Geoscience Reference
In-Depth Information
particular the dips on Fig.
15.5
c revealing that this area is less active than other
parts of the city in early morning). All these characteristics go in line with the
commercial and touristic nature of the place.
The Camden Market signatures have average patterns during working hours
and are typically characterized by a peak of SMS activity during the workday
evenings (consistent with the recreational nature of this location where people
may gather to share a drink) and high level of request activity around lunch time
and in the early afternoon in the weekend (in line with the popularity of the
markets).
The signatures of the Newham pixel are specifically characterized by low
weekday-to-weekend activity ratios (suggesting a constant population rate over
the week) and specific nonzero request activity at night (the automatic update of
the mobile devices revealing that people are sleeping at those locations). These
characteristics straightforwardly reveal the residential nature of this location.
Finally, the signatures in Ealing Broadway are rather comparable to Newham's
one - revealing the residential nature of the area - but also show specific peaks of
request activity during morning and evening commuting hours, in line with the
commuter hub property of the Underground station.
These observations show a correspondence between mobile traffic signatures
and the nature of the places studied, suggesting that we could infer the nature
of each area based on its signatures. As a side perspective, the identification
of the central business district and the residential areas can offer insights on
the nature of commuting flows. Urban planners are already aware of the spatial
relationship between business district and residential areas, but the visualization of
such properties on such a precise spatiotemporal scale has been possible only for a
few years thanks to the advancement of digital data gathering.
15.5
Cluster Analysis
15.5.1
Principles
In the previous section, we focused on single pixels, and we have shown how their
signatures could reveal human dynamics' features at the local level. In this section,
we investigate the question of whether we can use mobile device traffic data to
detect large areas with homogeneous properties. Our goal is to group local pixels
according to the similarity of their signatures and use these groups to map the urban
spatiotemporal structure of the cities.
Among the many different clustering techniques to extract clusters of pixels with
similar signatures, we chose a K-means approach, used in many previous studies
(Andrienko et al.
2013
; Pei et al.
2013
; Reades et al.
2007
). This approach ensures
that each pixel of a cluster has a signature as much like the one the other members of
the clusters and as different as possible from the signature of the pixels in the other
Search WWH ::
Custom Search