Group Title: comparison of the Normalized Difference and the Tasseled Cap Vegetation Indices
Title: A comparison of the Normalized Difference and the Tasseled Cap Vegetation Indices
CITATION PDF VIEWER THUMBNAILS PAGE IMAGE ZOOMABLE
Full Citation
STANDARD VIEW MARC VIEW
Permanent Link: http://ufdc.ufl.edu/UF00100779/00001
 Material Information
Title: A comparison of the Normalized Difference and the Tasseled Cap Vegetation Indices a case study of using satellite remote sensing imagery for assessment of environmental impact of a hydroelectric power project on the River Danube
Physical Description: Book
Language: English
Creator: Aufmuth, Joseph L
Publisher: University of Florida
Place of Publication: Gainesville Fla
Gainesville, Fla
Publication Date: 2001
Copyright Date: 2001
 Subjects
Subject: Civil and Coastal Engineering thesis, M.S   ( lcsh )
Dissertations, Academic -- Civil and Coastal Engineering -- UF   ( lcsh )
Genre: government publication (state, provincial, terriorial, dependent)   ( marcgt )
bibliography   ( marcgt )
theses   ( marcgt )
non-fiction   ( marcgt )
 Notes
Summary: ABSTRACT: In the fall of 1992, water from a section of the Danube River border between Hungary and Slovakia was diverted to an adjacent hydroelectric power system known as the Gabcikovo Barrage System (GBS). Originally, hydropower production in the common reach was a joint goal of the two countries. In 1989, Hungary voided the original 1977 agreement, sighting potential adverse environmental effects. Slovakia finished the project alone and the conflict was brought before the International Court of Justice. This study focused on a part of the Szigetköz (Hungarian side) and Csalloköz (Slovakian side) regions along the Danube that are located between two flood dikes. This area contains a unique wetland system with numerous river branches immediately downstream from the diversion. Four single remotely sensed late summer Landsat satellite images from 1988, 1992, 1993, and 1997, were used to detect, measure and compare changes in vegetation cover and condition with environmental moisture throughout the study area and three regional forested controls. The 1988 and 1992 images represented pre-diversion conditions, and the 1993 and 1997 images represented post-diversion conditions. Integrated supervised and unsupervised satellite image classification techniques produced 4 study area land cover classes; water, forest, grass, and exposed. Average regional monthly and annual rainfall records were collected. For each study area and control image, satellite derived Normalized Difference Vegetation Index (NDVI) and Tasseled Cap greenness (GVI) were used to indicate plant condition and Tasseled Cap wetness (WI) was used to quantify environmental moisture.
Summary: ABSTRACT (cont): Three separate zonal images, country, water buffer distance, and region, were created for study area analysis. Map algebra was used to combine the separate zonal images with each of the 4 land cover class images, which resulted in four new images each containing 144 unique analysis zones. The mean NDVI, GVI and WI value per unique analysis zone was calculated and weighted by the number of pixels per zone, or histogram. Mean weighted index values for the forested study area and water buffer zones were compared to forested control index values and average rainfall. Additional mean histogram weighted zonal NDVI, GVI and WI relationships were investigated for the study area. Results showed study area and control WI patterns were similar and followed rainfall patterns. Higher average rainfall corresponded with lower WI values. WI values for the study area were higher than controls, except for 1988, but exhibited the same trends as controls. Control NDVI and GVI values were higher than or equal to study area values. For study and control areas NDVI and GVI values, and NDVI and WI values were highly correlated. This study concluded that a number of important, but limited, environmental changes are detectable from the satellite imagery and so the imagery provides a suitable means to monitor future changes in the region.
Summary: KEYWORDS: NDVI, Tasseled Cap, GVI, WI, Danube River, dam, precipitation
Thesis: Thesis (M.S.)--University of Florida, 2001.
Bibliography: Includes bibliographical references (p. 76-81).
System Details: System requirements: World Wide Web browser and PDF reader.
System Details: Mode of access: World Wide Web.
Statement of Responsibility: by Joseph L. Aufmuth.
General Note: Title from first page of PDF file.
General Note: Document formatted into pages; contains xv, 82 p.; also contains graphics.
General Note: Vita.
 Record Information
Bibliographic ID: UF00100779
Volume ID: VID00001
Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
Resource Identifier: oclc - 47889632
alephbibnum - 002728642
notis - ANK6404

Downloads

This item has the following downloads:

thesis ( PDF )


Full Text











A COMPARISON OF THE NORMALIZED DIFFERENCE AND THE TASSELED
CAP VEGETATION INDICES: A CASE STUDY OF USING SATELLITE REMOTE
SENSING IMAGERY FOR ASSESSMENT OF ENVIRONMENTAL IMPACT OF A
HYDROELECTRIC POWER PROJECT ON THE RIVER DANUBE
















By

JOSEPH L. AUFMUTH


A THESIS PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
MASTER OF SCIENCE

UNIVERSITY OF FLORIDA


2001




























Copyright 2001

by

Joseph L. Aufmuth


























For all of the endless days and nights spent waiting; For all of those "Ground Hog Days";
For being interested in what I was doing; For believing in us; and most of all, For being
my wife and friend of 20 years, this is dedicated to Marcia and our children, Kitty, Annie,
Claudius, Marge, Jack, Andy and Robbie.















ACKNOWLEDGMENTS

In accomplishing this six-year project there were many individuals, familiar and

professional, who contributed support, insight and understanding. The support and

prayers of immediate and extended family reinforced the blessings given to me by God.

To individual Hungarian researchers and new friends, Lajos Horvath, Gyorgy

Bittner, Ferenc Szilagyi, and Zoltan Somogyi, as well as the following Hungarian

agencies, the North-Transdanubian Environmental Protection Authority, FOMI Remote

Sensing Center, and the Hungarian Forest Research Institute, I am deeply grateful for

your generous assistance in making available local field interpretations, research data and

satellite imagery.

For their tireless support and encouragement towards the completion of this

document, I would like to acknowledge and thank the University of Florida's George A.

Smathers Libraries, the Government Documents Library team, and specifically,

Government Documents department chair, Jan Swanbeck.

I will remember forever the faculty, staff and students of the geomatics program

for instilling an appreciation of the elements of the survey and mapping profession and its

relationship to GIS and remote rensing. I am further indebted to my friend, and fellow

graduate student Mark Lee for helping guide my "real world" perspective on spiritual,

mental and physical health during times of trial.

The patience and guidance of committee members Dr. Bon Dewitt and Dr.

Grenville Barnes is greatly appreciated. Their commitment to excellence was an









inspiration toward further developing my research topic, and interpreting and presenting

the data. My deepest thanks and heartfelt appreciation are reserved for committee

member and chair Dr. Scot Smith. He continually supported my academic development

and research, and most of all provided his valued friendship.
















TABLE OF CONTENTS

page

A C K N O W L E D G M E N T S ................................................................................................. iv

L IS T O F T A B L E S ................................. ...................................... ............ .. ................... v iii

L IST O F F IG U R E S .............................. ............................... ..................................... x

A B S T R A C T ..................................................................................... x iii

CHAPTERS

1 IN TRODU CTION .................................. ....... ... ........ .... ............. .

B background and R ationale for Study...................................... ...................... .. .......... 1
Stu dy A rea D description ......................................................... .................................. 5
Problem Statement and Objectives .......................... ......... .. ....... .... 6

2 L ITE R A TU R E R E V IEW ....................................................................... ..................10

Spectral and Spatial Resolution of the Satellite Sensor............................. ............. 11
Im ag e R ectific atio n ............................................................................... .................. 12
Im age to Im age N orm alization ....................... .............................. ............ .............. 13
A tm ospheric C correction .......................................................... .......................... 13
Seasonal Variability of Vegetation Canopy.............................................................. 14
H um an A activities ........................................ 14
Change D election A nalysis........................................................ .......................... 15
M ap A lgebra ......................................................... ..... ...... 15
NDVI and Tasseled Cap Vegetation Indices ......................................................... 16
Previous GB S Environm ental Studies ........................................ ...................... 18

3 MATERIALS AND METHODS......................................................... ...............21

Im ag ery .................................................... ......... ..... 2 4
Softw are and H ardw are.............................................................. .......................... 24
Image Pre-processing ......... ............................................... ... 24
R adiom etric C orrection ........................................................ ........... .............. 25
A tm ospheric C orrection ...................................................... .......................... 26
G eom etric C orrection .................................................... ........ ............. ....... 27
Im age P ost-P processing ...... ...... ............ ........................................................ 30









Im age C classification .......................... ....... ................ ... ....... .. ............ 3 1
V egetation and M oisture Indices ........................................ .................... ..... 33
Statistical A nalysis............................................ 34
C country Z ones ................................................................................................... 34
Buffer zones .............................................. 35
R regional zones ................................................................................................. 35
M ap A lgebra .................................................... 36
Statistics ........................................................................................................................ 37
Precipitation Data............................................. .............. 39

4 RESULTS AND DISCU SSION ................................. ....... ..............................44

Image Classification............................ ........... 44
Anecdotal Accuracy Assessment ....................... ................. .......... 46
Analysis of Land Cover Changes ......................................................... ....... 47
Vegetation and Moisture Index Images ................. ........ .............. 50
NDVI and GVI Im ages ................................. .......................... .. ........ 51
W I Im ag es ..................................................... 54
Observations and Comparisons............................ .............. 56
Precipitation Com prisons .............................. .......................... ............ 56
Control NDVI and GVI Observations and Comparisons ....................................... 57
C control W I O bservations..........................................................................59
Control GVI and W I Trends .............................................. .... .. ........ ............... 60
Control and Study Area Forest Index Value Comparison ..................................... 61
Control and Water Buffer Distance Comparison ............................................... 62
Control Com prison Sum m ary .............. ....................................................... 64
Study Area Indices by Buffer Distance ........................................ 64
Study Area Indices By Land Cover ........................................................... 68
Study Area Indices by Region ................................... .............. .............. 69
Study A rea Indices by C ountry...................................................................... ...... 71

5 C O N C L U SIO N S ........................................................................................ . 73

LIST O F R EFER EN CE S ............................................................ ...... .....................76

BIOGRAPHICAL SKETCH ................................................................ ............ 82
















LIST OF TABLES


Table Page

1 T asseled C ap M ultiplicative M atrix......................................................................... ..... 17

2 Tasseled Cap Additive M atrix. ...... ........................... ........................................17

3 Previous Rescaled NDVI Values (Smith, et al, 2000). ................................................ 21

4 Meteorological Conditions in the Study Area at Time of Satellite Overpass...................25

5 1988 Rectification Param eters. ...... ........................... .......................................... 28

6 V isual Interpretation Standard. ........................................ .............................................32

7 Digital Numbers by Image Band for a Sample pixel ......................................................34

8 Software vs. Hand Calculated Tasseled Cap Values. ........................................ ..............34

9 An Example of 1988 Histogram Weighted GVI Means (WTMN) by Zone. ..................38

10 Regional Monthly Precipitation Values Surrounding Image Dates.............................43

11 Average Monthly Precipitation Prior to Image Date. ................... ............................. 43

12 Study Area Land Cover Classes (ha and percent cover), 1988-1997. ..........................45

13 Szigetkoz Region Land Cover Classifications ...............................................45

14 Land Cover Classification Comparisons. .............................................. ............... 47

15 Annual Precipitation for the Water-Years 1989 to 1997. ..............................................56

16 M ean W eighted GVI and NDVI Values ........................................ ....... ............... 59

17 Histogram Weighted Mean Study Area Forest Indices to Control Indices Correlations.61

18 Vegetation Index to Wetness Index Correlations by Year.............................................67

19 Index Correlations Across Regions. ........................................ .......................... 71









20 Correlations W within Year and Across Regions..................................... ............... 71

21 M ean W eighted Index Values....................................................................... 72
















LIST OF FIGURES


Figure Page

1 Hungarian-Slovakian Danube River Border and Study Area Location ...........................

2 GBS System and the Surrounding Region................................. ...............4

3 Flow Rates of the Old Danube at Rajka, m3/s (Smith et al., 2000). .............................7

4 Water Levels at Rajka 1992 to 1998 (Smith et al., 2000).......................................8

5 Control and Study Area Locations........... ......................... .. ............... 22

6 Northern Controls 1988, 1992, 1993, and 1997....................................... 23

7 South-Central Controls 1988, 1992, 1993, and 1997.............. ................... ............... 23

8 Southern Controls 1988, 1992, 1993 and 1997 ............... ............................ 23

9 1997 Pre-Haze Reduction Cloud Cover and Shadow .............................................. 26

10 1997 Post Haze Reduction Image. .......... ................. ........................... 27

11 1988 False Color Composite (TM Bands 4,3,2). ....................................................... 29

12 1992 False Color Composite (TM Bands 4, 3, 2). ................................................. 29

13 1993 False Color Composite (TM Bands 4, 3, 2). ..... ..... ....................................... 30

14 1997 False Color Composite (TM Bands 4, 3, 2). ..... ..... ....................................... 30

15 Study Area Land Cover Classifications ........................................ ........ ............... 33

16 C country Statistical Z ones ........................................................................ ...................35

17 W after Buffer Statistical Zones ........... ................. ............................ ............... 35

18 R regional Statistical A analysis Zones........................................... .......................... 36

19 1988 M ap Algebra and Zonal Image Creation. .................................... .................37









20 Zonal Statistics Im age Exam ple, 1988....................................... .......................... 37

21 MSN Excel's Linear Correlation Formula. ........................................... ............... 39

22 Precipitation Recording Stations ...................................................................... 40

23 1988 M monthly Precipitation Totals. ............................................................................ 40

24 1992 M monthly Precipitation Totals. ....................................................... ............... 41

25 1993 M monthly Precipitation Totals. ............................................................................ 41

26 W ater Y ear Precipitation Totals ........................................... ..... ......................... 42

27 Pre- and Post-Danube Diversion, 1992 and 1993, Respectively. ....................................44

28 Percent Land Cover Composition by Year. ........................................ ............... 46

29 1993 W eir C construction .......................................................................... ................... 49

30 1997 Post W eir Construction ........... .................. ........ .................. ............... 49

