The Evolution of Twitter Messages Before, During, and After the May 20th Tornado in Moore, OK

Recent research has found that real-time data on social media platforms like Twitter enhance our ability to detect and monitor public attention to severe weather. Building upon this research, we use an original archive of almost two million public posts or “tweets” about tornados collected immediately before, during, and after the May 20th tornado in Moore, OK (May 19 - 22, 2013) to advance our understanding of how individuals and organizations use social media to communicate about severe weather. More specifically, we use the tweets published during this time period to develop and evaluate a multi-level coding scheme designed to categorize tweets according to their content. Then, we use this coding scheme to characterize and compare patterns of communication across three phases of the event, including: (a) the pre-storm phase, wherein information about forecasts, watches, warnings, and preparations predominate; (b) the storm phase, wherein postings reflect the arrival of the storm, the “active” behavior of those in its path, preliminary impact reports, and messages about response efforts targeted toward affected populations; and (c) the post-storm phase, wherein the majority of posts contain “official” reports about what happened (i.e., reports about injuries, fatalities, and damages), subjective commentary, expressions of support/encouragement, and information about how to help the victims. Following this characterization, the paper concludes with a general discussion of how data of this sort can be used to improve severe weather forecasting, communication, and post-event assessments.