s .
7. $4.“.
. ’

:5. i...

2%.?

It.u.
x, (2.3 ‘ ..
.vunflhvmWI-JHWQ ...
. i 1.1.: a 1i}
.4"... . ht

-

.~
”no...
~ cut. .
2 5 ”an? ‘
. \nlbv o
A“? v: .... .. “H!
as “by? \
HRS . . t ..\M. 7
if C “I”.
‘ , V ‘ ‘ ,3.
dinno (v ‘Laanufl
‘ . .. . : v,
. . 3?... ‘ . 57 in”...
‘7. 1, 9 £1
:9
AC.

.

t .11: xii.

. W a

.‘I \A u: ‘33: L.
‘1!~

a
.. uflu.
.L shhtbi
.{Nflnnnhnéh
1‘ I .
$.ufl1hasmi
3.) {I .
‘ Wm ‘-
A>¢JJ
. ‘ .rln
Extvn":
.:A&.« 1
‘ .X :3.
5““ “a‘ VI.
. :3 grufif 1. ‘\ 5‘s... .5."
A i 3.1.. . .3... klikwwyfluflu
» nun: t

st:

 

 

s 5-, n T II.
..1vm§uzduwn.

._ an.» 4‘

t: x 1.1.?

 

LIBRAFI
MICI’II; , 3tYate

University

 

 

 

This is to certify that the
thesis entitled

THE INFLUENCE OF LAND COVER/ LAND USE
CHARACTERISTICS ON
SHUTTLE RADAR TOPOGRAPHY MISSION (SRTM) ELEVATION
ERROR: CASE STUDIES FROM LOUISIANA AND THAILAND

presented by

TARA LOUISE LALONDE

has been accepted towards fulfillment
of the requirements for the

MS. degree in Geoggghy

 

 

  

 

   

 

Major rofessor’s Si ature
Z ' ‘1’ vZaaj
Date

MSU is an affirmative-action, equal-opportunity employer

_. ...-.—-—-a-—----—-.—--—--.-o—--u—-—-—--—-~—c—-—------—-

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

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

5/08 K:IProj/AocaPresIClRC/Dateoue.indd

 

THE INFLUENCE OF LAND COVER/ LAND USE CHARACTERISTICS ON
SHUTTLE RADAR TOPOGRAPHY MISSION (SRTM) ELEVATION ERROR:
CASE STUDIES FROM LOUISIANA AND THAILAND

By

Tara Louise LaLonde

A THESIS

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

MASTER OF SCIENCE
Department of Geography

2008

ABSTRACT

THE INFLUENCE OF LAND COVER/ LAND USE CHARACTERISTICS ON
SHUTTLE RADAR TOPOGRAPHY MISSION (SRTM) ELEVATION ERROR:
CASE STUDIES FROM LOUISIANA AND THAILAND
By
Tara Louise LaLonde

The Shuttle Radar Topography Mission (SRTM) DEM has near-global coverage
with its 3 arc second resolution (~ 90 m). This research explores the relationships
between the SRTM DEM and other DEMS both in the United States and elsewhere. This
thesis investigates the magnitude and variance of SRTM DEM error using online
reference data Light Detection and Ranging (LiDAR) data for two Louisiana study areas
and SRTM X-band (~25 m resolution) for a portion of Nang Rong, Thailand. The
correlations of slope values and elevation values among LiDAR, National Elevation
Dataset (NED), and SRTM DEMS were determined. The average SRTM DEM error was
found to be positive for all study areas, which indicates an elevation bias of the SRTM. In
comparison to land cover/land use characteristics from the National Land Cover Dataset
(N LCD) 2001, forested categories had greater average SRTM DEM error. MODIS tree
cover indicated significant differences between a high (2 50 %) and low percent (< 50 %)
tree cover for the Louisiana areas; however, relationships between MODIS tree cover and
SRTM error were insignificant for the Nang Rong area. A land use application developed
from criteria of flat topography and low topographic position indicated agreement
between LiDAR 90 m and NED 90 m DEMs; however, different results were found for
the SRTM data. This research contributes to understanding the characteristics of the

SRTM DEM, which will improve upon its use as it is adopted in applications.

Copyright by
Tara Louise LaLonde
2008

ACKNOWLEDGEMENTS

This thesis would not have been possible without the guidance, direction, and
support of my advisor, Dr. Asthon Shortridge, along with my committee members Dr.
Joe Messina and Dr. Catherine Yansa. The engaging conversations with Dr. Ashton
Shortridge during the completion of this thesis were an important part of the research
process and my graduate program experience. In addition, the faculty, staff, and my
fellow Michigan State University geography students provided much support to me
during the my graduate school experience. In addition, the Michigan State University
geography department provided a supportive environment to enrich my educational and
personal endeavors.

I would like to also express gratitude to my other friends and family. My fellow
Spartan geography graduate students provided much encouragement and support. A
Special thanks to Brad Blumer, who during our first semester of graduate school shared
his interest in Light Detection and Ranging (LiDAR) data and geography with me. This
sparked an interest in me to pursue the use of LiDAR data and other elevation models
further in my thesis research.

The Geography Department administrative assistants were helpful and
knowledgeable about logistical manners. The Center for Global Change and Earth
Observations staff and support provided a pleasant work environment. This was greatly
appreciated. Also during my time in graduate school, the MSU Geography Department
provided me with a graduate assistantship, which enabled me to gain valuable teaching

and research experience. The collection of the memories shared with students, faculty,

iv

and fellow graduate students; along with time spent in the classroom and in the computer

lab created my graduate school experience, which I will take with me into the future.

TABLE OF CONTENTS

 

 

 

 

LIST OF TABLES ................................................................................. ix
LIST OF FIGURES ............................................................................... xii
LIST OF ACRONYMS ........................................................................... xv
1. INTRODUCTION - - - -- -- - - - - - - - - 1
1.1. Motivation ................................................................................................................ 4
1.2. Research Questions ................................................................................................ 5
1.3. Significance ............................................................................................................. 8
1.4. Overview of thesis .................................................................................................. 8
2. BACKGROUND - - -- - 10
2.1. Digital Elevation Model Background ................................................................ 10
2.2. Elevation Model Sources ...................................................................................... 12
2.2.1. National Elevation Dataset (NED) ................................................................ 12
2.2.1.1. NED Production Methods ......................................................................... 13
2.2.2. Shuttle Radar Topography Mission (SRTM) .............................................. 14
2.2.2.1. SRTM: Production Issues ......................................................................... 20

2.3. SRTM Elevation Model Error Literature ........................................................... 22
2.4. Implications of Land Cover in SRTM ................................................................ 26
2.5. SRTM Application Use .......................................................................................................... 29
2.6. LiDAR: Light Detection and Ranging ................................................................ 30
2.7. Error Analysis ....................................................................................................... 35
2.7.1. Conceptualization ........................................................................................... 35
2.7.2. Resampling Techniques ................................................................................. 36
2.8. Vertical Datums in Terrain Models .................................................................... 37
3. STUDY AREAS - 39
3.1. Methodology for Site Selection ........................................................................... 39
3.2. N ang Rong, Thailand .......................................................................................... 40
3.3. Selection of United States Sites ........................................................................... 44
3.4. Louisiana study areas ......................................................................................... 46
3.4.1. Acadia study area ........................................................................................... 49
3.4.2. Webster study area ......................................................................................... 53
3.5. Land Use Application ......................................................................................... 58
4. METHODS- - - - - ........ - - 61
4.1. Processing Software .............................................................................................. 61
4.2. Naming conventions ............................................................................................ 61

vi

4.3. Louisiana study areas .......................................................................................... 62

 

4.3.1. LiDAR Data from Atlas, Louisiana Statewide GIS ................................... 62
4.3.2. National Elevation Dataset ........................................................................... 65
4.3.3. Shuttle Radar Topography Mission: C-Band ............................................. 65
4.3.4. Land Cover/Land Use Characteristics ......................................................... 66
4.4. N ang Rong, Thailand study area ......................................................................... 67
4.4.1. C-band SRTM ............................................................................................... 67
4.4.2. X-band SRTM ................................................................................................ 67
4.5. Datum Transformations ....................................................................................... 68
4.6. Terrain Derivatives ............................................................................................... 69
4.7. Error Analysis ....................................................................................................... 69
4.8. Sampling Procedure ............................................................................................. 71
4.8.1. Louisiana study areas .................................................................................... 71
4.8.2. Nang Rong study area .................................................................................... 75
4.9. Comparative Surface Properties ......................................................................... 76
4.10. Application Development ................................................................................... 76
5. RESULTS AND DISCUSSION - - - -- 80
5.1. Louisiana study areas ........................................................................................... 80
5.1.1. Acadia study area .......................................................................................... 80
5.1.1.1. Comparison of Elevation Values ............................................................ 85
5.1.1.2. Comparison of Slope Values .................................................................. 86
5.1.1.3. Error Analysis ........................................................................................ 87
5.1.1.3.]. SRTM-LiDAR DEM Error ............................................................... 9O
5.1.1.3.2. SRTM-NED DEM Error ................................................................... 92
5.1.1.4. Error Relationships to National Land Cover Dataset ........................ 94
5.1.1.5. Select Location Error Comparison ...................................................... 97
5.1.2. Webster study area ........................................................................................ 98
5.1.2.1. Comparison of Elevation Values .......................................................... 101
5.1.2.2. Comparison of Slope Values ................................................................ 102
5.1.2.3. Error Analysis ...................................................................................... 102
5.1.2.3.]. SRTM-LiDAR DEM Error ............................................................. 104
5.1.2.3.2. SRTM-NED DEM Error ................................................................. 107
5.1.2.4. Error Relationships to National Land Cover Dataset ...................... 109
5.1.2.5. Select Location Error Comparison .................................................... 113

5.2. N ang Rong study area ........................................................................................ 114
5.2.1. Comparison of Elevation Values ................................................................ 116
5.2.2. Comparison of Slope Values ...................................................................... 119
5.2.3. Error Analysis: SRTM C-band -SRTM X-band ..................................... 120
5.2.4. Specific Locations Error Comparison ....................................................... 122
5.3. Comparative Analysis: Louisiana study areas and Nang Rong, Thailand 123
5.4. Land Use Application ......................................................................................... 131
5.4.1. Acadia study area ........................................................................................ 137
5.4.2.Webster study area ....................................................................................... 139
5.4.3. Nang Rong study area .................................................................................. 144
5.5. Discussion: Lessons Learned ............................................................................. 151

vii

5.5.1. Louisiana study areas ................................................................................. 151

 

5.5.1.1. Elevation Characteristics ...................................................................... 151
5.5.1.2. Slope Characteristics ............................................................................ 153
5.5.1.3. Error Analysis: Land Cover/ Land Use Relationships ...................... 153
5.5.1.4. Land Use Application ............................................................................ 157
5.5.2. N ang Rong, Thailand ................................................................................... 158
5.5.3. Comparable Tree Cover Analysis ............................................................... 160
5.5.4. Global Implications ...................................................................................... 161
6. CONCLUSIONS - -- 162
6.1. Summary ............................................................................................................. 162
6.2. Limitations .......................................................................................................... 170
6.3. Future Direction of Research ............................................................................ 171
APPENDIX A
APPENDIX B

REFERENCES '

viii

Table 2.1.
Table 2.2.
Table 2.3.
Table 2.4.
Table 2.5.
Table 2.6.
Table 2.7.
Table 3.1.
Table 4.1.
Table 4.2.
Table 4.3.
Table 4.4.
Table 4.5.
Table 4.6.
Table 5.1.
Table 5.2.
Table 5.3.
Table 5.4.
Table 5.5.
Table 5.6.
Table 5.7.

Table 5.8.

LIST OF TABLES

Terrain Parameters ..................................................................... 11

Production methods of the NED DEMS ............................................. 14
NED, SRTM, and LiDAR DEM Characteristics ................................... 20
Summary of SRTM Accuracy Literature ....................................... 25-26
Summary of Publicly Available LiDAR-related online repositories. . . . . . . . ....34
Differences in Resampling Techniques ............................................. 37
Comparison of Datums of Elevation Sources ....................................... 38
Crop-Specific Area Calculated from National Cropland Data (2005) .......... 48
Naming Conventions of Data Types with Descriptions ........................... 62
Reclassification Scheme of the NLCD 2001 USGS Data .................... 71 -72
NLCD Summary Statistics for Louisiana study areas ......................... 73-74
Webster study area Reclassified Values ............................................ 74

Acadia study area Reclassified Values ............................................. 74

Addition Process for the Comparison of DEM Application Results ........... 79
Entire Acadia study area Characteristics ........................................... 83

Slope (degrees) of entire Acadia study area DEMS ................................ 84
Acadia study area Sample Elevation Characteristics (n= 1956).. ........ ........84
Acadia study area Sample Slope Characteristics (n= 1956) ..................... 85

Pearson Correlation Matrix of Acadia study area Elevation Data ............... 85
Pearson Correlation Matrix for Acadia study area Slope Grids. . . . . . . . . . . ........86
Entire Acadia study area Elevation Error Characteristics ........................ 88
Acadia study area Sample SRTM Error Characteristics (n= 1956) ............. 88

ix

Table 5.9. Acadia study area: One-sided T-test Results ....................................... 89

Table 5.10. Acadia study area: Sample Error Characteristics by Land Use/ Land

Cover .................................................................................................................................. 94
Table 5.11. Acadia study area Variances of SRTM-LiDAR error by NLCD 2001

class ................................................................................................... 97
Table 5.12. Elevation Characteristics of entire Webster study area .......................... 99
Table 5.13. Slope (degrees) Characteristics of entire Webster study area .................. 99
Table 5.14. Webster study area Sample Elevation Characteristics (n= 1484). . . . . . . . .....100
Table 5.15. Webster study area Sample Slope Characteristics (n= 1484) ................. 101

Table 5.16. Pearson Correlation Matrix for the Webster study area (n= 1484) ........... 101
Table 5.17. Pearson Correlation Matrix for the Webster study area Slope. . . . . . . . . . . ......102
Table 5.18. Error Analysis Results for Webster study area ................................. 103
Table 5.19. Webster study area Sample SRTM Error Characteristics (n= 1484).... . ....103
Table 5.20. Webster study area: One-sided T-test Results .................................. 105

Table 5.21. Webster study area: Sample Error Characteristics by Land Use/Land

Cover ................................................................................................ l 10
Table 5.22. Webster study area Variances of SRTM-LiDAR error by NLCD 2001

class .................................................................................................. 112
Table 5.23. Nang Rong study area DEM Characteristics .................................... 1 14
Table 5.24. Nang Rong Sample Characteristics (n = 1537) ................................. l 14
Table 5.25. Nang Rong study area Slope (degrees) Characteristics ........................ 1 15
Table 5.26. Nang Rong study area Slope Characteristics: Sample (n = 1537). . . . . . ..1 15
Table 5.27. Pearson Correlation Matrix of the Nang Rong study area Elevation

values ................................................................................................ 116
Table 5.28. Pearson Correlation Matrix of Slope Values of the Nang Rong Area. . .....1 19
Table 5.29. Nang Rong study area Sample DEM Error Characteristics .................... 121

Table 5.30. Nang Rong study area: One-sided T-test Results ............................... 122

Table 5.31. Sample Characteristics of MODIS Data ......................................... 124
Table 5.32. Variances of the SRTM Error by MODIS % Tree Cover Groupings

Per study area ....................................................................................... 130
Table 5.33. Suitable Areas from the Land Use Application Output ........................ 135

Table 5.34. Comparison of LiDAR 90 m and NED 90 m Results: Acadia study
area ................................................................................................................................... 139

Table 5.35. Comparison of LiDAR 90 m and SRTM 90 m Results: Acadia study
area ................................................................................................... 139

Table 5.36. Comparison of LiDAR 90 m and NED 90 m Results: Webster study
area ................................................................................................... 141

Table 5.37. Comparison of LiDAR 90 m and SRTM 90 m Results: Webster study
area ................................................................................................... 141

Table 5.38. Nang Rong Suitable Areas from the Land Use Application Output ......... 145

Table 5.39. Comparison of SRTM X-90 m and SRTM X-90 m Results: Nang
Rong ................................................................................................. 148

xi

LIST OF FIGURES

Figure 1.1. Comparisons of Digital Elevation Models ......................................... 4

Figure 2.1. Process of Digital Terrain Modelling ............................................. 11
Figure 2.2. Shuttle Radar Topography Mission Coverage .................................... 15
Figure 2.3. Space Shuttle Endeavor ............................................................... 16
Figure 2.4. Light Detection and Ranging (LiDAR) Data Collection ......................... 31
Figure 2.5. Comparison of Fine and Coarse Resolution Data ................................ 33
Figure 2.6. Procedure for creation of Error Grids .............................................. 36
Figure 2.7 . Representation of a Datum .......................................................... 38
Figure 3.1. Nang Rong, Thailand ................................................................ 41
Figure 3.2. Major stream network of the Nang Rong Study Area ........................... 42
Figure 3.3. Locations of study areas in Louisiana .............................................. 46
Figure 3.4. Availability of SRTM X-band Coverage in study areas ......................... 49
Figure 3.5. Acadia study area: National Land Cover Dataset 2001 ......................... 50
Figure 3.6. Reference Map of the Acadia study area .......................................... 52
Figure 3.7. Major stream network of the Acadia Study Area ................................. 53
Figure 3.8. Webster study area: National Land Cover Dataset 2001 ......................... 55
Figure 3.9. Reference Map of the Webster study area ......................................... 56
Figure 3.10. Major Stream Network of the Webster Study Area ............................ 57
Figure 3.11. Schematic of a Low Location on the landscape ................................. 60
Figure 4.1. Processing Steps for Louisiana Data ............................................... 64
Figure 4.2. Focal Operations ...................................................................... 67
Figure 4.3. Processing Steps for Nang Rong, Thailand study area ........................... 68

xii

Figure 4.4. Louisiana study area: Error Analysis .............................................. 70
Figure 4.5. Criteria for Application .............................................................. 77
Figure 4.6. Cartographic Model for Application ............................................... 78
Figure 5.1. Striping in the southeastern portion of the Acadia study area ................... 82
Figure 5.2. Histogram of SRTM 90 m Slope (degrees) (n= 1956)............. ...............87
Figure 5.3. Histogram Distribution of Elevation Error Grids (n = 1956) ................... 89
Figure 5.4. Acadia study area SRTM-LiDAR Error ............................................ 91
Figure 5.5. Acadia study area SRTM-NED Error ............................................... 93
Figure 5.6. Box plot of the SRTM-LiDAR Error and NLCD 2001 .......................... 95
Figure 5.7 Histogram of the SRTM 90 m Slope (n= 1484) .................................. 100
Figure 5.8. Histograms of the Sample points (n= 1484) ..................................... 104
Figure 5.9. Webster study area SRTM —LiDAR Error ....................................... 106
Figure 5.10. Webster study area: SRTM- NED Error ........................................ 108
Figure 5.11. Box plot of the SRTM-LiDAR Error and NLCD 2001 ....................... 110
Figure 5.12. Elevation Characteristics of the Nang Rong study area ....................... 117
Figure 5.13. Slope (degrees) of the Nang Rong study area .................................. 118
Figure 5.14. Histogram of SRTM C-band Slope Values (n =1537) ........................ 120
Figure 5.15. Histogram of the SRTM Error ................................................... 121
Figure 5.16. SRTM C-band —SRTM X-band Error ........................................... 123
Figure 5.17. MODIS Percent Tree Cover: Louisiana study areas .......................... 125
Figure 5.18. Box Plots of the Acadia and Webster study area: MODIS % Tree

cover ................................................................................................. 126
Figure 5.19. Nang Rong study area: MODIS: Percent Tree Cover ......................... 127
Figure 5.20. Suitable Low Slope Areas: Acadia/Webster study areas ...................... 133

xiii

Figure 5.21.
Figure 5.22.
Figure 5.23.
Figure 5.24.
Figure 5.25.
Figure 5.26.
Figure 5.27 .
Figure 5.28.

Figure 5.29.

Select depictions of the Focal Mean of Acadia study area .................... 134
Final Application Results: Acadia/Webster study area. . . . . . . . . . . . . . . . . .......136
Application Result Classified by Possible Categories ........................ 142
Suitable locations found in all DEMS of Application Analysis .............. 144
Final Application Results: Nang Rong study area .............................. 146
Suitable locations found in all DEMS of Application Analysis .............. 147
Comparison of SRTM X-90 m and SRTM C-9O m ........................... 150
Interaction of Landscape Feature with short wavelength radar .............. 156
Reflectivity of Radar from Tree Canopy ....................................... 157

xiv

LIST OF ACRONYMS

CGIAR-CSI: Consortium for Spatial Information
DEM: Digital Elevation Model

DLR: Deutsches Zentrum fur Lufi- und Raumfahrt (German Aerospace Center)
DTED: Digital Terrain Elevation Data

DTM: Digital Terrain Model

GIS: Geographic Information Systems

GIScience: Geographic Information Science

GLCF: Global Land Cover Facility

GPS: Global Positioning System

GTOPO30: Global 30 Arc-Second Elevation Data set

I PL: Jet Propulsion Laboratory

LiDAR: Light Detection and Ranging

MODIS: Moderate Resolution Imaging Spectroradiometer Sensor
NAD83: North American Datum of 1983

NASA: National Aeronautics and Space Administration
NAVD88: North American Vertical Datum of 1983
NED: National Elevation Dataset

NGA: National Geospatial-Intelligence Agency

RMSE: Root Mean Square Error

SAR: Synthetic Aperture Radar

SIR-C: Spacebome Imaging Radar-C

SRTM: Shuttle Radar Topography Mission

XV

USGS: United States Geological Survey

X-SAR: X-Synthetic Aperture Radar

xvi

1. Introduction

Digital elevation models (DEMS) are digital representations of the earth’s surface
and vital spatial data sources for understanding many environmental, social, and
geographic issues around the world. An understanding of differences of spatial resolution
and production method is important to end users and model selection in application use.
The focus of this thesis is on the differences of the Shuttle Radar Topography Mission
(SRTM) C-band 90 m DEM in relation to higher quality DEMS through case studies in
the United States. This investigation of DEM error Sheds light on its relationship with
land cover characteristics. A land use model is explored to examine the differences of
selected DEMS.

An important part of geographical study is terrain analysis because of its ability to
capture characteristics and spatial patterns of the Earth’s surface. A digital elevation
model (DEM) has been defined as “set of measurements that record the elevation of the
surface of the Earth, such that the Spatial proximity of, and spatial relationships between,
those measurements can be determined either implicitly or explicitly” (Fisher and Tate
2006: 468). The error of the DEM is analyzed by comparison to a more accurate
elevation model, termed the reference model (Fisher and Tate 2006). DEM error refers to
how closely the values of the model correspond to true elevation values in reality (or
what is termed “reference data”). DEM error literature has used this approach to measure
vertical differences at individual locations and to summarize overall error through global
metrics. In the DEM error literature, accuracy refers to how closely the elevations under

investigation match true values.

DEMS included in this research are the National Elevation Dataset (NED) from
the United States Geological Survey (USGS), elevation models derived from Light
Detection and Ranging (LiDAR) from ATLAS GIS in Louisiana (Atlas 2007), and
finished Shuttle Radar Topography Mission (SRTM) from the Earth Resource
Observation and Science (EROS) Data Center of the US. Geological Survey (USGS
2007). These DEMS comprise a range of Spatial resolutions: 30 m, 5 m, and 90 m
respectively. Of particular interest is the near-global SRTM DEM, which could be used
in many international contexts. Elaborating on the error characteristics of the SRTM
using US data would benefit analysis in locations without higher resolution data. Publicly
available LiDAR-derived DEMS from ATLAS, The Louisiana Statewide GIS, were
obtained as reference data to determine the error of the SRTM model.

This research focuses on investigating the relationships between SRTM error and
terrain characteristics to aid users in the proper employment of SRTM data in
environmental applications. This thesis research incorporates international data for Nang
Rong, Thailand and parts of Louisiana (USA) to gain greater insight into the details of
these diverse datasets. The Shuttle Radar Topography Mission (SRTM) 90 m resolution
elevation products have become available for locations +/- 60 degrees N/ S for public use
(Gesch, Muller, and F arr 2006). The SRTM DEM as a seamless radar elevation product
is a standard product across multiple scales of analysis from the local, regional, and
global level. This highlights an advantage of the SRTM: its consistency across national
borders and availability in inhospitable areas, such as steep mountain ranges and dense
tropical forests. Thus, this examination of SRTM error in relation to terrain

characteristics, such as slope, will aid the public, academics, and professionals in

assessing the fitness of SRTM data for specific applications interpreting the results of
SRTM environmental applications that apply it. A greater understanding of the
applications of SRTM is important to geography research because of the global interest in
the earth’s characteristics.

Geography consists of understanding thematic relationships across spatial scales
of analysis. Geography is the study of what is where and why. Thus, this applies to both
environmental characteristics and social demographics. Digital terrain analysis can be
used to understand many dynamics in geography. In relation to digital terrain analysis,
the availability of data also reflects different scales possible with each product. Scale is
defined as the “relative distance on the image to the corresponding distance on the
ground” (Bolstad 2005: 191). In large-scale maps, objects and landscape features appear
larger and with more detail than on small-scale maps (Bolstad 2005). Figure 1.1
illustrates the DEMS used in this thesis with their coverage extents and general resolution
characteristics. Extent refers to the area covered by a particular image (Bolstad 2005).
The quality of data used in geographic applications could impact the results of the
application. This research explores an agricultural suitability model related to rice
production in two different geographic areas of the world to examine the implications of
differences in DEMS in application use. An agricultural application was selected because
of the similar agricultural conditions and flat topography of both Nang Rong and

southwest Louisiana.

 

 

 

 

 

 

 

 

 

 

 

 

 

SRTM Near-Global Coverage F Coarse
C
NED Regional: United States 73 3%
3 '8
CD a:
m

LiDAR Local: Select States ,

and Regions Fme

 

 

 

 

Figure 1.1. Comparisons of Digital Elevation Models
1.1. Motivation

A motivation of this research is to expand understanding of important
characteristics of the SRTM DEM to improve its use in the public, private, and
governmental settings. Environmental and social issues related to environmental change
and earth monitoring faced can be addressed by using SRTM data (Wood 2004).
However, understanding the appropriate use of available digital terrain data is still needed
to have effective implementation of decisions made from its results. Thus, greater
understanding of SRTM gained from applications in data rich areas, such as the United
States, will strengthen the appropriate use of the SRTM model in applications
internationally.

This comparative study of locations in the United States and elsewhere illustrates
how research in an area with very rich spatial data resources, including high resolution
raster data, accurate vector datasets, and effective means of public dissemination, can be
used to aid research without these resources. The Nang Rong district of Thailand was
chosen as one location for this thesis work in conjunction with other research at Michigan

State University. Its flat topography and agricultural characteristics were used to identify

possible study areas in the United States. Based on these characteristics and the
availability of a high resolution reference DEM, two study areas in Louisiana were
selected: an area in southewestem Louisiana including mainly Acadia Parish, and an area
in northern Louisiana including parts of the Webster and Boosier Parishes. The Acadia
study area represents an agricultural area, while the Webster study area represents a
forested area. Thus, these study areas: the Acadia, Webster, and Nang Rong study areas,
enable research into the characteristics of the SRTM C-band 90 m DEM in relation to
land cover/ land use properties.
1.2. Research Questions

This research proposes to answer the following research questions (1 -6):
1. What are the relationships among elevations of NED, LiDAR, and SRTM DEMS?

A comparison of the elevations values of the NED, LiDAR, and SRTM DEMs is
important to complete in order to determine how much the values are similar or different
from one another. Since, the NED, LiDAR, and SRTM DEMS are available for the
Louisiana study areas, it is important to determine how the selection of one of these
DEMs could portray topographic characteristics similar or different from the others. The
elevations of the DEM products are expected to be similar for similar production
methods: LiDAR, NED, and SRTM.

2. What are the relationships among NED, LIDAR, and SRTM slopes?

Similar to the elevation comparison, the slopes of the NED, LiDAR, and SRTM

DEMs will be compared to determine similarities and differences. These similarities and

differences can be important to application use and pertinent to the type of landscape the

DEM is used in. The slopes of the NED, LIDAR, and SRTM DEMS are expected to be
correlated for the original and resampled versions from each production method.
3. Is the average SRTM DEM error greater than zero?

The average error is an indicator of whether the DEM is systematically higher or
lower than the reference DEM. The average error indicates bias of the DEM. It is
expected that the average SRTM DEM error will be different from zero for different land
uses. A SRTM error of zero would indicate no difference between the SRTM DEM and
the reference DEM.

4. What is the relationship of SRTM error in terms of different land cover/land use
classes (i.e. developed, forested, agriculture, water)?

Researchers (Shortridge 2006, Bhang, Schwartz, and Braun 2007, Carabajal and
Harding 2006), have recognized that the SRTM DEM has issues with tree canopy. The
SRTM DEM has been found to overestimate elevation in locations of dense tree canopy
(Shortridge 2006). My thesis research is focused on broad land cover/ land use classes
(i.e. developed, forested, agriculture, water, Shrub) to determine whether these classes
also have certain characteristics with SRTM DEM Error. Distinction is made between
non-forested and forested classes for comparative purposes.

5. Is the average SRTM DEM error greater for vegetated or non-vegetated areas?

AS seen above in the explanation of the land cover/ land use relationships research
question, interest is in the differences between the vegetated and non-vegetated areas.
Non-vegetated here is in reference to areas without tree canopy. The vegetated areas are
expected to have a larger average SRTM DEM error variance and magnitude. The SRTM

error for a vegetated surface is expected to be positive due to an elevation error bias.

6. How will the LiDAR, NED, and SRTM DEMS perform in a land use application?

The application of the SRTM compared to the LiDAR is expected to have a lower
overall accuracy. The application of the SRTM compared to the NED is expected to have
a lower overall accuracy. Since DEMs are used in geographical research and public
setting, it is important to understand how the use of one DEM differs from another. This
research question is aimed at identifying the different results of a DEM when used in an
application setting. This is valuable information to other researchers in terms of their
selection of a specific DEM to employ in their research. It is likely that the given the
coarser resolution 90 m of the SRTM DEM used in the application, it would not perform
the same as the LiDAR and NED DEMS.

The above research questions are aimed at expanding the knowledge of SRTM
elevation models. This examination is focused on the differences between the SRTM
model and other available elevation models including LiDAR-derived DBMS and NED
data. The limitations of the SRTM can be explored through analysis with higher quality
data (i.e. LiDAR and NED) in the United States context and SRTM X-band in the
international context. This research is aimed at assessing these error structures in relation
to landscape characteristics, such as land cover/land use. Given the nature of the actual
slopes of the selected study areas, the slopes of the NED, LIDAR, and SRTM DEMS are
investigated in a comparative sense; however, the relationships between their slope
properties and SRTM error is not a focus of this research. This research is valid both in
the United States and elsewhere where DEMS are employed in government use, research,

and environmental modeling.

1.3. Significance

Diverse landscapes and spatial analysis of large regions is possible with digital
terrain analysis, an important research topic in geography. This research supplements
existing error studies of SRTM data, which used diverse reference data, such as satellite
altimetry, Global Positioning Systems (GPS) point data, and National Elevation Dataset
DEMs (Berry, Garlick, and Smith 2007; Falorni et a1. 2005; Guth 2006). Trends in the
characteristics of a model in relation to slope and landscape characteristics can aid in
understanding its fitness of use. This research acknowledges that multiple data sources
are available to the user from various sources, and explores how variations among these
products are important in application use. Since the SRTM DEM has near- global
coverage, this research is significant to the international research community where the
SRTM model is the best available DEM.
1.4. Overview of Thesis

Chapter 1 provides an introduction to the key questions, and significance. This
chapter introduces the topic of elevation models in terms of various resolutions,
production methods, and error relationships. Images in this thesis are presented in color.

Chapter 2 presents background information on the digital elevation data sources
investigated to address the research questions stated in Chapter 1. Further information is
provided on data quality. A review is provided of elevation model sources employed in
this thesis: National Elevation Dataset (NED), Light Detection and Ranging (LiDAR),
and Shuttle Radar Topography Mission (SRTM) data. The characteristics, production
methods, and literature of the SRTM DEM are more extensively reviewed. A review of

the literature on SRTM error is presented to provide context for this thesis research in

relation to other known relationships between SRTM error and terrain characteristics.
Further discussion is provided on the conceptualization of error and resampling
techniques in error analysis.

Chapter 3 describes the methodology of site selection of study areas. The
characteristics of chosen study areas of Nang Rong, Thailand and portions of Louisiana
are examined in relation to geographical and environmental characteristics. This section
provides a framework for an agricultural application in both areas.

Chapter 4 outlines a detailed methodological narrative on the processing
procedures necessary for this research. The sources of the DBMS and resampling
techniques are outlined. Error analysis methodology and application implementation are
indicated in relation to study areas.

Chapter 5 discusses the results of the analysis of DEMS for the Acadia and
Webster study areas in the United States and Nang Rong in Thailand. Results of SRTM
error by land cover/land use characteristics are summarized. General trends and patterns
of error characteristics are identified per study area. In addition, a geospatial application
related to rice production is presented to illustrate how the use of different DEMS can
present different outcomes.

Chapter 6 summarizes the main conclusions of this research in terms of each case
study and makes inferences to other areas. Directions of research are discussed to indicate

future paths of SRTM error research.

2. Background

The United States’ well-structured system of dissemination and coordination of
geospatial information enables researchers to access data for anywhere in the US. This
research utilizes publicly available elevation data to determine the error characteristics of
DEMS. This section will focus on discussion of the characteristics of elevation models
used in this research, including National Elevation Dataset (NED), Shuttle Radar
Topography Mission (SRTM), and Light Detection and Ranging (LiDAR) DEMS. Issues
related to the SRTM production will be closely reviewed, along with research directed at
understanding the SRTM C-band 90 m DEM’S relationship to error and applications.
Following discussion of the sources of elevation data, key concepts related to DEM
analysis are explained.
2.1. Digital Elevation Background

DEMS are used to address hydrological, ecological, natural resource management,
and land use planning issues (Moore, Grayson, and Ladson 1991). Current uses of digital
elevation models include watershed delineation, automatic delineation of sub-watersheds,
watershed linkages, drainage networks, and overland paths (J enson and Domingue 1998).
More recently, land cover/land use investigations, urban studies, and medical
geographers have incorporated terrain variables into analysis (Wilson and Gallant 2000).
While elevation itself is important, terrain derivative parameters such as slope, aspect,
plan and profile curvature, wetness index, and solar radiation, may be even more valuable
for environmental analysis (CarliSle 2005; Wilson and Gallant 2000). Table 2.1 indicates
descriptions of key terrain derivatives. These terrain derivatives are obtained from DEMs

produced from various sources, including digitized topographic contour maps, scanned

10

 

aerial imagery, manual profiling techniques, as well as radar technology, such as

Synthetic Aperture Radar (SAR) data. The generation of DEMS is useful in

understanding landscape properties, visualizations, and applications (Figure 2.1).