31 Study Area and Control Example Image Locations. ...................................................... 50

32 Sample Study Area NDVI Images 1988, 1992, 1993 and 1997 ..................................51

33 Sample Study Area GVI Images 1988, 1992, 1993 and 1997 .....................................52

34 Southern Control Area NDVI Images 1988, 1992, 1993, 1997................. .......... 52

35 Southern Control Area GVI Images 1988, 1992, 1993, 1997. ..................................53

36 Sample Study Area WI Images 1988, 1992, 1993 and 1997 ......................................54

37 Southern Control Area WI Images 1988, 1992, 1993, 1997. ....................................55

38 M ean Control Area NDVI per Year ...................................................... .............. 58

39 M ean Control Area GVI per Year. ............................................................................58

40 Yearly M ean Control W I Levels........................................................ ............... 60

41 M ean NDVI, GVI and W I Comparison........................................ ....................... 60

42 Y early C control G V I M eans.............................................................................. ............62

43 Mean Control and Weighted Mean Forest WI Values................... .......................... 62

44 Mean Control, Weighted Mean Forested and Buffer Distance Forested GVI Values.....63









45 Mean Control, Weighted Mean Forest and Buffer WI Values. .......................................64

46 GVI Means by Buffer Distance from Water...............................................................65

47 WI Means By Buffer Distance From Water. ...................................... ............... 65

48 1988 Mean GVI vs. WI By Buffer Distance To Water. ...............................................66

49 1992 Mean GVI vs. WI By Buffer Distance To Water. ................................................66

50 1993 Mean GVI vs. WI By Buffer Distance To Water. ...............................................67

51 1997 Mean GVI vs. WI By Buffer Distance To Water. ................................................67

52 M ean W ater Class Index V alues.............................................. ............................ 68

53 M ean Forest Class Index V alues........................................................... ............... 68

54 M ean G rass Class Index V alues. .......................................................... .....................69

55 M ean Exposed Class Index V alues........................................... ........................... 69

56 U pper R egion M ean Index V alues........................................................ ............... 70

57 M iddle Region W eighted M ean Index Values...................................... ............... 70

58 Lower Region Weighted Mean Index Values....................................... ............... 70

59 Index Correlations between Countries ........................................................ ...............72















Abstract of Thesis Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Master of Science

A COMPARISON OF THE NORMALIZED DIFFERENCE AND THE TASSELED
CAP VEGETATION INDICES: A CASE STUDY OF USING SATELLITE REMOTE
SENSING IMAGERY FOR ASSESSMENT OF ENVIRONMENTAL IMPACT OF A
HYDROELECTRIC POWER PROJECT ON THE RIVER DANUBE

By

Joseph L. Aufmuth

May 2001

Chairman: Dr. Scot E. Smith
Major Department: Civil and Coastal Engineering

In the fall of 1992, water from a section of the Danube River border between

Hungary and Slovakia was diverted to an adjacent hydroelectric power system known as

the Gabcikovo Barrage System (GBS). Originally, hydropower production in the

common reach was ajoint goal of the two countries. In 1989, Hungary voided the

original 1977 agreement, sighting potential adverse environmental effects. Slovakia

finished the project alone and the conflict was brought before the International Court of

Justice.

This study focused on a part of the Szigetkoz (Hungarian side) and Csallokoz

(Slovakian side) regions along the Danube that are located between two flood dikes. This

area contains a unique wetland system with numerous river branches immediately

downstream from the diversion. Four single remotely sensed late summer Landsat

satellite images from 1988, 1992, 1993, and 1997 were used to detect, measure and









compare changes in vegetation cover and condition with environmental moisture

throughout the study area and three regional forested controls. The 1988 and 1992

images represented pre-diversion conditions, and the 1993 and 1997 images represented

post-diversion conditions.

Integrated supervised and unsupervised satellite image classification techniques

produced 4 study area land cover classes; water, forest, grass, and exposed. Average

regional monthly and annual rainfall records were collected. For each study area and

control image, satellite derived Normalized Difference Vegetation Index (NDVI) and

Tasseled Cap greenness (GVI) were used to indicate plant condition and Tasseled Cap

wetness (WI) was used to quantify environmental moisture. Three separate zonal images,

country, water buffer distance, and region, were created for study area analysis. Map

algebra was used to combine the 3 separate zonal images with each of the 4 land cover

class images, which resulted in four new, 144 zone images. The mean NDVI, GVI and

WI value per unique analysis zone were calculated and weighted by the number of pixels

per zone, or histogram. Mean weighted index values for the forested study area and

water buffer zones were compared to forested control index values and average rainfall.

Additional mean histogram weighted zonal NDVI, GVI and WI relationships were

investigated for the study area.

Results showed study area and control WI patterns were similar and followed

rainfall patterns. Higher average rainfall corresponded with lower WI values. WI values

for the study area were higher than controls, except for 1988, but exhibited the same

trends as controls. Control NDVI and GVI values were higher than or equal to study area









values. For study and control areas NDVI and GVI values, and NDVI and WI values

were highly correlated.

This study concluded that a number of important, but limited, environmental

changes are detectable from the satellite imagery and so the imagery provides a suitable

means to monitor future changes in the region.














CHAPTER 1
INTRODUCTION

A hydroelectric dam located in Slovakia, near the Danube River border with

Hungary, was completed in 1993. It's commissioning resulted in domestic and

international concern over its potential environmental impacts. Since the project diverts

water from the two countries' shared river border, much of the Hungarian concern was

over downstream impacts to a large forested wetland system, the Szigetkoz.

This thesis examined the validity of some of those concerns using Landsat TM

imagery as its basis. The study was part of a larger effort to understand the dam's

environmental impact in its entirety. This particular research effort focused on forested

wetland composition changes and vegetation response related to moisture conditions in

the area surrounding the Danube.




Background and Rationale for Study

As a result of the 1920 Trianon Treaty, the Danube forms the international

boundary between Hungary and Slovakia (formerly Czechoslovakia) (Figure 1). There

are 33 hydroelectric dams on the Danube from its headwaters in Germany to its delta in

Romania (ICPDR-PCU, 2000), but none previously resided in the portion of the river

shared by Hungary and Slovakia.



















Szkes ehrv r"
P / -:,.


.Miskole



BUDAPEST Debrecen.

HUNGRY

S Kecskemet


ROMANIA


f--' S~Szged. ---'
0 40 0 km
0 40 0wt
CROATIA Serbia
CA S era Source: CIA World Fact Book, 2000

Figure 1 Hungarian-Slovakian Danube River Border and Study Area Location.


In October of 1992, along a 58 kilometer (km) portion of the Danube River

between Rajka in Hungary and Sap in present day Slovakia, Czechoslovakia diverted

85% of the water from the river into a 27 km long concrete hydroelectric power channel.

This was one of only a few instances in modern history when an upstream country

unilaterally diverted a major river representing a natural border between countries (Smith

et al., 2000, Kurland et al., 1992). Other major ongoing international water disputes

involving dams exist between Iraq, Iran, Turkey, and Syria over the Tigris and Euphrates

River (Trade and Environment Database, 2000), and between Laos, Thailand, China,

Cambodia, Vietnam, and Myanmar, over the Mekong River (Samson and Charrier,

1997). However, the situation in Hungary and Slovakia caught international attention

when it ended up as the first environmental lawsuit before the International Court of

Justice ( aka the World Court) (International Court of Justice, 1997). As water becomes a

scarcer commodity these types of conflicts are very likely to escalate in the future (Smith

and Al-Rawaby, 1990).









Under provisions of a treaty agreed to in 1977, the construction and operation of a

two dam system for hydropower production, the Gabcikovo-Nagymaros Barrage System

(GNBS), was a joint agreement between Hungary and Czechoslovakia. One major dam

was built at Gabcikovo in eastern Czechoslovakia and the other would have been built

downstream at Nagymaros, Hungary. The entire project, presently known as the

Gabcikovo Barrage System (GBS), had four major objectives: (i) hydroelectric power

production (ii) improved navigation on a reach of the Danube (iii) reliable upstream

water supply and, (iv) economic development. In this region, flood dikes protect the

adjacent lands from the Danube on the Hungarian and Slovakian sides. The dikes

enclose a temperate forested flood plane called the Szigetkoz on the Hungarian side and

the Csallokoz (Zitny Ostrov in Slovakian) on the Slovakian side. A map of the GBS and

surrounding region is shown in Figure 2.

In May 1989, a controversy arose after Hungary, stating concerns over economic

viability and potential environmental impacts, suspended project participation. In June,

citing the need for further studies, the Hungarians stated "Having studied the expected

impacts of the construction in accordance with the original plan, the Committee [ad hoc,

set up for this purpose] of the Academy [Hungarian Academy of Sciences (HAS)] came

to the conclusion that we do not have adequate knowledge of the consequences of

environmental risks. . Thorough and time consuming studies are necessary" (ICJ,

1997).
































Figure 2 GBS System and the Surrounding Region.


On May 19, 1992 Hungary terminated the 1977 Treaty. Czechoslovakia

continued constructing a variation of the original plan, not requiring Hungarian

participation and, in 1996, Slovakia, now independent from the Czech Republic,

completed the project.

The Hungarian and Slovakian governments jointly brought the case before the

ICJ. The Hungarian government's position was that

Insufficient engineering and environmental studies were performed before the

construction of the GBS;

The source of drinking water for Budapest is located in the gravel plateau of the

Szigetkoz region;

A valuable forested wetland in the river's floodplain might be adversely impacted;









A potential ground water level decrease might cause a decline in local agricultural

production;

International law and policy issues regarding trans-boundary biodiversity were in

doubt (Dobson, 1992, Chelminski, 1993).



The Slovakian government's position was that

Adequate assessment of the project's environmental side effects took place.

The potential environmental impacts were outweighed by the benefits afforded by

flood protection, hydroelectric power production and improved navigation.



On September 25, 1997, the ICJ concluded that, while it was illegal for Hungary

to break the formal 1977 agreement to jointly build the GNBS, it was also illegal for

Czechoslovakia to divert the Danube unilaterally in 1992 and that the 1977 Treaty was

transferable from Czechoslovakia to Slovakia. The judgment obligates both parties to

take all necessary measures to ensure the achievement of the objectives of the September

16, 1977 Treaty and, according to the Court "the Parties together should look afresh at

the effects on the environment of the operation of the Gabcikovo power plant"

(International Court of Justice, 1997). A headline in the New York Times summed up

the decision: "World Court Leaves Fight Over Danube Unresolved" (Perlez, 1997).


Study Area Description

The study area shown in Figure 2, falls into the common Hungarian and

Slovakian stretch of the Danube located between Rajka and Sap. The area includes the

floodplain situated between the flood protection dikes in this river stretch. The study area

belongs to the 232,700 hectares (ha) Little Lowland Unit. The Little Lowland contains









two sub-areas, the 52,700 ha Hungarian Szigetkoz Region and the 180,000 ha Slovakian

Csallokoz Region. The total 10,100 ha study area represents 4.3 percent of the Little

Lowland Unit. Of the Total, 5,200 ha of the study area cover approximately 10 percent of

the Hungarian Szigetkoz and the remaining 4,900 ha cover approximately 3 percent of

the Slovakian Csallokoz.

One of the most important environmental factors in this region is the groundwater

level. The forestry and the agriculture of the Szigetkoz's Little Lowland Unit are based

on a very thin soil layer. A deep gravel bed, which has high water conductivity, exists

under this soil layer and forms one of Europe's largest freshwater reservoirs.

The study area delineation was based on it being directly downstream of the

diversion, having important environmental and commercial floodplain forests and

occurring between flood control dikes along the entire length. The area is a mixture of

natural wooded wildlife habitat and forestry tree plantations that were first established at

the end of the 1800s. Commercial forestry operations consisting of thinning, clear

cutting and selective harvesting are conducted throughout the region in both Hungary and

Slovakia.




Problem Statement and Objectives

Electrical demand governs continued operation of the dam for hydro-electric

power. Peak and off-peak electrical demand results in repeated cycles of river diversion

or flow. The continued Danube river diversion results in fluctuations in hydro-period, or

changes in the amount and duration of water in the forested wetland system. The hydro-

period fluctuations are hypothesized to cause quantifiable changes in vegetation condition

within the forested Szigetkoz and Csallokoz regions and consequently, changes in land









cover composition. In fact, reduced and erratic water flow downstream of the diversion

has been observed during and after diversion to the GBS power channel, (Figure 3 and

Figure 4).

Hungary's North-Transdanubian Environmental Inspectorate (NTDEI) provided

Figure 3 and Figure 4. The tabular data used to create the figures was not available for

this research. Trends prior to 1992 are not represented. The NTDEI provided an

interpretation of their data and figures, which concluded: 1) After the diversion in late

1992, flow rates dropped until 1995 when lateral channel weirs were constructed to

increase water supply (Figure 3); 2) In late 1992, a 4 m post diversion drop in water level

occurred at Rajka (Figure 4); and 3) Due to a 1995 agreement between Hungary and

Slovakia, less water was diverted from the Danube and water levels at Rajka increased

-300-400 m3/s (Figure 4).


3000


Diversion Water supply To Laterals
2000



1000




1992 1993 1994 1995 1996 1997

Figure 3 Flow Rates of the Old Danube at Rajka, m3/s (Smith et al., 2000).










H [ cm 1 000001 Rajka
Duna
450
400
35 Diversion 11/92

250 -I --
200
150


100 L __-
-50 IDJ
-58


-150
-200
-250
-30088
-358 IDd
1992 1993 1994 1995 1996 1997 1998
Jan-01 Jan-01 Jan-81 Jan-O1 Jan-O1 Jan-01 Jan-01

Figure 4 Water Levels at Rajka 1992 to 1998 (Smith et al., 2000).




In order to quantify forest vegetation responses solely due to water diversion,

naturally occurring environmental moisture conditions, such as precipitation or drought,

should be identified in forested control areas away from the diversion. Control area

vegetation response associated with naturally occurring moisture conditions should also

be measured. Simultaneous measurement of study area moisture condition and

vegetation response and comparison with control measurements would help identify

moisture related affects strictly due to the diversion. Monitoring studies such as these are

vital to implementing the ICJ decision.

