Anomalies of dwellers' collective geotagged behaviors in response to rainstorms: a case study of eight cities in China using smartphone location data

Jiawei Yi, Yunyan Du, Fuyuan Liang, Tao Pei, Ting Ma, Chenghu ZhouNatural 2019 · DOI

Abstract

Understanding city residents' collective geotagged behaviors (CGTBs) in response to hazards and emergency events is important in disaster mitigation and emergency response. It is a challenge, if not impossible, to directly observe CGTBs during a real-time matter. This study used the number of location requests (NLR) data generated by smartphone users for a variety of purposes such as map navigation, car hailing, and food delivery to infer the dynamics of CGTBs in response to rainstorms in eight Chinese cities. We examined rainstorms, flooding, and NLR anomalies, as well as the associations among them, in eight selected cities across mainland China. The time series NLR clearly reflects cities' general diurnal rhythm, and the total NLR is moderately correlated with the total city population. Anomalies of the NLR were identified at both the city and grid scale using the Seasonal Hybrid Extreme Studentized Deviate (S-H-ESD) method. Analysis results demonstrated that the NLR anomalies at the city and grid levels are well associated with rainstorms, indicating that city residents request more location-based services (e.g., map navigation, car hailing, food delivery, etc.) when there is a rainstorm. However, the sensitivity of the city residents' collective geotagged behaviors in response to rainstorms varies in different cities as shown by different peak rainfall intensity thresholds. Significant high peak rainfall intensity tends to trigger city flooding, which leads to increased location-based requests as shown by positive anomalies in the time series NLR.

Introduction

While sophisticated tools exist to measure meteorological phenomena like rainfall, quantifying the "human element" of a disaster in real-time remains a significant challenge. Traditional methods, such as post-disaster surveys or insurance claims, are often slow, expensive, and prone to recall bias. This study addresses the challenge of quantifying Collective Geotagged Behaviors (CGTBs)—the aggregate digital footprints left by urban dwellers as they navigate their daily lives. By analyzing the "digital exhaust" of a city—specifically the pings sent to servers for map navigation, ride-hailing, or food delivery—researchers can observe how a city’s rhythm breaks during extreme weather events. Understanding this "coping capacity" is vital for urban planners and emergency responders to identify vulnerable neighborhoods and assess recovery speeds.

Methodology and Data

The research framework utilizes the Number of Location Requests (NLR) as a proxy for human activity. To compare diverse urban environments, from megacities like Harbin to smaller cities like Zhuhai, the researchers employed median-interquartile normalization.

Map of the eight study cities in China showing precipitation levels and urban footprints.

The core analysis relies on the Seasonal Hybrid Extreme Studentized Deviate (S-H-ESD) algorithm. This method decomposes time-series data into trend, seasonality (the city's daily "heartbeat"), and residuals. Anomalies are identified where the residual significantly deviates from the expected daily rhythm. A composite anomaly score was then calculated based on the frequency, magnitude, and spatial density of these deviations.

The study utilized three primary datasets:

  • Tencent Big Data: Hourly NLR records from August 2017, representing nearly 1.1 billion monthly active users across apps like WeChat and DiDi.
  • GPM IMERG: 30-minute rainfall data at a 0.1° resolution.
  • Ground Truth: Verification of flooding events through mainstream Chinese media reports.

Results and Discussion

The City’s Heartbeat and Rain as a Catalyst

Every city exhibits a consistent diurnal rhythm, typically peaking at lunch (12:00 LT) and after work (20:00 LT). Interestingly, rainstorms often act as a catalyst for positive anomalies. Rather than activity ceasing, behavior shifts: residents transition from walking to hailing cars and from dining out to ordering delivery. The ratio of positive anomalies ($R_{pos}$) was found to jump from 0.12 during dry periods to 0.70 during violent rainstorms.

Regional Sensitivity and Thresholds

The behavioral response to rain is not uniform across cities. Coastal cities like Haikou, accustomed to typhoons, require much higher rainfall intensities to trigger behavioral shifts compared to inland cities like Xiangyang, suggesting varying infrastructural or social "coping capacities."

Peak rainfall intensity versus duration for the eight study cities, illustrating varying behavioral sensitivity thresholds.

Predictive Timing and Recovery

Approximately 31% of maximum anomaly scores were detected before rainfall reached peak intensity, suggesting that residents react to forecasts or the initial onset of a storm. While most cities recovered quickly, severe events like Category 3 Typhoon Hato in Zhuhai caused disruptions lasting over 72 hours, highlighting the impact of infrastructure damage on digital rhythms.

Conclusion

This study demonstrates that massive smartphone location data can serve as a "pulse check" for urban resilience. While the data naturally skews toward younger, tech-savvy populations and currently relies on a single platform (Tencent), it provides a powerful new metric for real-time disaster response. Future work will aim to integrate multi-source geospatial data to better isolate weather-driven behaviors from other social events, such as the "back-to-school" movements observed in late August.

Summary updated

Abstract

Understanding city residents' collective geotagged behaviors (CGTBs) in response to hazards and emergency events is important in disaster mitigation and emergency response. It is a challenge, if not impossible, to directly observe CGTBs during a real-time matter. This study used the number of location requests (NLR) data generated by smartphone users for a variety of purposes such as map navigation, car hailing, and food delivery to infer the dynamics of CGTBs in response to rainstorms in eight Chinese cities. We examined rainstorms, flooding, and NLR anomalies, as well as the associations among them, in eight selected cities across mainland China. The time series NLR clearly reflects cities' general diurnal rhythm, and the total NLR is moderately correlated with the total city population. Anomalies of the NLR were identified at both the city and grid scale using the Seasonal Hybrid Extreme Studentized Deviate (S-H-ESD) method. Analysis results demonstrated that the NLR anomalies at the city and grid levels are well associated with rainstorms, indicating that city residents request more location-based services (e.g., map navigation, car hailing, food delivery, etc.) when there is a rainstorm. However, the sensitivity of the city residents' collective geotagged behaviors in response to rainstorms varies in different cities as shown by different peak rainfall intensity thresholds. Significant high peak rainfall intensity tends to trigger city flooding, which leads to increased location-based requests as shown by positive anomalies in the time series NLR.
BibTeX
@inproceedings{yi2019anomalies,
  author = {Jiawei Yi and Yunyan Du and Fuyuan Liang and Tao Pei and Ting Ma and Chenghu Zhou},
  title = {{Anomalies of dwellers' collective geotagged behaviors in response to rainstorms: a case study of eight cities in China using smartphone location data}},
  booktitle = {Natural},
  year = {2019},
  doi = {10.5194/nhess-19-2169-2019},
}
Figure