Jong de

Jong de
ITC Journal 1998-1
Imaging spectrometry for monitoring tree damage caused
by volcanic activity in the Long Valley caldera, California
Steven M de Jong1
ABSTRACT
canic activity, which causes significant damage to the
pine and fir species. Multitemporal images acquired by
AVIRIS were used to survey damage to pine and fir
trees, and to map the spatial extent of diffuse volcanic
gas emissions. AVIRIS acquires images at an altitude of
20 km in the spectral range of 400 to 2500 nm, with a
pixel size of 20 x 20 m. It has 224 spectral bands with
a nominal bandwidth of 10 nm (Figure 1).
Developments in detector technology have triggered a new remote sensing technology: imaging spectrometry. Imaging spectrometers measure
reflected solar radiance on a pixel-by-pixel basis in many narrow spectral
bands, allowing the identification of materials or their properties by diagnostic absorption features. To date, only airborne imaging spectrometers
are available, but several imaging spectrometers are planned for the next
generation of space platforms. The abundance of information available
in the continuous spectral coverage makes it possible to address questions in numerous environmental disciplines. This paper describes a
study in the Sierra Nevada, using multitemporal images acquired by the
Airborne Visible/InfraRed Imaging Spectrometer (AVIRIS) for monitoring tree damage by volcanic activity. Diffuse volcanic gas emanations
deprive the roots of oxygen, resulting in trees that are under stress and
ultimately die. Imaging spectrometry yields important information on
tree conditions and on the presence of dead vegetative material. The
spatial and temporal extent of the dead and stressed tree areas were
mapped using AVIRIS data. The use of imaging spectrometry to map
the diffuse volcanic gas emissions was less successful. Although the
images yield noisy spatial patterns of carbon dioxide, it is difficult to
separate atmospheric gases from the diffuse soil emanations.
STUDY AREA
The research area is situated around Mammoth
Mountain. Mammoth Mountain is a volcanic cone rising
up to 3300 m; it forms the western rim of the Long
Valley caldera in the Sierra Nevada, California (Figure
2). The Long Valley caldera measures approximately 17
x 32 km, and was formed by a large eruption about
760,000 years ago [34]. After a period of rest (the last
signs of activity from Mammoth Mountain occurred
roughly 500 years ago), the area has since 1980 been
suffering from frequent earthquakes, hydrothermal activity and gas emissions [22, 26, 32]. Furthermore, the
resurgent dome in the center of the Long Valley caldera
is inflating; the U.S. Geological Survey has measured an
uplift of approximately 60 cm since 1980.
In 1990, areas of dying forests were found on the
flanks of Mammoth Mountain [22]. At first, the cause
of tree die-off was sought in the persisting drought of
the preceding years. However, trees died regardless of
age or species, as shown in Figure 3. Research [22]
revealed that high concentrations of carbon dioxide (30
In the last decennia, a new remote sensing technique was
developed through significant advances in detector technology: imaging spectrometry. An imaging spectrometer
collects narrow spectral bands on a pixel-by-pixel basis,
aiming to identify surface materials by using diagnostic
absorption features [12, 23, 37]. Figure 1 shows the
concept of imaging spectrometry. Conventional broadband sensors such as Spot-XS, Landsat MSS and
Landsat TM are not very suitable for mapping minerals
or soil properties because their bandwidth of 70 to 240
nm cannot resolve diagnostic spectral features of terrestrial materials. Often, absorption features of interest
have bandwidths of only 20 nm or less. Since the construction of the first spectrometer, the technique and the
sensors have been further developed and refined, and
software especially designed to analyze the large data
volumes generated by imaging spectrometers have
become available [31, 39]. These developments have
led to the successful applications of imaging spectrometry in several environmental disciplines, such as atmospheric science [6], ecology [36, 38, 44, 46, 47], geology
[29, 30, 31,37, 45], soil science [11, 15, 16], hydrology
and oceanography [5, 25, 35]. The importance of these
types of instrument may be indicated by the fact that
several proposals for launching spaceborne spectrometers in the near future have been approved. This paper
presents a practical application of imaging spectrometry
for vegetation survey in the Long Valley caldera in the
Sierra Nevada, California. This area suffers from vol-
each pixel has an
associated, continuous
spectrum that can be
used to identify the
surface materials
224 spectral
bands
crosstrack
(614 pixels x
20 m/pixel)
20 m
Reflectance
along track
(512 pixels
per scene)
100
kaolinite
50
10
0.4
1
Department of Physical Geography, Utrecht University, PO Box 80
115, 3508 TC Utrecht, The Netherlands
1 1.5 2 2.5
Wavelength (µm)
FIGURE 1 The concept of imaging spectrometry
1
Imaging spectrometry for monitoring tree damage
ITC Journal 1998-1
Mountain. The study area is covered by an open type of
montane forest and the dominant species are the lodgepole pine (Pinus contorta) and the red fir (Abies magnifica). The soil surface near Horseshoe Lake is generally fairly bright. The surface is covered by glacial
deposits (till) consisting mainly of weathered granitic
rocks. As a result of the volcanic gas emission around
Mammoth Mountain, there is an excellent sequence of
dead, stressed and healthy trees and bare surfaces. This
site provides excellent opportunities to study:
- the capabilities of imaging spectrometry for mapping
stressed (and dead) pine and fir species
- methods of separating the vigorous vegetation,
stressed vegetation and dead vegetation from the soil
background of glacial deposits and crystalline rocks
- the potential of AVIRIS to detect areas of volcanic
gas emissions.