Visualization techniques for DEMs include contour mapping, hypsometric Shading,

hillshading, and creating Triangular Irregular Networks (TIN).

Table 2.1. Key Terrain Parameters.

 

 

 

 

Attribute Description
Slope Gradient, Rise over Run (degree and percent); lSt Derivative
Aspect Direction of steepest slope

Curvature Plan curvature, Profile Curvature. Curvature indicates the

 

convexity and concavity of the surface. 2nd Derivative

 

This table defines main terrain attributes of DEMs. Modified from Moore, Grayson, and

Ladson (1991).

Figure 2.1. Process of Digital Terrain Modeling. This schematic illustrates how once
elevation and coordinate information are acquired, they are used to generate DBMS and

 

 

i
l

 

PE

Applications g

 

 

 

 

 

 

 

 

l

I
F Interpretations Visualization

 

 

it Tl

 

 

 

 

DEM Creation ~—

J

 

 

 

4 I
l

T Data Capture l
(X, Y, 2)
Reality "
J

 

 

 

 

 

 

 

 

 

 

 

 

used in applications. Modified from Hutchinson and Gallant (2005).

ll

 

Digital elevation models are continuous field models that are able to characterize
many natural and anthropogenic features of the earth’s surface, ranging from small
ditches to vast mountain ranges. The interpolation of X, Y, Z point elevation data to a
continuous field model enables the DEM to be incorporated into many GIS applications.
However, the interpolation of elevation models can be done in many ways, which can
result in models of various resolutions with different properties (ESRI 2006b). The cell
size of elevation models can be more or less appropriate for certain details of analysis.
Given the significance of resolution and production method of digital elevation models,
the following section provides a review of elevation model sources.

2.2. Elevation Model Sources

This section summarizes the elevation model sources for both the United States
and internationally used in this research. A review of the characteristics of the NED,
SRTM, and LiDAR data are the focus of this section. These elevation models differ in
terms of production method and resolution, which will be further explored in this thesis.
Recognition of the differences of these models is Si gnificant to understanding their fitness
for use in applications in a variety of settings
2.2.1. National Elevation Dataset (NED)

The United States Geological Survey (USGS) seamless distribution server
provides access to the seamless National Elevation Dataset (NED) elevation models at
1/9th are second (~3 m), 1/3rd arc second (~10 m), and 1 arc second (~30 m) for the
United States (USGS 2007). The NED l arc second data are available for the entire
United States, while the 1/9lh arc second data are created from LiDAR data for select

locations. The USGS elevation products correspond to 7.5 minute quadrangle extents

12

(Moore, Grayson, and Ladson 1991). However, the NED DEM product is not restricted
to the extents of the 7.5 minute quadrangles. Each raster cell of the NED DEM represents
one elevation posting from the source elevation information. Throughout the examination
of the USGS seamless web viewer for this research, consideration was given to its user-
friendliness and ease of accessibility (USGS 2007). An increase in the availability of the
products to the public, modifications made to public repositories to make them more
user-friendly, and the progress of web applications aided broader dissemination of spatial
data. The availability of elevation models and Spatial data have greatly expanded, which
warranted the creation of advanced dissemination methods to the public.
2.2.1.1. NED Production Methods

Advances in geospatial technology have led to the creation of more sophisticated
DEMS. NED data were created from manual profiling methods and more recently from
incorporated higher resolution LiDAR data (USGS 2007). Artifacts from the production
methods of a DEM can be seen by hillshading a DEM. This is valuable information
regarding possible production errors observed in the DEMs, such as striping. Striping is
seen as linear features in the DEM, which are not part of the actual landscape. Striping is
the identification of parallel lines in USGS DEMS from the manual profiling method of
production (Fisher and Tate 2006). The NED metadata indicates the production method
of the DEM. Metadata refers to the information regarding source, production methods,
and originator of these data. Table 2.2 shows the production methods of NED DEMS
corresponding to the value indicated in the metadata. The table indicates the names of
specialized DEM processing software by names such as (ANUDEM). NED represents

elevation models for the United States, while other digital elevation models from radar

l3

technology, such as SRTM, have been created over a larger area of the earth. The SRTM
model will be discussed in subsequent sections to highlight its main characteristics,

relevant research, and application use.

Table 2.2. Production methods of the NED DEMS

 

 

 

 

 

 

 

 

 

 

 

 

 

Production Meaning
Method
0 Unknown
1 Electronic Image Correlation (specifically Gestalt Photo Mapper II
(GPM II))

2 Manual Profiling
3 DLG2DEM: (Digital Line Graph)2DEM
4 DCASS: Digital Cartographic Software System (USGS Sofiware)
5 LT4X: Either LT4X or LTPlus software
6 Complex polynomial interpolation, such as ANUDEM
7 LiDAR or other active remote sensing
8 Photogrammetric mass points and breaklines

 

The USGS uses specialized software to create DEMS. From Oimen (2002).
2.2.2 Shuttle Radar Topography Mission (SRTM)

On February 11, 2000, the Shuttle Radar Topography Mission was flown on the
Space Shuttle Endeavor for 11 days and acquired 12 terabytes of radar data for 60°N to
60°S latitude, which is 80% of the Earth’s land surface (Grohman et a1. 2006; Korbick
2006). Figure 2.2. indicates the SRTM coverage of the world. The SRTM project was a
joint project of the National Geospatial-Intelligence Agency (N GA) and the National
Aeronautics and Space Administration (NASA). The SRTM data were used to create an
elevation model of the earth representative of the top of the earth’s surface (Guth 2006).

The vertical absolute height accuracy of the SRTM DEM is less than 16 m (NASA 2002).

14

'1

strbai

Vlm'fi‘iwufil‘lli' Hum: mun ~; nun-«re u 21;. u '; ' o - '. I33 v es *4: Isaak: r It;"1"‘12:"131"1"¢|l'l"‘

 

Figure 2.2. Shuttle Radar Topography Mission Coverage. From Landserf (2005).

The Space Shuttle Endeavor that acquired the radar data for the creation of SRTM
elevation products entailed sophisticated mechanisms and engineering components.
Figure 2.3. illustrates the Space Shuttle Endeavor, which was flown with a single-pass,
dual antenna, including a C-band interferometric synthetic aperature radar system and X-
band antenna (Slater et al. 2006). The SRTM hardware included a 60 m mast with C-
band and X-band antennas and advanced tracking devices (J PL 2005d). Other products
obtained from the SRTM included ascending and descending orthorectified image mosaic
(AOIM, DOIM), Terrain Height Error Data (THED), Seam Hole Composite Map
(SHCM) (Slater et a1. 2006). The US Geological Survey disseminates the SRTM version
1 (90 m), and SRTM version 2 (30 m) for only the United States (Slater et al. 2006). For
clarification purposes, the SRTM C-band DEM used primarily in this research differs
from the SRTM DTED data produced and released prior to the release of the finished
SRTM data. The SRTM data used in this research was the finished product from the
USGS (EROS) obtained in July of 2007. The completion of the processing of the SRTM

C-band data was 2 years (.1 PL 2005a). The C-band data were processed by the Jet

15

Propulsion Laboratory, and Deutsches Zentrum fur Lufi-und Raumfahrt (DLR), (the
German Aerospace Center) processed the X-band data at 25 m resolution, a finer

resolution compared to the C-band.

 

    

_ ‘ C-band Transmitter/Receiver
. .. antennas

X-band

Receiver antennas

 

 

 

Figure 2.3. Space Shuttle Endeavor. The Space Shuttle Endeavor carried the hardware
and software used to acquire the radar topographic information. Modified from Yastiki,
Kocak, and Buyuksalin (2006).

Consideration of how the SRTM data were created enables the user to better
determine whether a certain use will be appropriate. The SRTM data distributed by the
USGS EROS service used the subsampling method to resample the SRTM 30 m DEM to
the 90 m resolution, while data distributed from the Land Processes Distributed Active

Archive Center (LP DAAC) used an averaging method (NASA 2005). The subsampling

used the center cell of the 90 m area, while the averaging method took the average value

of the neighborhood of cells. This method decreases the noise in the initial radar data. In
addition to understanding the production method of the SRTM DEM, the nomenclature
of the various models from the SRTM mission warrants clarification because several
versions of the SRTM DEM were released during processing.

SRTM DEMS were created using radar interferometry. Radar interferometry
entails obtaining two radar images at different locations and using the differences
between the images to determine the elevation (J PL 2005e). The radar technology is able
to penetrate clouds, which makes it an advantageous imagery collection method (J PL
1999). These two images are taken by one radar antenna in the shuttle payload bay and
another one on the 60m mast (J PL 2005a). A calculation is made for the earth’s surface
elevation based on the differences of these radar images (J PL 2005 a). Possible errors in
the interferometric calculations relate to static errors, which are errors constant for the
entire dataset and time-varying errors, which are from the movement of the mast and
electronic beam steering (Rodriguez, Morris, and Belz 2006).

The C-band and X-band SRTM DEMS have slightly different characteristics,
which indicate slight variation in surface response to landscape properties. The C-band
radar antenna is able to transmit and receive 5.6 cm wavelengths (J PL 2005c). The 225
km swath width of the C-band radar antenna enabled its 80% coverage of the earth,
which is 119.56 M km2 (J PL 2005c; JPL 2005i). The X-band swath width was 50 km
(J PL 2005c). The X-band was able to receive and transmit data at a 3 cm wavelength
(JPL 20050). The different wavelengths of the SRTM DEMS are discussed in subsequent

paragraphs as they relate to the radar’s response to surface characteristics.

17

The wavelength of the radar determines how well it is able to penetrate the
atmosphere (Lillesand, Kiefer, and Chipman 2004). The radar of the SRTM mission did
not penetrate vegetation; rather the radar collects data from the top of the canopy cover
(J PL 2005b). However, if the vegetation is less dense, the radar may be able to detect
ground conditions (J PL 2005b). The small size of the wavelengths of the SRTM C-band
and X-band do not enable deep penetration into vegetation, like the wavelengths of 10 cm
to 30 cm (Lillesand, Kiefer, and Chipman 2004). Typically, radar at shorter wavelengths
from 2 cm to 6 cm, such as the SRTM C-band and X-band, are better for detecting crop
surfaces (Lillesand, Kiefer, and Chipman 2004). It is difficult to decipher differences
between the SRTM C-band and SRTM X-band regarding vegetation canopy. If the
wavelength they transmitted at was more dissimilar, possibly more of a distinction could
be made among vegetated surfaces. Also in comparison, the X-band scatters radar from
the top of tree canopies; while the P-band (another radar band that receives and transmits
at 74 cm) is able to penetrate vegetation (Andersen, Reutebuch, McGaughey 2006).

A study was found that examined the SRTM C-band and SRTM X-band DEM in
an area of Norway in terms of major land cover characteristics (Weydahl et al. 2007).
Weydahl et al. 2007 indicated the difference between the SRTM C-band and SRTM X-
band for agricultural and forest cover in an area of Norway. The RMSE of the SRTM X-
band DEM was 3.4 m for agricultural fields and 9.8 m for dense coniferous forest, where
the reference DEM was one created from digital vector data at a cell spacing of 5 m
(Weydahl et al. 2007). While the SRTM C-band 90 m had a RMSE of 4.6m in
agricultural fields and 9.1 m in deciduous coniferous forest (Weydahl et a1. 2007). A

main finding of Weydahl et al. 2007 was that the SRTM C-band seemed to penetrate

l8

slightly deeper than the SRTM X-band; however, as these researchers noted this slight
difference may be insignificant in actual use of the DEMS. Thus, despite slight
differences in the penetration of the SRTM C-band and SRTM X-band DEMS, they may
still be used effectively by researchers.

Recently the use of X-band SRTM data are appearing in scientific literature
(Hoffinann and Walter 2006, Weydahl et al. 2007, Kiel, Alsdorf, and LeFavour 2006)
The SRTM X-band Data was obtained for 40% of the land covered by the SRTM C-band
(DLR 2008). Table 2.3 indicates the characteristics of the X-band DEM. In comparison
to the C-band data, the absolute vertical accuracy is comparable; however, the relative
vertical accuracy is better. In addition to differences in accuracies, the two products were
processed in reference to slightly different vertical datums, which will be further
discussed in later sections. The SRTM X-band DEM was created by using the elevation
value of a 25 m by 25 m extent (Wagner 2007).

The most complete article investigating the differences between SRTM-X and
SRTM-C band data was Hoffmann and Walter (2006). Hoffrnann and Walter (2006) take
a global scale approach to investigate the root mean square error (RMSE) differences
between data products. In comparison of the SRTM C-band and SRTM X-band to a
gound reference, Hoffinann and Walter (2006) found the SRTM C—band to be more
accurate than the X-band for water areas in the Amazon and Ohio. In relation to studies
that incorporate the X-band data, such as an error assessment in the Norwegian forest
area using X-band and C-band SRTM data (Weydahl et al. 2007). In this comparison, a
difference of 15-17 m was seen between the SRTM elevations (SRTM X-band and

SRTM C-band and the ground value for forested areas (W eydahl et al. 2007). Another

19

study used the SRTM X-band data for analysis of watersheds in the Amazon; however,
the authors did not correct for the differences in geoids between the SRTM X-band and
other DEMS because their research was interested in variability of results not absolute
accuracy (Kiel, Alsdorf, and LeFavour 2006). These examples indicate that where the
SRTM X-band data are available, researchers are beginning to compare its fitness for use
to those of other products.

Table 2.3. NED, SRTM, and LiDAR DEM Characteristics

 

 

 

 

 

NED SRTM C- SRTM C- SRTM X— LiDAR 5 m
band 90 m band 30 m band
Horizontal NAD83 WGS 84 WGS 84 WGS 84 NAD83
Datum
Vertical NAD83 EGM96 EGM96 WGS 84 NAD83
Datum
RMSE 7m 10m 10m 6m 0.15-0.5m

 

 

 

 

 

 

 

From Wagner (2007), USGS (1999), NASA (2002), Atlas (2007).
2.2.2.1. SRTM: Production issues

Factors related to the SRTM hardware and operations have been explored in the
literature in relation to mechanical operations, data voids, and systematic sources of error
in the SRTM model (Rodriguez, Morris, and Belz 2006). Some possible contributors to
the SRTM error include baseline roll errors, phase errors, beam differential errors, timing,
and position errors that occurred in the initial retrieval of radar information (Rodriguez,
Morris, and Belz 2006). Also, errors from the extraction methods, data spacing, and
smoothing techniques are present from the model’s creation methodology (Grohman et
al. 2006). The SRTM elevation data distributed by the USGS had a half pixel shift in
comparison to the NED (Guth 2006). However, as of February 6, 2007, the USGS

corrected this half pixel shift (USGS 2007).

20

The SRTM product released contains voids in the data from its collection
methodology. Voids (i.e. holes in the data) have been investigated by researchers
(Grohman et a1. 2006). Methods have been investigated to fill the voids, such as a Fill
and Feather method and a Delta Surface method (Grohman, Kroenung, and Strebeck
2006). In the Fill and Feather method, the most accurate elevation model available is
used to fill the void value and is “feathered “into the SRTM to obtain a smooth transition
in values (Grohman, Kroenung, and Strebeck 2006). The Delta Surface method uses
SRTM values near the void to fill the void (Grohman, Kroenung, and Strebeck 2006).
The SRTM data the USGS distributes did not implement a filling algorithm. However,
the Consultative Group for International Agricultural Research Center (CGIAR-CSI) has
implemented a void-filling algorithm and use of other DEMS to fill in the voids in the
data (CGIAR-CSI 2004). Other SRTM sources, such as the Global Land Cover Facility,
have also enhanced their DEMs by filling the voids in the DEM (GLCF 2008). The
process of filling the voids alters the original data; thus, this research is focused on using
data that was not altered from its finished form.

Recognized systematic errors of SRTM DEMs include striping and mis-alignment
with local datums (Miliaresis and Paraschou 2005). Striping is seen as linear patterns in
the error characteristics of the DEM, which are unrelated to the Earth’s surface
characteristics. Striping has been observed in the SRTM DEM from the swath-like
manner of the radar collection. In an investigation of SRTM on the island of Crete, a
misalignment of the SRTM DEM was observed that was suggested to be from an
inability to align the SRTM DEM with the local datum in this location (Miliaresis and

Paraschou 2005). Datum issues of SRTM will be further discussed later in the chapter.

21

In comparison of SRTM to alternative elevation products, the radar technology
has potential advantages. SRTM provides consistent source information across a wide
land area, better resolution compared to the GTOPO30 (1 km), and better capabilities to
analysis surface characteristics. At a global scale before the SRTM mission, the finest
global elevation model, (GTOPO30), was available at 1 km (30 arc seconds) resolution
(Korbick 2006). The blunders or stripings from manual profiling present in the NED data
are not present in the SRTM. Thus, in areas where NED striping or production issues are
prevalent the SRTM C-band 30 m DEM could be a useful alternative. Geographic
analysis in locations where very few other spatial datasets are available can benefit from
the use of the SRTM DEM. Hence, SRTM data have great implications in international
land management. Furthermore, locations with poor reference data can benefit from this
thesis research.

2.3. SRTM Elevation Model Error Literature

Studies of error in digital elevation models are well-documented in the scientific
literature (Carlisle 2005; Fisher and Tate 2006). Rugged areas have less vertical accuracy
because of smoothing effects of elevation models (Grohman, Kroenung, and Strebeck
2006). The type of landscape (i.e. flat or rugged) can have different error characteristics.
Hydrological applications have been explored in terms of the elevation differences found
between USGS 30 m and ground truth points (Holmes, Chadwick, and Kyriakidis 2000).
A standard methodology of exploring DEM quality has been by comparing them to a
high quality reference model, which will be the methodology employed in this

research.(Guth 2006; Shortridge 2006).

22

More recently, studies began investigating the characteristics of SRTM DEM
error (Rodriquez, Morris, and Belz 2006; Guth 2006). A global assessment of the SRTM
model was performed using ground control points from kinematic global positioning
systems (Rodriquez, Morris, and Belz 2006). The use of kinematic GPS was limited to
roads only and lacked inhospitable locations, such as rough terrain, steep cliffs, and high
ridges (Rodriquez, Morris, and Belz 2006). Another disadvantage of the global
assessment using kinematic GPS was the omission of areas of high difference between
GPS ground truth points and SRTM heights, including bridges and heavily forested areas
(Rodriguez, Morris, and Belz 2006). This indicates how selective areas for error analysis
may not be fully representative of landscape characteristics. Thus, this demonstrates that
a different continuous model could be valuable as a reference model in error analysis.

In the United States, a large scale analysis of SRTM and NED investigated
geomorphic parameters at over 500,000 sample areas. Guth (2006) found correlations to
be high between elevation and relief, but low for curvature and skewness (Guth 2006). In
flat landscapes, for example, on alluvial fans, substantial noise has been found in SRTM
data, which increases with average slope (Guth 2006). Noise is the term given to
disturbances in digital imagery from issues related to the remote sensing technology
(Lillesand, Kiefer, and Chipman 2004). DEM errors are related to the physical
characteristics of the earth (Carlisle 2005). This illustrates how areas with different
topographic characteristics can have different representation issues with a DEM. Further
application analysis is needed with these data to determine if results depend on the

selected data, in terms of hydrology, slope processes, and viewsheds (Guth 2006).

23

In contrast to using massive datasets for reference in locations, smaller scale
studies on select areas used GPS data as reference. For the Little Washita River in
Oklahoma and the Tolt River in Washington, ground control points and USGS DEMs (30
m) were examined in relation to the SRTM C-band 30 m data (F alomi et al. 2005). An
analysis of SRTM and differentially corrected GPS elevation points for a Site in the
Catskill Mountains of New York and a site in Puket, Thailand, indicated correlations
between SRTM error and slopes greater than 10 degrees, overestimated elevations on
southeast slopes, and underestimated elevations on northwest slopes (Gorokhovich and
Voustianiouk 2006). Reference GPS data provide a limited sample of points for
comparison to the SRTM model, which may not be representative of an entire region.

DEMS generated from digitized contour lines have been used in locations without
GPS data. An alternative approach used to create DEMs is interpolation from contour
lines; however, hillshades of these models can have a terraced appearance. From an
analysis of the SRTM DTED level 1 for the island of Crete, higher errors were found on
mountain tops when using a reference elevation model generated from digitized contour
lines (Miliaresis and Paraschou 2005). The use of a DEM created from digitized contour
lines was also used in comparison of catchments in Australia (Hancock et al. 2006). In an
examination of SRTM and high-resolution models (25 m) for catchments of in the Tin
Creek, Swift Creek, and J emmys Creek of Australia, the SRTM DEM captured slope and
elevation characteristics; however, SRTM derivatives of catchment area, relief, and shape
were not as accurate with the SRTM DEM (Hancock et al. 2006). The lack of a
correspondence between the Australian catchments could have been from the method of

the creation of the reference DEM.

24

SRTM has been considered in relation to reference data sources, such as satellite
radar altimetry (Berry, Garlick, and Smith 2007), Ice, Cloud, and Land elevation Satellite
(ICESAT) data (Carabajal and Harding 2006). Berry, Garlick, and Smith (2007)
employed a continental analysis of SRTM C-band data using satellite radar altimetry.
ICESAT was used in a more select manner in representative landscapes to analyze in
relation to SRTM (Carabajal and Harding 2006). The C-band of the SRTM 30 m digital
elevation model was compared to data from the Ice, Cloud, and land-elevation satellite
(ICESat) laser altimetry for Otter Tail County, Minnesota (Bhang, Schwartz, and Braun
2007). NASA’S laser vegetation imaging sensor (LVIS) was used as a reference to SRTM
for sites in Maine, Massachusetts, Maryland, New Hampshire, and Costa Rica (Hofton et
al. 2006). These studies show the use of satellite sensor information, such as ICESat,
could be employed as a possible source of reference information to the SRTM model.

Overall, the literature published on SRTM error characteristics ranges in type of
reference data. Table 2.4 provides a summary of select literature reviewed in this section.
The SRTM DEM is also indicated since different sources enhance the DEM as indicated
earlier in the section. The reference data were often dependent on the scale of analysis
ranging from small watersheds to global scale analysis desirable by the researchers.

Table 2.4. Summary of SRTM Accuracy Literature

 

 

 

Reference Reference Data Study Areas SRTM Model

Guth (2006) USGS NED 500,000 sample areas SRTM 90 m C-
(USA) Band

Berry, Garlick, Satellite Radar By Continent SRTM 90 m C-

& Smith (2007) Altimetry band

 

 

Carabajal and Ice, Cloud, and Amazon, Tibetan Plateau- SRTM 90 m C-
Harding (2006) Land Elevation Himalayan Mountains, band

Satellite (ICESAT) East Afiica, western
data Australia, western USA

 

 

 

 

25

 

Table 2.4. continued.

 

 

 

 

 

 

 

 

 

 

 

Gorokhovich & Differential GPS Catskill Mountains, New CGIAR-CSI
Voustianiouk points York SRTM 90 m
(2006) Phuket, Thailand
Hancock et al. High resolution Catchments of Tin Camp SRTM 90 m
(2006) models Creek, Swift Creek,
J emmys Creek (Australia)

Miliaresis & SRTM DTED level Crete Island SRTM 90 m
Paraschou 1/ digitized contour
(2005) lines
Rodriquez, Kinematic global Global assessment SRTM 90 m
Morris & Belz positioning system
(2006)
Shortridge NED/ GPS Ground Oakland County Michigan SRTM 90 m
(2006) truth
Hofton et al. NASA’S laser Maine, Massachusetts, SRTM 90 m C-
(2006) vegetation imaging Maryland, New band

sensor (LVIS) Hampshire, Costa Rica
F alomi et al. Ground Control Little Washita River SRTM 30 m
(2005) Points, USGS (Oklahoma), Tolt River

DEMS (Washington)

 

This summary of select literature of SRTM Accuracy indicates the diversity of the range
of reference data used and the range of study sites possible for investigation.

2.4. Implications of Land Cover in SRTM

Researchers have investigated the relationship between SRTM elevation and
different land cover classes, with special focus on differences between forest and open
land. Land cover is defined as natural cover on the surface of the earth, while land use is
related to how land is used, such as for agriculture or urban development (Lambin et al.
2001) In a study of STRM in relation to satellite altimeter echoes from the ERS-l
Geodetic Mission, Berry, Garlick, and Smith (2007) found skewed results between the
land cover of the Amazon Basin for open land and forest.) They also observed greater
differences above the forest canopy compared to open land (Berry, Garlick, and Smith
2007). Similarly, for North America, the SRTM error histogram distribution was skewed

in relation to inland water and vegetation (Berry, Garlick, and Smith 2007). The SRTM

26

 

model was created by NGA and NASA was based on reflections from the earth’s surface,
thus vegetation canopy present on the surface would affect wavelength transmission
(Lillesand, Kiefer, and Chipman 2004).

Similar studies examined the different characteristics of surfaces in relation to
SRTM DEM characteristics (Bhang, Schwartz, and Braun 2007; Hofton et al. 2006). An
examination of the SRTM DEM using ICESat laser altimetry as reference found bare
ground from a classified Landsat-7 image did not impact the SRTM values, while the
forested and agricultural areas impact SRTM and ICESAT in different ways (Bhang,
Schwartz, and Braun 2007). The irregular surface created by forests was found to impact
the radar used in the creation of the SRTM data for a study area in Minnesota (Bhang,
Schwartz, and Braun 2007). Also, an accuracy assessment of SRTM C-band DEMS using
NASA’S laser vegetation imaging sensor (LVIS) as reference found a positive
relationship exists between the elevational difference and height of the canopy (Hofton et
al. 2006). Similarly, in the Amazon, the dense canopy increased the height of the DEM
and obscured the identification of fine scale topographic patterns on the landscape
(Valeriano et a1. 2006).

In an analysis using Michigan land cover data, an error assessment of SRTM with
NED and ground truth data from Oakland County Michigan indicated that the SRTM
DEM were overestimated in forested landscapes (Shortridge 2006). The research
literature indicated the possible biases of vegetation and SRTM, which would be
important to the applications related to forest cover. An application related to tree
inventory was created with the inclusion of SRTM and NED data to aid in the tree

inventory of loblolly pine (Pinus taeda) (Heo et al. 2006). Some researchers have

27

explored how the differences of SRTM and vegetation can be accounted for in offset
calculations (Kellndorfer et al. 2004). Additional analysis of canopy height and SRTM
elevation, indicated a minimum mapping unit of approx 1.8 ha (18,000 m2) to be used in
vegetation canopy height mapping from SRTM data for Georgia and California in
comparison to locations in Iowa and North Dakota identified as tree-less areas
(Kellndorfer et al. 2004). Thus, the reference data employed and the region under
analysis can find different SRTM error characteristics.

A worldwide assessment of the SRTM C-Band data of various topographic and
vegetative sites in the South American Amazon Basin, Asian Tibetan Plateau-Himalayan
Mountains, East Afiica, western Australia, and the western United States used the
Geoscience Laser Altimeter System (GLAS) instrument data (Carabajal and Harding
2006). This study found similar results for elevation values between the products
(Carabaj a1 and Harding 2006). In addition to comparisons of elevation values, Carabaj al
and Harding (2006) examined the SRTM error based on MODIS Vegetation Continuous
Fields (V CF) 500 m resolution data. This study confirmed earlier reports that found
SRTM elevation values to be less accurate for areas of higher tree cover.

As indicated in Carabajal and Harding (2006), a global land cover dataset was
utilized for comparison across study areas. Often as seen from these studies, satellite
imagery such as MODIS data is used to examine broad land cover classes to extrapolate
relationships of SRTM error and surface characteristics. A brief review follows of the
application domains in which SRTM elevation data have been employed to address

environmental issues.

28

2.5. SRTM Application Use

SRTM data have been employed in applications, such as hydrology in the
Amazon (Kiel, Alsdorf, and LeFavour 2006), forest mapping in the Everglades (Sirnard
et al. 2006), and lake level reconstruction in Africa (Leblanc et al. 2006). Environmental
studies and land management applications have included SRTM DEM information with
the desire to include a reliable topographic variable, where possibly not another DEM
exists.

Characteristics of geomorphic elements of the landscape have been identified
through the use of SRTM DEMS. SRTM elevation data were used to examine mega
dunes distribution to better understand the aeolian processes of their formation
(Blumberg 2006). Through the selection of multiple locations in northern Afiica and
across central Asia, the SRTM C-band was compared to SRTM X-band dune locations
(Blumberg 2006). This analysis of dune locations and form illustrated how the SRTM X-
band was able to identify small scale geomorphic characteristics, unlike work with the
SRTM C-band (Blumberg 2006). SRTM DEMs have also been employed to examine
valleys and ridges around the La Pacana Caldera, northern Chile (Bailey et a1. 2007) and
the Pocos de Caldas Alkaline Massif, a volcanic caldera located in southeastern Brazil
(Grohmann, Riccomini, and Alves 2007). The SRTM DEM has been previously used in
the study of landforms and processes related to landscape change.

Other hydrologic applications have explored the potential of SRTM. The extent of
SRT M application use has included hydrological analysis of the Upper Parana River
basin of Bolivia (Brandt and Townsend 2006), watershed analysis of the Asu watershed

in Brazilian Amazonia using SRTM 90 m (Valeriano et a1. 2006), and flood risk

29

identification in Izmir (11, 810 kmz), western Turkey (Demirkesen, Evrendisek, and
Berberglu 2006). The surface water elevations acquired fiom SRTM were used in a
regional groundwater flow model in a study area within the Northern Highland Lakes
region of Wisconsin, USA (Fredrick et a1. 2007). Hydrologic analysis of the Three
Gorges Reservoir, China (Wang et al. 2007) and the Minjiang drainage basin (Zhang et
al. 2006) have utilized the SRTM DEM. Most applications have found the SRTM 90 m
data to be suitable for analysis. However, Hancock et al. (2006) found the SRTM 90 m
DEM, in comparison to a DEM derived from photographic images, portray catchment
properties, network patterns, and runoff properties adequately. An additional caution was
presented in analysis of watershed hydrology of the Elbe Basin and its sub-basins that
spans the Czech Republic and Germany, in terms of the SRTM being less appropriate for
analysis of small-scale topographic features (Haase and F rotscher 2005).

The literature related to SRTM applications demonstrates how studies related to
aeolian processes and hydrological analyses have incorporated the SRTM DEM into their
models (Blumberg 2006). Also, as indicated in the above section on the relation of SRTM
to land cover characteristics, studies investigating tree cover have utilized the SRTM
DEM. Hence, studies around the world have been utilizing the SRTM DEM as an
available source of topographic information in a range of different landscapes.

2.6. LiDAR: Light Detection and Ranging

More recently Light Detection and Ranging (LiDAR) data have been utilized as a
source of high resolution topographic data for flood mapping and hydrologic analysis.
LiDAR represent the best publicly available high resolution data for certain sections of

the United States and is the reference data used in this research. LiDAR collection

30

involves a plane flying over an area with a GPS and laser rangefinder technology on
board collecting signals, which are used to determine the elevation for locations based on
coordinate information (Figure 2.4). LiDAR is collected by directing laser pulses at the
earth’s surface and using the time of the pulse return to calculate the distance (Lillesand,
Kiefer, and Chipman 2004). The LiDAR methodology collects point data for the surface
of the earth at a fine scale difference in spacing of points. An airborne scanning LiDAR
system consists of airborne GPS, an inertial measurement unit, a pulsing laser, a clock,

and onboard computer support (Lillesand, Kiefer, and Chipman 2004).

 

 

 

 

 

Figure 2.4. Light Detection and Rang'ng (LiDAR) Data Collection. LiDAR data are
acquired through the use of GPS ranging technology onboard a plane flying overhead.
Modified from BGMAPS 2008.

A high resolution 5 m DEM can be created as seen in Figure 2.5. with detail not
see at a the coarser resolution of 90 m. An advantage of LiDAR data is its ability to

collect data on steep topography and inaccessible areas (Lillesand, Kiefer, and Chipman

31

2004). LiDAR is also able to collect information on surface features at multiple levels
above the surface from multiple returns per laser pulse (Lillesand, Kiefer, and Chipman
2004). This distinguishes the first return LiDAR pulses fi'om feature such as tree canopies
and buildings, from processing done to obtain bare-earth models (Lillesand, Kiefer, and
Chipman 2004). Even though LiDAR can be used to create high-resolution terrain model,
some of its limitations also need to be acknowledged, such as canopy and collection
issues (Fowler 2007). I will briefly address some issues related to LiDAR, which are
acknowledged in the literature; however, the limitations of LiDAR are not a focus of this
research.

In previous research, LiDAR has been examined in relation to its resolution
implications and production issues implications. Climatic conditions such as rain, mist,
fog, or smoke can hinder LiDAR collection (Fowler 2007). Thus, the timing of LiDAR
collection is important to consider in data collection. In addition to data collection issues,
the resolution of the LiDAR has been investigated. In an investigation of LiDAR derived
DEMs (6 m) and 30 m DEMS for the Fishing River Watershed (1,642 kmz) and Neuse
River Watershed (1,440 kmz) in North Carolina, high relief areas were found to be
Similar for hydrologic analysis including watershed delineation, in comparison to low
relief areas (Barber and Shortridge 2005). Barber and Shortridge (2005) implemented a
resampling methodology to compare the LiDAR-derived data of a small resolution to
other datasets at a coarser resolution. Thus, it is necessary for datasets of different
resolutions to be aligned through resampling methodology to align with the resolution of

other datasets in error analysis.

32

In terms of the LiDAR-derived data I used for this research from the state of
Louisiana, it had already been processed and edited from its raw data version (Atlas
2007). The distributor performed the editing and created the data in raster format. As part
of the editing of the LiDAR data before public release, vegetation and obstructions in the
data were removed from the dataset (Cunningham, Gisclair, and Craig 2004). Thus,
features such as tree canopies and buildings were removed from the dataset prior to its
release to the public in raster format. This created a more reliable source of reference
elevation data for me use in comparison to the SRTM DEM because it did not contain
these surface features.

The state of Louisiana has been acquiring LiDAR data in six phases from 2000
(Cunningham, Gisclair, and Craig 2004). The Louisiana LiDAR project is funded by
FEMA and the state of Louisiana (Cunningham, Gisclair, and Craig 2004). This LiDAR
project in Louisiana was motivated from high flood loss rated from FEMA, the National
Flood Insurance Program (Cunningham, Gisclair, and Craig 2004). The company, 3001 ,
Inc, was in charge of acquiring the LiDAR data, and used the Leica Geosystems ALS40

LiDAR mapping system (Cunningham, Gisclair, and Craig 2004).

