H:
J H; 'P;::::t1"z '37;th it“ I ‘J":"":‘
5 I1‘[,{' 1:; I? :( W( ; %;3: 3; t: ' ‘
. Bk g? " "o ”.g ‘]i ”(13:9 <
l ) - '1'51'7§f§’~1'{"";:§;1
" 'i

‘. ’éVO“.. .g‘é“! '- .
;_ «new

     
     
  

   
 
 

 

 
 
  
  
   

       
  
 

  
    
  

- - .. .- ‘ 1‘
‘. ‘i..‘ “3 , 3104‘“ ~‘
m" 5P 3; .1, 11mm 't'fip
159%"? ' '. 1'1'II’:L! 14:1; "“4; I’. .;
:3; “I" it"? “H mm: "‘ ."'\" 527?; 323%
1:" ‘ I l

 

.
ligan
.. .

";3f ’ '
‘ ,4 ‘—
‘fi: ... o-

”. u-

,

   
 
 

  

    
   
 

    
  

   

   
     

'[ :E’Eizfifi'f
‘ .1 .:-' 3‘: $3. .
1. rmséa-x 4:. 1.
f 1.: 1 2 “if E; .414 _
s ‘ 335.3113

     

  
 
  
  
 
  
    

 
   
 

9
22:71?! ”'33,;
fi'?’ ’31?!"
r ”’3'"
£1": 2"}?
‘ - a;

ti”;

 
   

3 F03

é”; 2V3;

   
  
  

sin A” ‘
awtzzuis‘ E5. ..
1,

¢
n:.
M"

  
  

  
    

   
  
  
   
   
 
  

 
  

.‘ W... .
.._‘_- >vu A. ;. .
o

  
  

  
 

   
 
 

              
         

 
 
 
  
  
  
  
 

  

 

 

  

 

    

 

 

 
   
 

 

 
     
     
      

     

 

.gizkfi ;;’:]m . -g
. . y > ' ‘ OPT; ' ' a .
I. N?" "’ m ’ -o..:--..i;d 't’hg? 5'33 :213’ ~ .
" "i. ifldkfiul W2”; 'h’ Ehfifih . '37“; {égfififlfig'
= w w; u’f’fi“ 1’13?” ;; ‘ .; r221;
s" ‘ " 20431515,!er ';.i!"‘s‘ ‘f’i’i mm WW 'I' . NH!

- §g - 9m, -;; 51,“:‘H‘Nli’. "‘3‘; A f""’=~41"E-
:3 V} "ii£§:‘£€””’“;‘”§’i"i.quiz’gtmt"? ”’ "if 1521;; W“; 313‘?

. , ,;;~q . .‘a , .. . ' , U 2' ~ ;‘ in“ .259 "P :
’i i? §;;“;"";";;§ ‘ ”;}:~;;2’§;2§E‘;,’?§fi; '3’u"1;':u’ r53;
“ '1 ' W“: H Vim-m T‘s}; £41.13 {M m. J»
5’ ":2 "NW" 5 12m; : 5;, ’;-;‘;;’ WM" “his I innit 4.3;};

’1. ":Ergkhu ’fi ’15,; wig-’1" glam“! if” fig nil Au’ Jami” , "
E‘ .1; .' :mlJ-‘z i" ‘!’~’Jn':dk&3’ 'Ela"§d!....}iifl! @232“ M51. .

      

11-15513

,‘3. ’" :5 ”3

lllllllllllilllllilllHHHIllIIUHI‘HIIUI||||lll||||iIllili

31293 02058 6537

This is to certify that the

dissertation entitled

A DECISION SUPPORT SYSTEM
FOR WHITE-TAILED DEER MANAGEMENT

presented by

JIALONG XIE

has been accepted towards fulfillment
of the requirements for

Ph.D. degree in Fisheries and Wildlife

Jianguo Liu

 

Major professor

4

 

/ ,4 ”’2'“ , \\ 7
Date December 15,1999 k// ‘7 r ,

MS U is an Affirmative Action/Equal Opportunity Institution 04 2771

W? i 77

A DECISION SUPPORT SYSTEM FOR WHITE-TAILED DEER MANAGEMENT

By

Jialong Xie

A DISSERTATION

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Fisheries and Wildlife

1999

ABSTRACT
A DECISION SUPPORT SYSTEM FOR WHITE-TAILED DEER MANAGEMENT
By

Jialong Xie

Effective management of white-tailed deer (Odocoileus virginianus) can be
enhanced by the integration of wildlife science, modern technology, and wildlife
managers' expertise. The goal of this research was to integrate the population
modeling, landscape structure analyses, and socioeconomic considerations into

a decision support system for deer management in the Upper Peninsula (UP) of
Michigan.

This research consists of three interrelated studies. The first study was a
deer management options model (DeerMOM) designed to assess the effects of
harvest options on deer population size, sex and age structure. The simulation
results indicated that a high rate of buck harvest (similar to the current harvest
practices) resulted in high deer numbers with a low percentage of antlered bucks.
Moderate harvest of both sexes would control population growth and increase
the percentage of antlered bucks. Quality deer management objectives could be
reached only by lowering buck harvest rates while simultaneously increasing doe

harvest.

The second study was a landscape structure analysis of deer habitat in
the UP. Using 1991 Landsat Thematic Mapper imagery, the landscape was
classified into 30-, 21-, 8-, and 4-class of patches. Each of the 15 counties in the
UP was treated as a landscape and its landscape metrics were calculated using
these four classification schemes. The analysis indicated that deer density had
significant positive relationships with edge density, patch density, and landscape
Shape index even when different schemes were used. There were no Significant
positive relationships with patch diversity indices, contagion and interspersion
indices. These findings suggest that edges and spatial heterogeneity play
important roles in determining deer Spatial distribution at landscape scales.

The third study was a knowledge-based system for deer management
(DeerKBS). The main task of deer management was divided into 3 major sub-
tasks: assessments of population status, habitat conditions, and cultural carrying
capacity. Each sub-task was represented by a decision tree that related a
decision with relevant attributes; and each sub-task was developed and tested
separately. DeerKBS was able to capture and retain wildlife managers' expertise
and to standardize decision-making procedures, and could be an effective

communication tool for presenting the decision-making process to stakeholders.

Copyright by
Jialong Xie
1 999

Dedicated to my wife Hong Zhang for her love and support, and my daughter
Daisy for her delightful diversions

ACKNOWLEDGMENTS

I would like to thank Drs. Henry, Campa, Larry Leefers, Jack Liu, Jon
Sticklen, and Scott Winterstein for serving on my Guidance Committee and for
their valuable suggestions throughout this project. My deepest gratitude goes to
my Major Professor Jack for his time, patience, and encouragement. I am
indebted to his insightful comments to my manuscripts and advice to my personal
growth.

I would like to acknowledge the financial support of the Wildlife Division of
the Michigan Department of Natural Resource (MDNR) through Federal Aid in
Wildlife Restoration Act under Pittman-Robertson Project W—127-R.

I am grateful to Harry Hill at the Wildlife Division of the MDNR for his data,
insights, and meeting arrangements with other wildlife managers and biologists in
the MDNR. Without his help, this research would be much more difficult. Many
other MDNR experts helped this study in one way or another. Among them,
Robert Doepker, Michael Donovan, Edward Langenau, William Moritz, Dale
Rabe, and John Urbain deserve special recognition for their data and comments
to the programs that I developed. In addition, Carl Bennett, Jim Sikarskie, and
Stephen Schmitt provided valuable assistance in the beginning stage of this
project.

My appreciation is extended to Christopher Lepczyk and Daniel Rutledge
for their timely help with my conference preparation, critical comments on my
reports, and technical assistance with computers. Thanks are due to Kimberly

Mattson-Hansen, a former Master student at this Department, now a Ph.D.

vi

student at Virginia Tech for her data and comments on my deer population
model. Thanks also go to many fellow graduate students and faculty members
for their car-pooling to conferences and their teaching on fishing and target
shoofing.

I am grateful to Qiuyun Wang, Guijie Wang, Zhiyun Ouyang, Beicheng
Xia, and Li An for their friendship and help to my family, especially to our
daughter Daisy. Zachary Huang lured me into cyberspace and hosted my domain
free for years and deserves e-special thanks.

My deepest gratitude goes to my parents Pusheng Xie and Dingli Cai. I
would never forget the scene that my father saw me off at Shanghai lntemational
Airport, to where he traveled more than 1200 kilometers by train. He was very
worried that it might be his last chance to see me. Nine months later, he passed
away. My brothers Jianlong, Jinlong, and Lianglong Xie and sister Qing Xie took
good care of my parents and made it possible for me to finish my degree study
here. I am deeply indebted to all of them for their support and sacrifice
throughout my life.

From my wife’s viewpoint, deer is more important for me than ”dear”. Now
I Should thank my wife Hong Zhang for her love, patience, and support.

Finally, I thank my daughter Daisy. She brings me delightful diversions

from my computer modeling, a game she never likes to play.

vii

TABLE OF CONTENTS

LIST OF TABLES ................................................................................. x
LIST OF FIGURES ............................................................................... xi
CHAPTER 1: INTRODUCTION ................................................................ 1
CHAPTER 2: WHITE-TAILED DEER MANAGEMENT OPTIONS MODEL
(DeerMOM): DESIGN, QUANTIFICATION, AND APPLICATION .................... 8
2.1 METHODS ....................................................................................................... l
2.1 1 MODEL DESIGN ........................................................................................ 11
2.1 1.1 Simplicity .......................................................................................... 11
2.1.1.2Accuracy .......................................................................................... 12
2.1.1.3 Management orientation .................................................................. 12
2.1.2 MODEL STRUCTURE .................................................................................. 13
2.1.2.1 Female Sector ................................................................................. 13
2.1.2.2 Male Sector ...................................................................................... 15
2.1.2.3 Birth/Death Sector ........................................................................... 15
2.1.2.4 Harvest Sector ................................................................................. 15
2.1.2.5 Population Sector ............................................................................ 16
2.1.3 QUANTIFICATION OF THE MODEL ................................................................ 16
2.1.3.1 Reproductive rates ........................................................................... 16
2.1.3.2 Fetal sex ratio .................................................................................. 18
2.1.3.3 Mortality ........................................................................................... 18
2.1.3.4 Neonatal mortality ............................................................................ 22
2.1.4 MODEL INITIALIZATION ............................................................................... 22
2.1.5 MODEL TESTING ....................................................................................... 22
2.1.6 SIMULATION SCENARIOS ............................................................................ 23
2.2 RESULTS ..................................................................................................... 24
2.3 DISCUSSION ................................................................................................ 30
2.3.1 MODEL STRUCTURE .................................................................................. 30
2.3.2 QUANTIFICATION OF PARAMETERS .............................................................. 30
2.3.3 CURRENT MANAGEMENT PRACTICES ........................................................... 31
2.3.4 HARVEST RECOMMENDATIONS ................................................................... 31

CHAPTER 3: LANDSCAPE STRUCTURE AND SPATIAL DISTRIBUTION OF
WHITE-TAILED DEER .......................................................................... 33

3.1 METHODS .................................................................................................... 37

viii

3.1.1 IMAGE ACQUISITION, PROCESSING, AND CLASSIFICATION .............................. 37
3.

1.2 QUANTIFICATION OF LANDSCAPE STRUCTURE .............................................. 37
3.1.3 DEER DENSITY MEASURES ......................................................................... 41
3.1.4 RELATIONSHIPS BETWEEN DEER DENSITY AND LANDSCAPE METRICS ............. 42

3.2 RESULTS ..................................................................................................... 44
3.3 DISCUSSION ................................................................................................ 56

CHAPTER 4: A KNOWLEDGE-BASED SYSTEM FOR WHITE-TAILED DEER

MANAGEMENT (DeerKBS) .................................................................. 59
4.1 SYSTEM COMPONENTS OF DEERKBS .................................................... 63
4.2 SYSTEM DEVELOPMENT OF DEERKBS: AN OVERVIEW ........................ 64
4.3 DOMAIN PROBLEM DEFINITION ................................................................ 66
4.4 SELECTION OF KBS TOOL: XPERTRULE KBS ......................................... 68
4.5 KNOWLEDGE REPRESENTATION ............................................................. 69
4.5.1 KNOWLEDGE STRUCTURING ....................................................................... 69
4.5.2 KNOWLEDGE ACQUISITION ......................................................................... 70
4.5.3 KNOWLEDGE VALIDATION .......................................................................... 71
4.6 SYSTEM IMPLEMENTATION ...................................................................... 71
4.6.1 SUBTASK FOR ASSESSING POPULATION ....................................................... 72
4.6.2 SUBTASK FOR ASSESSING HABITAT ............................................................. 74
4.6.3 SUBTASK FOR ASSESSING CULTURAL CARRYING CAPACITY .......................... 75
4.6.4 SUBTASK FOR ASSESSING WEATHER ........................................................... 78
4.6.5 DECISION TREE AND TASK FOR DECISION MAKING ......................................... 79
4.6.6 DECISION PROCEDURE .............................................................................. 81
4.7 OPERATION AND DELIVERY OF DEERKBS ............................................. 83
4.8 CONCLUSIONS AND DISCUSSIONS ......................................................... 84
CHAPTER 5: SUMMARY ..................................................................... 90
APPENDIX A: DeerMOM USER’S GUIDE ................................................ 94
APPENDIX B: DeerKBS USER INTERFACES ......................................... 101
APPENDIX C: DECISION RULES IN DeerKBS ........................................ 107
REFERENCES .................................................................................. 113

LIST OF TABLES

Table 3.1. Four-level hierarchical classification schemes .............................. 38

Table 3.2. White-tailed deer density measures (number/100 kmz) (1991 data,
provided by the MDNR) ................................................................ 43

Table 3.3. Landscape metrics for the counties in the Upper Peninsula under 30-
class scheme. PD = patch density (number of patches I100 ha), ED =
edge density (sum of lengths (m) of all edge segments per ha), LSI=
landscape Shape index (ratio of the sum of landscape boundary and all
patch edge segments to total landscape area), TCAI = total core area
index (percentage of the landscape that is core area), SHDI = Shannon's
diversity index (a combined measure of richness and evenness), IJI =
interspersion & juxtaposition index (%, observed interspersion over the
maximum possible interspersion for the given number of patch types), and
CONTAG = contagion index (%, observed contagion over the maximum
possible contagion for the given number Of patch types ....................... 46

Table 3.4. Landscape-level edge densities sum of lengths (m) of all edge
segments per ha) under 4 classification schemes .............................. 51

Table 4.1. Attributes in DeerKBS and their assessment outcomes .................. 7O

LIST OF FIGURES

Figure 2.1. Part of the user interface in DeerMOM showing the five interrelated
sectors. ........................ 14

Figure 2.2. Relationship between age-specific density-dependent mortality and
deer density, assuming that there was no harvest. ...................................... 20

Figure 2.3. Comparison between population estimates from DeerMOM simulation
and from pellet group surveys ...................................................................... 24

Figure 2.4 (a) Population dynamics under different doe harvest rates when 50%
of bucks were harvested. (b) Dynamics of percentage of antlered bucks
under different doe harvest rates when 50% of bucks were harvested ........ 25

Figure 2.5 (a) Population size in 1996 under different management scenarios. (b)
Percentage of antlered bucks in 1996 under different management
scenaNOs. .................................................................................................... 27

Figure 2.6 (a) Effect of winter severity on population dynamics (Buck and doe
harvest rates = 20%). (b) Effect of winter severity on dynamics of the
percentage of antlered bucks (Buck and doe harvest rates = 20%). ........... 28

Figure 2.7 (a) Deer population dynamics in response to low harvest rates under
harsh winters (WSls=120) and high harvest rates under mild winters
(WSIs=80). (b) Percentage of antlered bucks in response to low harvest
rates under harsh winters (WSls=120) and high harvest rates under mild
winters (WSls=80). ...................................................................................... 29

Figure 3.1. Map of the study area showing the names and locations of all 15
counties in the Upper Peninsula of Michigan. .............................................. 36

Figure 3.2. Example of landscape patterns under 4 classification schemes. The
image was part of Menominee landscape with an extent of 300 x 300 cells
(83.61 kmz). ................................................................................................. 40

Figure 3.3. Relationship between deer harvest (number of harvested deer per
100 kmz) and (a) patch density (number of patches per 100 ha), (b) edge
density (sum of lengths (m) of all edge segments per ha), (c) landscape

xi

shape index (ratio of the sum of landscape boundary and all patch edge
segments to total landscape area), and (d) total core area index
(percentage of the landscape that is core area) ........................................... 47

Figure 3.4. (a) Relationship between deer-vehicle accidents and patch density
(number of patches per 100 ha). (b) Relationship between reported kills
and patch density (number of patches per 100 ha) ............................... 48

Figure 3.5. Relationships between deer density measures and landscape
metrics. * = p < 0.05, ** = p < 0.01, PD = patch density (number of patches
per 100 ha), ED = edge density (sum of lengths (m) of all edge segments per
ha), LSI = landscape shape index (ratio of the sum of landscape boundary
and all patch edge segments to total landscape area), TIAI = total core area
index (percentage of the landscape that is core area), SHDI = Shannon's
diversity index (a combined measure of richness and evenness), IJI =
interspersion and juxtaposition index (%, observed interspersion over the
maximum possible interspersion for the given number of patch types), and
CONTAG = contagion index (%, observed contagion over the maximum
possible contagion for the given number of patch types). ............................ 49

Figure 3.6. Sensitivity of edge density (sum of lengths (m) of all edge segments
per ha) to Changes in classification schemes. ............................................. 52

Figure 3.7. Relationship between deer harvest (number I100 kmz) and edge
densities (sum of lengths (m) of all edge segments per ha) under 4
Classification schemes. ................................................................................ 55

Figure 3.8. Highest associations between deer harvest (number I100 kmz) and
edge density (sum of lengths (m) of all edge segments per ha) of single
Class under 4 classification schemes ................................................. 56

Figure 4.1. Subtask for population Size assessment, where PopulationSize
represents population size, and PopDensity represents population density.
..................................................................................................................... 73

Figure 4.2. Subtask for antlered buck percentage assessment, where
BuckPercent represents the percentage of antlered buck in deer population.
..................................................................................................................... 74

Figure 4.3. Subtask for habitat assessment, where HabitatQuantity represents
habitat quantity, HabitatQuality represents habitat quality; Abundant,
Average, and Lack are the outcomes of habitat quantity assessment; High,
Medium, and Poor are the outcomes of habitat quality assessment; Poor,
Moderate, and Excellent are the outcomes of habitat assessment. ............. 75

xii

Figure 4.4. Subtask for CCC assessment, HunterExpected, FarmerExpected,
and PublicExpected represents the expectation for deer numbers by
hunters, farmers, and the other interest groups. Fewer, Same, and More are
the outcomes of CCC assessments ............................................................. 77

Figure 4.5. Subtask for winter weather assessment, Favorable, Average, and
Harsh are the outcome of the assessment of winter conditions. .................. 78

Figure 4.6. Part of a decision tree showing how a preliminary recommendation
was made based on the outcomes of assessments of deer density, habitat
conditions, cultural carrying capacity, and weather conditions. Note that
hexagon represents node and ellipse represents leaf (outcome). Left-pointed
triangles represent backward chaining, and right-pointed ones represent
forward chaining. ......................................................................................... 80

Figure 4.7. Diagram of a decision-making procedure. HarvestRecommendation
represents the preliminary harvest recommendations that were made based
on the assessments of population density, habitat, CCC, and weather. The
recommendations were further refined by considering antlered buck
percentage in deer population. Note that left—pointed triangles represent
backward chaining, and right-pointed ones represent forward chaining. ....82

Figure 4.8. Graphic user interface of DeerKBS showing that background
information and images are incorporated into the user interface. Explanation
is provided by clicking the "How" button. ..................................................... 83

Figure B1. Figure B1: User inputs deer population size in an area of interest...102

Figure B2: User inputs the area of interest in squar miles ............................ 102

Figure 83: User chooses the evaluation result of deer habitat quantity from a list
of values. .................................................................................. 103

Figure B4: User chooses the evaluation result of habitat quality from a list of
values. ..................................................................................... 103

Figure BS: User Chooses the evaluation result of deer hunters’ expectation from a
list of values ................................................................................ 104

Figure 86: User chooses the evaluation result of farmers’ expectation from a list
of values. ................................................................................... 104

Figure 87: User Chooses the evaluation result of other general public’s
expectation from a list of values. .................................................... 105

xiii

Figure 38: User inputs the percentage of antlered bucks in the fall
population .............................................................................. 105

Figure B9: DeerKBS makes harvest recommendations. ......................... 106

Figure B10: DeerKBS provides explanations by tracking the decision process.
............................................................................................ 106

xiv

Chapter 1

INTRODUCTION

The graceful and elusive white-tailed deer (Odocoileus virginianus) iS a
symbol of wildlife in North America, which reminds us of the importance of the
relationship between humans and nature. Because of their economic, ecological,
and aesthetic values, white-tailed deer are one of the most studied and generally
enjoyed large game Species (Halls 1984, Rue 1989).

The population dynamics and spatial distribution changes of white-tailed
deer exemplify the interaction between wildlife and human development. From
near extirpation in North America at the end of the nineteenth century, white-
tailed deer have rebounded almost over their entire distribution area (MCCabe
and McCabe 1984, Langenau 1994). This rebound may be attributed to the
deer’s strong adaptability to diverse habitats and wildlife managers' continuous
management efforts (Halls 1984, Ozaga et al. 1994).

The rebounding population provides more opportunities for hunting and
recreation, but also causes serious socioeconomic and ecological concerns,
such as deer-vehicle accidents (Allen and McCullough 1976, Conover et al.
1995), crop damage (Matschke et al. 1984, Braun 1996), disease transmission
(Schmitt et al. 1997, Wilson and Childs 1997), negative impacts on biodiversity
(DeCalesta 1994), changing succession trajectories (Ross et al. 1970, Alverson
et al. 1988) and impacting forest regeneration (T ilghman 1989, Mladenoff and

Streams 1993). Given the conflicting interests surrounding deer, it is essential to

establish management goals that consider the needs of different stakeholders as
well as the integrity of the ecosystem. Based on the established goals, deer
managers then can determine the appropriate measures needed to achieve their
specific objectives.

