Effective hurricane risk messaging

Scientists at the U.S. National Science Foundation National Center for Atmospheric Research (NSF NCAR) analyzed 2017 Twitter (now X) data related to two hurricanes -- Harvey and Irma. The researchers found that forecast information communicated in the early stages of storm development, when the threat posed by a hurricane is uncertain, sets the stage for how people react to subsequent warning messages.

The research team also identified ways authoritative sources, such as the National Weather Service, broadcast meteorologists, and emergency managers, can better engage at-risk community members.

"When we began this research, a lot of work had been done analyzing Twitter data in the post-disaster space, and we were interested in how people were responding to weather forecast information in earlier stages, especially as predictions change," said NSF NCAR scientist Rebecca Morss, lead author of the Hurricane Harvey paper. "Twitter provided a natural laboratory for us to look at what communications people are responding to and what information they are sharing. This kind of research can help the meteorological community learn what are the most important things to communicate and how to improve that messaging."

The research was supported by NSF NCAR and funded in part by NOAA and a grant from NSF.

Risk Information Ecosystem

Harvey and Irma both occurred in 2017, but the two storms behaved very differently from one another. Harvey was an atypical hurricane that rapidly intensified and then lingered along the Texas coast for a few days, leading to devastating flooding in the Houston region. Whereas Irma was forecast well in advance of making landfall and followed a more typical movement pattern over land, leading to significant risks from strong winds and storm surge. Combined, the two storms provided complementary data about the different types of hurricane risks that need to be communicated amid varying levels of uncertainty.

Twitter offered the NSF NCAR researchers a detailed log of how people were sharing and reacting to information in real time. The research team developed a dataset of tweets that was relevant to each hurricane and focused on tweets posted by authoritative sources.

To evaluate how the tweets communicated information about the storms to those in harm's way, the researchers categorized tweets based on the type of visuals used and looked for patterns in what image types were retweeted and interacted with the most. Additionally, the researchers examined the content of replies to forecast and warning tweets to explore how people in at-risk areas make sense of and respond to a hurricane's evolving threat.

"In early stages of the threat, we could see a really clear cadence that every six hours there would be an uptick in Twitter conversation about the hurricanes," said NSF NCAR scientist Robert Prestley, lead author of the Hurricane Irma paper. "This was driven by the National Hurricane Center putting out updated forecast information. That information would then be redistributed by broadcast meteorologists, emergency managers, news media, and weather enthusiasts, and the conversation would grow from there. It really highlighted the key role of the National Weather Service in leading this communication."

Morss and Prestley also observed that the role of the national and local National Weather Service (NWS) offices shifted over the course of a hurricane. As the storm developed, NWS centers such as the National Hurricane Center played the leading role in spreading information, but as the storms began to impact communities, local NWS offices took the lead in generating locally relevant forecast and warning content.

Viral Visuals

The research identified several types of images that were commonly used to convey information about the threat of a hurricane. Of these, the "cone of uncertainty" was the most retweeted graphic.

The cone of uncertainty shows the current location of a tropical storm as well as the probable track of the storm's center with a cone shape around it that represents the uncertainty of the track based on historical errors. Previous research has shown that cone images have several limitations such as a lack of information about the risks to those outside of the cone, which can lead to people falsely believing areas outside of it are not at risk.

"There's a clear need for better uncertainty visualizations, especially in the forecast and warning period when people are looking for information, but there's not yet enough certainty to say specifically where or what the impacts will be," Prestley said. "The cone is not necessarily equipped to do that. The question is how to maintain the visibility of an image people are accustomed to while more effectively communicating hurricane risks to different populations."

The researchers also identified ways image content varied across the two hurricanes. For instance during Hurricane Harvey, tweets that included images highlighting heavy rain and flooding were highly retweeted. The high engagement with those tweets demonstrates the value of authoritative sources tweaking their communication to emphasize the different risks posed by individual storms.

The image type least interacted with was watch and warning imagery. These images include information about watches and warnings issued by NWS for hazards like flash floods, high winds, or tornadoes. Often these images are generated by automated programs and include identical or similarly formatted text and visualizations.

Many of these watches and warnings are issued for narrow geographic areas over short time-frames, which could explain why most of them have few retweets. Morss and Prestley recommend more research into how to improve communication of watch and warning information on social media, including how and when to use automated tweets to rapidly disseminate more user friendly information.

Although several things about Twitter have changed since the NSF NCAR research team conducted their studies, the research conducted by Morss, Prestley, and colleagues using the platform built a new understanding of how people respond to evolving information about natural disaster threats. Currently, the researchers are using additional research methods, including surveys before, during, and after weather events, to study the way people react to pending natural disasters. The Twitter studies helped shape the questions used in this ongoing research.

The knowledge gained can help professional weather communicators improve how they use social media networks as a vibrant resource for improving disaster communication.