ESE: ‘IHHWHWIIIIIWIIHIWUHIIHIIHIHIHIIHIHHII

THIi‘SJQ

                                                                                                                           

31293 01555 6602

This is to certify that the

thesis entitled

TOPOGRAPHTC EFFECTS ON THE
NORMALIZED DIFFERENCE VEGETATION INDEX,
ROCKY MOUNTAIN NATIONAL PARK, COLORADO

presented by

David Frota Vaughan

has been accepted towards fulfillment
of the requirements for

MoA. degree in Geography

 

 

Major professor

Date 9/7/‘3Q

0-7639 MS U is an Affirmative Action/Equal Opportunity Institution

 

 

LIBRARY

Michigan State
University

 

 

 

* “ 'A

PLACE N RETURN BOX to remove this checkout from your recent.
TO AVOID FINES return on or betore date due.

DATE DUE DATE DUE DATE DUE

  

 

 

 

 

 

 

 

 

 

 

 

l__J-L__l
SEE]

MSU to An Atfinndlve Action/Equal Oppommlty lmtltulon
FIG-9.1

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

WEI,

TOPOGRAPHIC EFFECTS ON THE
NORMALIZED DIFFERENCE VEGE’I‘ATION INDEX.
ROCKY MOUNTAIN NATIONAL PARK, COLORADO
B y

David Frota Vaughan

A THESIS

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

MASTER OF ARTS

Department of Geography

1996

ABSTRACT

TOPOGRAPHIC EFFECTS ON THE
NORMALIZED DIFFERENCE VEGETATION INDEX,
ROCKY MOUNTAIN NATIONAL PARK, COLORADO

By

David Frota Vaughan

The normalized difference vegetation index (NDVI) has
become widely accepted because of its compensation for changing
illumination conditions and simplicity (Lillesand and Kiefer, 1994).
Although NDVI compensates, partially, for the effects of
topography on remote sensing measurements, the question
remains whether signal noise, attributed to the remaining
topographic effect, may explain variability in NDVI. Greater
understanding of the remaining topographic effect may yield more
accurate interpretation of NDVI in sensitive ecotones. Through the
examination of NDVI, solar incidence angle, and known vegetation
cover for a site in Rocky Mountain National Park, this thesis
explores the question of whether the calculated NDVI values are
related to topography and how strong is that relationship, if any,
relative to the vegetation influence. Statistical tests indicate a
significant relationship between incidence angle and NDVI. In
large, homogeneous areas, the “noise” of incidence angle may

account for up to 6% of the NDVI signal.

To my mother,
for encouraging me to continue

my education

iii

ACKNOWLEDGMENTS

I wish to thank my advisor, Dr. Daniel Brown, for his continued
support, encouragement, and guidance throughout this research
effort. In addition, I am grateful for the support of Dr. Jay Harman
and Dr. David Lusch, who gave willingly of their time and expertise
in order for me to complete the project.

Several others deserve to be mentioned. I would like to thank
Dr. Bruce Pigozzi for his patient and helpful explanations. Mike
Lipsey gave generously of his time whenever called upon.

Finally, I want to acknowledge Nat, Cath, Claudia, Dee, Jim, and
Reina for there ever-present friendship and unwavering support

throughout this project.

iv

TABLE OF CONTENTS

LIST OF TABLES ................................................................................................. vii
LIST OF FIGURES ................................................................................................ viii
CHAPTER:
1. INTRODUCTION AND RESEARCH QUESTION ....................................... 1
1.1 Remote sensing of vegetation ............................................... 1
1.2 Linking vegetation characteristics and radiance .......... 4
1.3 Vegetation indices .............................. , ....................................... 5
1.4 Background: Terrain's effect on remote sensing .......... 7
1.5 Problem. ......................................................................................... 1 1
1.6 Research question ...................................................................... 1 4
2. SITE DESCRIPTION ....................................................................................... 1 5
2.1 Study area ..................................................................................... 1 5
3. DATA AND METHODS .................................................................................. 1 9
3.1 Addressing the research question ...................................... 19
3.2 Data .................................................................................................. 20
3.2.1 Landsat Thematic Mapper image ............................ 20
3.2.2 Digital elevation model (DEM) .................................. 2 2
3.2.3 Vegetation cover map .................................................. 2 2
3 .3 Preparing the data ..................................................................... 2 4
3.3.1 Generating NDVI ............................................................. 24
3.3.2 Calculating incidence angle ........................................ 2 8
3.3.3 Shadow map ..................................................................... 33

3.3.4 Vegetation reclassification ......................................... 3 3

3.3.5 Raw and classed cos(i) ................................................. 4 1

3 .4 Methods .......................................................................................... 4 2

3.4.1 Sampling the data .......................................................... 4 2

3.4.2 Statistical techniques .................................................... 4 3

IV. RESULTS ........................................................................................................ 4 8
4.1 Statistical results ........................................................................ 4 8

V. DISCUSSION AND CONCLUSIONS ............................................................ 5 3
5 .1 Discussion ...................................................................................... 5 3

5 .2 Concluding remarks .................................................................. 5 6

5.3 Future research ........................................................................... 5 8

VI. APPENDICES ................................................................................................. 5 9
A. Characteristics of the Landsat TM ...................................... 5 9

B. Meta data of the Landsat image .......................................... 6 O

VII. LITERATURE CITED ................................................................................. 6 1

vi

Table
1.1

1.2
3.1

3.2
3.3
3.4
3.5
3.6
3.7
4.1

4.2

LIST OF TABLES

Correction of the topographic effect
by cosine manipulation ....................................................................... 1 0
Basic and modified vegetation indices .......................................... 12

Original groundcover classifications for Rocky Mountain

National Park ........................................................................................... 2 3
Vegetation cover classes ..................................................................... 3 4
Forest density classifications ............................................................ 3 5
Reclassified cover classes from original vegetation map ...... 3 7
Final classifications of composition/density classes ............... 3 8
Classification of cos (1'), range 0 - 1 ................................................ 4 1
Vegetation class occurrence within incidence classes ............ 4 7

ANOVA results for vegetation
and incidence angle groups ............................................................... 4 9
Regression results for

final 33 vegetation/density classes ............................................... 51

vii

LIST OF FIGURES

Figure

1.1 Deciduous and coniferous tree spectral signatures,

range 0.4 gm - 0.9 um ............................................................. 3
2.1 Study area in Rocky Mountain National Park ................ 16
3.1 Landsat TM image bands 4, 3, and 2
displayed in RGB ......................................................................... 2 1
3.2 NDVI image generated from radiance values ................ 27
3 .3 Solar Incidence angle, 1' ........................................................... 2 9
3 .4 Map of cos(i) ................................................................................. 3 2
3.5 Vegetation/density classification map and legend......39, 40

viii

CHAPTER 1
INTRODUCTION AND RESEARCH QUESTION

1.1 MW

Vegetation can be studied without physical contact or direct
observation because of its interaction with electromagnetic
radiation. Remote sensing is the science and art of obtaining
information about an object, area, or phenomenon through the
analysis of data acquired by a device that is not in contact with the
object, area, or phenomenon under study (Lillesand and Kiefer,
1994). This concept is now extended to include sensors aboard
earth-orbiting satellites. -

Electromagnetic energy registered by a passive remote
sensor1 originates primarily from the sun. As incoming solar
electromagnetic radiation (EMR) interacts with the earth's
atmosphere, three paths are possible: absorption by water or
aerosols in the atmosphere, reflection by clouds back into space, or
transmission through the atmosphere to the ground. Of the energy
that encounters the ground, some is absorbed and some energy is
reflected by ground cover. The reflected energy will be directed
back through the atmosphere and, possibly, to an earth-orbiting
remote sensing satellite. What is received by the satellite is known
as radiance, the total of energy radiated by a unit area per solid

angle of measurement (Lillesand and Kiefer, 1994). Not all ground

 

1 A passive sensor does not provide its own electromagnetic

energy source.

2
cover (e.g., vegetation) will interact with incoming EMR in the same

manner. The proportion of radiation that is reflected will vary with
both vegetation type and the portion of EMR (i.e., band wavelength)
in which the vegetation is being observed. The unique relationships
between vegetation characteristics and reflectance in the visible
and near-infrared portions (i.e., wavelengths of 0.4 - 0.9 um) of the
EM spectrum enable a passive satellite to efficiently record
information about vegetation.

The depiction of an object's reflectance characteristics across a
range of EMR is known as its “spectral signature.” The reflectance
for vegetation peaks in two separate wavelength ranges: the green
(0.5 - 0.6 pm) and near-infrared (0.7 - 0.9 pm) portions of the
spectrum (Figure 1.1). Reflectance within the green portion of the
spectrum is due to plant pigmentation (Lillesand and Kiefer, 1994).
Overall reflectance of visible light energy is diminished due to the
presence of chlorophyll-a and -b, which are highly absorptive of red
and blue light. In comparison with visible wavelengths, relative
reflectance from vegetation in near-infrared wavelengths is much
greater. Reflectance from plants within the near-infrared portion of
the spectrum is controlled by internal leaf structure. Large
airpockets within the leaf enable reflection of the longer
wavelengths (Curran, 1985; Gausman, 1977; Lillesand and Kiefer,
1994). Given sufficient spatial detail of the sensor, variation in
internal leaf structure and leaf morphology creates variations in
spectral signatures large enough to enable the classification of
vegetation, especially between major classes such as deciduous and

coniferous trees, from spectral information alone.

 

 

 

 

 

 

 

 

 

 

 

50
!
Deciduous trees
(Maple)
40 -
(Nate range of
spectral values)

“ 30 -
at .
7: Coniferous trees
8 (Pine)
0'
i3
"a
z

20 ~

10 -

O " 1'" I

Blue Green Red Reflected infrared
0.4 0.5 0.6 0.7 0.8 0.9

Wavelength (pm)

(Adapted from Kalensky and Wilson, 1975).

Figure 1.1. Deciduous and coniferous tree spectral signatures, range

0.4 um - 0.9 pm.

4
Other physical characteristics of vegetation affect the amounts

of EM energy reflected, absorbed, and transmitted by an individual
plant. Water content within the leaf, plant maturity and
senescence, and the presence of disease will all alter the
pigmentation and/or the structure of leaves and, thus, the amount

of energy reflected (Curran, 1985).

1.2 1.]. . l .. l l'

Reflected radiance measurements are related to physical
properties of vegetation over extents larger than individual plants
(e.g., of communities) if the ground resolution cell of the imaging
radiometer is much greater than the individual plant. Vegetation
type characteristics such as biomass, leaf area, species, and stress
have known effects on spectral response (Perry and Lautenschlager,
1984). Biomass is defined as the total amount of vegetation within
a specified region (Perry and Lautenschlager, 1984). Leaf Area
Index (LAI) is defined as the cumulative leaf area per unit on the
ground (Price, 1993).

The first to make the link between physical characteristics of
vegetation and radiance was Jordan (1969), who incorporated
vegetation response characteristics in the near-infrared (~0.8 um)
and red (~O.675 um) regions of EMR to derive, using a ratio of the
two bands, a measure of LAI (c.f., Tucker, 1979). Further
development of ratios between infrared and red radiance, as
reported by Tucker (1979), included the works of Colwell (1973)
and Rouse et a1. (1973, 1974). Colwell (1973) determined that the

ratio of infrared to red radiation normalized variation in soil

5
background reflectance and "was useful for estimating biomass"

(Tucker, 1979, p.128.).

