Twitter Storm: Analysis of Tweet Performance During the 23 February 2016 Louisiana Tornado Outbreak
Joshua Eachus, WBRZ-TV & Louisiana State University, Baton Rouge, LA
Michelle Meyer, Louisiana State University
Barry Keim, Louisiana State University
 ,
 ,

Abstract
Many weather forecasters turn to Twitter, among other social media platforms to rapidly disseminate warming information during high impact weather events. The continued use of Twitter as a channel for weather information inspires a deeper analysis of what kinds of messages resonate with users-knowledge that could improve the reach of warning messages in extreme events. The literature identifies multiple aspects of disaster warnings such as geography and personalization that improve the impact of warning messages to the public. The Protective Action Decision Model (PADM) describes what factors increase risk perception and drive protective actions in response to hazards. To assess the usefulness of Twitter as warning channel, this research uses the PADM as guide to develop a typology for weather tweets and identify what factors affect the reach or speed of warning messages via Twitter. The dataset of tweets comes from two television stations and three Twitter accounts within a broadcast television market during the 23 February 2016 Southeast Louisiana tornado outbreak. Retweets, impressions, and likes were compared across identified tweet types to assess attention and exposure to warning messages. The most popular tweets were then content analyzed to determine commonalities. Results indicate that photographs and official statements were popular tweet types, while value added messages performed surprisingly poor. Additionally, warning messages which offer a mitigating action perform better than those only mention a threat. These results will help weather forecasters develop an optimized messaging strategy for using Twitter during future high impact weather events.