..\~a,x‘u.
‘~.‘.‘v I
. .? x

 

,. v. ISI 3—?

2. . .. . 4 . , , f ‘ u : , 1.4.x. , . , . . . 1w.....i.n.
E. ...J .. . . ‘ . _ flat“. ; . 31% .mm.

4
M
.r: z

1

M!

.x‘ ..\. . ‘ V . .. z».

 

 

 

71.4

; i
.V'vg
a

-.

4'
I

a}:
-52;

This is to certify that the
thesis entitled

APPLICATION OF DIGITAL PHOTOGRAMMETRIC METHODS
IN THE INTERPRETATION OF LAND COVER CHANGE

ON THE COASTAL DUNES OF WARREN DUNES STATE PARK,

BERRIEN COUNTY, MICHIGAN, 1978-1999

presented by

Tiffiny Ann Rossi

has been accepted towards fulfillment
of the requirements for the

M. A. degree in Geography

 

 

(2M

 

Major Profes r’s Signature
I Z/zz/Oj

Date

MSU is an Affinnative Action/Equal Opportunity Institution

 

LIBRARY
Michigan State
University

 

 

 

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

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6/01 cJCIFIC/DateDue.p65-p.15

APPLICATION OF DIGITAL PHOTOGRAMMETRIC METHODS IN THE
INTERPRETATION OF LAND COVER CHANGE ON THE COASTAL
DUNES OF WARREN DUNES STATE PARK, BERRIEN COUNTY,
MICHIGAN, 1978-1999

By
Tiffiny Ann Rossi

A THESIS
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
MASTER OF ARTS
Department of Geography

2004

.rah ‘- ‘

. v- -_
I ‘??-\ n
I ‘upx... v

linen D
it Lake \Iic‘uga
Eli 31:39 SUPER“.
i 3 Wm of t
{6 also “PIC-reg

32:: Bea. Hs'

 

Winn a:

    

i .
“£595 at ‘an I

g

K.

1‘

“'«5 k.
«145.715. A p;

M.
h; a
\“1

Abstract

APPLICATION OF DIGITAL PHOTOGRAMMETRIC METHODS IN THE
INTERPRETATION OF LAND COVER CHANGE ON THE COASTAL DUNES OF
WARREN DUNES STATE PARK, BERRIEN COUNTY, MICHIGAN, 1978-1999

By
Tiffiny Ann Rossi

Warren Dunes State Park (W DSP) is part of an aggregation of dunefields along
the Lake Michigan coastline and is also one of the most visited state parks in Michigan.
Previous research on Lake Michigan coastal dunes discusses the link between vegetation
and dune stability. Land cover in WDSP was analyzed using digital orthophotography
in the context of this research. The possibility of human impact on coastal dunefields
was also explored by comparing land cover changes in a high-use area of the park to a
natural area. Historic aerial photography was orthorectified to produce the digital
orthophotography used in the land cover analysis.

Two methods of orthorectification were tested in this study. Extracting a Digital
Terrain Model (DTM) from the bundle-block adjusted aerial photography and using the
DTM to orthorectify the images produced orthophotos with the best positional accuracy.
Interpretation and analysis of the digital orthophotography revealed that land cover
changes at WDSP followed the model of ecological succession on Lake Michigan sand
dunes suggested in previous research. The results of this study also indicate that rates of
land cover change were variable over the period of study and may be tied to lake level
fluctuations. Additionally, this study suggests erosion is more prevalent in the high-use

area than in the natural area of the park.

Dedicated to the memory of Melissa Anne Smith,

August 22, 1977 - July 8, 2000.

iii

II"
t

I WOJAC .
fa: have aided
5:: "rate beret?-
fijrrsin, Dr

- V
0"... ‘IF TH"? m‘.

l .
.h"‘.“vu‘o 5‘.
V

 

 

ACKNOWLEDGEMENTS

I would like to express my sincere appreciation for the many individuals
that have aided, guided, or supported me. Without their help, this thesis would
not have become a reality. I especially wish to thank my advisor and committee
chairperson, Dr. Alan Arbogast for his patience, guidance, and encouragement
throughout my academic career. I would also like to thank Dr. Ashton
Shortridge and Dr. David Lusch for all the helpful and friendly technical
knowledge they provided to me while they served on my committee.

I wish to thank all of the people who provided me with materials,
technical assistance, and knowledge to pursue my research: Bob Goodwin (MSU
RS&GIS), MSU Remote Sensing and Geographic Information Science Research
and Outreach Services (MSU RS&GIS) for the Trimble ProXRS GPS units, Aaron
Krueger and Erin Kelley (field assistants), Scott Drzyzga, Scott Stevens (USGS
OSL), Phil Ross (USACE), the rangers and staff of WDSP, Sherman Hollander
(MDNR), and Penny Holt (MDEQ).

A special thanks goes to the entire US. Geological Survey Staff in Lansing
for all the support they provided to me during my time as a student trainee

Geographer.

iv

I
.o""
¥.u-“

~r~n

I
3""
at“
5
. I““

l
Le." "

‘5
~‘
51
‘ . Ab.

[Ia-‘3?

p

‘l c 1.. .l
«L. ..\ at.
«.l H A .IW
x a t.
tam ‘4U h“
I . w -
T: u:
A.“
at ..:
v.5. mlm .

 
 

 

 

 

Within the Geography Department, I appreciate the help from the
Geography Department Staff: Sharon Ruggles, Ellen Schuster, Marilyn Bria, Judy
Hibbler, Jim Brown and Wilson Ndovie. Special thanks goes to Dr. Randall
Schaetzl, graduate advisor, for the assistance he provided to me early in my
program. To the many graduate students who helped me along the way, thanks
for the ideas and inspiration, and making me laugh when I needed it.

I also wish to express thanks of a personal nature to those that provided
love, guidance or moral support to me throughout this entire process: To my
parents, who have encouraged me to succeed in school and all of life's
endeavors. To Raphy, my brother: Thanks for reminding me how to laugh. To
my ”Lansing friends” Reagan, Jeff, Wendy, Forest, Andy, Fred, Kris, Skinny
Dave, and Shelly: you're simply the best group of friends anyone could ask for. I
would especially like to thank my dear friend Wendy Tate, who went out of her
way to edit my thesis and print copies for my committee while I was in Finland.

Lastly (but certainly not least), I wish to express my profound gratitude to
my husband, Risto. Thank you for your never-ending love and patience. Mina

rakastan sinua aina.

 

munitions ..
LIST OF TABLEI
HST OI HGL’F

0.1me l
Irraduction .....
Pro-bier: '

LTD?
0

 

finial
literature Rem
Coastal S

E.

Ccas‘al l
‘n‘erra':

HLL

Lake .\‘1.

{brim
SIJdI Area m...
Late Qul

crate

 

........

(1i
v

Table of Contents

 

 

 

 

 

 

 

Abbreviations - - - -- - - _ _ viii

LIST OF TABLES - - ix

LIST OF FIGURES - - - ' ...... x
Chapter 1

Introduction - - - -- -- ........ - 1

Problem Statement ............................................................................................... 5

Purpose .................................................................................................................. 6
Chapter 2

Literature Review - - ...... 9

Coastal Sand Dunes ............................................................................................. 9

Eolian Sand Transport and Coastal Dune Formation ....................... 10

Coastal Dune Classifications and Morphology ............................................. 16

International Coastal Dune Studies ................................................................. 19

Lake Michigan Coastal Sand Dunes ................................................................ 24

Role of Vegetation in Lake Michigan Dune Environments ............. 24

Geomorphic Research on Lake Michigan Coastal Dunes ................ 28

Remote Sensing in Dune Areas ........................................................................ 45
ChapterB

Study Area - - 50

Late Quaternary Geology ................................................................................. 53

Climate ................................................................................................................. 56

Soils ...................................................................................................................... 57

Vegetation ........................................................................................................... 58

Historic Land Use .............................................................................................. 60
Chapter 4

Methods - - - - - - 62

Choice of Imagery .............................................................................................. 62

Orthorectification methods ............................................................................... 68

Interior and Exterior Orientation ......................................................... 70

Exterior Orientation ............................................................................... 80

Block Triangulation ............................................................................... 83

Orthorectification using USGS 30-meter DEM .................................. 86

DTM Extraction and Orthorectification .............................................. 88

Processing the Remaining Imagery ..................................................... 90

Classification System ......................................................................................... 94

Barren Sand ............................................................................................. 95

Barren with Isolated Plants ................................................................... 96

vi

 

 

 

Glam, 6

AWendiCec

list of Refe

 

 

 

Discontinuous Grass .............................................................................. 96
Grass and Shrubs ................................................................................... 98
Grass, Trees, and Shrubs ....................................................................... 99
Forest ...................................................................................................... 100
Interpretation Factors ...................................................................................... 102
Analysis and Mapping Procedures ............................................................... 105
Digitizing ............................................................................................... 105
Positional Accuracy Assessment ....................................................... 106
Thematic Accuracy Assessment ......................................................... 108
Measurement, Comparison and Presentation Procedures ............ 112
Limitations ........................................................................................................ 1 14
Chapter 5
Results and Discussion- -- - -- - -117
Accuracy ............................................................................................................ 117
Vertical Accuracy of the 1999 Input DEM and DTM ...................... 117
Positional Accuracy of the 1999 Test Imagery Sets ......................... 118
Vertical and positional accuracy: 1978 and 1988 imagery ............. 123
Thematic Accuracy .......................................................................................... 125
Land Cover Analysis ....................................................................................... 129
1978 Land cover: General Discussion ............................................... 130
1988 Land cover: General Discussion ............................................... 133
1999 Land cover: General Discussion ............................................... 136
Overlay Discussion .......................................................................................... 139
1978 - 1988 Visual Overlay Analysis ................................................. 141
1978 - 1988 Overlay Analysis: Quantitative Results ....................... 143
1988 — 1999 Visual Overlay Analysis ................................................. 148
1988 - 1999 Overlay Analysis: Quantitative Results ....................... 151
A comparison: 1978-1988 and 19884999 .......................................... 156
Possible Factors Influencing Land Cover Changes ......................... 158
High-Use Area Versus Natural Area ................................................ 166
Human Impact in the High-Use Area ............................................... 177
Chapter 6
Conclusions and Recommendations - -180
Appendices 187
List of References 194

 

vii

 

-z-"\.,9l’ n35: .

 

I
‘i-A_ ;( ,f _— -

 

:3 Vears 565..

“fin-O

fix.'}boevor
J ',_.y.;. A‘o
"u '- '- -
Vs?"
dé-‘u “Ahai A
P" e
_\ "W?’1"E"
\n-r ', .5“. a...
b A

 

N93 33‘. Xcri
XSTII‘D 88' NO

I _ .
“:3- Ntona

\‘H'w

“‘3 i: \Qm

‘-.t.

Abbreviations

B.P.: Years Before Present

DEM: Digital Elevation Model

DGPS: Differential Global Positioning System

DTM: Digital Terrain Model

GIS: Geographic Information System

GPS: Global Positioning System

GCP: Ground Control Point

INS: Inertial Navigation System

MDNR: Michigan Department of Natural Resources
NAD 83: North American Datum of 1983

NAVD 88: North American Datum of 1988.

NED: National Elevation Dataset

NDVI: Normalized Differenced Vegetation Index
NSSDA: National Standard for Spatial Data Accuracy
PDOP: Positional Dilution of Precision

PMT: Post Marine Transgression

RMSE: Root Mean Square Error

TM: Thematic Mapper

USGS: US. Geological Survey

WDSP: Warren Dunes State Park

viii

 

I
file 3:1. V856"
T114241. Suzi."
Tile 4:. - mi
labia 43. Ed: ‘
Ill‘lt 4'4. Eda]
labia 4:5 5:221
.355
Isle 4t. Lancl
Itble 3:1. Cak‘

Ft

 

at "'1
Title 5'2. Cad

'0 w

5".

 

Title 53. 5;?
link 3:4. Pass;A
Table 5-5. Sari
Izble 5:6. PCS;-
Iil‘it 3:7. Err!
lib]: 53. Iotj

. --

4”,:

Table 5.9. 5;."
Are.

LIST OF TABLES

Table 3:1. Vegetative species found within Warren Dunes State Park ................... 59
Table 4:1. Summary of Aerial Imagery Used in the Investigation .......................... 66
Table 4:2. Scanning and ground resolution for imagery in the study area ............ 68
Table 4:3. Fiducial mark calibration errors ................................................................. 73
Table 4:4. Fiducial mark calibration error for the 1978 and 1988 imagery ............. 91
Table 4:5. Summary of vegetation class, area, number of polygons per

class, and number of sampling points per class .................................... 111
Table 4:6. Land cover change categories and overlay classifications ................... 1 14
Table 5:1. Calculated differences between images and surveyed values

at checkpoints (1999 imagery) ................................................................... 119
Table 5:2. Calculated differences between images and surveyed values

at ground control points (1999 imagery) ................................................. 119
Table 5:3. Summary Statistics for Error Data of the 1999 imagery ........................ 120
Table 5:4. Positional Accuracy of the 1999 orthoimagery ....................................... 123
Table 5:5. Summary statistics for error data (1978 and 1988 imagery) ................. 124
Table 5:6. Positional Accuracy of the 1978 and 1988 orthoimagery ...................... 125
Table 5:7. Error Matrix for 1999 Land cover Classifications .................................. 126
Table 5:8. Total Number of Polygons and Total Area for Each Class:

1978, 1988, and 1999 .................................................................................. 133

Table 5:9. Summary of Largest Percentage Land Cover Changes by

Area ............................................................................................................... 175

ix

 

figure 1:1. 513'.
Firm 112- SW?
P?

I'm: 2:1. I)»
rim 2.2. it.»
Figure 2.3. Ca;
Igm 2:4. 9.3
Frat 3:1. Ma

“A

“Act
7-“- “-1 '9'.”
i153! )uo- u\\
w
. 4..
Y'I
Fight! 4:1. .....‘

q r
10-.

Egan 4:2. Ret
Him 43. Co:

 

 

“We 4.9. p:-
Eture 4:10, C
Figure 4:11, p

 

LIST OF FIGURES

Figure 1:1. State parks and coastal dune occurrence ................................................... 2
Figure 1:2. Swimming beach and Pike’s Peak dune at Warren Dunes State
Park ................................................................................................................. 4
Figure 2:1. Distribution of major dune coasts of the world ...................................... 10
Figure 2:2. Mechanisms of eolian sediment transport ............................................... 12
Figure 2:3. Coastal profile with incipient dunes ........................................................ 15
Figure 2:4. Diagram of coastal dune types .................................................................. 17
Figure 3:1. Map showing Warren Dunes State Park and study areas
labeled ”High Use Area” and ”Natural Area” ........................................ 51
Figure 3:2. IKONOS image of WDSP depicting a blowout near
Mt. Randall, 2001 .......................................................................................... 52
Figure 4:1. Illustration showing fiducial mark calibration in ERDAS
IMAGINE Orthobase .................................................................................. 72
Figure 4:2. Recommended ground control point distribution ................................. 76
Figure 4:3. Control and checkpoint distribution with respect to 1999
Imagery ......................................................................................................... 77
Figure 4:4. Illustration showing point measurement window in ERDAS
IMAGINE Orthobase ................................................................................. 82
Figure 4:5. Control and checkpoint distribution with respect to 1978
Imagery ......................................................................................................... 92
Figure 4:6. Control and checkpoint distribution with respect to 1988
Imagery ........................................................................................................ 93
Figure 4:7. Unvegetated dune depicting the barren sand classification
with discontinuous grasses in the foreground ....................................... 95
Figure 4:8. Photograph depicting the barren with isolated plants class ................ 97
Figure 4:9. Photograph depicting the discontinuous grass classification .............. 98
Figure 4:10. Grass and shrubs land cover class ........................................................ .99
Figure 4:11. Photograph illustrating the grass, trees, and shrubs
‘ vegetative cover class .............................................................................. 100
Figure 4:12. Photograph depicting the forest land cover class .............................. 101
Figure 5.1. Graph showing overlay percentage of unchanged land cover
by overlay classification, 1978-1988 ........................................................ 145
Figure 5.2. Graph showing overlay percentage of successive land cover
by overlay classification, 1978-1988 ....................................................... 146
Figure 5.3. Graph showing overlay percentage of regressive land cover by
overlay classification, 1978-1988 ............................................................ 147
Figure 5.4. Recession of forest cover as seen at areas H and I
(Appendix D) ............................................... , ............................................. 149

 

figure 325. G!
' cc
bl

first 5:7. G:
cc

figs: 53. F;
figure 5.9. L‘
1‘?

firm 5.10. :

Free 5211. 3
figure 5:12. .
Figure 5:13. I
figure 5:14. 4
figure 5:15 1

figure 5:16. T
Figure 5:17. T
Figure 5:18.

Hyatt 5:19.

Fsure 520.

Figure 5:5. Graph showing overlay percentage of unchanged land

cover by overlay classification,1988-1999 ................................................. 152
Figure 5:6. Graph showing overlay percentage of successive land cover
by overlay classification, 1988-1999 ........................................................ 154
Figure 5:7. Graph showing overlay percentage of regressive land
cover by overlay classification, 1988-1999 ............................................. 155
Figure 5:8. Forest recession 1978-1999 at (area I, Appendix E) .............................. 156
Figure 5:9. Land cover change as percent of entire classified area:
1978-1988 vs. 1988-1999 ............................................................................. 158
Figure 5.10. Average annual precipitation near the study area,
1978-1999 .................................................................................................. 160
Figure 5:11. Number of reported storms in Berrien County, 1978-1999 ............... 161
Figure 5:12. Average annual lake levels for Lake Michigan, 1978-1999 ............... 164
Figure 5:13. Estimated Park Visitor Numbers, 1978 to 1999 .................................. 166
Figure 5:14. Blowout healing from 1978 to 1999 ...................................................... 168
Figure 5:15. Ikonos image showing slip face of dune (Area A,
Appendix E) ............................................................................................. 170
Figure 5:16. Slip face of a dune (Area G, Appendix E) ........................................... 172
Figure 5:17. Slip face of a dune (Area 1, Appendix E) ............................................. 172
Figure 5:18. Slope at Pike’s Peak showing erosion (Area H,
AppendixE) .............................................................................................. 173
Figure 5:19. Land cover change as percent of entire classified area:
High-Use Area vs. Natural Area .......................................................... 174
Figure 5:20. Central parking lot and beach house in relation to areas of
regressive and successive land cover ................................................... 179

xi

 

 

""':'~ __ -—

335 =re es

9.1”..1... 5"-
undsuuuug.‘ .

' ‘-. r
‘9!» .n.' . -
\‘A V 5‘ \
-"0m. p v

o

u

u., I!

“"“‘.n~ ,.
’ A

"UH-Mucyhan ‘

n h.
{W‘U'Jw T

 

 

Chapter 1

Introduction

Lake Michigan may feature the largest aggregation of freshwater dunes in
the world (Buckler, 1979; Fig. 1:1). These coastal dunes are transient, dynamic
landforms that respond to environmental change in complex ways. Coastal
dunes are especially sensitive to variations in sediment supply, sea or lake-level
fluctuations, and changes in land cover. They serve a critical role in coastal
stability by storing and receiving sand blown from adjacent beaches.
Additionally, coastal dunes also supply sand back to the beaches. This continual
exchange of sediment between beaches and dunes is important in a coastal
system, as it supplies fresh sediment and nutrients for vegetative growth on
dunes (Carter, Nordstrom, and Psuty, 1990).

Due to Michigan’s humid climate, dense vegetation covers many coastal
dunes (Roethele, 1985). This vegetation effectively stops and traps eolian sand,
promoting dune stabilization (Woodhouse, 1978). If the protective land cover is
removed or disturbed by natural or anthropogenic causes, bare sand is
susceptible to eolian transport (Goudie et. al, 1999). This exposed sand can
subsequently bury anything in the path of its migration (Wilson, 1979, 2000).

Several state parks lie within dunefields along the Lake Michigan
coastline, providing numerous recreational opportunities for Michigan residents

(Fig. 1:1). Recreational use of these parks ranges from day-use to camping and

-k’

.44

/-K€@@@@@©@@

 

 

 

 

 

 

 

$ !W 84 1W
_ 45 ° N 46 ’ N .—
_ 44 , N 44 ° N ..
-42 ° N 42 . N-
0 so Miles
86 °W [I] 5'0 lGIometers 84 'W
I l
® Indiana Dunes @ Duck Lake ® Charlevoix
® Warren Dunes ® Silver Lake 69 Newegon
@ Grand Mere ® Charles Mears @ Tawas
® Van Buren @ Ludington @ Albert E. Sleeper
@ Saugatuck ® Orchard Beach ® Port Crescent
@ Holland Leelanau ® Lakeport
Grand Haven Fisherman’s Island
® PL Hoffrnaster Petoskey
@ Muskegon Wilderness

. Figure 1:1. State parks and coastal dune occurrence. From Loope and
Arbogast (1999) and MDNR (2003).

 

yaw-l

 

l

v
,.~v:"‘ 9"”8.‘
aunt‘u‘ .-~

I
an." firW‘I’"
”germ-w”
‘ O

V

I
n-r’w‘ : ‘5.""‘
.t3.- \Lneu

(ire r‘:

ushs r5

V ' re
“Cc!“ . ‘
~hu.é‘3n bed

N
ew- ’ . _
5.1.53.1) arm
5!. ‘ _' ‘
‘-~?3-'7-1I.CE'S.

A

Ffl‘~~"" ‘1
*‘wk. LE :1?

 

 

swimming. The day use parks (e.g. Saugatuck Dunes State Park and Grand
Mere State Park) are managed for low-intensity recreational use and generally
feature hiking trails, natural areas, picnic areas, and swimming beaches. These
parks primarily attract nature enthusiasts, birdwatchers, and hikers (MDNR,
2002) and attract less annual visitors than parks with campgrounds (Wells,
personal com., 2000).

State parks with campgrounds (e.g. P.J. Hoffrnaster State Park and
Ludington State Park) are managed for high-intensity recreational use and
generally attract high volumes of visitors for a variety of recreational
opportunities. Facilities at these parks generally include hundreds of campsites,
swimming beaches, hiking trails, and picnic areas (MDNR, 2002).

One of the most visited state parks in Michigan is Warren Dunes State
Park (WDSP; Fig. 1:2). Since the late 19705, more than one million people have
visited this park annually, with 1,511,369 park visitors in 2001 (Herta, written
com., 2002). The park is a popular recreational spot for campers, hikers,
naturalists, and hang gliders. WDSP has 180 modern dune campsites, several
picnicking areas and a beach with bathhouses and a concession stand. The park
also has six miles of hiking trails through forest and open sand-dune areas.
Many of these hiking trails cross through dune blowouts. The heavy recreational
use in this park thus destroys stabilizing land cover, consequently reactivating or

preventing stabilization of the sand dunes.

 

Figure 1:2. Swimming beach and Pike’s Peak dune at Warren Dunes State
Park (June 16, 2002).

 

 

-¢.I‘3 "'1 ‘tn'.fl\'
$4. “A|“"" -

. -
WA. 5 p: M D
2‘.’. . .L-L or

teachers iB'

05.4.5.1] . in
“Laud“ ‘ Lit

"v’nw mm
D‘ v.:.'~.a

k.

.fi'A
‘~ ‘. rimm

cob. .._
.

oiu.-.

Vienna."

 

|
' V

.t

"x

C

‘-

 

MDNR acknowledges that foot traffic and sustained westerly winds prevents
vegetation from growing in to stabilize the blowouts (MDNR, 1996).

Because vegetation plays such a crucial role in dune stability, land cover
can be monitored to detect changes in dune behavior. This can be facilitated
using remotely sensed imagery over several time periods (Fransaer, 1992). Past
researchers (Businksi, 1992; Fransaer, 1992; Hazlett, 1981; Stembridge, 1978) have
used traditional photogrammetric methods to monitor land cover and vegetative
conditions in dune environments. However, advancements in digital
photogrammetry have provided many advantages over traditional
photogrammetric methods (Chandler, 1999). Such advances include automated
tools to produce Digital Elevation Models (DEMs), a wide range of competitively
priced photogrammetric software, and less expensive PC or UNIX workstations,
as opposed to expensive photogrammetric stereoplotters (Chandler, 1999).
Additionally, production of digital orthophotography from historic aerial
photography can produce imagery to be handled on a GIS workstation for

quantitative analysis (thle, 1996).

Problem Statement

Much of the recent research regarding Lake Michigan coastal dunes has
related formation of these dunes with lake-level fluctuations (e.g. Dow, 1937;
Olson, 1958d; Dorr and Eschman, 1970; Buckler, 1979; Thompson and Baedke,

1997; Arbogast and Loope, 1999; Loope and ArbogaSt, 2000). These studies have

 

 

' r 1
y“ "n H - q
.3335“ “a: ”I
Eva-en .““I"
.....-..1. CL

Luv

'n.-fi \‘u fi‘.‘ ‘fi'
.53: .h..'....;.a.

"‘1 "- .
.r:.;>:embn.
22:23:13} ph

a 332:1 well

553335011 5

 

?
-.!I ‘ gram".
a “ '4.
L,':
U V

 

focused upon ancient processes and cover the entire lifespan of the dunes
(thousands of years). Uncertainties still exist with regard to the formation of
these dunes and the factors that influence coastal dune formation. Very little
research has been conducted within small time frames and within the period of
human record. To increase our understanding of the evolution and history of
Lake Michigan coastal dunes, more research needs to be conducted within the
period of record.

Previous studies have utilized historic aerial photography to monitor or
characterize coastal dune vegetation (Businksi, 1992; Fransaer, 1992; Hazlett,
1981; Stembridge, 1978). However, these previous studies have relied upon
traditional photogrammetric methods. Although the efficacy of these methods
has been well documented, they are difficult to implement in geographic
information system (GIS) analysis.

The literature available regarding the use of softcopy photogrammetry
and GIS in coastal dune environments is scarce. This study implements softcopy
photogrammetry to produce digital orthophotos. Since digital orthophotos can
be fully integrated into a GIS, data can be collected from the orthophotos,

quantitatively analyzed, and displayed using the GIS.

Purpose

The primary purpose of this study is to collect land cover data in a coastal

dunefield to establish whether the dunes or portions of the dune complex are

 

 

",‘.,'F W'
A“; 54-!

,..'.

It. ‘
.ucghg S;
u. :ww-uw

1
.~pn;u;4
g.‘..v, Lie¥

I

. .
unfit-(”3’s n
y.--‘IL L. $45 .4
5

3133‘s? :23:

2-.:~:'I,-
‘95-“... L5“ D \II
I
“In.
t. A
.r '3’2W‘n

nu-.-
.

an“. H,

n.,

va'lléie 3ft

‘ I

.55 ‘
dd.~fi. ‘ 4
*3“... 50 u

 

 

progressing towards ecological succession or eroding. This study will focus
upon a small, modern time scale to monitor land cover changes.

Additionally, this thesis analyzes the land cover trends in context of
resource management. To monitor the possibility of human impact on coastal
dunefields, land cover change in recreational and natural areas of a park will be
compared. These goals will be facilitated through the analysis of historic aerial
photography. One potential benefit of this study is a body of methods that the
MDNR may use to remotely monitor change in these coastal dune environments.
These methods may offer data to aid management strategies.

The dramatic variance in relief, coupled with a lack of permanent
landmarks, presents unique challenges when employing softcopy
photogrammetry to orthorectify aerial imagery of a dunefield. A secondary
purpose of the study is to test two methods of orthorectification of historic aerial
imagery to determine which method results in better positional accuracy. The
aerial imagery of the dunefield will be processed using softcopy
photogrammetric methods and will be quantitatively analyzed using a GIS.
Thus, it is intended that this research may also add to the body of remote sensing
literature that addresses coastal dune research.

Research questions that will be addressed are:

1. Which method of orthorectification produces the most accurate

imagery for vegetative change analysis on coastal dunes?

. Where has change occurred on a coastal dunefield and what types
of change have occurred, if any?

. Has change occurred at a constant rate, or has change been variable
over the period of study?

. What are the trends of land cover over the study interval in both

recreational and natural areas and how do the two areas compare?

 

51” 1"

. 4» LL.

‘ ‘14...“

Q.

1.» “IA-Infi-

. '_

"4“,.“ - V

9.1...fihg.
. -.

P -,-
l:’ 'I‘d'Ki
bu...

'. I a“.
' I
:T""I"' 0.-
low.‘ 5 “€
'

v
u-n.
lyp?

. L-‘ .
_~\‘\‘ s:4\

Chapter 2

Literature Review

Coastal Sand Dunes

An understanding of eolian processes is essential to the study of coastal
sand dunes. In this context, this section presents a background of the mechanics
of eolian sand transport and the formation of coastal sand dunes. According to
Carter (1988), sand dunes are eolian bedforms that develop where wind loses the
ability to transport sand. They are typically comprised of very well to
moderately well sorted sand-sized grains with average grain size ranging from
fine to medium (0.16-0.33 m) (Bauer and Sherman, 1999). Dunes range from
small, short-lived features to large, persistent conglomerations of dunes known
as dunefields (Carter, Nordstrom and Psuty, 1990).

In sandy coastal environments, dunes are a basic part of the geomorphic
system (Bauer and Sherman, 1996). Coastal dunes form in wide expanses above
the high water mark, and can sometimes extend up to 10 km inland (Carter,
1988). Their geographic distribution is worldwide (Fig. 2:1), but they are most
prevalent on dissipative coasts with strong onshore winds and an abundant
supply of sand-sized particles (0.2 to 2mm) (Carter, 1988; Carter, Nordstrom and
Psuty, 1990).

The necessary conditions for the formation of eolian dunes include: 1) an

extensive surface for sand deposition and dune evolution, 2) a readily available

 

 

 

 

I

-AnOV"

l..‘ 41

gears...“
‘ _ a

m" D) ‘r

‘h— 5- ud\

I
recs-race
bau‘Ji §5\‘

O

p 7

‘i

 

 

sand supply, and 3) winds of sufficient strength to entrain and transport sand in
a continuous direction for a sustained period of time. Although these conditions
are essential, they are not solely sufficient to explain the development of coastal
dunes (Bauer and Sherman, 1999). Livingstone and Warren (1996), suggest a
broader framework of coastal dune evolution that incorporates a complex mix of
eolian process, structural geology, and marine, estuarine, fluvial, slope,

pedological, ecological and cultural processes.

 

 
  
   

1 20.E 1 80. m.”

\_ ‘x.

 

" ,V-

 

 

 

 

 

 

 

 

 

Figure 2:1. Distribution of major dune coasts of the world (width of dune
fields not to scale). From Carter, Nordstrom and Psuty (1990).

Eolian Sand Transport and Coastal Dune Formation

The formation of coastal dunes is a complex process, but is essentially

dependent upon the integration of a sediment source, wind strength, a nearby,

10

 

.4
‘0
”numb nJ
D

\-;m‘“ Y‘
"“5.“ in. . ‘,

Ill”. 2‘

. . ‘
“‘9'". 4.. q

5.“. 5 “MAG,

f‘!‘ 4

334‘! .

tag '4': __
“NE.
D:._.._

4

.k

. , .

I‘15 Sr»

wave-protected site, and vegetation (Sherman and Hotta, 1990). Sediment is
supplied to coastal dunes from beaches, where waves and currents deposit sand.
When this sand becomes subaerially exposed by seasonal or climatic lake-level
changes, the sand dries and is exposed to aerodynamic processes (Carter, 1988).

Eolian sand transport results from the transfer of momentum from air to
sand (Sherman and Hotta, 1990). This energy originates from wind when it
shears against a beach or other surfaces (Livingstone and Warren, 1996).
Described in terms of shear stress, this process is closely linked to wind velocity
(Sherman and Hotta, 1990).

When gravity and cohesive forces (i.e. soil moisture) are overcome,
individual grains of sand will mobilize when the shear stress exceeds a threshold
value that is dependent upon surface roughness and wind velocity (Sherman and
Hotta, 1990). Sand movement begins with entrainment, which is characterized
by the processes of lift and / or drag. Lift can be described according to
Bernoulli's principle (1738), which states that there is an inverse relationship
between wind and air pressure (i.e. Livingston and Warren, 1996). Where high
velocity winds encounter slower moving air near the surface, a pressure gradient
is produced between air near the sand surface and air just slightly above. Grains
protruding into this zone of lower pressure are susceptible to lift (Livingstone
and Warren, 1996).

Drag, being more powerful than lift, is expressed as surface drag and form

drag. Surface drag is the friction between a sand particle and the air, resulting in

11

the rolling of sand particles. Form drag is caused by a pressure differential
between the windward and leeward sides of sand particles, culminating in the
rolling and sliding of sand particles (Livingstone and Warren, 1996).

Once entrainment occurs, sand grains migrate by these distinct methods:
creep, saltation, and suspension (Fig. 2:2). Sherman and Hotta (1990) suggest

that saltation and creep are very common on beaches and coastal dunes.

 

— WIND—

 

 

Suspension
- %— >
””
I’
’I
’/ Incoming saltating grain Saltation Ejected saltating grain
’ v — “ ‘ ~ ’ — - - ~
I ”’ ‘\\ I, ~‘\‘
I I \\ I ‘x
I, ,’ s‘ I \‘
’ I, ‘s 1’ \
’1 1’ Creep \\ ’1 ‘4‘
I I [I ‘ r - I’ ‘ \‘ I \\
I I . \.l \

 

 

 

 

Figure 2:2. Mechanisms of eolian sediment transport. Modified from
Livingstone and Warren (1996).

