d.

f...

‘1

ally-.44... “flay“...
i 3"" C (u I
. .“m . .n. .1.

V!
$5
.3

fink, .
uhtmnka. It. I.— W
.2 i
.1. .
. .
. .
leer .r. _

a.
{E .4 w:
h ’-
..‘.u.u _ 4....
gt.“ _
tidafla .
.. z.

. 7.1 M».
an... a

n ,

i... «1
‘ l.
.dkhiil l s.
.5 (Q it
«til; 5
5.! d
r n .0- v
:‘lr‘ 1i-
. r
.
\‘I ‘ '4'
«$.51. !.
i .
I .
.121. 1.: 4
. $333..” r
‘ . o...
3 out
i. D
.r

 

a. .31 : . 2. .v
_ .. my} .0. H1..$....:ua.....
1. 1,3”: .r...

M. ..:..l

u l.‘ n‘ ’ ,o :1-1 4
”$4 .v 5%.;

r

.. 9W a. , .

 

335T

 

LIBRARY
Michigan State
University

 

 

 

This is to certify that the
thesis entitled

SPATIAL INTEGRATION OF THREATENED
OR ENDANGERED SPECIES AND TART
CHERRY ORCHARD LOCATIONS IN MICHIGAN

presented by

CORINNA NICHOLE RUBECK-SCHURTZ

has been accepted towards fulfillment
of the requirements for the

MS. degree in Fisheries and Wildlife

 

M

 

Maer ProfessBr's Signature
,7 // 7 /o 7

Date

MSU is an affinnative-acnon, equal-opportunity employer

-.-4‘.-.-.- -.- —.— -

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE

DATE DUE

DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6/07 p:/CIRC/DateDue.indd-p.1

 

SPATIAL INTEGRATION OF THREATENED OR ENDANGERED SPECIES AND
TART CHERRY ORCHARD LOCATIONS IN MICIHGAN

By

Corinna Nichole Rubeck-Schurtz

A THESIS
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
MASTER OF SCIENCE
Department of Fisheries and Wildlife

2007

ABSTRACT

SPATIAL INTEGRATION OF THREATENED OR ENDANGERED SPECIES AND
TART CHERRY ORCHARD LOCATIONS IN MICHIGAN

By
Corinna Nichole Rubeck-Schurtz

In 2003, the US Environmental Protection Agency denied tart cherry growers in
Oceana County, Michigan, an experimental use permit for a “reduced risk” insecticide
because of the presence of the endangered Karner blue butterfly (KBB: Lycaeides melissa
samuelis). However, this decision was not based on orchard-specific KBB data; KBB
locations were portrayed at a county scale, instead of a more biologically relevant scale.
My objective was to demonstrate a process for integrating private lands commodity
production with TES conservation. This was completed by producing federally
threatened or endangered species (TES) habitat and tart cherry block (TCB) maps. In
addition, I developed a spatial integration method to allow for better identification of
potential overlap areas between pesticide drift from TC 83 and TES habitat which can be
usefiJl for improving policy decisions. TCB spatial coordinates were collected through
global positioning system technology, and a pesticide drift layer was created in a
geographic information system (GIS). A predictive model was used to create statewide
habitat-suitability maps for three TES because Michigan currently lacks statewide TES
surveys. All data layers were integrated in a GIS to identify which tart cherry growers
had the potential to affect TES. Two example approaches integrating the data were
derived. Future work is required to determine the most appropriate habitat-suitability

layer to be used in the integration process for each TES.

AC KNOWLEDGEMENTS

Thanks are extended to Project GREEEN for providing the funding to support this
research project. I want to thank my major advisor Dr. Kelly Millenbah for all of her
support, advice, and encouragement. My committee members Dr. Mark Whalon
(Entomology), Dr. Larry Olsen (State Extension Agriculture Program Leader), and Dr.
Gary Roloff (Fisheries and Wildlife) also deserve thanks for their work on the project. A
thank you also goes to all those who gave me the opportunity to present this research at
national and international meetings.

Acknowledgment and thanks go to the members of the ESC WG (see Key to
Abbreviations, p. xi) who helped shape this project: Antonio Castro-Escobar (MDA),
Mike DeCapita (USFWS), Todd Hogrefe (MDNR), Brian Hughes (MDA), Margaret.
Jones (USEPA), Phil Korson (CM’I), Ben Kudwa (Michigan Potato Industry), Ken
Rauscher (MDA), Barbara Vantil (USEPA), and Denise Yockey (Michigan Apple
Committee). Peny Hedin (CIAB), the FMO, and Michigan tart cherry growers also
deserve thanks for their support in collecting tart cherry block spatial locations. Thanks
to Justin Booth (MSU RS & GIS), Helen Enander (MNFI), and Ed Schools (MNFI) for
their GIS support and Bill McConnell (Fisheries and Wildlife) for modeling advice.
Also, thanks to all fellow Biomapper users who have provided knowledge and support.

I want to thank Dana Infante for her help and advice on my thesis. Thanks also
go to my fellow lab mates for their help and support throughout this process. I want to
thank my family and friends for their support. Special thanks go to my husband, Kasey,

who has supported and encouraged me throughout this entire process.

iii

TABLE OF CONTENTS

LIST OF TABLES ....................................................................................................... vii
LIST OF FIGURES ........................................................................................................ ix
KEY TO ABBREVIATIONS ......................................................................................... xi
INTRODUCTION ........................................................................................................... 1
Need for Project .......................................................................................................... 1
Project Background ..................................................................................................... 2
Benefits of Pesticides to Tart Cherry Growers .......................................................... 2
Examples of Pesticide Eflects on YES ....................................................................... 4
Potential Method to Spatially Identify TES and TCOs along with Spatial Data Issues
.................................................................... 5

Case Study ................................................................................................................... 6
OBJECTIVES ................................................................................................................. 9
STUDY AREA DESCRIPTION .................................................................................... 10
METHODS ................................................................................................................... l3
Identifiling and C onvening a Working Group .................................... . ........................ '13
Michigan Tart Cherry Spatial Data ........................................................................... I 5
Collecting Data ...................................................................................................... l 5
Loading TCB Spatial Data into a GIS .................................................................... 18
Federal TES Spatial Data .......................................................................................... 18
YES and HS Layers ................................................................................................ 18
Generation of TES Layers ...................................................................................... 19

TES Habitat ............................................................................................................... 21
Karner Blue Butterfly Habitat ................................................................................ 21
Pitcher 's Thistle Habitat ........................................................................................ 21
Indiana Bat Habitat ............................................................................................... 23
Sources for and Initial GIS Development of EGV Layers ........................................... 23
Further Raster/EGV Layers Developmental Processes ............................................... 26
Creating Analysis Mask and Analysis Extent ........................................ . ................. 26
Converting ESRI Rasters to Idrisi Rasters .............................................................. 26
Converting Qualitative Layers into Quantitative Layers ......................................... 27
Boolean Layers .................................................................................................. 27
CircAn and Frequency of Occurrence Percent Statistical Method ....................... 28
Biomapper Processes ................................................................................................. 30
TES andEGVLayers ............................................................................................. 30
Processes ............................................................................................................... 30
Generating the HS layers for the TES ................................................................... 3 3

Creation of Two Example HS Layers per Species ....................................................... 33

Biomapper Cross-validation ...................................................................................... 34
Making the HS Layersfor each T ES Useable in Arc View or ArcMap ......................... 36
Integration of TES HS Layers and T CBS .................................................................... 36
Creation of Pesticide Drift Layer ........................................................................... 36
Data Integration Process ....................................................................................... 36
RESULTS ..................................................................................................................... 38
ESC WG .................................................................................................................... 38
Michigan Tart Cherry Spatial Data ........................................................................... 39
Michigan TES Spatial Data ....................................................................................... 39
Biomapper Processes ................................................................................................. 39
EGV Layers ........................................................................................................... 39
Verification ............................................................................................................ 43
ENFA ..................................................................................................................... 44
TES HS Layers ....................................................................................................... 49
CV .................... . .................................................................................................... 53
Integration of TES HS Layers, T C 85, and Pesticide Drift Layer ................................. 6O
DISC USS ION ............................................................................................................... 65
Developing the Process to Unite Groups and Identifiz Potential Areas of Overlap ...... 65
ESC WG ................................................................................................................ 65
Tart Cherry Spatial Data ....................................................................................... 67
Biomapper Challenges and Solutions ..................................................................... 68
TES, EG Vs, and GIS Spatial Data Matters ............................................................. 70
T ES Outputs and HS Layers ................................................................................... 72
Identification of Potentially Overlapping TES Habitat and T CBs ........................... 75
Develop Process to Aid in Decisions .......................................................................... 77
APPENDICES ............................................................................................................... 80
APPENDIX A: Michigan Georef ............................................................................... 81
APPENDIX B: Initial GIS Processes for IFMAP ....................................................... 82
APPENDIX C: Initial GIS Processes for Total Precipitation ..................................... 84
APPENDIX D: Initial GIS Processes for Percent Sand in Soil ................................... 86
APPENDIX E: Initial GIS Processes for Tree Canopy Cover .................................... 87
APPENDIX F: Initial GIS Processes for Texture and Drainage ................................. 88
APPENDIX G: Initial GIS Processes for National Elevation Dataset ........................ 89
APPENDIX H: Initial GIS Processes for Slope .......................................................... 90
APPENDIX 1: Initial GIS Processes for Average Daily Maximum and Average Daily
Minimum Temperatures ............................................................................................. 91
APPENDIX I: Initial GIS Processes for Proximity to Great Lakes ............................. 92
APPENDIX K: Initial GIS Processes for Wetlands .................................................... 93
APPENDIX L: Initial GIS Processes for Hydrologic Features ................................... 94
APPENDIX M: Initial GIS Processes for Proximity to Hydrologic Features ............. 95
APPENDIX N: Biomapper Challenges and Solutions ................................................ 96

LITERATURE CITED .................................................................................................. 99

vi

LIST OF TABLES

Table 1. Michigan counties (and abbreviations) containing tart cherry blocks (TCBs)
examined for this project. .............................................................................................. II

Table 2. Names and types of organizations participating in the Endangered Species and
Commodity Working Group (ESC WG) meetings. ........................................................ 14

Table 3. Spatial data sources used to create ecogeographical variables (EGV)
representing habitat features for the Karner blue butterfly (KBB). The derived EGVs
were used in the Ecological Niche Factor Analysis (ENFA). The table includes title of
the spatial data, originator of spatial data, location the spatial data were obtained from,
the appendix describing GIS processes for each spatial data layer, and EGVs derived
from each. ..................................................................................................................... 22

Table 4. Spatial data sources used to create ecogeographical variables (EGV)
representing habitat features for Pitcher’s thistle (PT). The derived EGVs were used in
the Ecological Niche Factor Analysis (ENFA). The table includes title of the spatial data,
originator of spatial data, location the spatial data were obtained from, the appendix
describing GIS processes for each spatial data layer, and EGVs derived from each. ....... 24

Table 5. Spatial data sources used to create ecogeographical variables (EGV)
representing habitat features for Indiana bat (IB). The derived EGVs were used in the
Ecological Niche Factor Analysis (ENFA). The table includes title of the spatial data,
originator of spatial data, location the spatial data were obtained from, the appendix
describing GIS processes for each spatial data layer, and EGVs derived from each. ....... 25

Table 6. Ecogeographical variable (EGV) layers used during the Ecological Niche Factor
Analysis (ENFA) and corresponding species. Species include Karner blue butterfly
(KBB), Pitcher’s thistle (PT), and Indiana bat (IB) ......................................................... 31

Table 7. Total number of tart cherry blocks (TCB) with total, mean, maximum, and
minimum hectares (ha) for all TC Bs included in this study. ........................................... 40

Table 8. Number of presence locations (i.e., polygons, points) used for Karner blue
butterfly (KBB), Pitcher’s thistle (PT), and Indiana bat (IB). The mean, maximum, and
minimum areas for presence locations whose last observation date occurred after 1990
are also listed. Total number of cells represents presence locations after each species
layer was converted to raster .......................................................................................... 42

Table 9. Ecological Niche Factor Analysis (ENFA) results for Karner blue butterfly
(KBB). Column headings indicate the first 8 (out of 11) ecological factors (marginality,
specialization I through 7), and the percent of specialization explained. Table values

vii

include coefficients on the marginality factor (sorted by decreasing absolute value) and
specialization coefficients for the ecogeographical variables (EGVs). ............................ 45

Table 10. Ecological Niche Factor Analysis (ENF A) results for Pitcher’s thistle (PT).
Column headings indicate the first 4 (out of 6) ecological factors (marginality,
specialization 1-3) and the percent of specialization explained. Table values include
coefficients on the marginality factor (sorted by decreasing absolute value) and
specialization coefficients for the ecogeographical variables (EGVs). ............................ 47

Table 11. Ecological Niche Factor Analysis (ENF A) results for the Indiana bat (IB).
Column headings indicate the first 3 (out of IO) ecological factors (marginality,
specialization 1-2) and the percent of specialization explained. Table values include
coefficients on the marginality factor (sorted by decreasing absolute value) and
specialization coefficients for the ecogeographical variables (EGVs). ............................ 48

Table 12. Mean area-adjusted frequency values resulting from the cross-validation (CV)
process. Table values represent the mean area-adjusted frequencies for bins l and 2 for

V each species reclassification scenario, while the values in parenthesis represent the

standard deviation .......................................................................................................... 56

viii

LIST OF FIGURES

Figure 1. Project study area is the state of Michigan. The main focus lies within the
counties highlighted in black as these contain the tart cherry blocks (TCB) under study.
County abbreviations are listed in Table 1. ................. . .................................................. 12

Figure 2. Example layout of tart cherry orchard (TCO). The outside black line indicates
the boundary of the TCO, which contains 6 tart cherry blocks (TCB). Each block is
labeled with the same grower number (e. g., GOOOO) and a different block number (e. g.,
_a, __b, _014) .................................................................................................................. 17

Figure 3. Number of tart cherry blocks (TCB) by county considered in this study. ........ 41

Figure 4. Habitat-suitability (HS) map for Karner blue butterfly (KBB). HS values range
from O (i.e., less suitable, lighter areas) to 100 (i.e., more suitable, darker areas). The
gray outline represents the Michigan state boundary. ..................................................... 50

Figure 5. Habitat-suitability (HS) map for Pitcher’s thistle (PT). HS values range from O
(i.e., less suitable, lighter areas) to 100 (i.e., more suitable, darker areas). The gray
outline represents the Michigan state boundary. ............................................................. 51

Figure 6. Habitat-suitability (HS) map for Indiana bat (IB). HS values range from O (i.e.,
less suitable. lighter areas) to 100 (i.e., more suitable, darker areas). The gray outline
represents the Michigan state boundary .......................................................................... 52

Figure 7. Area-adjusted frequency graph for the conservatively reclassified KBB. The
horizontal dashed line at 1 on the y-axis represents the random frequency line. The area-
adjusted frequency values are along the y-axis. The dot indicates partition values where
the first set of dots represents the results for bin 1 and the second set of dots represents
the results for bin 2. This graph only shows dots representing results for 1 bin’s
partitions. ........................................................................................................... . .......... 54

Figure 8. Area-adjusted frequency graph for the liberally reclassified KBB. The 1 on the
y-axis represents the random frequency line. Area-adjusted frequency values are
indicated on the y-axis. The dots indicate partition values where the first set of dots
represents the results for bin 1 and the second set of dots (not shown, but close to where
the solid lines run off the graph at <1 and 650) represents the results for bin 2. .............. 57

Figure 9. Area-adjusted frequency graphs for PT where a) represents the conservative
reclassification scenario and b) represents the liberal reclassification scenario. The 1 on
the y-axis represents the random frequency line. Area-adjusted frequency values are
indicated on the y-axis. The dots indicate partition values where the first set of dots

represents the results for bin 1 and the second set of dots represents the results for bin 2.
.............................. 58

ix

Figure 10. Area-adjusted frequency graphs for [B where a) represents the conservative
reclassification scenario and b) represents the liberal reclassification scenario. The l on
the y-axis represents the random frequency line. Area-adjusted frequency values are
indicated on the y-axis. The dots indicate partition values where the first set of dots
represents the results for bin 1 and the second set of dots represents the results for bin 2.
...................................................................................................................................... 59

Figure 1 I. Reclassified habitat-suitability (HS) maps for Karner blue butterfly (KBB),
where (a) represents the conservative reclassification scenario and (b) represents the
liberal reclassification scenario. In both maps unsuitable habitat is white and suitable
habitat is black. In the conservative HS map (a) unsuitable habitat represents HS values
of O and suitable habitat represents HS values from I-IOO. In the liberal HS map (b)
unsuitable habitat represents HS values from 0-99 and suitable habitat represents HS
values of 100. The gray outline represents the Michigan state boundary ........................ 61

Figure 12. Reclassified habitat-suitability (HS) maps for Pitcher’s thistle (PT), where (a)
represents the conservative reclassification scenario and (b) represents the liberal
reclassification scenario. In both maps unsuitable habitat is white and suitable habitat is
black. In the conservative HS map (a) unsuitable habitat represents HS values of O and
suitable habitat represents HS values from 1-100. In the liberal HS map (b) unsuitable
habitat represents HS values from 0-99 and suitable habitat represents HS values of 100.
The gray outline represents the Michigan state boundary. .............................................. 62

Figure 13. Reclassified habitat-suitability (HS) maps for Indiana bat (IB), where (a)
represents the conservative reclassification scenario and (b) represents the liberal
reclassification scenario. In both maps unsuitable habitat is white and suitable habitat is
black. In the conservative HS map (a) unsuitable habitat represents HS values of 0 and
suitable habitat represents HS values from [-100 In the liberal HS map (b) unsuitable
habitat represents HS values from 0-99 and suitable habitat represents HS values of 100.
The gray outline represents the Michigan state boundary. .............................................. 64

KEY TO ABBREVIATIONS

Absolute Validation Index

Cherry Industry Administration Board

Cherry Marketing Institute

Circular Analysis

Cross-Validation

Contrast Validation Index

Ecogeographical Variable

Ecological Niche Factor Analysis

Endangered Species Act

Endangered Species and Commodity Working Group
Environmental Systems Research Institute, Inc.
Experimental Use Permit

Federal Marketing Order

Geographic Information System

Global Positioning System

Habitat-Suitability

Indiana Bat

2001 Integrated Forest Monitoring Assessment and Prescription
Integrated Pest Management

Karner Blue Butterfly

Michigan Department of Agriculture

Michigan Department of Natural Resources
Michigan Natural Features Inventory

Michigan State University

Michigan State University Remote Sensing and GIS
National Agricultural Statistics Service

Pitcher's Thistle

Tart Cherry Block

Tart Cherry Orchard

Threatened or Endangered Species

US Department of Agriculture

US Environmental Protection Agency

US Fish and Wildlife Service

US Geological Survey

US Geological Survey Earth Resources Observation and Science
Value Attribute Table

xi

AVI
CIAB
CMI
CircAn
CV
CVI
EGV
ENFA
ESA
ESC WG
ESRI
EUP

F MO
GIS
GPS
HS

IB
IFMAP
[PM
KBB
MDA
MDNR
MNFI
MSU
MSU RS & GIS
NASS
PT
TCB
TCO
TES
USDA
USEPA
USFWS
USGS
USGS EROS
VAT

INTRODUCTION

Need for Project

The Endangered Species Act (ESA; 16 U.S.C. 1531 et set.) seeks to protect
threatened or endangered species (TES) and the essential habitat of the species (U SFWS
1996), and its rules and regulations can influence commodity producers whose economic
livelihoods may be affected by TES on or near their property. While many people
acknowledge the importance of protecting TES from extinction, opinions on conservation
approaches differ when TES protection impacts people’s jobs or livelihoods. For
example, in New Mexico, a difference of opinion occurred between farmers needing to
irrigate crops and environmentalists (Scharpf2001). The farmers claimed rights to the
water while environmentalists believed the river should be managed in a way that
ensured the survival of the Rio Grande silvery minnow (Hybognathus amarus). This
same type of discord has occurred with other TES conservation activities including the
threatened northern spotted owls (Strix occidentalis caurina) in the forests of the Pacific
Northwest (Freudenburg et al. 1998, Carroll et al. 1999) and gray wolves (Canis lupus) in
the western US (i.e., Idaho, Montana, Yellowstone Ecosystem) (Hardy-Short and Short
2000). To ensure that people’s livelihoods and TES are protected there is a need for open
communication among all stakeholder groups (i.e., commodity producers, private
landowners, agencies) to ensure that the most accurate and relevant information is
available for decision-making.

