Integrating Multiple Data Sources for Bayesian Nowcasting of Subnational Populations in Ukraine

Andrea Aparicio-Castro, University of Oxford
Douglas R Leasure, Leverhulme Centre for Demographic Science, Department of Sociology, University of Oxford
Edith Darin, University of Oxford

The full-scale Russian invasion of Ukraine in February 2022 has caused widespread displacement and major challenges for the remaining population due to damaged infrastructure and disrupted services. To address the need for accurate and real-time population data in this ongoing conflict, we developed a Bayesian hierarchical model that integrates multiple data sources to nowcast subnational populations and mobility patterns. Our model uses a state-space framework to account for imperfect observations, combined with a population time-series model and a mobility matrix within a Bayesian approach. By incorporating real-time data from Facebook and Instagram active users, mobile phone data, humanitarian surveys, geolocated conflict events, road networks, and remotely-sensed building damage, we estimate daily subnational population sizes across Ukraine, providing uncertainty measures. This approach is especially useful in situations where conventional data collection is not possible due to ongoing conflict. We validated the model using simulated populations to compensate for the lack of validation data before applying it to real-world data. Our method enhances understanding of population dynamics in conflict zones and offers a scalable solution for subnational population assessment in humanitarian emergencies. This work illustrates how integrating population, displacement, and geospatial data can overcome the challenges of real-time population estimation in crises.

Keywords: Spatial Demography, Geo-referenced/geo-coded data, Population, Shocks and Pandemics, Bayesian methods

See extended abstract.