The pivotal process of eolian sand movement is saltation, since it drives all
of the other processes of eolian sand transport (Livingstone and Warren, 1996).
Saltation refers to the "leaping" of sand particles ejected into the air. These
ejected, airborne particles gain momentum from the higher wind velocities above

the sand surface, fall back to the ground and collide with other sand grains in a

 

w‘afl WINY'
ubLV sank. ‘e
A

I
23555319 56

01! - 3L
”La: TC. .0

parabolic trajectory (Livingstone and Warren, 1996; Gillette, 1999). The collision
of sand grains initiates a chain reaction, causing the impacted sand grains to
bounce into the air.

Creep involves coarser particles than saltation. Particles traveling by
creep move in two distinct ways. The first of these is the rolling of particles
across the sand surface as a result of impact from finer, saltating grains. The
other method is induced by gravity and results in particles rolling into tiny
impact craters caused by saltation (Livingstone and Warren, 1996).

Suspension is the condition in which entrained sand particles follow
turbulent motion (whereas in saltation, the particles do not). Although sands
rarely go into suspension, the undulating topography of dunes can induce
turbulence and help raise coarse particles into suspension. Suspension has been
found to be an important process in the formation of foredunes, where high
winds may carry coarse grains of sand inland for several meters (Livingstone
and Warren, 1996).

Sand transport is influenced by several factors, including, but not limited
to, slope, dune form, surface moisture and vegetation. Due to the effects of
gravity upon sand grain collisions, sand transport increases on downward slopes
and decreases on upward slopes (Lancaster, 1995). Additionally, as wind flows
over the surface of a dune, especially that with a sharp crestline, airflow velocity

increases, thus increasing shear and enhancing entrainment (Carter, 1988).

13

 

Surface moisture is also an important control on entrainment and
transport. Grain movement is halted by complete saturation even under very
windy conditions (Livingston and Warren, 1996). As surface moisture increases,
the threshold velocity required to entrain damp sand also increases (Lancaster,
1995).

Vegetation is an important control of sand transport and dune form on
coastal sand dunes (Carter, 1988). Vegetation restricts the movement of sand by
trapping it or by acting as an anchor to prevent mobilization. When wind flows
over a stand of vegetation, the wind encounters a false ground surface created by
the vegetation. This process creates a pool of still air beneath the false ground
surface, which is generally located around two-thirds the height of the
vegetation. As grains are blown into vegetation stands and meet these pools of
still air, the particles lose momentum and are deposited in or very near the stand
(Carter, 1988).

Coastal dunes generally develop on the beach backshore, where barriers,
such as vegetation or debris, inhibit the transportation of sand via wind (Carter,
1988; Fig 2:3). Small, pyramidal dunes form within clumps of vegetation and
shadow dunes develop behind debris. Vegetation mats reinforce the incipient
dunes, creating a hurnmocky surface. Overwash processes partially erode the
seaward side of the dunes, forming a scarped ridge. Scarping exposes more
sand to eolian transport, and creates an area favorable for sand deposition.

Vegetation stabilizes the hummocks again and new hummocks form. The

14

all. 5.4.1
‘ k.

"3""

 

j a
.-“O I

-q i ..
:;!;."C‘” U1

d- Q
“.cat I

*-v\1.u: *“ ‘

‘71
"3
:3
3.:

 

incipient dunes begin to grow and coalesce. Sand deposition shifts seaward
towards the scarped ridge and true foredunes begin to form (Bauer and

Sherman, 1999).

 

Predominant
Onshore Winds

 

Incipient Dunes ‘

Backshore Foreshore Nearshore’u ' ' Water ‘

A
'

 

 

4 AL A
r ’7

 

 

 

Figure 2:3. Coastal profile with incipient dunes (modified from Collins,
2002).

A wide range of environmental controls also influence coastal dune
development. Perhaps the two most important controls are sediment supply and
sea or lake-level changes. Over relatively short time intervals, sediment supply
is the most important environmental control. However, sea or lake-level
fluctuations become predominant controls over longer periods of time (Carter,
Nordstrum and Psuty, 1990).

When sea or lake-levels are high, coastlines are eroded by wave attack. In
this manner, vast amounts of sediment are transported alongshore and can
accumulate to form new coastal dunes or deposit in previously established
dunes. During periods of low sea or lake-level, sediment may also be supplied

from offshore deposits that become sub-aerially exposed (Carter, N ordstrum and

15

A “I...

 

 

n“
3-. .w
.233 U‘
, I
‘1;4nn

fl. 9.... -

Co
mien

A
'-'1ln\‘flh

-_.J:4 .kt’l

.59.
“N'W-‘l'.
Lu‘

\e.
a.
\
v.3"
- j .
‘r-J. lb

 

Psuty, 1990). This sediment transfer can lead to extensive mobilization of sand if

the dunes remain unvegetated.

Coastal Dune Classifications and Morphology

Coastal dunes can be broadly defined as primary dune forms and

secondary dune forms (Fig. 2:4). Primary dune forms are those that are directly
influenced by coastal processes (Livingston and Warren, 1996). Primary dunes
include small ephemeral dunes (Livingstone and Warren, 1996) and incipient
foredunes of the backshore (Carter, Nordstrom, and Psuty, 1990). Foredunes are
long, linear features with gentle, leeward slopes. The formation of foredunes is
favored when sand transport is moderate and beach grasses enhance deposition
of sand (Livingston and Warren, 1996). Foredunes commonly form in a seaward-
advancing manner, but may also form in a landward-advancing manner when in
conj unction with marine transgressions (Bauer and Sherman, 1999).

Dunes inland from the beach are considered secondary dunes (Bauer and
Sherman, 1990). Secondary dunes include established foredunes, blowouts and
Parabolic dunes. Blowouts are elongated wind hollows, depressions, troughs or
S“Vales that are often erosional features (Bauer and Sherman, 1999; Carter, 1988).
Blowouts are commonly thought to represent destabilization and reactivation of

Sand dunes usually in response to loss of land cover or an increase in sand

SuPply (Bauer and Sherman, 1999; Carter, 1988).

I6

 

 

 

 

 

513W

9
, '-. we
Jet-Jr froth:

_.W

1‘ '
~,“"

am? h .. -.

{37.5.7565 31

I
”can." "
uni-1-? _ n e U

I

" ,_. 1
‘5; J” '
vt-bana'fi“ t: A

. .4‘ V
k . ‘1‘ 411
"one. u

 

 

Blowouts begin to evolve in places where land cover is thin, forming an
initial notch in the landscape. The notch widens and deepens as sand is blown
away within the notch, forming a depositional lobe downwind. Erosion
continues and the notch grows into a blowout with a deflation hollow. It grows
until a deflation limit is reached (i.e. only coarse particles too heavy to be
transported via wind remain; or water has filled the deflation hollow; Bauer and
Sherman, 1999; Carter, 1988). Eventually, blowouts may close again through a
”healing” process (Gares and Nordstrom, 1995). Sand deposition creates a
shadow dune at the throat of the blowout, which grows and eventually seals off

the throat of the blowout, preventing further erosion (Bauer and Sherman, 1999).

- Parabollc dunes
fi (Secondary)
Blowouts
(Second: .

Foredunes
/ (Primary)

 

   

Figure 2:4. Diagram of coastal dune types. Modified from Livingston and
Warren (1996).

17

Parabolic dunes are believed to develop from blowouts that have reached
the deflation limit (Livingston and Warren, 1996). They are U- or V-shaped with
limbs that point upwind (Livingston and Warren, 1996). They are commonly
found in nested, concentric groups of dunes, appearing to have migrated along
the same path. These nested patterns of parabolic dunes are an indication of
alternating periods of dune activation and stability (Livingstone and Warren,
1996). Vegetation helps control the shape of parabolic dunes, taking root upon
the limbs where less sand deposition occurs. The vegetation stabilizes the limbs,
retarding or halting mobilization and allowing the central crest to migrate
forward (Livingstone and Warren, 1996).

Coastal dunes that are actively migrating downwind over previously
established terrain are broadly categorized as transgressive dunefields (Hesp and
Thom, 1990). Transgressive dunefields vary in size from small sheets of a few
hundred square meters to small sand seas of several square kilometers. They
commonly occur in temperate humid regions with an abundance of coastal sand
and high onshore winds and are mostly unvegetated while active (Hesp and
Thom, 1990).

The term ”transgressive dunes” may be applied to be both active dunes
and stabilized dunes where evidence shows the dunes have migrated inland
over vegetated terrain, bedrock or lagoons. Such evidence includes buried soils,
peat or tree stumps that are exposed on eroded dune faces (Hesp and Thom,

1990). The term may also be used in a generic manner, since it encompasses a

18

A... ls._

 

. 5'
73:65 '31

.lu " 1“.
unnet'w'l

.- up ”3‘
.1 4
-....u.. Mb“

5511525
:23; to :l‘.

v
‘- lv.
‘1’.“
.‘apl' A
on -

“~10;t LA\
0.55.

erss‘in i

vlks.

‘5‘.

say; c

‘»-
I3N S:
I;

ra

 

variety of coastal dune features including, but not limited to, blowouts, parabolic
dunes, and cliff-top dunes (Hesp and Thom, 1990).

Transgressive dunes are initiated by a supply of beach sand available for
eolian transport or the destabilization of pre—existing dunes. Beach erosion
associated with rising sea levels combined with an abundant supply of sand may
lead to the formation of transgressive dunefields (Hesp and Thom, 1990). Sand
supply may become abundant and available to the nearshore-beach-dune system
from the mouths of nearby rivers or from longshore drift (Hesp and Thom, 1990).
Preexisting dunes may destabilize due to changes in storminess or shoreline

erosion (Hesp. and Thorn, 1990).

International Coastal Dune Studies

Because coastal dunes have worldwide distribution, a large body of
coastal dune research exists. The majority of this research is centered upon
European, Australian and North American coastal dunefields. The context of
these studies ranges from ecological and management issues to geomorphic
research on process and evolution. The intent of this section is to provide a brief
overview of examples of coastal dune research that has been conducted in
different regions of the world.

In Australia, the eastern coastline is marked by transgressive and
parabolic dunes (Pye and Bowman, 1984). Pye and Bowman (1984) used

radiocarbon dates and paleosol development in transgressive dunefields on

19

 

 

rzrgrxqc.

um~~.»-

Australia’s east coast to determine the causes for dune building activity during
the Holocene. The results they obtained from five separate sites showed an
episode of transgressive dune instability (erosion or migration) between 10,000
and 6,000 B.P.

The timing of this instability corresponds to the postglacial marine
transgression (PMT), when sea levels rose 30 m to the present day elevation (Pye
and Bowman, 1984). The presence of well-developed paleosols indicates that
eolian deposition slowed around 7,000 B.P. as the rate of sea level rise also
declined. Their conclusion was that the rate of sea level change during the PMT
was the catalyst for dune instability from 10,000 to 6,000 B.P.

Short (1988) observed that most sandy shores on the southern coast of
Australia are backed by coastal dunes that evolved during the Quaternary. Dune
types found on the southern shores of Australia are prograding foredunes,
extensive parabolic dunes and blowouts, transverse dunes, and clifftop dunes.
The timing of dune formation on the southern coast of Australia was difficult to
establish due to scarcity of datable materials, the size of the dunes, the difficulty
of access, and multiple episodes of dune activity. Despite this limitation, Short
(1988) made correlations of the age of the clifftop dunes to similar dunes on the
eastern coast of Australia.

According to Short (1988), the first major episode of dune building on the
eastern coast of Australia corresponded to the PMT and occurred between 9,000

and 7,500 B.P. The clifftop dunes of southern Australia most likely formed

20

around the same time, when wave energy and sediment supply was high due to
the rising water levels. The clifftOp dunes likely formed sand ramps that topped
over the rocky cliffs of the southern shoreline. Once the PMT halted and
stillstand conditions developed, the sand ramps would have been eroded and
abandoned the dunes atop the rocky cliffs, far from a sediment supply (e. g. the
beach) (Short, 1988).

Other episodes of dune activity on the eastern coast of Australia occurred
between 3,500 and 2,000 B.P., and a present phase began about 1,000 B.P. The
few radiocarbon dates available on the southern coast of Australia were taken
from active dunes migrating over barriers that were formerly prograding during
the mid-Holocene. These deposits date between 2,700 and 1,500 B.P. and
represent a late Holocene phase of dune activity, roughly correlating the timing
of activity in the south with that of the east. The present dune activity on the
southern coast may, therefore, also be correlated to activity on the eastern coast
of Australia (Short 1988).

From his observations, Short (1988) developed a description of the
sequence of dune building on Australia’s southern coast. Primary dune forms,
or foredunes, probably began forming between 11,000 and 6,000 B.P. during the
PMT. Following the stillstand, sand supply diminished and foredune instability
led to the formation of the secondary parabolic dunes and blowouts.

In Europe, the majority of coastal dune research is conducted in the

Netherlands. Arens et a1 (1995) studied the flow of wind over dunes in the

21

 

 

'3 'o- Lo"!

 

. av - v_4.'
fl’.'1AU ‘.

y‘.
u ,l

Netherlands. Their research focused upon Dutch foredunes in three different
geographical locations. The Schiermonnikoog study site had low, vegetated
foredunes that had been stabilized over a period of 16 years using sand fences.
At Groote Keeten they observed high, unvegetated foredunes with steep
seaward-facing slopes. The third site at Nieuw-Haamstede had heavily
vegetated, stabilized foredunes that reached heights over 20 m (Arens et al.,
1995).

The methods employed by Arens et al (1998) included the collection of
relative wind speed and wind direction at different topographical positions on
the foredunes. Sampling in this fashion provided insight as to where deposition
and erosion was occurring on foredunes. Their research led to several
conclusions about the airflow over foredunes and its relationship to sediment
transport (Arens et al., 1995).

On dunes less than 20 m high, sand could be transported from the beach
to the foredune, even in the presence of dense vegetation. The foredune ridges
act like a sandtrap. Thus, landward transport of sand from the foredunes is
negligible. Topography is a predominant influence upon wind speed.
Maximum speed-up of the airflow increases with height of the foredune. This
results in a dramatic increase of the sand-carrying capacity of wind over
foredunes. A final conclusion drawn from the Arens et al (1995) study is that
there is a direct relationship between wind direction and relative wind speed.

Winds perpendicular to the beach cause deceleration of winds at the dunefoot

22

 

 

 

Frame

(u... a

.....L y

t
2" v a
«Tu .ea

' 32
.5:

 

and favors deposition there. This process decreases as the winds turn parallel to
the beach. Maximum speed-up also occurs with perpendicular onshore winds,
smaller with oblique winds, and absent with parallel winds (Arens et al., 1995).

A final example of international research of coastal dunes is in North
America. Orme (1990) studied the timing and causes of coastal dune activity at
Morro Bay, California. Here, coastal dune activity is divided into four main
phases: young paleodunes with a maximum age of 27,000 B.P.; older parabolic
dunes with maximum age of 4,160 B.P.; younger parabolic dunes and lobate
dunes with maximum age of 1,730; and active dunes that are less than 200 years
old. Thus, the coastal dunes in Morro Bay have a record of episodic deposition
and reactivation (Orme, 1990).

The Pleistocene deposits in the young paleodunes are most likely
associated with an interstadial interlude shortly after 27,000 B.P. The most
probable scenario for formation of the older parabolic dunes was the end of the
Flandrian transgression. At the end of the transgression, they most likely began
to form as barchnoid or transverse dunes. As sea level stabilized, sediment
supply to these dunes was probably lost and subsequent blowing out of these
dunes occurred, forming parabolic morphologies. The younger parabolic dunes
were most likely a result of erosion of the older parabolic dunes due to fires and
loss of stabilizing vegetaton (Orme, 1990).

The recent, active dunes are a result of several influences over the last 200

years, with the most intense dune activation occurring since 1947. Destruction of

23

 

 

' O

- h. ‘
>1 n ,
fiat-bud“

s I
”‘5";q' "‘
'lhl..’. ’ M i

. .

ol""“. 4"
,l.

.u.n‘..-\ ui

.-.

.91.: F398

 

E‘T‘n
‘J H
““4211
r... . J
N. “
"~¢ re:_
0
.55.
E‘I'-‘_)
' V
5’ ~\«
Eds.- !
. l,“
Ck.“

 

vegetation from fire, grazing, off-road vehicle, and military activities are
probable influences on the recent instability of this dunefield. Modification of
the Morro Bayentrance channel and the repositioning of Morro Creek is another
possible influence. Lastly, relative sea-level rise could be another possible
influence driving recent dune activation. The most relevant conclusion Orme
(1990) presented is that the impacts of natural forces are being augmented by

human activities and will present management challenges in the future (1990).

Lake Michigan Coastal Sand Dunes

The research regarding Lake Michigan coastal dunes can be broadly
categorized into two bodies of investigation: ecological and vegetation studies,
and geomorphic studies. The early research focused on the ecology of coastal
dunes and was mostly qualitative in nature. Subsequent studies were

geomorphological in nature and became increasingly quantitative.

Role of Vegetation in Lake Michigan Dune Environments

Cowles (1899) considered Lake Michigan dunes to be a unique
environment to study plant succession due to the harsh conditions for survival.
In this region, dune plants must endure temperature and sunlight extremes;
strong, sustained winds, nutrient-poor soils, and scanty water supplies due to
highly permeable sand (Cowles, 1899). Cowles (1899) studied dunes in Indiana

Dunes State Park, on the southern shores of Lake Michigan.

24

 

 

o \k“

‘. ..
l_*
he" ‘

efit “
5'“ "CA-
'~Q&.5--‘

f.)
(u

we

-<" ‘\
...l.‘

I
‘x
"I

vcoselj" (~-
-..C..‘C SI...
1

..,..,..

1
.r v.1“ it at.

5.. . ‘1. ..
“lax

{35:95 :
015

p

5")“ r
‘htug.“131_

A;
~4~¥ .

 

According to Cowles (1899), the coastal dune environment can be
described in terms of ecological stages. The most basic stage is the beach, then in
succession, stationary beach dunes, active or wandering dunes, arrested or
transitional dunes, and passive or established dunes. The beach zone is the
harshest of the environments. Plants that occupy the beach zone must endure
intense sunlight and strong, desiccating winds. The conditions for plant survival
become less severe with each successive stage. Established dunes must pass
through several stages before culminating in a diversified, deciduous forest
(Cowles, 1899).

Olson (1958a, 1958c) expounded upon Cowles’ description of dune
formation, characterizing the ecological stages as successional stages. Olson’s
dune succession consisted of three primary stages: pioneer, intermediate and
mature. According to Olson (1958a), pioneer stage communities are found on
dunes near the lake and must tolerate the constant influx of sand, while dunes
inland receive little input of sand and can support perennial vegetation (1958a,
1958c).

Olson’s (1958a, 1958c) studies concluded that the succession of plants on
Lake Michigan coastal dunes starts with the growth of American beachgrass
(Ammophilia breviligulata), which stabilizes incipient dunes so that other species of
plants may take root. American beachgrass, is tolerant of erosion and is
dependent upon sand accumulation for vigorous growth. Thus, American

beachgrass is found in areas that offer little stability and protection from wind

25

 

 

“cw-VA" a.

v
.5 H 'I'
“an... e‘I‘\ “

 

. .
anal W
‘ 'v

at;
big. -.
o L‘

33.533913

I
W:“u- _-
m: .3 a.»

e‘.-,

JOX‘,

 

.s,,‘ .

'V..v"‘?‘\‘1
. .v .

M -
u...“ 3.’ ‘x‘
..., k. C"...

I

"' .

-.v';-_‘
v-- ..‘\ -
“"htprt

(n

erosion and sand deposition. Because American beachgrass has adapted to such
harsh environments, it is associated with the formation of foredunes near the
lakeshore (Olson, 1958c).

Sand reed grass (Calamovifa longifolia) is often found with American
beachgrass and is considered a major stabilizer of eolian sand. This pioneer grass
is generally found where sand deposition rates are lower than in areas occupied
exclusively by American beachgrass or on nearly stabilized dunes. Sand reed
grass is also less tolerant to wind erosion than American beachgrass (Olson,
1958c). ‘

The presence of little bluestem bunchgrass (Andropogon scoparius var.
septentrionalis) on a dune indicates a great reduction in sand deposition as
compared to dunes inhabited by sand reed grass or American beachgrass.
Therefore, little bluestem bunchgrass indicates a dune that has been stable for a
few years (Olson, 1958c). Likewise, the presence of cottonwood (Populus deltoids)
and balsam poplar (Populus balsamifera) shrubs indicate a nearly stable dune.
These shrubs are also able to grow even if partially buried by sand (Olson,
1958c)..

Sand deposition greatly decreases or ceases with increasing distance from
the lake, allowing the growth of herbaceous perennials and woody shrubs and
trees to replace that of the pioneer plants. These plants indicate the intermediate
stage of succession and generally inhabit dunes that are mostly stable and

receive little sand deposition (Olson, 1958c). Red osier dogwood (Cornus

26

stonifera), choke cherry (Prunus viriana) and sand cherry (Prunus pumila) inhabit
dunes in the intermediate stage of succession. These shrub species establish
rapidly once sand deposition stops or slows to a few centimeters per year.

The last stage of plant succession on dunes is the mature stage, when
dunes are fully stabilized. Little or no deposition of sand occurs on stabilized
dunes. White pine (Pinus strobus) and jack pines (Pinus banksiana) become
predominant upon dunes stable for several decades or centuries, replacing the
intermediate plant communities. Hardwoods such as red, black and white oak
(Quercus rubra, Q. velutina, and Q. alba), and sugar and red maples, (Acer
saccharum and A. rubrum) replace the coniferous species once the dunes stabilize
(Olson, 1958c).

Although ecological and geomorphic research on Lake Michigan coastal
dunes continued after Olson, Lichter’s (1998) work was the next significant
research to build upon Olson’s methods and theories. His research was
conducted in northern Michigan at Wilderness State Park in Emmet County. The
dunes in his study area were primarily a sequence of dune-capped beach ridges
that formed during episodes of receding lake levels or during low lake levels
phases. The growth of vegetation on these beach ridges acts as a site for dune
formation and is also integral to secondary development of coastal dunes. The
ability of vegetation to stabilize the dunes during high lake levels and wave

erosion determines whether beach ridges retain their shoreline parallel

27

configuration or develop blowouts that will gradually form parabolic dunes
(Lichter, 1998).

Lichter focused on young beach ridges, less than 500 years old, because
vegetation changes are most pronounced on these young dunes. His research
revealed patterns of vegetation succession similar to those described in Cowles
(1899) and Olson (1958a), with the addition of timing associated with the
successional stages. The first dune ridge was less than 25 years old and was
dominated by beach grass, but other pioneer species (willow (Salix spp.) and
sand cherry) were also present. On older, inland dune ridges (55-175 years old),
beach grass declined and shrubs such as bearberry (Arctostphylos uva-ursi) and
juniper (Juniperus communis), and little bluestem bunchgrass dominated. This
vegetation was succeeded by mixed pine forest on dune ridges that were 225-440

years old (Lichter, 1998).

Geomorphic Research on Lake Michigan Coastal Dunes

Coastal dune research on the Lake Michigan shoreline has not been
limited to ecological research. Geomorphic research related to Lake Michigan
coastal dunes has concentrated on the processes controlling dune formation,
lake-level oscillations and the timing of dune evolution (e.g. Dow, 1937; Olson,
1958d; Dorr and Eschman, 1970; Buckler, 1979; Thompson and Baedke, 1997;

Arbogast and Loope, 1999; Loope and Arbogast, 2000).

28

 

7......2. .. ,.
$35.3? K
“quit: 3,:

a. 5. £4.“

. , ‘

M. 4"...
f .

~- But a

0'
3.... \r
I Q

‘Okcg‘,

 

 

 

Dow (1937) noted that many of the sand dunes of northern Lake Michigan
exist on high headlands; thus, he referred to them as ”perched dunes.” The term
perched dunes refers to eolian bedforms that overlay non-eolian sediments and
are elevated above the present lake level. This includes dunes that are perched
high above the lake on moraines and those dunes that overlay the postglacial
beaches of the Nipissing and the Algonquin Great Lakes. Dow (1937)
investigated perched dunes at Sleeping Bear Point in Empire, M1 to determine
the driving processes behind the formation of the dunes perched atop the Port
Huron Moraine.

Dow (1937) concluded that the majority of the sand supplied to the dunes
is fine sand blown up from the moraine. The fine sands from the moraine begin
their journey upward when the steep slip faces erode, exposing fresh material to
eolian transport. The coarse particles slide downward due to gravity, while the
fine particles are entrained by strong winds and deposit atop the bluff. Dow also
noted that erosion of the moraine is accelerated when lake levels are high and
when waves directly undercut the bases of the bluffs (1937).

Whereas Dow (1937) focused his research on large, perched dunes, Olson
(1958d) studied the processes involved in the development of foredunes. Olson’s
research was conducted on the southern shores of Lake Michigan. According to
Olson (1958d), foredune formation is dependent upon oscillations in lake level,

Which he predicted occurred roughly every 30 years.

29

 

-fl.
1‘ ‘

;.“;~3+I 1P1

10.ndnu‘x a I

. . 1
3:33:13

 

 

 

“t In.“
‘VAA.~.
I
N .
e A.
r«.,
A I‘.A

 

Foredune ridge formation commonly begins in the shallow waters
offshore where waves break and subaqueous ridges form. When lake levels
drop, these beach ridges become subaerially exposed. This drop in lake level
leaves wide expanses of beach open to eolian transport. An eolian cap forms on
the ridge and is colonized by pioneer vegetation. This in turn will initiate the
formation of new foredunes or widen pre—existing foredunes. When net sand
accumulation is too low to form offshore bars, beach ridges will form at the
former berm, or wave-cut cliff with a drop in lake level. These beach ridges then
become capped with eolian sand, thus forming foredunes (Olson, 1958d). As
vegetation stabilizes these foredune ridges they become preserved upon the
landscape.

A series of 90 such preserved beach ridges from Gary to Hammond,
Indiana provided a chronosequence for Olson (1958d) to determine the timing of
lake level oscillations that relate to foredune formation. Radiocarbon dating
provided a 2700-year period of time for the formation of these 90 beach ridges.
Because the beach ridges were spaced in relatively even intervals, Olson (1958d)
was able to estimate that the 30-year interval for lake level oscillations relevant to
the formation of foredunes.

After Olson’s (1958d) work, there is an extended gap in significant
research on the geologic characteristics of Lake Michigan coastal dunes until
Buckler’s work in 1979. In response to the State of Michigan Sand Dune

Protection and Management Act (Act No.222, RA. 1976), Buckler (1979)

30

 

mm
3
‘1

5.

A'. u .~ ‘ M

Fig.

.-‘ 1

~«.r~
. k

‘

:“"

“£35 a:

1‘.

established a dune classification scheme for Lake Michigan coastal dunes based
on dune form, relative relief, orientation, arrangement, and the relationship of
the dune formto the underlying material (Buckler, 1979).

Buckler (1979) identified nine dune forms in his classification scheme,
some of which are not genetically eolian deposits but were, nonetheless, desired
to be protected under the Sand Dune Protection and Management Act. The dune
forms Buckler (1979) defined in his classification scheme were parabolic dunes,
linear dune ridges, dune terraces, dune platforms, domal dunes, complex dune
fields, dune flats, marginal sand aprons, and interdune lowlands. The most
commonly identified features in his classification were parabolic dunes, linear
dune ridges, dune terraces, complex dune fields, dune flats and marginal sand
aprons. The discussion of Buckler’s (1979) research will focus upon these most
commonly identified dune forms.

Parabolic dunes are perhaps the most frequently noted dune forms in the
Lake Michigan coastal environment (Buckler, 1979). They are bow-, U-, or
hairpin-shaped with limbs that increase in height inland. The limbs of parabolic
dunes are generally oriented perpendicular to the shoreline, although at times
they are obliquely oriented. The windward slopes of parabolic dunes are
characterized by concave, relatively gentle slopes that steepen near the crest and

apex. The leeward slopes of parabolic dunes are convex and steep (Buckler,

1979).

31

 

w! ~qfi-n -

,. of". r‘
l e. .sU v
v.‘ ‘

, 0

Av.

8
av
7“

Q‘“ ‘.
(“it 'A

532910

.If

I
I-
I

A
f.
‘1

_A

~. I“)

‘1!

v

 

Linear dune ridges are long ridges of sand that are most commonly
oriented parallel to the present shoreline (Buckler, 1979). Frequently, linear dune
ridges have a gentle lakeward slope and a steeper leeward slope. However,
when ridges are found adjacent to the lake the lakeward slopes may be scarped.
Ridges may be found as singular features or in multiples with swales between
ridges. (Buckler, 1979).

A dune terrace is typically parallel to the shoreline and is bounded by a
higher bluff slope on one side (a dune form, moraine, or another terrace) and
lower relief on the opposite side (usually the beach; Buckler, 1979). The surface
is usually level, but may be somewhat hummocky. The terrace may have gentle
to rather abrupt slopes and is generally less than 15 feet (4.6 meters) above the
lower level. Dune terraces may also be somewhat irregularly shaped, especially
if associated with the margin of another dune type (Buckler, 1979).

Complex dune fields are generally characterized by hummocky, chaotic
terrain where the dunes lack orientation (Buckler, 1979). Slopes may vary from
gentle to steep and the relief may appear undulating, or rugged. Complex dune
fields may be transitional zones between two different dune types (Buckler,
1979).

A dune flat is a smooth, gentle or horizontally sloped depositional surface
(Buckler, 1979). Dune flats are not specifically eolian features and largely apply
to sandy deposits of lacustrine origin (Buckler, 1979). Marginal sand aprons are

transitional zones aldng the landward side of a sand dune area (Buckler, 1979).

32

 

. - 16m
:‘Pa: “ .‘u

.....4' n
..J....u "

Val

s....... .- W
p

- A.“ - ,,

l
‘ 1
3921.: L'-
1...'_: ..
““sakser l 4
32273111:

3.;C-Cer,

n
.fi‘Q.’
‘kIT‘fiu.

0,."

\{L

 

Dune sand may or may not be continuously dispersed over the surface. In some
areas it may only be a thin deposit and in other areas it may be deflated.
Generally, therelief in marginal sand aprons is not determined by dune forms
but by the underlying material. However, some isolated dune types might be
found on marginal sand aprons (Buckler, 1979).

Buckler (1979) also delineated ”barrier dunes,” which are additional
features protected under the Sand Dune Protection and Management Act of 1976.
Because the legal definition of a barrier dune (in Michigan) is ambiguous,
Buckler (1979) decided not to include it in his dune form classification scheme.
Rather, he listed the barrier dune classification as its own separate entity
(Buckler, 1979). According to Buckler (1979), a barrier dune is a sand dune
formation that separates the shorezone from the inland environments.

Buckler (1979) believed that the barrier dunes in Michigan are the largest
collection of freshwater dunes in the world. He also noted that once these
features are destroyed, they will not be replaced under current climatic and
geomorphic conditions (Buckler, 1979). According to Buckler (1979) these barrier
dunes mostly formed during the Nipissing glacial lake stage and also while the
Nipissing lake levels fell.

After Buckler’s (1979) investigation, the geomorphic research on Lake
Michigan dunes was not actively pursued again until the 1990’s. The research in
the early 1990’s used beach ridges as indicators for prehistoric lake level

fluctuations. This work built upon Olson’s (1958d) study of foredune ridges,

33

1

-.~—_ v77

 

'- 0
Oz‘vfi-D 'F
AdUau' ‘.

~.;; 36?.

337N983

. .i

p.
t I

. I
‘-
V ‘10“: ,
‘\e .|

 

NM ..

5‘ut
.

CL .

’C" -
“‘2”

.1

“I. ' .

“HRH-
. .

M1
V.

which are beach ridges with eolian caps. Thompson (1992) began this research in
his study of the Toleston Beach strandplain of southern Lake Michigan. The
strandplain is comprised of more than 150 beach ridges. Preserved within this
series of beach ridges is one of the longest records of late Holocene lake level
fluctuations in the Great Lakes (Thompson, 1992).

Thompson (1992) used vibracores to obtain material suitable for
radiocarbon dating in the wetland swales between beach ridges. This method
was performed under the assumption that the wetlands formed in the swales
between beach ridges, shortly after the beach ridges formed lakeward of the
swales. He also used vibracores from within the beach ridges to find foreshore
deposits associated with prior lake levels and to determine the elevation of these
deposits (Thompson, 1992).

Thompson’s findings enabled him to create lake level curves for the
southern Lake Michigan basin that extended back about 4,000 years. These
curves revealed three scales of quasi-periodic lake level fluctuation. The shortest
term and smallest scale fluctuation was approximately 31 years with an elevation
fluctuation of 0.5 to 0.6 meters. This short cycle supports Olson’s (1958d)
roughly 30-year lake level oscillation. Additionally, Thompson’s (1992) lake
level curves showed an intermediate scale fluctuation with an occurrence of
roughly 151 years and an elevation fluctuation of 0.8 to 0.9 m. The largest scale
fluctuations were found to occur at 500 to 600 year intervals and with an

elevation range of 1.8 to 3.7 m (Thompson, 1992).