Using Landsat Thematic Mapper (TM) imagery, this study compared control and

study area vegetation indices, Normalized Difference Vegetation Index (NDVI) and

Tassel Cap Greenness (GVI), with a moisture index, Tasseled Cap Wetness (WI), for pre-

and post-dam periods. Specifically, the three research objectives were: 1) Detect and

compare pre and post diversion land cover composition; 2) Detect control and study area









plant condition and hydro-period change using NDVI, GVI and WI as estimators of

vegetation condition and environmental moisture respectively; and 3) Compare and

potentially correlate control and study area vegetation and moisture indices and 4)

Identify trends in GVI and WI.

The research questions asked were as follows:

* Are WI values, as a measure of surface moisture, similar to precipitation patterns?

* Are study area forest NDVI, GVI and WI patterns similar to Control areas?

* Are GVI and NDVI correlated?

* Are there correlations between WI and NDVI or GVI?

* Are effects detectable by 1) Distance to Water, 2) Land Cover Class, 3) Region, or 4)

Country.














CHAPTER 2
LITERATURE REVIEW

Vegetation change, specifically loss of tree canopy due to changing

environmental conditions, is a global concern (Forseth, 1997, Fisher and Levine 1999,

Petch and Kolajka, 1993, Pineda, 1992, Anonymous, 1993, Erlich and Wilson, 1991,

Lauver and Whistler, 1993). The change can be a result of natural disasters, infestations

or societal alterations of the environment. As environmental conditions (surface water,

hydrology, soil moisture, nutrients, weather, etc.) change, vegetation can become stressed

(Lichtenthaler, 1996). Sufficient and constant stress resulting in extreme plant cellular

structure and chlorophyll change can eventually lead to plant mortality (Forseth, 1997,

Kay, 1991).

Measurable reflected sunlight energy (reflectance) of a plant, both visible and

near infrared wavelengths is based upon its chlorophyll content and cellular structure

(Lillesand and Kiefer, 1994). In response to environmental conditions, change in the

plant's physiology (i.e. cellular structure and chlorophyll content) affects the measurable

reflectance values. Consequently, as surrounding environmental conditions change, the

reflected sunlight energy from the earth's surface changes. Since remote sensing

satellites record reflectance values, remote sensing is an ideal tool for detecting and

quantifying environmental and vegetation health change (Muchoney and Hacck, 1994,

van Leeuwen and Huete, 1996, Todd and Hoffer, 1998, Todd et al., 1998, Gao et al.,

2000, Serrano et al., 2000).









Several factors influence the quantification of plant health and land surface

condition using satellite imagery. The DNs from bands of a satellite image represent an

instantaneous reflectance of ground features and environmental conditions. The values

contributed by plants are a result of cyclic environmental interactions that occur

throughout its life span. Trying to relate any one instantaneous measurement with the

current vegetation condition is a difficult task. A plant's elastic response to

environmental change has been well documented (Lichtenthaler 1996). An elastic

response is a mechanism of evolution, which allows genetic survival during times of

stress. A slow response allows compensation for slight changes that, if environmental

conditions persist, culminates in a response such as senescence, or ultimately, death. A

single observation may be the culmination of a series of daily, monthly, or yearly events

and trends. Correlation of these trends with environmental conditions requires numerous

and consistent field observations and data sets. Measuring plant response and vegetation

change directly related to environmental conditions are governed by a satellite sensor's

spectral and spatial resolution, percent cloud cover, image rectification, image to image

normalization, seasonal variability of vegetation canopy, and human activities.




Spectral and Spatial Resolution of the Satellite Sensor

When this study was started, the highest combined spectral and spatial resolution

available to civilians was Landsat Thematic Mapper (TM5) satellite imagery. TM5 has 7

spectral bands and a nominal pixel size of 28.5m x 28.5m. The TM image is typically re-

sampled to 30m x 30m and sometimes 25m x 25m (Jensen 1996). The spectral

sensitivity of TM's charge-coupled device (CCD) helps detect environmental and









vegetation reflectance. The CCD's spatial resolution influences the recorded digital

number (DN). The DN of the 30m x 30m pixel represents a mixed reflectance of sub-

pixel elements contained in the corresponding 30m x 30m ground area. For example, a

small house (70ft x 26ft or 22m x 8m) and the surrounding landscape with varying

canopy sizes all contribute different reflectance values to the individual 30m x 30m pixel.

Urban examples such as this one or other highly heterogeneous systems produce mixed

reflectance values (Jensen, 1981). Therefore, it is desirable to study large homogeneous

vegetation and land use areas to avoid mixing of sub-pixel elements. However, changes

to be studied frequently occur along habitat and land use edges where reflectance values

can be influenced by the surrounding pixels (Todd and Hoffer, 1998).




Image Rectification

While it is possible to interpret, analyze and classify remotely sensed data without

image rectification, assigning a known coordinate system to the image aids in relating

recorded field data coordinates to image coordinates and performing accuracy

assessments. In order to perform change detection analysis, images must be co-registered

to each other, or have a common coordinate system. Digital number values are affected

by the method of rectification, i.e. nearest neighbor, bilinear interpolation or cubic

convolution (Smith, et. al 1995). The success of change detection analysis is directly

related to the positional correlation between multi-date images and it is reflected in the

root mean square error (RMSE), of the two images. RMSE is a measure of residual

errors, or deviations, produced during column/row to map X,Y coordinate transformation.

Rectification is critical in "change detection" studies, especially when the image dates to









be compared vary in location by a pixel or more and the change occurs along a narrow

edge (Jensen, 1996).


Image to Image Normalization

Equally important in assessing vegetation changes is normalizing, or histogram

matching of multi-date imagery. Temporally, digital numbers between image dates can

vary at the same location due to sun angle, atmospheric conditions and satellite

radiometric quality (Jensen, 1996). By matching histograms of very bright or very dark

areas in one image to the same areas in another image, multi-date image variability can

be adjusted (Chavez, and MacKinnon, 1994.).

Two different radiometric normalization methods are widely accepted: "dark and

bright objects" (DBO) and pseudo invariant features (PIF) (Jensen 1981, Lillesand and

Kiefer, 1994). Both methods are based on statistical invariance of certain scene elements

and use linear functions for scene normalization. The DBO method uses statistics of

features having time independent reflectance. The PIF method relies on man made in-

scene elements present in urban areas (Schott et al., 1988). Other methods have been

developed by image processing software developers. ERDAS Imagine offers a histogram

matching algorithm (Erdas Field Guide 4th Ed.)




Atmospheric Correction

Cloud cover affects not only the ability to sense what is beneath the clouds using

certain bands, but can also increase the DN values of adjacent pixels. Clouds scatter light

thereby increasing brightness measured from ground features. Shorter wavelengths, such

as TM 5 bands 1, 2, and 3 (blue, green, and red) scatter more than longer wavelengths









(Jensen, 1996, Lillesand and Kiefer, 1994). Consequently, adjacent pixels can be

affected because clouds often have gradations of vapor moisture leading to the cloud

center. Additionally, cloud shadow may be cast further from the adjacent pixels, thereby

darkening the brightness component other pixels. Song et al. (2001) recommended dark

object subtraction with or without further atmospheric correction or relative atmospheric

correction for classification and change detection applications.




Seasonal Variability of Vegetation Canopy

When performing change detection analysis, the seasonal variability of canopy

cover can be accounted for by performing the analysis at a consistent time period across

multiple years (Schiever and Congalton, 1995). By performing canopy analysis during

full leaf out, change should be apparent. However, if climatic conditions have been

severe (i.e. extreme drought, flood, cold, or heat) and not measured or normalized,

natural climatic events may be confused with artificially induced environmental canopy

change (Lambin, 1996). A plant species diversity study in California concluded weather

variables account for the bulk of the diversity patterns in the models used and that mean

weather variables are generally more important than seasonality or irregularity

(Richerson and Lum, 1980). Campbell et al., (2001) reported that available moisture

during dry summers was the environmental variable that limits forests the most.




Human Activities

Lastly, human activities can directly impact the ability to assess environmental

and vegetation changes. According to Anderson (1991), the natural soil-topography-









vegetation wetness system can be disturbed by human intervention in the form of canopy

cover changes and artificial drainage. In areas where trees have died, forestry practices

often cull dead trees for paper/pulp or firewood. Forestry practices may also reintroduce

new trees to replace those removed (Sader, 1995). Other effects on change detection

studies may result from efforts to correct the original environmental alterations that

caused the vegetation stress, such as adding weirs to retain water.




Change Detection Analysis

There are several approaches to measuring environmental and tree canopy

change. One involves image classification techniques (unsupervised, supervised, cluster

busting, accuracy assessment, etc.) and others involve band ratio or indexing techniques

such as normalized vegetation index (NDVI) or Tasseled Cap (Bauer, et al 1994). With

either technique, the final step is a cross year comparison of classified or index results

and a quantification of change (Green et al.,, 1994). Many studies utilizing satellite

imagery for monitoring vegetation change have been conducted and the literature

synthesized (Van Niel, 1995).




Map Algebra

Map algebra (Tomlin, 1983) refers to calculating new spatial data from the

interaction of two or more existing layers. This study used map algebra to calculate

vegetation and moisture indices from multiple bands of satellite imagery and create

geographical zones for statistical analysis. Four cartographic modeling operations,

focalfunction, incrementalfunction, local function, and zonal function, were described by









Tomlin (1990). Of the four operations, the zonalfunction uses a secondary layer to create

zones for statistical analysis of a primary, or first layer. This study applied the

zonalfunction using the index values as primary layers and the geographic zones as the

secondary layer.




NDVI and Tasseled Cap Vegetation Indices

NDVI (Rouse et al., 1974) is a ratio of reflectance bands and has been found to be

an accurate and reliable means of detecting vegetation health or vigor (Kidwell 1990) and

was reported by Tucker (1979) to respond to green biomass changes. The NDVI ratio is

expressed as NDVI = (IR-R)/(IR+R); where IR is a near infrared band (band 4 in TM5)

and R is the visible red band (band 3 in TM5) (Rouse et al., 1974). Thus, the equation

this study used to calculate NDVI for the TM imagery is NDVI = (TM4-

TM3)/(TM4+TM3). NDVI values range between -1 and 1. In theory, healthy, or

vigorous plants have turgid cells and high chlorophyll. Because of the turgid cell

structure, more IR (band 4) is reflected and the high chlorophyll content absorbs more

red, i.e. less band 3 reflectance (Jensen, 1996, Lillesand and Kiefer, 1994).

The "Tasseled Cap" transformation (Kauth and Thomas, 1976) has been found to

be an accurate and reliable means of detecting vegetation health ("greenness"),

environmental condition ("brightness" and "wetness") and atmospheric condition (haze)

(Crist and Cicone, 1984, Crist and Kauth, 1986, van Leeuwen and Huete, 1996, Todd and

Hoffer, 1998). Tasseled Cap wetness has been used to stratify TM image ratios and

accounted for 78% of the variation in canopies (Kushla and Ripple, 1998). Results of

Serrano et al., (2000) indicated the landscape scale sensitivity of WI to variations in









canopy relative water content. The Tasseled Cap transformation multiplies a band's pixel

DN by a band's specific constant and adds an index specific factor to the sum of the

index for all bands (Crist and Cicone, 1984, Crist and Kauth, 1986). Multiplicative and

additive Tasseled Cap constants are presented in Table 1 and Table 2 (Crist et al., 1986,

ERDAS Field Guide).


Table 1 Tasseled Cap Multiplicative Matrix.
Tasseled
Cap TM TM TM TM TM TM TM
Index Bandl Band2 Band3 Band4 Band5 Band6 Band7
Brightness 0.2909 0.2493 0.4806 0.5568 0.4438 0.0000 0.1706
Greenness -0.2728 -0.2174 -0.5508 0.7221 0.0733 0.0000 -0.1648
Wetness 0.1446 0.1761 0.3322 0.3396 -0.6210 0.0000 -0.4186
Haze 0.8461 -0.0731 -0.4640 -0.0032 -0.0492 0.0000 0.0119
Other 0.0549 -0.0232 0.0339 -0.1937 0.4162 0.0000 -0.7823
Other2 0.1186 -0.8069 0.4094 0.0571 -0.0228 0.0000 -0.0220



Table 2 Tasseled Cap Additive Matrix.

Brightness Greenness Wetness Haze Otherl Other2
Scale 10.3695 0.7310 -3.3828 0.7879 -2.4750 -0.0336


The matrix formulas for Greenness and Wetness are

Greenness = (-0.2728(Bandl) + -0.2174(Band2) + -0.5508(Band3) +

0.7221(Band4) + 0.0733(Band5) + 0.0(Band6) + -0.1648(Band7)) + -0.7310



Wetness = (0.1446(Bandl) + 0.1761(Band2) + 0.3322(Band3) + 0.3396(Band4)


+ -0.6210(Band5) + 0.0(Band6) + -0.4186(Band7)) + -3.3828









The single most equivalent spectral index study reported to date is Todd and

Hoffer (1998). Their mathematically modeled study compared NDVI, and Tasseled Cap

GVI, WI, and brightness (BI) values calculated from composite reflectance (soil and

green vegetation) value estimates. Vegetation and soil spectral reflectance curves from

Hoffer (1978) and Bartolucci (1977) were used to develop the estimates. Simulated DN

values were calculated to 8-bit data, thereby simulating TM data. In the study, Todd and

Hoffer used estimates of percent vegetation cover (100%, 80%, 50% and 20%), and soil

types (silt, sand and clay) at two moisture levels to compare NDVI and GVI to BI and

WI. They also compared NDVI and GVI to percent change in green vegetation cover.

GVI varied little at 80% cover for both moisture levels. NDVI values were higher for the

more moist soils. Todd and Hoffer concluded that GVI was less affected by variation in

soil type and moisture when predicting green vegetation cover. They also found

substantial increases in NDVI with increases in WI. The study's discussion of results

suggests high wetness values are produced by absorption of mid-infrared (TM bands 5

and 7) reflectance in healthy green vegetation. The study concludes the index results are

representative of homogenous canopies and heterogeneous plant canopy application is

untested. This further demonstrates the need for quantifying the affect on vegetation

indices of a landscape feature's scale, distribution, condition and diversity, as well as

environmental condition.




Previous GBS Environmental Studies

Scientific literature describing the environmental impacts of the GBS diversion in

scientific journals is sparse. Newspaper and magazine articles, papers filed in the World









Court case and lengthy discussions on both sides of the controversy abound in informal

forums such as the Internet due to the highly politically charged nature of the subject.

