The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
256
2. STUDY SITE AND HYPERSPECTRAL
DATA COLLECTION
2.1 Study Site
The City of Tampa was selected as study area. It is the largest
city on the west coast of Florida consisting of approximately
285 sq. km. The population is increasing and is currently
estimated at approximately 335,000 people (www.tampagov.net
accessed on Nov. 26, 2007). The city is located at
approximately 28° N and 82° W (Figure 1). Historically, the
natural plant communities of the Tampa Bay region included
pine flatwoods, cypress domes, hardwood hammocks, high pine
forests, freshwater marshes, and mangrove forests. Based on the
City of Tampa Urban Ecological Analysis (Campbell and
Landry, 1999), important, dominant urban tree species include
American Elm (Ulmus americana), Bluejack Oak (Q. incana),
Crape Myrtle (Lagerstroemia indica), Laurel Oak (Q.
laurifolia), Live Oak (Q. virginiana), Southern Magnolia
(.Magnolia grandiflora), Persimmon (Diospyros virginiana),
Red Maple (Acer rubrum), Sand Live Oak (Q. geminata),
American Sycamore (Platanus occidentalis), Turkey Oak (Q.
laevis), Slash Pine (Pinus elliottii), and Longleaf Pine (Pinus
palustris). Other dominant tree species within the City of
Tampa include Cabbage Palm (Sabal palmetto), Queen Palm
(Syagrus romanzoffiana), Citrus (Citrus spp.), Goldenrain Tree
(Koelreuteria paniculata), Bottlebrush (Callistemon viminale),
and Water Oak (Q. nigra), etc. In this analysis, a total of 11
broadleaf species (Table 1) were selected for testing the
capability of in situ hyperspectral data for discriminating
between species. Note that shorthand abbreviations for tree
species discussed throughout this paper can be found in Table 1
in parentheses after the common name of the tree species.
Figure l.A location map of the study area
2.2 In Situ Hyperspectral Data Collection
A full-range Analytical Spectral Device (ASD) (FieldSpec®3,
Analytical Spectral Devices, Inc., U.S.) was used to collect
spectral reflectance measurements from the 11 broadleaf tree
species in the city area, which are a subset of urban forest
species within the Tampa Bay area. The ASD instrument
consists of three separate spectrometers and covers a spectral
range of 350 nm to 2500 nm.
Tree species
Number of trees Train samples Test samples
Total
American Elm (Elm)
25
22
12
34
Bluejack Oak (Blue)
20
17
8
25
Crape Myrtle (Crap)
20
27
13
40
Laurel Oak (Laur)
31
24
13
37
Live Oak (Live)
26
24
11
35
Southern Magnolia (Magn)
20
30
16
46
Persimmon (Pers)
29
22
11
33
Red Maple (Mapl)
27
24
12
36
Sand Live Oak (Sand)
23
25
12
37
American Sycamore (Syca)
16
25
12
37
Turkey Oak (Turk)
28
22
12
34
Total
265
262
132
394
Table 1. Spectral measurements taken from 11 broadleaf
species.
In the field, at least 20 trees of each species (except Syca) were
measured to account for spectral variation and spatial
distribution. The spectral measurements were collected from
top, middle and low foliage branches from the crowns of
individual species. Tree heights lower than 7 m for most
sampled trees were generally selected because of the logistical
difficulties with measuring spectra from the top of tall trees. A
ladder with an effective height of 5 m was used for collecting
spectral measurements. To ensure that relatively pure spectra
from individual trees of difference species were collected, the
data acquisition was executed with a careful selection of view
area from tree foliage branches to avoid or lessen the effect of
background on target spectra. One to three spectra were
collected from individual trees. Because of the difficulty in
separating a shadow/shaded area from a sunlit area from a tree
crown, only fully sunlit areas were measured. Each spectral
measurement was repeated ten times to obtain reliable mean
and variance estimates. In this manner, a total of 394 spectral
measurements were collected from the different foliage
branches of the 11 broadleaf species in the City of Tampa
(Table 1).
3. ANALYSIS METHODS
3.1 Preprocessing of Spectral Measurements
The following preprocessing of spectral measurement was
performed. First, spectral curves were truncated below 400 nm
and above 2400 nm because the measurements were extremely
noisy outside of this range. Approximately 2000 bands remain,
each with a width of about 1 nm. Next, curve smoothing was
used with a simple average over blocks of five neighboring
bands. The spectral curves for constant area were then
normalized by dividing the mean reflectance for that curve.
That is, a spectral reflectance curve p, was replaced
PO
, where, k represents the total bands of the
spectral reflectance. The benefit of such normalization is the
suppression of illumination differences. Figure 2(a) shows a
plot of unnormalized curves versus band wavelength for two
observations of each of the five oak species (Blue, Laur, Live,
Sand and Turk). Figure 2(b) shows the same curves of Figure
2(a) after normalization. Notice that the clearer separation
between the species over a wide range of wavelength in Figure
2(b). Figure 2(c) shows a plot of normalized curves versus band
wavelength for all the 11 species (Elm through Turk).
3.2 Extraction of Spectral Variables
Forty-six spectral variables (Table 2), including normalized
spectra, derivative spectra, spectral vegetation indices, spectral