34

 

q
'5‘ 37,4
;.- .Ȣ W

,.

W H‘ ‘

2:...“
L.

.
.4..;— ~

”
.‘ «an -C¢ A

.
r-”--
.L....C..
.
’4‘
.0 ”"i‘

""‘A‘ the:
v

P
h ‘
: V“
t-klvl.. ‘fig
A

 

 

Moreover, Thompson (1992) suggested that though lake levels followed a
general trend of decline after the Algoma highstand (3200 B.P.), there were long-
term high stands at about 2300, 1700, 1175 and 600 B.P. The shoreline of
southern Lake Michigan during these time periods was undergoing
progradation, or progressive, lakeward deposition, while on the southeastern
shores dunes were developed and building (Thompson, 1992).

Thompson and Baedke (1995) tested Thompson’s (1992) results in the
northern portion of the Lake Michigan basin at the Thompson Embayment beach
ridge complex near Manistique, MI. Using Thompson’s (1992) methods,
Thompson and Baedke (1995) found results in northern Michigan to corroborate
the three scales of quasi-periodic lake level fluctuations. However, the results
differed in that the highstand around 1700 B.P. caused erosion of beach ridge
profiles in the northern part of the basin at the Thompson Embayment, whereas
no erosion occurred in the south at Toleston Beach. Thompson and Baedke
(1995) concluded that this discrepancy was due to differences in sediment supply
between the two regions.

Sediment supply along the southern shores of Lake Michigan is high
because this region receives input from both the eastern and western shores of
the lake. The southern portion of the lake is therefore a sediment sink. Thus, the
high sediment supply offsets the landward progression of the shoreline as waters

rise. The Thompson Embayment in the north received less sediment flux than

35

 

I .
‘q... ,
.de aaui A A

hard

“rigmin
hybu‘aa .A.

9‘“. 9“
,.
't-L- 35x.

1

>655: me

Q 9

(,4
(I!
It)

\-

.\f

t7£~s
" m
L“c JA‘J

Fj‘e.
«rem c

T
i. r. :C? 33?;
l V

the Toleston Beach in the south, and thus, the rising water levels eroded the
beach during this highstand (Thompson and Baedke, 1995).

Subsequently, Thompson and Baedke (1997) continued this work at five
beach ridge complexes elsewhere in the Lake Michigan basin to determine long-
term patterns of each ridge development and the timing of high lake levels.
Similar methods to the previous studies were used, but a least-squares regression
was used to balance extreme outliers in the radiocarbon data that may be due to
contamination errors. This study links formation of small beach ridges to ~30-
year oscillations in lake-level. Groups consisting of approximately four to six
beach ridges were found to have developed over a longer term, on the order of
~150-year fluctuations of lake-level (Thompson and Baedke, 1997).

More recent research has shifted from foredune and beach ridge research
to the massive, transgressive dune complexes that so prominently mark the
eastern coastline of Lake Michigan (Arbogast and Loope, 1999). These dunes are
consistent in size and form as perched dunes (e.g., Dow, 1937), but have formed
on topographically lower lake plains associated with the proglacial lake history
(Arbogast and Loope, 1999). These lake-terrace dunes were believed to have
largely formed between 6,000 and 4,000 B.P. (Hansel et al., 1985) during the
Nipissing transgression of ancestral Lake Michigan (Dorr and Eschman, 1970;
Buckler, 1979). However, this hypothesis had not been systematically tested
until Arbogast and Loope (1999) conducted research on lake-terrace dunes along

the shore of central lower Michigan.

36

 

3:

«e ”"71
,'..C-4 .VU-r

\'_~v “‘Lfl“

_ g.wxuu

. ‘-
‘T:&: F'AJ

q .
ganai-Q'

.14th;

'n ~rm-
U0p500e3\
..

I

it

3, . .v -
:1: :18 ill

 

Arbogast and Loope (1999) used stratigraphical and radiocarbon dating
methods to determine the approximate timing of dune evolution at four sites:
Nordhouse Dunes (near Manistee, MT), the Nugent Quarry and Jackson Quarries
(near Muskegon, MI) and the Rosy Mound Quarry (near Grand Haven, MI).
Radiocarbon dates were obtained from buried soils formed within the
uppermost lacustrine surfaces found at all four sites. The radiocarbon dates
suggested that the onset of dune building was not concurrent between the sites.
At the northern-most site, Nordhouse Dunes, these soils were buried by eolian
deposits between 4,900 and 4,500 B.P. The cause of dune migration at this site
was likely the result of rising waters during the Nipissing transgression
(Arbogast and Loope, 1999). Rising lake levels would have destabilized the
lakeward face of the lake terrace in a manner consistent with the perched dune
model (e.g. Dow, 1937, Marsh and Marsh, 1987).

At the remaining sites, radiocarbon ages of buried soils found within the
uppermost proglacial lake sediments suggest that they were buried by eolian
sand after the Nipissing transgression and highstand. At the Nugent and
Jackson Quarries, the lake-terrace soils were buried by eolian deposits between
4,300 and 3,900 B.P. Dune building here began as lake levels fell from the
Nipissing highstand, but may be associated with bluff destabilization at a peak
within the regression (Arbogast and Loope, 1999).

Burial of the lake-terrace soil at Rosy Mound Quarry occurred between

3,300 and 2,900 B.P. Dune building here correlates nicely with a peak after the

37

 

H .a
.le'

5' ..
be

ii

 

I ‘4
{11
l

r—w

 

It

a

 

Nipissing, the Algoma phase (Larsen, 1985), and thus probably occurred due to
bluff destabilization and increased sand supplies caused by rising lake levels
(Arbogast and Loope, 1999).

Arbogast and Loope (1999) also concluded that the models for dune
evolution and behavior relative to lake levels was poorly understood and needed
further exploration. In 2000, Loope and Arbogast continued their research on
Lake Michigan coastal dunes. This research integrated the foredune and perched
dune models of dune building to explain dune behavior on the eastern shore of
Lake Michigan. Additionally, this research further established a link between
eolian activity and lake level oscillations developed by Thompson and Baedke
(1997). Most importantly, this study tested the hypothesis that dune building on
the eastern Lake Michigan shoreline is related to the Nipissing transgression (i.e.,
Dorr and Eschman, 1970; Buckler 1979).

Loope and Arbogast’s (2000) study area was comprised of 32 localities
spanning over the eastern coastline of Lake Michigan. Seventy-five buried soils
were identified throughout the study area and sampled for radiocarbon dating.
The radiocarbon dates provided an approximate time of burial of these soils by
eolian sediments. Of these samples, 64 had calibrated radiocarbon ages ranging
from 4750 and 110 B.P. and were included in their analysis. Eleven of the soils
dated as modern (Loope and Arbogast, 2000).

Loope and Arbogast (2000) concluded that dune building on the eastern

Lake Michigan coastline was episodic during the late Holocene and alternated

38

rm‘J

 

. .m- '16"
"out 7 ‘
i

. "fl "
:18. 3, c.

‘.

. .A¥'\ \
3...! 4.46..

 

9 u
A
use _‘
5
:“e. .4 .
“has HG.

I..“

Q ‘-\-‘;_

a.. k;
n

“‘3‘ I -
“HE‘I'AAZ

 

 

with periods of stability as evident from the presence of multiple paleosols
interspersed between eolian sediments. The timing of soils burial also shows
that a majority of dune building is linked to post-Nipissing lake level peaks.
Moreover, 75 to 80% of the sand volume in a portion of their study area overlaid
soils dated at 1,500 B.P. or younger.

Loope and Arbogast (2000) concluded that dune building at the Platte Bay
portion of their study area between 3,500 and 1,100 B.P. began as a series of
beach ridges (e. g. Thompson and Baedke, 1997); these beach ridges formed in
accordance with Olson’s foredune model (Olson, 1958d). When lake levels were
low, foredunes capped many of the beach ridges (Loope and Arbogast, 2000).

Radiocarbon data from the Platte Bay area showed that during subsequent
~150-year peaks, the beach ridges were eroded by waves and scarped the ridges
into lakeward-facing bluffs. Perched dunes formed in the lee of the bluffs as
waves continued to erode the bluff. This suggests that dune building associated
with the ~150-year peaks can be explained in accordance to the modified perched
dune model (Dow, 1937; Marsh and Marsh, 1987; Loope and Arbogast, 2000).
This model suggests that on proglacial lake—terraces, dune building occurs when
lake levels are high and waves erode beach ridges, forming a lakeward-facing
bluff. Sand dislodged from the bluff face via wave action is carried with the wind
to deposit on the lee of the bluff (Loope and Arbogast, 2000).

In 2001, Lepczyk conducted research on coastal dunes at Petoskey State

Park, outside the northern limits of Loope and Arbogast’s (2000) study area.

39

F”"

 

 

58 SE".

 

 

Lepczyk (2001) used radiocarbon dates obtained from organic material found in
buried paleosols and optically stimulated luminescence (OSL) dating of eolian
sands to establish limiting ages of the evolution of dunes within her study area.
Lepczyk (2001) identified five geomorphic units in the park, with each unit
successively located lakeward of the latter unit in a matter consistent with a
transgressive dunefield.

The easternmost geomorphic unit was the Lake Terrace. The elevation of
these sediments corresponds with Nipissing lake levels. These sediments were
found under the next geomorphic unit, the massive parabolic dunes (Lepczyk,
2001). The results from Lepczyk’s (2001) study reveals that dune building of the
massive parabolic dunes began around 4800 to 4400 B.P. as lake levels rose
during the Nipissing II transgression. OSL dates collected from the uppermost
unit of eolian sand suggest that dune building continued until approximately
2060-2300 B.P. at the southern end of the dunefield and 980-1080 B.P. at the
northern end of the park (Lepczyk, 2001).

The third geomorphic unit identified by Lepczyk (2001) in Petosky State
Park was the inset dunes. Field observations and radiocarbon dating suggest
that this unit is formed upon shorezone, lacustrine sediments from the Nipissing
II trangression. As lake levels dropped, lacustrine sands and an eolian cap were
deposited over the shorezone deposited. A soil formed at around 28002700 B.P.,
indicating a period of stability that corresponds with the Algoma transgression.

The soil was buried by eolian deposits, indicating that dune building had been

40

9’ *"3‘
lb“ val

r“‘1

O ‘
n“ '1‘”
.539 AH

.
”tn- ‘9

rhlaa. it.

 

.JZ. \ l '
l)...
(1.-
,-.- .
'44
S.)

”Whm- '
“he's—4.4“

 

' l
VCAI

 

reactivated at the end of the Algoma transgression. An OSL date suggests that
dune building ceased in this geomorphic unit at about 1750 B.P. (Lepczyk, 2001).

The fourth geomorphic unit identified by Lepczyk (2001) had no buried
soils to sample for radiocarbon dating. However, the elevations of the
underlying lake sediments suggests that these dunes formed over lacustrine
sediments from the Algoma transgression. Thus, dune building in this unit
likely occurred after the Algoma transgression. Because of the lack of datable
material, Lepczyk used results from Thompson and Baedke (1997) to provide a
minimum limiting age for the incipient dune unit of approximately 1300 B.P.
(Lepczyk, 2001).

The last unit identified by Lepczyk (2001) was the active dunes and beach
unit. Stratigraphic evidence correlates with work conducted by Thompson and
Baedke (1997) and suggests that this unit has formed within the last 1,000 years.
From radiocarbon dates obtained from buried soils within this unit, Lepczyk
(2001) determined that there were at least 3periods of eolian activity in the active
dunes and that dune building alternated with periods of stability.

The timing of dune building of these three periods was at 0-300 B.P., 470-
290 B.P. and 515-315 B.P. Lepczyk (2001) also notes that soil development in
these buried soils is weak, indicating the periods of stability were very brief. The
active unit was also problematic from a park management standpoint, as
migrating sand threatened to bury over some of the park’s campground area due

to natural and human induced processes (Lepczyk, 2001).

41

 

 

3.. q a

it? ~:
-i -,
SHORE.

""3- 5» a
A .\ ‘l
.5. 8- .
.. .

.sv‘. ‘

..~v" '-

.(r it?

.‘Yv
,.

43.x-

3‘51:

u.

‘s i
‘I
b-J 'fi‘fiz.
.N

n
‘ u

 

 

The conclusions of Lepczyk’s (2001) study suggest that dune building of
the massive dunes at Petoskey State Park occurred within the last 4,000 years,
corroborating the earlier work of Loope and Arbogast (2000). Additionally,
Lepczyk (2001) found that eolian activity within Petoskey State Park was cyclic,
alternating between periods of dune building and stability. Furthermore,
Lepczyk (2001) compared Thompson and Baedke’s (1997) lake level curves to
results from her study. The comparison suggested that these changes are most
likely related to lake level fluctuations (Lepczyk, 2001).

Van Oort et a1. (2001) also expanded the research of Loope and Arbogast
(2000) to determine the dune building of large, parabolic dunes on the
southeastern shore of Lake Michigan. Van Oort et al. (2001) studied dunes in
Van Buren State Park near South Haven, MI. In their investigation, Van Oort et
a1. (2001) mapped and classified buried soils in exposed dune faces, and collected
samples from the A-horizons of the buried soils for radiocarbon dating.

, The stratigraphic evidence and radiocarbon dates suggest that the dunes
in Van Buren State Park started building on lacustrine deposits around 5,500 B.P.
during the Nipissing high stand (Van Oort et al., 2001). Active dune growth
continued until about 3,500 B.P. Paleosol development and configuration
suggested that the dunes began as low foredunes that merged into dune
platforms (Van Oort et al., 2001). The lower sequence of alternating eolian sand
and weakly developed soils (Entisols) suggest that dune building during this

early period was episodic (Van Oort et al., 2001).

42

I“ nsfiO‘Q

..\~' “'
.

' l“.
oroen~
‘.\/

\4..U~Ab

,-
1 1. Ln»-

.‘
. use-5.5.-

‘Qp. ‘
.4 ,
‘ amt ‘7-

 

*— ‘-_._..._ .

'

F‘.‘l ‘
.Y! “n
\IL“...
C

5 .

J‘«
fi"':4'“
-.

kg“:

157‘} . '
3‘". "r I:

Above the lower Entisol sequence is a moderately developed soil (an
Inceptisol). Four samples collected across the study area revealed overlapping
radiocarbon dates in the range of 500-0 B.P. (Van Oort et al., 2001). The presence
of a buried Inceptisol at Van Buren State Park indicates a relatively long period
of stability (Van Oort et al., 2001). In the north central portion of their study area,
an Entisol underlying the Inceptisol was buried by eolian deposits between 2,145
and 1,955 B.P. These radiocarbon dates suggest a period of stability that lasted at
least 1,500 years (Van Oort et al., 2001).

An upper, alternating sequence of eolian sand and Entisols overlaid the
Inceptisol throughout the Van Buren State Park study area. The resolution of the
radiocarbon dating limited the ability to date these soils. The alternating
sequence of eolian sand and Entisols implies a alternating periods of dune
building and dune stability. Van Oort et al. (2001) were able to obtain a
radiocarbon date from the uppermost Entisol in the south central portion of their
study area, revealing a burial date between 200 and 0 B.P., with activity still
continuing at the time of publication (Van Oort et al., 2001).

From the results of their study, Van Oort et a1. (2001) concluded that the
dune evolution at Van Buren State Park was similar to dunes south of Holland,
MI studied by Arbogast et al., 1999. Both dune areas showed episodic dune
growth that began during the Nipissing high stand, followed by an Inceptisol
that marked a long period of stability, and lastly a new period of dune migration

that buried the Inceptisol and is still active at present (Van Oort et al., 2001).

43

 

 

a»

5'

 

 

However, radiocarbon date evidence suggests that the onset of the period of
stability, which formed the Inceptisol, may have been somewhat earlier in
Holland than at Van Buren State Park (Van Oort et al., 2001). Additionally, the
onset of the new period of dune migration may have begun slightly earlier at
Holland than at Van Buren State Park (Van Oort et al., 2001).

Van Oort et a1. (2001) also note that a study in Indian Dunes State Park
also indicates that dune building there began at approximately the same time
and an Inceptisol was observed there that was buried approximately 300 B.P. by
eolian sand (i.e. Gutschick and Gonsiewski, 1976). This evidence, coupled with
the evidence in Van Buren State Park and Holland, MI suggests that the barrier
dunes on the southeastern shore of Lake Michigan share similar geomorphic
histories (Van Oort etal., 2001).

The most recent research from Arbogast et a1. (2002) further tested the
results from Loope and Arbogast (2000) and Van Oort et al. (2001) in an
approximately 1-km section of coastline south of Holland, MI. For the Holland
study, Arbogast et a1 (2002) used dune stratigraphy and radiocarbon dates from
soils buried by eolian sediment to reconstruct the evolution of these dunes in
relation to lake level fluctuations.

Arbogast et al. (2002) concluded that dune building at the Holland site
was episodic and interrupted by brief periods of stability. Dune growth began
during the Nipissing high stand (~5,500 B.P.). After this initial growth, the dunes

grew rapidly between ~4,000 and 2,500 B.P. This early period of growth

44

 

out:

«'5

i

y .

‘1

K

H\
‘VNV

Iv.

H.

n k
L»

correlates best to conditions of falling lake levels. Thus, sand was likely supplied
to the dunes from the broad, exposed beach in accordance with Olson’s (1958d)
foredune model. Later periods of dune building at 3,200 B.P., 2,400 B.P., and 900
B.P. were most likely tied to high lake levels and is consistent with the perched
dune model suggested by Loope and Arbogast (2000).

Additionally, an extended period of stability was found at each dune site,
marked by a buried Inceptisol (Arbogast et al., 2002). The radiocarbon dates of
the underlying Entisols indicate that the period of stability lasted approximately
2,000 years. The Incepticol was buried between 900 and 500 B.P. (Arbogast et al.,
2002). In the past 500 years, the dunes have had episodic periods of migration
and stability, with dune migration observed at present (Arbogast et al., 2002).
These findings correlate well with the findings of Van Oort et a1. (2001),

suggesting that these dunes share a common geomorphic history with dunes in

Van Buren State Park.

Remote Sensing in Dune Areas

Most of the research regarding Lake Michigan coastal dunes has been
faCilitated through exhaustive and time-consuming field surveys and data
C01lection in the field. In many cases, remote sensing can be used to provide
quilntitative information, while reducing the amount of time spent in the field.

ReII'iote sensing is also an effective technique for many monitoring and inventory

Pm’poses (Somers et a1, 1991).

45

* ~0L| fl ‘
-riiw
SPECO'
r33...” ‘ :3
I
il
l
'1 m m

.i. .41..

6‘3”"

iota; ‘.

0. ‘5' -.r1-.

 

 

: 1.15—-

l
:57",- fl-
‘ v 4.-

I
"v

Remotely sensed data acquisition can be facilitated through the use of
aerial photography, satellite imagery, and spectral data collected from various
sensors, amongst other methods. Data collected using these remote sensing
methods may be analyzed to produce information about objects, areas, or
phenomena (Lillesand and Kiefer, 2000).

In coastal dune research, remote sensing has been applied as a precursor
to many of the field studies. For example, many researchers will collect and
examine aerial photography before conducting field surveys or to qualitatively
describe the landscape conditions for their study areas (e.g. Olson, 1958a;
Hazlett, 1981; Pye and Bowman, 1984; Jungerius et al., 1991; Lichter, 1994, 1998).

Remote sensing may also be applied to dune environments in quantitative
analyses. Stembridge (1978) used aerial infrared photography to explore
methods to detect dune growth upon vegetated coastal dunes in North Carolina.
He developed a method that used image intensity as a proxy for sand accretion.
Ammophila breviligulata grows vigorously where there is sand accretion. The
presence of both Ammophila and newly deposited sand produced areas of high
image-density on color infrared photography. A comparison of mean image
density and mean dune elevation showed a high positive correlation between
image density and sand accretion (Stemridge, 1978).

Remote sensing platforms can also provide quantitative data to study
phenomena over time. Temporal effects influence virtually all remote sensing

operations. The process of change detection is based upon the ability to measure

46

”4— “.51
I“ w V
I

.y A
of‘

"=‘ A .3.

“5d.

,..,.,3
‘Chutu
a

 

L“. Alli
I.
ll

' v
\I .h“.

l _t
K'- Ht

 

 

q‘dlb
‘ “His.

7.
px'a-
A ..~:

"ia - ,

5:. .t‘\

,V 4
WW» .
.N. .

temporal effects. Change detection uses multitemporal data to determine areas
of land cover change between successive dates of imagery (Lillesand and Kiefer,
2000).

For example, Businski (1992) used aerial photography to interpret
vegetative cover change from 1938 to 1987 on the Sleeping Bear Dunes complex
in Michigan. This research resulted in land cover maps in addition to estimates
of the acreage and fragmentation of the land cover. Businski (1992) concluded
that vegetation cover types became more fragmented through time.
Additionally, she determined that while certain land cover types lost acreage
(Forest, Continuous Grass and Unvegetated), other vegetation cover types
gained acreage (Discontinuous Grass and Grass and Shrubs) (Businski, 1992).

Additionally, Fransaer (1992) employed aerial photography in concert
with field surveys to produce differential dune vegetation maps of coastal dunes
in Belgium and the Netherlands. These maps were produced first by
interpreting vegetation cover classes from aerial photography from 1982-1991. A
differential dune vegetation map was then constructed that highlighted areas of
vegetation change. Similarly, topography was collected for the dunes during the
same time period and a differential topography map was created to show areas
of topographic change. When overlaid upon each other, the differential dune
vegetation and differential topography map revealed a direct link between

changes in land cover and evolution of topography (Fransaer, 1992).

47

rims

w:—

I
U M‘.

 

 

 

Brown and Arbogast (1999) also performed a change analysis. In their
study they used aerial photography to estimate sand flux from 1965 to 1987 on
coastal dunes in Ludington State Park, Michigan. In their research, Brown and
Arbogast (1999) used softcopy photogrammetry to produce digital elevation
models (DEM) representing different time periods. Through cut and fill analysis,
the DEMs were then employed to map and visualize topographic change, as well
as quantify volumetric changes in sand accumulation. Blowouts were observed
wherever dune elevations had substantially decreased between the two time
periods. Additionally, the cut and fill analysis revealed that sand aggradation in
Ludington State Park exceeded sand degradation (Brown and Arbogast, 1999).

Although the aforementioned research has integrated aerial photography
into the investigations, satellite imagery has also been used to quantify data in
dune environments. When Landsat 1 was launched on July 27, 1972, satellite
imagery improved the study of sand dunes around the globe (McKee, 1979). The
imagery from Landsat 1 was used in a project called ”The Morphology,
Provenance, and Movement of Desert Sand Seas.” This project facilitated
descriptions of dune form and distribution, in addition to comparing dunes in
different regions of the world (McKee, 1979).

More recent research involving Landsat imagery has also been conducted.
In their study of dunes in northwest Botswana, Jacobberger and Hooper (1991)
used Landsat Thematic Mapper (TM) imagery to analyze reflectance patterns of

land cover. Their research suggests a geomorphological link with reflectance

48

and Normalized Difference Vegetation Indices (NDVI). They also concluded
from NDVI that dune crests had higher seasonal variability than interdune
corridors, even though total vegetation density could be equal in both areas

(Jacobberger and Hooper, 1991).

49

r' 51.x-

 

I;

H“. .
4UJ\A'

In '

.nn .

h \
“ll '0

.
“b". .

LIES.“

.‘.

(I)

 

 

Chapter 3

Study Area

The study area for this investigation is Warren Dunes State Park (WDSP).
Various recreational activities at WDSP attract over one million visitors per year.
These activities include, but are not limited to: camping, swimming, hiking,
picnicking, hunting, and skiing. The park features massive, active coastal dunes,
some rising nearly 76 meters (240 feet) above the lake. Because WDSP is an
important recreational resource to the State of Michigan, one might assume that a
fair amount of research has been conducted within the park. However, very
little published research is available about WDSP.

The park is a large (7.9 kmz) park located along the southeastern shore of
Lake Michigan. It is located within Berrien County (T65, R20W) and extends 5.2
km in a northeast-southwest direction and 1.5 km from east to west. Red Arrow
Highway and Interstate 94 border the park to the east, the city of Bridgman abuts
the park to the north, and Browntown Road lies to the south of the park (Fig.
3:1).

Topographic relief within the park varies from 176.5 m (approximate
mean lake level for Lake Michigan, NGVD 1929) to approximately 245 m (USGS,
1978). The majority of the park can be described as barrier dune, according to
Buckler’s (1979) classification scheme. However, more specific

dune forms identified within the park are linear dune ridges, parabolic

50

 

Warren Dunes State Park
Berrien County, Michigan

    
      
  
  
  
  
   

EXPLANATION

————— Study Area
Road

Sand Blowout

BRlDGMAN

1.5 km/

 

MT
EDWARD

State Land

LAKE
MICHIGAN

  
  
       
    
 

% Q -
§/
\\ PIKES TOWER
PEAK HILL
SCALE

0 1000 20mm
2» 400 mm

<7.
\‘ MT FULLER 1/

l\‘ r
F

   

\.,

  

Park
Entrance

\\ / //

B ROWN TOWN ROAD
/

 

 

 

 

Figure 3:1. Map showing Warren Dunes State Park, including the ”High-
Use Area” and ”Natural Area" (Modified from MDNR, 1996).

51

dunes, dune platforms, a marginal sand apron and dune flats (Buckler, 1979).
The terrain is marked by massive, active blowouts, some extending
approximately 1,030 meters inland from the lakeshore (as measured at the
greatest extent of Mt. Randall; Fig. 3.2). The eastern portion of the park features
forested dunes with weakly developed soils and low areas behind the dunes

with poorly drained soils (Larsen et a1, 1980).

/|l\l
\/l( ll/l:l\

 

Figure 3:2. IKONOS image of WDSP depicting the extent of the blowout
near Mt. Randall and the beach in 2001.

From a management perspective, the park can be divided into two areas,
which will be referred to as the ”High Use Area” and the ”Natural Area” (Fig.

3:1). The High Use Area is a recreational area and thus experiences heavy foot

52

 

 

..
“w; 4.‘
nus. W

id’s:

i

”4..“
3C .n

a.

«6.- *o.
~

.A. ‘
1 "H
W H‘Lu

V
~.- ‘

-.~

v‘ ‘1
:b‘.lv‘
\

'5

."
L‘Ssli t‘

,

.. .

‘snp‘
it.

c-

'SI '3.

traffic, especially in the summer when the park beaches are open. The Natural
Area receives less foot traffic because it is located away from the beach areas.
This division was based upon information provided by the park manager (Hill,
Richard, Park Manager WDSP, personal com., 2000) and from the MDNR

(1996) map of the park.

Late Quaternary Geology

Fluctuations in Lake Michigan water level have played a significant role in
sediment supply to Lake Michigan coastal dunes, and thus, have been a
significant factor in the building of dunes (Arbogast and Loope, 1999). In order
to understand the formation of the dunes within WDSP, a discussion of the
geolOgic history of the Great Lakes is necessary.

The modern landscape of southwestern lower Michigan is mainly the
result of geomorphic processes since the latter portion of the Late-Wisconsin
glacial period. When the ice sheet advanced into the Great Lakes region, ice
moved via the path of least resistance into old river valleys and divided into
lobes. The ice lobes broadened and deepened these pre—existing valleys as the ice
advanced over Michigan. The lobe that enlarged the Lake Michigan basin is
known as the Michigan Lobe (Farrand, 1988).

Throughout most of the Late-Wisconsin glacial, Michigan was covered by
several thousand feet of ice (Dorr and Eschman, 1970). The last readvance of ice

into the study area occurred in approximately 14,000 B.P. and formed the Lake

53

7 .

17v ~

‘1",‘4: *v

“5

 

 

A
1"» ‘
\w

‘51

 

 

Border Moraine (Hansel et a1, 1985). This moraine occurs just outside the study
area, approximately 3.5 kilometers east (USGS, 1978) of the present Lake
Michigan shoreline (Dorr and Eschman, 1970). From that time until the mid-
Holocene, the lake levels of ancestral Lake Michigan fluctuated in response to
glacial readvance and retreat and isostatic rebound (Eschman, 1985; Hansel et a1,
1985; Farrand, 1988; Chrzastowski and Thompson, 1992).

The most prominent of these lake level stages relative to the WDSP study
area is the Nipissing Phase, which peaked around 4,700 to 4,000 B.P. (Hansel et
al, 1985). During the Nipissing phase of Lake Michigan intense wave action and
currents were probable. The Nipissing phase persisted roughly 1,000 years
before gradual outlet incision and differential uplift caused a lower lake stage,
the Algoma, around 3,800 B.P. Lake-level continued to drop to modern elevation
following the Algoma phase, with fluctuations attributed mostly to climatic
variability during the late Holocene (Hansel et a1, 1985).

Previous studies have linked the timing of dune building of the massive
coastal dunes along the eastern Lake Michigan shoreline almost exclusively to
the Nipissing transgression (6000 to 4000 years B.P.), when lake-level slowly rose
to levels above the modern lake elevation (Dorr and Eschman, 1970 and Buckler,
1979). However, recent research (Arbogast and Loope, 1999; Loope and
Arbogast, 2000; Van Oort et al., 2001; Arbogast et a1, 2002) on Lake Michigan
coastal dunes suggests that the dune building has been episodic and cannot be

linked solely to one event.

54

a
- Q'..

‘5‘“.
o
I». h
, I
.- U“

I
”a“.

 

 

.7
*A‘y.
I
«Q‘ J‘
4 .
‘03“
L
I‘_ I
{7 DA '1"
l .L' n...
4'
I
'
':
V
i ‘7.
1 “ '
. ‘11-- o».
l
.
..
’1
‘1
.
‘I
‘ l
' l-
v
.
‘ a
r
_ x‘
. .. iv
.. . . I:
"r... ..

 

IQN
I ~“ ~.

i“

llr

 

S

r

 

 

Van Oort et a1. (2001) conducted research approximately 64 km (40 miles)
north of WDSP on massive coastal dunes in Van Buren State Park. Their study
area has dunes similar to those found in WDSP. As indicated by soils buried
under eolian deposits, dune building at Van Buren State Park has been episodic
and interrupted by periods of stability. The dunes at Van Buren State Park began
to form during the Nipissing high stand, approximately 5,500 B.P (Van Oort et
aL,2001)

A lengthy period of stability followed the initial, episodic dune building at
Van Buren State Park, lasting approximately 1,500 years. This period of stability
resulted in the formation of well-developed soil. Dune building was reinitiated
approximately 500 B.P., burying the soil with eolian sand. The presence of
alternating paleosols and eolian sand indicates that dune building has been
interrupted by periods of stability. Dune building is presently active within Van
Buren State Park (Van Oort et al., 2001).

Van Oort et al. (2001) concluded that dunes of the southeastern Lake
Michigan shoreline might share a common geomorphic history. These findings
remain untested on the dunes within WDSP. However, paleosols were observed
on active, eroded dune surfaces during field surveys for this investigation. The
presence of such paleosols buried within eolian sand indicates episodic periods
of stability and dune growth. Radiocarbon dates collected from these buried

soils may reveal a geomorphic history similar to that of Van Buren State Park.

55

arr-y

 

l" G”

Ll rl'“

F3

(n-

pres:
b

I‘ ‘
”r u.
‘“ a.

‘5‘ P

"f-v-s.‘
.5‘L‘

 

Climate

The climate within the study area is humid cold, with no dry season and
hot summers (deBlij and Muller, 2000). Air temperature is moderated by
prevailing south-southwest winds from nearby Lake Michigan (Larson et al.,
1980). Humidity is also influenced by the lake, with average relative humidity in
midafternoon at 63 percent. Average relative humidity is higher at night, with
the average humidity reaching about 82 percent in the morning (Larson et a1,
1980).

The closest coastal weather station to the study area is at Benton Harbor,
MI (approximately 27.5 km north), which has been collecting temperature and
precipitation data since 1881 (NOAA, 2001). Mean monthly precipitation values
range from 110.74 mm in September to 44.96 mm in February, with a mean
annual precipitation of 948.18 mm. (NOAA, 1999). Average seasonal snowfall
for the study area is 170.18 cm (Larson et al, 1980).

Mean monthly temperature ranges from —4°C in January to 22°C in July,
with a mean annual temperature of 10°C (NOAA, 2001). The lowest recorded
temperature was on January 12, 1918 at —29°C. The highest temperature on
record was on June 25, 1937 at 43°C (Larson et al., 1980). Sunny conditions are
observed 67 percent of the time in summer and 37 percent of the time in winter

(Larson et al, 1980).

56

 

Jeane

'.Q.
1"”!‘“
.b

“~44”

Rik!

J
U D

Soils

The soils within Warren Dunes State Park are categorized within the
Spinks-Oakville-Oshtemo association. This association consists of well-drained
sandy and loamy soils on moraines, till plains, outwash plains and beach ridges.
A total of five soil series were identified by Larson et. al (1980) within the park:
Oakville fine sand, Gilford sandy loam, Pipestone sand, Granby loamy fine sand,
and Morocco loamy sand. However, the predominant soil series is Oakville,
found on the forested dune areas. The other four soils are in the lowland areas of
the park behind the dunes and outside the area of interest for this thesis.