 

5 m Resolution 90 m Resolution

      

‘L .- -

 

 

 

Figure 2.5. Comparison of Fine and Coarse Resolution Data. The 5 m DEM identifies
clear features, not present in the 90 m SRTM. Source NASA (2002); Atlas( 2007).

33

Table 2.5. Summary of Publicly Available LiDAR-related online repositories

 

 

Source Description
Kansas Geospatial Community LiDAR data are available for the Kansas River Corridor. The elevation
Commons (2007) datasets available are GPS-Base (ASCII text), LiDAR 2006- All returns

(LAS), LiDAR 2006-Bare Earth (ASCII text), LiDAR 2006: DEM
(DEM), LiDAR 2006-Intensity (GeoTIF, TIF).

 

Interactive Numeric & Spatial
Information Data Engine
(2007)

LiDAR Derived DEMS. The user is able to browse by category:
elevation and derived products. This site contains bare earth digital
elevation models created from LiDAR data. There is well-documented
metadata, which includes source information, accuracy, etc. These files
were created by Horizon’s Inc. and all canopy and structures were
removed.

 

Atlas: The Louisiana Statewide
GIS (2007)

This site enables the user to identify areas by USGSJD quads. A DEM,
contours, edited points, and raw points can be downloaded in feet. The
points are 5 m spacing.

 

 

 

Puget Sound LIDAR This site contains LiDAR for parts of Washington State. There is a

Consortium (2007) viewer which enables the user to select designated areas to download.
Data are available as LiDAR ASCII points (x,y,z) and DEMs as Bare
Earth and Top Surface. This data were surveyed by TerraPoint in feet.

NASA (2006) This is an index directory to files in tar gunzip format for Darrington,
Mt. St. Helens, Rainer, and San Andreas. There is a parent directory,
NASA TerraPoint product format information, and a description of
TerraPoint systems. Information is also provided on properly citing the
data, along with figures of the study areas.

North Carolina: Map Viewer enables the user to download zip files of x,y,z ASCII files.

NCFloodsmap.com (2007) The coordinate system used was NC Stateplane, NAD 83 feet, and

elevation values are in feet.

 

 

Red River Basin Decision
Information Network. North
Dakota State University.
(2007)

 

LiDAR data from the Army Corp of Engineers for the Red River Basin.
Spot elevations every 2 feet.

 

This summary of LiDAR data sources indicate their range of formats and select locations
of availability. Compiled in April 2007 by author.

A survey of online repositories for LiDAR data was undertaken by the author in

the spring of 2007. This investigation identified key Sites to obtain from public LiDAR

data. Table 2.5 summarizes state-specific sources of LiDAR data publicly available.

Online repositories for elevation models are summarized in (Sanders 2007). United

States coastal areas have flown LiDAR for coastal mapping and hydrology projects.

However, these projects could also be useful for the creation of elevation models for

reference to other elevation models, such as SRTM. The USGS Center for LiDAR

Information Coordination and Knowledge (CLICK) provides users with access to LiDAR

34

 

data collected from various locations in the United States in LAS Binary format and
American Standard Code for Information Interchange (ASCII) format. The LAS binary
format is the standardized binary format for distribution of LiDAR data. Other LiDAR
sources include NOAA LiDAR Data Retrieval tool (N OAA 2007), which contains
limited amounts of LiDAR for coastal areas. Select states in the United States have had
LiDAR flown and made available to the public for free. These include North Carolina,
parts of Washington, parts of Louisiana, and a few other areas of certain states as
indicated in Table 2.5. Additional information will follow in the methods section related
to the LiDAR data sources used in this investigation.
2.7. Error Analysis
2.7.1. Conceptualization

The reviewed literature illustrates how DEMs can be compared to other DEMs
through a differencing process (Shortridge 2006). Figure 2.6. illustrates a sample raster
subtraction that occurs between DEMS for error analysis. The difference between the
reference and the SRTM is termed error. Positive error indicates that DEMl is higher
than the reference DEM2, while negative errors indicate the opposite. Different
procedures can be used in the creation of the DEM cell values, such as through points or
averages. However, users need to recognize what is represented by the numerical
elevation value in a raster cell. In order to determine how the real world is conceptualized
in the digital sense, the user needs to have an understanding of projections, scale, and
resolution of the data sources to understand fully the meaning of the DEM. Thus,
acknowledging the meaning of the 90 m resolution is important to this research. The

methodology of the radar-derived SRTM product included averaging to obtain the

35

elevation values of the model. Hence, a resampling method that provides an average
value from the input would be appropriate for the comparison of elevation models at the
same cell size. A discussion of resampling methods and datums follow to highlight how
understanding the meaning of a given elevation value of a DEM is a Significant part of

DEM error analysis.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

5 7 7 7 7 7 -10 o -2
6 6 7 7 7 9 0 -1-2-2
6 6 7 8 7 7 9 9 -1-1-2-1

 

 

 

DEM 1 - DEM2 DEM Error

 

 

 

Figure 2.6. Procedure for creation of Error Grids. The subtraction of the reference DEM
(DEM 2) elevation values from the DEM of interest (DEM 1) results in the DEM error.

2.7.2. Resampling Techniques

High resolution data can be resampled (i.e. 5 m resolution to 90 m resolution)
various ways. Table 2.6 represents three various methods of resampling considered in this
research. The center cell of a given extent of a larger resolution area can be sampled
using the Nearest Neighbor technique (ESRI 2006b). In the Nearest Neighbor method,
the output grid is assigned the center cell value of the input grid. Thus, in this research,
the nearest neighbor method is used to take the center cell of the reference DEM (i.e.
LiDAR 5 m, NED 30 m, and SRTM X-band 25 m) for a 90 m by 90 m area, to assign to a
90 111 cell in the output grid. In addition, a weighted average can be taken, which puts
more emphasis on the center cells of the 90 m x 90 m resolution area of the higher quality

data. A method that includes a larger window to resample from will create a smoother

36

output. Thus, the resampling method of the higher quality data is done to determine the
error characteristics of the lower resolution data.

Table 2.6. Differences in Resampling Techniques

 

Technique Description
Nearest Neighbor Center of the input grid cell is computed for the nearest
neighbor in the output cell

 

 

 

 

 

 

 

Bilinear Uses four nearest cells, which provides a weighted average in
Interpolation the output.
Cubic Employs sixteen neighboring cells to provide a smoothed
Convolution output.

Modified from ESRI 2006b.

2.8. Vertical Datums in Terrain Models

Vertical datums are reference surfaces based on the shape of the earth for
calculations of elevation values. The reference base height is different with different
datums, as seen from Figure 2.7. Figure 2.7 depicts the centered WGS84 datum in
comparison to a local datum, which employs different earth models. Understanding the
differences of vertical datums is critical because of the comparison of data from multiple
sources. The same vertical datum should be used when combining datasets, Since the
datum represents the baseline of the information. The baseline is the level at which
elevation is determined from. Datum transformation was given less attention in the DEM
comparison literature. The vertical datum provides a base from which elevation is
referenced, while the horizontal datum has significance in terms of positional location. As
seen in Table 2.3 of the characteristics of elevation datasets, different datums are used in
the reviewed DEMs. Recognition of the differences in datums was found in few articles
(Walker et al. 2007), which warranted a greater investigation and is discussed in the
methods section. Walker et al. (2007) indicated that the elevation heights needed to be

converted to represent heights above the WGS84 (EGM96) geoid. WGS84 refers to the

37

World Geodetic Datum of 1984. EGM96 stands for Earth Geopotential Model 1996.
Datum transformations can be corrected by using the height of the ellipsoid and the geoid

undulation (Bhang, Schwartz, and Braun 2007). The issues of datum transformations are

further discussed in the Chapter 4.

 

——Earth Centered Datum
................ Local Datum
—Earth’s Surface

 

 

 

 

 

 

 

 

 

 

Figure 2.7. Representation of a Datum. The schematic Shows the earth centered datum
with a different central point compared to the local datum. Modified from Hurvitz (2004).

38

3. Study Areas

This section provides a background to the study areas of this research and
describes the methodology for site selection. International and domestic sites were chosen
so that the US study could represent a rich data location where a higher resolution (5 m)
DEM could be obtained, and an international location represents an area where only a
coarser DEM (90 m, 25 m) is available. This comparative work in important for a number
of reasons. First, the Nang Rong study area provides an example of an area, which has
the SRTM C-band 90 m and SRTM X-band 25 m available. There is not another publicly
available DEM with a finer resolution available for this location. Thus, the locations in
the United States where a higher resolution DEM is available can be used to aid the
interpretation of the Nang Rong elevation DEM characteristics. The areas in the United
States represent two main area with contrasting land cover/ land use characteristics (i.e.
one mainly agriculture, one mainly forested). The comparison of the United States areas
is important to see the different results between these land cover/land use characteristics.
A description of environmental characteristics of selected areas is included related to
topographic, environmental, and land cover/land use characteristics. This background
provides context for the application developed in this study to compare DEM
performance.
3.1. Methodology for Site Selection

A region in northeast Thailand was chosen due to complementary research
activities at Michigan State University and the availability of relevant data, such as
political boundary extents. Nang Rong a district in Thailand was selected. District data

related to its political boundary, hydrology, and classified satellite imagery were acquired

39

from the University of North Carolina, Carolina Population Center (University of North
Carolina 2007). The University of North Carolina had compiled a vast spatial dataset
related to Nang Rong, including both geographic and socioeconomic variables. The Nang
Rong, Thailand site, provides an international location where SRTM data can be easily
obtained from the USGS. Following the examination of the topographic and land
cover/land use characteristics of Nang Rong, Thailand, comparable locations in the
United States were chosen based on similar agriculture and topography characteristics.
Key environmental characteristics pertinent to the selection of study areas in the United
States are indicated in the next section. Following discussion of these key characteristics,
the methodology for site selection in the United States is elaborated on prior to describing
these areas.
3.2. Nang Rong, Thailand

The Nang Rong region is located in the Khorat Northeast Plateau region of
northeast Thailand, one of Six physiographic divisions of Thailand. Figure 3.1 indicates
the Nang Rong region in the Buriram province. The major stream networks are illustrated
in Figure 3.2.The Nang Rong study area extends from 102° 29’ 13” E, 14°30’45”N in
the southwest to 102°54’19”E, 14°49’23” N. Rice production is observed in flat areas
and valleys from a 2003 LandSat image from the University of North Carolina
Population Center (2007). The percent of area in capable of rice production was 55.0 %

based on reclassification of the imagery.

40

 

 

Source: Nang Rang Boundary. UNC Population Carter
ESRI Duh

Projection: UTH Zone 48M, WGS“ Denim

 

 

 

Figure 3.1. Nang Rong, Thailand Elevation and key features

41

 

 

Ma;orstreams

Source: Nang Rong Boundary. UNC Population Center
ESRI Duh

' 0 35 7 14 :1
A ‘rvmwlers

Projection: UTM Zone 48N, WGSS4 Datum

 

 

 

Figure 3.2. Major stream network of the Nang Rong Study Area

Research on the people-environment characteristics of the Nang Rong District
have been done by incorporating demographic information and geospatial information
(Entwisle et al. 2005). The Nang Rong District is relatively poor with marginal
agriculture compared to the rest of Thailand (Entwisle et al. 2005). In the Nang Rong
District, crops are grown both for subsistence and export. Cash crops grown on uplands
include cassava, corn, and sugar cane, all of which are not consumed locally (Entwisle et
al. 2005). Following WWII, the Nang Rong District went through land use/ land cover
changes, which altered the patterns on the landscape (Entwisle et al. 2005). The terraces
surrounding the alluvial plains were used to grow rice (Entwisle et al. 2005). In addition,

cash crops were grown on uplands (Entwisle et al. 2005). Paddy rice located at lower

42

elevations is primarily rainfed, while the higher elevations are covered by forest and
upland crops, such as cassava (Entwisle et al 2005). The Nang Rong District has poor soil
quality, which makes it suitable for only certain crops (Faust et al. 1999).

The climate is generally hot with a wet and dry season (Faust et al. 1999). These
wet and dry seasons correspond to monsoonal season and non-monsoonal season. The
monsoonal season provides moisture for agriculture. The typical farming activities in the
Northeast area of Thailand include preparation of the field from May to December during
the rainy season (Phongphit and Hewison 2001). The rainy season usually begins in May
until August and harvests begin in December (Phongphit and Hewison 2001). In
February, during the time the radar used for the SRTM was acquired, it is expected that
the rice fields would be fallow given the planting from May to July and harvesting season
from November to December (IRRI 2007).

Agriculture is a major component of Thailand culture, environment, and
lifestyles, so a highly relevant SRTM application issue is related to agricultural uses. As
seen from Thai history, connections to the land have played a pivotal role in the
development of villages and cultural traditions (Phongphit and Hewison 2001). Other
researchers have incorporated social survey data, and satellite imagery to examine the
land use/land cover change from deforestation and agricultural activities in Nang Rong
(Walsh et al. 2001). It was seen from statistical analysis that elevation changes in the
Nang Rong area were slight; however, they could have great implications in land use
practices and the suitability of soils (Walsh et al. 2001). Agricultural applications of
SRTM data could have worldwide significance because of the value of agricultural

products as food and commodities. The supply of food is an important concern of all

43

citizens regardless of locations, thus an application related to food production or land use
practices would be significant to research. Thailand, similar to nations around the world,
aims to gain the most yields from its resources. Hence, the use of geospatial data,
specifically elevation data, could be used to manage land effectively based on the optimal
growing conditions of crops. As of 2007, only Gorokhovich and Voustianiouk (2006)
have used Thailand SRTM data for comparison to GPS reference data. This research
includes an agricultural application, which illustrates the potential differences of DEMS
used in the identification of areas on the landscape. Now that the key characteristics of
the Nang Rong study area have been described, these will be referred to in the following
section on the selection of the United Study areas, as the characteristics of the Nang Rong
study area were identified prior to the selection of the United States study areas.
3.3. Selection of United States Sites

Following examination of the characteristics of the Nang Rong region, locations
in the United States with rice production were examined for possible comparison to Nang
Rong. The main areas of rice production in the United States include locations along the
Mississippi River, southeastern Texas, southern Louisiana, and parts of central California
(USDA 2005). A map of the areas of rice production from USDA (2005) was examined
to determine if these locations would also have high resolution elevation data available. It
was evident that climatic, soil, and water properties of only certain locations in the United
States area are appropriate for rice production. Although, the type of rice and agriculture
techniques of the US differs fiom the Thailand context, comparison can still be made
because the crop characteristics would still be similar on the surface during the time of

the SRTM. The main differences in terms of the rice production in Thailand and

44

Louisiana relate to rice species and irrigation method. The Thailand rice species are
mainly rainfed, while the Louisiana species are irrigated (Phongphit and Hewison 2001,
After the areas of rice production were identified and the locations of available
reference data (Table 2.5), the state of Louisiana was honed in on to determine possible
study areas. The National Land Cover Data (N LCD) 2001 was examined for the state of
Louisiana to determine the variability of land cover/ land use characteristics. Similarly,
the DEM of the Louisiana was examined for differences in topographic characteristics.
Based on these examinations, it was observed that the southwestern part of Louisiana was
mainly agriculture and wetland land covers, while the northern part of the state has more
forested areas. Following the identification of these characteristics fiom the NLCD 2001
and DEM of the entire state, the locations indicated in Figure 3.3 were selected.
Additional information follows regarding the characteristics of selected study areas. The
Webster study area in northern Louisiana extends from 93°37’23.1 ” W, 32°37’38.9” N
in the southwest corner to 93°15’12.7”W, 33°0’3.0”N in the northeast corner. The
Acadia study area in the southwestern portion of Louisiana extends from 92°37’36.4”W,
30°7’31.4”N in the southwest comer to 92°7’29.3”W, 30°29’54.7”N in the northeast

corner.

45

 

WEI Webster Study Area

 

Mira—4

I

 

I
if

.‘I'.

Acadia Study Area

 

 

I

 

Gulf of Mexico It.

 

 

 

 

I

.3“

U 30 50 120
—=—=
Idle meters

1

 

 

Figure 3.3. Locations of Study areas in Louisiana

3.4. Louisiana Study Areas

Of the possible areas in Louisiana with suitable land cover characteristics and

available reference data, two locations were chosen: an area in southwestern Louisiana

(Acadia study area) covering parts of Jefferson Davis, Evangeline, St. Landry, Lafayette,

46

Acadia, and Vermillion parishes, was chosen based on its rice production (9.0 %) and an
area in northern Louisiana (Webster study area) included parts of Webster and Bossier
Parishes with a percent of forest cover (62.2 %) were chosen (USDA 2006) (Table 3.1).
At the time of obtaining data for select study areas, some locations around the potential
forested areas in northern Louisiana were not available. Hence, the extent 567 mi2 of the
Webster study area is smaller than the Acadia study area 781 miz. The southwestern and
northern study areas in Louisiana represent two locations with different land cover/ land
use properties, which as seen from Chapter 2 may have different error characteristics. In
particular the Acadia study area is comparable to the Nang Rong agricultural
characteristics, since the Webster study area is mainly forested.

Since the Acadia and the Webster Study area crosses political county borders, a
remotely sensed imagery was used to identify the crop-specific characteristics of the
study areas. The National Cropland Data Layer-2005 was obtained and the percentage of
each class was calculated to create table 3.1. Table 3.1 indicates the percentage of rice in
the Acadia study area to be 9.0 % of the total area, which was greater the trace amount in
the Webster study area. Other key characteristics obtained from the crop specific data
included the percentages of woodland (12.2 % for the Acadia study area and 62.2 % for
the Webster study area), percentages of other crops in the study areas such as cotton and

grain in Acadia, and pasture percentages.

47

Table 3.1. Crop-Specific Area Calculated from National Cropland Data (2005)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Code Description Cell Counts Area (m2) Percent of area
Acadia Webster Acadia Webster Acadia Webster
1 Corn, all 19,974 1,178 17,976,600 1,060,200 0.9 % 0.1 %
2 Cotton 54,780 542 49,302,000 487,800 2. % 0 .0 %
3 Rice 202,584 2,760 182,325,600 2,484,000 9.0 % 0 .2 %
4 Sorghum 5,724 2 5,151,600 1,800 0.2 % O .0 %
5 Soybeans 39,939 15,464 35,945,100 13,917,600 1.8 % 1.0 %
24 Winter 191 139 171,900 125,100 0.0 % 0 0%
Wheat
25 Other 111,935 11,852 100,741,500 10,666,800 5.0 % 0.7 %
Grains/ Hay
45 Sugar Cane 65,799 1 59,219,100 900 2.9 % 0 .0 %
48 Watermelon 0 22 0 19,800 0.0 % 0 .0 %
50 State 560 9,112 0 8,200,800 0 0.4 % 0 .0 %
Other Crops
52 State 562 0 467 0 420,300 0.0 % 0 .0 %
Other Crops
53 State 563 2 0 1,800 0 0.0 % 0 .0 %
Other Crops
56 State 566 9 0 8,100 0 0.0 % 0 .0 %
Other Crops
61 Idle 101,039 4,552 90,935,100 4,096,800 4.5 % 0.3 %
Cropland
Fallow
62 Pasture, Non- 821,861 461,177 739,674,900 415,059,300 36.6 % 28.3 %
agriculture
63 Woodland 275,885 1,016,007 248,296,500 914,406,300 12.2 % 62.2 %
81 Clouds 228,257 49,950 205,431,300 44,955,000 10.1 % 3.1 %
82 Urban 103,024 23,566 92,721,600 21,209,400 4.6 % 1.4 %
83 Water 206,268 26,573 185,641,200 239,157,00 9.2 % 1.6 %
85 Waterway 0 17,860 0 16,074,000 0.0 % 1.1 %
87 Wetland 0 124 0 111,600 0.0 % 0.0 %
92 Aquaculture 0 159 0 143,100 0.0 % 0.0 %

 

 

 

 

 

 

 

 

Source: USDA 2006. This crop specific data was originally obtained from satellite
imagery from the Landsat 5 Thematic Mapper.

As indicated in Chapter 2, the SRTM X-Band data are not available for the entire

earth because of its smaller swath width. Figure 3.4 indicates the X—band coverage of

both study areas. The X-band SRTM 25 m resolution data for the Louisiana study areas is

not available; however, X-band SRTM data can be obtained for part of the Nang Rong

region, and these data were obtained.

48

 

 

 

 

Louisiana Nang Rong, Thailand

 

 

 

Figure 3.4. Availability of SRTM X-band Coverage in study areas
Modified from (DLR 2007).

The following section provides background information for the Louisiana study
sites, focused on its land cover/land use, agricultural production, and environmental
characteristics. These geographic and environmental characteristics are especially
important in the development of a suitable application where the elevation product would
be a significant factor in the analysis. A great deal of publicly available data can be
obtained to gain more knowledge about the environmental characteristics of these sites,
unlike that available for the Nang Rong site.

3.4.1. Acadia study area

The southwestern study area contains part of Jefferson Davis, Evangeline, St.
Landry, Lafayette, Acadia, and Vermillion parishes. This study area will be called the
Acadia study area in this research. This southwestern area of Louisiana is mainly rural.
The National Land Cover Dataset 2001 of the Acadia study area seen in Figure 3.5
indicates this area to be primarily agricultural and wetlands (USGS 2003). The two major

land characteristics of the southwestern Louisiana study area the 1) Gulf Coast Prairie,

49

 

 

 

”saw: . . . . .
cow R can on cow 9 com w omv n o

 

mn ”1.3.—on @042 .85an

<

Z

 

E. a: c 26:62, €82, a
3. RS 38.98 H
3.. 9.3 23.822 '_

 

 

:35 .82 «23.63
room .630 use; .2252

 

 

50

 

 

 

 

Figure 3.5. Acadia study area: National Land Cover Dataset (2001)

which is characterized by mainly cropland, pastureland, and loamy to clayey soils, that
are poorly drained; the 2) Southern Mississippi Valley Silty Uplands characterized by
mainly cropland and pastureland, and poorly drained soils (McDaniel, Trahan, and
Godfrey 2006). The natural vegetation of the Gulf Coast Prairies contains loblolly pine,
water oak, sweetgum, Southern red oak, and magnolia on the better drained soils in
higher positions and poorly drained soils in lower positions have water oak, willow oak,
overcup oak, swamp chestnut, green ash, and bald cypress present (McDaniel, Trahan,
and Godfrey 2006). The main transportation networks of the Acadia area include
Interstate 10 and US Highway 90. The main areas of development are located along these
corridors including the cities of Crowley and Eunice (Figure 3.6). The stream network in
Figure 3.7 illustrates the main drainage patterns in the Acadia study area. Generally, the
Acadia study area contains many oil and gas fields, in addition to agricultural land uses.
The climatic and soil properties of parts of the Acadia study area create suitable
conditions for rice cultivation. The average annual total precipitation is 60 in. (0.406 m),
of which 82 % occurs between the main growing period, from February to November
(McDaniel, Trahan, and Godfrey 2006). The temperature ranges from an average of 52
°F in the winter to 81° F in the summer (McDaniel, Trahan, and Godfrey 2006). For
southwestern Louisiana, the growing season consists is 280 days (Salassi and Breaux
2006). The average yearly relative humidity in this area in mid-afternoon is 71 %
(McDaniel, Trahan, and Godfrey 2006). The main soil map units (91%) in the Acadia
parish comprise of soil on level to gently sloping surfaces slopes of 0-3 % with relatively
poor drainage, which makes them well-suited for rice cultivation(McDaniel, Trahan, and

Godfrey 2006).

51

 

 

Evangeline [ /——-———/ St. Landry

Eunice

   
      

US 190

 

   

Jefferson Davrs

0
Church Pomt

 

 

 

        

 

 

 

 

Duson
Mermentai US 90 Estherwood
Lafayette
Vermilion
N
Legend
0 3 200 6400 12 800 19200
0 Place Names _ ‘ _' ' . A

— Major Highways

Source: LAGIC and US. Census 2007. Mineral Management Service 1999.

 

 

 

Figure 3.6. Reference Map of the Acadia study area

52

 

 

7&1

fir
rim

 

[UV/4*

AA

 

Legend N

Streams 0 3,000 6.000 12.000 18,000 A
-==—Meters

Source: US. Census 2006

 

 

 

 

Figure 3.7. Major stream network of the Acadia Study Area
3.4.2. Webster study area

The northern Louisiana study area contains part of Webster and Bossier parishes.
This study area is referred to the Webster study area in this research. Webster parish is
predominantly rural with many outdoor recreation opportunities related to the rich

environmental features such as forests and waterways (Webster Parish 2001). Springhill,

53

Cotton Valley, and Minden developed area are located along US Highway 371, which
runs north to south in this study area. Cotton Valley is bordered by the Bodcau State
Wildlife Management Area and the Cotton Valley Oil and Gas Fields. The predominant
crops in this study area include cotton, grain, sorghum, corn, wheat, hay, and sweet
potatoes (Kilpatrick, Godfrey, and Henry 1998).

The Webster study area is predominantly forested (Figure 3.8). The Bodcau
Wildlife Management Area in the northwestern portion of this study area and the
Kisatchie National Forest is located in the southeastern portion of this study area (Figure
3.9). The rich habitat of this area includes cypress swamps, upland pine, hardwood
forests, grassland, and open fields (LWLF 2005). The bottomland forests consist of bald
cypress, willow, and cow oaks, while Shortleaf and loblolly pine, white, red, and
cherryka oaks, sweetgum and elm trees dominate upland forests (LWLF 2005).
Streams run through these forests and upland areas as seen in Figure 3.10.

The Webster study area has climatic conditions similar to the Acadia study area.
Average annual temperature is 65°F in Webster Parish (Webster Parish 2001). Climatic
characteristics of this area include a summer average of 80°F, and an annual precipitation
of 52 inches (Kilpatrick, Godfrey, and Henry 1998). Webster Parish is comprised of level
to steep uplands and terraces, with soils that are not conducive to agriculture (Kilpatrick,
Godfrey, and Henry 1998). The soils located on uplands are gravelly, loamy, and sandy,
with poor drainage (Kilpatrick, Godfrey, and Henry 1998). Thus, the factors of sloping
topography, soil wetness, and low fertility in this area limit the growth of crops

(Kilpatrick, Godfrey, and Henry 1998).

54

 

 

 

National Land Cover 2001 (percent total area)

 

E Water (2 =7.) - Deciduous Forest (6 °/.) C Developed (2 °/.)

- Wetlands (15 %) [:j Forest/ Shrub (20 %)
- Evergreen Forest (50 %) Agriculture (6 0/o)

 

 

N
0 2,750 5.500 11,000 16.58%68 I
l

NLCD 2001. USGS
UTM Zone 15N. WGSB4

 

 

 

Figure 3.8. Webster Study area: National Land Cover Dataset (2001)

55

 

 

    

 

 

 

 

 

O
Shongaloo
Bodcau Sate
WlIdIIIe
Management Area
5
m
.D
E
E
I?
O
D]
Kisatchie
National Forest
Legend 0 3,550 7,100 14200 N
0 Place Names
—- Major Highway

Source: LAGIC and US. Census 2007. Mineral Management Service 1999.

 

 

 

Figure 3.9. Reference Map of Webster Study Area

56

 

 

 

 

 

 

 

Source: US. Census 2006

 

nae-S

-‘/4“"..\-;' -.

 

 

’ ..
g .
a.
*.
:1.
Legend 3500 7000 14.0.0.0.
—Streams

'3-2

{\‘h x'

f/.

 

 

Figure 3.10. Major Stream Network of the Webster Study Area

57

 

The examination of the land cover/land use characteristics indicated the potential
of an application related to elevation characteristics and land use being pertinent across
the Acadia, Webster, and Nang Rong study areas. The following section discusses
information related to a land use application that could be relevant in all study areas.

3.5. Land Use Application

Digital terrain models have much utility in diverse geographic application
domains, including urban growth models, land use, weather modeling, and vegetation
analysis (Wilson and Gallant 2000). An agricultural model was identified as an
appropriate application to explore in both the Thailand and Louisiana contexts given their
similar surface characteristics. This type of application highlights how the use of terrain
models can play a vital role in agricultural models worldwide to better understand land
management, food supply dynamics, irrigation methods, and soil management. This
section provides a brief review of the approaches to create a GIS model for rice
production given the available datasets for both locations and the goal of identifying
differences in model use.

Rice production is an important land use in the Acadia study area of Louisiana, in
contrast to the more forested land uses in the Webster study area. The availability of
water and the soil characteristics of this region create a suitable location for rice
production (McDaniel, Trahan, and Godfrey 2006). According to an Extension Rice
Specialist and Retired Extension Agronomist of Mississippi State University, rice
production is based on the following factors: soils with permeable subsoils, flat area, and

a gentle slope (Street and Miller 2004). These criteria for rice production identify the first

58

derivative of the DEM, slope, to be a major factor for rice production, along with
drainage characteristics.

Similar to the Louisiana study areas, the low-slope topography is a key suitability
factor for rice cultivation in Thailand. In a land suitability study for rice in the Namphong
watershed of Northeastern Thailand, slope, water availability, nutrient availability, and
salt hazard were variables input into the spatial model (Mongkolsawat, Thirangoon, and
Kuptawutinan 1997). Elevation was noted as an important factor for land suitability in the
Nang Rong district because of its ties to other land surface characteristics, such as soil
moisture (Entwisle et al 2005). The availability of soil moisture can play a role in the
cultivation of certain crops in Thailand, such as the growth of rice on the alluvial plain
areas versus the growth of cassava on the uplands (Entwisle et al 2005). Thus, not only
slope is important, but the relative position on the landscape is important to crop
cultivation.

Topographic position is identified as a relevant terrain characteristic by Deng
(2007). Low landscape position represents a criterion here for land suitable for
agriculture. These areas on the landscapes would have flow accumulation and be
appropriate for the production of rice and/or other agriculture in the Nang Rong area. The
schematic of a low location indicates the position of an area in relation to surrounding
upland areas (Figure 3.11). Through manipulation of an elevation surface, the low

topographic positions on the landscape can be incorporated into environmental analysis.

59

 

 

 

Locally Low Elevation

Figure 3.11. Schematic of a low location on the landscape. This low location criterion
will be used in the land use application.

 

 

In the context of this thesis, the model variables related to the DEM are most
relevant. Here a simplified model will be created based on identifying locations on the
landscape where potentially rice production or other agriculture can be located based on a
low slope and low relative location on the landscape. Analysis of sophisticated crop
models with multiple variables is outside the scope and aims of this application. The aims

of this application are to assess differences in DEM results.

60

4. Methods

The methods section indicates the procedures used for Louisiana and Nang Rong,
Thailand study areas, and application development. These sections are organized based
on data source and processing steps necessary for examining SRTM error properties for
each study area. National datasets are used for the United States to enable the possibility
of replication in different areas than the ones chosen for this analysis. This methodology
enables the examination of SRTM DEM error characteristics in relation to differences
between the DBMS (LiDAR and NED) and in relation to land cover/land use
characteristics.
4.1. Processing Software

Geographic information systems (GIS) and statistical software were used to
investigate the SRTM error characteristics in relationship to landscape parameters. The
methodological decisions and techniques employed in this research used ArcGIS 9.1—9.2
in ArcInfo versions, in a Windows platform, and ERDAS Imagine (8.7) software. The
statistical software used was SYSTAT (10.0).
4.2. Naming conventions

The following acronyms used in this thesis are indicated with their meanings
(Table 4.1). The resolution of the C-band is indicated when necessary, and the X-band
SRTM resolution is also indicated to distinguish between the original resolution of 25 m

and its resampled version at 90 m.

61

Table 4.1. Naming Conventions of Data Types with Descriptions

 

Description

LIDAR 5 m Original LiDAR-derived DEM
LIDAR 90 m Resampled LiDAR-derived DEM
NED 30 m Original NED DEM from the USGS
NED 90 m Resampled NED DEM

SRTM 30 m SRTM 30 m C-band DEM from USGS
SRTM 90 m SRTM 90 m C-band DEM from USGS
SRTM X-25 m SRTM X-band from DLR

SRTM X-90 m Resampled SRTM X-band

 

 

 

 

 

 

 

 

 

 

 

 

4.3. Louisiana study areas
4.3.1. LiDAR Data from Atlas, Louisiana Statewide GIS

The source of the LiDAR data for Louisiana was the Atlas, Louisiana statewide
GIS (Atlas 2007). The Atlas, Louisiana Statewide GIS online source of the LiDAR-
derived DEM data used as reference was carefully examined to determine whether it
would be suitable for analysis in this research (ATLAS 2007). The Louisiana Statewide
GIS distributes various raster and vector datasets. Digital topographic quads, LiDAR-
derived DBMS, and imagery are obtained though online viewers (ATLAS 2007).

The LiDAR data were identified by their 1: 24,000 Quadrangle USGS name.
Appendix A identifies the USGS quads (1: 24,000 scale) in each study area, along with
their unique identifiers. The Acadia study area covered 16 quadrangles (781 miz, 2,022
kmz) and the Webster study area included 12 quadrangles (567 miz, 1,469 kmz). During
this process, the decision was made to use LiDAR-derived raster data instead of point
LiDAR data. LiDAR point data were seen as less appropriate for this analysis given as it
would more processing time, which would not be appropriate for this research. Thus, the

LiDAR-derived data for the selected data were obtained in interpolated-raster format. The

62

LiDAR DEM was downloaded in USGS DEM format, converted to raster, integer
format, and mosaiced together to obtain seamless coverage of the desired area.

The Acadia study area data were flown in February/March of 2003 as part of the
Louisiana / Federal Emergency Management Agency (FEMA) Project - Phase 3 of
Louisiana LIDAR Data Development (Watershed Concepts, 2004a,b,c). The Webster
study area data were obtained in February 2006 as part of the Louisiana / Federal
Emergency Management Agency (F EMA) Project - Phase 5 of Louisiana LIDAR Data
Development: Bossier / Webster Parishes, at 5 at resolution (Watershed Concepts 2006a,
b; Watershed Concepts 2005). The processing contractor used the Arclnfo latticedem
command to create the DEMS from the point-collected LiDAR data (Watershed Concepts
2004a,b,c; Watershed Concepts 2006a,b; Watershed Concepts 2005). The seasonality of
the LiDAR acquisition was considered in relation to the timing of the SRTM mission to
obtain comparable surface conditions.

The LiDAR DEMS heights were converted from feet to meters by the raster
algebra: DEMMcters = DEMFeet *3.28084. Following the metric conversion, the DEM was
projected and resampled to 90 m by the nearest neighbor method. The nearest neighbor
method was chosen to resample the DEMS because it would provide the center cell of the
90 m cell extent of the SRTM resolution. This would represent an elevation value to align
with the SRTM 90 m cell extent. Figure 4.1 illustrates these processing steps of the

LiDAR DEMS.