1.3 mm

It was Rouse et al. (1974) who defined the term "vegetation
index": the ratio of radiance in one spectral band to that of another
for a particular satellite sensor (cf., Tucker, 1979). In general, a
vegetation index is a composite measure of spectral radiance
recorded in both the red and near-infrared regions of the
electromagnetic spectrum. Indices take advantage of the
differences in the spectral response characteristics of vegetation in
each of these wavelength channels. An index, therefore,
summarizes information recorded within several wavelength
channels into one variable/image that is representative of broad
vegetation characteristics such as species, leaf area, stress, or
biomass (Curran and Wardley, 1988; Perry and Lautenschlager,
1984). A wavelength channel of a particular satellite sensor is that
portion of the EM spectrum to which a sensor channel is sensitive.
In the case of Rouse et al. (1974), the sensor was the Multispectral
Scanner (MSS) of the early Landsat program and the index was
calculated by combining MSS channels 2 and 4
(0.6 - 0.7 pm and 0.8 - 1.1 um, respectively).

Early tests were performed to evaluate the relationships
between vegetation indices and vegetation (Tucker, 1979). Many
indices were "sensitive to the amount of photosynthetically active
vegetation present in the plant canopy" (Tucker, 1979, p. 134). The

width of the red and near-infrared band was shown to have little

6
effect on the radiance received, and thus, the derived vegetation

information. One effective index tested by Tucker (1979) was the
difference ratio:
(Infrared - Red) / (Infrared + Red)

This difference ratio became known as the normalized difference
vegetation index (NDVI) and has been widely utilized. Throughout
the development and testing of vegetation indices, combinations of
red and infrared were shown to be superior to the earlier ratios of
red and green in extracting canopy variables (Tucker, 1979).

Applications of vegetation indices have included monitoring
areally extents of food crops (Baret and Guyot, 1991; Guttrnan,
1991). The importance of global food production has led to the
daily monitoring of vegetation by the National Oceanic and
Atmospheric Association (NOAA) satellites, carrying the Advanced
Very High Resolution Radiometer (AVHRR). This instrument gathers
data in the red and infrared portions of the spectrum (Lillesand and
Kiefer, 1994), in addition to longer wavelength channels. The
ground resolution element (GRE) of an orbiting satellite system is
the area of land that the scanning sensor “sees” at any given time
and is equivalent to its spatial resolution. The GRE for the NOAA
series of satellites is 1.2 kmz. At such a resolution, the reflected
radiance, and therefore vegetation cover information, is
summarized with one value for an entire square kilometer. AVHRR
is appropriate for depicting large farming systems, such as those
present in the United States, because of the large areal extents of
the crop cover patches. The reduction of the data that occurs when

vegetation indices are used to summarize radiance recorded in

7
multiple wavelength channels allows for daily regional assessment

of crop phenology throughout the growing season (Brown et al.,
1993). The data reduction also enables global vegetation

monitoring (Goward et al., 1993; Gutman, 1991). The significance of
applications of vegetation indices to cropland is matched by
applications to forests and natural areas. In many naturally
vegetated areas, the ground is not as flat as cropland. More
undulating topography may have an impact upon the applications of

vegetation indices, specifically NDVI.

1.4 W

The topography of a region serves to inhibit the simple
application of spectral information derived from orbiting
multispectral scanner data (Karaska et al., 1986.) Many applications
to agriculture are not affected by topography due to the generally
flat nature of farmland. However, for non-agricultural applications,
specifically forestry, the influences of topography can complicate
interpretations (Leprieur et al., 1988, Meyer et al., 1993).

Variability in topography appears as differences in elevation,
hill slope, and aspect, the orientation of a slope face. Terrain
variability influences the amount of solar radiation striking any
given location by inducing shadowing and shading (Dubayah and
Rich, 1995). The geometric relationship between the Sun's position
in the sky and the orientation of the landscape will vary from one
location to another. Under clear sky conditions solar illumination
angle can be used to explain the amount of solar irradiance for any

given location on a landscape (Dubayah and Rich, 1995). Terrain

8
surrounding a given location may block direct solar radiation from

reaching that place: the area is, therefore, in shadow. Shading, on
the other hand, is a measure of the strength of the direct solar
radiation received at any given landscape position at some specified
date and time.

The impact of topography on multispectral sensor
measurements is termed the topographic effect: a phenomenon that
alters spectral reflectance from similar cover types due to
variations in slope and aspect of the terrain (Holben and Justice,
1980). The result is a greater variance than expected in satellite-
generated digital numbers (DN) for any given vegetation type due
to variations in terrain conditions. Digital numbers are the
nominally-scaled measurements at each of the satellite's sensors
and must be corrected for slight imbalances between each of the
detectors in order to calculate radiance (EOSAT NOTES, 1994).

A correction factor, based on the solar incidence angle and the
differences in radiance from a given cover type between flat and
inclined surfaces, may be calculated and applied to the original
scene to ameliorate the topographic effect. Solar incidence angle for
any given location within a landscape may be determined with the
use of a digital elevation model, a geometric representation of the
terrain surface using a grid system. The terrain orientation may
then be compared to sun positions to determine solar incidence
angle for each cell within the grid. The topographic correction
reduces the variance of measured radiance within known cover

types due to variations in topography. As a result, greater accuracy

9
is achieved by computer-conducted landcover classifications of

digital numbers (Civco, 1989; Teillet et al., 1982).

Attempts to correct for the topographic effect have taken
different forms. First, topographic data, and calculated maps of
incidence angle, have been used to adjust the digital numbers
directly (Civco, 1989; Frank, 1988; Holben and Justice, 1980; Meyer
et al., 1993) (Table 1.1). Although it is theoretically possible to do a
topographic correction to individual bands prior to NDVI calculation,
applications of NDVI do not include such manipulations, historically.
Topographically corrected data have not been used to calculate
NDVI because the topographic effect is wavelength dependent and
ratios of multiple topographically corrected bands may introduce
unknown biases into the derived values. In addition, the
topographic effect is reduced significantly by the nature of the
formulation of the NDVI.

The concept of using a ratio to reduce the topographic effect
was effectively demonstrated by Holben and Justice (1981). In
tests comparing red and infrared radiance values, ratios of
individual bands effectively reduced the topographic effect by a
factor of six (Holben and Justice, 1981). It was also reported that if
directional reflectance properties were wavelength dependent,
spectral band ratioing did not completely reduce the topographic
effect (Hoblen and Justice, 1981). In order for a ratio to be correct
in its application to reduce topographic influence, whether using
single bands or difference ratios, converting DNs to radiometric

units facilitates the accurate computation of NDVI (Price, 1987).

Table 1.1.

manipulation.

10
Correction of the tOpographic effect by cosine

 

Correction Method

Remarks

 

Statistic-empirical correction

LH=LT-cos(i)-m

Purely statistical approach
based on a linear relationship
between the original band and
the illumination. Oeometrically
the correction rotates the
regression line to the horizontal
to remove the illumination
dependence.

 

Cosine correction

l
L”: LT .cos(sz)]

l cos(i)

Trigonometric approach taking
into account the portion of
direct irradiance on the inclined
(pixel).
Objects are regarded as

surface element

Lambertian reflectors.

 

Minnaert correction
(semi-empirical)

Icos(sz) k
L =1. ——
H 1ilcosti)

Variation of the cosine
correction by introduction of a
Minnaert constant. simulating
the non-Lambertian behaviour
of the earth surface. With k=l
it is a normal cosine correction.

 

 

Cocorrection
(semi-empirical)

cos ($2) + c

f
L =L
H Tleos(i)+c

 

 

Modification of the cosine
correction by a factor c which
should model the diffuse sky
radiation. c is based on the
regression in the statistic-
empirical approach.

 

where:

82

i
k

c:

b/m =

Ln: radiance observed at horizontal surface
LT = radiance observed over sloped terrain
= solar zenith angle

sun's incidence angle

= Minnaert constant
correction parameter

In: inclination of regression line
b= intercept of regression line

(Reprinted from Meyer et al., 1993)

 

11
1.5P_r_Q_b_l_em

Of the many vegetation indices that have been developed,
several to measure vegetation (Table 1.2), NDVI has become widely
accepted because of its simplicity (Lillesand and Kiefer, 1994).
Although popular and widely applied, NDVI does not compensate
for all topographic effects (Guttman, 1991). Slight variations exist
in the bi-directional distributions (i.e., by illumination and viewing
angle) of red and near-infrared reflectance for a given landscape
(Guttman, 1991).

Evaluation of NDVI, as expressed in much of the literature, is
primarily focused on NDVI (generated from data acquired by the
AVHRR instrument on the NOAA series of polar-orbiting satellites.
AVHRR imagery covers a swath width of 2400 km (Lillesand and
Kiefer, 1994), making the view angle an important concern (Goward
et al., 1991; Guttman, 1993). View angle is the determinant of a
satellite's ability to detect surface illumination (Wardley, 1984). As
the earth's surface curves away from the satellite nadir position, for
any particular scan line, its ability to accurately detect surface

illumination is affected (Goward, 1991).

12

Table 1.2. Basic and modified vegetation indices.

10.

Normalized difference vegetation index (NDVI)
= (infrared - red) I (infrared + red)
Soil adjusted vegetation index (SAVI)
= [(infrared-red)/(infrared+red+L)] * (1+L)
where L = soil calibration factor
Modified soil adjusted vegetation index (MSAVI)
= { 2infrared +1 - [(2infrared + 1 )2 - 8(infrared-red)]-5}/ 2

Atmospherically resistant vegetation index (ARVI)
= (p*nir - p*rb)l(p* + p*rb)
where p*rb = p*r - ‘Y(p*b - p*r),
p*r = ozone absorption and molecular scattering

Modified soil and atmospherically resistant vegetation index
(MSARVI)

={2p*nir + 1 - [2p*nir + 1)2 - 8(p*nir - p*rb)]0-5}/ 2
Transformed vegetation index (TVI)

= (ND7 +0.5)-5

where ND7 = (CH7-CH5)/(CH7+CH5)
Modified TVI

= ((ND7+O.5)IABS(ND7+0.S)) *(ABS(ND7+O.5) -5
where ND7 = (CH7-CH5)/(CH‘7+CHS)
Difference vegetation index (DVI)

= 2.4CH7 - CH5
Ashburn vegetation index (AVI)
= 2.0CH7 - CH5

Tasseled Cap composed of 4 axes (Crist and Cicone, 1985)
Soil brightness index, SBI (brightness)

Green vegetation index, GVI (greenness)

= Yellow stuff, YVI

= Nonsuch, NSI

(Adapted from Ashburn, 1978; Deering et al., 1975; Huete, 1994; Kanth
and Thomas, 1976; Perry and Lautenschlager, 1984; Richardson and
Weigand, 1977; Tucker, 1980)

13
Vegetation indices are sensitive to solar elevation angle, solar

azimuth angle, and the look angle of the satellite platforms (Duggin,
1980; Kirchner and Schnetzler, 1981). An additional concern is the
coordination of NDVI generated from different platforms and even
the same platform at different times in its life cycle (Price, 1987).
Performance of satellite components is subject to conditions present
in orbit and the degradation of parts over repetitive usage. These
slight variations are significant when compared to the amount of
information within each pixel.