Areas mapped as dune land by Larson et al (1980) are found adjacent to
beaches and in blowouts (Larson et. a1, 1980). This material was likely deposited
within the late Holocene and is generally very pale brown (10 YR 7—8 / 4) to light
gray (10 YR 7/ 2) (Stone, 2001). The dune land consists of actively shifting sand
with little or no protective vegetative cover. Soils have not yet developed in
these areas due to the actively blowing sand. Conversely, the stable, forested
dunes are comprised of Oakville soils. Oakville fine sands are classified as
mixed,‘mesic, Typic udipsamments (Larson et. al, 1980). The parent material for
this soil is fine, yellow (10 YR 7/ 6) dune sand and is classified as a middle
Holocene deposit, possibly related to Glacial Lake Nipissing (6000—4000 B.P)

(Stone, 2001). These are weakly developed soils with A/ E/ B/ C horizonation.

57

. .
um rr“ .

usual

E ...t
1 Luau.
‘1
3‘

 

‘l
t)
f}
2.
.J.

‘g.$—~
’55-.va

512*“
Ind“.

These soils are well drained with rapid to very rapid permeability (Larson et. al.,

1980).

Vegetation

Prior to settlement, natural vegetation upon the dunes consisted of central
hardwood forest. These forested uplands were comprised of white oak (Quercus
alba), black oak (Q. velutina) and hickory (Carya species). The open dune areas
were not noted in the original survey of 1830 (Corner et al., 1995). However,
descriptions of massive expanses of open dunes in this area at the turn of the 20th
century (e.g., Janette, 1919) may suggest that these open dunes were present
during that initial survey but were neglected by the surveyors.

Current vegetation within the parkland is typical of that of Lake Michigan
coastal dunes (Hoagman, 1994). Although the intent of this study was not to
provide an exhaustive list of vegetative species found within the park, a minimal
field survey was conducted on June 8 and 9, 2002 to present a general overview.
Table 3:1 contains a list of vegetative species that were identified within the park,

in addition to species commonly found upon Lake Michigan coastal dunes.

58

IT!

’1'"): mm

 

3.
v ?-_
“so

Table 3:1

Vegetative species found within Warren Dunes State Park

 

GRASSES:
" A grostis gigantea
' A mmophilia breviligulata
' Andropogon scoparius
‘ Calamovilfa longlfolia

COMMON NAME:

redtop grass

American beachgrass (or marram grass)
Little bluestem

Sand reed grass

HERBACIOUS PLANTS: COMMON NAME:
‘ Arenaria stricta Rock sandwort
" Artemisia campestris wormwood
‘ Asclepias syriaca common milkweed
* Cakile edentula Sea rocket
' Equisetum arvense horsetail
" Equisetum hyemale scouring rush
‘ Lithospermum croceum hairy puccoon
*Smilacina stellata starry false solomon’s seal
‘ Solidago sp. goldenrod
‘ Toxicodendron radicans poison ivy
Vitis aestivalis summer grape
' Vitis ripon'a riverbank grape
SHRUBS AND TREES: COMMON NAME:
'Acer rubrum Red maple
" Acer saccharum sugar maple
A rctostaphylos uva-ursi bearberry
' Camus stolomfera Red-osier dogwood
' F raxinus americana white ash
Juglans cinerea butternut
'Junipem communis common juniper
Juniperus virginiana Red cedar
Larix Iaricina tamarack
‘ Lin‘odendron tulipzfera Tulip tree
Pinus strobus white pine
Populus balsamfera balsam poplar
" Populus deltoides cottonwood
' Prunus pumilia Sand cherry
' Populus tremuloides quaking aspen
" Quercus alba white oak
Quercus rubra northern red oak
’ Quercus velutina black oak
Salix syrticola dune willow
" Tilia americana American basswood
Tsuga canadensis eastern hemlock
Ulmus americana American elm

" Denotes species identified in field

Source: Developed from The Acorn, 1919; Olson, 1958a; Little, 1980; Marino, 1980; Hazlett, 1981 ;
Businski, 1992; Hoagman, 1994; MDNR, 1996.

 

59

 

 

lag"

E

71") 1
. V‘ssq

Historic Land Use

Edward K. Warren, a prominent inventor and businessman from the
village of Three Oaks (south of WDSP) purchased a 250—acre portion of the dune
land in Lake Township, two miles north of Sawyer in Berrien county in 1919 (The
Acorn, 1919). This purchase took place in an age when the railroad industry
placed high demand on the sands quarried from coastal dunes. After witnessing
the destruction of dunes near Michigan City, Indiana, Warren purchased the
sand dune land to preserve it for future generations (Berrien Springs Era, 1917).
As landowners reverted their agriculturally unproductive dune land back to the
state, E. K. Warren purchased this land as a preservation measure (Galien River
Watershed Project, 2002).

In the years prior to his death, Warren placed this dune land under the
jurisdiction of the Warren Foundation. By Michigan law, a foundation allowed
the land to be dedicated to public use and care while the landowner still held the
title to the land (Janette, 1919). According to Alexander G. Ruthven, a professor
at University of Michigan in 1919, this dune land had about 3,000 feet of Lake
Michigan shoreline and had probably the largest dunes in the State of Michigan.

Professor Ruthven also had little doubt that the original vegetation
prevailed in most areas of the property (The Acorn, 1919). Early visitors to
the park describe in detail the bald sides and crest of the Great Warren Dune,

providing evidence that the massive blowouts seen today were present even in

60

o"‘-‘
. d
A ‘1'
i'. - 5
$91:
u"'..

;n.‘~~“

‘.~:-bfil

..
V“ .4

v’l- 5.;

'4'."
"rm:

 

 

1919. Visitors also describe tamaracks, elms and butternuts that were buried or
threatened by moving sand.

George R. Fox, an early director of the Warren Foundation museum
attempted to monitor this sand movement by placing a series of wooden stakes
five feet apart at the foot of the Great Warren Dune. Fox estimated that the dune
was moving inland at a rate of 1 foot per month (Janette, 1919).

Since 1938, the State of Michigan has continued to buy land adjacent to the
Warren foundation land (Galien River Watershed Project, 2002). A small,
northeastern portion of the parkland was owned by the Martin Marietta Corp., a
sand mining operation, until the State of Michigan purchased it for Warren

Dunes State Park in 1987 (Rockford Map Publishers, 1954, 1964, 1987).

61

Ia-Iit‘i

a.

‘1'”! " AT“

 

Chapter 4

Methods

In order to study the land cover changes on WDSP over time, historic and
recent aerial photographs were digitized and orthorectified to create
planimetrically correct images. The orthorectification was performed using
ERDAS IMAGINE Orthobase and Orthobase Pro, version 8.5. Land cover was
digitized from the orthophotos using ArcView GIS 8.2. Further data analysis and
cartographic work was also produced using ArcView GIS 3.3 and ArcView GIS
8.2.

For the land cover change analysis, a period of at least 20 years of data
covering at least three points in time was desired. This would provide starting
and ending points, in addition to a middle point to compare whether rate of
change was constant or varied. The time frame chosen for this investigation was
from 1978-1999. This decision was made because supporting data for that time
frame could be acquired. The following section describes the imagery and

methods used to study land cover over this time frame.

Choice of Imagery

In the past, aerial phbtography has been the most commonly used method
of remote sensing (Tueller, 1989). However, aerial photography has limited

spectral sensitivity, which decreases its usefulness in the measurement of land

62

 

e‘l”
5““:

fl“.
. 5.. ~

.
7..-
a"-

B '
h—‘

97

(7 :t
‘a.

w:
I

cover (Tueller, 1989). Today numerous satellite platforms exist to collect
remotely sensed data, including the SPOT series of satellites, the LANDSAT
series of satellites, IKONOS and QuickBird to name a few.

The LANDSAT series of satellites collected imagery of the study area
throughout the entire time period for this investigation (Lillesand and Kiefer,
2000). However, the ground resolution of LANDSAT imagery is limited to 79-
meter for Multispectral Scanner data and 30-meter pixel size for the Thematic
Mapper (Lillesand and Kiefer, 2000). Given that the changes within WDSP were
hypothesized to be smaller than the resolution of the LANDSAT imagery,
LANDSAT images were not chosen for this study.

Alternatively, the SPOT series of satellites offer better ground resolution at
20-meter ground resolution for multispectral imagery, and 10 m resolution for
panchromatic imagery (Lillesand and Kiefer, 2000). However, because the first
SPOT satellite (SPOT-1) was launched in 1986, SPOT series imagery has not been
available throughout the desired period of study.

H<ONOS and QuickBird offer high-resolution imagery and were launched
in 1999 and 2000, respectively. Ground resolution for IKONOS imagery is 1-
meter for panchromatic imagery and 4-meter limited multispectral imagery
(Lillesand and Kiefer, 2000). QuickBird offers 0.61-meter ground resolution in
panchromatic imagery and 2.5-meter resolution with limited multispectral data

(DigitalGlobe, 2003). Although these satellite platforms offer high-resolution

63

1a.llpl.9zl O. 4

Au \.
ed

”h

. m.

imagery, both have been very recently launched and cannot be used for any
long-term historical analysis prior to launch dates.

Given the difficulties with ground resolution and the questionable
availability of long-term, repeat imagery within the study area, aerial
photography was chosen as the image source for this investigation. In context of
the goals of this study, aerial photography was chosen based upon several
factors, including: scale, tonal quality, camera calibration report availability, and
date of imagery. Scale was the overarching factor. Large-scale imagery is
desired because it provides the most amount of detail. The scale should also
remain constant, if possible, over the span of the imagery to minimize
inconsistencies between the temporal datasets.

Aerial photography of the study area was available through a few
different sources, with varying temporal coverage. In this context, the archives
that were investigated were The Center for Remote Sensing and GIS Aerial
Image Archive at Michigan State University; the MDNR, Land and Mining
Resources Division; and the US. Army Corps of Engineers (USACE), Detroit
District.

The Center for Remote Sensing and GIS Aerial Image Archive had historic
aerial photography at varying scales, with a temporal coverage from 1938-1999.
At the time this investigation began, documentation regarding orthorectification
without the use of a camera calibration report was not available. The camera

calibration report contains information critical to the orthorectification process.

64

 

 

fear

a.“
,. .
atria:

in;

.t5

”ru-
an...

Ill

r-v

As a result, no imagery could be used that was older than 1967, the earliest
available coverage year for which a calibration report was available. However,
the 1967 photographic prints had poor tonal quality, with high reflectance from
the open sand areas. The next available imagery from the Aerial Imagery
Archive was captured in 1974. The scale of this photography is 1:40,000 and was
ultimately not selected because larger-scale imagery was available elsewhere.

USACE had shoreline aerial photography coverage at various intervals
between 1973 and 2001. The scale of this imagery is 1:6,000. Because this
imagery showed an abundant amount of detail, this would have been a good
imagery source. However, two factors made this imagery undesirable to the
study. First, the camera calibration reports were unavailable for the early
imagery. Second, the USACE area of interest during the aerial photography
flights was the shore zone. Ultimately, the photographic coverage did not extend
far enough inland to show some of the more interesting areas in WDSP.

The MDNR, Land and Mining Resources Division had aerial photography
of the entire state flown in 1978, 1988, and 1999. This aerial photography was
ultimately chosen because of the availability of camera calibration reports, the
relatively even temporal spacing of the flights, and the minimal differences of
scale among the photographs. Additionally, at least two of the dates collected
(1978 and 1999) were available as diapositives. Diapositives (i.e. transparencies)
are preferred over paper prints because they are less susceptible to distortions

due to print degradation over long time periods (Brown and Arbogast, 1999).

65

 

”a:
.4
.431

up

OI

'WV

“I.

I):
K)
’_\

Lastly, the dates of the aerial photography flights provided an initial time
frame (1978) and an ending time frame (1999) that spanned a period of
approximately two decades. Because initial information regarding the
movement of the dunes estimated a sand migration rate of 1 foot per month
(Janette, 1919), it was hypothesized that twenty years would be an ample amount
of time to observe landscape and land cover changes. Additionally, the middle
time frame (1988) could be used to determine whether the changes were gradual
or at varying rates. Information regarding the aerial photographs used in this

investigation is summarized in Table 4:1.

Table 4:1. Summary of Aerial Imagery Used in the Investigation.

 

 

FLIGHT ,

D ATE FILM TYPE SCALE COMMENTS

9-11-1978 CIR 1:24,000 High reflectance from sand;
diapositives; early- fall leaf-on
conditions.

6-12-1988 BWP 124,000 Good tonal quality; paper prints;
late-spring leaf-on conditions.

6-17-1999 BWIR 1:15,840 Fair tonal quality; diapositives;

late-spring leaf-on conditions.

 

" CIR: Color Infrared; BWP=Black and White Panchromatic; BWIR: Black and White Infrared.

Overlapping pairs of images were selected in order to increase the
accuracy of the orthorectification process (ERDAS, 2001). Only two overlapping
images were needed to provide coverage for the study area for the 1978 and 1988
photography. Because the 1999 photography was at a larger scale, four images

were needed to provide full coverage of the area.

66

 

In order to process the imagery using softcopy photogrammetric
techniques, the aerial photography was digitized with a UMAX Mirage II
desktop scanner using Adobe Photoshop 5.0 software. Because scanning
resolution ultimately determines the ground resolution covered by each pixel in
a digital image, a desired ground resolution was determined before the imagery
was scanned. In order to capture the detail needed to detect features on the
digital imagery such as shrubs or small trees, a ground resolution below one
meter was determined to be an appropriate pixel size.

When scanning aerial photography for orthorectification, additional
considerations include the file size of the digital image and the pixel resolution
measured in micrometers (um). Hohle (1996) notes that pixel size will determine
the size of the digital image file and thus, will determine processing time. Large
files can have excessive processing times. Therefore, choosing pixel resolution is
often a compromise between accuracy of the imagery and economy (Hohle,
1996).

Chandler (1999) recommends a scanning resolution of 40 pm for softcopy
photogrammetry applications. Hohle (1996) used scanning resolutions of both 25
pm and 30 um. Based upon this previous research, it was concluded that
between 25 um and 40 um would be acceptable scanning resolutions. Accuracy
and file size economy were considered while scanning the photographs used in
this study. Scanning and ground resolutions for the imagery are summarized in

Table 4:2.

67

«L

 

.nru

l‘

t. ‘l

Table 4:2. Scanning and ground resolution for imagery in the study area.

 

 

DATE OF SCANNING SCANNING GROUND
IMAGERY RESOLUTION (gs) RESOLUTION (pm) RESOLUTION (m)
1978 .1000 25.40 0.606

1988 1000 25.40 0.606

1999 800 31.75 0.503

 

The scanned images were saved as Tagged-Image File Format (TIFF) files
and subsequently converted into the ERDAS IMAGINE native image format

(*.img) using IMAGINE software.

Orthorectification methods

Orthophotos are photographs that have been corrected for distortions due
to tilting of the camera during the photographic survey, distortions from the
camera lens, and relief distortions (Lillesand and Kiefer, 2000). Orthophotos
display all the valuable information of a photograph, but unlike a photograph,
true distances, angles and areas can be measured directly (Lillesand and Kiefer,
2000). Today, orthophotos in digital format are widely used because they are
ideal for image interpretation and feature encoding in a GIS environment
(Lillesand and Kiefer, 2000).

Digital orthophotos are produced through softcopy photogrammetry.
Photogrammetry is the science and technology of obtaining reliable spatial
measurements of objects or phenomena from photographs (Lillesand and Keifer,
2000). Softcopy photogrammetry is applied to digital images that are stored and

processed on a computer (ERDAS, 2001). This section of the investigation will

68

discuss the softcopy photogrammetric methods used and tested to produce the
digital orthophotos for this study.

Because ground conditions were most current in the 1999 imagery, two
methods of orthorectification were tested upon this imagery. The first method
used ERDAS IMAGINE Orthobase (version 8.5) to process and orthorectify the
imagery using a USGS 30—meter DEM. The second method employed ERDAS
IMAGINE Orthobase Pro (version 8.5) to process the imagery and to extract a
localized digital terrain model (DTM) from the imagery. This extracted DTM
was then applied to correct for relief distortion.

ERDAS IMAGINE Orthobase is a softcopy photogrammetry platform
limited to producing orthophotos with the use of a pre—existing DEM to correct
for relief distortion. ERDAS IMAGINE Orthobase Pro is a more sophisticated
program that employs photogrammetric methods to extract a DTM from the
input imagery. Therefore, orthorectification can be performed using a dataset
that is customized to the imagery and study area location. Both methods employ
the same initial steps until the DEM/DTM step of the process. Thus, the
methods will be described as one and the same until that juncture. The 1978 and
1988 images were orthorectified using the method that demonstrated the greatest

horizontal accuracy, as described later in this investigation.

69

 

'I' \~'-11r.e-

 

’ Y:

‘._"
I

6}

CC

Interior and Exterior Orientation

In both Orthobase and Orthobase Pro, image orthorectification is
performed within a ”block file.” Block files require overlapping pairs of
photographs, information about the frame camera from the camera calibration
report, the interior and exterior orientations of each image in the block file,
ground control points, tie points and a DEM (Harrell, 2001).

The first step in building the block file was to add the four overlapping
images from the 1999 image set. Next, the camera model was defined using
information from the camera calibration report that establishes the interior and
exterior orientation. The necessary data from the calibration report were camera
type, calibrated focal length, calibrated principle point, fiducial mark
coordinates, and radial lens distortion coefficients.

The calibrated principle point, calibrated focal length, and fiducial mark
coordinates are used to define the image-space coordinate system (Thieler and
Danforth, 1994). The image space is the geometry ”within” the camera at the
time of image capture (ERDAS, 2001). It is based upon a 3-dimensional Cartesian
coordinate system, with the origin at the calibrated principle point (Thieler and
Danforth, 1994).

Radial distortion can cause points on the images to be displaced from their
true positions and is symmetric around the principle point (Thieler and
Danforth, 1994). The effects of the radial distortion can be approximated using a

polynomial function:

70

 

W]

(0

Ar = Kor + 1(1r3 + K2r5,
where Ar is the change in radial distortion along a radial distance r from the
calibrated principal point (ERDAS, 2001). Values for r and the three ”kappa”
coefficients were extracted from the camera calibration report.

Once these data were input to the block file, the interior orientation for
each image in the block file was established. This procedure matched the image
coordinates of the fiducial marks in each image with the fiducial mark
coordinates given in the camera calibration report. Fiducial marks are features
within the camera body that are exposed onto the photograph when the image is
captured (ERDAS, 2001). Since the imagery was digitized using a scanner, a row
and column coordinate system defined the position of each image pixel.

The calibration of fiducial marks was done manually by digitizing a point
in the center of each image fiducial mark (Fig. 4:1). This digitized point indicated
to Orthobase and Orthobase Pro the pixel coordinates associated with the
corresponding fiducial mark coordinates (measured in mm) from the camera
calibration report. Because fiducial marks from the camera calibration report are
oriented according to the data strip position, the scanned images required a 270°
rotation to match this orientation.

Once the fiducial marks were digitized from the images, Orthobase and
Orthobase Pro performed an affine transformation to match each pixel on the
image with its corresponding image-space coordinate (ERDAS, 2001). The

quality of this transformation is measured by its root mean square error (RMSE).

71

 

 

 

 

 

 

 

 

 

 

 

R: L 1" PW [MAL A);
8. l I all.“ ....

manna ESL}.

F‘ 54‘215'}, Stain-a In)..~

filial: 55.1.. .73. .r...

 

1. Illustration showing fiducial mark calibration in ERDAS

Figure 4

IMAGINE Orthobase.

72

 

The
used"?

3110le

tester the.

 

ll‘r
mean:
The ME
block file a

mable to

 

of he lldl

|
onentatio
eolian;
mm. at

MES a

errOYS ca

 

The RMSE represents how well the image coordinates correspond to their
respective calibrated fiducial mark coordinates (ERDAS, 2001). The ERDAS
IMAGINE Orthobase manual (2001) suggests that RMSE for fiducial marks
greater than 0.5 pixels indicates errors associated with the image.

These errors can be attributed to film deformation, poor scanning quality,
mismeasured fiducial marks, or incorrectly calibrated fiducial mark coordinates.
The RMSE associated with the fiducial mark calibration of each image in the 1999
block file are shown in Table 4:3. The error values were rather high, but I was
unable to improve them. Repeated efforts were made to improve the digitization
of the fiducial marks with no remarkable improvement. Additionally, the
orientation of the imagery with respect to the data strip was triple-checked for
accuracy. Because a great amount of care was taken to calibrate the fiducial
marks, and the other imagery sets (1978 and 1988) also had relatively high
RMSEs associated with fiducial mark calibration, it is believed that most of the

errors can be attributed to scanning distortions.

Table 4:3. Fiducial mark calibration errors.

 

 

Image Number RMSE (in
pixels)
705-8-197 2.74
705-8-198 2.51
705-8-199 2.65
705-8-200 2.12

 

73

 

ldSlOfli
.
4 ext
m -

{ROLES

and D.-

Once the interior orientation was established for each image, the exterior
orientation parameters were entered into the block file. Exterior orientation
establishes the. link between the image space and the ground coordinate system,
and the angular orientation of the camera at the time of exposure (ERDAS, 2001).
During image capture, the camera is subject to the roll, pitch and yaw of the
airplane in flight (Thieler and Danforth, 1994). These measurements can be
collected during modern flights using an airborne Global Positioning System
(GPS) and Inertial Navigation System (INS) (ERDAS, 2001). However, during
historic aerial photography flights, these tools were not available for use. Thus,
the external orientation parameters are calculated during the block triangulation
process, which will be described in more detail in subsequent sections (Thieler
and Danforth, 1994).

The rotational information was defined during the block setup was
Omega, Phi, and Kappa representing roll, pitch and yaw, respectively (Thieler
and Danforth, 1994). Omega is rotation about the image-space x-axis, phi is
rotation about the image-space y-axis and kappa is rotation about the image-
space z-axis.

The positional orientation (latitude, longitude, and elevation) of the
images was established through the use of ground control points. A ground
control point (GCP) appears in one or more image and has known ground

coordinates (X,Y,Z) (Thieler and Danforth, 1994).

74

 

 

Be

ushers
mm Z

lutlierxll

 

 

 

 

Ground Control Point Measurement

Before measuring GCPs, unrectified aerial photographs were printed as
large, field-sized maps so that points could be easily identified. In order to
compute the external orientation a minimum of two GCPs with X, Y, and Z
positions are recommended per photograph. One additional point with at least a
known Z position is also needed (Thieler and Danforth, 1994; ERDAS, 2001).
Furthermore, when evaluating a strip of adjacent images, a minimum of two
GCPs for every third image, with an additional three measurements at the corner
edges of the strip is recommended (ERDAS, 2001).

The recommended arrangement of GCPs is depicted in Figure 4:2.
Collecting more than the minimum number of GCPs is always advantageous to
the quality and accuracy of the processing and is also recommended in order to
perform an accuracy check (ERDAS, Inc., 2001; Hohle, 1996). With these
recommendations taken into consideration, a total of 33 GCPs were selected from
the photo maps for ground coordinate measurement.

Ideally, the network of ground control should be evenly distributed to
ensure—that the image space is accurately modeled (ERDAS, 2001). However, the
ability to locate ground identifiable points, accessibility to points, and the
sensitivity of the GPS equipment were considerations that limited where GCPs
could be measured. The resulting ground control network with respect to the

imagery is shown in Figure 4:3.

75

 

 

Figure l

 

 

 

D

>>q
v

 

 

 

 

 

 

 

 

 

 

Figure 4:2. Recommended ground control point distribution (modified from
ERDAS, 2001).

The GCPs selected for this investigation were primarily road intersections
surrounding the area of study. The choice of these GCPs was mainly due to the
difficulties in identifying isolated landmarks that were visible on the photos in
the open dune and forested areas. However, some isolated trees and large
bushes were available to be used as GCPs within the actual study area.

Ground control was measured using GPS. The GPS is comprised of a
minimum constellation of 24 satellites in precisely monitored orbits, at an
altitude about 20,200 km above the earth’s surface (Lillesand and Kiefer, 2000).
GPS signals are used to determine position by satellite ranging. Using a code
correlation technique, the distance between a GPS satellite and a receiver can be
calculated by measuring the amount of time it takes the signal to reach the
receiver. The signals from at least four GPS satellites are needed to measure both
horizontal and vertical position, and to correct for clock bias between each of the

satellites and the GPS receiver.

76

 

 

 

 

A
mdnr705-8-200

 

 

mdnr705-8-1g9

 

 

 

 

A A
A mdnr705-8-198
O A
A

mdnr705-8-1 97

 

 

A Control Point
0 Checkpoint

 

Figure 4:3. Control and checkpoint distribution with respect to 1999

imagery.

77

 

GPS measurements are subject to errors that include uncertainties in the
satellite orbits, atmospheric conditions (differences in signal velocity depending
on time of day, season, and the angular direction of the signal through the
atmosphere), receiver errors (electrical noise), and multipath errors (the
scattering and reflection of signals from objects not in a straight-line path
between the satellite and the receiver). Differential GPS (DGPS) can be used to
correct most of these errors (multipath errors usually persist). DGPS works by
taking continuous position measurements at stationary base stations with known
positions. Because the positions at the stationary base stations are known, errors
can be measured and used to correct roving GPS receivers out in the field
(Lillesand and Kiefer, 2000).

Trimble Pro XRS GPS receivers were used to collect the GCPs with either a
TDC1 or a TSC1 datalogger. The Pro XRS is reported to have a horizontal
accuracy better than 75 cm + 1 part per million (ppm) (Trimble, 2003). Because
vertical accuracy is roughly 1.5 times the horizontal accuracy, vertical accuracy
specifications for this receiver are about 1.1 m (Yao and Clark, 2000). The field
accuracy of these instruments was not tested because at the time of GCP
measurement, a nearby survey benchmark with known horizontal and vertical
positions could not be located under forest litter. However, Schaetzl et a1. (2002)
tested a predecessor of the Pro XRS, the Trimble Pro XL, in northern Michigan

and determined that this earlier model had a vertical RMSE of 1.34 m.

78

 

 

9M??-
Genetic.
recezver. l
attitude
Planing

was pied

 

ill)? Trill
393511191
aPDOP

New

inten'a

ltsltic

Both units received real-time, differential corrections from OmniSTAR, a
subscription service that uses a network of base stations to obtain differential
correction information that is transmitted by communication satellite to the
receiver. Before the points were measured in the field, mission planning was
conducted using Trimble Quick Plan mission planning software. The mission
planning provided time windows when the overall dilution of precision (PDOP)
was predicted to be less than 4.0. Yao and Clark (2000) suggest that using a low
PDOP filter (in the 24 range) in the field may improve the accuracy of elevation
measurements made with the Pro XRS. The receivers were subsequently set with
a PDOP mask of 4.0, and GPS measurements were only conducted during time
windows pre-determined to have PDOP of less than 4.0.

At each pre-selected GCP, position fixes were taken at one-second
intervals for two minutes, resulting in 120 positions (WI DNR, 2001). The
positions were recorded in geographic, decimal degree coordinates on the WGS
1984 ellipsoid. When a GCP was located at a tree, the positions were offset by a
measured distance and direction (Brown and Arbogast, 1999). Each recorded
GCP was saved in the datalogger as an individual point feature file. The point
was then marked on the image photomap to aid in subsequent identification in
ERDAS IMAGINE Orthobase.

. The data were downloaded from the datalogger using Trimble GPS
Pathfinder Office software. Each GCP was opened and decomposed into the 120

individual position fixes associated with that point. The measurements were

79

evaluated for multipath errors, changes in PDOP, and number of satellites. If
multipath errors showed an obvious trend, extreme outliers were deleted from
the point cloud. A majority of the GCPs had standard deviations less than 0.5
meters in horizontal precision.

Once the GCPs were examined, the multiple position fixes were averaged
to create one X, Y, Z coordinate per GCP (Rees, 2000). The GCPs were then
exported in ArcView GIS shapefile format after being projected into Michigan
Georef, with the horizontal datum set to the North American Datum of 1983
(NAD 83), and the vertical datum set to the North American Vertical Datum of

1988 (NAVD 88).

Exterior Orientation

With ground control measured, the GCPs were imported into the ERDAS
IMAGINE Orthobase blockfile as a text file with X, Y, and Z measurements. In
order to determine the unknown exterior orientation parameters, the measured
ground coordinates needed to be paired with the corresponding pixels on the
images. This was facilitated in Orthobase by the Point Measurement Window
(Fig. 4:4) that displayed the images and the GCP coordinate text file. A GCP
record was selected from the text file, and a cross-hair cursor was used to create a
point on the image that corresponds to the point listed in the text file. GCP
positions marked on the image photomaps taken into the field and field notes

aided in the process of correctly locating the GCPs in the Point Measurement

80

Window. For the 1999 imagery, a total of 11 GCPs were used, with three selected
for use as checkpoints to test the accuracy of the camera model (Fig. 4:3).

After the GCPs were digitized in the Point Measurement Window, tie
points were collected. Tie points are distinct features that appear on two or more
images but their ground coordinates are unknown. They commonly include
features such as road intersections, buildings, or trees. Ground positions of the
tie points are solved in latter steps (during block triangulation) and thus extend
the ground control network to improve the accuracy of the camera model
(ERDAS, 2001). Tie points also establish the relative orientation of photos to each
other (Thieler and Danforth, 1994).

A total of 200 tie points per image were collected to produce the most
accurate solution to the model. This number increases data redundancy and
distributes the error throughout the images (ERDAS, 2001). The first ten tie
points were manually collected throughout the block of images in the Point
Measurement Window, which can show two images at a time. This allowed a tie
point to be digitized on one image and then the matching pixel could be
digitized on other overlapping images.

Because tie point collection can be time-consuming, the remaining tie
points were measured using the ”Automatic Tie Point Collection” feature in

Orthobase. This procedure uses image matching, which analyzes the

81

 

Figure 4:4. Illustration showing point measurement window in ERDAS

IMAGINE Orthobase.

82

correspondence of pixel brightness values between the manually placed tie and
control points in two or more images. Additional tie points are then created
based upon these correspondence values (Thieler and Danforth, 1994; ERDAS,
2001). Once the tie points were automatically generated, the quality of each was
visually inspected in order to discard poorly matched tie points. A total of 367

tie points were used in the 1999 imagery.

Block Triangulation

ERDAS (2001, p. 80) defines block triangulation as ”the process of defining
the mathematical relationship between the images contained within a block, the
camera or sensor model that obtained the images, and the ground.” This process
simultaneously calculates the exterior orientation of each image in the block at
the time of image capture, the ground coordinates of the tie points (thus adding
more ground control points to the model), and the interior orientation
parameters associated with the camera (ERDAS, 2001).

In Orthobase and Orthobase Pro, block triangulation is facilitated through
bundle block adjustment (ERDAS, 2001). Because the exterior orientation
parameters and the ground coordinates of the tie points are derived through the
same equations, they are derived as a ”bundled" solution. The entire ”block” of
images is processed simultaneously, using a least squares ”adjustment” to
calculate the bundled solution for the entire block (ERDAS, 2001). Block bundle

adjustment uses the collinearity condition to establish the exterior relationship of

83

each image (ERDAS, 2001). The collinearity condition is met when the camera
position at time of image capture, any point on the ground, and its
corresponding point on the image all lie in a straight line. The collinearity
condition is expressed for each image through a series of collinearity equations
(Lillesand and Kiefer, 2000).

The block bundle adjustment also incorporates a least squares adjustment.
Least squares adjustment is a statistical method used to determine the unknown
parameters in the photogrammetric solution while simultaneously distributing
and minimizing error (ERDAS, 2001). The least squares adjustment is an
iterative process where a solution is obtained when the errors have been
minimized. The least square adjustment integrates the solutions from the
collinearity equations (ERDAS, 2001).

In Orthobase and Orthobase Pro, the block triangulation can be optimized
by using statistical information pertaining to the quality of input data. For the
1999 imagery, some of these optimizations were employed in an effort to
improve the results of the block triangulation. These optimizations are specified
in the ”Aerial Triangulation” dialog box in Orthobase. An image point standard
deviation of 0.33 pixels for both x and y was selected based on the
recommendation of ERDAS (2001). This specified the amount of fluctuation the

image coordinates could have during the iterative bundle block adjustment

process.

84

A statistical weight was also assigned to the GCPs before the bundle block
adjustment was performed as recommended by ERDAS (2001). Weighted
standard deviation values were the same for all GCPs: 0.5 m X, 0.5 m Y and 2.0 m
2. These values were chosen based upon an informal examination of the
standard deviations provided from the GPS measurement of the GCPs. The
values are representative of the standard deviations in X, Y, and Z associated
with the GPS measurement.

Because the exterior orientation parameters were unknown, no statistical
weights were applied to these parameters (ERDAS, 2001). However, statistical
weights were applied to the interior orientation parameters because information
was available from the camera calibration report. The ”Same Weighted Values”
option was used on the interior orientation parameters because the number of
GCPs was limited (ERDAS, 2001). The standard deviation weights were: 0.003
pm for focal length, 0.003 pm for principal point x, and 0.003 pm for principal
point y. These values were obtained from the camera calibration report, which
reported the standard deviations in the measurements of these input parameters.