Scientific research is still in progress and unpublished in peer reviewed international

scientific literature, except Smith et al., 2000. Several reports on the topic have been

published by the Hungarian Academy of Sciences (CEC Working Group, 1993; HAS

Working Group, 1993, HAS Working Group, 1994).

The only publication written in English by Slovakian scientists and available to

previous studies was "Dams in Slovakia" by Abaffy et al., (1995). This book has a

detailed description of the GBS with detailed schematics and specifications. Cleminski

(1993) published an article describing the GBS in general terms as well as other dams

along the Danube. A 1999 private consulting publication supporting the Slovakian

claims appeared on the web, "Visit to The Area of The Gabcikovo Hydropower Project"

(Mucha et al., 1999). Petch and Kolejka (1988) presented a regional skeleton of

ecological stability of Slovakia based on Miklos (1988), which portrays the area as the

most threatened linear feature and an ecologically important landscape area outside the

main territorial system.

Two previous analyses of regional satellite data for the study area have been

performed. Both focused on generalized analysis of the area's vegetation composition

and non-weighted mean NDVI for the study area and some controls. The first research

project, conducted at the Hungarian Center for Remote Sensing (FOMI), used 1992, 1993

and 1994 TM imagery to map vegetation composition in the region between the Danube

dikes (Smith et al., 1996). The August 1992 imagery represented the region's pre-

diversion status and August 1993/1994 the post diversion status. Cloud cover obscured






20


the 1994 image and vegetation was not classified. The second project, conducted at the

University of Florida, expanded the 1996 study to include Septemberl988 and September

1997 TM imagery (Smith et al., 2000).















CHAPTER 3
MATERIALS AND METHODS

The methods used in this thesis surpasses the image processing and analysis

methods employed during an initial Joint US-Hungarian Fund project conducted shortly

after the Gabcikovo Dam was commissioned (Smith et al., 1996). A subsequent

publication concerning two additional satellite image dates (Smith et al., 2000) was also

used. Pre and post processing of Smith's image dates, 1988 and 1997, was a prerequisite

to this thesis. Both previous studies focused on generalized analysis of the area's

vegetation composition and non-weighted-mean, rescaled NDVI values (0 to 255) for the

study area and selected controls, Table 3.



Table 3 Previous Rescaled NDVI Values (Smith, et al, 2000).
Zone/Year 25 ma 50 ma 100 ma 200 ma All forest
( a) Upper zone
1988 169.5 178.2 182.2 182.2 181.0
1992 191.4 195.4 198.3 199.8 200.7
1993 161.9 169.4 175.0 177.7 179.2
1997 181.3 186.7 189.4 190.1 188.8
( b ) Middle zone
1988 166.0 176.7 181.5 185.1 183.0
1992 194.0 197.9 200.8 202.2 202.9
1993 159.4 166.7 172.5 175.3 177.0
1997 183.2 189.6 191.9 192.7 191.7
( c ) Lower zone
1988 159.8 173.1 184.5 189.0 185.4
1992 193.0 196.9 199.9 201.0 202.4
1993 160.0 168.6 174.6 176.4 177.6
1997 183.6 189.7 192.7 192.6 191.5
a Denotes distance from image classified left and right banks of the 1992 Danube River.









The methods of this thesis differ from the previous studies in that: 1) two

additional indices, GVI and WI, were evaluated; 2) in order to detect subtle changes,

floating point index values were calculated and not rescaled; 3) WI values were

compared to precipitation trends; 4) mean index values were weighted by histogram and

compared. Also, three new control areas were chosen because previous studies' control

boundaries were unavailable. Controls were chosen to measure regional forest vegetation

index values and non-diversion related moisture index values. Figure 5 shows the

location of the study area and the new controls, Northern (N), South-Central (SC) and

Southern (S). Figure 6 to Figure 8 depict control area conditions. In general, controls

exhibited visual characteristics similar to forested portions of the study area and controls

appeared similar across all image dates. Controls were presumed to be isolated from

potential river diversion related effects.


IR uIt" u *.~


Figure 5 Control and Study Area Locations.














Figure 6 Northern Controls 1988, 1992, 1993, and 1997.


Figure 7 South-Central Controls 1988, 1992, 1993, and 1997.


c J- : '111 11


Figure 8 Southern Controls 1988, 1992, 1993 and 1997.


The methods were developed to preprocess and classify satellite images according

to 4 land cover types (Water, Forest, Grass and Exposed), calculate NDVI, GVI and WI

indexes for all image years, create a zonal statistics image, compare precipitation patterns

and WI values, and lastly, compare mean-histogram-weighted NDVI, GVI and WI study

zonal values to control zonal values.









Imagery

TM scenes taken during the same seasonal time period were used to compare

vegetation regimes between years at approximately the same growth stage. A time series

of five TM images taken in August and September between 1988-1997 were obtained.

The 1988 and 1992 images were taken prior to diversion of the Danube, whereas the

other three images were taken post diversion. Due to excessive cloud cover and haze

over the northern portion of the 1994 study area image, it was not used in the analysis.

Similarly, cloud cover was limited to isolated regions in the northern portion of the 1997

image; however the cloud cover did not prohibit the use of the image.




Software and Hardware

Image processing was conducted using algorithms supplied with ERDAS'

Imagine image processing software (v. 8.3.1). Imagine's inability to perform raster zonal

summaries on continuous floating-point data, required the use of additional analysis and

graphic display software. For post processing map algebra functions, Environmental

Science Research Institute's (ESRI) Arc/Info (v 7.2.1) GRID software was chosen. Due

to the number of images and variables analyzed, processing models and analysis

automation programs were written. Microsoft EXCEL 2000 was used to calculate linear

index correlations.




Image Pre-processing

Standard image pre-processing techniques were applied to all images. Satellite

images taken at different dates were normalized to account for different acquisition









parameters (e.g., sun angle and sensor response) and environmental parameters (e.g.,

atmospheric attenuation). Some atmospheric conditions at the time of satellite overpass

were recorded (Table 4).



Table 4 Meteorological Conditions in the Study Area at Time of Satellite Overpass.
Average air Vapor Water Cloud
Year temperature content temperature cover
(C) (%) (C) (%)
1988 -a a a 0
1992 24.8 67 22.2 0
1993 19.8 55 19.1 0
1997 -a a a 6
aNot available.


After applying radiometric (sensor) and atmospheric corrections, a geometric correction

was applied, fitting the images to a common coordinate system.



Radiometric Correction

Because satellite sensor response varies over time, an invariant feature on the

ground one year can have a different DN the following year. Image to image radiometric

normalization was necessary to utilize the NDVI index derived from the red and near-

infrared bands of the imagery (Mather, 1987; Swain, and Davis 1978) as well as the

Tasseled Cap indices derived from the six reflective bands of TM.

FOMI applied the DBO normalization method to the 1992 and 1993 images.

Radiometric correction of the 1988 and 1997 imagery used ERDAS' Imagine histogram

matching algorithm. The histogram matching algorithm, which was developed and

implemented after the FOMI study, matches high and low areas of one image's histogram

to highs and lows of another image (ERDAS Field Guide, 4th Ed). Atmospheric









corrections were applied prior to 1997's radiometric correction. Both 1988 and 1997

images were histogram matched to the 1992 TM image.



Atmospheric Correction

Due to excessive cloud cover and haze over the northern portion of the 1994 study

area image, it was ultimately not used in the analysis. Even after FOMI applied haze

reduction algorithms, the central portion of the study area remained obscured. Similarly,

the 1997 image contained 6 percent cloud cover. However, the cloud cover was isolated

to limited regions in the northern portion of the image. Figure 9 shows the cloud cover

prior to haze reduction. Imagine's haze reduction algorithm was applied. Figure 10

shows the post haze reduction conditions. Areas of interest (AOI) were formed around

the remaining, visually apparent clouds and cloud shadows. The AOIs were used to

eliminate these pixels from the analysis.


9 1997 Pre-Haze Reduction Cloud Cover and Shadow.


























Figure 10 1997 Post-Haze Reduction Image.


Geometric Correction

The geometrically corrected TM images for 1992, 1993,1994, and 1997, were

provided by FOMI. Several 29 km by 24 km subset images of the TM scene covering

both the Hungarian and Slovakian sides of the Danube in the vicinity of Gabcikovo were

originally selected. FOMI determined the parameters for registering the sub-images to a

reference image using an automated feature mapping technique (Bittner and Parareda,

1993). A composite transformation formed by the image registration parameters and a

previously established reference image to map transformation was used to subsequently

transform the images to the Hungarian National Datum.

Ground control points were selected on 1:100,000 scale maps. Root mean square

error (RMSE) of image-to-image registration was reported to be approximately 12.5 m,

while the RMSE of image to map registration was approximately 25 m. FOMI used

nearest neighbor (NN) re-sampling with a 25 m pixel size. The NN method was chosen

over a bilinear interpolation or bi-cubic convolution because, comparatively, the process










alters the original input pixel values less than the others (Smith et al., 1995). The images

were imported to an Imagine file format for processing. Due to cloud cover, the 1997

image required additional atmospheric and radiometric preprocessing.

The 1988 image was supplied in a generic binary format. After importing to an

Imagine file format, the image was registered to the 1992 image through 21 ground

control points (GCP). The GCPs were evenly distributed across the scene and located at

features recognized in both images. An overall RMSE of less than one pixel, or 25 m,

was achieved. Positional accuracy was verified through visual comparison of the 1988

image with the 1992 image at recognized locations. Table 5 presents the GCPs used in

rectification.


Table 5 1988 Rectification Parameters.
X Y X Y RMSE
GCP Column Row Y X
Coordinate Coordinate Residual Residual (m)
GCP #2 1450.465 2820.544 516750.025 303794.369 -12.318 2.887 12.652
GCP #7 1373.460 2920.421 515139.180 307181.097 -0.813 0.985 1.277
GCP #9 1448.846 2788.479 516454.708 302890.973 27.169 -4.023 27.465
GCP #14 1893.557 2801.705 529573.264 300179.539 -13.158 -7.744 15.268
GCP #15 2030.773 2923.491 534457.833 302711.540 -0.001 1.889 1.889
GCP #19 2481.270 2877.476 547327.177 298215.743 8.055 2.580 8.458
GCP #25 2815.217 2716.840 555884.260 291236.933 -8.321 -3.000 8.845
GCP #28 2920.471 2342.666 556256.381 279606.163 11.518 4.063 12.213
GCP #29 2871.426 2063.550 552938.960 271823.286 -11.702 -3.723 12.280
GCP #34 2994.589 1626.830 553445.678 258252.735 3.470 0.856 3.574
GCP #36 2702.524 1603.134 544760.209 259598.706 -3.679 -0.844 3.774
GCP #38 2485.210 1583.125 538307.231 260557.809 -7.787 0.194 7.790
GCP #40 2372.638 1697.931 535814.066 264705.482 17.077 0.697 17.091
GCP #47 1782.605 1957.538 520427.038 276382.084 -9.528 13.572 16.582
GCP #45 1773.116 1812.952 519127.889 272287.008 -1.224 -7.203 7.307
GCP #48 1388.780 1988.881 509226.585 280089.129 3.627 2.768 4.562
GCP #49 1584.960 2248.984 516732.350 286264.759 1.543 -18.590 18.653
GCP #50 1464.729 2555.089 515388.143 295993.365 -1.360 19.267 19.315
GCP #51 1264.790 2676.613 510461.711 300959.591 -8.148 -5.720 9.955
GCP #52 1630.352 2513.871 519901.371 293649.028 1.821 -2.057 2.747
GCP #53 1949.091 2473.085 528897.788 290215.175 3.760 3.147 4.903









The rectification utilized a 4th Order polynomial. A single GCP, #9, had a high RMSE.

This point should have been deleted from the model and a lower polynomial used.

The final pre-processed images are presented in Figure 11 to Figure 14. The

images are a false color composite utilizing TM bands 4, 3 and 2.


Figure 11 1988 False Color Composite (TM Bands 4,3,2).


Figure 12 1992 False Color Composite (TM Bands 4, 3, 2).



























Figure 13 1993 False Color Composite (TM Bands 4, 3, 2).


Figure 14 1997 False Color Composite (TM Bands 4, 3, 2).


Image Post-Processing

The floodplain study area was delineated on the TM false color composite (Bands

4, 3 and 2) by following the dikes on both sides of the river and saved as an AOI. The

study area was extracted from the larger image to reduce the variance in the distribution









of digital numbers (DN) and thus reduce pixel class confusion. Three regional control

areas for this research consistently identified across all years were also extracted from the

larger image but the vegetation was not classified. Control and study locations were

shown in Figure 5. Image NDVI, GVI and WI values were generated for control and

study areas. Annual, statistical analysis zone images were developed which combined

distance from water, country of origin, study area region and vegetation class.



Image Classification

Previous studies used supervised classification techniques to classify the 1992 and

1993 images. Training areas were selected from color infrared aerial photographs taken

in 1992. Training areas were selected for 3 water classes, 2 forest classes, 2 grassland

classes and an exposed surface class. The 3 subcategories for water corresponded to the

main river, the two lakes and the laterals, all having different turbidity status. The forests

were classified into older and younger forest stands. Grasses were classified into wet and

dry categories. Highly reflective targets, such as disturbed areas associated with the dam

and weir construction or point bars within and next to the riverbed, made up the exposed

class. The selected categories were highly separable in terms of transformed divergence

(Swain and Davies, 1978), except some pairs of subclasses of the same category.

A separate unsupervised maximum likelihood classification was performed using

the six reflective channels of 1988 and 1997 images. A total of 16 unsupervised classes

were created. The sixteen classes were pared down to the 8 previously identified classes.









Classes were grouped based on visual interpretation and comparison of the 1992 image

and the 1992 classified result versus the 1988 image and its associated classes. The

visual interpretations are summarized in Table 6.