The evaluation of NDVI for finer-resolution satellites is
limited by factors other than those of the NOAA series of satellites.
Research into the local application of NDVI, generated by Landsat
MSS and TM, and satellites with spatial resolution better than 30 m2
has been directed toward the issue of noise. NDVI signals may be
affected by atmospheric conditions or the presence of soil patches
intermingled with ground cover within a pixel. Several
modifications to the standard NDVI have attempted to compensate
for these influences (Table 1.2). However, the question remains: to
what extent does the topography affects influence a fine-resolution
NDVI when it is applied to a localized area?

NDVI is a tool utilized to gather information about vegetation
in a particular region. Information gathered is input into studies
that examine global change (Baker et al., 1995; Overpeck et al.,
1990). Proper interpretation of NDVI, at the global or regional scale
of spatial resolution, is necessary for accurate understanding of
change. Baker et al. (1995) looked to the sensitive forest-tundra

ecotone as an indicator of global change. Understanding all sources

14
of noise within a fine-resolution NDVI signal is needed for accurate

assessment of possible climate change as indicated by altering
vegetation patterns in mountainous ecotones. Examination of the
remaining topographic effect within NDVI is one such possible

source of noise that needs to be investigated.

1.6 KW

The importance of vegetation indices necessitates further
research toward understanding those processes which affect the
signal received and the information content derived. The
tOpography of a region, the atmosphere through which the signal
must travel, and the spatial resolution of the satellite sensors all
have an impact upon the information obtained. Although all are
important areas of scientific inquiry, the scope of this study is
focused on the examination of topographic influence for a single
vegetation index, calculated for one time, under clear sky
conditions, and in a mountainous terrain. Formally expressed, the
research questions are: (1) Is NDVI sensitive to topographic effects?;
and (2) How strong is the topographic "noise" relative to vegetation
information ("signal") in the vegetation index image? In other
words, are the influences that are acting upon NDVI values, for a
mountainous terrain, solely related to vegetation or is there a

topographic influence as well?

CHAPTER 2
SITE DESCRIPTION

2.1 W
Rocky Mountain National Park (RMNP) is located within the

Colorado Front Range, northwest of Boulder, Colorado. This range
extends for 300 km from the Arkansas river in the south into the
state of Wyoming in the North(Peet, 1981). The study area is
located within RMNP, 40' 10' N to 40' 32' N latitude and 105' 31' to
105' 41’ west longitude (Peet, 1981). Specifically, the study site
includes a 27,000 hectare portion of Southeast RMNP (Figure. 2.1).

RMNP straddles the Continental Divide. Elevations range from
approximately 1830 m to roughly 4000 m. Much of the topographic
variation seen today is due to the differential weathering of
Pleistocene glaciated mountains (Allen et al., 1991). Richmond
(1960) reports that the protection of the unspoiled examples of
glaciation was a major impetus in the creation of RMNP. Underlying
geology is mostly Precambrian granites, gneisses, and schists (Peet,
1981).

Vegetation within RMNP has been described in the context of
life zones. A life zone attempts to organize vegetation within similar
regions based on moisture, winds, exposure, and topography (Nelson,
1953). One classification includes: foothills, montane, subalpine, and
alpine life zones (Peet, 1978b). One classification is based on climate
conditions at increasing elevation (Veblen and Lorenz, 1991). Each
zone contains characteristic vegetation that occurs due to climate
conditions which are chiefly controlled by elevation and moisture
(Peet, 1978a).

15

 

v.33
.1
‘2'? Mona” United States
~ c
has
. f’
Yv.’
l. 13’- do“
if}. if? ‘.'>~,_ o 20 miles
4.. ".1" Egg-2.; Morning
EkLE} .2.
r ('4 , 7%.?
; 3 ‘13:.
,' 9 .(
“‘3 " <3.
3::
~ i?
g; _*-r CO’Ofado
I; f?
1' E.
= as
Q33 .4
J ..

 

 

 

 

(Adapted from Brown, 1994.)

Figure 2.1. Study area within Rocky Mountain National Park,

17
Dominant vegetation types in Rocky Mountain National Park

include: Aspen (Papulus tremuloides) , Douglas-Fir (Pseudosuga
menziesii), Limber pine (Pinus flexilis), and Ponderosa pine (Pinus
ponderosa) forests (Chiou and Hoffer, 1994). Other forest
dominants include Blue Spruce and Subalpine Fir (Frank, 1988),
Alder (Alnus spp.), and Lodgepole pine (Pinus contorta). Non—forest
cover types include: alpine tundra, moist and dry meadows, bogs,
rock outcrops, krummholz, ponds, willow, grasses, and sedges
(Frank, 1988). Krummholz, a German word meaning twisted wood,
identifies the dense, low mats of spruce and fir trees in the
transition zone between forest and alpine tundra (Veblen and
Lorenz, 1991).

Climatic conditions will vary both with elevation and latitude
(Peet, 1978b). Vegetation response to climate conditions result in
species gradation between successive zones of climate. For any
particular species, the density and vigor of the individual examples
varies with minor gradients of climate. A given species may appear
on a variety of slopes and aspects. Appearance of a species will
vary in density and size according to presence or lack of ideal
conditions. These factors of topographic position and moisture
availability are closely related to forest composition (Peet, 1978b).
Limber pine, for example, occupies xeric sites between montane
forest levels and treeline (Peet, 1978). Higher, rockier elevations
within the park are most associated with the transition from forest
to tundra. Within this region can be found species of spruce, fir,
and occasionally Limber pine (Weisberg and Baker, 1995). Lower

elevations (i.e., montane) support more mesic environments with

1 8
other conifer species such as Douglas fir and Ponderosa pine

(Veblen and Lorenz, 1991). Variation within the general trend of
deciduous to subalpine to alpine and tundra species of vegetation
will be due to local conditions of climate. These variations include
maximum height attainment for a species, density of a stand, and
recovery from disturbances.

Remote sensing of the vegetation in RMNP is not new.
Previous studies have utilized Landsat data, combined with a
geographic information system (GIS) to analyze spatial patterns
within RMNP (Baker and Weisberg 1995). The utilization of digital
terrain information to complement the use of remote sensing and
GIS is exemplified by Brown (1994). Landsat TM data was
combined with topographic data to compare the relationship
between vegetation and topography at the sensitive alpine treeline
ecotone.

Analysis of vegetation through remote sensing serves both
current research initiatives and practicality. Baker and Weisberg
(1995) discuss the importance of understanding population
parameters to further comprehend the dynamic environment in
ecotones. Baker et a1. (1995) connect changes in the forest tundra
ecotone to global change. Determination of whether global change is
altering mountain vegetation communities is a monumental task
made more difficult by its remote physical environment. Remote
sensing, GIS, and digital terrain data offer a practical means to
understand vegetation communities in a terrain that is difficult to

access.

CHAPTER 3
DATA AND METHODS

3.1 W

This investigation will address the question of the degree to
which vegetation and topography influence NDVI by separately
modeling the influence of each factor using a remotely sensed
satellite image, a vegetation map, and a digital elevation model
(DEM). Topographic corrections have historically not been applied to
the satellite radiance values prior to calculation of the index. This
study will not stray from this precedent. The study is only plausible
due to the existence of detailed groundcover data with which to
compare and categorize the generated NDVI values. Again, the
research question is: Are the influences that are acting upon NDVI
values, for a mountainous terrain, solely related to vegetation or is
there a topographic influence as well?

Homogeneous vegetation and topography classes were
identified, using a digitized vegetation map and a DEM, respectively.
NDVI values within these areas were compared to assess their
influence. In previous studies, NDVI has been used as a surrogate
for leaf area index and biomass based on an assumption that
radiance is related to density and composition (Tucker, 1979).
However, with the use of detailed ground cover data, I examined the
possible influences of density and composition of vegetation and

topography on N DV1.

19

20
3.2 Data

3.2.1 LandsaLIhematiLMaDaer—image

Radiance data for RMNP are provided in a Landsat 4 Thematic
Mapper (TM) image (scene id # 425461765) acquired on July 5,
1989. Landsat TM provides ground resolution cells of 30 m x 30 m.
Specifically, channels 3 and 4 (0.63 - 0.69 um and 0.76 - 0.90 urn,
respectively) were used to calculate NDVI (Tucker, 1979).
Additional information about Landsat image channels may be found
in Appendix A. The image was georeferenced to the UTM
coordinate system, zone 13. The root mean square error (RMSE) for
the rectification was less than 30 meters (R. Thomas, unpublished).
RMSE is a measure of error between sample points in a rectified
image and their known locations on the ground; it is the distance
between input image control points and the same points after
rectification (Erdas, 1991). The areal coverage (677 rows by 451
columns) was a subset from the original scene and ranges from
(1061017 m, 405234 m) to (1051411 m, 394921 m) (Figure 2.1).
Additional information about the Landsat image may be found in
Appendix B.

The Landsat Thematic Mapper image (figure 3.1) of the test
site was originally displayed using bands 4, 3, and 2 (display colors
of red, green, and blue, respectively). Vegetation appears as shades
of red because of the dominance of near-infrared (band 4)

reflectance from vegetation.

21

 

Figure 3.1. Landsat TM image bands 4,3,2 displayed in RGB.

22
3.2.2 DigitaLElmtionJQdelJDEMl

A DEM for the RMNP study site was acquired from the United
States Geological Survey (USGS), corresponding to 1:24.000 scale 7-
1/2 minute topographic quadrangles (30 meter resolution). A
previous study by Brown and Bara (1994) has shown systematic
biases (striping) present in DEMs generated by photogrammetric
means. The production of DEMs involves the use of photographic
scanners and manual profiling which produce “striping” features in
the final product. A 1-by-3 filter was used to reduce the striping
effect present in the data (Brown and Bara, 1994). The vertical
accuracy of the DEM is reported at +/- 7 meters (USGS, 1987).

3.2.3 W

A detailed, digital vegetation map for Rocky Mountain
National Park was created by park conservation personnel and field
checked by Karl Hess (Colorado State University). Nine dominant
forest types and sixteen non-forest vegetation classes were
surveyed (Table 3.1). Forest types were broken down into multiple
sub-classes to include density data. In addition, the locations of
krummholz and rock outcrops were recorded where present.
Additional modifiers to vegetation classes included information on
disease, mountain beetle damage, disturbance, and rock outcrops. A
total of 409 different classes of vegetation was identified and
mapped. Landcover changes are assumed to be negligible for the

one year time duration between map creation and image capture.

23

Table 3.1. Original groundcover classifications for Rocky Mountain

National Park.

mum Size—Clams 11mm
Aspen O" - 5" O - 20%
Douglas Fir 5" - 10" 20 - 40%
Limber Pine 10" -15" 40 - 60%
Lodgepole Pine 15" - 20" 6O - 80%
Ponderosa Pine 20" + 80 - 100%
Spruce/Fir

Other—Tame:

Alder/Aspen

Bog

Cottonwood*

Wet meadow
Open water/pond
Blue Spruce“
Rushes/cattail
Willow

Wm WM
Disturbed/artificial Grass/forb
Rock Shrub/sage
Sandbar

Aim
Grass/forb
Willow

Modifications to the above classes:
disturbed, mountain beetle killed, krummholz, rock outcrop

* includes density and Diameter at Breast Height (DBH) data

24
3.3 W

33.1 W
Digital numbers (DN) of the Landsat TM bands were first

converted to radiance values for this investigation. The use of
radiance data is but one of several cautions necessary when
manipulating satellite imagery for analysis of vegetation indices.
Digital numbers are the nominally-scaled amplitudes of radiance
measured at each of the satellite's sensors and must be corrected
for slight imbalances between each of the detectors in order to
calculate radiance (EOSAT, 1994). The use of DNs is deemed
inappropriate for NDVI calculations due to the index's sensitivity to
intensity of irradiance and reflected radiance (Goward et al., 1993).