N
Reno
Sacramento
Mammoth
San Francisco
Fresno
Las Vegas
Californië
Bakersfield
Santa Barbara
Los Angeles
FIELD AND IMAGE SPECTRA
In October 1995, spectral field measurements were
carried out in two of the four dead tree sites, using a
portable field spectrometer [3]. The instrument covers
the 350 to 2500 nm spectral region, using three individual spectrometers. It samples every 2 nm, with a variable resolution between 10 and 11 nm. Together with
each set of target measurements, a white reference plate
and the instrument's dark current were measured, and
multiple target measurements were averaged to increase
the signal-to-noise ratio. The dark current was subtracted from the raw spectral measurements and the white
reference measurements were used to convert the raw
data to reflectance. The spectral behaviour of several
surfaces was measured: healthy pine and fir species,
stressed and dead trees, the litter layer, dead trunks, bare
soil surfaces, rock outcrops and some reference sites for
radiometric image corrections. Trees were considered
"stressed" if several branches were dead or if at least
some of the needles had lost colour.
The information content of the field spectra was then
analyzed in respect to healthy, stressed and dead trees,
and the soil background. The analysis methods used
included standard normalization procedures, first derivative transform, convex hull transform and automatic
absorption feature finding [16, 24]. First derivative
analysis is a common method of enhancing abrupt
changes in the spectral curve of objects, eg, the red edge
San Diego
0
250km
FIGURE 2 Location of the Mammoth study area in the Sierra
Nevada, California
to 96 vol percent) in the soil profile are the most probable cause of tree death. Although not much is known
about the effect on roots of high levels of carbon dioxide in the soil profile, it is most likely that the trees are
dying because of oxygen deprivation: the CO2 drives the
oxygen out of the soil [20, 33]. The most probable gas
source is the magma body approximately 7 km beneath
the Long Valley caldera. The gas surfaces via the
fumaroles and as diffuse soil emanations. It is only
recently that research at Vulcano Island in southern Italy
has shown that volcanoes release abundant carbon dioxide from their flanks as diffuse soil emanations [4].
To date, four dead tree areas have been mapped out
around Mammoth Mountain. The two largest dying tree
sites are located near Horseshoe Lake and near
Mammoth Mountain Main Lodge, covering areas of
approximately 10 ha. Figure 4 shows an AVIRIS image
cube of Mammoth Mountain, acquired on 23 August
1994. Snow-covered Mammoth Mountain is visible in
the center of the image. Horseshoe Lake and the largest
dying tree areas are located just south of Mammoth
A
B
FIGURE 3 The southern flank of Mammoth Mountain at Horseshoe Lake; (A) the forest in excellent health; (B) trees dying of
carbon dioxide asphyxia [22]
2
Imaging spectrometry for monitoring tree damage
ITC Journal 1998-1
Normalized Reflectance (%)
Reflectance (%)
for vegetation [13, 27, 46]. The
convex hull transform is a method
of normalizing spectra [16, 41].
The convex hull technique is analogous to fitting a rubber band over
a spectrum to form a continuum.
Figure 5 shows the concept of the
convex hull transform. The difference between the hull and the original spectrum is subtracted from a
constant to obtain a hull difference.
Such a normalization of the spectra
allows the application of quantitative absorption feature characterization in terms of feature depth,
surface area and asymmetry.
Figure 6 shows some examples
of the collected field spectra for
dead, stressed and healthy lodgepole pines. Figure 7 shows the
first derivative of the spectra in
Figure 6. The derivative computation tends to enhance not only the
absorption features but also the
noise [16]. Both figures clearly
show the presence/absence of
chlorophyll absorption near 680
nm in the healthy and dead lodgepole pine spectra, respectively.
Although the red edge [13, 16], the
steep spectral transition zone
between chlorophyll absorption at
FIGURE 4 AVIRIS image cube of Mammoth Mountain (acquired on 23 August 1994
and covering an area of approximately 12 x 12 km). X and Y axes show the geo680 nm and the high near-infrared
graphic position in the scene; the Z axis shows the spectral bands (224). Snowreflectance at 720 nm, is not very
covered Mammoth Mountain is visible in the center of the image; Horseshoe Lake
pronounced, it is visible in the origand the largest dying tree areas are just south of Mammoth Mountain
inal and derivative spectra. Figure
6 also illustrates the effect of
increasing brightness between 1400 and 1700 nm with
green canopy obscures the presence of woody material.