To better manage deer populations and their habitats, many field
observations and experiments have been carried out to collect data and
information about deer biology (Eberhardt 1960, Venne 1969), population
ecology (Hayne 1984, Dusek et al. 1989, Nixon et al.1991), movement and
habitat (Verme 1973, Larson et al. 1978, Nelson 1984), relationship with weather
(Verme 1968, Drolet 1976, Mech et al. 1987), and effects of management
(Crawford 1984, Creed et al.1984, Euler 1984). Through these studies,
knowledge about deer populations and habitats and management expertise have
been accumulated in wildlife science and management.

However, deer managers are facing more Challenges than ever before.
First, society demands that deer be managed to meet a wide range of goals.
Public involvement in deer management is an indispensable component in the
decision-making process. So people management is becoming important to
address socioeconomic issues (Kellert 1994). Second, the management
paradigm is shifting from traditional small-scale, single Species management to
ecosystem management. Deer management needs to be integrated with the
management of multiple natural resources system and biodiversity conservation
programs (NRC 1995, Swanson 1997), and to greatly expand the scale on which

ecosystems are analyzed and managed (Crow and Gustafson 1997). Third,

ecosystem management emphasizes sustainability and large-scale
management, recognizes humans as an ecosystem component, and
acknowledges “that current knowledge and paradigms of ecosystem functions
are provisional, incomplete, and subject to change” (Christensen et al. 1996).
Expertise in ecosystem management is even scarcer and more valuable it it
exists (Ludwig et al. 1993, Giles 1998).

Computer modeling can help wildlife managers understand wildlife
population dynamics and choose a better alternative to reach their Objectives
(Walters and Gross 1972, McCullough 1979, Starfield 1997). Empirical models
such as “stock-recruitment” models are constructed through empirical data fitting
instead of mechanistic functions (McCullough 1979). The drawback Of empirical
models iS their lack of description of population sex and age structure.
Mechanistic models, such as those based on POP-ll (Bartholow 1986, Bender
and Roloff 1996) and its predecessor ONEPOP (Walters and Gross 1972,
Meddin and Anderson 1979) track animals through each annual cycle. These
"accounting" models require detailed data, including pre- and post-harvest
season natural mortality, harvest and wounding loss, age-specific vulnerability,
and age-Specific reproductive rates (Bartholow 1986). Both empirical and
mechanistic modeling approaches have their advantages and disadvantages.
The empirical models incorporate the expertise of Wildlife managers, but neither
address some underlying mechanisms nor provide crucial information needed for
deer management. The mechanistic models, on the other hand, provide more

scientific reasoning and explanation but require much more data.

Deer are an abundant herbivore and habitat generalist that have unique
roles in ecological processes because they are often a key Species in the
ecosystem (Alverson et al. 1988, DeCalesta 1994). Understanding landscape
structure and its relationship with deer distribution is important in ecological
research and natural resources management (Turner 1989, Gustafson 1998).

A traditional approach to studying the responses of deer to their
environmental changes is habitat analysis of isolated patches at fine scales.
Based on deer's physiological and nutritional requirements, carrying capacity in a
study area is estimated (Wallmo et al. 1977). A more coarse-scale analysis is
the quantification of necessary habitat elements such as food, cover, water, and
space (Bender and Haufler 1990, Virgés and Telleria 1998). One of the most
comprehensive deer habitat models may be the habitat suitability index (HSI)
models developed by Short (1986) for use in the Gulf of Mexico and south
Atlantic coastal plain. These HSI models include 4 models. Model I estimates the
carrying capacity of habitats during autumn-winter based on the energy
requirements of deer during these seasons. Model II and Model III are derived
from Model I. Model ll determines the quantities of suitable forage and Model lll
provides general estimates of habitat quality on an evaluation area. Model IV
predicts the probable presence or absence of deer on an evaluation area. The
uniqueness of these models is that they take direct measurements of available
energy and forage quantity as indicators of habitat quality. Stauffer (1990)
evaluated the inputs and assumptions of those models. Banker (1994) did a field

test on Quantico Marine Corps Base, Virginia, and made some changes and

improvements for these HSI models. The obvious limitation of this approach is
that it requires intensive field samplings.

Other attempts are made to link deer age-specific birth rates with forage
nitrogen yield (Meddin and Anderson 1979), to link deer population with northern
forest vegetation (Cooperrider 1974) and with an aspen-type (Populus spp.)
forest (Mello 1983, Campa 1989). An individual-based modeling approach was
also used to study the effects of nutrition, variation in predation pressure,
bonding interruption, and variation of birth cover quality on deer population
dynamics (German 1992). The model results indicated that predation and
nutrition are the most important regulating factors and the size of the deer herd
could be increased by reducing the number of predators and by improving
birthing cover.

The models based on these habitat analyses have limited applications in
management because of intensive data requirements and/or low accuracy in
prediction (Morrison et al. 1992). New developments in remote sensing (RS),
global positioning system (GPS) and geographical information systems (GISS),
provide valuable tools to obtain, update, integrate, synthesize, and analyze
temporal and spatial data (Coulson et al. 1987, 1991, On'nsby et al. 1987, Elston
and Buckland 1993, Mladenoff 1994, Chang 1995). Linking GlS with resource
databases enables the generation and evaluation of multiple-use management
options (Brown et al. 1994, Mallawaarachchi et al. 1996).

Deer have relatively large and often seasonally migratory home ranges

(Verme 1973, Nelson and Mech 1984, Van Deelen et al. 1998), requiring the

understanding of landscapes in addition to isolated patches of habitat. Field
observations often indicate that deer densities differ dramatically even though
two areas have seemingly Similar habitats (Morrison et al. 1992). This implies
that there exists a landscape level of habitat Characteristics that influence deer
Spatial distribution.

With more data accumulating from research and new technology, the
complexity of decision making increases because it involves more issues and
needs to Offer more alternatives. Complexity makes it difficult for wildlife stewards
to have all the relevant information and knowledge (Starfield and Bleloch 1986).
However, decisions must be made, even with uncertainties and conflicting goals
and without complete data and scientific consensus (Ludwig et al. 1993, Ehrlich
and Daily 1993). This requires deer managers to take a new approach to address
the problem. The application of artificial intelligence (Al) in natural resources
management offers great potentials for aiding and improving decision making
(Coulson et al. 1987, 1991, Covington et al. 1988, Davis et al. 1989).

Knowledge-based systems (KBSS), a branch of artificial intelligence (Al),
can be used to help wildlife managers take advantage of the accumulated data
and information to make more informed decisions. Generally speaking, KBSS are
computer systems that use human knowledge to solve problems that normally
would require human intelligence. They differ from other Al programs in their
performance, domain-specific problem-solving strategies, and their justification

for conclusions (Edwards 1991). They differ from conventional software

programs in their symbolic representation, symbolic inference, and heuristic
search (Hayes-Roth et al. 1983).

KBSS for natural resource management have been developed Since the
beginning of KBSS research. The first successful KBS application in natural
resource management is Prospector for mineral exploration in the late 19705
(Duda et al. 1979). Most of the early applications are limited to the interpretation
and diagnosis task of geological and mineral problems. Production rules are the
predominant type of knowledge representation scheme. Since the late 19805,
KBSS have expanded applications to management of forests, water resources,
and ecosystems. Bremdal (1997) reviewed 27 KBSS for the management of
natural resources. Among them, only one is for Wildlife (whale) visual
identification.

A new direction of KBSS for natural resource management is to integrate
simulation models, GIS, and database management to provide a framework tool
for decision-making. The ecosystem management decision support system
(EMDS), which was developed by the USDA Forest Service Pacific Northwest
Research, integrates GIS and knowledge-based reasoning technologies
(Reynolds 1999).

With the development of KBSS, natural resource managers and ecological
modelers have begun to recognize their potentials to understand, manage, and
model natural resource systems and to integrate landscape data and knowledge
from diverse scientific disciplines into natural resource management (Starfield

and Bleloch 1986, Coulson et al.1987, 1991, Schmoldt and Rauscher 1996, Giles

1998). In fact, a few attempts with Al or KBSS techniques have been made to
study moose-forest interactions (Saarenmaa et al. 1988) and deer movements in
a heterogeneous habitat (Folse et al. 1989), to help resource agencies and
managers use computer models (Ritchie 1989), and to assess qualitative
information and infer logically valid conclusions (Davis et al. 1989).

Effective management of deer and other wildlife species can be enhanced
by integration of wildlife science, modern technology, and wildlife managers'
expertise. The goal of this research was to integrate deer population modeling,
landscape level habitat analysis, and socioeconomic considerations into a
decision support system for white-tailed deer management. Through this
research, I provided the wildlife community with tools to evaluate population
dynamics under different management options, to assess habitat conditions on
landscape scales, and to make informed decisions using knowledge-based
system. My research was done using the data from the Upper Peninsula (UP) of
Michigan.

Specific objectives of this research were to:

1. Assess the effects of management alternatives on deer population size, sex
and age structure by constructing a population model;

2. Identify the relationships between landscape structure and deer spatial
distribution; and

3. Provide a unifying framework for deer decision-making with a knowledge-

based deer management system

Chapter 2

WHITE-TAILED DEER MANAGEMENT OPTIONS MODEL (DeerMOM):
DESIGN, QUANTIFICATION, AND APPLICATION

The white-tailed deer is one of the most important game species in North
America. This species provides opportunities for hunting and recreation, but also
cause serious socioeconomic and ecological concerns, such as deer-vehicle
accidents, crop damage, and impacting forest regeneration (Caughley 1981,
McShea et al. 1997). Given the conflicting interests surrounding deer, it is
essential to establish management goals that consider the needs of different
stakeholders as well as the integrity of the ecosystem. Based on the established
goals, deer managers then can determine the appropriate measures needed to
achieve their specific objectives. Harvesting is a primary option for population
manipulation.

Deer management requires instruments that can assess the effects of
harvesting options on deer population size and structure. Computer modeling
can help wildlife managers to understand deer population dynamics and choose
a better alternative to reach their objectives (Walters and Gross 1972,
McCullough 1979, Starfield 1997). Good models incorporate essential
information and allow for evaluation of the consequences of different
management options.

There are basically two approaches to modeling deer population dynamics.
One approach is to build empirical models such as “stock-recruitment” models.

9

The stock-recruitment models are constructed through empirical data fitting
instead of mechanistic functions (McCullough 1979). The use of empirical models
is limited because they do not describe population sex and age structure, which
are important parameters to deer managers. The other approach is to build
mechanistic models, such as a density-dependent matrix harvest model (Jensen
1996) and models based on POP-ll (Bartholow 1986, Bender and Roloff 1996)
and its predecessor ONEPOP (Walters and Gross 1972, Meddin and Anderson
1979). The matrix model considers age-specific mortality and fecundity, but does
not describe sex structure. POP-ll and its predecessor track animals through
each annual cycle. These "accounting" models require detailed data, including
pre- and post-harvest season natural mortality, harvest and wounding loss, age-
Specific vulnerability, and age-specific reproductive rates (Bartholow 1986). Both
modeling approaches have their advantages and disadvantages. The empirical
models incorporate the expertise of wildlife managers but do not address some
underlying mechanisms and do not provide crucial information needed for deer
management. On the other hand, the mechanistic models provide more scientific
reasoning and explanation but require much more data.

I developed a hybrid (mechanistic and empirical) model, DeerMOM, to
evaluate the effects of deer management options on population Size, sex and
age structure. Mechanisms in DeerMOM‘are similar to those in POP-ll, but
DeerMOM also incorporates management empiricism. The model takes
advantage of the data available from field studies, yet it is not limited by data

incompleteness. DeerMOM is designed to simulate deer populations in different

10

locations but was parameterized and tested for the deer population in the Upper
Peninsula (UP), Michigan. The objectives of this study are to introduce the
design of the model and the quantification of the model parameters, and to
demonstrate its application in deer management under the Specific constraints of

management goals in the UP of Michigan.

2.1 Methods

2.1.1 Model Design

I followed 3 principles to design the model: simplicity, accuracy, and
management orientation. These principles were incorporated into the process of
model design, construction, validation, and Simulation. They are also

interconnected and not separable.

2.1.1.1 Simplicity

The model should be structured as simply as possible. The deer
populations in most areas are heavily hunted, which reduce their life expectancy
relative to slightly exploited deer populations (Burgoyne 1981). l grouped deer
into 3 age classes: fawn (<12 months), yearling (13-24 months), and adult (>24
months) to simplify model structure. This classification is widely accepted and
adopted by deer managers (McCullough 1979). l divided mortality into harvest
mortality and natural mortality (e.g. vehicle accidents, starvation, predation) and
used annual mortality instead of seasonal mortality to ease data collection and

model application.

11

2.1.1.2 Accuracy

The model Should Simulate and predict population dynamics as accurately
as possible. Knowledge of age and sex composition of the deer population is
essential for deer managers to establish their management objectives. Since
male deer are polygamous, only female deer numbers have a substantial impact
on Changes in population size. However, the percentage of bucks in the
population is of special interest to deer hunters and consequently deer
managers. In the model, I Simulated the dynamics of males and females
separately. | used age- and sex-specific natural mortality and harvest mortality. In
addition, I differentiated their reproductive rates and fetal sex ratios by age class

for female deer.

2.1 .1 .3 Management orientation

The final principle was to develop a model that is management oriented.
Management goals could be used as criteria to evaluate harvest strategies under
defined time frames. A user could set goals and run the model to determine
which harvest scenarios would achieve the management goals during a
predefined time period. Specific management goals include maximum sustained
yield, trophy management, and quality deer management. A user could also
change the model structure by adding or removing some components, or by
modifying parameter values. However, the model should be user friendly and
have a graphical user interface to assist users in learning and using the model. In

addition, model results should be easily transferred to other software programs

12

for further data analysis. Finally, the model should accommodate users With
different levels of expertise in modeling and mathematics.

To fulfil the above-mentioned requirements, I used Stella ®to develop
DeerMOM. Stella ® is a multi-level, hierarchical environment for constructing
models (High Performance Systems 1997) which allows DeerMOM to have a 3-
Ievel hierarchical structure. The first level is a user interface, where novice users
can change parameter values by using Sliders or numerical pads. Moreover, the
user can also inspect Simulation results using tables and graphs as well as
export them to other programs through dynamic data exchange. At the second
level, a more advanced user can access the model structure in order to modify
the model structure and manipulate parameter values. Finally, an expert user can
access the third level to review the mathematical equations and to learn the

model mechanisms in detail.

2. 1.2 Model Structure

DeerMOM consists of 5 interconnected sectors: Female, Male, Birth/Death,
Harvest, and Population (Figure 2.1). DeerMOM operates on an annual cycle

that starts on October 1 and ends on September 30 of the following year.

2.1.2.1 Female Sector

Females were grouped into 3 age classes: fawn, yearling, and adult as
previously mentioned. For each annual cycle, surviving newborn females enter
the female fawn group, and Similarly surviving fawns recruit to the female

yearling group, surviving yearlings to the female adult group. Adult females that

13

alumnae: V

 

 

 

BlrlhiDoth

 

Female V

 

 

Female

 

 

 

 

Harvest

Figure 2.1. Part of the user interface in DeerMOM showing the five interrelated
sectors.

14

survive hunting season and natural mortality remain in the adult age class.
Natural mortality for each age class was density— and harvest-dependent and

influenced by winter severity.

2.1.2.2 Male Sector

The male sector had the same model structure as the female sector, but
parameter values were quantified differently. For example, the harvest rate for
male adults was different from that for female adults. As with females, all

mortality rates were age- and sex-specific.

2.1.2.3 Birth/Death Sector

Changes in deer population size and structure depend on the dynamics of
both births and deaths. I assumed that immigration and emigration would not
affect age and sex structure of the population. Thus, any addition to the
population comes from the offspring of female fawns, yearlings, and adults. To
estimate newborn females and males, I used female number in spring,
reproductive rate, and sex ratio for each age class. The spring female number in
each age class was estimated by subtracting harvest and natural mortality from
the previous fall female number in each age class. Deaths were a result of

harvest mortality and natural mortality. Mortality rates were age- and sex-specific.

2.1.2.4 Harvest Sector

I included all harvest-related information in this sector. Specifically the
sector contained age and sex composition of harvest, total harvest, percentage

of antlered bucks and antlerless deer in the harvest.

15

2.1.2.5 Population Sector

I included population initialization, management goals, and all data related
to age and sex distribution of deer in this sector. The population was initialized by
Size, sex ratio, and percentage of deer in each age Class. This sector also
extended the female and male sectors by deriving population statistics of
interest. Specifically, it contained numbers and sex ratios for each age Class,
numbers of females and males, total population Size, sex ratio, density, and

percentage of antlered bucks in the population.

2.1.3 Quantification of the Model

The most important parameters for DeerMOM included age-specific
reproductive rates, fetal sex ratios, and neonatal mortality; age— and sex-specific
natural mortality, and harvest mortality. DeerMOM was parameterized to
represent the deer population in the UP of Michigan. AS winter weather
conditions greatly influence the population dynamics in the UP, winter severity
indices (WSIS), which measure air chill and snow hazard (Verme 1968), were
incorporated into the model. Data for estimating model parameters were obtained
from the Michigan Department of Natural Resources (MDNR) and recent field

studies (Van Deelen et al. 1997).

2.1 .3.1 Reproductive rates

Reproductive rates were defined as average fetuses per female based on
spring surveys (Friedrich and Schmitt 1988, Venne 1989). From March to May,

the MDNR personnel conducted necropsies of female deer that died from vehicle

16

collisions and other accidents. Female deer were aged and their fetuses were
counted if they were pregnant. Based on these data, reproductive rates were
calculated for fawnS, yearlings, and adults. The MDNR discontinued these
surveys in the UP after 1988. In DeerMOM, I used the data from the MDNR
spring surveys from 1973 to 1988 (Friedrich and Schmitt 1988) because these
data provided an adequate sample for analytical purposes. Reproductive rate
ranges were 0-0.14, 0.8-1.55, 1.67-2.13 with an average of 0.05 (0.01, SE), 1.30
(0.05), and 1.84 (0.05) for female fawns, yearlings, and adults, respectively.

As reproductive rates showed a great annual variability, regression
analyses and F-tests were used to detect whether or not the variability was
associated with population size. The estimates for population size were provided
by the MDNR based on POP-ll model (MDNR 1996). No Significant relationship
was detected between population and the reproductive rates of yearlings (F1,14 =
0.399, P = 0.538) or of adults (F1, 14 = 0.036, P = 0.852) from 1973 to 1988.
However, there was a Significant decrease for fawn reproductive rates as
population size increased (F144 = 7.891, P = 0.014). These findings were
consistent with the reproductive patterns found in southern Michigan (Verme
1989). Even so, as fawn deer had very low reproductive rates (0.05) in the UP,
changes in fawn reproductive rates had a minimal impact on the entire
population.

Regression analyses were also used to detect the relationships between
age—specific reproductive rates. The results Showed that reproductive rates for

fawns and for yearling were positively associated with reproductive rates for

17

adults (P = 0.014, P = 0.054, respectively). To account for the annual variability
of reproductive rates, a uniform random function of 95% confidence intervals of
the mean (from 1.78 to 1.90) was used to estimate the reproductive rates for
female adults. The estimates of reproductive rates for fawns and for yearling
were derived from regression equations Y1 = 0.2145 X - 0.3472 (R2 = 0.36, F1, 14
= 7.969, P = 0.014), Y2 = 0.9221 X- 0.3945 (R2 = 0.24, F1_14 = 4.407, P = 0.054),
where Y1 is the reproductive rate for fawns, Y2 is the reproductive rates for

yearling, X is the reproductive rate for adults.

2.1.3.2 Fetal sex ratio

Sex ratio was defined as the percentage of males in each age class. Fetal
sex ratio was the percentage of newborn males in offspring. Verrne’s (1983) data
were used to determine the fetal sex ratio. The fetal sex ratios for female fawnS,

yearlings, and adults were 62.5%, 52.6%, and 50.6%, respectively.

2.1.3.3 Mortality

Causes of mortality were Classified as harvest mortality and natural
mortality. The harvest mortality was estimated from the MDNR harvest survey
(Verme 1989) data from 1989 to 1996. The natural mortality was based on a
radio-collared deer study in the UP from 1992—1994, where the deer population
was exposed to hunting (Van Deelen et al. 1997). l partitioned the natural
mortality into 3 additive parts: base mortality, winter mortality, and harvest
compensatory mortality (HCM). The base mortality was the mortality when the

population density was low (< 5 deer/kmz) and winter weather was very mild.

18

Base mortality may be largely due to vehicle accidents or old age. The winter
mortality was mainly due to malnutrition during severe winter conditions.
Cumulative WSIS, which summed the weekly WSIS from December to April, were
used to relate winter severity to deer winter mortality. The cumulative WSIS in the
UP ranged from 65.7 in 1987 to 147.7 in 1979, with an average of 105.4 from
1969 to 1996. To estimate winter mortality, l scaled the difference between the
actual WSI and the lowest WSI in the UP (65.7) by a different adjustment factor
for fawns, yearlings, and adults. HCM was a function of deer density and harvest
mortality. Harvest mortality was assumed to be compensatory to density-
dependent mortality (DDM). In other words, DDM would be lower under a higher
harvest rate. HCM was mainly a result of predation, diseases, and other
unknown causes.

I used a 2-Step procedure to calculate HCM. The first step was to calculate
DDM, assuming that there was no harvest. If deer density was < 5 deer/kmz,
DDM(age) was assumed to be 0 for all age classes. Five deer/km2 was used as
the minimum density(Dmin) because DDM(age) could be negligible when deer
density was < 5 deer/km2 in the UP. When deer density was between 5 deer/km2
and 35 deer/kmz, it was assumed that there was a linear relationship between
DDM(age) for each age class and deer density (Figure 2.2). If deer density was
> 35 deer/kmz, the maximum DDM (DDMmax(age)) was assumed to be 0.5, 0.3,
and 0.1 for fawns, yearlings, and adults, respectively. Thirty-five deer/km2 was
used as the maximum density(Dmax) for DDM because it could be the biological

carrying capacity (BCC) in Michigan (McCullough 1979). In mathematical terms,

19

0.6 -

Density dependent mortality rate

 

 

 

Deer density (deer/km2)

Figure 2.2. Relationship between age-specific density-dependent mortality and
deer density, assuming that there was no harvest.