Other cautions are raised when comparing calculated NDVI
between sensors and between platforms. Any comparison of NDVI
between sensors or platforms should only be conducted with full
knowledge of the calibration procedures of each (Goward et al.,
1991). These cautions may seem to place extreme limitations on a
NDVI study of TM generated values. However, due to the nature of
this investigation, such cautions will not serve to limit the scientific
findings for two reasons. The first is that many of the above
cautions were based on findings of NDVI generated from Advanced
Very High Resolution Radiometer (AVHRR) data. This sensor has a
much larger field-of-view and, hence, much coarser spatial
resolution than the Landsat TM (Goward et al., 1991). By using one
Landsat image, assumptions of nadir (i.e., vertical) satellite position
may be made for all elements being investigated (i.e., pixels).

Secondly, this investigation does not include a comparison between

25
different sensors of the same satellite type nor is it a comparison of

NDVI generated from different platforms.

The objective is to investigate the possible systematic
influences of vegetation cover and topography on NDVI. The
combination of detailed ground data, a DEM, and a Landsat TM
image, converted to radiance values, enabled this investigation to
proceed without concern for many of the limitations cited in other
applications of NDVI (Goward et al., 1991; Guttman, 1993: Perry,
1984; Price, 1987). Conversion of DNs to radiometric units
facilitates the accurate computation of NDVI (Price, 1987). The

general equation for the conversion of digital numbers is:

DN'= Int[((DN-DNminx)/(DNmax7(-DNminx))*(maxi-minAH-minx] (3.1)
Where:
DN = digital number for an individual pixel,

DN' =_ recalculated digital numbers,

DNminx = minimum digital number in wavelength 1.,
DNmaxA = maximum digital number in wavelength 1,
min}, = mimimum radiance value recorded in wavelength A,

max), = maximum radiance value recorded in wavelength it.

Each transformation is wavelength band specific, although the
recalculated range of values (converted to integer range of O -
255) is based on the maximum range from both bands.

Numbers generated from the Landsat DN to radiance
transformation (Equation 3.1) were scaled to an integer range of 0-

255 for use in a geographic information system (GIS). The end

26
result of this calculation is the elimination of the slight mis-

calibrations of radiance measurements in each of the independent
sensors (EOSAT NOTES, 1994). Raw radiance values, used for the
calculation of NDVI, were adjusted only in scale to accommodate
analysis within an integer—only framework. Each pixel was
transformed using the ALGEBRA program within the ERDAS image
analysis software package.

Once rescaled to the same range of radiance, these two image
bands were then combined algebraically to create the NDVI index
(Figure 3.2). Lighter areas indicate a greater amount of vegetation
(i.e., photosynthetically active radiation, biomass, LAI, etc.) than
areas with darker shading. The influence of topography on NDVI
response is apparent. Lighter shades in the NDVI image tend to
correspond to the valley locations, whereas the ridge (i.e., higher

elevation) sites tend to have lower NDVI values.

27

 

Figure 3.2. NDVI image generated from radiance values.

28
3.3.2 CalculatinL'mcidenmugle

Irradiance, a measurement of available incident radiation for
any given location on the ground, may be used to determine the
extent of the topographic effect on NDVI. However, the term
irradiance implies that atmospheric conditions are known and
incorporated in its calculation. In this investigation, these
atmospheric conditions were assumed to be uniform due to limited
satellite scene size and clear sky conditions. Under clear sky
conditions, variability of incoming solar radiation is dominated by
direct irradiance (Dubayah and Rich, 1995). In place of the
irradiance measure, incidence angle will be used. Cosine of the
incidence angle, 1', is a measure of direct illumination as a function of
topographic position and is determined by the slope angle and
aspect at each pixel (Figure 3.3). It is the angle between the
normal to any given point and the direct path rays of the sun

(measured in degrees above the horizon and compass direction).

29

ll

Normal Vertleol

 

 

Solar Axlmuth I

/ H th
Axlmuth of Normal I or

to Slope /

Figure 3.3. Solar incidence angle, i. (Reprinted from Teillet et al.,
1982).

30
Other influences upon incoming solar radiation are diffuse sky

irradiance and reflected radiance, both direct and diffuse, from
nearby terrain (Dubayah and Rich, 1995; Proy et al., 1989;
Woodham and Lee, 1985). These additional influences on radiation
at the earth's surface are important, but not as influential as the
solar illumination angle. The simple application of mere solar
illumination to create radiation maps can only be accurate under
clear-sky conditions (Dubayah and Rich, 1995). Under any other
conditions, the radiation measurement must be augmented by
additional considerations of diffuse sky and reflected radiance
within a mountainous region. Again, for simplicity, the incident
radiation upon the topographic data were represented in cos(i)
form.

A map of cos(i) was first calculated using the DEM and
information about the sun's location at the time of image capture.
The incidence angle map was generated using the HILLSHADE

command in Arc/Info, without the shadow option (Equation 3.2):

cos(i) = 255 [ cos(S) sin(s) cos(a-A) + sin(S) cos(s) ] (3.2)
where:

s = terrain slope angle (calculated from DEM),

S = solar zenith angle (30’),

S = 90’- solar elevation (60'),

a = terrain slope aspect (calculated from DEM),

A solar azimuth angle (118').

Sun and sensor angles, necessary to calculate the incidence angle
map, were obtained from header information of the Landsat 4

seven band digital data.

3 l
The measure of incidence angle for each pixel is a variation of

a shaded relief map generated with the sun elevation and azimuth
positions corresponding to the time of satellite image capture.

Figure 3.4 is the cos(i) map utilized in this study.

3 1
The measure of incidence angle for each pixel is a variation of

a shaded relief map generated with the sun elevation and azimuth
positions corresponding to the time of satellite image capture.

Figure 3.4 is the cos(i) map utilized in this study.

32

 

Figure 3.4. Map of cos(i).

33
3.3.3 Slime

The calculation of the incidence angle map (Figure 3.4)does
not include the effect of shadows. A shadow is caused when
surrounding landforms completely block direct radiation from the
sun. A map of shadows was calculated from the DEM using sun
angles at the time of image capture. Each pixel of the DEM is
compared to its neighbors and the angle of direct sunlight (Dubayah
and Rich, 1995). Again, the command HILLSHADE was used in
Arc/Info to generate the shadow map. Pixels in shadow (a total of

33) were removed from further analysis.

3.3.4 W

In order to examine the relationships between vegetation
type and NDVI, the 409 cover type classes from the vegetation map
of RMNP were reclassified into groups of similar expected NDVI
response. Nineteen cover classes were formed by combining classes
according to similar ground cover types. For several of the forest
cover classes, density data and information On the presence of
krummholz and rock outcrops were detailed and used as
discriminating factors. Density classes of 0-20%, 20-40%, 40-60%,
60-80%, 80-100% were included for classes dominated by Douglas
Fir, Aspen, Limber Pine, Lodgepole Pine, Ponderosa Pine, Spruce/Fir,
and Blue Spruce. Eleven other classes of vegetation cover, forest
and non-forest categories, were included but were not augmented
by density data. Two classifications, one based on density and the
other on dominant vegetation type were cross-tabulated to produce
a total of 210 possible classes. Classes of dominant vegetation are

listed in Table 3.2, and density classes in Table 3.3.

Table 3 .2.

34
Vegetation cover classes.

C! E! El 'E' .

99:55”pr

10.
11.
12.
13.
14.
15.
16.
17.

18.
19.
20.

Other

Open water

Non-vegetated

Alpine

Dry meadow

Bog, wet meadow

Combinations of Lodgepole, Limber Pine, and

Spruce/Fir with the presence of krummholz

Combinations of Lodgepole, Limber Pine, and

Spruce/Fir without the presence of krummholz

Other combinations of conifers

Combinations of Ponderosa Pine and Douglas Fir

Blue Spruce

Mixtures of deciduous and conifer species

Combinations of Willow, Aspen, and Alder

Combinations of conifer forest with wet meadow

Combinations of deciduous forest with dry meadow

Empty classification

Alpine grass and willow appearing with rock outcrops

Combinations of Lodgepole, Limber Pine, and

Spruce/Fir with the presence of rock outcrops

Ponderosa Pine

Ponderosa Pine with presence of rock outcrops

Alpine species of grass and willow appearing with wet
meadow

35
Table 3.3. Forest density classifications.

Class Ilium
1. 0-20%
2. 0-20% with rock
3. 20-40%
4. 20-40% with rock
5. 40-60%
6. 40-60% with rock
7. 60-80%
8. 60-80% with rock

9. 80-100%

36
Of the 210 possible combinations of classes, only 39 were

shown to have actual representation on the ground. The list of
these classes may be found in Table 3.4. Several classes were
combined due to similarity of NDVI means, low standard deviations,
and similarity of vegetation. The final classification is listed in
Table 3.5.

Figure 3.5 depicts the final 33-class vegetation classification.
Notice that classes 7, 9, 11, 13, and 15 (i.e., combinations of Limber
pine, Lodgepole pine, Spruce fir, krummholz, and rock outcrops) are
the dominant forest types within this region. Therefore, the areal

extents of these classes is noticeably larger than the others.

37
Table 3.4. Reclassified cover classes from original vegetation map.

Class Descn'mian. ha. um 3112
0. Other 1831 74.5 13.0
1. Non-vegetated, no density data 10270 71.6 12.1
2. Alpine, no density data 2191 81.6 10.6
3. Alpine, 80 - 100% 232 80.5 13.8
4. Dry meadow, 80 400% 65 89.3 11.22
5. Bog, wet meadow, rushes, cattails, no density data 160 89 10.0
6. Bog, wet meadow, rushes, cattails, 80 -100% 59 91 14.2
7. Lodge Pole Pine, Limber Pine, Spruce Fir,

with presence of krummholz, O - 20% 82 84.1 9.8
8. Lodge Pole Pine, Limber Pine, Spruce Fir,

with presence of krummholz, 20 - 40% 28 86.6 7.1
9. Lodge Pole Pine, Limber Pine, Spruce Fir, with

presence of krummholz, 20 - 40% & rock outcrops 12 90.1 5.1
10. Lodge Pole Pine, Limber Pine, Spruce Fir, with

presence of krummholz, 40 - 60% 685 83.7 10.6
11. Lodge Pole Pine, Limber Pine, Spruce Fir, with

presence of krummholz, 60 - 80% 169 85.0 10.7
12. Lodge Pole Pine, Limber Pine, Spruce Fir, 0 - 20% 190 85.2 9.7
13. Lodge Pole Pine, Limber Pine, Spruce Fir, 20 - 40% 296 85.4 10.2
14. Lodge Pole Pine, Limber Pine, Spruce Fir, 20 - 40%,

with presence of rock outcrops 4 84.0 1.9
15. Lodge Pole Pine, Limber Pine, Spruce Fir, 40 - 60% 880 86.4 8.3
16. Lodge Pole Pine, Limber Pine, Spruce Fir, 40 - 60%,