respect to the reduced water content of healthy pines as
Furthermore, a convex hull transform was computed
compared with dead pines. Within the same spectral
from the field spectra and the feature-finding algorithm
range (about 1720 nm), absorption features associated
[16, 24] was applied. The results are presented in Table
with lignin and cellulose can be seen for the dead pines
1; water is the most dominant absorption feature (1900
and litter spectra [36, 44, 48]. These features are not
and 1400 nm) identified by the algorithm. Compared
visible in the case of the healthy spectrum because the
with the healthy lodgepole pine, the stressed tree shows
90
85
80
75
70
100
99
98
97
96
95
94
65
93
60
92
55
91
50
90
45
89
A
B
88
40
87
0.60
35
0.60
0.80
1.00
1.20
1.40
1.60
1.80
2.00
2.20
1.00
1.20
1.40
1.60
1.80
2.00
2.20
2.40 2.50
Wavelength (µm)
Hull
Wavelength (µm)
Hull
0.80
2.40 2.50
Transformed Spectrum
Original Spectrum
FIGURE 5 Concept of the convex hull transform; (A) a hull fitted over the original spectrum; (B) the transformed spectrum. The
example shows a laboratory spectrum of a weathered limestone rock with absorption features for iron near 900 nm, for water
near 1400 and 1900 nm, for clay minerals near 2200 nm and for calcite near 2350 nm [16]
3
Imaging spectrometry for monitoring tree damage
ITC Journal 1998-1
TABLE 1 Results of the automatic absorption feature
search on field spectra after convex hull normalization
Wavelength
Depth
Area
Asymmetry
Healthy lodgepole pine (Pinus contorta)
1.8040
1.4540
0.6860
1.9860
1.9700
25.2592
22.2009
18.4455
13.6994
13.6013
3.6183
4.7954
2.2596
0.3587
0.1553
0.0314
0.4094
2.6633
0.1768
1.9560
Stressed lodgepole pine (Pinus contorta)
1.8040
1.4460
0.6880
1.9640
1.7940
FIGURE 6 Examples of field reflectance spectra (400 to 1800
nm) for healthy, dead and stressed lodgepole pines (Pinus
contorta). The spectra were collected near Horseshoe Lake
and each spectrum is an average of eight measurements
18.5023
17.5540
14.2796
12.0939
11.8013
2.6410
3.5220
1.9255
0.2501
0.1222
0.0441
0.3128
3.1891
0.3853
4.3238
Dead lodgepole pine (Pinus contorta)
1.8040
1.9640
1.9560
1.4540
2.1060
1.7880
25.1266
14.9782
14.1485
13.5969
11.1379
10.8838
3.5749
0.1669
0.0831
2.3908
0.2055
0.1313
0.0278
0.5096
0.5139
0.3469
1.5145
0.5303
31.5548
11.5030
11.3799
9.1239
8.0103
4.6171
0.0454
0.1287
0.0661
1.1761
0.0400
1.0037
0.2122
0.8646
0.3211
4.6679
1.0844
3.4357
2.9338
0.2219
0.0950
20.5118
0.0349
2.9247
1.3108
Litter
1.8040
1.9560
1.9600
1.7880
1.4460
Healthy red fir (Abies magnifica)
1.4460
1.4280
1.8040
0.6880
1.9640
First derivative spectra (500 to 850 nm) of field
measurements of dead, stressed and healthy lodgepole
pines (Pinus contorta)
FIGURE 7
25.3815
25.2224
24.0184
23.6689
16.5759
a decrease in chlorophyll absorption of approximately 15
percent. The convex hull transform is a promising tool
to quantitatively analyze field and image spectra.
Based on these encouraging results with the field
spectra, the first derivative was computed for a transect
in the AVIRIS image spectra located over the dead tree
area near Horseshoe Lake (Figure 3B). Figure 8 shows
the 460 to 950 nm range of these first derivative computations. The transect covers a quarry, an area with
stressed trees, an area with dead trees, a bare soil region,
another zone with stressed and dead trees and finally an
area of healthy trees. The variation in the peak height in
the center of the graph shows the differences in chlorophyll absorption. The results of the transect analysis
show that information on the varying shape of the
chlorophyll absorption features of vegetation is present
in the AVIRIS images. Similar results were reported by
Elvidge et al [21] and Roberts et al [38].
IMAGE ANALYSIS
Three AVIRIS images (acquisition dates: 21 May
1994, 23 August 1994 and 22 June 1995) of the
Mammoth Mountain region were analyzed for this study.
The AVIRIS radiance data were converted to reflectance
by using the empirical line method: a linear regression
FIGURE 8 3D plot of the first derivative spectra of a transect
in the AVIRIS image of 23 August 1994 near Horseshoe Lake
(Figure 4). The transect crosses a gravel pit and healthy,
stressed and dead vegetation (Figure 3B). The X, Y and Z
axes represent the distance along the transect in pixels of 20
m, the AVIRIS spectral range from 460 to 950 nm and the
first derivative of reflectance, respectively
4
Imaging spectrometry for monitoring tree damage
ITC Journal 1998-1
function was computed between the field reflectance and
corresponding AVIRIS pixels for a dark and a bright
surface (a large asphalt parking lot and a gravel field).
The field spectra for empirical line correction were not
collected on the same dates that the images were
acquired, thus hampering the multitemporal comparison
of the AVIRIS images. Furthermore, differences in
snow cover and the phenological state of the vegetation
played a role. A comparison of the reflectance curves of
a number of reference sites in the three multitemporal
images showed that the deviation of these multitemporal
spectra was smaller than 0.05, which was considered
acceptable for further analyses [18]. Multitemporal
analyses and spectral unmixing techniques were then
used to survey the temporal changes of the dead tree
areas and to separate stressed, dead and healthy pine
trees from the soil background.
aims at unravelling the complex satellite spectrum by
using the pure endmember spectra, and is a useful tool
for surveying. The mathematical notation of the unmixing procedure is as follows [2]:
n
R i = ∑ (F j × RE ij )+ ε i
n =1
[1]
where:
R i = reflectance value of a pixel in band i
F j = fraction of endmember j (in terms of percentage of the pixel surface)
RE ij = reflectance value for band i of endmember j
ei
= residual error in band i
n
= number of endmembers.
Spectral unmixing techniques have been described and
applied in various contexts by several authors [1, 2, 28,
38]. Two approaches were followed when using spectral
unmixing techniques in this study:
SPECTRAL UNMIXING
- field spectra were used as endmember spectra
Spectral unmixing is a technique to evaluate the large
- endmember spectra were collected from the airborne
volume of spectral data in a spatial context. The basic
images.
assumption of the unmixing technique is that the signal
Selected endmembers from the field spectra included
recorded by the sensor is for each pixel, a linear comdead, stressed and healthy lodgepole pines and bare soil.
bination of spectral signatures of pure components
Endmembers selected from the images comprised areas
weighted to their abundance in that pixel [1, 2, 38]. The
of dead, stressed and healthy pine species and bare soil.
components (or endmembers) are those elements that
Unconstrained unmixing (ie, abundances of endmembers
represent the spectral variability of the landscape (eg,
are not forced to sum to one and may assume negative
vegetation, soil and shade). The unmixing approach
values) was applied in this investigation. The unmixing
algorithm was applied to a spectral
subset from 400 to 1800 nm for all
AVIRIS images because the 1800
to 2500 nm range suffers from
noise, which is detrimental to the
algorithm.