63

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

NLCD
2001
(30 m)

 

 

 

 

 

 

 

 

Focal
Majority

 

 

 

LiDAR NED
‘/4 Quad DEMs (30 m)
(5 m)
V
Resample Resample
Nearest Nearest
Neighbor Neighbor
LiDAR 90 m NED 90 m

I

 

 

 

 

l

NLCD
90 m

 

 

 

 

Comparable Cell Resolution of the
SRTM C-band DEM 90 m

 

 

Figure 4.1. Processing Steps for Louisiana Data

The following process was employed in each successive resampled DEM of this
research to ensure the alignment of raster cells following resampling. When the DEM
was resampled to the resolution of the SRTM C-band data, it was set to 90 m and the
raster environment was set to snap to the cells of the SRTM 90 m DEM. This raster cell
alignment was necessary for error analysis. The coordinate system used in analysis of the
Louisiana DEMs was UTM zone 15N, WGS84 datum (horizontal and vertical) in units of
meters. This system ensured that both study areas in Louisiana shared the same zone and

that consistency was maintained between the datum of the SRTM C-band data and the

other DEMS.

64

 

 

 

 

 

4.3.2. National Elevation Dataset

The National Elevation Dataset (NED) DEM was obtained from the USGS
(USGS 2007, USGS 1999). The NED 1 are second DEM was downloaded for selected
Louisiana study areas, projected with a cell size of 30 m, clipped to desired extents, and
resampled to 90 m by the nearest neighbor method to match the resolution of the SRTM
(Figure 4.1). The Acadia study area NED data were produced by methods 5 (LT4X
software) and 3 (Digital Line Graph2DEM) and the data for Webster study area were
produced by method 5 (LT4X software) (USGS 1999). The Line Trace 4X (LT4X)
software is a specialized software for the creation of DEMs. This information was
obtained from the metadata of the NED from the USGS and refers to the production
methodology software used in the creation of the NED (USGS 2007). Hence, the LiDAR-
derived DEMS have not been incorporated into the NED product as of July 2007;
however, have been incorporated into other areas of the state.
4.3.3. Shuttle Radar Topography Mission: C-Band

The finished SRTM 1 are second and 3 are second data were obtained from the
USGS seamless server for the Acadia and Webster, Louisiana study sites (U SGS 2007,
NASA 2002). The SRTM data used in this research were downloaded after the USGS
corrected the half pixel shift of the SRTM data (USGS 2007). The SRTM 1 are second
and 3 arc second data were projected with a 30 m and 90 m resolution, respectively, and
clipped to the desired extents. For the NED, LiDAR, and SRTM data, no filling operation
was employed to fill voids in the DEMs. Thus, the DEMS were used without further
altering their elevation values to account for sinks or voids in the data because this

research is interested in the DEM form obtained by the public or other researchers

65

originally from the USGS and/ or other Sites. The methods used to fill sinks or voids in
the DEM can differ from one another, and selection of the most appropriate filling
parameters or void adjusting method was not part of the research.
4.3.4. Land Cover/Land Use Characteristics

National Land Cover Data (N LCD) 2001 were used to examine the relationship of
different land cover/land use classes to SRTM error (USGS 2007, USGS 2003). The
specific dataset was the National Land Cover Database Zone 37B, which is the zone
covering Louisiana. The NLCD was resampled by using a focal majority (3 x 3)
operation to obtain the major land cover code in each cell of the SRTM 90 m C-band
DEM (Figure 4.2). The focal majority was chosen to identify the land cover code, which
represents the majority of the areal extent in the 90 m area of the 3 by 3 window. As
depicted in Figure 4.2 in a 90 m by 90 m neighborhood, the focal majority identifies the
code (i.e. 54) with the majority in the new raster. This would be possible with the focal
majority neighborhood operation, rather than a focal average operation, which computes
an average value of the codes.

The percentage of the land cover/land use codes for the Webster and Acadia study
areas were recalculated following the focal operation. The resampled NLCD data were
converted to polygons and used to stratify the elevation error samples using Hawth’s

tools (Beyer 2004).

66

 

Focal Average Focal Majority

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

11 11 22 53 53 54

11 22 22 53 54 54

22 22 22 54 54 55
18.3 54

 

 

 

 

 

 

 

 

 

Figure 4.2. Focal Operations
4.4. Nang Rong, Thailand Study Area
4.4.1. C-Band SRTM

The SRTM C-Band 90 m resolution data were downloaded for the Nang Rong
region from the USGS Seamless Server (USGS 2007, NASA 2002). This DEM was
projected and clipped to the extent of the coverage of the X-band data for the same
region. The coordinate system of the Thailand data were UTM Zone 48N, WGS84 datum
(horizontal and vertical) in units of meters (Figure 4.3).
4.4.2. X-Band SRTM

The SRTM X-SAR data were obtained in digital terrain elevation data (DTED)
format from GAF AG, the German X-band distributor (DLR 2007). The X-Band DEM
was projected and clipped to the extent of the overlap with the Nang Rong District
boundary (1,564.2 kmz). The X-band SRTM data (25 m) were resampled by the nearest

neighbor method to obtain the same cell resolution of the C-band SRTM data (Figure

67

4.3). The raster environment was set for the resampled 90 m cells to align with the

SRTM C-band cells.

 

 

 

Download Obtain
SRTM SRTM
C-band from USGS X-band from DLR

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Clip '0 Resample
X-band Extent Nearest Neighbor
I J

 

Difference

SRTM C-band —
Resampled SRTM X-band

 

Error Analysis

 

 

 

Figure 4.3. Processing Steps for Nang Rong, Thailand study area. The Nang Rong study
area used the SRTM X-25 rn resampled to 90 m as the reference to the SRTM C-90 m.

4.5. Datum Transformations

The NED, LiDAR, and SRTM DEM employ different vertical datums: NAVD88,
NAVD88, and WGS84, respectively. As indicated in the background section, these
datums needed to be adjusted to account for the differences in their base elevation
reference information. A possible way to account for datum difference is through the use
of online converters, such as the NGA EGM96 Geoid calculator (N GA 2006) to
determine the geoid heights (m) and the GEOID03 calculator (N GS 2007), which would
reflect the NAVD88 vertical datum. However, a different approach was employed in this

thesis to adjust for the differences in the vertical datums, which was the use of a graphical

68

representation of datum differences in North America. The average difference between
the WG884 and NADV88 was about 0.6 m for the Louisiana study areas (Roman 2007).
This value was added to the NED and LiDAR DEMS, so that their elevations would
reflect the WGS84 datum of the SRTM C-band DEM.

To correct for the vertical datum differences of the SRTM C-band and SRTM X-
band DEMS, DLR, the distributor of the X-band DEM, suggested the use of ERDAS
Imagine. The Recalculate Elevation Values tool was used in ERDAS IMAGINE. The X-
band SRTM DEM was recalculated to reflect the elevation values of the SRTM C-band
WGS84/ EGM96 Geoid values.

4.6. Terrain Derivatives

The first derivative of elevation, Slope (degrees), was generated in ArcGIS 9.2
using the Spatial Analyst extension for each elevation model indicated above. Slope is
defined as the rate of change from the steepest neighboring cell, calculated via the Horn
algorithm in Arc GIS 9.2 (ESRI 2006a).

4.7. Error Analysis

The LiDAR 90 m was subtracted from the SRTM 90 m to obtain the SRTM-
LiDAR error grid. Similarly, the elevations of the NED 90 m DEM were subtracted from
the SRTM 90 m DEM (Figure 4.4). The difference of the SRTM and the reference DEM
is the DEM error. A Similar subtraction method of the X-band and the C-band was

employed in the Nang Rong study area (Figure 4.3).

69

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Resampled Resampled
SRTM LiDAR SRTM NED
l I I I I |
SRTM-LIDAR SRTM-NED
Difference Difference

 

 

 

 

 

 

 

 

Figure 4.4. Louisiana study area: Error Analysis. For the Acadia and Webster study
areas, error analysis of the SRTM C-90 m DEM was done using both the LiDAR-90 m
and NED 90 m DEMS.

Each error grid was squared and the average value for each raster was calculated.

The square root of the mean squared error is the root mean square error (RMSE), which

is:

 

 

~ 2 —Z 2
RM S E: Jz( DEM Reference) (41)
n

The RMSE values are reported in the results section. Additional descriptive
statistics for the study areas include minimum, maximum, mean, and standard deviation
of the elevation errors (meters), which are presented in Chapter 5.

Error analysis should be developed with independent samples (Shortridge 2006).
However, spatial autocorrelation exists in the DEM and error grids; thus, a determination
of the distance at which this occurs would be useful in setting the minimum distance
between sample points. The ArcGIS Geostatistical Analyst was used to examine the
semivariogram of the error grid to measure its spatial autocorrelation. The lag size of the
semivariogram was set to 90 m, the cell resolution of the SRTM C-band data. The major
range indicates the average minimum distance between uncorrelated cells. This distance

was determined to about 650m for the SRTM-LiDAR error grid for the Webster study

70

area. Thus, a distance of 650 m was used as a minimum distance between sample points
in a 1500 point sample for the Webster study area. For the Acadia study area, a minimum
distance of 800 m was used to create the sample initially with 2000 sample points. Based
on the minimum inter-sample distance (i.e. the distance between sample points) and the
spatial extent of the areas, there was a limit to how many points could be obtained from
the study areas given the minimum distance set in sampling.

Hawth’s tool extension for ArcGIS (Beyer 2004) was used to sample the data.
This sample is stratified by land cover type of the NLCD focal majority (90 m). This
sample was exported for subsequent analysis in SYSTAT 10.0 software, and missing data
were removed from the dataset. Statistics are included in the results section. The
following subsection indicates the characteristics of the samples taken from the Webster
and Acadia study areas, and Nang Rong study area.
4.8. Sampling Procedure
4.8.1. Louisiana study area

The NLCD 2001 Land Cover/ Land use data were reclassified into groups to
represent different land cover classes, which could be relevant to the characteristics of
DEM error. Table 4.2 indicates the NLCD 2001 land cover/land use categories, along
with the reclassified class, as justified below.

Table 4.2. Reclassification Scheme of the NLCD (2001)

 

 

 

 

 

 

 

 

 

 

 

NLCD Name Reclassification Class
2001 Code
1 1 Open Water Water
21 Developed, Open space Developed
22 Developed, Low Intensity Developed
23 Developed, Medium Intensity Developed
24 Developed, High Intensity Developed
31 Barren Land (Rock/Sand/Clay) Agriculture/Barren
41 Deciduous Forest Deciduous Forest

 

 

71

Table 4.2. continued

 

 

 

 

 

 

 

 

 

 

 

 

42 Evergreen Forest Evergreen Forest
43 Mixed Forest Forest/ Shrub

52 Shrub/ Scrub Forest/ Shrub

7 l Herbaceous/ Grassland Agriculture/Barren
8 l Pasture/Hay Agriculture/Barren
82 Cultivated Crop Agriculture Barren
90 Woody Wetlands Wetland

95 Emergent Herbaceous Wetland Wetland

 

The NLCD were reclassified in 7 categories based on the understanding that
different land cover characteristics may respond differently with the SRTM, which would
ultimately lead to different error characteristics. Table 4.3 indicates the cell counts and
percentages of the NLCD for each study area. The open water class was left into a group
of its own, since the radar technology has been noted to respond to water different than
land surfaces. The NLCD 2001 values 21, 22, 23, and 24 representing the developed land
uses were grouped together, since they represent built-up surfaces on the landscape. In
terms of the natural features on the landscape, the barren land was grouped with the
pasture/hay, herbaceous/ grassland, and cultivated crop classes because all of these land
covers have little vegetation above the earth’s surface.

The forest classes evergreen and deciduous were left as distinct classes in the
dataset because their tree structure characteristics could potentially have different error
pattern in the SRTM DEM data. The coniferous and deciduous trees on the landscape
would have different leaf and surface characteristics, which could create different error
patterns when the SRTM and reference DEMS are compared. As seen in Walker,
Kellndorfer, and Pierce (2007), the SRTM radar does not penetrate trees completely
because of its short wavelength and is scattered by branches and leaves. Thus, whether

the trees are known to keep their leaves or not could play a role in the scattering potential

72

of the SRTM, and ultimately the SRTM DEM error characteristics. The mixed forest
class (N LCD code 43) is identified as a class with deciduous and coniferous tree species,
which in this research is identified as a mixed vegetation class. The mixed forest (N LCD
code 43) and the shrub/scrub (N LCD code 52) were grouped together as a forest/shrub
class in this reclassification. However, for analysis of the sample, the points identified
originally as shrub were removed from the sample because the height of the shrubs can
not be determined. In both the Webster and the Acadia study areas, the emergent
herbaceous wetlands represent very small portions of these areas. Thus, the herbaceous
wetland class (NLCD code 95) was grouped with the woody wetlands class (N LCD code
90). The wetlands class was a distinct land cover class because of its predominance in
Louisiana.

Following the reclassification of the NLCD 2001 (90 m) data, the percentages of
each reclassified class (column 3 of Table 4.2) was calculated. The reclassified NLCD
2001 (90 m) was converted to a vector format for use with the Hawth’s tool extension,
which requires this format to stratify the sample of the dataset (Beyer 2004). These
percentage values were input into the database and used to extract a random stratified
sample with the indicated minimum distance (i.e. 650 m and 800 m).

Table 4.3. NLCD Summary Statistics for Louisiana study areas

 

 

 

 

 

 

 

Webster Acadia

NLCD Name Reclassification Cell % of Cell % of

2001 Category count entire count entire

Value (90 m) area 4903) area

11 Open Water Water 2,693 2 % 494 0.2 %

21 Developed, Open Developed 1,540 l % 698 0.3 %

space

22 Developed, Low Developed 2,156 l % 9463 3.8 %
Intensity

23 Developed, Medium Developed 109 0 % 193 0.1 %
Intensity

 

 

 

 

 

 

 

73

 

Table 4.3. continued

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

24 Developed, High Developed 54 0 % 123 0.1 %
Intensity
31 Barren Land Agriculture/Barren 12 0 % 21 0.0 %
(Rock/Sand/Clay)
41 Deciduous Forest Deciduous Forest 9,702 6 % 14 0.0 %
42 Evergreen Forest Evergreen Forest 83,900 50 % 96 0.0 %
43 Mixed Forest F orest/Shrub 9,603 6 % 3 0.0 %
52 Shrub/Scrub Forest/Shrub 24,268 14 % 415 0.2 %
71 Herbaceous/Grassland Agriculture/Barren 0 0 % 901 0.4 %
81 Pasture/Hay Agriculture/Barren 9,848 6 % 29,222 11.9 %
82 Cultivated Crop Agriculture Barren 156 0 % 160,752 65.3 %
90 Woody Wetlands Wetland 24,523 15 % 43,818 17.8 %
95 Emergent Herbaceous Wetland 15 0 % 127 0 %
Wetland
Total 168,579 246,340

Table 4.4 Webster study area Reclassified Values

Reclassified Class # of Cells (90 Percent of Count in
m N LCD) area Sample
(n =1484)

1. Water 2,693 2% 23 (1.5 %)

2. Deciduous Forest 9,702 6% 38 (2.6 %)

3 Evergreen Forest 83,900 50% 740 (49.9 %)

4. Developed 3,859 2% 82 (5.5 %)

5. Forest/Shrub 33871 20% 300 (20.2 %)

6. Agriculture/Barren 10.016 6% 86 (5.8 %)

7. Wetland 24,538 15% 215 (14.5 %)

 

 

 

 

 

Table 4.5. Acadia study area Reclassified Values

 

 

 

 

 

 

 

 

 

 

Reclassified Class # of Cells (90 m Percent of Count in Sample

NLCD) area (n =1956)

1. Water 494 0.2 %

2. Deciduous Forest 14 0.0 %

3. Evergreen Forest 96 0.0 %

4. Developed 10,477 4.3 % 92 (4.7 %)

5. F orest/shrub 418 0.2 %

6. Agriculture/Barren 190,896 77.5 % 1,523 (77.9 %)

7. Wetland 43,945 17.8 % 341 (17.4 %)

 

 

 

 

 

Given the proportions of the above samples (Tables 4.4 and 4.5), a minimum

threshold of 20 points was set for each class, which eliminated some classes from being

74

represented. This minimum threshold was set so that the sample would be representative
of the study areas, for instance with the Acadia study area only 14 cells were deciduous
in the entire area, which is less than 1 % of the entire area. For the Webster study area, all
classes are represented in the sample; however, with the Acadia study area, the water,
deciduous, evergreen classes, and forest/shrub classes did not meet the minimum
necessary number of sample points to be represented in the sample (Tables 4.4 and 4.5).
The LiDAR 5 m, LiDAR 90 m, NED 30 m, NED 90 m, SRTM 30 m, and SRTM 90 m
elevation values and slope values were collected for the sample points. The SRTM-
LiDAR and SRTM-NED DEM error values were also collected. The NLCD 2001 and
MODIS Percent Tree cover (2000-2001) were also included in the sample. The MODIS
Percent Tree cover will be further discussed in a later section. Points with missing data
were removed from the sample. A similar methodology of using the semivariogram to
determine Spatial autocorrelation was employed with the Nang Rong study area.
4.8.2. N ang Rong study area

The semivariogram of the SRTM DEM error grid was examined to determine the
minimum distance between sample points for the Nang Rong study area. The cells were
spatially autocorrelated up to a distance of 400 m. Thus, a sample of 1537 points with a
minimum inter-point distance of 400 m was extracted from the Nang Rong study area
covered by both the SRTM-C- and X-band DEMS. The SRTM-90 m, SRTM X-25 m, and
SRTM X-90 m elevation and slope values were included in the sample, along with the
DEM error difference of the SRTM C-90 m and SRTM X-90 m DEM. The MODIS

2000-2001 data were also included in the sample, which is discussed in the next section.

75

4.9. Comparative Surface Properties

To compare surface cover characteristics of all study areas to SRTM, a data
source needed to be identified that was available in both locations. MODIS data (2000-
2001) were determined to be a possible source of data, which would be available for the
Louisiana and Thailand case studies. Continuous MODIS vegetation data (500 m)
available from the Global Land Cover Facility (GLCF) of the University of Maryland is
distributed by continental region in the Goodes Interrupted Homolosine projection
(GCLF 2007). North American data on percent tree, percent herbaceous, and percent bare
earth were downloaded from the GLCF (Hansen et al. 2003a). Similarly, Euroasian data
for percent tree, percent herbaceous, and percent bare earth were downloaded (Hansen et
al. 2003b). The projection information for the North America and Euroasian data was
defined in Arclnfo using the regional characteristics for the projection (Lethcoe and
Klaver 2007). The projected MODIS data were clipped to the extents of the study areas
for further analysis.
4.10. Application Development

Based on the discussion of relevant land uses in all study areas in Chapter 3, an
application was developed based on slope criteria and topographic position on the
landscape to determine how the selection of a certain DEM worked in an application
(Figure 4.5). The slope was reclassified for the LiDAR 90 m, NED 90 m, and SRTM 90
m DEMs. To determine the locally low elevations, a circular focal mean of a 900 m
radius (10 cells at 90 m resolution) was used in the focal operation. This decision was
based on observation of the semivariogram of the SRTM 90 m for Nang Rong, Thailand

DEM. For consistency, this distance was applied to all DEMs compared in the

76

application analysis. The actual elevation in each cell was subtracted from the smoothed
focal mean elevation. If the difference is positive, the focal mean is greater than the

actual, and the cell is a low location in the landscape.

 

0 if Slope > 1

Suitable Slope =
1 if Slope = 0-1
_ 0 if Mean Elevation- Actual Elevation 5 0
Suitable Low =
Posrtron 1 if Mean Elevation —Actual Elevation > 0

 

 

 

Figure. 4.5. Criteria for Land Use Application. This figure indicates the equations for the
suitable Slope and low position criteria.

This application addresses the second research question related to how differences
in DEMS impact the outcomes of applications. Figure 4.6 indicates the cartographic
model employed on the LiDAR 90 m DEM, NED 90 m, and SRTM 90 m DEM for the
Acadia and Webster study areas. Similarly for the Nang Rong study area, the model will
be employed with the SRTM C-90 m, SRTM X-25 m, and SRTM X-90 m to determine
how the DEMS perform. The multiplication of the suitable slope locations and the
suitable low locations on the landscape indicate the locations where the appropriate
criteria were found for both of these variables. The unsuitable and suitable cells for each
DEM were recoded with a different value, so that when they were added together, each
cell could be 1 of 4 different values. The cell counts were recorded for each combination.
An example is provided in Table 4.6 of the possible values from addition. In addition to

determining areas found suitable by all DEMS, contingency tables were made using the

77

LiDAR 90 m as reference in relation to the NED and SRTM application results. These
contingency tables indicate the unsuitable and suitable cell counts for the DEMS. For the
Nang Rong study area, the SRTM X-90 m was used as reference to the SRTM C-90 m.
The Kappa statistics were calculated to determine the agreement of the DEMS in terms of

their application results.

 

 

Flat Areas Low Areas on the Landscape
Focal
Mean
Subtract:

If Positive: Low Area

Slope Low

0- 1 deg Areas
Compare Results
LiDAR, NED, SRTM

Ovverlay Suitable Areas

 

 

 

 

Figure 4.6. Cartographic Model for Land Use Application. This flow chart illustrates the
creation of slope and low position variable, along with the overlay process to identify
suitable areas for the LiDAR 90 m, NED 90 m, and SRTM 90 m DEMS.

78

Table 4.6. Addition Process for the Comparison of DEM Application Results

 

 

 

 

 

 

 

 

 

LiDAR
Unsuitable Suitable
(10) (15)
SRTM Unsuitable(12) 22 27
Suitable (13) 23 28

 

This methodology illustrated the process of acquiring data from data repositories,
projecting, and processing the DEMS. The process of performing the error analysis was
outlined. Furthermore, the description of the land cover/ land use data from the NLCD
2001 was summarized, as the samples of the Louisiana study areas were stratified by the
NLCD 2001 codes. The flow chart of the application indicates the process employed on
the DEMS to determine how the different DEMS would result in different outcomes. The
methodology employed in the study areas in Louisiana and Thailand will aid in
identifying characteristics of the SRTM DEM. A better understanding of the properties of
the SRTM DEM will benefit future users of the DEM, especially in contexts without
another DEM. The following chapter reports the results of these methods and discusses

the overall themes across study areas.

79

5. Results and Discussion

This research is aimed at investigating the characteristics of the SRTM DEM in
relation to other available DEMS (i.e. LiDAR, NED, and SRTM X-band 25 m, where
available). After the selection of study areas in Thailand and parts of Louisiana (Acadia
and Webster study areas), the elevation and slope characteristics of these areas are
explored. In addition, the differences across land cover/ land use classes and SRTM error
are examined. This chapter contains the results of the DEM resampling, error analysis,
and application development. Results are organized by study area: Acadia, Webster, and
Nang Rong. The results and discussion of the Acadia and Webster study areas are
organized in a similar way. Results from the entire study area opposed to the sample are
denoted by the word “entire” in the figure or table title. Following discussion of the error
analysis of the SRTM models, the results of the land use application are presented with a
focus on comparing different DEM results. The discussion section addresses themes
across all study areas.
5.1 Louisiana study areas
5.1.1. Acadia study area

The Acadia study area is a relatively low-slope area in southwestern Louisiana.
Figures B.1-B.6 in Appendix B illustrate the elevation and slope characteristics of the
Acadia study area for the LiDAR, NED, and SRTM DEMS at their original and
resampled resolutions. The color scheme is the same for each map sequence for
comparative purposes. The LiDAR elevation model has finer detail of the geomorphic
characteristics of the earth, such as the properties of streams, compared to the SRTM 90

m resolution. Agricultural land use patterns and other small-scale topographic features,

80

such as canals, are differentiated from other features on the landscape using LiDAR. The
detail observable in the LiDAR DEM is also not seen in the NED DEM. In addition,
these landscape details are not preserved in the LiDAR 90 m or the SRTM C-band DEM.
The SRTM DEM 90 m contains striping on the southeastern portion of the study
area (Figure 5 .1). Similar striations have been noted by Bhang, Schwartz, and Braun
(2007) as an artifact from radar collection, and are interpreted as such here. The lower
elevation in the streams networks are represented as blocky patches of high and low error
in the SRTM C-band. In contrast to the blocky pattern and rectangular form of some
trends in the SRTM 90 m, the NED DEM has a comparatively smoothed appearance. The
NED DEM also includes the linear feature across the southern portion of the study area;
however, this is less prominent than in the LiDAR DEM. This could be attributed
possibly to an elevated road because it is higher than the surrounding cells and crosses

over a wetland area.

81

 

 

 

 

‘ Meters
0 3.400 6,800 13.803 20.4113
— — JMEIBYS Higr1161
LOW -27

 

 

 

 

 

 

Figure 5.1. Striping in the southeastern portion of the Acadia study area. Striping is an
artifact from the SRTM radar collection.

The contrasts in the elevation and slope (degrees) values of the LiDAR, NED, and
SRTM models are evident from comparison of the descriptive statistics (Table 5.1).
LiDAR and NED models have similar ranges, means, and standard deviations for both
their original and resampled resolutions. However, substantial differences are observed
between the SRTM DEM and the characteristics of the LiDAR and NED models. The
SRTM 30 m and 90 m DEMS contain much higher maximum elevation values (85 m and

61 m) in the study area, along with greater means and standard deviations. The higher

82

maximum values are an indicator of erroneous elevations. In addition, to the comparison
of ranges of values, the minimum values of the models are as negative as -32 m and -27
m in the case of the SRTM 30 m and SRTM 90 m DEMS. The negative elevation values
are unlikely because the lowest elevation in Louisiana is around 2 m below sea level near
New Orleans (MSN Encarta Encyclopedia 2008). A negative value would indicate the
elevation was below sea level.

Table 5.1. Entire Acadia study area Characteristics

 

 

 

 

 

 

 

Columns x Rows Min (m) Max (In) Mean (m) Std (m)
LiDAR 5 m 9660 x 8364 -6.1 22.1 8.9 3.9
LiDAR 90 m 537 x 465 -3.3 21.3 9.5 3.9
NED 30m 1618x1404 0.3 20.1 9.1 3.9
NED 90 m 539 x 468 1.1 20.7 9.7 3.9
SRTM 30 m 1617 x 1403 —32 85 11.8 5.1
SRTM 90 m 538 x 468 -27 61 12.3 4.8

 

 

 

 

 

 

 

 

In addition to the comparisons of elevation values, the slope (degrees) of the
DEMS can illustrate difference in the models (Table 5.2). In comparison of the LiDAR,
NED, and SRTM DEMS, a smoothing effect was observed in the resampling of their
original resolution to the 90 m resolution. For the LiDAR DEM its maximum slope was
46.3° at 5 m resolution, which was reduced to 45° for the LiDAR 90 m DEM. It is likely
the maximum of 463° was incorrect due to blunders in the LiDAR and was lost in the
resampling method. The NED 90 m and the LiDAR 90 m slope range were more
comparable than the LiDAR and SRTM ranges. Blunders (aka gross errors) are identified
as errors from the equipment used to create the DEM (Fisher and Tate 2006). However,
the SRTM 90 m slope range was much smaller than the SRTM 30 m resolution. These

observations of the differences in mean slope (i.e. 09° for the LiDAR 5 m, 03° for the

83

NED 30 m, and 06° for the SRTM 90 m) indicate the difficulty of stating the average
elevation of an area, since it can differ based on the model used.

Table 5.2. Slope (degrees) of entire Acadia study area DEMS

 

 

 

 

 

 

 

Columns x Rows Min Max Mean Std
LIDAR 5 m 9660 x 8364 0 46.3 1.0 1.6
LIDAR 90 m 537 x 465 0 4.5 0.2 0.3
NED 30 m 1618 x 1404 0 9.1 0.3 0.5
NED 90 m 539 x 468 0 3.1 0.2 0.3
SRTM 30 m 1617 x 1403 0 39.6 2.1 1.5
SRTM 90 m 538 x 468 0 10.9 0.6 0.7

 

 

 

 

 

 

 

 

In addition to reporting the descriptive statistics of the entire study area, the
sample descriptive statistics of the Acadia study area are reported in Tables 5.3 and 5.4.
From examination of the sample statistics, the SRTM DEMS have greater average
elevations (11.7 m for the SRTM 30 m and 12.3 m for the SRTM 90 m) compared to the
NED 30 m and 90 m (9.1 m and 9.6 m) and LiDAR 5 m and 90 m DEMS (8.7 m and 9.3
m). The average Slope values of the sample are comparable to the entire study area. The
standard deviations of the sample statistics are greater(4.0 m to 5.2 m) compared to the
standard deviations of the entire study area (3.9 m to 5.1 m) However, the standard
deviations of the SRTM 30 m and 90 m slope values are lower compared to the entire
area.

Table 5.3. Acadia study area Sample Elevation Characteristics (n = 1956)

 

 

 

 

 

 

 

 

LIDAR 5 m LiDAR NED 30 m NED 90 m SRTM SRTM
90 m 30 m 90 m

N of cases 1956 1956 1956 1956 1956 1956
Minimum 0.0 0.0 1.3 1.9 -4 0
Maximum 19.0 19.7 18.9 19.5 32 35
Range 19.0 19.7 18.9 19.5 36 35
Mean 8.7 9.3 9.1 9.6 11.7 12.3
Standard 4.0 4.1 3.9 4.0 5.2 5.0
Dev

 

 

 

 

 

 

 

84

 

Table 5.4. Acadia study area Sample Slope (degrees) Characteristics (n = 1956)

 

 

 

 

 

 

 

 

LIDAR 5 m LIDAR NED 30 m NED SRTM SRTM
90 m 90 m 30 m 90 m
N of cases 1956 1956 1956 1956 1956 1956
Minimum 0.0 0.0 0.0 0.0 0.0 0.0
Maximum 21.4 2.2 2.3 13.2 4.6 5.6
Range 21.4 2.2 2.3 13.2 4.6 5.6
Mean 0.9 0.2 0.2 2.0 0.6 0.3
Standard 1.6 0.3 0.3 1.4 0.6 0.5
Dev

 

 

 

 

 

 

 

 

5.1.1.1. Comparison of Elevation Values of LiDAR, NED, and SRTM

The pairwise correlation coefficients (r) of the LiDAR, NED, and SRTM DEMS
were investigated in SYSTAT to address research question 1: What are the relationships
among elevations of NED, LiDAR, and SRTM DEMS? It is expected that the original and
the resampled resolution data will be highly correlated given the nearest neighbor
method. The correlation values of the LIDAR, NED, and SRTM DEMS were determined
(Table 5.5) A high correlation exists between the NED 30 m and NED 90 m (r = 0.993)
and the LiDAR 5 m and LiDAR 90 m (r = 0.989) Also, a high correlation was noted
between the SRTM 90 m and the SRTM 30 m DEMS (r =0.936). The SRTM 90 m and
the SRTM 30 m showed much lower correlations with the LiDAR 5 m and 90 m (r =
0.633 and 0.625, respectively) and NED 30 m and NED 90 m DEMS (r =0.657 and
0.655).

Table 5.5. Pearson Correlation Matrix of Acadia study area Elevation Data (n =1956)

 

 

 

 

 

 

 

 

 

 

LIDAR 5 m LIDAR 90 m NED 30 m NED 90 m SRTM 30 m SRTM 90 m
LIDAR 5 m 1.000
LIDAR 90 m 0.989 1.000
NED 30 m 0.939 0.929 1.000
NED 90 m 0.941 0.933 0.993 1.000
SRTM 30 m 0.619 0.610 0.647 0.644 1.000
SRTM 90 m 0.633 0.625 0.657 0.655 0.936 1.000

 

 

 

 

 

 

85

5.1.1.2. Comparison of Slope Values

To determine how correlated the slope values of the datasets were in relation to
research question 2 (What are the relationships among NED, LIDAR, and SRTM
slopes?), a correlation matrix of the slopes of the original and resampled resolution
DEMS was generated (Table 5.6). It is expected that the slopes will not be highly
correlated for DEMS of the same production method, given the smoothing effects that
occurred during resampling. It is evident that the NED 90 m and NED 30 m have a high
correlation (r= 0.776). Medium correlations were found between the LiDAR 90 m and
the NED 30 m (r = 0.661) and the LiDAR 90 m and the NED 90 m (r= 0.752). Weak
correlations were found between the SRTM 30 m and SRTM 90 m DEMS (r =0.459),
along with between the SRTM 90 m and SRTM 30 m DEM and other DEMS. These
weak correlations occur because the resampling done to obtain new resolutions smoothed
the DEM and decreased the slopes. Generally, the slope values are skewed toward the
right. Figure 5.2 shows the histogram of the sample SRTM 90 m slope values.

Table 5.6. Pearson Correlation Matrix for Acadia study area Slope (n =1956)

 

 

 

 

 

 

 

 

 

LIDAR 5 m LIDAR 90 m NED 30 m NED 90 m SRTM 30 m SRTM 90 m
LIDAR 5 m 1.000
LIDAR 90 m 0.307 1.000
NED 30 m 0.332 0.661 1.000
NED 90 m 0.277 0.752 0.776 1.000
SRTM 30 m 0.067 0.120 0.124 0.128 1.000
SRTM 90 m 0.131 0.250 0.213 0.244 0.459 1.000

 

 

 

 

 

 

86

 

 

 

 

 

 

 

 

 

 

 

 

 

400 1 1 1 1 0.2
"l
320 :
n -o
e
240 _ ” o
E a
S — — o 1 8
160 — E
(D
e
80 e
0 Ln— 0.0
0 1 2 3 4 5
Slope (degrees)

 

Figure 5.2. Histogram of SRTM 90 m Slope (degrees) (n = 1956)
5.1.1.3. Error Analysis

The SRTM model was analyzed in relation to the LiDAR 90 m and NED 90 m
DEM. Table 5.7 summarizes the error descriptive statistics of the Acadia study area. The
minimum, maximum, mean, and standard deviation of the SRTM-LiDAR difference grid,
and the SRTM-NED difference grid produced very Similar results. Table 5.8 indicates the
sample SRTM error characteristics of the Acadia study area. The RMSE of the sample
characteristics of the SRTM-LiDAR DEM was 4.6 m and the SRTM-NED DEM was 4.6
m , which are comparable to the entire study area RMSE, 4.7 m and 4.6 m, respectively.
The main difference between the sample and the entire area is the range of the error
values. The range of the entire study area of the SRTM-LIDAR DEM error was -37.1 m
to 49.7 III, while the range of the sample was -5.7 m to 21.8 m. The sample range was
smaller and contained less extreme values compared to the entire area. Table 5.7 and
Table 5.8 display the means of the SRTM DEM error to be greater than zero, which

indicates an upward bias of the SRTM. Bias is defined as the over or underestimation of

87

elevation values of the DEM, which can be determined from the mean elevation error
being positive or negative (Fisher and Tate 2006). It is expected from findings of the
research literature, such as Carabajal and Harding 2006, that similar results will be seen
in this research.

