IEEE Geoscience and Remote Sensing Magazine - September 2015 - 103

Figure 2. A mosaic of ASAR Wide Swath Mode collected
from 2005 to 2010 and covering North America. The mosaic is
108,000 # 50,400 pixels wide.

including two different sets of human settlements classified according to their size, are provided in fig. 4, while fig.
3 shows the intermediate and final outputs of the procedure for the area around Edmonton, Canada.
The overall approach has been applied to the whole
Earth land surfaces, subdivided into 7 regions. The overall
results for the P.R. of China are shown in fig. 5. The same
figure shows, in an insert, a detail of the whole picture,
allowing the comparison between the extraction from
SAR data and a Landsat-based extent extraction, kindly
provided by Tsinghua University, although including only
the most populated cities. As a matter of fact, the extraction of the global ASAR WSM map of urban areas has been
recently validated [24] by comparison with extractions
performed by visual interpretation using Landsat data for
the 663 most populated cities in P.R. China, showing a
very low omission rate. Additionally, the results, as clearly
visible in the insert of fig. 5, highlight the capability of
SAR systems to characterize small settlements. Large cities need however to be better delineated, and this is where
extractions from finer spatial resolution SAR/optical data
can make the difference. Another approach to global urban mapping - Urban Area Detecting Parameter (UADP) exploiting the above mentioned multitemporal stack of
ASAR WSM data is presented in [25]. The UADP method is
based on modeling the co-occurence matrix of the pixel's
and the intensity and properties of its surrounding pixels under a neighborhood constraint via a neighborhood
Greylevel Dependency matrix (NGLD). While the method
is suited for global urban extractions due to its simplicity and used data characteristics and availability, it is connected to some limitations and drawbacks, similar to the
ones from the UEXT algorithm. Both algorithms perform
worse in arid regions where bare ground and rocks cause
high backscatter values which are very similar to the backscatter from artificial structures. Also steep slopes look
quite similar in ASAR WSM imagery, from a digital number (DN) point of view and need to be treated separately.
Due to these limitations UADP as well as UEXT require a
number of post-processing steps, including morphological filtering and DEM based corrections. However, in order to satisfyingly extract information about the mix of artificial and natural features in heterogeneous inner-urban
september 2015

ieee Geoscience and remote sensing magazine

Figure 3. Intermediate results of the human extent extraction
procedure for the area around Edmonton, Canada. Top row: the
brightest point selection results; middle row: results of the region
growing step; bottom row: the final extracted urban extents.

structures levels of detail below 75 meters, HR and VHR
data is necessary.
B. Finer resolution techniques
Although the double bounce effect is still valid as a peculiar
feature of SAR images of urban areas at high and very high
spatial resolution, the available finer details change the appearance of human settlements. Instead of having multiple
103

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