Using Google's Aggregated and Anonymized Trip Data to Estimate Dynamic Origin-Destination Matrices for San Francisco
Corresponding Author: Bhargava Sana, San Francisco County Transportation Authority
Presented By: Bhargava Sana, San Francisco County Transportation Authority
Historically, origin-destination (OD) data could only be collected using time-consuming and expensive vehicle intercept or license plate surveys. Recent technological advancements have resulted in use of Bluetooth detectors and cell phone call detail records (CDR) to passively collect data that can be used to estimate OD demand. In this talk, a new passive data source - Google’s Aggregated and Anonymized Trips (AAT) - will be presented. Google’s Better Cities program, which seeks to minimize traffic congestion, speed up journeys, improve safety, and reduce the amount of money spent on infrastructure, has partnered with the San Francisco County Transportation Authority (SFCTA) to make available aggregated and anonymized OD flow information from location reports. This dataset is derived from users who have chosen to store their location information from Google-enabled devices on Google servers. The aggregate data are anonymized by applying differential privacy algorithm. Since the dataset accounts for only a sample of travelers, Google provides relative trips rather than absolute counts.
Aggregate flow data (in terms of relative trips) to and from approximately 90 districts covering the 9-county San Francisco Bay Area, in 1-hour increments were provided to SFCTA to support the San Francisco Freeway Corridor Management Study (FCMS). The primary objective of this study is to assess strategies for improving the performance of the US-101 and I-280 corridor, which connects San Francisco and Silicon Valley. Google also provided origin-destination flows for four primary freeway segments on the US-101 and I-280 corridor in San Francisco for these same districts and time periods. A total of six months of daily and hourly data were provided, covering three months each from spring and fall 2015.
The talk will focus on the methods used and models estimated to convert the weights provided in AAT dataset to actual person trips resulting in hourly OD matrices. In addition, a comparison of the estimated OD demand to that indicated by regional travel demand model and household travel survey will be presented.