Earthquakes, Hurricanes, and Mobile Communication Patterns in the New York Metro Area: Collective Behavior during Extreme Events

We use wireless voice-call and text-message volumes to quantify spatiotemporal communication patterns in the New York Metro area before, during, and after the Virginia earthquake and Hurricane Irene in 2011. The earthquake produces an instantaneous and pervasive increase in volume and a ~90-minute temporal disruption to both call and text volume patterns, but call volume anomalies are much larger. The magnitude of call volume anomaly diminishes with distance from earthquake epicenter, with multiple clusters of high response in Manhattan. The hurricane produces a two-day, spatially varying disruption to normal call and text volume patterns. In most coastal areas, call volumes dropped anomalously in the afternoon before the hurricane’s arrival, but text volumes showed a much less consistent pattern. These spatial patterns suggest partial, but not full, compliance with evacuation orders for low-lying areas. By helping us understand how people behave in actual emergencies, wireless data patterns may assist network operators and emergency planners who want to provide the best possible services to the community. We have been careful to preserve privacy throughout this work by using only anonymous and aggregate data.

💡 Research Summary

This paper investigates how collective human behavior during extreme natural disasters can be inferred from aggregated mobile voice‑call and text‑message volumes. The authors focus on two high‑impact events that occurred in 2011: the magnitude‑5.8 Virginia earthquake on August 23 and Hurricane Irene, which threatened the New York metropolitan area on August 27‑28. Using anonymized, cell‑tower‑level data supplied by a major carrier, the study constructs a baseline of normal communication activity by averaging the preceding 30 days of 5‑minute interval call and SMS counts for each spatial unit. The baseline is transformed into z‑scores to capture deviations that are statistically significant while accounting for diurnal and weekly cycles.

For the earthquake, the analysis reveals an instantaneous surge in call volume that reaches roughly 150 % above the baseline within the first five minutes of the event and persists for about 90 minutes. Text volume also rises but peaks at only about 60 % of the call surge. The magnitude of the call anomaly decays with increasing distance from the epicenter, yet distinct high‑response clusters appear in Manhattan, northern Brooklyn, and eastern Queens—areas characterized by high population density and concentrated business activity. This spatial pattern suggests that voice calls serve as the primary channel for real‑time emergency reporting and coordination, especially in densely populated zones where information spreads rapidly.

In contrast, Hurricane Irene produces a more prolonged, spatially heterogeneous disruption. In the two days leading up to landfall, coastal neighborhoods (e.g., Brooklyn’s waterfront, southern Queens, and Long Island) exhibit a noticeable pre‑storm dip in call volume—approximately 20‑30 % below baseline—indicating partial compliance with evacuation orders. Inland areas show little deviation. Text‑message patterns are less consistent; some coastal cells display a rise in SMS volume even as calls fall, reflecting the asynchronous nature of texting for emotional expression, status updates, or non‑urgent coordination during evacuation.

Methodologically, the study emphasizes privacy protection by working exclusively with aggregated counts and by removing any personally identifiable information before analysis. The authors employ both regression techniques to test distance‑decay effects and clustering algorithms (K‑means, DBSCAN) to identify spatial hotspots of anomalous activity.

The findings have practical implications for both network operators and emergency managers. Real‑time detection of call‑volume spikes can trigger dynamic capacity allocation, priority routing, or targeted emergency alerts, thereby preserving service quality when demand surges. Conversely, observed call‑volume drops in evacuation zones can serve as a proxy for compliance monitoring, enabling authorities to allocate resources more efficiently. The divergent behavior of voice and text channels underscores the need for multimodal communication strategies in disaster preparedness plans.

Overall, the paper demonstrates that passive, large‑scale mobile communication data provide a valuable, near‑real‑time lens on population‑level responses to sudden shocks. By quantifying how call and text volumes deviate from expected patterns, researchers and practitioners can better understand evacuation dynamics, information diffusion, and the temporal evolution of collective stress. The authors argue that integrating such data streams into smart‑city and disaster‑response platforms could enhance situational awareness, improve resource distribution, and ultimately save lives during future extreme events.