Table 6 Visual Interpretation Standard.
Image Appearance Interpretation Original Class Reclassified
Light Blue/Grey "Turbid" Water 1 Water
Slate Gray-Light Purple "Deep/Turbid" Water 2 Water
Purple-Black "Deep" Water 3 Water
Light Brownish Red "Young" Forest 1 Forest
Brown-Deep Brown "Established" Forest 2 Forest
Bright Red-Magenta "Recent" Grass 1 Grass
Bright Red/Brown-Deep Magenta "Established" Grass 2 Grass
White-Bright Blue-Light Grey/Green "Barren" Exposed Exposed


The lack of pixel homogeneity in some areas caused classification confusion. In

those areas, the mixed pixels were extracted and a separate maximum likelihood

classification applied to the extracted pixels. The newly classed pixels were then

replaced into the final product, a process commonly called "cluster busting". The same

classification methods were applied to the 1997 image. For the purposes of this research,

classes for all years were aggregated to the 4 main classes, water, forest, grass and

exposed, Figure 15. Because field visits were not possible, classification accuracy

assessment was not performed. The lack of an accuracy assessment forced the use and

analysis of vegetation and moisture index values.



























-W 1992

Figure 15 Study Area Land Cover Classifications.


Vegetation and Moisture Indices

Using Imagine (v. 8.3) Tasseled Cap Greenness and Wetness indices as well as

NDVI values were calculated for control and study area pixels. In order to detect subtle

changes in NDVI and Tasseled Cap indices, floating-point values were calculated and

pixel values were not re-scaled to 8 bit integer (0,255) as they were in previous studies.

To avoid edge effects in the control areas, an area inside the border was selected for

analysis.

Using the previously discussed matrix formulas for GVI and WI, software

Tasseled Cap algorithms were verified by hand calculating sample pixel values for

Tasseled Cap Indices (Table 7). Results from hand calculations utilizing the tasseled cap

matrix approximated those from software algorithms (Table 8).










Table 7 Digital Numbers by Image Band for a Sample pixel.
Rectified
X-Coor Y-Coor Sample Valu
Value
Map 527243.881 286880.226Band 1 71
File 769 804 Band 2 29
Band 3 25
Band 4 82
Band 5 43
Band 6-dup7 11
Band7 11



Table 8 Software vs. Hand Calculated Tasseled Cap Values.
Tasseled Hand
Software Difference
Cap Band Calculated
Brightness 116.924 116.8857 0.0383
Greenness 20.346 20.3769 -0.0309
Wetness 16.884 16.8353 0.0487
Haze 44.916 44.8940 0.0220
Other -5.009 -4.9945 -0.0145
Other2 -1.320 -1.3183 -0.0017


Statistical Analysis

Separate statistical zone images were generated in an attempt to analyze potential

study area NDVI, GVI and WI changes based on 4 land cover classes, 2 countries of

origin, 6 buffer distances from water, and 3 study area regions. Using map algebra, the

separate zonal images were recorded and combined to form an image with 144 unique

statistical analysis zones. The process was repeated for each of the 4 classified TM

scenes.




Country Zones

The country border, provided in ESRI's Arc/Info format by FOMI, was used to

produce a 2 country zonal image (Figure 16).

























Figure 16 Country Statistical Zones.


Buffer zones

Using the 1992 water class, 6 water buffer zones were generated for Om (Water),

Om to 25m, 25m to 50m, 50m to 100m, 100m to 200m distances. The remaining

background area was classed as greater than 200m. The buffer zone image was limited to

areas within the flood plain dikes. An example of the buffer zones is shown in Figure 17.








I-, 1 ,:, r j w, = 7 1 -.1

'" 1.:. 1), IUO i

Figure 17 Water Buffer Statistical Zones.



Regional zones

Regional zones were chosen to study potential backwater affects. Backwater

affects occur where the diverted water channel meets the original Danube. Within the


Slo;iakia





H unary









study area, 3 regions, upper, middle and lower, were arbitrarily chosen along the diverted

portion of the Danube (Figure 18). Each region represented approximately one third of

the area.





Upper






Middle
Lower




Figure 18 Regional Statistical Analysis Zones.


Map Algebra

Through map algebra functions, the country, water buffers, and region images

were recorded and added to each year's classed image to produce 4 images containing 144

unique zonal classes (Figure 19). The 4 classes of the land cover images were assigned

values 1 (Water), 2 (Forest) 3 (Grass) and 4 (Exposed). The countries of the country

image were given values 0 (Slovakia) and 5 (Hungary). The 6 buffers of the water buffer

image (Water, 0-25m, 25-50m, 50-100m, 100m-200m, <200m) were recorded in

increments of 10 from 0 to 50. Lastly, the upper, middle and lower regions of the region

image were calculated equal to 0, 60 and 120, respectively.























Figure 19 1988 Map Algebra and Zonal Image Creation.


When added together, each of the 144 unique zonal classes reflected the land

cover, country, buffer and region for each image date (Figure 20). For example: Forested

Class (2) + Country Hungary (5) + Water Buffer 100-200m (40) + Upper Region (0) =

Zone 47.


Figure 20 Zonal Statistics Image Example, 1988.


Statistics

Due to Imagine's (v.8.3.1) zonal statistic limitations, the double floating point

vegetation indices and zonal images were exported to Arc/Info Grid format. For each

year's zonal and study area image, Arc/Info's GRID command Zonalstats generated









NDVI, GVI and WI zonal means. Control NDVI, GVI and WI statistics were obtained

using the same export conversion and grid command. The process generated an Arc/Info

INFO file containing the NDVI, GVI and WI zonal means. The INFO file was converted

to a dBase with Arc/Info's INFODBASE command. Using Microsoft's Excel program,

the dBase file data were imported and analyzed

Histogram weights were calculated by dividing the unique zone's histogram value

by the total histogram value. A unique zone's weighted index mean was obtained

through multiplying the zone's weight and its mean. Table 9 is an abbreviated table of

the 1988 histogram weighted GVI means for a sample of unique zones.



Table 9 An Example of 1988 Histogram Weighted GVI Means (WTMN) by Zone.

Mean Zonal WTMN
Class Country Buffers Region Area Histogram 88 GI Wegt 8
88 GVI Weight 88 GVI
Water Slovakia< 200m Upper 16.0625 257 1.7090 0.0015 0.0026
Forest Slovakia< 200m Upper 344.9375 5519 17.7180 0.0324 0.5737
Grass Slovakia< 200m Upper 72.8750 1166 15.3548 0.0068 0.1050
Exposed Slovakia< 200m Upper 6.3125 101 -12.1530 0.0006 -0.0072
Water Hungary < 200m Upper 22.0625 353 -1.4013 0.0021 -0.0029
Forest Hungary< 200m Upper 798.2500 12772 18.2146 0.0749 1.3650
Grass Hungary< 200m Upper 150.5625 2409 17.9269 0.0141 0.2534
Exposed Hungary < 200m Upper 6.1250 98 -18.3730 0.0006 -0.0106

Total 10,652 170433 1.0


The formula: Weighted Mean = (wti)(xi) + ... (wti)(xi)
Y(wtij)

was used to produce total histogram weighted mean values for control areas and study

vegetation classes, water buffer areas, regions, and countries. The weighted means for

these larger areas were obtained through dividing the sum of the products, unique zonal









weights times unique zonal means from element i to element j, by the sum of the unique

zonal weights from element i to element j. Total histogram weighted mean NDVI values

were compared to total histogram weighted mean GVI and WI values for the same zones.

Simple linear correlations across years and between indices were obtained using MSN's

Excel linear correlation formula (Figure 21).


co v(x, Y)
where
J1 (x, -)
and


Figure 21 MSN Excel's Linear Correlation Formula.



Precipitation Data

Since WI is a moisture related index, and it is presumed that plants have short and

long term responses to moisture fluctuations (NDVI, GVI), readily available precipitation

data for the study area was obtained to access possible trends in index values. Regional

precipitation station locations can be seen in Figure 22.

Monthly station precipitation totals for the years 1988, 1992, and 1993 were

supplied by the Hungarian Meteorology Department. Graphs of monthly precipitation

totals are presented in Figure 23, Figure 24 and Figure 25. Monthly precipitation totals

for 1997 were unavailable; however, annual precipitation data for the 1989-1997 water-

years were obtained from FOMI and are presented in Figure 26.

































Figure 22 Precipitation Recording Stations.


1988 Monthly Precipitation


160.0
140.0 E- BRATIS
120.0 -RAJKA
100.0
80.0 E MOSONM
80.0 -
60.0 GYOR
40.0 n SALA
20.0 O HURBAN
0.0




Month


Figure 23 1988 Monthly Precipitation Totals.










1992 Monthly Precipitation

140.0
S120.0
100.0 D0 BRATIS
o 80.0 DMOSONM
S60.0 -I 0 GYOR
z 40.0 0 HURBAN
& 20.0
0.0





Month

Figure 24 1992 Monthly Precipitation Totals.




1993 Monthly Precipitation


140.0
120.0
0 BRATIS
100.0 MOSONM
O MOSONM
80.0 -
.2 0 GYOR
S60.0 -GY
40.0 SALA
S]EO URBAN
20.0
0.0


Month

Month


Figure 25 1993 Monthly Precipitation Totals.











Study Area Precipitation Totals


900
800
700
600
500
S400
300--- -
^ 200
100
0




Water Year

Figure 26 Water Year Precipitation Totals.


0 Precipitation


It is important to consider the image date when using monthly precipitation totals.

Since the image dates used in this study fell at the beginning of the month, precipitation

station totals for the months prior to and the month during satellite overpass are provided

in Table 10. The Hurban station values were not used in the average calculation. The

station was not used because it is very distant, approximately 40 km from the study area,

and the 116.7mm July 1992 value compared to the other July 1992 station values

suggests a possible recording error. The average monthly rainfall for the month prior to

the image date is presented in Table 11.









Table 10 Regional Monthly Precipitation Values Surrounding Image Dates.
r S# N e July August September
Year Sta# Name
(mm) (mm) (mm)
1988 111816 BRATIS 14.5 90.7 61.4
1988 80 RAJKA 12.5 137.7 34.0
1988 91 MOSONM 16.7 88.9 73.7
1988 212822 GYOR 25.5 105.3 66.6
1988 142479 SALA 15.0 95.0 57.0
1988 111858 HURBAN 31.3 125.5 78.2
b
June July August
Year Sta# Name
(mm) (mm) (mm)
1992 111816 BRATIS 82.1 23.7 8.7
1992 91 MOSONM 74.3 33.4 25.5
1992 212822 GYOR 74.1 95.3 1.6
1992 111858 HURBAN 83.6 116.7 1.0
c
June July August
Year Sta# Name
(mm) (mm) (mm)
1993 111816 BRATIS 39.2 77.1 69.8
1993 91 MOSONM 69.7 65.3 53.6
1993 212822 GYOR 44.7 63.5 35.2
1993 142479 SALA 29.0 72.0 59.0
1993 111858 HURBAN 51.3 37.6 28.7


Table 11 Average Monthly Precipitation Prior to Image Date.
July August
Image precipitation precipitation
(mm) (mm)
9/88 --
8/92
8/93
9/97














CHAPTER 4
RESULTS AND DISCUSSION

Morphological changes in the original Danube were readily apparent through

observation of the satellite imagery. For example, reduced surface water and exposed

riverbanks can be seen throughout the Szigetkoz region (Figure 27).


Figure 27 Pre- and Post-Danube Diversion, 1992 and 1993, Respectively.


Image Classification

Land cover total aerial extent and percent composition for the four study area land

cover classes, are summarized in Table 12. Land cover composition for the Szigetkoz

portion of the study area is displayed separately in Table 13. Control areas were not

classified since they were visually similar to the study area forest class.









Table 12 Study Area Land Cover Classes (ha and percent cover), 1988-1997.
Category Water Forest Grassland IExposed Total
(a) Total Area in Hectares (ha)
1988 2,644 6,627 1,242 140 10,652
1992 2,422 5,252 1,858 1,120 10,652
1993 2,102 4,948 3,252 350 10,652
1997 2,329 5,499 2,005 223 10,056a
(b) Percent of Total Area (%)
1988 25 62 12 1 100
1992 23 49 17 11 100
1993 20 46 31 3 100
1997 23 55 20 2 100
aArea reduction due to cloud cover.


Table 13 Szigetkoz Region Land Cover Classifications.
Category Water Forest Grassland Exposed Total
(a) Szigetkoz Area in Hectares (ha)
1988 1,596 3,696 725 69 6,086
1992 1,398 2,941 1,090 657 6,086
1993 1,138 2,766 1,985 197 6,086
1997 1,290 3,077 1,079 110 5,556a
(b) Percent of Szigetkoz Area (%)
1988 26 61 12 1 100
1992 23 48 18 11 100
1993 19 45 33 3 100
1997 23 55 19 2 100
aArea reduction due to cloud cover.


Approximately 5% of the 1992 and 1993 classification from previous studies was

undetermined. Upon further image review, the 1992 portion was aggregated with the

water category and the 1993 portion was aggregated with the grass category for this


research. Figure 28 charts annual land cover percentages.










Percent Land Cover

70
60
50 Water
c 40 Forest
S30 - OGrassland
20 OExposed


0
1988 1992 1993 1997
Year

Figure 28 Percent Land Cover Composition by Year.



Anecdotal Accuracy Assessment

Due to lack of ground-based data, a formal accuracy assessment of the

classification was not performed. Instead, local experts reviewed results (Buttner,

personal communication). Field data reported in the Hungarian Forest Database

estimated managed forests in the study area on the Hungarian side at 3575 ha (Smith et

al., 2000). The area indicated as a forest on the 1997 image is 3,077 ha, which represents

a difference between the field observations and classification measurements of 14

percent. This difference is most likely due to 1) a 9% loss total area due to cloud cover;

2) recording of forest roads and clearances as forest in the database; 3) the cut, but not yet

reforested areas are not distinguishable in the database, however it was classified as

grassland in the image.

Forested area in the image closely fits the ground data. Furthermore, the water

surface measured on the image is similar to the actual ground surface area. There are

virtually no croplands within the dikes on the Hungarian side and so the area classified as










grassland is actually grassland on newly reforested area and marshland on the image.

The exposed category on the image is negligible in size, but is similar in both image and

ground measurement (Smith, et. al. 2000).

The land cover data as classified indicates the following; 1) the most important

type of 1997 land cover is forest representing 55 percent of the total area; 2) areas

covered with water and grassland are approximately 20 percent each; 3) the barren area is

a minor component (2-3 percent); and, 4) ratios of the land cover on the Slovakian and

the Hungarian side are similar. The conclusion by the expert forester was that these

classifications accurately represented the land cover conditions in the study area (Butner

and Somogyi, personal communication, 1998).




