Behavioral response to mobile phone evacuation alerts

TL;DR

This study analyzes how people respond to mobile evacuation alerts during the February 2024 Valparaíso wildfires by leveraging anonymized mobile-network data from about $580{,}000$ devices and a quasi-experimental controlled interrupted time-series design. It introduces the Relative Evacuation Index ($REX$) to quantify displacement across warned versus non-warned towers and across three socioeconomic groups, uncovering immediate strong responses to the first alert, widespread non-targeted evacuation, and pronounced SES-based differences in both evacuation timing and recovery. The findings highlight potential alert fatigue, inequities in evacuation capability, and spillover effects that can strain transportation networks, offering policy guidance for SES-calibrated, multi-stage alert strategies to improve safety and equity in climate-driven disasters. Overall, the work demonstrates how integrating high-resolution mobility data with rigorous causal designs can inform targeted, efficient, and equitable emergency communication and evacuation planning in wildfire contexts.

Abstract

This study examines behavioral responses to mobile phone evacuation alerts during the February 2024 wildfires in Valparaíso, Chile. Using anonymized mobile network data from 580,000 devices, we analyze population movement following emergency SMS notifications. Results reveal three key patterns: (1) initial alerts trigger immediate evacuation responses with connectivity dropping by 80\% within 1.5 hours, while subsequent messages show diminishing effects; (2) substantial evacuation also occurs in non-warned areas, indicating potential transportation congestion; (3) socioeconomic disparities exist in evacuation timing, with high-income areas evacuating faster and showing less differentiation between warned and non-warned locations. Statistical modeling demonstrates socioeconomic variations in both evacuation decision rates and recovery patterns. These findings inform emergency communication strategies for climate-driven disasters, highlighting the need for targeted alerts, socioeconomically calibrated messaging, and staged evacuation procedures to enhance public safety during crises.

Figures (14)

• Figure 1: Mobile phone tower connectivity patterns and location during a wildfire evacuation event.A. General geographic area of the fires (in red) along with the warned towers (green dots). B. Time series comparing average mobile tower connections during baseline days (blue) and fire week (orange) in February 2024, with 95% confidence intervals (shaded areas). Vertical lines indicate critical emergency communication times: first evacuation alert (red, February 2, 16:45) and first alert of day two (gray, February 3, 10:45). The data reveals distinct behavioral patterns following the initial evacuation alert, including an immediate spike in connectivity followed by a rapid decrease, suggesting population displacement.
• Figure 2: Snapshots of the Cumulative distribution functions (CDFs) of tower-level evacuation rates with respect to the baseline behavior, stratified by socioeconomic status. The figure presents six time-sliced CDFs capturing activity patterns before and after the intervention across low (blue), medium (orange), and high (green) socioeconomic areas. The top row of the panel shows the pre-intervention behavior at 90 minutes, 1 hour, and 30 minutes, respectively, while the second row in the panel indicates the post-intervention behavior displaying changes at 1 hour, 75 minutes, and 90 minutes after the intervention. The x-axis represents the percentage of the evacuation rate, where the "0" rate indicates a consistent behavior with the baseline days. The negative rates indicate an evacuation behavior with respect to the baseline, while the positive rates represent more connections with respect to the baseline. The y-axis expresses the cumulative percentage of towers exhibiting evacuation rates lower or equal to the corresponding evacuation rate in time (x-axis).
• Figure 3: Cumulative distribution functions of tower evacuations over time. The y-axis indicates the percentage of towers that reached the specified evacuation rates (50%, 75%, and 85%) at a specific time (x-axis). We stratified by socioeconomic groups: low (blue line), medium (orange line), and high (green line). The dashed line marks the time at which the first evacuation alert was sent.
• Figure 4: Relative Evacuation Index for non-warned (upper panel) and warned towers (lower panel), across SEGs before and after the first evacuation alert. The blue lines represent pre-warning periods, while the orange lines indicate post-warning periods. Vertical dashed lines correspond to the times when evacuation alerts were sent. Panels are divided by SEG (low, medium, high) and whether towers were warned or non-warned. The shaded regions indicate 95% confidence intervals. The figure illustrates behavioral differences in evacuation responses and recovery patterns across SEGs and between warned and non-warned areas.
• Figure 5: Differential REX over time stratified by three socio-economic groups (Low (left), Medium (center), and High (right)) from February 1-3, 2024. We depict the pre-intervention (blue) and post-intervention (orange) time series, with their respective 95% confidence intervals, and measurements taken every 15 minutes. The dashed lines are all the evacuation alerts sent.