20

for a deer population with density D, age-specific DDM can be expressed as

equation 1:
0 (if D s 5)
DDM(age) = (D - Dmin) * DDMmax(age) / (Dmax - Dmin) (if 5 < D < 35) (1)
DDMmax(age) (if D Z 35)

The second step to calculate HCM was to consider the compensatory
effect of harvest and to estimate adjusted DDM. Because deer harvest might
mitigate DDM, I then adjusted DDM by a factor of (1 - harvest rate(HR)) for each
respective age class. In mathematical terms, age specific HCM can be

expressed as equation 2:

HCM(age) = (l — HR(age)) * DDM(age) (2)

Take male yearling as an example. If the deer population density was 15
deer/kmz and the harvest rate for male yearlings was 50%, then the DDM for
male yearlings would be (15-5)*0.3/(35-5) = 0.1 (from Equation 1), and the HCM
was (1 -0.5) * 0.1 = 0.05 (from Equation 2).

To summarize, the calculation of annual natural mortality can be expressed

as equation 3:

NM(age) = BM(age) + (WSI — LWSI) * SF(age) + HCM(age) (3)

Where NM is natural mortality, BM is base mortality, WSI is winter severity
index, LWSl is the lowest winter severity index, and SF is a scale factor that

converts WSIS to winter mortality.

21

2.1.3.4 Neonatal mortality

Winter weather has a significant impact on fetal development during late
gestation and thus influences natal survival (Verme 1977). Because of this
impact, I partitioned neonatal mortality into 2 parts: base neonatal mortality and
neonatal mortality due to winter severity. Base neonatal mortality was the
mortality when winter weather was very mild (low WSI). Based on Verme’s data

(1977), I used the following equation to estimate neonatal mortality:

NnM = BNnM + (WSI — LWSI) * SF (4)

Where NnM is the neonatal mortality, BNnM is the base neonatal mortality,

and other variables are defined in previous equations.

2. 1.4 Model Initialization

The 1989 fall population in the UP of Michigan was used to initialize
DeerMOM. Population size was estimated from pellet group surveys. Sex ratio
and age distribution were based on MDNR deer checking station data (Hill and

Rabe 1989).

2. 1.5 Model Testing

DeerMOM was run 50 times to Simulate deer population from 1989 to 1996
using a one-year time step. The model was run only 50 times because the
variability of the population size was small. Annual harvest data and WSIS from
the MDNR were used in these Simulations. The 1989-1996 fall population

estimates from annual pellet group surveys (Hill 1996) were used to verify

22

Simulation results from DeerMOM. Paired t-tests were used to compare
population estimates (1990-1996) from the pellet group surveys and DeerMOM
simulations. After the model was verified, the 1996 fall population estimate,
harvest data during 1996-1997 hunting season, and WSI for the winter of 1996-

1997 were used to predict the 1997 fall population.

2. 1.6 Simulation Scenarios

I ran 50 simulations of 35 representative management scenarios resulting
from the combinations of 7 buck harvest rates and 5 doe harvest rates. Buck
harvest rates ranged from 10% to 70% with 10% intervals and doe harvest rates
ranged from 5% to 25% with 5% intervals. Based on the deer checking station
data, the harvest rates for female and male fawns were assumed to be 5%. l
Simulated deer population dynamics from 1989 to 1996 using actual WSIS. These
scenarios were evaluated by whether they could reach the management goals by
1996. The management goals included a quantity goal (372,500 deer in the fall)
and a quality goal (35% antlered bucks) (MDNR 1993).

I chose the optimal scenario, defined as the scenario that could most
closely achieve both goals by 1996, to project the 1996 deer population for the
next 5 years (from 1997 to 2001) under different weather conditions (mild,
moderate, harsh, and random). WSIS were set at 80, 100, and 120 for mild,
moderate, and harsh winters, respectively. WSIS were assumed to be constant
for 5 consecutive years, except when WSI was a random function, where the

WSIS were uniformly distributed random numbers between 80 and 120.

23

2.2 Results

DeerMOM simulated deer population well and there was no significant
difference (t= 1.24, df = 6, P =0.26) between the population estimates from
DeerMOM and those from pellet group surveys (Figure 2.3). DeerMOM predicted
that the average of 1997 fall population was 528,307, 11.5% higher than the
MDNR estimate of 474,000 deer (H. R. Hill, MDNR, personal communication).

When buck harvest rate was 50%, which was Similar to current buck
harvest rate, harvesting more does reduced the population Size dramatically
(Figure 2.4a) but only Slightly increased the percentage of antlered buck. Thus,

the quality goal was not reached (Figure 2.4b).

 

 

1,000,000 - p=0.26 (df = 6, paired two-tailed t-test)
900fl00
{a 800,000
3:; 700,000
g 600,000
% 500,000 " _
3 400,000 ‘ —°—Pellet group surveys
E 300,000 - —0—Average values of DeerMOM simulations
3 200.000 ‘ +95% lower bound ofsimulation results
100.000 ‘ —>K—95% upper bound 0 simulation results
0 r 1 r u 1 I .

 

1989 1990 1991 1992 1993 1994 1995 1996

Year

Figure 2.3. Comparison between population estimates from DeerMOM
simulations and from pellet group surveys.

24

 

 

 

 

 

 

 

 

 

 

1,000,000 _ +008 harvest rate = 5%
—B—D0e harvest rate =10%
900-000 ‘ +D0e harvest rate = 15% ’ 9
300,000 _ —-><——Doe harvest rate = 20° ’
a, —)I(—D0e harvestrat - 1" %
3 700-000 ‘ - - - - QuantIty go ’ = = ‘
5 600,000 . 5 , j “
g 500.000 - Li, ? V a
8 \: - fi' ’\ ~
0. 400,000 a _____________________ __ _ _ _
3 300,000 .
0
200,000 -
100,000 J
0 T I I T T l I
1989 1990 1991 1992 1993 1994 1995 1996
Year
40 _ b
35 a ...................................... _
:\°‘
3 30 -
9 S
‘3 E 25 ~
8 3 ° A
a o /; L x x
:5“ § \§- 0 55 K?
0) a) 15 -
35 —e—Doe harvest rate=5%
5 5 10 _ —E—Doe harvest rate= 10%
8 —A—D0e harvest rate: 15%
If —)(-—Doe harvest rate=20%
5 ‘ —)lé-—D0e harvest rate=25%
- - - - Quality goal
0 I j— j T l fi l
1989 1990 1991 1992 1993 1994 1995 1996

Year

Figure 2.4 (a) Population dynamics under different doe harvest rates when 50%
of bucks were harvested. (b) Dynamics of percentage of antlered bucks under
different doe harvest rates when 50% of bucks were harvested.

25

When doe harvest rates remained constant, decreasing buck harvest rates
did not change deer population sizes markedly (Figure 2.5a) but increased the
percentage of antlered bucks greatly (Figure 2.5b). The best scenario in terms of
management goals was the harvest of 20% of both bucks and does. It brought
the population down to the quantity goal level (Figure 2.5a) and increased the
percentage of antlered bucks near the quality goal level (Figure 2.5b).

Winter severity had a dramatic impact on the population. When 20% of
bucks and does were harvested, it took 5 years to reach the quantity goal under
moderate (WSIS = 100) winters (Figure 2.6a). When winter conditions were
continuously harsh (WSIS = 120), deer population decreased rapidly. When the
winter conditions were mild (WSls =80), the population remained stable (Figure
2.6a). However, regardless of WSI, the percentage of antlered bucks increased
to the goal level (Figure 2.6b) if 20% of bucks and does were harvested.

Winter weather conditions were important in determining appropriate
harvest rates for achieving the quantity goal. Under harsh winters, it was
necessary to lower the harvest rate from 20% t0 5% (Figure 2.7a). Under mild
winters, however, the harvest rate Should be increased from 20% to 30% (Figure
2.73) to reach the quantity goal. Both harsh and mild winter conditions allowed
deer to reach the quality goal level under either 5% or 30% harvest rates if

harvest rates were the same for both female and male deer (Figure 2.7b).

26

+ Doe harvest rate = 5%
—B— Doe harvest rate = 10%

 

 

 

 

 

 

 

 

800 000 a —A— Doe harvest rate = 15%
’ —-><— Boe names: rate = Egg;
,. + 0e arvesrae= 0
700,000 : C A- - - - Quantity goal
g 600,000 :: _ V V v *2
,5 500,000 .. . L” ‘3 3 :2 +3
3 - A A A A
g- 400,000 _ _ _ _____ :1 _____ ‘:‘ _____ 1:: _____ 1.1 _____ A
3 300,000 4 " ,. X x fix!
0)
3 200,000 -
100,000 -
0 T I T I T I
10 20 30 40 50 60 70
Buck harvest rate (%)
60.00 b —e— Doe harvest rate = 5%

 
 
  
 
   

—B— Doe harvest rate = 10%
-—A—- Doe harvest rate = 15%
—)(— Doe harvest rate = 20%
-)Ié— Doe harvest rate = 25%
- - - - Quality goal

Percentage of antlered bucks (%)
In the deer populatIon

 

 

O. 00 I T T l l 1
10 20 30 40 50 60 70
Buck harvest rate (%)

Figure 2.5 (a) Population Size in 1996 under different management scenarios.
(b) Percentage of antlered bucks in 1996 under different management scenarios.

27

 

  

 

800,000 1 a
700,000 ~ , ——-— _._
.. ’ , e- e- .9 _ _ _<_>_ _ _ _0
3 600,000 A
'6
.5 500,000 -
s 400,000 -_
Q. .
8. . ‘ - ~
5 300900 ‘ —<>— WSI= 80, mild winter 15* - . ‘ ~ A.
‘1’ —B—WSI = 100, moderate winter - ~ ~ .
D 200'000 —A - WSI: 120, harsh winter A
100,000 _ -—>(—-WSI= Rand0m(80,120)
- - - - Quantity goal
0 I I T l 1
1996 1997 1998 1999 2000 2001
Year
40 - b

35~
30-
25-

Percentage of antlered bucks (%)
In the deer population

 

    

 

20 —<>— WSI= 80, mild winter
15 _ -B—WSI= 100, moderate winter
—A - WSI=120,harshwinter
1° ‘ —)(—-WSI = Random(80,120)
5 — - - - - Quality goal
0 . . T . .
1996 1997 1998 1999 2000 2001
Year

Figure 2.6 (a) Effect of winter severity on population dynamics (Buck and doe
harvest rates = 20%). (b) Effect of winter severity 0n dynamics of the percentage
of antlered bucks (Buck and doe harvest rates = 20%).

28

1’000'000 —°' 'WSI = 120, 5% buck and doe harvest rate
900.000 ' +WSI = 100, 20°/o buck and doe harvest rate
800,000 - —A—WSI = 80, 30% buck and doe harvest rate
700,000 ‘ ' " " Quantity goal

600,000
500,000 ‘
400,000 4. ......................
300,000 ‘
200,000 ‘
100,000 i

 
  
 

Deer population size

 

 

0 F T T l7 1

1996 1997 1998 1999 2000 2001

Year

401 b
35*
30‘
25‘

 

Percentage of antlered bucks (%)
In the deer populatlon

 

 

20
15 _ _°‘ 'WSl = 120, 5% buck and foe harvest tate
+WSI = 100, 20% buck and foe harvest tate
10 ‘ -A-WSI = 80, 30% buck and foe harvest tate
5 ‘ ' ' ' Quality goal
0 I f r r r
1 996 1997 1 998 1999 2000 2001

Year

Figure 2.7 (a) Deer population dynamics in response to low harvest rates under
harsh winters (WSls=120) and high harvest rates under mild winters (WSls=80).
(b) Percentage of antlered bucks in response to low harvest rates under harsh
winters (WSls=120) and high harvest rates under mild winters (WSls=80).

29

2.3 Discussion

2. 3. 1 Model structure

In this study, I tried to balance model simplicity, accuracy, and generality
(Levins 1966) in order that the model is accessible to the wildlife management
community. I minimized data requirements for the model but still offered vital
population information, which is crucial for deer management decision
processes. The model structure could be easily modified to include new
information once it is available. For example, if there are field data available for

neonatal mortality, the empirical function can be easily replaced with actual data.

2.3.2 Quantification of parameters

Quantification of model parameters is usually the most challenging step in
systems modeling. I felt comfortable with age-specific reproductive rates. because
good historical data existed. There is controversy about how maternal age and
nutritional conditions affect Offspring sex ratios (Harder 1980, Venne 1983, Burke
and Birch 1995). Because of this controversy, I used Verme’s (1983) data as
they were based on research on deer in the UP. However, I did not incorporate
yearly variation in fetal sex ratios because the data were not available.

I took an empirical approach to estimate natural mortality and neonatal
mortality. Because harsh winter weather accounted for most of the deer mortality
in winters, winter mortality was linked with WSIS. This approach might
oversimplify the relationship between deer mortality and weather conditions,

however, simulations suggested these estimates are reasonable.

30

To apply DeerMOM to other deer populations, model parameter values
Should be changed to reflect the biological and ecological characteristics of those
populations (Euler and Morris 1984, Gavin et al. 1984, Dusek et al. 1989). The
most important changes might be the estimates of natural mortality and neonatal
mortality. Winter severity might not play such an important role in regulating deer

population in warmer geographical areas as it did in the UP.

2. 3.3 Current management practices

Although the MDNR tended to adapt ad hoc regulations in the UP because
of the unpredictability of harsh winters, they realized that more consistent
proactive management regulations Should and could be implemented in the field.
The winters of 1995-1996 and 1996-1997 were harsh and greatly decreased the
UP deer population by up to 50% (Urbain 1997). Nevertheless, the deer
population in the UP remains higher than the goal level. The MDNR continues to
issue antlerless deer permits to reduce the deer population (Urbain 1997). Some
local experiments, such as Hiawatha Sportsmans Club's 1996 resolutions, are
carried out to harvest more antlerless deer and to restrain buck harvest (T. R.
Minzey, MDNR, personal communication). However, no such statewide effort has
been implemented to discourage the harvest of antlered bucks. Under current
management practices, the quality goal can not be reached (Van Deelen et al.

1997).

2. 3.4 Harvest recommendations

The recommendations are mainly based on ecological considerations.

31

Which option is the best to implement in management practices also depends on
acceptability of a specific management option by stakeholders. However, the
advantages of DeerMOM is that it offers options that a wildlife manager can
assess their outcomes before the Specific harvest scenario is implemented, thus
it offers a logical and defensible rationale for wildlife decision making (Starfield
1997)

The recommendation of harvesting 20% of both bucks and does is based
on then existing population size, management goals, winter severity conditions,
and the time frame previously defined. Accordingly, the 20% harvest rates should
not be interpreted as a general guideline to other populations. However, the
model reveals that doe harvest is a more effective way to manage population
size and buck harvest is a more effective way to manage the percentage of
antlered bucks. This finding has important implications in quality deer
management. To reach both the quantity and the quality goals, managers must
balance both buck and doe harvest rates. Contrary to wildlife manager’s intuitive
viewpoint that simply increasing doe harvest would increase the percentage of
antlered buck to the goal level (McCullough 1979), the simulations Show that the
quality goal cannot be reached without also decreasing buck harvest
simultaneously. In other words, the quality goal can only be achieved by

restricting the buck harvest and increasing the doe harvest.

32

Chapter 3

LANDSCAPE STRUCTURE AND
SPATIAL DISTRIBUTION OF WHITE-TAILED DEER

Understanding landscape structure and its underlying ecological
consequences is a central theme in ecological research and natural resource
management (Turner 1989). Quantifying landscape structure has been important
in landscape analyses and has received considerable attentions in landscape
ecology (Gustafson 1998). Many attempts have been made to link landscape
attributes with spatial pattern of endangered species and common birds and
small mammals (Andrén 1994, Flather and Sauer 1996, Bayne and Hobson
1998). However, few have studied the relationship between landscape structure
and spatial pattern of abundant species, particularly large herbivores such as
white-tailed deer at landscape scales. These abundant herbivores are usually
habitat generalists and have unique roles in ecological processes because they
are often key species in the ecosystem.

In this study, I examined the relationship between landscape structure and
the spatial pattern of white-tailed deer, a habitat generalist and a keystone
Species in many ecosystems in North America (Alverson 1988, Campa 1989,
DeCalesta 1994). The importance of the study lies not only in that deer are the
most valuable game Species, but also in that their spatial dynamics in the past
100 years reveals well its response to spatial and temporal landscape Changes

(McCabe and McCabe 1984). Almost extirpated due to overexploitation in the

33

end of the nineteenth century, white-tailed deer have rebounded and expanded
in most areas due to landscape changes and wildlife management since the
beginning of the twentieth century.

A traditional approach to studying the responses of deer to environmental
changes is habitat analysis of isolated patches at two levels. The first level of
analysis, at a fine scale, deals with the physiological and energetic requirements
of the organism, such as Short's (1986) deer energetics model. The second level
of analysis, at a coarse-scale, is the quantification of necessary habitat elements
such as food, cover, water, and Space (Bender and Haufler 1990, Virgés and
Telleria 1998). The assumption for these studies is that a positive relationship
exists between the amount of necessary habitat elements and relative deer
abundance. It is assumed that investigators, with good estimates of habitat
components, can predict deer abundance in a defined area by evaluating the
habitat suitability.

The models based on these habitat analyses have limited applications in
management because of intensive data requirements and/or low accuracy in
prediction (Morrison et al. 1992, Van Deelen 1995). Deer have relatively large
and often seasonally migratory home ranges (Verme 1973, Nelson and Mech
1984, Van Deelen et al. 1998); these characteristics require wildlife managers to
understand deer's landscapes in addition to isolated patches of habitat (Hanley
1996). Field observations often indicate that deer densities differ dramatically
even though two areas have a similar habitat suitability index (HSI, a combined

measurement of habitat quantity and quality) (Morrison et al. 1992, Van Deelen

34

1995). This may imply that there exists a higher level of habitat properties that
influence deer spatial pattern. I hypothesized that the deer spatial distribution is
related to spatial characteristics of habitat patches at landscape level. The spatial
characteristics at the landscape level include spatial attributes of patch, area,
edge, and shape. To test this hypothesis, | quantified landscape attributes and
then linked these attributes with deer spatial distribution. The first objective of this
study was to determine which landscape attributes have good associations with
deer spatial distribution.

Many field observations and empirical studies at local scales have
indicated that deer prefer edges to interior areas (Leopold 1933, Verme 1965,
Krefting and Phillips 1970). It is a widely accepted principle in deer management
that increasing habitat edge would improve deer habitat conditions, thus would
increase deer population size. However, it is not clear whether this principle will
hold at landscape scales. The second objective of this study was to determine
whether deer Show edge preference at landscape scales.

The investigation included all 15 counties in the Upper Peninsula (UP) of
Michigan, which comprise a total area of 43,070 km2 (Figure 3.1). County areas
range from 1,368 km2 to 4,836 km2, with an average of 2,871 kmz. Physical and
biotic characteristics of the UP are described in detail by Albert et al. (1986). The
eastern UP has a cool lacustrine climate, low elevation, and flat, glacial lake plain
topography with poorly drained soils composed of sand and clay. The forests of
this region mainly comprise northern conifers and northern hardwoods. Northern

conifers, such as white spruce (Picea glauca), balsam fir (Abies balsamea), and

35

northern white-cedar (T huja occidentalis), are prevalent in swamp areas while
northern hardwoods are in well-drained moraine ridges. The western UP has a
continental climate, high elevations, and plateau topography with well-drained
soils. Northern hardwoods are dominant over most areas, whereas white cedar is
less common than in eastern UP. Commercial lumbering and fire have reduced
hemlock (Tsuga canadensis), white pines (Pinus strobus) and red pines (Pinus

resinosa) over the entire UP (Albert et al. 1986).

   
 

HI I ghll

  
 
  

Ontonagon

 

 

0 ogeblc

 

Mara uette
Alger
Sc hoolcratl

 

 

 

 

Delta

 

    

- it

-l§ “‘3

when;

- II‘I‘T”
a»

W

Figure 3.1. Map of the study area Showing the names and locations of all 15
counties in the Upper Peninsula of Michigan.

36

3.1 Methods

3. 1. 1 Image acquisition, processing, and classification

The Michigan Department of Natural Resources (MDNR) inventoried deer
habitat in the UP using Landsat Thematic Mapper (TM) imagery in the growing
season of 1991 (Maclean 1994). A hybrid unsupervised-supervised approach
was applied to classify the imagery (Lillesand and Kiefer 1994). Each of the 15
counties had a separate image with a grain size of 929 m2 (100 x 100 feet).
These images were first classified into 21 classes according to the MDNR deer
habitat classification scheme (Table 3.1). Each of these classes was considered
as a separate Class of habitat patch type. The coniferous classes in the 21-class
scheme were further differentiated (by pine Species or crown closures) and
resulted in a total of 30 classes of patch type. Aerial photos and ground truthing
were used to assess the accuracy of these classifications. The overall accuracy
of the 30-class classification scheme is 81%. On the other hand, the 21 classes
were re-coded into 8 and 4 classes to more reflect deer manager's perceptions of

deer habitat (food, water, space, and Shelter) (Table 3.1, Figure 3.2).

3. 1.2 Quantification of landscape structure

There are many metrics to quantify landscape composition and
configuration. I used FRAGSTATS (McGarigal and Marks 1994) to compute
metrics that measure area, patch, edge, shape, core area, diversity, interspersion
and contagion for each landscape. For a given landscape, FRAGSTATS (raster

version) accepts image file as input, and calculates about 60 landscape statistics

37

 

m: was 922,
m: we: xeee
m _ , SE S, ; lesion. .3 ,.,...,...E; 3: ease
.. . .ees L ,,.,.,.,.-,.._,.e,...>> e Lens, .2 ,,,,, Lem;
m ucm=o>> NF wucm=o>> NF mvcmzo>>
E 5.: 5350632 H.m.>> C 5:. amtcoofiznc eo>>
or 5E BEooEBuc >5 or x_E Eoe_coo\tu3nc >5
a 58322 98:53 a 38268; 2326..
e 55:82 m 55:83.
N xmo N xmo
m “no.0“. m 395me c 80265: 50552 9 89522 Eoctoz
v 983sz m 959sz m 98325
9.358,
m mzoooweo: v 29:03 900890: v 29:08 900035:
N 69252 N 2988582 N ucm_aoeo\m< m ucmfioeo>m<
....... . N o>_,§omo>-coz N . .0>:Eomo>-coz
_. o>=flomo>éoz F 02539-52 F :35 F :35
u 820 men seen. e 830 Se gene e 8&0 we: game e 820 83 gene

 

oEocom mmm_o-v

 

mEocom mmm_o-m

 

mEmcow wmm_o-FN

 

oEocow wmm_o-om

 

.onofim cosmoEmmflo .moEEEoE .0513". mm 2an

38

058.0 £590 ”00 it 68265: ”can: t .95 3.3.3:? Hm< ..

