Geoscience Reference
In-Depth Information
corresponding to several million subscribers in all studied cities (precise penetration
rates cannot be given for confidentiality reasons). Data were collected between April
1 and July 7, 2013 at fifteen-minute intervals across the three cities at cell level.
While the whole city of New York is covered, the dataset of Greater London is
mostly concentrated on the Inner London districts. The Hong Kong measurements
cover urban zones, while no data is available for the unpopulated mountainous parts.
The three months of data consist in counter data recording the numbers of calls,
SMS, and requests (for data communication initiated either by the users or some
applications running in the background in their mobile devices) as well as the
amount of data uploaded and downloaded by subscribers (measured in bytes and
thereafter denoted by “UL Data” and “DL Data”). The provided data was aggregated
at the cell level by the data providers and therefore did not reveal any individual user
information. Before receiving the data, the actual numbers were obfuscated by using
a secret scaling factor, such that we have only access to normalized amounts of each
type of counter data.
In addition to mobile phone data, we gathered various shapefiles, census data,
and land use data from open access sources.
1
We thus obtained land use data of
different nature and with different number of categories across the different cities
(9 categories for London, 6 for New York, 24 for Hong Kong). To better compare
the results obtained in our three cities, we converted these original categories into
seven land use types that best match them: High-Density Residential, Low-Density
Residential, Business and Commercial, Mixed (Residential and Commercial),
Infrastructures, Parks, and Other. Details of the procedure are available on request.
The location of the cells' recording mobile phone activity is given as longi-
tude/latitude pairs - the service area of a cell having a typical radius varying from
around 100 m (in dense area) to several kilometers (in rural zones), while the census
and land use data are provided in polygonal zones corresponding to administrative
divisions in the cities. In order to study the mobile phone usage patterns in the
different cities and their relationships to census and land use data, we chose to
transform the spatial representation of the different datasets by projecting them on
uniform lattice grids of 500 m by 500 m “pixels.”
To reduce the bias induced by the attribution of the activity within a cell's service
area to a single pixel location, we used a smoothing procedure: we defined the
activity on one pixel as the mean of activities of all cells within a 1,500 m by 1,500 m
square centered on the pixel center. Census data were similarly projected on the grid
by interpolating demographic data and most significant land use on each grid cell.
The length of 500 m was chosen after testing different grid sizes. It proved to be
coarse enough to reduce noise level and detailed enough to explore spatial patterns
of activities within the cities. At the end of this procedure, the mobile phone traffic
data was projected on around 2,700 pixels in Greater London and around 3,000
pixels in New York and Hong Kong.
1
such as
http://data.london.gov.uk/
for London,
https://nycopendata.socrata.com/
for New York or
http://www.census2011.gov.hk/
for Hong Kong.
Search WWH ::
Custom Search