With these statistical weights applied, the block triangulation was
performed with the ”No Additional Parameters” and ”Advanced Robust
Checking” options selected. An initial run was executed to solve for the
unknown exterior orientation parameters and to provide ground coordinates for
the tie points. The residual errors for the tie points were examined and points

were repositioned if the errors were high. The block triangulation was

85

performed once more with the updated exterior orientation parameters for each
image and the estimated ground coordinates for the 367 tie points. The overall
RMSE of the block triangulation for the 1999 imagery was 0.6120 pixels.

With the triangulation completed, the remaining distortion in the images
that needed correction was relief displacement. The final step in
orthorectification requires the use of a DEM. The remainder of the
orthorectification discussion will be separated into two different subsections to

represent the two methods tested. Both methods use the results obtained from

the aforementioned bundle block adjustment.

Orthorectification using USGS 30—meter DEM

This method combines the results of the bundle block adjustment, the

imagery and a USGS 30—m DEM of the area covered by the imagery using
ERDAS IMAGINE Orthobase. The DEM was obtained from the National
Elevation Dataset (NED), a seamless raster product that provides data in a
geographic coordinate system. The horizontal datum for the NED is NAD 83,
the vertical datum is NAVD 88 and the elevation units are reported in decimal
meters (Gesch et al., 2002). The vertical accuracy of the NED for the study area is
five meters.

In Orthobase, the NED DEM was applied to the bundle block adjustment
results and the imagery through a process called ortho resampling. This process

matches each pixel in the DEM to its corresponding position on the imagery and

86

assigns each location a brightness value based upon a resampling of surrounding
pixels. The brightness value is then combined with the elevation value and the
exterior orientation parameters to determine the relief-corrected ground location
on the imagery (ERDAS, 2001).

The method of ortho resampling chosen was bilinear interpolation.
Bilinear interpolation is a distance weighted technique that incorporates the
brightness values of the four nearest pixels (Lillesand and Kiefer, 2000). Bilinear
interpolation is the recommended ortho resampling technique in Orthobase
(ERDAS, 2001). The two other methods of resampling are nearest neighbor and
cubic convolution. Nearest neighbor resampling can cause pixel offsets of up to
one-half pixel (Lillesand and Kiefer, 2000) and cubic convolution is
computationally intensive (ERDAS, 2001).

The output cell size for the orthophotos was equivalent to the imagery
pixel size (0.5 meters). Once the individual orthophotos were generated, they
were mosaicked together to produce one full image of the study area, as opposed
to a few, disjointed images of the study area. Only two of the orthophotos were
chosen. for use in the final image mosaic (images 7058-198 and 705-8-199),
because the outlying photographs (images 705-8-197 and 705-8-200) were far
outside the area of interest for this investigation. The final image mosaic was

saved in the native IMAGINE file format (*.img) for analysis.

87

DTM Extraction and Orthorectification

DTM extraction is based upon the principal of parallax, the apparent
change in relative positions of stationary objects that is caused by imaging the
objects from two different positions (Lillesand and Kiefer, 2000). Elevation can
be derived by measuring the parallax between corresponding image points and
triangulating a solution using the parallax measurement and the interior and
exterior orientations of the camera and imagery (Lillesand and Kiefer, 2000).

Orthobase Pro allows for the extraction of a DTM from the results of the
bundle block adjustment and the imagery. For the 1999 imagery, a DTM cell size
of 3 meters was chosen. This decision was based upon the DEM resolution
chosen by Brown and Arbogast (1999) to analyze coastal dunes at Ludington, MI.
The output DTM type selected was DEM, as this format produces a file in the
native IMAGINE file format (*.img), which is useful when orthorectifying
imagery. The *.img file displays the DEM in gray scale, with brighter pixels
representing higher elevations and darker pixels representing lower elevations
(ERDAS, 2001). A single DTM mosaic was chosen as the output form to be used
later in' ortho resampling. The output coordinate system of the DEM was the
same as that used in the bundle block adjustment (Michigan Georef, NAD 83,
NAVD 88, with units measured in meters).

The ”Area Selection” tab in Orthobase Pro allows the user to digitize areas
of the imagery that will be included in the DTM extraction and the associated

”region strategies” with these areas, in addition to allowing the user to exclude

88

some areas from the DTM extraction. Region strategy refers to parameters and
methods that are used to optimize the performance of the DTM extraction.

In the case of the 1999 imagery, the portions of the image that captured
Lake Michigan were excluded from the DTM extraction by digitizing them using
the area selection feature of Orthobase Pro. This area was then assigned a
constant value of 176.2 meters, which corresponds to the average elevation of
Lake Michigan in 1999 (USACE, 2003). The areas of the imagery that featured
dunes were also digitized and a ”rolling hills” strategy was used for DTM
extraction in these areas. This strategy incorporates a 15 X 3 pixel search window
to find corresponding image points within the overlapping areas of the imagery.
The Y-value of the search window is reduced in this strategy. This is due to

smaller parallax between points located on more than one image. Lower

elevations result in smaller Y-parallax (ERDAS, 2001).
To analyze the accuracy of the extracted DTM, the block GCPs and check

points were used. The accuracy of the resultant DTM will be discussed in the

Results and Discussion section of this thesis. Once the extraction was
successfully completed, the output DEM was smoothed in an effort to
compensate for artifacts. For example, subtle differences in elevation between
adjacent pixels were detectable in the forested regions of the photographs, most

probably due to differences in canopy height of the trees. Additionally, pits and

hummocks resulted that were not observable in the field or on topographic

89

maps. In order to compensate for these errors, the DEM was smoothed using a 9
X 9 averaging filter (as in Brown and Arbogast, 1999).

With the DEM preparation complete, the extracted and processed DEM
was used to ortho resample the block of imagery via bilinear interpolation as
recommended by ERDAS (2001). The output cell size for the orthophotos was
equivalent to the imagery pixel size (0.5 meters). Once the individual

orthophotos were generated, they were mosaicked together according to the

aforementioned methods.

Processing the Remaining Imagery

The remaining imagery sets from 1978 and 1988 were processed according
to the DEM-extraction method. This method produced orthophotos with higher
positional accuracy, as explained in the Results and Discussion section of this
investigation.

The parameters and inputs to the photogrammetric models for the 1978
and 1988 imagery were the same used for the 1999 imagery, with a few
exceptions. The number of GCPs varied slightly in the 1978 and 1988. This was
because the images were a smaller scale than the 1999 imagery and thus, covered
more ground area. Ultimately, this meant that fewer images were used in the
photogrammetric solution, and it also allowed the addition of more GCPs into

the models used in the 1978 and 1988 image processing. Only two images were

90

used in the 1978 and 1988 bundle block adjustments. Table 4:4 shows the fiducial

calibration errors associated with each imagery set.

Table 4:4. Fiducial mark calibration error for the 1978 and 1988 imagery.

 

Imagery Year Image Number RMSE (in pixels)

 

1978 69-124-42 3.64
1978 69-124-43 2.89
1988 426-294-43 3.55
1988 426-294-44 3.63

 

Figures 4:5 and 4:6 show the 13 GCPs and 4 checkpoints used in the 1978
and 1988 imagery relative to the image coverage areas. The number of tie points
used to process the 1978 imagery was 211 and the number of tie points used to
process the 1988 imagery was 190.

The accuracy of the bundle block adjustments for the 1978 and 1988
imagery were 1.934 pixels and 3.7120 pixels, respectively. Lastly, extracting the
DTMs from the 1978 and 1988 imagery differed slightly from the 1999 imagery.
The image areas depicting Lake Michigan were excluded from the DTM
extraction, with constant lake elevations set to 176.6 meters, the approximate

average lake elevation for both time periods (USACE, 2003).

91

 

 

 

 

 

 

 

 

69-1 24-43

 

 

 

 

 

A o
A A
A
o 0
69-1 24-42
A A A

 

 

A Control Point
0 Checkpoint

 

Figure 4:5. Control and checkpoint distribution with respect to 1978 imagery.

92

 

 

 

426-294-43

 

 

 

 

 

 

 

 

 

 

 

 

A 426-294-44

O

A

 

 

 

A Control Point
0 Checkpoint

 

Figure 4:6. Control and checkpoint distribution with respect to 1988 imagery.

93

 

Classification System

According to Lachowski et al (1995), a well-defined classification system
should be able to accomplish the goals of the project using the technology and
resources available, and simplify all further processing. With those goals in
mind, a classification scheme was chosen that would meet the objectives of this
investigation, be appropriate for the imagery available, and reflect vegetation
conditions within WDSP. Businski (1992) created a classification scheme for
Lake Michigan dunes at Sleeping Bear National Park that fit the context of
Olson’s (1958a; 1958c) research. This classification system recognizes the
successional evolution of vegetation upon coastal dunes, ranging from barren

sand with little or no vegetation to the successional climax represented by closed-

canopy.
There are also geomorphological implications in Businski’s classification
scheme. Olson (1958c, p. 351) suggests that vegetation ”can be useful to the
geomorphologist who would translate its language.” The presence of different
types of vegetation can be associated with differing rates of sand deposition and
the disappearance of vegetation over time can be associated with erosion.
Consequently, different classes of land cover are associated with varying levels
of dune stability. For example, areas identified as barren sand represent

geomorphically active surfaces, where rates of erosion or sand deposition are too

high for vegetation to take root.

94

Because Businski’s classification system fit so eloquently into the goals of
this investigation, a modified version of this system was chosen. Ground
photographs of each of the six classes are used to illustrate the field conditions

represented in each class.

Barren Sand

This land cover class bares no stabilizing vegetation; therefore, the
surfaces are exposed to high winds with sufficient energy to entrain sand
particles. Due to the absence of vegetation, these areas have unstable surfaces
and can be reshaped readily by wind, water, and human foot traffic. These areas
are found near the lake and in active blowouts. Figure 4:7 illustrates the Barren

Sand classification.

 

Figure 4:7. Unvegetated dune depicting the barren sand classification (at
arrow) with discontinuous grasses in the foreground.

95

Barren with Isolated Plants

The Barren with Isolated Plants classification is slightly ambiguous
because it could possibly represent two conditions in the field. The first is barren
sand that has begun to be colonized by pioneer vegetation. However, the vast
majority of this class represents vegetation that is actively being buried by sand.
It represents an unstable environment, where the surface was once stable enough
to support vegetation, but has since destabilized. This class was developed from
an initial analysis of the imagery, which showed a trend in some areas with
stable, vegetated dunes in the older imagery, to isolated trees or shrubs
surrounded by barren sand. Figure 4:8 shows an example of the barren with

isolated plants class.

Discontinuous Grass

Discontinuous grasses are indicative of areas where sand deposition is
actively occurring, but vegetation has colonized the dune and acts as a
stabilization agent. Some species of dune grass require rapid sand deposition to
grow, whereas other species establish in conditions of very slow deposition
(Olson, 1958c). Regardless of the specific grass species, discontinuous grass
cover represents low dune stability. Most grasses growing in these
environments reproduce through rhizomes, as seeds may be buried too deeply to

germinate with such rapid deposition rates.

96

 

Figure 4:8. Photograph depicting the barren with isolated plants class.

These areas may have some humus at the surface, but no fully developed
soils or continuous land cover to entirely stabilize the dune surface. These areas
may also be mixed with herbaceous annuals and perennials that can tolerate
varying levels of sand deposition. Additionally, dune grasses are particularly
sensitive to even low levels of human disturbance (Roethele, 1985), making these
areas especially susceptible to destabilization in the ”High Use Area" area of

WDSP. Discontinuous grass cover is illustrated in Figure 4:9.

97

 

Figure 4:9. Photograph depicting the discontinuous grass classification.

Grass and Shrubs

These areas possess grass species, annual and perennial plants, woody
shrubs, and some small trees that grow as shrubs (such as cottonwoods) that
tolerate at least some sand deposition. Shrub species growing here may be
remnants from a time of rapid sand deposition, or may have seeded once sand
deposition slowed to a few millimeters per year (Olson, 1958a). Species that
reproduce through seed can be found in these areas, since sand deposition has
slowed enough to allow sufficient time for seed germination. These areas are
older and have been stable or nearly stable for a longer time than areas

represented by discontinuous grass. This land cover is illustrated in Figure 4:10.

98

 

Figure 4:10. Grass and shrubs land cover class.

Grass, Trees, and Shrubs

Shrubs, isolated trees, and continuous grass cover become established
rapidly when sand deposition has stopped or decreased to a few millimeters per
year (Olson, 1958c). Vegetation has become dense enough in these areas that a
well-defined humus layer has formed and a weakly developed soil can be
observed. These areas are similar to areas covered by grass and shrubs, with the
exception that the rate of sand deposition has stopped or slowed for a long
enough time that tall trees with large canopies are discemable in both the field
and on the aerial imagery.

As a precursor to forest development, these types of vegetation also act as

stabilization agents in dune environments. However, it should be noted that this

99

land cover is less stable and more sensitive to disturbance than forest cover. This

land cover is depicted in Figure 4:11.

 

Figure 4:11. Photograph illustrating the grass, trees, and shrubs vegetative
cover class.

Forest

Forests establish on dunes after other plant communities have thoroughly
stabilized the surface (Olson 1958a). Once closed canopies of forest develop on
dunes, vegetation and soil anchor the sand in place and prevent eolian transport.
As comparedto the other land cover classes, areas covered by forest are the

oldest surfaces because sand deposition has ceased long enough for communities

of large trees with undergrowth to develop. The forest land cover is illustrated

by Figure 4:12.

 

Figure 4:12. Photograph depicting the forest land cover class.

One additional class, unclassified, was used to mask out areas that
were outside the area of study or areas that had been urbanized. These areas
included the inland areas that were not covered by dunes, but were still inside
the park boundaries and the road, parking lot, and beach houses. From the
images, it was apparent that the paved areas and beach houses have remained
the same dimensions and in the same location. The unclassified areas were not

analyzed as part of this investigation.

lOl

Interpretation Factors

According to Lillesand and Kiefer (2000), an image interpreter examines
aerial imagery systematically, often referring to additional materials such as
maps and field observations to assist in the interpretation process. The success of
image interpretation relies heavily upon the experience of the interpreter, the
nature of the objects or phenomena under examination, and the quality of the
imagery used. It is important for the interpreter to be familiar with the objects or
phenomena under observation, in addition to having knowledge of the
geographic region under examination. Additionally, the interpreter must also be
able to evaluate several image characteristics to interpret features or phenomena
on the imagery. Factors important to this study include shape, size, tone, texture,
shadows, site, resolution, and image quality.

Shape refers to the general form of an individual object. This factor was
not exceptionally important to distinguish between land cover classes in this
study since it refers to individual objects (Lillesand and Kiefer, 2000). However,
the shapes of individual trees and bushes were helpful in discerning when a
particular area might be referred to as grass and shrubs versus discontinuous
grasses.

The size of objects is always considered in the context of image scale.

Additionally, relative sizes of objects to each other on the imagery are also an

102

important consideration (Lillesand and Kiefer, 2000). This factor was important
to this study, especially when classifying cover types with trees and shrubs.

Tone and texture were perhaps the most useful interpretation factors in
this study. Tone is the relative brightness or color of features on an image
(Lillesand and Kiefer, 2000). Tone helped distinguish between areas of barren
sand and areas of vegetation. Barren sand is bright white or light gray on the
black and white imagery and bright white to light cyan on the color infrared
imagery, whereas vegetated areas are darker shades of gray on the black and
white imagery and magenta to deep red on the color infrared imagery.

The aggregation of smaller features, some too small to be individually
discemable on an image, produces texture on an image (Lillesand and Kiefer,
2000). Tree leaves, branches, and twigs, in addition to tree shadows are
examples of features that produce texture. Texture can be defined according to
the overall coarseness and smoothness of image features. Discontinuous grass
cover presents a coarser texture than barren sand due to the aggregated blades of
grass, and forest cover represented the coarsest texture on the film from the
leaves or branches.

Shadows can both aid and hinder the interpreter. The shape and outline
of a shadow can help the interpreter recognize features. Alternatively, objects
within shadows are obscured and become difficult to interpret (Lillesand and
Kiefer, 2000). Shadow was also an important factor in interpretation in that it

helped distinguish between trees and shrubs. Shadows are also a limitation to

103

this investigation; especially adjacent to forest cover where canopy shadows
obscure the margins between land cover classes.

The geographic and topographic location of features is also a useful aid in
interpreting imagery. This factor is referred to as ”site” (Lillesand and Kiefer,
2000). Site played a minor role in interpretation in this study. Field observations
indicated a loose trend that the topographically low blowout areas had no
vegetation and that the higher crests were usually vegetated, very often with
trees. Additionally, relatively little forest cover was found directly adjacent to
the beach.

Resolution is the most limiting factor. At the scale of this imagery,
individual species of plants could not be discerned. Additionally, the resolution
at which the imagery was scanned limited the recognizable features to an
approximate size of one-half meter. Identification of individual species of plants
would likely provide a stronger link between this investigation and Olson’s
research (1958a; 1958c).

The quality of the film was also influential on the results of the
interpretation. In the 1988 and 1978 imagery, the areas with barren sand were
overexposed. This could have lead to some misinterpretation in these areas,
especially if small clumps of discontinuous grasses were obscured due to the
brightness of the surrounding pixels. Because the quality of the 1999 imagery is
slightly better than the other imagery, it is suspected that the interpretations of

the 1999 imagery could be more accurate.

104

Analysis and Mapping Procedures

Digitizing

Land cover polygons were head’s-up digitized from the *.img format
image mosaics using ArcView GIS 3.3. The images were contrast stretched to
compensate for areas where barren sand was overexposed. A minimum
mapping unit area of 4 meters2 was used based upon the resolution of the
imagery, and the lack of necessity for any smaller unit in this study. The land
cover classifications were saved in the native ArcView GIS shapefile format. As
each polygon was digitized, the associated land cover class was recorded in the
database file associated with the land cover shapefile in an attribute field named
”VegCover.”

By copying the data, two shapefiles of land cover for each time period
were created. One shapefile in each set represented the entire study area so that
the general trends in land cover in the park could be analyzed. The second
shapefile for each set divided the park into the ”High Use Area” and the
”Natural Area” so a comparative analysis could be performed.

Once each time period of imagery was completely digitized, the shapefiles
were converted to ArcInfo coverages using ArcInfo Workstation 8.2. Once the
shapefiles were converted to coverages, the area of each polygon in each
coverage was automatically calculated in square meters. The ”clean” and

”build” commands were performed on the land cover coverages in order to build

105

topologically correct polygon files. To eliminate errors in the digitization process
known as sliver polygons, the ”eliminate” command was utilized. The coverages
were then converted back into shapefile format in ArcView GIS 3.3 for further

analysis.

Positional Accuracy Assessment

It was hypothesized that:

H1: The DTM-extraction method of orthorectification will be the best

method to create digital orthophotos of the study area.

It was believed this method would be superior to the method that employs a
USGS 30-m DEM because the DTM is created from GPS measurements collected
at points visible on the orthophotos. Therefore, the DTM was localized to the
study area. The method used to test this hypothesis was to assess the positional
accuracy of the orthophotos produced by both methods.

Reporting positional accuracy supplies a measure of how well the spatial
relationships on the imagery model those in the real world. To measure
positional accuracy, the Geospatial Positioning Accuracy Standards from the
National Standard for Spatial Data Accuracy (NSSDA) were used as a guideline.

The NSSDA is intended for use with digital geospatial data derived from a
variety of sources and stored in raster, vector or point format (FGDC, 1998). The
NSSDA uses RMSE to report positional accuracy. To measure accuracy in

accordance with the NSSDA, a minimum of twenty, well-defined checkpoints

106

should be collected from an independent source of higher accuracy (FGDC,
1998). Well-defined checkpoints may be small isolated shrubs and bushes, or
right angle intersections of roads or buildings. These points may be surveyed
using a GPS to function as the independent source of higher accuracy.
Additionally, these points should not be included in the aerial triangulation
model used to produce the orthoimages (FGDC, 1998).

Several factors limited the number of both GCP and independent
checkpoints that were identifiable both on the imagery and in the field. These
factors included: terrain, property ownership, land cover, and a relatively low
number of road intersections in and around the study area. Therefore, collecting

20 independent checkpoints to test each set of imagery was not feasible. In lieu
Of this, the largest feasible number of checkpoints was collected for each set of
iII'lagery. In addition, the accuracy at ground control points used in the model
was reported to provide supplementary information about the positional
accuracy of the imagery.

The checkpoints were collected is the same manner as the GCPs, using
Trimble Pro XRS GPS receivers. The checkpoints were located at road
intersections, large isolated trees or shrubs, and building corners. The points
were marked on a photomap of the area for easy recognition during the accuracy
assessment process. Ten checkpoints were used to test the 1999 imagery, eight

C1leckpoints were used to test the 1988 imagery, and 12 checkpoints were used to

teSt: the 1978 imagery.

107

In ArcView GIS 3.3, the imagery was added, and a new point shapefile
was created. Using the field photomaps as reference, the positions of the
checkpoints were head’s-up digitized, zooming in at the points to assure accurate
placement of the points. The ”Add X,Y coordinates” script was run on the point
shapefile to add horizontal positions of the checkpoints, as displayed by the
orthophotos. These positions were then compared to the GPS positions of the
checkpoints.

Additionally, coordinates at GCPs were collected from the digital
orthophotos in the same manner as the checkpoints and compared to the GPS
coordinates measured in the field, providing nine additional points from the
1 999 and 1988 imagery, and 10 additional points from the 1978 imagery.

Errors were calculated by subtracting the GPS coordinates (actual) from
the orthophoto coordinates (measured). The results of the positional accuracy
assessment are reported in the Results and Discussion section of this thesis. The
re81.1lts are reported in RMSE in accordance with the NSSDA, with
Sllllbplementary information including: minimum error, maximum error, range of

eI'I‘cr, mean error, and standard deviation.

Thematic Accuracy Assessment

Reporting thematic accuracy establishes the level of confidence within the
reSuits of the land cover class interpretation. For thematic accuracy, standards

uSed by the USGS-NPS Vegetation Mapping Program were used to test the 1999

108

imagery only. Because the field conditions had changed greatly in some areas
since the 1978 and 1988 imagery had been acquired, it was deemed unfeasible to
assess the accuracy of these imagery dates. The landscape may also have
changed from the image capture date in 1999 to the field observation date in
2003, which is a limitation in this investigation. However, an independent
orthophoto of the study area from 1999 was available from the Michigan Center
for Geographic Information to assist in the accuracy assessment.

These USGS-NPS Mapping Program standards were deemed appropriate
for use in this investigation because they were specifically designed for
Vegetation classification schemes and the methods were supported through pre-
existing research. Stratified random sampling was used for thematic accuracy
assessment, with vegetation class as the stratifying variable. This method of
Sampling performs well and can be used when small, yet important, areas need
representation in the sample (Congalton, 1988).

When a study area has classes that are extremely abundant and classes
that are not very abundant, the USGS-NPS Vegetation Mapping Program
accuracy assessment standard recommends a stratified random sampling
Strategy based upon number of polygons and total area covered. Under this
SCl‘leme, the maximum number of points sampled in abundant classes is 30, while
the minimum number of points sampled in rare classes is 5. The most

abundantly present land cover class would cover greater than 50 hectares and

COIlsist of at least 30 polygons. This class would have 30 randomly chosen

109

sample points. Less abundant classes may also cover greater than 50 hectares but
consist of less than 30 polygons. These classes would have 20 randomly chosen
sample points. The rare classes would have greater than five but fewer than 30
polygons, and cover less than 50 hectares. These classes would have five
randomly chosen sample points (The Nature Conservancy and ESRI, 1994).

In order to determine the area in hectares per class, an attribute field was
added to the land cover class shapefiles called ”Hectares.” The area in hectares
for each polygon in each class was determined by using the ”calculate”
command in ArcView GIS 3.3. The total area of each class was determined by
first using the ”query” tool to select out each class, then using the field
”statistics” command. This also determined the number of polygons per land

cover class. The land cover classes and number of sampling points per class is

Summarized for the entire study area in Table 4:5.

With the necessary number of sampling points determined, the positions
of the sample points needed to be acquired randomly. This was facilitated in
ArcView GIS 3.3 using a random point generator extension that used a polygon
t1Ieme to generate a desired number of randomly determined points. Decimal
degree latitude and longitude for a total of 140 randomly generated points were

Output to a text file that could be input into a GPS as waypoints for accuracy

assessment.

The sample points were also plotted against a 1999 color infrared,

.OI'thorectified image of the park, obtained from the Michigan Center for

110

Geographic Information. This procedure produced a field map to assist in

identification of the points in the field.

Table 4:5. Summary of vegetation class, area, number of polygons per class,
and number of sampling points per class.

 

 

Class Area Number of Number Of
Polygons Sample
Points
Barren Sand >50 hectares 106 30
Barren With
Isolated Plants <50 hectares 75 20
scontinuous >50 hectares 76 30
Grass
Grass and Shrubs <50 hectares 63 20
Grass, Trees, and
S] bs <50 hectares 36 20
Forest >50 hectares 8 20

 

The sample point file was loaded as waypoints into a Trimble Pro XRS

GPS unit so that each point could be navigated to. As each point was visited, the
Observed land cover was recorded. In areas of dense forest cover, the terrain and
tree cover made it extremely difficult to navigate to sample points in these areas.
Thus, the orthorectified image provided by the Michigan Center for Geographic
Information was used to check the accuracy of these points, in addition to a small
Portion of points that were inaccessible due to the terrain. This orthophoto of the

Study area has an accuracy of + /-33 feet (10.06 meters). Although the positional

error is greater than the imagery produced in this investigation, it was the best

available independent source to aid in the accuracy assessment.

lll

Once the data wer collected from the field and the independent 1999
orthophoto, the sample point data was inserted into an error matrix to compute

the thematic accuracy results.

Measurement, Comparison and Presentation Procedures

One of the research objectives of this investigation was to determine if
land cover change occurred at a constant or a variable rate throughout the period
of study. It was hypothesized that:

H2: Given that coastal systems are constantly changing in response to
various factors, the rate of land cover change would not be the same from 1978 to
1 988 as it was for 1988 to 1999.

To explore this hypothesis, an overlay function was used to analyze any
Changes in land cover in the study area. Overlays are used to detect differences
between categorical data on two or more time periods in the same region
(Cl-misman, 1997). In ArcView 3.3, an ”intersect” was performed between the
1 978 and 1988 land cover classifications, the 1988 to 1999 land cover
Classifications, and the 1978 and 1999 land cover classifications. New fields were
cl‘eated in the shapefile databases to categorize the type of overlay change
Category and the area of the overlay polygons in square meters.

The overlay categories were divided into three classes to simplify the

discussion: successive, regressive, and unchanged. Table 4:6 breaks down these

three categories into their component land cover change classifications. For

112

analysis purposes, the regressive, successive, and unchanged areas are presented

as a percentage of the entire classified area. Individual cover changes within
each of these three categories are discussed as a percentage of total regressive,
successive, or unchanged area, depending on which category they belonged to.

An additional research objective of this study was to determine if

management strategies had an effect upon land cover change over the period of
study. It was hypothesized that:

H3: Given that the southern end of the park is more intensely used by
people than the northern end of the park, degradation of land cover would be
more apparent in the High-Use Area than in the Natural Area.

This hypothesis was explored by splitting the land cover interpretations
into a Natural Area and a High-Use Area. These areas were shown on Figure 3:1
in the Study Area section. This division was made so a comparison could be
drawn between an area under heavy recreational use and an area under light
recreational use. The determination of the boundary between the two
approximates the designation of the natural area according to the MDNR.

The Natural Area is used primarily by hikers and nature enthusiasts. It is
1()cated north of the beach and therefore away from the main activities there. The
I‘Iigh-Use Area is subject to heavy recreational use, especially in the summer

When temperatures are favorable for swimming. The campground is also to the

east of the High-Use Area, and campers must hike through the area in order to

113

reach the beach by walking. This area is subject to trampling of the land cover, in

addition to disturbances that could lead to sand mobilization.

Table 4:6. Land cover change categories and overlay classifications.

 

 

Re resslve
Stream Erma:
Changes

BWI to BS BS to BWI BS
DG to BS BS to DG BWl
DG to BWI BS to F 06
F to BS BS to GS F
F to BWI BS to GTS GS
F to DG BWI to DG GTS
F to GS BWI to F
F to GTS BWI to GS
GS to BS BWI to GTS
GS to BWI DG to F
GS to DG 06 to GS
GTS to BS 06 to GTS
GTS to BWI GS to F
GTS to DG GS to GTS
GTS to GS GTS to F

 

a"’BS, barren sand; BWI, barren with isolated plants; DG, discontinuous grass; F, forest; GS, grass
and shrubs; GTS, grass, trees, and shrubs.

Maps, tables, and graphs were used to present the results and aid in the
discussion of the section. Some images and maps presented in this thesis are

presented in color.

Limitations

This study has several limitations that will be discussed in this section.
Limitations within the study arise primarily from to the imagery used. Ideally,

photo scale, timing of the image capture, type of film, and the quality of the

114

images should have been the same for each time frame. However, when
utilizing historic aerial photography, it is difficult to control for all of these
factors. Often whatever is available must be used. Differences in photo scale,
month the imagery was captured, the type of film, and the differing quality of
the photographic prints undoubtedly had some impact upon both the
orthorectification of the images and the interpretation of the land cover from
these orthoimages.

Additionally, the temporal scale of this study is too coarse to be able to do
any correlative studies. Many of the possible factors influencing the results (such
as park visitor numbers, lake levels, wind, precipitation, and storms) are
collected at much finer temporal frequencies. In order to do any true correlative
studies, the images would need to have a finer temporal frequency, possibly one
photo every one to three years and over a longer time period than 20 years.

Moreover, because this is a historical study, no field measurements were
available regarding erosion or deposition rates in the dune field during the
period of study. Although the vegetation cover may be used as a means to detect
erosion or sand deposition, the relationship between vegetation and eolian sand
movement can only be speculated about using existing literature.

Positional and thematic accuracy also limit the results of this study.
Although quantitative results regarding land cover change are presented in this

study, it is difficult to say that these numbers are accurate. That being said, most

115

of the results discussed in this thesis were limited to the land cover classifications
that produced the best thematic accuracy values (user’s accuracy).

Lastly, experience of the interpreter is a limitation upon the results. Image
interpretation is a skill that can be improved through practice. Although I had
done land cover interpretation prior to this study, I would not deem myself an
expert in land cover interpretation. However, I spent extensive amounts of time
in the field using aerial photographs as maps to become familiar with the
landscape. This gave me a solid base knowledge of the vegetation in existence at

present day and how the vegetation correlated with landscape features.

116

Chapter 5

Results and Discussion

Accuracy

In this section, the accuracy of the orthoimagery and the data derived
from the orthoimagery will be discussed. Reporting vertical accuracy provides a
statement regarding the quality of the data used to prepare the orthoimages.
Measuring the positional accuracy of the resultant orthoimages alludes to the
quality of the measurements taken from the orthoimages. Describing thematic
accuracy of the land cover interpretation presents an estimate of how well a

sample of land cover interpretations matched conditions in the field.

Vertical Accuracy of the 1999 Input DEM and DTM

The vertical accuracy of the USGS 30—m DEM was reported in RMSE as
five meters within the study area. This was considerably better than the
accuracy of the DTM extracted from the 1999 imagery, which had an RMSE of
13.2 meters (measured from checkpoints and ground control combined). Based
upon the quality of the input DEM and DTM alone, it was believed that the
orthoimagery processed with the USGS 30-m DEM would have better positional
accuracy as a result. However, this was not the case as described in the

following section.

117

Positional Accuracy of the 1999 Test Imagery Sets

Positional accuracy describes how well the coordinates of features on the
digital orthoimagery match their true horizontal location. Positional accuracy
can be used, amongst other measurements of accuracy, to assess the quality of
data derived from the orthoimagery. If the positional accuracy of the
orthoimagery is high, then confidence can be placed in measurements collected
from the orthoimagery. Thus, positional accuracy was used to determine which
method of orthorectification should be used to interpret land cover in WDSP.
This section compares the positional accuracy of the 1999 imagery orthorectified
using a USGS 30-m DEM and the orthorectified images produced using a DTM
extracted from the imagery itself.

X and Y errors were calculated by subtracting the GPS coordinate
measurements of ten independent checkpoints (actual) from the image
coordinate values of those checkpoints (observed). This process was repeated for
the GCPs used in the orthorectification process. The checkpoint errors for both
datasets are shown in Table 5:1, and GCP errors are shown in Table 5:2.

Summary statistics of error are provided for both sets of 1999 imagery.
Table 5:3 presents the descriptive statistics and positional accuracy of the two test
imagery sets for both the checkpoints and the ground control points. It is clear
from these statistical measures that the orthoimagery produced by the DTM

extraction method had a smaller range of error in X and Y than the orthoimagery

118

produced by the USGS 30—m DEM method, indicating higher accuracy in the

DTM extraction-produced imagery.

Table 5:1. Calculated differences between images and surveyed values at

checkpoints (1999 imagery).

