)04
the
Wn
res,
the
han
lose
mal
In
ised
ther
dex
] in
LAI
the
LAI
ree,
Hi.
nple
:tral
vely
Was
the
tion
) the
ially
lues.
and
both
s by
cies,
ight.
the
; for
IOUS
ed at
are
n of
tand
only
ugh
tions
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
between LAI and ETM+ reflectance in visible and SWIR bands.
The negative correlation between the LA]
and SWIR
reflectance may explained by several factors including leaf
moisture content, shadow effects among trees, and understory
vegetation (Nemani et al., 1993),
Table 1. Correlation coefficients between field measured LAI
and spectral reflectance of ASTER and ETM+ bands.
Spectral band all conifers deciduous
NDVI 0.228 0.508 -0.125
Band | -0.114 -0.320 0.056
Band 2 -0.097 -0.302 0.070
Band 3 -0.287 -0.560 0.055
Band 4 0.086 0.296 -0.179
Band 5 -0.233 -0.277 -0.286
Band 7 -0.270 -0.574 -0.075
No significant correlations were found at mixed deciduous
stands. Unlike the plantation coniferous stands, the mixed
deciduous stands showed very little variation in the field
measured LAI value (mean=4.33, std=0.78). The subtle
differences in the actual LAI values were thought to be the
cause of such relatively low correlation.
Figure 5 shows the relationship between the field-measured
LAI and NDVI that was derived from the two ETM+ bands. In
overall, the correlation between the forest LAI and NDVI is
very weak. The forest stands were almost close canopy and
their LAI values were larger than three. The lack of relationship
in large LAI value corresponds several previous studies. This
general low correlation between NDVI and LAI at high LAI
vegetation has been noted in several studies (Chen and Cihlar
1996, Turner et al. 1999, Cohen et al. 2003). For over two
decades, NDVI has been a popular index with which to
estimate LAI across diverse systems, but these results suggest
that other indices may be more appropriate. Fortunately,
numerous recent studies have noted a strong contribution of
SWIR bands to the strength of relationships between
reflectance and LAI (Nemani et al. 1993, Brown et al. 2000).
0.70
0.60
ETM+ ND VI
050 +
0.40 1 A
20 30 4.0 5.0 6.0 7.0 8.0
LAI (field measurements)
Figure 5. Relationship between the field-measured LAI and
NDVI that was derived from the two ETM+ bands.
403
The primary difference between the spectro-radiometer and
ETM+ spectral measurements is that the spectro-radiometer
data has more and narrow bands. Strength of correlation
coefficients were much stronger with the spectro-radiometer
data although the general pattern of correlations were similar.
The low correlation with the ETM+ reflectance is probably
related to the structure of forest stand where the tree species,
size, and density vary. Further, the LAI values at the study
sites is larger than three. On the other hands, the simulated
vegetation samples measured by the spectro-radiometer have
the LAI values ranging from one to six. The vegetation
samples were basically horizontal layering of flat leaves, which
did not quite reflect the three dimensional structure of forest
stand. Further experiment will be focused to measure
reflectance spectra on the vegetation samples that has vertical
structure of leaf distribution.
CONCLUSIONS
Forest LAI has been a key variable to understand the
productivity and process of forest ecosystem at various spatial
and time scales. However, there are not enough evidences that
spectral reflectance in visible and near-IR region will be
enough to estimate the forest LAL, in particular at the close
canopy situation. In this study, we have conducted a simple
correlation analysis between LAI and spectral reflectance at
two different settings. From the spectral reflectance data
measured by the spectro-radiometer over the multiple-layers of
leaf samples, the stronger correlations were found at the visible
and SWIR wavelength region. Although positive correlations
were noticed at the near-IR wavelength region, they were not as
strong as the other wavelength spectrum.
When we extended the comparison to the actual forest stands.
the correlations between the field measured LAI and ETM+
reflectance were very weak. Significant correlations were only
found at the plantation coniferous stands. The overall patterns
of correlation was somewhat similar to the result obtained from
the spectro-radiometer experiment. In both cases, the
correlations with the SWIR and red reflectance were higher
than the near-IR reflectance.
Although the SWIR spectrum has been known for its
relationship with the moisture content of vegetation, it has been
rare to verify the information content of SWIR to derive
biophysical characteristics of vegetation. Since the launch of
the Landsat-1 in 1972, only a few satellite sensors have
comprised spectral bands that have been operating at SWIR
spectrum. Landsat Thematic Mapper (TM, ETM+) is probably
the most well-known sensor that has SWIR spectral bands. In
recent years, there has been increasing number of new satellite
sensors (such as MODIS, ASTER, SPOT) that include SWIR
bands. We expect that there will be additional studies to
evaluate the potential of SWIR bands for extracting vegetative
information.
REFERENCES
Badhwar, G.D., R.B. MacDonald, and N.C. Mehta. 1986.
Satellite-derived leaf area index and vegetation maps as
input to global carbon cycle models — a hierarchical
approach. International Journal of Remote Sensing
7(2):265-281.
