Analysis of post-disaster population movement by using mobile spatial statistics
Introduction
Countries around the world have experienced a significant increase in frequency, intensity, and impact of natural disasters over the past decades. Building effective evacuation and response plans after the occurrence of natural disasters is one of the most challenging issues for many local government agencies [1]. During the early response phase, the responsive agencies need to understand the dynamic distributions of human population in impact areas for the purpose of optimal placement of shelters and delivery of relief supplies. During later recovery stage, understanding and predicting the population movement is key to the reconstruction efforts, because many economic, social, and civic functions are largely dependent on population dynamics [2]. The critical roles of population dynamics after the occurrence of disasters call for in-depth investigations of this issue to derive useful implication for post-disaster response and recovery plans.
Traditionally, population data comes from government census data, which has a lot of limitations: (1) Census data represent only the nighttime population distribution, which cannot reflect dynamic change of population; (2) Information is not available at small scale; (3) It is difficult to determine the accuracy and reliability of the data. Other studies use questionnaire survey to get information about people's evacuation behavior after disasters. These methods rely on the accurate recall of activity and time by the participants and hence are a relatively crude measure with wide margins of error. Furthermore, it might introduce a lot of burden to participants after they experienced the disasters, as it requires considerable attention and effort of participants to complete the questionnaire. Particularly, some studies have also shown that people exposure to natural disasters may suffer from metal health problems, such as anxiety disorders and psychological stress [3].
Recently, the emerging big data offers a new way to analyze and model human dynamics in space and time. Especially, the increased use of mobile phone data provides a great research opportunity for researchers to map and analyze dynamic human behaviors, communications, and movements [4]. A variety of existing studies has examined travel behavior analysis by using mobile data [[5], [6], [7], [8]]. People use smart phones and mobile devices to build up their digital life and to leave their digital footprint. These human-made digital records provide a foundation for human dynamics research. In this research, we attempt to use the mobile spatial statistics created by NTT Docomo's mobile phone network to analyze post-disaster population dynamics. By comparing with the traditional census data and questionnaire survey, there are several advantages to use the mobile spatial statistics: (1) it is possible to obtain the population data of high resolution in time and space; (2) it is instantaneously available with not interview bias; (3) it provides longitudinal data for very large number of population.
Despite the above-mentioned advantages, there are some challenges in utilizing mobile spatial statistics to understand population dynamics. As the mobile spatial statistics are aggregate data of population in a target grid cell, the factors that influence the change of population are unclear. Therefore, the purpose of the study is twofold: (1) to propose the appropriate method to extract behavior patterns from the aggregated data; (2) to examine the suitability of using mobile spatial statistics to estimate the dynamic change of human population. The first objective falls in a class of latent variable analysis (LVA) problem that has been studied for decades [[9], [10], [11], [12]]. LVA is a problem where underlying components are separated from the observed data without (or limited) information on the sources and the mixing process. In our study, only aggregate population in each grid cell is available, while we try to recover latent components, e.g., evacuation behavior, rescue behavior, and so forth for a better understanding of how human behavior patterns decline and recover after the occurrence of a disaster. For the purpose of this study, we use four different Latent Variable Analysis (LVA) methods - Independent Component Analysis (including FastICA and Spatial colored ICA), Non-negative Matrix Factorization (LS-NMF), and Sparse Principal Component Analysis (SPCA) to analyze mobile statistics data of the disaster-affected area. The results indicate that each LVA methods has its pros and cons in extracting behavior patterns from the aggregated population. We conclude that, using multiple LVA methods and finding out the common patterns would be a robust way to understand and explain population dynamics. The results show that a number of common essential patterns can be extracted from different LVA methods. Finally, we argue that using mobile spatial statistics to estimate the dynamic change of human population would be a feasible and practical option, essentially because there would be no alternative way to do the same task without adding burdens to affected people.
The rest of this paper is organized as follows. Section 2 gives a brief review of relevant existing studies. Section 3 describes the research area and data that is used in this study. Methodology and the results of data analysis are discussed in Section 4 Methodology, 5 Estimation results and discussion. Finally, conclusions are summarized along with a discussion about important future research issues in the last section.
Section snippets
Review
In designing evacuation and response plans after the occurrence of natural disasters, it is critical for the responsive agencies to consider the dynamic change of human population within impact areas when designing traffic assignment plans, evacuation procedures, and shelter locations [1]. In order to understand population dynamic after the disasters, a number of studies focused on analysis of people's evacuation behavior. Findings from various disasters in the past suggest that the decision to
Study area and data
On August 20, 2014, a series of landslides and debris flows occurred around Hiroshima City, Japan due to the torrential rainfall. According to the Japan Meteorological Agency, a total of about 240 mm of rain fell on the city within 24 h which is equivalent of a month worth of rain. Debris flows took place at 107 valleys and landslides occurred at 59 locations between 3am and 4am. 74 people have been confirmed dead as a result of the disaster. More than 168,000 people living in the affected area
Methodology
Signals and datasets collected in scientific, engineering, medical, and social applications are often combinations of underlying data (so called latent variables). In the field of signal processing, different methods have been developed to extract latent variables from observed signals. The broad topic of separating mixed sources is referred to as Latent Variable Analysis (LVA), which is commonly used to restore a set of unknown latent variables from a set of observed signals which are mixtures
Estimation results and discussion
First, Independent Component Analysis (ICA) is adopted to analyze mobile statistics data of area that has been affected by landslides. Here, we have mobile statistics data of 24 zones. For each zone, the data is further divided into resident within the target area and non-resident from other area. Let matrix denote the observed population (here, is a row vector representing the observed population of resident in the given grid cell i along time
Conclusion
Understanding and predicting post-disaster human movements is critical for evaluating a population's vulnerability and resilience and developing plans for disaster evacuation, response and recovery. However, our knowledge of human movements after natural disasters is limited due to both a lack of empirical data and the low precision of available data. In this study, we attempt to analyze population movement by using mobile spatial statistics. The research findings are summarized as below:
In
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
References (44)
- et al.
An integrated evacuation decision support system framework with social perception analysis and dynamic population estimation
International Journal of Disaster Risk Reduction
(2017) - et al.
Understanding individual mobility patterns from urban sensing data: a mobile phone trace example
Transport. Res. C Emerg. Technol.
(2013) - et al.
The promises of big data and small data for travel behavior (aka human mobility) analysis
Transport. Res. C Emerg. Technol.
(2016) - et al.
Blind separation of sources, part I: an adaptive algorithm based on neuromimetic architecture
Signal Process.
(1991) - et al.
Predictive models for post disaster shelter needs assessment
International Journal of Disaster Risk Reduction
(2017) - et al.
The paradox of social resilience: how cognitive strategies and coping mechanisms attenuate and accentuate resilience
Global Environ. Change
(2014) - et al.
Race, class, and Hurricane Katrina: social differences in human responses to disaster
Soc. Sci. Res.
(2006) - et al.
Evacuation transportation modeling: an overview of research, development, and practice
Transport. Res. Part C
(2013) The impact of evacuating on short-term disaster recovery: a study of individuals affected by Hurricane Harvey living in Texas counties
International Journal of Disaster Risk Reduction
(2020)Geometrical methods for non-negative ICA: manifolds, Lie groups and toral subalgebras
Neurocomputing
(2005)
Predictability of population displacement after the 2010 Haiti earthquake
Proceedings of the National Academy of Sciences of the United States of Aamerica
Mental health implications for older adults after natural disasters–a systematic review and meta-analysis
Int. Psychogeriatr.
Research challenges and opportunities in mapping social media and Big Data
Cartogr. Geogr. Inf. Sci.
Understanding individual human mobility patterns
Nature
Towards estimating urban population distributions from mobile call data
J. Urban Technol.
Adaptive Blind Signal and Image Processing: Learning Algorithms and Applications
NMF versus ICA for blind source separation
Advances in Data Analysis and Classification
A review of blind source separation methods: two converging routes to ILRMA originating from ICA and NMF
APSIPA Transactions on Signal and Information Processing
Experiences during and Responses to the Loma Prieta Earthquake
Risk area accuracy and hurricane evacuation expectations of coastal residents
Environ. Behav.
Gender and evacuation: a closer look at why women are more likely to evacuate for hurricanes
Nat. Hazards Rev.
An assessment of change in risk perception and optimistic bias for hurricanes among Gulf Coast residents
Risk Anal.
Cited by (9)
Mobile positioning-based population statistics in crisis management: An Estonian case study
2023, International Journal of Disaster Risk ReductionQuantifying COVID-19 recovery process from a human mobility perspective: An intra-city study in Wuhan
2023, CitiesCitation Excerpt :For a long time, follow-up surveys (Wang, Li, et al., 2014; Stringfield, 2010), repeat photography (Burton et al., 2011), and field visits (Contreras et al., 2018) are the main approaches for recovery phase division. In recent years, mobile phone data has become a popular source due to its advantages of real-time recording, wide spatial coverage, and no interference with objects (Hong et al., 2021; Wu et al., 2021; Yabe, Tsubouchi, Fujiwara, Sekimoto, et al., 2020). Sufficient observations are conducive to informing the accurate recovery phase, especially when there is a comparable normal baseline.
Analysis of the impact of non-compulsory measures on human mobility in Japan during the COVID-19 pandemic
2022, CitiesCitation Excerpt :In particular, the large number of mobile-phone users worldwide makes mobile-phone data a representative proxy for population mobility patterns (Jeffrey et al., 2020). In comparison to traditional methods such as collecting census data and conducting a questionnaire survey, the use of mobile-phone data is advantageous because of the following reasons: ability to obtain high-resolution population data in relation to time and space; accessibility of data without interview bias; availability of longitudinal data for a very large population (Wu et al., 2021). However, even if such data provide unique opportunities, there are also important methodological challenges that must be addressed.
Detection of SARS-CoV-2 RNA in wastewater and importance of population size assessment in smaller cities: An exploratory case study from two municipalities in Latvia
2022, Science of the Total EnvironmentCitation Excerpt :Based on literature research, an HPLC-MS/MS-based assay of neurotransmitter metabolite 5-HIAA has been selected in our study since it is an endogenous compound that is not lifestyle or habit dependent. In addition to biological and chemical biomarkers, the use of mobile data as a real-time data source for population size measurements has become increasingly relevant (Arhipova et al., 2020), and allows an assessment of the trends of regional economic development or to determine the change in population mobility patterns (Chen et al., 2018; Wu et al., 2021). This can be achieved through the use of call detail record (CDR) data, which are collected by mobile network operators and contain information about when, where and how a mobile network user generates voice calls and text messages (Chen et al., 2018).
Impact of transport network disruption on travel demand: A case study of the July 2018 heavy rain disaster in Japan
2022, Asian Transport StudiesCitation Excerpt :Thus, a framework that does not require such methods would be preferable (Wu et al., 2021). In addition, Wu et al. (2021) noted some advantages of mobile phone data, e.g.: 1) high resolution of time and space of the population data is obtained, 2) the data can be obtained directly without any interview bias, and 3) they are longitudinal data of a huge population. Although these advantages do not directly show the benefit of mobile phone data in terms of accuracy, gathering passive data from mobile phones is a promising option for this purpose.