Single endmember images were
colour-coded for visual interpretation and combinations of three endmember images were displayed in
red, green and blue to study the
quantitative contribution of each
endmember. Figure 9 shows the
promising unmixing results for the
dead tree endmember of the 23
August 1994 image (Figure 4).
The dead tree endmember abundance is shown colour-coded in
blue-green-red-yellow. The dead
tree areas near Horseshoe Lake
(south of Mammoth Mountain; see
Figure 3) and northwest of
Mammoth Mountain clearly show
up in yellow. The triangular yellow spot west of Mammoth
Mountain is a 1992 forest fire area
and a tree-logging site. The yellow
spot at the right edge of the figure
is brown/yellow pastureland. The
spectral responses of these land
cover types show some similarity
with the dead tree endmember and
are identified as such. The 23
FIGURE 9 Unmixing results for the dead tree endmember of the 23 August 1994
August 1994 image yields the best
AVIRIS image shown in Figure 4. The dead tree endmember abundance is colourresults because at this time of the
coded in blue-green-red-yellow. The dead tree areas near Horseshoe Lake (south
year the vegetation is most vigorof Mammoth Mountain; see Figure 3) and northwest of Mammoth Mountain clearly
show up in yellow
ous and, hence, the contrast
5
Imaging spectrometry for monitoring tree damage
ITC Journal 1998-1
between dead trees, stressed trees and healthy trees is at
its maximum. The somewhat poor results for the other
two images can most probably be attributed to the snow
cover and the dormant phenological condition of the
trees. Furthermore, the sun angle and, consequently, the
illumination conditions are not optimal for the last two
images. The results for the endmember images, using
field spectra as the reference for the unmixing algorithm,
are less satisfying, although the radiometric correction
using the empirical line method yielded fairly satisfactory results. Confusion occurs between the healthy and
stressed tree endmembers and between the dead vegetation and soil. The fractions for healthy vegetation, for
example, are often underestimated.
area, but field checks revealed that it was a 1992 forest
fire area. The multitemporal analyses of the AVIRIS
scenes proved valuable for surveying the dynamics of
the dead tree areas.
STRESS INDICATORS
The detection of early vegetation stress by optical
remote sensing depends largely on the capabilities of the
sensor to detect the subtle changes in the reflectance in
spectral regions that are most responsive to unfavourable
growth conditions. One of the first signs that vegetation
is exposed to stress (eg, water deficiency, senescence or
herbicides) is a decrease in the absorption of radiance by
chlorophyll in the visible wavelengths [7, 8, 9, 10, 27,
40, 50]. The decrease in absorption at the edges of the
chlorophyll absorption feature around 605 nm and 695
nm is particularly suitable as an early stress indicator.
The field spectra of Figures 6 and 7 illustrate this phenomenon. Two approaches, with the same theoretical
basis, are feasible:
(1) to determine the blue shift (or red edge), ie, the
spectral shift towards the blue by a decrease in chlorophyll activity [9, 14, 27, 50] and, hence, an indication of
vegetation under stress
(2) to use the ratios 5of spectral bands located at the
critical spectral position at the edges of chlorophyll
absorption [7, 8].
The second method was applied to the field and image
spectra to determine its suitability for mapping the
stressed pines near Mammoth. When reflectance ratios
are used as stress indicators, as suggested by Carter [7],
bands 694 and 760 nm or 695 and 420 nm are used.
Applying Carter's first stress indicator, R694/R760, to
MULTITEMPORAL IMAGE ANALYSIS
Multitemporal analysis refers to the comparison of
remote sensing images acquired at different times. The
three available AVIRIS images were used to survey the
temporal changes of the dead tree areas. A prerequisite
for successful multitemporal comparison is a proper
geometric match and radiometric correction of the
images. The images were geometrically matched to one
another using ground control points, a first degree polynomial warping algorithm and a nearest neighbourhood
resampling technique. Multitemporal analysis included:
- comparing the spectral unmixing results
- comparing NDVI (Normalized Difference Vegetation
Index) values for each AVIRIS scene.
Large NDVI values indicate pixels with high proportions of green biomass; low values indicate pixels of
bare soil, water bodies or built-up areas; and intermediate values give an indication of differences in coverage
for green vegetation [17, 43, 50].
The
three
multitemporal
unmixed images and the three
NDVI images were displayed as
red, green and blue combinations
for visual interpretation. Figure 10
shows the multitemporal NDVI values in red, green and blue, corresponding to image acquisition dates
of 21 May 1994, 23 August 1994
and 22 June 1995, respectively.
Results show that the Horseshoe
Lake dead tree area was fairly constant over the time span studied.
Only the western part of this region
shows some minor enlargement.
Interviews with the forest rangers
revealed that this site rapidly
expanded in the period late 1989 to
late 1992 but then came to a standstill. The second dead tree area,
further uphill from Horseshoe
Lake, shows a significant decrease
in NDVI values over time as a red
spot in Figure 10, indicating a significant enlargement of the dead
tree area since May 1994. Another
area of reduced green coverage was
located in the image, uphill on the
FIGURE 10 Multitemporal NDVI scene of the Mammoth Mountain area. Red, green
western
flank
of
Mammoth
and blue are assigned to AVIRIS images acquired on 21 May 1994, 23 August 1994
Mountain. It was first identified as
and 22 June 1995, respectively. The reddish border is caused by the incomplete
coverage of the three scenes. Red tints indicate a loss of vegetative cover over time
a fifth, so far unknown dead tree
6
Imaging spectrometry for monitoring tree damage
ITC Journal 1998-1
the spectra of Figure 6 yields values for healthy, stressed
and dead pines of 0.22, 0.25 and 0.45, respectively.
