n an ambitious and provocative piece, Wurman et al. (2007, hereafter W07) explored the catastrophic potential of a violent tornado striking densely populated portions of Chicago, Illinois, and other major U.S. cities. This important paper was the first to predict the mortality of such an event, using tornadic wind fields, tornado path length and width, census data, housing types, degrees-of-damage (DOD) ratings, and probability-of-death (POD) estimates as predictors. In each of their simulations, Chicago suffered multiple thousands of deaths, with a worst-case estimate of 63,000. A recent comment and reply exchange between Brooks et al. (2008) and Wurman et al. (2008) dealt largely with the numerical accuracy of W07’s mortality projections in the context of previous urban tornadoes, and to a lesser extent, W07’s tornado model. I have modified this comment so as not to overlap with that debate. While I also question W07’s mortality estimates, I am most concerned with the underlying assumptions about the population. These assumptions, combined with several left-out factors, make the mortality estimates in W07 very difficult to believe, even after considering the additional reasoning in the Wurman et al. (2008) reply to the Brooks et al. (2008) comment. While modeling mortality at the city–tornado interface is, in itself, a very difficult undertaking, W07’s model could nevertheless be improved. The POD given a particular DOD should be used with other “death variables,” and a more sophisticated model of population dynamics. In simplified terms, the death toll for a given scenario in W07 is a function of 1) the population in low-rise, single family residential structures that face winds ≥76 m s–1, and 2) the population in highrise structures exposed to winds ≥102 m s–1. W07 use POD = 0.1 for all homes meeting criterion 1, and POD = 0.01 for all structures meeting criterion 2. In the case of their most lethal track (“HN”), they infer 630,000 criterion 1 residents, and thus, 0.1 × 630,000 = 63,000 deaths. The model assumes, however, that each censusregistered resident is at home when the tornado strikes. While it may be impossible to determine exactly where each of the 630,000 EF-4/5-affected residents in the HN scenario will be at the time of the tornado, one can be virtually certain that they will not all be at home. In fact, given the tendency for tornadoes to strike during the late afternoon and early evening—a time when many people are often in transit, coming from work, school, etc.—a W07-type tornado would likely affect many people not in their homes. Anyone familiar with Chicago knows that its “rush hour” often begins during midafternoon, and ends during the midevening; understanding the impacts of a rush-hour tornado therefore requires a thorough accounting of traffic densities and flows. Furthermore, a significant number of people would probably be outside, at least for a “prime time” tornado; this is Chicago after all. If one is modeling impacts, then one should account for the varying levels of impact among the mobilized population, not just the impacts to people in structures. To their credit, W07 did note that, “[t]he probability of dying in a tornado is likely sensitive to the time of day, the day of week, and the meteorological conditions, since these determine whether people are in their homes, driving, whether they are asleep or awake, and whether the tornado would be visible or surrounded by rain” (p. 40). It is unclear, however, why they otherwise ignored these important factors when estimating the impacts, especially since a study with a largely reproducible methodology already existed (and was cited in W07). Rae and Stefkovich (2000) used detailed economic, demographic, land use, and traffic data to model impacts from a major tornado in the Dallas–Fort Worth, Texas (DFW), area. While their tornado model merely superimposed tracks from the 3 May 1999 Oklahoma City tornadoes onto DFW, and was not as robust as the W07 tornado model, their treatment of the nontornado factors was superior to W07’s. Methodologies from “time geography,” which is concerned with daily population movements and has been employed by traffic modelers among others (e.g., Timmermans et al. 2002), could have further illuminated W07’s investigation. Incorporating more realistic variables into the model would, of course, require understanding critical wind thresholds for people who are stuck in traffic jams, driving in ordinary traffic, or are outDOI:10.1175/2008BAMS2670.1 I
Read full abstract
Nov 1, 2013
J Brotzge + 1
Open Access
