Estimating the presence of people in buildings using Call Detail Records

Abstract

As geographic data about individual movement become increasingly available, they open LIP the possibility of understanding and modeling urban mobility patterns. While no all-encompassing dataset regarding mobility is available, this study explores how Call Detail Records (CDRs), a highly ubiquitous dataset, can be leveraged to create models that can reproduce mobility patterns observed from time consuming, capital-intensive and infrequent travel surveys. While mechanisms have been proposed for reproducing particular characteristics of individual mobility, this is the first attempt to generate all mobility patterns at fine spatial and temporal scales at the level of individual buildings. Two shortcomings of any dataset include spatial uncertainty at very high resolution and the presence of high-fidelity traces for only a fraction of the population. While the proposed model addressed the former to some extent by providing high accuracy counts at the level of census tracts, a separate method has been explored to address this along with the latter phenomenon. To achieve this, the study leverages hyper-local datasets such as building footprints and places of interest. In the absence of primary datasets, the study is able to provide a model to estimate of the presence of people at the level of individual buildings. Hence, this study provides a pipeline to proceed from high fidelity location traces from a fraction of the population to building level occupancy profiles using fairly ubiquitous data sources.