Image Processing Reference
In-Depth Information
soil and parking lots. These land cover types might only be derived with insufficient
accuracy from spectral remote sensing mapping.
4.5
Most Suitable Wavelengths in Mapping Urban
Land Cover
One of the purposes of spectrometry is to identify the most important wavelengths
or spectral bands for mapping urban areas. This helps to evaluate limitations of
current multispectral systems or to design appropriate sensor systems, and is
required for specific image analysis algorithms. In land cover classification, for
example, having 224 bands from a sensor like AVIRIS
provides “too much” spectral information. Map accuracy
can actually decrease if too many spectral bands are
applied for such purposes (Landgrebe 2000 ). The derivation
of the most suitable bands also uses the B-distance spectral
separability measurements. The B-distance values identify
the wavelengths that contribute the most spectral contrast
and rank the band combinations that are optimal for sepa-
rating between the land cover classes. Related investiga-
tions of a ground spectral library and AVIRIS data resulted
in a set of most suitable bands that allow the greatest spectral separability of urban
land cover classes (Herold et al. 2003 ). In this case, the number of most suitable
bands was 14. This number was sufficient for this study, but does not imply that
exactly 14 bands should always be selected for hyperspectral urban mapping. In
this context, it is more important to look at the location of these bands. They appear
in nearly all parts of the spectrum with a fair number in the visible region (Fig. 4.5 ).
Narrow spectral bands are important in resolving small-scale spectral contrast in
the visible spectrum, e.g., from color, iron absorption features. Additional bands
appear in the near and short wave infrared. They represent the larger dynamic range
of reflectance values related to the increase in object brightness towards longer
wavelengths for several land cover types (e.g., tile roofs, wood shingle roofs, veg-
etation, soils, gravel surfaces). There also are specific absorption features that cor-
respond to some of the most suitable bands (Herold et al. 2003 ).
most suitable
spectral bands
can be selected
from hyperspec-
tral datasets
based on mea-
sures of spectral
separability
4.6
Effects of Spectral Resolution on Urban Land
Cover Mapping
The distribution of most suitable bands indicates that some of them are located
outside or near the edges of the Landsat TM spectral configuration. This indicates
possible limitations of this and similar sensor systems in mapping urban areas.
To evaluate the effects of spectral sensor resolution on urban mapping accuracies,
 
Search WWH ::




Custom Search