Analysis of Land Cover Changes

A comparison of land cover consistency and change for the period 1988 to 1997 is

presented in Table 14. The table is the product ofbetween-year cross-classification

matrices.



Table 14 Land Cover Classification Comparisons.
Year Constant Constant Constant Constant Former Former Other Total
Comparison water forest grassland exposed water forest changes Area
(ha) (ha) (ha) (ha) (ha) (ha) (ha) (ha)
1988/1992 1,844 4,717 554 94 799 1,910 734 10,652
1988/1993 1,650 4,434 817 48 993 2,193 517 10,652
1992/1993 1,697 4,327 1,463 198 725 924 1,317 10,652
1992/1997 1,628 3,962 601 88 681 1,102 1,993 10,056a
1993/1997 1,624 3,784 1,008 82 396 967 2,194 10,056a
1988/1997 1,753 4,727 444 16 767 1,598 750 10,056"
area reduction due to 1997 cloud cover.









In Table 14, changes before the diversion are represented by data for the years

1988 and 1992. The short/term changes are characterized by the data of 1992 and 1993.

The midterm changes after the diversion are demonstrated by the data of the years 1993

and 1997. Constant values represent areas classified the same in both years. Conversely,

former values represent areas previously classified as one land cover and classified

differently in the subsequent year. The "other changes" category reflects change from

areas previously classed as grass or exposed that shifted to any of the other 3 classes.

The data indicate a continued fluctuation in all classes. As surface water and

forested classes declined, grass and exposed classes increased over their 1988 values.

Some changes were substantial. For example, comparison of the data for 1988 and 1997

reveals some changes in aerial extent. Previous studies data showed that between 1990

and 1992, cuttings of dead trees were 214 m3/year, while the same value for the period of

1993-1997 was almost five times higher (1158 m3/year) (Smith et al., 2000). Whether

the increase in dead trees was a result of the diversion or climate is unclear. While the

data of Table 12, Table 13, and Table 14 summarize study area affects, they do not reflect

the over all spatial distribution of the changes.

Several explanations for class shifts are possible. One explanation for a forest-

exposed-grass-forest class shift is forestry practice. Harvested and cleared forested land

is first seen as exposed. It is then replanted with seedlings and grass is grown as cover.

As trees mature, the area is interpreted as forested.

Another explanation for some of the water-exposed-grass shift relates directly to

the diversion and ecological succession. Reduced water depth exposes banks. Exposed

banks are colonized by vegetation growth and subsequently classed as grass. As time









passes and water levels are stabilized, the grass areas may become established with trees

or other woody vegetation and classed as Forest.

Lastly, the forest-exposed-grass-water shifts can be related to the addition of side

channels and weirs to the branches of the Danube. During side channel excavation for

new weirs some areas previously classed as forest or grass become exposed. The new

weirs cause increased surface water retention and inundation and consequently a shift in

the next year's water class. Construction of the weirs is clearly visible in the 1993

imagery (Figure 29). In Figure 30, the 1997 image revealed the 1993 weir construction

area was inundated with water and the exposed area colonized by vegetation. Figure 29

also shows exposed banks along the Danube and Figure 30 shows their colonization by

vegetation.


Figure 29 1993 Weir Construction.


Figure 30 1997 Post-Weir Construction.









Vegetation and Moisture Index Images

For presentation and discussion ofNDVI, GVI and WI index image results,

portions representative of study area and controls were chosen. For the study area, a

section containing the lower half of the upper region and a small portion of the upper half

of the middle region was selected and the southern control was chosen to represent all

controls. Locations of examples are shown in Figure 31.



















Figure 31 Study Area and Control Example Image Locations.


Index results in Figure 32 through Figure 37 each contain 4 gray scale images.

From left to right, top images represent 1988 and 1992 index results and bottom images

represent 1993 and 1997 results respectively. Portions of the 1997 image excluded from

analysis due to cloud cover are seen in top left portions of the 1997 study area examples.

For display purposes calculated index image values were stretched from 0 to 255 to

produce the gray scale images. The gray scale is indicative of index values, where the

color range of white-gray-black equals high-medium-low index values.









NDVI and GVI Images

Study area NDVI and GVI image examples are presented in Figure 32 and Figure

33. Control NDVI and GVI images are presented in Figure 34 and Figure 35.


Figure 32 Sample Study Area NDVI Images 1988, 1992, 1993 and 1997.



































Figure 33 Sample Study Area GV1 Images 1


Figure 34 Southern Control Area NDVI Images 1988, 1992, 1993, 1997.


1992, 1993 and 1997.
































Figure 35 Southern Control Area GVI Images 1988, 1992, 1993, 1997.


Study area landscape pattern changes visible in Figure 27 and are also visible in

the NDVI and GVI gray scale images. Morphological changes in the main river channel

and side branches are visible. Also, large, well-defined straight edged rectangular and

triangular patterns indicative of silviculture practices are present in all the study and

southern control image examples; however, visually detectable evidence of silviculture

was absent from the northern and south-central controls.

In Figure 32, the 1993 NDVI study image (lower left) shows the newly exposed

riverbanks as dark gray, or low NDVI. The 1997 riverbank area, which had been

colonized by vegetation, appears light gray, indicating higher NDVI than the 1993

riverbanks.






54


The GVI results displayed in the study GVI and control GVI figures provide

higher contrast images and thus better visual interpretation than the NDVI images.

Water and exposed areas appear dark gray to black and had low NDVI and GVI values.

Forested areas appear as gray patches. Grass and newly reforested areas had high NDVI

and GVI causing them to appear white to light gray.



WI Images

Example study and control wetness index image results are shown in Figure 36

and Figure 37 respectively.



























Figure 36 Sample Study Area WI Images 1988, 1992, 1993 and 1997.
IL. I "-






:"












Figure 36 Sample Study Area WI Images 1988, 1992, 1993 and 1997.



















..,

1 .P












Figure 37 Southern Control Area WI Images 1988, 1992, 1993, 1997.


Wet surfaces such as exposed, grass and newly replanted areas, display as black

to dark gray, less exposed (forested) as gray and dry areas as light gray to white. Land

cover patterns distinguishable in NDVI and GVI images are not as apparent as in the WI

images; however some trends can be seen along the main Danube channel. In Figure 36,

the 1993 image contains areas immediately adjacent to the center channel, which were

recently exposed and remained very wet or black. In the 1997 image, while some dark

areas around the main river channel remain, a majority has become lighter gray, or dry.









Observations and Comparisons

Precipitation Comparisons

Annual precipitation amounts are needed to interpret trends in environmental

moisture (WI) and ultimately, to interpret measured vegetation responses (NDVI, GVI) to

those trends. FOMI provided regional precipitation summaries for the water-years 1989

to 1997 (Table 15). The 8-year mean precipitation was 587 mm and 1 standard deviation

was 121.026 mm. The annual deviation from the mean provided in Table 15 was

calculated and described.




Table 15 Annual Precipitation for the Water-Years 1989 to 1997.
m o Number of
Sum of
Period pn ( Standard Description
precipitation (mm) Deviations
Deviations
1989-1990 488 -0.821 Droughty
1990-1991 532 -0.458 Droughty
1991-1992 480 -0.887 Droughty
1992-1993 508 -0.656 Droughty
1993-1994 637 0.410 Wet
1994-1995 721 1.104 Wet
1995-1996 806 1.806 Extremely Wet
1996-1997 527 -0.499 Droughty


The spring and early summer months of 1992 were unusually arid. This

coincided with the time of the diversion and so the effects of the diversion might have

been accentuated by the dry climate conditions. The summer of 1994 had average

precipitation, however, arid periods were frequent. 1995, 1996 were wet years with

exceptional floods on the Danube. These meteorological observations are important to

this study in that the drought conditions in 1991 and 1992 probably would have caused









stress to the natural and planted vegetation irrespective of the diversion. The wet period

of 1994-1996 may have masked the effect of the diversion.

In summary:

The 1988 spring was wet and the month during and prior to satellite

overpass was also wet.

The 1992 spring was dry and the rainfall before the image date was below

average.

The 1993 spring was dry but the average rainfall before the image date

was higher than the 1992 average for the same month.

The 1997 yearly summary indicates below average rainfall for the year.

The annual mean precipitation and standard deviations reflect long-term

drought conditions from 1989 to mid 1994, a brief period of drought relief

from mid 1994 to mid 1996, and a return to conditions in mid 1996.



Control NDVI and GVI Observations and Comparisons

Mean NDVI and GVI values for all control areas were graphed and compared for

trends. Figure 38 and Figure 39 show control NDVI and GVI trends were similar. In

1988, a relatively wet year, and in 1992, a relatively dry year, NDVI and GVI levels were

low. In 1993, a year with moderate rainfall, mean values were highest and in 1997 values

dropped to levels similar to 88 and 92. With the exception of 1992, NDVI levels for all

controls were similar. The northern 1992 NDVIs approximated the 1988 levels. South-

central means were highest in all years except 1988, when the southern control mean was

the highest.












Yearly Control NDVI Values


0.7000
0.6000
0.5000
0.4000
0.3000
S0.2000
0.1000
0.0000


--NDVI-N
---NDVI-S
--NDVI-SC


Year


Figure 38 Mean Control Area NDVI per Year.




Yearly Control GVI Values


35.0000
30.0000
- 25.0000
20.0000
S15.0000
S10.0000
5.0000
0.0000


- Green-N
--- Green-S
-- Green-SC


88 92 93
Year


Figure 39 Mean Control Area GVI per Year.


The graph of mean control GVI values in Figure 39 shows greater differentiation


between yearly values. The amplitude of the difference between values may be a result


of the Tasseled Cap GVI calculation, which includes a scaling factor. Further difference


may be due to inclusion of a moisture component in NDVI values (Richardson and


Wiegand, 1977, van Leeuwen and Huete, 1996, Todd and Hoffer, 1998, and Fisher,


1999), which is separated in Tasseled Cap calculations. Northern and Southern control


GVI levels were similar in 1988. Southern and south-central means were similar in 1992









and 1993. The northern mean for 92' and 93' was lowest and then intermediate in 1997.

South-central GVI control means were highest in all years.

Simple linear NDVI and GVI correlations (r) presented at the end of Table 16

were similar for all analysis zones and controls. High correlations of NDVI and GVI are

consistent with Hill and Aifadopoulou's 1990 NDVI conclusions.


Table 16 Mean Weighted GVI and NDVI Values.
Buffer 1988 1992 1993 1997
(a) GVI
Om -6.82 -23.92 -23.66 -6.50
25m 3.28 -0.46 3.40 5.12
50m 9.78 12.34 19.27 12.73
100m 14.44 16.36 25.19 17.43
200m 14.09 17.37 23.31 16.31
Outside 17.82 19.24 26.40 19.31
(b) NDVI
Om 0.22 0.00 -0.10 0.27
25m 0.33 0.33 0.35 0.39
50m 0.39 0.45 0.52 0.45
100m 0.42 0.48 0.58 0.49
200m 0.43 0.50 0.57 0.48
Outside 0.46 0.51 0.60 0.50
(c) GVI to NDVI Correlation
r = 1.00 0.99 0.99 1.00


Control WI Observations

The yearly mean control WI values are presented in Figure 40. All south-central

yearly means were highest, or driest, while 1988 and 1992 southern control values were

lowest, or most wet, and northern values for the same years were intermediate. In 1993,

the northern control was lowest, or most wet, and the 1993 southern control was slightly

higher, or drier than the southern control. The south central control remained the driest in

all years.







60



Yearly Control WI Values

12.0000
10.0000
8.0000
S6.0000 Wet-N
g 4.0000 --Wet-S
S2.0000 ---Wet-SC
0.0000
-2.0000 2 93 97
-4.0000
Year


Figure 40 Yearly Mean Control WI Levels.



Control GVI and WI Trends

Tasseled Cap scale factors allow GVI and WI values to be concurrently graphed.


On the other hand, variations in calculated NDVI values, which range between -1 and 1,


cannot be mapped in the same value range as GVI and WI. A composite graph of Yearly


NDVI, GVI and WI means is presented in Figure 41. Since a high NDVI and GVI


correlation has been established, it is presumed that WI-GVI relationships are the same as


WI-NDVI trends.





Yearly Control Index Values

35.0000 NDVI-N
30.0000
UNDVI-S
25.0000
M -A-NDVI-SC
> 20.0000
Green-N
i 15.0000
= Green-S
10.0000Green-
S5.0000 ----Green-SC
.0000 Wet-N
0.0000 ----- ---+--- Wet-N
-5.0000 88 92 93 97 Wet-S
Year Wet-SC


Figure 41 Mean NDVI, GVI and WI Comparison.









The graph of yearly control index values shows that during extreme

environmental moisture conditions, i.e. very wet (1988) and very dry (1992), GVI control

means are similar. During moderate moisture conditions such as 1993, GVI values are

highest and with less moisture, as in 1997, the mean GVI declined.



Control and Study Area Forest Index Value Comparison

Total forest class weighted mean NDVI, GVI and WI values were highly

correlated with all control locations, range of r = 0.897 to 0.998 (Table 17).




Table 17 Histogram Weighted Mean Study Area Forest Indices to Control Indices
Correlations.
South-
North South
Index Central
(N) (S) (SC)
(SC)
NDVI 0.924 0.977 0.994
GVI 0.897 0.948 0.991
WI 0.899 0.998 0.977


The highest study area forest WI to control WI correlation was with the southern control

(r=0.998). Forest NDVI and GVI were most correlated with south-central control NDVI

and GVI. A graph of the control and forest mean comparison is shown in Figure 42.

The high WI correlations indicate the study area forest and control areas are following

the same "wetness patterns", however, Figure 43 shows histogram weighted mean study

forest WI levels higher, or drier, than controls. This higher value indicates dryer

conditions in the study area except in 1988 when heavy rainfall occurred and all areas

were inundated. As noted previously, the south-central control 1993 WI was higher, or

drier, than the northern 1993 control WI (Figure 40 and Figure 43).











Yearly Conrtol and Forest GVI Means

35.00
30.00
25.00 ---GVI-N
20.00 --- GVI-S
I 15.00 GVI-SC
10.00 Forested
5.00
0.00
88 92 93 97
Year

Figure 42 Yearly Control GVI Means.



