Abstract
While human mobility research often focuses on movement within national borders, timely data on international flows is critical for managing global crises like pandemics or migration. This study proposes using roaming data from mobile phones to monitor international mobility in near real-time. We introduce the COVID Gravity Model (CGM), an extension of traditional gravity models that incorporates government policy data (non-pharmaceutical interventions) to account for pandemic-era travel restrictions. Our results show that CGM improves prediction accuracy by up to 126.9% for incoming mobility and 63.9% for outgoing mobility compared to classical models.
The Challenge of Tracking International Flows
Most mobility research relies on intra-country data. Tracking international movement is significantly more difficult because official statistics, such as air passenger records, are often delayed by months. Social media data, while faster, often suffers from significant demographic biases. Furthermore, traditional mobility models—like the Gravity Model, which assumes travel is a function of population and distance—fail during a global health crisis. In a pandemic, physical distance matters less than political barriers; a short trip is impossible if a border is closed.
Data: Roaming Traces as a Real-Time Proxy
To bridge this data gap, researchers utilized anonymized, aggregated roaming traces from a major UK mobile network operator (MNO) with a 28% market share. This dataset tracks "incoming" mobility (foreign SIMs in the UK) and "outgoing" mobility (UK SIMs abroad).
As shown in Figure 1, roaming data provides high spatial resolution, identifying specific corridors of movement, such as the high volume of commuters from Poland to the UK. To validate this proxy, the researchers compared roaming trends against official UK Home Office air traffic records, finding a remarkably high correlation of 0.926. Unlike official statistics, roaming data can be processed with as little as a one-day delay.
Methodology: From Physics to Policy-Aware Models
The study compared three primary modeling approaches:
- The Gravity Model: Mimics Newton’s law of gravitation, assuming travel increases with population and decreases with distance.
- The Radiation Model: Assumes individuals travel to the nearest location offering a specific opportunity (e.g., jobs).
- COVID Gravity Model (CGM): The authors' novel extension. Using negative binomial regression, the CGM incorporates the Oxford Stringency Index (SI)—a metric of government restriction severity—for both origin and destination countries. This allows the model to "learn" how specific policies, like school closures or travel bans, suppress natural mobility flows.
Key Results and Performance Gains
The introduction of the Stringency Index transformed the model from a static physical simulation into a dynamic, policy-aware framework. While the accuracy of traditional models plummeted by approximately 10% as soon as lockdowns began in March 2020, the CGM remained robust.
As illustrated in Figure 5, the model successfully captures the dramatic collapse of international travel in mid-March. Quantitatively, the CGM outperformed traditional models by 126.9% for incoming mobility and 63.9% for outgoing mobility. Using the Common Part of Commuters (CPC) index, the CGM achieved high fidelity scores of 0.78 for incoming and 0.83 for outgoing flows.
Limitations and Future Outlook
Despite its success, roaming data has limitations. it may miss travelers who do not own mobile phones, those under 18, or individuals who purchase local SIM cards to avoid roaming charges. Additionally, the data is proprietary, which can limit accessibility for the broader scientific community. Future work aims to expand this model beyond the UK-centric view to evaluate sub-national or global mobility patterns, providing a vital tool for future pandemic preparedness and migration monitoring.
Summary updated