IEEE Geoscience and Remote Sensing Magazine - September 2015 - 101

spatial resolution (VHR) sensors provides fine object details, but with a limited capability to recognize materials
because of their, usually small, number of spectral bands.
Data collected from radar sensors mix geometrical and
material properties into urban area images which are
quite difficult to understand by a non-expert. Hyperspectral data offer hints to material recognition, but lack the
spatial details, resulting, within urban areas, mostly into
mixed pixels, i.e., pixels representing areas where multiple material spectra mix up in a linear or non-linear way.
Finally, three-dimensional sensors allows discriminating
between objects with similar material considering their
3D geometry, and to better understand the results by
two-dimensional sensors. Their spatial resolutions, however, do not usually match the one of the other sensors,
and their 3D capabilities and accuracy varies according to
the system, being it Interferometric SAR, LIDAR, or stereo
photogrammetry.
Additionally, it must be noted that the challenges for
spaceborne and airborne data fusion in urban areas depend on the final application. Data which are very useful in specific cases may be totally useless in other ones,
depending on the physics of the studied phenomena, the
spectral, spatial and temporal resolution of the sensors, the
quality of the data and the availability of effective algorithms for information retrieval. From the point of view of
the data fusion technique, following the standard nomenclature in [1], these challenges can be addressed at the raw
data, feature or decision level.
To deal with all of these sensors and organize the different ways to combine this wealth of information, a data
fusion framework specialized to urban areas was introduced in [2]. It includes the possibility to combine sensors
working at different wavelengths, with different spatial
resolutions, and adding whenever possible multitemporal
capabilities. The framework has been revised and the challenges updated in [3], noting than in 10 years the stress
has moved from the need to combine multiple sensors in
a limited geographical area to the necessity to combine
data sets on wide geographical scenes, possibly on a global
scale. This explains the large amount of papers dealing
with global urban analysis [4]-[12] posterior to the date of
the first review paper.
Combining multiple remote sensing data sources at the
national, continental and global level implies not only the
need to select the most effective sensors for a specific problem, but also to design techniques that allow to efficiently
manage large amounts of data. As mentioned in [13], the
time cost required to apply relatively complex algorithms
tested on a single scene to the whole surface of the Earth
would result impracticable without using parallel/cloud
computing, and efficient computing methodologies. Instead, the possibility to implement or run efficient algorithms allows near-real-time VHR spatial and spectral data
processing in urban areas (see for instance the case of hyperspectral data unmixing in [14]). Alternatively, simpler
september 2015

ieee Geoscience and remote sensing magazine

approaches can be implemented on reduced spatial and
spectral resolution data sets [15].
A big advantage of any of these techniques is the possibility to exploit archived data by reprocessing the massive
data stored and available for multitemporal and/or multiple geographical area analysis. In this line of research, the
current work specifically focuses on the use of global data
sets of spaceborne multispectral data, jointly with equally
global synthetic aperture radar (SAR) data, for the characterization of urban area extents. By combining efficiently
and at the global level the data currently available (or that
is going to be stored) in databases by ESA and NASA it
is expected to be able to characterize phenomena in the
past with a large relevance in our present, such as urban
sprawl and its connection with global warming, regional
global meteorological phenomena, as well as natural and
man-made disasters. According to [16] more than half of
the world population is currently living in cities and this
trend is going to continue, leading to an estimate of 70 %
of urban dwellers in 2050. The ongoing increase of urban
population is strongly connected to an expansion of urban extents. In turn, transition from natural to artificial
surfaces in areas of urban sprawl indisputably affects atmospheric conditions, local and regional surface temperature patterns as well as cycling of water, carbon, aerosols
and nitrogen in the climate system, on local, continental
as well as global scales. In order to understand the impact
of urbanization on global climate, information on how
and where urban areas develop is crucial. Therefore, the
urban footprint must be accounted for in emerging climate
modeling systems [17]. Urbanization and climate change
is also directly connected to other research fields including
natural hazards and risk assessment, humanitarian crisis
management or reinsurance economics, just to name a few.
Thus, accurate information about global urban land
cover and its change is unconditionally needed by a variety
of entities, including national authorities, international
organizations, reinsurance companies and scientists stemming from a wide range of research areas. This work builds
on the recent achievements in this area by the University
of Pavia, starting from the results published in [19]-[23].
Accordingly, Section II will describe the methodology used
to process global coarse resolution SAR data sets, as well as
to extract urban extents from finer resolution data available from past and current SAR sensors. Section III will
be instead devoted to showcase the procedure for urban
extent extraction using Landsat data on wide geographical reason and the Google Earth Engine platform. Finally,
Section IV provides some hints on the fusion of data sets at
multiple resolutions and SAR/multispectral sensors for the
same problem. Section V provides the reader with a general
discussion about the topic and concludes the paper.
II. GLOBAL MAPPING OF URBAN AREAS BY SAR
The big advantage of SAR for urban area mapping relates
to the peculiarity of this active microwave sensor, able to
101

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