[Paper Review] Unraveling the Temporal Importance of Community-scale Human Activity Features for Rapid Assessment of Flood Impacts

Faxi Yuan, Yang Yang+2 more|arXiv (Cornell University)|Jun 15, 2021

Disaster Management and Resilience4 citations

TL;DR

This study leverages community-scale big data—such as mobility, credit card transactions, and social media activity—to rapidly assess flood impacts during Hurricane Harvey in Harris County, Texas. Using random forest models, it identifies that fluctuations in human activity indices and road congestion are the most temporally important indicators for early flood impact estimation, significantly outperforming social media data.

ABSTRACT

The objective of this research is to explore the temporal importance of community-scale human activity features for rapid assessment of flood impacts. Ultimate flood impact data, such as flood inundation maps and insurance claims, becomes available only weeks and months after the floods have receded. Crisis response managers, however, need near-real-time data to prioritize emergency response. This time lag creates a need for rapid flood impact assessment. Some recent studies have shown promising results for using human activity fluctuations as indicators of flood impacts. Existing studies, however, used mainly a single community-scale activity feature for the estimation of flood impacts and have not investigated their temporal importance for indicating flood impacts. Hence, in this study, we examined the importance of heterogeneous human activity features in different flood event stages. Using four community-scale big data categories we derived ten features related to the variations in human activity and evaluated their temporal importance for rapid assessment of flood impacts. Using multiple random forest models, we examined the temporal importance of each feature in indicating the extent of flood impacts in the context of the 2017 Hurricane Harvey in Harris County, Texas. Our findings reveal that 1) fluctuations in human activity index and percentage of congested roads are the most important indicators for rapid flood impact assessment during response and recovery stages; 2) variations in credit card transactions assumed a middle ranking; and 3) patterns of geolocated social media posts (Twitter) were of low importance across flood stages. The results of this research could rapidly forge a multi-tool enabling crisis managers to identify hotspots with severe flood impacts at various stages then to plan and prioritize effective response strategies.

Motivation & Objective

• To address the critical time lag in obtaining definitive flood impact data (e.g., insurance claims, inundation maps), which hinders timely emergency response.
• To investigate the temporal importance of diverse community-scale human activity features in indicating flood impacts across response and recovery stages.
• To evaluate the relative predictive power of features derived from mobility, financial transactions, and online communications for rapid flood impact assessment.
• To support crisis managers with data-driven, real-time indicators to prioritize resource allocation during disasters.

Proposed method

• Utilized four categories of community-scale big data: mobility (traffic congestion), financial activity (credit card transactions), online communications (geolocated Twitter posts), and human activity indices derived from aggregated signals.
• Extracted ten temporal features reflecting daily variations in human activity, including average daily activity index, percentage of congested roads, transaction volume, and total spending.
• Applied multiple random forest models to assess feature importance in predicting flood impacts, using both flood insurance claims and flood inundation maps as ground truth.
• Conducted three-class classification of flood impacts (low, medium, high) across response and recovery stages to evaluate temporal stability of feature importance rankings.
• Employed feature importance functions within random forest models to rank the predictive contribution of each feature to flood impact estimation.
• Validated results across two impact metrics (insurance claims and inundation) and two flood stages to ensure robustness and consistency.

Experimental results

Research questions

• RQ1 Which community-scale human activity features are most temporally important for rapid flood impact assessment during the response and recovery stages?
• RQ2 How do the relative importance rankings of features derived from mobility, financial transactions, and online communications compare across flood stages?
• RQ3 To what extent do geolocated social media posts contribute to early flood impact detection compared to other human activity indicators?
• RQ4 Can fluctuations in human activity indices and traffic congestion serve as reliable proxies for flood damage before official data becomes available?

Key findings

• Fluctuations in the average daily human activity index (FE 1) and the daily maximum percentage of congested roads (FE 2) were the top two most important features for indicating flood impacts in both response and recovery stages.
• The daily average percentage of congested roads (FE 3) ranked third in importance, reinforcing the significance of mobility disruptions as early indicators of flood severity.
• Variations in credit card transactions—measured by number of cards, number of transactions, and total spending—ranked in the middle of the importance scale, indicating moderate predictive power.
• Geolocated social media posts (Twitter) contributed the least to impact prediction, with all four derived features (FE 7–10) consistently ranking lowest in importance across both stages.
• The feature importance rankings remained stable across both flood impact metrics (insurance claims and inundation), indicating robustness of the findings.
• In the absence of activity index or congestion data during the response stage, total spending (FE 6) emerged as a reliable alternative indicator for rapid flood impact assessment.