Control and Forest Class WI

20.00
15.00
-- WI-N
S10.00- -- WI-S
I 5.00 -A-WI-SC
0.00 Forested
0.00
-5.00 88 92 997
Year

Figure 43 Mean Control and Weighted Mean Forest WI Values.


Control and Water Buffer Distance Comparison

Histogram weighted water buffer distance index means for the forest class were

compared to weighted forest class means and control means. A graph of the GVI

comparison is presented in Figure 44. In the graph, the 1992 mean 0-meter buffer

distance (water) GVI equals 0 since the 1992 river boundary was used as the 0-meter

buffer class. Consequently, no forested vegetation occurred within the 1992 Om buffer

class. Higher Om means for all other years are a result of morphological changes in water

features. In general, the forested portions of the buffer distances follow control and total










forested trends. The 25m buffer distance more closely follows the northern control GVI

response. With the exception of the 0-meter class, means within the 25-meter zone are

lower than all buffer zones and years




Control and Forest Class GVI

35.00 GVI-N
30.00 GVI-S
3a o0-------,--- -* GVI-S
25.00 -- GVI-SC
20.00 Forested
15.00 -- Om
S10.00 -*- 25m
5.00 50m
0.00 100m
88 92 93 97 200m
Year Outside

Figure 44 Mean Control, Weighted Mean Forested and Buffer Distance Forested GVI
Values.


A graph of the WI comparison is presented in Figure 45. The graph shows all

forested buffer zone means were higher or dryer than controls. The 1992 WI mean

equaled 0 for reasons previously discussed in the buffer distance GVI comparison. The

graph shows 1988 was a relatively wet year, 1992 was an extremely dry year, 1993 was

intermediate and 1997 was slightly drier than 1993. While the wetness patterns are the

same as controls, they were drier than controls. Particularly, the 25m buffer distance

class was the driest in 1992, 1993, and 1997. Conversely, the outside buffer distance

class (200m+) was the most wet.











Control and Forest Study WI
--WI-N
20.00 --- WI-S
A-l- AWI-SC
15.00
S15.0 Forested
S10.00 -- Om
----25m
5.00
-I--- 50m
0.00 l100m
S5.00 88 92 93 97 200m
-5.00 200m
Year Outside


Figure 45 Mean Control, Weighted Mean Forest and Buffer WI Values.



Control Comparison Summary

The control and study area comparison can be summarized as follows:

Study area and control moisture changed across years.

1988 was the wettest year and 1992 was the driest.

Mean control WI patterns were similar to rainfall.

Study WI and GVI patterns approximated control WI and GVI patterns.

Study area forest WI's were higher, or dryer, than controls, except in

1988, a pre-diversion and very wet year.

South-central mean control GVI was higher than study area forest GVI.




Study Area Indices by Buffer Distance

The highest buffer distance GVI means occurred during 1993 and at distances

greater than or equal to 50 meters from water (Figure 46). For the same buffer distances,

1988 means were lowest. All means were aggregated at the 25-meter distance. High

turbidity visible in the 1988 image, and decreased river depth in 1997 may have

contributed to higher TM band 5 reflectance and lowered Tasseled Cap GVI.












GVI vs Water Buffer Zones


30.00
S20.00
S10.00
- 0.00
-a -10.00
-20.00
-30.00


- GVI-88
-W- GVI-92
GVI-93
GVI-97


Buffer Zone


Figure 46 GVI Means by Buffer Distance from Water.



As predicted by control WI means, 1992 study area WI buffer zone means were


highest, or driest, for all years (Figure 47). The WI means ranked from driest to wettest


are: 1) 1992, 2) 1997, 3) 1993, and 4) 1988. A lower WI mean for the Om 1997 buffer


distance may be a result of higher water levels reported at Rajka (Smith 2000) and


backwater effects occurring at the power channel and river channel confluence.





WI vs Buffer Zone


25.00

20.00

15.00

10.00

5.00

0.00

-5 00


-*- WI-88
-- WI-92
WI-93
WI-97


Buffer Zone


Figure 47 WI Means By Buffer Distance From Water.


0 5m 50m 100m 200m outside


(1992 Used as Om Start)

(1992, 1997 Drier than 1993)







Om 25m 50m 100m 200m outside








66



Separate yearly graphs of buffer distance GVI means versus WI means are


displayed in Figure 48 through Figure 51, and the negative correlations of GVI to WI and


NDVI to WI are presented in Table 18. Higher correlations were observed in 1992 and


1993, while lower correlations were observed in 1988 and 1997.





1988 GVI and WI vs Buffer


20.00

S15.00

10.00
5.00 -GVI-88
-u WI-88
0.00
o Om 25m 50m 100m 200m outside
S-5.00

-10.00
Distance


Figure 48 1988 Mean GVI vs. WI By Buffer Distance To Water.




1992 GVI and WI vs Buffer


30

20
10
GVI-92
0
7 -W- WI-92
-10 Om 5m 50m 100m 200m outside
20
2 -20
-30
Distance


Figure 49 1992 Mean GVI vs. WI By Buffer Distance To Water.











1993 GVI and WI vs Buffer


30
g 20
S10

0
-10
I -20
-30


-*GVI-93
-- WI-93


Distance


Figure 50 1993 Mean GVI vs. WI By Buffer Distance To Water.


1997 GVI and WI vs Buffer


--GVI-97
--- WI-97


Distance


Figure 51 1997 Mean GVI vs. WI By Buffer Distance To Water.



Table 18 Vegetation Index to Wetness Index Correlations by Year.
Comparison Correlation
(a) GVI to WI
1988-1988 -0.655
1992-1992 -0.921
1993-1993 -0.906
1997-1997 -0.610
(b) NDVI to WI
1988-1988 -0.687
1992-1992 -0.949
1993-1993 -0.940
1997-1997 -0.638


Om 25m 50m 100m 200m outside


0 ?5m 50m 100nnm n00m noitside












Study Area Indices By Land Cover


Graphs of index values by land cover class are presented in Figure 52 through


Figure 52. No simple linear correlations exist between index values based upon land


cover class.






Yearly Water Index Values


20.00
15.00
x 10.00
5.00
S0.00 -
o -5.00 93
-10.00
-15.00
-20.00

Water Category Year


Figure 52 Mean Water Class Index Values.




Yearly Forest Index Values


30.00
-
> 25.00
-
- 20.00

15.00
-O

10.00
-
5.00

0.00


A


--NDVI
--- GVI
WI


--NDVI
--- GVI
WI


88 92 93

Forest Category Year


Figure 53 Mean Forest Class Index Values.












Yearly Grass Index Values


25.00
20.00
15.00
10.00
5.00
0.00
-5.00
-10.00


+NDVI
--- GVI
WI


Grass Category Year


Figure 54 Mean Grass Class Index Values.




Yearly Exposed Index Values


5.00
S 0.00
g -5.00
-10.00
-15.00
-
-20.00
S-25.00
-30.00


-*NDVI
--- GVI
WI


Exposed Category Year


Figure 55 Mean Exposed Class Index Values.



Study Area Indices by Region


Upper, middle, and lower region GVI means and WI means were graphed and are


presented in Figure 56 to Figure 58. The graphed weighted mean upper and middle


region values, Figure 56 and Figure 57, indicate that as WI rises, or becomes drier, GVI


lowers and as WI lowers, or becomes wetter, GVI rises. Divergent lower region GVI and


WI means in 1997 may be due to backwater effects occurring at the power channel and


river channel confluence (Figure 58).


88 93 97


- 92 93 97
__ 88 _____ 22 ____ 3 ___ 97








70




Upper Region Yearly Index Values


j 20.000

x 15.000

- 10.000

5.000

S0.000

S-5.000


NDVI
--- GVI
---- WI


Up-88 Up-92 Up-93 Up-97
Index


Figure 56 Upper Region Mean Index Values.



Middle Region Yearly Index Values


S20.000

15.000

10.000

5.000

0.000


NDVI
--- GVI
-- WI


Mid-88 Mid-92 Mid-93 Mid-97
Mid-88 Mid-92 Mid-93 Mid-97


Index


Figure 57 Middle Region Weighted Mean Index Values.


Lower Region Yearly Index Values


14.000
12.000
10.000
8.000
6.000
4.000
2.000
0.000


("Back Flom








Low-88 Low-92 Low-93 Low-97
Low-88 Low-92 Low-93 Low-97


NDVI
-W- GVI
--- WI


Figure 58 Lower Region Weighted Mean Index Values.


-









Simple linear correlation between NDVI and WI or between GVI and WI does

not exist within a region and across years. Within an index and across image years,

regions were highly correlated with each other (Table 19).



Table 19 Index Correlations Across Regions.
Correlation (r)
Index Lower to Middle Lower to Uppe Middle to Upper
NDVI 0.905 0.875 0.985
GVI 0.895 0.749 0.893
WI 0.911 0.915 0.994


Within an individual year and across regions, NDVI to WI and GVI to WI correlations

were high except for 1997 (Table 20). In 1997 correlation was slight for NDVI to WI

and correlation did not exist for GVI to WI means. This may be due in part to the lower

region backwater effects.

Table 20 Correlations Within Year and Across Regions.
Year NDVI-WI GVI-WI
88-UML 0.973 1.000
92-UML 0.843 0.946
93-UML -0.984 -0.990
97-UML 0.699 -0.172


Study Area Indices by Country

Table 9 presents mean weighted index values for the study years. A graph of the

means is contained in Figure 59. The mean country WI values are closely matched and

the graph appears similar to the upper and middle region graphs, Figure 56 and Figure 57.

Slovakia GVI means for 1993 and 1997 were lower than Hungarian GVI means for the

same years. Correlations within an individual index across years and between countries

were very high, WI r = 0.99, GVI r = 0.99, and NDVI r = 0.98.













Table 21 Mean Weighted Index Values.
1988 1992 1993 1997
Hun-WI 0.763 11.593 5.852 8.660
Slov-WI -0.276 11.910 6.517 8.168
Hun-GVI 10.978 8.874 15.566 12.955
Slov-GVI 10.298 9.278 13.661 11.647
Hun-NDVI 0.390 0.392 0.443 0.445
Slov-NDVI 0.394 0.406 0.447 0.440


Country Index Values


S16 ovakia GV Lower iverson
S14


6 12




1988 1992 1993 1997
Year

Figure 59 Index Correlations between Countries.


- Hun-WI
- Slov-WI
-- Hun-GVI
Slov-GVI
-- Hun-NDVI
-*Slov-NDVI


The results of these comparisons contradict the mathematically modeled results of

Todd and Hoffer (1998). A lower WI indicates more infrared (IR) IR reflectance and

higher plant and background moisture. Higher WI equates with higher IR absorption and

less moisture. Surface water, which absorbs IR also produces higher WI values. A

positive GVI equates with higher IR reflectance and potentially healthier plant condition.

Similarly, Todd and Hoffer (1998) concluded their modeled results were different from

field study observations of a single plant species study with varying vegetation cover.














CHAPTER 5
CONCLUSIONS

There were 4 objectives of this research; 1) classify Landsat Thematic Mapper

images to assess land cover change, 2) assess plant condition and hydro-period change

using NDVI and Tasseled Cap GVI and WI as estimators; 3) compare and correlate

indices; and 4) identify trends in GVI and WI.

Through these objectives, five research questions were formulated: 1) were WI

values as a measure of surface moisture similar to precipitation patterns 2) were the study

area forest NDVI, GVI and WI patterns similar to the control area patterns 3) were GVI

and NDVI correlated 4) were there correlations between GVI or NDVI and WI 5) were

NDVI, GVI and WI patterns discernable with respect to proximity to water, land cover

class, region, or country?

The following conclusions were reached:

WI is a useful index for comparing environmental moisture to rainfall.

NDVI and GVI are well correlated, and changed across image years.

NDVI and GVI measurements indicate vegetation responds to

precipitation patterns.

Tasseled Cap GVI scaling factors better differentiate displayed and

measured changes in vegetation condition.

Histogram-weighted mean floating point WI values are inversely related

to environmental moisture patterns.









*Control WI and GVI may be used to normalize study area imagery for

variance in environmental conditions.



Satellite imagery can be a valuable tool for monitoring large-scale environmental

change, but the data must be properly pre-processed, interpreted and verified in order to

be useful. Pre-processing consisted of both geometric and radiometric corrections to the

imagery in order to "normalize" it over time.

Environmental variables, such as precipitation pattern, had to be taken into

account so that any change in vegetation as measured by the satellite imagery could be

attributed to the change in the hydrological regime caused by the diversion project

instead of a change in climate patterns. This was accomplished in this study by

incorporation of weather statistics and the use of "control" zones well outside the study

area. The weather statistics showed extremely wet and extremely dry conditions

occurred during the image years. Based on observations of control indices related to

these environmental moisture extremes, an elastic vegetation response may have occurred

and masked any satellite detectable impacts.

The conclusions of this study with respect to the environmental impacts of the

GBS on the Szigetkoz region were as follows:

Visible surface water area along the river channel and side branches

decreased.

Exposed Danube River banks were colonized by vegetation.

Satellite detectable moisture in the study area decreased.









Study area land cover changes were due to mostly forestry management

rather natural succession.

Post diversion hydrologic alterations such as side channel dams and weirs

increased study area moisture.



Therefore, it is the overall conclusion of this study that, while Landsat TM

imagery is, potentially, a very useful tool for assessment of large-scale environment

impact, a longer period of record is needed in order to ascertain the actual impacts of the

GBS on the Szigetkoz.















LIST OF REFERENCES


Abaffy, D., Lukac, M. and Liska, M. (1995). "Dams in Slovakia." Emil Holecka and
T.R.T. Medium, Bratislava, 1st Ed, 103.

Anderson, L. and Sivertun, L. (1991) "A GIS Supported Method for Detecting the
Hydrological Mosaic and the Role of Man as a Hydrologic Factor." Landscape
Ecology 5(2):107-124.

Anonymous (1993). "Biodiversity: There's a Reason for It." Science 262 (5139):1511.

Bauer, M.E., Burk, T.E., Ek, A.R., et al. (1994) "Satellite Inventory of Minnesota Forest
Resources." Photogrametric Engineering and Remote Sensing, 60(3):287-298.

Bartolucci, L.A., Robinson, B.F., and Silva, L.F. (1977) "Field Measurements of the
Spectral Response of Natural Waters." Photogrammetric Engineering and
Remote Sensing, 43(5):595-598.

