317
Symposium on Remote Sensing for Resources Development and Environmental Management / Enschede / August 1986
Relation between spectral reflectance and vegetation index
S.M.Singh
NERC Unit for Thematic Information Systems, Department of Geography, University of Reading, UK
ABSTRACT: Atmospheric corrections are applied to the Advanced Very High Resolution Radiometer (AVHRR)
channel-1 and channel-2 data. Both raw and atmospherically corrected Normalized Difference Vegetation
Indices (NDVIs) are calculated. A comparison between them shows a contrast enhancement by a factor of at
least two when atmospheric corrections are applied. Spectral reflectances and atmospherically corrected
NDVI are partially correlated indicating a possibility of improving surface cover classification using NDVI
and spectral reflectances instead of NDVI values alone. Raw NDVI and atmospherically corrected NDVI do not
have a unique relationship but are highly correlated. This indicates that atmospheric corrections be applied
to each scene of interest.
1 INTRODUCTION
The visible channel (0.58-0.68 pm; hereafter
referred to as channel-1) and near infrared channel
(0.725-1.1 Pm; hereafter referred to as channel-2)
data from the Advanced Very High Resolution Radio
meter (AVHRR) instrument flown on the Tiros-N/NOAA
meteorological satellites have been found to be
useful for monitoring health and vigour of photo-
synthetically active vegetation canopy. These data
have been used for mapping and monitoring
vegetation cover on local and continental scale (for
example, see Tucker et al., 1983, 1984 and Hayes and
Cracknell, 1984) as well as on a global scale
(Justice et al., 1985). There is a daily coverage
at higher latitudes but around the equator complete
coverage requires three days. This means that the
global coverage data could be obtained in three days
if there were no cloud covers. There is an
absorption band of chlorophyll within channel-1
wavelength range whereas wavelengths within
channel-2 spectral band width are strongly reflected
by green pigments. In principle, the data from
these two spectral channels should be correlated to
the abundance of vegetation. Many relations exist
in the literature for calculating vegetation index,
for example, see Hayes (1985). Most popular of all
relations is the so-called Normalized Difference
Vegetation Index (NDVI) which is defined as
NDVI
DN2 - DN1
DN2 + DN1
(1)
where DN1 and DN2 are channel-1 and channel-2 pixel
values, respectively. There are several advantages
of using equation (1) rather than channel-2 data
only; because of optical properties of photo-
synthetically active chlorophyll as noted above,
equation (1) results in enhanced values of NDVI,
which could be useful particularly for low
vegetation; the relation (1) partially compensates
for atmospheric interference, solar elevation,
changing solar irradiance on the surface and topo
graphic effects (see, Justice et al., 1985).
Ideally, one would have liked to remove atmos
pheric effects from these data first and then
calculated vegetation indices because band ratioing
does not remove atmospheric effects completely
(Holben and Justice, 1981). The reason is that
atmospheric contaminations in channel-1 and channel-2
are not proportional to each other. The larger the
view angle of the sensor the larger the atmospheric
contribution is expected to be. Therefore, even if
there were no topographic effects and if surfaces
were Lambertian in nature, the NDVI values calculated
from equation (1) have strong view angle
dependence (Duggin et al., 1982), a significant
fraction of which is expected to be due to atmos
pheric effects. Within the framework of vegetation
mapping and monitoring, the same surface area is
viewed from various viewing angles (from different
orbits) and it is evident from the work of Duggin
et al. (1982) and many others that the NDVI values
do depend on the view angle. However, it has not
yet been possible to come up with a perfect atmos
pheric correction algorithm because of the diffi
culties in estimating atmospheric contamination due
to aerosols. Nevertheless, an approximate estimate
of atmospheric contribution to remotely sensed data
can be made.
One finds the same NDVI value for several surface
cover types (Townshend and Tucker, 1985) and,
therefore, vegetation type classification using NDVI
values alone has limited success. Ideally, one would
like to have several spectral band data which are
partially or poorly correlated among themselves so
that each spectral channel data carries information
about the nature of surface cover which supplements
information carried in other spectral channels.
Using Landsat Thematic Mapper (TM) data Toll (1985)
demonstrates that the land cover classification
accuracy does not improve by adding spectral
channel data which are highly correlated to other
channel(s) data which are already used for classi
fication. Also, when photosynthetically active
chlorophyll amount increases (say, in dense forests)
the NDVI values calculated from equation (1) tend to
saturate thereby limiting the range of applicability
of equation (1). Under such circumstances it would
be interesting to see how channel-1 reflectivity
changes and whether or not this reflectivity is
still a sensitive function of vegetation abundance.
It is in this spirit that a brief summary of atmos
pheric correction technique will be presented,
channel-1 and channel-2 reflectivities will be
calculated, raw and atmospherically corrected NDVI
will be calculated and relationship betweèn
reflectivities and NDVI values will be examined in
order to find some uncorrelated or partially
correlated parameters which may prove to be
valuable for land cover classification.
