Towards a Comparative Science of Cities: Using Mobile Traffic Records in New York, London, and Hong Kong - Computational Approaches for Urban Environments - page 371

very similar to those of the London green cluster, with the exception of a specific
peak of activity on Saturday just before lunch time. This peak could be explained
by a recurring event (such as a market). In New York, a quick comparison with
detailed maps of the city reveals that the pastel cluster corresponds to parks.
The activity shares of this cluster are very low (in accordance with the fact that
there are no people residing or working full time on the park premises ), and its
signature shows high activity levels during the weekends (when people may go
for a walk within the nearer park). Finally in Hong Kong, the pink cluster appears
to be a second residential one.

To summarize, each city can be characterized by a gradation of clusters. A
feature common to all cities is the existence of core business areas and purely
residential areas, an already known urban fact that we were able to check thanks
to communication traffic data.
Compared to classical time-consuming and expensive field surveys, our approach
makes it possible to build automatic, quick, and relatively cheap way of preparing
land use maps of the cities. The maps we obtained are closely related to classic
land use maps, especially for the distinction between business and residential areas.
When it comes to other land use classes, some difference occurs. Indeed, while the
classical land use classes are related to what the land looks like (is the neighborhood
consisting of a retail store, bank, residential buildings, etc.), the clusters we found
are based on communication data revealing human dynamic behaviors (like working
in an office, shopping, eating, commuting, sleeping, etc.). Rather than emulating
classical land use mapping, our approach thus produces a complementary point of
view that enriches our understanding of the multiple dynamics at stake in the cities.
Although we used every type of activity in our clustering analyses, the simi-
larities between the signatures of different types (calls, SMS, request, UL or DL
data) displayed on Figs. 15.6 c- 15.8 c suggest that our finding would not qualitatively
change if we focused on only one activity type.

15.5.3 Revealing Universal Patterns

The results presented in the previous section suggest that mobile traffic patterns can
reveal a concentric structure of cities into clusters that can be interpreted the same
way. To what extent are these cities similar? In this section, we investigate this issue
by making a transversal analysis of our three studied cities. We performed a K-
means clustering analysis on all cities at once, grouping 500m by 500m grid pixels
with similar signature patterns.
Figure 15.9 displays the results of this transversal clustering analysis. As before,
we chose to display results obtained for K

D 6.

￿ All previously identified core business centers are gathered into a single (here

red) cluster, whose signature can be characterized as before.