Buttner, G., and Nador, G. (1992). Detection of Environmental Change by the
Gabcikovo Nagymaros Barrage Construction on the Danube River Using Landsat
Data.

Bittner, G., and Paraeda, S. (1993). "Using Randomly Selected Ground Control Points
for Automatic Registration of Satellite Images." Remote Sensing for Monitoring
the Changing Environment in Europe, Balkema Publishers, Budapest, 141-146.

Campbell, S., Ripley, K., and Snell, K. (2001) "Disturbance and Forest Health: Chapter
1, An Overview of Disturbance and Forest Health in Oregon and Washington."
United States Forest Service, http://www.fs.fed.us/r6/nr/fid/pubsweb/fhchapl.htm
(2/19/01).

Chavez, P.S. Jr., 1994. Automatic Detection of Vegetation Changes in the Southwestern
United States Using Remotely Sensed Images. Photogrametric Engineering and
Remote Sensing, 60(5):571-583.

Chelminski, R. (1993) "The Not-So-Blue Danube: A Storied Link between Europe's Old
and New." Smithsonian, 32-43.

Crist, E.P. and Cicone, R.C. (1984) "Application of the Tasseled Cap Concept to
Simulated thematic Mapper Data." Photgrammetric Engineering and Remote
Sensing 50(3):343-352.










Crist, E.P. and Kauth, R.J. (1986) "The Tasseled Cap de-Mystified." Photogrammetric
Engineering and Remote Sensing 52(1):81-86.

Crist, E. P., Laurin, P.R., and Cicone, R.C. (1986) Vegetation and Soils Information
Contained in Transformed Thematic Mapper Data." Proceedings of IGARSS
1986 Symposium, Zurich, 2:1465-1470.

Dobson, T. (1992) "Loss of Biodiversity: An International Environmental Policy
Perspective." North Carolina Journal of International Law, 17 (2) 277:xxx.

Ehrilich, P and Wilson E. (1991) "Biodiversity Studies: Science and Policy.", Science,
253(5021): 758-762.

ERDAS, (1997) "Field Guide." 4th edition. ERDAS, Atlanta.

Fisher, G.W. and Levine, E. (1999) "The Response of Vegetation to Change of Annual
Rainfall in the Sahel Region of Africa, and Its Dependence on Soil Type."
National Science Foundation, Project Varenious,
http://www.ncgia.ucsb.edu/varenius/ (2/19/01).

Forseth, I.N. (1997) "Plant Response to Multiple Environmental Stresses: Implications
for Climatic Change and Biodiversity." Chapter 12. Biodiversity II:
Understanding and Protecting Our Biological Resources, M. Reaka-Kudla, D.
Wilson and E. Wilson ed. Joseph Henry Press, Washington, D.C. 1997.

Gao, X., Huete, A.R., Ni, W. and Miura, T. (2000) "Optical-Biophysical Relationships
of Vegetation Spectra without Background Contamination." Remote Sensing of
Environment, 74(3):609-620.

Green, K., Kempa, D. and Lackey, L. (1994) "Using Remote Sensing to Detect and
Monitor Land-Cover and Land-Use Change." Photogrametric Engineering and
Remote Sensing 60(3):331-337.

Hoffer, R.M. (1978) "Biological and Physical Considerations in Applying Computer-
aided Analysis Techniques to Remote Sensor Data." In Remote Sensing: The
Quantitative Approach. P. Swain, and S. Davis editors, McGraw-Hill, New York.

Hungarian Academy of Sciences, CEC Working Group. (1993a) "Assessment of Impacts
of Gabcikovo Project and Recommendations for Strengthening of Monitoring
System." Report of Working Group of Monitoring and Water Management
Experts for the Gabcikovo System of Locks, Hungarian Academy of Sciences,
Budapest.









Hungarian Academy of Sciences Working Group (1993b) "The Szigetkoz:
Environmental Research, Environmental Status and Ecological Requirements."
Hungarian Academy of Sciences, Budapest, 145pp.

Hungarian Academy of Sciences Working Group (1994) "Environmental Risks and
Impacts Associated with the Gabcikovo-Nagymaros Project: Summary of the
Main Results of the Environmental Research Activities." Hungarian Academy of
Sciences, Budapest.

The International Commission for the Protection of the Danube River, Programme Co-
ordination Unit (ICPDR-PCU), (2000) "Hydroengineering." 6pp.
http://www.rec.org/DanubePCU/hydro.html (accessed 3/5/01).

International Court of Justice, (1997) "Case Concerning the Gabcicovo-Nagymaros
Project (Hungary/Slovakia)." International Court of Justice, September 25, 1997,
General List No. 92. pp. 65
http://www.lawschool.cornell.edu/library/cijwww/icjwww/idocket/ihs/ihsjudgeme
nt/ihs_ijudgment_970925_frame.htm (accessed 03/06/01).

Jensen, J.R. (1981) "Urban Change Detection Mapping Using Landsat Digital Data."
American Cartographer, 8:127-147.

Jensen, J.R. (1996) Introductory Digital Image Processing: A Remote Sensing
Perspective. 2nd ed. Prentice Hall, Upper Saddle River, NJ.

Kay, J.J. (1991) A Non-equilibrium Thermodynamic Framework for Discussing
Ecosystem Integrity." Environmental Management, 15(4):483-495.

Kidwell, K.B. (1990) "Global Vegetation Index User's Guide." U.S. Department of
Commerce, National Oceanic and Atmospheric Administration, National
Environmental Satellite Data and Information Service, National Climatic Data
Center, and Satellite Data Services Division, Washington, DC.

Kauth, R.J. and Thomas, G.S. (1976) The Tasseled Cap A Graphic Description of the
Spectral-Temporal Development of Agricultural Crops as Seen by Landsat."
Proceedings of the Symposium on Machine Processing of Remotely Sensed Data,
IEEE Catalogue 76, Ch. 1103-1, Purdue University, West Lafayette Indiana.
pp4B41-4B51.

Kurland, K., Fortunato, J., and Barcus, L. (1992) "Hungary Slovak Dam: Hungarian
Dam Controversy." Trade and Environment Data Base (TED), Case 34, 1(1),
16pp, http://www.american.edu/ted/HUNGARY.HTM (accessed 3/5/01).

Kushla, J.D. and Ripple, W.J. (1998) "Assessing Wildfire Effects with Landsat Thematic
Mapper Data." International Journal of Remote Sensing 19(13):2493-2507.









Lambin, E.F., 1996. Change Detection at Multiple Temporal Scales: Seasonal and
Annual Variations in Landscape Variables. Photogrametric Engineering and
Remote Sensing, 62(8):931-938.

Lauver, C.L. and Whistler, J.L. (1993) "A Hierarchical Classification of Landsat TM
Imagery to Identify Natural Grassland Areas and Rare Species Habitat."
Photgrammetric Engineering and Remote Sensing 59(5): 627-634.

Lichtenthaler, H.K. (1996) "Vegetation Stress: An Introduction to the Stress Concept in
Plants." Journal of Plant Physiology 148:4-14.

Lillesand, T.M., and Kiefer, R.W. (1994) Remote Sensing and Image Interpretation. 3rd
ed. John Wiley & Sons, New York.

Mather, P. (1987) Computer Processing of Remotely Sensed Images. John Wiley & Sons,
New York.

Mikos, L. (1988) "Space and Position: Scene of the Origin of Spatial Ecological
Landscape Problems." Proceedings VIII International Symposium, Problems of
Landscape Ecology, Bratislava: Slovakia Academy of Science, 1:52-72.

Muchoney, D.M., and Haack, B.N. (1994) "Change Detection for Monitoring Forest
Defoliation". Photogrametric Engineering and Remote Sensing, 60(10):1243-
1251.

Perlez, J. (1997) World Court Leaves Fight Over Danube Unresolved." New York
Times, September 26, Late Edition.

Petch, J.R., and Kolejka, J. (1993) "The Tradition of Lanscape Ecology in
Chzechoslovakia." Landscape Ecology and Geographic Information Systems,
Chapter 4. Taylor and Francis.

Pineda, F. (1992) "Biodiversity and the Quality of Human Life." Development 4:75-78.

Richardson, A.J. and Wiegand, C.L. (1977) "Distingushing Vegetation from Soil
Background Information." Photgrammetric Engineering and Remote Sensing,
43(12):1541-1552.

Richerson, P.J., and Lum, K. (1980) "Patterns of Plant Species Diversiy in California:
Relation to Weather and Topography." The American Naturalist, 116(4):504-
536.









Rouse, J.W., Haas, R.H., Schell, J.A., and Deering, D.W. (1974) "Monitoring
Vegetation Systems in the Great Plains with ERTS." Proceedings, Third
Resources Technology Satellite-1 Symposium, Goddard Space Flight Center,
NASA SP-351, Science and Technical Information Office, NASA, Washington
DC., pp 309-317.

Sader, S. (1995). "Spatial Characteristics of Forest Clearing and Vegetation Regrowth as
Detected by Landsat Thematic Mapper Imagery." Photogrametric Engineering
and Remote Sensing, 61(9):1145-1151.

Samson, P., and Charrier, B. (1997) "International Freshwater Conflict: Issues and
Prevention Strategies." Green Cross International, 36 pp.
http://www4.gve.ch/gci/water/gcwater/study.html (accessed 3/5/01).

Schott, J., Salvaggio, C. and Volchok, J. (1988) "Radiometric Scene Normalization
Using Pseudo Invariant Features." Remote Sensing of Environment, 26:1-16.

Schriefsver, J.R., and Congalton, R.G., 1995. "Evaluating Seasonal Variability as an Aid
to Cover-Type Mapping from Landsat Thematic Mapper Data in the Northeast."
Photogrametric Engineering and Remote Sensing, 61(3):321-327.

Serrano, L., Ustin, S.L., Roberts, D.A., Gamon, J.A., and Penuelas, J. (2000) "Deriving
Water Content of Chaparral Vegetation from AVIRIS Data." Remote Sensing of
the Environment 74(3):570-581.

Song, C., Woodcock, C.E., Seto, K.C., Lenney, M.P., and Macomber, S.A. (2001)
"Classification and Change Detection Using Lansat TM Data: When and How to
Correct Atmospheric Effects?" Remote Sensing of Environment, 75(2):230-244.

Smith, S., and Al-Rawaby, H. (1990) "The Blue Nile: Potential for Conflict and
Alternatives for Meeting Future Demands." Journal of Water Resources
International, 15, 217-22.

Smith, S., Buttner, G., Szilagyi, F., Horvath, L. (1996) "Analysis of Land Use/Land
Cover Change in the Szigetkoz Area Since Construction of the Gabcikovo Barage
System (Final Report)." University of Florida, Department of Civil Engineering
and the North-Transdanubian Environmental Inspectorate. Joint United States
Hungarian Research Fund 42pp

Smith, S.E., Bittner, G., Szilagyi, F., Horvath, L, and Aufmuth, J. (2000)
"Environmental Impacts of River Diversion: The Gabcikovo Barrage System."
Environmental impacts of river diversion; Gabcikovo barrage system." Journal
of Water Resources Planning and Management, 126(3): 138-145.









Smith, S., Dewitt, B., Gonzalez, P. and Hurt, G. (1995) "Georeferencing of Satellite
Imagery for Digital Soil Mapping." Journal of Surveying and Land Information
Systems, 55 (1) 13-20.

Swain, P. and Davis, S. (1978) Remote Sensing: The Quantitative Approach. McGraw
Hill, New York.

Todd, S.W., and Hoffer, R.M. (1998) "Responses of Spectral Indices to Variations in
Vegetation Cover and Soil Background." Photgrammetric Engineering and
Remote Sensing, 64(9):915-921.

Todd, S.W., Hoffer, R.M., and Milachunas, D.G. (1998) "Biomass Estimation on Grazed
and Ungrazed Rangelands Using Spectral Indices." International Journal of
Remote Sensing, 19(3):427-438.

Tomlin, C. D. (1983) "A Map Algebra." Proceedings of the Harvard Computer Graphics
Conference, Cambridge, MA.

Tomlin, C.D. (1990) Geographic Information Systems and Cartographic Modeling
(revised edition). Prentice-Hall, Englewood Cliffs, NJ.

Trade and Enviroment Data Base, (2000) "TED Case Studies: An Online Journal."
Cases 36, 69, 185, 186. http://www.american.edu/ted/ted.htm (accessed 3/5/01).

Tucker, C.J. (1979) "Red and Photographic Infrared Linear Combinations for
Monitoring Vegetation." Remote Sensing of Environment, 8:127-150.

Van Leeuwen, W.J.D., and Huete, A.R. (1996) "Effects of Standing Litter on the
Biophysical Interpretation of Plant Canopies with Spectral Indices." Remote
Sensing of Environment 55:123-138.

Van Niel, T.G. (1995) "Classification of Vegetation and Analysis of Its Recent Trends
at Camp Williams, Utah Using Remote Sensing and Geographic Information
System Techniques." Master of Science Thesis, Watershed Science, Utah State
University, Provo.















BIOGRAPHICAL SKETCH

Mr. Aufmuth has accepted a new Documents Library Faculty and GIS

Coordinator position with the University of Florida Libraries. While completing his

Master of Science degree Mr. Aufmuth has been a research and teaching assistant in the

geomatics program at the University of Florida's Department of Civil Engineering. Prior

to his enrollment in Civil Engineering, he spent 5 years as a geographic information

systems (GIS) manager and specialist in private consulting and government research.

Before becoming a Gator, he earned his Bachelor of Science degree in ecology, ethology

and evolution from the University of Illinois in 1984.

While working on his degree, Mr. Aufmuth spent a short time in Africa. He co-

instructed GIS and remote sensing short courses in Tanzania and Ethiopia. Amongst

other publications, he recently co-authored an article for ASCE 's Journal of Water

Resources Planning and Management on the effects of the Danube River diversion in

Hungary. GIS, remote sensing and data base development are his areas of concentration,

and he will provide a link to current University of Florida GIS efforts, the special needs

of those programs and the Library's mission to assist in research and education.




University of Florida Home Page
© 2004 - 2010 University of Florida George A. Smathers Libraries.
All rights reserved.

Acceptable Use, Copyright, and Disclaimer Statement
Last updated October 10, 2010 - - mvs