with presence of rock outcrops 119 83.8 14.4
17. Lodge Pole Pine, Limber Pine, Spruce Fir, 60 - 80% 1776 87.1 7.6
18. Ledge Pole Pine, Limber Pine, Spruce Fir, 60 - 80%,

with presence of rock outcrops 194 80.9 11.4
19. Lodge Pole Pine, Limber Pine, Spruce Fir, 80 - 100% 6380 86.6 7.8
20. Other conifer mixtures, 0 - 20% 28 85.4 11.5
21. Other conifer mixtures, 0 ~ 20%, rock outcrops 56 93.0 7.6
22. Other conifer mixtures, 20 - 40%, rock outcrops 2 85.7 2.6
23. Other conifer mixtures, 40 - 60% 10 71.0 5.4
24. Other conifer mixtures, 40 - 60%, rock outcrops 8 84.5 2.0
25. Ponderosa Pine, Douglas Fir, 20 -40% 6 98.0 10.2
26. Ponderosa Pine, Douglas Fir, 40 ~60%, 4 77.6 16.6
27. Blue Spruce, 0 - 20% 1 92.3 4.5
28. Blue Spruce, 40 - 60% 14 67.3 9.3
29. Blue Spruce, 60 - 80% 6 92.4 9.4
30. Deciduous, conifer mixture, 0 - 20% 16 89.0 3.9
31. Deciduous, conifer mixture, 40 - 60% 4 80.6 6.2
32. Deciduous, conifer mixture, 80 - 100% 100 95.9 11.3
33. Willow, Aspen, Alder, 0 - 20% 19 83.7 4.7
34. Willow, Aspen. Alder, 40 - 60% 142 86.7 7.0
35. Willow, Aspen, Alder, 6O - 80% 18 91.2 7.2
36. Willow, Aspen, Alder, 80 - 100% 11 87.4 3.9
37. Forest, wet meadow, conifer 15 79.3 8.2
38. Alpine with presence of rock, no density data 1301 77.0 11.4
39. Alpine, wet meadow, no density data 100 92.0 7.8

38
Table 3.5. Final classifications of composition/density classes.

Skip

Clam WW5 AmLhAJEacmr
0. Other 12100 50
1. Alpine 2423 25
2. Dry meadow 65 4
3. Bog, wet meadow, rushes, cattail, and no data 218 5
4. Lodge Pole Pine, Limber Pine, Spruce Fir,

Krummholz, 0-20% 82 5
5. Lodge Pole Pine, Limber Pine, Spruce Fir,

Krummholz, 20-40% 28 3
6. Lodge Pole Pine, Limber Pine, Spruce Fir,

Krummholz, 20—40% + rocks 12 2
7. Lodge Pole & Limber Pines, Spruce Fir, Krummholz,

40-60% 685 15
8. Lodge Pole & Limber Pines, Spruce Fir, Krummholz,

60-80% 169 5
9. Lodge Pole & Limber Pines, Spruce Fir, Krummholz,

0-40% 486 10
10. Lodge Pole Pine, Limber Pine, Spruce Fir,

20-40% + rocks 4 1
11. Lodge Pole Pine. Limber Pine, Spruce Fir, 40-60% 880 20
12. Lodge Pole Pine, Limber Pine, Spruce Fir, 40~60%

40-60%+ rocks 119 4
13. Lodge Pole Pine, Limber Pine, Spruce Fir, 60-80% 1776 20
14. Lodge Pole Pine, Limber Pine, Spruce Fir,

60-80% + rocks 194 5
15. Lodge Pole Pine, Limber Pine, Spruce Fir, 80-100% 6380 25
16. Other conifer mixtures, 0-20% 28 3
17. Other conifer mixtures, 0-20% + rocks 56 4
19. Other conifer mixtures, 40-60% 10 2
20. Other conifer mixtures, 40-60% + rocks 8 1
21. Ponderosa Pine, Douglas Fir, 20-40% 6 1
22. Ponderosa Pine, Douglas Fir, 40-60% 4 1
24. Blue Spruce, 40-60% 14 2
25. Blue Spruce, 60-80% 6 1
26. Deciduous, Conifer mixture, 0-20% 16 2
27. Deciduous, Conifer mixture, 40-60% 4 1
28. Deciduous, Conifer mixture, 80-100% 100 5
29. Willow, Aspen, Alder, 0-20% 19 2
30. Willow, Aspen, Alder, 40-60% 142 5
31. Willow, Aspen, Alder, 60-80% 18 2
32. Willow, Aspen, Alder, 80-100% 11 1
33. Forest, Wet meadow, Conifer, 60-80% 15 2
34. Alpine, without density data with rock 1301 20
35. Alpine, Wet meadow, no density data 100 5

 

Other

Alpine

Dry meadow

Bog. wet meadow. rushes. cattail. no density data

Lodge Pole 8; Limber Pines. Spruce Fir. Krummholz. 0—20‘?

Lodge Pole 8; Limber Pines. Spruce Fir. Krummholz. 2040‘}
Lodge Pole & Limber Pines. Spruce Fir. Krummholz. 20—409’. rocks
Lodge Pole 8; Limber Pines. Spruce Fir. Krummholz. 40—60‘72
Lodge Pole 8; Limber Pines. Spruce Fir. Krummholz. (30-80“?
Lodge Pole 8; Limber Pines. Spruce Fir. Krummholz. 040‘?

Lodge Pole & Limber Pines. Spruce Fir. 204052; rocks

Lodge Pole & Limber Pines. Spruce Fir. 40-6098

Lodge Pole & Limber Pines. Spruce Fir. 40-6002. rocks

Lodge Pole & Limber Pines. Spruce Fir. Oil-80%

Lodge Pole & Limber Pines. Spruce Fir. (30—8097. rocks

Lodge Pole & Limber Pines. Spruce Fir. 80—100‘76

Other conifer mixtures. 0—20‘}

Other conifer mixtures. O—ZO‘T. rocks

Other conifer mixtures. 40-609é

Other conifer mixtures. 40—6092; rocks ~ . .
Ponderosa Pine. Douglas Fir. 20-40‘3 ' . . .51." 1 p ‘ 3:532
Ponderosa Pine. Douglas Fir. 40—609'2 if i J
Blue Spruce. 40-60%

Blue Spruce. 6080?}

Deciduous. conifer mixture. 0—20‘72-
Deciduous. conifer mixture. 40—6098
Deciduous. conifer mixture. 80-lOOC/r
Willow. Aspen. Alder. 080%
Willow. Aspen. Alder. 40—60%
Willow. Aspen. Alder. 60—80‘7r
Willow. Aspen. Alder. 80—100‘52

Forest. wet meadow. conifer. 60-809?

 

Alpine without density data

Alpine. wet meadow. no density data

 

Figure 3.5. Vegetation/density classification map legend. Figure 3-5 (Cont d).

 

41
3.3.5 RamnulasseLEQsIMnsidemauglfl

Cos(i) data were incorporated into the analysis in two forms:
(1) as a raw measure of illumination for each pixel location and (2)
as classed groups of similar cos(i). The ways in which the two forms
of incidence angle were used in the analysis are discussed in the
statistical techniques section of this chapter. The raw cos(i) values
were calculated using Equation 3.2 and the ALGEBRA program of
ERDAS.

The classification of cos(i) is a compromise between equal area
and equal interval approaches to classification. A histogram of the
incidence angle revealed a large number of pixels with values
greater than cos(i) = 0.6. A cut-off at cos(i) = 0.6 was used as the
upper limit of the first class of cos(i). The remaining four classes
divided the range of 0.6 - 1.0 into four equal intervals. The

resulting classification is listed in Table 3.6.

Table 3.6. Classification of cos(i), range 0-1.

Class 223.01 Amine.)
1. <0.6 3348
2. 0.6 - 0.7 3135
3. 0.7 - 0.8 5966
4. 0.8 - .09 8807
5. 0.9 - 1.0 6223

42

3.4 mm
341an

In a final preparation step, data on plant cover and incidence
angle, masked for shadows, were combined with a map of NDVI in
order to address the research questions. A program in the ERDAS
image analysis software package was used to generate samples of
pixels for NDVI, vegetation class, incidence angle, and incidence
angle class.

The pixel values for each. of the variables are highly
correlated to those areas (pixels) immediately surrounding it
(apparent in their non-random patterns in Figures 4.1, 4.3, and 4.5).
This spatial autocorrelation is reduced with increased distance away
from the original pixel. To alleviate the potential biasing of
statistical tests associated with this phenomenon, a stratified,
systematic sampling was adopted. Pixels were sampled at regular
pixel intervals in the x and y directions using the following skip
factors: 50, 20, 15, 10, 5, 4, 3, 2, l. The skip factor used for each
class was set according to the largest interval that yielded
approximately 30 samples, a minimum value necessary for
statistical analysis. Many of the smaller classes required a very
small skip factor and even no skip factor at all (Table 3.5). The
adopted approach allowed for a reduction of the effects of spatial
autocorrelation to the maximum extent possible while including as

many classes as possible.

43
3.4.2 StatistisaLteshnigues

A combination of bivariate regression and analysis of variance
(ANOVA) statistical tests were performed to evaluate the
hypothesized relationships. Again, the research question: Are the
influences that were acting upon NDVI values, for a mountainous
terrain, solely related to vegetation or is there a topographic
influence as well? ANOVA tests were employed to determine if
sample groups of NDVI had significant differences. Bivariate
regressions were used to test the expected relationship between
NDVI and each of the variables tested. Results of these regressions
address the more specific question: “Of what form is the
relationship?” The strengths of the relationships were tested for
statistical significance.

Two groups of statistical tests were employed, one based on
classes of pixels by vegetation type and the second by incidence
angle class. Each group of statistical tests involved two steps. In
the first step, ANOVA statistical tests were conducted to examine
the similarity or difference of NDVI values by class. The first group
of statistical tests focused on the differences in NDVI for the 33
classes of vegetation/density. The second group of statistical tests
was directed at the differences in NDVI for classes of similar
incidence angle. In the second step, within each classification (i.e.,
by vegetation type and incidence class) the influence of the other
variable on NDVI was assessed using either bivariate regression (in
the case of incidence angle) or ANOVA (in the case of vegetation

class).

44
The first group of tests was based on vegetation type classes.

ANOVA was employed to test the differences within and between
groups. The ANOVA test determined whether the groups of

vegetation have significantly different NDVI values.

H0: in = u2 = 113 = = u35 The samples of NDVI for each
vegetation class are drawn from the same population

Ha: u1<>u2<>u3<>...<> u35 At least one sample is drawn from
a different population

Should the null hypothesis be rejected, signifying that at least one
sample is drawn from a different population, then it can be
concluded that NDVI is related to vegetation type.

The influence of incidence angle on NDVI for groups of similar
vegetation was then evaluated using bivariate regression. NDVI
values were extracted by the 33 vegetation/density classes. Within
each of the classes of vegetation, NDVI values were regressed with

values of incidence angle, cos(i).

Regression for vegetation groups 1-35:
NDVI = a + b[cos(i)]

Ho: b = 0; There is no relationship between NDVI and

incidence angle for pixels in this vegetation
class.

Ha: b <> 0; A relationship exists between NDVI and incidence
angle for pixels in this vegetation class.

