Visualize Temporal and Spatial Patterns of AirSage Trips Data
Huajie Yang, Wei Shi, Yi Wang, and Liming Wang (Faculty Advisor)
12/16/2016
Introduction
AirSage collects anonymous wireless signaling data and processes them into aggregated information of the location, movement and flow of people. For transportation research, AirSage data can be processed as OD flows at the common Traffic Analysis Zone (TAZ) level. Besides its nationwide coverage, the data is at a high temporal resolution over a long-time span, a rarity for other transportation data sources.
For this project, we are taking advantage of the data’s high temporal resolution and long-time span to visualize and explore the temporal and spatial patterns of trip flow. We visualize various aspects of the data, in particular, the temporal variations of trips and flow and the difference in patterns across trip purpose and subscriber class. We also visualize the temporal-spatial pattern of travel flow via an animation of net trips at the TAZ level.
This paper is available on github. We follow the practice of reproducible research (Gandrud 2015) for our work and this write-up is fully reproducible from a Rmarkdown file and R scripts, the source code of which is available on a github repository at https://github.com/cities-lab/TRB2017DataContest (the data files are withheld according to AirSage’s confidentiality agreement).
Visualization
Temporal Variation of Trips by Day of Week and Trip Purpose
We first visualize the variation of trip frequencies by time-of-day, day of week, and trip purpose. We expect trips to vary by all the three dimensions, likely more so for time-of-day and trip purpose than day of week.
Figure 1 displays trip frequencies for nine trip purposes by time-of-day and day of the week. We apply Loess smooth to draw these line charts, in which the solid lines are the mean trip frequencies over the same day of week in April 2014 and the shades represents the 95% confidence interval.
Trip frequencies for the OO (Other to Other) purpose are substantially higher than those for other purposes. It makes sense because the AirSage data only distinguishes home and workplace, and lumps all other locations into others. Even thought the height of the peaks are similar throughout different days of a week, it appears at a different hour of the day and the shape varies from day-to-day: it has a similar shape from Monday to Thursday, but a different one on Sunday, Friday, and Saturday.
Another noticeable pattern in Figure 1 is that, compared with weekdays (Monday-Friday), there is no morning peak of HW trips on weekends (Saturday, Sunday), which makes sense as most workers get these days off and don’t have to make HW trips.
As shown in Figure 1, the variance (shades) of OO trip frequencies is the largest, followed by HW, HH, HO and WH trips. The variance of OH, WO and OW trip frequencies is small. The variance of work-related (WO, WH, WW, OW, HW) trip frequencies on weekdays is larger than those on the weekends. The difference in variance between weekday and weekend seems to make sense that much fewer HW trips are made during the weekends than during weekdays.
To make it easier to see the difference between weekdays and weekends, we create Figure 2 that displays trip frequencies for the nine trip purposes over time-of-day on weekdays and weekends. Besides a wider peak on weekday, there is little difference in trip frequencies for OO purpose between weekday and weekend (even though there is higher variance on weekend). For work related trips, trip frequencies on weekend is much lower than that on weekday. This makes sense that few people go to work places during weekend. The distribution of HH trips are quite different on weekday and weekend: there are two peaks of trip frequencies for HH purpose on weekday, while there is only one sharp peak on weekend. For OH trips, there are higher trip frequencies during 12 to 17 hour (12-5pm) on weekend than on weekday. This makes sense that people are more likely to go back home during this period on weekend than on weekday.
Finally, we would like examine how the patterns vary by hours and day. Figure 3 displays trip frequencies for the nine trip purposes by time-of-day (X-axis) and date of April 2014 (Y-axis). Each rectangle in Figure 3 represents scaled trip frequencies (z-score) for a certain purpose during one hour period of certain day. Figure 3 further shows that there is no obvious difference in trip frequencies for WW, OO and OH purposes between weekday and weekend. The most obvious difference in trip frequencies between weekday and weekend are OW, WO, WH and HW trips. Figure 3 also shows that HW and WH trips on weekend are made later than those made on weekdays. This indicates that people go to work later on weekend than on weekday and they also leave later on weekend than weekday. They spend same time period on working during weekend as during weekday. This implies that workers have more flexible schedule on weekends. There is also difference in trip frequencies for HH and HO between weekday and weekend, but they are not as obvious as WW, OO and OH trips. Figure 3 shows there is almost no difference between two days if they are both weekday or weekend no matter what dates they are. This implies that there is little variation across date of month in April 2014. One exception is April 18th, which is a Friday but its trip frequencies for HO is more similar to weekend, because April 20th is the Easter and April 18th is Good Friday.
Temporal Variation of Trips by Day of Week and Subscriber Class
We repeat the above visualization for subscriber class. Comparing the difference in travel patternamong subscriber classes, we found that short term visitors made the most trips, followed by long term visitors, resident workers and home workers, whereas there were relatively less trips made by inbound and outbound commuters. In terms of time of day, the trip count is an apparent inverse U curve indicating more trips were made in daytime, with PM peak trend for resident workers. All subscriber classes seemed to have later starts in the mornings and earlier ends in the evenings in weekends than weekdays. Resident workers made relatively more trips in late afternoon , possibly after-work errands, in the weekdays than weekends.
To better examine the difference between weekdays and weekends, we compare the travel patterns between weekdays and weekends among different subscriber classes. It was clear that almost all subscriber class, except long term visitors made more trips in weekdays than in weekends. The resident workers’ travel patterns of more late afternoon trips in weekdays are more apparent in the figure below. In addition, the variance of trip frequecies is bigger in weekends than weekdays, indicating trips frequecies may be less predictable on weekends than on weekdays.
Spatial Variation of Trips by Hour (Average Day)
Next we visualize and examine the spatial flow patterns in the AirSage data. We are interested in the hour by hour net trips at the TAZ level (for each TAZ, net trips = trips destinated - trips originated) for an average weekday. It is clear that net trips come from fringe of the Orlando area in the morning, concentrates in a few hotspot TAZs during middle day, and then dissipates in the evening/night.
Animation of Average Net Trips on Weekdays
Animation of Average Net Trips on Weekends
