IEEE Geoscience and Remote Sensing Magazine - March 2015 - 27

be attained in the leaf-off season, and that a temporal resolution of one to two years would give a good
estimation of a road width [87].
◗ Building height: literature reports several researches
that focused on extracting building height as a subsequent step of extracting building footprints. Building
heights were used to estimate building floor number,
as an indicator of common importance for different
elements of risk assessment like exposure and vulnerability. In fact, the indicator is an attribute widely used
in building taxonomies like PAGER-STR (Prompt Assessment of Global Earthquakes for Response) [107]
or GEM (Global Earthquake Model) [108], due to its
direct relation to structural failure in extreme events.
Additionally, it helps disaggregating population and
computing population density as well as the average per
capita square footage in a given area. These outcomes
are of particular importance in areas where the access is
difficult, data are scarce, and in urban areas where the
population density increases due to building height rise
instead of built-up area extension. Previous literature
has indicated three main approaches used in estimating
building heights from high-resolution satellite images.
a) The first method used stereo pairs with two matching techniques, window-based and pixel-based.
The drawback of this approach is in the need to
find an optimum size for the matching-search
window, in addition to the over estimation of a
building size in the generated DSM (Digital Surface Model) [21].
b) The second methodology is shadow-based, and
it estimates building height using the acquisition
geometry of a single image. The classification outcome determines the building footprint and its corresponding shadow, and then the building height
is estimated using the imaging angle, sun azimuth
and sun elevation. The accuracy of the method
depends highly on the urban complexity and the
likelihood of detecting the correct shadow. These
methods deliver good results only if buildings are
correctly detected during the classification step and
if they are tall and isolated enough to generate a
definite shadow area around the building [109].
Moreover, overestimation of the building height
may rise from shadow areas being mixed up with
occluded areas [22].
c) LIDAR proved to have great potential, representing a unique data source apparently more effective
than satellite and aerial photographs in accurately
extracting urban features and their height, particularly for buildings [110]. However, acquisition cost
and limited coverage have curbed LIDAR usage to
extremely local scales [111]. Moreover, LIDAR technology offers a possible solution of the above-mentioned difficulties on extracting roads and building
footprint from multispectral data. In particular,
march 2015

ieee Geoscience and remote sensing magazine

the elevation information from LIDAR data can be
beneficial in enhancing discrimination between
buildings and background. Therefore, LIDARbased techniques would afford a great improvement in the extracted cartographic information
like road networks and buildings footprints [112],
but they are not viable for the large-scale analysis
required by risk management of natural disasters.
◗ Building age: a good estimation of single-building age
is of particular interest in the assessment of its vulnerability class. This indicator considers the building code
in force at the time of construction, the aging of the
structure, and the presence of factors that may have increased the building vulnerability (e.g. previous earthquakes). Building age estimation is a direct product
of the extraction of buildings footprint of consecutive
dates. Age estimation accuracy is highly dependent on
the accuracy of results on footprint extraction. Moreover, the regularity of single-footprint polygons is not a
requirement in determining the building age, as far as
matching of footprints from the same building at different dates can be ensured. This indicator is typically extracted by masking the building footprint of a sequence
of HR images. The presence of a building footprint in
a masked layer would let one derive the building age.
◗ Roof type: The roof type is another critical indicator in
building taxonomy that can be extracted remotely. In
extreme events, ceiling collapse rather than wall collapse is the most common cause for increases in fatalities and injuries. Building taxonomies include a large set
of roof types, which react differently in extreme events;
some of these types enhance the vulnerability to disasters like earthquakes.
a) From hyperspectral data: spectral resolution is the
critical parameter in extracting this indicator. Therefore, spectral bands number, location and bandwidth have to be considered carefully in selecting the
most suitable data for extracting this indicator. The
increasing number of spectral bands, ranging from
multi-spectral to hyper-spectral data has enhanced
the potential of recognizing the nature and properties of the different building roof materials [19].
b) From multispectral data: lots of efforts are reported
in the literature on roof type extraction from multispectral data. The main issue to consider here is
that reflected multispectral data does not allow the
classification of the roofs according to the material
characteristics but rather to their spectral color. Using multispectral data the roof types would basically reflect the color of the roof material.
◗ Footprint irregularity: The fuzzy concept here is that
irregular building footprints imply higher vulnerability [94]. What can be inferred from remote sensing imagery is the plan irregularity. The indicator is estimated
from building footprint shape and its possible torsion
response to horizontal mechanical shake created by an
27

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - March 2015