45
This statistical test aided in the determination of whether or not

NDVI is affected by incidence angle, how strong the relationship is,
and the direction of the relationship. These tests were performed
while controlling for the influence of vegetation type, because
vegetation type and incidence angle may be interrelated.

The second group of statistical tests was based on incidence
angle classes. Groups of incidence angle were reclassified to five
ranked classes (Table 3.6). ANOVA was employed to test the

differences within and between groups.

Ho: [11 = u2 = u3 = 114 = as The samples of NDVI for each .

incidenceangle class are drawn from

the population
Ha: u1<>u2<>u3 <>u4<>u5 At least one sample is drawn from a

different population

Should the null hypothesis be rejected, then it could be said that
incidence angle accounts for some of the variability of NDVI. This
information alone is not complete, because, although incidence angle
may influence NDVI measurements directly, it also may be related
to NDVI indirectly by affecting vegetation patterns. Brown (1994)
showed that vegetation patterns are sensitive to levels of exposure
to solar radiation.

Next, from the vegetation classification, each class of
vegetation was revalued to take on its average NDVI value (i.e., its

expected NDVI). This expected NDVI (NDVIe) was to be regressed

against actual NDVI values, for each incidence angle class, to

46
discover influences of vegetation on groups of similar incidence

angle.

Planned regression for incidence angle groups 1-5
NDVIe vs. NDVI For each ranked incidence group:

NDVI = a + b(NDVIe)

Ho: b = 0; There is no relationship between predicted NDVI, for
this class of incidence angle, and actual NDVI.

Ha: b <> 0; A relationship exists between expected NDVI, for this
class of incidence angle, and actual NDVI.

This statistical test was to aid in the determination of whether or
not the NDVI values for groups of similar incidence angles are
related to vegetation.

With incidence angle held constant, the relationship between
vegetation type and NDVI could not be tested with bivariate
regression. Too few vegetation classes per incidence class (Table

3.7) did not provide enough different NDVIe values to run the

bivariate regression. The analysis was, therefore, not conducted.

47
Table 3.7. Vegetation class occurrence within incidence classes.

Vegetation Incidence Class

Class 1 2 3 4 5 Total
0. 9 6 6 9 16 46
l. 6 4 11 17 2 40
2. 0 O 3 29 14 46
3. O 3 3 78 8 92
4. 0 O 40 10 20 34
5 0 0 10 14 16 31
6. O 0 0 21 14 35
7. 1 2 6 11 13 33
8. 0 3 16 20 30 69
9. 0 4 12 14 17 47
10. 0 3 35 2 0 40
ll. 3 2 6 6 9 26
12. 20 4 8 20 33 85
13. O 6 8 16 11 41
14. 15 25 22 17 8 87
15. 5 8 25 41 17 96
16. O 0 O 22 13 35
17. O 5 11 12 14 42
19. 17 10 0 3 0 3O
20. O 11 75 1 O 87
21. 0 O 0 0 63 63
22. 0 0 0 28 9 37
24. 0 O 1 l4 19 34
25. 0 7 18 22 14 61
26. 0 0 9 29 2 40
27. 0 0 3 34 1 38
28. 0 0 0 5 37 42
29. 9 15 26 0 0 50
30. l 12 17 24 6 6O
31. 0 69 0 22 20 111
32. 0 0 23 16 13 52
33. 33 0 25 0 15 73
34. 0 5 8 17 7 37
35. 0 1 18 22 2 43
totals 119 205 400 596 463 1783

CHAPTER 4
RESULTS

4.1 StatistiaaLLesults

Table 4.1 depicts the ANOVA results for the two groups of
classed pixels (i.e., by vegetation type and incidence class). These
tests were utilized to determine whether significant relationships
existed between NDVI and the tested variables of vegetation and
incidence class. Specifically, (I) is NDVI related to vegetation type?
and (2) is NDVI related to incidence angle?

Within each group of statistical tests, the ANOVA test was
applied to two different sets of pixels. The first group included all
sample pixels, regardless of skip factor. In the presence of spatial
autocorrelation, statistical tests are more likely to yield significant
results than when spatial autocorrelation is controlled. Therefore the
second group utilized only those pixels sampled with a skip factor of
210. This second test examined the ANOVA relationship on those
pixels that were sampled in a manner that limited the effects of
spatial autocorrelation. The results indicate that, in each case, NDVI
is significantly (p < 0.05) related to vegetation class and incidence
angle class.

These initial ANOVA tests are not without their potential for
misinterpretation. The relationships being reported, for groups of
similar vegetation class and incidence angle, may potentially be the
same interrelationship. To avoid misinterpretation, one variable
must be held constant while the other is tested. Bivariate
regressions were conducted to test the strength and directionality of

the relationships between NDVI and the tested variables of

48

49
Table 4.1. ANOVA results for vegetation and incidence angle

groups.

ANOVA A: Grouped by vegetation/density class

1. All sample pixels, outliers removed
Chiz = 540.075 DF = 33 p = 0.00
source sum-of-squares DF mean-square F—ratio p
between groups 67394.31] 33 2042.252 34.411 0
within groups 101428.523 1709 59.350

2. Sample pixels of with skip factor 210 (O, 1, 7, 9, 11, 15,

34).

Chi2 = 49.687 DF=7 p = 0.00
source sum-of—squares DF mean-square F—ratio p
between groups 12373 7 1767.573 23.621 0
within groups 26939.271 360 74.831

ANOVA B: Grouped by incidence class

1. All sample pixels, outliers removed
c1112 = 90.724 DF=4 p = 0.00
source sum-of—squares DF mean-square F-ratio p
between groups 16394.993 4 4098.748 46.734 0
within groups 152427.841 1738 87.703

2. Sample pixels with skip factor 210 (0, 1, 7, 9, 11, 15,

34).

C1112 = 15.836 DF=4 p = 0.003
source sum-of—squares DF mean-square F-ratio p
between groups 2164.932 4 541.233 5.289 0

within groups 37147.343 363 102.234

50
vegetation type and incidence angle. NDVI values were regressed

against corresponding raw cos(i) values for the samples within each
vegetation/density class. Table 4.2 lists the number of samples,
(x,y) skip factor, r2, f-ratio, p values, and the regression line
coefficients for each regression of vegetation class.

With vegetation held constant, a relationship between
incidence angle and NDVI was significant if it had a low p value (p<
.05). The direction of the relationship (i.e., whether NDVI increases
or decreases with any increase in incidence angle) is given by the
sign (+ or -) on the regression coefficient (b) for incidence. The
value of r2 is a measure of how strong the entire relationship is
between NDVI and incidence angle. Larger r2 values indicate that
more of the variation in NDVI is explained incidence angle. The
label It indicates the number of samples used for that regression.

A large skip factor present in a vegetation class indicated that
the vegetation class was sampled in a way that limited the effect of
spatial autocorrelation. The larger the skip factor, the more the
effect was limited. A vegetation class with a skip factor of 10 or
greater may be interpreted with more reliability than one with a
low skip factor (i.e., 5 or less). Four of these vegetation classes had

significant relationships between incidence angle and NDVI:

Class #‘7 (p =0.011) Lodge pole & Limber pines, Spruce Fir,

krummholz, 40-60%,
Class #9 (p =0.00S) Lodge pole & Limber pines, Spruce Fir, krummholz, 0-40%,
Class #11 (p =0.001) Lodge pole & Limber pines, Spruce Fir, 40-60%, and

Class #15 (p =0.001) Lodge pole & Limber pines, Spruce Fir, 80-100%.

51

Table 4.2. Regression results for final 33 vegetation/density
classes.

skip constant incidence

Class a. factor :3 F_-Rati.0 a. La.) £12.)

0. 46 50 0.078 3.704 0.061 57.507 +0071
1 4O 25 0 0.012 0.914 84.425 -0.006
2. 46 4 0.149 7.698 0.008 37.771 +0242
3. 89 5 0.032 2.875 0.094 55.669 +0.164
4. 28 5 0.088 2.515 0.125 99.34 -0.072
5. 31 3 0.061 1.882 0.181 107.094 -0.087
6. 35 2 0 0.012 0.913 87.264 +0012
7. 33 15 0.191 7.309 0.011 57.7 +0.13]
8. 69 5 0.008 0.551 0.461 80.761 +0033
9. 47 10 0.163 8.752 0.005 57.901 +0131
10. 39 1 0.06 2.358 0.133 94.687 -0.058
11. 27 20 0.388 15.878 0.001 68.53 +0.086
12. 85 4 0.333 41.399 0 62.474 +0.116
13. 41 20 0.049 2.06 0.165 75.697 40.058
14. 87 5 0.001 0.109 0.742 78.586 40.012
15. 96 25 0.061 6.94 0.015 76.284 +0053
16 35 3 0.056 1.944 0.173 127.129 -0.175
17 42 4 0.001 0.034 0.855 92.332 +0007
19. 30 2 0.019 0.537 0.47 75.659 -0.029
20. 87 1 0 0.022 0.882 83.784 +0.004
21. 63 1 0.363 34.777 0 -194.93 +1.186
22. 37 1 0.013 0.444 0.509 32.774 40.212
24. 34 2 0.002 0.049 0.826 76.377 -0.029
25. 61 1 0.31 26.446 0 51.6 +0.2

26. 40 2 0.093 3.902 0.056 71.04 +0.086
27. 38 1 0.031 1.144 0.292 112.90 1-0.149
28. 42 5 0.057 2.42 0128 37.663 +0.238
29. 49 2 0.255 16.08 0 67.843 +0095
30. 61 5 0.135 9.202 0.004 66.869 +0.1

31. 42 2 0.386 25.128 0 -21.806 +0494
32. 121 1 0.125 16.957 0 77.06 40.054
33. 41 2 0.609 60.641 0 25.753 +0245
34. 38 20 0.048 1.798 0.188 97.106 -0.107
35. 43 5 0.001 0.047 0.829 95.546 -0.017

52
These 4 classes are variations of one of the dominant vegetation

types within the study site. Large areal extent of vegetation cover
(i.e., large enough to allow sampling that limits the effect of spatial
autocorrelation) may limit the applicability of a vegetation
type/density combination for investigation. Discussion of the
results and conclusions of the ANOVA and bivariate regressions are

presented in Chapter 5.

CHAPTER 5
DISCUSSION AND CONCLUSIONS

5.1 Disgussjsm

Incidence angle influence on NDVI measurements for
vegetation classes was determined through (a) ANOVA based on
classified incidence values and (b) bivariate regression of NDVI
versus incidence angle for vegetation classes. ANOVA tests were
conducted for NDVI values within and between each of the
vegetation classes and for NDVI values between each of the classes of
incidence (Table 4.1). ANOVA tests were tabulated for all pixel
samples and for pixels from vegetation groups with skip factors 210.

Results from each group of tests showed that by increasing the
skip factor to partially compensate for spatial autocorrelation, f-
ratios and chi-squared values decreased. The relationship of
incidence angle and NDVI are only reliable for those vegetation
classes whose sampling skip factors partially compensated for spatial
autocorrelation.

ANOVA results indicated a significant relationship between
NDVI and vegetation type (p=0). Also indicated was the presence of
a significant relationship between NDVI and incidence class (p=
0.003). The NDVI relationship was stronger with vegetation class (F:
23.62) than with incidence angle (F=5.29). However, the
relationships between NDVI and incidence class were affected by the
possibility that incidence angle may have influenced vegetation type.

