Digital Urban Sensing: A Multi-layered Approach

No single dataset captures a whole city — so this one asks what happens when you stack six together.

Zhu E, Khan M, Kats P, Bamne S, Sobolevsky S · arXiv preprint, 2018 · Read the paper →

The widest-angle paper of the group, and a useful check on the others. If a neighborhood's signature depends on Twitter, or on 311, how much does the choice of data decide what you see? We lined up six layers of New York — taxi trips, subway, Citi Bike, a mobile-app feed, Twitter, and 311 — and compared the spatial and temporal patterns each one tells on its own.

They disagree more than you'd hope: no single source sees the whole city, and each over- or under-counts particular people and places. Combined into a multi-layered model, though, they sharpen both the zoning and the socioeconomic estimates. It stayed a preprint. My part was on the data and analysis side.