International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
3.4 Vegetation Indices
The vegetation monitoring and the observation of vegetation dis-
tribution in large areas can be done with a vegetation indices (VI).
They have been used for assessment of various plant biophysical
parameters, such as leaf area index (LAI), percent green cover,
green biomass, fractional absorbed photosynthetically active ra-
diation (fAPAR), net primary production (NPP), and so on. Ba-
sically, they have capability of monitoring for seasonal change,
phenological change, state, and condition of vegetation. The veg-
etation biophysical parameters should be provided with sufficient
accuracy to be used as an input of general circulation models,
where the satellite remote sensing also plays an important role.
Therefore, VI should have continuity for long-term assessment
and should be more accurate. Various VI products currently ex-
ist and more VI products are expected to be available simultane-
ously, the estimation of biophysical parameters through VI has to
be compatible by various sensors to ensure continuity of global
environmental simulation over generations. Normalised Differ-
ence Vegetation Index (NDVI) is most used for land from be-
fore. Therefore, NDVI should be land product for global land
monitoring. The NDVI has been the most widely used index in
global vegetation studies, including phenological studies of veg-
etation growing season and so on(e.g. Tucker and Seller, 1986).
Many studies have shown a qualitative and quantitative analysis
of NDVI in vegetation growth. The NDVI is described as the
following equations;
PNIR — PRED
PNIR + PRED
NDVI =
where pNrr is reflectance in near infrared region, and prep is
reflectance in visible red region (Pobs Or ps). This index can re-
duce noise and uncertainty associated with instrument character-
istics, topographical effect, and so on. On the other hand, this
index also has disadvantages. This index saturates at the high
biomass area, and has sensitivity to canopy backgrounds over
open canopy conditions (Huete et al, 1997). GLI land team
adopt more another new index, a new 'Enhanced Vegetation In-
dex' (EVI) for increased sensitivity over a wider range of veg-
etation conditions, removal of soil background influences, and
removal of residual atmospheric contamination effects present in
the NDVI. The soil background adjustment is based on the soil
adjusted vegetation index (SAVI) (Huete 1988). An atmospheric
resistant term is derived from the atmospherically resistant vege-
tation index (ARVI) (Kaufman and Tanré,1992). The EVI equa-
tion shows as the following;
PNIR — PRED
L+ pnir + C1 x pren — C2 X PBLUE
EVI=GX
where L is the canopy background and snow correction caused by
differential NIR and red radiant transfer (transmittance) through
a canopy; and C1 and C2 are the coefficients of the aerosol ‘resis-
tance’ term, which uses the blue channel (PBLUE;Pobs OT Ps) tO
correct for aerosol effects in the red channel. Huete et al.(1997)
shows that the currently used coefficients, G = 2.5; L= 1; Cl =6;
and C2 = 7.5, are fairly robust. Especially, aerosol variations are
considerably reduced via the self-correcting combination of the
red and blue channels as less prone to instrument noise compared
with the AVHRR. GLI 1km bands are much narrower than the
GLI 250m bands, and provide increased chlorophyll sensitivity
(band 13) and avoids water vapor absorption (band 19). The blue
band provides aerosol resistance in the EVI.
814
4 DISCUSSION AND CONCLUSION
Requirement of geometric correction algorithm for land products
is less than 1 pixel. The accuracy of precise geometric correc-
tion for both of GLI lkm and 250m is less than 1 pixel. Band-
to-band registration errors are less than 0.5 pixel. GPSR has
sometimes stopped in ADEOS-II operational period. However,
the error is around 70-80m, which is enough small for 1 pixel
size. Therefore, geometric accuracy is almost satisfied with the
requirements. GLI geometric calibration team will continue to
check error pattern. Especially, GLI 250m data and tilt data
should be evaluated. The saturation level for land is almost sat-
isfied with maximum radiance of specification. In case of GLI
Ikm land channel, Ch. 5,8,13,15,19 may be saturated in high
bright cloud and ice cloud. In land area, the saturation is not
confirmed. In case of GLI 250m channel, Ch.22 is sometimes
saturated on the part of desert bright area as estimated by the
pre-launch analysis. And Ch.23 is rarely saturated like 1km land
channels without the extreme bright cloud. The DN of VNIR2
(land channel Ch.13,19,22,23) turns down partly in the range ex-
ceeding the saturation level (over saturation). They will not be
critical problems in unsaturated areas. Detector sensitivity nor-
malization error, mirror reflectance normalization error, and elec-
tric system noise of MTIR are sometimes appeared. The LIB DN
of Ch.30 (3.7 4 m) frequently becomes zero in low-temperature
(«240K) areas, however this will be occurred at the top of high-
altitude clouds and in polar regions in the nighttime. Left im-
age of Figure.2 shows the example of precise geometric corrected
GLI 1km L1B image around Black Sea, and right image shows
250m. Both of these images are captured on June 3, 2003. GLI
Ikm/250m data is able to monitor global area and local area at
same time.
Figure.3 shows the example of GLI 1km and 250m NDVI. Right
image is atmospherically corrected 16-day composite NDVI by
using the above data. Unfortunately, in this paper, 250m atmo-
spherically corrected data is not shown in this paper, because
algorithm is under development. However, it is not so difficult
to complete this code, because it simply needs to change LUTs
for rayleigh scattering and ozone absorption. Scatterplot of rela-
tionship between satellite zenith angle and atmospherically cor-
rected Ch.5/Ch.8/Ch.19 reflectance using proposed algorithm in
this paper is shown in Figure.4, because NDVI and EVI use these
channels. Ch.5 (blue channel) reflectance is largely corrected,
and Ch.8 (red channel) is slightly corrected. On the other hand,
Ch.19 (NIR channel) is almost same as atmospherically uncor-
rected reflectance. Therefore, this is expected and appropriate
result. However, especially, Ch.5 reflectance indicates the depen-
dance on satellite angle. In blue channel, it is able to consider that
there is not so much BRDF effect on the ground surface. As men-
tion the above, GLI 1km/250m atmospheric correction algorithm
can remove only the rayleigh scattering and ozone absorption.
The algorithm proposed in this paper cannot remove the aerosol
over land, therefore, the aerosol effect might be one of the rea-
son for this characteristics. GLI lkm global 16-day composite
NDVI/EVI image, which is conducted atmospheric correction,
is shown in Figure.5. There are some pixels affected by satel-
lite zenith angle. Especially, these are distributed around central
Africa, south America, and so on.
In future, GLI land higher level algorithm should include cor-
rection for aerosol over land. And then, GLI 250m land higher
algorithm is also needed to complete to develop as soon as posi-
ible. As described in the above, 250m channel is expected to
be affected by water vapour, because these channels are broader
band than Ikm channels. 250m algorithm may have to be dif-
ferent from 1km algorithm. Moreover, those products should be
Inter.
evalu
Cihl:
tiona
iron..
Thuil
Gillo
tral i
SPEC
Solar
Huet:
"Ac
ages
Huet
Renx
Kauf
veget
Remc
Hash
Terra
and R
Mura
GLI
Grou
(http:
