adopting mobile phone data to justify urban models a case...
TRANSCRIPT
Land Use Individuals Households
Road Infrastructure
Real estate development model
New development
Zoning
Household mobility model
Firm mobility model
Household location choice model
Firm location choice model
Hedonic price model
Demographic + socioeconomic
model
Growth Rate
Urb
ansi
m M
od
el
Year t-1
OD Travel Time (TT) (AM Peak auto/transit)
Traffic assignment
Car ownership model
Trip generation (HBO)
Mode choice (HBO)
Mode choice (HBW)
Destination Choice (HBW)
Destination Choice (HBO)
Ite
rati
on
s
Accessibility (logsum model)
Trav
el D
em
and
Mo
de
l
OD trip matrices by mode (auto/transit)
Auto Network/ Path Model
Transit travel time model
Trip generation (HBW)
Mobile source emission model
VMT Total
emission
Emis
sio
n
Mo
de
l U
rban
Sim
Ou
tpu
t
Credits: Joseph Ferreira, P. Christopher Zegras, Weifeng Li, Yi Zhu, Shan Jiang, Jae Seung Lee, Mi Diao, Tyler Kreider, and Marianne Hatzopoulou.
Adopting Mobile Phone Data to Justify Urban Models—A Case Study for Lisbon
Students: Shan Jiang (DUSP) & Laura B. Viña-Arias (CEE); Faculty Advisors: Marta C. Gonzalez , Joseph Ferreira, and P. Christopher Zegras
1. MOTIVATION & INTRODUCTION
In recent years, massive amounts of data generated by mobile phone activity have been increasingly used to help understand human mobility patterns, trajectories, and travel behavior. Mobile phones are becoming devices to track movement with no cost added to their usage
and are greatly distributed worldwide. These capabilities provide a new source of information that can be used to improve model
estimation and validation for traditional metropolitan accessibility systems.
In this study, we use mobile phone data provided by a local service carrier to
justify travel demand and travel times generated from an integrated land use and transportation (LUT) model calibrated for the Lisbon Metropolitan Area, Portugal. From the mobile phone data, a morning home-to-work origin-destination (OD) matrix was
estimated based on the two most frequently visited mobile phone towers.
The results show the potential for using this new type of technology to improve the use of complex integrated land use and transportation models.
2. AN INTEGRATED LAND USE-TRAVEL DEMAND MODEL FOR LISBON
4. MOBILE PHONE DATA TO ESTIMATE OD AND TRAFFIC FLOW
In the LMA LUT model: Population and employment data are aggregated from the census block (BGRI) level (the LMA contains 32,762 BGRIs), to the freguesia level. We use 216 freguesias as the spatial units of analysis . As for the mobile phone data analysis, Voronoi lattice is used to derive service area, We dissolved the census blocks (BGRIs) into 601 zones based on the 601 Voronoi Lattice of mobile phone towers. This enables us to estimate population and employment data that fall in the tower service areas.
Spatial Analysis Units
Population and Employment Validation
Data
The mobile phone data used for this study
record the time and tower location of on-going calls
more than 0.3 million users for an entire month in 2009 (w.r.t. 2.8 million population in 2009).
a total of 601 towers were identified in the Lisbon Metropolitan Area.
A time frame was determined in order to identify two most frequently used towers, assuming that they were the residential and employment locations for a mobile phone user.
A tower routing a signal from a mobile phone user during evenings was inferred as home location, and during daytime as the employment location.
These locations were used to estimate a 601 by 601 seed OD matrix.
All trips detected by the mobile phone usage in an
average day in the Lisbon Metropolitan Area.
Extrapolated OD Matrix: Iterative Proportional Fitting (IPF)
•Density of Employed Residents (in 2001) at the Tower-BGRI (601) Level
• Density of Mobile Phone Users (in 2009) “residing” at the Tower-BGRI (601) Level
Note: (a) Total # of Employed Residents in 2001= 1.29 million (b) The classification method is quantile. (c) The original data set is at the BGRI level(327,632), and is then aggregated to Tower-BGRI level (601).
Note: (a) Total # of mobile phone users in 2009 = 320,531 (b) In this map, the total # of mobile phone users was scaled up to 2001 employed residents level, so as to make the density of the two comparable.
•Density of Workers (in 2001) at the Tower-BGRI (601) Level
•Density of Mobile Phone Users (in 2009) “working” at the Tower-BGRI (601) Level
Note: (a) Total # of mobile phone users in 2009 at work-end = 307,965 (b) In this map, the total # of mobile phone users was scaled up to 2001 total employment level, so as to make the density of the two comparable.
Note: (a) Total Employment in 2001 = 1.3 million (b) The classification method is quantile. (c) The original data set is at the Freguesia level (216), and is then disaggregated to Tower-BGRI level (601).
Estimated by Raw Mobile Phone Data Estimated by Extrapolated Mobile Phone Data
5. MODEL COMPARISON RESULTS
3. MODEL COMPARISON APPROACH
We are aiming at comparing the base year (2001) road network performance estimated from the integrated LMA LUT model and that from mobile phone data.
OD pairs generated by LUT model are not directly comparable to those in the mobile phone model, since the basic spatial analysis units for our LUT model and the mobile phone model are quite different the LUT uses units generated from census zonal boundaries (freguesia) while the mobile phone model uses the Voronoi Lattice of the mobile phone towers.
We created the same road network (consisting of both centroids of “freguesia” and towers as nodes in the network) and used the mobile phone data to validate the LUT model’s road network performance instead of comparing the OD matrices from the two models.
Traffic Assignment of the Mobile Phone Model (2001)
6. CONCLUSION
External sources are important for validating increasingly complicated LUT model Mobile phone data+ data mining & statistical techniques present a rich validation
opportunity We demonstrate new methods for validation of integrated urban LUT models by using
mobile phone data, which can be scaled in a proper way to capture journey-to-work flows in the road network. represents a very low cost alternative to traditional data collection that are expensive and with
small sample
can be used as a daily detector of travel demand in the future.
7. ACKNOWLEDGEMENT
The generous support of the Government of Portugal through the Portuguese Foundation for International Cooperation in Science, Technology and Higher Education and was undertaken as part of the MIT-Portugal Program.
We acknowledge the contributions of Dr. Pu Wang, Weifeng Li, Yi Zhu, and Jae Seung Lee at MIT.
OD by Time-of-Day and by Mode
JTW_Tripsz,t is the journey-to-work trips originated from zone z during time period t,
ß is the total employed residents in zone z. EMPz is the total employed residents in zone z.
OD by Mode
Link volume and speed comparison between the LUT model and mobile phone model.
We cannot detect the transportation mode that each mobile phone user chooses for his/her trips.
We adopt the same mode share in the LMA LUT model.
We convert the 216x216 mode share pairs into 601x601 pairs based on spatial relationship.
Time of Day ß t Adj R2
7:00-8:00 AM 0.333 63.47 0.951
8:00-9:00 AM 0.409 70.80 0.960