Foot traffic can predict COVID-19 spread in New York City neighborhoods

A new study published in the journal PLOS Computational Biology reveals how foot traffic data from mobile devices can enhance neighborhood-level COVID-19 forecasts in New York City. The research, led by researchers at Columbia University Mailman School of Public Health and Dalian University of Technology, provides a novel approach to predicting the spread of the SARS-CoV-2 virus and improving targeted public health interventions during future outbreaks.

The COVID-19 pandemic hit New York City hard, with infection rates varying dramatically across neighborhoods. While some areas experienced rapid transmission, others saw lower transmission rates and cases, largely due to differences in socioeconomic factors, human behavior, and localized interventions.

To address these inequities, the researchers developed a forecasting model that accounts for neighborhood-level mobility patterns to provide accurate predictions of disease spread. They analyzed anonymized mobile location data to track foot traffic in restaurants, retail stores, and entertainment venues across 42 neighborhoods. By integrating these movement patterns with an epidemic model, they identified where and when outbreaks are likely to occur.

“Our analysis clearly shows how routine activities like dining out or shopping became major COVID-19 transmission pathways,” explains senior author Sen Pei, Ph.D., assistant professor in the Department of Environmental Health Sciences at Columbia Mailman School. “These behavioral insights give our model significantly greater predictive power than conventional approaches.”

Precision forecasting for neighborhood COVID-19 spread

This study demonstrates how neighborhood-level COVID-19 modeling can help address health disparities by identifying hyperlocal transmission patterns. The research reveals that crowded indoor spaces—particularly restaurants and bars—played a significant role in early pandemic spread. By integrating real-time mobility data, the team developed a behavior-driven model that outperforms traditional forecasting methods in predicting cases at the community level.

Another critical component is the model’s incorporation of seasonal effects. Researchers confirmed winter’s heightened transmission risk, linking it to lower humidity levels that prolong virus survival in air. This seasonal adjustment enables more accurate short-term predictions, giving public health officials crucial lead time to prepare for infection surges.

A tool for equitable pandemic response

The behavior-driven model could empower health departments to distribute testing and clinical resources and direct public health interventions where they’re needed most, ensuring protection reaches vulnerable neighborhoods first.

By pinpointing exactly when and where transmission spikes will likely occur, the approach replaces guesswork with targeted prevention. For example, as cold weather drives people indoors, the model could identify gathering places that would require capacity restrictions.

Refining the model for future outbreaks

While the behavior-driven model has proven effective, researchers note that real-world implementation requires further refinement. A key challenge lies in ensuring consistent access to high-quality mobility and case data—a limitation faced during the pandemic’s early phases when information streams were unreliable.

The researchers are now enhancing the model to incorporate adaptive behavior change in response to infections and its feedback on disease transmission. These improvements will be especially vital for the preparedness and response to future pandemics, enabling more precise predictions of disease spread patterns.

“This model’s success with COVID-19 opens new avenues for combating future outbreaks,” explains Pei. “By mapping disease transmission at the community level, we can arm New York City—and potentially other locations, too—with information to make more informed decisions as they prepare for and respond to emerging health threats.”

The study’s first author is Renquan Zhang, Dalian University of Technology, Dalian, China. Additional authors include Qing Yao, Wan Yang, Kai Ruggeri, and Jeffrey Shaman at Columbia; and Jilei Tai at Dalian University of Technology.