International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
radiometrically calibrated, and converted to surface reflectance
value. Initially, ETM+ images were geo-referenced to the local
plane rectangular coordinates by using a set of ground control
points obtained from the 1:5,000 scale topographic maps.
Although DN value represents a certain amount of radiometric
quantity that was reflected from the canopy, it also includes
partial signal originated by atmospheric attenuation. After raw
DN value was converted to the sensor-received radiance by
applying gain and offset coefficients, the radiance value was
transformed to percent reflectance after the atmospheric
correction. Although atmospheric correction has become a
critical step for deriving any quantitative variables of
biophysical parameters from optical remote sensing data, it is
rather complex and difficult to apply the absolute correction of
atmospheric effects on multispectral data such as ETM+. We
used MODTRAN radiative transfer code to calculate the
atmospheric transmittance and other terms using a standard
atmospheric model and local meteorological data for the
atmospheric correction.
After the geometric and radiometric correction of the spectral
imagery, a vector file of the 30 forest stands was overlaid to the
geo-rectified ETM+ imagery. Three or four pixels spanning the
boundary of each field-measured forest stand were extracted
and their reflectance values were averaged. Due to the high
spatial autocorrelation, the variation of adjacent pixels was very
low to overcome the problem of the sub-pixel error from the
geometric registration.
RESULTS AND DISCUSSIONS
Correlation coefficients between the simulated LAI and
spectral reflectance obtained from the spectro-radiometer were
highly variable by wavelength (Figure 3). In general, the only
positive correlations were in the near-IR regions. For all other
spectral regions, correlation coefficients were negative.
Correlation coefficients between LAI and spectral reflectance
in visible and short-wave infrared (SWIR) wavelengths were
much stronger than in near infrared regions.
ü8 r
aa s
-a2 | \
à | A
correlation coefficient (r)
L M / N rm N 1.
aL >
400 700 1000 1300 1600 1900 2200 2500
wavelength (nm )
Figure 3. The correlation coefficients between sample LAI and
spectral reflectance measured by the spectro-
radiometer.
In this experiment where the vegetation samples simulated the
completely closed canopy, the relationship between LAT and
402
spectral values seems to exhibit rather unique pattern. It is
interesting to note that the contrast between the NIR and the
other region. The near-IR wavelength region has been known
for the essential part of deriving several vegetative features,
such as LAI. However, under the close canopy situation, the
strength of correlation in the near-IR region was weaker than
the other wavelength region. These results suggest that close
canopy vegetation systems may be explained by additional
wavelength reflectance other than the near-IR wavelengths. In
particular, the SWIR region shows strong potential to be used
for estimating LAI in close canopy situation that is rather
common in many dense forests.
NDVI has been the most widely used spectral vegetation index
to estimate LAI over diverse biomes. As has been reported in
many previous studies, the correlation between the sample LAI
and NDVI appears relatively high (Figure 4). However, the
positive correlation looks apparent only when the sample LAI
value is relatively small. When the LAI is larger than three,
there are no significant correlation between the LAI and NDVI.
L +
0.65 + * +
+
+
. * *
+
2 e
=z
08 F +
*
0.55 d
0.5 1.5 25 35 4,5 55 6.5
LAI
Figure 4. Relationship between the LAI of vegetation sample
and NDVI that was derived from the spectral
reflectance measurements at 655nm and 846nm.
Forest LAI values measured over the study area were relatively
high in which the lowest LAI was 2.74 and the highest was
7.11. LAI of plantation conifer stands were higher than the
natural stands of mixed deciduous stands. The LAI variation
was very low at the mixed deciduous stands as compared to the
plantation pine stands. Field measured LAI was initially
analyzed by its correlation with calibrated reflectance values.
Correlation coefficients between the spectral reflectance and
the field measured LAI were very low for all plots when both
species groups were combined (Table 1). Forest LAI varies by
several factors of stand structural parameters, such as species,
stand density, canopy closure, DBH, and tree height.
Considering the diverse groups of species composition in the
study area, such low correlations would not be surprising.
When we calculated the correlation coefficient separately for
each of two species groups of coniferous and mixed deciduous
forest, the absolute value of correlation coefficients increased at
the coniferous forest. The plantation coniferous stands are
rather homogeneous in species composition. The variation of
LAI in these stands is mainly due to the tree size and stand
density. As seen in the previous laboratory experiment, only
near-IR band (band 4) shows the positive correlation although
correlation was very weak. There are negative correlations
Inter
betw
The
refle
mois
vege
Tabl
Spe
NO S
stands
decidi
meast
differ:
cause
Figure
LAT a
overal
very \
their L
in larg
genera
vegeta
1996,
decade
estima
that o
numer
SWIR
reflects
0.80
0.70
0.60
ETM* ND VI
0.50
0.40
Figure .
