1. - A case study in Austin TX
A multi-resolution approach in investigating the impacts of pre-planned road capacity reduction based on smartphone trajectory data
- A case study in Austin TX
XianBiao Hu, PhD, R&D Director, Metropia, Inc.
Yi-Chang Chiu, PhD, Associate Professor, Univ. of Arizona
Presented to:
16th TRB National Transportation Planning Applications Conference
May 17, 2017

2. Introduction
Pre-planned events are commonly seen, e.g. football games, constructions, street events, etc.
lead to road capacity reductions
At system-level, impact traffic dynamics
At driver-level, users detour to alternative corridors
At behavior level, various factors affect user’s route choice behavior
Need to study in a holistic multi-resolution manner

3. Prior studies and limitations
Simulation approach
Suitable for system level impact analysis
Not based on real data
Model calibration is challenging
Trip-based SBDTA not the ideal tool to study travel behavior and other activities

4. Prior studies and limitations
Fixed location sensors
Good to measure the traffic impact quantitatively
Individual traveler’s behavior change cannot be captured
Nor can the reasons for the detour behavior be revealed
Survey-based approach
Reveal information on behavior changes
Not based on observation
Suffers from miss reporting and nonresponse bias
Challenges in quantitative analysis

5. Smartphone Data collection
Data collection via smartphone GPS module
Reserve
Validate
When a user starts a trip with the Metropia app, the internal GPS module built in the smartphone is activated and starts to record the second-by-second data. These data, including detailed position such as latitude, longitude velocity and acceleration will be collected at fine time interval and sent back to the cloud server, where they will be stored and used for further analysis.

6. System-level traffic impact analysis
Identify the links of interest, both local and corridors of interest
Compare the traffic conditions before and after a capacity reduction event
Visualize and analyze in GIS platform

7. Individual behavioral change analysis
User set and trip set selection
Link set LS for the road segment where capacity reduction happens
Trip sets Before (TSB), Trip set after (TSA), Trip set non-traversal (TSN)
User set USS (those stayed), USD (those detoured)
Trip pair TSS (Those stayed), TSD (those detoured)

8. Behavior changes statistical analysis
Functional Data Analysis (FDA) approach
Take a series of data points that could come from discrete or continuous observations,
Approximate them by a function or curve, and
Perform analysis on such function/curve.

9. Behavior changes statistical analysis
FDA based trajectory data analysis workflow

10. Case study setup Lane closure event on MoPac, Austin, Texas
Started from Feb
Temporarily reduced MoPac NB from three lanes to two lanes between Lady Bird Lake and Enfield Road,

11. Traffic impact analysis
AM peak
Dramatic speed drop on MoPac NB approaching the construction location
With an approximately 1-mile long queue
Speed drops were observed on alternative corridors, e.g. I-35 Parallel freeway, S Lamar, S 1st Street, and TX-183

12. Detour outcome analysis
Travel time changes
For those stayed
5.7 min travel time increase
Equivalent to 27%
79% of them have travel time increases
For those detoured
7 min travel time increases
Equivalent to 31%
Prediction accuracy
For those stayed
System underestimated travel time by 12.8%
For those detoured
System underestimated travel time by 8%
Travel time VS Reliability

13. Detour Behavior Analysis
Detour pattern analysis: 3 pattern identified
First, we applied the FDA approach based clustering method presented in section to cluster trajectories, and three types of detour patterns are identified in Figure 3.
1. The first detour pattern is the group of drivers who have options to use a local arterial, such as Lamar Blvd, to avoid freeway traffic and the MoPac construction site. These drivers may originate from somewhere south of the river to go to downtown Austin or further north to destinations located between MoPac and I-35.
2. The second detour pattern is the group of drivers who shifted to other alternative freeways or highways. They could use TX-360, I-35, TX-183, or TX-130 to avoid the construction zone on MoPac.
3. The third detour pattern is the group of drivers who either have origin or destination addresses located downtown Austin. . For these trips, most drivers choose to avoid the construction zone by leaving MoPac earlier or entering MoPac later, while some other drivers choose to use I-35 or other corridors.

14. Detour Behavior Analysis
Next, we build statistical models to explain what attributes influence detour decision
Four spatial features, four temporal features
A binomial logistic regression model, and a multinomial logistic regression are employed
All four spatial features are significant
Two temporal features appear to be significant

15. Detour Behavior Analysis
LASSO multinomial regression to select proper variables and build robust model

16. Detour Behavior Analysis
LASSO Model result
The first detour pattern (avoiding freeways) is largely dependent on OD distance (S1).
The second detour pattern (using other freeways) usually have longer travel times (T2) and shorter estimated travel times (T1), i.e. trips with significant travel time increase tend to detour to other freeways.
The third detour pattern (O or D in downtown and frequently leave MoPac earlier or enter MoPac later) is most related with S4 (OD’s distance to construction area)
From the final LASSO model result,
1. The first detour pattern (avoiding freeways) is largely dependent on OD distance. The negative sign of S1 coefficient suggests that a longer OD distance come with a lower possibility of detour for drivers who exhibit first detour pattern. On the other hand, whether a user can avoid freeway is highly correlated with trip distance.
2. The second detour pattern (using other freeways) appears more complicated with 4 features selected as the order: T2, T1, S1,S3. The coefficients T2 and T1 indicate that drivers who exhibit this pattern usually have longer travel times and shorter estimated travel times. In other words, those trips with significant travel time increase tend to detour to other freeways.
3. The third detour pattern (O or D in downtown and frequently leave MoPac earlier or enter MoPac later) is most related with S4 (OD’s distance to construction area), i.e. if the user origin or destination location is close to the capacity reduction location, the possibility of them using a different ramp to enter or leave MoPac freeway and avoiding construction site will be high. The other significant features are T3, S1 and S2. Another interesting finding is that drivers tend to detour with this pattern more on weekends (indicator T3), which is likely due to the light traffic conditions on weekends, where using urban roads to bypass the construction area is a good choice.

17. Thank You
Q&A