URBAN ANALYSIS FOR THE XXI CENTURY: USING PERVASIVE INFRASTRUCTURES FOR MODELING URBAN DYNAMICS

Urban Analysis for the XXI Century:  Using Pervasive Infrastructures for  Modeling Urban Dynamics

Enrique Frias‐MartinezTelefonica

Research, Madrid, [email protected]

Índice

• Introducción• Pervasive

Infrastructure

• Hotspot

Detection

• Land Use Classification• Commuting

Patterns

• Conclusiones

Introducción

Introducción

“The 19th century was a century of empires, the 20th century was a century of nation states, the 21st century will be a century of cities”

Wellington E. Webb, former mayor of Denver

Introducción

Digital Footprints For the first time in human history, we have 

access to large‐scale human behavioral  data at varying levels of spatial and 

temporal granularities

Pervasive

Infrastructure

Pervasive

Infrastructure

1

Cell Phone Netw ork

Cell Phone networks are built using Base Transceiver Stations (BTS).

Each BTS will be characterized by a feature vector that describes the calling behavior area.

Pervasive Infrastructure

1

CDR dataset

Our Dataset• 1 month of phone call interactions.

• 1100 Base Transceiver Stations.

• Each CDR contains:

› phoneSource

| phoneDestiny

| btsSource

| btsDestiny

| DD/MM/YYYY | hh:mm:ss | d

• Phone number are encrypted to anonymize user identities.

Traffic

Subscribers sample

Cell catalogue

Mobility algorithms

2233445566|15/02/2008|2233445567|15/01/2008|2233445568|15/07/2008|25/07/20102233445569|15/09/2008|

Hotspot

Detection

Hotspot

Detection

• What is a hotspot?– In this context a hotspot is understood as a 

concentration of people (or activities) over a  specific period of time and a specific geographic 

area.

• Interesting for urban planning, emergency relief,  public health, context‐aware services

• Approach– Greedy clustering algorithm seeded with local maxima

– Hotspots based on activity or on number of people.

Hotspot

Detection

• Data:– CDR from Mexico for a period of 4 months. 

• Output: – At a national level: cities. At an urban level: city 

blocks. Evolution of dense areas for urban  planning.

Hotspot

Detection

Hotspot

Detection

Weekdays Morning Weekdays Afternoon

Weekdays Evening Weekdays Night

Land Use Classification

Land Use Classification

• Aggregate and clean data for each BTS.– Obtain signature of each BTS (total number of 

calls every hour: 24 hours average week day and  24 hours average weekend day)

– BTS based Voronoi gives the tessellation for land  classification.

– Automatic Identification of clusters with similar  behaviour that maximize the compactness of the 

groups identified.

Land Use Classification

1

Re pre se nt a t ions

Activity signature vectors are built: each component contains the number of managed calls by the BTS in 5-minute intervals.

Land Use Classification

• Industrial Parks / Office Areas

Land Use Classification

• Commercial ‐

Residential

Land Use Classification

• Night Life Areas

Land Use Classification

• Weekend Activities

Land Use Classification

• Residential

Commuting Patterns

Commuting

Patterns

Commuting

Patterns

Conclusions

Conclusiones

• Traditional

approaches

are costly

and

based on

questionnaires.

• Urban Dynamics

can be modelled

using pervasive

infrastructures

• Reduction

in cost, increment

of

the

flexibility

• Possibility

of

real‐time modelling