Assessment of Social Media Usage during Hurricane Sandy and the Development of a Twitter-based, Web Application for Real-time Communication of Storm-related Information

To better understand the sources, quality, and perceived reliability of social media information sent and received by individuals affected by Hurricane Sandy, the research team took a threefold approach to exploring the issue: 1) A dual-frame (landline and cellphone), RDD, telephone-based survey was conducted with a representative sample of approximately 750 respondents from New York, New Jersey, and Connecticut who experienced Hurricane Sandy; 2) Over 13 million Twitter messages sent prior to, during, and immediately following Hurricane Sandy were captured and analyzed; and 3) A novel software application was developed for use by emergency managers and other professionals in the field of meteorology to gather real-time, social media information (photographs and messages) for coordinating disaster relief, planning emergency response, and better understanding the immediate effects of a weather-related event.

The survey data yielded the following results: Television was the most frequently referenced medium for receiving storm-related information. Upon the loss of electrical power, there was an increase in the use of Twitter and other social media platforms to send and receive storm-related information. Twitter users obtained more information from family, friends, and government agencies than non-Twitter users. Across multiple social media platforms, storm-related photographs were the most frequently shared form of information, followed by personal experiences with the storm, and information about storm damage.

To identify key impact factors affecting the dissemination of weather-related, social media information during Hurricane Sandy, a total of 13.7 million Twitter messages (or tweets) were collected before, during, and after Hurricane Sandy made landfall in New Jersey on October 29, 2012. The raw data were indexed and inserted into a distributed NoSQL (MongoDB) database for storage. This database served as the central repository for all subsequent analyses. A significant number of key influencers from disparate domains (politics, news agencies, relief organizations, and weather services) were identified. These key influencers and their followers participated in Twitter-based discussions related to Hurricane Sandy. The connectivity of the influencers and their followers on Twitter played a vital role in information sharing and dissemination throughout the hurricane. There was a statistically significant increase in the number of Twitter messages and users participating in weather-related discussions during the peak of Hurricane Sandy and the average social-mood (or sentiment) of the users' messages shifted toward the negative end of a polar measure of overall sentiment. The observation of an average negative sentiment value for a group of users this large is a rare occurrence that is clearly indicative of a catastrophic event.

Taken cumulatively, the findings provided strong support for using Twitter as a reliable medium through which users can effectively communicate weather-related information to their followers. Furthermore, Twitter was found to be a rich source of information for gaining a better understand of how individuals communicate during weather-related emergencies.

Based upon evidence in support of using Twitter as a medium for bi-directional communication during weather-related emergencies, the research team developed a web-based application to provide emergency managers with a system for identifying and responding to important Twitter-based messages and images sent by the general public during weather-related events. Initial focus was placed on the value of using geo-coded images sent by Twitter users during Hurricane Sandy. Thousands of images depicting storm damage and flooding in the coastal areas of Connecticut, New York, and New Jersey during Hurricane Sandy were captured in real-time. Since the science of machine-based image recognition has not progressed to the level of accurately identifying images of storm damage or the depth of floodwater, the application relies on human judgment to codify the pertinent images. It is also critical for the human codifier to determine whether the exact location of the image (as indicated by its accompanying lat-lon coordinates) can be accurately identified. To this end, the web-based application has been designed to provide emergency managers and first responders with real-time, exact-location images chronicling the effects of a weather-related disaster. These images can be accessed prior to the first responders safely enter a disaster zone, thereby providing invaluable information for the planning of rescue and relief efforts. This research showed that many people ignore evacuation warnings and decide to remain in place and share photographs of dangerous weather activity. While our researcher team certainly does not condone this behavior, we chose to develop a software application that will allow emergency managers to use these photographs for the greater good.