Contextualizing disaster phases using social media data: Hurricane risk visualizations during the forecast and warning phase of Hurricane Irma

Abstract

Abstract Common disaster-phase models provide a useful heuristic for understanding how disasters evolve, but they do not adequately characterize the transitions between phases, such as the forecast and warning phase of predictable disasters. In this study, we use tweets posted by professional sources of meteorological information in Florida during Hurricane Irma (2017) to understand how visual risk communication evolves during this transition. We identify four subphases of the forecast and warning phase: the hypothetical threat, actualized threat, looming threat, and impact subphases. Each subphase is denoted by changes in the kinds of visual risk information disseminated by professional sources and retransmitted by the public, which are often driven by new information provided by the U.S. National Weather Service. In addition, we use regression analysis to understand the impact of tweet timing, content, risk visualization and other factors on tweet retransmission across Irma’s forecast and warning phase. We find that cone, satellite, and spaghetti-plot image types are retweeted more, while watch/warning imagery is retweeted less. In addition, manually generated tweets are retweeted more than automated tweets. These results highlight several information needs to incorporate into the current NWS hurricane forecast visualization suite, such as uncertainty and hazard-specific information at longer lead times, and the importance of investigating the effectiveness of different social media posting strategies. Our results also demonstrate the roles and responsibilities that professional sources engage in during these subphases, which builds understanding of disasters by contextualizing the subphases along the transition from long-term preparedness to postevent response and recovery. Significance Statement Visual information is an important tool for communicating about evolving tropical cyclone threats. In this study, we investigate the kinds of visualizations posted by professional weather communicators on Twitter during Hurricane Irma (2017) to understand how visual information shifts over time and whether different visuals are more retweeted. We find that visual information shifts substantially in the days before Irma’s impacts, and these shifts are often driven by changes in Irma’s strength or forecast track. Our results show that cone, satellite, and spaghetti-plot visualizations are retweeted more frequently, while watch/warning imagery is retweeted less. These results help us to understand how visual information evolves during predictable disasters, and they suggest ways that visual communication can be improved.