The variables were then examined independently.

53

54
When vegetation was held constant, incidence angle was

shown to have an influence on the signal of NDVI. In all cases, the
r2 values for each of the 33 vegetation/density classes were low
(ranging from zero to 0.61). Only 13 of the 33 regressions, by
vegetation class were significant. In each case the relationship was
positive, meaning that an increase in incident angle tended to
correspond with an increase in NDVI. Of the 13 significant
regressions, only 4 proved significant while also having a skip factor
210 pixels. R2 values for these regressions suggested that incidence
angle influenced as much as 39% of the variation in NDVI.
Relationships of the vegetation classes, having significance but not
large skip factors, were not considered because spatial
autocorrelation in the samples can artificially inflate the significance
values.

As skip factor decreased within the vegetation classes, a
greater number of significant regression relationships between
incidence and NDVI appeared. The observation indicated the
importance of including skip factor as a means to limit the impacts
of spatial autocorrelation present. Only 4 regressions had
significance and skip factors 2 10. These four were:

Class #7 Lodgepole & Limber Pine, Spruce Fir, and Krummholz, 40-60%;
Class #9 Lodgepole & Limber Pine, Spruce Fir, and Krummholz, 0-40%:
Class #11 Lodgepole & Limber Pine, Spruce Fir, without Krummholz, 40-60%;
Class #15 Lodgepole & Limber Pine, Spruce Fir, without Krummholz, 80-

100%.

These results indicated, at least for the four significant classes at

skip factors of 210, that incidence accounts for some noise

55
associated with the NDVI signal. The observation that the four

vegetation classes came from the same dominant mix of vegetation
should not be overlooked. These classes cover large areal extents
and appeared under different topographic conditions (Table 4.3).
Neither the presence of krummholz nor the density of the
vegetation class appear to have much importance on incidence
affecting NDVI. However, these observations were limited to the
four significant relationships. The factor that may have influenced
these observations is areal extent and whether the skip factor was
large enough to compensate for autocorrelation. This limitation may
have been the reason that some other classifications were not
significant. It may also have limited the implications of this study;
because only large extents of similar vegetation were sampled at a
skip factor large enough to partially compensate for spatial
autocorrelation, many stands of cover type were excluded from
examination. For topographic influence upon NDVI to be detected
in this investigation, a stand must have been large enough to have
been sampled at a large distance using appropriate sampling
schemes.

The relationship between vegetation and NDVI may only be
interpreted as far as the ANOVA tests allow. The relationship
between vegetation class and NDVI was strong, and stronger than
the relationship between NDVI and incidence angle. The analysis of
the relationship between NDVI and vegetation type within
incidence classes was not conducted due to a small sample size in

some of the class combinations (Table 3.7).

56
5.2 Concluding—masks

The examination of topographic influence upon NDVI has
revealed that incidence angle may cause some “noise” within the
signal from vegetation in alpine environments. However, results of
this study are limited to one grouping of related vegetation types.
The spatial autocorrelation present within the variables inhibits the
examination of all vegetation groups. In applications of NDVI for
vegetation mapping, topographic influence may be a minimal source
of error for large extents of vegetation. Smaller extents of
vegetation were excluded from this investigation of the influence of
incidence angle on their NDVI.

The results of this study are also limited by the nature of the
data set. While validity of the NDVI signal may be assumed for the
more dense vegetation classes (i.e., density >60%), caution is
warranted for lower density classes. In vegetation classes of lower
density, understory vegetation was not included in the analysis.
This limitation of the data may explain the large degree of NDVI
variability within each of the vegetation classes of low density and
the apparent discrepencies for relative NDVI values across classes.
Therefore, the most trustworthy result of this investigation is that
of vegetation class #15, Lodgepole pine, Limber pine, Spruce Fir, 80-
100%, which covers an area of 6380 hectares. Thus, incidence angle
introduces noise within the NDVI signal for some expansive
vegetation classes of high density (r2 = 0.06) within a mountainous
terrain. Tests of the relationship between NDVI and vegetation
type indicated a strong, general trend, but were not tested for

strength and directionality.

57
Information about error of any amount within NDVI is

valuable for proper interpretation of the index. Investigations into
the influence of noise upon the vegetation radiance measured by
passive remote sensing satellites have led to the development of
soil and atmospheric corrections to radiance values. Modifications
to NDVI have addressed the same noise influences caused by the
atmosphere and presence of soil adjacent to vegetation. Noise
associated with topography has been assumed to be eliminated by
the ratio construction of the NDVI. However, variations in the NDVI
that are not related to ground cover may lead to misinterpretations
of the index in sensitive ecotones.

Scientists are looking to the forest-tundra ecotone as
indicators of global change (Baker et al., 1995). The vegetative
measures are gathered to monitor changes and adaptations that
may indicate large, global changes in climate. The hypothesis is that
should climate conditions change, it will affect the forest-tundra
ecotones first. Any noise within the vegetation indices must be
investigated for the proper interpretation. Within large,
homogeneously vegetated terrain, an NDVI variation as large as 6%
may be attributed to the topographic effect. Smaller, non-
homogeneous areas were not able to be accurately assessed.
Additional investigations are needed to further characterize the
remaining topographic effect in mountainous terrain. While a 6%
variation for large, homogeneously vegetated areas may appear
minimal, it is uncertain from this investigation the extent of the
remaining variation for other vegetated mountainous areas,

including the forest-tundra ecotone.

58

5.3 W

Additional, more encompassing investigations are needed to
solidify these initial findings. The common practice of accepting the
topographic compensation characteristics of NDVI needs further
questioning for applications in alpine and sub-alpine terrain. The
future research may include one or several of the following areas:
larger site investigation, examination over different areas, different
combinations of incidence angles and vegetation/density, and
additional directional statistical tests. Such research needs to
overcome some of the limitations present within this work. One
area of immediate concern is an investigation into the difference
between radiance and DN generated NDVI. Radiance values have
typically been utilized generate the NDVI from AVHRR data, and are
the true basis for correct NDVI calculation. Why are DN values used
for some Landsat TM applications of NDVI? The examination of the
differences is in order to determine if the practice is justifiable and
under what conditions. One additional area of immediate interest is
the examination of a suitable extent of each of the vegetation
classes, used within this study, for possible topographic influence
within NDVI. With such an investigation completed, future users of
NDVI may have a better understanding of the topographic noise

contained within the index data.

APPENDICES

APPENDIX A

59

APPENDIXA

Characteristics of Landsat TM (Lillesand and Kiefer, 1994)

Band Q.) Spectra—n; Prin i l ' i n

1. 0.45 — 0.52 Blue Designed for water body
penetration, making it useful for
soil/vegetation discrimination,
forest type mapping, and cultural
features identification.

2. 0.52 - 0.60 Green Designed to measure green

reflectance peak of vegetation for
vegetation discrimination and
vigor assessment. Also useful for
feature identification.

0.69 Red Designed to sense in a
chlorophyll absorption region
aiding in plant species
differentiation. Also useful for
cultural feature identification.

0.90 Near IR Useful for determining
vegetation types, vigor, and
biomass content, for delineating
water bodies, and for soil
moisture discrimination.

1.75 Mid-IR Indicative of vegetation moisture
content and soil moisture. Also
useful for differentiation of snow
from clouds.

12.5 Thermal-IR Useful in vegetation stress
analysis, soil moisture
discrimination, thermal mapping
applications.

2.35 Mid-IR Useful for discrimination of
mineraland rock types. Also
sensitive to vegetation moisture
content.

3 . 0.63

4. 0.76

5. 1.55

6. 10.4

7. 2.08

APPENDIX B

6 0
APPENDIX B

Meta data of the Landsat image

Scene ID =4254617165 WRS =034/032 ACQUISITION DATE =l9890705
SATELLITE =L4 INSTRUMENT =TM PRODUCT TYPE =MAP PRODUCT
TYPE OF GEODETIC PROCESSING =PASS THROUGH

RESAMPLING =CC

ORIENTATION = -3.01

PROJECTION =SOM

USGS PROJECTION # = 21 USGS MAP ZONE = 34

USGS PROJECTION PARAMETERS = 0.637838800000000D+07 0.635691200000000D+08
0.000000000000000D-1-00 0.980120000000000D+08 0.7604 60242 17 1249D+08
0.000000000000000D-1-00 0.000000000000000D+00 0.000000000000000D+00
0.988 841200000000D+02 0.5 20 1 61 300000000D+00 0.000000000000000D+00
0.000000000000000D-1-00 0.000000000000000D+00 0.000000000000000D+00
0.000000000000000D-t-00

EARTH ELLIPSOID =INTERNAL_1909

SEMI-MAJOR AXIS =6378388.000

SEMI-MINOR AXIS =6356912.000

PIXEL SIZE =28.50 PIXELS PER LINE =3510 LINES PER IMAGE =3510

UL 1061017.4284W 405234.3400N 533919.000 15534837000

UR 1050029.0005W 404221.0242N 633787.249 15540093.701

LR 1051411.5080W 394920.7874N 628530.547 15639961950

LL 1062307.5657W 395326.1515N 528662.299 15634705249

BANDS PRESENT 1234567

****** FACTOR = 1

RECORD LENGTH =3510

SUN ELEVATION =60

SUN AZIMUTH =118

SCENE CENTER = 570921.416 15590642306

LITERATURE CITED

61

LITERATURE CITED

Allen, R. B., R. K. Feet, and W. L. Baker. (1991) Gradient Analysis
of latitudinal variation in Southern Rocky Mountain
forests. Journal of Biogeography. Vol. 18, 123-139.

Arno, S. F., and R. P. Hammerly. (1984) Timberline Mountain
and Arctic Forest Frontiers. Seattle, Washington: The
Mountaineers., 304p.

Ashburn, P., (1978) The vegetative Index Number and crop

identification. Proceedings: The LACIE Symposium. 843-
856.

Baker, W. L., J. J. Honaker, and P. J. Weisberg. (1995) Using Aerial
Photography and GIS to Map the Forest-Tundra
Ecotone in Rocky Mountain National Park, Colorado,
for Global Change Research. Photogrammetric
Engineering and Remote Sensing. Vol. 61, No. 3, 313-320.

Baker, W. L., and RI. Weisberg. (1995) Landscape Analysis of
the Forest-Tundra Ecotone in Rocky Mountain
National Park, Colorado. Professional Geographer. Vol. 47,

No. 4. 361-375.

Barbour, M. G., and W. D. Billings. (1988) North American
Terrestrial Vegetation. New York: Cambridge University
Press., 434p.

Baret, F. and G. Guyot. (1991) Potentials and Limits of
Vegetation Indices for LAI APAR assessment. Remote
Sensing of Environment. Vol. 35, 161-173.

Botkin, D. B., J. E. Estes, R. M. Macdonald, and M. V. Wilson. (1984)
Studying the Earth's Vegetation from Space. Bioscience.
Vol. 34, No. 8, 508-514.

62

Brown, D. G. (1994) Comparison of Vegetation-Topography
Relationships at the Alpine Treeline Ecotone. Physical
Geography. Vol. 15, No. 2, 125-145.

Brown, D. G. and T. J. Bara. (1994) Recognition and Reduction
of Systematic Error in Elevation and Derivative
Surfaces from 7 1/2 Minute DEMS. Photogrammetric
Engineering and Remote Sensing. Vol. 60, No. 2, 189-194.