In Michigan, tart cherries are an important commodity that could potentially be

impacted by TES. Recently, some of Michigan’s tart cherry growers were denied a

pesticide experimental use permit (EUP) because of potential harm to a TES (pers.
comm, M. Whalon, Department of Entomology, Michigan State University (MSU), pers.
comm, P. Korson, Cherry Marketing Institute (CMI), Project GREEEN 2005). This
situation highlighted a need to develop a process that could identify the spatial locations
near tart cherry orchards (TCO) where pesticide application could potentially impact
TES. Such a process would help identify areas where mitigation efforts for TES
protection could be directed.
Project Background
Benefits of Pesticides to Tart Cherry Growers

Commodity producers experience benefits from the use of pesticides including
higher crop yields and higher quality crops by reducing diseases, pest insects, and
competition with weeds (Pope et al. 1998; Ragsdale I999; Tart Cherry Pest Management
Strategic Plan 2000, 2006; USEPA 2006). In Michigan, the tart cherry commodity
provides income for a large number of Michiganders and produces a plethora of tart
cherry products for the US. Michigan leads the nation in tart cherry production, and in
2004, produced 149 million pounds, or 70%, of the US tart cherry product, followed by
208 million pounds, about 77%, in 2005 (Pollack and Perez 2006). The next two highest
tart cherry producing states for 2004 and 2005 included Utah, which produced 22 million
and 28 million pounds, respectively, and Washington, which produced 17.5 million and
16.5 million pounds, respectively (Pollack and Perez 2006).

In Michigan, tart cherry growers use pesticides to protect their crops from
damage. Some insecticides are used to protect tart cherries from plum curculio

(Conotrachelus nenuphar) and cherry fruit fly (Rhagoletis cingulata and Rhagoletis

_fausta) larvae damage (Tart Cherry Pest Management Strategic Plan 2000, 2006; USEPA
2001). Infestations can prevent a crop from going to market and if a cherry is found to
contain either pest the grower’s fruit can be denied for sale (Whalon et al. 1999; Tart
Cherry Pest Management Strategic Plan 2000, 2006; USEPA 2001). Other protective
measures include the use of herbicides which can be used to reduce the amount of
competition for nutrients between tart cherry trees and other plants (e. g., weeds) (Tart
Cherry Pest Management Strategic Plan 2000, 2006). Fungicides can also be used to
prevent outbreak of disease (e. g., American brown rot (Moniliniafiucticola), cherry leaf
spot (Blumeriellajaapii)), in TCOs that could damage the fruit or the tree (Tart Cherry
Pest Management Strategic Plan 2000, 2006).

While use of pesticides benefit crops by protecting them from disease,
competition, and pest insects, proactive steps are already being taken in Michigan to
reduce the potential for interactions between pesticides and the environment. For
example, Michigan’s tart cherry grower community is part of an integrated pest
management (1PM) program (Tart Cherry Pest Management Strategic Plan 2000, 2006).
This program allows tart cherry growers to produce a high-quality product, but also to
reduce pesticide use and potential negative effects on the environment and workers in the
orchards (Tart Cheny Pest Management Strategic Plan 2000, 2006). Some examples of
IPM practices and research occurring in TCOs include using organophosphate
alternatives to control pests, adjusting pesticide application methods, changing pesticide
application equipment, using models to help determine when to apply pesticides, and
improving pest monitoring methods (Tart Cherry Pest Management Strategic Plan 2000,

2006; Tart Cherry Integrated Orchard Management 2006).

Examples of Pesticide lgflects on TES

Although pesticides protect crops they can potentially have negative effects on
TES. Several examples highlight the interactions between TES and pesticide exposure.
One of the better known examples involves the bald eagle (Haliaeetus leucocephalus).
The bald eagle was listed as endangered through most of its range in 1978 because of
declining populations, and one cause for these declining populations included effects
from pesticides (USFWS 1983). Bald eagles ingested pesticides through consumption of
contaminated prey (e. g, fish) (USFWS 2006). A study by Reichel et al. (1984)
investigated 293 dead bald eagles and found all carcasses were contaminated with DDE
(p,p’-dichlorodiphenyldichloroethylene). Also, dieldrin may have played a role in the
death of five bald eagles as high levels of dieldrin were found in their brains (Reichel et
al. 1984). Other research indicates pollutants may have led to diminished reproduction as
DDE has been linked with bald eagle egg shell thinning (USFWS 1983, Wiemeyer et al.
1984, USFWS 2006). DDT (dichlorodiphenyltrichloroethane) was used to control insects
that could transmit diseases along with pest insects for crops (USEPA 1972) and dieldrin
was also used to control crop pest insects along with locusts, mosquitoes, and termites
(USEPA 2007). Banning of the majority of uses of DDT throughout the US in 1972 in
combination with other conservation activities allowed bald eagle populations to increase
over time to the point where the bald eagle could be delisted (USFWS 2006).

Various studies have shown the harmful effects that pesticides have had on TES
bat species. One study on the food source of endangered gray bats (Myotis grisescens)
found that 86% of insect samples contained dieldrin and/or heptachlor epoxide (Clawson

and Clark 1989). Aldrin (i.e., parent compound for dieldrin) was used to control

cutworrns in corn fields (Clawson and Clark 1989), Another study by Clark et al. (1978)
examined the brains of 28 dead endangered juvenile gray bats from Missouri and found
some contained lethal amounts of dieldrin. O’Shea and Clark (2002) reported lethal
levels of dieldrin in the brains of endangered Indiana bats (IB: Myotis sodalis) from
Missouri. Juvenile endangered gray bats are also thought to be more sensitive than adults
(Clark et al. 1983), and Clawson and Clark (1989) found that two juvenile gray bats died
from dieldrin poisoning likely caused by consumption of contaminated milk from their
mothers.

A study by Herms et al. (1997) examined the effects of the pesticide Bacillus
thuringiensis var. kurstaki (Bt) on the endangered Karner blue butterfly (KBB:
Lycaeides melissa samuelis) larvae and found that 73% of larvae tested died at low levels
of exposure (30-37 Billion International Units/ha). Herms et al- (1997) also examined the
timing of KBB larvae development with regard to the temporal application of Bt and
found that KBB larvae could be present at the same time Bt applications are needed to
control gypsy moth (Lymantria dispar) outbreaks, suggesting that the KBB larvae may be
susceptible to Bt exposure.
Potential Method to Spatially Identify YES and 'I'COS along with Spatial Data Issues

Geographic information system (GIS) and global positioning system (GPS)
technologies can be used to aid decision makers, landowners, and commodity producers
with TES concerns by recording and displaying spatial data. While GIS and GPS
technologies are frequently used by researchers, GPS and GIS data are ofien lacking for
TES. For example, the spatial locations for a particular TES might be recorded for only

portions of its range or these data might include only presence data (i.e., area surveyed

and species found) and not absence data (i.e., area surveyed and species not found). With
incomplete TES spatial data, it becomes challenging to use these data for making good
management or policy decisions because not all locations of species occurrences are
known.

One potential mechanism to overcome a lack of TES spatial data is to model
where TES potentially suitable habitats are likely to occur. Engler et al. (2004) suggested
that predictive models can potentially play an important role in conserving TES, but few
studies on TES have used them possibly because of a lack of information (i.e., sightings
data), a lack of “defined sampling units”, and a lack of absence data. Many different
types of predictive models have been created including those that require only species
presence data as well as others that require a combination of species presence and
absence data (Guisan and Zimmermann 2000, Hirzel et al. 2001, Hirzel et al. 2002).

Besides a lack of species data, data describing environmental characteristics may
also be lacking (Engler et al. 2004). Microclimate data and specific plant community
type data are examples of data sources that would be useful in a GIS environment, but are
hard to find or portray in a spatial environment (Engler et al. 2004). Fine scale GIS
layers like these would require a considerable amount of time to collect and portray as
layers. Additionally, these types of spatial data would be expensive to collect and create.
Case Study

The negative effects experienced by TES from pesticide exposure and positive
effects of pesticide use for protecting commodity crops emphasizes the need to develop a
process that will allow for better protection of both TES and commodity production. In

Oceana County, Michigan, this opportunity was highlighted when tart cherry growers

sought use of the “reduced risk” insecticide AvauntTM (E. I. de DuPont de Nemours, Co.,
Newark, Delaware), through a pesticide EUP, which would have helped growers protect
their tart cherry crops from plum curculio (pers. comm, M. Whalon, pers. comm, P.
Korson, Project GREEEN 2005). The US Environmental Protection Agency (USEPA)
denied the pesticide EUP as the endangered KBB exists within Oceana County (pers.
comm, M. Whalon, pers. comm, P. Korson, Project GREEEN 2005). Although the
species was present in the county, [(885 or KBB habitat may be located far enough from
the TCOs to not be affected by pesticides thus not all tart cherry growers in Oceana
County should have been denied use of the pesticide EUP without a more detailed spatial
evaluation. “The failure to obtain the pesticide EUP jeopardized approximately 17% of
the US's tart cherry production because growers in the County were not able to
participate in University assisted research and transition to what were deemed by USEPA
as “reduced risk” and theoretically environmentally safer insecticide tools” (pers. comm,
M. Whalon).

For this project, I used a GIS to examine the spatial locations of Michigan’s tart
cherry commodity with regards to pesticides drifting from tart cherry blocks (TCB) and
its potential interaction with three TES: KBB, Pitcher’s thistle (PT; Cirsium pitcheri) and
1B. These three TES were chosen because 1) they represent a wide range of organisms
including insects, plants, and mammals, 2) they have confirmed presence locations in the
same counties as those with TCBs or those counties immediately surrounding the TCB
counties, and 3) they represent a variety of ways that organisms may be affected by

pesticide use (e.g., KBB larvae have died from pesticide exposure (Herms et al. 1997), IB

 

have contained high amounts of pesticides in their brain (O’Shea and Clark 2002), and
PT was selected because it is a plant and may be susceptible to herbicides).

The goal of this project was to demonstrate a process for improved integration of
commodity production and TES conservation. This was accomplished by producing TES
habitat-suitability (HS) and TCB maps and developing an integration process for these
entities to locate potential areas of overlap. The process outlined by this research can aid
agencies, commodity producers, and landowners in decision-making (including policy)
especially as it relates to pesticide use. This project serves as the pilot for establishing
relationships among commodity producers and groups involved with TES protection to
set a national example in proactive management techniques. This project also allowed

for the generation of next step recommendations to refine the process.

OBJECTIVES

Specific objectives of this project were to:

1.

Develop a process that unites TES conservation activities and TCB locations to
identify potential areas of overlap,

Develop a process that can provide assistance in making decisions and policy
regarding protection of TES and pesticide use,

Set a national example for the protection of quality commodity production and
TES, and

Make recommendations for next steps in the process.

 

STUDY AREA DESCRIPTION

The study area was the state of Michigan, with the main emphasis located on 15
counties in the western portion of Michigan’s Lower Peninsula due to the location of
TCBs (Table I, Figure 1). Major cities in or near the study area include Benton Harbor,
Kalamazoo, Grand Rapids, Muskegon, and Traverse City. This area produces tart
cherries, sweet cherries, apples, peaches, plums, blueberries, strawberries, pears,
brambles, and grapes (Michigan Fruit Districts No Date). This region is suitable for fruit
production primarily due to its proximity to Lake Michigan which helps to buffer
temperature extremes (Olmstead 1956; Tart Cherry Pest Management Strategic Plan
2000, 2006). Similarly, the landscape is composed of moraines and elevated till plains,
which, if orchards are located near the tops of these features, allow heavier, colder air to
flow away from elevated areas protecting crops from frost damage (Olmstead 1956).
Soils in the area are well-drained and sandy loamy which are well—suited for crops

(Olmstead 1956; Tart Cherry Pest Management Strategic Plan 2000, 2006).

10

Table 1. Michigan counties (and abbreviations) containing tart cherry blocks (TCBs)
examined for this project.

 

County Abbreviation
Allegan ALL
Antrim ANT
Benzie BEN
Berrien BER
Cass CAS
Grand Traverse GRTR
Ionia ION
Kalamazoo KAL
Kent KEN
Leelanau LEE
Manistee MAN
Mason MAS
Muskegon MUS
Oceana OCE

Van Buren VABU

 

ll

 

Ontario

     

Wisconsin

\ U" I,”

 
  
  

> N Illinois

 

Indiana

 

- Counties in Study V Great Lakes 0 35 70 140 Kilometers
l:l United States 3,3 Canada L——;l—I———J

Fig. I. Project study area is the state of Michigan. The main focus lies within the
counties highlighted in black as these contain the tart cherry blocks (TCB) under study.

County abbreviations are listed in Table 1.

METHODS

Identifying and Convening a Working Group

The Endangered Species and Commodity Working Group (ESC WG) was formed
to provide research guidance on developing the process for integrating TES information
and TCB data. The ESC WG participants were limited to people who had interests, jobs,
or involvement with commodity production and TES issues and included representatives
from government agencies, universities, and commodity groups (Table 2). This group
was used to resolve data needs and provide insights into the utility of the process being
developed, including whether it would be accepted by different commodity and
government groups, and whether it could be used by these groups in the fiiture. Finally,
this group established lines of communication among all parties involved who otherwise
might not have the chance to meet.

Creation of the ESC WG began with development of an organization list from
Michigan and the Chicago USEPA Region 5 whose personnel would be interested in
contributing time and information pertaining to commodity production, GIS, modeling,
laws and regulations, and TES. The commodity groups chosen were CM], Michigan
Potato Industry Commission, and Michigan Apple Committee. CMI was selected
because it is the coordinating entity for the tart cherry producers throughout Michigan
and the tart cherry commodity was selected as the example because of the fore mentioned
pesticide EUP denial. The Michigan Potato Industry Commission and Michigan Apple
Committee were asked to join because this project has the potential to expand to other
commodity organizations in the future and their early involvement allowed for more

holistic planning throughout the project. The US Fish and Wildlife Service (USFWS),

Table 2. Names and types of organizations participating in the Endangered Species and
Commodity Working Group (ESC WG) meetings.

 

 

Organizations Type
Cherry Marketing Institute (CMI) Commodity
Michigan Apple Committee Commodity
Michigan Department of Natural Resources (MDNR) Government
Michigan Department of Agriculture (MDA) Government
Michigan Natural Features Inventory (MNFI) University
Michigan Potato Industry Commission Commodity
Michigan State University (MSU) - Department of Entomology University
MSU - Department of Fisheries and Wildlife University
MSU - State Extension Agriculture Program Leader University
US Environmental Protection Agency (USEPA) Government
US Fish and Wildlife Service (USFWS) Government

 

l4

USEPA, Michigan Department of Natural Resources (MDNR; Wildlife Division), and
Michigan Department of Agriculture (MDA) were selected to be members because of
their understanding of laws and regulations pertaining to TES and pesticides at both the
state and federal level. The MDNR and USFWS were also selected as they are
responsible for TES protection. Michigan Natural Features Inventory (MNFI)
contributed TES spatial data and GIS support to the project. The project was coordinated
by researchers from MSU’s Department of Fisheries and Wildlife, Department of
Entomology, and the State Extension Agriculture Program Leader.
Michigan Tart Cherry Spatial Data
Collecting Data

Currently, Michigan tart cherry growers are not required to record the spatial
coordinates for their TCOs in a GIS format (pers. comm, P. Hedin, Cherry Industry
Administrative Board (CIAB)); therefore no GIS spatial data existed for TCO locations.
A request was made to the Federal Marketing Order (FMO) to encourage the FMO to
require tart cherry growers to use a GPS to record locations for their TCOs. While the
FMO did not require that GPS coordinates be recorded for all TCOs, they did approve of
GPS locations being recorded as a part of the 2006 TCO diversion process and used in
this project. The orchard diversion process occurs yearly and allows tart cherry growers
the option to leave their fruit in the fields when the fruit supply is greater than the
demand for the fruit (pers. comm, P. Korson). For this process to be carried out,
National Agricultural Statistics Service (NASS) employees enter each TCO and record

which TCBs the grower is diverting.

Researchers worked alongside the CIAB to train NASS employees in the
collection of TCB GPS coordinates throughout the orchard diversion process using
Magellan GPS units. NASS employees were given GPS data collection and GPS data
recording training while attending one of three regional orchard diversion process
meetings (i.e., southwest, west central, and northwest). Before GPS units were sent to
NASS employees, each was programmed with the WGS 84 datum to ensure consistency.

Corner coordinates (i.e., Lat/Long, degddddd) for each TCB were collected and
hand recorded by NASS employees throughout the 2006 TCB spatial data collection
season. Corner coordinates for TCBs were collected in place of TCO coordinates as
NASS employees were visiting TCOs to divert blocks and not the entire orchard. TCBs
exist within TC Os and TC Os can contain multiple blocks of tart cherries (Figure 2). The
data recorded for every TC B corner coordinate consisted of grower name, grower
number, block number, GPS unit number, elevation, accuracy, latitude, and longitude.
The NASS employees sent the completed data sheets to CIAB where the grower names
were removed. The names of the growers are held with the CIAB to protect private
landowner information.