 

ognu-----—------------—--------------------------------—----—-—-—--—------.

ou----------------------nan-----—a--------------—--------------------------u

_.N

ON

2

9.

t

or
me

5:50 35.2

.88 922,

E Emmfim

8an of;

museum xom_m

xooEo:
xomemEmH

_-----------------—---n-----------—--———onn.---—----—-—-—---—--—------u---—-ur

om
mN
mN
NN
0N
mN
VN
mN
NN
_.N
ON
9
9.

t.

00 Roonnueroo “.055.
00 «convugcoo nox=2
00 $22300 e.E>>
oo exeonvuacmo 923
00 «gonna: Emmfim
00 Reonv u: 683m
00 $076023 0E>>
OO #09828 922,
00 $onnnoosam xom_m
oo Sonvuooaam xoma
00 geonnuxooEoI

t... 00 nNee?U_oo_E.....I
SCENES.

eea A 855 .25

39

.ANEx 56.8 m__oo oom x com Go Eflxm cm 5:5
mamomucfl 85:20:05. ho tag mm; cams: OFF ..on28 5385830 v Lona: mEozmo. 09853. Co o._.amew ..N m. 059“.

                 

man 02 I
Em C .
..Mfih . noozéow I
«moeou . ._ Co~m>> I
50.52.62 .I; . . Cacao; E .

BOOB—Em:
Ucmfiacw wens
UCNEOQO 901
298.93 I
28500253535 I
wocoS.

o>=m~omo>cozEm€D ”If
vocmS.

.\
:Nmmv». ..
\n

 

mEocOm wwm_o-v oEocow mmfloé

mfltmemm. mmflo- FN ioEocom «8.0. on

x=2 5:805:51 6>>
5.2 5:80:05: to

40

at 3 levels: patch-, class-, and landscape—level. Patch-level metrics represent
information for each single patch, whereas, class-level metrics represent the
spatial patterns of a single patch type within a landscape and landscape-level
metrics represent the spatial pattern of all patch types within an entire landscape
(McGarigal and Marks 1994). Based on the assumption of an existing deer HSI
model (Bender and Haufler 1990), I used 100 m as edge width to calculate
indices related to core areas. I was particularly interested in the indices for
diversity and interspersion because it was traditionally believed that there was a
good association between diversity and deer density at local scales (Leopold
1933, Compton et al. 1988). Therefore, l calculated Shannon's, Simpson's,
modified Simpson's diversity index and evenness index, interspersion index, and
contagion index (for more details on these indices, see McGarigal and Marks

1994)

3. 1.3 Deer density measures

To measure the spatial distribution of white-tailed deer populations, I used
the following data from the MDNR: harvest, deer-vehicle accidents, and reported
kills from check stations. The harvest data were obtained by sampling about 10%
of licensed hunters by mail surveys. The deer-vehicle accidents were the
reported collisions caused by deer, which are compiled by the Michigan State
Police. The reported kills from check stations were collected through deer
hunters' voluntary reports to check stations in hunting seasons. These three
measures are commonly used indices of deer population sizes (Jahn 1959, Allen

and McCullough 1976, Verme 1989). As none of these indices provide an

41

absolute deer number, I presented these three population indices for
comparison. The deer data used in this study were for each county in 1991, in
order to coincide with the 1991 TM imagery. All the data were standardized by
the county area to the measures of deer harvest, deer-vehicle accidents, and

reported kills per 100 km2 (Table 3.2).

3. 1.4 Relationships between deer density and landscape metrics

As the primary interest was the relationship between landscape structure
and deer spatial pattern, I used class- and landscape—level indices to perform
statistical analysis. I used each of the deer density measures as a response
variable and each of the landscape metrics as an explanatory variable to do
regression analyses (Cody and Smith 1997). I tested for significance of the
relationships using F ratios. The statistical analyses were performed using SAS
statistical software (SAS Institute Inc. 1998). I examined the residuals from the
linear regression model to determine whether the linear model was appropriate. If
not, a nonlinear model was applied. As many of the metrics represent the same
information about the spatial pattern of patch types or landscape, they are
inherently redundant. To avoid this problem, I presented only those metrics
representing unique aspects of landscape structure and with highest associations

with deer density measures.

42

Table 3.2. White-tailed deer density measures (number/100 kmz) (1991

data, provided by MDNR).

 

 

County Deer harvest Deer-vehicle accidents Reported kills
Alger 73.4 7.9 5.4
Baraga 142.9 11.2 18.9
Chippewa 69.5 10.2 4.9
Delta 359.1 31.3 15.2
Dickinson 579.2 37.5 28.5
Gogebic 158.3 5.9 13.5
Houghton 92.7 6.3 9.1
Iron 417.0 19.1 30.2
Keweenaw 19.3 1.2 2.0
Luce 84.9 5.5 5.4
Mackinac 127.4 15.6 6.4
Marquette 208.5 12.2 17.1
Menominee 695.0 41.4 35.2
Ontonagon 181.5 11.6 14.8
Schoolcraft 131.3 6.6 5.5

 

43

To identify the relationships between deer density measures and
landscape metrics and to determine the effects of different measures of deer
density, l analyzed the associations between three deer density measures and
landscape-level indices calculated using the 30-class classification scheme. To
determine whether these associations would hold under other three classification
schemes (4-, 8-, and 21-class), I did analyses using the landscape index that
showed the highest association with deer density under the 30-class scheme.
Finally, I analyzed the associations between deer spatial measure(s) and class-
level indices under 4 different classification schemes to determine which class'
spatial pattern would have the highest associations with deer density.

Ideally, ecological units such as ecoregions (Albert et al. 1986) instead of
political boundaries such as county boundaries should be used to relate deer
spatial pattern with landscape structure. This boundary limitation in this study is
due to the fact that deer data were collected at a county level. Data regarding
deer spatial distribution across landscapes in ecological units are not available

for meaningful statistical analysis.

3.2 Results

Landscape structures differed among the 15 counties in the UP. Most
counties had 20 to 23 patch types, while Keweenaw had only 15 patch types.
The single largest patch, measured as the percent of the total landscape area in
a county, ranged from 1% in Dickenson to 40% in Gogebic with an average of

18.6% (i12.2%, SD). The mean patch size ranged from 6.8 ha in Dickenson to

16.0 ha in Keweenaw with an average of 11.9 ha (i 3.1, SD). Other metrics

44

showed similar variations among landscapes (Table 3.3). For example, Dickinson
had the highest patch density, whereas Keweenaw had the lowest patch density.

The regression analyses demonstrated a significant positive relationship
between deer harvest and patch density (Figure 3.3a), edge density (Figure
3.3b), and landscape shape index (Figure 3.30). However, deer harvest showed
a negative relationship with mean patch size (r=-0.84, p=0.0001) and all three
core area indices: mean core area per patch (r=-0.84, p=0.0001), mean area per
disjunct core (r=-0.78, p=0.006), and total core area index (r=-0.89, p=0.0001,
Figure 3.3d). Patch density and mean patch size represented the same
landscape attribute because both were derived from the total landscape area and
the number of patches. All the core area indices, which were computed from
patch size, patch shape, and edge width, described similar landscape attributes.
To avoid redundancy, I chose patch density and total core area index to do
further analysis because they are more ecologically meaningful and easier to
measure.

Like the deer harvest, deer-vehicle accidents and reported kills showed
similar positive relationships with patch density (Figure 3.4a, Figure 3.4b), edge
density, and landscape shape index (Figure 3.5). Similarly, both deer-vehicle
accidents and reported kills had a negative relationship with total core area index
(Figure 3.5). However, among the three deer density measures, deer harvest
showed the highest degree of association with patch density, edge density,
landscape shape index and core area index; while the reported kills had the

lowest association (Figure 3.5).

45

Table 3.3. Landscape metrics for the counties in the Upper Peninsula under 30-
class scheme. PD = patch density (number of patches I100 ha), ED = edge
density (m/ha, sum of lengths (m) of all edge segments per ha), LSI= landscape
shape index (ratio of the sum of landscape boundary and all patch edge
segments to total landscape area), TCAI = total core area index (percentage of
the landscape that is core area), SHDI = Shannon's diversity index (a combined
measure of richness and evenness), IJI = interspersion & juxtaposition index (%,
observed interspersion over the maximum possible interspersion for the given
number of patch types), and CONTAG = contagion index (%, observed
contagion over the maximum possible contagion for the given number of patch

types)-

 

 

 

County PD ED LSl TCAI SHDI IJI CONTAG
Alger 7.1 47.5 63.2 54.7 1.8 70.2 63.9
Baraga 6.3 50.7 65.5 51.0 1.7 63.5 64.4
Chippewa 7.5 57.1 101.9 45.3 2.4 76.2 52.5
Delta 10.7 67.0 104.5 40.1 2.3 73.8 53.0
Dickinson 14.7 84.9 96.3 26.8 2.2 71.2 53.2
Gogebic 7.8 57.9 82.3 45.0 1.7 62.3 64.9
Houghton 6.8 51.3 70.2 51.0 1.8 64.5 64.4
lron 10.7 74.1 105.4 34.8 1.9 62.1 60.9
Keweenaw 6.3 48.7 42.6 52.4 1.9 68.4 58.5
Luce 7.7 59.0 74.3 43.4 2.2 68.6 57.4
Mackinac 7.5 55.6 83.1 46.9 2.3 72.2 55.9
Marquette 11.3 66.9 120.1 41.3 2.1 71.8 58.0
Menominee 14.6 86.0 115.6 25.3 2.3 72.6 51.1
Ontonagon 6.9 56.9 85.3 45.3 1.7 54.5 65.4
Schoolcraft 10.0 68.5 100.0 36.6 2.2 69.9 54.0
Mean 9.0 62.1 87.4 42.7 2.0 68.1 58.5
SD 2.8 12.2 21.6 8.8 0.2 5.5 5.0

 

46

.38 98 m. 8:. 383:3 m... 06 $3588.

32. men 98 .32 .3 En .38 3838. .59 o. mEmEmom once :23 ..m ucm bongo: oomomucfl .0 Sam on. .0
one. 5:5 398 oaoomucm. 8. .8: :3 $5268 mono ..m “.0 .8. 25:0. .0 Sam. bacon omoo 3. Am: cow :3 8:23
.o .5955 385: :0st Am. Em ANEV. cow 5: Lowe 8898: .o 53:55 “moan: Loon :mofion a.:m:o=m.mm .m.m 9:9”.

30. 366:. no.6 98 .80... x35 oqmcm 383:3

 

 

 

 

 

8 8 cc 8 om 02 cm. 0: 8. 8 8 E 8 8 o4
_ . p . O . r O
r 00—. 1 Dow
- 8m 0 - 8N
. 28 m . :8 m
. 84 w . 84 w
- o8 mused" q .86 u am . 8m m.
I my? . I . Ir . . I: I a
386 u q .26 H mm 0 WWW £86230. 8.. s 3&8 o u q 8 o . am . WWW m
ovxme + xmmdm- u A a we .vmm - 369m u \A com
3 I com -
SEE V bacon omom Am: 8: t 38:23 :29.
8 8 E 8 8 2. B 3 2 we 5 2 a m m
. p O p p h . F . . p I O
I OOF I DO?
. com 0 . com
0
. 28 m - 08 a
. oov w. - oov W
. com m . 08 m
. q. . 1 com 1 886": .86 "mm 1 com .s.
o 88 o- 8 a - mm . o2 2.8m - E8 u s . OE
3.85.8.2 u. . o8

 

 

 

 

 

 

  

 

 

 

47

45-
40.
35-1

y=3.73x- 18.87
R2 = 0.73, p = 0.0001

    

30-
25‘
20'
15‘
10‘
5.1
0.
6

Deer-vehicle accidents ( # I100 kmz)

 

 

Patchdensity(#/100ha)

y = 2.81x - 11.24
re2 =0.59, p =0.0009

5.33.?

l

N
01

l

N
0010

l

A

l

—l

Reported kill ( # I100 kmz)

01
1

 

 

 

O

l r I

10 11 12
Patchdensity(#/100 ha)

13

14

15

Figure 3.4. (a) Relationship between deer-vehicle accidents and patch density
(number of patches per 100 ha). (b) Relationship between reported kills and
patch density (number of patches per 100 ha).

48

88>. :98 ..o 825: :82... 8:. .9 869:8 83.888 :5E.x8E 8:. 8>o 8.09:8 82888

.0... 8.8:. 859:8 0 05.200 8:8 .88.: :98 8 8:E:: :8>.m 8:. .9 8.888885. 85.888 E:E.me 8:. 88
8.88885. 8288 3. 88. 83.8888... 8:8 8.88885. 0 .2 88:88 8:8 888:8: u.o 85888E 8:.:E8 8.
88:. 3.8828 8.:o::m:m u .95 .888 8:8 8. 8:. 8888. 8:. 8 89:898. 88:. 888 8.8 .99 n _<_._.

..888 888:8. .99 9 9:8Em88 888 :98 ..8 8:8 58:8: 8888. 8 E8 8:. 8 2.9. 88:. 88:8 8888.

u .w._ .8: 8a 9:8Eu88 898 ..8 8 E. 8:68. 8 E8. 5.888 888 H mm. .8: cow 88 8:98 .o 885:. 3.888
:98 u an. .5... v a u 3. .36 v a u .. 88.98E 8888. :8 885888E 3.888 888 8828: 8.88.8.8”. .m.m 850.“.

8.98E 888888..

{:1 1 CO...

.2. r 8.?

 

 

1 00.0-
r 3.?

r 0N6-
.m.. cm on.

 

 

 

 

I: M

 

 

 

 

- 8.0

enleA a

1 cud

 

 

 

1 0‘0

 

 

 

 

1 00.0

 

 

 

 

 

 

 

 

u ow.o

 

as; 2: 21:; 8.88:0 .. _
3.5. 8: e. 2828... 2022-880 . 8..
:25. 8: 8 . 82:9. 880.

 

 

1 ON;

49

The deer density measures displayed no significant relationships with
landscape metrics for diversity and interspersion (Figure 3.5). However, to the
contagion index, deer-vehicle accidents showed a negative relationship (p=0.02),
deer harvest data displayed a marginal negative relationship (p=0.08), whereas
reported kills had no relationship (p=0.55) (Figure 3.5).

Among all landscape-level metrics, edge density showed the highest
associations with deer density measures under the 30-class scheme (Figure 3.3,
Figure 3.5). So I chose landscape-level edge density to examine its behavior
under 4 classification schemes. Edge density varied across the landscapes in the
UP (Table 3.4). Under all 4 classification schemes, Menominee and Dickinson
had the highest and second highest edge density, respectively. Keweenaw had
the lowest edge density under the 4- and 8-class schemes and second lowest
under the other two schemes, while Alger had the lowest edge density under
both the 21- and ISO-class schemes, and Luce and Ontonagn had the second
lowest edge density under the 4- and 8-class schemes, respectively.

For the same landscape, edge density showed a logarithmic increase as
class number increased from 4 to 21, however, the increases leveled off when
the class numbers increased from 21 to 30 (Figure 3.6). For different landscapes,
the patterns of increase in edge density behaved differently across landscapes.
For example, edge density of the Iron county landscape ranked the third highest
under the 30- and 21-class schemes, but the rank fell to the tenth under the 8-

class scheme and the sixth under the 4-class scheme.

50

Table 3.4. Landscape-level edge densities (sum of lengths (m) of all edge

segments per ha) under 4 classification schemes.

 

 

 

County 30-class 21-class 8-class 4-class
Alger 47.5 46.9 29.7 14.3
Baraga 50.7 50.4 24.9 12.4
Chippewa 57.1 56 40.1 18
Delta 67 65.8 47.7 22.6
Dickinson 84.9 84.8 60.7 36.2
Gogebic 57.9 57.9 35.8 15
Houghton 51.3 51.1 26.6 16.5
lron 74.1 73.6 34.3 18.4
Keweenaw 48.7 48 21.7 9.6
Luce 59 56.9 39.3 12.3
Mackinac 55.6 54.1 36.7 14.3
Marquette 66.9 66.4 44.7 26.1
Menominee 86 85.9 64.2 37.3
Ontonagn 56.9 56.8 22.4 13.8
Schoolcraft 68.5 67.3 50 20.2
Mean 62.1 61.5 38.6 19.1
SD 12.2 12.3 13.1 8.3

 

51

100-
90 ~*
80 -'
70 l
60 —~
50 4
4o

Edge density (m/ha)

30
20 3
10 l

O '.
1.00

+Alger
+ Baraga
/ ”\ “3" Chippewa
‘9‘“ Delta
_‘__ ‘9‘" Dickinson
‘ + Gogebic
:éf—‘é; "" Houghton

      

‘_

—" Keweenaw
—°- Luce
‘9'" Mackinac
., ‘9' Marquette
, , +Menominee
"*‘ Ontonagn
__fi , _____ . -__ .__ #"i‘fii‘LSchoolcraft

1.50 2.00 2.50 3.00 3.50 4.00

Number of classes (loge transformation)

Figure 3.6. Sensitivity of edge density (sum of lengths (m) of all edge segments
per ha) to changes in classification schemes.

52

As three deer density measures showed similar associations with
landscape metrics and deer harvest showed the highest associations with deer
density measures under the 30-class scheme, I further analyzed the relationships
between deer harvest and landscapes-level edge density under the other three
classification schemes. There was a significant linear relationship between deer
harvest and landscape-level edge density. However, the degree of associations
between them varied as classification schemes changed (Figure 3.7). The order
of associations was 21-, 30-, 4-, and 8- class scheme, from the highest to the
lowest.

The edge density of tamarack, agricultural land/cropland, and herbaceous
openland had the highest associations with deer density under 21- and 30-class
schemes as these 3 classes remained in the same classes under these two
classification schemes (Table 3.1, Figure 3.8). Under 8-class scheme, the edge
density of agricultural land/cropland and hardwood showed a strong association
with deer density measure. Under the 4-class scheme, the edge density of
forests and non-forests showed strong associations with deer density measure
as both forests and non-forests shared the most edges in the UP where the

areas of non-vegetative and water were minimal.

53

.moEozom con—85330 v 56:: Am; .8 35:63 omum
__m .6 EV 26cm. Co :53 3:6ch mono ncm ANEx 8: 59:55 60ch Emu comics azmcozflmm NM 659“.

 

     

 

 

  

mEocow mmm_o.¢ coca: 36:3 .33 9:28 mmm_o-w 56:: 369% 008
ov om om 2 o 8 oo 9. om . o
l i ll l - l + 0 ll l, . a O
m com m 2 com m
.. 08 w ‘ 8v w
M . _ M
883 n «m - com m . :80 u mm _ coo m
. 2.x: 8.2.31 A . 8.93-36»: us M
u - 8m .- 8m
0
9:28 320$ .225 32% menu 32% 8.20-8 52: £28 83
2: 8 om ow om o 2: 8 8 ow om o
.l llil i l . _ O D r 2
o A l l l l _ o
T 8N m 2- 8m «.3
. m u n...
. 03 w .4 Se w
. u M M M
.Bmmxo mm s a com m 88.0 H mm H com m
9 2» mod N3 3 r _ o mud: - x392 u x 2
.1 com com

a a

54

.3828 8385830 v 56:: mmm_o 29.8 Co Am; .3 EcoEmom mono __m
ho AEV 26%. Co :53 38:3 mono ucm ANEx co: 59:55 #5ch .mon c0058 228688 829... .m.m 9:9“.

9:28 mmm_o-m .695
89522 ho 38:3 33

9:28. www.cé 56::
8.98 .0 bacon omum

ov mm om mm ON or ovm o

    
 

 

 

 

 

     

  
    

 

  

 

.i. -i .i :l.|1 e . a -.-.ii.ii-L_fi O
2 a 0
fl CON m m
4 8w m m
82.8 H mm f 88 a o 88.8 H mm - o8 %
I . . 1. O . u . I. 1
mm 8: -88 «mus i 08 8 Eu :8 t is .. 08
u o
9:28 mmm_o-om ocm -Fm cone:
oEmcom wmm_o-w 56:: m
6:33.223 _Eazaotmm ho bacon omnm x8885 .0 bacon m pm.
2 S m o
ov mm om mm cm 3 or m o i i i: .. i ii i o
i l Iiiiiiillil o
c G i com m
i 08 w - 08 m
M o -
. 8mg H mm .- 08 a . 22.8 H mm _ 88 m
85mm + 388 n s i 28 8.2: + 5.58 n s . 08

n a

55

3.3 Discussion

Quantifying the relationships between landscape structure and ecological
consequences is a primary focus of landscape ecology (Turner 1989, Dunning et
al. 1992). The study revealed that the spatial pattern of patches had a strong
association with deer density, measured by deer harvest, deer vehicle accidents,
and reported kills. Furthermore, this study demonstrated significant relationships
between deer density measures and several landscape metrics (Figure 3.3 and
Figure 3.5). Among these metrics, patch density and edge density are good
heterogeneity indices, while landscape shape index measures the complexity of
patch shape (Li and Reynolds 1994). These findings suggest that spatial
heterogeneity plays an important role in characterizing the spatial distribution of
white-tailed deer at landscape scales.

The regression analyses also indicated that a single landscape index such
as patch density (R2=0.83) and edge density (R2=0.84) can explain more than
80% of variations of deer density measures. This suggests that these linear
regression models may be applied to predict the spatial pattern of deer
distribution at a landscape level. These models may also apply to other areas if
the same resolution of imagery and the same habitat classification system are
used. However, further studies are necessary to examine whether and how these
findings can be applied to identify deer concentrated areas within a landscape. A

crucial issue is how to scale down and to sample a landscape in order to detect

56

the areas with specified landscape attributes such as high edge density or patch
density in a landscape. Because many landscape indices are not independent
but inherently interrelated, I did not perform multiple regression analysis.

Deer density measures showed a negative relationship with total core
area index (Figure 3.3d). However, it should be emphasized that the total core
area index was obtained using an arbitrary edge width. Therefore, I recommend
using patch density, edge density, or landscape shape index to describe the
landscape structure for white-tailed deer. These indices showed strong
relationships to deer spatial distribution, are not dependent on the arbitrary edge
width, are more ecologically meaningful, and are easier to measure.