Table 5.7. Entire Acadia study area Elevation Error Characteristics

 

 

 

 

Error grids Columns x Min Max Mean Std Dev RMSE
Rows (m) (m) (m) (m)
SRTM- LiDAR 537 x 465 -37.1 49.7 2.8 3.7 4.7
SRTM- NED 537 x 468 -35.8 49.7 2.7 3.7 4.6

 

 

 

 

 

 

 

 

Table. 5.8. Acadia study area Sample SRTM Error Characteristics (n = 1956)

 

Sample Min (m) Max (In) Mean (m) Std Dev (m) RMSE

 

SRTM- LiDAR 1956 -5.7 21.8 2.8 3.7 4.6

 

 

 

 

 

 

 

 

 

SRTM-NED 1956 -6.7 22.8 2.6 3.8 4.6

 

To determine whether the SRTM-LiDAR and the SRTM-NED means were
significantly different from each other (Figure 4.5), the difference of means t-test was
conducted. The null hypothesis is that the mean errors of each sample are equal. The
research hypothesis is that the means are not equal. The difference of means test of the
SRTM-NED and SRTM-LIDAR (n= 1956) indicated a significant t-value (t= -4.133,
p<<0.0001), thus the null hypothesis is rejected. This indicates that there is a Significant
difference between the mean values of using the LiDAR as reference to the SRTM versus
the NED DEM as reference. In order to see if the mean values of the SRTM-LiDAR
DEM error and the SRTM-NED DEM error are significantly greater than zero, a t-test
was done with the alternative hypothesis set to the mean greater than 0, and the null
hypothesis of a mean equal to zero. As Table 5.9 indicates the SRTM-LiDAR and

SRTM-NED average errors were significantly greater than zero (t=33.007, p <<0.0001;

88

t= 30.500, p <<0.0001). This supports my initial research hypothesis that the average

SRTM DEM error would be positive indicative of an upward bias of the SRTM DEM.

Table 5.9. Acadia study area: One-sided T-test Results

 

 

 

 

 

 

 

Mean T p-value
SRTM-LiDAR 2.8 3 3 .007 0.000
SRTM-NED 2.6 30.500 0.000

 

 

The correlation coefficient was calculated for the SRTM-NED error and SRTM-

LiDAR error. It was anticipated that their error values were highly correlated given the

general similarity observed in the error grids spatially. The correlation (r) between the

SRTM- LiDAR Error and the SRTM-NED Error was 0.907. This result suggests that

using either the NED 90 m or the LiDAR 90 In DEM will yield similar SRTM error

characteristics. The histograms of the SRTM-LiDAR Error and the SRTM-NED Error for

the Acadia study area indicate a bell-shaped distribution (Figure 5.3), skewed towards the

 

 

 

 

 

 

 

 

 

 

 

 

 

right.
333 I I I - 0.3 633 I I I 0.3
500 - . 500 - _
3 P
400 - - 0.2 g 400 - ‘ 0-2 8
~ 3 - 8
C —. C _.
8 300 - 8 g 300 g
0 '8 O 8
200- -0.1 g 200- -0.1 St
100 - 100 -
C 0.0 0 ' 0.0
-10 0 10 20 30 -10 O 10 20 30
SRTM- LIDAR Error SRTM- NED Error

 

 

Figure 5.3. Histogram Distribution of SRTM-LiDAR and SRTM-NED Error (n = 1956)

89

 

5.1.1.3.1. SRTM-LIDAR DEM Error

The similarity observed in the descriptive statistics of the SRTM-LIDAR DEM
error and the SRTM-NED DEM error was also evident from Figures 5.4 and 5.5 . The
LiDAR has yet to be incorporated in the NED for this area, so data sources are
independent. The root mean square error for the SRTM DEM model is 5 m, which
indicates the dispersion of error around the mean. The root mean square error (RMSE)
represents a global statistic. In the error grids, the very negative error values are
associated with voids in the data, which can be seen towards the southeastern area of the
Acadia study area.

In terms of the general error characteristics of the SRTM to LiDAR DEM, the
patterns of error follow land cover/ land use patterns present on the landscape (Figure
5.4). The symbology of the mean and standard deviations indicate green at locations
greater than the average (2.8 m for the SRTM-LIDAR DEM error and 2.7 m for the
SRTM-NED DEM error), and purple in locations lower than the average. In addition to
the identification of error patterns from the SRTM-LiDAR difference grid, a striated
pattern was located in the lower right comer of the region, and is interpreted as was
earlier, interpreted by Bhang, Schwartz, and Braun (2007) to be a result of the SAR
scanning procedure. Also, Bhang, Schwartz, and Braun (2007) attribute spikes in the

SRTM to the radar data used to create the SRTM DEM.

90

 

 

oowhw CON?

4:.

m6- - use.
QN. - hm.
ad - ad.
Qt - 06
v6 - Ev
NNF - md
«.9 - 0dr
230.2

08.9

ad ..er

hetm ¢<n=._ .Emm

oovd oomd o
I

e .6 Seize eaves» B 38..» 2.; .65 «(94-5% 2:

 

 

 

Figure 5.4. Acadia study area SRTM- LiDAR Error

91

This thesis research indicates that wetlands along the stream networks are areas of
high error in the region producing some distinct patterns. These included rectangular
level patches, which could possibly correspond to fields or forest patches. The developed
urban areas appeared to have low error magnitudes. The negative errors occurred
throughout the flatter agricultural lands in the area. More Specific discussion of the
characteristics of SRTM DEM error and land cover characteristics are in section 5.1.1.4.
5.1.1.3.2. SRTM-NED DEM Error

Examination of the metadata viewer from the USGS indicated some areas of
Louisiana have already incorporated LiDAR into the NED product (USGS 2007).
However, selected study areas does not have LiDAR incorporated into the NED product,
thus this is not the reason for the similar appearance of the SRTM DEM error (Figures
5.4 and 5.5)

Based on the results obtained from the use of the NED 90 m and LiDAR 90 m,
both of these products will yield similar error results when used as a reference to the
SRTM. This implies that the use of the LiDAR data in this case does not further enhance
knowledge of the SRTM error characteristics, compared to the use of the NED. This may
be specific to the type of LiDAR data or may apply more broadly based on other LiDAR
products. This research thus demonstrates that spatial resolution is a main driver of the
representation of the elevation of this area for the different production methods.

Given the high correlation found between the SRTM-LiDAR Error values and the
SRTM-NED Error values, the remainder of this chapter will include discussion and
figures from the SRTM-LiDAR Error analysis. Any major deviations between the

SRTM-LiDAR Error values and the SRTM-NED error results will be noted.

92

 

 

806" 03.9 ooodw

0.0. - ddn-
md. - No.
0.0 . ad-

of - 0.. 0d caus—
vd - Ev
ddw - md
ode - ndw

20a”:

..otm sz -Ehmw

8+6 03.0

o

.w +0 2.830329 Emucmfi an uovwcm we? Lotw szrstkm 9F

 

 

 

Figure 5.5. Acadia study area SRTM-NED Error

93

5.1.1.4. Error Relationships to National Land Cover Dataset (NLCD) 2001

The relationships of SRTM DEM error and land cover from the NLCD 2001 were
investigated to address research question 4: What is the relationship of SRTM error in
terms of different land cover/land use classes? Table 5.10 indicates the average SRTM-
LiDAR error and average SRTM-NED error for each reclassed land category: developed,
agriculture, and wetland. The RMSE was also calculated for each of these land class.
From the average errors of the different land categories it is evident the wetland class has
a higher SRTM DEM average error (8.5 m for the SRTM-LiDAR DEM error and 8.6 for
the SRTM-NED DEM Error) than the developed (3 m and 2.9 m) and agriculture (1.4 m
and 1.2 m) categories. The RMSE of the wetlands (9.5 m) is also greater than the RMSE
of the developed (3.9 m) and agriculture classes (2.5 m) for the SRTM-LiDAR DEM
error.

Table 5.10. Acadia study area: Sample Error Characteristics by Land Use/ Land Cover

 

 

 

 

 

Reclassified Class Sample SRTM-LIDAR SRTM-NED
Average RMSE Average RMSE
Error (m) Error (m)
I!!!) (m)
Developed 92 3.0 3.9 2.9 3.8
Agriculture/Barren 1529 1 .4 2.5 1 .2 2.3
Wetland 344 8.5 9.5 8.6 9.6

 

 

 

 

 

 

 

 

The following section discusses characteristics of the SRTM DEM error in terms
of landscape characteristics. To examine the SRTM error characteristics in relation to
land cover, errors in the sample were grouped by the reclassed NLCD 2001 categories.
The box plot of these groupings of SRTM-LIDAR error illustrates their median values as
well as their distributions (Figure 5.6). This indicates the generally higher median error (8

m) for woody wetland locations. The agricultural and urban areas indicated lower

94

elevation errors, 1 m and 3 m respectively. The main contrast seen between the woody
wetlands and agricultural lands was further qualitatively confirmed with the vectorization
of the NLCD data to a shapefile, which when overlaid on the error grid aligned

approximately with the boundary patterns of high and low error.

 

 

30 I I I I l

N
O
I
it
1

 

 

_x
O
|
)X 2
Ian: (HI 1 o
I

I——I

SRTM-LIDAR

 

 

 

 

 

 

 

 

l _

l l I l l
Developed Agriculture Wetland

 

 

 

 

 

 

Figure 5.6. Box plot of the SRTM-LiDAR Error and NLCD 2001. The wetland class
indicates a higher median error (8 m) compared to the developed (3 m) and agriculture
classes (1 m).

In order to investigate research question 5 (IS the average SRTM DEM error
greater for vegetated or non-vegetated areas?), a difference of means t-test was
conducted. To test the significance of the differences of means between forested wetlands

and non-forested landscapes, the sample was coded 0 for non—forest, and l for forest. The

non-forest class included the developed and agricultural locations, while the forest class

95

included the woody wetlands. The null hypothesis is the difference of means is equal to
zero, while the alternative hypothesis is that the means do not equal zero. The mean
SRTM-LiDAR error of the non-forested locations was 1.5 m, and the mean SRTM-
LiDAR error of the forested locations was 8.5 m, with a t-value of -30.23 8, which was
significant at the 95% confidence level (p << 0.0001).

To determine if the average SRTM-LiDAR DEM error of the vegetated class was
significantly greater than the SRTM-LiDAR DEM error of the non-vegetated classes, a
one-sided t-test was conducted. This t-test was set up using the SRTM-LiDAR DEM
error of the vegetated classes. The mean was set to 1.5 m (the SRTM-LiDAR mean of the
non-vegetated classes) and the alternative hypothesis was set up as greater than the 1.5 m.
This would indicate whether the average error of the vegetated classes was Si gnificantly
higher than the mean of the non-vegetated classes. The results indicated a t value of
31.450 (p <<0.0001). This result confirms findings indicating a greater SRTM-LiDAR
DEM error for the vegetated classes compared to the non-vegetated classes for the
Acadia study area.

The variances of the land cover/ land use classes of the NLCD 2001 were also
determined to examine differences across the classes (Table 5.11). In the Acadia study
area, the variance of wetlands (6.2 m) was greater due to the greater difference between
the mean value and the individual values. The wetland class had greater variance because
of the changes in canopy height above the surface of the earth contributing to the DEM
error. In contrast, the agriculture class had the least variance (1.4 m), since the crop
heights would remain relatively constant over an area. These specific land cover/land use

classes are investigated more in the next section.

96

Table 5.11. Acadia Study Area Variances of SRTM-LiDAR error by NLCD 2001 class

 

 

 

 

 

 

 

 

 

 

 

 

n ‘ Mean Variance
Developed 92 3.0 6.2
Agriculture 1523 1.4 4.0
Wetlands 341 8.5 17.3
F-ratio 190.702
p-value 0.000

 

5.1.1.5. Select Location Error Comparison

Select areas were identified in the SRTM-LiDAR error grid to examine certain
land cover/ land use relationships to DEM error. The letters (A, B, C, D) located on
Figure 5.4 identify four locations chosen to represent various locations in the study area
with different error characteristics. As discussed above, the highest error values occurred
throughout the wetland land cover class, which is represented by Location A. In this
wetland area, there are intertwined darker patches with lighter ones, which reflect a range
of error values indicative of different heights of plants within the wetland area. The
woody wetland land cover class adds height to the earth’s surface because of the tree
Canopy above the surface. The SRTM radar would have been reflected from the tree
canopy, which would have biased the elevations in the positive direction. Thus, this likely
contributed to the greater difference seen in relation this class and the SRTM DEM. In
the predominantly agricultural, non-wetland class (Location B) a relatively consistent
lower average error value is depicted. This is likely due to the shorter grasses and/or
mom on the landscape, which would not have contributed to the bias within the
collection of the SRTM data. Average error was observed in landscapes surrounding the

wetland areas, and along the eastern side of the study area where a striping pattern in the

97

DEM error analysis was present (Location C). The urban land cover class was typically
associated with an average error value and had similar characteristics to the agriculture
class (Location D). Generalizations about the characteristics of the SRTM DEM error
were ascertained by examining these locations in context with the entire study area.
5.1.2. Webster study area

The Webster study area contains slightly higher relief compared to the Acadia
study area. Figures B.7-B.12 of Appendix B portray the elevation and slope
characteristics of the Webster study area. These figures employ with the same color
scheme for comparative purposes. Similar to the Acadia study area, the SRTM 90 m
Webster study area elevation surface appeared to have rectangular patterns of similar
elevation. Rectangular and raised features are observed in the SRTM DEM towards the
middle and western side of the study area, which is likely created from the forest canopy
in the area. The eastern portion of the study area contained uplands, which were evident
in the SRTM DEM. The curvilinear feature of the Bayou Bodcau Reservoir is Observed
on the eastern side of the study area in the lower wetland area. The LiDAR 5 m DEM for
the Webster study provides a more detailed interpretation of the study area. The LiDAR 5
m DEM clearly indicates two main lower elevation areas, along with branching drainage
networks. The uplands on both the western and eastern borders of this study area are
identified in the LiDAR DEM. The NED DEM resembles the LIDAR DEM more than
the SRTM DEM.

In terms of elevation ranges, the Webster SRTM, LiDAR, and NED DEMS are
comparable (Table 5.12). The mean values of the LiDAR and NED values were also very

similar, which was also the case for the Acadia study area. Again, the main distinction

98

was between the LiDAR and NED DEMS, and the SRTM DEM for each study location.
As with the Acadia study area, slopes decreased with an increase in cell size (Table 5.13).
The histogram of the SRTM 90 m slope is skewed to the right, which shows a
predominance of low Slope values (Figure 5.7).

Table 5.12. Elevation Characteristics of entire Webster study area

 

 

 

 

 

 

 

 

 

 

 

 

 

Columns x Min Max (m) Mean (m) Std
rows (m)

LiDAR 5 m 7055 x 8344 44.1 138.1 69.3 12.1
LiDAR 90 m 877 x 585 44.9 135.3 69.9 12.0
NED 30 m 1243 x 1469 43.8 138.3 69.5 12.1
NED 90 m 392 x 464 44.4 135.84 70.1 12.1
SRTM 30 m 1176 x 1391 40 146 77.0 12.4
SRTM 90 m 392 x 464 43 144 77.0 12.4

 

 

Table 5.13. Slope (degrees) Characteristics of entire Webster study area

 

 

 

 

 

 

 

Columns x Min Max Mean Std
rows

LIDAR S m 7055 x 8344 0 47.8 2.3 2.4
LIDAR 90 m 877 x 585 0 10.1 1.0 1.0
NED 30m 1243x1469 0 21.1 1.5 1.7
NED 90 m 392 x 464 0 10.1 1.0 1.0
SRTM 30m 1176x1391 0 27.3 3.1 2.0
SRTM 90m 392x464 0 11.4 1.7 1.1

 

 

 

 

 

 

 

 

99

 

 

 

 

 

2

50 l I I l l | l _ 0.16

200 _ - 0.14
— 0.12 :9
8
150 — — 0.10 0
E EL.
3 O
8 — 0.08 g
100 Q
— 0.06 to
93

50 — 0.04

— 0.02

0 ‘H 0.00

0 1 2 3 4 5 6 7 8
Slope (degrees)

 

 

 

Figure 5.7. Histogram of the SRTM 90 m Slope (n = 1484). The Webster study area
slope histogram is skewed to the right and a majority of low-slope areas.

Tables 5.14 and 5.15 indicate the sample elevation and slope characteristics. The
means of the SRTM DEMS are higher compared to the LiDAR and NED DEMS, which
was also the pattern seen with the Acadia study area. The slope mean values of the DEM
were similar for the entire study area. The sample statistics give a more representative
picture of the Webster study area, since the sample points are independent of one another.

Table 5.14. Webster study area Sample Elevation Characteristics (n = 1484)

m m m

 

100

Table 5.15. Webster study area Sample Slope (degrees) Characteristics (n = 1484)

 

 

 

 

 

 

 

 

LiDAR 5 m LiDAR NED 30 m NED 90 SRTM SRTM
90 m m 30 m 90 m

N or cases 1484 1484 1484 1484 1484 1484
Minimum 0.0 0.0 0.0 0.0 0.0 0.0
Maximum 20.3 8.0 12.3 6.9 15.5 7.2
Range 20.3 8.0 12.3 6.9 15.5 7.2
Mean 2.3 0.9 1.1 1.0 3.1 1.6
Standard 2.4 1.0 1.5 1.0 1.9 1'1
Dev

 

 

 

 

 

 

 

 

5.1.2.1. Comparison of Elevation Values of LiDAR, NED, and SRTM

To determine how closely the elevation values of the LiDAR, NED, and SRTM
elevations correspond to each other, the correlation coefficients were calculated between
LIDAR, NED, and SRTM DEMS (Table 5.16). This aided in addressing research
question 1: What are the relationships among elevations of NED, LiDAR, and SRTM
DEMS? It was expected that the original and the resampled resolution data will be highly
correlated given the nearest neighbor method of resampling that was chosen to resample
the DEM from its original resolution to the resolution of the SRTM 90 m DEM. A high
correlation between the NED 30 m and NED 90 m (r =, and the LiDAR 5 m and LiDAR
90 m resolution was observed. However, in this case, the SRTM DEM was also
correlated with the other DEMS, which was in contrast to the Acadia study area. Similar
to the Acadia study area, the SRTM 30 m and the SRTM 90 m were strongly correlated
with an r equal to 0.976.

Table 5.16. Pearson Correlation Matrix for the Webster study area (n = 1484)

 

 

 

 

 

 

 

 

 

 

 

LIDAR 5 m LIDAR 90 m NED 30 m NED 90 m SRTM 30 m SRTM 90 m
LiDAR 5 m 1.000
LiDAR 90 m 0.896 1.000
NED 30 m 0.961 0.927 1.000
NED 90 m 0.961 0.927 0.956 1.000
SRTM 30 m 0.822 0.876 0.851 0.851 1.000
SRTM 90 m 0.814 0.873 0.846 0.849 0.976 1.000

 

 

 

 

 

101

5.1.2.2. Comparison of Slope Values

To determine how correlated are the slope values of the datasets, the correlation
matrix of the slope of the original and resampled resolution DEMS was created to answer
research question 2: What are the relationships among NED, LIDAR, and SRTM slopes?
(Table 5.17). The correlation coefficients of the slope grids are not highly correlated. This
was similar to the Acadia study area, which had low correlation coefficients for slope
values. The LiDAR 90 m and NED 90 m were the only slope grids with a correlation
above 0.7, which indicates the importance of resolution regardless of the production
method. This shows that despite the differences in the original production method of the
DEM, when it is coarsen by resampling, the elevation values are comparable.

Table 5.17. Pearson Correlation Matrix for the Webster study area Slope (n = 1484)

 

 

 

 

 

 

 

 

 

 

LIDAR 5 m LmAR 90 m NED 30 m NED 90 m SRTM 30 m SRTM 90 m
LiDAR 5 m 1.000
LiDAR 90 m 0.467 1.000
NED 30 m 0.025 0.044 1.000
NED 90 m 0.553 0.711 0.060 1.000
SRTM 30 m 0.168 0.170 -0.017 0.235 1.000
SRTM 90 m 0.231 0.277 0.006 0.394 0.464 1.000

 

 

 

 

 

 

5.1.2.3. Error Analysis

The SRTM model was analyzed in relation to the LiDAR 90 m and NED 90 m
DEM. Table 5.18 summarizes the error descriptive statistics of the Acadia study area.
The minimum, maximum, mean, and standard deviation of the SRTM-LiDAR difference
grid, and the SRTM-NED difference grid were very similar. The histograms of the
SRTM-LiDAR Error and the SRTM-NED Error for the Webster study area indicate a
bell-shaped distribution (Figure 5.8). Table 5.19 indicates the sample SRTM DEM error
characteristics, along with the RMSE. These results of the sample are reflective of the

RMSE calculated for the entire study area seen in Table 5.18. The minimum (-7.8 m) and

102

maximum (29.7 m) of the SRTM —LiDAR DEM error sample were less extreme
compared to the entire area minimum (-18.0 m) and maximum (38.2 m). The sample does
not contain as extreme points in the entire study area, which must have been either
erroneous points or from other data production issues. As indicated by Table 5.18, the
average SRTM DEM error is 7.1 m for the entire SRTM-LiDAR DEM error and 6.9m for
the SRTM-NED DEM entire area. These mean values are greater than zero, which
suggests my hypothesis regarding research question 3 (Is the average SRTM DEM error
greater than zero?) of a positive average error is valid. More statistical analysis follows
that supports this hypothesis.

Table 5.18. Error Analysis Results for Webster study area

 

 

 

 

 

 

 

 

 

 

 

Error grids Columns Min Max Mean Std Dev RMSE
x Rows (E!) (111) (mL (111)
SRTM- LiDAR 392 x 464 -18.0 38.2 7.1 5.5 9.0
SRTM- NED 392 x 464 -17.1 29.3 6.9 5.1 8.6

 

Table 5.19. Webster study area Sample SRTM Error Characteristics (n = 1484)

 

Sample Min (m) Max (m) Mean (m) Std Dev (m) RMSE

 

SRTM- LIDAR 1484 -7.8 29.7 7.2 5.6 9.1

 

 

 

 

 

 

 

 

 

SRTM-NED 1484 -6.2 24.1 6.7 5.1 8.6

 

103

 

 

 

 

 

 

 

 

 

 

 

20 I I I 20 I I I

- 0.12 . - 0.12
150- 50.10 3 150- -o.1o-n
g 8
E - 0.08 o « - 0.088
3 100- 3' 8100— E“-
0 - 0.06 S o - 0.06 3
a 8
.. m _ a
50_ 0.04 e, 50- 0.04 g
- 0.02 - 0.02 "

‘31 o o 10 20 30"00 L310 0 1o 20 3800

SRTM-LIDAR Error SRTM-NED Error

 

Figure 5.8. Histograms of the Sample Points (n =1484).
5.1.2.3.1. SRTM-LIDAR DEM Error

The Webster study area SRTM-LiDAR DEM error grid contains large positive
error values (15.4 m -38.2 m) throughout the western portion of the study area (Figure
5.9). This area of higher DEM error corresponds with the woody wetland land cover on
the landscape. Generally, through the middle portion of the study area, there was a patchy
pattern of high (9.9 m - 15.4 m) and low error (-1 .2 m - 4.3 m) present on the landscape.
Towards the eastern side of the study area an alternating pattern of high and low error
patterns were present on the upland areas, which is likely due to the high frequency
changes in slope directions in this area. However, the uplands also contain groups of high
locations (15.4 m- 38.2 m) of error likely from the forests present on the landscape.

For the sample of 1484 cells, the difference of means test of the SRTM-NED and
SRTM-LIDAR was done, where the null hypothesis is that the means of each error set
was equal given independent samples. The difference of means indicated a significant
result (t = 3.809, p <<0.0001). The Pearson’s correlation coefficient was calculated to
see how correlated were the SRTM-NED and SRTM-LiDAR values. It was anticipated

that their error values were highly correlated given the general spatial similarity observed

104

in the error grid. The correlation (r) between the SRTM- LiDAR Error and the SRTM-
NED Error was 0.872 (n= 1484). This indicates there is a strong correlation between the
SRTM- LiDAR error and SRTM —NED error for the Webster study area, which was
similar to the Acadia study area. However, this correlation is slightly weaker for the
Webster study area compared to the Acadia study area.

In order to see if the mean values of the SRTM-LiDAR DEM error and the
SRTM-NED DEM error are significantly greater than zero, a t-test was done with the
alternative hypothesis set to the mean greater than 0, and the null hypothesis of a mean
equal to zero. As Table 5.20 indicates the SRTM-LiDAR and SRTM-NED average errors
were significantly greater than zero (t=49.498, p <<0.0001; t= 51.416, p <<0.0001). This
supports my initial research hypothesis that the average SRTM DEM error would be
positive indicative of an upward bias of the SRTM DEM. Similar results were found with
the Acadia study area.

Table 5.20. Webster study area: One-sided T-test Results

 

 

 

 

 

 

 

Mean T p-value
SRTM-LiDAR 7.2 49.498 0.000
SRTM-NED 6.7 51.416 0.000

 

 

105

 

13:} .
,Q

.‘ -‘ V ... . )-
‘ .7. .1 . . ~‘. f .
.‘ .. .l‘.' “TH‘ , 4.
‘1 I " A - I
c , .i“‘. . k .\ ‘.
. A sir-Th

K

 

 

 

15.5 - 38.2 N
10.0- 15.4
Mean 71 4.4- 93 0 2.700 5,400 10,811) 16.2@
-1.3 - 4.3
5.8 - -1.2
48.0 - -6.7

 

 

 

Figure 5.9. Webster study area SRTM —LiDAR Error

106

5.1.2.3.2. SRTM-NED DEM Error

Figure 5.10 of the SRTM-NED Error illustrates similar characteristics of the
SRTM-LiDAR DEM Error. Again, the USGS has yet to incorporate LiDAR data into the
NED DEM product in this area as of July 2007. Similar to the Acadia study area, the
SRTM-NED error and the SRTM-LiDAR error results were very similar. This again
suggests the LiDAR 90 m and the NED 90 m data contribute to similar error patterns
with the SRTM DEM. The high correspondence of elevation values of the NED 90 m
and LiDAR 90 m was also indicated with a high correlation of 0.93 (Table 5 .16).

Given the high correlation found between the SRTM-LiDAR Error values and the
SRTM-NED Error values, this section will include discussion and figures from the
SRTM-LiDAR Error analysis. Any major deviations between the SRTM-LiDAR Error

values and the SRTM-NED error results will be noted.

107

 

 

15.5 - 29.3 N
10.0 - 154

Mean 6,9 4.4 - 9,9 0 2550 5,100 10.200 15.300
‘1 3.43 — — 1Melers

6.9 - -1.2
-17.1 - -6.8

 

 

 

Figure 5.10. Webster study area: SRTM- NED Error

108

5.1.2.4. Error Relationships to National Land Cover Dataset (N LCD) 2001

To examine the SRTM error characteristics in relation to land cover
characteristics of the NLCD 2001, the land cover codes were reclassed to sequential
integers. This addresses research question 4: What is the relationship of SRTM error in
terms of different land cover/land use classes (i.e. developed, forested, agriculture,
water)? The relationship between SRTM DEM error and different land cover classes can
be seen in the box plot of the SRTM-LIDAR grouped by land cover (Figure 5.11). This
indicates the generally higher median errors for forested locations, including the woody
wetlands (10.1 m). Also, the median values for the deciduous (7.1m) and evergreen forest
(8.3 m) land covers are greater than the non-forest classes, such as agriculture (0.8 m).
This indicates that the areas where forests are present are prone to higher median errors
compared to areas without the presence of forests. From this type of analysis, a clear
distinction can not be made between the evergreen and deciduous forest land cover with
the use of the NLCD data.

The average SRTM DEM error was also calculated per land cover/land use class,
along with the RMSE (Table 5.21). This reiterates the patterns identified from the box
plots (Figure 5.4). The wetland class has the greatest average SRTM-LiDAR error (10.0
m) and SRTM-NED error (9.8 m), along with the greatest RMSE, 11.1 m and 10.9 m,
respectively. The second greatest average SRTM DEM error land cover/land use class
was the evergreen forest (8.5 m), then this was followed by the deciduous class (7.9 m).
The agriculture/barren class had the lowest SRTM DEM error (1.6 m) and a small RMSE

of 4.3 m compared to wetland class. However, the developed class was determined to

109

have the least RMSE value of 1.4 m for the SRTM-LiDAR error and 2.0 m for the
SRTM-NED error, which indicates the least Spread of error values.

Table 5.21. Webster study area: Sample Error Characteristics by Land Use/ Land Cover

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Reclassified Class Sample SRTM-LIDAR SRTM-NED
Average RMSE Average RMSE
Error (m) Error (m)
(m) (In)
Water 21 4.4 7.6 4.5 6.8
Developed 38 4.2 1.4 3.4 2 .0
Evergreen Forest 922 8.5 10.0 8.2 9.5
Deciduous Forest 76 8.0 9.4 7.6 8.8
Mixed Forest 134 7.2 9.2 7.4 8.9
Agriculture/Barren 79 1 .6 4.4 1.7 4.0
Wetland 211 10.0 11.1 9.8 10.9
30 I I | l f I
*1.
T r
)F
20 — I —
+.
a: x
(
Q
...1 1O — - —
2
j_
0:
“’ Q
0
5i< 1’ —
_10 I I l I 1 1
a Q «0 "og
\, 5’ e- b a, .9
Q Q N. O x. O S. \ta
8 3.9 Aggsgsfio s o
a 6’ e ,9 S
0° ‘0 o ‘5

 

 

 

Figure 5.11. Box plot of the SRTM-LiDAR Error and NLCD 2001

110

A difference of means t-test was performed to investigate research question 5: Is
the average SRTM DEM error greater for vegetated or non-vegetated areas? To test the
significance of the differences of means between forested and non-forested landscapes,
the sample was coded 0 for non-forest, and 1 for forest. The non-forest classes included
water, developed, and agriculture/barren. The forested classes were deciduous forest,
evergreen forest, mixed forest, and woody wetland. Given that the shrub/scrub class
includes vegetation at unknown heights, these sample points were not used in the
difference of means for the non-forest and forest classes. It is expected that the non-forest
and forested cells will have significantly different SRTM- LiDAR DEM error means. The
mean error of the non-forested class was 4.7 m and the mean error of the forested class
was 8.2 m, with a t-value of -7.7, which was significant at the 95 % confidence level (p
<< 0.0001). This indicates that the SRTM- LiDAR error is significantly different for
non-forest and forest classes. The positive average error indicates the SRTM has positive
elevation bias. Also, this result suggests that the forested area had a greater mean SRTM
DEM error, which supports my hypothesis regarding the average error of vegetated areas
being greater than the average error for non-vegetated areas.

To determine if the average SRTM-LiDAR DEM error of the vegetated class was
significantly greater than the SRTM-LiDAR DEM error of the non-vegetated classes, a
one-sided t-test was conducted. This t-test was set up using the SRTM-LiDAR DEM
error of the vegetated classes. The mean was set to 4.5 m (the SRTM-LiDAR mean of the
non-vegetated classes) and the alternative hypothesis was set up as greater than the 4.5.
This would indicate whether the average error of the vegetated classes was significantly

higher than the mean of the non-vegetated classes. The results indicated a t value of

111

19.913 (p <<0.0001). This result confirms findings indicating a greater SRTM-LiDAR
DEM error for the vegetated classes compared to the non-vegetated classes for the
Webster study area. Similar findings were found for the Acadia study area.

The variances of the land cover/ land use classes were also investigated for the
Webster study area (Table 5.22). Similar to the Acadia study area, in the Webster study
area, the agriculture class had a relatively low variance compared to the other classes.
The deciduous class did have a slightly lower variance. The evergreen, forests/shrub,
wetland, and deciduous forest classes all had similar variances. These ranged from the
wetland variance of 24.2 to the evergreen variance of 27.9. To account for the greater
variances in the vegetated classes, I suspect the evergreen class and other forest classes of
the NLCD 2001 are at different stages of succession. This would attribute to different
heights of the evergreen trees in an area identified as evergreen and have greater SRTM-
LiDAR variance for this class. In the Webster study area, the water had the greatest
variance. This could be attributed to the SRTM scatter from the water surface. These
characteristics will be fiirther investigated in the next section.

Table 5.22. Webster Study Area Variances of SRTM-LiDAR error by NLCD 2001 class

 

 

 

 

 

 

 

 

 

 

 

 

 

N Mean Variance
Webster
Water 23 4. 1 37.6
Developed 38 4.4 1 5 .4
Evergreen Forest 740 7.9 27.9
Deciduous Forest 82 7.8 24.8
Forest/ Shrub 300 5.4 31 .7
Agriculture/ 86 1 .7 16.3
Barren
Wetland 215 9.9 24.2
F-ratio 2.939
p-value 0.007

 

 

 

 

 

112

5.1.2.5. Specific Locations Error Comparison

The SRTM-LIDAR error for the Webster study area contains letters that
correspond to selected areas representative of the variety of error characteristics (Figure
5.9). Location A is indicative of the forested area, which dominates the study area and is
situated toward the center of the study area and corresponds with higher SRTM-LiDAR
error. The trees on the landscape add height to the surface of the earth, which is indicated
by the higher values in the forested areas. This is evident by the larger error values
throughout this section. However, in the bayous in the study area, the SRTM errors are
lower than average (Location B). In the eastern uplands, there are alternating patterns of
high and low error, which may be attributed more to changes in slope than to
characteristics of land cover (Location C). A similar alternating pattern was identified in
the Acadia study area. Throughout the uplands in this area, there are patches of high and
low error values. There are rectangular patterns of neighboring high and low error
situated in the southern, central part of the study area, which indicate that land cover
features, such as forests on the landscape, may be playing a role in creating these patterns
(Location D).

The Louisiana study areas represented areas in the United States rich with
elevation data. These locations were able to further illustrate characteristics of the SRTM
DEM given the LiDAR dataset as a reference. However, as seen from the literature,
LiDAR data and similar products are not readily available at the international scale. Thus,
the Nang Rong study area in Thailand will illustrate how a location with SRTM C-band

data was analyzed in relation to the SRTM X-band with a 25 m resolution.

113

5.2. Nang Rong, Thailand

The elevation ranges and characteristics of the SRTM DEMS of the Nang Rong
study area in Thailand indicate similar patterns when shaded with the same color scheme
(Figure 5.12). This Nang Rong study area is defined based on the availability of the
SRTM X-band data. The SRTM C-90 m and SRTM X-90 m indicate three main stream
valleys through the center of the region, flowing from higher elevations in the southern
part of the study area northwards. The higher mountain area is also still captured with the
X-band data at the 25 m and 90 m resolution. Similarly in the SRTM X-25 m these
features are identified in the DEM. This study area contains generally low elevations
(mean value 186 m), with a few locations of higher relief. Visually, the SRTM C-90 m,
SRTM X-25 m, and the SRTM X-90 m look very similar with only subtle differences.
Table 5.23 summarizes the elevation characteristics of the Nang Rong study area.

