Research ArticleSocial media crowdsourcing for rapid damage assessment following a sudden-onset natural hazard event
Introduction
Natural disasters are costly (Coronese, Lamperti, Keller, Chiaromonte, & Roventini, 2019). The rapid appraisal of damages related to natural hazards is essential to event response and recovery efforts. Post-event environments are characterized by incomplete and rapidly evolving information (Rouhanizadeh, Kermanshachi, & Nipa, 2020). As a result, a significant hindrance to emergency response and efficient event recovery is a lack of knowledge of the spatial extent and magnitude of damage. While certain events (e.g., hurricanes) may be forecasted, other hazards (e.g., earthquakes) occur with little or no warning (Vieweg, Castillo, & Imran, 2014). These sudden-onset events present a particular challenge due to the inability to pre-deploy response and recovery resources.
First responders, government agencies, and private sector entities need information to support effective response and decision-making. While satellite monitoring, ground-based sensors, and other modern technologies have provided benefits in tracking post-disaster conditions (Haq, Akhtar, Muhammad, Paras, & Rahmatullah, 2012; Monfort, Negulescu, & Belvaux, 2019; Wang, Qu, Hao, Liu, & Stanturf, 2010), crowdsourcing through social media presents an additional and novel source of such information (Simon, Goldberg, & Adini, 2015). While social media is an inherently imperfect information source, it provides rapid and geographically distributed information that complements that from other sources. Specifically, social media data are available in large quantities and at spatial densities that may exceed conventional sensor networks. Moreover, social media data can be rapidly collected in near-real-time, without the time required to deploy reconnaissance technologies or personnel. However, they are of relatively low fidelity compared to data from conventional sensors, aerial imagery, or the observations of trained inspection teams.
The goal of this study is to explore the utility of social media data to provide rapid indications of damage following sudden-onset natural hazard events. Multiple studies have demonstrated the potential of applying social media for a rapid damage assessment. These studies use the intensity of social media activities or the sentiment level as a metric to indicate the extent of damage in the affected areas rather than to quantitatively parse the damage levels (Resch, Usländer, & Havas, 2018; Wu & Cui, 2018; Yuan & Liu, 2020). Moreover, there has been minimal focus on leveraging text classification methods to estimate damage levels with social media data in the earthquake context. To fill this research gap, this study investigates the use of Twitter™ postings and builds text classification models based on the Modified Mercalli Intensity (MMI) Scale to make approximate but rapid early assessments of damages due to earthquakes. This approach can be extended to other natural hazards such as floods, coastal surges, wildfires, or tornadoes.
While assessments based on social media have neither the accuracy nor the precision of ground-based instruments and observations, they have advantages of rapidity, quantity, and spatial coverage. This study provides insights on useful natural language processing instruments and machine learning classifiers for textual analysis in the context of post-event rapid damage assessment. Through the exploration of results, this study further offers insights regarding potential challenges to using social media and offers opportunities for future research. Specifically, using relatively simple metrics of text characteristics, a great deal of information can be gathered about the patterns and timing of a hazard event. Combining these simple metrics with textual analysis allows a characterization, albeit rough for the present, of damage levels. The case of the Ridgecrest, CA earthquake sequence on 4 and 6 July 2019 is used to investigate the potential for rapid damage estimates from social media data.