These values are in quite good agreement with results
reported by Carter [7]; his laboratory values range from
0.13 to 0.47. The same is true for the transect of Figure
8; values for the stressed/healthy trees range from 0.24
to 0.51 (bare surface areas were not considered). Table
2 shows some results obtained using the stress indicators. The performance of the second index (R694/R420)
is less satisfactory. Values are much larger than those
reported by Carter [7].
These stress indicators were then applied to the
AVIRIS 23 August 1994 scene. Table 3 shows the
results of Carter's index applied to this image for some
areas with known healthy, stressed and dead trees.
Although their absolute value is a bit high, the relative
values for the R694/R760 index are quite good; they
seem to represent the field situation. The performance
of the other index (R694/R420) is less satisfactory; all
values are too low. A possible explanation for these
somewhat poor results might be that the empirical line
correction does not sufficiently compensate for the
strong atmospheric effects in the short wavelengths
around 420 nm.
The next step in this study was to compute a stress
index for the entire 24 August 1994 scene and display it
as a colour-coded image for visual interpretation (see
Figure 11). As expected, an irrigated golf course close to
Mammoth village does not show
any signs of vegetation stress. The
results are also very promising for
the triangular forest fire area in the
left part of the image, and for the
dead tree area at Horseshoe Lake,
where the complete gradient from
healthy to severely stressed trees is
shown in conformity with field
observations. However, areas such
as lakes and the snow-covered
summit of Mammoth Mountain
show up as "stressed vegetation"
because the spectral signatures of
these objects show a decrease in
reflectance similar to that of
stressed vegetation.
Further
research is needed to refine the
stress indicator approach and to
develop methods to separate vegetation-covered areas from other
surfaces. A combination of spectral unmixing and stress indicators
might be useful in tackling this
problem.
MAPPING
VOLCANIC
GAS
EMISSIONS
The next step in this research
was to investigate whether AVIRIS
is capable of locating the areas of
volcanic gas emissions in the study
area. The volcanic gas emissions
near Mammoth Mountain contain
mainly water vapour, carbon diox-
ide and some minor gases such as methane and sulphur
dioxide [19, 22, 42]. Two approaches can be followed
to map the volcanic fluxes:
- to detect volcanic gas types not normally present in
the atmosphere
TABLE 2 Carter's stress index computed from the field spec-
tra
H LP
S LP
D LP
R694/R420
CV
n
R694/R760
CV
n
4.53
3.21
3.17
0.16
0.20
0.22
8
6
7
0.21
0.27
0.62
0.23
0.47
0.11
8
6
7
H LP = healthy lodgepole pine; S LP = stressed lodgepole pine; D LP =
dead lodgepole pine; CV = coefficient of variance; n = number of samples
TABLE 3 Carter's stress index computed for image spectra
(23 August 1994) dominated by lodgepole pines
H TR
S TR
D TR
R694/R420
CV
n
R694/R760
CV
n
0.81
1.04
1.37
0.23
0.10
0.04
55
51
51
0.33
0.63
0.87
0.23
0.08
0.03
55
51
51
H TR = healthy pine trees; S TR = stressed pine trees; D TR = dead
pine trees; CV = coefficient of variance; n = number of samples
FIGURE 11 Carter’s [7, 8] stress index computed from the AVIRIS scene acquired on
23 August 1994. Values range from high to low, from yellow to red, green, blue and
black. The irrigated golf course at the center right of the image is completely black,
as might be expected. The stressed tree area of Figure 3 and the forest-fire areas
are yellow, which also matches expectations. Water and snow-covered Mammoth
Mountain also show up in yellow because of their spectral signature
7
Imaging spectrometry for monitoring tree damage
ITC Journal 1998-1
- to image the abundance of gases normally present in
order to detect spatial anomalies.
The most promising approach to trace the volcanic
emissions is using carbon dioxide (CO2) and/or methane
(CH4), because both gases have absorption features in
the AVIRIS spectral range [37, 49]. Figure 12 shows a
field spectrum of the gas discharge of a fumarole near
Mammoth Mountain. Clear absorption features show up
for CO2; the concentration of CH4, however, is probably
small as no features show up for this gas (confirmed by
Sorey et al [42]).
The radiance images and reflectance images of the
three available AVIRIS images of the Mammoth
Mountain region were analyzed. The continuum interpolated band algorithm (CIBR) was used to enhance CO2
absorption at 2005 and 2055 nm. The CIBR algorithm
uses the "shoulders" at either side of the absorption feature to interpolate a radiance continuum over the absorption feature of interest. A ratio is then computed between
the absorption band and the interpolated continuum [6]:
B2
CIBR =
[2]
c1 B1 + c2 B2
ble with AVIRIS to locate the CO2 gas emission spots or
to separate atmospheric CO2 from the diffuse soil emanations of CO2.
When using AVIRIS, several problems may hamper
the mapping of CO2 emissions:
(1) the temporal and spatial distribution of gases in the
atmosphere
(2) the unknown spatial and temporal variation of the
diffuse CO2 soil flux at Mammoth (the flux depends,
among other things, on magma activity, barometric pressure and changes in soil moisture content)
(3) the low-energy solar irradiance at the 2000 to 2500
nm spectral region, resulting in a low signal-to-noise
ratio around the carbon dioxide absorption bands.
