchsize*t) p,c(.)} 327.130 103.77 0.00000 0.0000 35.000 253.903 {Phi, lambda (t) p,c(.)} 333.733 110.37 0.00000 0.0000 33.000 264.867 {Phi, lambda (patchsize*t) p,c (.) N (.)} 334.355 110.99 0.00000 0.0000 27.000 278.439 {Phi, lambda (patchsize*t) p,c (all .)} 412.282 188.92 0.00000 0.0000 33.000 343.416 -------------------------------------------------------------------------------------------------------------------------------Table 20. 2004 Models Using Pradel Robust Design with Groups Delta AICc Model Model AICc AICc Weight Likelihood #Par Devian ce {Phi, lambda (gr*t) p,c,N (gr)} 697.008 0.00 0.99691 1.0000 28.000 638.947 {Phi, lambda (gr*t) p,c (gr) N (.)} 709.508 12.50 0.00192 0.0019 28.000 651.447 {Phi, lambda (gr*t) p,c (gr) N (t)} 710.626 13.62 0.00110 0.0011 35.000 637.399 {Phi, lambda (gr*t) p,c,N (.)} 717.490 20.48 0.00004 0.0000 26.000 663.713 {Phi, lambda (gr*t) p,c (.) N (gr)} 719.508 22.50 0.00001 0.0000 27.000 663.591 {Phi, lambda (gr*t) p,c all N (gr)} 719.508 22.50 0.00001 0.0000 27.000 663.591 {Phi (gr) lambda (gr*t) p,c (gr) N (gr*t)} 807.012 110.00 0.00000 0.0000 32.000 740.319 {Phi, lambda (gr+t) p,c,N (gr)} 858.455 161.45 0.00000 0.0000 20.000 817.400 {Phi (t) lambda (.) p,c (.)} 87 7.237 180.23 0.00000 0.0000 47.000 777.369 {Phi, lambda (.) p,c (.)} 882.777 185.77 0.00000 0.0000 42.000 794.110 {Phi, lambda (t) p,c (.)} 882.866 185.86 0.00000 0.0000 47.000 782.999 {Phi, lambda (gr) p,c (.)} 88 4.541 187.53 0.00000 0.0000 44.000 791.412 {Phi (.) lambda (t) p,c (.)} 884.961 187.95 0.00000 0.0000 47.000 785.094