Some TCB spatial data were collected by the researcher in 2005 using a Garmin
GPS unit (i.e., spatial data representing 25 TCBs), but the majority of TCB spatial data
were collected by NASS employees in 2006. The TCB spatial data collected in 2005
might consist of multiple TCBs combined into one as they were collected by the
researcher who did not have access to paper maps representing each individual TCB.
While spatial data collected in 2005 were collected using different methods, they were

used in this project to increase the number of TCB locations. The error term for polygons

G0000_i 00000
015

   
   

 

 

 

 

G0000_b
C0000
_014 4’16
00000_a
N
- Tart Cherry Blocks 0 0.05 0.1 Kilometers

l:l Tart Cherry Orchard 1. -. .1 , , ._.I

Fig. 2. Example layout of tart cherry orchard (TCO). The outside black line indicates the
boundary of the TCO, which contains 6 tart cheny blocks (TCB). Each block is labeled

with the same grower number (e.g., G0000) and a different block number (e.g., _a, _b,
_014)

17

is unknown.
Loading TCB Spatial Data into a GIS

Corner coordinates, grower number, and block number for each TCB were
entered into a Microsoft Excel spreadsheet along with the TCB spatial data collected in
2005. Spatial identification of TCBs was represented by the tart cherry grower number
and TCB number (e. g, G0000_001). The Microsoft Excel file was saved as a .dbf file
and uploaded into ArcView (Environmental Systems Research Institute (ESRI), Inc.
Redlands, CA) where it was converted to a spatial point file. Using ArcView, the TCB
corner coordinates were projected to the Michigan Georef projection and the corner
coordinate locations for each TCB were converted into individual TCB polygons using
the point to polygon fiinction. The TCB polygons were examined for errors (i.e.,
overlapping boundaries, incorrect comer coordinate positions), and if an error was found
that could not be fixed, that TCB was removed from the sample.
Federal TES Spatial Data

This project focused on three federally listed TES occurring in Michigan.
Locations of all federally listed species were initially assembled fi'om existing data
obtained from MNFI in digital form. Other groups were contacted for TES spatial data
(e.g., USFWS, researchers) to ensure that all TES spatial data available were used for this
project.
[ES and HS Layers

The TES data collected by MNFI represents only areas surveyed for TES as they
are a positive sighting group (pers. comm, E. Schools, MNFI). As such, MNF I only has

data for locations where the TES is known to occur as not every location in Michigan has

18

been surveyed for TES, thus Michigan lacks a statewide survey for TES. Care must be
taken when interpreting this data because areas not currently surveyed could be treated as
lacking TES. For certain locations this is incorrect and misinterpretations could occur if
only the current TES survey data were used as other places where the organisms exist
would not be included. This obstacle was overcome by modeling TES habitat based on
known locational data (i.e., presence or known locations) in Biomapper 3.1 software
(Hirzel et al. 2002, Hirzel et al. 2004), which uses the Ecological Niche Factor Analysis
(ENFA) (Hirzel et al. 2002). The ENFA identifies potential locations that may provide
suitable habitat for the species by overlaying known species locational data and
ecogeographical variables (EGV). EGVs are quantitative ecological, topographical, or
anthropogenic variables that are spatially defined and found throughout a study area
(Hirzel et al. 2002, Hirzel 2004). The ENFA takes the known species distribution on
EGVs and compares it with the EGVs throughout the study area (Hirzel et al. 2002)
ultimately allowing for the creation of HS maps.
Generation of TES Layers

The KBB, PT, and [B were the focal species for the project. A shapefile (i.e., data
set for ArcView (ESRI 1997)) was obtained from MNF I containing presence data
collected for all state and federally listed species in Michigan from 1831 to 2005. This
shapefile was queried three times (i.e., once for each selected species) in ArcView to
create the respective species layer whose last species observation occurred on or after
1990. Records older than 1990 were not considered because of the potential for
organisms to move or for their habitat to have been altered since that time. The KBB and

PT presence data were captured as polygons, while the [B presence data was captured as

19

both polygon and point data. Both the point and polygon data were used for the IB
analysis. The majority ofthe three TES presence data were not site specific, but were
represented as polygons where every location within a presence polygon was equally
likely to have the species in it (pers. comm, E. Schools).

In ArcView, each species layer was converted to a 30 m raster. A 30 m raster was
selected as it represented the finest resolution for EGV layers used in this project. The [B
polygon species raster and IB point species raster were merged into one raster creating
the IB species raster. In ArcMap, the KBB and PT species rasters were projected to
Michigan Georef(Appendix A).

Although each species raster did not occupy the spatial extent of the entire state, it
was necessary to expand each raster to this extent for use in Biomapper software as the
entire state of Michigan was the study area for this project. To overcome this issue, a
raster analysis mask was used to make each species raster statewide. Each species raster
analysis mask was set to a blank Michigan county raster (i.e., all values 0), while the
extent was set to each species’ corresponding species raster (i.e., for KBB the analysis
mask was the blank Michigan county raster and the extent was the KBB species raster).
A blank Michigan county raster was used to ensure no values from the species raster
changed during the process. The masking process was carried out as each species raster
was reclassified such that all presence locations were represented by a 1 and locations
with no data for the species were represented by a 0. KBB, PT, and IB species boolean
layers (i.e., 1 representing species presence locations and 0 representing locations with no
data for species) were created because it is the data format required for use in Biomapper

software.

20

TES Habitat
Karner Blue Butterfly Habitat

KBB habitat is associated with remnant oak - pine savanna barrens, openings in
forests, airports, military camps, old fields, forest roads and trails, and power line or
highway right-of-ways (Rabe 2001, U SFWS 2003). Important land cover types for the
KBB include aspen, herbaceous openland, low density tree, oak, pine, and upland mixed
forest (pers. comm, J. Kleitch, MDNR; pers. comm, J. Skillen, Community College of
Southern Nevada; pers. comm, E. Schools)- KBB larvae feed primarily on wild lupine
(Lupinusperennis) while adults feed on nectar from flowers (Rabe 2001, USFWS 2003).
Soil characteristics important to KBB habitat include well-drained and sandy soils (Rabe
2001, USFWS 2003) and the species has been shown to prefer areas with less tree canopy
cover but can be found in areas with up to 80% cover (USFWS 2003). Consequently,
EGV selections for KBB included important land cover types (i.e., aspen, herbaceous
openland, low density tree, oak, pine, and upland mixed forest), soil texture and drainage
(i.e., well-and moderately-drained soils), percent sand in soil, total precipitation, tree
canopy cover, and elevation (Table 3). A layer representing statewide lupine occurrences
could not be found.
Pitcher ’s Thistle Habitat

PT is a plant that grows on non-forested sand dunes around Lake Michigan, Lake
Huron, and Lake Superior (Higman and Penskar 1999, USFWS 2002). Other variables
important to plant species that were used in other predictive models include slope,
precipitation, and temperature (Zaniewski et al. 2002, Engler et al. 2004) and these were

also used in this study. Accordingly, EGV selections for PT included elevation, land

. . . .-
-.n‘-..«thfv. M .V\P‘. \
lllllll It‘ln“ ~ -

 

cougflm

o 5283

GmZSC @852: 8:5qu 12:32 53:32
.822m .3 ”835 SHED Sam mOMm ,mOmD

3ng
so§>2m 3:232

 

.358 30:8 08H m 5333‘ $0va ooaomom 3980
use someatomno mooSomom 5am mOmD £369 on; Sow Swansea
.3 A83 $83 3&5 3868 m: :38 23 38:2
eoesaaee Ex; 0 588% 955 2mm: 8%
“banana: 2% 835 .388 380 52a ._8_ 322 amass
Kém .985 “So:
”Am—«Eozv coufiasoi
£8 3 3mm Eoocom Q Egon“? “coaomng gonomi >20
aemnofi Ba wage: _sseafiém .am .23 - 83:28
Hot“ 832883 mom Ammma 82>? 98 SEE
£8 cocmw8$§8onofi m cannon“? DmE .ansomd 09350 98 23on
@5433 we xoqozvocm ”@530 336323 88m ammEBE £345 d
388 m 5333‘ Sagan“ 0592ng 330 was Banana
@838 “:83: Jim dogma Jae com §80 namEoME Amoomv H8635 8&3 use 3304
£85 36:3 32 £3353 8080352 2E Ba 3352 .398...“ .mZQE m<OE<2E 32..
30333: .«o moaosvfim
m>Om 3>EQ $3805 88.,— nofifino cosmos 382%ch ooBow Sea 329
mIHO RES

 

22

.53 88w 33.8w from 98 £98— 88
33% :80 8m mommoooa m5 wastage Sausage Bu .8on 3:330 203 33 33% 05 8282 .83 away. we save 65

33% 2: mo 0% 338m 033 as. .3va mini 55 £22 Eassm 2: a... no»: 29: “>8 E5 2: 38¢ causes
83 55 05 .8 353 9333 gage? came as“? Ezaamsmso 233 a was “85% 3% 39% .m “Ea

cover type (i.e., sand - soil), tree canopy cover, total precipitation, average daily
maximum temperature, average daily minimum temperature, proximity to the Great
Lakes, and slope (Table 4).
Indiana Bat Habitat

This study focused on the summer foraging and roosting habitat of the 113. Winter
habitat was not considered as most [B hibernate in more southern states, although a few
have been recorded hibernating at Tippy Dam in Michigan (Kurta et al. 1997, Kurta and
Rice 2002). Many of the roosting sites for the IB consist of dead or dying trees with
exfoliating bark (Humphrey et a1. 1977, Callahan et al. 1997, USFWS 1999, Farmer et al.
2002, Kurta et a1. 2002). Land types associated with foraging and roosting behaviors of
the [B include wetlands, upland forests, riparian, agricultural, areas with water, flood
plains, and coniferous forest (Humphrey et al. 1977, Clark et al. 1987, Kurta et al. 1996,
USFWS 1999, Farmer et a1. 2002, Kurta et a1. 2002). As a result the EGV selection for
[B included important land cover types (i.e., croplands, hydrologic features, wetlands,
and uplands), tree canopy cover, total precipitation, average daily maximum temperature,
average daily minimum temperature, elevation, and proximity to hydrologic features
(Table 5). Layers representing dead or dying trees or exfoliating bark were not available.
Sources for and Initial GIS Development of EC V Layers

The spatial data sources obtained to create the EGV layers were developed by
different organizations (Tables 3, 4, 5). See Appendices B-M for initial GIS processes

used to manipulate each layer. All resulting rasters had a cell size of 30 x 30 m2.

 

83%on “an;

$50 3098 02H

one—m £2855

8qu 320 3 bmfimxem
23anan

82558 baa omfio><

058383
:38qu8 33”. ommuo><

:8 - 28m .3 55:35

m>0m 33.35

 

03¢ .

m x%5&<
: .6 303834

_. x659“?

H 598%?
_ 523%

m 596an

8383 20 3:5

955 Swan V

5.3%: saw 535 .983 955 Emma
60% meme 238 .H £89 mum:

Ezz figm .3 x83

E80 35. 6.38 0828 9a 8%?530
season saw .883 35m 12380 ,3
AEZEV ESEEH

mega—“mom 3.582 59:22 200an .m b30523
955 Emma

SE32: 3% 835 .386 955 5:3

955 29%
520%: 33m 895 .383 35 Emma

892585
oEafimooO Sm ~8qu 53:32 ”383 “SEED

“5:53:22 BE was 132:2 Jmflom .MZQE
88m $5890 cosmos uofifiwto 8.50m Sam

88m

A 53 32:2 6935

. $8 .385 :8:
”$38.85 nouflmnzookm
3980

02% Son 838mm
350 33 15232
womanh—

aougfim 13332

$qu 820

Aowfiog
.Sém 99:3. .52 3me
owfio>$ ”Eng—ca
oafiomfioh SEES—z
Ewan:
$2 snack $2 $3
owfio><v ”$38876
2383809 85::me
$30 was 3355;
6&3 was $30;—
m<OE<EE 50m
82%

 

.n—owo Scum B>COU m>mvm

was #93. 8% 33% :80 SM mommoooa m5 wflfibmou 530%? 05 ARE 3:330 20>» 8% 33% 05 H.382 S3 33% mo
“gunfire .83 33% 2: mo 0:: 83:05 035 2F .szmc mam—SE “88m 0:32 Homwsoum 06 5 com: 803 m>0m Egret 2:
.CLV Bug 98:85 8% 858m “833 mama—omega $08 833:3 HoEmeoowooo 38.8 3 com: 3958 Saw Hummm .v 033.

24

 

 

coHHSEHooHQ ESP 0 inseam/H @380 Ewan 88m; EH 3382 dwfigm Sém
.meHPHHHHD 88m 835 A383 9..on Ewan .. .90on got .AmHaEHoZV 5:836on
H6303 M iguana... E22 . V @382: $5303 8:232
Ho 53.55 . . . 35% .HH [Hgadac ~HH...,HH.H§HHH HHHaHHoB
HHSHHHZ .HmBHHmB gum £553 Ea HHHHHH m:
33o 225 8; m i225 60mm meme 533 .H ASHE mom: 388 8;
Sou @3anme 530 can; _wcoHHaZ.
cocnkxm O vacuum“? 9233 H0852: 6325 Hawaii :25me
BHanm Eanz aawEoHE .533m .3 Ago:
5250 Sam 30va oouflom EB HHoHHHHEoBO
39:5QO Sam .AmOmDv 325m HHHonOHBO wD
8333 33.06»: 2 nonmgomfi $283:
8 9838.3 J 80:52:? ufiafiwooc 8% .5300 53:22 ”Good HxHoaoEfim oEmanOuO 535:2
28 .8 55:?on gunshot: oEquwooO How 5230 59:22
833388 H 522:3 @380 Ewan Emma;
EHHEHEE HmeHoZHHD 88m coonO Aoooomv 35.5 Emam Sém .mth .52 ESQ umm5>$
3:6 umfio>< ”Am—HHEHOZV 8383th 62552
0253352 H xEHHuHE< H320 2min A65 Hzém :9th .HaE EEG owfiu><v
.52: 3:6 owEo>< SHHmHoZHHD 83m 83.5 325.8 @380 $553 AmHHHEHoZV BBSoQEoH Esgxmz
muss—9H £23305 m 595%? HHoHHmHEomE oflmflwooo H260 9.31— 2:33am
Ho 3535 He 5&8 53on H983 8535 5&2 Ba 533 H<OHH<2HH so“
Hcofiomfiag 2E can 3.852 .580» .MZQE
m>0m BZBQ 8&8on 88m @8850 H5383 HHoHHHHHHwtO 850m «:5 329
ms 335

 

dome 82m 335v
m>0m Ba .59: 8% 33% :83 H8 mommoooa m5 wan—tomou 533% 05 nHHHo.H..H Bflfino 203 8% 33% 06 8:83 .Sac 38%
.Ho .HoHHHHHHmto 43% 33% 2: .Ho 0:: 39:05 038 SF A<mzmv £ng HoHoum 0:32 3233on “EH 5 68: 98>» m>©m 33.8w
2:. .35 H3 2362 How 85:3“ 533 wHfiHHomoHnoH 90E 833:5 EoEnszoowooo 88.8 3 wow: 38:8 8% 33mm .m ”Bah

25

Further Raster/E G VLayers Developmental Processes
( ‘reating Analysis Mask and Analysis Extent

Each species’ boolean layer (i.e., value of 1 representing species presence
locations and value of 0 representing locations with no data for species) was used to
create an analysis mask (define spatial area examined) and analysis extent for its species’
boolean layer and rasters representing its EGVs. This process ensured all rasters would
cover the same area, have matching spatial extents, and align for processing in
Biomapper software. KBB rasters included KBB boolean layer, elevation, texture and
drainage, percent sand in soils, IFMAP, tree canopy cover, and total precipitation. The
rasters for PT included PT boolean layer, proximity to Great Lakes shoreline, total
precipitation, IF MAP, slope, tree canopy cover, average daily maximum temperature,
average daily minimum temperature, and elevation. IB rasters included IB boolean layer,
elevation, 1F MAP, total precipitation, hydrologic features, proximity to hydrologic
features, wetlands, tree canopy cover, average daily maximum temperature, and average
daily minimum temperature.
C onverting ESRI Rasters to Idrisi Rasters

Biomapper software requires that all files be in Idrisi file format. This file format
is the same as Idrisi GIS software (Clark Lab, Worcester, MA) file format and consists of
a metadata file and data file (Hirzel 2004). All rasters for this project were ESRI rasters
and were converted to Idrisi file format using the GridConverter (Biomapper 3.1, Hirzel
et al. 2004). GridConverter is part of the Biomapper software package and allows for

conversion from ESRI rasters to Idrisi rasters and vice versa.

26

Converting Qualitative Layers into Quantitative Layers

Before Idrisi rasters could be used in the ENFA, each qualitative layer had to be
converted into quantitative layer (Hirzel 2006). An example of a qualitative layer would
be IFMAP, where land cover types are coded as integers (e. g., 1 represents low intensity
urban and 2 represents high intensity urban). IFMAP is considered a qualitative layer
because the data are not portrayed as ratio or continuous values, but instead are
represented by categorical or nominal values. The total precipitation layer is considered
quantitative because data are portrayed as continuous values. Qualitative data cause
problems during computations because average values are meaningless (Hirzel 2006).
Two steps were used to create quantitative layers from the qualitative layers and are
discussed below.

Boolean Layers

 

From the qualitative layers, boolean layers showing important TES habitat
features were created using Booleanisator (Biomapper 3.1, Hirzel et al. 2004). These
boolean layers were created to highlight important habitat features for the TES which
could subsequently be fiirther processed in Circular Analysis (CircAn; Biomapper 3.1,
Hirzel et al. 2004), which created quantitative values from the boolean values (Hirzel
2004) described later. Booleanisator and CircAn both come in the Biomapper software
package.

Qualitative layers for the KBB included soil texture and drainage and IF MAP
(i.e., land cover type). Boolean layers representing well-and moderately-drained soils,
herbaceous openlands, low density trees, oak, aspen, pine, and upland mixed forest were

created for the KBB. The qualitative layer for PT included [F MAP, and from it, a

27

boolean layer representing sand - soil was created. Qualitative layers for the 1B included
IFMAP, hydrologic features, uplands, and wetlands. One boolean layer from IFMAP
representing croplands (i.e., non-vegetated farmland, row crops, and forage crops) was
created. Booleanisator was not used to create boolean layers for hydrologic features,
uplands, and wetlands as they were already available as boolean layers (refer to Appendix
L for hydrologic features, below for uplands, Appendix K for wetlands).

