Image Processing Reference
In-Depth Information
classification tools that are usually used for mapping more spectrally homogeneous
land covers like forests and agriculture.
14.3.1
Normalized Difference Vegetation Index
The most commonly used technique for vegetation analysis is the normalized dif-
ference vegetation index (NDVI), which is computed as:
rr
rr
−
NDVI
=
NIR
red
(14.1)
+
NIR
red
r
are reflectance values derived from
the near-infrared and red channels, respectively. The ratio is
a measure of the deviations between chlorophyll absorption
minimum and the infrared plateau, and thus, an indirect
proxy for the amount of photosynthetically active green
biomass (Tucker and Seller
1986
). NDVI values can be
computed for Landsat TM and MSS, SPOT and AVHRR
imagery. In spite of its popularity, the NDVI suffers from
several shortcomings. The index generally saturates for
areal vegetation fractions higher than 40-50% (Elmore et al.
2000
; Small
2001
) and is non-associative (Price
1990
) -
limiting its utility for characterizing scale dependent vegeta-
tion abundance. It is also sensitive to spectral band definitions
- complicating comparisons of indices measured by different
sensors (Price
1987
). Sensors, such as Landsat TM and
ETM+, Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER)
and Moderate Resolution Imaging Spectroradiometer (MODIS) provide sufficient spec-
tral resolution to be used for spectral mixture analysis as described in the next section.
NDVI and other vegetation indices usually map vegetation cover as a continuous
quantity (as opposed to a thematic class) but these indices are not well suited to mapping
vegetation within the urban mosaic because of the heterogeneity of urban vegetation
distribution. In addition to the difficulty in establishing a quantitative calibration between
areal vegetation abundance and NDVI at sub kilometer scales, it has been shown that
NDVI increases nonlinearly with increasing vegetation fraction (Small
2001
).
where
NI
r
and
red
vegetation indices
are not well suited
to mapping
vegetation within
the urban mosaic
because of the
heterogeneity of
urban vegetation
distribution and
mixed pixels of
coarse
multispectral
images
14.3.2
Spectral Mixture Analysis
Spectral mixture analysis (SMA) provides a physically based methodology to quan-
tify spectrally heterogeneous urban reflectance. SMA is based on the observation
that, in many situations, radiances from surfaces with different “endmember”
