Potential of hyperspectral remote sensing for analyzing the urban environment - Urban Remote Sensing: Monitoring, Synthesis and Modeling in the Urban Environment - page 46

4.1 Introduction

mapping which is described in Section 4.3.2. Exemplary results
are discussed and evaluated in their potential for thematic appli-
cations in Section 4.4. This way, the complete chain of urban
hyperspectral data analysis ranging from primary spectral inves-
tigations to derivation of higher-level information products is
covered demonstrating the high potential of hyperspectral remote
sensing for analyzing the urban environment.

Remote sensing technologies have been used for analyzing the
urban environment covering a wide range of applications in the
fields of urban land cover/land use mapping, urban planning,
urban growth monitoring, urban hazard and risk analysis as
well as urban ecology studies including investigations of urban
climate and hydrology. They are mainly driven by the need for
up-to-date information about surface conditions in the rapidly
changing urban environment whereas special emphasis is put
on the analysis of the spatially dominating manmade materi-
als/structures since they significantly determine the ecological,
climatologic and human living conditions and are most vulner-
able to natural and man-made hazards. These studies have been
carried out at different spatial and temporal scales based on a vari-
ety of sensor systems whereas mediumand high spatial resolution
multispectral data, such as Landsat-(E)TM, SPOT and IKONOS
have played the biggest role. For this purpose a wide range of
automated methods for land cover/land use mapping and change
detection has been developed. Despite the demonstrated high
potential of these approaches, there have been limitations in
automated multispectral analysis of material properties mainly
for man-made urban surfaces restricting consequential material
mapping which is required for physically-based modeling of
environmental conditions and for the derivation of quantitative
indicators allowing objective evaluation of urban areas.
In this context, advances in imaging spectroscopy (Schaepman
et al ., 2009) have opened up new opportunities for spatially and
thematically detailed analysis of urban structures. Hyperspectral
image data are especially suitable for material mapping due to
their high spectral resolution allowing detailed per pixel assess-
ment of spectral reflectance characteristics which are determined
by the chemical composition of the materials. The resulting
material-specific spectral absorption and reflection features form
the basis for identification and characterization of urban surface
materials. The potential of these data has already been inves-
tigated in a number of studies (see Section 4.2), which is still
small in comparison to studies based on multispectral data. In
the near future new opportunities will open up with the opera-
tional availability of satellite-based hyperspectral remote sensing
systems, such as EnMAP (Environmental Mapping and Analysis
Program) (Kaufmann et al ., 2006), PRISMA (PRecursore Iper-
Spettrale della Missione Operativa) (Sacchetti et al ., 2010) and
HyspIRI (Hyperspectral Infrared Imager) (Green et al ., 2008).
However, the existing studies have already shown the high spectral
information content of hyperspectral data. At the same time they
have revealed the challenges in automated information extraction
mainly consisting of the big variety of urban surface materials
and the small-sized urban structures leading to a large spectral
variability and a high number of mixed pixels. This requires
the development of specific methods for automated information
extraction which are capable of meeting the thematic needs for
detailed urban material mapping.
This chapter discusses the current state of the art of urban
hyperspectral remote sensing with emphasis on comprehen-
sive and quantitative assessment of spectral characteristics of
urban surface materials (Section 4.2) as the main prerequisite
for their automated identification (Section 4.3). In this connec-
tion the methodological experiences of the authors in urban
hyperspectral remote sensing are presented. They have led to
the development of an automated system for urban material

4.2 Spectral characteristics
of urban surfacematerials

This section deals with comprehensive assessment of spectral
characteristics of urban surface materials using hyperspectral
imagery. In contrast to natural materials, such as soil, rock and
vegetation, the number of investigations for man-made materi-
als dominating the urban environment has been fairly limited
(e.g., Ben-Dor, Levin and Saaroni, 2001; Herold et al ., 2004;
Heiden et al ., 2007). Motivated by the big variety of urban
surface materials discussed in more detail in Section 4.2.1, the
authors have carried out systematic research on spectral char-
acteristics of these surfaces in major German cities (Berlin,
Dresden, Munich, Potsdam) using field and airborne hyper-
spectral imaging spectroscopy. The investigations have resulted
in field and image spectral libraries of urban surface materials
widespread in German cities (Section 4.2.2). They form the basis
for the derivation of robust spectral features as an important
prerequisite for automated mapping of urban surface materials
(Section 4.2.3).

4.2.1 Categories of interest
for material mapping

Spatial inventories of surface materials are required for a wide
range of thematic applications ranging from mapping of imper-
viousness to characterizing thermal conditions (Section 4.4).
Therefore, several classification schemes have been developed
whereas the majority is structured in a hierarchical way (e.g.,
Roessner et al ., 2001; Herold et al ., 2004; Heiden et al ., 2007).
Most of them can be traced back to the land cover/land use
classification scheme of Anderson et al . (1976) allowing a flexi-
ble and application-oriented categorization of the earth surface
at different hierarchical levels based on the main land cover
classes, vegetation, water, built up/man-made areas, water and
bare ground. This scheme has been adapted by the authors to the
needs of comprehensive mapping of urban surface materials with
special emphasis on urban ecological aspects leading to primary
differentiation between manmade/artificial and natural surfaces
at level 1 of Table 4.1.
At level 2 main land cover types are distinguished whereas
man-made surfaces are further subdivided into buildings and
open spaces. Natural surfaces are categorized into vegetation, bare
ground and water bodies. In level 3 these categories are further
specified according to material types determining the appear-
ance of surfaces in hyperspectral imagery. Man-made roofing
materials are grouped into mineral, metallic, hydrocarbon, and
biomass ones. Artificial open spaces are distinguished according