Landscape diversity, measured by Shannon's and Simpson's diversity
index and evenness index, could not characterize deer spatial distribution.
Similar findings are reported by other studies (Compton et al. 1988). The lack of
any significant relationships between deer density measures and diversity and
interspersion indices might result from inappropriate loss of information of
diversity metrics. All of these diversity indices are basically dependent on the
patch richness (number of patch types in a landscape) and evenness (proportion
for each patch type). Although proportion for each patch type may differ, most
landscapes have similar patch richness. Thus the indices might fail to measure
the diversity of the landscape, and could not characterize the spatial pattern of
deer distribution.

The landscapes containing well interspersed patches, as indicated by a

lower contagion value, tended to have a high density of deer vehicle accidents.

57

This correlation is consistent with traditional observations that the areas with
large and contiguous patches have few deer vehicle accidents, while the areas
with small and dispersed patches have more deer vehicle accidents (Allen and
McCullough 1976, Bashore et al. 1986).

Depending on classification schemes, deer showed discriminating
preference to edges. The edge density of tamarack had a higher association than
other coniferous species had with deer density measured by deer harvest under
21- and 30-class schemes. The edge density of hardwood and agricultural land
had higher associations than softwood and wetland had.

Applying the study results from one scale to another scale is a challenging
issue in landscape ecology (Turner et al. 1989). Previous observations have
indicated that deer prefer edges at local scales probably because the edges
provide browse and are close to good shelter (Leopold 1933, Verme 1965,
Krefting and Phillips 1970). This study showed that edges played a significant
role in deer distribution at landscape scales as well. Thus, I concluded deer show
edge preference at large as well as small scales. This suggests that deer may be
scale-independent regarding habitat selection. However, it is unclear how edges
at local scales and at landscape scales interact to affect deer spatial distribution.
Both empirical and theoretical studies at multiple scales are warranted to explore
the effects of edge distances, areas, and configurations on deer spatial

distribution across heterogeneous landscapes.

58

Chapter 4

A KNOWLEDGE-BASED SYSTEM
FOR WHITE-TAILED DEER MANAGEMENT (DeerKBS)

Wildlife management is a blend of science and art. The previous two
chapters emphasize the importance of the scientific investigations of deer
population and their habitats. DeerMOM is developed to evaluate different
management options to see how harvest scenarios would affect the population
dynamics (Chapter 2). Deer managers can use this model to select the best
scenario in order to reach their objectives in a defined area during a specific time
frame (Xie et al. 1999). The spatial analyses are performed to evaluate habitat
conditions on landscape scales to see which landscape attributes would be more
associated with deer distribution than others (Chapter 3). Based on these
analyses, wildlife managers can plan and manipulate landscapes either to
improve habitat conditions in order to increase deer population or to deteriorate
habitat conditions in order to decrease crop damage and/or deer-vehicle
accidents.

What makes wildlife management as a practice differ from wildlife science
as a discipline is the importance of art in management. Some argue that "too
much of wildlife management is today still more of an art than a science" (Walker
1998). To make effective management, wildlife managers need to employ
expertise to balance the delicate relationships among all relevant factors to
address an array of complex issues. For example, wildlife stakeholders such as

hunters, farmers, and environment conservationists are demanding more

59

involvement in the process of management decision making. Global changes,
rapid species endangerment and extinction awake the public, and many
government agencies make biodiversity conservation mandatory in any natural
resources management (NRC 1995, Swanson 1997). As more and more
problems emerge from the humans shortsightedness or mismanagement, the
management paradigm is shifting from small scale, single species management
to large-scale adaptive ecosystem management (Walters 1986, Crow and
Gustafson 1997). Although this new paradigm sounds promising in solving the
problems caused by traditional management systems, decision-making risk is
exponentially increasing with the increase in management scales. Experience
and expertise are scarce or not available. Mismanagement may result in
ecological and economic disasters.

New technologies offer promises to overcome some difficulties for
adaptive ecosystem management. Global positioning systems (GPS), remote
sensing (RS), and geographic information systems (GIS) can be utilized to collect
and store huge amounts of spatial and temporal data of wildlife species and their
environment (Franklin 1994). Database management tools, such as data mining
and data warehousing, offer the capacity of finding patterns from historical data,
thus transforming a deluge of data into valuable information, and then from
information to useful knowledge (Singh 1997).

In science, it is a principle to let data speak out. In management, however,
data do not make decisions, only humans do. The overwhelming amount of data

and information can help wildlife managers better understand the system, but

60

also add another dimension of complexity to the decision making processes.
Complexity makes it difficult for stewards of the wildlife to have all the relevant
information and knowledge (Starfield and Bleloch 1986). However, decisions
must be made, even with conflicting goals and scientific uncertainties but without
complete data or scientific consensus (Ludwig et al. 1993, Ehrlich and Daily
1993). It is the wisdom of wildlife managers who apply the accumulated data,
information, and knowledge from researches and from the technologies to their
decision-making processes.

Knowledge-based systems (KBSs), a branch of artificial intelligence (Al),
can be used to help wildlife managers take advantage of the accumulated data
and information to make more informed decisions. Generally speaking, KBSs are
computer systems that use human knowledge to solve problems that normally
would require human intelligence (Edwards 1991). KBSS differ from other Al
programs in their performance, domain-specific problem-solving strategies, and
their justification for conclusions. KBSs differ from conventional software
programs in their symbolic representation, symbolic inference, and heuristic
search (Hayes-Roth et al. 1983).

KBSs for natural resource management have been studied since the
beginning of KBS research. The first successful KBS application on natural
resources is Prospector for mineral exploration in the late 19705 (Duda et al.
1979). Most of the early applications are limited to interpretation and diagnosis
task of geological and mineral problems. Production rules are the predominant

type of knowledge representation schemes. Since the late 1980s, KBSs have

61

expanded applications to management for forestry, water resources, and
ecosystems. The book written by Schmoldt and Rauscher (1996) may be the first
one that systematically introduces the concepts of Al, KBS, and knowledge
engineering life cycle to and integrates them into the system development for
natural resource management. The authors introduced the whole life cycle of the
development of a Red Pine Forest Management Advisory System (RP-FMAS)
using Prolog in which predicate logic was selected for knowledge representation.
A nice feature for this system is that it uses hypertext so that the system can be
used as an alternative to a manual. Bremdal (1997) reviewed 27 KBSs for
management of natural resources. Among them, 11 systems are for geological
exploration and mineral analysis, 6 for ecosystem management, 5 for forestry
management, 4 for water resources management, and only 1 for wildlife
taxonomy (whale visual identification).

A noticeable trend in KBS development is the integration of system
modeling, database management, and GIS into a decision support system to
provide a framework tool for decision making. The Ecosystem Management
Decision Support System (EMDS), which is developed by the USDA Forest
Service Pacific Northwest Research Center, integrates GIS and knowledge
based reasoning technologies. It uses NetWeaver engine for knowledge-based
reasoning and object- and fuzzy logic-based networks for the knowledge
representation (Reynolds 1999).

In this study, I constructed a deer knowledge-based system (DeerKBS)

that concerns white-tailed deer management in the Upper Peninsula (UP) of

62

Michigan. The major objectives of this study were three-fold: to provide a unifying
decision-making framework for deer management, to help users identify their
management issues and make appropriate decisions, and to standardize the
management procedures and use the system as a communication and

educational tool.

4.1 System Components of DeerKBS

The main components of a KBS include knowledge base, reasoning
engine, and user interface. As a database is a collection of data, a knowledge
base is a collection of knowledge. There are many different ways to represent
knowledge, such as rules (Newell and Simon 1973, Shortliffe 1976, Davis et al.
1977, McDermott 1982), frames (Minsky 1975), scripts (Schank and Abelson
1977), objects (Goldberg and Robson 1983), and networks (Quinlan 1986). In
DeerKBS, I use rule-based knowledge representation because it has some
desirable properties for system development. For example, each rule represents
a "chunk" of knowledge for a specific domain (Duda and Gaschnig 1981). The
rules' properties of uniformity (rules have the same structure) and modularity
(rules are grouped) also make the development and maintenance easier
(Schmoldt and Rauscher 1996). In practice, wildlife managers utilize their "rules
of thumb" in their decision making. It is relatively easy to ask experts to express
their thoughts out as rules, to code their thinking processes into rules, and to ask

the user to test the rules.

63

Reasoning engine is a set of reasoning methods that determine the
mechanisms by which the system makes decisions based on knowledge stored
in the knowledge base. Decision making or problem solving, in terms of artificial
intelligence, is a process of search for a solution (a state) from a collection of
solutions (a state space). Two mechanisms are involved in this process:
inference and control. In DeerKBS, a decision tree search is used as inference
that generates new assertions. A decision tree is a hierarchical structure with
each node (a branching point in a tree) determining which lower node will be
searched, and each leaf (terminal point) representing a solution. The nice
property of a decision tree is that the tree structure offers the control strategy and
the nodes allow for inferences (Quinlan 1986). Both backward chaining and
forward chaining are used in DeerKBS as control methods that determine how
the system seeks solutions. Forward chaining is a data—driven search strategy,
whereas backward chaining is a goal-driven search strategy. The last component
of KBS is the user interface. In DeerKBS, graphical user interface (GUI) is used

to ease the system usage.

4.2 System Development of DeerKBS: An Overview

A traditional life cycle for software engineering follows a waterfall model
(Boehm 1976) that includes 5 major phases: specification, design,
implementation, testing, and maintenance. The model requires that complete
specifications and assessment criteria for each phase can be provided before

system design and implementation, so the overall development remain linear. A

64

knowledge engineering life cycle usually follows a spiral model (Boehm 1988)

that has similar components with the waterfall model but is characterized by

iterative design and prototyping. One of the widely used spiral models in KBSs is

the life cycle model proposed by Buchanan et al. (1983) that includes the

following 5 iterative steps:

1.

2.

Identification: identify problem characteristics
Conceptualization: find concepts representing knowledge
Fonnalization: design structure organizing knowledge
Implementation: formulate rules embodying knowledge, and
Testing: validate rules organizing knowledge

In Buchanan’s model (1983), knowledge acquisition is an ongoing central

activity across different development stages and the system development is

cyclic processes. A variant of this model (Awad 1996), which focus on structured

tasks, includes the following stages:

1.

2.

Problem identification (system requirements and knowledge analysis)

Selection of KBS tools

. Representation (knowledge acquisition, prototype, verification and

validation)
Implementation, and
Operation and maintenance.

The development of DeerKBS roughly follows this model. The first phase

is domain problem identification that leads to the system requirements and

problem definition. Once the problem is defined, a KBS tool is determined based

65

on system requirements. Then the concern of the development focuses on
knowledge acquisition, prototyping, and validation. Following this phase are

system implementation, operation, and maintenance.

4.3 Domain Problem Definition

Deer management is the management of deer population, habitat, and
people (Giles 1978, Decker et al.1992). It usually takes a long time for a deer
manager to gain enough experience to effectively manage deer population and
habitats and to gain support from all stakeholders. Moreover, as government
agencies encourage employees to retire early, many experienced wildlife
managers have retired, or will retire soon. Retaining their valuable expertise is
important for management agencies to prevent loss of institutional expertise and
to provide more consistent decisions.

Problem defining is the process that identifies the scope, requirements,
and objectives of the problem-solving system. Before I define the problem, some
background for deer management is necessary to understand the problem.
Although it is often difficult to make distinctive judgments about how many deer
are best in an area, deer managers make their judgment based on their own
observations and experience. The areas with few deer usually have insufficient
or low quality deer habitat. These areas can be selected for deer habitat
improvement if more deer are desirable. Hunting recommendations can be
restrictive to antlerless deer harvest and thus to increase deer population size.

For those areas with too many deer (in terms of habitat and social carrying

66

capacity), hunting recommendations can be harvesting more antlerless deer. In
the areas with adequate population, hunting antlerless deer can control the
population as deer population in Michigan has potential to increase at a rate of
20-40 percent each year (McCullough 1979, Xie et al. 1999).

Deer managers often establish deer objectives at hierarchical
management levels (deer management unit (DMU), district, region, and state)
based on their assessments of deer population, habitat status, and stakeholders’
attitudes. The long-range objective in Michigan is a fall population of 1.3 million
deer, 35% of which should be antlered bucks (MDNR 1993). Each year, deer
managers recommend to have more, fewer or the same number of deer for each
management level compared with that in the previous years. These
recommendations are based on the following considerations and assessments
(Langenau 1994, Ozaga et al. 1994):

. Deer status: including deer population density and the percentage of
antlered bucks in the population.

0 Habitat status: including habitat quantity and quality

. Cultural carrying capacity (CCC): the range of population levels
acceptable to all stakeholders such as deer hunters, farmers, and other

interest groups (Minnis 1996)

. Environment conditions: winter conditions measured by winter severity

index (Verme 1965)

DeerKBS divides the whole task of decision-making for deer management

into a few small subtasks. Each subtask can be further divided into sub-subtasks.

67

lf DeerKBS is viewed as an intelligent agent, an application performing some
tasks for the user (Prerau et al. 1997), then each sub-task can be assigned to a
subagent that is an expert who makes his or her own decisions in a specific field.
For example, the subagent for deer population management can be an expert or
a panel of experts who specialize in deer population evaluation and decide
whether the population is good, too high, or too low. The outcomes of DeerKBS
can be a suite of recommendations for hunting, habitat modification, and
education based on all the inputs from one or many panels of experts. This study
will focus on recommendations for deer hunting. One major objective of DeerKBS
is to provide a general KBS framework for deer management, although the
knowledge base is built based on deer management in the UP of Michigan.

DeerKBS is intended for use by wildlife managers, researchers, and educators.

4.4 Selection of KBS tool: XpertRule KBS

XpertRule KBS (Attar Software Limited 1996) was chosen for DeerKBS
development. XpertRule KBS is a shell tool for the graphical development and
maintenance of knowledge-based applications. An XpertRule application is
constructed graphically as a hierarchy of chained tasks under Windows
environment. A task can consist of a decision tree representing a flow chart
controlling procedures, dialogs, reports or other tasks. Complex knowledge can

be structured into a hierarchy of chained tasks.

68

4.5 Knowledge Representation

Knowledge engineering is the process that centers on the process of
transforming the expert knowledge and expertise to a knowledge base. This
process can be further divided into 3 processes: knowledge structuring,

knowledge acquisition, and knowledge validation.

4. 5. 1 Knowledge structuring

Knowledge structuring is the process of transferring the problem-solving
conceptual model into a structure hierarchy of decision-making tasks that relate
the management alternatives with all the considerations. In DeerKBS, I identified
8 attributes that are related to 4 sub-tasks: assessment of population, habitat,
CCC, and weather. These attributes are: (1) deer population size, (2) buck
percentage in the population, (3) deer habitat quantity, (4) deer habitat quality,
(5) deer population size expected by deer hunters, (6) deer population size
expected by farmers, (7) deer population size expected by other interest groups,
and (8) winter severity index.

The outcomes of assessment for deer population size are low, good, or
high. The outcomes of assessment for other attributes are listed in Table 4.1.
Once assessment for each attribute is finished, further assessment for some
attributes can be performed to obtain a higher level assessment. For example,
habitat quantity can be lack, average, or abundant; and habitat quality can be
poor, medium, or high. The outcomes for overall habitat are assessed as poor,

moderate, or excellent (Table 4.1).

69

Table 4.1. Attributes in DeerKBS and their assessment outcomes

 

 

 

 

 

 

 

Attributes Outcomes Outcomes at a higher level
Deer population size low, good, high
Buck percentage low, average, high
Habitat quantity lack, average, abundant

poor, moderate, excellent

Habitat quality poor, medium, high
Hunters expectation fewer, same, more
Farmers expectation fewer, same, more fewer, same, more

 

General public expectation fewer, same, more

 

 

Winter severity index favorable, average, harsh

 

 

 

4. 5.2 Knowledge acquisition

Knowledge acquisition is the process of acquiring the knowledge of the
individual tasks from expert and literature. I acquired knowledge from experts by
interviewing deer managers and biologists and from documentation by searching
literature and the Michigan Department of Natural Resources (MDNR) reports.
The knowledge was represented as a decision rule. For example,

Rule l:

IF population density is high, and
habitat condition is moderate or poor, and
social carrying capacity is fewer deer
THEN recommend harvesting more deer
As deer managers gave only some typical management scenarios, I used

these typical scenarios as examples (referred to as example table) to generate

all the possible combinations of attributes (referred to as truth table). Based on

70

 

these tables, decision trees were generated by induction. If the information or
knowledge was not available to generate the rules, I made some educated

guesses.

4. 5.3 Knowledge Validation

Testing knowledge base involves examining the correctness and
completeness of the overall knowledge base. Tests were performed at two
levels. The first level was the validation of each individual decision rule. I present
the attributes for each rule to deer managers and then compare the decisions
from them and from DeerKBS. The second level was the validation of all rules to
eliminate redundant and conflicting rules. The decision tree that XpertRule
induced from example table could detect conflicting rules visually, so the
completeness of knowledge base could be guaranteed. During the development
of DeerKBS, I also demonstrated the system to deer biologists and wildlife
managers in the Wildlife Division at MDNR, compared the decision results from
them and from DeerKBS, and requested their suggestions for system

improvement.

4.6 System Implementation

DeerKBS was developed using XpertRule KBS under Microsoft
Windows® environment. The deer management decision making task was
structured into 4 sub-tasks: assessment of population, habitat, CCC, and

weather. Once these subtasks were finished successfully, decisions were made

71

to give pertinent harvest recommendations. These recommendations were

further refined by considering buck percentage in deer population.

4. 6. 1 Subtask for assessing population

Population assessment is the first task that deer biologists and managers
need to deal with. Generally speaking, population assessment can be divided
into two subtasks: assessment of population size and assessment of population
structure. Population size and structure are vital parameters in deer
management because they are bases for decision-making. DeerMOM can be
used to project population size and structure under different harvesting scenarios
(Chapter 2). Although many efforts are made to estimate the absolute number of
deer in the wild, many biologists and wildlife managers emphasize the
importance of relative abundance of deer. Management decisions are often
made based on the population size relative to biological carrying capacity (BCC),
CCC, and the trends of deer population changes, rather than the absolute
number of deer population (Hayne 1984).

On the basis of McCullough's (1979) long-term study in George Reserve
of Michigan, the maximum carrying capacity was estimated to be 35 deer/km2 (90
deer/miz), and the optimum carrying capacity to be 19 deer/km2 (50 deer/miz). On
some farmlands in Wisconsin, the carrying capacity was estimated to be 40
deer/km2 (McCafferey 1989). The management objective for population size in
the UP of Michigan is 372,500 in the fall, or at an average density of 9 deer/km2
(23 deer/miz) (MDNR 1993). In DeerKBS, deer population density is classified as

high if it is above 35 deer/miz, low if below 20 deer/miz, and good if between 20

72

and 35 deer/mi2 (Figure 4.1). For users' convenience, I used square miles rather

than square kilometers as area unit.

PopulationSize

     

 

”’PopDensityD>
< 20.00—-

>= 20.00—(PopDensityn>

< 35.00—-
>= 35.00@

 

 

 

Figure 4.1. Subtask for population size assessment, where PopulationSize
represents population size, and PopDensity represents population density.
Population structure is the other aspect of population assessment. As the
deer population increases, there are more hunting opportunities for deer hunters
and more recreational opportunities for the public. However, the cost of abundant
deer is high in some areas. Relatively high deer population have unacceptable
economic impact on the society caused by crop damages (Allen and McCullough
1976, Conover and Decker 1991) and deer-vehicle accidents (Conover et al.
1995), disease transmission (Wilson and Childs 1997) and ecological impact on
the ecosystem caused by influencing forest regeneration (Alverson et al 1988).
To improve the quality of deer hunting and to minimize the negative impact of
high deer densities, deer management in Michigan is aiming at producing more
bucks but fewer deer. One of the objectives in quality deer management in

Michigan is to have a 35% of antlered bucks in deer population. In DeerKBS, the

73

percentage of antlered buck (or buck ratio) was classified as high if it is above

35, low if below 20, average if between 20 and 35 (Figure 4.2).

 

<BuckPerce nil}

(_-
>= 20.0—<BuckPercent;/
F 35.0—-
>= 35.0—-

 

 

 

Figure 4.2. Subtask for antlered buck percentage assessment, where
BuckPercent represents the percentage of antlered buck in deer population.

4. 6.2 Subtask for assessing habitat

There are many approaches to assess deer habitat. The spatial analysis
of the deer distribution across the UP is an approach to assessing deer habitat at
the landscape level (county scale). Using a single landscape index such as edge
density can quickly assess the overall habitat condition at large scales. If
management recommendations need to be made at the landscape level, the
landscape analysis can give an assessment of overall habitat using some
landscape metrics. However, deer biologists and deer managers often assess
the habitat in a much smaller scale (Krefting and Phillips 1970, Ozaga et al.
1994, Braun 1996) by following habitat assessment procedures and habitat
suitability index (HSI) (Short 1986, Bender and Haufler 1990). In DeerKBS, only
high level evaluations of habitat quantity and habitat quality were considered.

Habitat quantity was evaluated as lack, average, or abundant. Habitat quality

74

was evaluated as poor, medium, or high. Overall habitat was evaluated as poor,
medium, or excellent (Figure 4.3). For example,

Rule ll:
IF habitat quantity is abundant, and

habitat quality is high or medium
THEN overall habitat is excellent

 

{:HabltatQuantltyl)
FAbundant—<HabltatQualltyn:>

High, Medlum-
Poor-

HabitatQuality.
High-
Medium—-
Poor—-

flack—cfHabitatQualitylj‘;

”'9"-
Medlum, Poor—-

 

 

 

     

 

 

 

Figure 4.3. Subtask for habitat assessment, where HabitatQuantity represents
habitat quantity, HabitatQuality represents habitat quality; Abundant, Average,
and Lack are the outcomes of habitat quantity assessment; High, Medium, and
Poor are the outcomes of habitat quality assessment; Poor, Moderate, and
Excellent are the outcomes of habitat assessment.

4. 6.3 Subtask for assessing Cultural Carrying Capacity

Cultural carrying capacity is the wildlife population level that is acceptable
to people (Decker and Purdy 1988, Minnis 1996). Measuring public attitudes has
been the subject of considerable interest by managers (Arthur and Wilson 1979).

While objectives based on BCC are an ecological consideration made mainly by

75

biologists, objectives based on CCC are a more popular decision guided by
wildlife managers (Strickland et al. 1994). Usually, the population objectives
imposed by CCC are well below those imposed by BCC. For example, on some
farmlands in Wisconsin, CCC is about 12 deer/kmz, while BCC is about 40
deer/km2 (McCafferery 1989). As expected, CCC is highly subject to how much
weight the decision-makers put on each stockholder' preferences. In a region
where deer hunters are well organized, they often exert greater influence than
other groups do in the decision-making processes. On the other hand, in a region
where farmers are well organized, they may have more impact than other groups
in shaping the deer management decisions. In DeerKBS, deer hunters, farmers,
and other interest groups (therefore other public) were considered to be the
major stakeholders. The expected deer population size for each group was
evaluated to have more, to have the same, or to have fewer deer. Once the
assessment for each stakeholder was finished, the overall CCC was evaluated to
have more, the same, or fewer deer compared with the previous years (Figure
4.4). For example,
Rule lll:
IF deer hunters expect to have more, and
farmers expect to have fewer, and
other public expect to have fewer
THEN overall of CCC is to have the same number of deer
In this rule, more weight was put on the expectation of deer hunters and

less weight on the expectation of farmers and the other public. In other words,

not every stakeholder was given the same consideration. In this example,

76

 

HunterExpected

 

 

—More—c;f Farm erExpected.)

tSame. More-

Fewer—(PublicExpected;

Same, Morr-
Femu--

L-Same-—'.';PuhlicExpectedy}

were-

-Same-<I:FannerExpected!>

More-
Same—-

Fewer Fewer

4mm

—Fewer—<FarmerExpected;/'=
~More—<PublicExpected;‘>

M...-
Sarne. Fewer—-

t-Sarne, Fewer-

 

 

 

 

 

 

Figure 4.4. Subtask for CCC assessment, HunterExpected, FannerExpected,
and PublicExpected represents the expectation for deer numbers by hunters,
farmers, and the other interest groups. Fewer, Same, and More are the

outcomes of CCC assessments.

77

if equal weight was given to each stakeholder, the outcome should favor the
groups who expected fewer deer because two groups expected fewer deer and
only one group expected more deer. These rules could be easily changed to
reflect the reality in the decision-making process specific to different

management units and management objectives.

4. 6.4 Subtask for assessing weather

There are many environmental variables that influence deer population
dynamics by affecting their movement and food availability. Winter conditions are
considered to be a critical factor that impacts deer natural mortality and
productivity (Verme 1969). DeerMOM incorporates winter conditions into the
model and can be used to project the population dynamics under different winter
conditions measured by WSI (Chapter 2). Historically, WSI in the UP ranged from
about 50 to 150 with an average of approximately 100. In DeerKBS, weather
condition is evaluated as favorable if WSI is below 80, harsh if above 120, and

average if between 80 and 120 (Figure 4.5).

tum
>= 80.0— WSID

tum
>= 120.0-

Figure 4.5. Subtask for winter weather assessment, Favorable, Average, and
Harsh are the outcome of the assessment of winter conditions.

78

4. 6.5 Decision tree and task for decision making

Putting all the information together, a decision tree was generated as
partly shown in Figure 4.6. The tree structure offers the control strategy and the
nodes allow for inferences. Each node, represented as a hexagon in Figure 4.6,
determines which lower node will be searched. Each leaf, represented as ellipse
in Figure 4.6, indicates a solution. For example, if population density (node
PopulationDensity) is high, the search will evaluate habitat (node Habitat). If
habitat is poor, then recommendation is made to harvest more deer (leaf More).
If habitat is excellent, then CCC (node CulturalCarryingCapacity) is further
considered, and harvest recommendations (leaf Same or More) are made based
on the outcome of CCC evaluations.

An emerging feature in the decision tree is that not all decision outcomes
follow the same path, as demonstrated in the aforementioned example and in
Figure 4.6. This feature simulates some decision-making processes in wildlife
management. For example, if population density is high and deer habitat is poor,
wildlife managers may recommend harvesting more deer without considering

CCC for the purpose of long-term deer sustainability.

79

 

PopulationDensity Q
—High—<yabitat‘><
t-ExceIIant-{:0ulturalCarryingCapacity>d

tMore, Samr-b
Fewer-b

-ModerateficulturalCarryingCapacity")d

Mme-»
Same. Fewer-.5

8......»

—-Good—<::}_Iabitat > 4
—Excellent—{CulturalCarryingCapacity><l
Hume-<7 Weather <1

Favorable-D
Average. Harsh-b

a, Weather <I
FavorabIr-D
Average-D
Harsh—.D

 

 

 

 

 

 

 

   

 

 

 

Figure 4.6. Part of a decision tree showing how a preliminary recommendation
was made based on the outcomes of assessments of deer density, habitat
conditions, cultural carrying capacity, and weather conditions. Note that hexagon
represents node and ellipse represents leaf (outcome). Left-pointed triangles
represent backward chaining, and right-pointed ones represent fonivard chaining.

80

4. 6.6 Decision procedure

Once every attribute has been evaluated, each sub-task in a specific sub-
domain has its evaluation outcome. Population density assessment can have an
outcome of low, good, or high. Habitat assessment results in an overall outcome
of poor, moderate, or excellent. CCC evaluation results in an outcome of more,
fewer, or the same. Weather evaluation results in an outcome of favorable,
average, or harsh. Now it is up to deer managers to make decisions based on
their expertise and knowledge. As Ed Langenau (1991) stated, 'Wildlife
managers assess deer populations and public attitudes to establish deer herd
objectives at a local level. Annual recommendations are made to have more,
fewer, or the same number of deer within each DMU". Obviously, these
recommendations are made in the context of previous years' deer management
practices. All harvest quotas are compared with the previous years' quotas. Then
they recommend the hunting licensee quotas for antlered bucks and antlerless
deer at each DMU based on their experience in previous years and the
management agency's policies and mission.

In DeerKBS, preliminary recommendations are made based on all the
outcomes of assessments of population density, habitat conditions, CCC, and
weather conditions (backward chaining process). DeerKBS then refines its
management recommendations by considering the buck ratios in deer population
(forward chaining process) (Figure 4. 7). For example,

Rule lV:

IF preliminary recommendation is to harvest more deer, and

buck ratio is low
THEN recommend harvesting more antlerless but fewer bucks

81

 

[HarvestRecommendation]

 

—<l-180cialCarryin§£apacltyl
“
_D—{Hawestlvlorel
—l>-lHarvestSame]

—l>-IHarvestLess]
4 BuckRatlo

 

 

 

 

 

 

Figure 4.7. Diagram of a decision-making procedure. HarvestRecommendation
represents the preliminary harvest recommendations that were made based on
the assessments of population density, habitat, CCC, and weather. The
recommendations were further refined by considering antlered buck percentage
in deer population. Note that left-pointed triangles represent backward chaining,
and right-pointed ones represent forward chaining.

82

4.7 Operation and Delivery of DeerKBS

DeerKBS works under Microsoft Windows® environment using XpertRule.
Background data, information and images were incorporated into the user
interface (Figure 4.8). Explanation mechanism, which is built in the XpertRule
KBS, keeps track of the decision process. The user can use this feature to
understand how DeerKBS works. DeerKBS can be modified through example
tables or decision trees. System inputs include attributes of deer population size,

size of interest area, evaluation results for habitat quantity and quality, evaluation

{1 Deer population size in the fall B

DeerKBS: a knowledge-based system for white-tailed deer management
Jialong X ie, December 19991

 

 

i 'This system captures deer managers' expertise on
iwhitetailed deer management. It is desimied to help
users to make more informed decisions based on the
evaluations of deer population status. habitat
conditions. and socioeconomic factors.

  

 

Data: The fall deer herd in the Upper
Peninsula of Michigan: 897,000 in
1995, 624,000 in 1996, 474,000 in
1997. The fall deer herd in the entire
state is 1.8 million in 1997.

FYI: "Wildlife managers assess dev
populations and public attitudes to
establish deer herd objectives at a
local level. Annual rerommendatior
are made to have more, less. or the
same number of deer within each
DML -- _ Ed Langenau (DNR Big Please input fall population size for
i G ante Specialist) area of interest.

Huck; Exit l Savel Printl How I 650000 I BE.”

 

Figure 4.8. Graphic user interface of DeerKBS showing that background
information and images are incorporated into the user interface. Explanation is
provided by clicking the "How" button.

83

results of expected deer population size for deer hunters, farmers, and other
interest groups, percentage of antlered buck, and WSI. Through dialogs, users
input numbers for numerical attributes or select a choice from a list of values for
logical attributes. System output is a set of harvest recommendations based on
the assessment of all the attributes. The output can be printed out or saved into a
file. A demonstration of the operation is included in Appendix B.

The system can be delivered to end users as a binary file with or without
password protected. End users can use XpertRun - the XpertRule KBS run time
program from Attar Software Limited - to run DeerKBS. The end users can use
DeerKBS but cannot change the system's structure or knowledge base without a
fully copyrighted XpertRule KBS system.

The knowledge base in DeerKBS can be exported to production rules both
as a text file (Appendix C) or a Visual BASIC function code. So these rules can

be embedded in other programs as an intelligent agent (Prerau 1997).

4.8 Conclusions and Discussions

In DeerKBS, the task of deer management decision making is divided into
an array of decision-making subtasks. These structured decision tasks can help
decision-makers better utilize the knowledge and expertise from different specific
fields. The final decisions are made based on the inputs from experts in deer
population, habitat, and human dimensions. Thus the decisions are more
informative, democratic (Schmoldt and Rauscher 1996), reliable, and effective

than a decision made by a single specialist.

84

The "top down" technique used in building DeerKBS offers the flexibility to
include more information once it is available. For some sub-tasks such as habitat
assessment, only the highest level attributes are included because the data for
lower levels are not available. If there is more detailed information available,
another level of details can be easily incorporated into the system. For example,
if I have sampled a specific area and can estimate the biomass for each species
of food, I can calculate the total amounts of biomass of food and convert the
amounts to a relative assessment of habitat quantity. I can also estimate how
much of the biomass is preferred food for deer and then evaluate the habitat
quality. If HSI is available for a specific area, a decision tree can be added to
convert the numerical HSI to qualitative assessment of habitat. Landscape
metrics such as edge density in the landscape analysis as demonstrated in
Chapter 3 can also be converted to qualitative assessment of habitat. This
feature is important in deer management because wildlife managers have to
make decisions even without complete data and scientific consensus. Very often,
they resort to their judgment, intuition, and experience to make decisions
(Starfield and Bleloch 1986).

Communication between policy makers and stakeholders is important to
achieve management objectives effectively (Kellert 1994). However, deer
managers lack an instrument to effectively communicate with other stakeholders.
DeerMOM can be used to demonstrate how management options and winter
weather conditions influence population size and structure (Chapter 2). But very

often, stakeholders want to be involved in the decision-making process.

85

DeerKBS provides a framework of the decision-making process for deer
management. Major ecological and socioeconomic factors are considered in this
system. The user can easily identify the management issues related to deer
harvest, habitat management, and different interest groups. As DeerKBS
provides a platform (common vocabulary) for all stakeholders, it can be an
effective communication tool for deer managers to present their decision-making
processes to other stakeholders and improve their accountability. With this tool,
deer managers can also exchange their decision-making processes with their
peers so that wildlife management can enhance collaboration among managers,
scientists, and interested stakeholders (Franklin 1994).

DeerKBS can be used for other purposes. For educators, it can be an
educational tool to help students understand the complexity of wildlife
management and the decision-making processes of wildlife management. For
wildlife management agencies, DeerKBS can be an effective tool to standardize
the management decision procedures. When this system was demonstrated to
the wildlife biologists and managers at the Wildlife Division of the MDNR, it was
considered as a pioneer study to standardize the diverse management decision
processes happening in field offices of the MDNR (Dale Rabe, MDNR, personal
communication).

Though DeerKBS has the potential in many aspects of deer management,
there are some limitations. First, the knowledge base in DeerKBS is based on
interviews with deer managers, literature, and some educated guesses. The

validity of its knowledge base is not fully tested. This is mainly due to the difficulty

86

of knowledge engineering. The process of transferring the expert knowledge into
a knowledge base is still the bottleneck for developing a KBS. Since the 1980s,
research efforts have been targeted to tackling this problem. Based on
knowledge-level hypothesis (Newell 1982, Sticklen 1989), some new approaches
such as heuristic classification, generic task (Chandrasekaran 1993), and role
limiting methods (McDermott 1988) are proposed to bridge the gap between
expert knowledge and representation constructs. Founded on these hypotheses
and approaches, system shells such as CSRL (Bylander and Mittal 1985) and
KADS (Breuker and Wielinga 1989) are developed. These tools may have broad
applications, but they are still more research-oriented and not commercially
available.

In this system, XpertRule KBS shell is used because it is an application-
oriented tool for KBS application development and maintenance (Attar Software
Limited 1996). XpertRule is a useful tool for rapid application design and
prototype. Its graphic user interface and decision tree are easy to learn and use
and helpful to communicate with wildlife managers. However, no matter how
technologies are advanced, the difficulty of eliciting the expertise and knowledge
from experts is not yet overcome. In DeerKBS, interviews and written documents
are two major knowledge sources. Although there are many interview techniques
(Guida and Tasso 1994) to stimulate the expression of knowledge and expertise,
most of these techniques are time-consuming and not efficient and require rich
personal experience. So DeerKBS relies much on written documents and my

educated guesses. The problem with this technique is that the knowledge elicited

87

must be verified by some appropriate ways (Guida and Tasso 1994), preferably
by deer managers. However, many wildlife managers are often reluctant to
share their decision-making processes. Like other KBS applications, validity of
recommendations from DeerKBS is dependent on the validity of the decision
rules in the knowledge base. The decision rules are difficult to validate, as the
rules are not universal. Each manager has his or her own "rules of thumb" in
practice, making cross-examination more difficult.

In DeerKBS, eight attributes are considered in the process of decision-
making. If each attribute has 5 discrete values, there are 390625 (58)
combinations. To avoid this kind of combinatorial exploration, 2 methods are
used in DeerKBS. First, the main task is structured into 4 subtasks. Second, for
logic attributes, only three discrete values (e.g., low, good, or high for population
density) are used. So the total combinations are reduced to 81 (34). To further
reduce the possible outcomes (or number of rules), some attributes are ignored
on purpose, as demonstrated in the example presented in section 4.6.5. and in
Figure 4.6.

The rules in the knowledge base need to be modified to reflect the
management reality in a specified area and under a specified time frame. For
example, winters in southern Michigan are much warmer than in the UP, so the
evaluation criteria for winter conditions should be changed. Moreover, DeerKBS
has no capacity to learn, so it is not so intelligent (Schank 1987) and cannot
replace deer managers. Nevertheless, DeerKBS can be an assistant or

educational tool to deer managers because it integrates all relevant information

88

into the decision-making process and provides a unifying decision-making

framework for deer management.

89

Chapter 5

SUMMARY

Effective management of white-tailed deer should integrate wildlife
science, modern technology, and wildlife managers' expertise into the
management decision processes. This research combined deer population
modeling, landscape structure analyses, and socioeconomic considerations into
the decision support system for deer management in the Upper Peninsula (UP)
of Michigan.

DeerMOM is a computer simulation model designed to assess the effects
of management options on population size, sex and age structure of white-tailed
deer. In this model, deer were grouped into three age classes: fawn, yearling,
and adult. Reproductive rates and fetal sex ratios were age-specific, while natural
and harvest mortality rates were both age- and sex-specific. DeerMOM was
parameterized to represent the deer population in the Upper Peninsula (UP) of
Michigan. Effects of winter severity were incorporated into the model. Population
estimates derived from annual pellet group surveys were used to validate the
model. Different management options were evaluated using two criteria: a
quantity goal (number of deer) and a quality goal (percentage of antlered bucks
in the deer population).

Simulation results from DeerMOM indicated that current management
practices (with a high rate of buck harvest) resulted in high deer numbers with a

low percentage of antlered bucks. Under the condition of high buck harvest rate,

90

increasing doe harvest did not achieve both the quantity and the quality goals
simultaneously. Moderate harvest of both sexes would control population growth
and increase the percentage of antlered bucks. The simulations also showed that
winter weather conditions and doe harvest shaped deer population trends but
buck harvest determined the percentage of antlered bucks. Although increasing
doe harvest was an effective method to control deer population, its effect on buck
ratio was limited. On the contrary, lowering buck harvest would increase buck
ratio and would have minimal impact on population size. These findings indicate
that quality deer management objectives could be reached only by lowering buck
harvest rates while simultaneously increasing the doe harvest. The best option
for achieving both the quantity and the quality goals was moderate harvest of
bucks and does without sex bias.

Deer population size and deer spatial distribution across landscapes are
determined not only by harvest, but also by landscape attributes. To evaluate
how deer spatial pattern is associated with landscape attributes, the spatial
structure of landscapes in the UP was quantified. Each of the 15 counties in the
UP was treated as a landscape. Using 1991 Landsat Thematic Mapper imagery,
the landscape was classified into 30-, 21-, 8-, and 4-classes of patches and its
landscape metrics was calculated. Deer harvest, deer-vehicle accidents, and
reported kills were used to measure deer density, and related to landscape
metrics in each landscape.

Deer spatial distribution showed significant positive relationships with

patch density, edge density, and landscape shape index, but had a significant

91

negative relationship with total core area index. There were no significant positive
relationships with patch diversity indices, contagion, or interspersion indices.
These findings suggest that spatial heterogeneity plays an important role in the
spatial distribution of white-tailed deer at landscape scales. Although it is well
known that deer prefer edges at local scales, this study shows that edges are
also important to deer distribution at landscape scales, indicating deer habitat-
selection may be scale-independent. To increase deer population size, one
should consider deer habitat characteristics and aim at increasing patch density
and edge density. On the contrary, to control crop damage and deer-vehicle
accidents, a reverse approach may be used to decrease spatial heterogeneity.

Recognizing that wildlife managers are facing many challenges (e.g.,
more stakeholders involvement) and decisions must be made with conflicting
goals and without complete data, I designed a knowledge-based system for
white-tailed deer management (DeerKBS) to help the user make informed
decisions. This system was an attempt to capture and retain wildlife manager
expertise and knowledge, and to standardize decision-making procedures for
management. A “top down” technique was used to divide the main task of deer
management into 3 major sub-tasks: population status evaluation, habitat
condition evaluation, and social carrying capacity evaluation. Each sub-task was
represented by a decision tree, which related a decision with relevant attributes.
Each sub-task was developed and tested separately.

The development of DeerKBS indicated that structured decision tasks

could help deer managers better organize their knowledge and expertise. The

92

“top-down” technique offers the flexibility to include more information once it is
available. DeerKBS can be an effective communication tool for presenting the
decision-making process to various stakeholders. It can also be used as an
educational tool to help students understand the complexity of management
decision-making process. However, like any other KBSs, the validity of

recommendations is dependent on the decision rules in the knowledge base.

93

APPENDIX A

DeerMOM USER’S GUIDE

This guide is developed using HTML. Included here is the main page that
has links to the detailed information for model background and model

components.

94

DeerMOM User's Guide (Main page)

1. Introduction

2. Model Design
3. Model Structure

4. Quantification of the Model

5. Model Initialization

6. System Reguirements
7. Get Started Quickly

8. More Informatjon about the Model

 

1. Introduction

DeerMOM is a computer simulation model designed to assess the effects
of management options on population size, sex and age structure of
white-tailed deer (Odocoileus virginianus). In this model, I grouped deer
into three age classes: fawn, yearling, and adult. Reproductive rates and
fetal sex ratios were age-specific, while natural and harvest mortality rates
were both age- angLsex-sp_ecific. DeerMOM was parameterized to
represent the deer population in the Upper Peninsula of Michigan. Effects
of winter severity were incorporated into the model. Population estimates
derived from annual pellet group surveys were used to validate the model
in the Upper Peninsula. The model was developed in Stella ® for

Windows.

95

2. Model Design

I followed 3 principles to design the model: simplicity, accuracy, and

management orientation:

Simplicity: l grouped deer into 3 age classes: fawn (<12 months),
yearling (13-24 months), and adult (24 months or older) to simplify the
model structure. I divided mortality into harvest mortalitv and gnu—raj
mortally (e.g., vehicle accidents, starvation, predation). I used annual
mortality instead of seasonal mortality to simplify data collection and
model application.

Accuracy: l simulated the dynamics of males and females separately. l

 

used age- and sex-specific natural mortality and harvest mortality. In
addition, I differentiated their reprodgctive rates and fetal sex ratios by age
gla_s_s_ for female deer.

Management orientation: Management goals could be used as criteria to
evaluate harvest strategies under defined time frames. A user could also
change the model structure by adding or removing some components, or

by modifying parameter values.

3. Model Structure

Model hierarchical structure

Stella is a multi-level, hierarchical environment for constructing models

(High Performance Systems, 1997) which allows DeerMOM to have a 3-

level hierarchical structure. The first level is a user interface, where

novice users can change parameter values by using sliders. Moreover, the

96

user can also igspect simglation results_ using tables and graphs as well as
export them to other programs through dynamic data exchange. In the
second level, a more advanced user can access the model structure
in order to modify the structure and manipulate parameter values. Finally,
an expert user can access the third level to review the mathematical

eguations and to learn the model mechanisms in detail.

Model components

DeerMOM consists of 5 interconnected sectors: Female, Male,
Birth/Death, Harvest, and Population. DeerMOM operates on an annual
cycle that starts on October 1 and ends on September 30 of the following

year

. Female Sector: Females are grouped into 3 age classes: fawn, yearling,
and adult as previously mentioned. For each annual cycle, surviving
newborn females enter the female fawn group, and similarly surviving
fawns recruit to the female yearling group, surviving yearlings to the
female adult group. Adult females that survive the hunting season and
natural mortality remain in the adult age class. Natural mortality for each
age class is density- and harvest-dependent and influenced by winter
severity.

. Male Sector: The male sector had the same model structure as the

female sector, but parameter values were quantified differently.

97

. Birttheath Sector: Changes in the deer population size and structure
depend on the dynamics of both births and deaths. I assumed that
immigration and emigration would not affect age and sex structure of the
population. Thus, any addition to the population comes from the offspring
of female fawns, yearlings, and adults. To estimate newborn females and
males, I used female number in spring, reproductive rate, and sex ratio for
each age class. The spring female number in each age class was
estimated by subtracting harvest and natural mortality from the previous
fall female number in each age class. Deaths were a result of harvest
mortality and natural mortality. Mortality rates were age- and sex-specific.