Table 5.23. Nang Rong study area DEM Characteristics

 

 

 

 

 

 

 

 

SRTM C-90 m SRTM X-25 m SRTM X-90 m
Rows x Columns 387 x 499 1374 x 1771 387 x 499
Cell Size 90 25 90
Min (m) 149 29 120
Max (m) 382 387 383
Mean (m) 186.6 185.7 186.0
Std Dev (m) 14.9 15.5 15.3

 

 

 

 

Table 5.24. Nang Rong Sample Elevation Characteristics (n = 153 7)

 

 

 

 

 

 

 

 

 

 

SRTM C-90 m SRTM X-25 m SRTM X-90 m
N of cases 1451 1451 1451
Minimum (9) 163.0 144.0 163.0
Maximum (m) 353.0 347.0 324.0
Mean (m) 186.6 185.1 185.8
Standard Dev (m) 14.6 17.2 14.6

 

ll4

 

 

Table 5.25. Nang Rong study area Slope (degrees) Characteristics

 

 

 

 

 

 

 

 

SRTM C-90 m SRTM X-25 m SRTM X-90 m

Rows x Columns 387 x 499 1374 x 1771 387 x 499
Cell Size 90 25 90

Min 0.0 0.0 0.0

Max 21.3 72.4 22.3
Mean 1.0 4.0 1.0
Standard 0.9 3.1 1.0
Deviation

 

 

 

 

Table 5.26. Nang Rong study area Slope Characteristics: Sample (11 = 1537)

 

 

 

 

 

 

 

 

SRTM C-90 m SRTM X-25 m SRTM X-90 m
N of cases 1451 1451 1451
Minimum 0.0 0.0 0.0
Maximum 9.7 45.8 1 1.4
Mean 0.9 4.0 0.9
Standard Dev 0.8 2.9 0.9

 

 

 

 

 

The Nang Rong slope characteristics indicate the generally level slope topography
of this area, with a mean slope of less than 5° for all DEMS (Table 5.24). Figure 5.13
suggests that distinctions can be made between the SRTM C-90 m and SRTM X-25 m
slope. The SRTM C-90 m indicates a generally flat study area with slopes less than 28°.
However, the SRTM X-25 m DEM is characterized by higher slopes (7.0 °-22.3 °)
throughout most of the study area. However, these higher slopes are lost with the increase
in resolution to the SRTM X-90 m. The summary slope statistics for the SRTM X-90 m
generally resembles those of the SRTM C-90 m. However, with the SRTM X-90 m there
are a few locations of greater slope in the 7.0 °-22.3 ° range, which could be due to the
processing of the DEM or other artifacts of the data. Tables 5.25 and 5.26 indicate the ‘

sample characteristics of the Nang Rong study area. These tables indicate the average

115

means are similar to the means found for the DEMS and slope grids of the entire study

area.

5.2.1. Comparison of Elevation Values of SRTM C-band and X-band

To determine how correlated the elevation values of the Nang Rong DEMS were
to answer research question 1: What are the relationships among elevations of NED,
LiDAR, and SRTM DEMS? The correlation coefficients were calculated between the
different elevation models (Table 5.27). It is expected the SRTM 25 m and the SRTM X-
90 m will be similar given the resampling. It is expected that the original and the
resampled resolution data will be highly correlated. The SRTM C-90 m was highly
correlated with the SRTM X-90 m (0.977) and the SRTM X-25 m (0.713). The resolution
seems to play a greater role than the data product based on the high correlations found
between the SRTM C-90 m and SRTM X-90 m DEMS.

Table 5.27. Pearson Correlation Matrix of the Nang Rong study area Elevation values

 

 

 

 

 

SRTM C-90 m SRTM X-25 m SRTM X-90 m
SRTM C-90 m 1.000
SRTM X-25 m 0.724 1.000
SRTM X-90 m 0.977 0.713 1.000

 

 

 

 

116

 

 

E8 653* 2.55

 

m: -mm
572%
8.-me
mm— -5—
~34»?

:5 co_uo>o_m

Emu 659x $2.2m E8 3259.0 2.5—w

om man 9 wk mud n

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 5.12. Elevation Characteristics of the Nang Rong study area

117

 

EEmEQEHHI
8 m mm m. o L. E. m a

53 newsflx Ema

 

a. .o ..
mm; _
a m - m N
g4:
78.5.3235

 

 

 

 

 

Eco uc~4x 2.5—m

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 5.13. Slope (degrees) of the Nang Rong study area

118

5.2.2. Comparison of Slope Values

To determine how correlated the slope values of the datasets were to address
research question 2: What are the relationships among NED, LIDAR, and SRTM slopes?
A correlation matrix of the slope of the DEMS was created to answer this question (Table
5.28). From this table it is evident that the SRTM C-90 m and SRTM X-25 m have weak
correlations (0.136), although the SRTM X-90 m and SRTM C-90 m are more similar
(0.517). This indicates the importance of the spatial resolution rather than the data
product. The histogram of slope values indicates the sample points were generally low in
slope with slopes less than 10 ° (Figure 5.14). Thus, as in the Louisiana study area, the
relationships between SRT M DEM error and slope can not be clearly related because of
the lack of much variation in topography.

Table 5.28. Pearson Correlation Matrix of Slope Values of the Nang Rong Site (n =1537)

 

 

 

 

 

 

SRTM C-90 m SRTM X-25 m SRTM X-90 m
SRTM C-90 m 1.000
SRTM X-25 m 0.136 1.000
SRTM X-90 m 0.517 0.276 1.000

 

 

 

 

ll9

 

 

800 I I 1' l I I I l I

500* A
_ 0.3 13
400“ .8
e — a
8 300— 40.2 g'
Q _. 'U
_ (D
200- ED
‘01 9"

 

 

 

 

 

 

m! IL.
0 1_ L1 15L [—

0123456789180
SRTMC-Band

 

 

 

 

Figure 5.14. Histogram of SRTM C-band Slope Values (n =1537)
5.2.3. Error Analysis: SRTM C-band — SRTM X-band

The error patterns of the SRTM C-band for the Nang Rong study area was
investigated through the difference between the SRTM C-90 m and the SRTM X-90 m
DEM (Figure 5.15). The range of SRTM error was —79 to 67 m with a mean of 0.7 m (3.2
m standard deviation). The RMSE was 3.3 m. The greater range of the SRTM error is
spread over a mainly flat topographic landscape, with some steeper slopes in the
southeastern part of the study area. Figure 5.15 indicates the histogram of the sample
SRTM error points, which had a generally bell-shaped in distribution. Table 5.29
summarizes the Nang Rong sample characteristics and the RMSE. The RMSE was
comparable to the RMSE found for the entire study area. The range of the sample was
less compared to the entire Nang Rong area given the sample accounted for spatial
autocorrelation. As shown in Table 5.27, the mean was 0.7 m, which indicates the

average SRTM DEM error is only slightly greater than zero. Thus, in relation to research

120

3 regarding the average SRTM DEM error, the differences between the SRTM X-90 m
and the SRTM C-90 m are only slightly positive.

Table 5.29. Nang Rong study area Sample DEM Error Characteristics

 

Sample Min (m) Max (m) Mean (m) Std Dev (m) RMSE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

SRTM Error 1537 -23.0 35.0 0.7 3.2 3.2
700 I I I I I I
600 - ‘ 0-4
500 —— a 1?
~ 0.3 g
.. 400— 3
C _.
8 _ — 8
0 300 — —o.2 g
a,
200 —- 2
— 0.1
100 — J ,
o 1 ~ L 4 o 0
-3o -20 -1o 0 1o 20 30 4o
SRTM 0 Band- SRTM x- Band

 

 

 

Figure 5.15. Histogram of the SRTM Error

In order to see if the mean value of the SRTM C-90 m —SRTM X-90 m error was
significantly greater than zero, a t-test was done with the alternative hypothesis set to the
mean greater than 0, and the null hypothesis of a mean equal to zero. As Table 5.30
indicates the SRTM DEM average error was significantly greater than zero (t=9.135, p
<<0.0001). This supports my initial research hypothesis that the average SRTM DEM
error would be positive indicative of an upward bias of the SRTM DEM. Similar results

were found with the Acadia and Webster study areas.

121

Table 5.30. Nang Rong Study Area: One-sided T-test Results

 

Mean T p-value
SRTM C-90 m-SRTM X-90 m 0.735 9.135 0.000

 

 

 

 

 

 

 

5.2.4. Specific Locations Error Comparison

The error characteristics identified fi'om Figure 5.16 indicate that the majority of
the Nang Rong study area had low negative error. The letters (A, B, C) identified in
Figure 5.16. correspond with specific locations discussed here. There is a slight diagonal
pattern of greater DEM error throughout the flatter topographic region (Location A). This
is unlikely from land cover/land use characteristics, and may be from production of the
DEMS. However, low error values were found throughout this area of flat topography
towards the western part of the study area (Location B). Towards the southeastern portion
of the study area where the inactive volcanic areas are present, greater positive error (4.1-
69.0 m) is identified for this area (Location C). This is likely from the steeper slopes in

this area rather than land cover/land use characteristics.

122

 

 

Mean 0.7—

-79 - -2.0

0 2,300 4.600 9.200 13.800 10,400
—;

 

 

 

Figure 5.16. SRTM C-band —SRTM X-band Error
5.3. Comparative Analysis: Louisiana study areas and Nang Rong, Thailand
MODIS
Since the SRTM C-90 m DEM is available for 80 % of the earth, an investigation
into land cover relationships was done across all study areas in Louisiana and Thailand.

This analysis with the MODIS tree cover aids in answering research question 5: Is the

123

average SRTM DEM error greater for vegetated or non-vegetated areas? MODIS data for
the land and water were acquired from the Terra and Aqua spacecraft (GLCF 2007). The
500 m resolution of MODIS is a larger spatial resolution compared to the NLCD data
used primarily in the United States. However, since the MODIS data were created from
satellite imagery, it is available for much larger extents, which will be available for all
study areas. Thus, its greater coverage enables its use in wider ranges of study areas.
However, it needs to be recognized that the coarser resolution provides a broader
understanding of the landscape in contrast to higher resolution land cover products. The
ranges of MODIS data varied across study areas, which impacted the categories used per
study area (Table 5.31). The lower values for the percent tree cover MODIS data in the
Acadia study area correspond with agricultural fields and non-wetland classes. The
wetland areas have higher values of percent forest cover in both study regions (Figure
5.17). For the Nang Rong study area, the higher percentages of tree cover correspond
with steeper elevation and smaller patches throughout the area.

Table 5.31. Sample Characteristics of MODIS Data

 

Acadia Webster N ang Ron&
N of cases 1956 1484 1537
Minimum 0.000 0.000 0.000
Maximum 100.00 86.000 45.000

Range 100.00 86.000 45 .000
Mean 19.936 53.486 7.962
Standard Dev 20.170 14.877 7.150

 

 

 

 

 

 

 

 

 

 

 

 

The box plots for the Acadia and Webster study areas indicate that areas with
MODIS tree cover greater than 2 50 %, the median error value was 7.4 % and 7.5 %,

which were greater than the median of the areas less then 50 % (1.4 % and 5.6 %) (Figure

124

5.18). A greater distinction can be made between the more and less forested areas in the
Acadia study area compared to the Webster study area. For the Nang Rong study area,
the range of percent forest cover from the sample was from 0 to 45 % (mean 8.0 %
standard deviation 7 .2). Thus, the Nang Rong study area could not be partitioned at 50 %,
as were the Louisiana study areas. For the Nang Rong study area, the MODIS tree cover
percentages were reclassed at 22.5%. From this classification, no clear relationship can
be determined from the relationship between MODIS tree cover and SRTM error
relationships. As the box plot in Figure 5.19 indicates, the median error remains near 0
for both classes of MODIS tree cover, 1.0m for the lower than 22.5 % and 0.0 for the

greater than 22.5 % class.

 

Acadia Study Area Webster Study Area

 

% Tree Cover

101-253
72- 100
58-71
46-57
34-45
13- 33
0-12

     

 

 

 

 

 

 

Figure 5.17. MODIS % Tree Cover for Acadia and Webster Study Areas

125

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Acadia Study Area
30 l I l T I
3:
n:
g 20 - -
.9
2
E 10 - -
(n
or I J -
_10 l 1 l 1 L
% Tree % Tree
< 50 2 50
Webster Study Area
30 I I I I I I
it: 20 - ° " -
Q
—.'
E 10- —
D:
(n
o_ I -
4:
_1 0 1 1 I l 1 l L
% Tree % Tree
< 50 2 50

 

 

 

Figure 5.18. Box Plots of the Acadia and Webster Study Area: MODIS % Tree Cover

126

 

Nang Rong Study Area

 

A
O

 

 

 

 

 

 

30"
20"
10-
0- 5% PELI
-10- 4‘
-20""
l I l
'30 % Tree % Tree
<22.5 2 22.5
': .r.- ‘
" .: I . j.
n u:- E..- I.

 

 

[:3 0-22.4

 

 

Figure 5.19. Nang Rong study area: MODIS: Percent Tree Cover

127

 

 

 

The difference of means t-test was also used to test for significant differences
between the means of the percent tree cover classes for each study area. The null
hypothesis is that the means are equal for both percent tree cover categories. The research
hypothesis is that the mean percent tree cover is different from each other. The results for
the Louisiana study areas indicate significant differences between the percent tree
categories. For the Acadia study area (11 = 1956), for the class of percent tree cover < 50
% had a SRTM-LiDAR mean of2.1 m and a class on 50 % had a mean of 7.5 m
(t= -17.085, p << 0.0001). For the Webster study area, for the class of percent tree cover
< 50 % had a SRTM-LiDAR mean of 4.6 m and a class of Z 50 % had a mean of 7.5 m
(t= -6.897, p<< 0.0001). However, for the Nang Rong study area, the difference of means
t-test had insignificant results. The class of percent tree cover < 22.5 % had a SRTM
DEM error mean of 0.8 m and the class of 2 22.5 % had a mean of —0.4 (t= 1.085, p=
0.283). Thus, the p value of the Nang Rong study area indicates the means of the two
percent tree groups are not significantly different from each other.

To determine if the higher MODIS grouping of SRTM-LiDAR DEM error was
significantly higher than the lower MODIS grouping, a one-sided t-test was conducted.
For the Acadia study area, the mean was set to 2.1, the mean of the SRTM-LiDAR DEM
error of the lower (<50 % Tree Cover), and the alternative hypothesis was set to greater
than this mean. The results indicated a t value of 17.497 (p <<0.000). This indicates that
the mean of the SRTM-DEM LiDAR DEM error of the locations with MODIS % Tree
Cover 2 50 % was significantly higher than the mean of the locations with MODIS %
Tree Cover < 50 %. Similarly for the Webster study area, a one-sided t-test was set up

with the mean equal to 4.6 m, and the alternative hypothesis set to be greater than this

128

mean of the MODIS % Tree Cover < 50 % SRTM-LiDAR DEM error. The results of the
Webster study area indicated the SRTM-LiDAR error of the MODIS % Tree Cover 2 50
% was significantly different (t =17.985, p << 0.0001) from the mean of the locations
will less tree cover, < 50 %. However, the findings that the locations with higher percent
tree cover in the Acadia and Webster study areas contrasted with the results of the Nang
Rong study area. For the Nang Rong study area, the SRTM C-90m —SRTM X-90m DEM
error was set to the mean of 0.8 m, the mean of the lower MODIS % Tree Cover class of
< 22.5 %, and the alternative hypothesis was set to greater than this mean. Similar to the
difference of means results of the MODIS classes, one-sided t-test for the Nang Rong
study area indicated insignificant results (t= -1.110, p= 0.864). Thus, the MODIS classes
for the Nang Rong study area does not indicate that the areas with greater tree cover have
significantly higher DEM error compared to locations with less tree cover.

The variances of the MODIS groupings of the study areas was investigated also to
determine whether these differed significantly based on the percent tree cover classes I
chose in this research (Table 5.32) For the Acadia study area, the variance between the <
50 % tree cover and Z 50 ”/0 tree cover was found to be significant (F-ratio = 0.376,
p<<0.0001). This shows the variance of areas with more vegetation is significantly
different from areas with less vegetation. This could be from the differences in tree
heights contributing to a wider range of SRTM DEM error. However, with the Webster
study area, the difference of variances across the MODIS groups (% Tree Cover <50 %
and % Tree Cover 2 50 %) was found to be insignificant (F-ratio =0.991, p= 0.910). I
suspect this was found because the Webster study area is heavily forested, a break at 50

% did not make a significant difference in terms of vegetation characteristics and tree

129

height differences on the surface. In the Nang Rong study area, the variances of the
MODIS % Tree Cover < 22.5 % and MODIS % Tree Cover 2 22.5 % groups were found
to be significantly different (F-ratio= 0.130, p <<0.0001). Similar as I noted above with
the Acadia study area, the differences in the variances of the MODIS groupings could be
attributed to the greater difference seen in the more vegetated class from the various
heights of the canopy on the surface. However, with the Nang Rong study area, the
SRTM was in reference to the SRTM X-90 m DEM.

Table 5.32. Variances of the SRTM Error by MODIS % Tree Cover Groupings Per

 

 

 

 

 

 

 

 

 

 

 

 

 

Study Area.

N Mean Variance F -ratio p-value
Acadia Study Area
°/o Tree Cover <50 % 1711 2.1 8.7 0.376 0.000
% Tree Cover 2 50 % 245 7.5 23.2
Webster Study Area
% Tree Cover <50 °/o 546 6.0 30.1 0.991 0.910
°/o Tree Cover 2 50 ”/0 938 7.8 30.4
Nang Rong Study Area
% Tree Cover < 22.5 1490 0.8 8.2 0.130 0.000
%
% Tree Cover 2 22.5 51 -0.4 62.8
%

 

 

 

 

 

 

 

For the Acadia and Webster study areas, the SRTM-LiDAR error was used in the above
table. The Nang Rong study area used the SRTMC-90 m —SRTM X-90 m DEM error.
The Louisiana study areas and the Nang Rong study area illustrate different
outcomes from examination of the MODIS data because of the different land- cover
characteristics of these areas (Figure 5.18). The data show the Thailand area with
significantly less percent forest cover, which is expected given other knowledge about the
agriculture base. In contrast with the Webster study area in Louisiana, where a large

portion of the study area is identified in multiple sources as a national forest, this was

130

indicated as forests. The Acadia study area was not predominantly forest, so the percent
tree cover was less of an indicator of land cover characteristics for this area. However,
this use of the MODIS percent tree data was an example of how a global dataset could be
used across study areas on different continents to explore relationships to the SRTM
DEM. As significant as using a global dataset to compare properties across study areas, a
land use application was created to analyze across study areas.
5.4. Land Use Application
This section summarizes how the LiDAR, NED, and SRTM DEMS perform to

address research question 6: How will the LiDAR, NED, and SRTM DEMS perform in a
land use application? The suitability model created different outcomes based on the
production and resolution of the elevation model. Of the DEMs used in the application,
only the Nang Rong study area varied the resolution. The identification of flat areas on
the landscape by the reclassification of slope from 0 to 1 degree indicates differences
across elevation products (Figure 5.20). For the LiDAR 90 m, the Acadia study area is
depicted as being mainly flat, with some steeper slopes (> 1 0) along the wetland areas.
Similar representation of the flat topography is illustrated with the NED 90 m. In
addition, for the Webster study area, the LiDAR 90 m and NED 90 m are very similar.
For the Webster study area, the LiDAR 90 m, NED 90 m, and NED 30 m, depict the
general patterns of the valleys in between greater slope upland areas.

For the SRTM elevations, suitable slopes depicted the general patterns of flatter
agricultural lands, and some steeper areas along the wetland. However, there are a few
locations where there are slopes greater than 1 degree on the landscape. For the Webster

study area, the SRTM 90 m slope reclassification illustrated a more regular, rectangular

131

pattern of slope between 0 -1°. This representation is not as smooth as the LiDAR 90 m
and NED 90 m representations of the landscape properties. The pattern of the suitable
SRTM 90 m slope indicates some grouping where expected in the valley area because of
the flat topography and low position.

The other criterion of the land use application was topographic position.
Topographic position was highly dependent on the characteristics of the focal mean
output for the study areas. The results seen in Figure 5.21 of the focal mean of the Acadia
study area indicate a much smoother depiction of the mean elevation characteristics of
the study area. The focal mean depicts the lower elevation in the southwest around 3 m
compared to the higher elevations around 16 m in the northeast. The focal mean
operation produced a surface, which was smoothed in terms of elevation; however, once

differenced with the actual elevations the low locations on the surface could be identified.

132

 

 

omodv So. on Soda acad— oam.m o

 

   

...a. .. 5..

E8 2.5—m Ecm omz

   

_ no; 203m 5325

ES «<9:
new >35 «:32

 

 

Suitable Low Slope Areas: Acadia/ Webster study areas

Figure 5.20

133

 

LiDAR 90 m SRTM 90 m

N

i n High:19.6

Low: 1.3

 

0 13,500 27,000
—=Melers

 

 

 

Figure 5.21. Select depictions of the Focal Mean of Acadia study area

The results of the application are summarized by cell count and area (Table
5.33). This table shows that using the LiDAR 90 m and the NED 90 m for the suitability
model resulted in overall similar areas. The LiDAR and NED suitability overall cell
counts differed by 3 % (Acadia study area) and 1 % (Webster study area). The Webster
study area identified less suitable areas with all models, which was expected given the
flat topography criterion. Table 5.33 indicates the difference in the total area between the
Webster and Acadia study areas. The Webster study area had less flat topography than
the Acadia study area. The results were found to be highly dependent on the results of
the focal mean operation in relation to the topographic position on the landscape. The
subsequent sections discuss the specific application result characteristics for the study

areas, along with contingency matrices.

134

 

Table 5.33. Suitable Areas from the Land Use Application Output

 

 

 

 

 

Acadia study area Webster study area Difference in km2
Cell Counts Area (kmz) Cell Area Webster-Acadia
Counts (ka) (ka)
LiDAR 90 m 110,540 895.4 58,689 475.4 420
NED 90 m 102,917 833.6 56,592 458.4 375
SRTM 90 m 136,684 1,107.1 31,411 254.4 853

 

 

 

 

 

 

135

 

 

Swami—HI 0 5m...—
83m 809 a as. 3 D 25.5 I

 

 

 

 

 

‘ ,
My“ :9.

.M

D

s

t
. .s .m ..7
graham...” ,
T L \rhany...

..InN/YKI \

an

im-

‘1

 

Eco 53:

 

 

Final Application Results: Acadia! Webster study area

Figure 5.22

136

5.4.1. Acadia study area

For the Acadia study area, the LiDAR 90 m and NED 90 m are quite similar
based on observations of the suitable areas (Figure 5.22). The wetland area along the
southwestern portion of the study area was indicated as suitable in the LiDAR 90 m and
the NED 90 m DEMs; however, with the SRTM 90 m this area was seen as unsuitable.
This is likely because the SRTM did not identify the wetland as flat given the forests on
the landscape give the SRTM a raised appearance in these areas, unlike the LiDAR 90 m
and NED 90 m. The LiDAR 90 m and the NED 90 m output contained many linear-like
features, which may be a result from the focal mean operation. In contrast, the SRTM 90
m result included the vast majority of the study area excluding the woody wetlands on the
landscape. The striping seen in the southeastern part of the study area was also seen as
unsuitable here. The striping could have resulted in slope values exceeding the 0-1
degrees, which would have made these areas unsuitable in the final output. Also, in terms
of the woody wetlands on the landscape, this would have been identified as not a low
location on the landscape and unsuitable in this context. However in reality, the wetlands
may be in a depressional area, with a neighboring agriculture field.

From the addition of the suitable results of the SRTM, NED, and LiDAR
DEMS, the following contingency matrices were created (Table 5.34 and Table 5.35).
The producer’s and user’s accuracy was calculated to see how the DEMS performed in
relation to the LiDAR, used as a reference. The user’s accuracy is calculated by the
number of cells correctly classified in each category divided by the number of cells in
that category (row total). The user’s accuracy represents the probability a category is

represented as that category in the reference data (Lillesand, Kiefer, and Chipman 2004).

137

The producer’s accuracy is the number of cells correctly classified by the reference data
divided by the number of cells in that specified category (column total). The producer’s
accuracy relays how well the reference data of a given category were classified
(Lillesand, Kiefer, and Chipman 2004). In terms of the LiDAR and NED comparison,
there were relatively high producer and user accuracy.

The overall accuracy is determined from the correctly classified cell counts of
the suitable and unsuitable LiDAR cells divided by the total number of cells. When the
LiDAR was compared with the NED, the overall application accuracy of the NED was
74.6 %, which was higher than the SRTM, 51.7 %. The Kappa coefficient is a measure of
the degree to which the percent of correctly classified values of the contingency matrix
are from true agreements compared to chance agreements (Lillesand, Kiefer, and
Chipman 2004). The Kappa coefficient is considered the corrected agreement based on
chance. A value closer to 1 indicates perfect agreement, while a value closer to -1
indicates greater disagreement (Lillesand, Kiefer, and Chipman 2004). The Kappa
coefficient was 0.5 for the LiDAR/N ED comparison, while the LiDAR/SRTM
comparison had a much lower Kappa coefficient of 0.04. These Kappa coefficients
indicate greater classification accuracy for the NED-based application than for the
SRTM-based application. The SRTM-based application was no better than chance at

identifying suitable and unsuitable cells, since its kappa coefficient was 0.04.

138

Table 5.34. Comparison of LiDAR 90 m and NED 90 m Results: Acadia study area

 

 

 

 

 

 

 

 

 

 

 

 

 

LiDAR
Unsuitable Suitable Row User’s
Total Accuracy
NED Unsuitable 109,155 35,810 144,965 0.7
Suitable 26,767 74,626 101,393 0.7
Column Total 135,922 110,436 246,358
Producer’s 0.8 0.7
Accuracy i
Overall 74.6%
Accuracy
Kappa 0.4
Coefficient

 

 

 

 

 

 

Table 5.35. Comparison of LiDAR 90 m and SRTM 90 m Results: Acadia study area

 

 

 

 

 

 

 

 

 

 

 

 

 

 

LiDAR
Unsuitable Suitable Row User's
Total Accuracy
SRTM Unsuitable 64,173 47,179 111,352 0.6
Suitable 71,749 63,257 135,006 0.5
Column 135,922 110,436 246,358
Total
Producer’s 0.472 0.573
AceuraLey
Overall 51.7%
AccuracL
Kappa 0.04
Coefficient

 

 

 

 

 

 

5.4.2. Webster study area

The locations found suitable for the Webster study area were much more sporadic
in comparison to the large areas found suitable in the Acadia study area (Figure 5 .20).
There was an area in the southwestern portion of the study area along the wetland area,

which was found suitable. However, there are some noticeable blocky patterns to the

139

result towards the southwestern portion of the Webster study area. This may have to do
with the presence of forests on the landscape neighbored by agriculture or pasture fields,
thus the pasture field would have been seen as a low location on the landscape, and
suitable in this application output. These rectangular patterns could be created from
forests being cut out of the landscape.

The contingency matrices in Table 5.36 and Table 5.37 illustrate the
comparisons of the LiDAR to NED application results and the LiDAR to SRTM
application results for the Webster study area. The LiDAR/N ED comparison had an
overall accuracy of 82.0 %, which was greater than the overall accuracy of the
LiDAR/SRTM of 66.6 % for the Webster study area. Overall, the Webster study area had
greater percentages of correctly classified cells for both the LiDAR/N ED comparison
(82.0 %) and the LiDAR/SRTM comparison (66.6 %) compared to the Acadia study area.
This could be from the different land cover characteristics of these study areas, which are
accounted for differently in the models, such as the differences between non-forested and
forested areas. As indicated above regarding the more forested landscape of the Webster
study area created different suitable patterns compared to the more agricultural Acadia
study area. The Kappa coefficient was higher for the LiDAR/NED (0.5) comparison than
for the LiDAR/SRTM (0.1) comparison. This indicates there was more agreement
between the LiDAR and NED application results compared to the LiDAR and SRTM
application results. For the LiDAR/NED comparison, there were relatively high
producer’s and user’s accuracy. However, for the SRTM/LiDAR comparison, the
producer’s and user’s accuracy for the unsuitable category was high compared to the

suitable category.

140

Table 5.36. Comparison of LiDAR 90 m and NED 90 m Results: Webster study area

 

 

 

 

 

 

 

 

 

 

 

 

 

LiDAR
Unsuitable Suitable Row User's
Total Accuracy
NED Unsuitable 106,998 17,191 124,189 0.9
Suitable 15,313 41,217 56,530 0.7
Column 122,31 1 58,408 180,719
Total
Producer's 0.9 0.7
Accuracy
Overall 82.0 %
Accuracy
Kappa 0.6
Coefficient

 

 

 

 

 

Table 5.37. Comparison of LiDAR 90 m and SRTM 90 m Results: Webster study area

 

 

 

 

 

 

 

 

 

 

 

 

 

LiDAR
Unsuitable Suitable Row User's
Total Accuracy
SRTM Unsuitable 106,048 44,019 150,067 0.7
Suitable 16,647 14,670 31,317 0.5
Column 122,695 58,689 181,3 84
Total
Producer's 0.9 0.3
Accuracy
Overall 66.6%
Accuracy
Kappa 0.1
Coefficient

 

 

 

 

 

Figure 5.23 illustrates the different combinations of unsuitable and suitable cells
classified per study area for the LiDAR/SRTM and LiDAR/NED comparisons. For both
Louisiana study areas, the areas identified as unsuitable areas border the wetlands. Most
of the Webster study area was seen as unsuitable, which can be attributed to much of the

area not meeting the slope criteria. Fewer areas were identified as unsuitable by the

141

LiDAR and suitable by the SRTM in the Webster study area compared to the Acadia

study area.

 

SRTMI LiDAR Comparison LiDARl NED Comparison

 

 

Legend
N
- Both Unsuitable
A - LIDAR Unsunable, SRTMJNED Suitable
- LIDAR Suitable, SRTM/NED Unsuitable
0 11 ,000 22,000
_=Meters |:] Both Suitable

 

 

 

 

 

 

Figure 5.23. Application Result Classified by Possible Categories

142

These locations of unsuitable by the LiDAR and suitable by the SRTM in the
Acadia study area were found surrounding the wetlands. The locations identified as
suitable in both the LiDAR and SRTM were found in some water areas in Webster and in
patches throughout the Acadia study area. In terms of the LiDAR/N ED comparison,
locations identified as unsuitable in both study areas were outside the woody wetlands.
There were distinct patterns identified as both suitable in the LiDAR and NED. These
were identified in valleys and present in agriculture areas in the Acadia study area.
Another noticeable pattern in both study areas was the presence of circular borders
around suitable areas. This could be attributed to the low topographic position criterion,
which used a circular radius in its operation.

The suitable locations were also determined for all the DEMS (i.e. NED, LiDAR,
and SRTM suitable) through the multiplication of the grids (Figure 5.24) For the Acadia
study area, locations found suitable in all DEMS appear to be clumped throughout the
study area. In contrast to the Webster study area, wetland areas are not found to be
suitable. The Webster study area indicates some locations to be suitable for all the DEMS.
This is likely due to use of the low topographic position criteria in the suitability analysis.
The total area found suitable by all the DEMS for the Acadia study area was 331,063,200
m2 (40,872 cells) (331.1 kmz), and the total area found suitable for the Webster study area
was 91335600 m2 (11, 276 cells) (91.3 kmz). In examination of the areas found suitable
in all DEMS, the Acadia study area found a greater area suitable compared to the Webster

study area.

143

 

 

 

 

 

 

 

 

 

 

 

" T w" I .Q I‘,
F: s r , . .' I
.’ ‘ . A r.‘ ‘ .
'7‘; «T. ' , . .a ,
' - v_ . ' If;
"r , .e - \ 5
I; - ‘. (z '1‘
3 -. 9, > J it
t , ‘ . .t . 1 '
I g . ' . “
I a x . '2; l
"bl; an." ’
‘~‘r- ' . r” x; m— v
I" .‘ h‘ .' > f
. ., ,' w ..l f L
-5';...‘ § 7“ _,,._“~"‘
I.-‘--. 5 - . J’ a r "I.”
. ‘ fl. _ 2 «r .3;
h 3 . ’ _
r‘ I fe‘ V | 5
- ' 1
. ‘f . 41 \ s- ", , fl
. .2 ii
i sire," ‘ . _ u}
N
- Suitable :I Unswtable . 0 15.000 30'9hligters

 

 

Figure 5.24. Suitable locations found in all DEMS of Application Analysis
5.4.3. N ang Rong study area
The Nang Rong application results were similar for the SRTM C-band and the
SRTM X-band 90 m (Figure 5.25). The total area identified as suitable for the Nang
Rong DEMS (Table 5.38) indicate the SRTM X-25 m DEM identified much less area
than the SRTM C-90 m and SRTM X-90 m resulting in a difference of 311.5 km2 and
362.3 kmz, respectively. The areas found suitable in the Nang Rong region are generally
found throughout the lowland, low slope areas. The higher elevations in the southeastern
part of the study area were found to be unsuitable. For the SRTM C-90 m and the SRTM
X-90 m, the main valleys were found to be suitable. These areas were identified as low
slope and low topographic position. In contrast to the patterns identified with the SRTM
C-90 m and the SRTM X-90 m, the SRTM X-25 m DEM indicated generally individual

cells as suitable rather than large areas. There is a grouping of cells towards the center of

144

 

the study area, which likely corresponds with a valley in the study area. This is further
identified in the multiplication of the DEMS in Figure 5.26, which identifies all areas
(1407 cells ~11.4 kmz) found suitable in the Nang Rong study area DEMS.

Table 5.38. Nang Rong Suitable Areas fiom the Land Use Application Output

 

Suitable Cells Suitable Area (kmz)

 

 

 

SRTM C- 90 m 42,469 344.0
SRTM X-25 m 51,914 32.4
SRTM C-90 m 48,733 394.7

 

 

 

 

 

145

 

 

SRTM X-band 90m

 

 

 

SRTM X-band 25m
16

 

0
—

 

 

- Suitable |:] Unsuitable

SRTM c-band 90m

 

 

 

 

 

 

 

Figure 5.25. Final Application Results: Nang Rong study area

146

 

    

 

- Unsuitable

 

 

 

- SJitable
0 3.000 6.000 12,000
DZMETBFS

 

 

 

Figure 5.26. Suitable locations found in all DEMS of Application Analysis

147

The contingency matrix used the SRTM X-90 m as the reference compared to
the SRTM C-90 m (Table 5.39). The contingency matrix of the SRTM X-90 m and
SRTM C-90 m indicate the overall accuracy was 68.61 %. Thus, this indicates that there
was high overall agreement between the SRTM X—90 m and SRTM C-90 m. The Kappa
coefficient was 0.3290. The producer’s and user’s accuracy were a little higher for the

unsuitable identification.