In ArcView, a boolean layer representing uplands was created from [F MAP, as
the IF MAP land cover types selected to represent uplands (i.e., northern hardwoods, oak,
mixed upland deciduous, and upland mixed forests) were not numbered consecutively
and thus could not be processed by Booleanisator. To create the uplands boolean layer,
the IFMAP raster was reclassified such that 1 represented northern hardwoods, oak,
mixed upland deciduous, and upland mixed forests, and 0 represented all other land cover
types.

CircAn and Frequency ofOccurrence Percent Statistical Method

The final step for creating quantitative EGV layers involved creating continuous
values from the boolean layers (i.e., values 0 and 1). This was accomplished by
processing each boolean layer through CircAn using the frequency of occurrence percent
statistical method. Hirzel (2004, 2006) suggests using CircAn and the frequency of
occurrence percent statistical method when EGV layers represent resources used by the
species, thus the reason this method was selected to process these EGV layers. The
frequency of occurrence percent statistical method computes the frequency of occurrence
of number of pixels or area comprised by a particular habitat feature in a circle around the

focal cell (Hirzel et al. 2002, Hirzel 2004). For CircAn to process the boolean layers, a

28

circle radius had to be established. The radii for KBB and PT analysis were based on the
average spatial extent of mapped MNFI data, while data from other studies was used to
determine the radius for the 18.

The KBB and PT average spatial extents were calculated by determining the
average area occupied by all presence records whose last observation date was after 1990
within the species original presence file. The average area mapped by MFNI for KBB

presence locations occurring after 1990 was 1 12,890 m2 (1 1.289 ha). The radius for a
circle of this size was calculated and converted to a number of 30 m cells (n=6). The
precision estimate at this level is unknown. The frequency of well-and moderately-
drained soils, frequency of herbaceous openlands, frequency of low density trees,
frequency of oak, frequency of aspen, frequency of pine, and frequency of upland mixed
forests layers were calculated based on this area. The average area mapped by MFNI that
corresponded to PT presence locations whose last observation date occurred after 1990
was 240,010 m2, which corresponded to a radius cell count of 9. The frequency of sand -
Soil layer for PT was calculated based on this area.

The [B’s average spatial extent was not determined based on MFNI mapped
POlygons. The radius for the [B CircAn process was determined to be 1 km according to
Farmer et a1. (2002). Dividing the radius by 30 resulted in a radius of 33 cells
encompassing an area of 3,168,900 m2. This process created the frequency of uplands,

fl"'EECIuency of wetlands, frequency of hydrologic features, and frequency of croplands

laYers for the IB.

Ifiomapper Processes
7 '1‘: S and [50 V Layers
The ENFA in Biomapper software was used to create statewide HS layers for
each TES. Each TES was processed separately, but followed a similar process. Multiple
EGV layers were used for each TES (Table 6).
I’rocesses
As suggested in the Biomapper software manual (Hirzel 2004) and the Biomapper
Frequently Asked Questions (Hirzel 2006) the EGV layers were normalized. For each
TES, corresponding EGV layers were normalized by the box—cox bytes transformation
method. Next, each TES’s EGV layers were verified by examining EGV layers for errors
that could cause problems and cell value discrepancies (Hirzel 2004). An example of a
discrepancy is when one EGV layer has a cell value while another EGV layer for the
same cell has a “no data” value (e.g., 255 represents the background value “no data” for
a byte map) (Hirzel 2006). “No data” cells are not used in the ENFA and are removed
From processing. After viewing each TES’s discrepancy layer, I allowed the Biomapper
SOfiware to remove discrepancies from the analysis as most occurred along the edges of
Mi chigan, most likely caused by variations in shoreline interpretation.

The ENFA, similar to a Principal Component Analysis, creates uncorrelated
factors from input variables where the first factor represents marginality of the species
and the other factors represent species specialization (Hirzel et al. 2002; Hirzel and
A11 ettaz 2003a, b). Marginality looks at how much the mean for the species distribution
Cliffers from the mean for the entire study area (Hirzel et al. 2002, Hirzel 2006).

Marginality is calculated by comparing the mean cell values corresponding to known

30

Table 6. Ecogeographical variable (EGV) layers used during the Ecological Niche Factor
Analysis (ENFA) and corresponding species. Species include Karner blue butterfly
(KBB), Pitcher’s thistle (PT), and Indiana bat (1B).

 

 

EGV Layers Species
Frequency of Aspen KBB
Frequency of Croplands“ [B
Elevationb KBB, [B
Frequency of Herbaceous Qpenlands KBB
Frequency of Hydrologic FeaturesC [B
Frequency of Low Density Trees KBB
Average Daily Maximum Temperature PT, IB
Average Daily Minimum Temperature [B
Frequency of Oak KBB
Percent Sand in Soil KBB
Frequency of Pine KBB
Total Precipitation KBB, PT, [B
Proximity to Great Lakes PT
Proximity to Hydrologic FeaturesC [B
Frequency of Sand - Soil PT
Slope PT
Tree Canopy Cover KBB, PT, [B
Frequency of Upland Mixed Forest KBB
Frequency of Uplandsd [B
Frequency of Well and Moderately Drained Soils KBB
Regency of Wetlandsc IB

 

anon-vegetated farmland, row crop, and forage crop

belevation was represented by 8 categories ranging from 10-80 where ArcView separated
heights by natural breaks ranging between 141—602 m

Clakes, rivers, streams, creeks, ditches, drains, and ponds

dnorthern hardwood, oak, mixed upland deciduous, and upland mixed forest

elacustrine, palustrine, and riverine wetlands

31

species locations to the mean cell values for the entire study area (Hirzel et al. 2002,
Hirzel 2004). The larger the absolute value for the marginality coefficients the more the
species differs from mean study area conditions for that EGV (Hirzel et al. 2002, Hirzel
2004). [f the marginality coefficient is positive it indicates that the species prefers areas
with higher cell values for that EGV than the average location in the study area and if the
marginality coefficient is negative it indicates the species prefers areas with a lower cell
value for that EGV than the average location in the study area (Hirzel et al. 2002, Hirzel
2004). The error term is unknown. Specialization is determined by comparing the
variance of cell values associated with known species locations to the variance for the
entire study area (Hirzel et al. 2002, Hirzel 2004). The larger the absolute value for
specialization coefficients the more the species range is restricted by that EGV (Hirzel et
al. 2002). Global marginality, tolerance (inverse of specialization), and specialization
values were also calculated for each TES. For more details about how these calculations
were completed refer to Hirzel et al. (2002) and Hirzel (2006).

After the ENFA was computed for PT, a large eigenvalue warning was
encountered. All eigenvalues must be 3 0 in Biomapper and large eigenvalues could be
caused when two EGV layers are correlated (Hirzel 2004, 2006). The correlated EGV
layers were listed by Biomapper and in an attempt to eliminate this large eigenvalue
warning, the elevation and average daily minimum temperature EGV layers were
removed from the PT analysis, and the ENFA was recomputed. The elevation layer was
correlated with proximity to Great Lakes layer and removed because the distance PT is
located from the Great Lakes seemed more biologically important than elevation. The

proximity to the Great Lakes seemed more important because most PT grow on the dunes

32

along the shorelines of the Great Lakes (Higman and Penskar 1999, USFWS 2002).
Average daily minimum temperature was removed because it was correlated with
average daily maximum temperature. The average daily minimum and average daily
maximum temperature layers shared a similar spatial pattern, thus average daily
minimum temperature was selected to be removed.
Generating the HS Layers for the 774.3

The next Biomapper step involved computing HS layers for each TES. The
medians algorithm was used to compute these layers because it can perform well and
provide results in a timely manner (Hirzel and Arlettaz 2003a, b; Hirzel 2004). Details
about the median algorithm can be found in Hirzel et al. (2002), Hirzel and Arlettaz
(2003a, b), and Hirzel (2004). The number of factors (i.e., created by the ENFA) selected
to become factor maps for each TES were determined based on suggestions provided by
Biomapper, which uses Mac-Arthur’s broken stick distribution (Hirzel et al. 2002, Hirzel
and Arlettaz 2003a, Hirzel 2006). Factor maps (i.e., . .a map that summarize all
ecogeographical variables according to the score matrix computed by the ENFA.” (Hirzel
2004)) were evaluated with the median algorithm to create individual HS layers for each
TES. The final HS layer values range from 0-100 where 0 represents less suitable habitat
and 100 represents more suitable habitat.
Creation of Two Example HS Layers per Species

Two HS layers were created for each species. One layer represented a
conservative reclassification scenario and the other layer represented a liberal

reclassification scenario. Under the conservative reclassification scenario any potentially

suitable habitat for the TES was identified, while under the liberal reclassification
scenario only the most suitable habitat for the TES was identified.

The conservative and liberal HS layers for each species were created in ArcMap
through the reclassification process. The conservative HS layers for each TES were
reclassified such that HS values of 0 represented unsuitable habitat and HS values from
1-100 represented suitable habitats. The liberal HS layers for each TES were reclassified
such that HS values from 0-99 represented unsuitable habitats and HS values of 100
represented suitable habitat. These reclassification scenarios were used during the
integration process described later. Emphasis needs to be placed on the fact that these
reclassification scenarios were created simply to show how the data could be used and
further work is needed to determine the most suitable reclassification scenario for each
species.

Biomapper Cross-validation

Cross—validation (CV) is important because it allows the user to determine the
usefiJlness or the predictive power of the model output. The default k—fold CV procedure
in Biomapper was used to evaluate the predictive power of the HS layers for each TES
(Hirzel 2004). The CV procedure was conducted twice for each TES HS layer once to
reflect the conservative scenario and once to reflect the liberal scenario. In the CV
procedure KBB and PT species known location data were partitioned into 10 equal-sized
subsets, where 9 subsets were used to calibrate an HS layer and the last was used for
validation. The [B species known location data were partitioned into 5 equal-sized
subsets, where 4 subsets were used to calibrate an HS layer and the last was used for

validation. This process was repeated 10 times for KBB and PT and 5 times for TB, each

34

time with a different subset being held out. Resultant CV HS layers were each

partitioned into 2 bins. where each bin covered a proportion of the maps’ area (A,) and

contained a proportion of the validation cells (Ni) (Hirzel 2004). Two bins were selected

to reflect suitable and unsuitable habitat. The bin boundary was set at HS 1 for the CV
processes reflecting the conservative reclassification scenario, while the bin boundary
was set at HS 100 for the CV processes reflecting the liberal reclassification scenario.

From this process, area-adjusted frequencies were calculated for each bin using the

equation F ,—= N,-/A,, where if all bins have a F, value of 1 it indicates a random HS map

(Hirzel 2004). Models with good predictive power have F ,-<1 for low HS and F ,->l for

high HS (Hirzel and Arlettaz 2()()3a, Hirzel 2004). More details on the CV procedure can
be found in Boyce et al. (2002). Hirzel and Arlettaz (2003a), Reutter et al. (2003), and
Hirzel (2004).

Before area-adjusted frequency calculations were used to evaluate the models, the
predictive power of these models could be evaluated using the absolute validation index
(AVI) and contrast validation index (CV1) found in Biomapper (Hirzel 2007) (also
calculated during CV procedure). The AVI indicates how well the model performs by
determining if TES presence locations are associated with high HS values, while the CVI
indicates if the model is predicting better than chance (Hirzel 2007). The AVI score was
calculated by computing the proportion of validation raster cells with HS values greater
than 50 for each of the 10 partitions (Hirzel and Arlettaz 2003b; Hirzel 2006, 2007). The
CVI scores were also computed for each partition, and the CV1 values were obtained by
subtracting from AVI the proportion of all raster cells with a HS score greater than 50

(Hirzel and Arlettaz 2003b; Hirzel 2006, 2007).

b.)
U1

Making the HS Layers for each TES Useable in Are View or ArcMap

For the HS layer for each TES to be useable for further analysis in GIS software,
each layer was individually converted from Idrisi file format to an ESRI raster using
DIVA-GIS 5.2 software (Hijmans et al. No Date). Each TES HS layer was imported as
an Idrisi file into DIVA-GIS and exported as an ESRI ASCII raster. In ArcView, the
ESRI ASCII rasters were imported as an ASCII file with integer cell values.
Integration of TES HS Layers and TCBs
Creation of Pesticide Drift Layer

A pesticide drift layer was created to demonstrate how tart cherry growers could
be affecting potential TES suitable habitat near their orchards. Pesticide drift distance
was based on pesticide drift studies conducted by the Spray Drift Task Force (1997),
which measured pesticide drift up to 549 m (1800 ft) outside an orchard. The Spray Drift
Task Force (1997) found that the majority of pesticide drift deposited within 183 m (600
ft). In ArcMap, a 210 m pesticide buffer (i.e., pesticide drift layer) was created around
the perimeter of each TCB. A 210 in pesticide drift distance was used in place of 183 m
to accommodate the 30 m resolution of TES HS layers. The pesticide drift layer created
did not consider wind direction or pesticide application equipment used. Rather the
intent was to provide a snapshot in time and a starting point for this process.
Data Integration Process

The data integration process was used to produce data layers (i.e., one each for
KBB, PT, and [B) for identifying which tart cherry growers had TES potentially suitable
habitat within 210 m of their TCB. In ArcMap each TES HS layer was projected to the

Michigan Georef projection and analysis mask and analysis extent was set to the TCB

36

pesticide drift layer. This step reduced the amount of data portrayed for each TES HS
layer to only show potential habitat within the pesticide drift layer and ensured that the
raster cells aligned. During data integration, each TES HS layer was reclassified. As
stated previously for this project, the example HS layers for each TES were reclassified
twice, once conservatively and once liberally (see section on Creation of Two Example
HS Layers per Species above). Reclassified TES HS potential habitat areas were
intersected with the TCB pesticide drift layer and boundaries were dissolved by TCB
name. Again, these HS reclassification scenarios are just examples of the outputs that
can be derived from the process. Selecting the correct HS reclassification scenario for
each species is really important and should be refined in the future based on CV results.
However, these HS layers do not represent a failed product rather that more work is
needed to refine the proper HS reclassification scenario for each species.

This process resulted in three TES layers (i.e., one for each TES) for each
reclassification scenario and a table identifying growers within 210 m of the TES
potential habitat. Per reclassification scenario, the three TES layers were condensed into
one table to simplify the output. In each condensed table, three new fields were added to
represent KBB, PT, and TB potentially suitable habitat. The original TCB and each
TES’s attribute tables were joined by the TCB name field. All TCBs found to contain a
value other than null were selected and given the value 1. A 1 indicates a particular TCB
was identified to be within 210 m of TES suitable habitat. The remaining null values
were set to equal 0, identifying no TES suitable habitat was within 210 m of the TCB.
This process produced two final tables, one for each reclassification scenario, indicating

which TCBs are within 210 m of TES potential habitat,

RESULTS

ESC WG

The ESC WG consisted of 17 members and met 4 times over the course ofthis
study (March 2005, July 2005, February 2006, and September 2006). The first ESC WG
meeting allowed participants to discuss the USEPA pesticide EUP denial previously
described, voice their level of support for the project, and talk about concerns for
obtaining spatial location data. At the second meeting, discussions occurred about the
results that would be distributed to TCO growers (e.g., map, list of potentially affected
TES) from this project and the lack of statewide surveys for TES. The working group
also discussed how to obtain TCO spatial data and whether both federal and state listed
species should be investigated. It was decided to focus on federally listed species.
Topics discussed at the third meeting included initial modeling effort results for KBB,
collection of additional TCB spatial location data, priority of TES investigated for the
pilot study, and landowner’s sensitivity toward the use of predictive models. The fourth
meeting covered items related to modeling effort results for KBB and PT. In addition, it
was decided to use a generic modeling approach to simulate pesticide drift rather than a
site specific pesticide drift model. In place of the site specific pesticide drift model, the
ESC WG decided that a layer representing the distance pesticide drifts beyond TCBs
would be created as a first step to identify locations where potential interactions could
occur. The members also agreed that as the project progresses, a site specific pesticide
drift model (e.g., based on grower application equipment, methods, and weather

conditions) could be developed to more accurately portray pesticides drifting off site.

38

Michigan Tart Cherry Spatial Data

Spatial data for 213 TCBs were collected representing 97 different tart cherry
growers in Michigan (Table 7). The greatest number of TCB spatial data coordinates
were collected from Berrien County (n = 56), while the smallest number was from Ionia,
Manistee, and Muskegon Counties each having 1 (Figure 3). The TCB spatial data
collected accounted for 5.91% ofall tart cherry hectares occurring in Michigan based on
2003 tart cherry total hectares.
Michigan TES Spatial Data

I examined three federally listed TES, the endangered KBB and [B and the
threatened PT. With the data available from MNFI, the KBB had the greatest number of
documented presence locations throughout Michigan while the IB had the fewest (Table
8). The number of cells represented presence locations after each species layer was
converted to raster. Some ofthe PT and IB polygon presence locations may have been
point presence locations that were buffered by MNFI, thus not all locations within the
buffer are a “true” presence location, but each cell within these polygons was used as a
presence location after it was converted to raster. The PT had the greatest number of
cells throughout Michigan, while the IB had the fewest (Table 8). The PT likely had
more presence location cells than the KBB because the PT presence locations tended to
encompass larger areas.
Biomapper Processes
EGVI.ayers

Twenty-one EGV layers were created. The KBB required 1 1 EGV layers, while

PT and IB required 6 and 10, respectively (Table 6).

39

Table 7. Total number of tart cherry blocks (TCB) with total, mean, maximum, and
minimum hectares (ha) for all TCBs included in this study.

 

% of all Mean

 

Number of Total TCB TCB ha in TCB Size Maximum Minimum
TCBs Hectares (ha) MT (ha) TCB Size (ha) TCB Size (lij_)__
213 765.]0 5.91 3.59 27.44 0.14

_—

40

 

 

 

 

 

 

 

 

F. p @QOQ
>90 A».
m e o A
M 00 O
e0 ,mveo
. 9w» s»
M Q
. ,va

6?

.

 

 

 

 

 

[ll] [:1, ,1: H3353 [l .mlmflg,

 

50 '~

m0... 30:52

10
0

County

 

 

Fig. 3. Number of tart cherry blocks (TCB) by county considered in this study.

41

Table 8. Number of presence locations (i.e., polygons, points) used for Karner blue
butterfly (KBB), Pitcher’s thistle (PT), and Indiana bat (IB). The mean, maximum, and
minimum areas for presence locations whose last observation date occurred after 1990
are also listed. Total number of cells represents presence locations after each species
layer was converted to raster.

