Lifestyle pattern analysis unveils recovery trajectories of communities impacted by disasters

Disasters disrupt community dynamics by impacting population lifestyles, encompassing both essential (e.g., grocery shopping) and non-essential activities (e.g., shopping at malls). Disruption of essential lifestyles negatively impacts populations' health and well-being, while the restoration of non-essential services signals a return to normalcy. Recovery, a vital component of disaster management, is a multifaceted and nonlinear process involving the return of formal and informal systems to a normal state. Existing recovery frameworks focus on infrastructure, social, economic, and environmental factors, but few have examined lifestyle activities at a granular level. This study introduces the concept of lifestyle recovery as a quantifiable and dynamic aspect of recovery, proposing the use of location intelligence data to quantify lifestyle patterns. Location intelligence data offers advantages over traditional datasets by providing greater spatial granularity, larger sample sizes, and shorter data collection lags. This research leverages location intelligence data, specifically POI data and human mobility data, to examine lifestyle activities at the Census Block Group (CBG) level, focusing on Hurricane Harvey in Harris County, Texas, from August to November 2017. The study aims to assess the extent of disaster impact on lifestyle activities, identify patterns of recovery trajectories, and examine variations in recovery based on flooding and demographic attributes.

Existing literature highlights the multidimensional and nonlinear nature of disaster recovery, emphasizing the need for quantitative measures and spatiotemporal perspectives. Several recovery frameworks have been developed, focusing on ecological, social, economic, infrastructure, and community competence factors. Prior research has examined recovery through population returns, infrastructure functionality, housing damage, and neighborhood poverty rates, but few have analyzed lifestyle activities at the granular scale of CBGs. Location intelligence data has been used to study disaster impacts, damage reporting, evacuation patterns, and rapid impact assessments. Studies have shown distinctive human mobility patterns during stable and disruptive periods, and differential recovery rates for Points of Interest (POIs). However, little work has leveraged location intelligence data for characterizing community recovery based on lifestyle patterns. This study addresses the knowledge gap in specifying, characterizing, and quantifying recovery milestones by focusing on the dynamic nature of lifestyles and their impact on community recovery trajectories.

This study analyzed location intelligence data from Spectus and SafeGraph, combined with building footprint data and FEMA flood data, to determine daily visitations from each home CBG to different POIs in Harris County, Texas during and after Hurricane Harvey (August-November 2017). Data processing involved creating a 'home_CBG to stop table' associating anonymous devices with home CBGs, a 'POI polygon table' specifying polygons for each POI using SafeGraph and building footprint data, and a 'home_CBG to POI table' classifying POIs into essential and non-essential categories based on NAICS codes and FEMA definitions. A baseline was established using the first two weeks of August 2017. CBG recovery was defined as when visits reached 90% of the baseline value. Two-step k-means clustering was employed: primary clustering based on baseline POI visitation patterns to identify lifestyle clusters, and secondary clustering within each primary cluster based on recovery features (maximum disruption point and recovery duration for essential and non-essential POIs). The elbow method determined the optimal number of clusters. ANOVA testing was used to assess statistical significance of differences between lifestyle clusters and recovery trajectories. Demographic data from the US Census were integrated to analyze variations in recovery based on flooding and demographic attributes. A sensitivity analysis of recovery thresholds (90%, 80%, 70%) was also performed.

Primary k-means clustering identified four distinct lifestyle patterns based on POI visitation frequencies. ANOVA testing confirmed statistically significant differences (p < 0.05) in the relative frequency of essential and non-essential lifestyles across these clusters. Each cluster exhibited unique dependencies on various POIs, indicating diverse community characteristics. Secondary k-means clustering within each primary cluster revealed four recovery trajectories: immediate recovery duration (IRD), short recovery duration (SRD), moderate recovery duration (MRD), and extreme recovery duration (ERD). ANOVA testing showed statistically significant differences (p < 0.05) in recovery trajectories within each primary cluster, indicating disproportionate recovery rates even within similar lifestyles. Analysis of demographic data revealed some variations in recovery related to income and minority population percentages, but no statistically significant overall patterns emerged. However, exploratory analysis showed potential links between extreme recovery durations, lower incomes, and higher minority populations in some clusters. Crucially, 59% of CBGs with extreme recovery durations had less than 1% direct flooding impact, highlighting the spatial spillover effects of flooding on lifestyle disruptions. The study found that approximately 89% of non-flooded home CBGs still experienced at least one week of lifestyle disruption.

The findings demonstrate the value of quantifying lifestyle recovery as a critical milestone in community recovery. The novel methodology, integrating multiple data sources and using k-means clustering, effectively characterizes distinct lifestyle patterns and quantifies the extent and speed of recovery after Hurricane Harvey. The identification of four distinct recovery trajectories highlights the variability in community resilience and the need for targeted interventions. The observation that flooding impacts extend beyond directly flooded areas emphasizes the importance of considering spatial spillover effects and the broader community network. While the study didn't find statistically significant relationships between all demographic factors and recovery trajectories, the exploratory analysis suggests potential social disparities warranting further investigation. The results provide data-driven insights for public officials and emergency managers to improve resource allocation and recovery strategies.

This study provides a novel approach to quantifying community recovery using location intelligence data to analyze population lifestyle patterns. The results emphasize the importance of understanding community dependencies on essential and non-essential services, the existence of differential recovery rates within similar lifestyles, and the spatial reach of disaster impacts beyond directly affected areas. Future research could explore the role of trip distance to POIs, use supervised learning methods to predict recovery based on demographic factors and hazard attributes, and improve the analysis with more comprehensive data. These advancements would enhance our understanding of social, physical, and environmental disparities in lifestyle recovery.

The study's analysis is limited to the period after Hurricane Harvey's landfall. A longer time frame would provide more robust insights into long-term recovery. The baseline period is limited to two weeks before the hurricane which may not fully capture seasonal variation in lifestyle patterns. Future studies can mitigate these limitations by including data from earlier time periods and using alternative baselining approaches. Additionally, this research considers flooding at the household level rather than at the facility level which may have contributed to the disproportionate flooding impacts. Future studies can incorporate data on facility level disruptions to better evaluate the implications.