Exploring the potential of social media crowdsourcing for post-earthquake damage assessment

Abstract

Following disaster events, a significant hindrance to emergency response is a lack of information on the spatial extent and severity of damages. Social media crowdsourcing has been studied as a potential source of information to help support damage assessment following various disaster events. However, the challenges and opportunities associated with using social media crowdsourcing to support post-earthquake response may differ from other hazards due to the sudden-onset of earthquakes (relative to events with more substantial warning time, such as hurricanes). This study aims to explore the potential of social media data, principally short-form communications such as Twitter™ postings, to support damage assessment. Such information can complement more traditional sources of information, such as inspections, surveys, and remote sensing, but may also be associated with unique challenges. Natural language processing (NLP) techniques and machine learning (ML) classifiers were applied to classify damage levels from contemporary Twitter short-form communication postings. Six earthquake sequences were studied to investigate the potential opportunities and challenges associated with using this approach. The damage estimates’ temporal characteristics were studied, and results were compared against USGS and news media reports to understand the potentials and challenges of social media crowdsourcing. This study offers insights into the application of the crowdsourcing approach, specifically in terms of its timing and spatial coverage. Additionally, this study explores how the quantity and quality of social data could potentially complement conventional technologies for damage assessment and to support decision-making.