| dc.description.abstract | The risks posed by rare and severe weather events are both the most impactful to the public and the most difficult to communicate, due to their low absolute probability of occurrence but high impact at a given point. Recent research on rare risk communication has found that probabilistic information is the most effective way to communicate the risk posed by weather to members of the public, but also that small absolute probabilities can be misinterpreted due to known cognitive biases. One potential solution to communicating rare event likelihoods is a value called relative risk, which this work holistically investigates in the context of tornado risk communication. Relative risk is defined as the ratio of the forecast likelihood of an event to the background likelihood of that event occurring, the quotient of which describes how many times more likely than “normal” an event is for a given forecast. To achieve a broader understanding of how relative risk for tornadoes might impact different aspects of tornado risk communication, a 1950-2021 climatology of tornado relative risks was studied alongside focus group and survey data collection that investigated the reception of relative risk information by broadcast meteorologists and members of the public.
First, the 1950-2021 US observed tornado dataset maintained by the Storm Prediction Center (SPC) was used to calculate relative risks for tornadoes for all events across the 71 year period. Observed tornado reports were used to calculate the climatological likelihood of a tornado within 25 miles of a point for every day of the year, as well as Practically Perfect Hindcasts (PPHs) for tornadoes for every day in the 71-year dataset. Dividing these PPHs by the daily 1950-2021 tornado climatology produced a full series of relative risk values. Analysis of the 71 years of relative risk data revealed that the highest values of relative risk occur across the western and northern regions of the contiguous US where tornadoes can be rare, while the lowest values of relative risk were observed in the southern US where tornadoes occur year-round. Relative risks values of 5, 20, 50, 100, 250, and 700 times more likely than normal were observed to occur across the 71-year dataset at similar rates to the absolute likelihood values of 2, 5, 10, 15, 30, and 45% used in the SPC probabilistic outlook, and were thus chosen to be used in in mapped presentations of relative risk to potential users. Overall, these results suggest that relative risks likely have the greatest value for communicating tornado events that occur in areas with established tornado climatology when tornadoes occur during off-season times of year or in places that they are more infrequent.
Next, focus group interviews with broadcasters revealed a great deal of suspicion of towards viewers’ ability to interpret probability information in any form. Broadcasters in these interviews believed that relative risk information might be seen as overblown, potentially inducing unnecessary panic in their viewers. Relative risk was seen by broadcasters as a useful tool for their personal use in quantifying how unusual a tornado event was, and a few suggested they would present relative risk during tornado for events occurring outside of tornado season or outside of tornado-prone regions.
Finally, members of the public were also surveyed about their perceived level of concern and likelihood of response when shown different levels of relative risk information. Participants on average reported large increases in concern when shown only relative risk information but presenting both absolute and relative risk information led to better differentiation in level of concern across increasing absolute likelihood risk levels. Later experiments added nuance to these findings, and showed that although relative risk information at values as high as 500 times more likely than normal had no effect on participant trust in future forecasts if that forecast were a false alarm, there was also no change to participant concern or likelihood of response across increasing levels of relative risk with a constant absolute risk value.
Through a series of three investigations, this dissertation attempts to develop and test a potential forecast product in a rigorous and methodical way that combines key understandings from both the meteorological and social sciences. Following this process helps ensure that both the potential range of values the product may contain and the range of responses that product recipients will display are well-understood. These rigorous efforts have revealed that although relative risk information could be valuable in specific situations for communicators like broadcast meteorologists, it does not appear to have significant effects on individual risk assessment or decision-making for members of the public. Thus, relative risk does not appear to offer a silver bullet for improving the communication of rare events like tornadoes. | en_US |