PAGE 78

68 Appendix B: (Continued) Table 21. 2004 Models Using Link Barker with Individual Covariates Delta AICc Model Model AI Cc AICc Weight Likelihood #Par Devianc e -------------------------------------------------------------------------------------------------------------------------------{Phi (t) p (t) f (PS*t)} 1346.410 0.00 0.43380 1.0000 23.000 1297.458 {Phi (t) p (.) f (PS*t)} 1346.979 0.57 0.32629 0.7522 19.000 1306.969 {Phi (t) p (t) f (PS+t)} 1349.474 3.06 0.09371 0.2160 18.000 1311.670 {Phi (PS+t) p (t) f (PS+t)} 1349.649 3.24 0.08586 0.1979 19.000 1309.639 {Phi (PS+t) p (.) f (PS+t)} 1351.091 4.68 0.04176 0.0963 15.000 1319.835 {Phi (t) p (t) f (PS)} 1353.958 7.55 0.00996 0.0230 15.000 1322.701 {Phi (PS*t) p (t) f PS*t)} 1354.433 8.02 0.00785 0.0181 29.000 1291.705 {Phi (t) p (t) f (t)} 1359.058 12.65 0.00078 0.0018 18.000 1321.253 -------------------------------------------------------------------------------------------------------------------------------Table 22. 2004 Models Using Link Barker with Groups Delta AICc Model Model AICc AICc Weight Likelihood #Par Deviance --------------------------------------------------------------------------------------------------------------------{Phi (gr) p (t) f (gr*t)} 1351.883 0.00 0.38923 1.0000 19.000 169.329 {Phi (t) p (t) f (gr*t)} 1353.494 1.61 0.17392 0.4468 22.000 164.252 {Phi (t) p (.) f (gr*t)} 1353.805 1.92 0.14891 0.3826 18.000 173.456 {Phi (gr) p (t) f (gr+t)} 1354.901 3.02 0.08608 0.2212 16.000 178.929 {Phi (gr) p (t) f (t)} 1355.115 3.23 0.07735 0.1987 14.000 183.475 {Phi (gr) p (.) f (gr*t)} 1355.707 3.82 0.05751 0.1478 14.000 184.067 {Phi (t) p (t) f (t)} 1356.864 4.98 0.03226 0.0829 17.000 178.710 {Phi (gr) p (gr) f (gr*t)} 1357.666 5.78 0.02160 0.0555 15.000 183.866 {Phi (t) p (t) f (gr)} 1358.658 6.77 0.01316 0.0338 15.000 184.858 --------------------------------------------------------------------------------------------------------------------Table 23. 2005 Models Using Conventional Robust Design Delta AICc Model Model AICc AICc Weight Likelihood #Par D eviance -------------------------------------------------------------------------------------------------------------------------------{Phi (a1) p,c (.) N (.) Markov migr} -1 69.681 0.00 0.52935 1.0000 20.000 0.000 {Phi (a1) gammas (.) p,c (.) N (.) Markov migr} -167.341 2.34 0.16436 0.3105 17.000 0.000 {Phi (a1) p,c (.) N (.) no migr} -166.620 3.06 0.11457 0.2164 15.000 0.000 {Phi (a1+PS) p,c (.) N (.) Markov migr} -165.792 3.89 0.07573 0.1431 22.000 0.000 {Phi (a1) p,c (.) N (.) random migr} -165.683 4.00 0.07173 0.1355 18.000 0.000 {Phi (a1+PS) p,c (.) N (PS) Markov migr} -163.610 6.07 0.02544 0.0481 23.000 0.000 {Phi (.) p,c (.) N (.) Markov migr} -1 61.396 8.28 0.00841 0.0159 19.000 0.000 {Phi (a1+PS) p,c (.) N (t) Markov migr} -159.997 9.68 0.00418 0.0079 27.000 0.000 {Phi (a1) p,c (.) N (t) Markov migr} -1 59.505 10.18 0.00327 0.0062 27.000 0.000 {Phi (a1+PS) p,c (all .) N (.) Markov migr} -158.075 11.61 0.00160 0.0030 10.000 0.000 {Phi (a1*PS) p,c (.) N (t) Markov migr} -157.772 11.91 0.00137 0.0026 28.000 0.000 --------------------------------------------------------------------------------------------------------------------------------

PAGE 79

69 Appendix B: (Continued) Table 24. 2005 Models Using Pradel Robust De sign with Individual Covariates Delta AICc Model Model AICc AICc Weight Likelihood #Par Devianc e -------------------------------------------------------------------------------------------------------------------------------{Phi (t) lambda (t) p,c (.) N (.)} 263.740 0.00 0.50239 1.0000 23.000 215.600 {Phi (t) lambda (t) p,c (.) N (t)} 264.720 0.98 0.30768 0.6124 27.000 207.767 {Phi (t) lambda (PS+t) p,c (.) N (t)} 266.875 3.14 0.10477 0.2085 28.000 207.697 {Phi (t) lambda (PS*t) p,c (.) N (t)} 268.250 4.51 0.05268 0.1049 32.000 200.084 {Phi (t) lambda (.) p,c (.) N (t)} 270.195 6.46 0.01992 0.0397 24.000 219.865 {Phi (.) lambda (t) p,c (.) N (t)} 272.323 8.58 0.00687 0.0137 24.000 221.993 {Phi (t) lambda (t) p,c (all .) N (.)} 273.938 10.20 0.00307 0.0061 13.000 247.246 {Phi (PS*t) lambda (t) p,c (.) N (t)} 274.249 10.51 0.00262 0.0052 32.000 206.084 -----------------------------------------------------------------------------------------------------Table 25. 2005 Models Using Pradel Robust Design with Groups Delta AICc Model Model AICc AICc Weight Likelihood #Par Devian ce -------------------------------------------------------------------------------------------------------------------------------{Phi (t) lambda (t) p,c (.) N (t)} 670.382 0.00 0.46722 1.0000 27.000 613.429 {Phi (t) lambda (t) p,c (.) N (.)} 670.440 0.06 0.45405 0.9718 23.000 622.300 {Phi (gr*t) lambda (t) p,c ( .) N (.)} 675.384 5.00 0.03833 0.0820 28.000 616.205 {Phi (t) lambda (gr) p,c (.) N (t)} 675.636 5.25 0.03378 0.0723 24.000 625.306 {Phi (gr) lambda (t) p,c (.) N (t)} 679.061 8.68 0.00609 0.0130 25.000 626.532 {Phi (gr*t) lambda (gr*t) p,c (.) N (.)} 684.978 14.60 0.00032 0.0007 33.000 614.543 {Phi (gr*t) lambda (gr*t) p,c (.) N (gr)} 685.867 15.48 0.00020 0.0004 34.000 613.154 {Phi (gr*t) lambda (gr*t) p,c (.) N (gr*t)} 694.860 24.48 0.00000 0.0000 42.000 603.592 -------------------------------------------------------------------------------------------------------------------------------Table 26. 2005 Models Using Link Barker with Individual Covariates Delta AICc Model Model AICc AICc Weight Likelihood #Par Deviance -------------------------------------------------------------------------------------------------------------------------------{Phi (t) p (.) f (.)} 918.824 0.00 0.46775 1.0000 7.000 904.453 {Phi (.) p (.) f (.)} 920.705 1.88 0.18267 0.3905 3.000 914.626 {Phi (t) p (.) f (PS)} 920.920 2.10 0.16401 0.3506 8.000 904.442 {Phi (t) p (.) f (t)} 922.594 3.77 0.07102 0.1518 11.000 899.708 {Phi (PS) p (.) f (.)} 922.680 3.86 0.06805 0.1455 4.000 914.549 {Phi (t) p (.) f (PS+t)} 924.733 5.91 0.02438 0.0521 12.000 899.682 {Phi (.) p (.) f (t)} 925.637 6.81 0.01552 0.0332 7.000 911.266 {Phi (t) p (t) f (t)} 928.582 9.76 0.00356 0.0076 15.000 896.950 {Phi (t) p (.) f (PS*t)} 928.887 10.06 0.00306 0.0065 15.000 897.254 -------------------------------------------------------------------------------------------------------------------------------

PAGE 80

70 Appendix B: (Continued) Table 27. 2005 Models Using Link Barker with Groups Delta AICc Model Model AICc AICc Weight Likelihood #Par Deviance -------------------------------------------------------------------------------------------------------------------------------{Phi (t) p (.) f (.)} 918.824 0. 00 0.36181 1.0000 7.000 94.339 {Phi (gr) p (.) f (.)} 920.431 1.61 0.16199 0.4477 4.000 102.186 {Phi (t) p (.) f (gr)} 920.515 1.69 0.15540 0.4295 8.000 93.922 {Phi (gr) p (.) f (gr)} 920.636 1.81 0.14626 0.4042 5.000 100.324 {Phi (.) p (.) f (.)} 920.705 1.88 0.14130 0.3905 3.000 104.512 {Phi (gr*t) p (.) f (.)} 924.762 5.94 0.01858 0.0514 12.000 89.598 {Phi (.) p (.) f (t)} 925.637 6.81 0.01200 0.0332 7.000 101.151 {Phi (gr*t) p (.) f (t)} 928.684 9.86 0.00261 0.0072 16.000 84.714 {Phi (gr*t) p (.) f (gr*t)} 937.216 18.39 0.00004 0.0001 21.000 81.894 {Phi (gr*t) p (t) f (gr*t)} 941.389 22.56 0.00000 0.0000 24.000 79.064 -------------------------------------------------------------------------------------------------------------------------------