Brown, J. F., T. R. Loveland, J. W. Merchent, B. C. Reed, and D. O.
Ohlen. (1993) Using Multisource Data in Global Land-
Cover Characterization: Concepts, Requirements, and

Methods. Photogrammetric Engineering and Remote Sensing.
Vol. 59, No. 6, 977-987.

Butera, M. K. (1986) A Correlation and Regression Analysis of
Percent Canopy Closure Versus TMS Spectral Response
for Selected Forest Sites in San Juan National Forest,

Colorado. IEEE Transactions on Geoscience and Remote
Sensing. Vol. GE-24, No. 1, 122-129.

Chiou, C. and R. Hoffer. (1994) A Sequential Sampling
Approach for Selecting Training Blocks of Landsat TM
Data in the Rocky Mountain National Park.
Proceedings: 1995 ACSM/ASPRS Convention and Exposition.
Vol. 3, 816-825.

Civco, D. L. (1989) Topographic Normalization of Landsat
Thematic Mapper Digital Imagery. Photogrammetric
Engineering and Remote Sensing. Vol. 55, No. 9, 1303-1309.

Colwell, J. E. (1973) Bidirectional spectral reflectance of grass
canopies for determination of above ground standing
biomass. Ph.D. thesis, University of Michigan, University
microfilm. 174 p.

Crist, E. P., and R. C. Cicone. (1984) A Physically-Based
Transormation of Thematic Mapper Data-The TM

Tasseled Cap. IEEE Transactions on Geoscience and Remote
Sensing. Vol GE-22, NO. 3, 256-263.

Curran, P. J. (1985) Principles of Remote Sensing. London:
Longman Group, 282p.

63

Curran, P. J. and N. W. Wardley. (1988) Radiometric leaf area
index. International Journal of Remote Sensing. Vol. 9, No. 2,
259-274.

Deering, D. W., J.W. Rouse, R. H. Haas, and J. A. Schell. (1975)
Measuring “forage production” of grazing units from
Landsat MSS data. Proceedings: 10th international
Symposium Remote Sensing of Environment. Vol. 2, 1169-

1178.

Dubayah, R., J. Dozier, and F.W. David. (1989) The Distribution of
Clear-Sky Radiation Over Varying Terrain. Proceedings,
IGARSS '89, 885-889.

Dubayah, R., and P. M. Rich. (1995) Topographic solar radiation
models for GIS. International Journal of Geographic
Information Systems. Vol. 9, No. 4, 405-419.

EOSAT. (1994) A Process of Radiometry. Eosat News, Vol. 9, No.
2, 10.

Erdas. (1991) Field Guide. Atlanta: ERDAS, Inc., 250p.

Frank. T. D. (1988) Mapping Dominant Vegetation
Communities in the Colorado Rocky Mountain Front
Range with Landsat Thematic Mapper and Digital
Terrain Data. Photogrammetric Engineering and Remote
Sensing, Vol. 54, No. 12, 1727-1734.

Gausman, H. W. (1977) Reflectance of leaf components. Remote
Sensing of Environment., Vol. 6,1-9.

Goward, S. N., B. Markham, D. G. Dye, W. Dulaney, and J. Yang. (1991)
Normalized Difference Vegetation Index
Measurements from the Advanced Very High

Resolution Radiometer. Remote Sensing of Environment.
Vol. 35. 257-277.

Goward, S. N., D. G. Dye, S. Turner, and J. Yang. (1993) Objective
assessment of the NOAA global vegetation index data

product. International Journal of Remote Sensing. Vol. 14,
No. 18, 3365-3394.

64

Guttman, G. G. (1991) Vegetation Indices from AVHRR: An
Update and Future Prospects. Remote Sensing of
Environment. Vol. 35, 121-136.

Hansen-Bristow, K. J., and J. D. Ives. (1984) Changes in the
forest-Alpine Tundra Ecotone: Colorado Front Range.
Physical Geography. Vol. 5., No. 2, 186-197.

Holben, B. N., and C. 0. Justice. (1980) The Topographic Effect on
Spectral Response From Nadir-Pointing Sensors.

Photogrammetric Engineering and Remote Sensing. Vol. 46,
No. 9, 1191-1200.

Holben, B. N., and C. 0. Justice. (1981) An examination of
spectral band ratioing to reduce the topographic
effect on remotely sensed data. International Journal of

Remote Sensing. Vol. 2, No. 2, 115-133.

Huete, A. R. (1994) An Error and Sensitivity Analysis of the
Atmospheric- and Soil-Correcting Variants of the
NDVI for the MODIS-EOS. IEEE Transactions on Geoscience
and Remote Sensing. Vol. 32, No. 4, 897-904.

Jordon, C. F. (1969) Derivation of leaf area index from quality
of light on forest floor. Ecology. Vol. 50, 663-666.

Kalensky, Z., and D. A. Wilson. (1975) Spectral Signatures of
Forest Trees. Proceedings, Third Canadian Symposium on
Remote Sensing. 155-171.

Karaska, M. A., S. J. Walsh, and D. R. Butler. (1986) Impact of
environmental variables on spectral signatures
acquired by the LANDSAT thematic mapper.
International Journal of Remote Sensing, Vol. 7, No. 12, 1653-
1667.

Kanth, R. J., and G. S. Thomas. (1976) The Tassel Cap - A
graphic description of the spectral-temporal
development of agricultural crops as seen by Landsat.
Proceedings: Symposium on Machine Processing of Remotely
Sensed Data. IEEE catalog No. 76, Ch. 1103-1. MPRSD, LARS,
Purdue University, West Lafayette, Indiana.

65

Kawata, Y., S. Ueno. (1988) Radiometric correction for
atmospheric and topographic effects on Landsat MSS
images. International Journal of Remote Sensing. Vol. 9, No.
4, 729-748.

Leprieur, C. E., J. M. Durand, and J. L. Peyron. (1988) Influence of
Topography on Forest Reflectance Using Landsat
Thematic Mapper and Digital Terrain Data.

Photogrammetric Engineering and Remote Sensing. Vol. 54,
No. 4, 491-496.

Lillesand, T. M. and R. W. Kiefer. (1994) Remote Sensing and
Image Interpretation. New York: John Wiley & Sons, Inc.,
750p.

Meyer, P., K. I. Itten, T. Kellinger, S. Sandmeier, and R. Sandmeier.
(1993) Radiometric corrections of topographically
induced effects on Landsat TM data in an alpine

environment. ISPRS Journal of Photogrammetry and
Remote Sensing. Vol. 48, No. 4, 17-28.

Nelson, R. A. (1953) Plants of Rocky Mountain National Park.
Washington, DC: US. Government Printing Office, 201p.

Overpeck, J.T., D. Rind, and R. Goldberg. (1990) Climate-induced
changes in forest disturbance and vegetation. Nature.
Vol. 343, No. 4, 51-53.

Peet, R. K. (1978a) Forest Vegetation of the Colorado Front
Range: Patterns of Species Diversity. Vegetatio. Vol. 37,
No. 2., 65-78.

Peet, R. K. (1978b) Latitudinal variation in southern Rocky
Mountain forests. Journal of Biogeography. Vol. 5, 275-
289.

Peet, R. K. (1981) Forest vegetation of the Colorado Front
Range. Vegetatio. Vol. 45, No. 1, 3-75.

Perry, C. R. Jr., and L. F. Lautenschlager. (1984) Functional

Equivalence of Spectral Vegetation Indices. Remote
Sensing of Environment. Vol. 14, 169-182.

66

Price, J. C. (1987) Calibration of Satellite Radiometers and
the Comparison of Vegetation Indices. Remote Sensing
of Environment. Vol. 21, 15-27.

Price, J. C. (1993) Estimating Leaf Area Index from Satellite
Data. IEEE Transactions on Geoscience and Remote Sensing.
Vol. 31, No. 3, 727-734.

Price, L. W. (1981) Mountains and Man. Berkeley: University of
California Press., 506p.

Proy, C., D. Tanre, and P.Y. Deschamps. (1989) Evaluation of

Topographic Effects in Remotely Sensed Data. Remote
Sensing of the Environment. Vol. 30, 21-32.

Richardson, A. J., and C. L. Weigand. (1977) Distinguishing
vegetation from soil background information.

Photogrammetric Engineering and Remote Sensing. Vol. 43,
1541-1552.

Richmond, G. M. (1960) Glaciation of the east slope of Rocky
Mountain National Park, Colorado. Bulletin of the
Geological Society of America. Vol. 71, 1371-1382.

Rouse, J. W., R. H. Haas, J. A. Schell, and D. W. Deering. (1973)
Monitoring vegetation systems in the great plains
with ERTS. Third ERTS Symposium, NASA SP-351. Vol. 1,
309- 317.

Rouse, J. W., R. H. Haas, J. A. Schell, D. W. Deering, and J. C. Harlan.
(1974) Monitoring vernal advancement and
retrogradation (greenwave effect) of natural
vegetation. NASA/GSFC Type 111 Final Report. Greenbelt,
Maryland: Goddard Space Flight Center, 371 p.

Sellers, P. J. (1985) Canopy reflectance, photosynthesis and
transpiration. International Journal of Remote Sensing. Vol.
6, No. 8, 1335-1372.

Singh, S. M. (1988) Simulation of solar zenith angle effect on
global vegetation index (GVI). International Journal of
Remote Sensing. Vol. 9, No. 2, 237-248.

67
Tarpley, J. D., S. R. Schneider, and R. L. Money. (1984) Global
Vegetation Indices from the NOAA-7 Meteorological
Satellite. Journal of Climateand Applied Meteorology. Vol.
23, 491-494.

Teillet, P. M. (1982) On the Slope-Aspect Correction of
Multispectral Scanner Data. Canadian Journal of Remote
Sensing. Vol. 8, No. 2, 84-106.

Teillet, P. M. (1986) Image correction for radiometric effects
in remote sensing. International Journal of Remote

Sensing. Vol. 7, No. 12, 1637-1651.

Thomas, R. (n.d.) UTM Registration of Landsat TM Data of
Rocky Mountain National Park. Colorado State University.
unpublished paper. Estes Park, CO. 5p.

Tucker, C. J. (1979) Red and Photographic Infrared Linear
Combinations for Monitoring Vegetation. Remote
Sensing of Environment. Vol. 8, 127-150.

U.S. Geological Survey. (1987) Digital Elevation Models. Data
User's Guide 5. Reston, Virginia: U.S. Geological Survey, 180p.

Veblen, T. T., and D. C. Lorenz. (1991) The Colorado Front Range.
Salt Lake City, Utah: University of Utah Press. 115p.

Wardley, N. (1984) Vegetation index variability as a function
of viewing geometry. International Journal of Remote
Sensing. Vol. 5, No. 5, 861-870.

Weisberg, P. J., and W. L. Baker. (1995) Spatial variation in Tree
Seedling and Krummholz Growth in Forest-Tundra
Ecotone of Rocky Mountain National Park, Colorado,

USA Arctic and Alpine Research. Vol. 27, No. 2, 116-129.

Woodham, R. J. and T. K. Lee. (1985) Photogrammetric method
for radiometric correction of multispectral scanner

data. Canadian Journal of Remote Sensing. Vol. 11, No. 2,
132-161.

"Il'llllllllllllllli "