Coastal Hurricane Prediction and Communication Dynamics in the Modern Information Environment

Many coastal areas in the U.S. face significant risks from hurricanes. Hurricane predictions have improved significantly in recent decades, providing additional forecast and warning information to coastal residents. At the same time, advances in information and communication technology are dramatically changing how people access, combine, and share information when hazardous weather threatens. We will discuss a project that aims to understand and improve hurricane information communication and use by integrating knowledge and methods from atmospheric and related sciences, computer and information science, and social and behavioral sciences. The project focuses around investigating how interactions among different actors and types of information influence risk interpretations and behavioral responses as a hurricane approaches and arrives, in the context of evolving meteorological predictions and the modern information environment. The research combines study of the real-world hazard information system with computational physical and social modeling. To understand the real-world system, we are collecting and analyzing data about how people communicate, perceive, and respond to hurricane threats from social media streams, complemented by interviews and focus groups. This is combined with computational modeling research that includes high-resolution ensemble hurricane and storm-surge modeling and agent-based modeling of social actors who pursue, process, and transmit information in coastal areas.

We will present the project paradigm and initial results, with an emphasis on empirical analysis of data from social media streams collected during Hurricane Sandy. Research has considered how the use of social media, and specifically Twitter, by members of the public before, during and after a disaster can be used opportunistically to detect the presence of a hazard and derive situational status information about hazardous impacts. The methods for such derivation, however, are being outstripped by the exponential adoption of Twitter, especially during disasters. The volume of Twitter posts in anticipation of hazards like hurricanes that hit highly populated coastal areas is so vast (and growing exponentially) that even advanced human- and machine- techniques cannot sift through the global response to this event to find the “signal in the noise.” As a result, techniques that rely on either human or machine language detection will fail at isolating critical pieces of information that researchers and practitioners need. In this research, we are developing new techniques for detection of protective-decision-making activity by geographically vulnerable Twitterers. By examining movement over time, we can detect those who evacuated and those who sheltered-in-place, and how well their public social media activity aligns with their actions. From these “found” users, we can then perform empirical content analysis of their tweet activity to investigate socio-linguistic behaviors with respect to information use and protective decision-making.