87th AMS Annual Meeting

Monday, 15 January 2007: 2:15 PM
A mortality-based heat wave climatology for U.S. cities
206A (Henry B. Gonzalez Convention Center)
Robert E. Davis, Univ. of Virginia, Charlottesville, VA; and P. C. Knappenberger, P. J. Michaels, and W. M. Novicoff
Although no widely accepted specific definition for a “heat wave” exists, one approach is to focus on a response variable, with the implication that heat waves must impart some negative societal impact. Given that heat is the primary weather-related killer in the United States, we here examine various aspects of human mortality during heat waves. Specifically, we attempt to answer four fundamental questions. First, is there evidence for long-term trends in heat-wave mortality? Second, what is the minimum duration of oppressive conditions needed to induce heat-related mortality? Third, does the timing of the heat wave within a given warm season influence the resulting mortality? Fourth, how do the mortality impacts arising from large positive anomalies compare to absolute conditions? An clear understanding of these fundamental questions is necessary prior to the development of heat wave mortality mitigation models.

In an attempt to answer these questions, we compiled a daily mortality data base comprised of all-causes mortality for each of 28 large metropolitan areas over 29 non-consecutive years from 1964-1998. These data were standardized by age to account for changing demographics within a given city and to facilitate inter-city comparisons. First order weather station data from each city were used to calculate 1600 LST apparent temperature, a heat index that combines air temperature and humidity into temperature-equivalent units.

For each city, a suite of “heat events” are defined using a very liberal definition of consecutive above-normal days. Using this data set, extreme heat events are defined as those events with a maximum apparent temperature that exceeds one standard deviation calculated using a centered moving five-year window. A similar definition is used to define high mortality events, based on heat events with unusually high total mortality during the event. Given that these two definitions—extreme heat events and high mortality events—are independently defined, we can determine if the overlap (e.g., high mortality during extreme heat, or “killer heat waves”) is statistically different from random chance. Using a similar approach, we also compare “absolute” versus “relative heat waves”; i.e., extreme heat events defined based on absolute conditions or departures from normal.

With respect to question one, trends in killer heat wave mortality, we find that mortality is declining significantly in 21 of the 28 cities during killer heat waves. The impact is most pronounced in the typically most susceptible regions of the Northeast and Midwest. Over the same period of time, while apparent temperatures are generally increasing, the rise is statistically significant in only seven of the 28 cities.

For question two, the minimum duration needed to induce a significant mortality response is 3-4 days in most cities. Events with this minimum duration account for 90% of all excess mortality. Minimum durations tend to be shorter in the southern and western United States.

Previous research on within-season timing, our third question, strongly suggested that early season heat waves have a greater impact than subsequent events, often regardless of the event's intensity or duration. However, we find that there is no statistical evidence to support this relationship in major U.S. cities. In a comparison of the number of “first (in season) killer heat waves” to non-killer heat waves, we find that the number does not exceed random chance across our sample. This result applies to both relative and absolute heat waves.

Finally, both absolute and relative killer heat waves are evident. Although the number of each exceed random chance, absolute heat waves are much more common, and many of these are also classified as “relative” based upon their apparent temperature departures. Nevertheless, there exist early- (April-May) and late-season (September) killer heat waves that should be carefully examined.

Our findings regarding heat wave mortality trends, minimum durations, within-season adaptations, and the impacts or relative vs. absolute heat waves could be incorporated into the planning efforts of municipalities to offset the impacts of summer heat on the resident population. New and ongoing modeling efforts should incorporate these factors into predictive models.

Supplementary URL: