Citation
The demography and movements of snail kites in Florida

Material Information

Title:
The demography and movements of snail kites in Florida
Creator:
Bennetts, Robert E
Publication Date:
Language:
English
Physical Description:
xv, 197 leaves : ill. ; 29 cm.

Subjects

Subjects / Keywords:
Animal nesting ( jstor )
Bird nesting ( jstor )
Birds ( jstor )
Breeding ( jstor )
Breeding value ( jstor )
Drought ( jstor )
Juveniles ( jstor )
Parametric models ( jstor )
Snails ( jstor )
Wetlands ( jstor )
Bird populations -- Florida -- Everglades ( lcsh )
Dissertations, Academic -- Wildlife Ecology and Conservation -- UF ( lcsh )
Everglade kite -- Florida ( lcsh )
Wildlife Ecology and Conservation thesis, Ph.D ( lcsh )
Miami metropolitan area ( local )
Genre:
bibliography ( marcgt )
non-fiction ( marcgt )

Notes

Thesis:
Thesis (Ph.D.)--University of Florida, 1998.
Bibliography:
Includes bibliographical references (leaves 187-196).
General Note:
Typescript.
General Note:
Vita.
Statement of Responsibility:
by Robert E. Bennetts.

Record Information

Source Institution:
University of Florida
Holding Location:
University of Florida
Rights Management:
The University of Florida George A. Smathers Libraries respect the intellectual property rights of others and do not claim any copyright interest in this item. This item may be protected by copyright but is made available here under a claim of fair use (17 U.S.C. §107) for non-profit research and educational purposes. Users of this work have responsibility for determining copyright status prior to reusing, publishing or reproducing this item for purposes other than what is allowed by fair use or other copyright exemptions. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder. The Smathers Libraries would like to learn more about this item and invite individuals or organizations to contact the RDS coordinator (ufdissertations@uflib.ufl.edu) with any additional information they can provide.
Resource Identifier:
029219566 ( ALEPH )
40097277 ( OCLC )

Downloads

This item has the following downloads:


Full Text










THE DEMOGRAPHY AND MOVEMENTS OF SNAIL KITES IN FLORIDA













By

ROBERT E. BENNETTS














A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY

UNIVERSITY OF FLORIDA

1998













ACKNOWLEDGMENTS

I am deeply indebted to my major advisor Wiley M. Kitchens, who stuck it out and

provided support through all of the ups and downs. I am also grateful to my committee

members, Drs. Crawford S. Holling, James D. Nichols, Kenneth M. Portier, Frank J.

Mazzotti. Financial support was provided by the U.S. Fish and Wildlife Service

(USFWS), National Park Service (NPS), U.S. Army Corps of Engineers (USACOE),

South Florida Water Management District (SFWMD), St. Johns River Water

Management District (SJRWMD), and the Biological Resources Division (BRD) of the

U.S. Geological Service. John Ogden (SFWMD) and David Wesley (USFWS) were

largely responsible for getting this project started and continued to provide strong support

throughout its duration. I am also grateful to Reid Goforth (BRD), Steve Miller

(SJRWMD), Ed Lowe (SJRWMD), Mary Ann Lee (SJRWMD), Jon Mouldling

(USACOE), Lewis Hornung (USACOE), Peter David (SFWMD), Dale Gawlik

(SFWMD), Paul Warner (SFWMD), and Jim Brown (USFWS), and Donald DeAngelis

(BRD). I greatly appreciate the help of our field biologists Phil Darby, Patty Valentine-

Darby, Katie Golden, Steve McGehee, Scott Severs, Hilary Maier, David Boyd, James

Conner, and Lynn Bjork. Their ability to work independently for long hours, and get the

job done made our job much easier.

This project has been a cooperative effort among biologists and agencies from the



ii








outset. For their help in the field and/or logistic support I am grateful to Brian Toland

(USFWS), Tim Towles (GFC), Peter Frederick (UF), Marilyn Spalding (UF), Mary Beth

Mihalik (West Palm Beach Solid Waste Authority), Al Vasquez (West Palm Beach Solid

Waste Authority), Deborah Jansen (NPS), Mike Wilson (NPS), Sue McDonald (NPS),

Vivie Thue (NPS), Fred Broerman (Arthur R. Marshall Loxahatchee National Wildlife

Refuge), Angela Chong (SFWMD), Vicky Dreitz (Univ. Miami), F.K. Jones (Miccosukee

Tribe of Indians), and Steve Terry (Miccosukee Tribe of Indians). The banding of Snail

Kites was conducted in cooperation with the GFC. In this effort, I appreciate the

cooperation of James Rodgers Jr. (GFC), Jon Buntz (GFC), and Brian Toland (USFWS).

I greatly appreciate the effort of Charlie Shaiffer (Mingo National Wildlife Refuge) who

took the time to travel to Florida to share his knowledge of trapping and handling birds. I

am grateful to Patuxent Wildlife Research Center, particularly Jim Nichols and Jim Hines,

for housing and assistance during the analysis phase of this project. I also appreciate the

help of Laura Brandt, Cynthia Loftin, Kenny Rice, and Cynthia Sain for keeping me

registered for credits when I had procrastinated to the last hours. Laura Brandt also

served as an unofficial academic advisor for which I am grateful.

For allowing me access to areas used by kites, which were closed to public access,

I am grateful to the Miccosukee Tribe of Indians, the City of West Palm Beach, and the

Arthur R. Marshall Loxahatchee National Wildlife Refuge. I am grateful to our pilots

Karen Dunne and Morton Sund for many hours of safe flying. Finally, I am grateful to

the employees of the Florida Cooperative Fish and Wildlife Research Unit, particularly

Barbara Fesler and Debra Hughes, for their help with administration of this study.


iii














TABLE OF CONTENTS

page
ACKNOWLEDGMENTS ............................................................................................

LIST OF TABLES ......................... ................................vii

LIST O F FIG U RE S ....................................................... .................................. ..x

A B STR A C T .................................................................................................. ..... xiv

CHAPTERS

1. IN TR O D U CTION ..................................................... ............. ................. 1

Overview ........................... ....................... ........ 1
O bjectiv es ............................................................................ .............. .......3

2. ST U D Y A R E A ..................................................................... ........................ 7

Spatial Scales .................................................... ........... ... ...........7
Regions ......................................................... ..........8
H abitat T ypes ................................................ ............................... ..........9

3. ANNUAL SURVIVAL OF SNAIL KITES IN FLORIDA WITH
COMPARISONS BETWEEN RADIO TELEMETRY AND
CAPTURE-RESIGHTING DATA ................................................................ 14

Introduction ....................................................................................................14
M ethods .............. ........................................................ .....................................16
Estimation of Survival from Radio Telemetry ........................................... 16
Estimation of Survival from Banding Data ............................................. 17
Influences on Survival .................................................. 19
Hypothesis Testing and Model Selection ..................................................22
Results ...................... ........................... ...23
A ge E ffects .................................................................. .................. 23
T im e E ffects .................................................................. .................24
R egional Effects ........................................................... .................. 25
Param eter Estim ates ............................................ ........................... 26
Censoring of Radio-tagged Birds .................. ............ .................. 27

iv









D iscussion ........................................................ ...........................................28
Comparisons Between Data Obtained Using Radio Telemetry and
Capture-resighting ............................................ .......................... 28
Parameter Estimates ..............................................30
Effects of Age, Time, and Region on Survival ........................................ 32
Implications of Resighting Probabilities .................................................... 33

4. WITHIN-YEAR SURVIVAL PATTERNS OF SNAIL KITES IN FLORIDA ...... 43

M ethods ......................................... ........... ....... ..... ....... 44
R esults ........................................................ ....................... .. ............. 46
D iscussion ........................................ ........... .. ....... ................. ......... 48

5. CAUSES OF MORTALITY OF POST-FLEDGING JUVENILE AND
ADULT SNAIL KITES IN FLORIDA ................................................54

6. REPROD U CTION .................................................... ............................... 60
Sem antics .......................................................................... .......... ... .. 61
B reeding A ttem pts ............................................. ............................ 61
Successful N ests ................................................. ......... ....... ............... 64
The Breeding Season ........................................................... ................ 64
The Breeding Population ............................................ ......................... 65
Age of First Reproduction ................................................ 65
Proportion of Birds Attempting to Breed ................................................ 65
N est Success ......................... ...... ...... .... ............................ .. ................... 69
Areas of Disagreement Regarding Estimation of Nest Success ................... 70
Estimates of Nest Success and its Process Variance ................................... 75
Influences of N est Success ......................................................................... 76
Number of Young per Successful Nest ............................................... 78
Number of Breeding Attempts per Year ............................. ................. 79
Conditional Probability of Attempting to Breed ...................................... 85
Number of Successful Broods per Year .................................................. 85

7. DISPERSAL PROBABILITIES OF SNAIL KITES IN FLORIDA .......................100

Introduction .............................................. .......................... ....................... 100
Methods ................................... .................... ....... 101
Term inology ....................................... ............. ...................................... 101
Field M ethods ..................................................................... ... ................. 102
Estimation of Natal Dispersal .............................................. ................. 103
Estimation of Dispersal Probabilities ...............................................103
Food Availability ...................................................................................105
W ater L evels ........................................................ .............................. 106
R esults ............................................................................... .................... ... 108


V









N atal D ispersal .............................................................. ....................... 108
Dispersal Probabilities ............................. ........ ................. 109
Pooling of Locations .............................. ........................ 111
Hydrologic Effects on the Probability of Dispersal .................................... 115
Food R esources ............................................... ............................ 116
D iscussion .................................................................. .......... ...... ........ ....... 116

8. IMPLICATIONS TO MANAGEMENT AND CONSERVATION .....................139

Drought Semantics ............................... ...................... 139
Intensity .............................................................. ........................ 141
Spatial Extent ................................................... .............................. 143
Tem poral Extent ..................................................................................... 144
Critical H abitat .................................................................. ...................145
Currently Designated Critical Habitat ................................................... 145
The Habitat Network .............................. ............. ......................146
Management of the Snail Kite in Florida: Beyond a Reductionist Paradigm .......150
The Reductionist Paradigm .................................... ................... 151
Conflicts and Limitations of the Existing Paradigm .................................... 154
The Importance of Spatial and Temporal Scales ........................................ 155
The Dynamic Landscape Hypothesis ...............................................156
Persistence of Snail Kites in a Dynamic Landscape ................................... 161
C o nclu sion s .................................................. ............................................. 163

APPENDICES

1 ESTIMATES OF CUMULATIVE NATAL DISPERSAL (*), NUMBER OF
ANIMALS AT "RISK" OF DISPERSAL DURING INTERVAL j (rj),
AND STANDARD ERROR (SE) OF THE ESTIMATE FOR EACH
STUDY YEAR .................... .......... .................... 182

2 ESTIMATES OF CUMULATIVE NATAL DISPERSAL (*i), NUMBER OF
ANIMALS AT "RISK" OF DISPERSAL DURING INTERVAL j (r),
AND STANDARD ERROR (SE) OF THE ESTIMATE FOR EACH
STUDY YEAR IN AREAS AFFECTED AND UNAFFECTED BY
THE PREVIOUS DROUGHT ........................................ ....................... 184

LITERA TURE CITED .......................................................................................... 187

BIOGRAPHICAL SKETCH ................................................. .................................... 197






vi














LIST OF TABLES


Table page

3-1 Capture-resighting summary of adult and juvenile Snail Kites in Florida from
1992-1997. ................................................. ........................... . ........ 34
3-2 Description of single-stratum Cormack-Jolly-Seber (CJS) models and their
corresponding Akaike Information Criteria (AIC) scores. Parameter structure
indicates whether survival (4) and/or resighting probability (p) was dependent
on time (t) and/or age(a). ........................................ ....................... 35
3-3 Likelihood ratio tests (LRTs) between Cormack-Jolly-Seber (CJS) models
used to test whether survival (4) or resighting (p) probabilities differed among
age classes or years (time). ........................................ ....................... 36
3-4 Description of multi-strata models and their corresponding Akaike
Information Criteria (AIC) scores. Parameter structure indicates whether
survival (4), resighting probability (p), and/or transition (movement)
probability (ir/) was dependent on age(a), time (t), and/or region (r) ............ 37
3-5 Annual estimates and standard errors for adult and juvenile survival (4) of
Snail Kites for study years (SYs) 1992, 1993, and 1994 using data from radio
telem etry. ................................................................................................. 38
3-6 Parameter estimates for the Cormack-Jolly-Seber (CJS) model SS10 in which
survival (4,) differed between adults and juveniles. Under this model, survival
was constant among years for adults, but differed among years for juveniles.
Resighting probabilities (p) differed among years. .................. ............. 39
3-7 Parameter estimates for my most parsimonious multi-stratum model (MS 13),
in which survival differs between adults and juveniles, survival is constant
among years and regions for adults, and survival differs among years and
regions for juveniles. Resighting probability in this model differs among years
and regions ....................................................... ............... .. ........... ......... 40
5-1 Probable causes of mortality of Snail Kites recovered in Florida from 1992-
1995 ................ .................................................. 59
6-1 The number of nest initiations reported in each month during studies from
1966 through 1995. ................. ............................... 88


vii









6-2 Mean annual nest success from major studies conducted since 1968. Also
shown are the estimator used to estimate success, whether or not nests placed
in nest baskets were included in the estimate, and whether or not nests found
in each of 3 stages were included in the estimate. .......................................... 89
6-3 Annual nest success reported during studies from 1968 through 1995. Nest
success was based on nests found during the egg stage. ................................ 91
6-4 Summary statistics from the final most parsimonious (based on AIC and
LRTs) logistic regression model for the factors effecting nest success. ............ 92
6-5 The annual mean and overall (i.e., all years and locations combined) number
young per successful nest from major studies conducted since 1968. A
successful nest was considered a nest in which at least one young fledged. ...... 93
6-6 The number of successful nests, young fledged, and number of young per
successful nest reported for each year from 1968 through 1995 ................. 94
6-7 Number of nesting attempts and number of attempts that were successful for
each of 23 adult Snail Kites during the 1995 breeding season. ...................... 95
7-1 Summary statistics for conditional logistic regression models for potential
seasonal groupings affecting the probability of movement between times t and
t + 1 (at monthly time steps), given that an animal was alive at time t and its
location known. Shown are the model description, number of estimable
parameters (np), relative deviance (-21n[9]), and Akaike's Information
Criteria (AIC). The model shown in bold would be the one selected from
these potential models based on AIC. .............................. ..................... 121
7-2 Summary statistics for conditional logistic regression models for potential
temporal effects on the probability of movement between times t and t + 1 (at
monthly time steps), given that an animal was alive at time t and its location
known. Shown are the model description, number of estimable parameters
(np), relative deviance (-21n[f]), and Akaike's Information Criteria (AIC).
The model shown in bold would be the one selected from these potential
models based on AIC criteria. .................................................................... 122
7-3 Summary statistics for conditional logistic regression models of the probability
of movement between times t and t + 1. The model with the lowest AIC
(bold) would be selected if based solely on this criterion. .............................. 123
7-4 Summary statistics for conditional logistic regression models of the probability
of movement between times t and t + 1 to evaluate the pooling of some
parameters. A failure to reject a LRT indicates that the additional parameters
of the more general (unconstrained) model may not be supported by these
data. The model with the lowest AIC (bold) would be selected if based solely
on A IC .......................................................................... ......... . ....... ..... 124


viii









7-5 Maximum likelihood analysis of variance table for univariate models (i.e., each
source term represents a separate model) of potential sources of variation of
the probability of dispersal between times t and t + 1................................ 125
7-6 Summary statistics used for model selection of conditional logistic regression
models of the probability of movement between times t and t + 1. The model
with the lowest AIC (bold) would be selected if based solely on AIC criterion. 126
7-7 Likelihood ratio tests (LRTs) comparing conditional logistic regression
models of the probability of movement between times t and t + 1. The null
hypothesis (Ho) from a LRT is that the reduced model (i.e., the model with
fewer parameters) fits the data equally well as the more general model (i.e.,
with more parameters). Thus, a rejection of Hi favors the more general model
and a failure to reject H. favors the more reduced model. ............................ 127
7-8 Analysis of variance table from model of foraging time per capture as the
dependent variable. Mean square (MS) and F values are based on type III
partial sums of squares (i.e., they are adjusted for all other terms in the model
and are not dependent on the order ofentry)(SAS Inc. 1988). ...................... 129
8-1 Drought intensity scores (standard deviations from mean annual minimum) for
most major wetlands used by Snail Kites in Florida. Scores > 1 sd below
mean are considered as drought years (bordered cells) and scores >2 sd below
mean are considered as extreme drought years (cells bordered with double
line). Spatial extent of a drought can be evaluated by how many areas in a
given year have scores >1 sd below the mean. ........................................... 166
8-2 The number of days that water stage was > 1 standard deviation below the
average minimum stage for a 10-gauge average from Lake Okeechobee for
each year from 1969-1994. This corresponds to a stage of< 11.17 ft MSL.
The intensity for a given drought year is shown as the number of standard
deviations below the mean. ...... .......................... .. ............... 168

















ix














LIST OF FIGURES


Figure page
1-1 The demographic cycle of Snail Kites showing three age classes (Juveniles=J,
Subadults=S, and Adults=A). Parameters for survival (4() and fecundity (f)
are shown for each age class. Adapted from Caswell (1989), Beissinger
(1995) and Legendre and Clobert (1995). ...................................... .......... 5

1-2 Conceptual framework for this study. Reliable estimation of parameters is the
first step toward the development of a wide variety of demographic models. ... 6
2-1 Major wetlands of South Florida referred to in this report. Wetlands are
Everglades National Park (ENP), Big Cypress National Preserve (BICY),
Water Conservation Areas 3A, 3B, 2B, 2A, Loxahatchee National Wildlife
Refuge (LOX), Holey Land Wildlife Management Area (HOLEY), West Palm
Beach Water Catchment Area (WPB), Lake Okeechobee (OKEE), Upper St.
Johns [Blue Cypress] Marsh (SJM), Lake Kissimmee (KISS), Lake
Tohopekaliga (TOHO), and East Lake Tohopekaliga (ETOHO). ................ 12
2-2 South Florida showing geographic regions used for some analyses in this
report. All areas not within a region shown were assigned to a peripheral
region. ......................................... .. ........... ............... 13
3-1 Percentage of radio-transmittered adult and juvenile Snail Kites that were
censored in each 60-day time interval from the time of attachment. ............... 41
3-2 Percentage of adult and juvenile Snail Kites from each sampling cohort (i.e.,
the year that they fledged or were captured) that died or were censored during
the first 180 d after radio attachment each year. ........................ ................. 42
4-1 Survivorship functions of adult and juvenile Snail Kites from a pooled sample
of 3 years (top). Because I was interested in temporal pattern rather than
magnitude, I aligned the functions without regard to magnitude, to illustrate at
what point in time the functions become similar (bottom). ............................ 51
4-2 Estimated hazard functions constructed at monthly intervals for adult and
juvenile Snail K ites ................................................................................. 52
4-3 Estimated hazard function for juveniles based on age, rather than time. This
function was constructed at monthly intervals starting at the time of fledging. 53


x









6-1 Conceptual diagram of reproductive parameters used to estimate fecundity.
Show here for simplicity is model for 2 nesting attempts; although more
attempts are possible within a given year. ............................................... 96
6-2 The proportion of nest initiations for each month of the year based on
cumulative data reported by Sykes (1987c), Snyder et al. (1989a)(1970-1982
only), Toland (1994), and this study. ................................................. ........... 97
6-3 The percentage of nests that were successful during each month. Data used in
this analysis were from Bennetts et al. (1988)(1986-1987), Toland (1994,
unpubl. data) (1990-1993), and this study (1994-1995). .............................. 98

6-4 The percentage of nests that were successful during each month of each year.
Data used in this analysis were from Bennetts et al. (1988)(1986-1987),
Toland (1994, unpubl. data)(1991-1993), and this study (1994-1995). ......... 99
7-1 Kaplan Meier estimates for the overall cumulative probability of dispersal
(solid line). Also shown is a 95% confidence interval (dotted line) for the
probability function. ...................... ............................. ................................... 130
7-2 Kaplan Meier estimates for the cumulative probability of dispersal in each of
the three years. Confidence intervals for estimates are not shown to minimize
cluttering, but are provided in detail in Appendix 1...................................... 131

7-3 Kaplan Meier estimates for the cumulative probability of dispersal from
wetlands that were and were not affected by the preceding drought in each of
the three study years. Confidence intervals for estimates are not shown to
minimize cluttering, but are provided in detail in Appendix 2 ........................ 132

7-4 Conditional probabilities that adult and juvenile Snail Kites that were alive and
their location known at time t, were in the same location (or conversely at a
different location) at time t + 1 during each season of each study year. Also
shown are the standard errors (rectangles) and 95% confidence intervals
(vertical lines). .................................................... ......... ........... .............. 133
7-5 Adjusted residuals from a cross tabulation of dispersal and location at time t.
Residuals >0 indicate that birds in this area moved more frequently than
expected and residuals <0 indicates that birds in that area moved less
frequently than expected. ............................ ... ................... 134
7-6 Standardized residuals for probability of dispersal during each study year for
adult (top) and juvenile (bottom) snail kites. ............................................ 135
7-7 Standardized residuals for probability of dispersal during each season for adult
(top) and juvenile (bottom) snail kites. ........................................................... 136
7-8 The mean ( SE) foraging time to capture snails. Sample sizes (number of
complete bouts observed) are also shown .......................................... 137


xi









7-9 The mean ( SE) foraging time to capture snails by radio-transmittered birds
before (location 1) and after (location 2) moving. Sample sizes (number of
individuals birds observed) are also shown. ................................. ............. 138
8-1 The minimum annual water stage for gauge 3-28 in Water Conservation Area
3A (WCA-3A) for the period of 1968-1988. Shown for reference are the
minimum and maximum ground elevation in WCA-3A, and ground elevation
at the 3-28 gauge. Points mark with an "S" were years identified by Snyder et
al. (1989) as drought years and those mark with a "B" were identified by
Beissinger (1995) as drought years ......................................................... 169
8-2 The minimum annual water stage for a 10-gauge average at Lake Okeechobee
for the period of 1968-1988. Shown for reference are the minimum and
maximum ground elevation for the littoral zone at Lake Okeechobee (based
on Pesnell and Brown [1977]). Points marked with an "S" were years
identified by Snyder et al. (1989a) as drought years and those mark with a "B"
were identified by Beissinger (1995) as drought years. ................................. 170
8-3 The minimum annual water stage for gauge 3-28 in Water Conservation Area
3A (WCA-3A) for the period of 1969-1994. Shown for reference are the
average annual minimum stage, 1 standard deviation, and 2 standard
deviations ............... ......................................................................... ....... 17 1
8-4 The currently designated critical habitat identified in the Snail Kite Recovery
Plan (after 50 CFR Ch. 1 [10-1-94 Edition]) .............................. ............... 172
8-5 South Florida showing the inter-wetland movements of individual radio-
tagged adult snail kites during 1992 and 1993. These movements illustrate a
basic habitat network used by snail kites (also shown). We have shown only a
limited subset of this network (and moments) to minimize cluttering, and
because a complete synthesis of the peripheral habitats has not been done. The
complete movements, and consequently the complete network, would include
all movements and habitats used by kites throughout the state ...................... 173
8-6 The percentage of Snail Kite nests (N=745) that were initiated in each 10 cm
water depth class. Data are from Bennetts et al. (1988), B. Toland (unpubl.
data), and this study. ........... ...................................................... 174
8-7 The distribution of Snail Kite nests in Water Conservation Area 3A during
each year from 1992 through 1996. During 1995 this area experienced
exceptionally high water levels as a result of Tropical Storm Gordon and the
distribution of nesting kites shifted to higher elevations. ................................ 175







xii









8-8 Conceptual model of relative habitat quality in relation to the time since a
drying event at a given location. In the absence of a drying event (A), habitat
quality initially increases as the apple snail population recovers, but declines
after 5-6 years as the plant communities comprising nesting and foraging
habitat begin degradation. If drying events occurs too frequently (B), the
apple snail population will have been unlikely to have recovered to its full
potential. If drying events occur at longer intervals (C) then a cumulative
process of slow and incremental degradation will occur as plant communities
undergo transition. ................................................................................... 176
8-9 Primary plant communities and their corresponding species that comprise
Snail Kite habitat in relation to elevation and a hydrologic gradient. ................ 177
8-10 The reported nesting distribution of nesting snail kites (shaded) in Water
Conservation Area 3A (WCA3A) from 1965 to present. Birds nesting in
southeastern WCA3A during the 1992-1996 period were foraging primarily in
Everglades National Park and the "Pocket" between the L-67A and L-67C
levees, both of which have shorter hydroperiods than the nesting area. ........... 178
8-11 A conceptual hydrologic window for a long (e.g., WCA-3A) and short (e.g.,
Big Cypress N.P.) hydroperiod wetlands. This window can shift over time
depending on the hydrologic conditions at different scales. ........................... 179
8-12 The distribution of Snail Kite nests in Big Cypress National Preserve during
each year from 1992 through 1995 (no nests were observed from 1992-1994). 180
8-13 Hypothesized relationship between the spatial extent of droughts and whether
the response by Snail Kites is likely to be behavioral (i.e., movement) or
numerical (i.e., change in survival and/or reproduction). ............................... 181




















xii














Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy

THE DEMOGRAPHY AND MOVEMENTS OF SNAIL KITES IN FLORIDA

by

Robert Edwin Bennetts

May, 1998

Chairman: Dr. Wiley M. Kitchens
Major Department: Wildlife Ecology and Conservation

Several authors have suggested that changes in Snail Kite (Rostrhamus sociabilis)

populations correspond with changes in hydrology. The primary objectives of this study

were to estimate survival and movement probabilities and to evaluate the influences on

those probabilities. I estimated survival of Snail Kites in Florida using the Kaplan-Meier

estimator with data from 271 radio tagged Snail Kites over a three-year period and

capture-recapture models with data from 1319 banded birds over a six-year period.

Results from these data sources were similar in their indications of the sources of variation

in survival, but differed in some parameter estimates. Both data sources indicated that

survival was higher for adults than for juveniles, but did not support delineation of a

subadult age class. Our data also indicated that survival differed among years and regions

for juveniles, but not adults.

My results indicated that the probability of movement is influenced by age, time,

and location. Although relatively high water conditions persisted throughout this study,

xiv














water levels (independently of location) did not appear to influence monthly movement

probabilities. Dispersal is generally thought to be favored when local resources are low or

better conditions exist elsewhere. In contrast, my results from both within-year and

between year comparisons suggest that higher probabilities of movement occur when food

resources were high. I suggest a hypothesis that this may be a reasonable strategy given

the dynamic and unpredictable nature of a kite's environment.

My data are consistent with previous views that the habitats used by Snail Kites in

Florida are considerably more extensive than the currently-designated-critical habitat.

Thus, protection of only the currently designated critical habitat may be insufficient to

maintain viable populations of Snail Kites over the long term. The use of habitats can be

characterized as an extensive network and I present a hypothesis of how the spatial and

temporal patterns of this network might influence viability of Snail Kites in Florida.




















xv














CHAPTER 1
INTRODUCTION

Overview

Florida's wetlands have undergone extensive anthropogenic change over the past

century including drainage, impoundment, changes in water flow regimes, increased

nutrient loadings, and invasion by exotic plants and animals. The Snail Kite (Rostrhamus

sociabilis), like many other species, is potentially influenced by these environmental

changes. Snail Kite populations during this century have changed considerably in number

and distribution and several authors (e.g., Sykes 1984; Beissinger 1988; Bennetts et al.

1994, Sykes et al. 1995) have suggested that changes in kite populations correspond with

changes in environmental conditions, particularly hydrology. Our knowledge, however,

of demographic processes and their influences is far from complete (Bennetts and Kitchens

1994).

Changes in the size of all populations are a sum of births and immigration minus

deaths and emigration. The Florida population of Snail Kites, however, is perhaps simpler

in that all evidence suggests that this population is closed with respect to immigration and

emigration. Snail Kites in Florida have long been known for their nomadic tendencies

(Stieglitz and Thompson 1967, Sykes 1979, Bennetts 1993), leading to suggestion that the

Florida population is not comprised of discrete subpopulations, but instead, is one

population that frequently shifts in distribution throughout the state (Bennetts and


1








2

Kitchens 1992, 1993). There has been speculation about exchange between populations

of the United States and Cuba (e.g., Sykes 1979, Beissinger et al. 1983, Sykes et al. 1995);

however, no evidence supporting this hypothesis has emerged. Thus, from a demographic

perspective, I view the Florida Snail Kite population as geographically closed, although

movements within Florida may play an important role in the population dynamics of this

species.

The birth and death processes can be conceptualized as part of the demographic

cycle represented by parameters for survival (0,) and fecundity (f) (Beissinger 1995,

Legendre and Clobert 1995)(Figure 1-1). Reliable estimates of these demographic

parameters would enable a wide range of demographic modeling (e.g., viability analyses

and risk assessments) with a higher degree of confidence. It also would increase our

predictive capability regarding the response of Snail Kites to changes in water

management. Given the scope of projects currently being planned or implemented (e.g.,

the Central and South Florida Project, the South Florida Ecosystem Restoration Initiative,

Kissimmee River Restoration, Upper St. Johns River Basin Project, Kissimmee chain of

Lakes Fishery Restoration) an improved predictive capability would be highly beneficial

and would greatly reduce controversies.

The goal of this study was to better understand Snail Kite population and spatial

dynamics and how they are affected by both natural and anthropogenic processes. I

believe that demographic models play an important role in the refinement of our

understanding of these dynamics. However, it is also my belief that reliable parameter

estimates, particularly if a model is sensitive to those parameters, are an essential basis for








3

reliable model outputs. Lastly, I believe that our models, as well as our knowledge,

should be an iterative and adaptive process (Walters 1986). As we acquire new

information or better parameter estimates, or if our predictions are falsified, we need to

adjust our models, as well as our thinking, to adapt to new information (Figure 1-2).

Objectives

In as much as there appears to be general agreement that changes in Snail Kite

populations are more sensitive to survival than to reproduction (Nichols et al. 1980,

Beissinger 1995, Sykes et al. 1995), data to estimate survival are very limited (Snyder et

al. 1989a) and as a result, reliable estimates of survival are sorely lacking (Beissinger

1995). Consequently, the first objective of this study was to estimate adult and juvenile

survival and to evaluate the influences of environmental conditions (e.g., hydrology) on

survival. In addition to this primary goal, I also recorded supplementary information on

reproductive parameters to the extent that it did not conflict with accomplishing my

primary goals and in areas where such information was not already being collected.

Most previous research on the demography of Snail Kites has focused on

reproduction. Nesting success, in particular, has received considerable attention in recent

years (e.g., Sykes 1987b, 1987c, Bennetts et al. 1988, 1994; Snyder et al. 1989a). There

remains considerable debate about what factors influences nesting success (Bennetts et al.

1994, Sykes et al. 1995); however, compared to other species, the relationship between

nesting success of Snail Kites and environmental conditions is relatively well understood.

Other reproductive parameters are less well known. Unsubstantiated estimates or

speculations have been made regarding the proportion of birds attempting to breed each








4

year and the number of nesting attempts per year (e.g., Nichols et al. 1980, Snyder et al.

1989a, Beissinger 1995); however, reliable estimates for these parameters have been

lacking.

In addition to demographic parameters, movements of Snail Kites are also

poorly understood and have been the subject of recent controversy during the planning of

marsh restoration within central and southern Florida. While long term changes in Snail

Kite distribution tend to coincide with changes in hydrologic regimes, shorter term (e.g.,

annual and seasonal) shifts do not always coincide with local hydrologic conditions

(Bennetts et al. 1994). It has been hypothesized that dispersal of kites may be in response

to hydrologic conditions (Takekawa and Beissinger 1989), localized food depletion

(Bennetts et al. 1988), or localized environmental conditions (e.g., dissolved oxygen in the

water) that may influence apple snail (Pomacea paludosa) availability (Bennetts et al.

1994). To what extent movements reflect long-term changes in habitat quality versus

short-term environmental dynamics is poorly understood, as is the bird's ability to locate

and re-colonize wetlands that have been, or will be, restored. Thus, movements are

critical to understanding Snail Kite population dynamics leading to my second primary

objective to evaluate the movement patterns of Snail Kites in Florida including rates,

locations, and what environmental conditions are correlated with movements.







5













fA



fs Adults
1-~ 1-0A

/ /Subadults ,

Juveniles












Figure 1-1. The demographic cycle of Snail Kites showing three age classes (Juveniles=J,
Subadults=S, and Adults=A). Parameters for survival (()) and fecundity (f) are shown for
each age class. Adapted from Caswell (1989), Beissinger (1995) and Legendre and
Clobert (1995).








6













Hypothesea&

Conservation
Strategy


Parameter Population
Estimation Modeling
Estimation Projection



Risk Assessment













Figure 1-2. Conceptual framework for this study. Reliable estimation of parameters is the
first step toward the development of a wide variety of demographic models.














CHAPTER 2
STUDY AREA


Snail Kites within the United States occur only in Florida (Sykes 1984). It has

been suggested (Bennetts and Kitchens 1992, 1993, 1994, Beissinger 1995) that Snail

Kites comprise one population that shifts in distribution throughout the state, rather than

there being separate subpopulations within the state. Data from studies on movements

(this study) and genetics (Rodgers and Stangel 1996) support that there is considerable

interchange of birds among wetlands in Florida. Consequently, it was deemed essential for

the scope of this study to include the entire population of Snail Kites in Florida and my

study area comprised a network of wetlands throughout central and southern Florida

within the entire documented range of Snail Kites (Figure 2-1).

Spatial Scales

Because the scale of my study is statewide, I did not focus on movements within

individual wetlands. For the purpose of this study, I considered wetlands to be distinct if

they were separated by a physical barrier (e.g., ridge or levee) and/or were under a

different hydrologic regime either through natural or managed control. Thus, adjacent

wetlands, which were once hydrologically continuous (e.g., WCA-2A and WCA-2B),

were considered separate units if they were under different water regulation schedules.

Although I recorded locations of animals by specific wetland, for many analyses I



7








8

had insufficient data to consider wetlands individually. For example, in the highly

fragmented agricultural areas, there were more than 50 wetlands used by kites during this

study. Consequently, some pooling of locations was required. For most analyses

agricultural areas were pooled into a single class of wetlands. It was not uncommon for

kites to frequent several such wetlands in immediate proximity, and I seldom (if ever)

would have had sufficient data to support estimating parameters (i.e., survival or

movement probabilities) for each of these wetlands. Other cases of pooling are described

below, or are reported on a case-specific basis based on model selection criteria.

Regions

For some analyses (e.g., survival) I treated location at a regional scale because it

was infeasible to estimate separate parameters for all wetlands. Based primarily on

watersheds, climatic factors, physiography, and management regimes, I assigned each

location to one of five primary regions (Figure 2-2). Locations not included in these five

regions (e.g., agricultural areas and isolated peripheral wetlands) were assigned to a sixth

region I call the peripheral region. Undoubtedly, there are differences in the quantity and

quality of habitats within this sixth "catch all" region (and within the 5 primary regions as

well); however, the amount of data required to partition the effects of this within-region

variability would be enormous and require significantly more effort that the scope of this

study. However, whenever the data supported partitioning beyond a regional scale I did

so.

The Everglades and Big Cypress Region is comprised of Water Conservation

Areas 1,2, and 3, Everglades National Park, and Big Cypress National Preserve. The








9

Loxahatchee Slough Region is comprised of wetlands in the drainage system of the

Loxahatchee Slough and vicinity including the Corbitt Wildlife Management Area, Pal-

Mar Water Control District, private wetlands owned by Pratt-Whitney Corp., and

wetlands within the Loxahatchee Slough owned by the City of West Palm Beach (i.e., the

West Palm Beach Water Catchment Area and vicinity). The Okeechobee Region is

comprised of Lake Okeechobee within the Herbert Hoover Dike. The Kissimmee Chain-

of-Lakes Region was comprised of all lakes within this chain including Lakes Kissimmee,

Tohopekaliga, East Tohopekaliga, Marion, Marian, Tiger, Pierce, Jackson,

and Hatchineha. The Upper St. Johns Region includes wetlands within the Upper St.

Johns River Basin, but most Snail Kites used the Blue Cypress Marsh Water Conservation

Area, Blue Cypress Water Management Area, and surrounding wetlands in private

ownership. Agricultural areas (e.g., citrus groves, canals, agricultural fields, or agricultural

retention ponds) within each of these regions, as well as all other areas not included in one

of the above regions, were assigned to the peripheral region.

Habitat Types

Snail Kites inhabit freshwater wetlands throughout central and south Florida.

There is considerable variation in the physiographic characteristics and specific plant

communities that comprise Snail Kite habitat (reviewed by Sykes et al. 1995). My

objectives did not warrant documentation of micro-habitat use by kites, nor was my

sampling (often by aircraft) conducive to recording such data. However, for some

analyses I wanted to incorporate the effects of at least a broad classification of habitats

being used by kites. This classification had to be broad enough to enable assignment of








10

locations obtained from aircraft to a given habitat type and sufficiently broad such that

micro-habitat variation did not confound the assignment given normal daily movements of

foraging birds. Consequently I assigned each location to one of five habitat types: (1)

graminoid marsh, (2) cypress prairie, (3) Okeechobee, (4) northern lakes, and (5)

miscellaneous peripheral. Graminoid marshes were generally slough and wet prairie

communities (Loveless 1959). I distinguished cypress prairies in that a dominant feature of

the landscape profile was comprised of cypress. This habitat occurred primarily in western

WCA3A, and portions of the Big Cypress National Preserve and Loxahatchee Slough.

The littoral zone of Lake Okeechobee is an extensive system of diverse marsh habitats,

and consequently had elements of at least three of my other habitat types (i.e., graminoid

marsh, northern lake, and highly disturbed). Because of this high local diversity I was

unable to assign locations to a particular type without extensive ground verification. Even

then, birds often used more than one of these habitat types within a given day. Thus, I

assigned each location at Lake Okeechobee to its own habitat type. The northern lake

habitat type consisted primarily of lakes within the Kissimmee Chain-of-Lakes, but also

included a few lakes along the Lake Wales Ridge. In contrast to Lake Okeechobee, this

habitat type generally was comprised of a narrow littoral zone (usually < 200 m) on the

periphery of these lakes. This littoral zone had a relatively steep elevation gradient

compared to other habitat types; the zone used by foraging kites often was a band of <

100 m usually dominated by maidencane (Panicum spp) interspersed with patches of

bulrush (Scirpus spp) or cattail (Typha spp). Primary nesting areas were often a zone of

cattail and/or willow (Salix spp) in the shallower zone adjacent to foraging areas. The








11

peripheral habitat type was comprised primarily of agricultural areas. These included

retention ponds for citrus groves, agricultural ditches, and other miscellaneous, usually

highly disturbed, habitats. Larger canals, not necessarily associated with agriculture, were

also included in this habitat type.

For some analyses I had insufficient data to partition locations into each of these

habitat types. Consequently, for some analyses I assigned locations to an even broader

category of lakes (i.e., Lake Okeechobee, the northern lake habitat type, and permanently

flooded canals [<0.01% of my locations]) and marshes (any non-lake habitat). This was

intended to distinguish habitats that had a permanent water source component available

(even if it was not used) from those that dried periodically.









12




E. Lake ...



Oce an:./..



Marsh














WCA
I 2A


M0XkCO 1. .....s.
at a Miami
Preserve 36












Figure 2-1. Major wetlands of South Florida referred to in this report. Wetlands are
Everglades National Park (ENP), Big Cypress National Preserve (BICY), Water
Conservation Areas 3A, 3B, 2B, 2A, Loxahatchee National Wildlife Refuge (LOX), Holey
Land Wildlife Management Area (HOLEY), West Palm Beach Water Catchment Area
(WPB), Lake Okeechobee (OKEE), Upper St. Johns [Blue Cypress] Marsh (SJM), Lake
Kissimmee (KISS), Lake Tohopekaliga (TOHO), and East Lake Tohopekaliga (ETOHO).
National Miami








. .. . .. .






Lan Wldif MnaemntAra HOE'),Wet al Bac Wte Ctcmet re
(WBLk kehbe(KE) pe t on Bu Cpes as SMLk
Kisme KS),Lk ooeafg TH) n Es aeThpeaia(TH)







13





Kissimmee
Chain-of-Lakes



Ri e.

Lake Loxaha' :e
Sloug
Okeechobee u
r
*- L*..


I

Gulf of I I
Me c Everglades

& Big Cypress I........










Figure 2-2. South Florida showing geographic regions used for some analyses in this
report. All areas not within a region shown were assigned to a peripheral region.














CHAPTER 3
ANNUAL SURVIVAL OF SNAIL KITES IN FLORIDA WITH COMPARISONS
BETWEEN RADIO TELEMETRY AND CAPTURE-RESIGHTING DATA

Introduction

For many long-lived avian species, population persistence is more sensitive to

annual survival than fecundity (Mertz 1971, Nichols et al. 1980, Beisssinger 1995).

Despite this, reliable estimates of survival remain unavailable for many species, while

extensive effort often is expended estimating reproductive parameters. Investigators also

must choose among available techniques for estimating parameters of interest. This

selection often is based on logistic constraints, or unfamiliarity with potential estimators,

rather than how procedure selection might influence resulting parameter estimates. Given

current threats to many populations, reliable demographic data are essential for effective

conservation arguments in the context of alternative management scenarios.

The Snail Kite (Rostrhamus sociabilis) is an endangered raptor whose range in the

United States is limited to central and southern Florida (Sykes et al. 1995). Florida's

wetlands have been severely altered over the past century by drainage, impoundment,

changes in water flow regimes, increased nutrient loadings, and invasion by exotic plants

and animals (Walters et al. 1992, Davis and Ogden 1994). This has resulted in what is one

of the largest ecosystem restoration projects ever undertaken (Davis and Ogden 1994).

The Snail Kite, like many other species, is potentially influenced by these, as well as other,



14








15

changes (Bennetts et al. 1994). Consequently, reliable estimates of demographic

parameters are essential to understanding population responses to environmental change

(Nichols et al. 1980).

There have been several previous reports of annual survival of Snail Kites in

Florida, although none has used reliable statistical estimators to derive estimates. Snyder

et al. (1989) estimated minimum annual survival of Snail Kites by using the number of

birds that were banded over a 10-year period from 1968-1978 that were observed alive in

1979. They did not use available capture-recapture estimators for these data because of

limited effort to resight banded birds in all but one year (Snyder et al. 1989). Hence, their

approach provides a crude indication of minimum annual survival, but does not provide

estimates that reliably can be used for demographic assessments. Several other authors

have reported estimates of Snail Kite survival based on differences between annual surveys

conducted in consecutive years (e.g., Sykes 1979, Beissinger 1988, 1995). This approach

fails to account for a high potential for confounding changes in detection probability with

changes in population size (Bennetts and Kitchens 1997a). Problems with using count

data without accounting for detectability have been well recognized (Burnham 1981,

Nichols 1992, Johnson 1995, Link and Sauer 1997). Bennetts and Kitchens (1997a)

found that the number of Snail Kites counted during these surveys was strongly influenced

by differences in observers, effort, and sites, all of which potentially influence detection

probabilities. None of these influences has been taken into account for any survival

estimates using these data. Thus, I believe that using the annual survey to estimate

survival, without accounting for detection, fails to provide reliable estimates.








16

Here, I estimate survival using reliable statistical estimators. This study also

was part of a larger study focused on both survival and movement. This provided the

opportunity to estimate survival using data obtained from both radio telemetry (the

primary tool I used to examine movement patterns) and capture-recapture (resighting);

enabling comparison of survival estimates derived from these independent data sources.

I was also able to test hypotheses about factors likely to influence survival using these

data sources.

Methods

Estimation of Survival from Radio Telemetry

Adult kites were captured using a net gun (Mechlin and Shaiffer 1979), which uses

22 caliber blank cartridges to propel a 10-foot triangular nylon net. Juveniles were

captured just prior to fledging, at approximately 30-35 d old, without a net gun. Fifteen-

gram radio transmitters were attached to birds with backpack harnesses. My goal was to

annually capture and radio tag 100 snail kites of which 60% were adults and 40%

juveniles for three consecutive years from April 1992 through April 1995. My targeted

ratio of adults to juveniles was intended to emphasize adult survival because

demography of Snail Kites probably is more sensitive to adult rather than juvenile

survival (Nichols et al. 1980, Beissinger 1995). To maintain independence of my

sample, a maximum of one juvenile per nest was equipped with a radio transmitter. I

targeted a 50:50 sex ratio of adults to keep my sample balanced. The proportion of

samples from each area was based on the annual survey to approximate the statewide

distribution (Bennetts and Kitchens 1997a). My targeted annual sample size of 100 was








17

based on having sufficient statistical power (e.g., > 0.8) to distinguish differences

(e.g. A ~=0.1-0.2) among groups (e.g., age or sex) or time periods from a

hypothesized survival estimate (4) of 0.90 (Bennetts and Kitchens 1997a). Radio-

tagged birds were located at approximately 14-day intervals from aircraft or ground

searches to determine their location and whether they were alive. All radios were

equipped with mortality censors which changed pulse rates if the transmitter had not

moved for 6-8 h. Birds with a transmitter emitting a mortality signal were then located on

the ground to verify their fate.

I estimated survival (0) of radio-tagged kites using a staggered entry design

(Pollock et al. 1989) with the Kaplan-Meier product limit estimator (Kaplan and Meier

1958). I used an arbitrary starting date of 15 April 1992 for annual survival estimates. By

this time during my first year I had a sample (n = 16) sufficient to allow reasonable

estimates of survival. Subsequent evaluation of annual survival was based on study years

(SY) from 15 April to 14 April of consecutive years (Bennetts and Kitchens 1997a).

Estimation of Survival from Banding Data

My sample of banded birds for survival analyses was obtained through a

cooperative banding effort with the Florida Game and Fresh Water Fish Commission

(GFC). My sample also was supplemented by resightings of birds banded during two

previous studies by REB (unpubl. data) and J. A. Rodgers (unpubl. data) that were

observed during this study. A previously-banded bird observed alive during my study

at time t was treated as a newly-marked individual.

I estimated annual survival from banding data using the capture-recapture








18

(resighting) models originally developed by Cormack (1964), Jolly (1965) and Seber

(1965). The basic Cormack-Jolly-Seber (CJS) approach has undergone extensive

advancement in recent years to become a flexible and unified framework capable of

handling simple to complex models of survival (Lebreton et al. 1992, Nichols 1992).

Recent approaches enable evaluation of effects attributable to individual characteristics

(e.g., age and sex) and environmental variables (e.g., weather). Additional models

have capability to incorporate transition probabilities and multiple strata (e.g.,

exchanges of individuals among geographically stratified populations)(Brownie et al.

1993, Nichols et al. 1993). All analyses of capture-recapture data were conducted in

either Program SURVIV (White 1983, White and Garrott 1990) or MSSURVIV (Hines

1994).

I conducted capture-resighting over six sampling occasions from 1992-1997.

My capture and resighting occasions corresponded with the peak fledging time of Snail

Kites (March-June)(Bennetts and Kitchens 1997a). Thus, survival estimates can be

roughly interpreted as survival from one breeding season to the next, regardless of

whether a given animal was breeding. Snail Kites have a relatively long breeding

season and are not synchronous in their breeding attempts (Snyder et al. 1989, Bennetts

and Kitchens 1997a). Consequently, the time span over which fledging, and therefore

banding, occurred was relatively long. I tried to minimize the time span of my

sampling by limiting my capture and resighting period to the peak four months of

fledging.








19

Influences on Survival

Survival of young birds tends to be lower than that of adults in many species (e.g.,

Ricklefs 1973, Loery et al. 1987). However, Ricklefs (1973) points out that "Just how

much experience the young need to attain adult behavior and physiological capabilities

(and thus adult survival rates) is open to question." Beissinger (1995) suggested that

Snail Kites have three age classes with respect to survival (juveniles or age 0-1, subadults

or age 1-2, and adults or age > 2 yr); nevertheless, the survival estimates he used for his

demographic modeling were the same for subadults and adults. I predicted lower survival

for juvenile Snail Kites compared to adults or subadults. I further hypothesized that, if

subadult survival differed from that of adults and juveniles, that it would be intermediate

between the two. To test these hypotheses, I first considered kites as adults after their

first year post fledging. Juvenile Snail Kites are capable of breeding at 9 months of age

(Snyder et al. 1989). For my capture-recapture models, resighting probability at the first

resighting period after initial capture (time 2) was considered to be equal for juveniles and

adults. Bennetts and Kitchens (1997a) tested this assumption by comparing models in

which juveniles and adults had different resighting periods at time 2 to models in which

resighting was equal for the two ages. They concluded that separate estimates for

resighting probability were not warranted. I then tested the hypothesis that adult and

subadult survival does not differ by parameterizing a CJS model such that birds banded as

juveniles were considered to have three age classes with respect to survival rates (i.e.,

juvenile survival during their first year, subadult survival their second year, and adult

survival after year two). In addition to age effects, there is substantial variability in habitat








20

quality for Snail Kites over space and time (Bennetts and Kitchens 1997a, b), which could

result in differences in survival. However, because Snail Kites are highly mobile they have

the potential to escape to other areas when conditions are poor. Adults, having had more

experience at alternative sites and the corresponding selective pressures of environmental

variability, may be less susceptible to temporal variation than younger birds.

Consequently, I hypothesized that, if temporal variation in survival exists, that it would be

greater for younger birds than for adults. I tested temporal effects using a sequence of

models analogous to Models A, B, C, and D described by Jolly (1982) and Pollock et al.

(1990). Model SS1 (Pollock's Model A) treats both survival (0) and resighting

probabilities (p) as variable over time (i.e., separate estimates of each parameter were

derived for each year). Model SS2 (Pollock's Model B) treats resighting probability, but

not survival, as variable over time. Model SS3 (Pollock's Model C) treats survival, but

not resighting probability, as variable over time. Model SS4 (Pollock's Model D) treats

both survival and resighting probabilities as constant over time. I then incorporated age

effects into this sequence of models (Pollock et al. 1990).

Because variability in habitat quality occurs over both space and time (Bennetts

and Kitchens 1997a, b), I was interested in regional (spatial) effects of survival in addition

to annual (temporal) effects. For the same reasons as temporal variation, I predicted that,

if regional variation in survival exists, that it would be greater for younger birds than for

adults. I tested for regional differences in survival using radio telemetry data two ways.

First I tested the hypothesis that differences in juvenile survival were attributable to their

region of natal origin. For this analysis, a bird was assigned to its natal region, regardless








21

of whether or not it moved subsequent to initial capture. In most cases I did not know the

natal origin of adults or their history of locations prior to capture. Consequently, I limited

this approach to juveniles.

The second approach I used for testing regional differences in survival was based

on time at risk in each region, rather than focusing only on natal region. Thus, I tested the

hypothesis that survival was affected by current location (e.g., by local factors such as

predation risk). For this analysis, a bird that moved from a given region to another was

censored (White and Garrott 1990) from the number of animals at risk for the region from

which it moved and added to the number of animals at risk in the region to which it

moved. All movements and corresponding changes in the number of animals at risk were

assigned at the midpoint of the time interval between locations. All deaths were assigned

to the region where the dead bird was found.

To test for regional effects of survival and resighting probabilities from capture-

recapture data, I generated a suite of multi-strata models analogous to the models

described above, except that they enabled stratum-specific parameter estimation (in this

case strata= 4 of the 6 regions of capture -- no captures occurred in the peripheral region

and I had too few observations in the Loxahatchee Slough to include it in the analysis)

(Brownie et al. 1993, Nichols et al. 1993). As above, I generated models with and

without age dependency, enabling us to test hypotheses that 0 and/or p were affected by

age, time, and region. Regional effects were tested only in relation to the region of

capture or resighting because capture-recapture/resighting data do not reveal where a bird

has been between resightings. Estimates for the transition probabilities among strata (i.e.,








22

the probability that an animal in stratum r at time t was alive in stratum s at time t +1,

given that it was alive at t + 1) were also generated from these models; however, my

primary interest here was site-specific estimates of 0 and p. Radio telemetry provides a

more comprehensive assessment of movement probabilities and these data are presented

elsewhere (Bennetts and Kitchens 1997a, Bennetts et al. unpubl. data).

Hypothesis Testing and Model Selection

All comparisons among survivorship curves generated by the Kaplan-Meier

estimator for radio telemetry data were made using log-rank tests (Savage 1956, Cox and

Oakes 1984). All comparisons were made using SAS (SAS Inc. 1988, White and Garrott

1990). For banding data I used a combination of likelihood-ratio tests (LRTs), Akaike's

Information Criterion (AIC)(Akaike 1973, Shibata 1989), and goodness-of-fit tests to

determine the most parsimonious model based on all combinations of effects. My testing

procedures and philosophy were described in detail elsewhere (e.g., Burnham and

Anderson 1992, Lebreton et al. 1992, Brownie et al. 1993, Nichols et al. 1993). In

contrast to LRTs, which were used for pairwise comparisons of nested models to test for

specific effects, AIC was used more as an optimization tool for any number of models,

nested or not (Lebreton et al. 1992, Spendelow et al. 1995). Models with AIC scores

differing by <1 -2 were not considered statistically different (Sakamoto et al. 1986). All

test statistics were generated using Program SURVIV (White 1983, White and Garrott

1990) or MSSURVIV (Hines 1994).








23

Results

I attached 282 radio transmitters on 271 individual Snail Kites; 11 birds were

recaptured in a subsequent year and their radios replaced. I attached 82 radios during SY

1992 and 100 each during SYs 1993 and 1994. Of the 282 radios, 165 (59%) were placed

on adults and 117 (41%) on juveniles. Of the adults, 82 (49.7%) were males and 83

(50.3%) were females. My total sample of banded birds used in CJS models was 1319.

Of these, 164 were initially banded as adults and 1155 as juveniles. However, an

additional 301 resightings of birds initially banded as juveniles supplemented my sample of

adults (Table 3-1).

Age Effects

My results from both radio telemetry and capture-recapture data indicated that

survival differed between adult and juvenile Snail Kites. Based on log-rank statistics using

radio telemetry, survival differed between these age classes for SYs 1992 (X-4.61, 1 df,

P=0.032) and 1994 (X229.52, 1 df, P<0.001), but not 1993 (X2=0.027, 1 df, P=0.869).

In both years where they differed, estimates of adult survival were higher than estimates of

juvenile survival. My capture-recapture data also indicated that survival differed between

adults and juveniles. All models that included age effects on survival had lower AIC

scores than corresponding models without age effects (Table 3-2), and LRTs between

models with and without age effects on survival strongly rejected the more reduced

models further supporting the inclusion of age effects (Table 3-3).

I used two variations of my most parsimonious model (Model SS10) to test the

hypothesis that survival of subadult (i.e., age 1-2) Snail Kites differed from adult survival.








24

Both of these models had separate parameter estimates for subadult survival; however, in

one model subadult survival was held constant among years and for the other it was

variable among years. LRTs between Model SS10 and each of these more general models

failed to reject the more reduced model (X2=2.37, 1 df, P=0.124 and X22.38, 3 df,

P=0.498 for each LRT, respectively), indicating that separate parameter estimates for

subadult survival were not warranted for these data.

Time Effects

Both data sources indicated that survival differed among years for juveniles, but

not adults. Estimates of survivorship functions of adult Snail Kites using radio-telemetry

data did not differ among years at (a = 0.05) between SYs 1992 and 1993 (X22.84, 1 df,

P=0.092), 1992 and 1994 (X2=1.76, 1 df, P=0.184), or between 1993 and 1994 (X =0.48,

1 df, P=0.486). In contrast, my estimates of survivorship of juveniles differed between

SY 1992 and 1994 (X2=6.16, 1 df, P=0.013), 1993 and 1994 (X2=12.41, 1 df, P<0.001),

but not between 1992 and 1993 (X2=1.43, 1 df, P=0.231). I also found strong evidence,

using capture-recapture data, for the inclusion of time (year) effects for juvenile, but not

adult survival. The AIC scores of models with time effects were lower than corresponding

models without time effects. LRTs between models with and without time effects also

supported this conclusion, except when time effects were limited to adult survival. Based

on my results from radio-telemetry data I generated two models in which 4b differed

between adults and juveniles and was variable among years for juveniles, but not adults.

For Model SS9, p was constant among years, and for Model SS10, p differed among

years. Model SS10 had the lowest AIC score of any model, goodness-of-fit was








25

reasonable (G=26.326, 19 df, P=0.121), and a LRT between Model SS10 and SS5 (an

identical model, except that 5 differed among years for both juveniles and adults) failed to

reject the more reduced model (SS10). These results indicated that survival differed

among years for juveniles, but not adults; and that resighting probabilities also differed

among years.

Regional Effects

There was little indication of regional differences in adult survival using either

radio-telemetry or capture-recapture data. Of 15 pairwise comparisons, using radio-

telemetry data, of adult survival between regions during each year (for which I had

sufficient data) only one differed at a=0.05. Adult survival differed between the

Everglades and Okeechobee regions during SY 1994 (X2=4.06, 1 df, P=0.044). If the a

level was adjusted for inflation due to simultaneous comparisons (e.g., using a Bonferonni

correction), none of the 15 comparisons was significant at a=0.05. For juveniles, none of

eight survivorship functions (for which I had sufficient data), based on actual time in each

region, was significant at a=0.05. For survivorship functions based on natal region, 1 of

10 comparisons was significant. There was a difference between the Okeechobee and

Everglades regions during SY 1992 (X2=4.58, 1 df, P=0.032); however this result also

would not be significant at a=0.05 if adjusted for simultaneous comparisons.

I also had some data limitations using multi-strata capture-recapture models. Two

regions (Loxahatchee Slough and the peripheral region) had insufficient data to enable

estimates. However, data from the remaining four regions supported the conclusion that

survival did not differ among regions for adults, but indicated that survival did differ








26

among regions for juveniles. A model (Model MS 13) in which survival (1) differed

among age classes, (2) differed among years for juveniles, but not adults, and (3) differed

among regions for juveniles, but not adults had the lowest AIC score (Table 3-4). An

LRT between this model and an analogous model (Model MS7) in which survival differed

among years and regions for both age classes failed to reject (2=l 1.42, 15 df, P=0.722),

further supporting that these effects were warranted for juveniles, but not adults. Similar

to the single stratum models, Model MS 13 indicated that resighting probabilities differed

among years, but also indicated differences among regions.

Parameter Estimates

Overall estimates of adult survival were very similar using the Kaplan-Meier

estimator with radio telemetry data (Table 3-5) and the CJS models with capture-

recapture data (Table 3-6). In contrast, estimates of juvenile survival tended to differ both

in the overall estimates and even in the rank order of estimates among years. Overall

estimates using multi-strata models tended to be lower for both age classes than estimates

derived from either Kaplan-Meier or CJS estimators (Table 3-7). Estimates of resighting

probabilities also differed substantially between single-stratum or multi-strata models.

The precision of individual parameter estimates ranged from 3 to 92% coefficient

of variation (CV), depending on the number of parameters being estimated and the

distribution of my sample for a given estimate. CVs for my estimates of adult survival

were 3.2% using the Kaplan-Meier estimator, 6.0% from my final single-stratum model

(Model SS10), and 4.1% using my final multi-strata model (Model MS13). Average

CVs for juvenile survival were 13.4% using the Kaplan-Meier estimator, 18.9% my final








27

single-stratum model (Model SS10), and 36.7% using my final multi-strata model (Model

MS13).

Censoring of Radio-tagged Birds

Censoring is the removal of radio-transmittered animals from a sample when the

radio-transmitter signal can no longer be detected (White and Garrott 1990). An

assumption for an unbiased estimate using the Kaplan-Meier estimator is that censoring is

random with respect to fate (Pollock et al. 1989); that is, the probability of a bird being

censored is not related to its fate. In the case of simple radio failure this assumption

probably is valid; however, when a radio is destroyed when an animal dies (e.g., during

predation or scavenging) this assumption may not be valid (White 1983). Censoring due

to radio failure would not be expected to differ among adults and juveniles. My results

indicated that the mean time to censoring differed strongly from this expectation (t=3.77,

df-179, P<0.001). Juveniles, but not adults, had a substantial surge in the number of

censored animals within the first 60 days after radio attachment (Figure 3-1). This result

would have been expected if juveniles either left the study area or experienced undetected

mortality. Dead Snail Kites were usually found in water where radio signal strength was

strongly diminished. I suspected that some mortality went undetected as a result.

Consequently, during SY 1994 I increased my search effort for missing birds. I then

examined the proportions of censored and dead birds during the first 180 d after radio

attachment (i.e., before radio batteries likely died). The proportions of adults censored

and confirmed dead remained relatively constant among years (X= 1.02, 2 df,

P=0.601)(Figure 3-2). In contrast, the proportions of juveniles censored and confirmed








28

dead were similar during SY 1992 and 1993, but differed during SY 1994, when search

effort was increased (X230.25, 2 df, P<0.001). During SY 1994, the proportion of birds

confirmed dead substantially increased and the proportion of censored birds substantially

decreased. The proportion of censored juveniles during 1994 also closely matched the

proportion of censored adults, which it had not during 1992 or 1993.



Discussion

Comparisons Between Data Obtained Using Radio Telemetry and Capture-resighting

The results from radio telemetry and banding data were generally consistent in

identifying sources of variation. Both data sources indicated that survival differed between

age classes and among years for juveniles, but not adults. Single- and multi-strata

capture-recapture models also indicated similar sources of variation for survival and

resighting probabilities, except that the multi-strata models indicated additional regional

effects. In contrast to sources of variation, there were considerable differences in some

parameter estimates among data sources. Although both sources of data indicated

differences among years for juvenile survival, the parameter estimates from these two data

sources differed markedly and were not even consistent in their relative ranking among

years. Estimates of juvenile survival during 1992 and 1993 were higher using radio-

telemetry data than for either capture-recapture models. I believe that this was due to a

bias for my estimates using radio telemetry data during those years. My results from

censored radio-tagged birds indicated that I was finding dead juveniles during 1994 when

search effort was increased; whereas a substantial number of dead birds may have gone








29

undetected during 1992 and 1993. Thus, my survival estimates using radio telemetry

probably were biased high for juveniles. I had no such evidence for adults.

Another assumption using radio telemetry to estimate survival is that the radio

transmitter does not affect survival (White and Garrott 1990). There has been substantial

evidence in recent years to suggest that, for some species, radio transmitters may

negatively affect survival (e.g., Marks and Marks 1987, Burger et al. 1991, Paton et al.

1991). Bennetts and Kitchens (1997a) tested the hypothesis that radio transmitters

negatively affect survival of Snail Kites using capture-recapture of birds with and without

radio transmitters. They had reasonable power to detect any substantial differences, yet

found no effect.

In contrast to radio telemetry, I had no reason to suspect that violations of my CJS

models significantly biased my results. Probably the most substantial violation was for the

assumption that capture and release of animals occurs over brief time intervals (Pollock et

al. 1990). This assumption enables a clear definition of the interval over which survival is

measured and helps to standardize intervals being compared. The life history of Snail

Kites makes this assumption difficult to meet. However, I do not believe that violation

of this assumption caused substantial bias to my estimates. For adults, the highest risk

of mortality appeared to be during the fall and winter (Bennetts and Kitchens 1997a,

Bennetts et al., unpubl. data). Thus, animals within a given study year all experience

the same period of high risk. For juveniles, the highest risk of mortality occurs over

the first few months post fledging and all juveniles within a given cohort also were

exposed to that period of high risk.








30

Band loss probably was negligible on my study because all but 19 (99%) birds

were marked with riveted aluminum bands that were extremely unlikely to have been

lost. The remaining 1 % were made of PVC and anecdotal evidence suggests that band

loss from these bands also was negligible. I also believe that capture and release did

not substantially influence the subsequent resighting of animals. Snail Kites are

relatively tolerant of human presence and often allow humans to approach relatively

close (Beissinger 1988). In addition, most birds were nesting at the time of resighting

and tended to stay close enough to their nest to enable bands to be read with minimal

difficulty.

Parameter Estimates

Because of the potential for biased estimates of juvenile survival using radio

telemetry, I was more confident in my estimates using capture-recapture for this

parameter. I also have greater confidence that my parameter estimates using single-

stratum reflects actual survival. My data indicated that, at least for juvenile survival,

regional effects were warranted. However, capture-recapture models estimate apparent

survival, such that permanent emigration (i.e., permanent for the study) is confounded

with actual survival. Because my data were insufficient to partition among two age

classes and all six regions using multi-strata models, there is a potential for increased

confounding of these two components of apparent survival. First, the four regions for

which I had sufficient data were those with a greater number of sightings. This could be

due either to greater use of these regions and/or a greater probability of observing birds

that were present. This could account for the higher estimates of resighting probability








31

observed from my multi-strata models. Similarly, any permanent emigration to these

regions would have been included in the resulting estimates as decreased apparent

survival. My single-stratum models included these regions because I was not attempting

to derive separate parameter estimates. Thus, although I would expect my estimates of

apparent survival using multi-strata models to be less biased because I was accounting for

regional heterogeneity, there also may have been greater confounding of actual survival

and permanent emigration in these estimates. This would explain the lower estimates of

survival from my multi-strata models compared to estimates from radio telemetry or

single-stratum models.

Nichols et al. (1980) reported that adult survival of Snail Kites in Florida was 0.90.

This was not based on a statistical estimator; rather, it was their "best guess" for

demographic modeling. Similarly, Snyder et al. (1989) suggested that during non-drought

years annual adult survival of Snail Kites probably exceeds 0.90, although this value also

was not derived using any specified estimator. Beissinger (1995) later reported adult

survival during non-drought years as 0.95 based on Snyder et al.'s suggestion. My

estimates were similar to these previous reports of adult survival (4,=0.89 and 0.92 from

Kaplan-Meier and CJS estimators, respectively), but were based on reliable statistical

estimators. In contrast to adults, my estimates of juvenile survival were not consistent

with some previous estimates. Beissinger reported juvenile survival during non-drought

years as 0.90. Nichols et al. (1980) reported a "best guess" of 0.58 for juvenile survival.

My data suggest that juvenile survival may be substantially lower than Beissinger's

estimate, but similar to the "best guess" reported by Nichols et al. (1980).








32

Effects of Age. Time, and Region on Survival

As predicted, I observed differences in survival between juvenile and adult Snail

Kites, although separate estimates of subadult survival were not indicated by my data.

Younger birds may have lesser foraging skill than adults (e.g., Verbeek 1977, Bennetts

and McClelland 1997) and also may be more vulnerable to predation due to a lack of

experience. My results also supported my hypothesis that younger birds are more

susceptible to environmental variation than adults. Survival ofjuveniles, but not adults,

differed among both years and regions. Environmental conditions, and consequently

habitat quality for Snail Kites, may be quite variable in central and south Florida

(Beissinger 1986, Bennetts and Kitchens 1997a). Adult kites are well adapted to this

variability and are quite capable of moving throughout their range in response to changing

conditions (Bennetts and Kitchens 1997a, b). In contrast, juveniles that have not yet

experienced alternative locations, may be less efficient at locating alternative sites when

local conditions are not favorable. Consequently, juveniles may be more sensitive to both

spatial and temporal variation in the environment.

Although my data indicate that juveniles, but not adults were sensitive to

environmental variability, it has been suggested that survival during drought years may be

substantially lower than during high-water years (Beissinger 1988, Takekawa and

Beissinger 1989). Beissinger (1995) found drought-year survival to be one of the most

sensitive parameters of his population viability model. Thus, adults may be susceptible to

this more extreme case of environmental variability. Because I did not encounter drought

conditions during this study, my results can not reliably be extended to drought years.








33

Thus, there remains a need for reliable estimates of survival during drought years (see also

Beissinger 1995).

Implications of Resighting Probabilities

Although emphasis of capture-recapture models is usually on survival, annual

differences in resighting probabilities of Snail Kites also may have implications for a

statewide monitoring program. An annual survey of Snail Kites was conducted every year

from 1969-1994. Reported uses of these data include estimation of survival based on

differences in counts between consecutive years (e.g., Beissinger 1988, 1995), and as an

index of population size for comparisons between areas or years (Rodgers et al. 1988).

Using count data for these purposes requires an assumption that either the survey

represents a complete census, or that the proportion of birds detected does not differ

between years or areas being compared (Lancia et al. 1994). The resighting probabilities I

estimated suggest that the annual survey fails to meet either of these assumptions. My

overall resighting probability using CJS models was 0.19; whereas, a census is a complete

count of all animals (Lancia et al. 1994). My results also indicated that resighting

probability differed among years and regions; a result inconsistent with the assumption

that the proportion of birds detected during the annual survey is constant. It must be

taken into account that my estimates were derived during spring, whereas the annual

survey is conducted during autumn. However, my results certainly supports concern for

the validity of using these data for these applications.









34

Table 3-1. Capture-resighting summary of adult and juvenile Snail Kites in Florida from
1992-1997.


Year of Next Resighting
Year of
Last Birds Banded as Adults Birds Banded as Juveniles"
Capture or Never Never
Resighting 92 93 94 95 96 97 e e 92 93 94 95 96 97
esi g Resighted Resighted
1992 -- 4 10 5 1 6 23 -- 11 14 8 5 7 104

1993 14 4 2 10 26 10 9 17 14 206

1994 13 11 8 45 21 12 21 88

1995 -- 5 5 14 -- -- 36 59 148

1996 6 13 -- -- -- -- -- 46 158

Total No
To l 0 4 24 22 19 27 0 11 24 38 70 147 -
Resighted

Total New
taNew 49 52 53 2 0 8 149 245 118 205 134 304
Captures

Total No.
Releasedb 49 56 77 24 19 35 149 256 142 243 204 451
a Considered to be adults at time 2 of each cohort.
b Includes total resighted and new captures; however analysis is parameterized such that
juveniles resighted as adults also contribute to estimation of adult surival.








35

Table 3-2. Description of single-stratum Cormack-Jolly-Seber
(CJS) models and their corresponding Akaike Information
Criteria (AIC) scores. Parameter structure indicates whether
survival ($) and/or resighting probability (p) was dependent
on time (t) and/or age(a).
Modl Parameter No.
Model AIC
Structure Parameters

SS1 p, P 9 198.68

SS2 Obp, 6 214.07

SS3 4 p 6 236.04

SS4 Op 2 289.37

SS5 ~tPt 14" 171.77

SS6 ,p 11 195.87

SS7 P, 7 194.21

SS8 ,p 3 275.15

SS9 O(,,),pb 7 192.16

SS10 ow.xaptb 11 166.34

'Because b and p were both variable over time, we were
only able to estimate a product of the two for the last time
period (Lebreton et al. 1992).
b Survival was time dependent for juveniles, but not adults.








36


Table 3-3. Likelihood ratio tests (LRTs) between Cormack-Jolly-Seber (CJS) models
used to test whether survival (p) or resighting (p) probabilities differed among age
classes or years (time).
General Reduced Parameter Effect 2 X2
X df P > i
Model Model Tested Tested

SS8 SS4 Age 16.219 1 <0.001

SS7 SS2 Age 21.861 1 <0.001

SS6 SS3 Age 50.168 5 <0.001

SS1 SS2 Time 21.393 3 <0.001

SS3 SS4 Time 61.335 4 <0.001

SS6 SS8 Time 95.284 8 <0.001

SS6 SS9 Time' 4.289 4 0.368

SS5 SS10 Time' 0.570 3 0.903

SS2 SS4 p Time 83.300 4 <0.001

SS1 SS3 p Time 43.358 3 <0.001

SS7 SS8 p Time 88.942 4 <0.001

SS10 SS9 p Time 33.814 4 <0.001

STests for time variation of survival of adults only.








37

Table 3-4. Description of multi-strata models and their corresponding
Akaike Information Criteria (AIC) scores. Parameter structure indicates
whether survival (0), resighting probability (p), and/or transition
(movement) probability (tI) was dependent on age(a), time (t), and/or region
(r).

Parameter No.
Model AIC
Structure Parameters

MS 1 Ar p, 176 989.173

MS2 O, pr gr 96 982.051

MS3 r ArP r 36 940.507

MS4 5br pA, r ,r 52 900.544

MS5 p, o 20 1032.120

MS6 a,,Pr, A, 36 982.580

MS7 pAr, 80 881.293

MS8 ,ar P O, 36 934.653

MS9 A P, Os, 46 896.930

MS10 pat() P r ,r 50 881.364

MS11 ,Oa Pt Ar 52 894.622

MS12 ,OtX Pr A 68 867.264

MS13 ,,.. .,) P, r r 65 862.713








38


Table 3-5. Annual estimates and standard errors for adult and juvenile
survival (4) of Snail Kites for study years (SYs) 1992, 1993, and 1994
using data from radio telemetry.
Adults Juveniles
Study
Year 4 SE (4) # SE (0)

1992 0.962 0.038 0.825 0.080

1993 0.858 0.063 0.867 0.088

1994 0.883 0.042 0.439 0.090

Overall* 0.894 0.029 0.671 0.059

"Estimated using a pooled sample of all years. The arithmetic mean
gives equal weight to each annual estimate, whereas the pooled sample
essentially weights by sample size.








39


Table 3-6. Parameter estimates for the Cormack-Jolly-Seber (CJS) model SS10 in
which survival (4) differed between adults and juveniles. Under this model, survival
was constant among years for adults, but differed among years for juveniles. Resighting
probabilities (p) differed among years.
Adults Juveniles Adults
Study SA S A
Year SE (E ) SE (P) p SE (p)

1992 0.922 0.055 0.527 0.075 0.121 0.031

1993 0.922 0.055 0.350 0.051 0.198 0.030

1994 0.922 0.055 0.783 0.122 0.166 0.027

1995 0.922 0.055 0.942 0.187 0.168 0.034

1996 0.922 0.055 0.701 0.211 0.322 0.087

Overall 0.922* 0.055a 0.530b 0.047" 0.189" 0.020C

Adult survival in Model SS10 is constant over time.
b Estimated using Model SS7, which is identical to my selected model (SS10) except
that f, is constant over time. This approach is equivalent to using a weighted mean
estimate where weights are based on the variance-covariance matrix.
c Estimated using Model SS9, which is identical to my selected model (SS10) except
that p is constant over time. This approach is equivalent to using a weighted mean
estimate where weights are based on the variance-covariance matrix.








40

Table 3-7. Parameter estimates for my most parsimonious multi-stratum model
(MS 13), in which survival differs between adults and juveniles, survival is constant
among years and regions for adults, and survival differs among years and regions for
juveniles. Resighting probability in this model differs among years and regions.


Adults Juveniles Adults

Study Region' 1 SE() SE() p SE(p)
Year

1992 EVER 0.822 0.034 0.487 0.214 0.000 0.000
1992 OKEE 0.822 0.034 0.740 0.142 0.053 0.031
1992 KISS 0.822 0.034 0.367 0.116 0.358 0.135
1992 USJ 0.822 0.034 0(447 0.128 0.280 0.113
1993 EVER 0.822 0.034 0.404 0.097 0.222 0.057
1993 OKEE 0.822 0.034 0.436 0.089 0.110 0.037
1993 KISS 0.822 0.034 0.102 0.049 0.527 0.132
1993 USJ 0.822 0.034 0.343 0.128 0.113 0.065
1994 EVER 0.822 0.034 0.720 0.107 0.245 0.043
1994 OKEE 0.822 0.034 0.301 0.276 0.081 0.035
1994 KISS 0.822 0.034 0.275 0.097 0.248 0.080
1994 USJ 0.822 0.034 <0.00 <0.001 0.304 0.114
1995 EVER 0.822 0.034 0.454 0.074 0.199 0.037
1995 OKEE 0.822 0.034 0.437 0.198 0.387 0.103
1995 KISS 0.822 0.034 0.921 0.188 0.194 0.068
1995 USJ 0.822 0.034 1.000 0.317 0.368 0.127
1996 EVER 0.822 0.034 0.234 0.074 0.568 0.095
1996 OKEE 0.822 0.034 <0.00 0.412 0.389 0.121
1996 KISS 0.822 0.034 0.613 0.298 0.749 0.204
1996 USJ 0.822 0.034 0.248 0.175 0.756 0.242
Overall 0.822 0.034 0.441b 0.036b
Regions are Everglades (EVER), Okeechobee (OKEE), Kissimmee (KISS), and
Upper St. Johns (USJ). There were insufficient sightings to include the Loxahatchee
Slough Region.
b Based on model MS9 for which survival is considered constant among years and
regions.










41









50
mo Adult

)40

0
30-
0.
CO
E



) 10-
(D
a)

0 60 120 180 240 300 360 420 480 540 600 660 720 780
Day From Attachement


50
S Juveniles




S30
E
C/ 20


10-



O 60 120 180 240 300 360 420 480 540 600 660 720 780
Day From Attachement









Figure 3-1. Percentage of radio-transmittered adult and juvenile Snail Kites that were
censored in each 60-day time interval from the time of attachment.









42







50
SAdults Censored
) MU Confirmed Dead
4. 40
E

.. 30-
c)
10
0










JUVenileS Censored
II Confirmed Dead
a 40-
a)













30 5
S20














i^,,




1992 1993 1994

Year






Figure 3-2. Percentage of adult and juvenile Snail Kites from each sampling cohort (i.e.,
Juvthe year that they fledged or were captured) that died or were censored during the first
180 d after radio attachment each year.Dead
'a 40
E


(D















Figure 3-2. Percentage of adult and juvenile Snail Kites from each sampling cohort (i.e.,
the year that they fledged or were captured) that died or were censored during the first
180 d after radio attachment each year.














CHAPTER 4
WITHIN-YEAR SURVIVAL PATTERNS OF SNAIL KITES IN FLORIDA



A common generalization, that recent evidence supports, is that juvenile survival

of many avian species tends to be lower than that of adults (e.g., Loery et al 1987, Nichols

et al. 1992). This difference may be attributable to a lack of experience of younger birds

for foraging, dispersal, and avoidance of predators. There is considerably less evidence

for the time at which the survival rate of younger birds becomes similar to that of adults

(Ricklefs 1973, Loery et al. 1987). In chapter 3 I examined annual survival of Snail Kites

(Rostrhamus sociabilis) in Florida using a combination of radio telemetry and capture-

recapture data. They found that annual survival of juveniles (birds of age <1 yr) was

lower than that of adults (birds of age >lyr), but that delineation of a subadult age class

(age 1-2 yrs) was not supported by my data. This suggests that survival rates between age

classes were similar after one year, but does not indicate whether or not it requires less

than a year to become similar. Neither capture-recapture nor telemetry data enables any

evaluation of the patterns of survival within the sampling intervals. However, the intervals

using radio telemetry are usually very short compared to capture-recapture data, even

though estimates are often reported on an annual basis. Thus, within-year patterns can

provide considerable information regarding the time over which juvenile and adult survival

rates converge. These patterns also may provide considerable insights about how the risk


43








44

of mortality changes over time, which can further our understanding of what factors

influence survival. Here I use survivorship functions derived from radio-telemetry data to

refine our knowledge of the time at which juvenile survival becomes similar to that of

adults and to evaluate the seasonal patterns of risk for each age class.

Methods

My goal was to annually capture and radio tag 100 snail kites of which 60% were

adults and 40% juveniles for three consecutive years from April 1992 through April 1995.

My targeted ratio of adults to juveniles was intended to emphasize adult survival because

demography of Snail Kites probably is more sensitive to adult rather than juvenile survival

(Nichols et al. 1980, Beissinger 1995). To maintain independence of my sample, a

maximum of one juvenile per nest was equipped with a radio transmitter. I targeted a

50:50 sex ratio of adults to keep my sample balanced. The proportion of samples from

each area was based on an annual survey to approximate the statewide distribution

(Bennetts and Kitchens 1997a).

Adult kites were captured using a net gun (Mechlin and Shaiffer 1979), which uses

22 caliber blank cartridges to propel a 10-foot triangular nylon net. Juveniles were

captured just prior to fledging, at approximately 30-35 d old, without a net gun. Fifteen-

gram radio transmitters were attached to birds with backpack harnesses. Radio-tagged

birds were located at approximately 14-day intervals from aircraft or ground searches to

determine their location and whether they were alive. All radios were equipped with

mortality censors which changed pulse rates if the transmitter had not moved for 6-8 h.








45

Birds with a transmitter emitting a mortality signal were then located on the ground to

verify their fate.

I used logistic regression as a preliminary analysis for the influences on survival. I

found (Chapter 3) that annual survival differed between adults and juveniles, and among

years for juveniles, but not adults. However, here I was interested in the timing of

mortality, rather than the magnitude of annual estimates. Thus, from my preliminary

analysis, I particularly wanted to determine if there was evidence for a within- and

between-year interaction, which would have indicated that the within-year patterns

differed from year to year. Based on these results and because I was interested in the

timing of mortality, rather than annual differences in magnitude, I pooled my samples from

each year for each age class to better illustrate the overall timing of mortality. However, I

did not pool age classes because of apparent differences in timing. It should be noted,

however, that using an unpooled sample does not alter any of the conclusions reported

here.

I generated survivorship functions using a staggered entry design (Pollock et al.

1989) with the Kaplan-Meier product limit estimator (Kaplan and Meier 1958). I used an

arbitrary starting date of 15 April for survivorship functions. By this time during my first

year I had a sample sufficient (n = 16) of adults to allow reasonable estimates of survival.

This time also corresponded with the beginning of peak fledging. Thus juveniles were

added to my sample as they fledged, rather than some having fledged during the previous

calender year. Thus survivorship functions are based on a study year (SY) from 15 April

of calender year t to 14 April of calender year t+1. All comparisons among survivorship








46

curves generated by the Kaplan-Meier estimator for radio telemetry data were made using

log-rank tests (Savage 1956, Cox and Oakes 1984) with a modified version of the SAS

code (SAS Inc. 1988) reported by White and Garrott (1990).

Cox (1972) proposed a nonparametric model for the instantaneous probability of

an animal's death, or hazard function. The hazard function is a measure of instantaneous

risk of mortality as a function of age or time. I estimated the hazard function ( fi) for

discrete one-month intervals as the number of animals dying during the interval divided by

the number of animals surviving over that interval (Lee 1980). In contrast to survivorship

functions, which were intended to assess if there were differences in survivorship, the

hazard function better illustrates how the risk of mortality changes over time. Because

some risk of mortality of juveniles may have been more related to their age, than time, I

estimated an additional hazard function for juveniles based on age (i.e., time since

fledging). I did not do this for adults because, in most cases, I did not know their age, and

age was less likely to have been a factor for adults.

Results

I captured a total of 271 individual Snail Kites and attached 282 radio transmitters.

Eleven of the 271 birds were recaptured in a subsequent year and their radios replaced.

Of the 282 radios, 165 were placed on adults; of which, 45 were during SY 1992 and 60

each during SYs 1993 and 1994. I attached a total of 117 radios on juveniles; of which

37 were during SY 1992 and 40 each during SYs 1993 and 1994. Of the adults, 82

(49.7%) were males and 83 (50.3%) were females.








47

My preliminary logistic regression model confirmed what had been previously

reported (Bennetts and Kitchens 1997a). Based on Akaike's information criterion

(AIC)(Akaike 1973, Burnham and Anderson 1992), the most parsimonious model was

one that included the effects of age, year, season, an age*year interaction, and an

age*season interaction (Bennetts and Kitchens 1997a). The general effects of age and

annual differences have been discussed in detail elsewhere (Bennetts and Kitchens 1997a,

Chapter 3). Of particular interest here was an age*season interaction, which indicates that

the seasonal patterns of survival were not similar between age classes. Also of interest

were that a season*year interaction or a three-way interaction of age*season*year were

not warranted for these data, which supports my belief that the seasonal pattern for each

age class does not differ among years.

A log rank test supported the conclusion that survivorship functions differed

between adult and juvenile Snail Kites (X2=20.76, df=l, P<0.001). A further analysis

indicated that these differences were substantial for the first four months of the year

(X2=33.69, df=l, P<0.001), but not the remaining eight months (X 0.47, df=l, P=0.494).

Juvenile survivorship dropped sharply for the first four months, after which it leveled off

considerably to a much slower rate of decline until a relatively less dramatic decrease at

about eight months (Figure 4-1). In contrast, adult survivorship declined at a relatively

slow rate for approximately the first eight months, after which there was a relatively

moderate decrease in survival similar to that of juveniles at this time. Thus, survivorship

functions of the two age classes became quite similar after the first four months.








48

It was also apparent from hazard functions that the risk of mortality was not

constant over time for either age class (Figure 4-2). Juveniles had the highest risk of

mortality during the first four months; whereas this, was a period of low risk for adults.

Both age classes then experienced increased risk of mortality during winter and early

spring. The hazard function for juveniles based on age indicated that the highest risk of

mortality occurred between 30 and 60 days post fledging, although it was nearly as high

for the first 30 days (Figure 4-3).

Discussion

My data indicate striking differences in seasonal patterns of survivorship between

adult and juvenile Snail Kites. Juveniles were at greatest risk of mortality during late

spring and early summer; adults were at greater risk during winter and early spring. The

period of high risk for juvenile Snail Kites during late spring is not surprising. My

estimated hazard function based on age, rather than time, revealed that the period of

greatest risk is between 30 and 60 days after fledging. Although juveniles are least

experienced during the first 30 days, they also are still attended by their parents. Snyder et

al. (1989) suggested that the post-fledging dependency period lasts for about six weeks.

Hence, during the period between 30 and 60 days after fledging, juveniles are becoming

independent of their parents, are foraging on their own, and dispersing into unfamiliar

areas (Bennetts and Kitchens 1997a). Juveniles that survived the first few months post

fledging appeared to be most vulnerable at the same time as peak mortality for adults.

In contrast to juveniles, it was less clear why adult mortality was highest in winter,

although I offer several hypotheses. During late winter and early spring there is a








49

potential for increased risk of predation, which probably was the most frequent cause of

adult mortality (Bennetts and Kitchens 1997a, Chapter 5). Late winter and early spring

corresponds to the time that adults begin courtship. Consequently, at this time they are

exhibiting behaviors more conspicuous to potential predators such as courtship flights and

vocalizations. Adults also may be less wary of predators if their attention is on

procurement of a mate. Evidence also suggests that Great-homed Owls (Bubo

virginianus), which forage at night, were the most common predators. During winter,

leaves are absent from willows, which is the most commonly used species for communal

roosting by Snail Kites (Sykes et al. 1995, Darby et al. 1996). Thus, concentrations of

roosting kites may have been more visible to nocturnal predators during this period.

Apple snails (Pomacea paludosa) are the almost exclusive food of Snail Kites.

These snails are aquatic, and are most vulnerable to kites while at the water surface.

Apple snails have both gills and lungs and the frequency at which they come to the surface

for air is inversely related to the amount of dissolved oxygen in the water (McClary 1964).

Colder temperatures result in higher levels of dissolved oxygen and reduced activity and

oxygen consumption by snails (Freiburg and Hazelwood 1977, Hanning 1978).

Consequently, cold temperatures during winter result in fewer foraging bouts and lower

capture success by Snail Kites (Cary 1985, Sykes et al. 1995).

Communal roosting during winter also could result in a greater potential for the

transmission of infectious disease. Although the extent of disease as a source of mortality

of Snail Kites is unknown, I did find at least one dead bird that was diagnosed with an

infection of the coelomic cavity (D.J. Forrester and M.G. Spalding, Laboratory of Wildlife








50

Disease Research, University of Florida, unpubl. data, Bennetts and Kitchens 1997a,

Chapter 5).

My data indicate that survivorship of juvenile Snail Kites becomes similar to that of

adults after about four months. Estimated survivorship and hazard functions based on

time, and hazard functions of juveniles based on age all indicated that the period of

greatest risk for juveniles was during their first four months, after which survivorship

became remarkably similar to that of adults. However, I do emphasize that my study was

conducted during a period of favorable environmental conditions. During periods of food

shortage (e.g., widespread droughts), it would not be surprising for there to be a greater

disparity between age classes or an increased time for the risk of juvenile mortality to

become similar to that of adults.









51








1.0-


0.8 -......



0.6-


-Adults
0.4 ... Juveniles


15APR 15JUN 15AUG 15OCT 15DEC 15FEB 14APR
Date











C,.


Adults
SJuveniles


15APR 15JUN 15AUG 15OCT 15DEC 15 FEB 14APR
Date



Figure 4-1. Survivorship functions of adult and juvenile Snail Kites from a pooled sample
of 3 years (top). Because I was interested in temporal pattern rather than magnitude, I
aligned the functions without regard to magnitude, to illustrate at what point in time the
functions become similar (bottom).









52
















0.20
... Adult
0 -*- Juvenile
S0.15
U-
"0

0.10 -
I


E 0.05


S .. ........... .
0.00 .f I I
A M J J A S O N D J F M A
Month













Figure 4-2. Estimated hazard functions constructed at monthly intervals for adult and
juvenile Snail Kites.








53














0.12

0 0.10

u. 0.08
"2\
0.06-

i 0.04

U 0.02-

0.00 1 I
15 45 75 105 135 165 195 225 255 285 315 345
Days Post Fledging












Figure 4-3. Estimated hazard function for juveniles based on age, rather than time. This
function was constructed at monthly intervals starting at the time of fledging.













CHAPTER 5
CAUSES OF MORTALITY OF POST-FLEDGING JUVENILE
AND ADULT SNAIL KITES IN FLORIDA


Previous demographic studies of Snail Kites (Rostrhamus sociabilis) have focused

primarily on reproduction, though survival may be the most important demographic

parameter for this species (Nichols et al. 1980, Beissinger 1995, Sykes et al. 1995). There

have been several reports of causes of death of nestling Snail Kites (e.g., Sykes 1987,

Bennetts et al. 1988), but such information has been largely speculative for post-fledging

juveniles and adults (Beissinger 1986, Sykes et al. 1995). The purpose of this paper is to

provide an indication of the relative frequencies of different causes of mortality of post-

fledging juvenile and adult kites during my period of study.

My study was conducted between April 1992 and April 1995 in central and south

Florida as part of a larger study of survival and movements of Snail Kites in Florida

(Bennetts and Kitchens 1997). I attached 282 radio transmitters on 271 individual Snail

Kites, with 11 birds having been recaptured in a subsequent year and their radios

replaced. I monitored birds primarily by aircraft, although verification and retrieval of

dead birds was conducted by airboat or on foot. The average interval between

consecutive locations of birds was 13.5 (+7.9 SD) days (Bennetts and Kitchens 1997). All

radio transmitters were equipped with a mortality switch that, upon prolonged lack of

motion (-6 h), altered the pulse rate enabling us to remotely determine if a bird was dead


54








55

or had dropped its radio. I attempted to find all birds emitting a mortality signal the same

day that the signal was detected, although logistic constraints sometimes precluded

attaining this goal.

I found 47 dead or moribund Snail Kites, of which 31 were post-fledging juveniles

and 16 were adults. Of these 47, I assigned a likely cause of death to 24 (51%). Most

(81%) dead birds were found using radio telemetry, although I have included birds that

were found more incidentally during this study (19%). Carcasses (n=7) in which the

fleshy parts had not been consumed or that did not exhibit extreme autolysis were sent for

necropsy either to the Laboratory of Wildlife Disease Research, College of Veterinary

Medicine at the University of Florida, or the U.S. Geological Survey, National Wildlife

Health Research Center. I emphasize from the outset that the exact cause of death can

seldom be determined with certainty without finding each carcass while fresh and

conducting a necropsy. Even when a necropsy was performed, a conclusive diagnosis was

seldom possible and several contributory factors were often confounded. Despite my

effort to minimize the time until a necropsy could be performed, severe autolysis was

common due to the time interval between death and detection of the mortality signal.

Thus, my intention here is to provide a crude indication of the relative frequencies of

different causes of death, rather than a definitive assessment.

Beissinger (1986, 1988) suggested that adult mortality due to predation is

probably rare. In contrast, I found predation to be the most frequently suspected cause of

death, when it could be identified (Table 5-1). However, birds in which I assigned

predation to be the probable cause were limited to those in which I had ancillary








56

information, in addition to the carcass having been eaten. Otherwise, the death was

classified as unknown. Four carcasses were found at sites (e.g., a feeding or plucking

perch) with carcasses of other species. Six others had been plucked, a behavior not

associated with any local scavengers. The predator most frequently suspected was the

Great Homed Owl (BWhu virginianus). In four cases (not including those listed above)

adults had been found decapitated on their nest, which is a common signature of Great-

homed Owl predation (Nesbit 1975). In several other cases, a Great-homed Owl had

been seen frequenting the area. Barred Owls (Stri varia) also have been previously

suspected to have killed at least one adult female based on feathers left at the nest (Sykes

et al. 1995). Peregrine Falcons (Falco peregrinus) also are occasionally observed in the

area during migration, but have not been reported to take Snail Kites.

Starvation was the second most frequently suspected cause of death for juveniles,

although only two cases were diagnosed (N. Thomas, National Wildlife Health Center,

unpubl. data). In the two cases where starvation was assigned to be the probable cause of

death, each bird was found alive, but in a severely weakened state from which they did not

recover. Also in each case the birds were found in marine environments where apple

snails, the primary food of kites, were completely lacking. These deaths were attributed to

inexperienced birds that dispersed in a direction where they were unable to find sufficient

food. One band return on a previous study (Bennetts et al. 1988) also was a juvenile

found dead in a marine environment (Sanibel Island, Florida). One additional juvenile

was severely emaciated at the time of death, but the diagnosis was not conclusive and the

bird may also have had an intestinal disease (N. Thomas, National Wildlife Health Center,








57

unpubl. data). In this case, emaciation may have been a symptom of illness, rather than a

cause of death. Starvation also may have been underestimated in my sample. For

example, juvenile birds dispersing to habitats atypical of adults may have had a lower

probability of detection due to less intensive searches of these habitats. These areas also

may be more likely to have less predictable food resources. In addition, starvation may be

a more frequent cause of death for both age classes during drought years, when food may

be scarce (Beissinger 1986); my data were collected only during non-drought years.

Other causes of death included vehicle collisions, disease, and one probable

gunshot. Vehicle collisions were observed for both age classes and occurred where birds

had been observed foraging or nesting adjacent to roadways. One adult female probably

died of an infection of the coelomic cavity (D.J. Forrester and M.G. Spalding, Laboratory

of Wildlife Disease Research, University of Florida, unpubl. data). The skeletal remains of

one juvenile had a probable gunshot (shotgun) hole through its sternum, but I was unable

to confirm conclusively if this was the cause of death.

It has been suggested that most adult mortality of Snail Kites in Florida occurs

during droughts and is likely caused by starvation or risks encountered during dispersal

(Beissinger 1986, 1995, Snyder et al. 1989). This inference was derived primarily from

changes in the number of kites counted during an annual survey, rather than from

empirical evidence of actual mortality. Although I agree with these authors that an

increased risk of mortality during widespread droughts is likely, the actual extent of

mortality attributed to specific causes is not known (Bennetts et al. 1994, Sykes et al.








58

1995, Bennetts and Kitchens 1997). However, I emphasize that my study was conducted

during non-drought conditions and my inferences are limited accordingly.

My inferences also were likely influenced by my methods. For example, two of

three birds I found that had been hit by vehicles were found without using radio

transmitters. This probably is because they were along roadways where detection of the

birds was likely. Finding birds that died of other causes would have been far less likely

without the use of radio transmitters. It is also imperative that, if a study goal is to

determine the causes of mortality, sampling intervals of radio-tagged birds be frequent.

Even with my relatively intensive sampling effort, severe autolysis precluded much of the

information that could have been derived had the carcasses been found fresh. In Florida's

subtropical environment, decomposition occurs quickly. Consequently, causes of death

that required examination of soft tissues for diagnosis are likely to have been

underestimated in my sample.








59

Table 5-1. Probable causes of mortality of Snail Kites recovered in Florida from 1992-
1995.


Juveniles Adults Total

Probable Cause No. % No. % No. %

Predation 10 32.2 7 43.8 17 36.2

Starvation 2 6.5 0 0.0 2 4.3

Disease 0 0.0 1 6.3 1 2.1

Vehicle Collision 1 3.2 2 12.5 3 6.4
Gun Shot" 1 3.2 0 0.0 1 2.1

Unknown 4 12.9 3 18.8 7 14.9
(undisturbed)b
Unknown' 13 41.9 3 18.8 16 34.0

Total 31 100.0 16 100.0 47 100.0

SBird banded by Jon Buntz (Florida Game and Fresh Water Fish Commission).

b These birds were found with carcasses intact, indicating that predation was not a likely
cause of death. One juvenile was determined to be in excellent nutritional health,
indicating that emaciation was also not a likely cause. Another juvenile was determine
to be severely emaciated, which could have been either a symptom of illness or a cause
of death.

c These birds were too severely decomposed and the carcass had been potentially
disturbed, such that there was no evidence at all as to the potential cause of death.













CHAPTER 6
REPRODUCTION


Because the focus of this dissertation is on demography, I have concentrated my

attention in this chapter on estimates of reproductive parameters. There are numerous

papers on other aspects of reproduction (e.g., behavioral ecology) which I have given less

attention but encourage interested readers to seek out (e.g., reviewed by Beissinger 1988,

Sykes et al. 1995).

Reproduction of Snail Kites (Rostrhamus sociabilis) has been well studied;

although significant gaps remain in our knowledge. There also exist several areas of

disagreement among researchers regarding interpretation of existing data and literature.

Here, I present a combination of original data and a synthesis of the existing literature on

reproduction. I have attempted to explicitly point out any areas where disagreement

among researchers exists, and to provide detailed explanations for my interpretations.

From a demographic perspective, what is ultimately of interest is the mean

fecundity rate of females in each age class (Caughley 1977); that is, the mean number of

young (or sometimes just the mean number of female young female) produced per female

of each age class in the population. Unfortunately, for many species, including Snail Kites,

I cannot estimate this parameter directly. Rather, it is derived from the proportion of birds

attempting to breed (a), the proportion of breeding attempts that are successful (S,), and



60








61

the number of breeding attempts per year (f,). For successful nesting attempts, I also need

to know the number of young produced (Y) and the sex ratio of the young produced (RP)

(Brown 1974, Caughley 1977)(Figure 6-1). In this chapter I review the information that

has been previously reported on each of these parameters, as well as present estimates

based on new data.



Semantics

Misunderstandings about measures of reproduction can frequently be attributable

to a lack of clear definitions of what is being measured and/or to what is appropriate to be

measured. I will address the latter type of misunderstanding in my discussions below;

however, to avoid the former type, I will begin my assessment of reproduction by

providing operational definitions for terms discussed below.



Breeding Attempt

There has been considerable disagreement among researchers regarding what constitutes a

breeding attempt. For the purposes of this chapter, I consider a breeding attempt to begin

with the laying of the first egg Steenhof(1987). Snyder et al. (1989a) considered a

breeding attempt to begin with nest building, prior to the laying of the first egg. They

suggested that to ignore the period before egg laying in analyses of reproductive success

would be ill-advised because of the high proportion (>0.33) of nests they observed that

failed prior to egg laying. I agree entirely with Snyder et al. (1989a) that, for many

questions, the failure of nests prior to egg laying may have important biological








62

implications. These failures may provide insight as to environmental conditions at the time

of egg laying and also may provide information regarding behavioral aspects of mate

choice. However, I disagree that nests during the nest-building stage for this species

should be considered as a nesting attempt for estimation of reproductive parameters. I

have several reasons for this conclusion.

First, inclusion of "pre-laying" failures may include nests in which a pair bond has

not even been established between a male and female. Nest building is initiated by the

male as part of courtship (Beissinger 1988, Bennetts et al. 1988) and more than one male

may direct courtship toward a single female (Beissinger 1987, pers. obs.). Thus, if two

males initiated nest building as part of the courtship toward a single female, this would be

considered as two nesting attempts using the definition of Snyder et al. (1989a), even

though only one of these nests may produce young. Similarly, my observations indicate

that a single male may exhibit courtship behavior, including nest building, towards several

females in succession. This behavior may last as little as a few hours or may last several

days and may then be redirected to a new female if a pair bond is not established. I

observed a single radio-tagged male direct courtship to as many as five different females

before a pair bond was established that resulted in egg laying. Using the definition of

Snyder et al. (1989a), each of these courtship attempts would have been interpreted as a

failed breeding attempt. In contrast, I view this as part of the mate selection (courtship)

process rather than as a demographic parameter. Second, the passage of cold fronts and

corresponding temperature change often results in reduced food availability (Cary 1985).

Consequently, courtship is often terminated with the passage of cold fronts and resumed








63

(often at a new location) when temperatures return to pre-front conditions (Beissinger

1988, Bennetts et al. 1988). Thus, if two cold fronts passed before eggs were actually

laid, the pair would have been considered to have made three separate breeding attempts

(with two failures) even if the pair successfully raised a brood. For demographic

purposes, I view these postponements as courtship interruptions, rather than multiple

breeding attempts with each interruption being considered as a breeding failure. Third,

because nest building begins with the placement of the first stick and many more courtship

nests are probably initiated than are ever detected, it creates a substantial bias in the

estimate of success if these early starts are not detected (Mayfield 1961, Miller and

Johnson 1978, Johnson 1979, Hensler and Nichols 1981).

Finally, it is well known that nesting raptors tend to be considerably more sensitive

to disturbance early in the nesting cycle (Grier and Fyfe 1987, Steenhof 1987). Although

previous investigators have reported a high proportion of nest abandonment by Snail Kites

prior to egg laying (e.g., Beissinger 1986, Snyder et al. 1989a), I have seen no accounting

for how much of this abandonment might have been attributable to disturbance by the

investigators themselves. In contrast, abandonment of eggs or young by Snail Kites is

extremely rare (Bennetts et al. 1994, Sykes et al. 1995). Thus, measuring nesting success

after the first egg has been laid can reduce this potential source of confounding and

minimize disturbance to this endangered species.

Based on these concerns, I defined a breeding attempt to begin with the laying of

the first egg. Thus, unless otherwise stated, references to nests in this chapter implies the

presence of eggs or young.








64

Successful Nest

For the purposes of this chapter, a successful nest is one in which at least one

young reaches fledging age (Steenhof 1987). Because birds after fledging may or may not

be present at the nest, I defined fledging age as 80% of the average age of first flight

(Steenhofand Kochert 1982). Snail Kites are capable of first flight at approximately 30

days of age (Chandler and Anderson 1974, Beissinger 1988, Bennetts et al. 1988); thus, I

considered a nest as having been successful if it produced young that survived to at least

24 d (Bennetts et al. 1988). At this age, animals are reasonably assured of still being at

the nest and mortality for most raptors between this time and fledging is minimal (Milsap

1981, Steenhof 1987). In addition, I banded birds at the time they were determined to be

of fledging age. Consequently, any mortality that occurred after this age would have been

included in my estimates of juvenile survival from my mark-resighting program.



The Breeding Season

The initiation of nests (i.e., egg laying) has been documented in all months of the

year (Sykesl987c); although, for any given year, Snyder et al. (1989a) observed a

maximum breeding season (interval over which nests were initiated) of 31.7 weeks (7.9

months) during an 18-year study in Florida. Although Snail Kites in Florida can

potentially lay eggs in all months of the year, there is a very distinct seasonal distribution

of nest initiations (Table 6-1) (Figure 6-2). Nest initiations begin as early as November,

but in most years widespread initiations usually do not begin until January or February.








65

Peak initiations usually occur in March, but are often several weeks later, peaking in April,

in the northern habitats (Toland 1994).

The Breeding Population

Age of First Reproduction

Sykes (1979) reported that Snail Kites are capable of breeding at 3 years of age.

However, Sykes (1979) suggested that some birds possibly breed at a younger age.

Beissinger (1986) later reported both male and female birds breeding at one year of age

and Snyder et al. (1989a) reported one female breeding at nine months. My data are

consistent with Beissinger (1986) and Snyder et al. (1989a). During this study, I

commonly observed yearling Snail Kites attempting to breed.

Proportion of Birds Attempting to Breed

Nichols et al. (1980) suggested that the proportion of birds that attempted to breed

during favorable conditions was quite high. They suggested that there was no reason to

suspect that it was not 1.0 and, consequently, assumed that value for their demographic

model. They reported, however, that this was a crude estimate for lack of a better one.

Beissinger (1995) similarly reported that the proportion of adult Snail Kites attempting to

breed during high-water years was 1.0, but also provided no empirical evidence. Although

my data for this parameter are very limited, they are consistent with these earlier estimates.

During 1995, I closely monitored 23 radio-transmittered adult Snail Kites for breeding

activity in order to assess the proportion attempting to breed and the number of breeding

attempts per year. Of these 23 adults, 14 were females and 9 were males. During the 1995

nesting season, I located each bird on the ground approximately bi-weekly to determine its








66

breeding status (e.g., a nest, courtship, not breeding). Birds in which no breeding activity

was detected were generally observed for >2 hrs and subsequent visits, usually within 10

days, were required to confirm a non-breeding status and to confirm any nests for birds

exhibiting courtship. During 1995 (a relatively high water year throughout the kite's

range), all 23 (100%) adults attempted to breed at least once. My estimate is based on a

relatively small sample (N=23) and on only one year; however, it does provide an

empirical basis that most, if not all, adults may attempt to breed in some years.

Sykes (1979) reported that he observed no nesting attempts during 1971 (a

widespread severe drought). Based on this observation, Nichols et al. (1980) assumed

that no birds nested during 1971 for their demographic modeling effort. Beissinger (1986)

reported that during the 1981 drought 80-90% of the kites did not attempt to nest, and

Beissinger (1995) later reported that only 15% of adult Snail Kites attempt to breed

during drought years. However, no empirical evidence was presented in support of these

estimates. Based on anecdotal evidence, I believe that the proportion of birds attempting

to breed during drought years may be highly variable depending on the spatial extent of

the drought. I agree that during a severe widespread drought, most birds probably do not

attempt to breed. However, in cases of more localized droughts, where portions of the

kite's range may not be experiencing dry conditions, the proportion of birds attempting to

breed may remain very high. For example, during 1991, the Everglades region was at the

end of a 2-3 year drought (whether it was a 2 or 3 year drought depends on how a

drought is defined). During this year almost no nesting activity was observed in the

Everglades region (J.A. Rodgers Jr., pers. comm.). This would appear consistent that a








67

small proportion of birds had attempted to breed. However, during this year, record

numbers of birds were breeding on Lake Tohopekaliga (J.A. Rodgers and J. Buntz, pers.

comm.) and in the upper St. Johns Marshes (B. Toland, pers. comm.), areas not influenced

by the drought conditions in the Everglades. My data on movement strongly suggest that

the Florida population is one population that moves frequently throughout its range, rather

than a meta-population of quasi-isolated subpopulations. Thus, in years where drought is

not widespread, birds may merely shift the location of nesting activities. Consequently, I

suggest that this parameter may be quite variable and needs to take into account the

severity and spatial extent of a given drought.

Sykes (1979) observed relatively few (n=6) nests during 1972, the lag year

following the 1971 drought. The average number of nests per year that Sykes (1979)

reported from 1968-1976, excluding 1971, was 23. Based on this observation of reduced

nesting during this lag year, Nichols et al. (1980) assumed a proportion of 0.5 adults

attempted to nest during 1972. Beissinger (1995) reported that a proportion of 0.8 adults

attempt to breed during lag years, although I could find no empirical support for this

estimate in any of the sources cited. I suspect that, similarly to drought years, this

parameter may be highly variable depending on the specific drought. Thus, I view this

parameter as also being unknown and subject to high variability.

Snail Kites have been reported to breed as young as 9 months old (Snyder et al.

1989a); thus, by a calendar-year definition Snail Kites are capable of breeding as juveniles

(i.e., < 1 year old). However, these cases are ones in which the birds attempted to breed

during the nesting season following the nesting season of their hatch year. Thus, this








68

parameter should be defined as the proportion of birds attempting to breed during their

second breeding season (the first is the one in which they hatched).

Based on data from Snyder et al. (1989a), Beissinger (1995) reported that 25% of

subadults attempt to breed during high water conditions. Snyder et al. (1989a) observed 8

banded subadults breeding during 1979 out of a minimum of 74 that had survived from

their hatching year of 1978. Because Snyder et al. (1989a) only checked 50.8% of the

nests for bands, they estimated that there were probably 16 subadult breeders out of a

minimum of 74 banded subadults (22%). Of course, this estimate assumes that only the

74 subadults observed alive in 1979 had survived and that there was an equal probability

of detecting a banded subadult that was breeding in the sample of nests that were checked

and those that were not checked.

During 1992, I estimated a similar percentage of 17% of the subadult birds

attempting to breed (Bennetts and Kitchens 1992). My estimate was based on only 2

breeding birds of 12 banded yearlings that I observed during the 1992 breeding season.

Consequently, my estimate requires similar assumptions that I suggested above for Snyder

et al. (1989a). During 1995 (a high water year throughout the kite's range), I also closely

monitored 9 radio-transmittered juvenile Snail Kites for breeding activity (as described

above). Of these 9 birds 3 (33%) attempted to breed. All of the estimates derived from

the data of Snyder et al. (1989a), as well as from my own data, are very limited (i.e., small

samples each from one year); however, they do consistently suggest that a relatively small

proportion of subadults do attempt to breed during some years.








69

Beissinger (1995) also reported that the proportion of subadults attempting to

breed during drought years and lag years was 0.15. I could find no empirical basis for this

estimate in any of the sources cited; but I agree with Beissinger that the average

percentage would probably be lower when conditions are poor in part or all of their range.



Nest Success

Nest success has been among the most widely estimated parameters of

reproduction of Snail Kites. However, it has probably also been among the most

confusing. There are several areas of disagreement among researchers regarding

estimation of nest success. The disagreements center primarily on which nests should be

included in the sample and what estimator should be used. Consequently, nest success has

been difficult to compare because different researchers have used different estimators and

have included or excluded different categories of nests within their respective data sets. I

have attempted to summarize below the major issues of contention. I have also

summarized the literature on nest success and explicitly pointed out which estimator was

used and what categories of nests were included or excluded in the sample. Thus, readers

can make comparisons among studies and decide for themselves which estimates are most

appropriate for their particular needs.








70

Areas of Disagreement Regarding Estimation of Nest Success

Inclusion or exclusion of nests found at different stages

At what stage a given nest is found can greatly influence its probability of success.

Nests found late in the nesting cycle have a higher probability of success because they

have less observation time during which they are at risk. A Snail Kite nest requires at

least 57 days to fledge young (27 days of incubation and 30 days for nestlings to reach

fledging age). Thus, a nest found during egg laying will have potentially >50 days "at

risk" (provided it does not fail earlier) to be considered successful. In contrast, a nest

found close to the time of fledging may have only a few days "at risk" to be considered

successful. Consequently, estimates of nest success that were derived using nests found

late in the nesting cycle tend to be biased high (Mayfield 1961, 1975, Miller and Johnson

1978, Hensler and Nichols 1981, Hensler 1985).

Nests at different stages also are vulnerable to different risks. For example, rat

snakes (Elaphe obsoleta) are believed to be one of the major predators of Snail Kite nests

(Bennetts and Caton 1988). Rat snakes will readily take eggs or young that are less than

one week old; however, the larger size of older nestlings largely precludes predation by rat

snakes. Consequently, nests found when young are >1 week have an inherently lower risk

of predation by rat snakes.

Some researchers (e.g., Beissinger 1986, Snyder et al. 1989a) also have included

nests prior to eggs having been laid (i.e., during nest building) in deriving estimates of

success. I disagree with this practice for the reasons previously discussed.








71

Because of these biases, estimates of nest success can be substantially influenced

by what nests (i.e., found at what stage) are included or excluded for deriving a given

estimate. This makes comparison of previous estimates of nest success for Snail Kites

difficult because researchers have not used the same criteria for inclusion or exclusion of

nests found at different stages when deriving their estimates. Steenhof and Kochert

(1982) suggested three ways to minimize this type of sampling error for estimating nest

success. First, they suggest estimating success based on a pre-determined sample of

territorial pairs. However, because Snail Kites do not maintain nesting territories from

one year to the next, this solution is not feasible for this species. Secondly, they suggested

using estimates derived only from nests that were found during incubation (by definition,

they considered a breeding attempt to have begun after they laying of the first egg). This

suggestion is feasible for kites; but of the previously reported estimates, only Snyder et al.

(1989a) reported estimates using this criterion. Their third suggestion was to use the

Mayfield Estimator, which is intended to account for the bias imposed by not finding all

nests during early stages. Of the previously reported estimates, only Bennetts et al. (1988)

reported estimates using this estimator.

Given the differences in what nests were included or excluded in previous studies,

I urge caution in making comparisons among previous studies. I also agree with Steenhof

and Kochert (1982) that estimates of success should be derived either using only nests that

were found during incubation or using the Mayfield estimator. Of these two approaches I

prefer the latter, although there remains disagreement among researchers regarding this

conclusion.








72

Manipulated nests

Nests that occur in cattails may have a tendency to collapse under conditions of

high winds or waves (Sykes and Chandler 1974). This led to a previous practice of

placing nests that were subject to this type of failure in artificial nest baskets (Chandler

and Andeson 1974, Sykes and Chandler 1974). Because this may influence the outcome

of a given nest, whether to include or exclude these nests has been the subject of some

debate (e.g., Beissinger 1986, Snyder et al. 1989a). Similarly, when these nests have been

included in samples from which estimates of nest success were derived, there have been

differences among researchers (e.g., Sykes 1979, Snyder et al. 1989a) as to how these

nests were treated in the derivation of nest success.

Sykes (1979, 1987b) included 43 nests that were placed in artificial nest baskets in

his sample for estimating success. These nests were not treated differently than other

nests. Snyder et al. (1989a) later criticized this use of manipulated nests. They suggested

that the success of manipulated nests was higher than if they had not been manipulated,

and that this would have biased Sykes's estimate of success upward. Snyder et al. also

presented estimates of nest success using 94 manipulated nests. They argued that because

these nests were in imminent danger of collapse, they considered them all as failures.

They suggested that to exclude them, as was done by Beissinger (1986) and Beissinger

and Snyder (1987), would have also biased success upward because these manipulated

nests were not a random sample with regard to their probability of success (i.e., that they

would have failed). In contrast to their suggestion, I have observed collapsed several

nests containing older (>10 d old) nestlings that have been successful. Although I agree








73

with Snyder et al. (1989a) that exclusion of these nests probably would have biased

success upward, I also believe that including them all as failures probably would have

biased their estimate downward. My tendency is to agree with the solution of Snyder et

al. (1989a), but to recognize that there might be a slight bias toward underestimation of

success.

An additional concern that has not been addressed by previous authors is that the

susceptibility of nests to collapse may be influenced by the investigators themselves. The

vulnerability of nests to collapse can be greatly influenced by the paths of airboats while

conducting nests visits, particularly in cattails (Bennetts 1996). Airboat trails are often

wide enough to allow increased susceptibility to wind damage and/or to weaken the

structural support provided by the cattails adjacent to the nest. This type of damage can

be minimized, if not eliminated, by maintaining a substantial distance from the nest during

an approach and either wading in to nests or using a mirror pole from a distance to check

them (Bennetts 1996). Nest baskets have not been used in recent years and I do not

anticipate (or advocate) a recurrence of their use. Although some nest collapse still

occurs in some areas, particularly on lakes (J.A. Rodgers, pers. comm.), I do not believe

that the benefits of nest baskets warrant the effort or disturbance for their use as a general

management tool. They may, however, be warranted for isolated special circumstances.

Previous use of nest baskets had been initiated when numbers of Snail Kite probably were

much lower than are currently found. I do, however, advocate that researchers exercise

extreme care to avoid influencing the outcome of nests being monitored.








74

Mavfield vs conventional estimator

Mayfield (1961, 1975) proposed an estimator for nest success that was based on

daily exposure (risk) such that a daily probability of success was derived using only those

days in which a given nest was under observation. Overall success is then derived by

applying the daily success over the length of the interval being estimated. This approach

provides an estimate of success that is unbiased with respect to when a given nest was

found, but requires an assumption that the probability of success is constant over the

period (e.g., incubation) being estimated. Hensler and Nichols (1981) later showed, using

Monte Carlo simulations, that this estimator was superior to the conventional estimator

under a wide variety of conditions.

Bennetts et al. (1988) used the Mayfield estimator for nest success of Snail Kites

and found it to perform favorably for this species. They found some violation of the

assumption of constancy (e.g., success differed between incubation and nestling stages);

however, this assumption can be overcome by using separate estimates for periods that

differ (Hensler and Nichols 1981). Snyder et al. (1989a) later argued that the Mayfield

estimator was inappropriate for Snail Kites because the interval length for nest building

was too variable to apply this estimator. I agree with Snyder et al. (1989a) that the

Mayfield estimator would be inappropriate for estimation of success during the nest

building stage. However, I also have argued that the nest-building period is inappropriate

to include in estimates of nesting success for this species. Consequently, I disagree with

Snyder et al. (1989a) that the Mayfield estimator is inappropriate for estimating nesting

success of Snail Kites. Rather, I agree with Hensler and Nichols (1981), Miller and








75

Johnson (1978), Steenhof and Kochert (1982), and Steenhof(1987), that this estimator is

preferable to conventional estimates of nesting success because of its ability to produce

unbiased estimates of nesting success.



Estimates of Nest Success And Its Process Variance

Given the wide disagreement among researchers regarding nesting success, I

suggest that future researchers be specific about what is being included or excluded, and

that consideration be given to reporting success both by conventional and Mayfield

estimators so readers have the ability to compare their results. I have also provided a

summary of previously reported estimates, showing what nests (i.e., found at what stage)

were included in each estimate, whether or not manipulated nests were included, and

which estimator was used (Table 6-2).

I estimated the mean annual nest success (S,) as 0.32 based on reported nest


success from each year using estimates that were based on nests, in which at least one egg

has been laid, that were found during the egg stage (Table 6-3). However, some years had

extremely low sample sizes, which may have precluded a reliable estimate for that year. If

I had excluded estimates for those years with <10 nests, I would have estimated mean

A
annual nest success (S,) as 0.28.


It is also important to recognize that there are several distinct variance components

associated with demographic parameters (White et al. 1982, Burnham et al. 1987). A

demographic parameter (e.g., survival) may vary over time (temporal variation) or among








76

locations (spatial variation). There is also likely to be heterogeneity among individual

(individual variation) in their probability of survival due to genetic or phenotypic variation

(DeAngelis and Gross 1982). Each of these sources of variation are a type of population

variation (Burnham et al. 1987). There is also variation attributable to sampling

populations. Unlike these previous sources of variation, sampling variation is not a

measure of population variability, but rather is a measure of sampling error. This latter

source of variation is important because it provides a measure of the certainty for a given

parameter estimate. However, for demographic modeling, what is important is the actual

variability of parameter over time, space, and among individuals (collectively called

process variance). For modeling populations, sampling variation is a source of noise and

should be removed from the overall variance estimate. Burnham et al. (1987) provided

the theoretical framework and formulae for estimating process variance. I used this

framework to estimate process variance for nest success based on estimates reported from

1968-1995 using only nests found after the first egg was laid, but before hatching. Based

on the data from table 6-3, I estimated 62=0.08 and 6=0.28.



Influences of Nest Success

There are a multitude of factors that could potentially influence the outcome of

Snail Kite nests. Factors that have been reported to significantly affect nest success include

location (i.e., area)(Snyder et al. 1989a), water levels (Sykes 1987b, Bennetts et al. 1988,

Snyder et al. 1989a, Toland 1994), date of initiation (Bennetts et al. 1988), nest substrate

(Snyder et al. 1989a, Toland 1994), nest height (Bennetts et al. 1988, Toland 1994),








77

distance to land (Sykes 1987c), and interspecific coloniality (Snyder et al. 1989a). I used

logistic regression to test for the influence of each of these effects, except interspecific

coloniality, on a sample of 854 nests using data from Bennetts et al. (1988), Toland (1994,

unpubl. data), and this study. My preliminary univariate analysis, which had a liberal

rejection criterion of a=0.25 for each effect, indicated that all of these effects warranted

retention for further analysis. However, my results indicated that the specific substrate,

rather than herbaceous versus woody, was warranted for further consideration. Similarly,

my results indicated that a categorical threshold distance to land of less than or greater

than 200m (Sykes 1987c) was warranted for further consideration, rather than the actual

distance. Although my preliminary univariate analysis supported the retention of these

effects, a multivariate analysis with each of the retained effects (but lacking interaction

terms) indicated that only year and date of initiation were warranted at more restrictive

rejection criteria of a =0.05.

My final model indicated an area, but not a year effect, as was indicated by my

preliminary analyses (Table 6-4). However, area and year effects were highly confounded

in these data because the studies included in this analyses that were conducted during

different years were also conducted at different areas. Thus, I do not believe that I can

reliably distinguish between these effects. Differences in success among areas and years

are not surprising given the many causes of nest failures (Sykes 1987c, Bennetts et al.

1988, Snyder et al. 1989a). My data indicated an effect from the date of initiation in all

phases of this analysis; although it was not completely clear as to whether this effect was

quadratic or linear. Overall nest success (all years combined) was highest during January








78

with a generally decreasing trend over time (Figure 6-3). However, the overall trend is

somewhat misleading because it was heavily influenced by one year (1987) of

exceptionally high success in January (Figure 6-4). Most years had the peak of success in

February (3 of 7) or March (2 of 7). In only one year was peak success in January (1987)

and one year in April (1993). In only one year did I observe nesting during December

(1985), and success was lower than during January, February or March of that year.

These temporal effects of success were undoubtedly confounded with year effects

because studies conducted from 1991-1993 by Toland (1994, unpubl. data), which were

included in this analysis, were conducted in the northern part of the kite's range where the

date of initiation was often several weeks later than in the southern portion of their range.

In contrast, most of the data from other years were from the southern portion of the kite's

range. This probably also accounts for the interaction effect of year with date of initiation.



Number of Young per Successful Nest

In contrast to nest success, the number of young per successful nest probably is

one of the least variable and has been the least controversial of the reproductive

parameters. The relative lack of variability for this parameter is not surprising since it is

not unusual for raptors to produce normal numbers of young per successful nest even

when other aspects of reproduction (e.g., proportion of population attempting to breed or

nest success) are depressed (Brown 1974, Steenhof 1987). For this reason, the number of

young per successful nest is not particularly informative in the absence of these other

reproductive parameters (Brown 1974).








79

Several studies have reported estimates for the number of young per successful

nest (Table 6-5) and the average from 20 years of reported data is 1.9. Annual estimates

reported have ranged from a low of 1.4 (Sykes 1979, 1987b, Bennetts et al. 1988) to a

high of 2.5 (Sykes 1979, 1987b)(Table 6-6).



The Number of Breeding Attempts per Year

For Snail Kites, the success per breeding attempt and the number of young

produced per attempt are relatively well known (Sykes 1979, Bennetts et al. 1988, Snyder

et al. 1989). In contrast, there has been little evidence for the number of breeding

attempts per year. Snail Kites are capable of raising >1 brood per year and attempts at

multiple brooding may be fairly widespread (Snyder et al. 1989). Snyder et al. (1989)

suggested that individuals have the potential to successfully raise four broods per year,

although there have been no documented cases of individuals successfully raising >2

broods in a given year. Snyder et al. (1989) estimated the number of nesting attempts per

pair to be 2.7 per pair. Their estimate was derived using the number of Snail Kites

counted on an annual survey at two locations (Lake Okeechobee and Water Conservation

Area 3A) during late fall of 1977 as an estimate of the potential breeding population of

1978. They then used the number and success of nests found at those locations the

following breeding season to estimate the number of breeding attempts by that breeding

population. However, there are several assumptions inherent in their calculation which

may have greatly influenced their estimate. Beissinger (1995) later used a more

"conservative" estimate of 2.2 attempts per pair in a population viability analysis because








80

the estimate by Snyder et al. (1989) was reported to be "under the best conditions". Here,

I estimate the proportion of the population attempting to breed and the number of

breeding attempts per year for individual Snail Kites during the 1995 breeding season

using radio-telemetry.

During 1995, I closely monitored the breeding activity of these kites in order to

assess the number of breeding attempts per year. I defined a breeding attempt to begin

with the laying of the first egg (Steenhof 1987). However, I recorded all activity

associated with breeding, including courtship behaviors to enable more comprehensive

record of each individual. During the nesting season, I located each bird approximately

biweekly by airboat and determined its breeding status (e.g., not breeding, courtship, or

breeding). Birds in which no breeding activity was detected were generally observed for >

2 hrs and subsequent visits, usually within 10 days, were required to confirm a non-

breeding status and to confirm any nests for birds exhibiting courtship behavior.

I was able to successfully monitor the breeding status of 23 adult Snail Kites for

the entire 1995 season. Of these, 14 were females and 9 were males. I was able to

monitor the breeding status of an additional 9 subadults. The average interval between

successive observations of breeding status was 14.1 d ( 8.1 SD).

All adult birds monitored attempted to breed at least once, and I observed an

average of 1.4 ( 0.6 sd) breeding attempts per bird. Of the 23 adult birds, 15 (65%)

made only one breeding attempt, 7 (30%) made two breeding attempts, and 1 (4%)

attempted three times (Table 6-7). I observed only one bird (4%) which successfully

raised two broods. In contrast to adults, not all subadults attempted to breed. Of the nine








81

birds monitored, only 3 (33%) were confirmed to have a nest in which at least one egg

was laid, and none were observed attempting to breed more than once.

Our data are consistent with reports by Snyder et al. (1989) that >1 breeding

attempt by Snail Kites in Florida is common during some years. However, my data are

not consistent with previous estimates of 2.7 attempts per year by Snyder et al. (1989) and

even the more "conservative" estimate of 2.2 attempts per year used by Beissinger (1995),

which was based on Snyder et al.'s estimate. A combination of differences in my

estimation procedures, difference in our respective definitions of a breeding attempt, and

annual variability of this parameter probably account for these discrepancies between these

two data sources.

Semantic differences in defining a breeding attempt likely contributed to the

disparity between the estimate of Snyder et al. (1989) and the lower estimates of this

study. Snyder et al. (1989) considered a breeding attempt to begin with nest building,

prior to the laying of the first egg. Thus, their estimate for the number of attempts

included courtship by my my definition. In contrast, I agree with Steenhof (1987) and

defined a breeding attempt to begin with the laying of the first egg. If this definition

were applied to the data reported by Snyder et al. (1989) their estimate would have been

reduced from 2.7 to 1.9 breeding attempts per pair.

The assumptions required by Snyder et al.'s (1989) and this study's estimates can

also have a dramatic influence on the resulting estimates. The primary assumption of my

estimates was that no breeding attempts went undetected during the breeding season. The

interval of my breeding status checks could have resulted in failure to detect an occasional








82

bird that initiated a nest that failed early during laying or incubation. Consequently, my

estimate may have been slightly low for 1995. However, exclusion of birds from my

sample in which I had gaps in the known breeding status helped to minimize this potential

bias. In addition, most breeding failures occur during the first week after hatching

(Bennetts et al. 1988), which occurs after 4-5 weeks of breeding activity, including

courtship. Consequently, the potential bias from having missed breeding attempts

probably was very low. To further assure that this bias was minimal, I also tried a more

restrictive criterion for my sample, such that the average interval between visits was -8 d,

with a maximum of21 d between any 2 visits. This more restrictive criterion reduced my

sample size (n=10), but did not alter my estimate of the number of breeding attempts per

adult (5=1.38). In addition, 1995 had very favorable water conditions throughout the

Snail Kite's range in Florida. For this reason, I might also have expected my 1995

estimate to be higher than an annual average.

As pointed out by Snyder et al. (1989), their procedure assumed that the 1977

annual survey was an accurate census (i.e., a complete count of all kites). Recent

evidence suggests that this assumption was highly unlikely to have been met. Bennetts

and Kitchens (1997a) and Darby et al. (in review) found that during late fall, when the

annual survey was conducted, a substantial portion (up to 60%) of the population may be

in areas not included in the survey or in habitats (e.g., cypress) where detection is difficult.

In addition, I found that the average probability of detecting marked individuals during

spring, when birds are more concentrated, was quite low (<25%). In contrast to the

procedure used by Snyder et al. (1989), my approach required no assumptions about the








83

annual survey because I used radio transmitters to locate individual birds and I eliminated

all birds from my sample whose breeding status was unknown for prolonged periods.

Snyder et al. (1989) reported that they also assumed that no birds died between the

1977 annual survey and the end of the 1978 breeding season. However, because they

used the 1977 survey as an estimate of the number of potentially-breeding pairs during the

1978 breeding season, their approach actually required a more rigorous assumption that

WCA-3A and Lake Okeechobee represented a closed population. Thus, the assumption is

not only that there were no deaths, but also that there were no births, immigration, or

emigration. The assumption regarding births does not impose a serious problem because

the survey was conducted before the primary breeding season. The assumption of no

deaths poses a slightly greater problem since annual mortality of adults is approximately

10%, most of which occurs during the time period between the survey and the following

breeding season (Chapter 4). However, the most serious problem probably is the

assumption of no immigration or emigration. Data from 271 radio-tagged Snail Kites in

Florida indicated that the probability of a bird moving from one wetland to another during

a given month is approximately 0.25 (Bennetts and Kitchens 1997a, Chapter 7). Given

that the time between the 1977 survey and the end of the 1978 breeding season was

approximately 8-9 months, there is a strong likelihood of substantial immigration and

emigration. Further, Bennetts and Kitchens (1997a) and Darby et al. (in review) found

that there was a substantial shift from peripheral habits, during the time of the survey, to

breeding habitats during spring. Snyder and Snyder (1991) also reported that birds left

Lake Okeechobee after the breeding season and returned during fall. Thus, there was very








84

likely a net increase in the breeding population which could have substantially inflated their

estimate of the number of breeding attempts per pair. In contrast to the approach used by

Snyder et al. (1989) my approach did not require assumptions about closure because my

estimates were based on individual radio-tagged birds, regardless of their location.

Snyder et al. (1989) also recognized that their approach required an assumption

that all birds counted during the 1977 survey were potential breeders during the 1978

breeding season. This assumption also was unlikely to have been met because the annual

survey does not distinguish between adult and subadult birds. Previous reports (e.g.,

Snyder et al. 1989), and my data confirm, that not all subadults are potential breeders. In

contrast, my approach did not require this assumption since my estimates were specific to

each age class.

An additional assumption, required if an estimate is applied to years other than one

from which it was derived, is that the estimate be from a "representative" year of the

conditions to which the estimate is being applied. Estimates derived from both my data

and that of Snyder et al. (1989) were based on a single year. Based either on the number

of nests found or the number of young banded, 1978 was an exceptional year for

reproduction and Snyder et al. (1989) correctly limited their inference to that year. To

extend the inferences of Snyder et al. (1989) to other years would likely result in an

inflated estimate. To account for this bias, Beissinger (1995) used what he considered a

"conservative" estimate of 2.2 breeding attempts per pair in a population viability analysis.

my data suggest that even this "conservative" estimate was likely to be inflated if used as








85

an annual average, although my estimate also was based on a single year of relatively

favorable conditions.



Conditional Probability of Attempting to Breed

An alternative way to look at the number of attempts per year is using conditional

probability for the proportion of birds attempting to breed (a.). That is, ai would be the

probability that a bird attempted to breed, given that it had not attempted previously

during that nesting season. Of 23 birds I monitored for breeding activity during 1995, all

23 attempted to breed at least once. Thus, my estimate of a, would be 1.0. Similarly, a2

would be the probability that a bird attempted to breed, given that it had previously made

1 attempt during that breeding season. Based on my data from 1995, I would estimate a2

to be 0.34 (8 of 23). The probability that a bird attempted to breed, given that it had

previously made two attempts during that breeding season (a3) was 0.13 (1 of 8). The

variance for these estimates could be derived based on a binomial distribution, although

the formula traditionally used for this estimate is intended for large samples (White and

Garrott 1990). Hollander and Wolfe (1973) provide alternative procedures that could be

used for smaller samples.



Number of Successful Broods per Year

Snyder et al. (1989a) suggested that in some years it was possible for Snail Kites

to successfully raise four broods. This was based on the length of the breeding season for

certain years (e.g., 1978 and 1979) and the assumption that it would take 10 weeks (70 d)




Full Text
xml version 1.0 encoding UTF-8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID ETTSJYUAX_GA5P2M INGEST_TIME 2015-01-09T20:16:39Z PACKAGE AA00026591_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES



PAGE 1

THE DEMOGRAPHY AND MOVEMENTS OF SNAIL KITES IN FLORIDA By ROBERT E. BENNETTS A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 1998

PAGE 2

ACKNOWLEDGMENTS I am deeply indebted to my major advisor Wiley M. Kitchens, who stuck it out and provided support through all of the ups and downs. I am also grateful to my committee members, Drs. Crawford S. Holling, James D. Nichols, Kenneth M. Portier, Frank J. Mazzotti. Financial support was provided by the U.S. Fish and Wildlife Service (USFWS), National Park Service (NPS), U.S. Army Corps of Engineers (USACOE), South Florida Water Management District (SFWMD), St. Johns River Water Management District (SJRWMD), and the Biological Resources Division (BRD) of the U.S. Geological Service. John Ogden (SFWMD) and David Wesley (USFWS) were largely responsible for getting this project started and continued to provide strong support throughout its duration. I am also grateful to Reid Goforth (BRD), Steve Miller (SJRWMD), Ed Lowe (SJRWMD), Mary Ann Lee (SJRWMD), Jon Moulding (USACOE), Lewis Hornung (USACOE), Peter David (SFWMD), Dale Gawlik (SFWMD), Paul Warner (SFWMD), and Jim Brown (USFWS), and Donald DeAngelis (BRD). I greatly appreciate the help of our field biologists Phil Darby, Patty ValentineDarby, Katie Golden, Steve McGehee, Scott Severs, Hilary Maier, David Boyd, James Conner, and Lynn Bjork. Their ability to work independently for long hours, and get the job done made our job much easier. This project has been a cooperative effort among biologists and agencies from the

PAGE 3

outset. For their help in the field and/or logistic support I am grateful to Brian Toland (USFWS), Tim Towles (GFC), Peter Frederick (UF), Marilyn Spalding (UF), Mary Beth Mihalik (West Palm Beach Solid Waste Authority), Al Vasquez (West Palm Beach Solid Waste Authority), Deborah Jansen (NPS), Mike Wilson (NPS), Sue McDonald (NPS), Vivie Thue (NPS), Fred Broerman (Arthur R. Marshall Loxahatchee National Wildlife Refuge), Angela Chong (SFWMD), Vicky Dreitz (Univ. Miami), F.K. Jones (Miccosukee Tribe of Indians), and Steve Terry (Miccosukee Tribe of Indians). The banding of Snail Kites was conducted in cooperation with the GFC. In this effort, I appreciate the cooperation of James Rodgers Jr. (GFC), Jon Buntz (GFC), and Brian Toland (USFWS). I greatly appreciate the effort of Charlie Shaiffer (Mingo National Wildlife Refuge) who took the time to travel to Florida to share his knowledge of trapping and handling birds. I am grateful to Patuxent Wildlife Research Center, particularly Jim Nichols and Jim Hines, for housing and assistance during the analysis phase of this project. I also appreciate the help of Laura Brandt, Cynthia Loftin, Kenny Rice, and Cynthia Sain for keeping me registered for credits when I had procrastinated to the last hours. Laura Brandt also served as an unofficial academic advisor for which I am grateful. For allowing me access to areas used by kites, which were closed to public access, I am grateful to the Miccosukee Tribe of Indians, the City of West Palm Beach, and the Arthur R. Marshall Loxahatchee National Wildlife Refuge. I am grateful to our pilots Karen Dunne and Morton Sund for many hours of safe flying. Finally, I am grateful to the employees of the Florida Cooperative Fish and Wildlife Research Unit, particularly Barbara Fesler and Debra Hughes, for their help with administration of this study. iii

PAGE 4

TABLE OF CONTENTS page ACKNOWLEDGMENTS ii LIST OF TABLES vii LIST OF FIGURES x ABSTRACT xiv CHAPTERS 1. INTRODUCTION 1 Overview 1 Objectives 3 2. STUDY AREA 7 Spatial Scales 7 Regions 8 Habitat Types 9 3 ANNUAL SURVIVAL OF SNAIL KITES IN FLORIDA WITH COMPARISONS BETWEEN RADIO TELEMETRY AND CAPTURE-RESIGHTING DATA 14 Introduction 14 Methods 16 Estimation of Survival from Radio Telemetry 16 Estimation of Survival from Banding Data 17 Influences on Survival 19 Hypothesis Testing and Model Selection 22 Results 23 Age Effects 23 Time Effects 24 Regional Effects 25 Parameter Estimates 26 Censoring of Radio-tagged Birds 27 iv

PAGE 5

Discussion 28 Comparisons Between Data Obtained Using Radio Telemetry and Capture-resighting 28 Parameter Estimates 30 Effects of Age, Time, and Region on Survival 32 Implications of Resighting Probabilities 33 4. WITHINYEAR SURVIVAL PATTERNS OF SNAIL KITES IN FLORIDA 43 Methods 44 Results 46 Discussion 48 5. CAUSES OF MORTALITY OF POST-FLEDGING JUVENILE AND ADULT SNAIL KITES IN FLORIDA 54 6. REPRODUCTION 60 Semantics 61 Breeding Attempts 61 Successful Nests 64 The Breeding Season 64 The Breeding Population 65 Age of First Reproduction 65 Proportion of Birds Attempting to Breed 65 Nest Success 69 Areas of Disagreement Regarding Estimation of Nest Success 70 Estimates of Nest Success and its Process Variance 75 Influences of Nest Success 76 Number of Young per Successful Nest 78 Number of Breeding Attempts per Year 79 Conditional Probability of Attempting to Breed 85 Number of Successful Broods per Year 85 7. DISPERSAL PROBABILITIES OF SNAIL KITES IN FLORIDA 100 Introduction 100 Methods 101 Terminology 101 Field Methods 102 Estimation of Natal Dispersal 103 Estimation of Dispersal Probabilities 103 Food Availability 105 Water Levels 106 Results 108 v

PAGE 6

Natal Dispersal 108 Dispersal Probabilities 109 Pooling of Locations Ill Hydrologic Effects on the Probability of Dispersal 115 Food Resources 116 Discussion 116 8. IMPLICATIONS TO MANAGEMENT AND CONSERVATION 139 Drought Semantics 139 Intensity 141 Spatial Extent 143 Temporal Extent 144 Critical Habitat 145 Currently Designated Critical Habitat 145 The Habitat Network 146 Management of the Snail Kite in Florida: Beyond a Reductionist Paradigm 150 The Reductionist Paradigm 151 Conflicts and Limitations of the Existing Paradigm 154 The Importance of Spatial and Temporal Scales 155 The Dynamic Landscape Hypothesis 156 Persistence of Snail Kites in a Dynamic Landscape 161 Conclusions 163 APPENDICES 1 ESTIMATES OF CUMULATIVE NATAL DISPERSAL (v|f), NUMBER OF ANIMALS AT "RISK" OF DISPERSAL DURING INTERVAL j (r } ), AND STANDARD ERROR (SE) OF THE ESTIMATE FOR EACH STUDY YEAR 182 2 ESTIMATES OF CUMULATIVE NATAL DISPERSAL (i|r), NUMBER OF ANIMALS AT "RISK" OF DISPERSAL DURING INTERVAL j (rj), AND STANDARD ERROR (SE) OF THE ESTIMATE FOR EACH STUDY YEAR IN AREAS AFFECTED AND UNAFFECTED BY THE PREVIOUS DROUGHT 184 LITERATURE CITED 187 BIOGRAPHICAL SKETCH 197 vi

PAGE 7

LIST OF TABLES Table page 3-1 Capture-resighting summary of adult and juvenile Snail Kites in Florida from 1992-1997 34 3-2 Description of single-stratum Cormack-Jolly-Seber (CJS) models and their corresponding Akaike Information Criteria (AIC) scores. Parameter structure indicates whether survival () differed between adults and juveniles. Under this model, survival was constant among years for adults, but differed among years for juveniles. Resighting probabilities (p) differed among years 39 3-7 Parameter estimates for my most parsimonious multi-stratum model (MS 13), in which survival differs between adults and juveniles, survival is constant among years and regions for adults, and survival differs among years and regions for juveniles. Resighting probability in this model differs among years and regions 40 51 Probable causes of mortality of Snail Kites recovered in Florida from 1 9921995 59 61 The number of nest initiations reported in each month during studies from 1966 through 1995 88 vii

PAGE 8

6-2 Mean annual nest success from major studies conducted since 1968. Also shown are the estimator used to estimate success, whether or not nests placed in nest baskets were included in the estimate, and whether or not nests found in each of 3 stages were included in the estimate 89 6-3 Annual nest success reported during studies from 1968 through 1995. Nest success was based on nests found during the egg stage 91 6-4 Summary statistics from the final most parsimonious (based on AIC and LRTs) logistic regression model for the factors effecting nest success 92 6-5 The annual mean and overall (i.e., all years and locations combined) number young per successful nest from major studies conducted since 1968. A successful nest was considered a nest in which at least one young fledged 93 6-6 The number of successful nests, young fledged, and number of young per successful nest reported for each year from 1968 through 1995 94 67 Number of nesting attempts and number of attempts that were successful for each of 23 adult Snail Kites during the 1995 breeding season 95 71 Summary statistics for conditional logistic regression models for potential seasonal groupings affecting the probability of movement between times t and t + 1 (at monthly time steps), given that an animal was alive at time t and its location known. Shown are the model description, number of estimable parameters (np), relative deviance (-21n[9?]), and Akaike's Information Criteria (AIC). The model shown in bold would be the one selected from these potential models based on AIC 121 7-2 Summary statistics for conditional logistic regression models for potential temporal effects on the probability of movement between times t and t + 1 (at monthly time steps), given that an animal was alive at time t and its location known. Shown are the model description, number of estimable parameters (np), relative deviance (-21n[S£]) ; and Akaike's Information Criteria (AIC). The model shown in bold would be the one selected from these potential models based on AIC criteria 122 7-3 Summary statistics for conditional logistic regression models of the probability of movement between times t and t + 1. The model with the lowest AIC (bold) would be selected if based solely on this criterion 123 7-4 Summary statistics for conditional logistic regression models of the probability of movement between times t and t + 1 to evaluate the pooling of some parameters. A failure to reject a LRT indicates that the additional parameters of the more general (unconstrained) model may not be supported by these data. The model with the lowest AIC (bold) would be selected if based solely on AIC 174 viii

PAGE 9

7-5 Maximum likelihood analysis of variance table for univariate models (i.e., each source term represents a separate model) of potential sources of variation of the probability of dispersal between times t and t + 1 125 7-6 Summary statistics used for model selection of conditional logistic regression models of the probability of movement between times t and t + 1 The model with the lowest AIC (bold) would be selected if based solely on AIC criterion. 126 7-7 Likelihood ratio tests (LRTs) comparing conditional logistic regression models of the probability of movement between times t and t + 1 The null hypothesis (Hq) from a LRT is that the reduced model (i.e., the model with fewer parameters) fits the data equally well as the more general model (i.e., with more parameters). Thus, a rejection of H,, favors the more general model and a failure to reject H 0 favors the more reduced model 127 78 Analysis of variance table from model of foraging time per capture as the dependent variable. Mean square (MS) and F values are based on type III partial sums of squares (i.e., they are adjusted for all other terms in the model and are not dependent on the order of entry)(SAS Inc. 1988) 129 81 Drought intensity scores (standard deviations from mean annual minimum) for most major wetlands used by Snail Kites in Florida. Scores > 1 sd below mean are considered as drought years (bordered cells) and scores >2 sd below mean are considered as extreme drought years (cells bordered with double line). Spatial extent of a drought can be evaluated by how many areas in a given year have scores >1 sd below the mean 166 8-2 The number of days that water stage was > 1 standard deviation below the average minimum stage for a 10-gauge average from Lake Okeechobee for each year from 1969-1994. This corresponds to a stage of < 11.17ft MSL. The intensity for a given drought year is shown as the number of standard deviations below the mean 168 ix

PAGE 10

LIST OF FIGURES Figure page 1-1 The demographic cycle of Snail Kites showing three age classes (Juveniles=J, Subadults=S, and Adults=A). Parameters for survival () and fecundity (f) are shown for each age class. Adapted from Caswell (1989), Beissinger (1995) and Legendre and Clobert (1995) 5 1-2 Conceptual framework for this study. Reliable estimation of parameters is the first step toward the development of a wide variety of demographic models. ... 6 2-1 Major wetlands of South Florida referred to in this report. Wetlands are Everglades National Park (ENP), Big Cypress National Preserve (BICY), Water Conservation Areas 3 A, 3B, 2B, 2A, Loxahatchee National Wildlife Refuge (LOX), Holey Land Wildlife Management Area (HOLEY), West Palm Beach Water Catchment Area (WPB), Lake Okeechobee (OKEE), Upper St. Johns [Blue Cypress] Marsh (SJM), Lake Kissimmee (KISS), Lake Tohopekaliga (TOHO), and East Lake Tohopekaliga (ETOHO) 12 22 South Florida showing geographic regions used for some analyses in this report. All areas not within a region shown were assigned to a peripheral region 13 31 Percentage of radio-transmittered adult and juvenile Snail Kites that were censored in each 60-day time interval from the time of attachment 41 32 Percentage of adult and juvenile Snail Kites from each sampling cohort (i.e., the year that they fledged or were captured) that died or were censored during the first 180 d after radio attachment each year 42 41 Survivorship functions of adult and juvenile Snail Kites from a pooled sample of 3 years (top). Because I was interested in temporal pattern rather than magnitude, I aligned the functions without regard to magnitude, to illustrate at what point in time the functions become similar (bottom) 51 4-2 Estimated hazard functions constructed at monthly intervals for adult and juvenile Snail Kites 52 4-3 Estimated hazard function for juveniles based on age, rather than time. This function was constructed at monthly intervals starting at the time of fledging. 53 x

PAGE 11

6-1 Conceptual diagram of reproductive parameters used to estimate fecundity. Show here for simplicity is model for 2 nesting attempts; although more attempts are possible within a given year 96 6-2 The proportion of nest initiations for each month of the year based on cumulative data reported by Sykes (1987c), Snyder et al. (1989a)( 19701982 only), Toland ( 1 994), and this study 97 6-3 The percentage of nests that were successful during each month. Data used in this analysis were from Bennetts et al. (1988)( 19861987), Toland (1994, unpubl. data) (1990-1993), and this study (1994-1995) 98 64 The percentage of nests that were successful during each month of each year. Data used in this analysis were from Bennetts et al. (1988)( 19861987), Toland (1994, unpubl. data)(1991-1993), and this study (1994-1995) 99 71 Kaplan Meier estimates for the overall cumulative probability of dispersal (solid line). Also shown is a 95% confidence interval (dotted line) for the probability function 130 7-2 Kaplan Meier estimates for the cumulative probability of dispersal in each of the three years. Confidence intervals for estimates are not shown to minimize cluttering, but are provided in detail in Appendix 1 131 7-3 Kaplan Meier estimates for the cumulative probability of dispersal from wetlands that were and were not affected by the preceding drought in each of the three study years. Confidence intervals for estimates are not shown to minimize cluttering, but are provided in detail in Appendix 2 132 7-4 Conditional probabilities that adult and juvenile Snail Kites that were alive and their location known at time t, were in the same location (or conversely at a different location) at time t + 1 during each season of each study year. Also shown are the standard errors (rectangles) and 95% confidence intervals (vertical lines) 133 7-5 Adjusted residuals from a cross tabulation of dispersal and location at time t. Residuals >0 indicate that birds in this area moved more frequently than expected and residuals <0 indicates that birds in that area moved less frequently than expected 134 7-6 Standardized residuals for probability of dispersal during each study year for adult (top) and juvenile (bottom) snail kites 135 7-7 Standardized residuals for probability of dispersal during each season for adult (top) and juvenile (bottom) snail kites 136 7-8 The mean ( SE) foraging time to capture snails. Sample sizes (number of complete bouts observed) are also shown 137 xi

PAGE 12

7-9 The mean ( SE) foraging time to capture snails by radio-transmittered birds before (location 1) and after (location 2) moving. Sample sizes (number of individuals birds observed) are also shown 138 8-1 The minimum annual water stage for gauge 3-28 in Water Conservation Area 3 A (WCA-3A) for the period of 1968-1988. Shown for reference are the minimum and maximum ground elevation in WCA-3 A, and ground elevation at the 3-28 gauge. Points mark with an "S" were years identified by Snyder et al. (1989) as drought years and those mark with a "B" were identified by Beissinger (1995) as drought years 169 8-2 The minimum annual water stage for a 10-gauge average at Lake Okeechobee for the period of 1968-1988. Shown for reference are the minimum and maximum ground elevation for the littoral zone at Lake Okeechobee (based on Pesnell and Brown [1977]). Points marked with an "S" were years identified by Snyder et al. (1989a) as drought years and those mark with a "B" were identified by Beissinger (1995) as drought years 170 8-3 The minimum annual water stage for gauge 3-28 in Water Conservation Area 3A (WCA-3 A) for the period of 1969-1994. Shown for reference are the average annual minimum stage, 1 standard deviation, and 2 standard deviations 171 8-4 The currently designated critical habitat identified in the Snail Kite Recovery Plan (after 50 CFR Ch. 1 [10-1-94 Edition]) 172 8-5 South Florida showing the interwetland movements of individual radiotagged adult snail kites during 1992 and 1993. These movements illustrate a basic habitat network used by snail kites (also shown). We have shown only a limited subset of this network (and moments) to minimize cluttering, and because a complete synthesis of the peripheral habitats has not been done. The complete movements, and consequently the complete network, would include all movements and habitats used by kites throughout the state 173 8-6 The percentage of Snail Kite nests (N=745) that were initiated in each 10 cm water depth class. Data are from Bennetts et al. (1988), B. Toland (unpubl. data), and this study 174 8-7 The distribution of Snail Kite nests in Water Conservation Area 3 A during each year from 1992 through 1996. During 1995 this area experienced exceptionally high water levels as a result of Tropical Storm Gordon and the distribution of nesting kites shifted to higher elevations 175 xii

PAGE 13

8-8 Conceptual model of relative habitat quality in relation to the time since a drying event at a given location. In the absence of a drying event (A), habitat quality initially increases as the apple snail population recovers, but declines after 5-6 years as the plant communities comprising nesting and foraging habitat begin degradation. If drying events occurs too frequently (B), the apple snail population will have been unlikely to have recovered to its full potential. If drying events occur at longer intervals (C) then a cumulative process of slow and incremental degradation will occur as plant communities undergo transition 176 8-9 Primary plant communities and their corresponding species that comprise Snail Kite habitat in relation to elevation and a hydrologic gradient 177 8-10 The reported nesting distribution of nesting snail kites (shaded) in Water Conservation Area 3 A (WCA3A) from 1965 to present. Birds nesting in southeastern WCA3A during the 1992-1996 period were foraging primarily in Everglades National Park and the "Pocket" between the L-67A and L-67C levees, both of which have shorter hydroperiods than the nesting area 178 8-1 1 A conceptual hydrologic window for a long (e.g., WCA-3 A) and short (e.g., Big Cypress N.P.) hydroperiod wetlands. This window can shift over time depending on the hydrologic conditions at different scales 179 8-12 The distribution of Snail Kite nests in Big Cypress National Preserve during each year from 1992 through 1995 (no nests were observed from 1992-1994). 180 8-13 Hypothesized relationship between the spatial extent of droughts and whether the response by Snail Kites is likely to be behavioral (i.e., movement) or numerical (i.e., change in survival and/or reproduction) 181 xiii

PAGE 14

Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy THE DEMOGRAPHY AND MOVEMENTS OF SNAIL KITES IN FLORIDA by Robert Edwin Bennetts May, 1998 Chairman: Dr. Wiley M. Kitchens Major Department: Wildlife Ecology and Conservation Several authors have suggested that changes in Snail Kite ( Rostrhamus sociabilis ) populations correspond with changes in hydrology. The primary objectives of this study were to estimate survival and movement probabilities and to evaluate the influences on those probabilities. I estimated survival of Snail Kites in Florida using the Kaplan-Meier estimator with data from 271 radio tagged Snail Kites over a three-year period and capture-recapture models with data from 1319 banded birds over a six-year period. Results from these data sources were similar in their indications of the sources of variation in survival, but differed in some parameter estimates. Both data sources indicated that survival was higher for adults than for juveniles, but did not support delineation of a subadult age class. Our data also indicated that survival differed among years and regions for juveniles, but not adults. My results indicated that the probability of movement is influenced by age, time, and location. Although relatively high water conditions persisted throughout this study, xiv

PAGE 15

water levels (independently of location) did not appear to influence monthly movement probabilities. Dispersal is generally thought to be favored when local resources are low or better conditions exist elsewhere. In contrast, my results from both within-year and between year comparisons suggest that higher probabilities of movement occur when food resources were high. I suggest a hypothesis that this may be a reasonable strategy given the dynamic and unpredictable nature of a kite's environment. My data are consistent with previous views that the habitats used by Snail Kites in Florida are considerably more extensive than the currently-designated-critical habitat. Thus, protection of only the currently designated critical habitat may be insufficient to maintain viable populations of Snail Kites over the long term. The use of habitats can be characterized as an extensive network and I present a hypothesis of how the spatial and temporal patterns of this network might influence viability of Snail Kites in Florida. xv

PAGE 16

CHAPTER 1 INTRODUCTION Overview Florida's wetlands have undergone extensive anthropogenic change over the past century including drainage, impoundment, changes in water flow regimes, increased nutrient loadings, and invasion by exotic plants and animals. The Snail Kite (Rostrhamus sociabilis ). like many other species, is potentially influenced by these environmental changes. Snail Kite populations during this century have changed considerably in number and distribution and several authors (e.g., Sykes 1984; Beissinger 1988; Bennetts et al. 1994, Sykes et al. 1995) have suggested that changes in kite populations correspond with changes in environmental conditions, particularly hydrology. Our knowledge, however, of demographic processes and their influences is far from complete (Bennetts and Kitchens 1994). Changes in the size of all populations are a sum of births and immigration minus deaths and emigration. The Florida population of Snail Kites, however, is perhaps simpler in that all evidence suggests that this population is closed with respect to immigration and emigration. Snail Kites in Florida have long been known for their nomadic tendencies (Stieglitz and Thompson 1967, Sykes 1979, Bennetts 1993), leading to suggestion that the Florida population is not comprised of discrete subpopulations, but instead, is one population that frequently shifts in distribution throughout the state (Bennetts and 1

PAGE 17

2 Kitchens 1992, 1993). There has been speculation about exchange between populations of the United States and Cuba (e.g., Sykes 1979, Beissinger et al. 1983, Sykes et al. 1995); however, no evidence supporting this hypothesis has emerged. Thus, from a demographic perspective, I view the Florida Snail Kite population as geographically closed, although movements within Florida may play an important role in the population dynamics of this species. The birth and death processes can be conceptualized as part of the demographic cycle represented by parameters for survival (0,) and fecundity (f) (Beissinger 1995, Legendre and Clobert 1995)(Figure 1-1). Reliable estimates of these demographic parameters would enable a wide range of demographic modeling (e.g., viability analyses and risk assessments) with a higher degree of confidence. It also would increase our predictive capability regarding the response of Snail Kites to changes in water management. Given the scope of projects currently being planned or implemented (e.g., the Central and South Florida Project, the South Florida Ecosystem Restoration Initiative, Kissimmee River Restoration, Upper St. Johns River Basin Project, Kissimmee chain of Lakes Fishery Restoration) an improved predictive capability would be highly beneficial and would greatly reduce controversies. The goal of this study was to better understand Snail Kite population and spatial dynamics and how they are affected by both natural and anthropogenic processes. I believe that demographic models play an important role in the refinement of our understanding of these dynamics. However, it is also my belief that reliable parameter estimates, particularly if a model is sensitive to those parameters, are an essential basis for

PAGE 18

3 reliable model outputs. Lastly, I believe that our models, as well as our knowledge, should be an iterative and adaptive process (Walters 1986). As we acquire new information or better parameter estimates, or if our predictions are falsified, we need to adjust our models, as well as our thinking, to adapt to new information (Figure 1-2). Objectives In as much as there appears to be general agreement that changes in Snail Kite populations are more sensitive to survival than to reproduction (Nichols et al. 1980, Beissinger 1995, Sykes et al. 1995), data to estimate survival are very limited (Snyder et al. 1989a) and as a result, reliable estimates of survival are sorely lacking (Beissinger 1995). Consequently, the first objective of this study was to estimate adult and juvenile survival and to evaluate the influences of environmental conditions (e.g., hydrology) on survival. In addition to this primary goal, I also recorded supplementary information on reproductive parameters to the extent that it did not conflict with accomplishing my primary goals and in areas where such information was not already being collected. Most previous research on the demography of Snail Kites has focused on reproduction. Nesting success, in particular, has received considerable attention in recent years (e.g., Sykes 1987b, 1987c, Bennetts et al. 1988, 1994; Snyder et al. 1989a). There remains considerable debate about what factors influences nesting success (Bennetts et al. 1994, Sykes et al. 1995); however, compared to other species, the relationship between nesting success of Snail Kites and environmental conditions is relatively well understood. Other reproductive parameters are less well known. Unsubstantiated estimates or speculations have been made regarding the proportion of birds attempting to breed each

PAGE 19

4 year and the number of nesting attempts per year (e.g., Nichols et al. 1980, Snyder et al. 1989a, Beissinger 1995); however, reliable estimates for these parameters have been lacking. In addition to demographic parameters, movements of Snail Kites are also poorly understood and have been the subject of recent controversy during the planning of marsh restoration within central and southern Florida. While long term changes in Snail Kite distribution tend to coincide with changes in hydrologic regimes, shorter term (e.g., annual and seasonal) shifts do not always coincide with local hydrologic conditions (Bennetts et al. 1994). It has been hypothesized that dispersal of kites may be in response to hydrologic conditions (Takekawa and Beissinger 1989), localized food depletion (Bennetts et al. 1988), or localized environmental conditions (e.g., dissolved oxygen in the water) that may influence apple snail ( Pomacea paludosa ) availability (Bennetts et al. 1994). To what extent movements reflect long-term changes in habitat quality versus short-term environmental dynamics is poorly understood, as is the bird's ability to locate and re-colonize wetlands that have been, or will be, restored. Thus, movements are critical to understanding Snail Kite population dynamics leading to my second primary objective to evaluate the movement patterns of Snail Kites in Florida including rates, locations, and what environmental conditions are correlated with movements.

PAGE 20

5 Figure 1-1. The demographic cycle of Snail Kites showing three age classes (Juveniles=J, Subadults=S, and Adults=A). Parameters for survival ((b) and fecundity (f) are shown for each age class. Adapted from Caswell (1989), Beissinger (1995) and Legendre and Clobert (1995).

PAGE 21

6 Figure 1-2. Conceptual framework for this study. Reliable estimation of parameters is the first step toward the development of a wide variety of demographic models.

PAGE 22

CHAPTER 2 STUDY AREA Snail Kites within the United States occur only in Florida (Sykes 1984). It has been suggested (Bennetts and Kitchens 1992, 1993, 1994, Beissinger 1995) that Snail Kites comprise one population that shifts in distribution throughout the state, rather than there being separate subpopulations within the state. Data from studies on movements (this study) and genetics (Rodgers and Stangel 1996) support that there is considerable interchange of birds among wetlands in Florida. Consequently, it was deemed essential for the scope of this study to include the entire population of Snail Kites in Florida and my study area comprised a network of wetlands throughout central and southern Florida within the entire documented range of Snail Kites (Figure 2-1). Spatial Scales Because the scale of my study is statewide, I did not focus on movements within individual wetlands. For the purpose of this study, I considered wetlands to be distinct if they were separated by a physical barrier (e.g., ridge or levee) and/or were under a different hydrologic regime either through natural or managed control. Thus, adjacent wetlands, which were once hydrologically continuous (e.g., WCA-2A and WCA-2B), were considered separate units if they were under different water regulation schedules. Although I recorded locations of animals by specific wetland, for many analyses I 7

PAGE 23

8 had insufficient data to consider wetlands individually. For example, in the highly fragmented agricultural areas, there were more than 50 wetlands used by kites during this study. Consequently, some pooling of locations was required. For most analyses agricultural areas were pooled into a single class of wetlands. It was not uncommon for kites to frequent several such wetlands in immediate proximity, and I seldom (if ever) would have had sufficient data to support estimating parameters (i.e., survival or movement probabilities) for each of these wetlands. Other cases of pooling are described below, or are reported on a case-specific basis based on model selection criteria. Regions For some analyses (e.g., survival) I treated location at a regional scale because it was infeasible to estimate separate parameters for all wetlands. Based primarily on watersheds, climatic factors, physiography, and management regimes, I assigned each location to one of five primary regions (Figure 2-2). Locations not included in these five regions (e.g., agricultural areas and isolated peripheral wetlands) were assigned to a sixth region I call the peripheral region. Undoubtedly, there are differences in the quantity and quality of habitats within this sixth "catch all" region (and within the 5 primary regions as well); however, the amount of data required to partition the effects of this within-region variability would be enormous and require significantly more effort that the scope of this study. However, whenever the data supported partitioning beyond a regional scale I did so. The Everglades and Big Cypress Region is comprised of Water Conservation Areas 1,2, and 3, Everglades National Park, and Big Cypress National Preserve. The

PAGE 24

9 Loxahatchee Slough Region is comprised of wetlands in the drainage system of the Loxahatchee Slough and vicinity including the Corbitt Wildlife Management Area, PalMar Water Control District, private wetlands owned by PrattWhitney Corp., and wetlands within the Loxahatchee Slough owned by the City of West Palm Beach (i.e., the West Palm Beach Water Catchment Area and vicinity). The Okeechobee Region is comprised of Lake Okeechobee within the Herbert Hoover Dike. The Kissimmee Chainof-Lakes Region was comprised of all lakes within this chain including Lakes Kissimmee, Tohopekaliga, East Tohopekaliga, Marion, Marian, Tiger, Pierce, Jackson, and Hatchineha. The Upper St. Johns Region includes wetlands within the Upper St. Johns River Basin, but most Snail Kites used the Blue Cypress Marsh Water Conservation Area, Blue Cypress Water Management Area, and surrounding wetlands in private ownership. Agricultural areas (e.g., citrus groves, canals, agricultural fields, or agricultural retention ponds) within each of these regions, as well as all other areas not included in one of the above regions, were assigned to the peripheral region. Habitat Types Snail Kites inhabit freshwater wetlands throughout central and south Florida. There is considerable variation in the physiographic characteristics and specific plant communities that comprise Snail Kite habitat (reviewed by Sykes et al. 1995). My objectives did not warrant documentation of micro-habitat use by kites, nor was my sampling (often by aircraft) conducive to recording such data. However, for some analyses I wanted to incorporate the effects of at least a broad classification of habitats being used by kites. This classification had to be broad enough to enable assignment of

PAGE 25

10 locations obtained from aircraft to a given habitat type and sufficiently broad such that micro-habitat variation did not confound the assignment given normal daily movements of foraging birds. Consequently I assigned each location to one of five habitat types: (1) graminoid marsh, (2) cypress prairie, (3) Okeechobee, (4) northern lakes, and (5) miscellaneous peripheral. Graminoid marshes were generally slough and wet prairie communities (Loveless 1959). I distinguished cypress prairies in that a dominant feature of the landscape profile was comprised of cypress. This habitat occurred primarily in western WCA3A, and portions of the Big Cypress National Preserve and Loxahatchee Slough. The littoral zone of Lake Okeechobee is an extensive system of diverse marsh habitats, and consequently had elements of at least three of my other habitat types (i.e., graminoid marsh, northern lake, and highly disturbed). Because of this high local diversity I was unable to assign locations to a particular type without extensive ground verification. Even then, birds often used more than one of these habitat types within a given day. Thus, I assigned each location at Lake Okeechobee to its own habitat type. The northern lake habitat type consisted primarily of lakes within the Kissimmee Chain-of-Lakes, but also included a few lakes along the Lake Wales Ridge. In contrast to Lake Okeechobee, this habitat type generally was comprised of a narrow littoral zone (usually < 200 m) on the periphery of these lakes. This littoral zone had a relatively steep elevation gradient compared to other habitat types; the zone used by foraging kites often was a band of < 100 m usually dominated by maidencane (Panicum spp) interspersed with patches of bulrush (Scirpus spp) or cattail ( Typha spp). Primary nesting areas were often a zone of cattail and/or willow (Salix spp) in the shallower zone adjacent to foraging areas. The

PAGE 26

11 peripheral habitat type was comprised primarily of agricultural areas. These included retention ponds for citrus groves, agricultural ditches, and other miscellaneous, usually highly disturbed, habitats. Larger canals, not necessarily associated with agriculture, were also included in this habitat type. For some analyses I had insufficient data to partition locations into each of these habitat types. Consequently, for some analyses I assigned locations to an even broader category of lakes (i.e., Lake Okeechobee, the northern lake habitat type, and permanently flooded canals [<0.01% of my locations]) and marshes (any non-lake habitat). This was intended to distinguish habitats that had a permanent water source component available (even if it was not used) from those that dried periodically.

PAGE 27

12 Figure 2-1 Major wetlands of South Florida referred to in this report. Wetlands are Everglades National Park (ENP), Big Cypress National Preserve (BICY), Water Conservation Areas 3 A, 3B, 2B, 2A, Loxahatchee National Wildlife Refuge (LOX), Holey Land Wildlife Management Area (HOLEY), West Palm Beach Water Catchment Area (WPB), Lake Okeechobee (OKEE), Upper St. Johns [Blue Cypress] Marsh (SJM), Lake Kissimmee (KISS), Lake Tohopekaliga (TOHO), and East Lake Tohopekaliga (ETOHO).

PAGE 28

13 Figure 2-2. South Florida showing geographic regions used for some analyses in this report. All areas not within a region shown were assigned to a peripheral region.

PAGE 29

/ CHAPTER 3 ANNUAL SURVIVAL OF SNAIL KITES IN FLORIDA WITH COMPARISONS BETWEEN RADIO TELEMETRY AND CAPTURE-RESIGHTING DATA Introduction For many long-lived avian species, population persistence is more sensitive to annual survival than fecundity (Mertz 1971, Nichols et al. 1980, Beisssinger 1995). Despite this, reliable estimates of survival remain unavailable for many species, while extensive effort often is expended estimating reproductive parameters. Investigators also must choose among available techniques for estimating parameters of interest. This selection often is based on logistic constraints, or unfamiliarity with potential estimators, rather than how procedure selection might influence resulting parameter estimates. Given current threats to many populations, reliable demographic data are essential for effective conservation arguments in the context of alternative management scenarios. The Snail Kite (Rostrhamus sociabilis ) is an endangered raptor whose range in the United States is limited to central and southern Florida (Sykes et al. 1995). Florida's wetlands have been severely altered over the past century by drainage, impoundment, changes in water flow regimes, increased nutrient loadings, and invasion by exotic plants and animals (Walters et al. 1992, Davis and Ogden 1994). This has resulted in what is one of the largest ecosystem restoration projects ever undertaken (Davis and Ogden 1994). The Snail Kite, like many other species, is potentially influenced by these, as well as other, 14

PAGE 30

15 changes (Bennetts et al. 1994). Consequently, reliable estimates of demographic parameters are essential to understanding population responses to environmental change (Nichols et al. 1980). There have been several previous reports of annual survival of Snail Kites in Florida, although none has used reliable statistical estimators to derive estimates. Snyder et al. (1989) estimated minimum annual survival of Snail Kites by using the number of birds that were banded over a 10-year period from 1968-1978 that were observed alive in 1979. They did not use available capture-recapture estimators for these data because of limited effort to resight banded birds in all but one year (Snyder et al. 1989). Hence, their approach provides a crude indication of minimum annual survival, but does not provide estimates that reliably can be used for demographic assessments. Several other authors have reported estimates of Snail Kite survival based on differences between annual surveys conducted in consecutive years (e.g., Sykes 1979, Beissinger 1988, 1995). This approach fails to account for a high potential for confounding changes in detection probability with changes in population size (Bennetts and Kitchens 1997a). Problems with using count data without accounting for detectability have been well recognized (Burnham 1981, Nichols 1992, Johnson 1995, Link and Sauer 1997). Bennetts and Kitchens (1997a) found that the number of Snail Kites counted during these surveys was strongly influenced by differences in observers, effort, and sites, all of which potentially influence detection probabilities. None of these influences has been taken into account for any survival estimates using these data. Thus, I believe that using the annual survey to estimate survival, without accounting for detection, fails to provide reliable estimates.

PAGE 31

16 Here, I estimate survival using reliable statistical estimators. This study also was part of a larger study focused on both survival and movement. This provided the opportunity to estimate survival using data obtained from both radio telemetry (the primary tool I used to examine movement patterns) and capture-recapture (resighting); enabling comparison of survival estimates derived from these independent data sources. I was also able to test hypotheses about factors likely to influence survival using these data sources. Methods Estimation of Survival from Radio Telemetry Adult kites were captured using a net gun (Mechlin and Shaiffer 1979), which uses 22 caliber blank cartridges to propel a 10-foot triangular nylon net. Juveniles were captured just prior to fledging, at approximately 30-35 d old, without a net gun. Fifteengram radio transmitters were attached to birds with backpack harnesses. My goal was to annually capture and radio tag 100 snail kites of which 60% were adults and 40% juveniles for three consecutive years from April 1992 through April 1995. My targeted ratio of adults to juveniles was intended to emphasize adult survival because demography of Snail Kites probably is more sensitive to adult rather than juvenile survival (Nichols et al. 1980, Beissinger 1995). To maintain independence of my sample, a maximum of one juvenile per nest was equipped with a radio transmitter. I targeted a 50:50 sex ratio of adults to keep my sample balanced. The proportion of samples from each area was based on the annual survey to approximate the statewide distribution (Bennetts and Kitchens 1997a). My targeted annual sample size of 100 was

PAGE 32

17 based on having sufficient statistical power (e.g., > 0.8) to distinguish differences (e.g. A 0*0. 1-0.2) among groups (e.g., age or sex) or time periods from a hypothesized survival estimate (<^) of 0.90 (Bennetts and Kitchens 1997a). Radiotagged birds were located at approximately 14-day intervals from aircraft or ground searches to determine their location and whether they were alive. All radios were equipped with mortality censors which changed pulse rates if the transmitter had not moved for 6-8 h. Birds with a transmitter emitting a mortality signal were then located on the ground to verify their fate. I estimated survival () of radio-tagged kites using a staggered entry design (Pollock et al. 1989) with the Kaplan-Meier product limit estimator (Kaplan and Meier 1958). I used an arbitrary starting date of 15 April 1992 for annual survival estimates. By this time during my first year I had a sample (n = 16) sufficient to allow reasonable estimates of survival. Subsequent evaluation of annual survival was based on study years (SY) from 15 April to 14 April of consecutive years (Bennetts and Kitchens 1997a). Estimation of Survival from Banding Data My sample of banded birds for survival analyses was obtained through a cooperative banding effort with the Florida Game and Fresh Water Fish Commission (GFC). My sample also was supplemented by resightings of birds banded during two previous studies by REB (unpubl. data) and J. A. Rodgers (unpubl. data) that were observed during this study. A previously-banded bird observed alive during my study at time t was treated as a newly-marked individual. I estimated annual survival from banding data using the capture-recapture

PAGE 33

18 (resighting) models originally developed by Cormack (1964), Jolly (1965) and Seber (1965). The basic CormackJolly-Seber (CJS) approach has undergone extensive advancement in recent years to become a flexible and unified framework capable of handling simple to complex models of survival (Lebreton et al. 1992, Nichols 1992). Recent approaches enable evaluation of effects attributable to individual characteristics (e.g., age and sex) and environmental variables (e.g., weather). Additional models have capability to incorporate transition probabilities and multiple strata (e.g., exchanges of individuals among geographically stratified populations)(Brownie et al. 1993, Nichols et al. 1993). All analyses of capture-recapture data were conducted in either Program SURVIV (White 1983, White and Garrott 1990) or MSSURVIV (Hines 1994). I conducted capture-resighting over six sampling occasions from 1992-1997. My capture and resighting occasions corresponded with the peak fledging time of Snail Kites (MarchJune) (Bennetts and Kitchens 1997a). Thus, survival estimates can be roughly interpreted as survival from one breeding season to the next, regardless of whether a given animal was breeding. Snail Kites have a relatively long breeding season and are not synchronous in their breeding attempts (Snyder et al. 1989, Bennetts and Kitchens 1997a). Consequently, the time span over which fledging, and therefore banding, occurred was relatively long. I tried to minimize the time span of my sampling by limiting my capture and resighting period to the peak four months of fledging.

PAGE 34

19 Influences on Survival Survival of young birds tends to be lower than that of adults in many species (e.g., Ricklefs 1973, Loery et al. 1987). However, Ricklefs (1973) points out that "Just how much experience the young need to attain adult behavior and physiological capabilities (and thus adult survival rates) is open to question." Beissinger (1995) suggested that Snail Kites have three age classes with respect to survival (juveniles or age 0-1, subadults or age 1-2, and adults or age > 2 yr); nevertheless, the survival estimates he used for his demographic modeling were the same for subadults and adults. I predicted lower survival for juvenile Snail Kites compared to adults or subadults. I further hypothesized that, if subadult survival differed from that of adults and juveniles, that it would be intermediate between the two. To test these hypotheses, I first considered kites as adults after their first year post fledging. Juvenile Snail Kites are capable of breeding at 9 months of age (Snyder et al. 1989). For my capture-recapture models, resighting probability at the first resighting period after initial capture (time 2) was considered to be equal for juveniles and adults. Bennetts and Kitchens (1997a) tested this assumption by comparing models in which juveniles and adults had different resighting periods at time 2 to models in which resighting was equal for the two ages. They concluded that separate estimates for resighting probability were not warranted. I then tested the hypothesis that adult and subadult survival does not differ by parameterizing a CJS model such that birds banded as juveniles were considered to have three age classes with respect to survival rates (i.e., juvenile survival during their first year, subadult survival their second year, and adult survival after year two). In addition to age effects, there is substantial variability in habitat

PAGE 35

20 quality for Snail Kites over space and time (Bennetts and Kitchens 1997a, b), which could result in differences in survival. However, because Snail Kites are highly mobile they have the potential to escape to other areas when conditions are poor. Adults, having had more experience at alternative sites and the corresponding selective pressures of environmental variability, may be less susceptible to temporal variation than younger birds. Consequently, I hypothesized that, if temporal variation in survival exists, that it would be greater for younger birds than for adults. I tested temporal effects using a sequence of models analogous to Models A, B, C, and D described by Jolly (1982) and Pollock et al. (1990). Model SSI (Pollock's Model A) treats both survival (0) and resighting probabilities (p) as variable over time (i.e., separate estimates of each parameter were derived for each year). Model SS2 (Pollock's Model B) treats resighting probability, but not survival, as variable over time. Model SS3 (Pollock's Model C) treats survival, but not resighting probability, as variable over time. Model SS4 (Pollock's Model D) treats both survival and resighting probabilities as constant over time. I then incorporated age effects into this sequence of models (Pollock et al. 1990). Because variability in habitat quality occurs over both space and time (Bennetts and Kitchens 1997a, b), I was interested in regional (spatial) effects of survival in addition to annual (temporal) effects. For the same reasons as temporal variation, I predicted that, if regional variation in survival exists, that it would be greater for younger birds than for adults. I tested for regional differences in survival using radio telemetry data two ways. First I tested the hypothesis that differences in juvenile survival were attributable to their region of natal origin. For this analysis, a bird was assigned to its natal region, regardless

PAGE 36

21 of whether or not it moved subsequent to initial capture. In most cases I did not know the natal origin of adults or their history of locations prior to capture. Consequently, I limited this approach to juveniles. The second approach I used for testing regional differences in survival was based on time at risk in each region, rather than focusing only on natal region. Thus, I tested the hypothesis that survival was affected by current location (e.g., by local factors such as predation risk). For this analysis, a bird that moved from a given region to another was censored (White and Garrott 1990) from the number of animals at risk for the region from which it moved and added to the number of animals at risk in the region to which it moved. All movements and corresponding changes in the number of animals at risk were assigned at the midpoint of the time interval between locations. All deaths were assigned to the region where the dead bird was found. To test for regional effects of survival and resighting probabilities from capturerecapture data, I generated a suite of multi-strata models analogous to the models described above, except that they enabled stratum-specific parameter estimation (in this case strata= 4 of the 6 regions of capture ~ no captures occurred in the peripheral region and I had too few observations in the Loxahatchee Slough to include it in the analysis) (Brownie et al. 1993, Nichols et al. 1993). As above, I generated models with and without age dependency, enabling us to test hypotheses that


PAGE 37

22 the probability that an animal in stratum r at time / was alive in stratum s at time t +1, given that it was alive at / + 1) were also generated from these models; however, my primary interest here was site-specific estimates of


PAGE 38

23 Results I attached 282 radio transmitters on 271 individual Snail Kites; 1 1 birds were recaptured in a subsequent year and their radios replaced. I attached 82 radios during SY 1992 and 100 each during SYs 1993 and 1994. Of the 282 radios, 165 (59%) were placed on adults and 117 (41%) on juveniles. Of the adults, 82 (49.7%) were males and 83 (50.3%) were females. My total sample of banded birds used in CJS models was 1319. Of these, 164 were initially banded as adults and 1 155 as juveniles. However, an additional 301 resightings of birds initially banded as juveniles supplemented my sample of adults (Table 3-1). Age Effects My results from both radio telemetry and capture-recapture data indicated that survival differed between adult and juvenile Snail Kites. Based on log-rank statistics using radio telemetry, survival differed between these age classes for SYs 1992 (x 2 =4.61, 1 df, P=0.032) and 1994 (x ?=B 29.52, 1 df, P<0.001), but not 1993 (x 2 =0.027, 1 df, P=0.869). In both years where they differed, estimates of adult survival were higher than estimates of juvenile survival. My capture-recapture data also indicated that survival differed between adults and juveniles. All models that included age effects on survival had lower AIC scores than corresponding models without age effects (Table 3-2), and LRTs between models with and without age effects on survival strongly rejected the more reduced models further supporting the inclusion of age effects (Table 3-3). I used two variations of my most parsimonious model (Model SS10) to test the hypothesis that survival of subadult (i.e., age 1-2) Snail Kites differed from adult survival.

PAGE 39

24 Both of these models had separate parameter estimates for subadult survival; however, in one model subadult survival was held constant among years and for the other it was variable among years. LRTs between Model SS10 and each of these more general models failed to reject the more reduced model (x 2 =2.37, 1 df, P=0.124 and X 2= 2 38, 3 df, P=0.498 for each LRT, respectively), indicating that separate parameter estimates for subadult survival were not warranted for these data. Time Effects Both data sources indicated that survival differed among years for juveniles, but not adults. Estimates of survivorship functions of adult Snail Kites using radio-telemetry data did not differ among years at (a = 0.05) between SYs 1992 and 1993 (x 2 =2.84, 1 df, P=0.092), 1992 and 1994 faM.76, 1 df, P=0.184), or between 1993 and 1994 (x^O.48, 1 df, P=0.486). In contrast, my estimates of survivorship of juveniles differed between SY 1992 and 1994 (x^-16, 1 df, P=0.013), 1993 and 1994 (x 2 =12.41, 1 df, P<0.001), but not between 1992 and 1993 (x 2 =1.43, 1 df, P=0.231). I also found strong evidence, using capture-recapture data, for the inclusion of time (year) effects for juvenile, but not adult survival. The AIC scores of models with time effects were lower than corresponding models without time effects. LRTs between models with and without time effects also supported this conclusion, except when time effects were limited to adult survival. Based on my results from radio-telemetry data I generated two models in which differed between adults and juveniles and was variable among years for juveniles, but not adults. For Model SS9, p was constant among years, and for Model SS 10, p differed among years. Model SS 10 had the lowest AIC score of any model, goodness-of-fit was

PAGE 40

25 reasonable (G=26.326, 19 df, P=0. 121), and a LRT between Model SS10 and SS5 (an identical model, except that differed among years for both juveniles and adults) failed to reject the more reduced model (SS10). These results indicated that survival differed among years for juveniles, but not adults; and that resighting probabilities also differed among years. Regional Effects There was little indication of regional differences in adult survival using either radio-telemetry or capture-recapture data. Of 1 5 pairwise comparisons, using radiotelemetry data, of adult survival between regions during each year (for which I had sufficient data) only one differed at a=0.05. Adult survival differed between the Everglades and Okeechobee regions during SY 1994 (x 2 =4.06, 1 df, P=0.044). If the a level was adjusted for inflation due to simultaneous comparisons (e.g., using a Bonferonni correction), none of the 15 comparisons was significant at a=0.05. For juveniles, none of eight survivorship functions (for which I had sufficient data), based on actual time in each region, was significant at a=0.05. For survivorship functions based on natal region, 1 of 10 comparisons was significant. There was a difference between the Okeechobee and Everglades regions during SY 1992 (x 2 =4.58, 1 df, P=0.032); however this result also would not be significant at
PAGE 41

26 among regions for juveniles. A model (Model MS 13) in which survival (1) differed among age classes, (2) differed among years for juveniles, but not adults, and (3) differed among regions for juveniles, but not adults had the lowest AIC score (Table 3-4). An LRT between this model and an analogous model (Model MS7) in which survival differed among years and regions for both age classes failed to reject (x 2= l 1 42, 1 5 df, P=0.722), further supporting that these effects were warranted for juveniles, but not adults. Similar to the single stratum models, Model MS 13 indicated that resighting probabilities differed among years, but also indicated differences among regions. Parameter Estimates Overall estimates of adult survival were very similar using the Kaplan-Meier estimator with radio telemetry data (Table 3-5) and the CJS models with capturerecapture data (Table 3-6). In contrast, estimates of juvenile survival tended to differ both in the overall estimates and even in the rank order of estimates among years. Overall estimates using multi-strata models tended to be lower for both age classes than estimates derived from either Kaplan-Meier or CJS estimators (Table 3-7). Estimates of resighting probabilities also differed substantially between single-stratum or multi-strata models. The precision of individual parameter estimates ranged from 3 to 92% coefficient of variation (CV), depending on the number of parameters being estimated and the distribution of my sample for a given estimate. CVs for my estimates of adult survival were 3.2% using the Kaplan-Meier estimator, 6.0% from my final single-stratum model (Model SS10), and 4. 1% using my final multi-strata model (Model MS13). Average CVs for juvenile survival were 13 .4% using the Kaplan-Meier estimator, 18 .9% my final

PAGE 42

27 single-stratum model (Model SS10), and 36.7% using my final multi-strata model (Model MS13). Censoring of Radio-tagged Birds Censoring is the removal of radio-transmittered animals from a sample when the radio-transmitter signal can no longer be detected (White and Garrott 1990). An assumption for an unbiased estimate using the Kaplan-Meier estimator is that censoring is random with respect to fate (Pollock et al. 1989); that is, the probability of a bird being censored is not related to its fate. In the case of simple radio failure this assumption probably is valid; however, when a radio is destroyed when an animal dies (e.g., during predation or scavenging) this assumption may not be valid (White 1983). Censoring due to radio failure would not be expected to differ among adults and juveniles. My results indicated that the mean time to censoring differed strongly from this expectation (f=3.77, df=179, P<0.001). Juveniles, but not adults, had a substantial surge in the number of censored animals within the first 60 days after radio attachment (Figure 3-1). This result would have been expected if juveniles either left the study area or experienced undetected mortality. Dead Snail Kites were usually found in water where radio signal strength was strongly diminished. I suspected that some mortality went undetected as a result. Consequently, during SY 1994 I increased my search effort for missing birds. I then examined the proportions of censored and dead birds during the first 180 d after radio attachment (i.e., before radio batteries likely died). The proportions of adults censored and confirmed dead remained relatively constant among years (x 2 =1.02, 2 df, P=0.601)(Figure 3-2). In contrast, the proportions of juveniles censored and confirmed

PAGE 43

28 dead were similar during SY 1992 and 1993, but differed during SY 1994, when search effort was increased (x 2 =30.25, 2 df, P<0.001). During SY 1994, the proportion of birds confirmed dead substantially increased and the proportion of censored birds substantially decreased. The proportion of censored juveniles during 1994 also closely matched the proportion of censored adults, which it had not during 1992 or 1993. Discussion Comparisons Between Data Obtained Using Radio Telemetry and Capture-resighting The results from radio telemetry and banding data were generally consistent in identifying sources of variation. Both data sources indicated that survival differed between age classes and among years for juveniles, but not adults. Singleand multi-strata capture-recapture models also indicated similar sources of variation for survival and resighting probabilities, except that the multi-strata models indicated additional regional effects. In contrast to sources of variation, there were considerable differences in some parameter estimates among data sources. Although both sources of data indicated differences among years for juvenile survival, the parameter estimates from these two data sources differed markedly and were not even consistent in their relative ranking among years. Estimates of juvenile survival during 1992 and 1993 were higher using radiotelemetry data than for either capture-recapture models. I believe that this was due to a bias for my estimates using radio telemetry data during those years. My results from censored radio-tagged birds indicated that I was finding dead juveniles during 1994 when search effort was increased; whereas a substantial number of dead birds may have gone

PAGE 44

29 undetected during 1992 and 1993. Thus, my survival estimates using radio telemetry probably were biased high for juveniles. I had no such evidence for adults. Another assumption using radio telemetry to estimate survival is that the radio transmitter does not affect survival (White and Garrott 1990). There has been substantial evidence in recent years to suggest that, for some species, radio transmitters may negatively affect survival (e.g., Marks and Marks 1987, Burger et al. 1991, Paton et al. 1991). Bennetts and Kitchens (1997a) tested the hypothesis that radio transmitters negatively affect survival of Snail Kites using capture-recapture of birds with and without radio transmitters. They had reasonable power to detect any substantial differences, yet found no effect. In contrast to radio telemetry, I had no reason to suspect that violations of my CJS models significantly biased my results. Probably the most substantial violation was for the assumption that capture and release of animals occurs over brief time intervals (Pollock et al. 1990). This assumption enables a clear definition of the interval over which survival is measured and helps to standardize intervals being compared. The life history of Snail Kites makes this assumption difficult to meet. However, I do not believe that violation of this assumption caused substantial bias to my estimates. For adults, the highest risk of mortality appeared to be during the fall and winter (Bennetts and Kitchens 1997a, Bennetts et al., unpubl. data). Thus, animals within a given study year all experience the same period of high risk. For juveniles, the highest risk of mortality occurs over the first few months post fledging and all juveniles within a given cohort also were exposed to that period of high risk.

PAGE 45

30 Band loss probably was negligible on my study because all but 19 (99%) birds were marked with riveted aluminum bands that were extremely unlikely to have been lost. The remaining 1 % were made of PVC and anecdotal evidence suggests that band loss from these bands also was negligible. I also believe that capture and release did not substantially influence the subsequent resighting of animals. Snail Kites are relatively tolerant of human presence and often allow humans to approach relatively close (Beissinger 1988). In addition, most birds were nesting at the time of resighting and tended to stay close enough to their nest to enable bands to be read with minimal difficulty. Parameter Estimates Because of the potential for biased estimates of juvenile survival using radio telemetry, I was more confident in my estimates using capture-recapture for this parameter. I also have greater confidence that my parameter estimates using singlestratum reflects actual survival. My data indicated that, at least for juvenile survival, regional effects were warranted. However, capture-recapture models estimate apparent survival, such that permanent emigration (i.e., permanent for the study) is confounded with actual survival. Because my data were insufficient to partition among two age classes and all six regions using multi-strata models, there is a potential for increased confounding of these two components of apparent survival. First, the four regions for which I had sufficient data were those with a greater number of sightings. This could be due either to greater use of these regions and/or a greater probability of observing birds that were present. This could account for the higher estimates of resighting probability

PAGE 46

31 observed from my multi-strata models. Similarly, any permanent emigration to these regions would have been included in the resulting estimates as decreased apparent survival. My single-stratum models included these regions because I was not attempting to derive separate parameter estimates. Thus, although I would expect my estimates of apparent survival using multi-strata models to be less biased because I was accounting for regional heterogeneity, there also may have been greater confounding of actual survival and permanent emigration in these estimates. This would explain the lower estimates of survival from my multi-strata models compared to estimates from radio telemetry or singlestratum models. Nichols et al. (1980) reported that adult survival of Snail Kites in Florida was 0.90. This was not based on a statistical estimator; rather, it was their "best guess" for demographic modeling. Similarly, Snyder et al. (1989) suggested that during non-drought years annual adult survival of Snail Kites probably exceeds 0.90, although this value also was not derived using any specified estimator. Beissinger (1995) later reported adult survival during non-drought years as 0.95 based on Snyder et al.'s suggestion. My estimates were similar to these previous reports of adult survival (&=0.89 and 0.92 from Kaplan-Meier and CJS estimators, respectively), but were based on reliable statistical estimators. In contrast to adults, my estimates of juvenile survival were not consistent with some previous estimates. Beissinger reported juvenile survival during non-drought years as 0.90. Nichols et al. (1980) reported a "best guess" of 0.58 for juvenile survival. My data suggest that juvenile survival may be substantially lower than Beissinger's estimate, but similar to the "best guess" reported by Nichols et al. (1980).

PAGE 47

32 Effects of Age. Time, and Region on Survival As predicted, I observed differences in survival between juvenile and adult Snail Kites, although separate estimates of subadult survival were not indicated by my data. Younger birds may have lesser foraging skill than adults (e.g., Verbeek 1977, Bennetts and McClelland 1997) and also may be more vulnerable to predation due to a lack of experience. My results also supported my hypothesis that younger birds are more susceptible to environmental variation than adults. Survival of juveniles, but not adults, differed among both years and regions. Environmental conditions, and consequently habitat quality for Snail Kites, may be quite variable in central and south Florida (Beissinger 1986, Bennetts and Kitchens 1997a). Adult kites are well adapted to this variability and are quite capable of moving throughout their range in response to changing conditions (Bennetts and Kitchens 1997a, b). In contrast, juveniles that have not yet experienced alternative locations, may be less efficient at locating alternative sites when local conditions are not favorable. Consequently, juveniles may be more sensitive to both spatial and temporal variation in the environment. Although my data indicate that juveniles, but not adults were sensitive to environmental variability, it has been suggested that survival during drought years may be substantially lower than during highwater years (Beissinger 1988, Takekawa and Beissinger 1989). Beissinger (1995) found drought-year survival to be one of the most sensitive parameters of his population viability model. Thus, adults may be susceptible to this more extreme case of environmental variability. Because I did not encounter drought conditions during this study, my results can not reliably be extended to drought years.

PAGE 48

33 Thus, there remains a need for reliable estimates of survival during drought years (see also Beissinger 1995). Implications of Resighting Probabilities Although emphasis of capture-recapture models is usually on survival, annual differences in resighting probabilities of Snail Kites also may have implications for a statewide monitoring program. An annual survey of Snail Kites was conducted every year from 1969-1994. Reported uses of these data include estimation of survival based on differences in counts between consecutive years (e.g., Beissinger 1988, 1995), and as an index of population size for comparisons between areas or years (Rodgers et al. 1988). Using count data for these purposes requires an assumption that either the survey represents a complete census, or that the proportion of birds detected does not differ between years or areas being compared (Lancia et al. 1994). The resighting probabilities I estimated suggest that the annual survey fails to meet either of these assumptions. My overall resighting probability using CJS models was 0. 19; whereas, a census is a complete count of all animals (Lancia et al. 1994). My results also indicated that resighting probability differed among years and regions; a result inconsistent with the assumption that the proportion of birds detected during the annual survey is constant. It must be taken into account that my estimates were derived during spring, whereas the annual survey is conducted during autumn. However, my results certainly supports concern for the validity of using these data for these applications.

PAGE 49

34 Table 3-1. Capture-resighting summary of adult and juvenile Snail Kites in Florida from 1992-1997. Year of Next Resighting Year of Last Birds Banded as Adults Capture or Never Resighting 9 2 93 94 95 96 97 Birds Banded as Juveniles 8 Resighted 92 93 94 95 96 97 Never Resighted 1992 4 10 5 1 23 11 14 8 104 1993 14 4 2 10 26 10 9 17 14 206 1994 1995 13 11 8 5 5 45 14 21 12 21 88 36 59 148 1996 13 46 158 Total No. Resighted Total New Captures 0 4 24 22 19 27 49 52 53 2 0 8 Total No Released' b 49 56 77 24 19 35 0 11 24 38 70 147 149 245 118 205 134 304 149 256 142 243 204 451 Considered to be adults at time 2 of each cohort. b Includes total resighted and new captures; however analysis is parameterized such that juveniles resighted as adults also contribute to estimation of adult surival.

PAGE 50

35 Table 3-2. Description of single-stratum Cormack-Jolly-Seber (CJS) models and their corresponding Akaike Information Criteria (AIC) scores. Parameter structure indicates whether survival (0) and/or resighting probability (p) was dependent on time (/) and/or age(a). Model Parameter No. Structure Parameters SSI P 3 275.15 SS9 ti^uP b 7 192.16 SS10 $Guv),aA 11 166.34 a Because


PAGE 51

36 Table 3-3. Likelihood ratio tests (LRTs) between Cormack-Jolly-Seber (CJS) models used to test whether survival (x 2 SS8 SS4 d> A op 16 219 1 1 <0 001 \J\J 1 SS7 SS2 6 Ace 21.861 1 <0.001 SS6 SS3 r Aee 50.168 5 <0.001 SSI SS2 Time 21.393 3 <0.001 SS3 SS4 f Time 61.335 4 <0.001 SS6 SS8 V Time 95.284 8 <0.001 SS6 ch

Time" 0.570 3 0.903 SS2 SS4 P Time 83.300 4 <0.001 SSI SS3 p Time 43.358 3 <0.001 SS7 SS8 p Time 88.942 4 <0.001 SS10 SS9 p Time 33.814 4 <0.001 a Tests for time variation of survival of adults only.

PAGE 52

37 Table 3-4. Description of multi-strata models and their corresponding Akaike Information Criteria (AIC) scores. Parameter structure indicates whether survival (0), resighting probability (p), and/or transition (movement) probability {iff) was dependent on age(a), time (t), and/or region (r). Model Parameter Structure No. Parameters AIC MSI &,,r Ar ft* 176 989.173 MS2 At Ar ftr 96 982.051 MS3 4>< Pur ft 36 940.507 MS4 V P,,r K 52 900.544 MS5 4>rPr$< 20 1032.120 MS6 4> V Pr ft,r 36 982.580 MS7 80 881.293 MS8 V Pi ft, 36 934.653 MS9 , Pt,r K 46 896.930 MS10 0a,^juv) Pxf fta,r 50 881.364 MS11 ^l(juvXr Pt ft,r 52 894.622 MS12 0a,t(juvXr Ar ft 68 867.264 MS13 0a,l(juv), ifluv) Pt,r fta,r 65 862.713

PAGE 53

38 Table 3-5. Annual estimates and standard errors for adult and juvenile survival (#) of Snail Kites for study years (SYs) 1992, 1993, and 1994 using data from radio telemetry. Adults Juveniles 4> SE(#) $ SE(#) 1992 0.962 0.038 0.825 0.080 1993 0.858 0.063 0.867 0.088 1994 0.883 0.042 0.439 0.090 Overall 4 0.894 0.029 0.671 0.059 a Estimated using a pooled sample of all years. The arithmetic mean gives equal weight to each annual estimate, whereas the pooled sample essentially weights by sample size.

PAGE 54

39 Table 3-6. Parameter estimates for the Cormack-Jolly-Seber (CJS) model SS10 in which survival () differed between adults and juveniles. Under this model, survival was constant among years for adults, but differed among years for juveniles. Resighting probabilities jp) differed among years. Adults Juveniles Adults Study Year X SE(#) X 9 SE ($) P SE(p) 1992 0.922 0.055 0.527 0.075 0.121 0.031 1993 0.922 0.055 0.350 0.051 0.198 0.030 1994 0.922 0.055 0.783 0.122 0.166 0.027 1995 0.922 0.055 0.942 0.187 0.168 0.034 1996 0.922 0.055 0.701 0.211 0.322 0.087 Overall 0.922 3 0.055 a 0.530" 0.047" 0.1 89 c 0.020 4 Adult survival in Model SS10 is constant over time. Estimated using Model SS7, which is identical to my selected model (SS10) except that is constant over time. This approach is equivalent to using a weighted mean estimate where weights are based on the variance-covariance matrix. c Estimated using Model SS9, which is identical to my selected model (SS10) except that p is constant over time. This approach is equivalent to using a weighted mean estimate where weights are based on the variance-covariance matrix.

PAGE 55

40 Table 3-7. Parameter estimates for my most parsimonious multi-stratum model (MS 13), in which survival differs between adults and juveniles, survival is constant among years and regions for adults, and survival differs among years and regions for juveniles. Resighting probability in this model differs among years and regions. Adults Juveniles Adults Studv kJ lUUy Year Region" SE(0) P SEfc) i aao iyyz EVER A OT) U.oZZ A AO A U.UJ4 A a on U.4o / A 1 1 A U.Z14 A AAA U.UUU A AAA U.UUU iyyz UK. be, u.ozz A AO A U.UJ4 A 1/IA U. /4U U. 14z A AC O U.UDJ A AO 1 U.UJ 1 iyyz U.5ZZ A AO A U.Uj4 A O/^l U.JO/ U. 1 10 U.jjo A 1 0 Z U. 1 J j iyyz T TC T A OH U.oZZ A AO A U.U34 A*1 /IT IT44 / a no U. Izo a ion U.zoU A 1 1 O U. 1 1 j 1 QQ1 iyyj C \ AC D IlVllK U.oZz A AO A U.UJ4 A A C\A U.4U4 A AQT u.uy / A 111 U.ZzZ A ACT U.Uj / 1 AAO iyyj UKtb a coo U.ozz a m a U.UJ4 U.4J0 A AOA u.usy A 1 1 A U. 1 1U A AOI U.UJ / iyyj VTCC U.oZZ A AO/1 U.UJ4 n i ao U. 1 Uz A A/IO U.U4y a U.jZ / A 111 U. I 51 1 QQ1 TTCT U3J U.oZZ A C\1A U.UJ4 A O/IO U. J4J A 1 1Q U. IZo A 1 1 0 U. 1 1 J A AAC U.UOj iyy4 H VxlK A on U.oZZ A AO /I U.UJ4 A HA U. 1 ZU A 1 AT U. 1U/ A 1/1 C U.Z45 A A/IO U.U4j 1994 OKEE 0.822 0.034 0.301 0.276 0.081 0.035 1994 KISS 0.822 0.034 0.275 0.097 0.248 0.080 1994 USJ 0.822 0.034 <0.00 <0.001 0.304 0.114 1995 EVER 0.822 0.034 0.454 0.074 0.199 0.037 1995 OKEE 0.822 0.034 0.437 0.198 0.387 0.103 1995 KISS 0.822 0.034 0.921 0.188 0.194 0.068 1995 USJ 0.822 0.034 1.000 0.317 0.368 0.127 1996 EVER 0.822 0.034 0.234 0.074 0.568 0.095 1996 OKEE 0.822 0.034 <0.00 0.412 0.389 0.121 1996 KISS 0.822 0.034 0.613 0.298 0.749 0.204 1996 USJ 0.822 0.034 0.248 0.175 0.756 0.242 Overall 0.822 0.034 0.441 b 0.036 b a Regions are Everglades (EVER), Okeechobee (OKEE), Kissimmee (KISS), and Upper St. Johns (USJ). There were insufficient sightings to include the Loxahatchee Slough Region. b Based on model MS9 for which survival is considered constant among years and regions.

PAGE 56

41 50 o Adult S I 40 H c S O 30 Q. E (0 W 20 8 10 IIiIIHu.. 0 60 120 180 240 300 360 420 480 540 600 660 720 780 Day From Attachement 50 0> 2 40 c a) O a) 30 CO 20 § io H CD 0 60 120 180 240 300 360 420 480 540 600 660 720 780 Day From Attachement Figure 3-1. Percentage of radio-transmittered adult and juvenile Snail Kites that were censored in each 60-day time interval from the time of attachment.

PAGE 57

42 a E c5? 50 Adults 4020 h_ 30 H o CD O) CD •4— c a) Q_ Censored Confirmed Dead 1992 1993 1994 50 _CD D. E cc C/5 O CD O) cd c CD O }K 10CD D_ Juveniles 40 30 20 I Censored Confirmed Dead 1994 Figure 3-2. Percentage of adult and juvenile Snail Kites from each sampling cohort (i.e., the year that they fledged or were captured) that died or were censored during the first 1 80 d after radio attachment each year.

PAGE 58

CHAPTER 4 WITHINYEAR SURVIVAL PATTERNS OF SNAIL KITES IN FLORIDA A common generalization, that recent evidence supports, is that juvenile survival of many avian species tends to be lower than that of adults (e.g., Loery et al 1987, Nichols et al. 1992). This difference may be attributable to a lack of experience of younger birds for foraging, dispersal, and avoidance of predators. There is considerably less evidence for the time at which the survival rate of younger birds becomes similar to that of adults (Ricklefs 1973, Loery et al. 1987). In chapter 3 I examined annual survival of Snail Kites ( Rostrhamus sociabilis) in Florida using a combination of radio telemetry and capturerecapture data. They found that annual survival of juveniles (birds of age <1 yr) was lower than that of adults (birds of age >lyr), but that delineation of a subadult age class (age 1-2 yrs) was not supported by my data. This suggests that survival rates between age classes were similar after one year, but does not indicate whether or not it requires less than a year to become similar. Neither capture-recapture nor telemetry data enables any evaluation of the patterns of survival within the sampling intervals. However, the intervals using radio telemetry are usually very short compared to capture-recapture data, even though estimates are often reported on an annual basis. Thus, within-year patterns can provide considerable information regarding the time over which juvenile and adult survival rates converge. These patterns also may provide considerable insights about how the risk 43

PAGE 59

44 of mortality changes over time, which can further our understanding of what factors influence survival. Here I use survivorship functions derived from radio-telemetry data to refine our knowledge of the time at which juvenile survival becomes similar to that of adults and to evaluate the seasonal patterns of risk for each age class. Methods My goal was to annually capture and radio tag 100 snail kites of which 60% were adults and 40% juveniles for three consecutive years from April 1992 through April 1995. My targeted ratio of adults to juveniles was intended to emphasize adult survival because demography of Snail Kites probably is more sensitive to adult rather than juvenile survival (Nichols et al. 1980, Beissinger 1995). To maintain independence of my sample, a maximum of one juvenile per nest was equipped with a radio transmitter. I targeted a 50:50 sex ratio of adults to keep my sample balanced. The proportion of samples from each area was based on an annual survey to approximate the statewide distribution (Bennetts and Kitchens 1997a). Adult kites were captured using a net gun (Mechlin and Shaiffer 1979), which uses 22 caliber blank cartridges to propel a 10-foot triangular nylon net. Juveniles were captured just prior to fledging, at approximately 30-35 d old, without a net gun. Fifteengram radio transmitters were attached to birds with backpack harnesses. Radio-tagged birds were located at approximately 14-day intervals from aircraft or ground searches to determine their location and whether they were alive. All radios were equipped with mortality censors which changed pulse rates if the transmitter had not moved for 6-8 h.

PAGE 60

45 Birds with a transmitter emitting a mortality signal were then located on the ground to verify their fate. I used logistic regression as a preliminary analysis for the influences on survival. I found (Chapter 3) that annual survival differed between adults and juveniles, and among years for juveniles, but not adults. However, here I was interested in the timing of mortality, rather than the magnitude of annual estimates. Thus, from my preliminary analysis, I particularly wanted to determine if there was evidence for a withinand between-year interaction, which would have indicated that the within-year patterns differed from year to year. Based on these results and because I was interested in the timing of mortality, rather than annual differences in magnitude, I pooled my samples from each year for each age class to better illustrate the overall timing of mortality. However, I did not pool age classes because of apparent differences in timing. It should be noted, however, that using an unpooled sample does not alter any of the conclusions reported here. I generated survivorship functions using a staggered entry design (Pollock et al. 1989) with the Kaplan-Meier product limit estimator (Kaplan and Meier 1958). I used an arbitrary starting date of 15 April for survivorship functions. By this time during my first year I had a sample sufficient (n = 16) of adults to allow reasonable estimates of survival. This time also corresponded with the beginning of peak fledging. Thus juveniles were added to my sample as they fledged, rather than some having fledged during the previous calender year. Thus survivorship functions are based on a study year (SY) from 15 April of calender year / to 14 April of calender year t+l. All comparisons among survivorship

PAGE 61

46 curves generated by the Kaplan-Meier estimator for radio telemetry data were made using log-rank tests (Savage 1956, Cox and Oakes 1984) with a modified version of the SAS code (SAS Inc. 1988) reported by White and Garrott (1990). Cox (1972) proposed a nonparametric model for the instantaneous probability of an animal's death, or hazard function. The hazard function is a measure of instantaneous risk of mortality as a function of age or time. I estimated the hazard function ( h\) for discrete one-month intervals as the number of animals dying during the interval divided by the number of animals surviving over that interval (Lee 1980). In contrast to survivorship functions, which were intended to assess if there were differences in survivorship, the hazard function better illustrates how the risk of mortality changes over time. Because some risk of mortality of juveniles may have been more related to their age, than time, I estimated an additional hazard function for juveniles based on age (i.e., time since fledging). I did not do this for adults because, in most cases, I did not know their age, and age was less likely to have been a factor for adults. Results I captured a total of 271 individual Snail Kites and attached 282 radio transmitters. Eleven of the 271 birds were recaptured in a subsequent year and their radios replaced. Of the 282 radios, 165 were placed on adults; of which, 45 were during SY 1992 and 60 each during SYs 1993 and 1994. I attached a total of 1 17 radios on juveniles; of which 37 were during SY 1992 and 40 each during SYs 1993 and 1994. Of the adults, 82 (49.7%) were males and 83 (50.3%) were females.

PAGE 62

47 My preliminary logistic regression model confirmed what had been previously reported (Bennetts and Kitchens 1997a). Based on Akaike's information criterion (AIC)(Akaike 1973, Burnham and Anderson 1992), the most parsimonious model was one that included the effects of age, year, season, an age*year interaction, and an age* season interaction (Bennetts and Kitchens 1997a). The general effects of age and annual differences have been discussed in detail elsewhere (Bennetts and Kitchens 1997a, Chapter 3). Of particular interest here was an age*season interaction, which indicates that the seasonal patterns of survival were not similar between age classes. Also of interest were that a season*year interaction or a three-way interaction of age*season*year were not warranted for these data, which supports my belief that the seasonal pattern for each age class does not differ among years. A log rank test supported the conclusion that survivorship functions differed between adult and juvenile Snail Kites (x 2 =20.76, df=l, P<0.001). A further analysis indicated that these differences were substantial for the first four months of the year (X 2 =33.69, df=l, P<0.001), but not the remaining eight months (x 2 =0.47, dfM, P=0.494). Juvenile survivorship dropped sharply for the first four months, after which it leveled off considerably to a much slower rate of decline until a relatively less dramatic decrease at about eight months (Figure 4-1). In contrast, adult survivorship declined at a relatively slow rate for approximately the first eight months, after which there was a relatively moderate decrease in survival similar to that of juveniles at this time. Thus, survivorship functions of the two age classes became quite similar after the first four months.

PAGE 63

48 It was also apparent from hazard functions that the risk of mortality was not constant over time for either age class (Figure 4-2). Juveniles had the highest risk of mortality during the first four months; whereas this, was a period of low risk for adults. Both age classes then experienced increased risk of mortality during winter and early spring. The hazard function for juveniles based on age indicated that the highest risk of mortality occurred between 30 and 60 days post fledging, although it was nearly as high for the first 30 days (Figure 4-3). Discussion My data indicate striking differences in seasonal patterns of survivorship between adult and juvenile Snail Kites. Juveniles were at greatest risk of mortality during late spring and early summer; adults were at greater risk during winter and early spring. The period of high risk for juvenile Snail Kites during late spring is not surprising. My estimated hazard function based on age, rather than time, revealed that the period of greatest risk is between 30 and 60 days after fledging. Although juveniles are least experienced during the first 30 days, they also are still attended by their parents. Snyder et al. (1989) suggested that the post-fledging dependency period lasts for about six weeks. Hence, during the period between 30 and 60 days after fledging, juveniles are becoming independent of their parents, are foraging on their own, and dispersing into unfamiliar areas (Bennetts and Kitchens 1997a). Juveniles that survived the first few months post fledging appeared to be most vulnerable at the same time as peak mortality for adults. In contrast to juveniles, it was less clear why adult mortality was highest in winter, although I offer several hypotheses. During late winter and early spring there is a

PAGE 64

49 potential for increased risk of predation, which probably was the most frequent cause of adult mortality (Bennetts and Kitchens 1997a, Chapter 5). Late winter and early spring corresponds to the time that adults begin courtship. Consequently, at this time they are exhibiting behaviors more conspicuous to potential predators such as courtship flights and vocalizations. Adults also may be less wary of predators if their attention is on procurement of a mate. Evidence also suggests that Great-horned Owls ( Bubo vireinianusV which forage at night, were the most common predators. During winter, leaves are absent from willows, which is the most commonly used species for communal roosting by Snail Kites (Sykes et al. 1995, Darby et al. 1996). Thus, concentrations of roosting kites may have been more visible to nocturnal predators during this period. Apple snails (Pomacea pahjdosa) are the almost exclusive food of Snail Kites. These snails are aquatic, and are most vulnerable to kites while at the water surface. Apple snails have both gills and lungs and the frequency at which they come to the surface for air is inversely related to the amount of dissolved oxygen in the water (McClary 1964). Colder temperatures result in higher levels of dissolved oxygen and reduced activity and oxygen consumption by snails (Freiburg and Hazelwood 1977, Harming 1978). Consequently, cold temperatures during winter result in fewer foraging bouts and lower capture success by Snail Kites (Cary 1985, Sykes et al. 1995). Communal roosting during winter also could result in a greater potential for the transmission of infectious disease. Although the extent of disease as a source of mortality of Snail Kites is unknown, I did find at least one dead bird that was diagnosed with an infection of the coelomic cavity (D.J. Forrester and M.G. Spalding, Laboratory of Wildlife

PAGE 65

50 Disease Research, University of Florida, unpubl. data, Bennetts and Kitchens 1997a, Chapter 5). My data indicate that survivorship of juvenile Snail Kites becomes similar to that of adults after about four months. Estimated survivorship and hazard functions based on time, and hazard functions of juveniles based on age all indicated that the period of greatest risk for juveniles was during their first four months, after which survivorship became remarkably similar to that of adults. However, I do emphasize that my study was conducted during a period of favorable environmental conditions. During periods of food shortage (e.g., widespread droughts), it would not be surprising for there to be a greater disparity between age classes or an increased time for the risk of juvenile mortality to become similar to that of adults.

PAGE 66

51 CO 1.0 0.8 0.6 0.4 — Adults Juveniles i 1 i 15 APR 15 JUN l l 15 AUG I I I I I l 15 OCT Date 15 DEC 15 FEB 14 APR 1 f CO Adults Juveniles i 1 i 15 APR I I 1 I 15 JUN I 1 I I 15 AUG I 1 I 1 I 15 OCT Date i 1 i 1 i 15 DEC I 1 I I 15 FEB I I 14 APR Figure 4-1. Survivorship functions of adult and juvenile Snail Kites from a pooled sample of 3 years (top). Because I was interested in temporal pattern rather than magnitude, I aligned the functions without regard to magnitude, to illustrate at what point in time the functions become similar (bottom).

PAGE 67

52 Figure 4-2. Estimated hazard functions constructed at monthly intervals for adult and juvenile Snail Kites.

PAGE 68

53 Figure 4-3. Estimated hazard function for juveniles based on age, rather than time, function was constructed at monthly intervals starting at the time of fledging. This

PAGE 69

CHAPTER 5 CAUSES OF MORTALITY OF POST-FLEDGING JUVENILE AND ADULT SNAIL KITES IN FLORIDA Previous demographic studies of Snail Kites (Rostrhamus sociabilis ) have focused primarily on reproduction, though survival may be the most important demographic parameter for this species (Nichols et al. 1980, Beissinger 1995, Sykes et al. 1995). There have been several reports of causes of death of nestling Snail Kites (e.g., Sykes 1987, Bennetts et al. 1988), but such information has been largely speculative for post-fledging juveniles and adults (Beissinger 1986, Sykes et al. 1995). The purpose of this paper is to provide an indication of the relative frequencies of different causes of mortality of postfledging juvenile and adult kites during my period of study. My study was conducted between April 1992 and April 1995 in central and south Florida as part of a larger study of survival and movements of Snail Kites in Florida (Bennetts and Kitchens 1997). I attached 282 radio transmitters on 271 individual Snail Kites, with 1 1 birds having been recaptured in a subsequent year and their radios replaced. I monitored birds primarily by aircraft, although verification and retrieval of dead birds was conducted by airboat or on foot. The average interval between consecutive locations of birds was 13.5 (7.9 SD) days (Bennetts and Kitchens 1997). All radio transmitters were equipped with a mortality switch that, upon prolonged lack of motion (~6 h), altered the pulse rate enabling us to remotely determine if a bird was dead 54

PAGE 70

55 or had dropped its radio. I attempted to find all birds emitting a mortality signal the same day that the signal was detected, although logistic constraints sometimes precluded attaining this goal. I found 47 dead or moribund Snail Kites, of which 3 1 were post-fledging juveniles and 16 were adults. Of these 47, 1 assigned a likely cause of death to 24 (51%). Most (81%) dead birds were found using radio telemetry, although I have included birds that were found more incidentally during this study (19%). Carcasses (n=7) in which the fleshy parts had not been consumed or that did not exhibit extreme autolysis were sent for necropsy either to the Laboratory of Wildlife Disease Research, College of Veterinary Medicine at the University of Florida, or the U.S. Geological Survey, National Wildlife Health Research Center. I emphasize from the outset that the exact cause of death can seldom be determined with certainty without finding each carcass while fresh and conducting a necropsy. Even when a necropsy was performed, a conclusive diagnosis was seldom possible and several contributory factors were often confounded. Despite my effort to minimize the time until a necropsy could be performed, severe autolysis was common due to the time interval between death and detection of the mortality signal. Thus, my intention here is to provide a crude indication of the relative frequencies of different causes of death, rather than a definitive assessment. Beissinger (1986, 1988) suggested that adult mortality due to predation is probably rare. In contrast, I found predation to be the most frequently suspected cause of death, when it could be identified (Table 5-1). However, birds in which I assigned predation to be the probable cause were limited to those in which I had ancillary

PAGE 71

56 information, in addition to the carcass having been eaten. Otherwise, the death was classified as unknown. Four carcasses were found at sites (e.g., a feeding or plucking perch) with carcasses of other species. Six others had been plucked, a behavior not associated with any local scavengers. The predator most frequently suspected was the Great Horned Owl (Bubo virpinianus ). In four cases (not including those listed above) adults had been found decapitated on their nest, which is a common signature of Greathorned Owl predation (Nesbit 1975). In several other cases, a Great-horned Owl had been seen frequenting the area. Barred Owls ( Strix varia ) also have been previously suspected to have killed at least one adult female based on feathers left at the nest (Sykes et al. 1995). Peregrine Falcons (Falco peregrinus ) also are occasionally observed in the area during migration, but have not been reported to take Snail Kites. Starvation was the second most frequently suspected cause of death for juveniles, although only two cases were diagnosed (N. Thomas, National Wildlife Health Center, unpubl. data). In the two cases where starvation was assigned to be the probable cause of death, each bird was found alive, but in a severely weakened state from which they did not recover. Also in each case the birds were found in marine environments where apple snails, the primary food of kites, were completely lacking. These deaths were attributed to inexperienced birds that dispersed in a direction where they were unable to find sufficient food. One band return on a previous study (Bennetts et al. 1988) also was a juvenile found dead in a marine environment (Sanibel Island, Florida). One additional juvenile was severely emaciated at the time of death, but the diagnosis was not conclusive and the bird may also have had an intestinal disease (N. Thomas, National Wildlife Health Center,

PAGE 72

57 unpubl. data). In this case, emaciation may have been a symptom of illness, rather than a cause of death. Starvation also may have been underestimated in my sample. For example, juvenile birds dispersing to habitats atypical of adults may have had a lower probability of detection due to less intensive searches of these habitats. These areas also may be more likely to have less predictable food resources. In addition, starvation may be a more frequent cause of death for both age classes during drought years, when food may be scarce (Beissinger 1986); my data were collected only during non-drought years. Other causes of death included vehicle collisions, disease, and one probable gunshot. Vehicle collisions were observed for both age classes and occurred where birds had been observed foraging or nesting adjacent to roadways. One adult female probably died of an infection of the coelomic cavity (D.J. Forrester and M.G. Spalding, Laboratory of Wildlife Disease Research, University of Florida, unpubl. data). The skeletal remains of one juvenile had a probable gunshot (shotgun) hole through its sternum, but I was unable to confirm conclusively if this was the cause of death. It has been suggested that most adult mortality of Snail Kites in Florida occurs during droughts and is likely caused by starvation or risks encountered during dispersal (Beissinger 1986, 1995, Snyder et al. 1989). This inference was derived primarily from changes in the number of kites counted during an annual survey, rather than from empirical evidence of actual mortality. Although I agree with these authors that an increased risk of mortality during widespread droughts is likely, the actual extent of mortality attributed to specific causes is not known (Bennetts et al. 1994, Sykes et al.

PAGE 73

58 1995, Bennetts and Kitchens 1997). However, I emphasize that my study was conducted during non-drought conditions and my inferences are limited accordingly. My inferences also were likely influenced by my methods. For example, two of three birds I found that had been hit by vehicles were found without using radio transmitters. This probably is because they were along roadways where detection of the birds was likely. Finding birds that died of other causes would have been far less likely without the use of radio transmitters. It is also imperative that, if a study goal is to determine the causes of mortality, sampling intervals of radio-tagged birds be frequent. Even with my relatively intensive sampling effort, severe autolysis precluded much of the information that could have been derived had the carcasses been found fresh. In Florida's subtropical environment, decomposition occurs quickly. Consequently, causes of death that required examination of soft tissues for diagnosis are likely to have been underestimated in my sample.

PAGE 74

Table 5-1 Probable causes of mortality of Snail Kites recovered in Florida from 19921995. Juveniles Adults Total Probable Cause Ma 1NU. /o No. % No. % Predation 10 19 9 7 43.8 17 36.2 Starvation 7 6 5 0 0.0 2 4.3 Disease 0 0.0 1 6.3 1 2.1 Vehicle Collision 1 3.2 2 12.5 3 6.4 Gun Shot* 1 3.2 0 0.0 1 2.1 Unknown (undisturbed) b 4 12.9 3 18.8 7 14.9 Unknown 0 13 41.9 3 18.8 16 34.0 Total 31 100.0 16 100.0 47 100.0 a Bird banded by Jon Buntz (Florida Game and Fresh Water Fish Commission). b These birds were found with carcasses intact, indicating that predation was not a likely cause of death. One juvenile was determined to be in excellent nutritional health, indicating that emaciation was also not a likely cause. Another juvenile was determine to be severely emaciated, which could have been either a symptom of illness or a cause of death. c These birds were too severely decomposed and the carcass had been potentially disturbed, such that there was no evidence at all as to the potential cause of death.

PAGE 75

CHAPTER 6 REPRODUCTION Because the focus of this dissertation is on demography, I have concentrated my attention in this chapter on estimates of reproductive parameters. There are numerous papers on other aspects of reproduction (e.g., behavioral ecology) which I have given less attention but encourage interested readers to seek out (e.g., reviewed by Beissinger 1988, Sykesetal. 1995). Reproduction of Snail Kites (Rostrhamus sociabilis ) has been well studied; although significant gaps remain in our knowledge. There also exist several areas of disagreement among researchers regarding interpretation of existing data and literature. Here, I present a combination of original data and a synthesis of the existing literature on reproduction. I have attempted to explicitly point out any areas where disagreement among researchers exists, and to provide detailed explanations for my interpretations. From a demographic perspective, what is ultimately of interest is the mean fecundity rate of females in each age class (Caughley 1977); that is, the mean number of young (or sometimes just the mean number of female young female) produced per female of each age class in the population. Unfortunately, for many species, including Snail Kites, I cannot estimate this parameter directly. Rather, it is derived from the proportion of birds attempting to breed (a^, the proportion of breeding attempts that are successful (S,), and 60

PAGE 76

61 the number of breeding attempts per year (fi). For successful nesting attempts, I also need to know the number of young produced (Y^) and the sex ratio of the young produced (Ity (Brown 1974, Caughley 1977)(Figure 6-1). In this chapter I review the information that has been previously reported on each of these parameters, as well as present estimates based on new data. Semantics Misunderstandings about measures of reproduction can frequently be attributable to a lack of clear definitions of what is being measured and/or to what is appropriate to be measured. I will address the latter type of misunderstanding in my discussions below; however, to avoid the former type, I will begin my assessment of reproduction by providing operational definitions for terms discussed below. Breeding Attempt There has been considerable disagreement among researchers regarding what constitutes a breeding attempt. For the purposes of this chapter I consider a breeding attempt to begin with the laying of the first egg Steenhof (1987). Snyder et al. (1989a) considered a breeding attempt to begin with nest building, prior to the laying of the first egg. They suggested that to ignore the period before egg laying in analyses of reproductive success would be ill-advised because of the high proportion (>0.33) of nests they observed that failed prior to egg laying. I agree entirely with Snyder et al. (1989a) that, for many questions, the failure of nests prior to egg laying may have important biological

PAGE 77

62 implications. These failures may provide insight as to environmental conditions at the time of egg laying and also may provide information regarding behavioral aspects of mate choice. However, I disagree that nests during the nest-building stage for this species should be considered as a nesting attempt for estimation of reproductive parameters. I have several reasons for this conclusion. First, inclusion of "pre-laying" failures may include nests in which a pair bond has not even been established between a male and female. Nest building is initiated by the male as part of courtship (Beissinger 1988, Bennetts et al. 1988) and more than one male may direct courtship toward a single female (Beissinger 1987, pers. obs). Thus, if two males initiated nest building as part of the courtship toward a single female, this would be considered as two nesting attempts using the definition of Snyder et al. (1989a), even though only one of these nests may produce young. Similarly, my observations indicate that a single male may exhibit courtship behavior, including nest building, towards several females in succession. This behavior may last as little as a few hours or may last several days and may then be redirected to a new female if a pair bond is not established. I observed a single radio-tagged male direct courtship to as many as five different females before a pair bond was established that resulted in egg laying. Using the definition of Snyder et al. (1989a), each of these courtship attempts would have been interpreted as a failed breeding attempt. In contrast, I view this as part of the mate selection (courtship) process rather than as a demographic parameter. Second, the passage of cold fronts and corresponding temperature change often results in reduced food availability (Cary 1985). Consequently, courtship is often terminated with the passage of cold fronts and resumed

PAGE 78

63 (often at a new location) when temperatures return to pre-front conditions (Beissinger 1988, Bennetts et al. 1988). Thus, if two cold fronts passed before eggs were actually laid, the pair would have been considered to have made three separate breeding attempts (with two failures) even if the pair successfully raised a brood. For demographic purposes, I view these postponements as courtship interruptions, rather than multiple breeding attempts with each interruption being considered as a breeding failure. Third, because nest building begins with the placement of the first stick and many more courtship nests are probably initiated than are ever detected, it creates a substantial bias in the estimate of success if these early starts are not detected (Mayfield 1961, Miller and Johnson 1978, Johnson 1979, Hensler and Nichols 1981). Finally, it is well known that nesting raptors tend to be considerably more sensitive to disturbance early in the nesting cycle (Grier and Fyfe 1987, Steenhof 1987). Although previous investigators have reported a high proportion of nest abandonment by Snail Kites prior to egg laying (e.g., Beissinger 1986, Snyder et al. 1989a), I have seen no accounting for how much of this abandonment might have been attributable to disturbance by the investigators themselves. In contrast, abandonment of eggs or young by Snail Kites is extremely rare (Bennetts et al. 1994, Sykes et al. 1995). Thus, measuring nesting success after the first egg has been laid can reduce this potential source of confounding and minimize disturbance to this endangered species. Based on these concerns, I defined a breeding attempt to begin with the laying of the first egg. Thus, unless otherwise stated, references to nests in this chapter implies the presence of eggs or young.

PAGE 79

64 Successful Nest For the purposes of this chapter a successful nest is one in which at least one young reaches fledging age (Steenhof 1987). Because birds after fledging may or may not be present at the nest, I defined fledging age as 80% of the average age of first flight (Steenhof and Kochert 1982). Snail Kites are capable of first flight at approximately 30 days of age (Chandler and Anderson 1974, Beissinger 1988, Bennetts et al. 1988); thus, I considered a nest as having been successful if it produced young that survived to at least 24 d (Bennetts et al. 1988). At this age, animals are reasonably assured of still being at the nest and mortality for most raptors between this time and fledging is minimal (Milsap 1981, Steenhof 1987). In addition, I banded birds at the time they were determined to be of fledging age. Consequently, any mortality that occurred after this age would have been included in my estimates of juvenile survival from my mark-resighting program. The Breeding Season The initiation of nests (i.e., egg laying) has been documented in all months of the year (Sykes 1987c); although, for any given year, Snyder et al. (1989a) observed a maximum breeding season (interval over which nests were initiated) of 3 1 .7 weeks (7.9 months) during an 18-year study in Florida. Although Snail Kites in Florida can potentially lay eggs in all months of the year, there is a very distinct seasonal distribution of nest initiations (Table 6-1) (Figure 6-2). Nest initiations begin as early as November, but in most years widespread initiations usually do not begin until January or February.

PAGE 80

65 Peak initiations usually occur in March, but are often several weeks later, peaking in April, in the northern habitats (Toland 1994). The Breeding Population Age of First Reproduction Sykes (1979) reported that Snail Kites are capable of breeding at 3 years of age. However, Sykes (1979) suggested that some birds possibly breed at a younger age. Beissinger (1986) later reported both male and female birds breeding at one year of age and Snyder et al. (1989a) reported one female breeding at nine months. My data are consistent with Beissinger (1986) and Snyder et al. (1989a). During this study, I commonly observed yearling Snail Kites attempting to breed. Proportion of Birds Attempting to Breed Nichols et al. (1980) suggested that the proportion of birds that attempted to breed during favorable conditions was quite high. They suggested that there was no reason to suspect that it was not 1.0 and, consequently, assumed that value for their demographic model. They reported, however, that this was a crude estimate for lack of a better one. Beissinger (1995) similarly reported that the proportion of adult Snail Kites attempting to breed during high-water years was 1.0, but also provided no empirical evidence. Although my data for this parameter are very limited, they are consistent with these earlier estimates. During 1995, 1 closely monitored 23 radio-transmittered adult Snail Kites for breeding activity in order to assess the proportion attempting to breed and the number of breeding attempts per year. Of these 23 adults, 14 were females and 9 were males. During the 1995 nesting season, I located each bird on the ground approximately bi-weekly to determine its

PAGE 81

66 breeding status (e.g., a nest, courtship, not breeding). Birds in which no breeding activity was detected were generally observed for *2 hrs and subsequent visits, usually within 10 days, were required to confirm a non-breeding status and to confirm any nests for birds exhibiting courtship. During 1995 (a relatively high water year throughout the kite's range), all 23 (100%) adults attempted to breed at least once. My estimate is based on a relatively small sample (N=23) and on only one year; however, it does provide an empirical basis that most, if not all, adults may attempt to breed in some years. Sykes (1979) reported that he observed no nesting attempts during 1971 (a widespread severe drought). Based on this observation, Nichols et al. (1980) assumed that no birds nested during 1971 for their demographic modeling effort. Beissinger (1986) reported that during the 1981 drought 80-90% of the kites did not attempt to nest, and Beissinger (1995) later reported that only 15% of adult Snail Kites attempt to breed during drought years. However, no empirical evidence was presented in support of these estimates. Based on anecdotal evidence, I believe that the proportion of birds attempting to breed during drought years may be highly variable depending on the spatial extent of the drought. I agree that during a severe widespread drought, most birds probably do not attempt to breed. However, in cases of more localized droughts, where portions of the kite's range may not be experiencing dry conditions, the proportion of birds attempting to breed may remain very high. For example, during 1991, the Everglades region was at the end of a 2-3 year drought (whether it was a 2 or 3 year drought depends on how a drought is defined). During this year almost no nesting activity was observed in the Everglades region (J. A. Rodgers Jr., pers. comm.). This would appear consistent that a

PAGE 82

67 small proportion of birds had attempted to breed. However, during this year, record numbers of birds were breeding on Lake Tohopekaliga (J.A. Rodgers and J. Buntz, pers. comm.) and in the upper St. Johns Marshes (B. Toland, pers. comm.), areas not influenced by the drought conditions in the Everglades. My data on movement strongly suggest that the Florida population is one population that moves frequently throughout its range, rather than a meta-population of quasi-isolated subpopulations. Thus, in years where drought is not widespread, birds may merely shift the location of nesting activities. Consequently, I suggest that this parameter may be quite variable and needs to take into account the severity and spatial extent of a given drought. Sykes (1979) observed relatively few (n=6) nests during 1972, the lag year following the 1971 drought. The average number of nests per year that Sykes (1979) reported from 1968-1976, excluding 1971, was 23. Based on this observation of reduced nesting during this lag year, Nichols et al. (1980) assumed a proportion of 0.5 adults attempted to nest during 1972. Beissinger (1995) reported that a proportion of 0.8 adults attempt to breed during lag years, although I could find no empirical support for this estimate in any of the sources cited. I suspect that, similarly to drought years, this parameter may be highly variable depending on the specific drought. Thus, I view this parameter as also being unknown and subject to high variability. Snail Kites have been reported to breed as young as 9 months old (Snyder et al. 1989a); thus, by a calendar-year definition Snail Kites are capable of breeding as juveniles (i.e., < 1 year old). However, these cases are ones in which the birds attempted to breed during the nesting season following the nesting season of their hatch year. Thus, this

PAGE 83

68 parameter should be defined as the proportion of birds attempting to breed during their second breeding season (the first is the one in which they hatched). Based on data from Snyder et al. (1989a), Beissinger (1995) reported that 25% of subadults attempt to breed during high water conditions. Snyder et al. (1989a) observed 8 banded subadults breeding during 1979 out of a minimum of 74 that had survived from their hatching year of 1978. Because Snyder et al. (1989a) only checked 50.8% of the nests for bands, they estimated that there were probably 16 subadult breeders out of a minimum of 74 banded subadults (22%). Of course, this estimate assumes that only the 74 subadults observed alive in 1979 had survived and that there was an equal probability of detecting a banded subadult that was breeding in the sample of nests that were checked and those that were not checked. During 1992, 1 estimated a similar percentage of 17% of the subadult birds attempting to breed (Bennetts and Kitchens 1992). My estimate was based on only 2 breeding birds of 12 banded yearlings that I observed during the 1992 breeding season. Consequently, my estimate requires similar assumptions that I suggested above for Snyder et al. (1989a). During 1995 (a high water year throughout the kite's range), I also closely monitored 9 radio-transmittered juvenile Snail Kites for breeding activity (as described above). Of these 9 birds 3 (33%) attempted to breed. All of the estimates derived from the data of Snyder et al. (1989a), as well as from my own data, are very limited (i.e., small samples each from one year); however, they do consistently suggest that a relatively small proportion of subadults do attempt to breed during some years.

PAGE 84

69 Beissinger (1995) also reported that the proportion of subadults attempting to breed during drought years and lag years was 0. 15. I could find no empirical basis for this estimate in any of the sources cited; but I agree with Beissinger that the average percentage would probably be lower when conditions are poor in part or all of their range. Nest Success Nest success has been among the most widely estimated parameters of reproduction of Snail Kites. However, it has probably also been among the most confusing. There are several areas of disagreement among researchers regarding estimation of nest success. The disagreements center primarily on which nests should be included in the sample and what estimator should be used. Consequently, nest success has been difficult to compare because different researchers have used different estimators and have included or excluded different categories of nests within their respective data sets. I have attempted to summarize below the major issues of contention. I have also summarized the literature on nest success and explicitly pointed out which estimator was used and what categories of nests were included or excluded in the sample. Thus, readers can make comparisons among studies and decide for themselves which estimates are most appropriate for their particular needs.

PAGE 85

70 Areas of Disagreement Regarding Estimation of Nest Success Inclusion or exclusion of nests found at different stages At what stage a given nest is found can greatly influence its probability of success. Nests found late in the nesting cycle have a higher probability of success because they have less observation time during which they are at risk. A Snail Kite nest requires at least 57 days to fledge young (27 days of incubation and 30 days for nestlings to reach fledging age). Thus, a nest found during egg laying will have potentially >50 days "at risk" (provided it does not fail earlier) to be considered successful. In contrast, a nest found close to the time of fledging may have only a few days "at risk" to be considered successful. Consequently, estimates of nest success that were derived using nests found late in the nesting cycle tend to be biased high (Mayfield 1961, 1975, Miller and Johnson 1978, Hensler and Nichols 1981, Hensler 1985). Nests at different stages also are vulnerable to different risks. For example, rat snakes (Elaphe obsolete) are believed to be one of the major predators of Snail Kite nests (Bennetts and Caton 1988). Rat snakes will readily take eggs or young that are less than one week old; however, the larger size of older nestlings largely precludes predation by rat snakes. Consequently, nests found when young are >1 week have an inherently lower risk of predation by rat snakes. Some researchers (e.g., Beissinger 1986, Snyder et al. 1989a) also have included nests prior to eggs having been laid (i.e., during nest building) in deriving estimates of success. I disagree with this practice for the reasons previously discussed.

PAGE 86

71 Because of these biases, estimates of nest success can be substantially influenced by what nests (i.e., found at what stage) are included or excluded for deriving a given estimate. This makes comparison of previous estimates of nest success for Snail Kites difficult because researchers have not used the same criteria for inclusion or exclusion of nests found at different stages when deriving their estimates. Steenhof and Kochert (1982) suggested three ways to minimize this type of sampling error for estimating nest success. First, they suggest estimating success based on a pre-determined sample of territorial pairs. However, because Snail Kites do not maintain nesting territories from one year to the next, this solution is not feasible for this species. Secondly, they suggested using estimates derived only from nests that were found during incubation (by definition, they considered a breeding attempt to have begun after they laying of the first egg). This suggestion is feasible for kites; but of the previously reported estimates, only Snyder et al. (1989a) reported estimates using this criterion. Their third suggestion was to use the Mayfield Estimator, which is intended to account for the bias imposed by not finding all nests during early stages. Of the previously reported estimates, only Bennetts et al. (1988) reported estimates using this estimator. Given the differences in what nests were included or excluded in previous studies, I urge caution in making comparisons among previous studies. I also agree with Steenhof and Kochert (1982) that estimates of success should be derived either using only nests that were found during incubation or using the Mayfield estimator. Of these two approaches I prefer the latter, although there remains disagreement among researchers regarding this conclusion.

PAGE 87

72 Manipulated nests Nests that occur in cattails may have a tendency to collapse under conditions of high winds or waves (Sykes and Chandler 1974). This led to a previous practice of placing nests that were subject to this type of failure in artificial nest baskets (Chandler and Andeson 1974, Sykes and Chandler 1974). Because this may influence the outcome of a given nest, whether to include or exclude these nests has been the subject of some debate (e.g., Beissinger 1986, Snyder et al. 1989a). Similarly, when these nests have been included in samples from which estimates of nest success were derived, there have been differences among researchers (e.g., Sykes 1979, Snyder et al. 1989a) as to how these nests were treated in the derivation of nest success. Sykes (1979, 1987b) included 43 nests that were placed in artificial nest baskets in his sample for estimating success. These nests were not treated differently than other nests. Snyder et al. (1989a) later criticized this use of manipulated nests. They suggested that the success of manipulated nests was higher than if they had not been manipulated, and that this would have biased Sykes's estimate of success upward. Snyder et al. also presented estimates of nest success using 94 manipulated nests. They argued that because these nests were in imminent danger of collapse, they considered them all as failures. They suggested that to exclude them, as was done by Beissinger (1986) and Beissinger and Snyder (1987), would have also biased success upward because these manipulated nests were not a random sample with regard to their probability of success (i.e., that they would have failed). In contrast to their suggestion, I have observed collapsed several nests containing older (>10 d old) nestlings that have been successful. Although I agree

PAGE 88

73 with Snyder et al. (1989a) that exclusion of these nests probably would have biased success upward, I also believe that including them all as failures probably would have biased their estimate downward. My tendency is to agree with the solution of Snyder et al. (1989a), but to recognize that there might be a slight bias toward underestimation of success. An additional concern that has not been addressed by previous authors is that the susceptibility of nests to collapse may be influenced by the investigators themselves. The vulnerability of nests to collapse can be greatly influenced by the paths of airboats while conducting nests visits, particularly in cattails (Bennetts 1996). Airboat trails are often wide enough to allow increased susceptibility to wind damage and/or to weaken the structural support provided by the cattails adjacent to the nest. This type of damage can be minimized, if not eliminated, by maintaining a substantial distance from the nest during an approach and either wading in to nests or using a mirror pole from a distance to check them (Bennetts 1996). Nest baskets have not been used in recent years and I do not anticipate (or advocate) a recurrence of their use. Although some nest collapse still occurs in some areas, particularly on lakes (J. A. Rodgers, pers. comm.), I do not believe that the benefits of nest baskets warrant the effort or disturbance for their use as a general management tool. They may, however, be warranted for isolated special circumstances. Previous use of nest baskets had been initiated when numbers of Snail Kite probably were much lower than are currently found. I do, however, advocate that researchers exercise extreme care to avoid influencing the outcome of nests being monitored.

PAGE 89

74 Mavfield vs conventional estimator Mayfield (1961, 1975) proposed an estimator for nest success that was based on daily exposure (risk) such that a daily probability of success was derived using only those days in which a given nest was under observation. Overall success is then derived by applying the daily success over the length of the interval being estimated. This approach provides an estimate of success that is unbiased with respect to when a given nest was found, but requires an assumption that the probability of success is constant over the period (e.g., incubation) being estimated. Hensler and Nichols (1981) later showed, using Monte Carlo simulations, that this estimator was superior to the conventional estimator under a wide variety of conditions. Bennetts et al. (1988) used the Mayfield estimator for nest success of Snail Kites and found it to perform favorably for this species. They found some violation of the assumption of constancy (e.g., success differed between incubation and nestling stages); however, this assumption can be overcome by using separate estimates for periods that differ (Hensler and Nichols 1981). Snyder et al. (1989a) later argued that the Mayfield estimator was inappropriate for Snail Kites because the interval length for nest building was too variable to apply this estimator. I agree with Snyder et al. (1989a) that the Mayfield estimator would be inappropriate for estimation of success during the nest building stage. However, I also have argued that the nest-building period is inappropriate to include in estimates of nesting success for this species. Consequently, I disagree with Snyder et al. (1989a) that the Mayfield estimator is inappropriate for estimating nesting success of Snail Kites. Rather, I agree with Hensler and Nichols (1981), Miller and

PAGE 90

75 Johnson (1978), Steenhof and Kochert (1982), and Steenhof (1987), that this estimator is preferable to conventional estimates of nesting success because of its ability to produce unbiased estimates of nesting success. Estimates of Nest Success And Its Process Variance Given the wide disagreement among researchers regarding nesting success, I suggest that future researchers be specific about what is being included or excluded, and that consideration be given to reporting success both by conventional and Mayfield estimators so readers have the ability to compare their results. I have also provided a summary of previously reported estimates, showing what nests (i.e., found at what stage) were included in each estimate, whether or not manipulated nests were included, and which estimator was used (Table 6-2). I estimated the mean annual nest success (S„ ) as 0.32 based on reported nest success from each year using estimates that were based on nests, in which at least one egg has been laid, that were found during the egg stage (Table 6-3). However, some years had extremely low sample sizes, which may have precluded a reliable estimate for that year. If I had excluded estimates for those years with <10 nests, I would have estimated mean annual nest success (S s ) as 0.28. It is also important to recognize that there are several distinct variance components associated with demographic parameters (White et al. 1982, Burnham et al. 1987). A demographic parameter (e.g., survival) may vary over time (temporal variation) or among

PAGE 91

76 locations (spatial variation). There is also likely to be heterogeneity among individual (individual variation) in their probability of survival due to genetic or phenotypic variation (DeAngelis and Gross 1982). Each of these sources of variation are a type of population variation (Burnham et al. 1987). There is also variation attributable to sampling populations. Unlike these previous sources of variation, sampling variation is not a measure of population variability, but rather is a measure of sampling error. This latter source of variation is important because it provides a measure of the certainty for a given parameter estimate. However, for demographic modeling, what is important is the actual variability of parameter over time, space, and among individuals (collectively called process variance). For modeling populations, sampling variation is a source of noise and should be removed from the overall variance estimate. Burnham et al. (1987) provided the theoretical framework and formulae for estimating process variance. I used this framework to estimate process variance for nest success based on estimates reported from 1968-1995 using only nests found after the first egg was laid, but before hatching. Based on the data from table 6-3, 1 estimated r) 2 =0.08 and 6=0.28. Influences of Nest Success There are a multitude of factors that could potentially influence the outcome of Snail Kite nests. Factors that have been reported to significantly affect nest success include location (i.e., area)(Snyder et al. 1989a), water levels (Sykes 1987b, Bennetts et al. 1988, Snyder et al. 1989a, Toland 1994), date of initiation (Bennetts et al. 1988), nest substrate (Snyder et al. 1989a, Toland 1994), nest height (Bennetts et al. 1988, Toland 1994),

PAGE 92

77 distance to land (Sykes 1987c), and interspecific coloniality (Snyder et al. 1989a). I used logistic regression to test for the influence of each of these effects, except interspecific coloniality, on a sample of 854 nests using data from Bennetts et al. (1988), Toland (1994, unpubl. data), and this study. My preliminary univariate analysis, which had a liberal rejection criterion of a=0.25 for each effect, indicated that all of these effects warranted retention for further analysis. However, my results indicated that the specific substrate, rather than herbaceous versus woody, was warranted for further consideration. Similarly, my results indicated that a categorical threshold distance to land of less than or greater than 200m (Sykes 1987c) was warranted for further consideration, rather than the actual distance. Although my preliminary univariate analysis supported the retention of these effects, a multivariate analysis with each of the retained effects (but lacking interaction terms) indicated that only year and date of initiation were warranted at more restrictive rejection criteria of a =0.05. My final model indicated an area, but not a year effect, as was indicated by my preliminary analyses (Table 6-4). However, area and year effects were highly confounded in these data because the studies included in this analyses that were conducted during different years were also conducted at different areas. Thus, I do not believe that I can reliably distinguish between these effects. Differences in success among areas and years are not surprising given the many causes of nest failures (Sykes 1987c, Bennetts et al. 1988, Snyder et al. 1989a). My data indicated an effect from the date of initiation in all phases of this analysis; although it was not completely clear as to whether this effect was quadratic or linear. Overall nest success (all years combined) was highest during January

PAGE 93

78 with a generally decreasing trend over time (Figure 6-3). However, the overall trend is somewhat misleading because it was heavily influenced by one year (1987) of exceptionally high success in January (Figure 6-4). Most years had the peak of success in February (3 of 7) or March (2 of 7). In only one year was peak success in January (1987) and one year in April (1993). In only one year did I observe nesting during December (1985), and success was lower than during January, February or March of that year. These temporal effects of success were undoubtedly confounded with year effects because studies conducted from 1991-1993 by Toland (1994, unpubl. data), which were included in this analysis, were conducted in the northern part of the kite's range where the date of initiation was often several weeks later than in the southern portion of their range. In contrast, most of the data from other years were from the southern portion of the kite's range. This probably also accounts for the interaction effect of year with date of initiation. Number of Young per Successful Nest In contrast to nest success, the number of young per successful nest probably is one of the least variable and has been the least controversial of the reproductive parameters. The relative lack of variability for this parameter is not surprising since it is not unusual for raptors to produce normal numbers of young per successful nest even when other aspects of reproduction (e.g., proportion of population attempting to breed or nest success) are depressed (Brown 1974, Steenhof 1987). For this reason, the number of young per successful nest is not particularly informative in the absence of these other reproductive parameters (Brown 1974).

PAGE 94

79 Several studies have reported estimates for the number of young per successful nest (Table 6-5) and the average from 20 years of reported data is 1.9. Annual estimates reported have ranged from a low of 1.4 (Sykes 1979, 1987b, Bennetts et al. 1988) to a high of 2.5 (Sykes 1979, 1987b)(Table 6-6). The Number of Breeding Attempts per Year For Snail Kites, the success per breeding attempt and the number of young produced per attempt are relatively well known (Sykes 1979, Bennetts et al. 1988, Snyder et al. 1989). In contrast, there has been little evidence for the number of breeding attempts per year. Snail Kites are capable of raising >1 brood per year and attempts at multiple brooding may be fairly widespread (Snyder et al. 1989). Snyder et al. (1989) suggested that individuals have the potential to successfully raise four broods per year, although there have been no documented cases of individuals successfully raising >2 broods in a given year. Snyder et al. (1989) estimated the number of nesting attempts per pair to be 2.7 per pair. Their estimate was derived using the number of Snail Kites counted on an annual survey at two locations (Lake Okeechobee and Water Conservation Area 3 A) during late fall of 1977 as an estimate of the potential breeding population of 1978. They then used the number and success of nests found at those locations the following breeding season to estimate the number of breeding attempts by that breeding population. However, there are several assumptions inherent in their calculation which may have greatly influenced their estimate. Beissinger (1995) later used a more "conservative" estimate of 2.2 attempts per pair in a population viability analysis because

PAGE 95

80 the estimate by Snyder et al. (1989) was reported to be "under the best conditions". Here, 1 estimate the proportion of the population attempting to breed and the number of breeding attempts per year for individual Snail Kites during the 1995 breeding season using radio-telemetry. During 1995, 1 closely monitored the breeding activity of these kites in order to assess the number of breeding attempts per year. I defined a breeding attempt to begin with the laying of the first egg (Steenhof 1987). However, I recorded all activity associated with breeding, including courtship behaviors to enable more comprehensive record of each individual. During the nesting season, I located each bird approximately biweekly by airboat and determined its breeding status (e.g., not breeding, courtship, or breeding). Birds in which no breeding activity was detected were generally observed for z 2 hrs and subsequent visits, usually within 10 days, were required to confirm a nonbreeding status and to confirm any nests for birds exhibiting courtship behavior. I was able to successfully monitor the breeding status of 23 adult Snail Kites for the entire 1995 season. Of these, 14 were females and 9 were males. I was able to monitor the breeding status of an additional 9 subadults. The average interval between successive observations of breeding status was 14.1 d ( 8.1 SD). All adult birds monitored attempted to breed at least once, and I observed an average of 1.4 ( 0.6 sd) breeding attempts per bird. Of the 23 adult birds, 15 (65%) made only one breeding attempt, 7 (30%) made two breeding attempts, and 1 (4%) attempted three times (Table 6-7). I observed only one bird (4%) which successfully raised two broods. In contrast to adults, not all subadults attempted to breed. Of the nine

PAGE 96

81 birds monitored, only 3 (33%) were confirmed to have a nest in which at least one egg was laid, and none were observed attempting to breed more than once. Our data are consistent with reports by Snyder et al. (1989) that >1 breeding attempt by Snail Kites in Florida is common during some years. However, my data are not consistent with previous estimates of 2.7 attempts per year by Snyder et al. (1989) and even the more "conservative" estimate of 2.2 attempts per year used by Beissinger (1995), which was based on Snyder et al.'s estimate. A combination of differences in my estimation procedures, difference in our respective definitions of a breeding attempt, and annual variability of this parameter probably account for these discrepancies between these two data sources. Semantic differences in defining a breeding attempt likely contributed to the disparity between the estimate of Snyder et al. (1989) and the lower estimates of this study. Snyder et al. (1989) considered a breeding attempt to begin with nest building, prior to the laying of the first egg. Thus, their estimate for the number of attempts included courtship by my my definition. In contrast, I agree with Steenhof (1987) and defined a breeding attempt to begin with the laying of the first egg. If this definition were applied to the data reported by Snyder et al. (1989) their estimate would have been reduced from 2.7 to 1.9 breeding attempts per pair The assumptions required by Snyder et al.'s (1989) and this study's estimates can also have a dramatic influence on the resulting estimates. The primary assumption of my estimates was that no breeding attempts went undetected during the breeding season. The interval of my breeding status checks could have resulted in failure to detect an occasional

PAGE 97

82 bird that initiated a nest that failed early during laying or incubation. Consequently, my estimate may have been slightly low for 1995. However, exclusion of birds from my sample in which I had gaps in the known breeding status helped to minimize this potential bias. In addition, most breeding failures occur during the first week after hatching (Bennetts et al. 1988), which occurs after 4-5 weeks of breeding activity, including courtship. Consequently, the potential bias from having missed breeding attempts probably was very low. To further assure that this bias was minimal, I also tried a more restrictive criterion for my sample, such that the average interval between visits was ^8 d, with a maximum of 2 1 d between any 2 visits. This more restrictive criterion reduced my sample size (n=10), but did not alter my estimate of the number of breeding attempts per adult (x=1.38). In addition, 1995 had very favorable water conditions throughout the Snail Kite's range in Florida. For this reason, I might also have expected my 1995 estimate to be higher than an annual average. As pointed out by Snyder et al. (1989), their procedure assumed that the 1977 annual survey was an accurate census (i.e., a complete count of all kites). Recent evidence suggests that this assumption was highly unlikely to have been met. Bennetts and Kitchens (1997a) and Darby et al. (in review) found that during late fall, when the annual survey was conducted, a substantial portion (up to 60%) of the population may be in areas not included in the survey or in habitats (e.g., cypress) where detection is difficult. In addition, I found that the average probability of detecting marked individuals during spring, when birds are more concentrated, was quite low (<25%). In contrast to the procedure used by Snyder et al. (1989), my approach required no assumptions about the

PAGE 98

83 annual survey because I used radio transmitters to locate individual birds and I eliminated all birds from my sample whose breeding status was unknown for prolonged periods. Snyder et al. (1989) reported that they also assumed that no birds died between the 1977 annual survey and the end of the 1978 breeding season. However, because they used the 1977 survey as an estimate of the number of potentially-breeding pairs during the 1978 breeding season, their approach actually required a more rigorous assumption that WCA-3 A and Lake Okeechobee represented a closed population. Thus, the assumption is not only that there were no deaths, but also that there were no births, immigration, or emigration. The assumption regarding births does not impose a serious problem because the survey was conducted before the primary breeding season. The assumption of no deaths poses a slightly greater problem since annual mortality of adults is approximately 10%, most of which occurs during the time period between the survey and the following breeding season (Chapter 4). However, the most serious problem probably is the assumption of no immigration or emigration. Data from 271 radio-tagged Snail Kites in Florida indicated that the probability of a bird moving from one wetland to another during a given month is approximately 0.25 (Bennetts and Kitchens 1997a, Chapter 7). Given that the time between the 1977 survey and the end of the 1978 breeding season was approximately 8-9 months, there is a strong likelihood of substantial immigration and emigration. Further, Bennetts and Kitchens (1997a) and Darby et al. (in review) found that there was a substantial shift from peripheral habits, during the time of the survey, to breeding habitats during spring. Snyder and Snyder (1991) also reported that birds left Lake Okeechobee after the breeding season and returned during fall. Thus, there was very

PAGE 99

84 likely a net increase in the breeding population which could have substantially inflated their estimate of the number of breeding attempts per pair. In contrast to the approach used by Snyder et al. (1989) my approach did not require assumptions about closure because my estimates were based on individual radio-tagged birds, regardless of their location. Snyder et al. (1989) also recognized that their approach required an assumption that all birds counted during the 1977 survey were potential breeders during the 1978 breeding season. This assumption also was unlikely to have been met because the annual survey does not distinguish between adult and subadult birds. Previous reports (e.g., Snyder et al. 1989), and my data confirm, that not all subadults are potential breeders. In contrast, my approach did not require this assumption since my estimates were specific to each age class. An additional assumption, required if an estimate is applied to years other than one from which it was derived, is that the estimate be from a "representative" year of the conditions to which the estimate is being applied. Estimates derived from both my data and that of Snyder et al. (1989) were based on a single year. Based either on the number of nests found or the number of young banded, 1978 was an exceptional year for reproduction and Snyder et al. (1989) correctly limited their inference to that year. To extend the inferences of Snyder et al. (1989) to other years would likely result in an inflated estimate. To account for this bias, Beissinger (1995) used what he considered a "conservative" estimate of 2.2 breeding attempts per pair in a population viability analysis, my data suggest that even this "conservative" estimate was likely to be inflated if used as

PAGE 100

85 an annual average, although my estimate also was based on a single year of relatively favorable conditions. Conditional Probability of Attempting to Breed An alternative way to look at the number of attempts per year is using conditional probability for the proportion of birds attempting to breed (a^. That is, a, would be the probability that a bird attempted to breed, given that it had not attempted previously during that nesting season. Of 23 birds I monitored for breeding activity during 1995, all 23 attempted to breed at least once. Thus, my estimate of a, would be 1.0. Similarly, a 2 would be the probability that a bird attempted to breed, given that it had previously made 1 attempt during that breeding season. Based on my data from 1995, 1 would estimate a 2 to be 0.34 (8 of 23). The probability that a bird attempted to breed, given that it had previously made two attempts during that breeding season (a 3 ) was 0.13 (1 of 8). The variance for these estimates could be derived based on a binomial distribution, although the formula traditionally used for this estimate is intended for large samples (White and Garrott 1990). Hollander and Wolfe (1973) provide alternative procedures that could be used for smaller samples. Number of Successful Broods per Year Snyder et al. (1989a) suggested that in some years it was possible for Snail Kites to successfully raise four broods. This was based on the length of the breeding season for certain years (e.g., 1978 and 1979) and the assumption that it would take 10 weeks (70 d)

PAGE 101

86 to raise a brood if mate desertion occurred and 16 weeks (1 12 d) if no mate desertion occurred. Although this is certainly theoretically possible, I believe that the probability of a Snail Kite successfully raising even three broods in a given year is very close to zero. I base my conclusion on several points. First, empirical data do not support Snyder et al.'s (1989a) conclusion. There have been no documented cases of Snail Kites successfully raising > 2 broods in a given year, and the occurrence of successfully raising 2 broods appears quite rare. Out of an 18-year study including 666 nesting attempts, Snyder et al. (1989a) documented only 3 cases of Snail Kites successfully raising 2 broods. Similarly, only 1 of 23 (4%) radio-transmittered birds that I closely monitored for breeding activity during 1995 (a good year), successfully raised two broods. I believe that Snyder et al. (1989a) overlooked some critical aspects of the breeding biology of Snail Kites when making this suggestion. First, although the inclusive dates from the first nest initiated in a given year to the last may span a period of 6-7 months, the initiation of nests is not evenly distributed throughout that period. The majority of nests (82% of the nests reported by Snyder et al. [1989a]) were initiated during a five month period from January through May. This is sufficient time for only two successful broods even when mate desertion occurs. The longest nesting season (time span over which nest initiations were observed) reported by Snyder et al. (1989a) over an 18-year period was only 31.7 weeks. Given that they suggest that it takes 10-16 weeks per successful brood, the longest nesting season they observed did not even have sufficient time to successfully raise four broods (even with mate desertion for all broods), and barely had sufficient time to successfully raise two broods without mate desertion. There also has been no consideration given to energetic

PAGE 102

87 costs of raising successive broods. Thus, I believe that a small percentage of birds (e.g., < 10%) may successfully raise two broods during some years; however, there is currently no empirical evidence to support the conclusion that Snail Kites successfully raise > 2 broods per year.

PAGE 103

88 Table 6-1 The number of nest initiations reported in each month during studies from 1966 through 1995. Sykes Snyder et al. Snyder et al. Bennetts et Toland This Proportion OCT 5 0 0 0 0 5 0.00 NOV 9 9 8 0 0 17 0.01 DEC 5 35 35 0 7 47 0.04 JAN 26 90 81 14 3 74 184 0.16 FEB 35 98 78 102 22 66 201 0.17 MAR 36 147 119 125 49 106 310 0.26 APR 21 114 103 102 53 40 217 0.18 MAY 10 64 56 32 19 17 102 0.09 JUN 5 44 38 0 16 0 59 0.05 JUL 1 27 25 0 5 0 31 0.03 AUG 2 1 1 0 0 3 0.00 SEP 1 0 0 0 0 1 0.00 Total 156 629 544 375 167 310 1177 a Includes all years reported by Snyder et al. (1989a). b Includes only years reported by Snyder et al. (1989a) to have wide seasonal coverage (1970-1982). c Unpublished data from nests reported by Bennetts et al. (1988). Seasonal coverage was limited to January through July. d Unpublished data courtesy of B R. Toland from nests reported by Toland (1994). c Based on data from 1994 and 1995. f Based only on data with wide seasonal coverage (Sykes 1987c, Snyder et al. 1989a [1970-1982 only], Toland 1994, and this study.

PAGE 104

oo f) oj CD GO E <*•3 00 rCD o O H £: § rV 1 o OO O T3 C 3 u od ca 5 E on .5 — -*-* J J -o .S 11 8.1 3 u a> O E X) E o C/3 .3 GO __ 0 o o. to ai J II (jf ** § C w S £ ** 3 53 S 4> j22 *> tN 3 1 O c/j c CO — 1 H 3 CD tCD u cu I o c = c IX c 3' 8 2 T3 4) 3 — go U CO £ PQ c o 1 o o -J go 5 0 ON On CO CD on go 0 on CD go CU VO IO tN oo On CO CD M >> c/5 co CD n cu O O Z Z GO SO tN oo ON cu s M go '53 CQ o Z / — > / — — i_ 00 00 00 0? 00 00 00 00 On ON On ON S — s — s — 15 15 ts % go oo on 2 t-> 2 2 cu u u g I cq PQ PQ GO CD o o >Z Z OS ON 00 00 ON ON tt "3 "3 kg T3 C/3 kg i ON 00 on CU kg W) IO 3 4> 4) 4) >H S O O £ z z on oo O z o o z z >z o o z z o o z z o S Z Z O 1> oo 4> O f> NO NO m oo t~tN NO tN so* NO tN so" kW rNO' kff ON ON -a c o H o o Z Z 8 o ^ Z o Z On On no 00 3 oo E— o Z CO (U o Z o Z \ a E -o \ \ % \ \ o U o U o U 1 o u o U o U o U o U o U o U s o tN so" in* tN NO 00 rm oo r 00 00 1^ 00 00 m 00 m 00 00 rn ON in On i oo NO oo NO 1 00 r1 NO 00 1 NO oo 1 NO OO 1 NO 00 1 00 NO 1 00 NO i 00 NO 1 o 0 s 4 ON -85% s5 o x 00 1 3 NO m s o^ NO N o i 0 s 0 s O m sP 0 s O O 0 s o o sP 0 s tr> iri \= 0 s f> NO 1 ro m tN o m tN so tN o i o 1 o 00 NO in s? o x O m o v 00 ON s 0 s m M v? O x ON M s? 0 s ON tN 0 s rn tN sP o x ON tN N 0 s 00 m CJ J* NO u Q oo oo CU M J m cu CU JO o J= o cu CU M O cu _^ 3 < PQ oo U c m oo cu J3 O 00 -iS oo E 3 Z o < 00 oo i„ |3 O O cu z c2 CO CO CU O o 3 3 ? 3 ?t. < U 2 8 3 T3 Q o CO kg _Cl U "~ a. -a S D < U cu u 03 oo C O 'S o o oo CU oo 00 U O 3 5 c o3 D 2 o Q. Oh ,
PAGE 105


PAGE 106

91 Table 6-3. Annual nest success reported during studies from 1968 through 1995. Nest success was based on nests found during the egg stage. No No. ^est Sampling Year Nests Successful Success (%) Var(§)* Source lyoo 1 1 i l.UU A AAAA U.UUUU anyder et ai. (lyoyaj 1 Q70 ly /U 1 i i i nn i .uu n nnnn U.UUUU onyuer ei hi. ^i^oyd^ 1 Q71 ly 1 1 1 077 ly 1 A -i j i i n n n 0741 onyuer ei oi. ^i^o^dj 1 Q7T 1 y 1 j 1 s 1 o A n t ? u.zz U.UU7U onyuer ei y 23 n /lo U.4Z n nn/i i U.UU41 C„,,J of oi / 1 OOOo\ inyder et ai. ^lyoyaj 1 mn iy /y lob /o 4Z n c /i U. 54 U.UUJZ Cvtirsfo** of oi nOQOn\ anyder et ai. ^lyoyaj i aoa iyou 2 U A AA U.UU A AAAA U.UUUU anyder et al. (iyya) iyi <* U n nn U.UU U.UUUU snyder et ai. ^iyyaj 1 OC7 iy&z 1 7^ 1Z i 1 n no U.Uo n nn^/i U.UU04 anyder et ai. (^lyoyaj 1 OB7 lysj 1 0*" IB c J n to U.Zo n ni 1 1 U.U1 1 1 onyder et ai. ^lysyaj 1 OQ/1 iys4 i no c i n o a iyo 1 at 1U/ Zs n o£ U.Zo n nn i o U.UUlo rJennetts et al. (iyo) 1987 210 92 0.44 0.0012 Bennetts et al. (1988) 1988 1989 1990 26 2 0.08 0.0027 Toland(1994) 1991 39 8 0.21 0.0042 Toland (1994) 1992 59 33 0.55 0.0042 Toland (1994) 1993 43 14 0.33 0.0051 Toland (1994) 1994 57 36 0.63 0.0041 This study 1995 176 94 0.53 0.0014 This study a Sampling variance was not reported by the source authors, but was estimated based on a binomial distribution. b Annual success for 1978-1983 was reported separately for lakes and WCA-3A by source authors, but combined here for estimate of annual success.

PAGE 107

92 Table 6-4. Summary statistics from the final most parsimonious (based on AIC and LRTs) logistic regression model for the factors effecting nest success. Source at X X 1 Year b 7 9.80 0.200 Area c 3 10.78 0.013 Date of Initiation (DOI) 1 5.34 0.021 DOI DOI d 1 2.66 0.103 YR*DOI 1 16.63 0.020 1 Chi square was based on a LRT between models with and without the source term. b Although the main effect of year was not significant at
PAGE 108

93 u o u 3 0 on <2 O O 3 00 o — ft H T3 £ 3 c > O Oh d On ON &0 00 ON 05 &0 NO 00 On 0O OO On On 00 ON I s 03 La "TO m no On S 2 2 2 On ON O H 3 dy NO ^— £ /~> r00 00 00 On ON ON ON On o ON ON ON 00 1 00 00 i NO l 00 i O 1 NO NO c OO NO ON On On On On ON On On On */*> rOn CN i ri i CN i i ( i Tt; J 1 NO On o NO -d O 00 CN j CN o CN o CNi O >/"> CN — On CN 00 5 g X) o J= o 1) o u M in c/i 1> O o a> u c ^ oo TO •o "S o a. c o Q. 8.S J3 C § tN 3 o o o > u to c3 E 1 g 3 H o •= o. •— DO T3 o a. c ll

PAGE 109

94 Table 6-6. The number of successful nests, young fledged, and number of young per successful nest reported for each year from 1968 through 1995 v No. Year Successful No. Young No. Young per 1 (A (TA/1 r icuycu Source iyoo 1 1 24 2.2 Sykes(1979, 1987b) iyoy o 5 1 *5 13 1.6 Sykes (1979, 1987b) 1 Q7n iy /u o O 12 1.5 Sykes (1979, 1987b) iy /i V 0 a Sykes (1979, 1987b) iy /z 3 7 2.3 Sykes (1979, 1987b) 1973 12 29 2.4 Sykes (1979, 1987b) 1 074 iy m 0 1 1 1 1 1 s 1 .5 Sykes (1979, 1987b) 1 Q7^ iy o 1 A 35 7 5 Z.J Sykes (1979, 1987b) 77 1 A 1 .4 Sykes (1979, 1987b) 1 077 iy / / O O ZU 7 5 Z.J / 1 A O '7L\ Sykes (1987b) 1 Q70 iy /a 1 1 1 1 ZU 1 O 1.0 Sykes (1987b) 1 Q7Q iy /y ,)4 i nfi b 1U5 7 n z.U T"> • • / i no/"\ Beissinger (1986) i ocn iysu c c c C i no i (J 0 a Beissinger (1986) ly&z z /lb 4 2.0 Beissinger (1986) 1 Qfil 1 r> 1U ZU 7 n. z.U Beissinger (1986) i no a lys4 c c c c 1 no c c c c c i no/: A C 45 65 1 A 1.4 Bennetts et al. (1988) iy / 1 (\A 104 172 1.7 Bennetts et al. (1988) 1 Q8fi iyoo c c c c 1989 c __c c c 1990 2 3 1.5 Toland (1994) 1991 8 17 2.1 Toland (1994) 1992 33 68 2.1 Toland (1994) 1993 14 26 1.9 Toland (1994) 1994 50 80 1.6 This study 1995 94 181 1.9 This study a No successful nests from which to estimate b Not reported, but inferred from number of successful nests and number young per successful nest c No reported information

PAGE 110

Table 6-7. Number of nesting attempts and number of attempts that were successful for each of 23 adult Snail Kites during the 1995 breeding season. Frequency Sex Number of Attempts No. Attempts Successful 1 CO £LC\Q 152.698 F 1 1 152.584 F 1 0 153.496 F 2 0 153.860 F 2 1 152.739 F 3 1 153.931 F 1 1 152.039 F 2 1 153.969 F 2 2 152.494 F 1 1 152.169 F 2 1 153.979 F 2 1 152.777 F 1 1 152. 499 F 1 0 i co 'i/cn 152. Joy F 1 1 1 CO o^cn 152.869 M 1 1 153.900 M 1 1 152.128 M 1 1 JV1 1 1 153.290 M 1 1 152.848 M 1 0 152.858 M 1 1 152.539 M 1 0 152.379 M 2 1

PAGE 111

Figure 6-1. Conceptual diagram of reproductive parameters used to estimate fecundity. Show here for simplicity is model for 2 nesting attempts; although more attempts are possible within a given year.

PAGE 112

97 Figure 6-2. The proportion of nest initiations for each month of the year based on cumulative data reported by Sykes (1987c), Snyder et al. (1989a)( 19701982 only) Toland (1994), and this study.

PAGE 113

98 Figure 6-3. The percentage of nests that were successful during each month. Data used in this analysis were from Bennetts et al. (1988)(1986-1987), Toland (1994 unpubl data) (1990-1993), and this study (1994-1995)

PAGE 114

99 ONDJFMAMJJAS 100 I 80 8 60 3 Z 20 0 1991 I i ONDJFMAMJJAS Figure 6-4. The percentage of nests that were successful during each month of each year. Data used in this analysis were from Bennetts et al. (1988)(1986-1987) Toland (1994 unpubl. data)(1991-1993), and this study (1994-1995)

PAGE 115

CHAPTER 7 DISPERSAL PROBABILITIES OF SNAIL KITES IN FLORIDA Introduction For birds living in stable and predictable environments there may be many advantages of site tenacity or fidelity. Knowledge of a location may better enable an individual to locate available food resources, avoid predators, and resist competitive intrusion (Alerstam and Enckell 1979). However, as environments become less stable and predictable, the advantages of dispersal may outweigh the advantages of not dispersing (Wiens 1976). One of the most commonly held beliefs for why animals move is availability of food or other resources (e.g., Krebs et al. 1974, Greenwood and Swingland 1984, Pyke 1984). Animals may move if resources are low or if there is potential for better resources elsewhere (Pyke 1984). Nomadic species in particularly, may disperse widely in response to sporadic food conditions (Andersson 1980). Snail Kites (Rostrhamus sociabilis ) in Florida exhibit nomadic tendencies and often move several times per year (Bennetts 1993, Bennetts and Kitchens 1997a, 1997b). Like many other species, a commonly suggested reason for Snail Kites to move is low food resources (others, Bennetts et al. 1994). Water levels are another commonly cited reason for kites to move, but low water levels are generally implied to represent low food availability (e.g., Beissinger 1988, Takekawa and Beissinger 1989, Sykes et al. 1995). The basis for much of this perception of why Snail Kites move is that if an area dries out, food for kites 100

PAGE 116

101 becomes unavailable, and individuals must disperse or perish. However, droughts in Florida occur at periodic intervals of approximately 5-10 years (Thomas 1974, Duever et al. 1994). Thus, most years are not drought years and dispersal would presumably be less advantageous under the current perception. In this chapter, I examine the probability of dispersal by Snail Kites during non-drought conditions and explore whether dispersal under these conditions is associated with local water levels or food availability. Methods One of the most commonly reported parameters of dispersal is distance. For many species, distance moved can have important genetic or population implications (e.g., Dobzhansky and Wright 1943, Shields 1982). However, Snail Kites in Florida exhibit nomadic tendencies and are likely to use and breed in numerous wetlands throughout their lifetime, including their natal one. Thus, distance per se is not an especially meaningful parameter for this species. Consequently, I focused primarily on dispersal between distinct wetlands. I considered wetlands to be distinct if they were separated by a physical barrier (e.g., ridge or levee) and/or were under a different hydrologic regime either through natural or managed control. Based primarily on watersheds, climatic factors, physiography, and management regimes, I also assigned each location to one of six regions (Bennetts and Kitchens 1997a). Terminology Many of the terms commonly used to describe types of dispersal are well suited to species that are territorial and/or migratory, but are poorly suited to species exhibiting

PAGE 117

102 nomadic tendencies. Thus, I have intentionally avoided some of the commonly used terms related to types of dispersal because they do not accurately describe the dispersal patterns of Snail Kites. For example, natal dispersal usually refers to the movement of an individual from its birth site to the place that it reproduces or would have reproduced had it survived and found a mate (Greenwood 1980, Greenwood and Harvey 1982, Johnson and Gaines 1990). Juvenile Snail Kites may disperse from their natal wetland and reproduce in numerous different wetlands during their lifetime including their natal site (Bennetts and Kitchens 1997a). Thus, for this paper, natal dispersal was defined as the initial movement of a juvenile from its natal wetland to a different wetland regardless of its resulting breeding status. Breeding dispersal has been defined as the movement of individuals between successive breeding attempts (Greenwood 1980, Greenwood and Harvey 1982, Johnson and Gaines 1990). Snail Kites do exhibit breeding dispersal, as defined by these authors, but it constitutes only a very small subset of the post-juvenile movements of this species. Consequently, I refer to the more general term of dispersal to describe the movement of individuals from one wetland to another, regardless of their breeding status. Field Methods Dispersal of Snail Kites was monitored over a 3 -year period between April 1992 and April 1995 in central and south Florida using radio telemetry. Adult kites were captured using a net gun (Mechlin and Shaiffer 1979), which uses 22 caliber blank cartridges to propel a 10-foot triangular nylon net. Juveniles were captured just prior to fledging, at approximately 30-35 d old, without a net gun. Fifteen-gram radio transmitters

PAGE 118

103 were attached to birds with backpack harnesses. Transmitters had a battery-life of approximately 9-18 months. I attempted to locate all radio-tagged kites at 14-day intervals. Tracking was accomplished from the air in a fixed-wing aircraft, and from the ground in an airboat. Estimation of Natal Dispersal I estimated the cumulative probability of natal dispersal of radio-transmitted birds using the Kaplan-Meier product-limit estimator (Kaplan and Meier 1958). A juvenile was considered to have dispersed when it was located alive outside its natal wetland. The time of dispersal was estimated as the midpoint between its previous location, in its natal wetland, and the first location outside of its natal wetland. This analysis only relates to the initial dispersal from the natal wetland. Movements subsequent to initial dispersal were not included in this analysis, but were included in my estimation of dispersal probabilities. Birds were censored (Lee 1980, White and Garrott 1990) if either I was unable to locate their radio signal or if they were known to have died prior to dispersal from their natal area. Log-rank tests were used for comparisons among dispersal functions. Estimation of Dispersal Probabilities I considered the probability of radio-tagged kites moving or staying at a given location^ over a one-month-time interval t to t + 1 as a binomial random variable, where the probability of an animal dispersing from a given location was estimated as the proportion of birds that dispersed of the birds that were available to disperse (i.e., conditional upon the animal being alive and its location known at / and / + l)(Nichols 1996). A one-month time interval was based on my sampling frequencies. The average

PAGE 119

104 time between consecutive locations was 13 .5 (7.9 SD) days with the upper limit of a 95% confidence interval of 29 days (Bennetts and Kitchens 1997a). Thus, I was reasonably certain to have located all birds within my study area every 29 days. For the ease of interpreting my results I used calendar months as the time step for this analysis. To explore the influences on whether or not a given bird moved during a time interval I used a conditional logistic regression model (Nichols 1996). Thus, my model expressed the conditional probability that, given a bird was alive and its location known at time /, it would be in the same location, or conversely at a different location, at time t+ 1. I then explored several potential effects on this probability. I began my model selection with a univariate analyses of each main effect. Because the potential contribution of main effects to interactions may be masked at this step, I initially used a liberal rejection criteria of a=0.25 (Hosmer and Lemeshow 1989). I then constructed a model including all main effects meeting the above criteria. At this and all subsequent steps of the analysis, I used a rejection criteria of a=0.05. I then used a combination of likelihood ratio tests and AIC to test for the inclusion or rejection of additional terms (Hosmer and Lemeshow 1989). I compared dispersal probabilities between adult and juvenile Snail Kites. Juvenile Snail Kites are capable of breeding at 9 months of age (Snyder et al. 1989) and survival of juveniles is similar to that of adults after the first few months post fledging (Bennetts and Kitchens 1997a). Consequently, I considered kites as adults after their first year post fledging. Annual differences were compared based on a study year from 15 April to 14 April of consecutive years (Bennetts and Kitchens 1 997a). Adults were trapped during spring and I usually had a reasonable sample each year by mid April. This also enabled

PAGE 120

105 analysis of three complete study years, rather than two complete years (1993 and 1994) and two partial years (1992 and 1995) if I had based my analysis on a calendar year. Seasonal differences were evaluated with respect to three 4-month seasons (January-April, May-August, and September-December). This designation of seasons was based on preliminary model selection using a combination of Akaike's Information Criterion (AIC)(Akaike 1973) and likelihood ratio tests to determine the most parsimonious designation of seasons using alternative lengths and using a sliding window approach to determine which months should be included (Bennetts and Kitchens 1997a). Food Availability During 1993 and 1994 I conducted 343 hours of foraging observations, which included 814 prey captures, to assess the influence of food resources on dispersal. I compared prey capture rates among seasons and years using the foraging time per capture as the dependent variable in an ANOVA model. To minimize confounding variation, all foraging observations were conducted on adult birds between two hours after sunrise and two hours before sunset. In addition, I restricted my observations to days that were not unseasonably cold (i.e., during the passage of cold fronts), were not raining, and winds did not exceed 20 kph. For comparisons of food acquisition, I used only complete observations of foraging bouts; that is, observations in which I observed an individual for the entire length of time it took to capture a snail from a previous capture. Course hunting (hunting by low flight over the marsh) is the most commonly used method of prey capture in Florida (Beissinger 1983, Sykes 1987a), and accounted for 82% of the captures I observed (Bennetts and Kitchens 1997a). Perch hunting accounted for only 18% of

PAGE 121

106 captures and was generally restricted to cypress prairie habitats (Bennetts and Kitchens 1997a, Darby et al. in press). Consequently, my comparisons also were limited to course hunting to reduce confounding between these different behaviors. In addition to comparisons of capture rates among seasons and years, I also compared the capture rates of individual radio-tagged birds for which I had foraging observations prior to subsequent dispersal. To reduce confounding, I restricted this analysis to those observations where the movement was within 30 days of obtaining the first foraging observation. This reduced the potential for seasonal differences in food acquisition to be confounded with differences between locations. To further reduce confounding, I matched the time of observations before and after moving. Thus, if the foraging observations before moving were conducted between 1 100 and 1300 h, then observations after moving were also conducted between 1 100 and 1300 h. I then tested the null hypothesis that the difference in mean foraging time per capture before and after moving was zero. Water Levels I tested the influence of relative water levels on the probability of dispersal in relation to water levels over a 27-year period from 1969-1995. For this analysis, I was limited to a subset of my data for which I had applicable hydrologic data (Bennetts and Kitchens 1997a). My subset for the 27-year period included most of the major wetlands used by kites, with the most notable exception being the Upper St. Johns marsh. Data that were applicable to the areas kites used most during this study (i.e., the Blue Cypress Water Management Area and Blue Cypress Marsh Water Conservation Area) were

PAGE 122

107 available only after 1990. Water levels were determined using site-specific gauges at each location maintained by the South Florida Water Management District, St. Johns River Water Management District, U.S. Army Corps, of Engineers, U.S. Geological Survey, and City of West Palm Beach. Specific gauges were reported by Bennetts and Kitchens (1997a). Because water depth can be highly variable within sites, and reliable ground elevation data to estimate site-specific depth are lacking, I used variation in stage (i.e., water surface elevation) as the basis for my assessment of water levels. I estimated an average of the monthly annual stage (i.e., water surface elevation relative to mean sea level) over a 27-year period. I then used the number of standard deviations above or below that average, for any given year, as my measure of water levels. This measure provides an objective assessment of water levels, which can be applied to all areas (Bennetts and Kitchens 1997a). Because water depth can be highly variable within sites, and reliable ground elevation data to estimate depth are lacking, I used the standard deviations of the average monthly water surface elevation (stage relative to mean sea level) over a 27-year period from 1969-1994 as my measure of water levels. The period of record was used for areas that did not have reliable records for all 27 years. This measure provides an objective assessment of water level, rather than the subjective designation of drought years by unspecified criteria used in previous studies (e.g., Snyder et al. 1989, Beissinger 1995). A major drought had occurred in the southern portion of the Snail Kite's range during 1989 and 1990, with relatively low water persisting through 1991. Droughts may have residual effects on food availability that extend beyond their occurrence (Snyder et al.

PAGE 123

108 1989). Consequently, I compared natal dispersal between areas that were highly influenced by the preceding drought (i.e., wetlands within the Everglades, Loxahatchee Slough, and Lake Okeechobee regions) and areas which were relatively unaffected by the drought (i.e., the Kissimmee Chain-of-Lakes, and Upper St. Johns River Basin)(Bennetts and Kitchens 1997a). Results I equipped 282 Snail Kites with radio transmitters representing 271 individuals; 1 1 birds were recaptured in a subsequent year and their radios replaced. Of these radios, 165 (59%) were placed on adults and 1 17 (41%) on juveniles. I obtained 5,299 locations of these radio-tagged birds, of which 3,618 were of adults and 1,681 were of juveniles. Natal Dispersal The overall cumulative probability of juvenile Snail Kites dispersing from their natal wetland during their first year was 0.81 (Figure 7-1). Only 8 of 65 (12%) radiotransmittered birds over a three-year period that survived their entire first year and whose locations were known remained in their natal wetland for their entire first year. Of the birds that dispersed during their first year (n=57), most (60%) did so within the first 60 days after fledging and all did so within the first 240 days. Dispersal of juveniles from their natal wetland was lowest during 1992 and relatively higher in both 1993 and 1994 (Figure 7-2). Differences were significant between 1992 and 1993 (x 2 =5.25, 1 df, P=0.022) and between 1992 and 1994 ( x 2 =4.049, 1 df, P=0.044), but not between 1993 and 1994 (x^O.129, 1 df, P=0.720). Natal

PAGE 124

109 dispersal from regions most affected by the previous drought (n=13) was substantially lower than that from other regions (n=21) during 1992 (x 2 =7.53, 1 df, P=0.006), but not during 1993 (x^O.635, 1 df, P=0.426) or 1994 (x 2< 0.001, 1 df, P=0.994)(Figure 7-3). I had insufficient sample sizes to statistically compare juvenile dispersal among individual wetlands or regions during 1992; however, all the three regions affected by the drought had lower estimates of dispersal than either of the two regions that were unaffected by the drought. Dispersal Probabilities I began preliminary testing for an overall time effect using a univariate model based on separate parameter estimates for each month of each year of my study (i.e., 12 months for each of 3 years =36 parameters). This test showed a strong effect of time (x 2 =90.79, df=36, P<0.001). I then tested whether this effect could be accounted for with a more parsimonious model using separate months, but not for each year (i.e., 1 parameter for each month [12] rather than 36 for the previous model). This model also showed a significant monthly effect (x 2 =32.36, df=l 1, P<0.001), but a LRT indicated that the more general model (with 36 parameters) was warranted (LRT=73.81, 24 df, P< 0.001). I next explored a series of models in which time was expressed as a seasonal, rather than monthly effect. These models reflected various combinations of 3 and 4 seasons in a sliding window approach (i.e., each iteration shifted the months included by one month) to determine how the months should be divided into seasons. This analysis indicated that one of the 3 -season models (of 4 months/per season) was the most parsimonious based on AIC (Table 7-1); however, 2 additional models (the model with a separate parameter for each

PAGE 125

110 month and one of the 4-season models) had similarly low AIC values and could not be rejected based solely on AIC criteria. Next, I compared a suite of models including combinations of the above effects including their interaction terms. This analysis indicated that a model with season (the 3 -season model selected from the above analysis), year, and the interaction between season and year was the most parsimonious based on AIC (Table 7-2). As above, two alternative models (individual month model with 36 parameters and the model with season and year without an interaction term) had similarly low AIC values and could not be rejected based solely on AIC. In contrast to AIC, a likelihood ratio test indicated that the model with 36 parameters was warranted (x 2 =50.72, df=27, P=0.004) and that a season x year interaction was not warranted (x 2 =9.55, df=4, P=0.049). Thus, my preliminary analysis indicated that movement probabilities were influenced by season, year, and possibly an interaction between season and year (Figure 7-4), but also that additional significant sources of variation likely exist. I next explored whether the location of a given bird influenced whether or not the bird moved between times t and / + 1 For this analysis, I was not concerned with the destination of the bird, only its location at the time that dispersal occured. I began this analysis with the general null hypothesis that the specific wetland where a bird was located at time / did not influence the probability of whether or not it moved to a different location at time / + 1 I rejected this null hypothesis based on a conditional logistic regression model (x 2 =107.38, 16 df, P <0.001)(Figure 7-5). I then tested the same hypothesis using the region, rather than the specific wetland, to determine if this might provide a more parsimonious model. This test also rejected the null hypothesis (x 2 =53 .82, 5 df, P

PAGE 126

Ill <0.001); however, a comparison of these two models indicated that my data supported use of the more general model (i.e., specific wetlands) based on both a LRT (x 2 =52.04, 1 1 df, P <0.001), and on AIC (Table 7-3). Pooling of Locations I next explored whether I could improve my model by some limited selective pooling based on a combination of biological and statistical criteria. My goal for this exploration was to determine if I could obtain a more parsimonious model by pooling areas in which the overall relative use and seasonal patterns of use were similar enough so as not to warrant separate parameter estimates. I did not attempt to pool some wetlands whose use patterns I felt were biologically different (e.g., wetlands that were used primarily during non-breeding with wetlands used primarily for breeding) even though I could have done so strictly based on statistical criteria. Thus, although a more parsimonious model for the effects of location on movement probability was possible, I preferred to maintain separate parameter estimates for some areas to better ensure the biological integrity of these models. I began my exploration of potential pooling with areas in the Southern Everglades. The first pooling I considered was Everglades National Park (ENP) and Northeast Shark River Slough (NESRS). Each of these these areas are administered by the National Park Service, are part of the Shark River Slough (ENP has areas not within the Shark River Slough, but these areas were not used by radio-transmittered kites during my study), receive low to moderate kite use, and are not impounded at their outflow (each have levees at their inflow). A statistical comparison indicated that separate parameter

PAGE 127

112 estimates for each of these areas was not warranted based on LRTs and AIC (Table 7-4). Next, I considered including Water Conservation Area 3B (WCA-3B) with ENP and NESRS. WCA-3B is also within the Shark River Slough, but is impounded at its outflow and is administered by state agencies. However, its overall relative use and its seasonal patterns of use (each of these areas tended to be used most during early summer) were quite similar to ENP and NESRS. A statistical comparison indicated that separate parameter estimates for each of these areas also was not warranted based on LRTs and AIC. I considered pooling Big Cypress National Preserve (BICY) with ENP, NESRS, and WCA3B; however, BICY was used more extensively during fall and early winter than these other areas and much of areas used in BICY consisted of cypress prairie habitat which was not generally available in these other areas. Consequently, I did not include BICY with these other areas, even though I probably could have justified doing so on a statistical basis. Next, I considered pooling the A R M. Loxahatchee National Wildlife Refuge (WCA-1), Water Conservation Area 2 A (WCA2A), and Holey Land Wildlife Management Area (HOLEY). Each of these areas represents northern Everglades habitats, although their water management histories have differed. My statistical comparison indicated that separate parameter estimates for each of these areas was not warranted based on LRTs and AIC. I next considered areas within the Kissimmee Chain-of-Lakes. First I considered pooling Lakes Tohopekaliga (TOHO) and East Lake Tohopekaliga (ETOHO). My statistical comparison indicated that separate parameter estimates for each of these areas

PAGE 128

113 was not warranted based on LRTs and AIC. Next I considered including Lake Kissimmee (KISS) with TOHO and ETOHO. Lake Kissimmee received moderately heavy use compared to the other lakes, but all are in close proximity, the seasonal patterns of use were similar, and there was considerable interchange among these lakes. My statistical comparison indicated that separate parameter estimates for each of these areas again was not warranted based on LRTs and AIC. In contrast, other lakes within the KissimmeeChain-of-Lakes (e.g., Lakes Marion, Tiger, Walk-in-the Water, and Marian) received substantially less use than KISS, TOHO, and ETOHO and the seasonal pattern of use was quite different (i.e., they were used most frequently during non-breeding periods). The seasonal pattern of use of these smaller lakes more closely resembled that of the peripheral habitats. Consequently, I next considered pooling the smaller lakes of the KissimmeeChain-of-Lakes with the peripheral habitats. My statistical comparison indicated that separate parameter estimates for each of these areas were not warranted based on LRTs and AIC. Finally, I compared various combinations of pooling to the general unconstrained model and among each other. This analysis indicted that the model containing all of my proposed pooling was a substantial improvement over the unconstrained model (i.e., with no pooling). Additionally, the model with all of my proposed pooling had the lowest AIC; although overall differences among all of the models I compared were relatively small. Consequently, I used the most parsimonious model (Model 16) of this set of models in further analyses of the influence of location on movement probability. This model had 10 parameters compared to the unconstrained model with 17 parameters; but was still a

PAGE 129

114 substantial improvement (x 2 =47.61, 7 df, P <0.001) over the 6-parameter model using regions, rather than location. Using the effects indicated from my preliminary analyses, I began my overall model selection with a preliminary univariate analysis of each potential term (Table 7-5). As above, this analysis indicated that dispersal probabilities were influenced by age, season, year, and location. The results from using AIC (Table 7-6) were not in complete agreement with LRTs (Table 7-7). The model with the lowest AIC was a model with all main effects and only one 2-way interaction term (Season*Location); whereas, LRTs supported the inclusion of an additional interaction term for (Year*Location). Differences between models indicated by LRTs and AIC are not uncommon and represent conceptual differences between an approach of hypothesis testing (LRTs) and optimization (AIC)(Spendelow et al. 1996). Both approaches were consistent in that they indicated the same main effects of age, season, year, and location and differed only in their respective indications of which interactions were supported by the data. Although univariate analyses indicated that dispersal probabilities differed among areas affected or not affected by the preceding drought (i.e., overall probabilities of dispersal were lower in areas affected by the drought), my model selection indicated that the most parsimonious models included the effect of individual wetlands, rather than region or whether or not they were affected by the drought. An examination of the residuals indicated that adults generally had higher dispersal probabilities than juveniles. The residuals also indicated that annual effects were primarily

PAGE 130

115 attributable to less than expected movements during 1992 for both age classes (Figure 76), and that seasonal effects were primarily attributable to higher than expected dispersal probabilities during summer (Figure 7-7). The effect of individual wetlands varied among years and seasons. Hvdrologic Effects on the Probability of Dispersal I began my analysis with a univariate approach to logistic regression (Hosmer and Lemeshow 1989) using the departure from average stage (described above) as a continuous independent variable. This analysis initially indicated an effect of relative water level (x 2 =8.27, 1 df, P=0.004). However, because location (i.e., wetland) and water level are confounded and my previous analyses indicated an effect of location, I next tested a model that included the effects of both location and water level. This test was consistent with my earlier analysis indicating a location effect (x^SS.88, 9 df, P<0.001); however when included in a model with location, the effect of water level on the probability of dispersal was no longer apparent (x 2 =0.01, 1 df, P=0.934). Although these results indicated that relative water level was not a major influence on the probability of dispersal, I must emphasize that the hydrologic conditions under which my study was conducted were generally high water conditions throughout the study area. Consequently, low water conditions that might have triggered dispersal generally did not occur during this study and inferences regarding the effects of lowwater conditions on dispersal probabilities could not be made. However, previous studies (e.g., Beissinger and Takekawa 1983, Takekawa and Beissinger 1989) have indicated substantial dispersal of Snail Kites during low-water conditions. Given that apple snails

PAGE 131

116 may die or become unavailable to kites during dry conditions, these previous reports are certainly reasonable, and may indicate that the lack of an effect of water levels I observed applies only to generally high water conditions. Food Resources My data indicated a difference among years, seasons, and an interaction between year and season in the time required to capture apple snails (Table 7-8). Capture times were lowest during summer, relatively higher during spring, and still higher during autumn of each year (Figure 7-8). Capture times also tended to be lower during 1994 compared to 1993 for each season. My data indicated no difference in food acquisition before and after moving (f=0.60, P=0.57). Furthermore, in 4 of the 8 movements I observed, birds increased the time required to capture snails (Figure 7-9). In the remaining 4 cases, birds decreased the time required to capture snails. Discussion Snail Kites in Florida range across a network of wetlands covering thousands of square kilometers. Thus, direct estimates of food availability were not feasible over their entire range. However, the temporal patterns of dispersal I observed were not consistent with the hypothesis that food availability was a proximate cue to initiate moving from one wetland to another. My analysis of residuals indicated that dispersal probabilities during 1992 were lower than expected compared to 1993 and 1994. My analysis of juvenile dispersal also showed that dispersal was lower during 1992, particularly in the southern

PAGE 132

117 portion of the kite's range that was influenced by the preceding drought. Although I did not begin formal foraging observations until 1993, anecdotal evidence strongly suggested that food resources during 1992 were lower following the preceding drought. My foraging observations during 1993 and 1994, as well as numerous previous reports (reviewed by Sykes et al. 1995) indicated that the capture of snails generally requires < 10 minutes. In contrast, birds I observed in the southern portion of their range during 1992 often would forage in excess of 30 minutes without capturing a snail. In a few instances during my trapping period, I observed several foraging birds for hours without observing a single snail having been captured. The seasonal patterns of dispersal I observed also were not consistent with low food resources having been the proximate cue for dispersal. My analysis of residuals indicated that dispersal probabilities were highest during summer, the season that my foraging observations indicated that the foraging time spent capturing snails was lowest. The evidence provided by temporal patterns that food availability was not a proximate cue to initiate dispersal was further supported by foraging observations of radio-tagged birds before and after dispersal. Although my sample size was small, half of the birds I observed actually did worse at obtaining food at their new location compared to the location that they left. Although one bird that moved to a new location where it took a particularly long time to capture snails moved again shortly after arriving to a new location where it took less time. This suggests that there might be a threshold of food acquisition, below which birds will move if there are more favorable sites elsewhere.

PAGE 133

118 Low food availability not having been a proximate cue to initiate dispersal during this study does not preclude its role as a cue to dispersal under other environmental conditions or it being an ultimate factor for the patterns of dispersal I observed. My study was conducted entirely during a period of relatively high water. If water levels become low enough to negatively affect food availability (e.g. < 20 cm), then kites will likely move from that area. However, during most years there is considerable dispersal that appears to be independent of water levels. The high rate of dispersal during favorable conditions may, however, be indirectly related to food availability on a broader scale. I suggest a hypothesis that these movements may be a reasonable strategy given the dynamic and unpredictable nature of a kite's environment. A virtual certainty about any wetland inhabited by Snail Kites is that it will go dry at some point in time. Florida apple snails, the primary food of Snail Kites, are aquatic and have a limited capacity to survive dry conditions (Little 1968, Darby et al. 1996). Consequently, dry conditions result in a reduction of local food resources. What is not certain is which wetlands will go dry in which years. Thus, there may be an advantage for kites to have experience regarding the availability of wetlands throughout their range so that when a local drying event does occur, past experience reduces the need for "blind" searching for suitable alternative habitats. Thus, moving when food availability is high may enable kites to "explore" their potential habitats with little risk of starvation. The resulting experience from many locations may then help kites to locate food resources faster during times when food is scarce.

PAGE 134

119 Dispersal patterns of animals may be highly influenced by the distribution of resources and may be integrally associated with social behavior. When resources are predictable and evenly distributed in space and time there may be advantages of tenacity, such as knowledge of available food resources, avoidance of predators, and resistance competitive intrusion (Alerstam and Enckell 1979). Under such conditions, territoriality also may have advantages, provided that the expense resource defense are not excessive (Wiens 1976). As the distribution of resources becomes more patchily distributed and less predictable in space and time, nomadic tendencies may emerge. My data indicate that Snail Kites exhibit a combination of dispersal patterns that express both of these characteristics, but at different scales. Apple snail populations may be patchily distributed and may exhibit substantial annual and/or seasonal variability (Darby et al. 1997). However, local areas of high snail availability are likely to persist for weeks or even months until environmental conditions, predation pressure, or their natural population dynamics that includes an annual die off affects their availability to kites. During times of high food availability, I observed radio-tagged kites using and defending small areas for weeks and even months. These birds exhibited territorial behavior including vocalizations at approaching conspecifics and ensuing chases if the intruding bird entered the foraging territory (see also Snyder and Snyder 1970, Sykes 1987a, Sykes et al 1995). However, the seasonal and annual changes in prey availability would preclude tenacity as a viable strategy over longer time scales. My data indicate that kites often move to different wetlands several times per year and, although there is some indications of site tenacity and/or philopatry, most birds exhibit a fair degree of less habitual movements (Bennetts i

PAGE 135

120 and Kitchens 1997a). Even within-wetlands during most times, kites exhibited a fair degree of wandering, except when tied to a nest. Thus, at longer time scales the dispersal patterns of Snail Kites would be better described as nomadic.

PAGE 136

121 Table 7-1 Summary statistics for conditional logistic regression models for potential seasonal groupings affecting the probability of movement between times t and t + 1 (at monthly time steps), given that an animal was alive at time t and its location known. Shown are the model description, number of estimable parameters (np), relative deviance (-21n[Sf]), and Akaike's Information Criteria (AIC). The model shown in bold would be the one selected from these potential models based on AIC. Season Model" np -2/(9?) AIC (JFMAM JJASOND) 12 2474.44 2498.44 (J F M A) ( M J J A) ( S O N D) 3 2490.66 2496.66 (FMAM)(JJ AS) (OND J) 3 2498.75 2504.75 (MAMJ)(JASO)(NDJF) 3 2506.24 2512.24 (AMJJ)(ASON)(D JFM) 3 2498.14 2504.14 (JFM) (AM J) (J AS) (OND) 4 2491.16 2499.16 (F M A) (M J J) (A S 0) (N D J) 4 2497.98 2505.98 (M AM) (J J A) (S 0 N) (D J F) 4 2497.36 2505.36 a Months are abbreviated by their respective first letter and listed sequentially.

PAGE 137

122 Table 7-2. Summary statistics for preliminary conditional logistic regression models for potential temporal effects, ignoring age effects, on the probability of movement between times t and t + 1 (at monthly time steps), given that an animal was alive at time t and its location known. Shown are the model description, number of estimable parameters (np), relative deviance (-21n[££]), and Akaike's Information Criteria (AIC). The model shown in bold would be the one selected from these potential models based on AIC criteria. Model np -2fo(S£) AIC Time 36 2400.63 2472.63 Month 12 2474.44 2498.44 Season 3 2490.66 2496.66 Year 3 2477.48 2483.66 Month Year 14 2446.67 2474.67 Month Year Month* Year 36 2402.74 2474.74 Season Year 5 2460.90 2470.90 Season Year Season* Year 9 2451.35 2469.35

PAGE 138

123 Table 7-3. Summary statistics for conditional logistic regression models of the probability of movement between times t and t + 1. The model with the lowest AIC (bold) would be selected if based solely on this criterion. Source -2/w(9£) np AIC Specific Wetland 2394.12 17 2428.12 Region 2446.16 6 2458.16

PAGE 139

124 Table 7-4. Summary statistics for conditional logistic regression models of the probability of movement between times t and t + 1 to evaluate the pooling of some parameters. A failure to reject a LRT indicates that the additional parameters of the more general (unconstrained) model may not be supported by these data. The model with the lowest AIC (bold) would be selected if based solely on AIC. No. Constraints (Pooling) LRT* (x 2 ) df p>x 2 np AIC 1 Unconstrained — — — 17 2428.13 2 NESRS = ENP 2.78 1 0.10 16 2428.90 3 NESRS = ENP=WCA-3B 2.80 2 0.25 15 2426.93 4 HOLEY = WCA1 0.07 1 0.79 16 2426.20 5 HOLEY = WCA1=WCA2A 0.14 2 0.93 15 2424.26 6 TOHO= ETOHO 1.46 1 0.23 16 2427.59 7 TOHO= ETOHO=KISS 1.48 2 0.48 15 2425.61 8 PERIPHERAL = KISSCH 0.00 1 0.99 16 2426.12 9 Reduced Model (2,4,6,8) 4.31 4 0.36 13 2424.44 10 Reduced Model (2,4,7,8) 5.39 6 0.49 11 2421.52 11 Reduced Model (2,5,6,8) 4.38 5 0.50 12 2422.51 12 Reduced Model (2,5,7,8) 4.40 6 0.62 11 2420.52 13 Reduced Model (3,4,6,8) 4.34 5 0.50 12 2422.46 14 Reduced Model (3,4,7,8) 4.36 6 0.63 11 2420.48 15 Reduced Model (3,5,6,8) 4.41 6 0.62 11 2420.53 16 Reduced Model (3,5,7,8)" 4.43 7 0.73 10 2418.55 a Based on comparison with unconstrained model b Includes all proposed constraints (pooling)

PAGE 140

125 Table 7-5. Maximum likelihood analysis of variance table for univariate models (i.e., each source term represents a separate model) of potential sources of variation of the probability of dispersal between times t and t + 1 Source x 2a df p>x 2 Age 5.38 1 0.020 Season 17.85 2 <0.001 Year 28.28 2 <0.001 Location 103.36 9 <0.001 1 Based on Wald x* statistic (SAS inc. 1988).

PAGE 141

126 Table 7-6. Summary statistics used for model selection of conditional logistic regression models of the probability of movement between times t and t + 1 The model with the lowest AIC (bold) would be selected if based solely on AIC criterion. Model Model Description 1 -2/(a!) np AIL 1 Age n zjUo.y.) I Season z4yu.oo T J z4yo.oo 5 Year n Ann a o Z47/.4o 3 24oJ.4o A 4 Location 2398.55 10 2418.55 5 A O V7" T A S Y L 2339.23 15 2369.23 6 A O ~\T T A *T A S Y L A*L 2325.62 24 ^^^^ 2373.62 7 A S Y L S*L 2294.01 33 2360.01 o 0 A C VA T V7"j|cT A a Y L Y*L 2308.41 33 2374.41 Q y A C V T A *C A a Y L A*a 2338.22 17 2372.22 1 u A C V T A *V 15 Jo.z4 1 T 1 / OOTA *>A 15 10.14 1 1 1 1 A C V T C*V A a l L 5 I Til 111 iy zjoy. 1 1 1 9 1 z A 5 V T Q*T V*T A a I J L I L ZZjy. j 1 5 1 zJoi.3 1 A C V T C*T C*V a a y .l a L, a i 228y.J7 37 2363.37 1 A 14 A a Y L a Y Y L 2302.20 37 2376.20 1 J A a Y L a*L Y*L a*Y 2257.06 55 2367.06 16 A S Y L with all 2-way Interactions 2237.34 68 2373.34 17 A S Y L S*L Y*L A*S*Y 2254.41 55 2364.41 18 A S Y L S*L Y*L A*S*L 2238.06 69 2376.06 19 A S Y L S*L Y*L A*Y*L 2226.81 69 2364.81 20 A S Y L S*L Y*L S*Y*L 2199.63 87 2373.63 21 A S Y L S*L Y*L S*Y*L A*Y*L 2166.04 105 2376.04 22 A S Y L with all 2 & 3 -way Interactions 2126.47 144 2414.47 23 Fully Saturated 2091.45 180 2451.45 "Abbreviated model terms are A=Age, S=Season, Y=Study Year, L=Location

PAGE 142

C O 55 > I gsf 9 2 o o 5? cO .-3 4> £ T3 £0 O X> c o c fa 00 • £ af o E o a 2 =3 CD M 03 CO u § S3 CO oo o v CN ts o o 00 00 N CN oo d CN " 00 < —i CN -J > H-l oo < > H-l # > oo O0 A* H-l H-l # < O0 00 00 oo OO H-l -J hJ >< > > >H > O0 oo oo 00 00 oo < < < < < < < o IT) no ON >< > # CO CO hJ >>< > hJ — 1 hJ # oo oo oo hJ hJ > oo oo oo <: < CN t*1 oo r^CN oo oo ^ tn 00 "* NO m CN rr' 00 CN CN cn' rn 5m CN o o c i (0 i CN 15 + -J oo < H-l # > oo S*Y Y*L hJ H-l oo Y*L oo >* oo hJ CO Y*L H-l OO H-l oo S*L hJ H-l H-l hJ hJ H-) H-l hJ H-l hJ hJ >> >" > > > > > >< >H >^ CO oo 00 oo in oo CO oo CO CO CO CO CO oo CO < < < < < < < < < < < < < < < # oo < hJ hJ oo hJ > CO

PAGE 143

128 -a 2 I o ri t-(N 00 (N O 00 o o V o d o d V no oo no > oo < -J 1—1 • > > > -J # 00 00 00 00 00 hJ -J >H > > >H >OO on oo oo < < < < < > hJ < < # oo > < < 00 > Xoo Y*L Y*L S Y*L * OO oo oo oo oo >— 1 — 1 >* > > >• 00 00 oo oo < < < < < so 00 ON o 1—1 CN O V NO NO ON r o SO IT) r CN o> rn o On (N m CO NO m < # 00 00 00 < C o o u >1 I CN + 00 < 00 c/5 c _o 5 E i CN + -1 oo < c/5 c o 1 c 1 I oy (N + > OO < o V On NO n f! NO 00 — 1 oo < c g O u i E 3 € 00 — CN ro N N N CN ir> NO ro CN O CO XS1 C o a E 0) I m CN 13 + > oo < c o • — *-> o 3 u i 0) i 00 3 CN

PAGE 144

129 Table 7-8. Analysis of variance table from model of foraging time per capture as the dependent variable. Mean square (MS) and F values are based on type III partial sums of squares (i.e., they are adjusted for all other terms in the model and are not dependent on the order of entry)(SAS Inc. 1988). Source df MS F P>F Year 1 386.79 29.65 <0.001 Season 2 374.80 28.73 <0.001 Year x Season 2 166.65 12.78 <0.001 Error 175 13.04

PAGE 145

130 Figure 7-1 Kaplan Meier estimates for the overall cumulative probability of dispersal (solid line). Also shown is a 95% confidence interval (dotted line) for the probability function.

PAGE 146

131 i I i i i i i i 1 1 1 r APR MAY JUN JUL AUG SEP OCT NOV DEC JAN FEB MAR APR Date Figure 7-2. Kaplan Meier estimates for the cumulative probability of dispersal in each of the three years. Confidence intervals for estimates are not shown to minimize cluttering, but are provided in detail in Appendix 1.

PAGE 147

132 CT3 10 Date Figure 7-3. Kaplan Meier estimates for the cumulative probability of dispersal from wetlands that were and were not affected by the preceding drought in each of the three study years. Confidence intervals for estimates are not shown to minimize cluttering, but are provided in detail in Appendix 2.

PAGE 148

133 0.50 — r~ SP — r~ FA SU 1993 Season/Study Year a c 0.50 0.40 2 0.30 O 5 0.20 to 2 0.10 0.00 Juveniles — i 1 ( — 1 r SP SU FA SP SU FA 1992 1993 Season/Study Year SP SU 1994 I FA Figure 7-4. Conditional probabilities that adult and juvenile Snail Kites that were alive their location known at time t, were in the same location (or conversely at a different location) at time t + 1 during each season of each study year. Also shown are the standard errors (rectangles) and 95% confidence intervals (vertical lines).

PAGE 149

134 10 -5 WCA1 WCA2B WCA3B BICY HOLEY OKEE TOHO K_CH PERIP WCA2A WCA3A ENP NESRS WPB KISS ETOHO SJM Location Figure 7-5. Adjusted residuals from a cross tabulation of dispersal and location at time t. Residuals >0 indicate that birds in this area moved more frequently than expected and residuals <0 indicates that birds in that area moved less frequently than expected.

PAGE 150

135 1992 1993 Study Year 1994 1992 1993 Study Year 1994 Figure 7-6. Standardized residuals for probability of dispersal during each study year for adult (top) and juvenile (bottom) snail kites.

PAGE 151

136 co -g "w CD DC o CD N "2 CO "D c CO *-> C7) Summer Season co g CO CD DC S co Spring Summer Season Fall Figure 7-7. Standardized residuals for probability of dispersal during each season for adult (top) and juvenile (bottom) snail kites.

PAGE 152

137 30 c 25 a 20 15 10 5 SPR SUM FALL SPR Season (Year) SUM FALL Figure 7-8. The mean ( SE) foraging time to capture snails, complete bouts observed) are also shown. Sample sizes (number of

PAGE 153

138 6 5 4 8 6 3 2 1 15 + 10 11 3 — + i i i r~ 12 12 152.079 152.539 i r 1 2 153.101 18 I I 1 2 153.140 Location (Frequency) i i r 1 2 3 153.564 — i — r~ 1 2 153.930 9 + 1 2 153.940 Figure 7-9. The mean ( SE) foraging time to capture snails by radio-transmittered birds before (location 1) and after (location 2) moving. Sample sizes (number of individuals birds observed) are also shown.

PAGE 154

CHAPTER 8 IMPLICATIONS TO MANAGEMENT AND CONSERVATION There has been considerable written regarding the management and conservation of Snail Kites in Florida, most of which has focused on water management. In this chapter I begin with some semantic issues that have been previously ignored, but in my view, must precede any discussion of the implications of water management on the Florida population of Snail Kites. I then discuss management in the context of the entire Snail Kite population, including a network of wetland habitats that comprise its range, and the spatial and temporal dynamics that characterize the persistence of this species in a fluctuating environment. Drought Semantics There has been considerable discussion in the literature about the influence of drought on Snail Kite populations (e.g., Sykes 1979, 1983a, Beissinger 1986, 1995, Snyder et al. 1989a). Given the potential importance for this influence, it is surprising that these authors have not even defined a drought sufficiently to enable an independent observer to designate a given year as a "drought year" based on objective criteria. Drought can be measured in several ways and be based on either water levels or rainfall (Duever et al. 1994). Snyder et al. (1989a) and Beissinger (1995) designated each year over their respective periods of study as a either a drought year, lag year (the first year 139

PAGE 155

140 post drought), or high water year (i.e., non drought) based on water levels. However, the water levels that were used to assign a given year as a drought year were not defined by these authors, nor was there consistency between these authors as to which years that were designated as drought years. Of the years that drought status was reported by both studies (1969-1983), Snyder et al. (1989a) determined that 1971, 1974, 1981, and 1982 were drought years; whereas, Beissinger (1995) determined that 1971, 1973, 1974, and 1981 were drought years. It seems a logical starting point that before any serious evaluation of the influence of drought on Snail Kite populations can be made, at least it be clear as to what constitutes a drought. I believe that an absence of clear definitions has led to considerable misunderstanding among researchers and managers. For example, Beissinger (1995) suggested that Snail Kite populations would decline if the interval between droughts was less than 3 .3 years. I have some serious concerns about the parameter estimates he used to derive this conclusion; however, I do not necessarily disagree with his conclusion; provided he is referring to widespread droughts. I would strongly disagree, however, if he is suggesting that local drying events at intervals less than 3 .3 years would result in population declines. In fact, much of the habitat used by kites during my study (and over the past decade) has dried out (on average) more frequently than every 3.3 years and all evidence (i.e., the annual survey and our parameter estimates) suggests that the overall population has been stable or increasing. Thus, I believe that this may, or may not, be a valid recommendation depending on how it is interpreted. There are three essential characteristics of droughts that should at least be considered and operationally defined for effective evaluation of droughts (Lin et al. 1984)

PAGE 156

141 and their influence on Snail Kite populations. They are the intensity, spatial extent, and temporal extent (e.g., duration and/or number of consecutive drought years) of the droughts being considered. Intensity I consider the intensity of a drought as being a measure of "how dry it was". This is probably best measured from the "well" type of gauges, which are capable of recording water stage even below ground level at the recording station. One approach to measuring intensity is to assess water levels at a given location in relation to a reference elevation, (e.g., ground elevation at that location). I have shown as an example the annual minimum stages for WCA-3A as recorded at the 3-28 gauge, (the gauge reportedly used by Beissinger [1995] because of its proximity to areas used by kites)(Figure 8-1). I have also shown some reference elevations that might be used to designate a given year as having been a drought year. To illustrate ambiguities that can result from not defining criteria used to designate a given year as a drought, I have also shown which years were determined to be drought years by Snyder et al. (1989a) and Beissinger (1995). It quickly becomes apparent that Beissinger (1995) could have defined a drought year (for WCA3 A) as a year in which the minimum annual water stage was<2 m. A similar criteria based on this gauge was not apparent for the assignment of drought years by Snyder et al. (1989); however, it is quite possible that they may have based their designation on a different location or using different criteria. Since WCA-3 A and Lake Okeechobee were the two primary areas in each of these studies, I conducted a similar assessment of the stages of Lake Okeechobee using a standard 10-gauge average (provided by the

PAGE 157

142 U.S. Army Corps, of Engineers) to account for spatial variation among gauges (Figure 82). The designation of drought years by Snyder et al. (1989) is more apparent based on data from Lake Okeechobee (all years they assigned as drought years had stages less than 3.57 m). However, the designation of drought years by Beissinger (1995) becomes less apparent using data from Lake Okeechobee. For example, Beissinger (1995) determined 1973 to be a drought year but not 1975, 1976, and 1982, all of which had lower stages at Lake Okeechobee. My point is not to agree or disagree with which years these authors determined to be drought years; but rather, to point out that without defining the criteria, the designation of drought years becomes quite subjective. If the intensity of a drought is used to measure the behavioral or demographic response of the population, then it is also not always the case that drying to ground elevation should be used as the measure. For example, anecdotal evidence suggests that Snail Kites move from a drying area well before water levels reach ground level. In some cases it may be more meaningful to distinguish a functional dry down (i.e., the water level at which the response occurs) from a physical dry down (i.e., water level < ground elevation). But again, I suggest that the criteria for such a designation be defined. An alternative approach for assessing the intensity of a given drought is to use a statistical measure of the variability (e.g., the standard deviation) of the actual water data. For example, I determined the minimum annual water level for gauge 3-28 for each year from 1969 through 1995. I then calculated the mean and standard deviation for these minimum annual water levels. If I use this approach and define a drought year as any year in which water levels were lower than 1 standard deviation below the mean minimum

PAGE 158

143 water level, then I would have selected exactly the same years as Beissinger (1995) for the period that he used in his analysis (Figure 8-3). Although, as above, this approach would not necessarily have been applicable to wetlands other than WCA-3 A. The advantage of this approach compared to assigning a reference elevation is that the same criteria (e.g., > 1 standard deviation below the mean) can be used for any area. This eliminates the subjectivity imposed by assigning a different reference elevation for each area and all areas determined to be drought years have an equal relative intensity. Spatial Extent In virtually every year, the water level in some portion of the habitat reaches ground elevation during the dry season. Additionally, the spatial and temporal variability of rainfall in Florida (Mac Vicar and Lin 1984) results in spatial and temporal variability in droughts (Duever et al. 1994). Consequently, it is necessary to identify and define the spatial extent of a particular drought being considered if meaningful conclusions are to be drawn, particularly for a species such as Snail Kites that uses the entire South Florida landscape. For a preliminary exploration of the spatial relationships of droughts among wetlands within Snail Kite habitat I used water levels from major wetlands (where reliable water data were available) to determine whether or not a drought had occurred during each year from 1969-1994 (the years in which the annual Snail Kite survey were conducted). The specific gauges used for this analysis (and all subsequent analyses of hydrology) are presented in Appendix 8-1. For this exercise, I defined a drought year as any year in which the minimum water levels were >1 sd below the 26-year mean as a

PAGE 159

144 drought year and an extreme drought year as any year in which the water levels were >2 sd below the mean. The 1977 minimum annual stage for Lake Kissimmee was not used to calculate the standard deviation because it was an extreme outlier (9.4 sd below the mean if 1977 value is not included) due to an intended drainage of the Lake for management. Inclusion of this value would have artificially lowered the mean and inflated the variance and resulting in some biologically important droughts to have been overlooked (i.e., fewer years would have been scored as drought years because of the inflated variance). The results from this analysis indicated that there is considerable variability in the spatial extent of droughts among the most frequently used habitats (Table 8-1). Some years (e.g., 1971) were relatively widespread droughts and encompassed many of the major kite habitats within the range of Snail Kites in Florida, while others (e.g. 1985) were considerably more local in their spatial extent. Temporal Extent The temporal extent of a drying event should also be considered when evaluating the effects of droughts. This includes both within-year extent (i.e., the duration of the drying event) and the between-year extent (i.e., whether or not drying events occurred in consecutive years). The duration of a given drying event may affect the survival of apple snails. Darby et al. (1996b, 1996c) found that apple snails experiencing a drying event survived on average 3.9 weeks (2.2 weeks sd) and 3.9 weeks (3.1 weeks sd) for marshes in the upper St. Johns River Basin and Lake Kissimmee, respectively. Thus, not surprisingly, droughts of greater duration may have more of an impact on apple snail populations than those of short duration. Although research is lacking for post-drought

PAGE 160

145 recovery of apple snail populations, some less direct evidence (e.g., post-drought return rates) suggests that apple snail populations may take >1 year to recover to pre-drought population levels. Consequently, consecutive drought years also may slow the recovery process. I used Lake Okeechobee as an example for examining the temporal extent of droughts from the period of 1969-1995. Using the definition (above) of a drought year being any year in which the minimum annual stage was <1 sd below the mean (i.e., minimum annual stage <1 1. 17 ft MSL), Lake Okeechobee experienced six droughts during this period (Table 8-2 ). Of these six droughts, there were two single-year events and two consecutive-year events of two years each (i.e., four droughts). The duration of these droughts (i.e., number of days below 1 1.17 ft MSL) ranged from 8-139 days. Thus, 1981 was a severe drought in intensity (-1.89 sd), was a two-year consecutive drought (followed by the 1982 drought), and had the longest duration at this level of intensity. Although not shown in this table, 1981 also had a relatively large spatial extent. In contrast, the 1989 drought was relatively low intensity (-1.07 sd) and was of shorter duration at an intensity of 1 sd below the mean, although 1989 was also a two-yearconsecutive drought (followed by the 1990 drought). Critical Habitat Currently Designated Critical Habitat Critical habitat for the Snail Kite was determined in 1977 (Federal Register 42 [155]:40685-40688)(50 CFRCh. 1 [10-1-94 edition])(USFWS 1986). It includes (1) St John's Reservoir, (2) Cloud Lake Reservoir, (3) Strazzulla Reservoir, (4) western portions

PAGE 161

146 of Lake Okeechobee, (5) A R M. Loxahatchee National Wildlife Refuge, (6) Water Conservation Area 2A, (7) Water Conservation Area 2B, (8) Water Conservation Area 3 A south of Highway 84, and (9) a portion of Everglades National Park (Figure 8-4). Sykes (1983 a) suggested the need to supplement existing protected habitat with "scattered islands" of habitat throughout the kites' range as a means of reducing the impacts of droughts. Takekawa and Beissinger (1989) also suggested that the currently designated critical habitat was inadequate because it lacked sufficient "drought-related" areas. Beissinger and Takekawa (1983) and Takekawa and Beissinger (1989) further suggested that habitats could be divided into primary, secondary, and drought-related habitats. They defined primary habitat as having been used extensively over the previous decade, while secondary habitats received irregular or sporadic use (Takekawa and Beissinger 1989). They defined drought-related habitats as having been used as a result of dry periods. Additionally, some management agencies have described Snail Kite habitats relative to their function (e.g., "breeding habitat", "wintering habitat", or "drought habitat"). The Habitat Network My data are consistent with the views of these authors in that they indicate that the habitats used by Snail Kites in Florida are considerably more extensive than the currentlydesignated-critical habitat. Approximately 40% of the locations where I observed radiotransmittered Snail Kites were in habitats outside of the currently designated critical habitat. Additionally, 67% of the radio-transmittered adults used habitats outside of the critical habitat sometime during my observations. That only 67% of the birds I observed

PAGE 162

147 used areas outside of the critical habitat probably reflects the short duration of my observations. The number of areas used by kites was highly correlated with how long I observed the birds, and the duration of my observations for birds that had used areas outside of the currently designated critical habitat (x=356 days, n=106) was significantly shorter than for birds who had not used areas outside of the critical habitat (x=267 days, n=53)(t=-3.94, P< 0.001). Thus, I believe that most, if not all, birds that live an average adult life span will use, and probably require, habitat outside of the currently designated critical habitat. Consequently, I strongly agree with the conclusions of Sykes (1983a) and Takekawa and Beissinger (1989) that smaller wetlands peripheral to the currently designated critical habitat need to be protected if Snail Kite populations are to persist. However, my data suggest that the concept of primary, secondary, and drought-related habitats needs to be revised. I believe that the emphasis of research during nesting and the inaccessibility of many areas has greatly limited our understanding of habitats used by kites. I would certainly agree with Sykes (1983a) and Takekawa and Beissinger (1989) that some habitats (e.g., Lake Okeechobee) have been used more consistently than others. I would also agree that some habitats are used more extensively during droughts. However, Snail Kite use of all habitats in Florida, including "primary" habitats, exhibits considerable fluctuation depending on local (and statewide) conditions and may at times have very few, if any, birds present. For example, during the annual survey of 1994 (a high water year several years after the previous drought)(Bennetts et al. unpubl. data), no kites were found in WCA-2A (a "primary" habitat). During that same count 46 birds were counted at Lake Kissimmee (a "drought-related" habitat) and 43 were counted at

PAGE 163

148 Lostman's Slough of Big Cypress National Preserve. The latter area is not even listed as "primary", "secondary", or "drought-related" by Beissinger and Takekawa (1983) or Takekawa and Beissinger (1989); nor is it within the designated critical habitat (USFWS 1986). In addition, several of the "drought-related habitats (e.g., Lake Kissimmee and Loxahatchee Slough [West Palm Beach water catchment area]) described by Beissinger and Takekawa (1983) and Takekawa and Beissinger (1989) have had kites present in every year that they have been included in the annual count regardless of whether or not it was a drought year. Approximately 25% of all my locations of radio-transmittered birds were in habitats considered to be "drought-related" habitats by Beissinger and Takekawa (1983) and/or Takekawa and Beissinger (1989) even though drought conditions did not occur during my study. My data indicate that agricultural areas, agricultural and roadside canals, and peripheral marshes both seasonal and permanent (all described as "droughtrelated" areas) are used each year, primarily outside of the breeding season. I have even observed kites breeding in these "drought-related" habitats. Consequently, I believe that designations of specific habitats with respect to their function (e.g., "breeding habitat" or "drought habitat) can be misleading and under-represents the use of these habitats by Snail Kites. All habitats can be "drought" habitats, and most can be "breeding" habitats. For example, if drought conditions exist in the northern portion of the kite's range, then birds inhabiting those areas might be using WCA-3 A (a "primary" habitat) as a "drought" habitat. Similarly, when local conditions are good, birds will often breed in "droughtrelated" habitats. Designations based on relative use (e.g., "primary" and "secondary" habitats) can also be misleading because there are often substantial shifts in the distribution

PAGE 164

149 of birds over time. Consequently, land (and water) management planning may often use outdated information based on prior assessments of relative use. For example, most investigators working on Snail Kites today, agree that the habitat assessment of the Snail Kite Recovery Plan (U.S.F.W.S. 1986) is in need of revision even though it is only 10 years old. Based on my data on the movements of Snail Kites from 1992-1995, 1 suggest that the use of habitats can be better characterized as an extensive network (Figure 8-5). This network is comprised of habitats that may or may not be physically connected, but are connected through extensive movements of kites among the individual habitats. The network consists of local habitats ranging from large lake or marsh tracts to small agricultural ponds or canals. Some of the habitats may be used by hundreds of kites at a time, while others may be used by a single bird on rare occasions. My data, as well as the annual count and numerous anecdotal observations and reports, suggest that the use of local habitats is highly variable over time. Even areas that are used consistently for a number of years have dramatic fluctuations in the number birds present at a given time. The number of birds using a given local habitat also fluctuates seasonally within years. Areas previously described as "secondary" or "drought-related" habitats may have numerous birds present at a time when "primary" habitats may have few birds (not always during droughts). Consequently, I suggest that the patterns of use are better described as a continuum that is highly variable in space and time.

PAGE 165

150 Management of the Snail Kite in Florida: Beyond a Reductionist Paradigm In science and problem-solving in general, it is easiest to seek a simple explanation for an observed phenomenon. In a book on the psychology of understanding and managing complex systems, Doerner (1996) illustrates through many examples how great the temptation is to use simple although sometimes erroneous hypotheses, or to push a correct one far beyond the limits of its validity. Doerner describes experiments in which subjects were presented with artificial, but realistically complex ecological and economic systems. The subjects were charged with running these systems over a period of time, with the goal of maintaining their viability, by making decisions, observing the consequences, making further decisions, and so forth. Time and again the subjects grasped at a single simple hypothesis or factor that resulted in a failure of the system they were charged with managing. Scientific practice, heeding Ockham's razor, encourages a reductionist approach that is often successful when the scope of inference is limited. However, the initial success of an hypothesis that reduces a phenomenon to one or a small number of factors may blind one to other factors that may be important or become important later, particularly if inferences are extended beyond their original scope. Too often in science, reductive approaches are used in the hope of avoiding having to deal with the complexity of such systems. Doerner' s conclusions have been repeatedly exemplified in resource management. A recent review of 23 case histories of resource management that were considered failures indicated that the type of reductionist approach described by Doerner was a common pathology that contributed to these failures (Holling 1995, Holling and Meffe

PAGE 166

151 1996). In these cases, the emphasis was placed on finding a target variable and attempting to stabilize that variable such that normal fluctuation would not impose a threat to the resource of interest. The result was that key elements of these systems changed over time. These changes, however, generally either occurred so slowly as to escape detection, or they were not considered of major concern because the resource problem was narrowly defined in such a way as not to encompass these changing elements. The net result of this reductionist approach was that, over time, these ecosystems tended to become more spatially homogenous and, consequently, were less resilient to disturbance. Here, we present a case history of endangered species management for the Snail Kite in Florida that we believe strongly parallels this type of pathology, which we refer to as the "reductionist paradigm". The Reductionist Paradigm Descriptive accounts of snail kite populations during the late 1800s and early 1900s indicated that snail kites were relatively abundant in Florida, at least at some locations. Scott (1881) described snail kites as being "abundant" at Lake Panasofkee. Bailey (1884) quotes from a letter he received from an egg collector, Mr. E.W. Montreuil, describing snail kites as being "found in numbers in the Everglades". Wayne (1895) described kites as "exceedingly common" on the Wascissa River in the Florida Panhandle. Later, Howell (1932) described "scattered flocks of a hundred or more" birds frequently having been found. By the 1920s and continuing through the 1950s, virtually all reports began to describe Snail Kites in Florida as declining or rare (Howell 1932, Sprunt 1945,

PAGE 167

152 Wachenfeld 1956). These declines were attributed primarily to widespread drainage that occurred throughout Florida (Howell 1932, Sprunt 1945, Sykes 1983a, Bennetts et al. 1994). Since that time Snail Kite numbers have rebounded from what was considered to be a low of about 25-50 pairs (Sprunt 1945) to a current estimate of approximately 1,500 birds (V.J. Dreitz, unpubl. data). This rebound has been largely attributed to the creation of several water conservation areas, whose primary purpose was flood protection and water storage, but which also provided long-hydroperiod marshes in areas where they had been substantially reduced by drainage (e.g., Sykes 1983b). Snail Kites in Florida have been monitored since 1969 via an annual survey (Sykes 1979, Rodgers et al. 1988, Bennetts et al. 1994). Several authors have viewed the survey as a census of population size, thus, enabling inferences to be made about changes in population size, particularly in response to water levels (e.g., Sykes 1979, 1983a, Snyder et al. 1989, Beissinger 1995). For example, the difference between the 1980 and 1981 surveys has been widely cited as an estimate of mortality during a widespread drought in 1981 (e.g., Beissinger and Takekawa 1983, Beissinger 1986, Takekawa and Beissinger 1989). Reproductive success also has been reported to be influenced by water levels. For example, Snyder et al. (1989) presented a summary of nesting data over an 18-year period and concluded that nest success was lower during years of low water; however, at least part of this effect was due to use of different nest substrates during lowwater years Beissinger (1986) also reported that a large proportion of Snail Kites did not attempt to nest during a drought in 1981, and that nest success was low both in 1981 and 1982.

PAGE 168

153 More recently, Beissinger (1995) presented a population viability analysis (PVA) for snail kites. This model used the annual survey as a primary basis for verification and as a source of data for its most sensitive parameter, survival during drought years. The model focused on the occurrence of droughts as the primary environmental effect for predicting changes in population size. Projections from the model indicated that the Florida population of snail kite kites would not be viable unless the interval between successive droughts exceeded 4.3 years. The cumulative effect of these observations, regardless of the validity of specific approaches, was a perception of water levels being the primary influence on the Florida snail kite population (Beissinger and Takekawa 1983, Sykes 1983a, 1983b, Takekawa and Beissinger 1989). In particular, droughts have been portrayed as demographic catastrophes in which both survival and reproduction plummet (Takekawa and Beissinger 1 989, Beissinger 1 986, 1 995). Management recommendations under the existing paradigm have focused primarily on increasing the interval between droughts and the maintaining permanent water levels (e.g., Stieglitz and Thompson 1967, Martin and Doebel 1973, Beissinger 1983, 1995), although a need for refugia during droughts also has been recognized (Sykes 1983 a, Takekawa and Beissinger 1989). I suggest that this management paradigm closely parallels the pathology described above for failed resource management in other ecosystems. A target variable was identified, in this case water levels. The proposed solution for maintaining viable populations of snail kites was then to minimize the occurrence of droughts. Thus, stabilization of water levels was intended to buffer the population from potential

PAGE 169

154 demographic consequences of hydrologic variability. However, as in other systems, this approach has ignored effects beyond the species of interest, particularly plant communities, and hence the long-term indirect effects on the snail kite through other changes to the ecosystem. Conflicts and Limitations of the Existing Paradigm There are a number of conflicts inherent in the existing paradigm that warrant consideration. For example, under the existing paradigm, conflicts have been perceived among species with differing hydrologic requirements. Suitable conditions for snail kites have been portrayed as being in conflict with species such as wood storks (Mycteria americana XGraham 1990) or white-tailed deer (Odocolieus virginianus) (Beissinger 1983), even though these species have coexisted for centuries. There also are conflicts with native flora. For example, virtually none of the plant species used as nest substrates by snail kites can tolerate continuous flooding (Craighead 1971, Gunderson et al. 1988), which is often asserted to be a requirement of suitable habitat. The primary graminoid species that comprise foraging habitat also cannot tolerate permanent inundation, even though they can tolerate longer inundation than most woody species (Gunderson 1994). There are other limitations of the current paradigm. Assessments of the effects of drought (e.g., Snyder et al. 1989, Beissinger 1995) have made no distinction regarding the spatial extent or severity of a given drought, even though the effect on survival or reproduction of kites probably depends greatly on the spatial extent of the drought (Bennetts and Kitchens 1997a, 1997b). Management recommendations under the existing paradigm also have largely overlooked the influence of stabilizing water levels on other

PAGE 170

155 species within central and southern Florida, and even on habitat quality for snail kites beyond a limited time frame. Recent assessments (e.g., Beissinger 1995) have merely assumed that habitat quality will not change in response to changing hydrologic regimes despite substantial evidence to the contrary. These assessments have focused only on a lower limit of drying frequencies, which implies that an upper limit does not exist, or that it is of relatively little concern. In contrast, there is a considerable body of evidence regarding the tolerances of the plant species, that comprise suitable habitat, to prolonged inundation (e.g., Craighead 1971, U.S. Department of Interior 1972, McPhereson 1973, Worth 1983, Gunderson 1994). Similarly, the occurrence of drought is the only hydrologic effect considered. This implies that factors such as water depth are not important; again, all evidence is to the contrary (e.g., Steiglitz and Thompson 1967, Sykes 1987, Bennetts et al. 1988). The spatial distribution of birds, and how that distribution has changed over time, provides important insights about how kites have responded to changes in habitat quality. However, the spatial distribution of birds has been virtually ignored under the existing paradigm. The Importance of Spatial and Temporal Scales One of the possible reasons for failures in resource management is a lack of attention to all of the spatial and temporal scales that are relevant to a problem. There is a substantial body of ecological literature describing the linkages between patterns and processes and how these linkages are related to scale (e.g., Wiens 1989a, Holling 1992). I suggest that a lack of consideration of some relevant scales has been an inherent feature of the reductionist paradigm of snail kite management. Recommendations have focused on

PAGE 171

156 stabilizing the conditions that seem most suitable for kites at a given time, while ignoring the consequences of that action over longer time scales. Similarly, management recommendations have focused on a limited spatial scale by considering the impacts of management to individual wetlands (e.g., Bennetts et al. 1988), while ignoring effects to the entire network of habitats that comprise the system as a whole. I suggest that explicit incorporation of scale can, at least conceptually, resolve many of the conflicts and limitations of the existing paradigm. I suggest a conceptual framework, described below, that goes beyond the limitations of the existing paradigm to enable a more comprehensive evaluation of snail kite population dynamics. This framework involves the interaction between spatial and temporal scales of snail kite habitat and population dynamics and their long-term persistence in a fluctuating environment. Although many of the features I describe in this conceptual model have a strong empirical basis, my intention here is to represent my ideas as hypotheses to be tested, rather than as a verified model. The Dynamic Landscape Hypothesis I agree with proponents of the current paradigm that water levels are a critical component of snail kite habitat and viability. However, I argue that favorable habitat for the snail kite depends on three different aspects of hydrology that can be related to different temporal scales. When these temporally explicit hydrologic factors are considered across a broad spatial scale, a dynamic pattern emergences that I call the "dynamic landscape hypothesis". The first scale is that of current water levels (depth). The empirical relationship between snail kite use of a given habitat and water depth has been well recognized and has been illustrated by the distribution of nests or foraging birds

PAGE 172

157 with respect to water depth (e.g., Steiglitz and Thompson 1967, Sykes 1987, Bennetts et al. 1988, Bennetts and Kitchens 1997a)(Figure 8-6). The response of snail kites to changing water depth also can be seen in shifts in spatial distribution. For example, the spatial distribution of nesting kites within Water Conservation Area (WCA) 3 A, a 237,000 ha impoundment used extensively for nesting during the past three decades, was similar for 1992, 1993, and 1994 (Figure 8-7). During the 1995 breeding season water depths were at record high levels throughout the Everglades as a result of tropical storm Gordon the previous fall (Bennetts and Kitchens 1997a). The distribution of nesting kites within WCA-3A shifted dramatically to the north during 1995, to an area of higher elevation than had been observed during the previous 3 years. When water levels receded the following year, the distribution of nesting birds shifted back to the south where they had been prior to the high water. Similar shifts also have been observed in other areas. Water depth probably is important for snail kites because of how it affects their primary food source, apple snails (Pomacea paludosaV particularly with respect to snail behavior. Water depths that are too shallow may restrict the movement of snails, as submergent vegetation is densely compacted within the water column (Darby et al. 1997). Shallow water during certain seasons also may result in water temperatures rising above the tolerance level of snails (Darby et al. 1997). Water that is too deep may lack sufficient oxygen to support apple snails (Hanning 1978) and/or lack sufficient vegetation that would enable snails to climb near the surface, where they are available to kites (Darby et al. 1997).

PAGE 173

158 The second hydrologic factor is the time since a dry-down at a given location. This factor contributes both to apple snail population dynamics and to the maintenance of plant communities Comprising Snail Kite habitat. Florida apple snails are aquatic and have a limited capacity to survive dry conditions (Little 1968), although the timing of drying may be more important to the overall population dynamics than just the occurrence of drying (Darby et al. 1997). However, drying events do result in periodic reductions in the availability of snail kite food resources regardless of whether or not there are substantial effects on snail survival. Based on preliminary comparisons of numbers of kites counted during the annual survey before and after drying events in several wetlands, relative habitat quality on average is about 50% of pre-drying conditions the year following the drying event, 85% two years following and fully recovered by three years. Although the occurrence of drying events may affect apple snail populations, the absence of drying results in changes in plant communities. Observable changes in the plant communities in the absence of drying have occurred by about 5-6 years (Ager and Kerce 1970, U.S. D.I. 1972) and some plant communities comprising kite habitat can be replaced by other communities in as little as 9-10 years (Milleson 1987). The relationship between relative habitat quality for snail kites and time since a drying event can be shown with respect to the time since drying using a conceptual model (Figure 8-8). The third hydrologic factor is a cumulative effect of the longer time scale pattern of repeated drying events. In particular, the frequency of drying events is expressed as a "hydrologic regime" and is measured as long-term hydroperiod (the proportion of time an area is inundated). This long-term pattern is the primary hydrologic scale at which plant

PAGE 174

159 communities are regulated; although vegetation is also regulated by still slower processes that affect climatic regimes and sea-level rise (Gunderson 1994). Although rapid degradation of habitat occurs if a site is kept inundated, most sites experience drying at intervals less than that which would result in a direct transition of a plant community. Habitat changes often occur slowly and incrementally, with periods of at least partial rejuvenation resulting from periodic drying. Because of the extreme lack of topographic relief across the central and southern Florida wetland landscape, relatively small changes in elevation correspond to relatively large changes in hydrology. Consequently, differences of a few centimeters in elevation can have profound effects on plant communities and ultimately on the quality of habitat for kites (Figure 8-9). The response of Snail Kites at this scale also can be illustrated by changes in their spatial distribution over longer time scales (Figure 8-10). Available evidence suggests that suitable conditions at each of these hydrologic scales is necessary, but none is sufficient by itself to constitute suitable habitat for snail kites. The hydrology at each of these scales regulates a different aspect of the environment important to snail kites. In combination, these factors regulate (1) the behavior of apple snails, and consequently their availability to kites, (2) apple snail population dynamics, and (3) plant community change. Thus, if a given location has the appropriate water depth, but has experienced a drying event within the past few months, the apple snail population may be too low to be used by kites. Similarly, if a location has been inundated for a suitable period of time, but water depth is too high or low, the availability of prey may be unsuitable for kites, and so on. Thus, I hypothesize that an

PAGE 175

160 indication of overall habitat quality can be determined as an alignment of suitable conditions at each of these three scales. This alignment can be viewed as a dynamic "window" of hydrologic conditions at a given point in time and space in which Snail Kites occur (Bennetts and Kitchens 1997a). This window represents the combined effects of hydrology from all time scales at a given point in space and time. The shifts in distribution described above illustrate the response of kites to this window. The area used by kites in recent years is not the longest hydroperiod within WCA3 A. Thus, with respect to the hydrologic regime, the current "window" is a relatively shorter hydroperiod portion of a long-hydroperiod wetland (Figure 8-11), although the window has and will shift over time. A quite different window exists in the Stairstep Unit of Big Cypress National Preserve over this same time period that further illustrates my concept in a shorthydroperiod wetland. Wet prairie habitats within the Stairstep Unit are generally shorter hydroperiod than those which occur in WCA3A. Although birds have used this area regularly during the non-nesting season for foraging, I had no indication of nesting activity prior to 1995. This is not surprising because, except for the very wettest portions, these prairies dried out for short periods of time during most years. In contrast to WCA3A, the areas used by kites (primarily in Lostman's, Dixon, and East Sloughs were generally the longer hydroperiod habitats within this wetland (Figure 8-12)). During the nesting season of 1995, 1 observed 24 nests in this area. Thus, our window shifted from suitable foraging habitat during the non-nesting (rainy) season, to suitable nesting habitat during this high water event (Figure 8-11). During 1996 when more typical spring dry downs occurred only 8 nests were found in this region. Thus, the window was shifting back down. My

PAGE 176

161 main point from these examples is that not all Snail Kite habitat needs to be managed for some "optimal" hydroperiod. Spatial and temporal variation is an integral part of the Florida wetlands landscape and Snail Kites are well adapted to this variability. If all areas were managed as long hydroperiod wetlands, then Snail Kites would have limited habitat available during high water events. Similarly, if all areas were managed as short hydroperiod wetlands, then kites would only have habitat during high water events. I believe that it is this mosaic of hydrologic regimes and local conditions that enable kites to have habitat available during a variety of hydrologic conditions. Persistence of Snail Kites in a Dynamic Landscape The second element of my dynamic landscape hypothesis was initially described (Bennetts and Kitchens 1997a, 1997b) and is based on the concept that persistence of the Florida population of Snail Kites is enhanced more by spatial extent of habitat than by prolonging local inundation. I agree with Beissinger (1995) that there is a threshold mean periodicity of droughts, below which kite population will not be sustained. However, this applies more to droughts that affect the whole snail kite range, than to more localized drying. In fact, periodic disturbance events such as fire, hurricanes, and local and regional droughts are integral parts of southern Florida's landscape patterns (Davis et al. 1994). In virtually every ecosystem where disturbance processes have been markedly reduced, there has been a subsequent realization of their ecological importance (Pickett and White 1985). In southern and central Florida droughts occur at periodic intervals of about 5-10 years (Thomas 1974, Beissinger 1986, Duever et al. 1994). However, like most disturbance processes, the frequency and spatial extent of such events are not independent (Delcourt

PAGE 177

162 et al. 1983). Rainfall patterns across Florida are quite variable and localized drying events occur at a relatively high frequency, while widespread droughts that encompass all, or most of the snail kite's range in Florida, occur much less frequently (Mc Vicar and Lin 1984, Duever et al. 1994, Bennetts and Kitchens 1997a, 1997b). It is these widespread droughts that have the greatest impact on the snail kite population. When a localized drying event occurs (e.g., the "drought" of 1985), kites usually are able to escape its effects by moving to a another location. However, during a widespread drought refugia are less available (Takekawa and Beissinger 1989) and a numeric response (i.e., change in survival and/or reproduction) becomes increasingly likely (Bennetts and Kitchens 1997a, 1997b)(Figure 8-13). Thus, periodic local drying, necessary to maintain plant communities and thus kite habitat, occurs without catastrophic effects on the snail kite population. These local drying events, at natural frequencies, should be considered as an essential component of a functioning ecosystem, rather than as catastrophic events requiring stabilization. Stabilization results in a slow but steady conversion of wetlands to a more homogeneous aquatic state, degrading their habitat value for snail kites. The heterogeneity of rainfall patterns throughout the kite's range plays a crucial role in enabling snail kites to persist in a dynamic environment. Because of the ability of kites to escape the effects of local drying events by moving, demographic consequences of most such events are buffered. Consequently, conservation of habitat over a broad spatial extent (e.g., in several watersheds) that encompasses natural heterogeneity of rainfall patterns is essential, if refugia are expected to be available during droughts. Reducing the spatial extent of habitat, reduces the mechanism for resilience of the population to local

PAGE 178

163 drying events. Others have recognized a need for drought refugia (e.g., Sykes 1983a, 1983b, Beissinger and Takekawa 1989), but the importance of the spatial pattern of such refugia has been largely overlooked. Conclusions I am not arguing for a more spatially and temporally complex description of snail kite dynamics purely for the sake of complexity. This framework is essential to understand kite dynamics and to make long-term predictions and prudent management decisions. The principle of parsimony suggests that a model should contain enough factors to describe the system of interest and to eliminate excessive parameters that do not significantly contribute to an understanding of that system (Burnham and Anderson 1992, Lebreton et al. 1992). It is my belief that the existing reductionist paradigm goes beyond parsimony and has resulted in an inadequate understanding of snail kite habitat and population dynamics by virtue of its narrow view of spatial and temporal time scales. It is extremely unlikely that regulating mechanisms inferred from studies of very limited spatial or temporal scope can be successfully applied to predict system-wide behaviors over long periods of time. This limitation on scope of inference is a basic principle of research design that is often overlooked when managers are seeking solutions to guide their management actions. It is easy to overlook subtle, but critical, vegetation changes that occur at a temporal scale of one or two decades, when one is looking at year to year changes in kite numbers. Reversing some the negative effects of stabilization also may be extremely difficult and require up to several decades.

PAGE 179

164 Variability, both temporal and spatial, is and always has been, an intrinsic feature of the environment of snail kites in southern and central Florida. Temporal variability of the individual wetlands that make up the range of the snail kite likewise is an intrinsic feature and is not in itself a danger to this species. In fact, it is this variability that likely enables the coexistence of species (e.g., snail kites and wood storks) with seemingly different hydrologic requirements. These species merely don't tend to occur at the same location at the same time. Thus, I believe that the reduction of natural variability, through the reduction of natural fluctuations in water level or hydroperiods, will ultimately erode the landscape and biotic diversity and will ultimately reduce the value of individual wetlands as habitat for snail kites. The temptation to "freeze" in place snail kite habitat through artificial ponding may be stimulated by a narrow reductionist view of the wetlands in southern and central Florida that focuses on the temporal frequency of droughts, without considering spatial extent and heterogeneity, and the snail kite's adaptation to a spatially and temporally varying environment. Under an historic pattern of localized droughts, snail kites have been able to persist through movements from poor to good habitat. As a nomadic species, the snail kite is behaviorally adapted to this mechanism of individual survival. Because the kite makes frequent flights among wetland areas in its range, it can quickly recolonize habitats that it abandoned because of temporarily bad conditions when local conditions change for the better. This concept of snail kite adaptation to a spatially and temporally varying environment (or a "dynamic landscape") is well substantiated by data on its natural history. It is also consistent with theory that has been developed for many other species

PAGE 180

165 (den Boer 1968, 1971, 1981, Delcourt et al. 1983, Wiens 1989a, 1989b, Holling 1992). Management policies are beginning to recognize the ecological importance of natural variability instead of trying to fix a static environment. Regeneration of habitat through fire is now part of the long-term environmental management in many ecosystems. Managers have come to recognize fire as an important process necessary for the functioning of these ecosystems. A similar perspective is now needed such that natural disturbance processes can be incorporated into management of the snail kite and other species of southern and central Florida. There are at least two crucial management implications of the dynamic landscape view for the snail kite. The first is that artificial attempts to create stable habitat by reducing variability will be harmful in the long run. The second is that the focus of management should be on maintaining the existence of as much as possible of the diverse and geographically distributed network of wetland habitat areas for the snail kite. The primary danger to the Snail Kite is from severe system-wide droughts. Droughts that affect the entire range of the snail kite are rare (about once every 10-20 years), because climatic conditions are not usually correlated over such a large area (Bennetts and Kitchens 1997a). However, reduction in the range of the snail kite would increase the probability of droughts affecting the entire range. Thus, the management emphasis, in my view, should be strongly oriented to conserving suitable wetlands across a wide geographic area of southern and central Florida.

PAGE 181

C/5 r "5 S ^ 0) c/3 £ "S fa 1 5 S'i o 2 1 on on C O E 9 P 0/5 t t§ 1 <£ Esc 3 4> 5 S a3 T3 '-t; § I & § ? eg B S u — E Jg if 1 cm.*; O A > '£ TD J, •g Sz 3 4> sis B 1 111 ,S — < € I o o S S .2 • _: 00 -c | •2 S H .9 3 c M C o o H w 00 (N o IT) i ir> ri O d so On ro CN O d 1 d 1 no o cn r o d — 1 o 1 On i no NO ^ rd — On o ro o ~ 3 o. 00 CN oo 00 en d On. o o o O -h o ~ 00 ~ 00 "\ SO 9 d d CN CN SO d d 00 o o NO SO O so o o o o so d m ^ so — ^ in ; (N m CN On o" ~ — .' d d d O On d d o — in On ^ d d d CN 00 so m d — < ^ ^ ON O0 C^. CN SO d 9 d d On 00 ON ON On O ~ 00 00 On On CN 00 ON m oo ON m no t — oo oo oo oo On On On On

PAGE 182

167 O X o H w b x o H 2 t/3 w a o x o < < u h— I CQ NO cs 1 1 o 00 in o — i >1 NO CS t~fl o tsi O oo oo m 1t o o o o o o m (N On m m >— < O O -H On m m r<"> On in no NO 9 O ~ on r\o r~00 ~ vo P 9 — o I h On d o oo o ~ ^ H CN NO 9 9 n 2 2 £ o On d —J F-i o o CN — m oo oo 9 d .bp — (N r-) 3 ^ a 8 8 J 5 -< ? 1 1 Ik. Ik. ii •B -s: 1 5 S ^ J 'o 8 •3 N i s I 0 ^ S .1 I, #3 -V ft ^3^ roe \ —

PAGE 183

168 Table 8-2. The number of days that water stage was > 1 standard deviation below the average minimum stage for a 10-gauge average from Lake Okeechobee for each year from 1969-1994. This corresponds to a stage of < 1 1.17 ft MSL. The intensity for a given drought year is shown as the number of standard deviations below the mean. Year Drought Intensity No. Days 1969 No 1970 No 1971 Yes -1.55 88 1972 No 1973 No 1974 Yes 1 Co -1 1 14 1975 Ma iNU Nn 1NO 1 077 17 / / INO 1 078 17/0 i>o 1979 17/7 1NU 1980 Nn 1981 -1 88 -l .00 1J7 1982 I Co 1 1.J7 1 1Q Ijy 1983 J. 70 J Nn 1984 1 70*T Nn 1 no c 1985 No 1986 No 1987 No 1988 No 1989 Yes -1.04 8 1990 Yes -1.45 85 1991 No 1992 No 1993 No 1994 No

PAGE 184

169 3.0 2.5 3 .5 2.0 Highest Ground Elevation in WCA-3A Ground Elevation at Gauge 3-28 Lowest Ground Elevation in WCA-3A BS BS B S BS B ~i i i i i i i i i i i i i i i i i i i i r 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 Year Figure 8-1. The minimum annual water stage for gauge 3-28 in Water Conservation Area 3 A (WCA-3A) for the period of 1968-1988. Shown for reference are the minimum and maximum ground elevation in WCA-3 A and ground elevation at the 3-28 gauge. Points mark with an "S" were years identified by Snyder et al. (1989) as drought years and those mark with a "B" were identified by Beissinger (1995) as drought years.

PAGE 185

170 5.0 £ 4.5 4.0 3.5 .= 3.0 Highest Elevation of Littoral Zone BS D B3 Lowest Elevation of Littoral Zone D BS 2.5 n — i — i — i — i — i — i — i — i — i — i — i — i i i i i i i i r 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 Year Figure 8-2. The minimum annual water stage for a 10-gauge average at Lake Okeechobee for the period of 1968-1988. Shown for reference are the minimum and maximum ground elevation for the littoral zone at Lake Okeechobee (based on Pesnell and Brown [1977]). Points marked with an "S" were years identified by Snyder et al. (1989a) as drought years and those mark with a "B" were identified by Beissinger (1995) as drought years.

PAGE 186

171 X + 2SD X + 1 SD D X Minimum Stage D X-1 SD X-2SD n i i i i i i i i i i i i i i i i i i — i — i — i — i — i — i — r 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 Year Figure 8-3. The minimum annual water stage for gauge 3-28 in Water Conservation Area 3 A (WCA-3A) for the period of 1969-1994. Shown for reference are the average annual minimum stage, 1 standard deviation, and 2 standard deviations.

PAGE 187

Figure 8-4. The currently designated critical habitat identified in the Snail Kite Recovery Plan (after 50 CFR Ch.l [10-1-94 Edition]).

PAGE 188

173 Figure 8-5. South Florida showing the interwetland movements of individual radiotagged adult snail kites during 1992 and 1993. These movements illustrate a basic habitat network used by snail kites (also shown). We have shown only a limited subset of this network (and moments) to minimize cluttering, and because a complete synthesis of the peripheral habitats has not been done. The complete movements, and consequently the complete network, would include all movements and habitats used by kites throughout the state.

PAGE 189

174 0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 Water Depth at Nest Initiation (cm) Figure 8-6. The percentage of Snail Kite nests (N=745) that were initiated in each 10 cm water depth class. Data are from Bennetts et al. (1988), B. Toland (unpubl. data), and this study.

PAGE 190

175 Figure 8-7. The distribution of Snail Kite nests in Water Conservation Area 3 A during each year from 1992 through 1996. During 1995 this area experienced exceptionally high water levels as a result of Tropical Storm Gordon and the distribution of nesting kites shifted to higher elevations.

PAGE 191

176 SrmllRacoviy Habitat Degradation Q 0 1 2 3 4 5 8 7 8 S 10 11 12 13 Year Since Drying Event Figure 8-8. Conceptual model of relative habitat quality in relation to the time since a drying event at a given location. In the absence of a drying event (A), habitat quality initially increases as the apple snail population recovers, but declines after 5-6 years as the plant communities comprising nesting and foraging habitat begin degradation. If the drying event occurs too frequently (B), the apple snail population will have been unlikely to have recovered to its full potential. If drying events occur at longer intervals (C) then a cumulative process of slow and incremental degradation will occur as plant communities undergo transition.

PAGE 192

177 Figure 8-9. Primary plant communities and their corresponding species that comprise Snail Kite habitat in relation to elevation and a hydrologic gradient.

PAGE 193

178 Figure 8-10. The reported nesting distribution of nesting snail kites (shaded) in Water Conservation Area 3 A (WCA3A) from 1965 to present. Birds nesting in southeastern WCA3A during the 1992-1996 period were foraging primarily in Everglades National Park and the "Pocket" between the L-67A and L-67C levees, both of which have shorter hydroperiods than the nesting area.

PAGE 194

179 Long-Hydroperiod Wetland Hydrologic Window Current Conditions Intermediate Hydro-history Hydrologic Regime z 1 Relative Hydroperiod Short-Hydroperiod Wetland Hydrologic Window Current Conditions Intermediate Hydro-history Hydrologic Regime Relative Hydroperiod Figure 8-11. A conceptual hydrologic window for a long (e.g., WCA-3A) and short (e. Big Cypress N.P.) hydroperiod wetlands. This window can shift over time depending the hydrologic conditions at different scales.

PAGE 195

180 Figure 8-12. The distribution of Snail Kite nests in Big Cypress National Preserve during each year from 1992 through 1995 (no nests were observed from 1992-1994).

PAGE 196

181 Local Widespread Spatial Extent of Drought Figure 8-13. Hypothesized relationship between the spatial extent of droughts and whether the response by Snail Kites is likely to be behavioral (i.e., movement) or numerical (i.e., change in survival and/or reproduction).

PAGE 197

APPENDIX 1 ESTIMATES OF CUMULATIVE NATAL DISPERSAL (ijr), NUMBER OF ANIMALS AT "RISK" OF DISPERSAL DURING INTERVAL j (ij), AND STANDARD ERROR (SE) OF THE ESTIMATE FOR EACH STUDY YEAR Date 1992 1993 1994 r j Hr SE(^) Hf r j SE(^) 04/15 0.000 3 0.000 13 0.000 4 04/18 0.000 3 — 0.067 15 0.064 0.000 4 — 04/26 0.000 5 — 0.138 13 0.091 0.000 5 — 04/28 0.000 8 — 0.354 12 0.128 0.000 5 — 05/13 0.000 8 — 0.354 12 0.128 0.063 16 0.061 05/19 0.000 18 — 0.392 17 0.126 0.063 16 0.061 05/23 0.000 17 — 0.422 20 0.123 0.063 17 0.061 05/26 0.000 17 — 0.422 20 0.123 0.115 18 0.076 05/31 0.000 17 0.422 20 0.123 0.167 17 0.088 06/02 0.000 17 0.422 20 0.123 0.271 16 0.103 06/03 0.048 21 0.046 0.454 18 0.120 0.271 14 0.103 06/05 0.095 20 0.064 0.454 17 0.120 0.271 14 0.103 06/10 0.095 20 0.064 0.454 17 0.120 0.316 16 0.106 06/11 0.095 22 0.064 0.479 22 0.117 0.316 16 0.106 06/12 0.095 22 0.064 0.504 21 0.114 0.316 21 0.106 06/14 0.095 22 0.064 0.504 21 0.114 0.343 26 0.105 06/26 0.186 20 0.084 0.504 21 0.114 0.343 25 0.105 06/28 0.186 18 0.084 0.528 21 0.111 0.343 25 0.105 182

PAGE 198

183 Appendix 1. Continued Date 1992 1993 1994 f r j r j SE(ft r j 07/01 0.186 17 0.084 0.552 19 0.108 0.343 25 0.105 07/04 0.234 17 0.091 0.552 18 0.108 0.343 23 0.105 07/07 0.282 16 0.097 0.552 18 0.108 0.371 23 0.105 07/09 0.282 16 0.097 0.552 18 0.108 0.428 22 0.103 07/14 0.282 16 0.097 0.552 18 0.108 0.514 20 0.098 07/19 0.282 16 0.097 0.552 18 0.108 0.547 15 0.097 07/27 0.282 16 0.097 0.552 18 0.108 0.611 14 0.093 07/30 0.282 13 0.097 0.612 14 0.102 0.611 14 0.093 08/06 0.282 13 0.097 0.702 11 0.090 0.611 14 0.093 08/12 0.282 13 0.097 0.702 11 0.090 0.644 12 0.091 08/13 0.337 13 0.104 0.702 10 0.090 0.644 11 0.091 08/23 0.337 13 0.104 0.702 10 0.090 0.676 11 0.088 09/05 0.392 12 0.109 0.768 9 0.082 0.676 10 0.088 09/08 0.392 11 0.109 0.801 7 0.076 0.676 10 0.088 09/14 0.447 11 0.113 0.801 4 0.076 0.709 10 0.085 09/26 0.503 10 0.114 0.801 3 0.076 0.709 9 0.085 09/29 0.503 10 0.114 0.801 3 0.076 0.741 9 0.082 10/01 0.503 9 0.114 0.867 3 0.074 0.741 9 0.082 10/03 0.503 9 0.114 0.934 2 0.060 0.741 9 0.082 10/04 0.503 9 0.114 0.934 2 0.060 0.773 8 0.078 10/26 0.558 9 0.114 0.934 1 0.060 0.773 7 0.078 11/10 0.558 9 0.114 0.934 1 0.060 0.806 7 0.073 11/11 0.558 9 0.114 0.934 1 0.060 0.838 6 0.068 04/14 0.558 6 0.114 0.934 0 0.060 0.838 5 0.068

PAGE 199

APPENDIX 2 ESTIMATES OF CUMULATIVE NATAL DISPERSAL (tjr), NUMBER OF ANIMALS AT "RISK" OF DISPERSAL DURING INTERVAL j (i^), AND STANDARD ERROR (SE) OF THE ESTIMATE FOR EACH STUDY YEAR IN AREAS AFFECTED AND UNAFFECTED BY THE PREVIOUS DROUGHT Unaffected by Drought Affected by Drought Date Hi r i SEftf) Hi r i A SE (Hh Study Year 1992 04/15/92 0.000 1 0.000 -J J 04/19/92 0.111 9 0.105 0.000 13 — 04/21/92 0.222 8 0.139 0.000 13 {jouo/yZ ft 0.139 0.167 1 z ft IDS /\"T lf\A lf\^ 07/04/92 0.319 o 8 0.152 0.167 1U A 1 ftQ U. lUo f\H lf\H ICl*) U.417 0.158 0.167 iu ft 1 ftS U. lUo 08/13/92 0.533 5 0.164 0.167 9 A 1 AO 0.108 09/05/92 0.650 4 0.159 0.167 9 0.108 09/14/92 0.767 3 0.143 0.167 9 0.108 09/26/92 0.883 2 0.109 0.167 9 0.108 10/26/92 0.883 1 0.109 0.259 9 0.129 01/10/93 0.883 0 0.109 0.259 6 0.129 04/14/93 0.883 0 0.109 0.259 6 0.129 Study Year 1993 04/15/93 0.000 0 0.000 13 04/18/93 0.000 0 0.067 15 0.064 04/26/93 0.000 0 0.138 13 0.091 184

PAGE 200

185 Appendix 2. Continued Unaffected by Drought Affected by Drought Date r i SE (Uh 04/28/93 0.000 0 — 0.354 12 0.128 05/19/93 0.000 5 — 0.408 12 0.128 05/23/93 0.000 8 — 0.457 12 0.126 06/03/93 0.143 7 0.132 0.457 11 0.126 06/11/93 0.143 6 0.132 0.491 16 0.123 06/13/93 0.143 6 0.132 0.525 15 0.119 06/28/93 0.143 8 0.132 0.561 13 0.116 07/01/93 0.143 7 0.132 0.598 12 0.112 07/29/93 0.250 8 0.153 0.598 7 0.112 07/30/93 0.357 7 0.165 0.598 7 0.112 08/04/93 0.464 6 0.168 0.655 7 0.109 08/06/93 0.571 5 0.165 0.655 6 0.109 09/05/93 0.786 4 0.135 0.655 5 0.109 09/08/93 0.786 2 0.135 0.724 5 0.107 10/01/93 1.000 1 — 0.724 2 0.107 10/03/93 1.000 0 — 0.862 2 0.111 04/14/94 1.000 0 — 0.862 2 0.111 Study Year 1994 04/15/94 0.000 0 — 0.000 4 — 05/13/94 0.000 2 — 0.071 14 0.069 05/26/94 0.000 5 — 0.143 13 0.094 05/31/94 0.000 5 0.214 12 0.110 06/02/94 0.000 5 0.357 11 0.128 06/09/94 0.200 5 0.179 0.357 11 0.128 06/14/94 0.200 10 0.179 0.397 16 0.126 07/09/94 0.500 7 0.177 0.397 15 0.126

PAGE 201

186 Appendix 2. Continued Unaffected by Drought Affected by Drought Date 07/14/94 0.500 5 0.177 0.518 15 0.119 07/19/94 0.600 5 0.167 0.518 10 0.119 07/27/94 0.600 4 0.167 0.614 10 0.113 08/12/94 0.600 4 0.167 0.663 8 0.109 08/23/94 0.700 4 0.152 0.663 7 0.109 09/14/94 0.700 3 0.152 0.711 7 0.103 09/29/94 0.700 3 0.152 0.759 6 0.097 10/04/94 0.800 3 0.130 0.759 5 0.097 11/10/94 0.800 2 0.130 0.807 5 0.088 11/11/94 0.800 2 0.130 0.855 4 0.078 01/22/95 0.800 1 0.130 0.855 3 0.078 04/14/95 0.800 1 0.130 0.855 3 0.078

PAGE 202

LITERATURE CITED Ager, H.A., and K.E. Kerce. 1970. Vegetation changes associated with water level stabilization in Lake Okeechobee, Florida. 24 th Ann. Conf. of S.E. Assoc. Game and Fish Comm. 338-351. Akaike, H. 1973. Information theory and an extension of the maximum likelihood principle. Pages 267-281 in B. Petrov and F. Czakil (Eds ), Proc. 2ndlnt. Symp. Inf. Theory. Akademiai Kiado Budapest. Bailey, H.B. 1884. Breeding habits of the Everglade Kite. Auk 1:95. Beissinger, S.R. 1983a. Nest failure and demography of the Snail Kite: effects of Everglades water management. Annual Report to the U.S. Fish and Wildlife Service. 22 pp. Beissinger, S.R. 1983b. Hunting behavior, prey selection, and energetics of Snail Kites in Guyana: consumer choice by a specialist. Auk 100:84-92. Beissinger, S.R. 1986. Demography, environmental uncertainty, and the evolution of mate desertion in the Snail Kite. Ecology 67: 1445-1459. Beissinger, S. R. 1988. The Snail Kite. Pages 148-165 in R.S. Palmer (ed ). Handbook of North American Birds. Volume IV. Yale University Press, New Haven, CT. Beissinger, S.R. 1995. Modeling extinction in periodic environments: Everglades water levels and Snail Kite population viability. Ecological Applications 5:618-63 1. Beissinger, S. R, and J. E. Takekawa. 1983. Habitat use and dispersal by Snail Kites in Florida during drought conditions. Florida Field Naturalist 1 1 : 89106. Beissinger, S.R., A. Sprunt, and R. Chandler. 1983. Notes on the Snail (Everglade) Kite in Cuba. American Birds. 37:262-265. Bennetts, RE. 1993. The Snail Kite: a wanderer and its habitat. Florida Nat. 66: 12-15. Bennetts, R. E., M. W. Collopy, and S. R. Beissinger. 1988. Nesting ecology of Snail Kites in Water Conservation Area 3 A. Dept. Wildl. and Range Sci., Univ. Florida, 187

PAGE 203

188 Florida Coop. Fish and Wildl. Res. Unit, Tech. Rep. No. 31. Gainesille. Florida Bennetts, R. E., M. W. Collopy, and J. A. Rodgers, Jr. 1994. The Snail Kite in the Florida Everglades: a food specialist in a changing environment. Pages 507-532 in S.M. Davis and J. C. Ogden (eds ). Everglades: the ecosystem and its restoration, St. Lucie Press, Delray Beach, FL. Bennetts, R E. and W. M. Kitchens. 1992. Estimation and environmental correlates of survival and dispersal of snail kites in Florida. 1st Annual Progress Report. Florida Cooperative Fish and Wildlife Research Unit, University of Florida, Gainesville. 48 pp. Bennetts, R.E. and W. M. Kitchens. 1993. Estimation and environmental correlates of survival and dispersal of snail kites in Florida. 1993 Annual Progress Report. Florida Cooperative Fish and Wildlife Research Unit, University of Florida, Gainesville. 48 pp. Bennetts, R.E. and W. M. Kitchens. 1994. Estimation and environmental correlates of survival and dispersal of snail kites in Florida. 1st Annual Progress Report. Florida Cooperative Fish and Wildlife Research Unit, University of Florida, Gainesville. 41 pp. Bennetts, R. E. and W. M. Kitchens. 1997a. The Demography and Movements of Snail Kites in Florida. U.S. Geological Survey/Biological Resources Division, Florida Cooperative Fish & Wildlife Research Unit. Technical Report No. 56. Gainesville, Florida. Bennetts, R. E. and W. M. Kitchens. 1997b. Population dynamics and conservation of Snail Kites in Florida: the importance of spatial and temporal scale. Colonial Waterbirds. 20: 324-329. Brownie, C, J. E. Hines, J. D. Nichols, K. H. Pollock, and J. B. Hastbeck. 1993. Capture-recapture studies for multiple strata including non-Markovian transitions. Biometrics 49:1173-1187. Burger, L. W., M. R. Ryan, D. P. Jones, and A. P. Wywailowski. 1991. Radio transmitters bias estimation of movments and survival. Journal of Wildlife Management 55:693-697. Burnham, K. P. 1981. Summarizing remarks: environmental influences. Pages 324-325 in Estimating Numbers of Terrestrial Birds (C. J. Ralph, J. M. Scott, Eds.). Studies in Avian Biology 6. Allen Press, Lawrence, Kansas.

PAGE 204

189 Burnham K.P., and D R. Anderson. 1992. Data-based selection of an appropriate biological model: the key to modern data analysis. Pages 16-30 in D. R. McCullough and R.H. Barrett. Wildlife 2001 :Populations. Elsevier Applied Science, New York. 1 163pp. Cary, D. M. 1985. Climatological factors affecting the foraging behavior and ecology of Snail Kites ( Rostrhamus sociabilis plumbeus Ridgeway) in Florida. M.S. Thesis, Univ. Miami. 58pp. Cormack, R. M. 1964. Estimates of survival from the sighting of marked animals. Biometrika 51:429-438. Cox, D. R., and D. Oakes. 1984. Analysis of survival data. Monographs on Statistics and Applied Probability 21. Chapman and Hall, New York. Cox, D. R., and D. Oakes. 1984. Analysis of survival data. Monographs on Statistics and Applied Probability 21. Chapman and Hall, New York. Craighead, F.C. 1971. The trees of South Florida. Vol. 1., The natural environments and their succession. University of Miami Press, Coral Gables, FL. Darby, P. C, P. L. Valentine Darby, R. E. Bennetts. 1996. Spatial relationships of foraging and roost sites used by Snail Kites at Lake Kissimmee and Water Conservation Area 3 A. Florida Field Nat. 24:1-9. Darby, PC, P. L. Valentine Darby, RE. Bennetts, J.D. Croop, H. F. Percival, and W. M. Kitchens. 1997. Ecological studies of apple snails ( Pomacea paludosa. Say). Final Report prepared for South Florida Water Management District and St. Johns River Water Management District. Contract # E-6609, Florida Cooperative Fish and Wildlife Research Unit, Gainesville, Florida. Davis, S. M., and J. C. Ogden, Editors. 1994. Everglades, the ecosystem and its restoration. St. Lucie Press, Delray Beach, Florida, USA. 826pp. Davis, S.M. 1994. Phosphorus inputs and vegetation sensitivity in the Everglades. Pages 357-378 in S. Davis and J. Ogden (eds ). Everglades: ecosystem and its restoration. St. Lucie Press. Delray Beach, Florida. 848 pp. Delcourt, H.R., P A. Delcourt, and T. Webb EL. 1983. Dynamic plant ecology: the spectrum of vegetational change in space and time. Quaternary Science Reviews 1:153-175. Den Boer, P.J. 1968. Spreading of risk and stabilization of animal numbers. Acta Biotheoretica. 18:165-194.

PAGE 205

190 Den Boer, P.J. 1981. On the survival of populations in a heterogeneous and variable environment. Oecologia. 50:39-53. Doerner, D. 1996. The logic of failure: recognizing and avoiding error in complex situations. Addison-Wesley, Reading MA. Duever, M. J., J. F. Meeder, L.C. Meeder, and J. M. Mcollom. 1994. The climate of South Florida and its role in shaping the Everglades ecosystem. Pages255-248 in S.M. Davis and J. C. Ogden (eds.). Everglades: the ecosystem and its restoration, St. Lucie Press, Delray Beach, FL. 848 pp. Freiburg, M. W., and D. H. Hazelwood. 1977. Oxygen consumption of two amphibious snails: Pomacea paludosa and Marisa cornuarietis (Prosobranchia: Ampullariidae). Malacologia 16:541-548. Graham, F. Jr. 1990. Kite vs stork. Audubon 92:104-110. Gunderson, L.H. 1994. Vegetation of the Everglades: determinants of community. Pages 323-340 in S.M. Davis and J.C. Ogden (eds.). Everglades: the ecosystem and its restoration, St. Lucie Press, Delray Beach, FL. 848 pp. Gunderson, L.H. J R. Stenberg, and A.K. Herndon. 1988. Tolerance of five hardwood species to flooding regimes in south Florida. Pages 1099-1 1 1 lin R.R. Sharitz and J.W. Gibbons (Eds.),Freshwater wetlands and wildlife. Ninth Ann. Symp., Savannah River Ecology Laboratory, Charleston, S C. Hanning, G. W. 1978. Aspects of reproduction in Pomacea paludosa (Mesogastropoda: Pilidae). M.S. Thesis. Florida State Univ., Tallahassee. 149 pp. Hines, J. E. 1994. MSSURVIV Users manual. National Biological Survey. Patuxent Environmental Science Center, Laurel, Maryland. 21 pp. Holling, C. S. 1992. Cross-scale morphology, geometry, and dynamics of ecosystems. Ecological Monographs 6:447-502. Holling. C.S. 1995. What barriers? What bridges? Pages 3-34 in L.H. Gunderson, C.S. Holling, and S .S. Light. Barriers and bridges to the renewal of ecosystems and institutions. Columbia University Press, New York. Holling, C.S., and GK. Meffe. 1996. Command and control and the pathology of natural resource management. Conservation Biology 10:328-337. Hosmer, D.W., and S.L. Lemeshow. 1989 Applied logistic regression John Wiley and

PAGE 206

191 Sons, New York. Howell, A H. 1932. Florida bird life. Coward-McMann Inc. 579 pp. Johnson, D. H. 1995. Point counts of birds: what are we measuring? Pages 1 17-123 in Monitoring Bird Populations by Point Counts (C. J. Ralph, J. R. Sauer, and S. Droege, Eds.) U S D. A. Forest Service, Pacific Southwest Research Station General Technical Report PSW-GTR-149, Albany, California. Jolly, G. M. 1965. Explicit estimates from capture-recapture data with both death and immigration-stochastic model. Biometrika 52:225-247. Jolly, G. M. 1982. Mark-recapture models with parameters constant in time. Biometrics 38:301-321. Kaplan, E. L., and P. Meier. 1958. Nonparametric estimation from incomplete observations. Journal American Statistical Association 53:188-193. Krebs, J. R., J. C. Ryan, and E. L. Charnov. 1974. Hunting by expectation or optimal foraging? A study of patch use by chickadees. Anim. Behav. 22:953-964. Kushlan, J. A. 1975. Population changes of the apple snail (Pomacea paludosa) in the southern Everglades. Nautilus 89:21-23. Lancia, R. A, J. D. Nichols, and K. H. Pollock. 1994. Estimating the number of animals in wildlife populations. Pages 215-253 in Research and Management Techniques for Wildlife and Habitats (T. A. Bookhout, Ed ). The Wildlife Society. Bethesda, Maryland. Lebreton, J. D., K. P. Burnham, J. Clobert, and D. R. Anderson. 1992. Modeling survival and testing biological hypotheses using marked animals: a unified approach with case studies. Ecological Monographs 62:67-1 18. Legendre, S, and F. Clobert. 1995. ULM, a software for conservation and evolutionary biologists. Journal Applied Statistics 22:817-834. Link, W. A, and J. R. Sauer. 1997. Estimation of population trajectories from count data. Biometrics 53:63-72. Little, C. 1968. Aestivation and ionic regulation of two species of Pomacea (Gastropoda, Prociobranchia). Journal of Experimental Biology. 48:569-585. Loery, G, K. H. Pollock, J. D. Nichols, and J. E. Hines. 1987. Age-specificity of Black-

PAGE 207

192 capped Chickadee survival rates: analysis of capture-recapture data. Ecology 68:1038-1044. MacVicar, T.K., and S.S.T. Lin. 1984. Historical rainfall activity in central and southern Florida: average, return period estimates and selected extremes. Pages 477-509 in P.J. Gleason (ed.) Environments of South Florida past and present II. Miami Geological Society. 551 pp. Marks, J. S., and V. S. Marks. 1987. Influence of radio collars on survival of Sharptailed Grouse. Journal of Wildlife Management 51:468-471. Martin, T. W., and J. H. Doebel. 1973. Management techniques for the Everglade Kite, preliminary report. Proc. Ann. Conf. Southeastern Assoc. Game and Fish Comm. 27:225-236. McClary, A. 1964. Surface inspiration and ciliary feeding in Pomacea paludosa (Prosobranchia: Mesogastropoda: Ampullariidae). Malacologia 2:87-101. Mechlin, L. M., and C. W. Shaiffer. 1979. Net firing gun for capturing breeding waterfowl. U. S.D.I. Fish & Wildlife Service. Northern Prairie Wildl. Res. Center. Jamestown, N.D. 14 pp. Mertz, D. B. 1971. Life history phenomena in increasing and decreasing populations. Pages 361-399 in G.P. Patil, E C Pielou, and W E. Waters (Eds ), Statistical Ecology, Vol. 2, Sampling and Modeling Biological Population Dynamics., Pennsylvania State Univ. Press. University Park, PA. Milleson, J.T. 1987. Vegetation changes in the Lake Okeechobee littoral zone 19721982. Technical Publication No. 87-3. South Florida Water Management District. West Palm Beach, Fl. Nesbit, I.C.T. 1975. Selective effects of predation in a tern colony. Condor 77:221-226. Nichols, J. D. 1992. Capture-recapture models. Bioscience 42:94-102. Nichols, J.D. 1996. Sources of variation in migratory movements of animal populations: statistical inference and a selective review of empirical results for birds. Pages 147-197 in Population dynamics in ecological space and time. O.E. Rhodes, R.K. Chesser, and M.H. Smith (eds ). Univ. Chicago Press. Nichols, J. D., G. L. Hensler, and P. W. Sykes Jr. 1980. Demography of the Everglade kite: implications for population management. Ecological Modelling 9:215-232.

PAGE 208

193 Nichols, J. D. J. Bart, R. J. Lampert, W. J. L. Sladen, and J. E. Hines. 1992. Annual survival rates of adult and immature eastern population Tundra Swans. J. Wildl. Manage. 56:485-494. Nichols, J. D., C. Brownie, J. E. Hines, K. H. Pollock, and J. B. Hastbeck. 1993. The estimation of exchanges among poulations or subpopulations. Pages 265-279 in Marked individuals in the study of Bird Population^ J.D. Lebreton and P.M. North, Eds ). Birkhauser Verlag Basel Switzerland. Paton, P. W. C, C. J. Zabel, D. L. Neal, G. N. Steger, N. G. Tilghman, B. R Noon. 1991. Effects of radio tags on Spotted Owls. Journal of Wildlife Management 55:617-622. Pickett, S. T. A., and P.S. White (eds.). 1985. The ecology of natural disturbance and patch dynamics. Academic Press, New York. Pollock, K. H, S. R. Winterstein, C. M. Bunck, and P. D. Curtis. 1989. Survival analysis in telemetry studies: the staggered entry design. Journal of Wildlife Management 53:7-15. Pollock, K. H., J. D. Nichols, C. Brownie, and J. E. Hines. 1990. Statistical inference for capture-recapture experiments. Wildlife Monographs 107. 97 pp. Ricklefs, R. E. 1973. Fecundity, mortality, and avian demography. Pages 366-435 in D. S. Farner (Ed). Breeding biology of birds: proceedings of a symposium on breeding behavior and reproduction physiology in birds. National Academy of Sciences, Washington, D C. Rodgers, J.A, Jr., S.T. Schwikert, and A.S. Wenner. 1988. Status of the Snail Kite in Florida: 1981-985. Am. Birds 42:30-35. Rodgers, J. A. Jr., and P. W. Stangel. 1996. Genetic variation and population structure of the endangered Snail Kite in South Florida. Journal of Raptor Research. 30:111-117. Sakamoto, Y M. Ishiguro, and G. Kitagawa. 1986. Akaike information criterion statistics. KTK Scientific Publishers, Tokyo, Japan. 290pp. Savage, I. R. 1956. Contributions to the theory of order statistics~the two sample case. Ann. Math. Stat. 27:590-615. Scott, W. E. D. 1881. On birds observed in Sumpter, Levy, and Hillsborough counties, Florida Bull. Nuttall Ornithol. Club 6:14-21.

PAGE 209

194 Seber, G. A. F. 1965. A note on the multiple recapture census. Biometrika 52:249-259. Shibata, R. 1989. Statistical aspects of model selection. Pages 215240 in J.C. Williams (ed.). From data to model. SpringerVerlag, New York. Shields, W.M. 1984. Optimal inbreeding and the evolution of philopatry. Pages 132-159 in Swingland, I.R., and P.J. Greenwood (eds.) The ecology of animal movement. Clarendon Press, Oxford. Snyder, N. F. R., Beissinger, S. R., and R. Chandler. 1989a. Reproduction and demography of the Florida Everglade (Snail) Kite. Condor 9 1 :300-3 1 6. Snyder, N. F. R., Beissinger, S. R., and M. R. Fuller. 1989b. Solar radio-transmitters on Snail Kites in Florida. J. Field Ornithology 60:171-177. Spendelow, J. A, J .D. Nichols, I. C. T. Nisbet, H. Hays, G. D. Cormons, J. Burger, C. Safina, J. E. Hines, and M. Gochfeld. 1995. Estimating annual survival and movement rates of adults within a metapopulation of Roseate Terns. Ecology 76:2415-2428. Sprunt, A, Jr. 1945. The phantom of the marshes. The Florida Naturalist. 4:57-63. Stieglitz, W.O., and R.L. Thompson. 1967. Status and life history of the Everglade Kite in the United States. Special Sci. Rep.., Wildl. No. 109.,U.S.D.I., Bur. Sport Fisheries and Wildl.,Washington, D.C. Sykes, P.W., Jr. 1979. Status of the Everglade Kite in Florida, 1968-1978. Wilson Bull 91:495-51 1. Sykes, P. W., Jr. 1982. Everglade Kite. Pages 43-44 in CRC Handbook of census methods for terrestrial vertebrates. D.E. Davis (Ed ). CRC Press, Boca Raton, Florida. Sykes, P.W., Jr. 1983a. Snail Kite use of the freshwater marshes of South Florida. Florida Field Naturalist 1 1 :73-88. Sykes, P.W., Jr. 1983b. Recent population trends of the Snail Kite in Florida and its relationship to water levels. Journal Field Ornithology 54:237-246. Sykes, P. W., Jr. 1984. The range of the Snail Kite and its history in Florida. Bulletin Florida State Museum 29:21 1-264. Sykes, P.W., Jr. 1987a. Snail Kite nesting ecology in Florida. Florida Field Naturalist 15:57-70.

PAGE 210

195 Sykes, P.W., Jr. 1987b. Some aspects of the breeding biology of the Snail Kite in Florida. J. Field Ornithol. 58:171-189. Sykes, P.W., Jr. 1987c. The feeding habits of the Snail Kite in Florida, USA. Colonial WaterbirdslO: Sykes, P. W., Jr., J. A. Rodgers, Jr., and R. E. Bennetts. 1995. Snail Kite (Rostrhamus sociabilis). Pages 1-32 in The Birds of North America, No. 171 (A. Poole and F. Gill Eds.). The Academy of Natural Sciences, Philadelphia, and The American Ornithologists' Union, Washington, D C. Takekawa, J. E., and S. R. Beissinger. 1989. Cyclic drought, dispersal, and conservation of the Snail Kite in Florida: lessons in critical habitat. Conservation Biology 3:302-311. Thomas, T.M. 1970. A detailed analysis of climatological and hydrological records of South Florida with reference to man's influence upon ecosystem evolution. University of Miami, Coral Gables, FL. U.S. Department of Interior. 1972. A preliminary investigation of the effects of water levels on vegetative communities of Loxahatchee National Wildlife Refuge, Florida. U.S.D.I. Bureau of Sport Fisheries and Wildlife. 20 pp. Valentine-Darby, P. L., R. E. Bennetts, and W. M Kitchens. 1998. Seasonal patterns of habitat use by Snail Kites in Florida. Journal of Raptor Research (in press). Wachenfeld, A. W. 1956. Present status of the Everglade Kite. Linnaean News-Letter. 10:1. Walters, C. L. 1986. Adaptive management of renewable resources. MacMillan, New York. Walters ,C, L. Gunderson, and C. S. Holling. 1992. Experimental policies for water management in the Everglades. Ecological Applications 2: 189-202. Wayne, A. T. 1895. Notes on the birds of the Wacissa and Aucilla River regions of Florida. Auk 12:362-367. White, G. C. 1983. Numerical estimation of survival rates from band-recovery and biotelemetry data. Journal of Wildlife Management 47:716-728. White, G. C, and R.A. Garrott. 1990. Analysis of wildlife radio-tracking data. Academic Press Inc.,New York. 383 pp.

PAGE 211

196 Wiens, J. A. 1976. Population responses to patchy environments. Ann. Rev. Ecol. Syst. 7:81-120. Wiens, J. A. 1989a. Spatial scaling in ecology. Functional Ecology 3:385-397. Wiens, J.A. 1989b. The ecology of bird communities. Cambridge University Press, New York. Worth, D. 1983. Preliminary responses to marsh dewatering and reduction in water regulation schedule in Water Conservation Area-2A. Tech. Publ. 83-6. South Florida Water Management District. 63pp.

PAGE 212

BIOGRAPHICAL SKETCH Robert Edwin Bennetts was born in Concord, California on 19 March 1953. His parents are Dorothy Lee Bennetts and Stewart Bennetts. He received his elementary education at Oak Park and Los Ranchos Elementary. He completed his secondary education at San Ramon High School and Del Amigo High School, Danville, California in 1971. He attended two semesters at Feather River College, Quincy California in 1972 and 1973. He was employed with the U.S. Forest Service in California as a seasonal firefighter from 1973 through 1975, after which he gained career status. He remained with the U.S. Forest Service as a firefighter, where he worked on engine crews, interregional "hotshot" crews, and the California Smokejumpers, until 1980. In 1980 he entered the undergraduate program in wildlife biology at the University of Montana, where he completed is BA in zoology and wildlife biology in 1988. During this period he also worked on several research projects including work on Spotted Owls in Washington, Bald Eagles in Montana, migratory land birds in Mexico and Guatemala, waterfowl in Montana and Canada, Snail Kites in the Florida Everglades, and several projects relating to radio nuclide transport while an intern at the Idaho National Engineering Laboratory. Mr. Bennetts entered the graduate program at Colorado State University in 1988, during which time he conducted research on the effect of dwarf mistletoe on bird communities. He completed his MS in 1991. He entered the Ph.D. program at University of Florida in 1991 to study the demography and movements of Snail Kites. 197

PAGE 213

I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. Wiley M. jfitchens, Chair Associate Professor of Wildlife Ecology and Conservation I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. Crawford S. Holling ( Arthur R. Marshall, Jr., Professor of Zoology I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. v 'enneth M. Portier Associate Professor of Statistics I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. Frank J. Ma^otti Assistant Professor of Wildlife Ecology and Conservation I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. Ja/hes D. Nichols Associate Professor of Wildlife Ecology and Conservation

PAGE 214

This dissertation was submitted to the Graduate Faculty of the College of Agriculture and the Graduate School and was accepted as partial fulfillment of the requirements for the degree of Doctor of Philosophy. May 1998 Dean, College/of Agriculture Dean, Graduate School