CONCLUSIONS
The use of remote sensing in environmental studies
has always been hampered by the difficulties in quantifying image-derived information. Reflectance properties of
broad-band sensors are not easy to correlate quantitatively with field or laboratory measurements. The development of imaging spectrometers is an important step forward in remote sensing because it allows the identification and quantitative assessment of several environmental
characteristics. The technique of imaging spectrometry is
currently finding its way from the specialized laboratories into the environmental science community to serve
as a powerful tool in studying earth science problems.
In this case study, AVIRIS images were used to monitor the deteriorating effect of diffuse volcanic gas emissions on the conditions of trees. Field spectral measurements revealed the presence of chlorophyll and lignincellulose associated absorption features in the spectra of
healthy, stressed and dead pine species. Spectral unmixing techniques proved useful in mapping areas of dead
and stressed trees. Multitemporal techniques were successfully applied to determine which dead tree areas
were still expanding and which were unchanged over the
period studied. However, the different image acquisition
dates presented problems, eg, snow cover and the dormant state of the vegetation in some images hampered
the spectral identification of tree conditions. The first
results with image-based vegetation stress indicators are
promising, although further research is necessary to
refine this approach.
Using AVIRIS to trace diffuse volcanic gas emissions
was less satisfactory. Field spectra and image spectra
contain distinct absorption features (eg, for carbon dioxide), but so far it has not been possible to find a distinct
correlation between areas showing vegetation stress and
the diffuse soil emanations, or to separate the atmospheric CO2 from the diffuse soil emanations, despite the
considerable volume of the soil flux. The solar energy
reflected around the 2003 absorption feature might be
too small, and the resulting AVIRIS spectrum too noisy
to reveal the subtle variations in CO2 concentration.
Another explanation might be that the temporal variability of the CO2 does not coincide with the three AVIRIS
data acquisition dates.
where:
B2
= radiance at the absorption feature position
B1 , B3 = radiance values at the shoulders
c1, c2 = symmetry coefficients.
Values for c1 and c2 depend on the symmetry of the
continuum at either side of the absorption feature.
Complete symmetry of the continuum results in c1 and
c2 values of 0.5.
The analysis results for the radiance and reflectance
images of all three data acquisition dates are poor. A
noisy salt-and-pepper pattern is visible, with a dim spatial pattern. It is striking is that all the lakes in the
images show high CIBR values. This is not caused by
high CO2 concentrations in the lakes or the fluxes forming the lakes, but is merely because the noise in the low
radiance spectrum for water is strongly enhanced. It was
expected that the dead tree areas would be clearly identified on the CIBR images since field surveys [22] yielded a strong correlation between dead/stressed trees and
gas emanations. Unfortunately, these areas do not show
up in any of the three radiance or reflectance images.
Although the volumes of the volcanic gas flux are considerable (estimated at 120 ton/day/ha by Farrar et al
[22]), the results show that in this case it was not possi-
ACKNOWLEDGMENTS
The research described in this paper was carried out at
the Jet Propulsion Laboratory, California Institute of
FIGURE 12 Field spectrum of the gas discharge of a fuma-
role near Mammoth Mountain
8
Imaging spectrometry for monitoring tree damage
ITC Journal 1998-1
Technology, Pasadena, California, and was financially
supported by the Netherlands Organization for Scientific
Research (NWO Talent Stipend S76-175). Special thanks
go to the staff of the AVIRIS Data Facility of the Jet
Propulsion Laboratory for their valuable help during the
processing of the data.
20 Drew, M C. 1991. Oxygen deficiency in the root environment and
plant mineral nutrition. In: M B Jackson (ed), Plant Life under
Oxygen Deprivation. SPB Academic Publ, The Hague, pp 303-326.
21 Elvidge, C D, Z Chen and D P Groeneveld. 1993. Detection of trace
quantities of green vegetation in 1990 AVIRIS data. Rem Sens of
Environ 44, pp 271-279.
22 Farrar, C D, M L Sorey, W C Evans, J F Howle, B D Kerr, B M
Kennedy, C Y King and J R Southon. 1995. Forest-killing diffuse
CO2 emission at Mammoth Mountain as a sign of magmatic unrest.
Nature 376, pp 675-678.
23 Goetz, A F H, G Vane, J E Solomon and B N Rock. 1985. Imaging
spectrometry for earth remote sensing. Science 228, pp 1147-1153.
24 Grove, C I, S J Hook and E D Paylor II. 1992. Laboratory
Reflectance Spectra of 160 Minerals, 0.4 to 2.5 Micrometers. JPLPubl 92-2, Pasadena, California, 406 pp.
25 Hamilton, M K, C O Davis, W Joseph Rea, S H Pilorz and K L
Carder. 1993. Estimating chlorophyll content and bathymetry of
Lake Tahoe using AVIRIS data. Rem Sens of Environ 44, pp 217230.
26 Hill, D P, R A Bailey and A S Ryall. 1985. Active tectonic and
magmatic processes beneath Long Valley caldera, eastern
California: an overview. J Geophysical Res 90-B13, pp 1111111120.
27 Hoque, E and P J S Hutzler. 1992. Spectral blue-shift of red edge
monitors damage class of beech trees. Rem Sens of Environ 39, pp
81-84.
28 Johnson, P E, M O Smith and J B Adams. 1992. Simple algorithms
for remote determination of mineral abundances and particle sizes
from reflectance spectra. J Geophysical Res 97-E2, pp 2649-2657.
29 Kruse, F A and P L Hauff. 1991. Identification of illite polytype
zoning in disseminated gold deposits using reflectance spectroscopy
and X-ray diffraction: potential for mapping with imaging spectrometers. IEEE Trans on Geoscience and Rem Sens 29, pp 101-104.
30 Kruse, F A, K S Kierein-Young and J W Boardman. 1990. Mineral
mapping at Cuprite, Nevada, with a 63-channel imaging spectrometer. PE&RS 56, pp 83-92.