 

 

Checkpoint 1999 DTM 1999 DTM 1999 USGS 1999 USGS

ID Extraction Extraction 30-m Method: 30-m Method:
Method: Method: Error X Error Y
Error X Error Y (meters) (meters)
(meters) (meters)

2 -0.50 0.89 -9.64 2.74

15 2.48 2.52 3.22 9.05

20 -4.42 -0.68 6.13 6.19

24 1.11 0.87 4.20 5.71

27 -2.70 2.49 3.40 13.82

28 -2.34 1.15 14.58 18.27

31 2.64 ‘ -3.41 -2.37 2.01

32 0.92 -0.79 -0.06 4.92

34 4.73 -0.86 8.11 9.45

35 0.04 -1.08 -5.12 3.17

 

Table 5:2. Calculated differences between images and surveyed values at ground
control points (1999 imagery).

 

 

GCP ID 1999 DTM 1999 DTM 1999 USGS 1999 USGS
Extraction Extraction 30-m Method: 30-m Method:
Method: Method: Error X Error Y
Error X Error Y (meters) (meters)
(meters) (meters)
' 21 .-4.83 0.01 11.94 13.56

14 0.31 1.32 2.14 0.15

16 1.66 0.30 1.79 2.40

17 1.05 2.65 -18.15 4.76

29 -0.54 -1.15 8.78 3.58

30 -0.47 - -2.32 -6.38 -3.95

3 -1.28 9.49 0.59 13.77

26 -1.36 1.23 -4.08 2.63

33 3.48 -4.76 ~6.75 -11.46

 

119

Table 5:3. Summary Statistics for Error Data of the 1999 imagery.

 

 

1999 1999 1999 1999
DTM DTM USGS USGS
Extraction Extraction 30-m 30-m
Method: Method: Method: Method:
Error X Error Y Error X Error Y
(meters) (meters) (meters) (meters)
Checkpoints Minimum -4.73 -3.41 -9.64 2.01
Maximum 2.64 2.52 14.58 18.27
Range 7.37 5.93 24.22 16.26
Mean -0.75 0.11 2.24 7.53
Sta‘ld‘lrd 2.54 1.73 6.91 5.22
DeV1ation
RMSE 2.65 1.73 6.93 9.02
GCP Minimum -4.83 -4.76 -18.15 -11.46
Maximum 3.48 9.49 11.94 13.77
Range 8.31 14.25 30.09 25.23
Mean -0.22 0.75 -1.12 2.83
Stalldalrd 2.32 3.95 9.00 7.88
Dev1ation
RMSE 2.20 3.80 8.56 7.95

 

The mean error in X was slightly negative (west-biased) in the DTM-

produced imagery, whereas the mean of Y errors was slightly positive (north-

biased). However, both values (-0.75 and 0.11) were very close to zero. For the

USGS 30-m DEM produced imagery, the means for both X and Y were farther

from zero (2.24 and 7.53, respectively). The X errors had an eastward bias,

whereas Y errors exhibited a northward bias that was greater in magnitude than

the eastward bias. Because all Y errors were positive in the USGS 30-m DEM

produced orthoimagery (Table 5:1), a northward bias can be confirmed.

120

Standard deviation and RMSE were smaller in both X and Y in the
orthoimagery produced by the DTM extraction method than the orthoimagery
produced by the USGS 30—m DEM method. This pattern implies that the
dispersion and magnitude of errors were lower in the imagery produced by
DTM extraction than in the imagery corrected with the USGS 30-m DEM.

Overall, the summary statistics show that the errors in X and Y on the
DTM Extraction-produced imagery were less than the errors in X and Y on the
USGS 30-m DEM-produced imagery. This trend was also evident at ground
control points (Table 5:3). The statistical summary also shows that ground
control error was greater in both orthoimagery sets than checkpoint error. This
result was most likely due to the manner in which GPS measurements were
collected.

The first of these limitations is that the GPS surveys for GCPs and
checkpoints were conducted on different dates. Second, the GCP network was
more evenly distributed over the entire image surface than the checkpoints.
Some extreme positional outliers were located towards the edge regions of the
orthophotos, whereas nearly all the checkpoints were in the interior portions of
the images, near the dunes. The sampling distribution of the checkpoints could
have biased the positional accuracy, making it appear better than the GCP

accuracy.

121

Once the errors had been determined, a RMSE was calculated to describe
positional accuracy in accordance with NSSDA guidelines (FGDC, 1998). RMSE
is determined by the following equation:

RMSEr = sqrt[ D((Xdata i — Xcheck i )2 +(Ydata i - Ycheck i )2)/ n],
where Xdata i , Ydata i are the coordinates of the i-th check point in the dataset,
Xcheck i , Ycheck i are the coordinates of the i-th check point in the independent
source of higher accuracy, n is the number of check points tested, and i is an
integer ranging from 1 to n. The NSSDA (FGDC, 1998) also states that the RMSE
should also be reported to a 95-percent confidence interval, determined by this
equation:

Accuracy, ~ 2.4477 * 0.5 * (RMSEx + RMSE), ).

Table 5:4 shows the results of the RMSE and accuracy calculations for both
imagery sets. Checkpoint and GCP accuracies were better in the orthoimagery
produced by DTM extraction than in the orthoimages produced using the USGS
30—m DEM. Therefore, this method was chosen to produce the remaining
imagery sets (1978 and 1988).

Positional accuracy could have varied between the two test imagery sets
for a few reasons. Shortridge and Clarke (1999) concluded that reprojection of
square raster cells may alter elevation data. The USGS 30-m DEM was
reprojected from geographic to Michigan Georef, using a nearest neighbor

resampling strategy. Although Shortridge and Clarke (1999) concluded that

122

nearest neighbor and cubic convolution preserve spatial characteristics in

elevation data to a greater degree, these methods still change the data.

Table 5:4. Positional Accuracy of the 1999 orthoimagery.

 

1999 Orthoimagery 1999 Orthoimagery
produced by DTM produced by USGS 30-

 

Extraction Method m Method (meters)
(meters)
Checkpoints RMSE; 3.17 11.37
Accuracy; 5.36 19.51
GCP RMSE; 4.39 11.68
Accuracyr 7.33 20.20

 

Additionally, raster cell size may have played a role in the resultant
accuracy of the orthoimagery. The USGS DEM had a cell-size of 30 m X 30 m,
whereas the extracted DTM had a cell size of 3 m X 3 m. The terrain becomes
more generalized with large cell sizes because the same elevation value is
applied over a large area. It is likely that the 3 m DTM more closely represented
the shape of the landscape than the 30-m DEM, despite the large errors in

elevation values.

Vertical and positional accuracy: 1978 and 1988 imagery

Extracted DTMs for the 1978 and 1988 imagery had vertical accuracies in
RMSE of 12.4 m and 12.9 m, respectively (as determined from GCPs and
checkpoints). These DTMs were used to produce the 1978 and 1988

orthoimagery.

123

The surmnary statistics for positional error in 1978 and 1988 imagery are
displayed in Table 5:5. Positional accuracy of the imagery sets is summarized in
Table 5:6. These results show that the 1978 and 1988 orthoimages were less
accurately positioned than the 1999 imagery. The images may be less accurately

positioned due to the difference in scale between the 1999 imagery and the 1978

and 1988 imagery.

Table 5:5. Summary statistics for error data (1978 and 1988 imagery).

 

 

1978 1978 1988 1988
Error X Error Y Error X Error Y
(meters) (meters) (meters) (meters)
Checkpoints Minimum -2.64 -7.86 -4.13 -1.06
Maximum 9.82 11.64 7.27 11.54
Range 12.46 19.49 11.40 12.60
Mean 1.74 1.56 0.38 3.79
Sta‘ldalrd 2.49 2.31 3.27 4.11
Dev1ation
RMSE 3.57 4.79 2.91 4.70
GCP Minimum -8.93 -21.07 -8.07 -20.88
Maximum 4.03 7.26 4.29 4.27
Range 12.96 28.33 12.36 25.15
Mean -0.22 0.98 1.00 -0.33
Starldird 3.85 7.97 4.04 7.78
Dev1atlon
RMSE 3.66 7.62 3.94 7.34

 

As was the case for the 1999 orthoimagery, the checkpoints were more
accurately positioned than the ground control points. This was most likely due
to a biased distribution of the checkpoints similar to that of the 1999 imagery.

Additionally, some shrinkage and/ or curling may have occurred on the paper

124

prints of the 1988 images. Shrinkage or expansion of paper prints by even 1-2
m can result in errors of from 24 m to 48m on 1:24,000 scale aerial photographs.
The 1978 diapositive imagery was on stable polyester material that resist
expansion and shrinkage. However, the quality of the 1978 imagery appeared to
be the worst of the imagery sets. This was possibly due to slightly overexposed
areas in and around barren sand. The reduced quality in this imagery set could
have hindered the image matching techniques used by ERDAS IMAGINE

Orthobase to create the orthophotos.

Table 5:6. Positional Accuracy of the 1978 and 1988 orthoimagery.

 

1978 Orthoimagery 1988 Orthoimagery

 

 

(meters) (meters)
Checkpoints RMSE; 6.60 5.87
Accuracy; 11.37 9.31
GCP RMSE: 8.45 8.33
Accuracy; 13.80 13.80
Thematic Accuracy

A classification error matrix is one of the most common methods of
presenting thematic accuracy. An error matrix compares ground truth to the
results of image classification on a category-by-category basis. Table 5.7 shows
the classification error matrix for the interpretation of the 1999 imagery. Each

column of errors represents omission errors (areas excluded from a category),

125

whereas the rows represent errors of commission (areas included in the category,

but are incorrectly classified).

Table 5:7. Error Matrix for 1999 Land Cover Classifications.

 

Reference Data

 

(Field)
BS BWI DG GS GTS F Row
Total
Classification
Data
BS 28 0 2 0 0 0 30
BWI 1 14 2 1 1 1 20
DC 2 0 26 2 0 0 30
GS 0 0 4 16 0 0 20
GTS 0 0 1 2 14 3 20
F 0 0 0 0 2 18 20
Column total 31 14 35 21 17 22 140
Producer’ 5 Accuracy User’ 8 Accuracy
BS 90% BS 93%
BWI 100% BWI 70%
DC 74% DC 87%
GS 76% GS 80%
GTS 82% GTS 70%
F 82% F 90%
Overall Accuracy = 83%
KHAT = 0.79

 

’BS, barren sand; BWI, barren with isolated planted; DG, discontinuous grasses; GS, grass and
shrubs; GTS, grass, trees and shrubs; F, forest.

Producer’s accuracy indicates the percentage of each category in the
sample set that was correctly classified and is a measure of omission errors
(Congalton, 1991). User’s accuracy refers to the probability that a point in the

sample set actually represents that category in the field and is a measure of

126

commission errors (Congalton, 1991). Overall accuracy indicates the percentage
of sample points that were correctly classified.

The error matrix reveals that the barren with isolated plants and the grass,
trees, and shrubs categories presented the most difficulties in the land cover
interpretation. The producer’s accuracy for barren with isolated plants was
100%, meaning there were no areas that were incorrectly classified as barren with
isolated plants during the interpretation that should be of another category.
However, the user’s accuracy for the barren with isolated plants classification
was only 70 %. This indicates that there is only a 70% chance that an area
interpreted as barren with isolated plants will be of that category in the field.

Likewise, the grass, trees, and shrubs classification had a fairly high
producer’s accuracy (82%), meaning that points from this classification were
correctly classified as grass, trees, and shrubs 82% of the time. However, this
classification had a user’s accuracy of only 70%, meaning, again, that there is
only a 70% chance that an area interpreted as grass, trees, and shrubs will be of
that category in the field. In terms of user’s accuracy, the most accurately
identified land cover classes were barren sand (93%) and forest (90%), meaning
that these categories had the highest probabilities of being that category if visited
in the field.

The overall accuracy of the sample set (83%) meets the USGS-NPS
Mapping Program standard of 80%. However, overall accuracy is a limited

measure of accuracy because some, or many, classifications may be correct due to

127

random chance. The KHAT statistic is used to determine the percentage of
correctly classified locations that are due to actual agreement and not random
chance agreement (Lillesand and Kiefer, 2000). The KHAT statistic is calculated

as:

KHAT= .

N2 - 2 (KM * x+,-)

i:

r
N Z X,’,'- Z (X[+ * X+;')
1 l i=

where r is the number of rows in the matrix, Xii is the number of observations in
row i and colurrm i, Xi+ and xn are the marginal totals of row i and column i, and
N is the total number of observations (Congalton, 1991). The KHAT value for the
sampled set of classified data is 0.79. This indicates that the land cover class
interpretation was 79% better than a random classification.

Once the accuracies for the sample set were determined, this information
was used to reclassify areas that were rnisclassified according to field verification
during the accuracy assessment. The accuracy assessment also aided in a re-
evaluation of the classifications over the entire study area. Special attention was
given to troublesome categories that presented the most confusion. With the
information gained from the classification of the 1999 imagery, the 1978 and 1988
images were classified.

Some factors may have influenced the thematic accuracy assessment.

Firstly, some of the sampling locations may have been incorrectly classified due

128

to the positional accuracy of the imagery, and not as a fault of the interpretation
itself. This is especially a concern when samples were drawn at or near land
cover polygonboundaries. Additionally, GPS navigation rarely led to the
precisely desired sampling location. A small fraction of the classification errors
may be attributed to sampling in the wrong location. The difference between the
date of image capture and the date of field verification could also have skewed
the results of the thematic accuracy assessment. For example, an area classified
as barren sand in 1999 may have appeared to be barren from the photograph, but

by the time of field verification in 2003, grasses had grown into the area.

Land Cover Analysis

Maps of land cover interpreted from the digital orthophotos of the entire
park area for 1978, 1988, and 1999 are shown in Appendices A, B, and C
respectively. These maps are presented to demonstrate the land cover at each of
the studied time periods and to demonstrate the general trends in land cover. It
is important to note that although specific locations are discussed throughout
this section, the quantitative measures are displayed on a park- wide scale. At
this scale, a variety of geomorphic and ecological conditions are detectable and it
is difficult to discern whether a pattern is dominant or not.

To place the land cover in context of landscape features, observations in
the field and a topographic map aided in the interpretation of the land cover

trends. To quantify these results, Table 5:8 shows the total number of polygons

129

and total area mapped for every land cover class in each time period for the

entire park area.

1978 Land cover: General Discussion

As demonstrated on the 1978 map of land cover, massive areas of barren
sand covered the blowouts in all areas of the park at this time. Within the
blowouts, vegetated areas were small and fragmented. The high number of
polygons interpreted from the imagery is evidence of the high amount of land-
cover fragmentation in 1978. For example, the barren with isolated plants,
discontinuous grasses, and grass, trees and shrubs classifications each have over
100 polygons (Table 5:8).

The majority of these fragmented areas of vegetation were found at the
lakeward throats of the blowouts, and mostly likely corresponded with
foredunes. More dense areas of vegetation existed within depressions, probably
where moisture was collecting. Recent field observations have determined that
these areas are deflation hollows on the present landscape.

Deflation basins are wind-eroded hollows formed by erosion of sand
(Hesp and Thom, 1990). According to Olson (1958a), deflation hollows are damp
because they are often located at or near the water table. The damp conditions at
deflation hollows are effective areas for initial vegetation development on dunes.
The presence of dense, more mature vegetation at areas A, B, and C (Figure 5:1)

suggests that the deflation hollows were protected from erosion in 1978.

130

Despite the general trend of fragmented vegetation in 1978, some rather
dense, contiguous areas of vegetation existed at areas D, E, and F (Figure 5:1). At
areas D and B, field surveys and examination of a topographic map show that
these areas correspond with the limbs of parabolic dunes. Thus, in 1978,
vegetation was growing up the slopes of the limb ridges.

Area F also had dense, contiguous areas of vegetation. Field observations
at F revealed that the topography at this site was unusually flat. Further
examination of this area on aerial photographs taken prior to 1978 provides
evidence that the flat area at F was previously mined. Observations in the field
suggest that the mined area was most probably the southern limb of a parabolic
dune. Just to the north of this flat area is a ridge of sand that is normal to the
lake and gradually rises toward the east in a manner similar to that of the limbs
of parabolic dunes found elsewhere in the park. Although the area had been
disturbed prior to 1978, vegetation had colonized the flat, mined area and the
ridge to the north by 1978. It is uncertain if the vegetation naturally colonized
the mined area, or if vegetation had been planted in an attempt to reclaim the
previously mined land.

With respect to specific classifications of land cover, barren sand and
forest account for the largest percentages of the total classified area in 1978, with
approximately 29.5% and 62.2%, respectively (Table 5:8). The vast majority of
forest cover was located on large, inland dunes to the east of the blowouts. Field

observations have shown that these are mature forest communities as described

131

by Olson (1958a) and thus, it is estimated that these dunes have been stable for at

 

least several hundred years. Areas of barren sand include the beach and
massive, open blowouts that are connected to the beach system. Because the
blowouts were still connected to the beach system in 1978, sand could still move
into the back dune areas with few obstacles provided that winds were strong
enough to support sand transport. Additionally, the large amount of barren

surfaces meant that the blowouts were very susceptible to erosion in 1978.

 

The remaining land cover classes represent much smaller percentages of

 

the total classified area. Grass, trees, and shrubs cover is the most prominent of 3
the remaining classes, representing approximately 2.8% of the total classified
area. The largest areas of grass, trees, and shrubs cover were located at areas A,
B, C, and F (Appendix A). As previously stated, the locations of areas A, B, and
C correspond with present day deflation hollows and area F was located at an
area that was previously mined.

Because the surface at area F is extremely low, it may be near the water
table. Mining could have created conditions similar to the deflation hollows,
producing a low, moist area. According to Olson (1958c) most tree and shrub
species require moist conditions for their seeds to germinate. It would appear
that in 1978, these low, moist areas were favorable for the growth of shrubs and
trees. Being near the water‘table, these areas would have been more favorable
for vegetation growth than the dry, sun exposed areas on windswept slopes of

the blowouts, especially if the rainfall and snow accumulation was low at that

132

time. Additionally, the moisture in these areas would have protected the sand

from eroding, thus creating a stable surface for the growth of plants.

1988 Land cover: General Discussion

In 1988, several of the large spans of barren sand found in blowouts in
1978 had been colonized by pioneer vegetation in accordance with Olson’s
(1958a) model of sand dune ecological succession. Discontinuous grasses or
grass and shrubs (Appendix B, areas A, B, F, E) clearly represented a bulk of the
pioneer growth in these formerly barren areas. The throats of the blowouts also
grew over with stabilizing vegetation, the majority of which was grass and
shrubs (areas C and D, Appendix B).

Compared to 1978, the land cover in 1988 was less fragmented and
covered larger, more contiguous areas. Field surveys and examination of a
topographic map suggested that the vegetated areas on blowouts spread from
the slopes of ridges into the interior of the blowouts from 1978 to 1988. Because
most pioneer grass species require at least some burial to grow more vigorously
(Olson, 1958 a, 1958c) this suggests that sand deposition was quite active from
1978 to 1988. Although some grass species tolerate low rates of erosion, the
majority of grass species need sand deposition to support vigorous growth
(Olson 1958a). Thus, the dense areas of grass growth suggest that erosion rates

were low from 1978 to 1988.

133

Present day field surveys show that area B is located on the slope of a
blowout limb. This area saw marked growth of discontinuous grass from 1978 to
1988. This would suggest sand deposition was active, conditions might have
been moist, and that erosion in the area was low or not present. At area G there
was a large area covered by barren with isolated plants that was adjacent to
barren sand on the lakeward boundary and forest to the west, indicating possible
burial of the forest cover by actively blowing sand.

Barren sand, discontinuous grasses, and forest accounted for the largest
percentages of the total classified area in 1988, with 17.7%, 11.1% and 63.2%,
respectively (Table 5:8). Barren sand decreased greatly between 1978 and1988
from 29.5% to 17.7% of the total classified area, while discontinuous grasses
increased sharply from 1.6% to 11.1% (Table 5:8). This sharp decrease in barren
sand cover and simultaneous sharp increase in discontinuous grass cover may
signify that erosion was occurring at a slower in 1988 than it was in 1978 and / or
that conditions were moister and more favorable for vegetative growth during
the years leading up to 1988 than the years prior to 1978.

Grass and shrubs also increased slightly for a difference of 1.5% (Table
5:8). Though the change was small, it suggests that a small percentage of the
total classified area progressed through succession from 1978 to 1988 in a manor
congruent with Olson’s (1958a) model on coastal dunes. Barren with isolated
plants, and grass, trees and shrubs showed little or no change between 1978 and

1988.

134

 

.mmemmfim dog uwmwuwom £me 39$sz .wan n 3.24 Uocwmmmfiv Each.

 

 

 

5qu 4883. «88$ 8a m3 an 33.

$85 38ka 8&2 « a a woaamaecb

as we me 382 2083 883: a S a each

. . . £53m

m m m N m N 888 843 09.8 mm mm. a 98 .889 .880

9m .3 3 $92 2.8: 88% 8 B o: 58$ 28 880

an a: 3 83mm 883 5.88 n we a: 8880 38588.5

. . . mafia—AH

o H E o H Baa Baa .808 E a g 3283? ceam

0.3 5: Ba 88% 84% 82a «2 8 8 28m seam
33 $2 $3 $2 82 «SH 83 83 3.3 .36

«34 $5 «34 13H. Bowen—om
Vomwmnnd—Umo «Goawm HO hwenuq—z wan—H.
82 Ba £2 ES 886 seem as 8;. 33 9a Bowboa Lo Lanaaz ES

.wum «Bah.

135

The total number of polygons has decreased from 548 to 375 between 1978
and 1988 (Table 5:8). This decrease verifies that in 1988, land cover was less
fragmented than in 1978. More specifically, nearly all the land cover
classifications with vegetation decreased with regard to number of polygons,
while barren sand increased only slightly. This means that vegetative cover
expanded from 1978 to 1988 (Table 5:8).

In 1988, it is interesting to note that the total area was smaller than in the
1978 and the 1999 interpretations. This was due to higher lake levels recorded on
the 1988 imagery. The high lake level decreased the size of the beach, therefore
decreasing the total area of the park. Though 1978 and 1988 had approximately
the same average annual lake level, the 1988 imagery may have been captured
just after a storm, explaining why the lake level might have been higher than

what was captured in the 1978 imagery.

1999 Land cover: General Discussion

Land cover in 1999 was even less fragmented than in 1978 or 1988.
Present day field studies showed that area A (Appendix C) is located within a
blowout, thus, in 1999 must have also been within a blowout. This blowout was
nearly filled in with vegetation in 1999. Grasses and shrubs dominated the
lakeward throats of these blowouts, a possible indication that these blowouts are
healing. The blowouts at areas B and C (Appendix C) also have extensive areas

of grass and shrubs or discontinuous grasses at the lakeward throats, implying

136

 

 

that these blowouts may also be healing as described in Gares and Nordstrom
(1995)

At area D (Appendix C) there was a very large area covered by the barren
with isolated plants class, with an open area of barren sand to the west. Field
investigations show that this area is currently the slip face of a parabolic dune.
Slip faces undergo sand deposition if the dune is active. Barren with isolated
plants cover at D suggests that in the years leading up to 1999, this dune was
active. As sand slipped down the lee slope of this dune, it buried forest cover
behind it. However, field evidence showed that while this dune may have been
migrating over pre—existing forest cover, new, pioneer species of trees were
taking root on the slip face as well. Olson (1958a, 1958c) noted that trees such as
cottonwoods are often pioneer tree species, especially on the steep lee sides of
dunes and can tolerate moderate rates of sand deposition.

The area of barren with isolated plants cover at area E (Appendix C) is
also adjacent to a lakeward area of barren sand, but is abutted by forest to the
east. However, field investigations of this area show that this area corresponds
with the windward slope of a dune. Exposed tree roots and fallen trees observed
during present day field studies are a testament of the severe erosion in this area.
Thus, in 1999, the large area of barren with isolated plant cover here indicates
that the area was most likely undergoing erosion in 1999 also. Sand was eroding

away at a high rate, exposing tree roots and undermining trees.

137

At area F (Appendix C), a large area of grass, trees and shrubs cover
existed, though this area was still rich with trees. Field studies showed that this
area also corresponded with the slip face of a large parabolic dune. A likely
scenario here was that sand had been sliding down the slip face of the dune at a

slow rate, buried some of the more sensitive understory vegetation and made the

 

 

vegetation less dense in 1999 than in the 1988 and 1978 imagery.

Field surveys concluded that discontinuous grasses at area G (Appendix

 

C) were located on a ridge that gradually rises running from west to east (as in
Appendix A, area F). Grasses and shrubs dominated the flatter, previously E
mined topography to the south of the ridge. The presence of grasses and shrubs
in this area indicates a slow rate of sand deposition and erosion. It also indicates
a maturing plant community at this location in accordance with Olson’s (1958a)
model of ecological succession.
Barren sand, discontinuous grasses, and forest accounted for the largest
percentages of classified area in 1999, with 16.0%, 12.4% and 61.8%, respectively
(Table 5:8). Barren sand decreased only slightly between 1988 and 1999 for a
difference of 1.7%, while discontinuous grasses increased slightly from 11.1% in
1988 to 12.4% in 1999 (Table 5:8). This suggests that over a park-wide scale, very
little of the most stable and mature cover (forest) regressed to other, less mature
land cover. However, at specific regions within the park, changes from forest

cover to less mature cover are noted.

138

Grass and shrubs also increased from 3.7% in 1988 to 5.0% in 1999 (Table
5:8). Additionally, grass, trees and shrubs increased slightly from 2.8% in 1988 to
3.3% in 1999 (Table 5:8). The expansion of these more mature vegetation covers

suggests some ecological succession had occurred in a manner similar to that

described by Olson (1958a).

 

 

Overlay Discussion

 

Overlay maps of the land cover changes from 1978-1988 and 1988-1999 are
presented in Appendices 4 and 5, respectively. These maps are included to show i
and compare the change in land cover for each time period studied. The maps
have been simplified into three categories of land cover changes in context of
Olson’s (1958 a, 1958c) research on the ecological succession of Lake Michigan
sand dunes. The three categories are successive land cover changes, unchanged,
and regressive land cover changes.

The presence of plant cover can provide a relative measure as to the
geomorphic conditions of dunes (Olson, 1958c). Land cover changes from a less
mature or barren stage in an earlier time to a denser, more mature form of
vegetation cover in a latter time period indicate a successive land cover change
had occurred. If vegetation communities had progressed forward through
ecological succession, erosion rates may have decreased to a rate to support the
succession of vegetation communities. Additionally, some land covers indicate

moderate rates of sand deposition (discontinuous grasses, grasses and shrubs for

139

example), while others are dominant when deposition has slowed or is non-
existent (such as grasses, trees, and shrubs cover or forest). Regardless of
deposition rates, an increase in vegetation density indicates succession has
occured. Areas undergoing succession are progressing towards dune
stabilization in accordance with Olson’s model of ecological succession (1958a;
1958 c).

In unchanged areas, land cover had not changed during the indicated
decade of study. Conditions in these areas were not suitable for vegetation to
succeed and mature, nor for pioneer vegetation to establish. Alternatively, not
enough time had passed for vegetative cover to mature to the next stage (Olson,
1958a). Additionally, conditions in unchanged areas did not allow for the
degradation of the existing vegetation.

Regressive land cover signifies that vegetation that had existed at an
earlier time period had been reduced. Conditions at these areas had caused
vegetation communities to degrade to a less mature state or to disappear.
Erosion or deposition rates in these areas may have increased to reflect this
reversal in the ecological succession as described by Olson (1958a). Thus,
because stabilizing vegetative cover had been lost, these areas were destabilizing
and geomorphic processes that cause the migration of sand were activated.

To quantify the results, figures and graphs are provided along with the
discussion of each time frame of the study. The changes from one land cover

type to another are discussed in terms of whether the change was successive,

140

 

unchanged or regressive. The breakdown of these categories is listed in Table 4:6,
of the methods section, for review. Unless stated otherwise, percentages of
change from one classification to another are reported as a percent of total
successive, regressive or unchanged land cover, and not as a percentage of the
total park classified area. Land cover changes from 1978 to 1988 are discussed

first, followed by changes from 1988 to 1999.

1978 - 1988 Visual Overlay Analysis

Appendix D shows the changes in land cover for the 1978—1988 overlay
analysis. During this time frame, the northernmost blowouts at area A
(Appendix D) had filled in with vegetation cover. The majority of species were
most likely pioneer vegetation such as Ammophilia breviligulata (American beach
grass; Olson, 1958a). Field investigations suggest that these areas were either the
limbs of parabolic dunes, topographically lower areas near blowout hollows, or
flatter, interior dune regions.

The regressive land cover found at area B (Appendix D) corresponds to a
slip-slope of a parabolic dune, according to field investigations at that site. Here,
sand was being deposited along the forest margin, burying pre—existing trees.
However, some pioneering species of trees invaded the slope from 1978, to 1988,
explaining the presence of some small trees seen on the slope in the 1988 imagery
that were undetectable on the 1978 imagery. This suggests that while the dune is

migrating and burying over pre-existing, mature forest cover, conditions here are

141

 

still favorable for the growth of new, pioneering trees (Olson 1958a, 1958c).
Perhaps the new trees on the slope will be able to stabilize the movement of this
dune in the future.

Thin, linear bands of regressive land cover at areas C, D, and E (Appendix
D) corresponded with the present day locations of foredunes. Because the land
cover changes were regressive and not successive, the rate of sand deposition
may have greatly increased from 1978 to 1988 to exceed the tolerance level of the
vegetation species located there. Alternatively, the regression of vegetative cover
may indicate the rate of erosion on these foredunes increased from 1978 to 1988.
Upon closer examination, it was evident that the majority of these foredunes had
regressed from grass, trees, and shrubs in 1978 to grasses and shrubs in 1988.
The relatively quick loss of trees during this decade suggests that this area was
most likely undergoing erosion.

In the southernmost blowouts at areas F and G (Appendix D) successive
land cover changes occurred. Field investigations suggest that these areas
corresponded with the limb slopes of parabolic dune and the flatter, interior
areas of the blowouts. The presence of successive cover here suggests that these
limbs were progressing towards stabilization.

The large areas of regressive land coverthat surrounded area H (Appendix
D) are in an area of flat topography. Field investigations suggest that area H was
at a windward slope of a dune. A small band of regressive land cover at H was

abutted by barren sand to the west, and forest to the east. This pattern suggests

142

that pre—existing mature forest cover was being destroyed by erosion on the
windward slope (Baurer and Sherman, 1999).

Field studies show that Area I coincides with the slip face of a parabolic
dune. In 1978, the regressive land cover here suggests that sand was actively
burying pre-existing forest cover at area I. In the 1988 imagery, extremely large,
tall trees were notably surrounded by barren sand at this spot. No new growth
appeared to have occurred from 1978 to 1988, suggesting that this area was
experiencing high rates of sand deposition.

At area] (Appendix A), a large area of successive land cover occurred in
an area that corresponds with the flatter, interior region of a blowout and also up
the northern slope of a limb. The presence of successive vegetation at this
location suggests that the area was progressing towards stabilization from 1978

to 1988.

1978 - 1988 Overlay Analysis: Quantitative Results

Although the former section discussed qualitative results in land cover
change in specific areas around the park, this section details the quantitative
results lof land cover change from 1978 to 1988 for the entire park. From 1978 to
1988, the largest percentage of classified land was in the unchanged category and
amounted to 82%. The breakdown by land cover classification is displayed in
Figure 5.1. The classifications that held the largest percentages of unchanged

areas from 1978 to 1988 were forest (75.1%) and barren sand (20.0%).

143

 

 

80.0

70.0

60.0

40.0

30.0

Percent of Unchanged Area

20.0

10.0

0.0

 

 

75.1

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

20.0
0.4 1-6 1.0 1'9
r": fl m 7“]
BS BWI DG F GS GTS

Overlay Classlflcatlon

 

Figure 5.1. Graph showing overlay percentage of unchanged land cover

by overlay classification, 1978-1988. *BS: barren sand, BWI: barren with
isolated plants, DG: discontinuous grasses, F: forest, GS: grasses and shrubs, GTS:

grass, trees, and shrubs.

The percentage of classified area that was covered by successive land
cover was the next largest portion of the study area, accounting for 13.8% of the
total classified area. Within the successive land cover changes category, the four

largest changes were from barren sand in 1978, to vegetated land in 1988 (Figure

5.2). Barren sand to discontinuous grasses was the most common type of

successive land cover change from 1978 to 1988, accounting for 59.1% of all the

areas that experienced ecological succession.

144

 

 

This value is distinctly higher than changes from barren sand to grasses
and shrubs, which accounted for the next largest percentage of the total area
covered by successive land cover (14.6%). The high percentage of barren sand to
discontinuous grasses suggests that grasses were most likely to colonize barren
sand before other plant species and that ecological succession was proceeding as
described by Olson (1958a).

Many dune grasses require sand deposition for vigorous growth (Olson
1958 a, 1958c). The large percentage of barren sand to discontinuous grasses
implies that this decade was marked by a rate of sand deposition that did not
exceed the growth rate of the grasses, and even favored this growth. It also
implies that the rate of sand deposition was too rapid in most areas to allow the
growth of larger plants, such as shrubs.

It is interesting to note that barren sand to forest accounts for 5% of the
successional land cover changes (Figure 5:2). It is highly unlikely that within 10
years barren sand cover would convert to forest. The majority of this conversion
is most likely explained by changes in tree canopy. A small portion of the areas
that changed from barren sand in 1978 to forest in 1988 was at the margins of
blowouts where sand abuts forest cover. Tree canopy changes add a small
amount of bias to the results only where the changes are from very immature
ecological stages (barren sand or discontinuous grasses) to forest and the areas

are at the margins of dense forest cover.

145

 

59.1

 

 

 

 

30.0 -

 

20.0

 

 

Percent of Successlve Vegetative Cover

10.0 ~

 

. 2.1
20 1-2 0.4 0.5 0.7 0.8

 

 

 

 

0.0- '-

 

BWI BWI BWI BWI DG DG 06 GS GS GTS
to to F to to to F to to to F to to F
06 GS GTS GS GTS GTS

  

 

 

Overlay Classification

 

Figure 5.2. Graph showing overlay percentage of successive land cover

by overlay classification, 1978-1988. *BS: barren sand, BWI: barren with
isolated plants, DG: discontinuous grasses, F: forest, GS: grasses and shrubs, GTS:
grass, trees, and shrubs.

Regressive land cover changes account for 3.5% of the total classified area
from 1978 to 1988. The largest portion of regressive land cover was from grasses
and shrubs in 1978 to discontinuous grasses in 1988 (26.0%; Figure 5.3). This
suggests that in a majority of the areas undergoing degradation, the rate of sand

d8position or erosion exceeded the tolerance of some shrub species.

146

 

 

 

 

 

 

 

 

  

 

 

 

  

 

 

 

 

26.0
25.0 -7 . . - ALVMA-
S
>
0
g 20.0 . __ ,,.____
E, .
i-
> f.‘
3 15.0
I
8
§ 1o_g 11.0
1 f
h
3 10.0 9‘5 :¢A_-A_
‘ § g: 7.8
s j.
l I l
; 5.0 ‘ a”, E - __,z
. " V:
l ..
0.0l »" .-

.13 -

BWI DG 06 Fto Fto Fto Fto Fto GS GS GS GTS GTS GTS GTS
to to to BS BWI DG GS GTS to to to to to to to
BS 88 BWI BS BWI 06 BS BWI DG GS

Overlay Classlflcatlon

 

 

Figure 5.3. Graph showing overlay percentage of regressive land cover by
overlay classification, 1978-1988. ”BS: barren sand, BWI: barren with isolated
plants, DG: discontinuous grasses, F: forest, GS: grasses and shrubs, GTS: grass,
trees, and shrubs.

The next largest portion of regressive land cover was from grass, trees,
and shrubs in 1978 to grass and shrubs in 1988. It is likely that this type of
change implies erosion in these areas. When compared to the digital
orthophotos, a majority of these areas were found on windswept slopes of
blowouts, suggesting that the sand was being swept from under the roots of very

small, immature trees, thus undermining the trees until they fell over. Some of

147

the grasses and shrub species that may have accompanied these trees in 1978
were more tolerant to erosion than the small trees that had disappeared by 1988
and thus, were still detectable in 1988. .

Another interesting pattern is the amount of change from forest cover in
1978 to barren sand in 1988. A small portion of this type of degradation was
biased by changes in tree canopy, but the bias was only detectable in small bands
at the margins of the forest. As noted before, Appendix D shows an area at H
and at I where sand was burying forest. Qualitatively, this recession of the forest
boundary is detectable on the digital orthophotos at these areas (Figure 5:4).
Additionally, field observations revealed that dead wood littered the barren sand
near the forest boundary, suggesting that sand is being eroded away from the

root systems of large trees.

1988 - 1999 Visual Overlay Analysis

Appendix E shows the changes in land cover for the 1988-1999 overlay analysis.
Long, linear regions of successive land cover occurred in the majority of
foredunes within the park at A, B, and C (Appendix E). Sand deposition
probably occurred at a moderate rate from 1988 to 1999 to support the growth of
vegetation on these foredunes (Olson, 1958 c; 1958d).

However, the foredunes at area D (Appendix E), which also corresponds

with the High-Use Area, have regressed from 1988 to 1999. This would suggest

148

 

that the foredunes at area D had experienced erosion during this decade,

possibly from the presence of beach visitors in this area.

 

 

 

 

EXPLANATION ? 2‘1” “1’0 6"” F9“
l
0

Forest Boundary. 1978 50 100 1&0 Meters

Forest Boundary. 1988

 

Figure 5.4. Recession of forest cover as seen at areas H and I (Appendix D).

In the northernmost blowouts at area B (Appendix E), some large areas of
regressive land cover occurred in the central region of these blowouts. Modern
field investigations suggest these areas corresponded with a windward slope.
The regressive land cover implies that erosion occurred in this blowout between
1988 and 1999. According to Baurer and Sherman (1999) a blowout undergoing
erosion has not yet reached it's deflation limit and will continue to erode and

migrate until it does.

149

The small blowout at area F (Appendix E) shows a large area of successive
land cover in the backdune area, suggesting that for the most part, sand
deposition or erosion decreased from 1988 to 1999 in the backdune area of this
blowout. However, degradation of the land cover occurred through some of the
interior, windswept slope of this blowout and near the lake.

Within the southern blowout at area G (Appendix B), were large areas of
successive land cover. Field studies show that these areas coincide with the
slopes of blowout limbs. Vegetation plays a role in the formation of parabolic
dunes by stabilizing the limbs while allowing the central part to advance with
the wind (Livingstone and Warren, 1996). This finding suggests that the dunes
here behave in a matter consistent with other dunes.

Some of the successive land cover at area G (Appendix E) also
corresponds to an area of flat topography. As previously stated, field
investigations and examination of historic aerial photography suggest this area
had been mined. The surface here is in a topographic low and may collect
moisture. This low, moist area would encourage the growth of trees that need
moisture and lengths of time to germinate from seed (Olson, 1958a).

To the west, at area H (Appendix E), is a pronounced area of regressive
land cover. Field investigations suggest that this area coincides with the slip face
of a parabolic dune. On slip faces of dunes, sand particles get deposited near the
crest. When the amount of deposited sand exceeds a critical angle, slope failure

occurs, resulting in a'flow of sand down the slip face (Livingstone and Warren,

150

 

1996). At area H, sand deposition and slope failure from 1988 to 1999 was high
enough to bury massive trees.

At areal (Appendix E), an additional, massive area of regressive land
cover was observed. This was located on the main, windward slope of a high
dune and therefore was experiencing erosion (Livingstone and Warren, 1996).
The erosion here was strong enough to undermine trees and therefore the land
cover regressed.

Lastly, area J was another large area of regressive land cover. This area,
like I, was on the slip face of a dune. However, the rate of sand deposition was
not apparently high enough to constantly bury large trees. The area had actually
converted from forest cover in 1988 to grass, trees and shrubs cover in 1999,
suggesting that sand may have buried understory vegetation, making the cover

less dense in this area.

1988 - 1999 Overlay Analysis: Quantitative Results

From 1988 to 1999, the largest percentage of total classified land was in the
unchanged category, amounting to 86.2%. The breakdown of the unchanged
category by land cover classification is displayed in Figure 5:5. The land cover
classifications that held the largest percentages of unchanged areas were forest
(71.3%) and barren sand (15.0%).

The percentage of total classified area that was covered by successive land

cover represented the next largest portion at 8.6%. Within the successive land

151

 

cover areas, barren sand to discontinuous grasses was the most common change

from 1988 to 1999, accounting for 36.2% of the total area covered by successive

 

 

80.0

 

 

70.0

 

60.0

 

 

 

 

 

 

 

30.0

 

Percent of Unchanged Area
‘5
O

 

 

 

 

 

 

 

 

 

 

 

 

 

 

15.0 1
’1
10.0 1' '51:;
‘33“: 0 5 2.3 1.9
0.0 I " . ' , W . F's—J
as BWI F GS GTS
Overlay Classification

 

 

 

Figure 5:5. Graph showing overlay percentage of unchanged land cover by

overlay Classification,1988-1999. *BS: barren sand, BWI: barren with isolated
plants, DG: discontinuous grasses, F: forest, GS: grasses and shrubs, GTS: grass,
trees, and shrubs.

vegetation (Figure 5:6). This value was considerably higher than the next highest
type of successive land cover change, from discontinuous grasses in 1988 to

grasses and shrubs in 1999 (19.3% of the total area experiencing succession).

152

 

Therefore, in areas where successive land cover was found, conditions were
more favorable for grasses to colonize barren sand than they were for shrubs to
grow into areas already stabilized by grasses. In accordance with Olson’s (1958a)
model of ecological succession on Lake Michigan dunes, the dominance of grass
as pioneer vegetation suggests that sand deposition occurred in a majority of the
areas where succession had occurred.

The third highest percentage of successive land cover changes was from
barren sand to grasses and shrubs (7.7% of the total area undergoing succession).

This finding suggests that within the cumulative areas that had successive land

 

cover changes from 1988 to 1999, there was a modest percentage of area where
ecological succession happened at a faster rate than for example, areas that had
progressed from barren sands to discontinuous grasses. This is evident from the
growth of both grass species and shrubs on barren sand within a decade.
Regressive changes in land cover account for 5.3% of the total classified

area from 1988 to 1999. Therefore, over the entire park area, degradation of the
vegetation was occurring at a slower rate than vegetation could mature. The
largest .change in regressive cover was from discontinuous grasses in 1988 to
barren sand in 1999, which accounted for 14.6% of the cumulative areas
undergoing degradation (Figure 5:7).

However, grasses and shrubs in 1988 to discontinuous grasses in 1999 was a
nearly equivalent type of land cover change (14.5% of the cumulative areas

undergoing degradation; Figure 5:7). This suggests that from 1988 to 1999, the

153

most likely types of vegetation to be lost were pioneers. This could imply that in
this decade, the rate of sand deposition within some areas of the park was too
high to support plant growth, or, conversely, that the rate of erosion was too

high for these pioneering species to grow.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

60.0
50.0
S
0
5 40.0
g 36.2
> _:
.3; 300
g
2 .
e 7 19.3
20.0 5: f
a. F ';
.fi :21
10.0 7“: i:
5.0 5-2 4.5
[I 33 25 25 t an 1.8 .
'1 3:: . ;
0.0 [] Unsung-«,1n
BS 38 BS BS BS BWI BWI BWI BWI DG DG 06 68 GS GTS
to to to F to to to to F to to to F to to to F to to F
BWI 06 GS GTS DG GS GTS GS GTS GTS
Overlay Classification

 

 

 

Figure 5:6. Graph showing overlay percentage of successive land cover

by overlay classification, 1988-1999. *BS: barren sand, BWI: barren with
isolated plants, DG: discontinuous grasses, F: forest, GS: grasses and shrubs, GTS:
grass, trees, and shrubs.

The next common types of recessive land cover changes were both from

forest in 1988 to a less dense vegetative cover in 1999. Forest in 1988 to grass,

154

 

trees and shrubs in 1999 occurred in 8.6% of the total regressive area, while forest
in 1988 to barren sand in 1999 occurred in 8.5% (Figure 5:7). Two large areas of
these types of changes were found at areas I and] (Appendix B). As explained

earlier, area I was on the windward slope of a high dune and is undergoing rapid

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

erosion.
!
16.0
14.6 14.5
"1 r- x
14.0 i
g 12.0
0
§ 10.0
:8: 86 8.7
i F _
a 8.0 67 71
2 - i
9 if 7'1 6.0 6.2
s so 51— — — 1..
a r-'-
4.1
3.8
E 40 *— r-— P. —‘ -— 36 fl — 34 A —1
20¢ ~— —- —1 ~— —- —— — — —
1.0
0.0 V I l V I I l I D! U I I T
BWI DG 06 Fto Fto Fto Fto Fto GS GS GS GTS GTS GTS GTS
to 88 to as to as BWI DG GS GTS to BS to to to BS to to to
BWl BWI DG BWI DG GS
Overlay Classification

 

 

 

Figure 5:7. Graph showing overlay percentage of regressive land

cover by overlay classification, 1988-1999. *BS: barren sand, BWI: barren
with isolated plants, DG: discontinuous grasses, F: forest, GS: grasses and shrubs,
GTS: grass, trees, and shrubs.

Figure 5:8 shows the forest boundary at each time period studied and

depicts the rapid rate of forest recession at this area. Area J, as mentioned

155

previously, was on the slip face of a dune. Because this area had converted from
forest cover in 1988 to grass, trees, and shrubs in 1999, there is the implication of
some possible deposition of sand over understory vegetation. However, the rate
of burial on this slope is probably moderate, given that the land cover

degradation has not been dramatic.

 

 

EXPLANATION <3 I I
I I
0

Forest Boundary, 1978

 

-—-— Forest Boundary, 1988
Forest Boundary. 1999

Figure 5:8. Forest recession 1978-1999 at (area 1, Appendix E).
A comparison: 1978-1988 and 1988-1999

The intent of this section is to compare the rates of land cover change over
the entire study area from 1978 to 1988 and 1988 to 1999. The middle time frame,

1988, was included to determine whether changes in land cover occurred evenly

156

over the period of study or if change had occurred at different rates. Comparing
results on a decadal scale, as opposed to the entire period of study, provides
insight as to whether land cover changes in the park are over the short term or
long term. Generalized land cover changes (successive, unchanged, or
regressive) for each decade are shown in Figure 5:9 as a graph comparing
percentage of total classified area.

The results of this study suggest that the rate of change was not constant
throughout the period of study. Figure 5:9 shows that from 1978 to 1988, 13.8%
of the total classified area was successive land cover, while 3.5% of the total
classified area was regressive land cover. From 1988 to 1999, the amount of
successive land cover had decreased to 8.5% of the total classified area, a
difference of 5.3% from the previous decade. Regressive land cover increased
slightly from 1988 to 1999 to 5.3%. The percentage of land cover remaining
unchanged (on a decadal scale, not over the entire period of study) also increased
from 82.8% of the total classified area in 1978-1988 to 86.2% in 1988-1999.

These results indicate that in both decades, successive land cover changes
covered a greater area than regressive land cover changes. However, a greater
percentage of successive land cover changes occurred from 1978 to 1988 than
from 1988 to 1999. Additionally, both decades saw some regression of land
cover, though the percentages of land cover regression were relatively low in
both decades. The findings of this study suggest, however, that degradation of

land cover occurred at a greater rate in 1988 to 1999 than from 1978 to 1988.

157

 

 

 

90
8O -

60‘
50-
40~
30~

2° ‘ 13.8

10 - ' has . if
0

Successive Unchanged Regressive
Land Cover

 

[:1 1978-1 988
I 1 988-1999

 

 

Percent of Classified Area

 

 

 

 

 

 

 

 

 

 

Figure 5:9. Land cover change as percent of entire classified area: 1978-1988
vs. 1988-1999.

Possible Factors Influencing Land Cover Changes

This section offers some environmental and cultural factors that could
explain the apparent differences in land cover change between 1978 to 1988 and
1988 to 1999 for the entire park. Vegetative growth or degradation may be
impacted by climatic factors such as precipitation or intense storm winds, or
indirectly affected by cultural factors such as the number of visitors who
contribute to disturbance of the vegetative cover. Additionally, since vegetation
responds to sand deposition and erosion, the succession or regression of
vegetation can also be attributed to geomorphic factors such as lake level

changes and sand nourishment.

158

Climate

The extent and type of vegetation can result from the amount of
precipitation in the region. Extremely wet years can encourage very vigorous
vegetative growth, while dry years can inhibit or discourage the growth of
stabilizing vegetation. Precipitation values used in this study were collected

from the Benton Harbor or Eu Claire stations, located near the study area.

m..- 'm

As seen from Figure 5.10, though average precipitation values vary from

year to year, the general trend in average precipitation is steadily falling over the

 

entire period of study. Within the region, the first time frame of the study (1978- a
1988) had a slightly higher average precipitation value (100.51 mm) than the later
time frame, 1988 to 1999 (92.04 mm). Additionally, precipitation was at a relative
peak within the study area in 1981. This could, in part, possibly explain the
greater percentage of successive vegetation from 1978 to 1988 than from 1988 to
1999.

Storms may also have an impact upon vegetation. The high winds during
storms can destroy vegetation on sand dunes. Patches of sand that are laid bare
by high storm winds may lead to the development of blowouts (Livingstone and
Warren, 1996). Thus, storms, especially those accompanied by high winds, may
initiate erosion in some areas of the dunefield.

During the period of study, storms were classified as thunderstorms with

high winds, hail storms, winter storms, flood events and tornadoes. The

159

intensity of winds measured during these storm events was generally over 74

km / h (46 mph).

 

140

 

120

100

80

60

40

Average Annual Precipitation (mm)

20

 

 

 

«9’ 93° ‘6" 12:“ 8° 15" 9° <8" <2)" <8" 9"
°-’ '9 '9 '9 '9 '9 '9 r9 '3’ '9

Year

 

 

 

Figure 5.10. Average annual precipitation near the study area, 1978-1999.
Source: National Oceanic and Atmospheric Administration, 1978-
1999.

Over the entire period of study, it is evident that amount of storms in
Berrien County had increased rapidly (Figure 5:11). In the later time frame (1988
to 1999), the number of storms had more than doubled from 56 in the first decade
(1978-1988) to 126 from 1988 to 1999 (NCDC, 2003). The increase in storms
during the latter decade may be partially accountable for the increase in

regressive land cover observed from 1988 to 1999.

160

 

 

(as)
O

 

N
01

N
O

 

 

 

Number of Reported Storm:
8 a

01

 

 

0 v

'9 ’9 >9 '9 '9 \°-’ '9 \Q’ '3 '3’ '9

 

Year

 

 

 

 

Figure 5:11. Number of reported storms in Berrien County, 1978-1999. Source:
NCDC, 2003.

Lake-level Fluctuations and Sand Nourishment

As discussed in the literature review of this thesis, lake level is one of

 

several factors that control sand supply to dune systems of Lake Michigan. As
lake levels fall, the widened beach supplies sand to the foredunes (Olson, 1958d).
Thus, foredunes receive sand from the beach and build when lake levels fall.
Because some plant species require sand deposition to thrive, an increase in
pioneering types of vegetation, especially grass species (Olson, 1958a, 1958c), on
dunes may be a response to the increased sand deposition as lake levels fall.

Additionally, while sand is deposited in the foredunes and at the beach, the

161

 

foredunes act as a sand trap, preventing sand nourishment into the inland areas
(Arens et al., 1995). This would create an environment where the blowouts are
sediment starved, and thus, gradually degrade (Livingstone and Warren, 1996).

Conversely, when lake levels are high, waves erode beach ridges and/ or
foredunes, forming a lakeward facing bluff. Dunes may migrate inland as sand
dislodged from the bluff is blown on the leeward side (Marsh and Marsh, 1987;
Loope and Arbogast, 2000). In turn, as foredunes erode, the vegetation cover
will degrade in response to the high rates of erosion. Furthermore, sand can
deposit further inland during high lake levels, thus feeding the large, open
blowout areas. Therefore, in response to this sand nourishment, pioneer species
that require sand deposition for vigorous growth (such as grasses) may colonize
these areas.

Figure 5:12 shows the trends in lake level for Lake Michigan during the
period of study. Lake level rose throughout most of the first decade of the study
(1978-1988) and peaked at record levels in 1986. After that peak, lake levels
began to fall. As described earlier, marked regression of land cover occurred at
foredunes between 1978 and 1988 (areas C, D, and E, Appendix D). This
suggests that the foredunes were possibly being eroded during the high lake
levels before the image in 1988 was captured, thus, degradation of the land cover
occurred in response.

Additionally, this decade of the study (1978 to 1988) was marked by a

higher percentage of successive land cover than the later time frame (1988 to

162

 

 

 

1999; Figure 5:9), especially with regard to changes from barren sand in 1978 to
discontinuous grasses in 1988. Most of these areas were inside blowouts (as in
area A, Appendix D). Because pioneer species of grasses require some amount of
sand deposition for vigorous growth (Olson 1958a, 1958c), this finding suggests
that inland sand deposition may have increased in a manner described by Marsh
and Marsh (1987) and Loope and Arbogast (2000) as a response to the high lake
levels during this decade of study.

Conversely, the later time frame of the study was characterized by low

 

lake levels (Figure 5:12). After the record high in 1986, lake level fell off rapidly
until 1991. Lake level then began a gradual rise, but the peak did not reach the
high levels of the earlier decade, nor was it long lived. Because lake levels were

lower from 1988 to 1999 than in 1978 to 1988, beaches would have been broader

 

as discussed in Olson (1958d) and Loope and Arbogast (2000).

The broad beaches would have supplied sand to the foredunes and
initiated foredune growth (Olson 1958d). Growth of the foredunes is detectable
through changes in land cover that show succession of vegetation (areas A, B,
and C on Appendix E). The majority of these land cover changes were from
barren sand in 1988 to discontinuous grasses in 1999. The growth of grass
species in these areas would imply moderate rates of sand deposition because
these species are dependent upon sand deposition for growth (Olson, 1958a;

1958c).

163

 

 

177,40

177,20 1

177,00 1

176.80 1

 

176.60 1

176.40 1

176.20 1

Lake Level (m, IGLDBS)

176,00 1

175.80 1

 

 

 

175.50 x v r n 1 fi a v 7 Y V v v r
«$49898‘81868‘898993‘8%ege°e‘eme5e“ebebé\e%eg
'3 '3‘ ’3‘ '9 '3‘ \°-’ '9 '9 '3’ '3’ ’9 '9 ’3‘ \q’ :9 ’9 '9 \°-’ '3' '9 '3’ \9

Year

 

 

 

 

Figure 5:12. Average annual lake levels for Lake Michigan, 1978-1999. Source:
USACE, 2003.

Because the sand was being stored within the foredunes, this would have

 

created a negative sand budget in the throats of the blowouts. A negative sand
budget on blowouts causes erosion on the windward slopes. This time frame of
study saw increased rates of land cover degradation over the earlier decade,
especially on windward slopes of blowouts and on leeward slopes where sand
was burying mature vegetation cover. This suggests that during the lower lake
levels from 1988 to 1999 the blowouts underwent rapid rates of erosion on

windward slopes, with sand depositing in the lee.

164

Park Visitor Numbers

Park visitors can contribute to changes in land cover by trampling
vegetation, therefore causing degradation of vegetative cover and erosion. The
presence of recreational users can also prevent the growth of vegetation in areas
that are barren and unstable, thus preventing the stabilization of dunes.

As seen in Figure 5:13 average number of park visitors increased from
1,346,978 within the first decade of study (1978-1988) to 1,317,082 in the later

decade of study (1988-1999). Thus, park visitor numbers increased slightly over

 

the entire period of study, taking a slight drop in 1986, but working up to a peak
in 1999 of 1,827,025 (Herta, Harold, personal com., 2002).

Although visitor numbers increased over the period of study, it is difficult
to establish a link between number of park visitors and land cover change at
park-wide level. Additionally, it is difficult to make any solid correlations
between park visitor numbers and land cover changes at the temporal scale of
this study. To make correlative studies, a finer temporal scale would be

desirable, possibly at a scale of 1-3 years between image capture dates to provide

 

more data for land cover change.

However, the possibility human impact on land cover may still be
explored by examining the area where park visitors are concentrated. The area
of high recreational use can be compared with the area managed as natural area.
The intent of the following section is to employ this method to further explore

the relationship between human impact and land cover changes.

165

 

2000000
1 800000
1 600000
1 400000

 

 

  
  
 

 

 

 

 

 

 

 

 

 

 

 

Estimated Number of Park Visitors

 

 

 

 

 

Figure 5:13. Estimated Park Visitor Numbers, 1978 to 1999. (No bar indicates
data unavailable for that year). Source: Herta,
Harold, personal com., 2002.

High-Use Area Versus Natural Area

As stated previously in the methods section of this thesis, the study area

 

was divided into a High-Use Area and a Natural Area to determine if
recreational use had any impact on the land cover changes within the park over
the period of study. For simplification of the results, only the overlay from 1978
to 1999 was compared with regard to the High-Use and Natural Areas.
Appendix F demonstrates how these two areas compare visually.

In the Natural Area (Appendix F), it is clear that successive land cover

changes were extensive over the northern blowout complex. Comparison with

166

the original imagery and field studies show that these areas that had undergone
succession coincide with north-facing slopes of ridges that extend eastward
through the blowout complex, backdune areas, swales between dune ridges near
the mouth of the blowout and blowout hollows.

Because the sun shines more intensely upon southern slopes in Michigan,
southern slopes are hotter and drier than northern slopes. The presence of
denser vegetation might suggest that conditions on the northern slopes were
moister and cooler, thus, favoring vegetative growth on these slopes.
Additionally, the initial growth of vegetation in swales and blowout hollows
conforms to Olson’s (1958a) research.

The northernmost blowout area appears to have been progressively
healing throughout the study interval. Figure 5:14 demonstrates this pattern. In
1978, large areas of barren sand were apparent in the throat region and were
interconnected with the beach area. Scattered discontinuous grasses and grasses
and shrubs provided some stabilizing land cover. By 1999, these areas had filled
in with denser vegetation. Thus, as vegetative cover expanded over the study
period, areas of barren sand within the blowout interior were progressively cut

off from the beach system and sand supply.

167

 

 

 

 

 

1999

1988
EXPLANATION

1978

 

 

Figure 5:14. Blowout healing from 1978 to 1999.

168

[:1 Discontinuous Grasses . Grass,Trees. and Shrubs

I: Barren Sand

E—

- Forest

Barren with Isolated Plants [3 Grass and Shrubs

 

Area A on this map (Appendix F) is a large area of regressive land cover
that corresponds with the slip face of a dune. Over the two decades of study, a
substantial amount of sand has blown over the crest of this dune and deposited
on the leeward side, as described by Livingstone and Warren (1996). When sand
deposition reached a threshold, slope failure occurred and progressively buried
the abutting forest on the lee side of the dune (Figure 5:15).

However, further examination of this area in the field and on the original

imagery shows that new trees have taken root upon this slope. According to

 

Olson (1958 a, 1958c) cottonwoods and willows may act as pioneer species if
conditions are moist and the rate of sand deposition does not impede seed
germination. Although it is not discemable from the aerial photography, these
small trees may be cottonwoods or willows. This would suggest that although
sand deposition was active on this slope, the surrounding tree cover may have

provided cooler, moister conditions for pioneer tree species to grow.

 

Area B (Appendix F) also corresponds with the slip slope of a dune, where
sand blown over from a high ridgeline is burying land cover located on the
leeward side of the ridge. There was no evidence of pioneer tree species on this
slip face. This would suggest that sand deposition is too rapid and/ or conditions
are to dry for tree seeds to germinate. This same pattern occurred at areas C and
D (Appendix F) in the High-Use Area.

A smaller extent of the blowouts within the High- Use Area was marked

by successive land cover changes. Thus, from a qualitative standpoint, it appears

169

that ecological succession was retarded in the High-Use Area relative to the
Natural Area.

The successive land cover at area E (Appendix F) appears to be most
dense on the north-facing slope of a ridge that elevates as it runs eastward. As
discussed previously, vegetation growth was probably favored on this slope
because it receives less sun than the southward-facing slopes; therefore more

moisture is retained in here.

   
   

6
.m-
:A.

   

5!.

A

K

   

    
   

   

  

. L ' . raw-.3 1
. 'Slrp face ovendr -

- ‘4‘.

forest cover":

  

 

Figure 5:15. Ikonos image showing slip face of dune (Area A
Appendix E).

The successive land cover found at area F (Appendix F)was in a
previously mined, topographically flatter region between limb ridges. Though

previously disturbed by industrial activity, this area grew in with stabilizing

170

 

 

vegetation fairly quickly. It is uncertain, however, whether vegetation was
planted in an attempt to reclaim the mined land, or whether the ecological
succession that occurred in this area was a natural phenomenon. As previously
mentioned, the low topography in this area may have created a cooler, edaphic
environment where vegetation thrived.

Unlike the Natural Area, extensive areas of regressive land cover are

 

observable in the High-Use Area. Large areas of regressive land cover at areas G,

I, and] (Appendix F), are located on the slip faces of parabolic dunes, which are

 

associated with depositional environments. Figure 5:16 is a present day [9:
photograph of the slip face at area G. Small trees can be seen growing upon the
slip face. Upon further examination of the imagery, grasses also once coexisted
with these trees (in 1978 and 1988).
However, by 1999 and until present, the grasses had nearly completely

died away. This would suggest that sand deposition had increased enough over

 

the period of study to prevent the growth of grasses and to bury trees rooted in
the slip face. Figure 5:17 is a modern photograph of the slip face at area I. Here,
tall trees are being buried by sand, and the dune encroaches upon a picnic area.
It is evident that this area is undergoing rapid rates of sand deposition.

At area H (Appendix F), the large area of regressive land cover is most
likely under an erosional environment due to its location on a windward slope of
a dune (Livingstone and Warren, 1996). This is also an area that gets heavy

human foot traffic in the summer, due to the location of the dune with respect to

171

the beach area. Figure 5:18 is a modern photograph taken on this slope. It is
evident from the photograph that erosion is very high in this area. The roots of

living trees are exposed as sand is eroded away from the surrounding area.

 

 

 

 

Figure 5:17. Slip face of a dune (Area I, Appendix E).

172

 

Figure 5:18. Slope at Pike’s Peak showing erosion. Note exposed roots (Area
H, Appendix E).

Figure 5:19 quantifies the differences between the High-Use Area and the
Natural Area. Within the Natural Area, land cover had undergone succession in
a manner described by Olson (1958a) over 19.0% of the total classified area. In
contrast, 14.8% of the total classified area had progressed into a more ecologically
mature state within the High-Use Area.

These changes are less than the successive land cover changes found in
the Natural Area. Thus, successive land cover changes are greater in the Natural

Area than in the High Use Area by 4.2% from 1978 to 1999. Because successive

173

 

 

land cover changes are a sign that dunes are progressing towards stabilization
(Olson 1958a), it could be stated that a greater percentage of the dunes in the

Natural Area are progressing toward stabilization than in the High-Use Area.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

90
79.4
30 1M ___.__ _ 7751‘“ __.__
70 111
< so .1 _
a so +1
g Dngh-Use Area
'5 4° 1 INaturalAree
3 so r
°- 19.0
20 11 1 —-~ - <~~v~~r
14.8
10 1
0 .
Successive No Change
Land Cover Charge

 

 

 

Figure 5:19. Land cover change as percent of entire classified area:
High-Use Area vs. Natural Area.

 

When considering regressive land cover changes, the High-Use Area has a
higher percentage of total classified area than the Natural Area by a margin of
2.5% (Figure 5:19). Thus, a larger percentage of the dunes in the High-Use Area
are progressing towards destabilization. Additionally, the area remaining
unchanged in the High-Use Area is also slightly higher than the Natural Area (by

1.9 percent).

Table 5:9 provides a summary of the greatest percentage of land cover

changes within each area of the park from 1978 to 1999. Forest in 1978 to barren

174

sand in 1999 accounted for the highest percentage of the total area undergoing
land cover regression in the High-Use Area (16.3%). Because this type of change
is so dramatic,,it implies that areas that had undergone degradation in the High-
Use Area had rapid rates of erosion, sand deposition, or both. As described
before, a large area of regressive land cover changes experiences high
recreational activity and is located on windward slopes (Area H, Appendix F).
Thus, erosion (both naturally occurring and human induced) may be the most

dominant process that causes land cover degradation than sand deposition in the

 

High-Use Area.

Table 5:9. Summary of Largest Percentage Land Cover Changes by

 

 

Area.
i-iqh-Use Area Natural Area

Percent of Regressive

Area F>BS 16.3% GS>DG 22.6%
Paw“ firsumss'v" es>oe 44.0% es>oe 67.6

ea

Percent of Unchanged F 585% F 365%

Area BS 26.1% BS 9.8%

 

 

*BS: barren sand, DG: discontinuous grasses, F: forest, GS: grasses and shrubs.

Conversely, the largest percentage of regressive land cover changes in the
Natural Area was from grass and shrubs in 1978 to discontinuous grasses in
1999. This type of land cover change accounted for 22.6% of the total regressive
area in the Natural Area. The environmental changes were great enough that
within regressing areas, shrubs could not sustain growth. However, grasses,

which are especially tolerant to sand deposition, could maintain growth (Olson,

175

1958a). This suggests that in the Natural Area, sand deposition may be the most
dominant process causing land cover regression in the Natural Area.

With regard to successive types of land cover changes, discontinuous
grasses in 1978to barren sand in 1999 was the largest percentage of total
successive land cover area in both the High-Use and Natural Areas (Table 5:6).
However conversion occurred over a greater amount of area in the Natural Area
than in the High-Use Area (by 23.6%). The higher percentage suggests that
conditions were more favorable for grasses to colonize barren sand in the

Natural Area than in the High-Use Area. The reasons could be two-fold. First,

 

because grasses grow vigorously during active sand deposition (Olson 1958a,
1958c), this may suggest that sand deposition is a dominant process in the
Natural Area. Second, it may also suggest that the area is less disturbed by foot
traffic than in the High-Use area, since grasses are sensitive to trampling
(Roethle, 1985).

With respect to unchanged land cover, forest accounted for the highest

 

percentage of total unchanged area in both the Natural Area and the High-Use
Area. However, only 68.5% remained forest in the High-Use Area, as compared
to 86.5% in the Natural Area. This suggests that the most mature and therefore
most stable land cover type changed to a greater extent in the High-Use Area
than in the Natural Area. 1

Barren sand covered the second greatest amount of unchanged area in

both regions of the park. In the High-Use Area, barren sand covered 26.1% of the

176

unchanged areas as compared to 9.8% of the unchanged area in the Natural
Area. This suggests that greater areas of unstabilized sand were exposed in the
High-Use Area than in the Natural Area from 1978 to 1999. Therefore, more land
cover was vulnerable to mobilization and erosion in the High-Use Area than in

the Natural Area.

Human Impact in the High-Use Area

Human impact is an explanation for at least some of the higher percentage
of regressive land cover changes in the High-Use Area. Recurrent trampling of
plant species by park visitors may have destroyed stabilizing plant species, thus
leaving sand exposed to eolian transport. Additionally, as dune climbers
scoured the face of Pike’s Peak (see H on Appendix F) and the surrounding
areas, each footstep would have caused at least a small amount of slumping
along the windward face of the dune. This process would easily expose sand to
eolian mobilization.

Perhaps the greatest area of interest in terms of human impact upon the
High-Use Area was adjacent to the park beach and the beach parking lot. Aerial
photography reveals that in the early 1970’s, the parking lot was expanded from
one to include three parking areas. With this construction, foredunes that had
once existed on the beach were destroyed. Since foredunes store excess sand
provided by the beach, there was no longer an area to store excess sediment.

Recreational use of the beach inhibits new foredune formation, as does the

177

 

 

presence and maintenance of the beach parking lot. Without foredunes to
capture the influx of sand from the beach environment, the sand movement to
the blowout areas east of the parking lots is virtually unchecked.

Most of the recreational activity on the beach is focused upon the central
parking lot and beach house (Figure 5:20). The concession stand is at this central
beach house and draws many people towards this central area. This central
beach house contains a restroom that is also open more often during the year
than the southern and northern beach houses. Additionally, the northern
parking area is closed for a large part of the year, centralizing recreational
activities. The massive areas of regressive land cover changes (Figure 5:20) lie
just to the east of this central parking area. Heavy recreational activity focused in

this area may have contributed to regression of land cover.

178

 

 

 

   

:: Successive Land Cover Changes E Regressive Land Cover Changes

Figure 5:20. Central parking lot and beach house in relation to areas of
regressive and successive land cover.

179

 

Chapter 6

Conclusions and Recommendations

This study had several goals. The first goal was to determine which
method of orthorectification produced imagery with the best positional accuracy
for use in land cover change analysis of Lake Michigan coastal dunes. In this
thesis, two methods to orthorectify historic aerial photography were tested. One
method was to use a USGS 30 m DEM to produce the orthoimagery. The second
method was to extract a 3 m DTM from the imagery once it had been block
triangulated and to use the extracted DTM to orthorectify the imagery.

Results from this study suggest that extracting a DTM from the imagery
and using it to produce the orthoimages produced the most positionally accurate
results. Therefore, the DTM extraction method was used to produce orthoimages
for the land cover interpretation analysis. Although this method produced
superior results in terms of positional accuracy, it should be noted that this
method required extensive amounts of time to collect GPS measurements in the
field. The terrain makes it physically challenging to collect GPS measurements.
Additionally, some GPS measurements were impossible to collect due to terrain
difficulties, signal scatter from forest cover or surrounding ridges, or land
ownership issues.

If this method were to be implemented to monitor land cover change in

additional parks with massive coastal dunes, careful mission planning would

180

 

 

need to be utilized before GPS measurements are made. Furthermore, ample
amounts of time and funding would be required to allow field personnel to
obtain the GPS measurements.

The second goal of this study was to determine where change had
occurred on a coastal dunefield and what types of change have occurred in
context of previous work conducted by Olson (1958a; 1958c). The results of this
study suggest that the major areas where land cover changes had occurred over
the period of study were on foredunes, windward slopes of blowouts, and the

leeward slopes of parabolic dunes and blowout crests.

 

Land cover changes essentially followed the model of ecological
succession on Lake Michigan sand dunes suggested by Olson (1958a). That is,
pioneer grasses were usually the first to colonize barren sand when conditions

were suitable, followed by pioneer tree species that are small enough to appear

 

as shrubs on the photographs, or shrub species that grow when sand deposition
or erosion has greatly decreased or stopped. As land cover progresses through
succession, vegetation density increases.

In the context of literature regarding the ecology of Lake Michigan coastal
dunes, the vegetation cover on coastal dunes can be an indicator of the relative
rates of sand deposition (e.g. Olson 1958a, Olson 1958c; Olson 1958d). Likewise,
regression from a more mature state of land cover in one time period to a less
mature state of land cover can indicate increases in the rate of sand deposition or

erosion. Because of the apparent link between geomorphic activity and

181

vegetation growth, observations in land cover change between time frames
placed the results of this study in the context of research that links dune building
to lake level fluctuations (Olson, 1958d; Marsh and Marsh, 1987; Arbogast and
Loope, 1999; Loope and Arbogast, 2000; Arbogast et al., 2002).

Overlaying land cover interpretations between time periods addressed the
third goal of this thesis: to determine whether change occurred at a constant rate
or was variable over the period of study. The results of this study suggest that

rates of land cover change were variable over the period of study and may be

 

tied to changes in lake level.

The first period of the study (1978 to 1988) experienced a greater
percentage of successive land cover changes than the later time frame (1988 to
1999), especially with regard to changes from barren sand in 1978 to
discontinuous grasses in 1988. The appearance of grass species indicates an

increase of deposition in areas of successive land cover. The majority of these

 

areas were within blowouts. Additionally, from 1978 to regressive of land cover
changes were observed in foredune areas. This time period also coincided with a
period .of record lake level highs. These observations suggest that the foredunes
were probably being eroded during the high lake levels. While the foredunes
were eroded during the high lake levels, sand was transported inland to nourish
the blowout areas, as evident by the increase in vegetation that requires sand

deposition for vigorous growth.

182

The second time frame of the study (1988 to 1999) had a greater
percentage of regressive land cover changes than the first decade, with a majority
of the regression occurring on windward slopes of blowouts and on slip faces of
dunes. Additionally, the latter time frame also saw successive land cover
changes in the foredunes as opposed to regression in the earlier decade. This
time frame was characterized by lower lake levels than the earlier decade.

Low lake levels created broader beaches and sand deposition was directed
into the foredunes from 1988 to 1999. As a response, pioneer species that thrive
during sand deposition colonized the foredune areas. Because foredunes trap
sediment, the blowout areas became sediment starved, thus the blowout areas
underwent erosion. Degradation of forest cover on windward slopes and burial
of vegetation on slip faces is evidence of this erosion from 1988 to 1999.

The last goal of this thesis was to determine how land cover changes
compare between the High-Use Area and Natural Area over the entire study
interval. Within each region, succession and regression were both observed,
while most of the areas remained unchanged.

However, from the data collected from the land cover interpretation and
overlay analysis, it would appear that the Natural Area experienced a greater
percentage of successive land cover changes from 1978 to 1999 than the High-Use
Area. Moreover, the High-Use Area saw greater percentages of regressive land

cover changes than the Natural Area.

183

 

 

 

Evidence from historic newspapers suggests that these areas were active
prior to the foundation of the park. However, the results from this investigation
suggest that recreational use of the dunes may be promoting the destruction of
stabilizing vegetation in these areas. Land cover regression was focused at a
location east of the central beach house and parking lot, on the windward slope
of Pike’s Peak. Ultimately, these areas will probably continue to degrade so long
as the slopes are open for recreational use.

Erosion will continue to occur on the windward slopes and sand
deposition on the leeward slopes will cause burial of mature forest cover to the
east of these intensely used areas. At present, the only park infrastructure and
services in danger of being buried on the opposite side of this erosion is a small
picnicking area and small access road to the picnic area.

Although there may not be many structures in danger of burial at this
park, the results of this study may have implications in other parks where coastal
dunes attract high numbers of visitors. For example, in Petoskey State Park,
active, unstable dunes have been encroaching upon a campground in the park
(Lepczyk, 2001). Remotely sensed land cover change studies could be used by
the MDNR to monitor the succession or regression of vegetation on coastal dunes
and thus, serve as a proxy measurement for the movement of sand. By
understanding the natural and human induced processes occurring on coastal
dunefields, the MDNR can prevent loss of infrastructure and property, such as at

what may occur at Petoskey State Park.

184

 

 

An immense amount of potential exists for further studies within WDSP,
and for future research upon other dune fields of the Lake Michigan coastline.
Observations in the field revealed several exposed paleosols in blowouts around
the park. At least one of these paleosols appeared to have strong development,
as evident by a thick, organic-rich A-horizon and brownish colored B-horizon.
This paleosol appears to be consistent in development with the well-developed
soil in the upper part of dunes in the Holland area (Arbogast et al., 2002).

Sedimentological studies of the dunes and radiocarbon-dates obtained from the

 

paleosols could provide better insight as to how and when these massive dunes
evolved. Additionally, this type of research would place the dunes at WDSP
within the context of other dunefields on the Lake Michigan coastline, thus
providing a better understanding of coastal dune evolution in the region.

For future remote sensing studies of coastal dunefields, it is recommended
that the imagery be captured at a less coarse time interval, possibly yearly, to do

correlation or regression studies with quantitative data that may explain changes

 

in the dune environment (such as lake levels, precipitation, wind, temperature,
number of visitors, storm events, fires, sediment supply, and erosion and sand
deposition rates). With advances in satellite technology, it is now easy to obtain
very high-resolution images without the need of ordering costly aerial
photography flights.

Ikonos or QuickBird imagery could be substituted in place of aerial

photography. Regardless of the method of image capture, it is highly

185

recommended that ground targets are set up on the date of image capture and
that GPS measurements are taken at these targets to increase the available points
in the ground control network. This will likely improve the positional accuracy
of the orthoimages.

Additionally, it is recommended that the link between vegetation cover
interpreted from remotely sensed imagery and geomorphic processes be tested.
Sand traps and erosion pins could be used on the dunes to measure rates of
deposition and erosion (Livingston and Warren, 1996; Arens et a1. 2002).
Remotely sensed imagery could be used simultaneously with these sand
transport measures to test the relationship between geomorphic processes and
land cover change.

This investigation is a beginning step toward future studies that could
examine the dynamics of Lake Michigan dunes on a small time scale using
remote sensing techniques. The goal of such studies would be to study present-
day phenomena in a manner consistent with the ideas of uniforrnitarianism.
Observing coastal dune behavior in the present might increase our
understanding of ancient processes that have formed these dune fields.
Additionally, these studies would allow us to predict forward into the future to
project how these dunes may change in response to changes in climatic or
cultural influences. An extensive body of knowledge regarding Lake Michigan
coastal dunes will lead to sound management strategies and policy decisions to

conserve and/ or preServe these valuable natural and recreational resources.

186

 

 

Appendices

187

 

___— . .""'tJ-:v.~" _

 

Appendix A. Land cover in WDSP, 1978.

  

 

F’H—ffige

Land Cover
Warren Dunes State Park Study Area
1978

High-Use Area

 

    
 
 
 
  
    
   
   
    
   

Eli-DID

EXPLANATION

Barren Sand

Barren with Isolated Plants
Discontinuous Grasses
Grass and Shrubs

Grass, Trees, and Shrubs
Forest

Unclassified

 

Warren Dunes State Park

   
 
      

 

 

" High-Use Area

 

 

Appendix B. Land cover in WDSP, 1988.

 

Land Cover EXPLANATION j
Warren Dunes State Park Study Area Barre” sand l
1988 Barren with isolated Plants ‘

Discontinuous Grasses
Grass and Shrubs

  
  
  
 
     
   
 

Grass, Trees, and Shrubs

 

Forest

Ell-EL:D

Unclassified

High-Use Area

 

   

Natural Area

Warren Dunes State Park

 

  
  

Natural Area

I
0 200 400 600 800 Meters

 

 

1' HighAUse Area

 

 

t
i
1
1
§ 0 1,000 2.000 3,000 Feet
1
t
t

 

 

189

 

 

Appendix C. Land cover in WDSP, 1999.

 

 
 
   
      
    
   
   
 

Land Cover EXPLANATION
Warren Dunes State Park Study Area Barren Sand
1999 ; Barren with Isolated Plants

Discontinuous Grasses
Grass and Shrubs
Grass, Trees, and Shrubs

Forest

   

Unclassified

High—Use Area
,..,ng a“ p

 
   
 
   
   

Natural Area

Warren Dunes State Park

Natural Area

0 1,000 2,000 3,000 Feet
0 200 400 600 800 Meters

High-Use Area

 

190

 

 

 

Appendix D. Land cover change in WDSP, 1978—1988.

 

1 Land Cover Change EXPLANATION
Warren Dunes State Park Study Area
1978—1988

I Successive Land Cover

- Unchanged

- Regressive Land Cover

D Unclassified

 

      
  

High-Use Area

Warren Dunes State Park

Natural Area

Natural Area

 
 

 

 

0 1,000 2,000 3,000 Feet
I | ‘ l II |
0 200 400 600 800 Meters

HighAUse Area

191

 

 

 

Appendix E. Land cover change in WDSP, 1988-1999.

 

Land Cover Change
Warren Dunes State Park Study Area
1988—1999

 
 
 

 

Natural Area

0 1,000 2,000 3,000 Feet
i

i I
| | l I
0 200 400 600 800 Meters

 

192

High-Use Area

 

 

EXPLANATION

I Successive Land Cover

I Unchanged

- Regressive Land Cover

D Unclassified

  
  

Warren Dunes State Park

 

  
  

Natural Area

 

High~Use Area

 

 

 

 

 

Appendix F. Land cover change in the High-Use Area compared to the Natural Area, 1978-1999

 

 

 

Land Cover Change

Warren Dunes State Park Study Area
1978-1 999 I Successive Land Cover

I Unchanged

I Regressive Land Cover

D Unclassified

EXPLANATION

Natural Area

   

High-Use Area

  
     
  

Warren Dunes State Park

 

  

Natural Area

0 1,000 2,000 3,000 Feet
l i

l l l | l

O 200 400 600 800 Meters

 

 

High-Use Area

 

 

193

 

 

 

 

 

List of References

The Acorn. (1919). ”E. K. Warren,” from The Acorn, 16 February 1919.
Volume XXVII, No, 10, in the Warren Family Papers, Michigan State
University Archives and Historical Collections, East Lansing, MI.

Arbogast, A. F.and Loope, W. L. (1999). Maximum-lirniting ages of Lake

Michigan coastal dunes: their correlation with Holocene lake level
history. I. Great Lakes Res, 25(2):372—382.

Arbogast A.F., Hansen, E.C., Van Oort, MD. (2002). Reconstructing the
geomorphic evolution of large coastal dunes along the southeastern
shore of Lake Michigan. Geomorphology, 46:241-255.

Arens, S.M., Slings, Q., and de Vries, C.N. (2002). ”Remobilisation of a
parabolic dune in Kennemerland, the Netherlands,” in Lee, J. A. and
Zobeck, T. M., Proceedings of ICAR5/GCTE-SEN Joint Conference,
International Center for Arid and Semiarid Lands Studies, Texas Tech
University, Lubbock, Texas, USA Publication 02-2 pp. 431-435.

 

Arens. S. M. Van Kaam-Peters, H. M. E., Van Boxel, J. H. (1995). Air flow over
foredunes and implications for sand transport. Earth Surf. Process.
Landforms, 20:315-332.

 

Bauer and Sherman. (1999). ”Coastal dune dynamics: problems and
prospects,” Chapter 4 in Goudie et al. (Eds.), Aeolian Environments
Sediments, and Landforms (pp. 71-74). West Sussex, England: John
Wiley & Sons Ltd.

 

Berrien Springs Era. (1917). ”E. K. Warren Offers 20 Mile Park to Michigan,”
a clipping from the Berrien Springs Era, 1917, in Warren Family
Papers, Michigan State University Archives and Historical Collection,
East Lansing, MI.

Brown, D.G., and Arbogast, A.F. (1999). Digital photogrammetric change
analysis as applied to active coastal dunes in Michigan. PE&RS, 65(4):
467-474.

Buckler, R.W. (1979). ”Dune Inventory Barrier Dune Classification Study of

Michigan’s Lake Michigan Shore.” Michigan Department of Natural
Resources, Geological Survey Division Investigation 23, Lansing, MI.

194

Businksi, S. A. (1992). Airphoto Interprietation of Vegetative Cover Change
on the Sleepingfiear Dunes Comlex, Leelanau County, Michigan,
1938-1987. Unpublished master’s thesis, Michigan State University.

Carter, R.W.G. (1988). Coastal Environments. San Diego, CA: Academic Press
Inc.

Carter, R. W. G., Nordstrom, K. F., and Psuty, N. P. (1990). ”The study of
coastal dunes,” Chapter 1 in Carter, R. W. G., Nordstrom, K.F. and
Psuty, N. P. (Eds), Coastal Dunes: Form and Process (pp. 1-14). West
Sussex, England: John Wiley 8: Sons Ltd.

Chandler, J. (1999). Effective application of automated digital
photogrammetry for geomorphological research. Earth Surf. Process.
Landforms 24:51-63.

Chrisman, N. (1997). Exploring Geographic Information Systems New
York: John Wiley 8: Sons, Inc.

Chrzastowski M. J., and Thompson, '1‘. A. (1992). Late Wisconsinan and
Holocene coastal evolution of the southern shore of Lake Michigan.

Quaternary Coasts of the United States: Marine and Lacustrine
Systems, SEPM Special Publication No. 48, pp. 397-413.

Collins, J. [Internet] (2002). Military Geography for Professionals and the
Public. [Last updated September, 30, 2002. Accessed August 26, 2003].
Available from http: / / www.ndu.edu/ inss / books / Books%20-
%201998 / Military%20Geography°/o20March%2098 /
milgeocontentshtml

Congalton, R. G. (1988). A comparison of sampling schemes used in
generating error matrices for assessing the accuracy of maps generated
from remotely sensed data. Photogrammetric Engm' eering 8: Remote
SensingI 54(5):593—600.

Corner, P.J., Albert, D.A., Wells, H.A., Hart, B.L., et. al., (1995). Michigan’s
Native Landscape, As Integpreted from the GLO Surveys 1816-1856.
Lansing, MI: Michigan Natural Features Inventory. Digital Media.

Cowles, H.C. (1899). The ecological relations of the vegetation on the sand

dunes of Lake Michigan. Eotanical Ga_zette_, 27:95-117, 167-202, 281-308,
361-391. -

195

 

deBlij, H. D. and Muller, P. O. (2000). Geography: Realms, Regions and
Concepts. Ninth edition. New York: John Wiley 8: Sons, Inc.

DigitalGlobe, (2003). [Internet]. Longmont, CO. QuickBird Imagery Products
FAQ [updated 7/18/03; accessed August 28, 2003]. Obtainable online
as Adobe Acrobat PDF at http: / / www.digitalglobe.com/

Dorr, J. A., and Eschman, D. F. (1970). Geology of Michigfl Ann Arbor,MI:
The University of Michigan Press.

Dow, K. W. (1937). The Origin of Perched Dunes on the Manistee Moraine, F“
Michigan. 5

ERDAS, Inc. (2001). IMAGINE OrthoBASE User’s Guide including IMAGINE
Orthobase Pro. Atlanta, GA: ERDAS, Inc.

 

Eschman, D. F. (1985). Summary of the quaternary history of Michigan, Ohio, 1
and Indiana. Iournal of Geological Education, 33:161-167.

Farrand, W. R. (1988). The Glacial Lakes around Michigan. Bulletin 4.
Lansing, MI: Michigan Department of Environment Quality, Geologic
Survey Division.

Fransaer, D. (1992). Hydrographic and photogrammetric surveys in coastal
environments. Zeitschg‘ft furfiVermessungswesen, 8-9z487-497.

Federal Geographic Data Committee (FGDC). (1998). Geospatial Positioning
Accuracy Standards Part 3: National Standard for Spatial Data
Accuracy. FGDC-STD-007.3-1998. Available online as Adobe Acrobat
PDF at:: http: / / www.fgdc. gov / standards/ documents/ standards/
accuracy/ chapter3.pdf

Galien River Watershed Project. (2002). ”A Pioneering Conservationist.”
_Gilien R_iver Watershed Project, Newsletter. Volume 1. January, 2002.

Gares, PA. and KP. Nordstrom, 1995. A cyclic model of foredune
blowout evolution for a leeward coast: Island Beach, New
Jersey. finals of the A_ssocigtion of American Geographers.
85(1): 1-19.

Gesch, D., Oimoen, M., Greenlee, 8., Nelson, C., Steuck, M., Tyler, D. (2002).

The National Elevation Dataset. Photogrammetric Engm' eering 8:
Remote Sensing 68(1):5-32.

196

Gillette, D. A. (1999). ”Physics of Aeolian Movement Emphasising Changing
of the Aerodynamic Roughness Height by Saltating Grains (the Owen
Effect).” Chapter 6 in Goudie A.S., Livingstone, I., and Stokes, S. (Eds).
(1999). Aeolian Environments Sediments and Landforms. West
Sussex, England: John Wiley and Sons.

 

Goudie A.S., Livingstone, I., and Stokes, S. (Eds). (1999). Aeolian
Environments, Sediments andJLandforms. West Sussex, England: John
Wiley and Sons.

Gutschick, RC. and Gonsiewski, J. (1976). Coastal geology of the Mt. Baldy
Area, Indian Dunes National Lakeshore. In: Wood, W.L. (Ed)., Coastal
Sedimentation and Stability in Southern Lake Michigfl Geol. Soc.
Am. No. Centen. Meet, Field Trip Guidebook, Kalamazoo, MI, pp. 38-
90.

Hansel, A. K., Mickelson, D.M., Schneider, A. F ., Larsen, C. E. (1985). Late
Wisconsinan and Holocene history of the Lake Michigan basin.
Quaternary Evolution of the Great Lakes, Geological Association of
Canada Special Paper 30.

Harrell, M. H. (2001). Development of a digital protocol for vegetation
mapping. Unpublished master’s thesis, North Carolina State

University.

Hazlett, B. T. (1981). The I-Iistogy, Vegetation and Flora of Nordhouse Dunes

Man_istee National Forest, Mason County, Michigan. Unpublished
master’s thesis, Michigan State University.

 

 

Herta, Harold, Operations Unit Supervisor, Parks and Recreation Bureau,
Michigan Department of Natural Resources. Written communication,
September 6, 2002.

Hesp, P. A. and Thorn, B.G. (1990). Geomorphology and evolution of active
transgressive dunefields. Chapter 12 in Nordstrom, K.F. and Psuty, N.
P. (Eds), Coastal Dunes: Form and Process (pp. 253-288). West Sussex,
England: John Wiley 8: Sons Ltd.

Hill, Richard, Park Manager, Warren Dunes State Park. Telephone
communcation, November 29, 2000.

Hoagman, W. J. (1994). A Field Guide: Great Lakes Coastal Plants. Tawas
City, MI: Michigan State University Extension.

197

 

Hohle, J. (1996). Experiences with the production of digital orthophotos.
PhotgLammetric Engineering 8: Remote Sensing, 62(10):1189-1194.

Jacobberger, PA. and Hooper, D.M. (1991). Geomorphology and reflectance
patterns of vegetation-covered dunes at Tsoldilo Hills, north-west
Botswana. Int. J. Remote Sensing, 12(11) :2321-2342

Janette, F. E. (1919). ”Great Warren Foundation Little known Wonderland,”
a clipping from The Detroit News, 23 March 1919, in the Warren
Family Papers, Michigan State University Historical Collections, East
Lansing, MI.

Jungerius, P. D., Witter, J. V., van Boxel, J. H. (1991). The effects of changing
wind regimes on the development of blowouts in the coastal dunes of
The Netherlands. Landscape Ecology, 6(1 / 2):41-48.

Lachowski, H. et a1. (1995). Guidelines for the Use of Digital Imagery for
Vegetation Mapping. U. S. Government Printing Office, Washington,
D. C.: United States Department of Agriculture, Forest Service.

Lancaster, N. (1995). Geomorphology of _Desert Dunes. New York, NY:
Routledge.

Larsen, GE. (1985). Lake level, uplift, and outlet incision, the Nipissing and
Algoma Great Lakes. In: Karrow, RF. and Calkin, P.E. (Eds)
mammary Evolptjon of the Great Lakes. Geological Association of
Canada Special Paper 30: pp. 63-77.

Larson et. al. (1980). Soil Survey of Berrien Coung. US. Government
Printing Office, Washington, DC: United States Department of
Agriculture, Soil Conservation Service.

Lepczyk, X.C. (2001). Geomorphic History of Sand Dunes at Petoskey State
Parlg Michigan. Unpublished Masters Thesis, Michigan State
University.

Lillesand, T.M., and Kiefer, R.W. (2000). Remote Snesing and Imagg
Integpretation. Fourth Ed. New York: John Wiley and Sons, Inc.

Litcher, J. (1994). Lake Michigan beach-ridge and dune development, lake

level, and variability in regional water balance. wterryary Research,
44:181-189.

198

 

 

Litcher, J. (1998). Primary succession and forest development on coastal Lake
Michigan sand dunes Ecological Monographs, 68(4):487-510.

Little, E. L. (1980). National Audobon Society FieLd Guide to Trees Eastern
Region. New York, NY: Alfred A. Knopf, Inc.

Livingstone, I. and Warren, A. (1996). Aeolian Geomorphology. Essex,
England: Addison Wesley Longman Ltd.

Loope W. L. and Arbogast, A. F. (2000). Dominance of an ~150—year cycle of
sand-supply change in late Holocene dune-building along the eastern
shore of Lake Michigan. Quaternary Research, 54:414-422.

Marino, M. L. (1980). Sand Dune Sguccession: A Comparison of Plant Life
History Chagctenfiics. Unpublished master’s thesis, Michigan State
University.

Marsh, W.M., and Marsh, 3D. (1987). Wind erosion and sand dune
formation on high Lake Superior bluffs. Geogr. Ann. 69A (3-4): 379-
391.

McKee, E.D. (Ed.) (1979). A Sgdy of Global Sand Seas. Geological Survey
Professional Paper 1052. Washington DC: US. Government Printing
Office.

Michigan Department of Natural Resources [Internet]. Lansing, MI: Map of
State Park and Recreation Areas, [copyright 2001-2003, accessed
August 28, 2003]. Available from
http: / / www.michigandnr.com/ publications/ pdfs/ recreationcamping
/ parkmap.asp

Michigan Department of Natural Resources. (1996). Mrren D_unes
State Park. Pamphlet. Lansing, MI: Parks and Recreation Division.

National Oceanic and Atmospheric Administration (NOAA). (1978-2001).
Climatological Data: Michigan. Asheville, NC: US. Department of
Commerce.

National Climatic Data Center (NCDC) [Internet]. (2003). Storm Events
Database. [Updated 2003, Accessed August 26, 2003]. Available from
http: / / www4.ncdc.noaa. gov / cgi-win/ wwcgi.dll?wwevent~storms

199

 

 

The Nature Conservancy and ESRI [Internet]. (1994). USGS-NPS Vegetation
Mapping Program Standardized National Vegetation Classification
System; [updated September 25, 2000, accessed April and June, 2003]
Available from: http: / / biology.usgs gov / npsveg/ aa/ aa.html

Olson, J. S. (1958a). Rates of succession and soil changes on southern Lake
Michigan sand dunes Botanical Gazette, 119(3):125-168.

- (1958b). Lake Michigan dune development. 1. Wind-velocity profiles.
Iournal of Geology, 66(3):254-263.

--(1958c). Lake Michigan dune development. 2. Plants as agents and tools
in geomorphology. Iggrnal of Geology, 66(4):345-351.

--(1958d). Lake Michigan Dune Development: 3. Lake-level, beach, and
dune oscillations. Journiof Geology, 66(5):473-483.

Orme, A. R. (1990). ”The instability of Holocene coastal dunes: the case of the
Morro dunes, California,” Chapter 14 in Carter, R.W.G., Nordstrom,
K.F. and Psuty, N. P. (Eds), Coastal Dunes: Form and Process (pp. 315-
336). West Sussex, England: John Wiley 8: Sons Ltd.

Pye, K. and Bowman, G. M. (1984). The Holocene marine transgression as a
forcing function in episodic dune activity on the eastern Australian
coast. Chapter 9 in Thom, B. G. (Ed). _Co_astal Geomorphology'g
Australia. Australia: Academic Press.

Rees, WC. (2000). Improving the accuracy of low-cost GPS measurements
for remote sensing applications. Int. I. of Remote Sensing, 22(5):871-
881.

Rockford Map Publishers (1954, 1964, 1987, 1991). Berrien County, Michigan
plat book. Rockford, IL: Rockford Map Publishers.

Roethele, Jon. 1985. More, much more than shifting hills of sand are
Michigan's dunes. Reprint from Michigan Natural Resources.
July / August, 1985. Division of Land Resource Programs, DNR.
Lansing, MI.

Schaetzl et a1. (2002). Measurement, Correlation, and Mapping of Glacial Lake

Algonquin Shorelines in Northern Michigan. Annals of t_he Association
of Amergan Geogi'aphers, 92:399-415.

200

 

 

Sherman, DJ. and Hotta, S. (1990). ”Aeolian sediment and transport: theory
and measurement,” Chapter 2 in Nordstrom, K.F. and Psuty, N. P.
(Eds), Coastal Dunes: Form afirocess (pp. 15-38). West Sussex,
England: John Wiley 8: Sons Ltd.

Short, AD. (1988). Holocene coastal dune formation in southern Australia: A
case study. Sedimengy Geology, 55:121-142.

Shortridge, A.M., and Clarke, KC. (1999).” On some limitations of square
raster cell structures for digital elevation modeling”. In: Lowell, K.,
and Jaton, A. (Eds). Spatial AccurJacy Assessment: Lapd Information
Uncertainty in Natural Resources, Ann Arbor, MI: Ann Arbor Press.

Somers, D., Smith, J. and Wissmar, R. (1991). Watershed and stream channel
cumulative effects analysis using aerial photography and ground
survey data : interim report, TFW-SH8-91-001. Olympia, WA: Timber,
Fish and Wildlife Project, Washington State DNR.

Stembridge, E.J. (1978). Vegetated coastal dunes: Growth detection from
aerial infrared photography. Remote Sensig of the Egironment, 7:73-
76.

Stone, Byron D. 2001. "Surficial Geologic Map of Berrien County, Michigan."
Scale 1 to 100,000. US. Geological Survey Open-File Report 01-0156.

Thieler E.R.and Danforth, W.W. (1994). Historical shoreline mapping (I):
Improving techniques and reducing positioning errors. |oumal of
Coastal Research 10(3):549-563.

 

Thompson, TA. (1992). Beach-ridge development and lake-level variation in
southern Lake Michigan. Sedimenta_ry Geology, 80:305-318.

Thompson, TA. and Baedke, SJ. (1995). Beach-ridge development in Lake
Michigan: shoreline behavior in response to quasi-periodic lake-level
events. Marine Geology 129:163-174.

Thompson, TA. and Baedke, SJ. (1997). Strand-plain evidence for late
Holocene lake-level variations in Lake Michigan. GSA Bulletin 109(6):
666-682.

Trimble Navigation Ltd, (2003). GPS Pathfinder Systems. Sunnyvale, CA:

Trimble Navigations Ltd. (PDF Version of document downloaded May
25, 2003).

201

 

 

 

 

Tueller, RT. (1989). Remote sensing technology for rangeland management
applications. Journal of Range Management, 42(6) 442-453.

United States Army Corps of Engineers (USACE) [Internet](2003). Great
Lakes Hydrographs. Lake Michigan / Huron Hydrograph. [Accessed
on August 26, 2003. Available at:
http: / / www.1re.usace.army.mil / index.cfm?chn_ID=13838:lake_id=2

USGS. ”Bridgman Quadrangle.” 1:24,000

Van Oort, M., Arbogast, A., Hansen, E. C., Hansen, B. (2001).
Geomorphological history of massive parabolic dunes, Van Buren
State Park, Van Buren County, Michigan. Michigan Academician
XXXIII:175-88.

Wells, Mitch. Park Manager, Saugatuck Dunes State Park. Telephone
interview October 1, 2000.

Wisconsin Department of Natural Resources (WI DNR), 2001. ”Global
Positioning System (GPS) Accuracy Report,” 02/ 07/ 01.

Wilson, Steven, E. (1979, 2000). Michigan’s Sand Dunes. Pamphlet 7.
Michigan. Department of Natural Resources, Geologic Survey
Division. Lansing, MI.

Woodhouse, W. W. Jr. (1978). Dune Building and Stabilization With
Vegetation. Special Report No. 3. US. Army, Corps of Engineers.
Coastal Engineering Research Center. Fort Belvoir, VA.

Yao, H. and Clark, R.L. (2000). Evaluation of sub-meter and 2 to 5 meter

accuracy GPS receivers to develop digital elevation models. Precision

Agg'culture, 2:189-200.

202

f.‘
l