Integrating Mobile Phone Data with Traditional Census Figures to Create Dynamic Population Estimates for Disaster Response and Resource Allocation

Abstract

Accurate and up-to-date information on where people are and in what numbers is essential for disaster management and the fair distribution of public goods and services. However, official population counts are typically only available from censuses, which are often outdated and lack temporal and spatial resolution. Here we develop and validate a scalable approach that combines anonymized, aggregated mobile phone data (call detail records and passive network events) and census baselines to produce up-to-date, error-quantified population estimates at sub-district spatial resolutions and hourly to daily temporal frequencies. We normalize mobile activity to resident population with a Bayesian hierarchical model that accounts for carrier market share, multi-SIM ownership, inactive devices, and diurnal/weekly seasonality effects. We allocate the normalized activity across space with dasymetric mapping using ancillary layers (land use, building footprints, night-time lights, road networks) to restrict to likely inhabited areas and disentangle through-mobility from residential population. A state-space data assimilation framework fuses multiple activity signals (tower handovers, app pings, emergency alert traffic) to provide near–real-time population updates in the presence of hazard-induced mobility changes.
We assess performance against ground truth from household surveys, administrative registries, smart-meter rollups, and post-event field counts, reporting accuracy as mean absolute percentage error, coverage probabilities for credible intervals, and resilience to simulated network outages. We illustrate applications for storm evacuation and flood response, showing how the estimates can inform the pre-positioning of supplies, dynamic routing of ambulances, siting of mobile clinics and shelters, and prioritization of damage assessment teams. The framework incorporates privacy-by-design measures (k-anonymity thresholds, differential privacy noise, strict governance and retention policies) and generates confidence surfaces to enable risk-aware decision-making.
Our method effectively transforms static census baselines into living population surfaces, helping responders and planners see where people are—and how they move—before, during, and after a disaster. Beyond the emergency context, the approach can support the routine distribution of health, education, and transportation resources, offering a practical, ethically sound pathway to data-driven public service delivery in rapidly urbanizing and peri-urban environments. This integrated methodology advances state-of-the-art approaches by delivering higher accuracy in complex urban contexts and enabling more impactful healthcare planning during post-disaster recovery periods.