Image Processing Reference
In-Depth Information
grain size effects). Methods to extract, compare and analyze such spectral features
are described in Swayze et al. (
2003
). A successful application of hyperspectral
remote sensing for material mapping in urban areas is presented by Clark et al.
(
2001
). This research utilized high-resolution AVIRIS data with continuum-removed
spectral feature analysis to derive a detailed map of material/dust accumulations in
connection with the World Trade Center attack on September 11, 2001.
Land Cover versus Materials
Many urban remote sensing applications focus on mapping urban land cover
rather than materials. Although both are related, i.e., red tile roofs have
unique material characteristics, mapping land cover types requires a different
perspective in urban area remote sensing. Land cover considers characteristics
that not only come from the material itself. The surface structure (roughness)
affects the spectral signal as much as usual variations within the land cover
type (i.e., cracks in roads, buildings with different roof angles, age or cover).
Two different land cover types (e.g., asphalt roads and composite shingle
roofs) can be composed of very similar materials. From a material perspective
these surfaces would map accurately (Fig.
4.3
). The discrimination of the land
cover types roads and roofs, however, is limited by this similarity.
The land cover perspective is the most common remote sensing based approach
in urban area mapping and will be the focus of the rest of this chapter. Table
4.1
presents a hierarchical scheme to structure the diversity of land cover types within
urban areas. Broad land-cover classes are considered Level 1. Level 2 further sub-
divides the Level 1 classes based on their use, function or other generic surface
characteristics. Level 3 further separates the functional land cover classes based on
their material properties for built up classes. This classification schemes does not
claim to be complete and even more detailed levels of categories might be differen-
tiated, e.g., based on surface color or other characteristics. However, remote sensing
data with higher spectral and spatial resolution, and more specialized image analysis
techniques, are usually required when a more detailed level of land cover class
discrimination is desired in the mapping process.
4.4
Spectral Separability of Urban Land Cover Types
Measures of spectral separability are used to quantify the degree of discrimination
between land cover types. Several of these statistical measures have been developed
(see Schowengerdt
1997
) with the Bhattacharyya distance (B-distance) being very
useful in the analysis of hyperspectral data (Landgrebe
2000
). The spectral comparison
is based on several spectral samples for each land cover class from spectral libraries