Section snippets
Social media in disaster management
Social media data from Twitter™, Facebook™, Instagram™, and other web platforms are now used for emergency response and recovery, natural hazards planning, and risk mitigation. Several efforts to use social media data for natural hazards response and civil infrastructure planning have recently appeared (e.g., Kryvasheyeu et al., 2016; Leykin, Lahad, & Aharonson-Daniel, 2018; Niles, Emery, Reagan, Dodds, & Danforth, 2019). These studies have evaluated social media volume and the use of targeted
Hypothesis development
The intent of this study is to quantitatively explore the value using of social media data to provide rapid indications of damage following sudden-onset natural hazard events, specifically earthquakes. Specifically, we seek to quantitatively parse the damage levels through the use of Twitter™ postings and text classification models. To assess the proposed approach, we establish a series of hypotheses to: (1) test whether social media users are shown to react significantly to earthquake events,
Data and methods
This study investigates the aforementioned hypotheses by parsing social media (Twitter) data to identify (1) indicators of various levels of damage and (2) the geospatial location of the damage (even when explicit geospatial information is not available). Given the engineering relevance of language related to damage, a special-purpose library is constructed, leveraging existing damage scales. A series of candidate models are developed and tested for identifying, processing, and analyzing tweets
Results
The subsections that follow explore the analysis of Twitter data associated with the Ridgecrest earthquake sequences from three perspectives, which are linked the exploration of the two hypotheses identified in Section 3. First, Section 5.1 presents an initial descriptive analysis of tweet volumes and characteristics, which provide interesting insights regarding the nature of responses to earthquakes on Twitter and supports Hypothesis 1(a) and Hypothesis 1(b) related to user actions following
Discussion
Sudden-onset natural hazards, especially earthquakes, occur with little or no warning and may result in significant damage. The rapid appraisal of losses related to these natural hazards events is of significance to residents, government agencies, insurance companies, and other stakeholders. Damage can vary from things being broken and minor injuries to complete collapse of buildings and deaths. The spatial distribution of impact on large regions makes place-by-place, in-person assessment
Conclusions
This research uses the Ridgecrest, CA earthquakes of July 2019 as a test case to illustrate the potential utility of using social media data for rapid damage assessment and to provide insights regarding potential challenges. To verify the feasibility of the proposed methodology, this study leverages the insights from MMI Scale and develops a simple four-step scale of earthquake damage. It then trains and tests a series of candidate models for parsing tweet text to extract information about the
Author statement
- •
Lingyao Li: Conceptualization, Methodology, Data Curation, Formal Analysis, Writing – Original Draft, Writing – Review & Editing.
- •
Michelle Bensi: Conceptualization, Validation, Writing – Original Draft, Writing – Review & Editing.
- •
Qingbin Cui: Conceptualization, Writing – Review & Editing.
- •
Gregory B. Baecher: Conceptualization, Writing – Original Draft, Writing – Review & Editing, Supervision.
- •
You Huang: Data Curation.
Data availability
The full dictionaries of words to filter the data and the temporal and spatial results can be found at https://data.mendeley.com/datasets/z9xjcmg6s2/3, an open-source online data repository hosted at Mendeley Data. (L. Li, 2020).
Acknowledgement
We acknowledge the help of Kaveh Faraji Najarkolaie with map data generation in Section 5. This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.
References (78)
- et al.
Facebook and Whatsapp as disaster management tools during the Chennai (India) floods of 2015
International Journal of Disaster Risk Reduction
(2019) - et al.
Extracting patterns from Twitter to promote biking
IATSS Research
(2019) - et al.
Assessing disaster impacts and response using social media data in China: A case study of 2016 Wuhan rainstorm
International Journal of Disaster Risk Reduction
(2019) - et al.
Techniques of Remote Sensing and GIS for flood monitoring and damage assessment: A case study of Sindh province, Pakistan
The Egyptian Journal of Remote Sensing and Space Science
(2012) - et al.
Social media data and post-disaster recovery
International Journal of Information Management
(2019) - et al.
How engaging are disaster management related social media channels? The case of Australian state emergency organisations
International Journal of Disaster Risk Reduction
(2020) - et al.
Social network analysis: Characteristics of online social networks after a disaster
International Journal of Information Management
(2018) - et al.
Emergency information diffusion on online social media during storm Cindy in U.S
International Journal of Information Management
(2018) - et al.
Gauging urban resilience from social media
International Journal of Disaster Risk Reduction
(2018) - et al.
Feeling alone among 317 million others: Disclosures of loneliness on Twitter
Computers in Human Behavior
(2019)