Characterizing urban subpixel composition using spectral mixture analysis - Urban Remote Sensing: Monitoring, Synthesis and Modeling in the Urban Environment - page 117

April 16, 2003 August 19, 2002

(a) (b)

(c) (d)

12 km

10-35%

% Cover

35-70%

0%

<10% 70-100%

FIGURE 8.7 Subset of shade-normalized V-I-S fractions for Denver, Colorado. Panels (a) and (b) display fractions results as
RGB composites: non-photosynthetic vegetation (R), green vegetation (G), and soil (B). Panels (c) and (d) display the green
vegetation fractions results, binned into classes that highlight change in vegetation cover.

grouped into generalized V-I-S classes, and continuous maps of
each component are generated. Accuracy is assessed in terms of
agreement between modeled fractions and reference fractions as
measured from finer spatial resolution imagery or from the field.
A potential disadvantage of this approach is that the process
will never be fully automated, as iterative revisions and contex-
tual meaning must be supplied by an analyst in the context of
the goals of each project; however, the examples presented here
highlight a major strength of this approach - that model imple-
mentation can easily be customized to best capture local spatial
and spectral variations. The output can be generalized within the
V-I-S framework, facilitating the comparison of urban landscapes

across regions and through time. Maps that summarize model
(i.e., spectral) complexity demonstrate the potential to delineate
urban extent at regional and global scales. Because SMA fractions
model the landscape in terms of continuous variables, landscape
change can be quantified in terms of land-cover conversion (i.e.,
urban expansion) and land-cover modification (i.e., internal
changes within the urban fabric itself) (Rashed et al ., 2005).
While the examples presented in this chapter analyzed multi-
spectral datasets, future applications of MESMA will increasingly
be applied to hyperspectral datasets, especially when such systems
become operational on satellite platforms. The primary advantage
of hyperspectral data is the increased spectral information content