Table 5.39 Comparison of SRTM C-90 m and SRTM X-90 m Results: Nang Rong

 

 

 

 

 

 

 

 

 

 

 

 

SRTM X-90 m
Unsuitable Suitable Row User's Accuracy
Total
SRTM Unsuitable 57,243 21,919 79,162 0.7
lC-90 ml Suitable 16,260 26,192 42,452 0.6
Column 73,503 48,111 121,614
Total
Producer's 0.8 0.5
Accuracy
% Correctly 68.6 %
Classified I
Kappa 0.3
Coefficient

 

 

 

 

 

 

 

 

Figure 5.27 illustrates the possibilities of the SRTM X-90 m and SRTM C-90
m comparison. Much of the agricultural area was identified as unsuitable for both the
SRTM C-90 m and SRTM X-90 m. There were isolated locations (i.e. single cells)
identified as unsuitable for the SRTM X-90 m and suitable by the SRTM C-90 m. A
concentration of cells was identified as suitable by the SRTM X-90 and unsuitable by the
SRTM C-90 m on the eastern side of the study area. Suitable locations identified by both

the SRTM C-90 m and SRTM X-90 m were found in the valleys. Other suitable locations

148

in the study area had circular edges, which could be attributed to the low topographic
position criterion. Thus, the Nang Rong application illustrated how the constraints of flat
topography and low topographic position would result in similar areas between the
SRTM C-90 m and the SRTM X-90 m. However, when the SRTM X-25 m was
employed, the results differed much from the SRTM C-90 m and SRTM X-90 m. This

suggests the importance of spatial resolution in this application.

149

 

 

9.0352?— HIHH'

m: NF w v N

4

Z

          

o

 

 

Begum 50m I
29853: 580 5%.
353%. 58-x 2me
asssm 580 5%
35352: 58-x 2me I

2935?: Sam I

 

 

 

 

Figure 5.27. Comparison of SRTM X-90 m and SRTM C-90 m

150

5.5. Discussion: Lessons Learned

This analysis of the SRTM DEM in relation to LiDAR-derived and NED DEMS
proved to be useful in identifying pertinent SRTM characteristics. From a geographic
perspective, this investigation of the elevation and slope properties of the SRTM DEM
showed how similar raster processing and statistical analysis could be employed in
different geographical areas. The utility of investigating the properties of different DEMS
(LiDAR, NED, and SRTM) was seen in the land use application, which demonstrated
how the selection of one DEM could differ from another. This thesis research illustrated
some important lessons regarding digital terrain analysis, which are pertinent to
geographers and other researchers who incorporate digital elevation models into their
research. Themes regarding comparisons of elevation models, error analysis, and land
cover/ land use relationships are summarized here.
5.5.1. Louisiana Study Areas

From the analysis of the elevation models in Louisiana, my findings suggest that
at the coarser resolution of the LiDAR data, it did not result in significantly different
results from the coarsened NED data. This has implications in the utility of the LiDAR in
comparison to the NED data. For instance, it may be suitable to use the NED data at a
certain coarser resolution and have the same findings as if the LiDAR was used. More
research would need to be done regarding resampling the LiDAR data at various
resolutions to compare them to the NED DEM at the same resolution.
5.5.1.1. Elevation Characteristics

The elevation characteristics of the LiDAR and NED had similarities; however,

had slightly different range and mean elevations (Tables 5 .1 and 5 .12). Figures B.1-3 and

151

B.7-9 in the Appendix illustrate a striking difference among the SRTM, LiDAR, and
NED DEMS. In the Acadia study area, the woody wetlands are seen as raised in the
SRTM DEM in comparison to the surrounding elevations. The NED and LiDAR DEM,
on the other hand, depict the wetlands as a low area on the landscape. Also, the striping
of the SRTM DEM as is an artifact of the DEM from its radar collection methodology.
This is in contrast to the LiDAR and NED DEMS. This pattern in the wetlands of the
SRTM is important to consider in terms of using the SRTM DEM in highly vegetated
areas, as it could portray inverse relationships between vegetated and non-vegetated
areas. Also, the user needs to be aware of the possible erroneous values of the SRTM
elevation negative values, as these are not realistic elevation values. As Hancock et al.
2006 warned about the use of the SRTM in hydrology studies because of its coarser
resolution, this thesis work cautions the user about working with SRTM in highly
vegetated areas.

In order to examine the error characteristics of the SRTM DEM, the reference
DEMS needed to be resampled to a comparable resolution. The nearest neighbor
resampling technique was used to create DEMS at a coarser resolution. At this coarser 90
m resolution, some key results appeared across the study areas. In terms of the elevation
values of the Acadia study area, the LiDAR 90 m and NED 30 m had a correlation of
0.929 and the LiDAR 90 m and NED 90 m had a correlation of 0.933 (Table 5.5). This
indicates that the use of the NED 90 m or the LiDAR 90 m would not differ greatly from
each other, despite being derived from different source DEMS. Similar results occurred
with the Webster study area, which would lead me to believe possibly similar results

would occur with the slope characteristics.

152

5.5.1.2. Slope Characteristics

The slopes of the LiDAR, NED, and SRTM DEMS decreased with an increase in
resolution. This is the typical inverse relationship is typical with the coarsening of
elevation data. However, despite similarities noted in elevation, smaller slope correlation
coefficients were found between the NED 30 m and LiDAR 90 m (0.661) and the NED
90 m and LiDAR 90 m (0.752) (Table 5.6). This suggests that when using a certain DEM,
the user needs to be aware of the possible slope/ elevation differences in the underlying
data and determine whether these differences are relevant to their use. However, from
examination of the slope of the study areas, the user needs to be cautious with
interpretation of slope values of the SRTM given the raised areas of vegetated locations
seen in the wetlands of the Acadia study area.
5.5.1.3. Error Analysis: Land Cover/ Land Use Relationships

The analysis with the use of the boxplots, difference of means t-tests, comparisons
of means, RMSE, and variances were all done to answer my research questions related to
the land cover/land use characteristics and relationships of vegetated and non-vegetated
characteristics of the SRTM DEM (research questions 3, 4, and 5). As seen with the
comparison of elevation values the LiDAR 90 m and NED 90 m were similar, thus,
similar results were found from the SRTM error analysis with these products in the
Louisiana study areas. This suggests that with the LiDAR DEM coarsened to 90 m, it
would be similar to the NED 90 m. However, in consideration of the error results, the
Acadia study area had lower average SRTM-LiDAR error (2.8 m) compared to the

Webster study area (7.2 m). The error analysis of the Louisiana study areas illustrated the

153

influence of vegetation on the SRTM DEM. From the box plots (Figure 5.6 and 5.11) and
significant difference of means of the non-vegetated and vegetated groups for the Acadia
and Webster study areas, the vegetated and non-vegetated areas indicated significant
differences in SRTM-LiDAR DEM error. The land cover/land use comparisons of the
average SRTM DEM errors for the non-forest classes (developed, water, and agriculture)
and forest classes (evergreen, deciduous, mixed forest, and woody wetland) show
significant differences between non-forest and forest classes.

These positive mean errors of the Acadia and Webster study areas indicate a
positive bias of the SRTM DEM compared to the NED and LiDAR DEMS. This positive
bias of the SRTM DEM was attributed to the radar scattering from the vegetation cover
on the landscape. Significant one-sided t-tests confirmed the positive bias of the SRTM
DEM. The relationships between the vegetation cover and the SRTM DEM error were
made qualitatively by visual comparison of the error patterns and the NLCD 2001 land
cover/ land use and quantitatively from comparisons of the average errors and RMSE
values per land cover/land use class. Similar to research findings from (Bhang, Schwartz,
and Braun 2007; Carabajal and Harding 2006; Shortridge 2006) regarding the
relationships of vegetation and SRTM, this research found higher average errors for
forested areas. From examination of the means and variances of the NLCD 2001 classes,
it was evident that the wetlands variances (17.3 m for the Acadia study area and 24.2 m
for the Webster study area) and other vegetated classes for the Webster study area
(evergreen 27.9m, forest/ shrub 31.7 m) had high variances compared to the agriculture
variance of 16.3 m for the Webster study area and 4.0 m for the Acadia study area

(Tables 5.11 and 5.22). This is suggested to be from the various heights of trees in these

154

areas, which would have been recorded at different heights in the radar collection
process. Also as seen with the case of the water variance (3 7.6 m) in the Webster study
area, its variance could be from the scattering from the surface. Editing has been done to
the SRTM in terms of the boundary of land and water to create a more smooth water
surface (NASA 2005).

The variances of the land cover/ land use classes from the NLCD differ because
of the manner which the radar used to create the SRTM was collected. The SRTM C-
band transmits and receives short wavelengths (5.6 cm), which do not completely
penetrate tree canopy. According to Walker, Kellndorfer, and Pierce (2007), the
vegetation on the surface including the twigs and branches are able to scatter the radar
energy. The SRTM C-band receives and transmits at a wavelength slightly greater than
the X-band (3 cm); thus this small difference enables the SRTM C-band to penetrate
slightly further into tree canopy (Walker, Kellndorfer, and Pierce 2007). This accounts
for the greater variances seen in the vegetated classes in the Acadia and Webster study
area. The SRTM C-band and SRTM X-band used to collect data for the SRTM DEM
were of short wavelengths, which show less capability to penetrate tree canopy compared
to the longer wavelength (74 cm) of the P-band (Andersen, Reutebuch, McGaughey
2006). As longer wavelengths have a greater ability to penetrate tree canopy.

Figure 5.28 illustrates a schematic of the SRTM radar striking tree canopy and a
nearby field and the more scattering that would occur from the tree canopy compared to
the field. Thus, the agriculture class would have lower SRTM DEM variances as
illustrated with my research in the Acadia and Webster study areas. To illustrate how the

variance of the SRTM DEM error would differ within a vegetated class, Figure 5.29

155

shows how the radar striking the surface of a vegetated surface, such as an area identified
as deciduous forest, would scatter and record the height differently for locations within
that area. These schematics show the tree canopy with leaf on conditions. This would
lead to more differences with the same deciduous area in an SRTM DEM error analysis,

as my research illustrated greater variances in the vegetated classes.

 

 

 

Tree Canopy Crops/ Pasture

 

 

 

Figure 5.28. Interaction of Landscape Feature with short wavelength radar. This
schematic shows the radar used to create the SRTM DEM reflecting off the tree canopy
in leaf on conditions, while less scattering is observed with the short grasses or crops in
the nearby pasture. Modified from Andersen, Reutebuch, and McGaughey (2006).

156

 

eierew

 

 

 

 

 

 

Deciduous Forest J

 

Figure 5.29. Reflectivity of Radar from Tree Canopy. This schematic of a deciduous
forest with leaves on illustrate how the radar from the SRTM mission would reflect from
the forest canopy at different heights above the surface. These various height differences
would be reflective as DEM error in my analysis and result in greater variances for the
vegetated classes.
5.5.1.4. Land Use Application

The land use application demonstrated how the choices of the user can impact the
results from various DEMS. Specifically, the land use application is important because it
highlights the results of the SRTM in comparison to other models. In locations, with only
the SRTM DEM as a reliable DEM, there would not be any DEM to compare the results

of an application to. However, from this research it can be seen how the characteristics of

157

the SRTM influence an application result. The criteria of the land use application
warrants some further discussion. In terms of identifying an agricultural land use practice,
slope is a significant factor, given the relationships to soil moisture characteristics
(McDaniel, Trahan, and Godfrey 2006). However, given the limited data sources I had on
the Thailand area, the topographic position was used to identify relative position on the
landscape. By the employment of focal operations, this was done to identify these low
areas on the landscape. The use of the focal operations in this way illustrates how these
can be combined to identify pertinent locations of an area. For instance in terms of how
the SRTM responds to vegetation by overestimating the elevation in these areas, as seen
in the wetlands of Acadia, impacted the slope classification in the SRTM DEM. In the
end to the results from the LiDAR and NED DEMS had an overall accuracy of (74.6 %
and 82.0 %) for the Acadia and Webster study areas, which differed from the SRTM
DEM results of 51.7 %, and 66.6 %, respectively (Tables 5.34-37).

LiDAR DEMS show promise in terms of their analytical capability; however, only
time will tell about how successful their widespread use will be. As the fine-resolution of
the LiDAR DEMS do require more processing time and larger data storage capabilities.
So, given that LiDAR is becoming more mainstream on national, state, and local
geographic data repositories indicates to me, students and researchers will have to
become more aware of this type of data’s uses and limitations. However, the United
States seems to be advancing at greater speeds in terms of its LiDAR collection compared
to other countries, since the literature using airborne LiDAR is mostly with United States
study areas from those sources identified from Table 2.5. Internationally, satellite sources

of LiDAR have been used as reference to the SRTM DEM and other DEMS (Hofien et

158

al. 2006). Thus, given the SRTM X-band is available for part of the world, this thesis
research investigated its utility in elevation analysis compared to the SRTM C-90 m,
which is publicly available.
5.5.2. Nang Rong, Thailand

The flat topographic characteristics of the Louisiana and Thailand data illustrate
some key findings to the research community. As in the case of the Nang Rong area, the
SRTM DEM appeared bumpy in a relatively flat area, which was not realistic of the
terrain characteristics. This was further accentuated with the Nang Rong SRTM X-25 m
which appeared to have much noise in the DEM. Researchers should consider the utility
of the SRTM X-25 m DEM in this context of a mainly flat topographic region given the
noise seen in the slope of this DEM. In terms of the Nang Rong study area, when the
nearest neighbor method of resampling was used on the SRTM X-band 25 m, the SRTM
90 m was more comparable to the SRTM C-90 m (elevation correlation coefficient of
0.977) and removed the noise seen in the SRTM X-25 m DEM (Table 5.27). Future
consideration of the use of a spatial filter or a resampling method to smooth the X-band
DEM may be warranted to have a more realistic interpretation of the earth’s surface.

The SRTM C-90 m and SRTM X-90 m had a mean error of 0.7 m, which showed
little bias of the SRTM C-90 m. It is expected that the SRTM C and SRTM X would have
similar responses to land cover/ land use classes on the landscape given their collection
methods. However, a global land cover dataset was not incorporated into this research.
Even with the examination of the SRTM C-90 m DEM, it appeared bumpy in areas of
low relief, thus knowledge of the region of interest and DEM characteristics are needed

to interpret these elevation models correctly. From the land use application of the SRTM

159

X-90 m and SRTM C-90 m, there seemed to be in agreement in the majority of the study

area; however, not the southeastern portion of the study area (Figure 5.27). This land use

application in the context of the Nang Rong study area illustrated how DEMS of the same
resolution but fiom different original production methods could identify some but not all

areas as appropriate for the application of interest.

5.5.3. Comparable Tree Cover Analysis

Similar to Carabajal and Harding (2006) who incorporated a global land cover
dataset into their investigation to compare SRTM characteristics across land cover types,
this thesis research used MODIS % tree cover to examine vegetation characteristics
across study area. From this thesis research, a few significant findings presented itself. As
expected, the Webster study area had higher average % tree cover (53.5 %) compared to
the Acadia study area (19.9 %). However, the ranges of the Webster and Acadia study
areas differed slightly from 86 % for the Webster study area and 100 % for the Acadia
study area (Table 5.31). The Nang Rong study area has a smaller range (45 %) compared
to the Louisiana study areas. This indicated that mainly forested areas occur at the higher
elevations or near villages in the Nang Rong District.

From the Acadia study area, it was learned that the portioning of the MODIS %
tree cover at 50 % would show significant differences between mean SRTM-LiDAR
error and variances of the groupings. However, the variances of the Webster study area
were not significantly different (Table 5.32) Future research could involve the
identification of a certain percentage that would yield significantly different results. In
terms of the MODIS data for the Nang Rong study area the variances were found to be

significantly different between the groups. The MODIS % tree cover gave some pertinent

160

information regarding the vegetation cover. Future research may involve the
identification of a land cover dataset, which would have more comparable cell resolution
of the SRTM DEM for more specific analysis by land cover class or vegetation species.
5.5.4. Global implications

In geographical research analysis of different types of environments is important
to discovering common themes across regions. The SRTM DEM is a near-global dataset
created from consistent processing, and overlaps national boundaries. Through this
dataset, it is able to be incorporated in local, regional, and national level research. As
seen from the applications (Blumberg 2006; Brandt and Townsend 2006; Fredrick et al.
2007) the SRTM DEM is making research possible over broad geomorphic areas, in
relevant areas of physical geography, and people/environment interaction research. By
learning about the SRTM characteristics from comparisons of the Acadia and Webster
study areas in Louisiana with high resolution DEMS, other applications in areas without
another DEM can be better informed. Thus, global environmental research that
incorporates digital terrain analysis shows much potential through the incorporation of

the SRTM DEM, advancement of geospatial technology, and human intuition.

161

6. Conclusions

The aim of this research was to gain a greater understanding of SRTM elevation
error characteristics in terms of its magnitude and its relationships to land cover. The
correlations of elevations and slopes were examined to determine the similarities between
the data products: NED, SRTM (C-band/X-band), and LiDAR. SRTM C-band data were
explored through error analysis using LiDAR DEM data for two locations in Louisiana,
including an agricultural area in southwestern Louisiana and a mainly forested area in
northern Louisiana. Also, SRTM data (C-band and X-band) were examined in a portion
of the Nang Rong region of Thailand. This analysis illustrated differences in the DEM
characteristics through comparable analysis with land cover/land use data, NLCD 2001
for two areas in Louisiana, and MODIS (2000-2001) data for the two areas in Louisiana
and Nang Rong study area. In addition to differences across land cover characteristics,
the land use application illustrated how the selection of a certain DEM can give different
results in an analysis. This section summarizes the key findings, limitations, and suggests
directions for future research.
6.1. Summary

The survey of available DEMS in the United States indicates that select locations
have high-resolution DEMs available (i.e. LiDAR DEMS) for geographical analysis. The
predominance of LiDAR availability in coastal areas was due to the concern of coastal
changes and hazard management. Greater availability of LiDAR in the United States and

elsewhere would enable more fine-scale geographical analysis. Access to high resolution

162

DEM information in the United States enabled this research into SRTM DEM error
characteristics.

The first research question was: What are the relationships among elevations
of NED, LiDAR, and SRTM DEMs? This was investigated with the creation of
correlation matrices. As expected the original and resampled DEMS of the same
production method were found to be similar to each other; however, similarities were also
noted between the NED and LiDAR production methods. For the Acadia study area, high
correlations were observed between the NED 90 m and LiDAR 90 m DEMS (0.933) and
between the NED 90 m and LiDAR 5 m (0.941). Similarly, for the Webster study area, a
high correlation was observed between the LiDAR 90 m and NED 90 m (0.927). For both
study areas, the SRTM 90 m DEMS had lower correlations when compared to the LiDAR
90 m (0.625 for the Acadia study area and 0.873 for the Webster study area) and NED 90
m (0.655 for the Acadia study area and 0.849 for the Webster study area). The similarities
between the NED and LiDAR DEMS were visually observed through comparable color
schemes. For the Nang Rong study area, the SRTM X-90 m and SRTM C-90 m DEMs
were highly correlated (0.977).

Similar to the methods used to address research question 1, a correlation matrix
was created to address research question 2: What are the relationships among the
slopes of NED, LIDAR, and SRTM DEMS? Weaker correlations were observed among
the slope grids compared to the correlations observed with the elevation values. For
instance, the slope of the NED 30 m and NED 90 m slopes had a correlation of 0.060 for
the Webster study area. However, this was expected given the resampling done to obtain

the 90 in resolution from the original datasets. For the Webster study area, the LiDAR 90

163

m and NED 90 m had a correlation coefficient above 0.7. The Nang Rong study area had
similar SRTM X-90 m and SRTM C-90 rn slope ranges (0-22.3°). However, from
examination of the slope distributions from all study areas, the predominance of
relatively flat topography limited investigations of SRTM DEM error patterns in terms of
slope. The Webster study area contained slightly greater relief (43 m-130 m); however,
the gradients were limited to upland areas. Similarly the Nang Rong study region was
mainly not a rugged area, since slope values were relatively small (< 10°). However,
some distinction could be made between the flatter part and steeper slopes of the volcanic
areas. Future research could be directed to investigation of more rugged study areas and
SRTM DEM error.

Following the investigation of the similarities of the elevation and slope values
across datasets, the differences of the DEMS was taken to consider the SRTM DEM error
for the study areas. This was done to address research question 3: Is the average SRTM
DEM error greater than zero? For the Acadia and Webster study areas, the average
SRTM-LiDAR DEM errors of the entire study areas were 2.8 m and 7.1 m, respectively.
The average SRTM DEM error was greater for the Webster study area compared to the
Acadia study area, which can be attributed to the forested cover on the Webster study
area landscape. This research found the SRTM- LiDAR DEM error for the Acadia and
Webster study areas, and the SRTM-NED DEM error to illustrate very similar spatial
patterns of error. This is due to the similarity of the NED 90 m and LiDAR 90 m DEMS.
A positive average elevation error indicates a bias in the SRTM DEM. For the Nang
Rong study area, the average SRTM DEM error was 0.66 m, which was only slightly

greater than 0. In the Nang Rong site, the general error pattern identified on the landscape

164

was of low error throughout the majority of the flatter, agricultural valleys in the northern
part of the site and higher error on the steeper slopes of the volcanic areas in the
southeastern part of the area. To determine if the average SRTM DEM error was
significantly greater than zero, the t-test was done with the alterative hypothesis that the
average SRTM DEM error was greater than zero. For the Acadia and Webster study
areas, this t-test was significant indicating the upward bias of the SRTM. However, for
the Nang Rong study area, this t-test was insignificant.

Given the vertical accuracy of the SRTM DEM to be less than 16 m, it is noted
that some of the higher extremities of the ranges exceeded 16 m. The greatest SRTM-
LiDAR DEM error for the Acadia study area was 49.7 m for the entire area and 21.8 m
for the sample. For the Webster study area the highest SRTM-LiDAR DEM error was
38.2 m for the entire area and 29.7 m for the sample. The Nang Rong study area also
exceeded the 16 m at times, with a range of SRTM error of -79 m to 67 m for the entire
area. Thus, this suggests that at certain locations the DEM error exceeds the 16 m
accuracy expectations of the SRTM.

Another key error statistic calculated for all study area was the root mean square
error (RMSE). The RMSE of the SRTM-LiDAR error analysis for the Acadia study was
about 5 m. However, the RMSE of the SRTM-LiDAR error analysis of the Webster
study area was about 9 m. These RMSE values are within the accepted error values of the
SRTM DEM of 10 m (NASA 2002). The RMSE of the Nang Rong study area was about
3 m. This is more comparable to the RMSE found in the Acadia study area, which had a
predominantly flat and agricultural landscape. Despite the differences in the reference

model used in both the United States and the Thailand context, this research illustrates

165

how the determination and analysis of DEM error relationships to landscape
characteristics can better illustrate the properties of the SRTM model.

This research expands on the options of reference data being used in DEM error
work. As seen reference data ranges from kinematic GPS from Rodriguez, Morris, and
Belz 2006 to satellite sources, such as Hofton et al. 2006. Based on examination of the
large scale error analysis completed with satellite altimeter data the result, there was a
3.60 in global mean difference, with a 3.18m mean difference for North America and a
2.54 m mean difference for Euroasia (Berry, Garlick, and Smith '2007). Other global
assessments have reported different statistics to determine height differences for the
SRTM DEM and reference DEMS. For instance Rodriguez, Morris, and Belz 2006,
indicated for North America there is an absolute height error of 9.0 m and for EuroAsia
there is an absolute height error of 6.2 m. Since other researchers have not done error
analysis with LiDAR-derived data in my selected Louisiana study areas, I ca not make
direct comparisons. However, I can speculate on comparisons regarding overall
differences. However, as my results indicate the general trend of the Nang Rong study
area in Asia had a lower mean error compared to the United States areas is valid;
however, there is a difference in the reference DEMS used.

The fourth research question: What is the relationship of SRTM error across
different land cover/land use classes? was investigated through the examination of the
SRTM error with NLCD 2001 in Louisiana. Generally, the locations of more forested
areas compared to agricultural areas were associated with higher SRTM DEM error. The
SRTM DEM error was higher and positive, which indicates biases in forested areas. In

the Acadia study area, the wetlands had greater average error (8.5 m), while the

166

developed (3.0 m) and agriculture (1.4 m) categories had lower average error and lower
RMSEs. For the Webster study area, the wetland class had a mean error of 10.0 m, which
was than other classes. Both Louisiana case studies reflected rectangular, regular patterns
in relation to agriculture and forest patches evident in the error grid. For the Nang Rong
study area, no direct relationship to land cover was made with this research, since a land
cover/ land use dataset with broad classes was not incorporated into this analysis.

The variances of the SRTM-LiDAR error of the land cover/ land use classes from
the NLCD 2001 were considered for the Acadia and Webster study areas. It was found
that the wetland class had greater variance (17.3 m) compared to the agriculture (4.0 m)
and developed (6.2 m) classes for the Acadia study area. Similarly, in the Webster study
area, the wetland class had a relatively higher variance (24.2 m) , along with the other
vegetated classes of evergreen (27.9 m) and forest/shrub (31.7 m). The water class of the
Webster study area was also seen to have a high variance (3 7.6 m). It is thought that the
variations in the canopy height of the vegetated areas would contribute to the greater
variances of the vegetated classes. The variations in heights of vegetation classes would
not be indicated from the satellite data used to create the NLCD 2001. Similarly, the
scattering that occurs with the radar collection used in the SRTM DEM would attribute to
greater variance in this class (Andersen, Reutebuch, McGaughey 2006).

In subsequent analysis in terms of the SRTM DEM error and the vegetated and
non-vegetated areas, a difference of means t-test was conducted to address research
question 5: Is the average SRTM DEM error greater for vegetated or non-vegetated
areas? For both the Acadia and Webster study areas, the difference of means t-test was

significant. The one-sided t-test was used to determine if the SRTM DEM error was

167

significantly greater than zero. The results of the one-sided t-test showed that the
vegetated classes had significantly greater mean 3 than the non-vegetated classes for the
Acadia and Webster study areas.

The MODIS data with its 500 m resolution did not provide as useful information
as expected to investigate the relationships about the percent tree cover and error
characteristics. The range of the MODIS percent tree cover data for the Nang Rong study
area (0- 45 %) was not as large as the Louisiana study areas’ MODIS range (0-100 % for
the Acadia study area and 0-86 % for the Webster study area). In addition to the
difference of means t-test with data from the NLCD (2001), MODIS data was used to
investigate patterns of vegetation by grouping the percent tree cover of less than 50 %
together and the percent tree cover 2 50 % together for the Acadia and Webster study
areas, and Z 22 % tree cover and < 22.5 % tree cover for the Nang Rong study area.
Significant differences were observed for the Acadia and Webster study areas; however,
not for the Nang Rong study area. So for the Nang Rong study area, the relationships
between more (2 22 % tree cover) or less (< 22.5 % tree cover) forested areas and error
could not be made definitely.

The land use application indicated how the selection of a specific DEM for use
could identify varied application results. The land use application addresses research
question 6: How will the LiDAR, NED, and SRTM DEMs perform in a land use
application? The criteria of flat slope (0-1°) and a low topographic position on the
landscape indicated the LIDAR 90 m, NED 90 m, and SRTM 90 m would identify
different suitable areas on the landscape. The LiDAR 90 m and the NED 90 m did result

in similar suitable results in terms of application use. The contingency matrices used the

168

LiDAR 90 m as reference for the Louisiana study areas, and the SRTM X-90 m as
reference for the Nang Rong study area to confirm these findings. The overall accuracy
of the LiDAR 90 m and NED 90 m was 74.6 % and 82.0 % for the Acadia and Webster
study areas. However, in terms of the comparison of the LiDAR 90 m and SRT M 90, the
overall accuracy for the Acadia and Webster study areas were 51.7 % and 66.6. %,
respectively. Thus, the SRTM had greater differences compared to the LiDAR and NED
outcomes. The SRTM C-90 m and the SRTM X-90 m also identified similar areas in the
application for the Nang Rong site.

Agricultural models and land use analysis with different DEMS can indicate
different outcomes, which may have implications in policies and recommendations for
future land use. As seen from the application literature (land cover/ land use, hydrology,
geomorphology), the incorporation of the SRTM DEM has been used to understand
environmental phenomena better. My research illustrated how a land use application
could differ in results based on DEM choice. Hence, understanding the resolution and
production methods (i.e. LiDAR, NED, or SRTM) of the input model is necessary for
choosing the appropriate DEM for use in analysis. In terms of policy decisions,
identification of key biophysical characteristics of the earth including factors related to
soil moisture, climatic characteristics, and agricultural growth controls, along with
socioeconomic factors are necessary (Bouman et al. 2000). Policy decisions are made
following research on past, present, and projected trends in land use dynamics. Thus, land
use models and applications incorporating the SRTM DEM are likely to occur in the

future.

169

The research is significant to the global community, academic professionals, and
users. This research illustrates the characteristics and potential SRTM has in addressing
environmental issues, and also challenges of its use. The consideration of greater use of
the SRTM also implies greater access to geographic information. Thus, trends in spatial
data and infrastructure to support data sharing, user-friendliness, and analysis
methodologies can benefit from the use of topographic datasets. However, caution must
be employed with the selection of the DEM for application purposes. Knowledge of
production methods and their limitations are important for interpretation of application
results. Thus, in regards to applications of SRTM, it is important to consider the
properties of the model under investigation to properly interpret its results.

6.2. Limitations

This research uses the standard methodology of error assessment, which involves
the alignment of raster cells and differencing of DEMS with consistent projections and
vertical datums. However, this process has limitations related to the data sources and
auxiliary data used in analysis of the DEM error. LiDAR was chosen as a higher quality
DEM in relation to the coarser SRTM DEM. However, LiDAR can come from different
sources and be collected and processed in different ways; thus, those factors need to be
considered in the replication of this type of analysis in other locations with available
LiDAR data. In addition, this research involved the assessment of DEM error in
relationship to land cover from remotely sensed land cover information both from
LandSat (USGS 2003) at the US. sites and MODIS (both in the United States and
elsewhere). As seen with other research, Landsat and/or MODIS land cover products

have been used to assess error differences in the SRTM (Bhang, Schwartz, and Braun

170

2007; Carabajal and Harding 2006). These data have accuracy and resolution limitations
that were not investigated in this research. Also, the MODIS data used at a 500 m
resolution was not downscaled to obtain values at the same resolution of the SRTM
model. These limitations are acknowledged as factors that may result in different
outcomes if data from other areas are employed in a similar process.
6.3. Future Direction of Research

This research illustrates the potential directions geography research can take in
relation to the investigation of DEMS and land use characteristics. Future research can be
directed into the following areas to strengthen understanding of the implication of error in
spatial data analysis and employment of SRTM data in international case studies as a
source of elevation information. This SRTM DEM error research can be advanced in
terms of the application chosen to investigate, reference data chosen in analysis, and
selection of study areas for investigation. Also, the development of land cover sensitive
elevation error models could be used to create more accurate depictions of the earth.

Geomorphic and other types of application domains are utilizing SRTM data in
various landscapes. This research could be furthered by the expansion of the application
domain to include a geomorphic application, which could have limitations based on
resolution of the landscape feature. These geomorphology applications could relate to
study of river hydrology changes or alluvial fan development, as these features would be
observed with SRTM DEM data. For example, certain small topographic features may be
not completely represented with the 90 m resolution of the SRTM. If the SRTM X-band

data were available at a greater extent and at no cost, greater use of this product would

171

enable more detailed analysis. The higher spatial resolution data could be employed for
different types of applications, which are more appropriate with this level of resolution.

Future research directions could explore the relationships between SRTM and
LiDAR in other locations in the United States with different topographic characteristics,
for example to coastal ranges of Washington or the Appalachians of North Carolina, to
determine if such similarities exist between NED 90 m and LiDAR 90 in elevation grids
when used in error assessment. This type of analysis would be utilizing different LiDAR
data as a reference; however, the overall methodology employed would be same.
Currently, the United States in working on a national LiDAR project, with each state
developing projects to obtain lidar for large sections (Stoker et al. 2007). As seen with the
LiDAR distributed from the USGS and other repositories this data can be valuable to
obtaining hi gh-resolution elevation data (Stoker et al. 2006). However, with the greater
extent of coverage of LiDAR other geospatial technology changes would also need to
occur for greater public and researcher use of this data. An effective means of
dissemination would need to be coordinated among agencies collecting LiDAR data. The
USGS along with other state agencies are currently collaborating and addressing the
many challenges associated collecting large datasets across large areas (Stoker et al.
2007)

This work with relatively flat topographic areas contributes to understanding the
SRTM DEM characteristics in this type of landscape. This is relevant to coastal areas
globally, since these areas have been the focus of geographical research, hazard research,
and environmental change research. International research has been devoted to analysis

of coastal areas given the occurrence of natural disasters such as hurricanes and tsunamis.

172

Thus, this research shows that if high resolution data is available, it can be used in
comparative purposes. In addition, this thesis research shows the global research
community how the SRTM can be employed and analyzed in a flat topographic area.

Investigations in more rugged terrains would further illustrate the characteristics
between slope and SRTM DEM error. As seen with work by Miliaresis and Paraschou
2005, SRTM has certain characteristics with more rugged terrain. Internationally,
investigations of more rugged areas could have implications in mountain geomorphology
studies and investigations in inaccessible areas. As geomorphology studies have an
international focus concerned about characterizing landforms in all parts of the world.
Since, mountains often can not be fully surveyed with the use of traditional GPS
technology, the use of the radar collected SRTM data or possibly LiDAR data opens new
ways to analyze these rugged environments.

The future of elevation data in the United States is tending towards LiDAR data;
however, at the international level, LiDAR data is not being acquired at great speed. As
seen in the literature, LiDAR studies are predominantly reported using United States
study area opposed to area outside the United States. Thus, satellite-derived data remain
the main source of elevation data. The SRTM C-90 m DEM improved upon the
resolution of the GTOPO30; however, as seen with this research the SRTM X-25 m did
not prove to be more useful in terms of application use or elevation/slope comparisons in
the case of Nang Rong. Thus, it may not me useful to obtain this data if the SRTM C-90
m DEM would be sufficient for the application intended by the user. The SRTM X-25 m
does have more detail than the SRTM C-90 m DEM; however, their may be noise in the

DEM that could obscure results from the use of this DEM.

