3. Tweeting and tornadoes: seek shelter now #okwx

Social media like Facebook and Twitter are popular for connecting with friends and sharing information about politics, celebrities, and cultural trends, but in a disaster, can a tweet really save a life?

As a way to understand how people use Twitter in a disaster situation, a team of geographers at Penn State, led by Justine Blanford, used Twitter to assess the effectiveness of warning messages sent during the Moore, Oklahoma tornado of May 20, 2013, by exploring the spatial and temporal relationships of real-time reactions of the general public as the storm system developed into a tornado.

During late May 2013, a series of tornado-producing storm systems swept through the greater Oklahoma City, Ok., region. A particularly violent tornado touched down at 2:56 p.m. local time near Newcastle, 16 minutes after a tornado warning was issued by the National Weather Service (NWS) Norman office. The tornado rapidly strengthened, tracking directly over the city of Moore. The tornado traveled nearly 14 miles in 39 minutes, with a maximum path width of just over one mile.


Moore tornado

Radar reflectivity factor around the time of the tornado reaching its maturity.

The pink/purple (really large values) at approximately the center of the image

is an indication of debris swirling around in the tornado.

Image created by Matthew Kumjian, assistant professor of meteorology at Penn State.



“Using GIS Tools, we analyzed 86,100 geo-located tweets collected on May 19, May 20, and May 21 for Oklahoma,” Blanford, a research associate with the GeoVISTA Center, said. “We started by selecting tweets with keywords such as ‘tornado,’ ‘storm,’ ‘weather,’ ‘shelter,’ ‘emergency,’ and used an iterative process to add relevant tweets with additional keywords such as ‘destruct,’ ‘help,’ and others,” she added.

“We found that people in Oklahoma were generally interested in weather and tornadoes, and on the day of the tornado, the number of tweets increased significantly.”

Content analysis of the tweets found that they were used to provide situational updates such as relaying media reports, which included television stations KOCO, KJRH, and KFOR, the University of Oklahoma emergency alert system, and re-tweets of information from the NWS (e.g., “KOCO is talking about the south, guys. Wall cloud. No tornado yet.” ); providing real-time weather observations (e.g., “This tornado is about a mile wide. Oh dang.” ); providing the location of shelters (e.g., “If you’re on campus, seek shelter in Residence Hall basements, Union basement or Huff. #OU #okwx”); and communicating personal safety and locating family and friends (e.g., “We are ok. F4 tornado hit about 2 miles from us. Don’t have power right now. Hope this posts.”).





tweet maptweet map legend



 Map (by Justine Blanford) showing location and density of tweets in relation to the tornado path.





After the tornado passed, messages included situational awareness and damage reports (e.g., “Heavy tornado damage near SE 4th and Bryant. Homes are gone,” and  “Children trapped in #Moore guess ill be #volunteering all night #oklahomatornadoes #oklahoma #okiepride #help #tornado #redcross” ).

“Twitter is an effective messaging system that enables information to be received and posted in a timely manner,” Blanford explained. “During the tornado, Twitter was useful for providing updates and relaying information. By analyzing the text of each tweet and using a list of keywords, we gained insights into what happened on the ground and understood people’s interest and reactions, both spatially and temporally. For unpredictable and destructive/hazardous weather events, such as that tornado, the time between issuing a warning and the tornado touchdown can be short, emphasizing the need for clear communication.”

Blanford noted that information passed on throughout Twitter is straightforward and easy to access, but is just one source of information. People obtain information from a variety of sources.

The study presented here highlights the need for additional research that should include strategies for prompting retweets, identifying effective versus ineffective messaging, and assessing the role of volunteer communities in communicating risk obtained from both formal and informal information sources.

“In this study we analyzed data for a single event. A long-term goal for the research reported here is to provide insights to forecasters and emergency response personnel concerning the impact of warnings and other advisory messages,” Blanford said.

This paper was presented at The 11th International Conference on Information Systems for Crisis Response and Management, held at University Park, Pennsylvania, USA, May 2014.