 

 

Total
Presence Total Mean Max. Min. Total
Locations Presence Presence Presence Presence Presence
(#polygons, Location Location Location Location Location
Spp. # points) Size (ha) Size (ha) Size (ha) Size (ha) Cells
KBB 175,0 1,975.54 11.29 120.53 0.01 21,907
PT 115, 0 2,760.06 24.00 643.99 0.13 28,722
IB 6, 18 —* — — —— 14,180

 

* indicate lacking area measurements because also contained point data

42

Verification

Verification is done to identify EGV layers that could be problematic during
analysis procedures. Aspects of verification included identifying problematic EGV
layers and identifying discrepancies between EGV layers. Problematic layers included
those that did not contain a wide range of values and, as a result, resembled a boolean
map surface (i.e., 0 and 1 values) (Hirzel 2004, 2006). Problematic layers were not found
for the KBB. The PT verification process identified the sand - soil layer as “not
continuous enough,” and the 1B verification process identified the proximity to
hydrologic features layer as “not continuous enough.” While problematic, I continued to
use these layers as per Hirzel‘s (2006) instructions where the first option is to continue to
include them through the ENFA process.

The EGV layers for KBB produced 1,372,366 discrepancies (0.3%), while PT and
IB had 17,201,514 (3.4%) and 30,745,825 (6.0%) discrepancies, respectively. The
discrepancies for all three TES EGV layers occurred primarily along the edges of
Michigan’s state boundary. For the KBB EGV layers the percent sand in soil and well-
and moderately-drained soils led to the majority of discrepancies along the interior
boundary of the state. For PT and IB EGV layers, a large portion of the discrepancies
were likely caused by the elevation layer that had a 3.2 km buffer around the perimeter of
the state. This would cause a problem because the buffer exceeded the spatial extent of
other EGV layers, making those overlapping areas incompatible. The elevation layer did
not cause many discrepancies for the KBB EGV layers because the buffer, which came
with the original layer, had been removed. The reason the buffer from the original layer

was not removed from the elevation layer for the [B and PT is not known, but may have

43

been due to a different GIS process being used. The frequency of wetlands layer may
have also contributed to the large number of [B EGV layer discrepancies as it also
exceeded the spatial extent of the other EGV layers. Afler examining discrepancy layers,
I allowed the Biomapper software to disregard cells with discrepancies.

ENFA

The ENF A produced 1 1 factors for the KBB. The first factor explains marginality
and a portion of specialization (i.e., 20%), while the remaining factors explain more of
the specialization (Table 9). The marginality coefficients indicated that KBB locations
were associated with higher cell values on frequency of oak, percent sand in soils, total
precipitation, frequency of herbaceous openlands, frequency of low density trees,
frequency of well—to moderately-drained soils, frequency of upland mixed forests,
frequency of pine, and frequency of aspen than the average location in Michigan (Table
9). KBB locations were also associated with lower elevations than the average location
in Michigan (Table 9). KBB locations displayed no difference from average locations in
Michigan for tree canopy cover (Table 9). Frequency of well-and moderately-drained
soils in Specialization 1 and total precipitation in Specialization 2 are important factors
for specialization (Table 9).

A large eigenvalue was encountered after performing the ENF A for PT and can
be caused by correlated EGV layers (Hirzel 2006). Two pairs of EGV layers were found
to be correlated: elevation and proximity to Great Lakes, average daily minimum
temperature and average daily maximum temperature. The elevation and average daily
minimum temperature EGV layers were removed from the analysis, but the large

eigenvalue persisted. Large eigenvalues are acceptable when combining multiple factors

44

 

 

 

00.0 :0 00.0- 00.0 _00- 00.0- 00 .0- 00.0 :38 0025 8:-
000 00.0- 00.0 3.0 2.0. 00.0 00.0- 20 500-4 0o 08802.0
_00 00.0 00.0 00.0- 00.0- 0 _ .0 00.0- 00.0 050 00 0050020
0.0.0- _00- 00.0- 00.0 _ _0 0 _0- 0.0.0- 0 0 .0 000200 0052 0520: 00 005000:
00.0 40.0 0.0- 04.0- 00.0 00.0- 00.0- 00.0 0000 02:05 .002 05 :03 00 080500: .4:
00.0- 00.0 00.0- :0 0.0- 00.0- 00.0- 00.0- 800.6:
0 0 .0- 00.0 0 0 .0- 0 _ .0 00.0 00.0- 00.0 00.0 0020 000800 33 0o 005:3:
00.0- 0 _ .0 0.0.0- 00.0 0 0 .0 00.0 00.0- 00.0 005.500 08800500 00 0050020
0 _w .0 00.0- 0 0 .0- 00.0 00.0- 00.0- 00.0 00.0 8020320 020.0
0 _ .0- 00.0 00.0 00.0 000- 00.0 2.0 00.0 :00 E 0000 60200
00.0 0 0 .0- 00.0- 00.0- 00.0- 0 0 .0 00.0- 00.0 sec .00 080:3:

00mm 20 80 $60
003 0.0.3 00000 0.000 0.00: 0.0.0 : 30000 0.0000 eoeasaeam

0 .00on e 0on m .oon .0. .ooam m .qum N doom _ :0on 00:05?va ”0583 .oom
000600 0200:?

_moEqflwoowOQQ of 00% 0056508 20230030on new $29» 820020 mammmoeooe 3 e853 88$ 03:00:92: 05 no 0056508
20305 002? 2an 605398 00230010620ch .5ch 05 28 An swoon: _ 5:00:06on .bmfifiwaev 00820.0 Huge—o8 C 0 co
Soc w 00:.“ 2t 8039: mwfiemo: 5:300 .Ammvc 05.50.05 0:3 550M 08 £38 04000025 00030904 .5me 0:22 Ezwofiom .o 280-0

(Yu No Date), thus the analysis was continued with a large eigenvalue. Six factors were
produced after the ENFA for PT. The marginality coefficients imply PT locations were
associated with higher values on frequency of sand - soil and slope than the average
location in Michigan (Table 10). PT locations were closer to the Great Lakes, had lower
average daily maximum temperatures, received less total precipitation, and had less tree
canopy cover than the average Michigan location (Table 10). The majority of the
specialization (i.e., 99%) is explained by the marginality factor for PT (Table 10).

The ENFA produced 10 factors for the IB. Marginality coefficients showed that
[B locations were associated with higher cell values of total precipitation, frequency of
croplands, frequency of hydrologic features, average daily maximum and average daily
minimum temperatures, and elevation than the average location in Michigan (Table l 1).
IB locations are also closer to hydrologic features and were associated with lower cell
values for frequency of uplands and tree canopy cover than the average location in
Michigan (Table 11). The [B displayed little difference from average locations in
Michigan for frequency of wetlands. Proximity to hydrologic features in Specialization 1
and average daily minimum temperature in Specialization 2 were driving factors for
specialization (Table 11).

The ENFA produced global marginality values greater than 1 for the KBB
(1.441), PT (2.128), and IB (1.528) indicating that their habitats associated with known
locations differed from the average conditions available in Michigan (Hirzel et al. 2002,
Hirzel 2006). The ENFA also produced global tolerance and global specialization
(inverse of global tolerance) values closer to 1 for the KBB (T20617, S=l.621)

suggesting it is specialized and can live in a wide range of habitat conditions (Hirzel

46

Table 10. Ecological Niche Factor Analysis (ENFA) results for Pitcher’s thistle (PT).

Column headings indicate the first 4 (out of 6) ecological factors (marginality,

specialization 1-3) and the percent of specialization explained. Table values include

coefficients on the marginality factor (sorted by decreasing absolute value) and

specialization coefficients for the ecogeographical variables (EGVs).

 

ECO. FactOFSI Marginality Spec. 1 Spec. 2 Spec. 3
Specialization: (99%) (0%) (0%) (0%)
EGVs for the PT
Frequency ofSand — Soil 0.83 -0.07 0.05 0.24
Proximity to Great Lakes -0.46 -0.43 0.39 0.66
Average Daily Maximum Temp. -0.23 0.81 0.12 -0.28
Total Precipitation -0.14 -0.38 -O.82 0.10
Slope 0.13 -0.09 0.40 -0. 12
Tree Canopy Cover -0. 10 -0.06 0.06 -O.64

 

47

Table 1 1. Ecological Niche Factor Analysis (ENFA) results for the Indiana bat (IB).
Column headings indicate the first 3 (out of 10) ecological factors (marginality,
specialization 1-2) and the percent of specialization explained. Table values include
coefficients on the marginality factor (sorted by decreasing absolute value) and
specialization coefficients for the ecogeographical variables (EGVs).

 

500- Factors: Marginality Spec. 1 Spec. 2
Specialization: (67%) (16%) (10%)
EGVs for the [B __
Total Precipitation 0.59 0.06 0.14
Average Daily Maximum Temperatures 0.41 0.01 -0.47
Average Daily Minimum Temperatures 0.39 -0.28 0.72
Frequency of Croplands 0.39 -0.10 -0.36
Frequency of Hydrologic Features 0.22 0.00 -0.01
Frequency of Uplands -0.22 0.00 -0.05
Tree Canopy Cover -0.22 0.00 0.00
Elevation 0.13 0.00 -O.32
Proximity to Hydrologic Features -0.1 l -0.95 -0.08
Frequency ofWetIands 0.04 0.00 -0.05

 

48

2006). Global tolerance values range from 0 to 1 (with 0 representing a more specialized
species using a narrow range of habitat conditions and 1 representing a species that can
live in a wider range of habitat conditions) and global specialization (inverse of global
tolerance) values range from 1 to infinity (Hirzel 2006). Global tolerance values for PT
(0.073) and IB (0.035) were near 0 and specialization values for PT (13.750) and IB
(28.858) were greater than 1 suggesting they are more specialized and live in a more
narrow range of habitat conditions (Hirzel et al. 2002, Hirzel 2006).

T ES HS Layers

Based on Mac-Arthur‘s broken stick distribution (Hirzel et al. 2002; Hirzel and
Arlettaz 2003 a, b; Hirzel 2006) eight factor maps were created and used to compute the
final HS layer for the KBB, four factor maps were created for PT and used to compute
the final lHS layer for PT; and three factor maps were created to compute the final HS
layer for IB. lEach derived HS map for the three TES ranged in values from 0-100, where
0 represents less suitable habitat and 100 signifies more suitable habitat.

A majority of the more suitable habitat for the KBB occurs in the west central part
of Michigan’s Lower Peninsula, while the Upper Peninsula and eastern Lower Peninsula
tend to contain the less suitable habitat (Figure 4). The HS map for PT identifies the
majority of more suitable habitat occurring along the coasts of Michigan, although a
small amount of the more suitable habitat does occur within the interior of Michigan
(Figure 5). The majority of less suitable habitat for the PT occurs in the interior (Figure
5). The HS map for the [B shows the bulk of more suitable habitat occurring in the
southern part of the Lower Peninsula (Figure 6). The Upper Peninsula and northern

Lower Peninsula contain the majority of the less suitable habitat for the [B (Figure 6).

49

 

KBB
HS Value

 

 

0 15 50 100 Kilometers
l 1 1 l 1 l

1 I l

 

Fig. 4. Habitat-suitability (HS) map for Karner blue butterfly (KBB). HS values range
from 0 (i.e., less suitable, lighter areas) to 100 (i.e., more suitable, darker areas). The
gray outline represents the Michigan state boundary.

50

 

 

   
 

PT
HS Value

”High : 100
Low : 0

[0 25 50 100 Kilometers

 

 

| r I r l iAnJ
Fig. 5. Habitat-suitability (HS) map for Pitcher’s thistle (PT). HS values range from 0
(i.e., less suitable, lighter areas) to 100 (i.e., more suitable, darker areas). The gray

outline represents the Michigan state boundary.

5]

 

IB
HS Value _ 7». J
”High : 100 . _--' '

 

Low : 0
N
+ 9 l?-
0 25 50 100 Kilometers
- ‘ -

 

 

 

Fig. 6. Habitat-suitability (HS) map for Indiana bat (IB). HS values range from 0 (i.e.,
less suitable, lighter areas) to 100 (i.e., more suitable, darker areas). The gray outline
represents the Michigan state boundary.

52

('l"

The k-fold CV results for the KBB generated a mean AVI of0.80454 (SD
0.00861) and a mean (‘Vl 01075321 (SD 0.00856). Useful models should have an AVI
score >075 while models not trustworthy have AVI scores <0.50 (Hirzel 2007). A useful
model should also have a CV1 score >030 (Hirzel 2007) while models with a contrast
index of 0 indicates performance comparable to random (Hirzel and Arlettaz 2003b).
The CV results for PT and 18 indicated that the models performed acceptably as PT had a
mean AVI of0.658 (SD 0.012) and a mean CV1 of 0.64004 (SD 0.01248) and the IB had
a mean AVI of0.64267 (SD 0.01562) and a mean CVI of0.60837 (SD 0.01562).

The area-adjusted frequency graph for the conservatively reclassified KBB layer
included data for only one bin, even though 2 bins had been specified (Figure 7). The
summary ofCV results for conservatively reclassified KBB displayed data for 2 bins
where the mean area-adjusted fi‘equency was listed as 0 (SD 0) for bin 1, and for bin 2,
the mean area-adjusted frequency value was 1 (SD 0). Based on the area-adjusted
frequency graph all of the partitioned area and species validation points are likely
included in potentially suitable habitat based on this reclassification scenario. When the
HS bin boundary was placed at HS 2, the area-adjusted frequency graph displayed two
bins with bin 1 having area-adjusted frequency values of 0 and bin 2 having area-adjusted
frequency values of 1. These outputs matched the area-adjusted frequencies listed in the
summary of CV results. When the bin boundary was placed at HS 20 the area—adjusted
frequencies for bin 1 were below 1 and for bin 2 above 1. This suggests a bin boundary
at HS 20 could be a better model because bin 1 had an area-adjusted frequency less than]

and bin 2 had an area-adjusted frequency greater than 1. These results suggest that an

53

 

 

 

 

 

 

 

Area-adjusted frequency

 

 

0 10 20 30 40 50 60 70 80 90 100
Habitat suitability bins

Fig. 7. Area-adjusted frequency graph for the conservatively reclassified KBB. The
horizontal dashed line at 1 on the y-axis represents the random frequency line. The area-
adjusted frequency values are along the y-axis. The dot indicates partition values where
the first set of dots represents the results for bin 1 and the second set of dots represents
the results for bin 2. This graph only shows dots representing results for 1 bin’s
partitions.

54

acceptable model could be achieved by using a different reclassification scenario.

Based on the mean area-adjusted frequencies for bins 1 and 2 of the liberally
reclassified KBB layer (Table 12), the values suggested this was not a useful model.
Upon examination of the summary of CV results and area-adjusted frequency graph
(Figure 8), the values in bin 1 were close to 1 and the majority of values in bin 2 were
closer to 0 (i.e., only one number was above 1), and bin 2 also had a very high standard
deviation, all ofwhich suggest it was not a usefiil model. The area-adjusted frequency
values listed in the summary of CV results did not match the values on the area-adjusted
frequency graph. The reason for this is unknown, but is thought to be a Biomapper error.
With such an extreme HS bin boundary, the line continues off the graph and may cause
problems when the data is being recorded. When all lines end on the graph the values
match those recorded in the summary of CV results. It is unknown how this error may
have affected other CV results, but the AVI and CVI values were the same for both the
conservative and liberal scenarios.

Based on the mean area-adjusted frequencies for bins 1 and 2 of the
conservatively reclassified PT layer (Table 12), this was a usefirl model. This is also true
for the liberally reclassified PT layer, conservatively reclassified IB layer, and liberally
reclassified IB layer (Table 12). For the conservatively and liberally reclassified PT and
1B area-adjusted frequency graphs refer to Figures 9 and 10, respectively. Again, like
with the liberally reclassified KBB layer, the liberally reclassified PT and IB layers each
had area-adjusted frequency values listed in the summary of CV results that did not
match the values on their area-adjusted frequency graphs. The reason for this is thought

to be an error in the Biomapper software when the values were recorded in the summary

55

Table 12. Mean area-adjusted frequency values resulting from the cross-validation (CV)
process. Table values represent the mean area-adjusted frequencies for bins 1 and 2 for
each species reclassification scenario, while the values in parenthesis represent the

standard deviation.

 

Species and Reclass. Scenario Bin 1 Bin 2 _
Conservative KBB * 1.00 (0.00)

Liberal KBB 0.99982 (0.00058) 66.59 (210.59)
Conservative PT 0.00019 (0.00024) 4.62 (0.0009)
Liberal PT 0.57626 (0.01417) 85.98 (2.8418)
Conservative IB 0.00000 (0.00) 2.20 (0.005)

Liberal [B 036617 (0015504) 51.54 (1.24)

 

*No bin was shown on the area-adjusted frequency graph

56

 

 

 

 

/
600 //,
>5 500 /
0
fl ' .
o .
3.). 400 1
a: .
3 300 /- .
U) -
.2, /
‘3 200 /’
<3
E . /
100 i /
0....- ........ . ...... ‘ -....... . ....
0 10 20 30 40 50 60 70 so 90 100
Habitat suitability bins

Fig. 8. Area-adjusted frequency graph for the liberally reclassified KBB. The 1 on the y-
axis represents the random frequency line. Area-adjusted frequency values are indicated
on the y-axis. The dots indicate partition values where the first set of dots represents the
results for bin 1 and the second set of dots (not shown, but close to where the solid lines
run off the graph at <1 and 650) represents the results for bin 2.

57

(a)

 

Area-adjusted frequency

 

 

 

0 10 20 30 40 50 60 70 80 90 100
Habitat suitability bins

(b)

 

 

Area-adjusted frequency
A
O

 

 

 

0 10 20 30 40 50 60 70 80 90 100
Habitat suitability bins

Fig. 9. Area—adjusted frequency graphs for PT where a) represents the conservative
reclassification scenario and b) represents the liberal reclassification scenario. The I on
the y-axis represents the random frequency line. Area-adjusted frequency values are
indicated on the y-axis. The dots indicate partition values where the first set of dots
represents the results for bin 1 and the second set of dots represents the results for bin 2.

58

(a)

 

 

 

 

 

 

 

 

 

 

 

i
2
>~.
g
“:3
U‘
3:3
__ .....................................................................
"5
N
o . . . .
o 10 20 30 4o 50 60 70 so 90 100
Habitat suitability bins
(b)
50
>.. 40
g.
“0’
0'
.E 30
'8
,3 7 .
.2, 20 _ 7 , ,
a; . 7 . .
510 ' - . ' /
0 ........ . ' . 3
o 10 20 30 so 60 70 so 90 100

40
Habitat suitability bins

Fig. 10. Area-adjusted frequency graphs for [B where a) represents the conservative
reclassification scenario and b) represents the liberal reclassification scenario. The 1 on
the y-axis represents the random frequency line. Area-adjusted frequency values are
indicated on the y-axis. The dots indicate partition values where the first set of dots
represents the results for bin 1 and the second set of dots represents the results for bin 2.