31 Kruse, F A, A B Lefkoff, J W Boardman, K B Heidebrecht, A T
Shapiro, P J Barloon and A F H Goetz. 1993. The spectral image
processing system (SIPS), interactive visualization and analysis of
imaging spectrometer data. Rem Sens of Environ 44, pp 145-163.
32 Langbein J, D P Hill, T N Parker and S K Wilkinson. 1993. An
episode of reinflation of the Long Valley caldera, eastern California:
1989-1991. J Geophysical Res 98, pp 15851-15870.
33 Marschner, H. 1995. Mineral Nutrition of Higher Plants. Academic
Pr, London, 889 pp.
34 Miller, C G. 1985. Holocene eruptions at the Inyo volcanic chain,
California: implications for possible eruptions in Long Valley
caldera. Geology 13, pp 14-17.
35 Nolin, A W and J Dozier. 1993. Estimating snow grain size using
AVIRIS data. Rem Sens of Environ 44, pp 231-238.
36 Peterson, D L, J D Aber, P A Matson, D H Card, N Swanberg, C
Wessman and M Spanner. 1988. Remote sensing of forest canopy
and leaf biochemical contents. Rem Sens of Environ 24, pp 85-105.
37 Pieters, C M and P A J Englert. 1993. Remote Geochemical
Analysis: Elemental and Mineralogical Composition. Cambridge
Univ Pr, New York, 593 pp.
38 Roberts, D A, M O Smith and J B Adams. 1993. Green vegetation,
non-photosynthetic vegetation, and soils in AVIRIS data. Rem Sens
of Environ 44, pp 255-269.
39 Rubin, T D. 1993. Spectral mapping with imaging spectrometers.
PE&RS 59, pp 215-220.
40 Ruth, B, E Hoque, B Weisel and P J S Hutzler. 1991. Reflectance
and fluorescence parameters of needles of Norway spruce affected
by forest decline. Rem Sens of Environ 38, pp 35-44.
41 Sedgewick, R. 1983. Algorithms. Addison-Wesley, Reading
Massachusetts.
42 Sorey, M L, B M Kennedy, W C Evans, C D Farrar and G A
Suemnicht. 1993. Helium isotope and gas discharge variations associated with crustal unrest in Long Valley caldera, California, 19891992. J Geophysical Res 98, pp 15871-15899.
43 Tucker, C J. 1979. Red and photographic infrared linear combinations for monitoring vegetation. Rem Sens of Environ 8, pp 127150.
44 Ustin, S L, C A Wessman, B Curtiss, E Kasischke, J Way and V C
Vanderbilt. 1991. Opportunities for using the EOS imaging spectrometers and synthetic aperture radar in ecological models. Ecology
72, pp 1934-1945.
REFERENCES
1
Adams J B, D E Sabol, V Kapos, R A Filho, D A Roberts, M O
Smith and A R Gillespie. 1995. Classification of multispectral
images based on fractions of endmembers: application to land-cover
change in the Brazilian Amazon. Rem Sens of Environ 52, pp 137154.
2
Adams J B, M O Smith and A R Gillespie. 1993. Imaging spectroscopy: interpretation based on spectral mixture analysis. In: C M
Pieters and P A J Englert (eds), Remote Geochemical Analysis:
Elemental and Mineralogical Composition. Cambridge Univ Pr,
New York, pp 145-166.
3
ASD. 1994. FieldSpec FR User's Guide. Analytical Spectral Devices
Inc, Boulder, Colorado.
4
Baubron J C, P Allard and J P Toutain. 1990. Diffuse volcanic
emissions of carbon dioxide from Volcano island, Italy. Nature 344,
pp 51-53.
5
Carder K L, P Reinersman, R F Chen, F Muller-Karger, C O Davis
and M Hamilton. 1993. AVIRIS calibration and application in
coastal environments. Rem Sens of Environ 44, pp 205-216.
6
Carrère, V and J E Conel. 1993. Recovery of atmospheric water
vapor total column abundance from imaging spectrometer data
around 940 nm: sensitivity analysis and application to airborne visible/infrared imaging spectrometer (AVIRIS) data. Rem Sens of
Environ 44, pp 179-204.
7
Carter G A. 1994. Ratios of leaf reflectances in narrow wavebands
as indicators of plant stress. IJRS 15, pp 697-703.
8
Carter, G A and R L Miller. 1994. Early detection of plant stress by
digital imaging within narrow stress-sensitive wavebands. Rem Sens
of Environ 50, pp 295-302.
9
Chappelle E W, M S Kim and J E McMurtrey III. 1992. Ratio
analysis of reflectance spectra (RARS): an algorithm for the remote
estimation of the concentrations of chlorophyll A, chlorophyll B,
and carotenoids in soybean leaves. Rem Sens of Environ 39, pp
239-247.
Clark, R N, T V V King, C Ager and G A Swayze. 1995. Initial
vegetation species and senescence/stress indicator mapping in the
San Luis Valley, Colorado, using imaging spectrometer data. Proc
5th annual JPL airborne earth science workshop, Jan 23-26. JPL
Publ 95-1, Jet Propulsion Laboratory, Pasadena, CA, pp 35-38.
Crowley, J K. 1993. Mapping playa evaporite minerals with
AVIRIS data: a first report from Death Valley, California. Rem
Sens of Environ 44, pp 337-356.
Curran, P J. 1994. Imaging spectrometry. Progress in Physical
Geography 18, pp 247-266.
Curran, P J, J L Dungan, B A Macler, S E Plummer and D L
Peterson. 1992. Reflectance spectroscopy of fresh whole leaves for
the estimation of chemical concentration. Rem Sens of Environ 39,
pp 153-166.
