Potential of hyperspectral remote sensing for analyzing the urban environment - Urban Remote Sensing: Monitoring, Synthesis and Modeling in the Urban Environment - page 48

TABLE 4.1 ( continued ).

Level 1 Level 2: Land cover types Level 3: Material types Level 4: Surface materials

Vegetation Trees Coniferous

Deciduous

Shrubs/bushes
Meadow/lawn Meadow - dry

Meadow - fresh
Ornamental lawn
Sports turf

Bare ground Soil Soil - dark

Soil - bright

Sand Sand - coarse

Sand - fine

Rock

Water bodies Ocean/sea

Inland waters

Lakes
Ponds
Rivers

Surface materials (level 4) marked in italics have been spectrally investigated by the authors.

the hierarchical structure of the developed scheme allows for
successive extensions at levels 3 and 4. The surface materials
listed at level 4 are characterized by a large number of spectral
variations due to varying material properties and bidirectional
reflectance distribution function (BRDF) effects resulting from
the imaging process. In case of man-made surfaces material-
inherent differences are caused by color, coating, weathering
and usage. For natural surfaces possible variations are the
distinction between green and non-photosynthetic vegetation,
the determination of main plant communities, rock and soil
types as well as the determination of the eutrophic state of water
bodies. Since the resulting number of categories is very large, only
selected ones are included in Table 4.1. However, comprehensive
material mapping requires systematic investigation of such
variations for the materials of interest in order to determine
their effects on spectral reflectance (Section 4.2.2).

result, field spectra for about 80 spectral targets were stored in
the field spectral library comprising about 50 targets for roofing
and 30 targets for open surface materials. All of these targets
were measured under open sky and direct sun illumination in
nadir position. A detailed methodological description of spectral
data acquisition assessed material variations and description of
the spectral behavior of selected man-made materials in relation
to their spectrally relevant chemical compounds can be found in
Heiden et al . (2007).
Since the establishment of a comprehensive field spectral
library is very time consuming and does not cover the large
spectral variety occurring in hyperspectral imagery, an image
spectral library has been built using the field spectral library to
assign surfacematerials to the image spectra. They were identified
interactively in the hyperspectral image database comprising
11 flight lines acquired for test sites in four cities between
1999 and 2007 (Table 4.2). The data were recorded by the
hyperspectral HyMap sensor operated by the German Aerospace
Center (DLR) and the obtained radiances have been transformed
into reflectances for normalization purposes (Heiden et al ., 2007).
For each material the number of spectra varies between a few
hundreds and several thousands depending on its frequency of
occurrence and spectral variability. This way a total of about
60 000 image spectra has been stored in the image spectral
library representing those surface materials of Table 4.1 marked
in italic print. Subsequently, this database has been used for
determination of robust spectral features allowing discrimination
between these materials (Section 4.2.3).

4.2.2 Establishment of urban
spectral libraries

In spectral libraries detailed knowledge about spectral material
characteristics is stored in form of spectra. They are recorded by
laboratory, field and imaging spectroscopy allowing the analysis
of materials at different spatial scales, spectral resolutions and
environmental conditions. In hyperspectral remote sensing they
serve as important input information for pre-processing and
classification procedures and contribute to the general spectral
understanding of the phenomena of interest. Spectral libraries
presented in this paper have been established during the last 10
years based onfield and imaging spectroscopy for various test sites
in the German cities of Berlin, Dresden, Munich and Potsdam.
Assessment of field spectra was performed using a field spec-
trometer (Analytical Spectral Device (ASD) Field SpecPro FR)
recording data with 1 nm spectral sampling distance (SSD) in
the wavelength range between 350 nm and 2500 nm. In the

4.2.3 Determination of robust
spectral features

A spectral feature represents a specific spectral behavior within
two wavelength positions in the reflectance spectrum. It can be