173

This research into the characteristics of a near-global model has many
implications for global change research. Globalization of the world has made the need to
address environmental issues on much larger scales. The Intemational-Geosphere-
Biosphere Program: A Study in Global Change was created to focus on global issues and
would benefit from the incorporation of a near-global elevation model into studies
(Kates 1987). Similarly in land cover/ land use change and environmental change
research, the incorporation of a near-global elevation source would complement global
land cover analysis. In addition to environmental change and earth monitoring analysis,
interests in global sustainability could benefit from the incorporation of a near-global
DEM (Wood 2005). Global issues can incorporate geospatial technology, geographic
research, and scientific studies to address issues that cross country and regional borders,
and that have wide-reaching implications. Hence, future studies about the SRTM model
are necessary because of its utility in environmental analysis across multiple scales and
application domains. Results obtained in the United States regarding the characteristics of
the SRTM DEM could apply elsewhere given similar topographic and landscape
characteristics. Future research could be directed into more high resolution comparisons
of elevation datasets with near-global land cover/ land use datasets to determine their

appropriateness of use.

174

Appendix A

175

Table A.1. Louisiana study areas: LiDAR Data

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Quad Name Number: Name ID
USGS ID

Acadia Parish Webster Parish

Basile 3009236 Taylor 3309361

Euricesouth 3009237 Walkerville 3309362

Richard 3009238 Shongaloo 3209306

Churchpoint 3009239 Cullen 3209305

Mire 3009247 Carterville 3209304

Branch 3009246 Ivan 3209312

Iota 3009245 Cotton Valley 3209313

Evangeline 3009244 Leton 3209344

Mementau 3009252 Minden North 3209322

Crowley West 3009253 Hortman 3209321

Crowley East 3009254 Brodrau Lake 3209320

Dusch 3009255 Benton 32093 19

Gueydan 3009260

Wright 3009261

Kaplan 3009262

Lafayette

Parish

Carencro 3009248

Breaux Bridge 3009441

Broussard 3009149

Lafayette 3009256

Download Date 6/27/07 7/1/07

Atlas 2007

 

Source: ATLAS (2007).

176

 

Appendix B

177

 

 

 

 

Elevation (m)

’ - 13.3-85
E: 102-13.2
. ~ [:3 7.2 -1o.1
'- . ‘ D41-“
~ i' ' - -32 - 4.0
. N
i 19,...“ «i
0 4,500 9.000 18,000 27,000 36,090

 

 

Figure B.1. Acadia Study Area LiDAR 5m and LiDAR 90m DEM (meters)

178

 

 

Elevation (m)

- 13.3-85
102-13.2

‘ ‘12-101
. 4.1-7.1

- -32 - 4.0

 

 

 

 

 

3' .. '1»?- .
..‘fh ‘
0 4,500 9,000 18,000 27,0“) 36,090

— — ‘n‘nm

 

 

 

 

Figure 8.2. Acadia Study Area NED 30m and NED 90m DEM (meters)

179

 

Elevation (m)

-13.3-85
102-13.2
|:] 7.2-10.1
[:3 4.1-7.1
- -32-4.o

 

0 4,500 9.000 18,000 27,000 36,000;

 

 

 

Figure B.3. Acadia Study Area SRTM 30m and SRTM 90m DEM (meters)

180

 

 

 

A

.g _ .I 7,,
P") \: VI i—fv'y— I2
I“. , ‘r ,
I . Jet-T
J M g-
L 3 '
l '-"
. , i" .
' "‘1!
fl /,r, .
r ”Q . '.
-( .;' {, I‘vr'

' e ‘ I
(‘9 All I. e V
2’. ~. ~ _ ~_, 1- .
I ,~ ' y. , ,— .—
' /.\

J3: ..._ a J" u

p .. ..
I ‘. /: u‘ )1

p“ ,. \ ’55.“: a 1':

if. ”3.51 , T .

0 4,500 9,000 18,000 27,0(X) 36,030

luV|VIv

 

 

 

Figure B.4. Slope: Acadia Study Area LiDAR 5m and LiDAR 90m DEM

181

 

, .
.4,
7 .
\2
K ;
2‘. ‘, L
i. -’ '
I ‘ '
..5 Are; - r
.2; f i?" V
I ,-
. ,W r“
K. ,2]
7’! b
., , ., JI‘
' 71
/ x»
‘ /. I
.. ' ’f‘ 51.1
4’ i
._ “7’
('17
e...
.. , ‘

0 4,500 9,000

 

18,000

27,000

 

36,o_o_o

 

182

Figure B.5. Slope: Acadia Study Area NED 30m and NED 90m DEM

 

 

 

   

Slope (degrees)

 

4.1 - 46.3
1.1 - 4.0
0 - 1.0
N
0 4,500 9,000 18,000 27,011) 36,000
— ‘- Meters

 

 

 

 

Figure B.6. Slope: Acadia Study Area SRTM 30m and SRTM 90m DEM

183

 

 

DE.»—

aov. Z

comm EQN a

 

 

 

Figure B.7. Webster Study Area LiDAR 5 m and LiDAR 90 m DEM

184

 

 

 

 

 

1.. _ . .I I 25-9.

comma Bat. 89 : 2: m ommw a 92 . @400 I
98 .m2 I
an: -03 I
33 -30.?

a: 8.53m

 

185

 

Figure B.8. Webster Study Area NED 30 m and NED 90 m DEM

 

 

.
”g,
.t g .

‘ .
. . .

~’r..
I:

'#3 fight-I.

1

 

Elevation (m;

A

N
...A

22,

17,100

11 ,4 00

 

0 2.050 5.700

104.0-144
07.0-1040
75.9-87.0
54.3-75.9
43- 64 8

 

Figure B.9. Webster Study Area SRTM 30 m and SRTM 90 m DEM

186

 

 

 

.20! .
com NH

8:...

not:

005 .m 0mm. N

o

 

 

Oelo
OWL;
WET—av

A3293 onsw

  

 

 

 

 

Figure B.10. Slope: Webster Study Area LiDAR 5 m and LiDAR 90 m DEM

187

 

22.9.90.

17,100

11,400

2,850 5,700

 

0

4.1-47.8
1.1-4.0

0-1.0

A
H
o
2
a}
o
3
o
2
U)

   

 

 

 

Figure B.11. Slope: Webster Study Area NED 30 m and NED 90 m DEM

I88

 

 

w“
an.
r
e.
....,

 

<1.

Swan

8::

8... e

I I
8% 8a.“ a

    

0e - o

O? - _.._.

mute - :w
539503 ono_m

 

 

Webster Study Area SRTM 30 m and SRTM 90 m DEM

Figure 8.12. Slope

189

References

190

References

Andersen, H., S. Reutebuch, and R. McGaughey. 2006. Chapter 3 Active Remote
Sensing in Shao, G.; Reynolds, K.M., eds. Computer Applications in Sustainable

Forest Management: Including Perspectives on Collaboration and Integration
The Netherlands: Springer-Verlag: 43-66.

Atlas. The Louisiana Statewide GIS. 2007. Maintained by LSU CADGIS.
Accessed 12 January 2008. http://atlas.lsu.edu/

 

Bailey, J. E., S. Self, L.K. Wooller, and P. J. Mouginis-Mark. 2007. Discrimination of
fluvial and eolian features on large i gnimbrite sheets around La Pacana Caldera,
Chile, using Landsat and SRTM-derived DEM. Remote Sensing of Environment
108: 24-41.

Barber, C., and A. Shortridge. 2005. Lidar Elevation Data for Surface Hydrologic
Modeling: Resolution and Representation Issues. Cartography and Geographic
Information Science 32 (4): 401-410.

Berry, P., J. Garlick, and R.G. Smith. 2007. Near-global validation of the SRTM DEM
using satellite radar altimetry. Remote Sensing of Environment 106: 17-27.

Beyer, H. L. 2004. Hawth's Analysis Tools for ArcGIS. Accessed 29 December 2007.
Available at http://www.spatialecologv.com/htools.

 

Bhang, K. J ., F.W. Schwartz, and A. Braun. 2007. Verification of the Vertical Error in
C-Band SRTM DEM Using ICESat and Landsat-7 Otter Tail County, MN. IEEE
Transactions on Geoscience and Remote Sensing. 45 (1): 36-44.

Blumberg, D. 2006. Analysis of larger Aeolian (wind-blown) bedforrns using the Shuttle
Radar Topography Mission (SRTM) digital elevation data. Remote Sensing of
Environment 100: 179-189.

Bolstad, P. 2005. GIS Fundamentals: A First Text on Geographic Information Systems.
Second Edition. Eider Press, White Bear Lake Minnesota.

Bouman, B.A.M., R. Roetter, R.A Schipper, and AG. Laborte. 2000. Regional land-use
analysis to support agricultural and environmental policy formulation. In
Characterizing and Understanding Rainfed Enviromnents. Edited by T.P. Tuong,
S.P. Kam, L. Wade, S. Pandey, B.A.M. Bouman, and B. Hardy. International Rice
Research Institute (IRRI) Philippines.

191

Brandt, J. and RA. Townsend. 2006. Land use— land cover conversion, regeneration and
degradation in the high elevation Bolivian Andes. Landscape ecology 21: 607-
623.

Carabajal, C., and D. Harding. 2006. SRTM C-Band and ICESat Laser Altimetry
Elevation Comparisons as a Function of Tree Cover and Relief. Photogrammetric
Engineering & Remote Sensing 72 (3): 287-298.

Carlisle, B. H. 2005. Modelling the Spatial Distribution of DEM Error. Transactions in
GIS 9 (4): 521-540.

Consortium for Spatial Information (CGIAR-CSI). 2004. (Last accessed 7 March 2007).
http://csi.cgiarorg/indexasp

Cunningham, R., D. Gisclair, and J. Craig. 2004. The Louisiana Statewide LiDAR
Project. Accessed 8 July 2008.
http://atlas.lsu.edu/central/la lidar proiectpdf

Demirkesen, A.C., F. Evrendilek, and S. Berberglu. 2006. Coastal Flood Risk Analysis
Using Landsat-7 ETM+ Imagery and SRTM DEM: A Case Study of Izmir,
Turkey. Environ Monit Assess. Springer.

Deng, Y. 2007. New Trends in digital terrain analysis: landform definition,
representation, and classification. Progress in Physical Geography 31 (4): 405-
41 9.

DLR.Cluster Angewandte F emerkundung (DLR). German Aerospace Center. 2007.
SRTM Coverage. Accessed 12 January 2008.
http://www.dlr.de/srtm/leve11/data_en.htm

DLR. Cluster Angewandte Femerkundung (DLR). German Aerospace Center. 2008.
Shuttle Radar Topography Mission. Accessed 8 July 2008.
http://www.dlr.de/caf/en/desktopdefault.aspx/tabid-2657/3990_rcad-5914/

Entwisle, B., S. J. Walsh, R.R. Rindfuss, and L. K. VanWey. 2005. Population and
Upland Crop Production in Nang Rong, Thailand. Population and Environment.
26 (6): 449-469.

ESRI. 2006a. ArcGIS version 9.2. How Slope Works. Spatial Analyst. Environmental
Systems Research Institute. Redlands California, USA.

-------- 2006b. ArcGIS version 9.2. Resampling Techniques. Environmental
Systems Research Institute. Redlands California, USA.

Falomi, G., V.Teles, E. Vivoni, R. Bras, and K. Amaratunga. 2005. Analysis and
characterization of the vertical accuracy of digital elevation models from the

192

Shuttle Radar Topography Mission. JOurnal of Geophysical Research 110.

Falvey, L. 2000. Thai Agriculture: Golden Cradle of Millenia. Bangkok: Kasetsart
University Press.

Faust, K., B. Entwisle, R. R. Rindfuss, S. J. Walsh, and Y. Sawangdee. 1999. Spatial
arrangement of social and economic networks among villages in Nang Rong
District Thailand. Social Networks 21: 311-337.

Fisher, P. and N. Tate 2006. Causes and consequences of error in digital elevation
models. Progress in Physical Geography 30 (4): 467-489.

Fredrick, K. C., M. W. Becker, L. S. Mattot, A. Daw, K. Bandilla, D. M. Flewelling.
2007. Development of numerical groundwater flow model using SRTM
elevation. Hydrogeology Journal. 152171-181 .

Fowler, R. 2001. Chapter 7 Topographic LiDAR. in Digital Elevation Model
Technologies and Applications: The DEM Users Manual. Ed. David Maune.
American Society for Photogrammetry and Remote Sensing: Maryland, 207-236

Gesch, D., J. Muller, and T. Farr. 2006. Forward: The Shuttle Radar Topography .
Mission-Data Validation and Applications. Photogrammetric Engineering &
Remote Sensing 72 (3): 233-235.

Global Land Cover Facility. GLCF. 2008. Accessed 18 January 2008.
http://glcf.umiacs.umd.edu/data/srtm/

Gorokhovich, Y., and A. Voustianiouk. 2006. Accuracy assessment of the processed
SRTM-based elevation data by CGIAR using field data from USA and Thailand
and its relation to the terrain characteristics. Remote Sensing of Environment 104:
409-415.

Grohman, G., G. Kroenung and J. Strebeck. 2006. Filling SRTM Voids: The Delta
Surface Fill Method. Photogrammetric Engineering & Remote Sensing 72 (3):
213-216.

Grohmann, CH, C. Riccomini, and F. M. Alves. 2007. SRTM-based morphotectonic
analysis of the Pocos de Caldas Alkaline Massif, southeastern Brazil. Computers
& Geosciences 33: 10-19.

Guth, P. 2006. Geomorphometry from SRTM: Comparison to NED. Photogrammetric
Engineering & Remote Sensing 72 (3): 269-277.

Hancock, G. R., C. Martinez, K. Evans, and DR. Moliere. 2006. A Comparison of

SRTM and hi gh-resolution digital elevation models and their use in catchment
geomorphology and hydrology: Australian examples. Earth Surface Processes

193

and Landforms 31: 1394-1412.

Hansen, M., R. DeFries, J.R. Townshend, M. Carroll, C. Dimiceli, and R. Sohlberg.
2003a. Vegetation Continuous Fields 2001 Percent Tree Cover. North America.
Collection 3. University of Maryland, College Park, Maryland, 2001.
fip://fip.glcfumiacs.umd.edu/modis/VCF/2001-v02/Goodes.NA.2001

 

-------- 2003b. Vegetation Continuous Fields 2001 Percent Tree Cover. Europe/Asia.
Collection 3. University of Maryland, College Park, Maryland, 2001.
ftp://ftp.glcf.umiacs.umd.edu/modis/VCF/200l -v02/Goodes.EUAS.2001/

 

Heo, J., J. Woo Kim, S. Pattnaik, and H. Sohn. 2006. Quality improvement of loblolly
pine (Pinus taeda) plantation inventory GIS using Shuttle Radar Topography
Mission (SRTM) and the National Elevation Dataset (NED). Forest Ecology and
Management 233: 61-68.

Hoffinann, J. and D. Walter. 2006. How Complementary are the SRTM-X and —C Band
Digital Elevation Models? Photogrammetric Engineering & Remote Sensing. 72
(3): 261-268.

Hofion, M., R. Dubayah, J .B. Blair, and D. Rabine. 2006. Validation of SRTM
Elevations Over Vegetated and Non-vegetated Terrain Using Medium

Footprinting Lidar. Photogrammetric Engineering & Remote Sensing 72 (3): 279-
285. ‘

Holmes, K., O.A. Chadwick, and RC. Kyriakidis. 2000. Error in a USGS 30-meter
digital elevation model and its impact on terrain modeling. Journal of Hydrology
233: 154-173.

Hurvitz, 2004. University of Washington. UW Spatial Information Technology, GIS, and
Remote Sensing. 2004. Phil Hurvitz 1998-2004. Last updated 2007. Last
Accessed 29 April 2008.
http://gis.washington.edu/esrm250/lessons/proiection/images av3/datum.gif

 

Hutchinson, MP. and J .G. Gallant. 2005. Representations of Terrain. in Geographical
Information Systems: Principles, Techniques, Management, and Applications.
(Ed) Longley, P. A., Goodchild, M.F., Maquire, D.J., and Rhind, D.W 2nd ed.
Hoboken, New Jersey: John Wiley & Sons.

Inside Idaho. 2007.1nteractive Numeric & Spatial Information Data Engine. University of
Idaho. 2007. Accessed 7 March 2007. http://inside.uidaho.edu

 

International Rice Research Institute (IRRI). 2007. Thailand. Rice Around the World.
International Rice Research Institute. Accessed 13 January 2008.
http://www.irri.org/science/cnvinfo/thailand.asp

 

I94

 

 

Jenson, S., and J. Domingue. 1988. Extracting Topographic Structure from Digital
Elevation Data from Geographic Information System Analysis. Photogrammetric
Engineering and Remote Sensing 54 (1 1): 1593-1600.

Jet Propulsion Laboratory. J PL. 1998. Seeing earth’s surface in 3-D. J PL. Jet Propulsion
Laboratory. Last accessed 8 July 2008
http://southport.ipl.nasa.gov/html/proiects/srtm-whathtml

 

Jet Propulsion Laboratory. J PL. 2005a. NASA. Accessed 12 January 2007. Shuttle Radar
Topography Mission. Data Products. http://www2.ipl.nasa.gov/srtrn/dataprod.htm

-------- 2005b. NASA. Shuttle Radar Topography Mission. Frequently Asked Questions.
SRTM Radar Penetration. Accessed 3 June 2008.
http://www2.ipl.nasa.gov/srtm/faq.html

-----2005c. NASA. Shuttle Radar Topography Mission. Instrument. Accessed 12
January 2007.http://www2.ipl.nasagoWsrtnflmainantennahtml

 

-------- 2005d. NASA. Shuttle Radar Topography Mission. Instrument Overview

Accessed 12 January 2007.
http://www2.ipl.nasa.gov/srtm/instrumcntoverview.html

 

--------- 2005c. NASA. Shuttle Radar Topography Mission. Mission Overview. Accessed
8 July 2008.

http://www2.ipl .nasagov/srtm/missionoverviewhtml

 

--------- 2005f. NASA. Shuttle Radar Topography Mission. Data Products: Data
Coverage. Accessed 8 July 2008
http://www2.ipl.nasa.gov/srtnr/datacovcrage.html

 

Kansas Geospatial Community Commons. 2007. Kansas Geological Survey. University
of Kansas. Accessed 7 March 2007. hm)://www.kansasgis.org/catalog/catalog.cfm

 

Kates, R.W. 1987. The Human Environment: The Road Not Taken, The Road Still
Beckoning. Annals of the Association of American Geographers. 77 (4): 525-534.

Kellndorfer, J., W. Walker, L. Pierce, C. Dobson, J. Fites, C. Hunsaker, J. Vona, and M.
Clutter. 2004. Vegetation height estimation from Shuttle Radar Topography
Mission and National Elevation Datasets. Remote Sensing of Environment: 93:
339-358.

Kiel, B., D. Alsdorf, and G. LeFavour. 2006. Capability of SRTM C- and X- band DEM

Data to Measure Water Elevations in Ohio and the Amazon. Photogrammetric
Engineering & Remote Sensing 72 (3): 313-320.

195

 

Kilpatrick, W.W., C. Godfrey, and C. Henry. 1998. Soil Survey of Webster Parish,
Louisiana. United States Department of Agriculture, Natural Resources
Conservation Service.

Korbick, M. 2006. On the Toes of Giants- How SRTM was Born. Photogrammetric
Engineering & Remote Sensing 72 (3): 206-207.

Lambin E.F., Agbola S.B., Angelsen A., Bruce J .W., Coomes O.T., Dirzo R., Fischer G.,
Folke C., George P.S., Homewood K., Imbernon J ., Leemans R., Li X., Moran
E.F., Mortimore M., Ramakrishnan P.S., Richards J .F ., Skanes H., Steffen W.,
Stone G.D., Svedin U., Veldkamp T.A., Vogel C., Xu J ., Turner B.L., and Geist
H]. 2001. The causes of land-use and land-cover change: Moving beyond the
myths. Global Environmental Change. 11 (4):261-269. -

Landserf. 2005. Image Gallery. Global SRTM Coverage. Maintained by J 0 Wood.
http://wWw.soi.city.ac.uk/~iwo/landserf’landserfi!10/doc/howto/images/srtmCove

ragegif

Leblanc, M., G. Favreau, J. Maley, Y. Nazoumou, C. Leduc, F. Stagnitti, P. van Oevelen,
F. Delclaux, and J. Lemoalle. 2006. Reconstruction of Megalake Chad using
Shuttle Radar Topographic Mission data. Paleogeography, Paleoclimatology,
Paleoecology 239: 16-27.

Lethcoe , K. and Klaver. R. 2008. Simulating the Interupted G‘oode Homolosine
Projection with Arclnfo. EROS Data Center. Accessed May 6, 2008.
http://gis.esri.com/library/userconf/proc98/PROCEED/TO850/PAP844/P844.HT

M

Lillesand, T., R. Kiefer, and J. Chipman. 2004. Remote Sensing and Image
Interpretation. 5th Edition. John Wiley & Sons, Inc.

Louisiana Department of Wildlife and Fisheries (LWLF). 2005. Accessed 2 January
2008. http://www.wlf.louisiana.gov/hunting/wmas/wmas/list.cfm?wmaid=14

Louisiana Geographic Information Center (LAGIC), and US. Department of Commerce,
US. Census Bureau, Geography Division, 2007. TIGER/Line, Census Places
(points) of Louisiana, Geographic NAD83, LAGIC (2007) Accessed 16 July
2008. http://lagic.lsu.edu/data/losco/tiger_places la pts lagic 2007.zip

McDaniel, D., Trahan, G., and CL. Godfrey. 2006. Natural Resources Conservation
Service. Acadia Soil Survey. Accessed 8 January 2008.
http://soils.usda.gov/survechmlinegurvevs/louisiana/

Miliaresis, G., and C. Paraschou. 2005. Vertical Accuracy of the SRTM DTED level 1 of
Crete. International Journal of Applied Earth Oberservation and Geoinformation
7: 49-59.

196

Minerals Management Service. 1999. Gulf-Wide Information System, Louisiana
Highways, Geographic NAD83, LDWF (2001). Accessed 16 July 2008.
http://lagiclsu.edu/data/losco/esi pfimarwoads_LDWL2001.zip

Mongkolsawat C., P. Thirangoon and P. Kuptawutinan. 1997. A Physical Evaluation of
Land Suitability for Rice: A Methodology Study using GIS. GISdevelopment.net.
Asian Association of Remote Sensing (AARS). Asian Conference on Remote
Sensing (ACRS) Proceedings. Last Accessed 5 June 2008.
http://www.gisdevelopment.net/aars/acrs/1997/tsl l/tsl 1004.asp

Moore, 1., RB. Grayson, and A. Ladson. R. 1991. Digital terrain modelling: a review of
hydrological, geomorphological, and biological applications. Hydrological
Processes 5: 3-30.

Mount Saint Helens-LiDAR Data. 2006. Washington State Geospatial Data Archive
University of Washington Libraries. Accessed 29 May 2008.
http://wagda.lib.washin gtonedu/data/type/ elevation" lidar/st_he1ens/toutle01 .html

MSN Encarta Encyclopedia. 2008. Louisiana. Microsoft Corporation. Accessed 28 May
2008. http://encarta.msncom/encyclopedia 761560280/louisiana.html

National Aeronautics and Space Administration (NASA). 2002. Shuttle Radar
Topography Mission (SRTM) spatial metadata dataset. Sioux Falls, SD.: US.
Geological Survey.

National Aeronautics and Space Administration (NASA). 2006. Index of Lidar/terrapoint.
Planetary Geodynamics Laboratory. (Last accessed 2 April 2007).
http://core2.gsfc.nasa.gov/lidar/terrap_oint/

National Aeronautics and Space Administration (NASA). 2005. SRTM Topography:
Accessed 8 January 2008.
flp://e05rp01u.ecs.nasagov/srtm/versi(ml/Documentation/SRTM Topopdf

National Geodetic Survey. NGS. 2007. Computation of GEOID03 Geoid Height.
Accessed 12 January 2008.
http://www.ngs.noaa.gov/ggi-bim’GEOID STUF F / geoid03 promptl .prl

National Geospatial-Intelligence Agency. NGA. 2006. NGA EGM96 Geoid Calculator.
Accessed 12 January 2008.
http://earth-info.nima.mil/GandG/wgs84/gravitvmod/cgm96/intpt.html

NOAA Topographic Change Mapping. 2007. Accessed 12 January 2008. NOAA Coastal

Services Center
http://maps.csc.noaa.gov/TCM/

197

North Carolina NCFloodsmap.com. 2008.NCSU Libraries. NC State University.
Elevation Data Sources. 2008. 29 April 2008.
http://www.lib.ncsu.edu/gis/elevation.html

 

Oimoen, M. 2006. National Elevation Dataset. Metadata Field Definitions. Accessed 8
January 2008. http://ned.usgsugov/N ed/N ED DataDictionarypdf

 

Phongphit, S. and K. Hewison. 2001.Village Life Culture and Transition in Thailand’s
Northeast. White Lotus: Bangkok, Thailand.

Puget Sound LIDAR Consortium. 2007. Accessed 29 May 2008.
http://pugetsound1idar.ess.washington.edu/About_LIDAR.htm

 

Red River Basin Decision Information Network. 2008. North Dakota State University.

Accessed 29 April 2008.
http://www.rrbdin.org/data

 

Rodriguez, E., C. Morris, and J.B. Belz. 2006. A Global Assessment of the SRTM
Performance. Photogrammetric Engineering & Remote Sensing 72 (3): 249-260.

Roman, D. R. 2007. Datum Shifis and Geoid Height Models. Session C of Datum, Height
and Geodesy. Accessed 29 April 2008.
http://www.ngs.noaa.gov/GEOID/PRESENTATIONS/2007 02 24_CCPS/Roma
n_C_PLSC2007.ppt

 

Salassi, ME. and J .B. Breaux. 2006. Rice Production in Louisiana, Soybeans, Wheat and
Sorghum Production in Southwest Louisiana. 2006 Projected Commodity Costs
and Returns. Farm Management Research & Extension, Department of
Agricultural Economics & Agribusiness A.E.A. Information Series No. 236-
January 2006. LSU Ag Center Research & Extension. Accessed 26 December
2007. http://www.lsuagcentercom

 

Sanders, B.F. 2007. Evaluation of on-line DEMS for flood inundation modeling.
Advances in Water Resources. 30: 1831-1843.

SBGMAPS. 2008. Spencer B. Gross, Inc. Mapping and Aerial Photography Accessed 29
April 2008. http://www.sbgmaps.com/imagcs/Image1.ipg

 

Shortridge, A. 2006. Shuttle Radar Topography Mission Elevation Data Error and its
Relationship to Land Cover. Cartography and Geographic Information Science
33(1): 6 5-75.

Simard, M., K. Zhang, V. Rivera-Monroy, M. Ross, P. Ruiz, E. Castaneda-Moya, R.
Twilley, and E. Rodriguez. 2006. Mapping Height and Biomass of Mangrove
Forests in Everglades National Park with SRTM Elevation Data.
Photogrammetric Engineering & Remote Sensing 72 (3): 299-311.

198

Slater, J., G. Garvey, C. Johnston, J. Haase, B. Heady, G. Kroenung, and J. Little. 2006.
The SRTM Data "Finishing" Process and Products. Photogrammetric
Engineering & Remote Sensing 72 (3):237-247

Stoker, J ., S.K. Greenlee, D.B. Gesch, and J .C. Mening. 2006. CLICK: The New USGS
Center for Lidar Information Coordination and Knowledge. Photogrammetric
Engineering & Remote Sensing. 613-616.

Stoker, J ., J. Parrish, D. Gisclair, D.Harding, R. Haugerud, Ma. Flood, H.Andersen, K.
Schuckman, D. Maune, P. Rooney, K. Waters, A. Habib, E. Wiggins, '
B.Ellingson, B.Jones, S. Nechero, A. Nayegandhi, T. Saultz, and G. Lee. Report
of the First National Lidar Initiative Meeting. February 14-16, 2007, Reston, Va.
US. Geological Survey. Accessed 8 July 2008.
http://lidar.cr.usgs.gov/presentations/NLImeetingReport-final .pdf

Street, J .E., and T. Miller. 2004. Rice Production, Mississippi State
University.Mississippi State University. Publication 1427. Extension Service of
Mississippi State University. Accessed 8 January 2008.
http://msucares.com/pubs/publications/p1427.htm

United States Geological Survey. USGS. 1999. National Elevation Dataset. Sioux Falls,
SD, EROS Data Center. Accessed 23 September 2007.

United States Geological Survey. USGS. 2003. National Land Cover Database Zone 37B
Land Cover Layer. 2003. Sioux Falls, SD, EROS Data Center. Accessed 23
September 2007.

United States Geological Survey. USGS. 2007. Seamless Data Distribution System,

Earth Resources Observation & Science (EROS). Accessed 2 April 2007.
http://seamless.usgs.gov/website/seamless/viewer.php

United States Department of Agriculture. USDA. 2005. Accessed 26 December 2007.
http://www.nass.usda.gov/Charts and Maps/Crops County/2005/images/A
R-YlOS-RGBChor.gif.

 

United States Department of Agriculture. USDA. 2006. National Agriculture Statistics
Service's 1:100,000-sca1e 2005 Cropland Data Layer, A Crop-Specific Digital
Data Layer for Louisiana, 2006 July 3. USDA (United States Department of
Agriculture (USDA), National Agriculture Statistics Service (NASS). Geospatial
Data Gateway. Last Accessed 6 July 2008. http://datagatewaynrcsusda.gov/

199

 

United States Department of Commerce, US. Census Bureau, Geography Division. 2006.
TIGER/Line 2006 Second Edition, Hydro graphy of Louisiana, Geographic
NAD83, CENSUS (2006). Accessed 16 July 2008.
http://lagic.lsu.edu/data/losco/tiger 1a hvdrographv CENSUS 2006.2ip

 

University of North Carolina 2007. UNC Carolina Population Center. Nang Rong. Last
accessed 29 December 2007.
http://www.cpc.unc.edu/proiccts/nargong

 

Valeriano, M., T. Kuplich, M. Storino, B. Amara], J. Mendes, and D. Lima. 2006.
Modeling small watersheds in Brazilian Amazonia with shuttle radar
topographic mission-90 m-data. Computers & Geosciences 32: 1169-1181.

Verstraeten, G. 2006. Regional scale modeling of hillslope sediment delivery with
SRTM elevation data. Geomorphology. 81: 128-140.

Wagner, M. 2007. SRTM DTED Format: Product Description: SRTM/PD-01/03, version.
DLR.

Walker, W, S., J. M. Kellndorfer, and LE. Pierce. 2007. Quality assessment of SRTM C-
and X-band interferometric data: Implications for the retrieval of vegetation
canopy height. Remote Sensing of Environment 106: 428-448.

Walsh, 8., T‘. Crawford, W. Welsh, and K. Crews-Meyer. 2001. A multiscale analysis of
LULC and NDVI variation in Nang Rong district, northeast Thailand.
Agriculture, Ecosystems, and Environment 85: 47-64.

Wang, Y., M. Liao, G.Sun, and J. Gong. 2005. Analysis of the water volume, length, total
area and inundated area of the Three Gorges Reservoir, China using the SRTM
DEM data. International Journal of Remote Sensing. 26: (18, 20): 4001-4012.

Watershed Concepts. 2004a. Digital Elevation Model (USGS DEM), Task Area 14 -
New Iberia, Louisiana, Louisiana/ Federal Emergency Management Agency
(FEMA) Project - Phase 3 of Louisiana LIDAR Data Development: St. Martin
and Iberia Parishes, Louisiana under the Watershed Concepts contract number
EMT-2002-CO-0048. [dem_######%%.dem].

Watershed Concepts. 2004b. Digital Elevation Model (USGS DEM), Task Area 15 -
Acadia / Lafayette, Louisiana, Louisiana / Federal Emergency Management
Agency (F EMA) Project - Phase 3 of Louisiana LIDAR Data Development:
Acadia and Lafayette Parishes, Louisiana under the Watershed Concepts contract
number EMT-2002-CO-0048. [dem_#######%%.dem].

Watershed Concepts. 2004c. Digital Elevation Model (USGS DEM), Task Area 16 -
Jefferson Davis, Louisiana, Louisiana / Federal Emergency Management Agency

200

(FEMA) Project - Phase 3 of Louisiana LIDAR Data Development: Jefferson
Davis Parish, Louisiana under the Watershed Concepts contract number EMT-
2002-CO-0048. [dem_#######%%.dem].

Watershed Concepts. 2005. Digital Elevation Model (USGS DEM), Task Area 25 -
Shreveport-Minden, Louisiana, Louisiana / Federal Emergency Management
Agency (FEMA) Project - Phase 4 of Louisiana LIDAR Data Development:
Caddo, Bossier, & Webster Parishes, Louisiana under the Watershed Concepts
contract number EMT-2002-CO-0048. [dem_##i#¢##%%.dem].

Waterhsed Concepts. 2006a. Digital Elevation Model (USGS DEM), Task Area 28 -
Springhill, Louisiana, Louisiana/ Federal Emergency Management
Agency (FEMA) Project - Phase 5 of Louisiana LIDAR Data Development:

Bossier / Webster Parishes, Louisiana under the Watershed Concepts contract
number EMT-2002-CO-0048. [dem_#######%%.dem].

Watershed Concepts. 2006b. Digital Elevation Model (USGS DEM), Task Area 29 -
Vivian, Louisiana, Louisiana/ Federal Emergency Management Agency (FEMA)
Project - Phase 5 of Louisiana LIDAR Data Development: Bossier and Caddo
Parishes, Louisiana under the Watershed Concepts contract number EMT-2002-
CO-0048. [dem_#######%%.dem].

Webster Parish. 2001. Webster Parish Convention and Visitors Bureau. Accessed 2
January 2007. http://visitwebster.com/aboutwebster.htm

Weydahl, D.J., Sagstuen, J., Dick, OB, and Ronning, H. 2007. SRTM DEM accuracy
assessment over vegetated areas in Norway. International Journal of Remote
Sensing 28 (16): 3513-3527.

Wilson, J. P., and J .C. Gallant (Eds.). 2000. Terrain Analysis: Principles and
Applications. New York: John Wiley & Sons, Inc.

Wood, William B. 2004. American Geography and International Research: A
Sustainable-Development Agenda. The Professional Geographer 56 (1): 53-61.

Yastiki, N., G. Kocack, and G. Buyuksalin, 2006. International Society for
Photogrammetry and Remote Sensing. (ISPRS) Ankara Workshop 2006.
Topographic Mapping from Space (with Special Emphasis on Small Satellites).
Ankara, Turkey, Feb 14-16, 2006. Last updated Feb 11, 2006.
http://www.isprs.org/commission1/ankara06/makaleler/Yastikli_DEM-
Accuracypdf

Zhang, H.P., S.F. Liu, N. Yang, Y. Zhang, and G. Zhang. 2006. Geomorphic
characteristics of the Minjiang drainage basin (eastern Tibetan Plateau) and its
tectonic implications: New insights from a digital elevation model study. Island
Are: 15: 239-250.

201

 

 
 

lllllllilll

6 8973

 

uljrljjjrjjjj