Danson, F M and S E Plummer. 1995. Red-edge response to forest
leaf area index. IJRS 16, pp 183-188.
De Jong, S M. 1992. The analysis of spectroscopical data to map
soil types and soil crusts of Mediterranean eroded soils. Soil
Technology 5, pp 199-211.
De Jong, S M. 1994. Applications of Reflective Remote Sensing for
Land Degradation Studies in a Mediterranean Environment.
Netherlands Geographical Studies 177, KNAG Utrecht, 240 pp.
De Jong, S M. 1994. Derivation of vegetative variables from a
Landsat TM image for erosion modelling. Earth Surface Processes
and Landforms 19, pp 165-178.
De Jong, S M. 1996. Surveying dead trees and CO2-induced
stressed trees using AVIRIS in the Long Valley caldera. Proc 6th
airborne earth science workshop, March 4-8, Jet Propulsion
Laboratory, Pasadena, California, pp 53-63.
De Jong, S M and T G Chrien. 1996. Mapping volcanic gas emissions in the Mammoth Mountain area using AVIRIS. Proc 6th airborne earth science workshop, March 4-8, Jet Propulsion
Laboratory, Pasadena, California, 8 pp.
10
11
12
13
14
15
16
17
18
19
9
Imaging spectrometry for monitoring tree damage
ITC Journal 1998-1
Des émanations diffuses de gaz volcanique privent les racines de leur
oxygène, il s’ensuit que les arbres souffrent et finalement meurent. La
spectrométrie d’images fournit une information importante sur les conditions des arbres et sur la présence de matériaux végétaux morts. L’étendue spatiale et temporelle de la zone d’arbres morts ou en mauvaise
condition a été cartographiée à l’aide de données AVIRIS. L’utilisation
de la spectrométrie d’images pour la cartographie d’émissions diffuses
de gaz a eu moins de succès. Bien que les images apportent des échantillons spatiaux bruyants de dioxyde de carbone, il est difficile de séparer
les gaz atmosphériques des émanations diffuses du sol.
45 Van der Meer, F D. 1995. Imaging Spectrometry and the Ronda
Peridotites. PhD thesis, Wageningen Agricultural University, The
Netherlands, 309 pp.
46 Wessman, C A, J D Aber and D L Peterson. 1989. An evaluation of
imaging spectrometry for estimating forest canopy chemistry. IJRS
10, pp 1293-1316.
47 Wessman, C A, J D Aber, D L Peterson and J M Mellilo. 1988.
Remote sensing of canopy chemistry and nitrogen cycling in temperate forest ecosystems. Nature 335, pp 154-156.
48 Weyer L G. 1985. Near infrared spectroscopy of organic substances.
Applied Spectroscopy Reviews 21, pp 1-43.
49 Wolfe, W L and G J Zissis. 1989. The Infrared Handbook. ERIM:
Environmental Research Institute of Michigan, Ann Arbor,
Michigan.
50 Yoder, B J and R H Waring. 1994. The normalized difference vegetation index of small Douglas-fir canopies with varying chlorophyll
concentration. Rem Sens of Environ 49, pp 81-91.
RESUMEN
Los desarrollos en la tecnología de detectores han fomentado una nueva
tecnología de teledetección: la espectrometría de formación de imágenes.
Los espectrómetros miden la radiancia solar reflectada sobre una base de
pixel por pixel en un gran número de bandas espectrales estrechas, permitiendo la identificación de materiales o de sus propiedades gracias a
rasgos diagnósticos de absorción. Hoy en día, se dispone solamente de
espectrómetros aeroportados, pero varios espectrómetros están previstos
para la próxima generación de plataformas espaciales. La abundancia de
la información disponible a través de la cobertura espectral continua permite analizar problemas referentes a numerosas disciplinas ambientales.
Este artículo describe un estudio en la Sierra Nevada, en el cual se usaron imágenes multitemporales adquiridas por el Airborne
Visible/InfraRed Imaging Spectrometer (AVIRIS) para monitorear daños
causados a árboles por actividad volcánica. Las emanaciones difusas de
gas volcánico privan las raíces de oxígeno, lo cual pone los árboles bajo
tensión y últimamente causa su muerte. La espectrometría genera información importante sobre las condiciones de los árboles y sobre la presencia de material vegetativo muerto. Mediante el uso de datos de AVIRIS, se mapeó la extensión espacial y temporal de las áreas con árboles
estressados y muertos. El uso de la espectrometría para mapear las emisiones difusas de gas volcánico fue menos exitoso. Aun cuando las imágenes producen patrones espaciales confusos de dióxido de carbono,
resulta difícil separar los gases atmosféricos de las emanaciones difusas
saliendo de los suelos.
RESUME
Les développements dans la technologie des détecteurs ont déclenché
une nouvelle technologie de télé-détection: la spectrométrie d’image.
Les spectromètres d’images mesurent la radiance solaire reflétée sur la
base de pixel par pixel dans plusieurs bandes spectrales étroites, en permettant l’identification de matériels ou de leurs propriétés par le diagnostic de leurs caractéristiques d’absorption. Aujourd’hui, seuls les
spectromètres d’images aériens sont disponibles, mais plusieurs de ces
spectromètres d’images sont prévus pour la prochaine géneration de
plates-formes spatiales. L’abondance d’information disponible dans la
couverture continue spectrale permet de poser des questions dans de
nombreuses disciplines d’environnement. Cet article décrit une étude
dans la Sierra Nevada, avec l’utilisation d’images multi-temporelles
acquises par Spectromètre aérien d’image visible infra-rouge (AVIRIS)
pour l’observation de trois lieux sinistrés suite à l’activité volcanique.
10
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement