Abhishek Damera, Data Scientist & Dr. Mike Flaxman, Spatial Data Science Practice Lead, OmniSci
Human mobility patterns have been studied for millennia and are of obvious interest to transportation infrastructure and retail planners among others. It is only in the last few years that consumer GPS and cell phones have allowed observations of large samples over wide areas and long periods of time. Yet the mechanisms used by these services vary widely and should be expect by default to contain demographic sampling biases. We have conflated these GPS-derived spatial event stream data sources with traditional census block group data in order to quantify and calibrate them. We have used these adjustments to produce the first known map of US populations at an hourly basis before and during the COVID-19 epidemic. In order to protect individual privacy, we have deployed an explicit aggregation system using variable-sized hexagonal bins. Nonetheless, our maps represent much higher resolution data than conventionally available, ranging from 50m in dense urban areas to 1km in non density rural areas. Because sampling along roads is much higher than elsewhere, we are able to provide traffic estimates at high spatial precision nationally. In addition to overall population density, we have used overnight device dwell times in residential areas to infer the hourly demographic characteristics of areas based on device trajectories.