. Harvest Sector: I included all harvest-related information in this sector.
Specifically the sector contained age and sex composition of harvest, total
harvest, percentage of antlered bucks and antlerless deer in the harvest.

. Population Sector: I included Mon initialization, management
ggjs, and all data related to age and sex distribution of deer in this sector.
The population was initialized by gize, sex ratio, and_percentgge of deer
numtfir in each age class. This sector also extended the female and male
sectors by deriving population statistics of interest.

4. Quantification of the Model

The most important parameters for DeerMOM included age-specific
reproductive rates, fetal sex ratios, neonatal mortalitv, age- and sex-

specific natural mortality, and harvest mortality.

98

. Reproductive rates: defined as average fetuses per female. To see how I
quantified the reproductive rates in the UP, please follow this link.

. Fetal sex ratio: defined as the percentage of males in each age class.
Fetal sex ratio was the percentage of newborn males in offspring. To see
Verme's (1983) data, follow this link.

. Mortality: Causes of mortality were classified as harvest mortality and
natural mortalitv. To see how I quantified the mortality rates in the UP,
follow this link.

. Neonatal mortality: Winter weather has a significant impact on fetal
development during late gestation and thus influences natal survival
(Verme, 1977). Because of this impact, I partitioned neonatal mortality into
2 parts: base neonatal mortality and neonatal mortality due to winter
severity. Base neonatal mortality was the mortality when winter weather
was very mild (low WSI). To see how I quantified the neonatal mortality
rates in the UP, follow this link.

5. Model Initialization

The model should be initialized by population size and age- and sex-
distribution. To see how I initialized the population in the UP, follow this

link.

6. System Requirements

Operating system: Microsoft Windows® 95/98

Stella ® Research Version 5 for Microsoft Windows®

99

7. Get Started Quickly

To open Stella: Click Start- Programs-- Stella5 (or Stella 5.11)-- Stella.

The program should be in Model mode.

To open file: From the menu, click File- Close Model. Then click File--
Open, choose the Drive, Folder, and then the File name. Wait for loading

the file DeerMOM.stm.

To change parameter values: Click Cont_rol Panel button, then click on
Initialization or Mortality button to change parameter values for initial

p_qpuation and harvest .

To run the model: From the menu, click Bpn— Time Specs to change
your time setting. Make sure that time step is set at one year. Then click
Run— Run.

To check the results: Click "Simulation Results" button to see the model

simglation resulfi in figures, tables, or numerical displays.

8. More lnforrnation about the Model

100

APPENDIX B

DEERKBS INTERFACE

The following figures are the screenshots of DeerKBS.

101

 

V Der-r population sum in the lull

? DeerKBS: a knowledge-based system for white-tailed deer management

Jialong Xie, December 1999

This sxstem captures die-er managers‘lemertise on 7
iwhitetailed deer management. It is designed to help
users to make more informed decisions based on the
evaluations of deer population status. habitat
lconditions, and socioeconomic factors.

‘WH _, i WW Data: The fall deer herd in the
FYI: "Wildlife managers assess deer Upper PenmSUIa 0f M'Ch'gan:

populations and public attitudes to 897'000 in 1995' 6241000 in
establish deer herd objectives at a 1996' 474000 'n 1997' The fa”

local level. Annual recommendations deer herd in the entire State i5 13
are made to have more, less, or the million in 1997-

same number of deer within each

DMU" . Ed Langenau (DNR Big Please input fall population size
Game Specialist) for area of interest.

----------- _L__... I

Figure B1: User inputs deer population size in an area of interest.

 

 

It}: Area m

 

 

I Please input the area of interest in sq. miles I Data: Area in the upper Peninsuja 6f

 

 

16446 i OK l Back] How] Michigan is16,446 sq. miles.

       

 

 

 

Figure B2: User inputs the area of interest in square miles.

102

it Deer habitat quantity

Habitat quantity and quality determine I
how many deer ran be supported in a I
defined area. Deer habitat quantity J
determines the deer population size in a I
defined area. i

l

l

 

 

[Please choose your evaluation about the“
ideer habitat quantity:
1

Habitat quantity is abundant
Habitat quantity is average
Habitat quantity is lack

Back] HW’ ll

 

 

 

Figure 83: User chooses the evaluation result of deer habitat quantity from a list
of values.

Zi- Deel habitat quality in

 
  
  
  
  
 
 
 
 
  
   
 

Usually habitat quality
determines deer quality. Good

' deer habitat provides condition 3
for healthy deer population.

figfiifiafi‘éfifiéfi

. :about deer habitat quality:

’ Habitat quality is high
Habitat quality is medium
Habitat quality is poor

Back I

1 .
n . 5 1 .

 

 

 

Figure B4: User chooses the evaluation result of habitat quality from a list of
values.

103

’: Hunlelx' expectation ul [leer nllllllllfl m the [mill

 
    
 
   
     
 
 
 
 
  
 
  

Deer hunters almost always i
expect to have more deer in
the field. But in recent years,l
the deer hunters expect more 5
bucks. (

l
l

 

ilj'lease choose your evaluation
labout hunters' expectation:

‘ More

Same

 

 

Figure BS: User chooses the evaluation result of deer hunters’ expectation from a
list of values.

Vi -
3.1. Farmers expectation of deer number In the Imhl

‘ FYI: Farmers usually complain ma
there are too many deer in the field iPlease choose 3'01}! evaluation about
‘ because deer damage their crops. ‘ farmer" “Puma“:
In Michigan, about 2000

 

 

. . More
landowners have serious economic . S
losses on their farms and orchards ame
due to deer croo damage. Fewer

 

Figure 86: User chooses the evaluation result of farmers’ expectation from a list
of values.

104

Vl’ublllil‘» cxriirilulmn (ll ili-ur nurnbrrr In ”It lll‘ltl

 

Some people expect to have more
deer in the field for sighting and
photograph. Others expert to have
less deer to avoid possible deer
vehicle accidents.

 

 

ease choose your evaluation about
lic's expectation:

More
Same
Fewer

     

FYI: The deer-vehicle
accidents in Michigan:
47832 in 1993, 56578 in .
1994, and 62493 in 1995

 

Figure B7: User chooses the evaluation result of other general public’s
expectation from a list of values.

7“" Buck Percentage in the population

 

Buck ratio is one of the most important
indices to measure the health of the The buck ratio is around 18% in 1997
deer population. Quality deer
management (QDM) requires that the
deer population maintain a balanced sex _ _
ratio and age distribution. The ‘ Base “‘13“? Percentage of bud“ m
management goal for QDM is that 35% all Populauom

of the fall population should be antlered
bucks.

 

 

   

 

Figure BS: User inputs the percentage of antlered bucks in the fall population.

105

3_ :1 x
DeerKBS RECOMMEND THnT:

Harvest Hore antlerless but FeuerBucks.

This recoaaendation is based on evaluations of deer populaiton
‘status, deer habitat,and socioeconoaic considerations.

 

 

 

 

l
hum-“A. M'-

"‘?2 __JBncr ._l.£i_M_H—l°w __l‘ffff.w__l°331mi.lip _lswe

m. , r. . is .w..-i-....._~..-.w.— n” " ""

 

 

Figure 89: DeerKBS makes harvest recommendations.

 

Given that PopulationSize has been processed
and area has been processed
and PopDensity >= 26.6
9nd PopDensity >= 35.6
The outcome of PopulationDensitg is High
Given that PopulationDensitg is High
Given that HabitatQuantity is Average
and HabitatQuality is Hediua
The outcome of Habitat is Moderate
and Habitat is Moderate
Given that HunterExpected is More
and FaraerExpected is Power
and PublicExpected is Sale or More
The outcoee of CulturalCarryingCapacitg is More
and CulturalCarryingCapacitg is More
The outcoae of HarvestRecoeaendation is Sale
Given that BuckPercent < 20
The outcome of BuckRatio is Lou
and BuckRatio is average or Lou
The outcoee of HarvestSaee is Harvest More antlerless but Feuer Bucks

 

 

Figure B10: DeerKBS provides explanations by tracking the decision process.

106

APPENDIX C

DECISION RULES IN DeerKBS

Population Density:

If population density < 20.00
Then Population Density is Low

If population density >= 20.00
And population density < 35.00
Then population density is Good

If population density >= 35.00
Then population density is High

Buck Ratio:

If buck ratio < 20.0
Then buck ratio is Low

If buck ratio >= 20.0
And buck ratio < 35.0
Then buck ratio is Average

If buck ratio >= 35.0
Then buck ratio is High

Habitat:

If habitat quantity is Abundant
And habitat quality is High or Medium
Then habitat is Excellent

If habitat quantity is Abundant
And habitat quality is Poor
Then habitat is Moderate

If habitat quantity is Average

And habitat quality is High
Then habitat is Excellent

107

If habitat quantity is Average
And habitat quality is Medium
Then habitat is Moderate

If habitat quantity is Average
And habitat quality is Poor
Then habitat is Poor

If habitat quantity is Lack
And habitat quality is High
Then habitat is Moderate

If habitat quantity is Lack
And habitat quality is Medium or Poor
Then habitat is Poor

Cultural Carrying Capacity:

If hunters’ expectation is More
And farmers’ expectation is More or Same
Then cultural carrying capacity is More

If hunters’ expectation is More

And farmers’ expectation is Fewer

And other interest groups’ expectation is More or Same
Then cultural carrying capacity is More

If hunters’ expectation is More

And farmers’ expectation is Fewer

And other interest groups’ expectation is Fewer
Then cultural carrying capacity is Same

If hunters’ expectation is Same
And other interest groups’ expectation is More
Then cultural carrying capacity is More

If hunters’ expectation is Same

And other interest groups’ expectation is Same
And farmers’ expectation is More

Then cultural carrying capacity is More

If hunters’ expectation is Same

And other interest groups’ expectation is Same
And farmers’ expectation is Same

Then cultural carrying capacity is Same

108

lf hunters’ expectation is Same

And other interest groups’ expectation is Same
And farmers’ expectation is Fewer

Then cultural carrying capacity is Fewer

If hunters’ expectation is Same
And other interest groups’ expectation is Fewer
Then cultural carrying capacity is Fewer

If hunters’ expectation is Fewer

And farmers’ expectation is More

And other interest groups’ expectation is More
Then cultural carrying capacity is Same

If hunters’ expectation is Fewer

And farmers’ expectation is More

And other interest groups’ expectation is Same or Fewer
Then cultural carrying capacity is Fewer

If hunters’ expectation is Fewer
And farmers’ expectation is Same or Fewer
Then cultural carrying capacity is Fewer

Weather Conditions:

If WSI < 80.0
Then Weather is Favorable

lf WSI >= 80.0
And WSI < 120.0
Then Weather is Average

If WSI >= 120.0
Then Weather is Harsh

Preliminary Harvest Recommendations:
If Population density is High
And habitat is Excellent
And cultural carrying capacity is More or Same
Then Preliminary harvest recommendation is Same

If Population density is High
And habitat is Excellent

109

And cultural carrying capacity is Fewer
Then Preliminary harvest recommendation is More

If Population density is High

And habitat is Moderate

And cultural carrying capacity is More

Then Preliminary harvest recommendation is Same

If Population density is High

And habitat is Moderate

And cultural carrying capacity is Same or Fewer
Then Preliminary harvest recommendation is More

If Population density is High
And habitat is Poor
Then Preliminary harvest recommendation is More

If Population density is Good

And habitat is Excellent

And cultural carrying capacity is More

And Weather is Favorable

Then Preliminary harvest recommendation is Same

If Population density is Good

And habitat is Excellent

And cultural carrying capacity is More

And Weather is Average or Harsh

Then Preliminary harvest recommendation is Fewer

If Population density is Good

And habitat is Excellent

And cultural carrying capacity is Same

And Weather is Favorable

Then Preliminary harvest recommendation is More

If Population density is Good

And habitat is Excellent

And cultural carrying capacity is Same

And Weather is Average

Then Preliminary harvest recommendation is Same

If Population density is Good

And habitat is Excellent

And cultural carrying capacity is Same

And Weather is Harsh

Then Preliminary harvest recommendation is Fewer

110

If Population density is Good

And habitat is Excellent

And cultural carrying capacity is Fewer

And Weather is Favorable or Average

Then Preliminary harvest recommendation is More

If Population density is Good

And habitat is Excellent

And cultural carrying capacity is Fewer

And Weather is Harsh

Then Preliminary harvest recommendation is Same

If Population density is Good

And habitat is Moderate

And cultural carrying capacity is More or Same or Fewer
And Weather is Favorable or Average

Then Preliminary harvest recommendation is More

If Population density is Good

And habitat is Moderate

And cultural carrying capacity is More or Same or Fewer
And Weather is Harsh

Then Preliminary harvest recommendation is Same

If Population density is Good

And habitat is Poor

And cultural carrying capacity is More or Fewer
And Weather is Favorable or Average

Then Preliminary harvest recommendation is More

If Population density is Good

And habitat is Poor

And cultural carrying capacity is More or Fewer
And Weather is Harsh

Then Preliminary harvest recommendation is Same

If Population density is Good

And habitat is Poor

And cultural carrying capacity is Same

And Weather is Favorable

Then Preliminary harvest recommendation is More

If Population density is Good

And habitat is Poor
And cultural carrying capacity is Same

111

And Weather is Average or Harsh
Then Preliminary harvest recommendation is Same

If Population density is Low
And habitat is Excellent or Moderate
Then Preliminary harvest recommendation is Fewer

If Population density is Low
And habitat is Poor
Then Preliminary harvest recommendation is Same

Further Harvest Recommendations:

If buck ratio is High
And harvest recommendation is More
Then further recommendation is Harvest More Antlerless and More Bucks.

If buck ratio is Average

And harvest recommendation is More

Then further recommendation is Harvest More Antlerless and Same
Bucks.

If buck ratio is Low

And harvest recommendation is More

Then further recommendation is Harvest More Antlerless but Fewer
Bucks.

If buck ratio is High
And harvest recommendation is Same
Then further recommendation is Harvest Same Antlerless and Bucks.

If buck ratio is Average or Low

And harvest recommendation is Same

Then further recommendation is Harvest More Antlerless but Fewer
Bucks.

If buck ratio is High or Average

And harvest recommendation is Less

Then further recommendation is Harvest Fewer Antlerless but Same
Bucks

If buck ratio is Low

And harvest recommendation is Less
Then further recommendation is Harvest Fewer Antlerless and Bucks.

112

REFERENCES

113

REFERENCES

Albert, D. A., S. R. Benton, and B. V. Barnes.1986. Regional landscape
ecosystem of Michigan. School of Natural Resources, University of
Michigan. Ann Arbor, MI. 32pp.

Allen, R. E. and D. R. McCullough. 1976. Deer-car accidents in southern
Michigan. Journal of Wildlife Management 40:317-325.

Alverson, W. S., D. M. Waller, and S. L. Solheim. 1988. Forests too deerzedge
effects in northern Wisconsin. Conservation Biology 2:248—258.

Andrén, H. 1994. Effects of habitat fragmentation on birds and mammals in
landscapes with different proportions of suitable habitatza review. Oikos
71 :355-366.

Arthur, L.M. and W. R. VWlson. 1979. Assessing the demand for wildlife
resourcesza first step. Wildlife Society Bulletin 7(1):30—34.

Attar Software Limited. 1996. XpertRule User Guide. Lancashire, United
Kingdom.

Awad, E. M. 1996. Building expert systems. West Publishing Co.

Banker, M. E. 1994. Modeling white-tailed deer habitat quality and vegetation
response to succession and management. MS. Thesis. Virginia Tech
University. Blacksburg, Virginia. 147pp.

Bartholow, J. 1986. POP-ll system documentation. Fossil Creek Software, Ft.
Collins, CO.

Bashore, T. L., W. M. Tzilkowski, and E. D. Bellis.1986. Analysis of deer-vehicle

collision sites in Pennsylvania. Journal of Wildlife Management 49:769-
774.

Bayne, E. M., and K. A. Hobson. 1998. The effects of habitat fragmentation by
forestry and agriculture on the abundance of small mammals in the
southern boreal mixedwood forest. Canadian Journal of Zoology 76:62-69.

Bender L. C. and J. B. Haufler.1990. A white-tailed deer HSI for the upper Great

Lakes Region. Unpublished report. Department of Fisheries and Wildlife,
Michigan State University. East Lansing, MI.

114

Bender, L. C. and G. J. Roloff, 1996. Modeling deer harvest alternatives in
Indiana. Transactions of the Illinois State Academy of Science 89:63-71.

Boehm, B. W. 1976. Software engineering. IEEE Transactions on Computers C-
25 (12):1226-1241.

. 1988. A spiral model of software development and enhancement.
Computer 21 (5):61-72.

Braun, K. F.1996. Ecological factors influencing white-tailed deer damage to
agricultural crops in northern Lower Michigan. M.S.Thesis. Michigan State
University. East Lansing, MI.

Bremdal, B. A. 1997. Expert system for management of natural resources. Pages
30:1-47. in J. Liebowitz, ed. The Handbook of Applied Expert Systems.
CRC Press LLC.

Breuker, J. and B. Wielinga. 1989. Models of expertise in knowledge acquisition.
Pages 265-295 in G. Guida and C. Tasso, eds. Topics in expert systems
design: methodologies and tools. North Holland Publishing Company,
Amsterdam.

Brown, 8., H. Schreier, W. A. Thompson, and |. Vertinksy. 1994. Linking multiple
accounts with GIS as decision support system to resolve forestry/wildlife
conflicts. Journal of Environmental Management 42:349-364.

Buchnanm, B. G., D. Barstow, R. Bechtal, J. Bennett, W. Clancy, C. Kulikowski,
T. Mitchell, and D. A. Waterman. 1983. Constructing an expert system.
Pages 127-167. in F. Hayses -Roth, D. A. Waterman, and D. B. Lenat,
eds. Building expert system. Addison-Wesley Publishers. Reading, MA.

Burgoyne, G. Jr, 1981. Observations on a heavily exploited deer population.
Pages 403-413 in C. W. Fowler. ed.. Dynamics of Large Mammal
Populations. John Wiley & Sons. Inc.

Burke, R. L. and J. M. Birch, 1995. White-tailed deer vary offspring sex-ratio
according to maternal condition and age. Ecological Research 10:351-
357.

Bylander, T. and S. Mittal. 1985. CSRL:a language for classificatory problem
solving and uncertainty handling. AI Magazine 7(3):66-77.

Bylander, T., T. R. Johnson, and A. Goel. 1991. Structured matchingL a task-
specific technique for making decisions. Knowledge Acquisition 3:1-20

115

Campa, H. Ill. 1989. Effects of deer and elk browsing on aspen regeneration and
nutritional qualities in Michigan. Ph.D. Dissertation. Michigan State
University. East Lansing, MI.

Caughley G .1981. Overpopulation. Pages 7-20 in P. A. Jewell, and S. Holt. eds.
Problems in management of locally abundant wild mammals. Academic
Press, NY.

Chandrasekaran, B. and T. R. Johnson. 1993. Generic tasks and task
structure:history, critique, and new directions. Pages 232-272 in J. M.
David, J. P. Krivien, and R. Simmons, eds. Second generation expert
system. Springer-Verag. Berlin.

Chang, K., D. L. Verbyla, and J. J. Yeo. 1995. Spatial analysis of habitat
selection by Sitka black-tailed deer in southeast Alaska, USA.
Environmental Management 19(4):579-589.

Christensen N. L., A. M. Bartuska, J. H. Brown, S. Carpenter, C. DAntonio, R.
Francis, J. F. Franklin, J. A. MacMahon, R. F. Noss, D. J. Parsons, C. H.
Peterson, and M. G. Turner. 1996. The Report of the Ecological Society of
America Committee on the Scientific Basis for Ecosystem Management.
Ecological Applications 6(3):665-691.

Clancy, W. J. 1983. The epistemology of a rule-based expert system - a
framework for explanation. Artificial Intelligence 20:215-251.

Cody R. P. and J. K. Smith.1997. Applied statistics and the SAS programming
language. 4th edition. Prentice-Hall, Inc. Upper Saddle River, NJ.

Compton B. B., R. J. Mackie, and G. L. Dusek. 1988. Factors influencing
distribution of white-tailed deer in riparian habitats. Journal of Wildlife
Management 52:544-548.

Conover, M. R. and D. J. Decker. 1991. Wildlife damage to cropszperceptions of
agricultural and wildlife professionals in 1957 and 1987. Wildlife Society
Bulletin 19:46-52.

Conover M. R., W. C. Pitt, K. K. Kessler, T. J. DuBow, and W. A. Sanbom .1995.
Review of human injuries, illness, and economic loses caused by wildlife
in the United States. Wildlife Society Bulletin 23:407-414.

Cooperrider, A. Y. 1974. Computer simulation of the interaction of a deer
population with northern forest vegetation. Ph. D. Dissertation. 229p. State
University of New York.

Coulson, R. N., L. J. Folse, and D. K. Loh. 1987. Artificial intelligence and natural

116

resource management. Science 237:262-267.

Coulson, R. N., C. N. Lovelady, R. O. Flamm, S. L. Spradling, and M. C.
Saunders. 1991. Intelligent geographic information systems for natural
resource management. Pages 153-172 in Ecological Studies
82:Mathematical methods in landscape ecology.

Covington, W. W., D. B. Wood, D. L. Young, D. P. Dykstra, and L. D. Garrett.
1988. TEAMS:A decision support system for multiresource management.
Journal of Forestry 86(8):25-33.

Crawford, H. S. 1984. Habitat management. Pages 629-646 in L. K. Halls. ed.
White-tailed deerzecology and management. Stackpole Books, Harrisburg,
PA

Crawford, H. S., and R. L. Marchinton. 1989. A habitat suitability index for white-
tailed deer in the Piedmont. Southern Journal of Applied Forestry
13(1):12-16.

Creed, W. A., F. Haberland, B. E. Kohn, and K. R. McCaffery. 1984. Harvest
managementzThe Wisconsin Experience. Pages 243-260 in L. K. Halls.

ed. White-tailed deerzecology and management. Stackpole Books,
Harrisburg, PA

Crow, T. R. and E. J. Gustafson. 1997. Ecosystem managementzmanaging
natural resources in time and space. Pages 215-228 in K. A. Kohm, J. F.
Franklin , eds. Creating a Forestry for the 21St Century, Island Press.

