Authors: Christa Brelsford*, Oak Ridge National Laboratory, Kevin Sparks, Oak Ridge National Laboratory
Topics: Urban Geography, Spatial Analysis & Modeling, United States
Keywords: Mobility Data, Distribution, Neighborhoods, Cities
Session Type: Virtual Paper
Start / End Time: 1:30 PM / 2:45 PM
Room: Virtual 9
Presentation File: No File Uploaded
Even with an excellent, high resolution measure of resident population, estimating the population density of visitors to a building, outdoor space, neighborhood or region is a vexing empirical challenge; and one that is of substantial practical importance. The number of people in a location can vary dramatically over the course of a day, week or year, which in turn influences risk assessment.
For the first time, data on human mobility patterns is becoming widely available through data collected from mobile devices. We use this data to explore visits to businesses in the US by hour of the week, capturing the full distribution of visit patterns. These distributions can be estimated under a number of different aggregation strategies: business type, various geographic or temporal scales, and under exceptional circumstances.
Measures of the central tendency of a distribution give an estimate of the most likely outcomes, but worst-case scenarios are captured primarily by the tail of the visitor density distribution. We expect that the probability of a place experiencing any given population density is asymmetrical with a strong left skew and bounded on both sides. Previous research on bounded distributions has been shown to result in interesting properties for the heterogeneity of outcomes in urban neighborhoods. If we can describe general patterns in these empirical distributions of localized population density, we may develop a richer understanding of fine temporal resolution population dynamics across the world, and develop a better understanding of human interaction patterns overall.