Characterizing urban subpixel composition using spectral mixture analysis - Urban Remote Sensing: Monitoring, Synthesis and Modeling in the Urban Environment - page 103

8
Characterizing urban subpixel
composition using spectral mixture
analysis
Rebecca Powell
Deriving accurate measures of urban land cover from remote sensing data is fundamentally a challenging endeavor
due to the high spectral and spatial heterogeneity of urban areas. One strategy to address spatial variability is
mapping sub-pixel components of land cover using spectral mixture analysis (SMA), which models each pixel as the
linear sum of spectrally ''pure'' endmembers. Spectral variability can be addressed by multiple endmember spectral
analysis (MESMA), which allows the number and type of endmembers to vary on a per-pixel basis. The high
spectral diversity of urban materials can be generalized in terms of three fundamental components - vegetation
(V), impervious surfaces (I), and bare soil (S) - facilitating comparisons of urban landscapes across regions and
through time. This chapter presents an overview of the SMA technique to characterize urban landscapes, followed
by two case studies that highlight the flexibility of MESMA to map V-I-S components. The first study characterizes
the morphology of settlements along the ''arc of deforestation'' in the Brazilian Amazon, leveraging assumptions
about the unidirectional trajectory of urbanizing land cover to improve accuracy of fraction estimates. The second
study quantifies urban tree and grass cover in a Western US city, integrating phenological information to
discriminate dominant vegetation types.