59

ofCV results they were placed in the wrong location.

At the time the reclassification scenarios were selected the CV process had not
been performed on any of the TES models, therefore, it was unknown that the KBB
models were not useful models. The best set of resulting maps for the KBB were not
determined for this portion of the project, but can be completed in the future. Again, the
focus ofthis portion ofthe pilot project was on creating the process. In the future the CV
process should be completed after the species HS layer has been created to determine if
the model is useful and to select the best reclassification scenario.

Integration of TES HS Layers, TCBs, and Pesticide Drift Layer

The conservatively reclassified KBB HS map (i.e., HS values of 0 represent
unsuitable habitat and HS values from 1-100 represent suitable habitat) shows the
majority ofthe state as suitable habitat (Figure 1 la). The liberally reclassified KBB HS
map (i.e., HS values from 0-99 represent unsuitable habitat and HS values of 100
represent suitable habitat) shows suitable habitat occurring in the west central Lower
Peninsula (Figure 1. 1b). The conservatively reclassified PT HS map (i.e., HS values of 0
represent unsuitable habitat and HS values from 1-100 represent suitable habitat) shows
high concentrations of suitable habitat occurring along the shorelines in Michigan (Figure
12a). There is some suitable habitat in the interior portions of the state also (Figure 123).
The liberally reclassified PT HS map (i.e., HS values from 0-99 represent unsuitable
habitat and. HS values of 100 represent suitable habitat) shows the majority of the state as
being unsuitable with suitable habitat occurring along the coastlines (Figure 12b). The
conservatively reclassified IB HS map (i.e., HS values of 0 represent unsuitable habitat

and HS values from 1-100 represent suitable habitat) shows most of the suitable habitat

60

(a) (b)

    

KBB KBB . .
llnblm Habitat i

Unsuitable Unsuitable 3
- Suitable -Sultable l

4

u l— LII—am n I II I'M

 

7+
1 .... . 10000.0.

Fig. 11. Reclassified habitat-suitability (HS) maps for Karner blue butterfly (KBB),
where (a) represents the conservative reclassification scenario and (b) represents the
liberal reclassification scenario. In both maps unsuitable habitat is white and suitable
habitat is black. In the conservative HS map (a) unsuitable habitat represents HS values
of 0 and suitable habitat represents HS values from 1-100. In the liberal HS map (b)
unsuitable habitat represents HS values from 0-99 and suitable habitat represents HS
values of 100. The gray outline represents the Michigan state boundary.

 

61

(b)

   

PT

Habitat
Unsuitable

- Suitable

PT

Habitat
Unsuitable

- Suitable

,L

I u s Int—m

 

Fig. 12. Reclassified habitat-suitability (HS) maps for Pitcher’s thistle (PT), where (a)
represents the conservative reclassification scenario and (b) represents the liberal
reclassification scenario. In both maps unsuitable habitat is white and suitable habitat is
black. In the conservative HS map (a) unsuitable habitat represents HS values of 0 and
suitable habitat represents HS values from 1-100. In the liberal HS map (b) unsuitable
habitat represents HS values from 0-99 and suitable habitat represents HS values of 100.
The gray outline represents the Michigan state boundary.

62

occurring in the southern Lower Peninsula (Figure 13a). The liberally reclassified [B HS
map (i.e., HS values from 0-99 represent unsuitable habitat and HS values of 100
represent suitable habitat) shows suitable habitat occurring in the south central Lower
Peninsula (Figure 13b).

When the conservatively reclassified TES HS layers were intersected with the
TCBs and the pesticide drift layer, all TCBs were found to be within 210 m (i.e.,
pesticide drift distance) of potential habitat for at least one TES. One-hundred and
twenty seven TCBs (60%) were within 210 m of PT, 201 TCBs (94%) were within 210 m
of KBB, and 97 TCBs (46%) were within 210 m ole. Ofthe 213 TCBs, 21 TCBs
(9.9%) intersected only one TES potential habitat, 172 TCBs (80.8%) intersected 2 TES
potential habitats, and 20 TCBs (9.4%) intersected all 3 TBS potential habitats.

When the liberally reclassified TES HS layers were intersected with the TCBs and
the pesticide drifi layer, 5 TCBs representing 4 tart cherry growers, were found to be
within 210 m (i.e., pesticide drifl distance) of at least one TES. The KBB suitable habitat
was the only TES intersected. The results pertaining to the KBB under both
reclassification scenarios should be taken with caution as these were considered to be not
useful models by the CV results.

The main focus for this project was developing the process. The process
developed does work and portrays a simple interaction occurring. In the fiJture work can
be completed (site specific pesticide drifi model, useful TES HS models) to enhance the

realism for this process.

63

  
   

.3
1 "I fi|
l [B ....- ~ . . . ‘ m
1 . ~ . . |.- . 7‘ t
, Habitat “ ‘ 7* _ , Habitat _
l Unsuitable ’ ‘ . i i : i Unsuitable ‘9?
, -Suitable .. _,‘_ I 1 -Suitlbl¢ ._.r l
, i ; ‘ ' ‘
V . . . ,11'9‘ v, . | a z
I i. x - .M ‘

o u :- lain—an t
‘ “a" JJ 1

Fig. 13. Reclassified habitat-suitability (HS) maps for Indiana bat (IB), where (a)
represents the conservative reclassification scenario and (b) represents the liberal
reclassification scenario. In both maps unsuitable habitat is white and suitable habitat is
black. In the conservative HS map (a) unsuitable habitat represents HS values of 0 and
suitable habitat represents HS values from 1-100. In the liberal HS map (b) unsuitable
habitat represents HS values from 0-99 and suitable habitat represents HS values of 100.
The gray outline represents the Michigan state boundary.

 

64

DISCUSSION

Developing the Process to Unite Groups and Identify Potential Areas of Overlap
ESC WG

The first step in developing a process for uniting TES conservation and
commodity production is to involve individuals and groups having concerns or
knowledge pertaining to the system under study. This step is important as it brings
together groups having an investment in the product and allows them the opportunity to
express ideas, fears, hopes, and other comments as the project develops. For this project,
this was accomplished by the ESC WG.

A variety of groups (Table 2) were involved with development of this process and
helped address issues as they arose throughout the project. One important topic
addressed by the ESC WG members included fears about maintaining the privacy of tart
cherry grower’s information (e.g., names, addresses). As a result of this discussion, steps
were taken to protect the tart cherry grower’s identities. This was accomplished by
employing NASS employees (through CIAB) to collect and record TCB spatial data.
CIAB subsequently removed grower names and addresses from the spatial data sheets
before sending them to MSU.

Another theme discussed by ESC WG members was the potential reaction of tart
cherry growers to the use of predictive HS maps to represent potential habitat that might
support a TES location. Some ESC WG members were concerned about how tart cherry
growers would respond because predictive maps do not necessarily represent the actual

presence location of a TES, but instead specify potentially suitable habitat for a species in

65

particular locations. While this is a valid concern, using current TES spatial data was not
possible because a statewide survey for TES does not exist. If the TES presence spatial
data had been used, less the modeling effort, results from the project would have
misrepresented potential and highly probable areas of overlap between TES and
commodities. Potential and highly probable areas of overlap may not have been
represented or shown in the resulting maps because the current data does not represent a
complete survey for all TES occurrences. An important part of this process was to help
commodity producers identify those areas of potential TES concern and as such, the HS
maps offer a first approximation that should be verified with field surveys. Aerial photos
could also be used to determine if the landscape surrounding a TCO is suitable for the
TES.

Another topic discussed by the ESC WG members was design of the pesticide
drift model. At the onset of the project, it was proposed that a highly specialized (i.e.,
parameterized for weather, topography, spray apparatus) pesticide drift model be built in
a GIS to represent pesticides moving beyond TCO/TCB boundaries. ESC WG members
decided as part of this pilot project a site specific drift model would be premature,
because the overall integration process was still being developed. In place of a
specialized pesticide drift model a 210 m buffer (pesticide drift layer) was created around
every TCB as a first attempt to represent pesticide drift distances and its potential
influence on TES potential habitat. The ESC WG members also agreed that in fiiture
stages of this project, a site specific pesticide drift model (e. g., variable parameters to

represent weather conditions during pesticide application, pesticide application

66

equipment efficiency, and pesticide types applied) would become important as it would
allow a more accurate representation of pesticide drift.

The previous examples demonstrate how ESC WG members and researchers
worked collaboratively to develop the project. As this project evolves, the hope is that
the ESC WG will continue to convene meetings and play a role in refinements, including
expansion to other commodities.

Tart ( ‘herry Spatial Data

Tart cherries were chosen as the commodity for this pilot project for many
reasons. First, the majority of TCOs occur in the western counties of Michigan’s Lower
Peninsula, creating a more concentrated spatial extent. Second, TCO owners have shown
dedication and interest in protecting the environment by employing [PM practices (Tart
Cherry Pest Management Strategic Plan 2000, 2006). Finally, the tart cherry commodity
was selected because it had recently been affected by denial of a pesticide EUP due to the
presence ofTES (pers. comm, M. Whalon, pers. comm, P. Korson, Project GREEEN
2005)

The process for collecting spatial data on TCB locations worked effectively. It is
important to recognize that these spatial data represent TCBs that occur within TCOs.
This is relevant because TCBs were overlaid with the TES potential habitat and not the
TCO boundaries. Actual TCO perimeters are not known and thus individual TCBs could
actually occur in the middle of an orchard. In the future, when perimeters of most TCBs
have been mapped it would be interesting to combine each grower’s TCBs into one and
create the actual TCO boundary polygons to observe how they overlay with TES

potential habitat.

67

While only a small percentage of tart cherry acreage was sampled it was sufficient
to demonstrate how this process would work. In the fixture as NASS workers become
familiar with GPS units it would be more efficient if TCB spatial coordinates were saved
on the GPS units’ memory cards, which could be sent in for direct download onto the
computer. This would reduce the risk for errors in data collection and data entry and
reduce the amount oftime spent entering data.

The majority of TCB locations were used throughout this project, but a few had to
be discarded as they appeared abnormal (i.e., overlapping boundaries, one comer
extended abnormally from others) when portrayed in GIS. Abnormal TCBs were sent
back to CIAB and a few were fixed by NASS employees returning to the TCB and re-
recording the corner coordinates; most abnormalities could not be corrected and these
TCBs were eventually discarded. In the future, continual collection of TCB locations is
important to increase the total number ofTCBs and tart cherry growers in the sample.
This will provide more growers the opportunity to know if they could potentially be
impacting TES or TES habitat. Another potential option to increase the TCO/TCB
sample size would be the generation of a NASS cropland dataset for Michigan. This
dataset would likely allow for identification of TCOs/TCBS in Michigan and is currently
under construction at MSU’s Land Policy Institute.

Biomapper Challenges and Solutions

Biomapper software was selected for this project for multiple reasons. First,

Biomapper software functions using only presence locational data for a species, which is

the type of data Michigan has for its TES. Second, Biomapper is a free software that is

()8

readily accessible for download over the internet, thus if this process is to expand to other
states or commodities they will be able to obtain this modeling software.

While Biomapper is a useful modeling software program, challenges were
encountered while preparing spatial layers for or using layers in this software. One of the
first challenges was in converting ESRI rasters to Idrisi file format. Different methods
were explored, including GridConverter and ESRI extensions (Hirzel 2006). The ESRI
extensions did not work, likely because ESRI rasters were too large for the program. The
GridConverter worked and was the method employed here.

Challenges were also encountered when importing EGV layers into the
Biomapper software. After the first EGV layer was added and the second EGV layer was
in the process of being added, an error occurred stating they could not be overlaid. To
bring all EGV layers into Biomapper software without this error, the EGV layers had
their analysis extent and analysis mask set to its corresponding species boolean layer in
ArcMap. This ensured that the EGV layers covered the same spatial extent and the raster
cells lined up. This should not be viewed as a “problem”, but rather a good error check to
make sure the data layers are aligned.

During the verification process discrepancies between EGV layers and potential
problematic EGV layers were identified (Hirzel 2004, 2006). Discrepancies must be
corrected because they can cause raster cells in the EGV layers to be eliminated from the
analysis (Hirzel 2006). Several EGV layers contained discrepancies and were
reclassified. For example, reclassifying the temperature layer to different units (e.g.,
Celsius to Fahrenheit) allowed for cell values to range between 0 and 250, which were

byte layers with a background value of 255 (Hirzel 2006). The elevation layer presented

69

a different scenario and was reclassified with values between 10-80, where 10
represented low elevations (i.e., 141-201 m) and 80 represented high elevations (i.e., 458-
602 m). Converting the elevation layer to different units (e. g., meters to miles) did not
work because it led to values of 255 which led to discrepancies. In the future, a different
method could be attempted to eliminate discrepancies between EGV layers that would
not require reclassifying the EGV layer values or converting the EGV layer values to
different units. Instead of getting all EGV layers into the byte data type, maybe EGV
layers could be integer or real data types or the background values could be changed for
EGV maps. For other concerns, challenges, and solutions refer to Appendix N.
7 ES EG Vs, and (1'1 S Spatial Data Matters

Spatial locations for TES were obtained from MNFI, which is a positive sighting
organization that surveys for rare organisms where they have received funding to search
for them (pers. comm, E. Schools). This project required a statewide perspective of TES
to help identify potential interactions with TCBs. The statewide perspective was needed
because there were no guarantees that MNFI sampled TES in areas near TCBs. To
overcome this shortcoming, a predictive model was used to create statewide HS maps.

Creation ofthe HS maps involved using species presence data, and EGVs
reflecting each species’ habitat requirements and needs. The EGV layers were created by
and obtained from different entities (Table 3, 4, 5) often with different spatial resolutions.
For example, the total precipitation layer had a spatial resolution of 800 m while the tree
canopy cover layer had a spatial resolution of 30 m. Because 30 m was chosen as the
analysis resolution, the 800 m raster had to be resampled to a 30 m cell size. The

resampling process did not cause a loss of information, but also did not improve the

70

quality of the total precipitation layer. Thus any map that used the total precipitation
layer only has a useful accuracy of 800 m.

The majority of EGV layers required to model TES habitat were identified and
obtained directly for the project. However, several unavailable data layers would have
greatly enhanced the quality of final HS maps. For example, lupine is the food source for
KBB larvae (Rabe 2001, USFWS 2003) thus a layer representing the statewide
distribution of lupine would have been valuable. A layer representing lupine was
available, but it was not completed at a statewide level (pers. comm, E. Schools). In the
future a statewide lupine survey could be completed and this layer could be incorporated
into the KBB habitat model. A more practical alternative would be to use lupine
presence locations to build a predictive model in support of the KBB assessment.

Along with locating spatial data layers, at the heart of GIS technology lies the
concern for spatial data accuracy. The outputs produced by GIS are a product of the data
input, therefore if data input to a GIS are not accurate the output will reflect that quality.
In many cases extremely accurate spatial data are not available thus the spatial data
currently available had to be used. In the future when new or more accurate spatial data
layers are created these spatial data layers can be substituted and replace the older less
accurate spatial data layers used in this modeling process. For example, if new layers are
produced that have a higher accuracy or a finer resolution (i.e., 10 m instead of 30 m)
than the older layers, these new layers can be replaced within the models and the model
outputs updated.

The National Land Cover Dataset 2001 Tree Canopy layer had an attribute

accuracy value of 93% (these values were obtained from the original layers metadata).

71

The IF MAP layer had an overall accuracy value of 88% for major land cover classes,
while it had an accuracy of8 % for non-forested classes and 68% for forested classes
(Space Imaging 2004). Attribute accuracy values were not able to be determined or
located for the remaining original GIS layers.
TES Outputs and HS I avers

The KBB model indicated that their habitat differs from average habitat
conditions available in Michigan. Model results suggest that within these habitats KBB
can live in a wide range of conditions (based on the global tolerance and specialization

values). The model outputs are not biological intuitive as KBB is an endangered species

 

 

and has specific habitat needs (USFWS 2003). The lack of a lupine layer may have
resulted in a more generalized model output, but this is not known for fact. Lupine is a
major determinate for KBB presence, and its inclusion could alter the models global
tolerance and specialization values. Incorporating a lupine layer into the KBB model
might also refine its HS layer to better identify suitable areas.

The final PT HS layer showed highly suitable habitat for PT occurring primarily
along the coasts in Michigan, which would be expected as it grows on dunes (USFWS
2002). There were some areas modeled as suitable habitat towards the interior of the
state which might not be likely places to encounter PT. These areas may have been
identified by the IFMAP frequency of sand - soil layer as this variable tended to occur in
these interior areas. Eliminating these suitable areas in the interior part of the state is
possible if the PT HS layer is reclassified such that suitable habitat is _>_ 40. Although it is
less likely for PT to be occurring in the interior portions of Michigan it would be worth

surveying these areas to see if the species occurs there.

72

The final IB HS layer predicted lower HS values for [8 presence locations (i.e.,
points) found further northeast than the other [8 presence locations (i.e., points and
polygons). This might be caused by one large [8 presence location (i.e., polygon) being
located in the south central part of Michigan. This large [B polygon presence location
might have been created by MNFI buffering an 18 point location, thus not all areas within
the polygon are “true” IB presence locations. When the large [B polygon presence
location was converted to raster, every 30 m location within the polygon was recorded as
a presence location. It would be informative to examine model performance if only point
presence locations were used to see how the results would change. The polygon presence
locations would not have to be discarded, but a random point could be assigned within
each presence polygon to represent it rather than the fill] polygon.

The CV results for the conservatively reclassified HS layer for KBB suggested
that all the area and species validation points were within suitable habitat based on this
reclassification scenario as only 1 bin was shown (i.e., random map). As the HS
boundary is increased (i.e., HS 20, HS 50) the area-adjusted frequencies for bin 1 remain
under 1 and for bin 2 the area-adjusted frequencies increase to values greater than 1
suggesting they could be usefiil models. This suggests that the conservative
reclassification scenario for KBB HS layer is not the best, but the reclassification
scenario could be changed (i.e., HS values from 0-20 represent unsuitable habitat and HS
values from 21-100 represent suitable habitat) to create a better model. The conservative
reclassification scenario for KBB was still used for this project as demonstration of the

process.

73

The CV results for the liberally reclassified HS layer for KBB suggest that it was
not a useful model. When the HS bin boundary was placed at HS 99, bin 1 had area-
adjusted frequency values slightly less than one while bin 2 had values greater than 1
(i.e., ranging between 20 and 90), suggesting this could be a better model, although
variation was still high. Again, the reclassification scenario can be changed in the future.

