Weekly Mobility Model Methodology (Trends)

Replica's mobility weekly data release provides near-real-time insight into mobility patterns for each census tract in the country. The output tables include O-D pairs by mode and residential VMT. The data is produced from the start of 2019 and onwards, enabling year-over-year comparisons.

In order to produce weekly mobility outputs, Replica runs a modified version of its seasonal mobility (Places) pipeline on a daily basis. The weekly pipeline leverages the same synthetic population and travel personas as described in the seasonal mobility model (Places) methodology. However, there are three primary differences in the data processing pipeline:

  • First, because the pipeline is run for the entire country on a daily basis, practical limitations in computing power mean that trip mode is modeled with a modified router that produces an origin and destination pair for each trip at the census tract level in lieu of network link or transit line-level detail.
  • Second, because the weekly delivery cadence makes it unrealistic to receive and process authoritative ground-truth data from customers for calibration, weekly mobility outputs are calibrated internally from different sources with national coverage (e.g., comparing mobile location data to in-vehicle GPS data). We also compare our weekly data to previously released, calibrated seasonal model outputs.
  • Third, our weekly mobility data does not currently include visitor or freight trips. We plan to add these trip estimates in a future update.

In effect, these differences in the pipeline reflect a trade-off of granularity and detail for speed of delivery, in order to maintain consistent quality. To that end, Replica's weekly data is best used for monitoring changes over time, and Replica's seasonal data is best used for in-depth point-in-time analysis that may benefit from the ability to do cross-tabular filtering and individual trip analysis.