Davis, R., B. G. Buchanan, and E. H. Shortliffe. 1977. Production rules as a
representation for a knowledge-based consultation system. Artificial
Intelligence 8:15-45.

Davis, J. R., P. M. Nanninga, J. R. L. Hoare, and A. J. Press. 1989. Transferring
scientific knowledge to natural resource managers using artificial
intelligence concepts. Ecological Modelling 46:73-89.

DeCalesta, D. 8.1994. Effect of white-tailed deer on songbirds within managed
forests in Pennsylvania. Journal of Wildlife Management 58:71 1-718.

Decker, D. J. and K. G. Purdy. 1988. Toward a concept of wildlife acceptance
capacity in wildlife management. Wildlife Society Bulletin 16:53-57.

Decker, D. J., T. L. Brown, N. A. Connelly, J. W. Enck, G. A. Pomerantz, K. G.
Purdy. and W. F. Sieer. 1992. Toward a comprehensive paradigm of
wildlife managementzintegrating the human and biological dimensions.

117

 

Pages 33-54 in W. R. Mangun, ed. American fish and wildlife policyzthe
human dimension. Southern Illinois University Press, Carbondale, ll.

Duda, R. 0., P. E. Hart, K. Konolige, and R. Reboh. 1979. A computer-based
consultant for mineral exploration (Technical report). SRI lntemational.

Duda, R. O. and J. G. Gaschnig. 1981. Knowlege-based expert systems come of
age. Byte, March, 238-249.

Dunning, J. B., B. J. Danielson, and H. R. Pulliam 1992. Ecological processes
that affect populations in complex landscapes. Oikos 65:169-175.

Dusek, G. L., R. J. Mackie, J. D. Herriges Jr., and B. B. Compton. 1989.
Population ecology of white-tailed deer along the lower Yellowstone River.
Wildlife Monograph 104:1-68.

Drolet, C. A. 1976. Distribution and movement of white-tailed deer in southern
New Brunswick in relation to environmental factors. Canadian Field-
Naturalist 90:123-136.

Eberhardt, L. L.1960. Estimation of vital characteristics of Michigan deer herds.
Ph.D. Dissertation. Michigan State University, East Lansing, MI. 192pp.

Edwards, J. S. 1991. Building knowledge-based systemsztowards a
methodology. Halsted Press, New York.

Ehrlich, P. R. and G. C. Daily. 1993. Science and the management of natural
resources. Ecological Applications. 3(4):558-560.

Euler, D. and M. Morris, 1984. Simulated population dynamics of white-tailed
deer in an any-deer hunting system. Ecological Modeling 24:281-292.

Flather, C. H. and J. R. Sauer. 1996. Using landscape ecology to test
hypotheses about large-scale abundance patterns in migratory birds.
Ecology 77:28-33.

Folse, L. J., J. M. Packard, and W. W. Grant, 1989. AI modeling of animal
movement in a heterogeneous habitat. Ecological Modelling 46:57-72.

Franklin, J. F. 1994. Developing information essential to policy, planning, and
management decision-makingzthe promise of GIS. Pages in V. A. Sample
Edited Remote sensing and GIS in ecosystem management. Island
Press. Washington, DC

118

Friedrich, P. D and S. M. Schmitt, 1988. Doe productivity and physical
conditionz1988 Spring survey results. Michigan Department of Natural
Resources. Wildlife Division Report No. 3083. Lansing, Michigan.

Gavin, T. A., L. H. Suring, P. A. Vohs Jr., and E. C. Meslow, 1984. Population
characteristics, spatial organization, and natural mortality in the Colombian
white-tailed deer. Wildlife Monograph 91 :1-41.

German, C. D. 1992. Individual based modeling of animal populations using
object oriented simulation techniques: Investigating the effects of bonding,
predation, and birth cover on white-tailed deer. MS. Thesis. Texas A&M
University. 149pp.

Giles, R. H. Jr. 1978. Wildlife Management. W. H. Freeman, San Francisco, CA.

. 1998. Natural resource management tomorrow: four currents. Wildlife
Society Bulletin 26(1):51-55.

Ginsberg, J. R. and E. J. Milner—Gulland. 1994. Sex-biased harvesting and
population dynamics in ungulates: implications for conservation and
sustainable user. Conservation Biology 8:157-166.

Goldberg, A. and D. Robson. 1983. Smalltalk-80:the language and its
implementation. Addison-Wesley. Reading, MA.

Guida, G. and C. Tasso. 1994. Development of the prototype. Chapter 62157-236
in Design and development of knowledge-based systems: from life cycle
to methodology. John Wiley & Sons Ltd. West Sussex, England.

Gustafson, E .1998. Quantifying landscape spatial pattern: what is the state of
the art? Ecosystems 11143-156.

Halls, L. K. 1984. Research problems and needs. Pages 783-790 in L. K. Halls,
ed. White-tailed deer: ecology and management. Wildlife Management
Institute. Stackpole Books. Harrisburg, PA.

Hanley, T. A. 1996. Potential role of deer (cervidae) as ecological indicators of
forest management. Forest Ecology And Management 882(1-2) 199-204.

Harder, J. D, 1980. Reproduction of white-tailed deer in the north central United
States. Pages 23-25 in R. L. Hine and S. Nehl. eds., White-tailed deer
population management in the north central states. Proceeding of a
symposium held at the 41st Midwest Fish and Wildlife Conference,
Urbana, Illinois, December 10, 1979.

119

Hayne, D. W. 1984. Population dynamics and analysis. Page 203-210 in L. K.
Halls ed. White-tailed DeerzEcology and Management. Stackpole Books.
Harrisburg, PA.

Hayes-Roth, F., D. A. Waterman, and D. B. Lenat. 1983. Building expert system.
Addison-Wesley Publishers. Reading, MA.

High Performance Systems, Inc, 1997. STELLA. II Technical Documentation.
Hanover, N H.

Hill, H. R. and D. L. Rabe, 1989. Deer checking station data -1989. Michigan
Department of Natural Resources. Wildlife Division Report No. 3126.
Lansing, MI.

Hill, H. R. 1996. The 1996 Deer pellet group surveys. Michigan Department of
Natural Resources, Wildlife Division Report. Lansing, MI.

Jahn L.R.1959. Highway mortality as an index of deer population change.
Journal of Wildlife Management 23: 1 87-1 97.

Jensen, A. L, 1996. Density-dependent matrix yield equation for optimal harvest
of age-structured wildlife populations. Ecological Modeling 88:125-132.

Kellert S. R. 1994. Managing for biological and sociological diversity, or 'deja vu,
all over again'. Wildlife Society Bulletin, 23.2.:274-278.

Krefting L. W. and R. S. Phillips.1970. Improving deer habitat in upper Michigan
by cutting mixed-conifer swamps. Journal of Forestry 68:701-704.

Langenau, E. E. Jr. 1991. Few deer, more bucks. Michigan Natural Resources,
November/December, 1991, 38-39.

.1994. 100 years of deer management in Michigan. Michigan Department
of Natural Resources. Wildlife Division Report. No. 3213. Michigan
Department of Natural Resources. Lansing, MI.

Leopold A .1933. Game management. Charles Scribner's Sons, New York.

Levins, R. 1966. The strategy of model building in population biology. American
Scientists 54:421-431.

Larson, T. J., O. J. Rongstad, and F. W. Terbilcox. 1978. Movement and habitat

use of white-tailed deer in southern central Wisconsin. Journal of Wildlife
Management 42:113-117.

120

Li, H. B. and J. F. Reynolds. 1994. A simulation experiment to quantify spatial
heterogeneity in categorical maps. Ecology 75:2446-2455

Lillesand, T. M. and R. W. Kiefer. 1994. Remote sensing and image
interpretation. 3rd ed. Wiley & Sons, New York.

Ludwig, D., R. Hilborn, and C. Walters. 1993. Uncertainty, resource exploitation,
and conservation: lessons from history. Science 260:(5104) 17.

Maclean, GA .1994. Inventory of deer habitat in Michigan's upper peninsula
utilizing Landsat Thematic Mapper data. A final report to the Michigan
DNR, Wildlife Division, Michigan Department of Natural Resources.
Lansing, MI.

Mallawaarachchi, T., P. A. Walker, M. D. Young, R. E. Smyth, and H. S. Lynch.
1996. GIS-based integrated modeling systems for natural resource
management. Agricultural Systems 50:169-189.

Matschke G., D. S. deCalesta, and J. D. Harder.1984. Crop damage and control.
Pages 647-654 in L. K. Halls. ed. White-tailed deer: ecology and
management. Stackpole Books, Harrisburgh, PA,

McCabe, R. E. and T. R. McCabe.1984. Of slings and arrows: An historical
retrospection. Pages 19-72 in L. K. Halls, ed. White-tailed deer: ecology
and management. Stackpole Books, Harrisburg, PA.

McCaffery, K. R. 1989. Deer population dynamics and management in
Wisconsin. Proceeding of Eastern Wildlife Damage Control Conference
4: 1 55-161 .

McCullough, D. R. 1979. The George Reserve deer herd: population ecology of a
K-selected species. University of Michigan Press, Ann Arbor, MI.

McDermott, J. 1982. R1 :A rule-based configurer of computer systems. Artificial
Intelligence 19(1):39-88.

. 1988. Preliminary steps towards a taxonomy of problem-solving
methods. Pages 225-256 in S. Marcus, ed. Automatic knowledge
acquisition for expert systems. Kluver Academic Publishers, Boston.

McGarigal K. and B. J. Marks.1994. FRAGSTATS 2.0:Spatial pattern analysis

program for quantitative landscape structure. Oregon State University.
Corvallis, OR.

121

 

McShea, W. J, H. B.Underwood, J. H. Rappole, eds. 1997. The science of
overabundance: deer ecology and population management. Smithsonian
Institution Press, Washington and London.

Mech, L. D., R. E. McRoberts, R. 0. Peterson, and R. E. Page. 1987.
Relationship of deer and moose populations to previous winters’ snow.
Journal of Animal Ecology. 56:615-627.

Meddin, D. E. and A. E. Anderson. 1979. Modeling the dynamics of a Colorado
mule deer population. Wildlife Monograph 68, 1-77.

Mello, P. J.1983. A computer simulation model for the study of deer-aspen forest
interactions. M. S. Thesis. Michigan State University. East Lansing,
Michigan.

Michigan Department of Natural Resources. 1993. Statewide DRIP goal.
Michigan Department of Natural Resources, Wildlife Diversion. Lansing,
MI.

. 1996. Michigan deer population figures (in thousands). Michigan
Department of Natural Resources, Wildlife Diversion. Lansing, MI.

Miller, S. G., S. P. Brattton, and J. Hadidian. 1992. Influence of white-tailed deer
on endangered and threatened vascular plants. Natural Areas Journal
12:67-74.

Milne, B. T., K. M. Johnston, and R. T. T. Forman .1989. Scale-dependent
proximity of wildlife habitat in a spatially-neutral Baysian model.
Landscape Ecology 22101-110.

Minnis, D. L. 1996. Cultural carrying capacity and stakeholders’ attitudes
associated with the deer crop damage issue in Michigan. Ph.D.
Dissertation. Michigan State University, East Lansing, MI.

Minsky, M. L. 1975. A framework for representing knowledge. Pages 211-277 in
P. H. Winston, ed. The psychology of computer vision. McGraw-Hill, NY.

Mladenoff D. J. and F. Steams. 1993. Eastern hemlock regeneration and deer
browsing in the northern Great Lake Region: A re-examination, and model
simulation. Conservation Biology 7:889-900

Mladenoff, D. J. and G. E. Host. 1994. Ecological perspective: current and
potential applications of remote sensing and GIS to ecosystem analysis.
Pages 218-242 in V. A. Sample ed. Remote sensing and GIS in
ecosystem management. Washington D. 0., Island Press.

122

 

Morrison, M. L., B. C. Marcot, and R. W. Mannan. 1992. Wildlife-habitat
relationships: concepts and applications. The University of Wisconsin
Press. Madison. 343pp.

National Research Council. 1995. Science and Endangered Species Act.
Washington, DC.

Nelson M. E. and L. D. Mech.1984. Home-range formation and dispersal of deer
in northern Minnesota. Journal of Mammalogy 65:567-575.

Nelson, M. E. 1994. Winter range arrival and departure of white-tailed deer in
northern Minnesota. Canadian Journal of Zoology 73:1069-1076.

Nixon, C. M., L. P. Hansen, and P. A. Brewer. 1988. Characteristics of winter
habitats used by white-tailed deer in Illinois. Journal of Wildlife
Management 52(3):552-555.

Nixon, C. M., L. P. Hansen, P. A. Brewer, and J. E. Chelsvig. 1991. Ecology of
white-tailed deer in an intensively farmed region of Illinois. VVIldlife
Monograph 118:1-77.

Newell, A. and Simon, H. 1973. Human Problem Solving. Englewood Cliffs,
NJzPrentice-Hall.

Newell, A. 1982. The knowledge level. Artificial Intelligence 18:87-127.

Nixon, C. M., L. P. Hansen, P. A. Brewer, and J. E. Chelsvig. 1991. Ecology of
white—tailed deer in an intensively farmed region of Illinois. VVIldlife
Monograph 11821-77.

Ormsby, J. P. and R. S. Lunetta. 1987. White-tailed deer food availability maps
from thematic mapper data. Photogrammetric Engineering and Remote
Sensing 53(8):1081-1085.

Ozaga, J. J. 1968. Variations in microclimate in a conifer swamp deeryard in
northern Michigan. Journal of Wildlife Management 32:574-585.

, R. V. Doepker, and M. S. Sargent. 1994. Ecology and management of
white-tailed deer in Michigan. Michigan Department of Natural Resources,
Wildlife Division Report 3209. 73pp.

Peyton, R. B. 1987. Mechanisms affecting public acceptance of resources

management policies and strategies. Canadian Journal of Fish and
Aquatic Science 44:306-312.

123

Potvin, F. and J. Huot. 1983. Estimating the carrying capacity of a white-tailed
deer wintering area in Quebec. Journal of Wildlife Management 47:463-
475.

Prerau, D., M. Adler, D. K. Pathak, and A. Gunderson. 1997. Intelligent agent
technology. Pages 1621-13 in J. Liebowitz, ed. The Handbook of Applied
Expert Systems. CRC Press LLC.

Quinlan, J. R. 1986. Induction of decision trees. Machine Learning.1:81-106.

Reynolds, K. M. 1999. EMDS users guide (version 2.0):knowledge-based
decision support for ecological assessment. Gen. Tech. Rep. PNW-GTR-
470. Portland, OR: US. Department of Agriculture, Forest Service, Pacific
Northwest Research Station. 63 pp.

Ritchie, J. R. 1989. An expert system for a rangeland simulated model.
Ecological Modeling 46:91 -1 05.

Ross, B. A., J. R. Bray, and W. H. Marshall .1970. Effects on long-term deer
exclusion on a pinus resinosa forest in North-Central Minnesota. Ecology
51 :1089-1093

Rue, L. L. Ill. 1989. The deer of North America, Grolier Books Club Inc.,
Danbury, Connecticut.

Saarenmaa, H., Stone, N. D., Folse, L. J., Packard. J. M., Grant, W. E., Makela,
M. E. and Coulson, R. N. 1988. An artificial intelligence modeling
approach to simulating animal/habitat interactions. Ecological Modeling.
44:125-141.

SAS Institute Inc. 1998. SAS Language Reference. Version 7. First Edition. Cary,
NC.

Schank, R. C. and R. P. Abelson. 1977. Script, plans, goals, and understanding.
Lawrence Erlbaum Associates Inc. Hillsdale, NY.

Schank, R. C. 1982. Dynamic memoryza theory of learning in computers and
people. Cambridge University Press.

. 1987. What is Al, anyway? AI Magazine 8(4):57-66.
Schmitt, S. M., S. D. Fitzgerald, T. M. Cooley, C. S. Bruning-Fann, L. Sullivan, D.
Berry, T. Carlson, R. B. Minnis, J. B. Payeur, and J. Sikarskie .1997.

Bovine tuberculosis in free-ranging white-tailed deer from Michigan.
Journal of Wildlife Disease 33:749-758.

124

Schmoldt, D. and H. M. Rauscher. 1996. Building knowledge-based systems for
natural resources management. Chapman & Hall. New York, NY.

Short, H. L.1986. Habitat suitability index modelszwhite-tailed deer in the Gulf of
Mexico and South Atlantic coastal plains. Biological Report 82.10.123.
Washington, DC.

Shortliffe, E. H.1976. Computer-based medical consultationszMYCIN. American
Elsevier. NY.

Singh, H. Data warehousing: concepts, techniques, implementations, and
management. Prentice Hall PTR. Upper Saddle River, NJ.

Sparrows, R. D. and P. F. Springer. 1970. Seasonal activity patterns of white-
tailed deer in eastern South Dakota. Journal of Wildlife Management
34(2):420-444.

Stanley, T. R. 1995. Ecosystem management and the arrogance of humanism.
Conservation Biology 92255-262.

Starfield, A. M. and A. L. Bleloch. 1986. Building models for conservation and
wildlife management. Macmillan Publishing Company. New York. S

Starfield, A. M. 1997. A pragmatic approach to modeling for wildlife management.
Journal of Wildlife Management 61 :261-270.

Stauffer, D. F. 1990. Field evaluation of an HSI model for white-tailed deer in the
coastal plain. Unpublished report. Department of fisheries and wildlife
sciences. Virginia Tech University

Strickland, M. D., H. J. Harju, K. R. McCaffery, H. W. Miller, L. M. Smith, and R.
J. Stoll. 1994. Harvest Management. Pages 445-473 in T. A. Bookhout ed.
Research and Management Techniques for Wildlife and Habitats. The
Wildlife Society. Bethesda, MD.

Suring, L. H. and P. A. Vohs, Jr .1979. Habitat use by Columbia white-tailed
deer. Journal of Wildlife Management 43(3):610-619.

Swanson, T. 1997. Global action for biodiversity. Earthscan Publications Ltd.
Londson.

Sticklen, J. 1989. Problem solving architecture at knowledge level. Journal of
Experimental and Theoretical Artificial Intelligence 1(1):1—52.

125

Tierson, W. G., G. F. Mattfeld, R. W. Sage Jr., and D. F. Behrend. 1985.
Seasonal movements and home ranges of white-tailed deer in the
Adirondacks. Journal of Wildlife Management 49:760-769.

Tilghman, N. G.1989. Impacts of white-tailed deer on forest regeneration in
northwest Pennsylvania. Journal of Wildlife Management 53:524-532.

Turner, M. G.1989. Landscape ecology: the effect of pattern on process. Annual
Review of Ecology and Systematics 20:171-197.

Turner, M. G., V. H. Dale, and R. H. Gardner.1989. Predicting across scales:
Theory development and testing. Landscape Ecology 3:245-252.

Ullrey, D. E., W. G. Youatt, H. E. Johnson, L. D. Fay, and B. E. Brent. 1967.
Digestibility of cedar and jack pine browse for the white-tailed deer.
Journal of Wildlife Management 312448-454.

Urbain, J. 1997. Michigan deer hunting prospects. Michigan Department of
Natural Resources, Wildlife Diversion. Lansing, MI. 6pp.

Van Deelen. T. R. H. 1995. Seasonal migrations and mortality of white-tailed
deer in Michigan’s Upper Peninsular. Ph.D. Dissertation. Michigan State
University. East Lansing. MI. 158pp.

, H. Campa Ill, J. B. Haufler, and P. D. Thompson, 1997. Mortality patterns
of white-tailed deer in Michigan's Upper Peninsula. Journal of Wildlife
Management 61 (3):901-910

, T, R, H. Campa, M. Hamady, J. B. Haufler. 1998. Migration and seasonal
range dynamics of deer using adjacent deeryards in northern Michigan.
Journal Of Wildlife Management 62(1): 205-213.

Verme, L. J.1965. Swamp conifer deeryards in northern Michigan: their ecology
and management. Journal of Forestry 63:523-530.

,1968. An index of winter weather severity for northern deer. Journal of
Wildlife Management 32:566-574.

,1969. Reproductive patterns of white-tailed deer related to nutritional
plane. Journal of Wildlife Management 332881 -887.

,1973. Movement of white-tailed deer in Upper Michigan. Journal of
Wildlife Management 37:881-887.

, 1977. Assessment of natal mortality in Upper Michigan deer. Journal of
Wildlife Management 41:700-708.

126

 

, 1983. Sex ratio variation in Odocoileuza critical review Journal of Wildlife
Management 37:545-552

, 1989. Decline in doe fawn fertility in southern Michigan deer. Canadian
Journal of Zoology 69:25-28.

, and J. J. Ozaga. 1971. Influence of winter weather on white-tailed deer in
upper Michigan. Michigan Department of Natural Resources, Research
and Development Report No. 237. Lansing, MI 14pp.

Virgés E. and J. L. Telleria. 1998. Roe deer habitat selection in Spain:
constraints on the distribution of a species. Canadian Journal of Zoology
76:1294-1299.

Walker, B. 1998. The art and science of wildlife management. Wildlife Research.
25:1-9.

Wallmo, O. C., L. H. Carpenter, W. L. Regelin, R. B. Gill, and D. L. Baker. 1977.
Evaluation of deer habitat on a nutritional basis. Journal of Range
Management 30(2):122-127.

Walters, C. J. and J. E. Gross, 1972. Development of big game management
plans through simulation modeling. Journal of Wildlife Management.
3621 19-128

Walters, C.J. 1986. Adaptive Management of Renewable Resources.
MacMilIan, N.Y.

Wilson, M. L. and J. E. Childs. 1997. Vertebrate abundance and the
epidemiology of zoonotic diseases. Pages 224-248 in W. J. McShea, H. B.
Underwood, and J. H. Rappole. eds. The science of overabundance: deer
ecology and population. Smithsonian Institution Press, Washington and
London.

Woolf, A. and J. D. Harder. 1979. Population dynamics of a captive white-tailed
deer herd with emphasis on reproduction and mortality. Wildlife
Monograph 6721-53.

Xie, J., H. R. Hill, S. R. Winterstein, H., III Campa, R. V. Doepker, T. R. Van
Deelen, and J. Liu.1999. White-tailed deer management options model
(DeerMOM): design, quantification, and application. Ecological Modelling
124 (2-3):121 -130.

127

“IIIIIIIIIIIIIIJIIIIII