The area-adjusted frequencies listed in the summary of CV results for the liberally
reclassified HS maps for KBB, PT, and [B did not match the values shown on the area-
adjusted frequency graphs. This problem did not occur with the conservatively
reclassified PT or [B HS maps as the area-adjusted frequencies shown on the graph
matched the list in the summary of CV results. A possible explanation might be that
when the Biomapper software was recording the area-adjusted frequency values shown in
the area-adjusted frequency graph it was not fimctioning properly and the values were not
recorded in the correct locations in the summary of CV results. With the extreme HS bin
boundary (i.e., HS value 100) the lines extended past the boundary of the graph and the
area-adjusted frequencies were recorded in incorrect locations in the summary of CV
results. When the HS bin boundary was placed at HS 99 for the IB, the lines on the graph
moved, but were not yet completely contained on the graph, and in the summary of CV
results, the area-adjusted frequencies still were not in the correct positions, but they did
move. When the HS bin boundary was placed at HS 98 for the IB, all lines ended on the
graph, and the area-adjusted frequencies were located in the correct place in the summary
of CV results. This suggests that it was just an issue with the recording process. The
newest version of the Biomapper software has developed a new CV result which is less

sensitive to the number of bins selected and where the bin boundaries are placed (Hirzel

74

et al. 2006). This new method could be used to gain a better understanding and refine the
current CV results and TES reclassification scenarios (Hirzel et al. 2006).

To better represent locations of more suitable habitat for the TES, changing the
reclassification classes should be attempted (e. g, unsuitable habitat represented by HS
values from 0-9 and suitable habitat represented by HS values from 10-100). The two
extreme scenarios used in this project were a first attempt to include as much and as little F
potentially suitable habitat as possible. Hirzel et al. (2006) lists a variety of methods on
how to reclassify HS layers in a meaningful way based on examining the curves produced

in the CV process. "

 

The species models created for this portion of the pilot project should not be
viewed as definitive or the final models that should be used to base decisions on. My
goal was to show a process not develop the “best” habitat models. If this process is
deemed a good one, then the investment in creating really good habitat models should be
made.

Identification ofPolentially Overlapping T ES Habitat and TCBs

Again, I want to stress that this is a crude first approximation for identifying
potential areas of overlap between TES habitat and TCBs. Based on results produced in
GIS all TCBs examined were found to be within at least 210 m of one TES under the
most conservative habitat identification scenario. Most TCBs were found intersecting
KBB suitable habitat as its modeled habitats were most widespread. It is important to
also remember that the KBB model is flawed and not useful when reclassified under this
scenario, thus these numbers are likely incorrect. PT had the second greatest number of

TC 83 intersecting suitable habitat, although this number was likely artificially high

75

because some TCBs identified occurred away from the coasts and more towards the
interior of the state where PT is not likely to occur. This artifact is probably caused by
the areas identified as suitable habitat towards the interior part of state because of the
frequency of sand — soil layer. The [B had the fewest TCBs intersecting its suitable
habitat. This result is logical as the majority of [B modeled suitable habitat occurs in the
southern portion ofthe state and does not continue into northwestern Michigan where
some TCB locations tend to occur. Four TCBs were identified in the northern Lower
Peninsula as intersecting [B suitable habitat.

Based on results produced in GIS under the most liberal habitat reclassification
scenario, only 4 tart cherry growers and 5 TCBs were within 210 m of suitable TES
habitat. The KBB was the only TES found to intersect with TCBs when this
reclassification scenario was used. These interactions occurred in the northwest and west
central parts of Oceana County, but again should be interpreted with caution because the
KBB models were found not useful under the extreme reclassification scenarios used.
This variability in results stresses the importance of habitat model reclassification on
implementing this process. Accurate habitat models are critical to implementation of the
commodity-TES process.

The pesticide drift layer used in this pilot project allowed for creation of a zone of
interaction (i.e., area of overlap) that portrayed pesticides leaving the TCBs. While this
layer served its purpose there are concerns regarding its use that could be addressed in the
future if site specific pesticide drift models are created. Currently the pesticide drift layer
surrounds each TCB uniformly. To better reflect pesticide drift it would be important to

include dominant wind direction as a factor influencing how pesticides may be deposited.

76

The pesticide drift layer also does not allow for differences in pesticide application
equipment and this could potentially influence how much pesticide drifts from the site.
Growers use pesticide application equipment that varies in its accuracy and thus
pesticides could drift varying distances depending on equipment. This factor was not
portrayed when using the fixed width buffer for each TCB.

While the results for the conservative scenario most likely over predicts the actual E
number of TES-TCB interactions, results for the liberal interpretation most likely under
predicts the actual number of interactions. These results, however, are an important place

to start. The conservative scenario portrays the largest amount of potential habitat and it

 

mum- ‘. ' ‘
'l

is likely that fewer positive TES occurrences would be found in the areas with very low
HS values. In contrast, the liberal scenario portrays only the best habitats, but positive
TES occurrences would most likely occur in more marginal habitat. Biomapper software
does offer a process and the associated statistics for measuring model performance under
different reclassification scenarios. In the future as the process continues to be refined,
Hirzel et al. (2006) suggests ways to reclassify HS maps based on CV curve outputs. For
example, reclassifying the HS values (e. g., HS values from 0—10 represent unsuitable
habitat and HS values from 11-100 represent suitable habitat) could be a better model and
reduce the amount of habitat put into the suitable category possibly lowering the number
of tart cherry growers found within 210 m of TES suitable habitat.
Develop Process to Aid in Decisions

The processes outlined by this pilot research project can be useful for other
commodities and states that want to take proactive measures to protect commodity

production, private landowners, and TES. These types of proactive processes could

77

become increasingly important as the human population continues to increase and
consume more land that was once used by these species, thus increasing the potential for
TES interactions.

This project has the potential to help commodity producers and government
agencies make more informed decisions with regards to TES and pesticide use. The
process could aid commodity producers by informing them of potential TES locations
based on modeled habitat and help evaluate pesticide application options. For example,
if a grower is identified as overlapping with potential TES habitat they could change their

pesticide application methods or timing or put in a hedgerow or buffer area to reduce the

 

potential for interaction between the pesticide and TES. Government agencies might also
benefit from the process described herein because it would allow them to base their
pesticide use or TES protection decisions or survey requirements on more resolute spatial
relationships between TES potential habitat and TCBs. Under this process decisions are
based on a more biologically meaningful rather than political (i.e., county) scale.

Future work to improve the process is still needed. Current TES models should
be updated (e.g., using newest version of Biomapper software for CV processes,
changing the reclassification scenarios to better represent areas of suitable habitat for
each TES, incorporating new TES presence locations, adding new and refining currently
used EGV layers). Alternative habitat modeling approaches could be examined to
determine which habitat model is best suited for this process (e.g., MaxEnt). HS maps
should be created for other TES and the process expanded to include state listed species.
The amount of TCB (and other commodity) spatial data should increase allowing for

more comprehensive evaluations of potential TES interactions. The TCBs identified as

78

potentially affecting TES should be overlaid with aerial photos to help verify if TES
habitat actually occurs. Results from the habitat models should be used to identify
priority areas to search for TES search areas. A more site specific pesticide drift model
should be parameterized to reflect weather conditions and each TCB grower’s pesticide
application methods (e. g, equipment, pesticide type). Refinement of the pesticide drift

model would provide a more accurate representation of pesticide drift behavior thereby

d' 1|.)

improving identification of TES-commodity production interactions. The TES
phenologies could also be examined to determine if TES are likely to be present when

pesticides are applied.

 

H’feh'l I. an"? «:31. ..:’.".'k"

Once these steps have occurred narrowing where potential interactions may be
occurring, then meetings with the individual grower’s identified to potential be affecting
TES/T ES habitat can be scheduled and mitigation procedures (i.e., prevent TES and
pesticide interactions) discussed. Finally, this process could expand to other
commodities in the state and throughout the nation. For this project to expand to other
commodities it would require TES spatial data and commodity spatial data for those of
interest to be known along with EGVs for the entire study area. Again, I want to re-stress
that my work was simply to develop the process and not to ensure that the “best” models
were developed. As such my work needs refinement before this is implemented at
operational scales. The utility of my project was to show some of the capabilities that

can be brought to bare on the issue of TES conservation and commodity production.

79

APPENDICES

80

 

. Q gt th F‘Z‘; I

if

APPENDIX A:

Michigan ('Ieoref
In ArcMap, the KBB and PT species rasters were projected to Michigan Georef
with the nearest sampling method. The projection file for each species raster was edited
such that the Azimuth at center of projection was 337.25556 instead of 337 15 20.016.
This is an error that occurs when working with rasters in ArcMap and if not fixed can
cause a 25 km shift between rasters (pers. comm, E. Schools). The project file was

checked for this error for all rasters created in ArcMap throughout this project.

81

 

APPENDIX B:

Initial GIS Processes for [FA/{AP

Important land cover types were extracted from the 2001 30 111 Integrated Forest
Monitoring Assessment and Prescription (IFMAP/GAP) Lower and Upper Peninsula land
cover images. The Upper and Lower Peninsula IFMAP images were each converted into
rasters in ArcView at a 30 m cell size. A 30 m cell size was selected for all rasters
because it was the smallest resolution of the EGV rasters. In ArcMap, each IFMAP
raster was projected into Michigan Georef with the nearest sampling method at a 30 m
cell size. When working with rasters in ArcMap, the projection file has to be checked
after each process and the azimuth corrected or the rasters would not line up correctly
(pers. comm, F.. Schools) and this was completed for all rasters created.

Each IFM AP raster was processed to show only data in the state of Michigan
boundary. In ArcMap, the Upper and Lower Peninsula IFMAP rasters were each masked
with county rasters. For both the Upper and Lower Peninsulas, IFMAP raster’s analysis
mask was set to its corresponding county raster while the extent was snapped to its
corresponding IFMAP raster. This process was carried out as IFMAP rasters were
evaluated.

Once the above steps were completed, the Upper and Lower Peninsula IF MAP
rasters were merged in ArcMap with the extent snapped to the Lower Peninsula IFMAP
raster because the TCBs are located in the Lower Peninsula and this area was more
important for the raster cells to remain in their current location. The merging process

caused the Upper Peninsula IFMAP raster to shift in location, but ESRI examined the

82

 

data and determined that the Upper and Lower Peninsula IFMAP rasters had different
starting extents and thus one of the rasters would have to shift slightly. This did not

affect the modeling process.

83

APPENDIX C:

Initial (ii/S Processes for Total Precipitation

Data pertaining to total precipitation amounts in Michigan was portrayed by the
800 m total precipitation (normals; total precipitation, 30-year (1971—2000 average))
raster. The total precipitation layer was in mm. This layer was imported into ArcView as
an ASCII raster where its values were set to integer to convert cell values to whole
numbers and not decimals (i.e., not floating point). In ArcMap the total precipitation
ASCII projection was defined as WGS72 as this was the datum specified in the metadata
(PRISM Group 2006a). It was projected as a raster to WGS72 and bilinearly resampled
with a cell size of‘0.008333, which is the resolution in decimal degree units for the total
precipitation layer (PRISM Group 2006a). Next, the total precipitation raster was
projected to Michigan Georef and bilinearly resampled with a cell size of 2,088. The
2,088 cell size was automatically used by ArcMap and kept to ensure the process worked.
In Arclnfo a value attribute table (VAT) was built for the total precipitation raster to be
able to examine raster cell values. In ArcView the total precipitation raster was clipped
by a complete Michigan county shapefile to eliminate data for the rest of the US. The
total precipitation raster was converted to a shapefile for fithher clipping and converted
back to a raster at a 30 m cell size. The total precipitation raster values were converted
from mm to inches. This was done to avoid discrepancies in Biomapper software as
many of the EGV rasters created for this project when converted to Idrisi file format were

byte maps. Byte maps values range between 0-255 (i.e., 255 is background value), and

84

 

with total precipitation amounts in mm the values exceeded this range and led to

discrepancies.

85

APPENDIX D:

Initial GIS Processes for Percent Sand in Soil

The percent of sand in soil was portrayed by the 1 km Conus - Soil, Sand, Silt,
Clay fraction layer. The percent sand in soil layer included data for the entire US and
was converted to show only data from Michigan. This layer consisted of soil samples
taken at l 1 different ground depths where the percent of sand, silt, and clay were
recorded (Miller and White 1998). The soils attribute table was edited and a total sand
field was added to create a new value scheme for this layer that would show the amount
of sand found in each soil type. The total sand field was populated by taking the sum of
the first 9 of the l 1 sand values measurements taken at different levels below ground to
determine the amount of sand in each soil. The last two sand value measurements
underground were not used as the majority of them were 0. A new percent sand field was
added to the attribute table and was filled by values when the total sand field values were
divided by 900 and multiplied by 100 to get the percent of sand in each layer. The value
in the total sand field was divided by 900 as 9 sand values were summed for each layer
and the sand value could reach 100 at each of the 9 distances where it was sampled. This
was completed to obtain values between 0-255 and not lead to discrepancies in
Biomapper software. In ArcMap the percent sand in soils raster was projected to

Michigan Georef and resampled in a bilinear method.

86

 

APPENDIX E:

Initial GIS Processes/or 'I'ree (’anopy Cover
The amount of tree canopy cover was portrayed by the 30 m National Land Cover
Database 2001 forest canopy image. In ArcMap, the tree canopy cover layer was
projected as a raster with a bilinear sampling method and converted from an image to a
raster with a 30 m cell size. The tree canopy cover raster was masked to a blank

Michigan county raster (i.e., all values 0) to show only data occurring within Michigan.

87

APPENDIX F:

Initial ( iIS I’racesses for Texture and Drainage

Well and moderately drained soils were extracted from the texture and drainage
shapefile. This shapefile consisted of three drainage categories including well and
moderately drained, somewhat poorly drained, and very poorly drained soils. As the
shapefile contained qualitative names it had to be reclassified such that the names were
changed to numbers (e. g, 3 = well and moderately drained, 2 = somewhat poorly
drained, 1 = poorly drained). This step had to be completed to create values that could be
selected to produce a boolean map. Next, the well and moderately drained soils shapefile
was converted to raster and in ArcMap projected to Michigan Georef using the nearest
sampling method. The nearest sampling method was chosen as it performs well for

categorical data. as it does not change the cell value (ArcMap help file).

88

APPENDIX C:

Initial GIS Processes for National It'levation Dataset
Elevation was portrayed by the 30 m national elevation dataset layer. The values
for this raster were in m multiplied by 1000. A 3.2 km buffer occurred around the
perimeter of the state. To obtain raster values ranging from 0-255, the raster was
classified into 8 natural break groups and reclassified to get low elevation values
represented by 10 and high elevations represented by 80 with the other elevations ranging

between them (i.e., 10 represents 141-201 m).

89

 

APPENDIX H:

Initial GIS Processes for Slope
The slope was also derived from the national elevation dataset. In ArcMap’s
spatial analyst surface analysis a raster representing slope was created for the entire state
of Michigan. The output was represented in degrees. The z-factor was set to 0.001 as the
original national elevation dataset raster was in m multiplied by 1000. The cell size was
set to 30 m. This process created a floating point raster, which was converted in

ArcView to an integer raster.

90

 

APPENDIX I:

Initial ( H S Prac,'e.s'.sesfl)r Average Dain Il/Iarimum and A verage Daily Minimum
Temperatures

Average daily maximum temperatures were portrayed by the 800 m average daily
maximum temperature (normals; average daily maximum temperature, 30-year (1971-
2000) average) layer and average daily minimum temperatures were portrayed by the 800
m average daily minimum temperature (normals; average daily minimum temperature,
30-year (1971 - 2000) average) layer. Both layers had temperature values recorded in
degrees Celsius. This layer was manipulated in a similar fashion to the total precipitation
layer (Appendix C) as it was obtained from the same source. Each layer was imported
into ArcView as an ASCII raster and the values were set to integer. In ArcMap, the
ASCII projection was defined to WGS72 and then projected as a raster to WGS72 where
it was bilinearly resampled with a cell size of 0.008333. Next, each raster was projected
to Michigan Georef and bilinearly resampled with a cell size of 2,088. In ArcView, each
raster was clipped by a complete Michigan county shapefile to eliminate extraneous data
while further clipping occurred after they were converted to a shapefile. Each shapefile
was then converted back to a raster with a 30 m cell size. The attribute table was edited
to get values between 0-255 and this was accomplished by converting the current degrees

Celsius values into degrees Fahrenheit.

91

APPENDIX J:

Initial ( ilS Processes for Proximity to Great Lakes

A Great Lakes shapefile was used to create a layer portraying proximity to Great
Lakes. The five Great Lakes were selected and converted to a 30 m raster where all the
Great Lakes had a common value of I. In ArcMap the Euclidean distance from the Great
Lakes to inner areas of Michigan was determined, thus creating the proximity to Great
Lakes raster. This raster was converted from a floating point raster to an integer raster
and masked to a blank Michigan county raster to display only values occurring in
Michigan. In Arclnfo a VAT was built to view the distance values associated with this
raster. To obtain the distance values ofthis raster between 0-255 (byte values) in Arc the

values \VCFC converted ITO”) m [0 mi.

92

 

APPENDIX K:

Initial GIS Processes for Wetlai'icls
Wetlands were portrayed by the National Wetlands Inventory raster (created at
1:24,000 resolution). The wetlands layer consisted of 5 types of systems including
lacustrine wetlands, palustrine wetlands, riverine wetlands, uplands, and unknown. The
attribute table was edited such that each type of system had a numeric code associated
with it. The raster was reclassified such that all wetlands’ (i.e., lacustrine, palustrine,
riverine) new values were set to l and the unknown and uplands were 0 creating a

boolean raster.

93

APPENDIX L:

Initial GIS Processes for H ydrologic Features

Hydrologic features were portrayed by the Michigan geographic framework
hydrology shapefile (created at 124,000 resolution). The hydrologic features layer
consisted of rivers, lakes, ponds, creeks, ditches, drains, and streams. The attribute table
of the hydrologic features shapefile was edited such that each hydrologic feature had a
common value of I. In ArcView the hydrologic features shapefile was converted to a 30
m raster and reclassified in ArcMap to get all hydrologic features represented by 1 and all
other values represented by 0, creating a boolean raster. It was masked to a blank

Michigan county raster to show only data within Michigan.

94

' a .. 'JhAaI—

 

APPENDIX M:

Initial GIS Processes for I ’roximity to H ya'rologic Features
A proximity to hydrologic features layer was created. In ArcMap the hydrologic
features boolean raster (created in Appendix L) was reclassified such that all hydrologic
features were a l and all other were no data instead of 0. The other values had to be no
data for the Euclidean distant analysis to function. Next, the raster was processed
through the Euclidean distance analysis at a 30 m raster cell size. In ArcView it was

clipped by a Michigan county shapefile to show only data for Michigan and in Arclnfo

 

the values were converted from m to km to get values between 0-255.

95

APPENDIX N:

Biomapper Challenges and Solutions

Computing times and computer space and memory became factors when using
Biomapper software. Depending on the algorithm selected, the amount of time to
compute the HS map and CV results can vary (Hirzel and Arlettaz 2003a, b). For
example, I attempted to use the distance geometric mean algorithm to create an HS map,
but after 24 hours the process had barely progressed. In place of the distance geometric
mean algorithm I chose the medians algorithm which for the KBB took 4.5 hours to
produce the HS map and 8.5 hours for the CV procedure. These times exclude time spent
gathering data, preparing data for use, verifying EGV layers, and normalizing EGV
layers.

The median and distance geometric mean algorithms were both tried as the HS
algorithm. The medians algorithm makes an assumption about the species distribution
(i.e., . .that the best habitat is at the median of the species distribution on each factor,
and that these distributions are symmetrical.” (Hirzel 2004)), but was computed for all 3
TES as it can provide results in a timely manner (Hirzel and Arlettaz 2003a, b; Hirzel
2004). The distance geometric means algorithm was tried because the assumption about
species distribution is not made (Hirzel 2004), but was not selected because it performed
too slowly. Studies by Hirzel and Arlettaz (2003a, b) also noted the distance geometric
mean, distance harmonic mean, and minimum distance algorithms had slower computing
times than the medians. Hirzel and Arlettaz (2003b) suggested that since the median

algorithm can be computed faster it could be used first and then the data can be examined

96

to determine if the distance geometric mean needs to be used. In the future it might be
useful to attempt the distance geometric mean algorithm again to observe differences
between output HS maps.

The output files created were also large and for one complete TES project the files
could consume 20 to 30 GB of hard drive space on the computer. The use of external
hard drives becomes important in this case for storage and backup copies. #-

I also experienced “out of memory” errors during the CV process and this was an ‘
inconvenience as many times it takes 30 minutes per partition and each time the error

occurs the process starts over from the beginning. In many situations to get the CV

 

process to successfiilly finish I would have to reduce the number of partitions (e. g., 5
partitions instead of 10 partitions). This made the goal of achieving CV results with 10
partitions, the number recommended by Hirzel (2006), hard to achieve and in some cases
it could not be reached. The other option provided by Hirzel (2006) to decide on the
number of partitions was Hubert’s rule, which bases its result on the number of EGV
layers used (Hirzel 2006). The KBB and PT were cross-validated using 10 partitions, but
the [B was cross-validated with only 5 partitions. The five partitions were between the
10 partitions recommended by Hirzel (2006) and the 4 partitions recommended by
Hubert’s rule. I have found no way to get around the “out of memory” error that occurs.
Recently a newer version of the Biomapper software was made available and the CV
methods are one of the main differences between these two versions and should be

looked at in the fixture.

97

LITERATURE CITED

98

LITERATURE CITED

Boyce, M. S., P. R. Vemier, S. E. Nielson, and F. K. A. Schmiegelow. 2002. Evaluating
resource selection functions. Ecological Modelling 157: 281-300.

Callahan, E. V., R. D. Drobney, and R. L. Clawson. 1997. Selection of summer roosting

sites by Indiana bats (Myotis sodalis) in Missouri. Journal of Mammalogy 78:
818-825.

Carroll, M. S., C. W. McKetta, K. A. Blatner, and C. Schallau. 1999. A response to

“Forty years of spotted owls? A longitudinal analysis of logging industry job
losses.” Sociological Perspectives 42: 325-333.

Clark, B. K., J. B. Bowles, and B. S. Clark. 1987. Summer status of the endangered
Indiana bat in Iowa. American Midland Naturalist 118: 32-39.

Clark D. R., Jr., C. M. Bunck, E. Cromatrie, and R. K. LaVal. 1983. Year and age
effects on residues of dieldrin and heptachlor in dead gray bats, Franklin County,
Missouri-1976, 1977, and 1978. Environmental Toxicology and Chemistry 2:
387-393.

Clark, D. R., Jr., R. K. LaVal, and D. M. Swineford. 1978. Dieldrin—induced mortality in
an endangered species, the gray bat (Myotis grisescens). Science 199: 1357-1359.

Clawson, R. L, and D. R. Clark, Jr. 1989. Pesticide contamination of endangered gray
bats and their food base in Boone County, Missouri, 1982. Bulletin of
Environmental Contamination and Toxicology 42: 43 1-43 7.

Engler, R., A. Guisan, and L. Rechsteiner. 2004. An improved approach for predicting
the distribution of rare and endangered species from occurrence and pseudo-
absence data. Journal of Applied Ecology 41: 263-274.

Environmental Systems Research Institute, Inc. [ESRI]. 1997. ESRI shapefile technical
description: an ESRI white paper—July 1998.

<www.esri.com/library/whitepapers/pdfs/shapefi1e.pdf>. Accessed 11 May 2007.

Farmer, A. H., B. S. Cade, and D. F. Stauffer. 2002. Evaluation of a habitat suitability
index model for the Indiana bat. Pages 172-179 in Proceedings of the Indiana bat:
biology and management of an endangered species. Bat Conservation
International, Austin, Texas, USA.

Freudenburg, W. R., L. J. Wilson, and D. J. O’Leary. 1998. Forty years of spotted owls?

A longitudinal analysis of logging industry job losses. Sociological Perspectives
41: 1-26.

99

 

Guisan A, and N. E. Zimmermann. 2000. Predictive habitat distribution models in
ecology. Ecological Modelling 135: 147-186.

Hardy-Short, D. C., and C. B. Short. 2000. Science, economics, and rhetoric:

environmental advocacy and the wolf reintroduction debate, 1987-1999. Pages
65-72 in USDA Forest Serve Proceedings RMRS-P-l 5-VOL-2.

Herms, C. P., D. G. McCullough, L.S. Baue, R. A. Haack, D. L. Miller, and N. R.
Dubois. 1997. Susceptibility of the endangered Karner blue butterfly
(Lepidoptera: lycaenidae) to bacillus thuringiensis var. kurstaki used for gypsy
moth suppression in Michigan. The Great Lakes Entomologist 30: 125-141. .

Higman, P. J ., and M. R. Penskar. 1999. Special plant abstract for Cirsium pitcheri.
Michigan Natural Features Inventory, Lansing, Michigan, USA. <
httpz/lweb4.msue.msu.edu/mnfi/pub/abstracts.cfm#Plants>. Accessed 1 March
2006.

Hijmans, R. J., L. Guarino, A. Jarvis, R. O’Brien, and P. Mathur. No Date. DIVA-GIS
5.2. International Potato Center and International Plant Genetic Resources
Institute, Lima, Peru. <http://www.diva-gis.org/>.

 

Hirzel A. H. 2004. BioMapper 3 user’s manual. Lab. of Conservation Biology,
Department of Ecology and Evolution, University of Lausanne, Switzerland. <
httpzl/wwwunil.ch/biomapper>.

Hirzel, A. H. 2006. BioMapper - frequently asked questions version 22.02.2006. <
http://www2.unil.ch/biomapper/faq.html>. Accessed 3 April 2006.

Hirzel, A. H. 2007. Evaluation. <http://biomapper.wikispaces.com/Evaluation>.
Accessed 28 February 2007.

Hirzel, A. H., G. LeLay, V. Helfer, C. Randin, and A. Guisan. 2006. Evaluating the
ability of habitat suitability models to predict species presences. Ecological
Modelling 199: 142-152.

Hirzel, A. H, J. Hausser, D. Chessel, and N. Perrin. 2002. Ecological-niche factor
analysis: how to compute habitat-suitability maps without absence data?
Ecology 83: 2027-2036.

Hirzel, A. H., J. Hausser, and N. Perrin. 2004. Biomapper 3.1. Lab. of Conservation
Biology, Department of Ecology and Evolution, University of Lausanne,
Switzerland. < http://www.unil.ch/biomapper>.

Hirzel, A .H., and R. Arlettaz. 2003a. Environmental-enveIOpe based habitat-suitability

models. Pages 67-76 in [st Conference on Resource Selection by Animals (B.F.J.
Manly, ed.). Omnipress, Laramie, Wyoming, USA.

100

Hirzel, A. H., and R. Arlettaz. 2003b. Modeling habitat suitability for complex species
distributions by environmental-distance geometric mean. Environmental
Management 32: 614-623.

Hirzel, A. H., V. Helfer, and F. Metral. 2001. Assessing habitat-suitability models with
avirtual species. Ecological Modelling 145: Ill-I21.

Humphrey, S. R., A. R. Richter, and J. B. Cope. 1977. Summer habitat and ecology of
the endangered Indiana bat, Myotis sodalis. Journal of Mammalogy 58: 334-346.

Kurta, A., K. J. Williams, and R. Mies. I996. Ecological, behavioral, and thermal
observations of a peripheral population of Indiana bats (Myotis sodalis). Pages
102-117 in Bats and Forests (R. M. R. Barclay and R. M. Brigham, eds.)
Research Branch, Ministry of Forests, Province of British Columbia, Victoria,
British Columbia.

Kurta, A, J. Caryl, and T. Lipps. 1997. Bats and Tippy Dam: species composition,
seasonal use, and environmental parameters. Michigan Academician XXIX: 473-
490.

Kurta, A., and H. Rice. 2002. Ecology and management of the Indiana bat in Michigan.
Michigan Academician XXXIV: 175-190.

Kurta, A., S. W. Murray, and D. H. Miller. 2002. Roost selection and movements across
the summer landscape. Pages 118-129 in the Indiana bat: biology and
management of an endangered species (A. Kurta and J. Kennedy, eds.) Bat
Conservation International, Austin, Texas, USA.

Michigan Center for Geographic Information. 2006. Michigan geographic framework:
state of Michigan: Hydro, version 6b. Michigan Center for Geographic
Information. < http://www.mcgi.state.mi.us/mgdl/?action=thm>.

Michigan Department of Natural Resources, Forest, Mineral and Fire Management
Division [MDNR, Forest, Mineral and Fire Management Division]. 2003.
IFMAP/GAP lower and upper peninsula land cover. Michigan Center for
Geographic Information. < httpzllwwwmcgi.statemi.us/mng/?action=thm>.

Michigan Fruit Districts, Michigan fruit inventory 2003-2004: all fruits [Michigan Fruit
Districts]. No Date. < www.nass.usda.gov/mi/mi_fruit04/allfruit.PDF>.
Accessed 15 December 2006.

Miller, D. A, and R. A. White. 1998. A conterminous United States multi-layer soil

characteristics data set for regional climate and hydrology modeling: sand, silt,
clay fraction. Earth Interactions. <http://Earthlnteractions.org>.

101

 

Olmstead, C. W. 1956. American orchard and vineyard regions. Economic Geography
32: 189-236.

O’Shea T. J ., and DR. Clark, Jr. 2002. An overview of contaminant in bats with special
reference to insecticides and the Indiana bat. Pages 237-253 in the Indiana bat:
biology and management of an endangered species, (A. Kurta and J. Kennedy,
eds). Bat Conservation International, Austin, Texas, USA.

Pollack, S., and A. Perez. 2006. Fruit and tree nut outlook FTS-323. US Department of
Agriculture, Electronic Outlook Report from the Economic Research Service. <
www.crs.usda.gov/Publications/FTS/2006/07JullFTS323.pdf>. Accessed 27

 

September 2006. F
Pope, R., S. Brown, and J. Ellerhoff. 1998. Why do farmers use pesticides? Iowa State

University: University Extension, Ames, Iowa, USA.
PRISM Group. 2006a. Precipitation (normals): total precipitation (30-year average,

annual 1971-2000; created 16 June 2006). Oregon State University. L

(:http://www.prismc1imate.org>.

PRISM Group. 2006b. Maximum temperature (normals): average daily maximum
temperature (30-year average, annual 1971-2000; created 28 June 2006). Oregon
State University. <http://www.prismclimate.org>.

PRISM Group. 2006c. Minimum temperature (normals): average daily minimum
temperature (30-year average, annual I971-2000; created 28 June 2006). Oregon
State University. <http://www.prismclimate.org>.

Project GREEEN. 2005. Partnering with growers, industry to protect endangered
species. Challenges. . . innovations. . .advancements. . . growth. Michigan State
University, East Lansing, Michigan, USA. <
httpzllwww.greeen.msu.edu/2005_Annual_REPORT.pdf>. Accessed 26
February 2007.

Rabe, M. L. 2001. Special animal abstract for Lycaeides melissa samuelis (Karner blue).
Michigan Natural Features Inventory, Lansing, Michigan, USA. <
http://web4.msue.msu.edu/mnfi/pub/abstracts.cfm >. Accessed 9 March 2006.

Ragsdale, N. N. 1999. The role of pesticides in agricultural crop protection. Annals of
the New York Academy of Sciences 894: 199-205.

Reichel, W. L., S. K. Schmeling, E. Cromartie, T. E. Kaiser, A. J. Krynitsky, T. G.
Lamont, B. M. Mulhern, R. M. Mulhern, R. M. Prouty, C. J. Stafford, and D. M.
Swineford. 1984. Pesticide, PCB, and lead residues and necropsy data for bald
eagles from 32 states - 1978-81. Environmental Monitoring and Assessment 4:
395-403.

102

Reutter, B. A, V. Helfer, A. H. Hirzel, and P. Vogel. 2003. Modeling habitat-suitability
using museum collections: an example with three sympatric Apodemus species
from the Alps. Journal of Biogeography 30: 581-590.

Scharpf, C. 2001. Farmers vs. fish: the battle of the Rio Grande silvery minnow.
American Currents 27: 8-12.

Space Imaging. 2004. Review of remote sensing technologies used in the IFMAP
project: final report. Ann Arbor, Michigan, USA.

Spray Drift Task Force. 1997. A summary of airblast application studies. <
http://www.agdrift.com/Text%20pageslPub_PDF.htm>. Accessed 8 November
2005.

Tart Cherry Integrated Orchard Management. 2006. About the project. <
http://www.ipm.msu.edu/tcabout.htm>. Accessed 20 October 2006.

Tart Cherry Pest Management Strategic Plan. 2000. Tart cherry pest management in the
future: development of a strategic plan.
<http://pestdata.ncsu.edu/pmsp/pdf/mitartcherry.pdf>. Accessed 23 June 2005.

Tart Cherry Pest Management Strategic Plan. 2006. Tart cherry pest management in the
fiiture: development of a strategic plan.
<www.ipmcenters.org/pmsp/pdf/MITartCherry2.pdf>. Accessed 8 May 2007.

US Environmental Protection Agency [USEPA]. 1972. DDT ban takes effect; EPA
press release. US Environmental Protection Agency. <
http://www.epa.gov/history/topics/ddt/O1.htm>. Accessed 30 April 2007.

US Environmental Protection Agency [USEPA]. 2001. Tart cherry initial benefits
assessment for azinphos-methyl and phosmet. Office of Prevention, Pesticides
and Toxic Substances. <
www.epa. gov/pesticides/op/phosmet/bead_Cherry_Tart1 .pdf>. Accessed 20
October 2006.

US Environmental Protection Agency [USEPA]. 2006. What is a pesticide? US
Environmental Protection Agency. <www.epa.gov/pesticides/about/index.htm>.
Accessed 8 November 2006.

US Environmental Protection Agency [USEPA]. 2007. Aldrin/Dieldrin. US
Environmental Protection Agency. <
http://earth1 .epa.gov/oppt/pbt/pubs/aldrin.htm>. Accessed 30 April 2007.

US Fish and Wildlife Service, National Wetlands Inventory [USFWS]. 1979-1994.

National wetlands inventory. US Fish and Wildlife Service, National Wetlands
Inventory.

103

US Fish and Wildlife Service [USFWS]. 1983. Northern states bald eagle recovery plan.
US Fish and Wildlife Service, Fort Snelling, Minnesota, USA.

US Fish and Wildlife Service [USFWS]. 1996. The endangered species act: history and
evolution ofthe endangered species act of 1973, including its relationship to
CITES. US Fish and Wildlife Service.
<http1/lwwwfws. gov/endangered/esasum.html>. Accessed 27 July 2005.

US Fish and Wildlife Service [USFWS]. 1999. Agency draft Indiana bat (Myotis
sodalis) revised recovery plan. US Fish and Wildlife Service, Fort Snelling,
Minnesota, USA.

US Fish and Wildlife Service [USFWS]. 2002. Pitcher’s thistle (Cirsium pitcheri)
recovery plan. US Fish and Wildlife Service, Fort Snelling, Minnesota, USA.

US Fish and Wildlife Service [USFWS]. 2003. Final recovery plan for the Karner blue
butterfly (Lycaeides melissa samuelis). US Fish and Wildlife Service, Fort
Snelling, Minnesota, USA.

US Fish and Wildlife Service [USFWS]. 2006. Bald Eagle: recovery, biologue, bald
eagle (Ha/invents leucocephalus). US Fish and Wildlife Service. <
http://www.fws.gov/midwest/eagle/recovery/biologue.htm1>. Accessed 16 March
2007.

US Geological Survey, EROS Data Center [USGS]. 1999. National elevation dataset.
US Geological Survey.

US Geological Survey [USGS]. 2003. National land cover database zone 60 tree canopy
layer, edition 1.0. US Geological Survey.
<http://www.mr1c.gov/mr102k_nlcd.asp>.

Whalon, M. E., B. J. Jacobson, S. D. Rawlins, D. Ricks, and S. M. Swinton. 1999.
Agriculture impact of the sudden elimination of key pesticides under the food
quality protection act. Council for Agricultural Science and Technology: Issue
Paper 11: 1-14.

Wiemeyer, S. N., T. G. Lamont, C. M. Bunck, C. R. Sindelar, F. J. Gramlich, J. D.
Fraser, and M. A. Byrd. 1984. Organochlorine pesticide, polychlorobiphenyl,
and mercury residues in bald eagle eggs—1969-79—and their relationship to shell

thinning and reproduction. Archives of Environmental Contamination and
Toxicology 13: 529-549.

Yu, C. H. No date. Gabriel biplot for principle component analysis/factor analysis.
<http://www.amstatorg/publications/jse/v10n1/yu/biplot.htm1>. Accessed 28
February 2007.

104

Zaniewski, A. E., A. Lehmann, and J. McC. Overton. 2002. Predicting species spatial
distribution using presence-only data: a case study of native New Zealand ferns.
Ecological Modelling 157: 261-280.

105

   

ttittjjtitjjtttiltI