The ecological and economic implications of the greenhouse effect have been the subject of discussion within the scientific community for the past three decades. In recent years, members of Congress have held hearings on the greenhouse effect and have begun to examine its implications for public policy. This interest was accentuated during a series of hearings held in June 1986 by the Subcommittee on Pollution of the Senate Environment and Public Works Committee. Following the hearings, committee members sent a formal request to the EPA Administrator, asking the Agency to undertake two studies on climate change due to the greenhouse effect.
One of the studies we are requesting should examine the potential health and environmental effects of climate change. This study should include, but not be limited to, the potential impacts on agriculture, forests, wetlands, human health, rivers, lakes, and estuaries, as well as other ecosystems and societal impacts. This study should be designed to include original analyses, to identify and fill in where important research gaps exist, and to solicit the opinions of knowledgeable people throughout the country through a process of public hearings and meetings.
To meet this request, EPA produced the report entitled The Potential Effects of Global Climate Change on the United States. For that report, EPA commissioned fifty-five studies by academic and government scientists on the potential effects of global climate change. Each study was reviewed by at least two peer reviewers. The Effects Report summarizes the results of all of those studies. The complete results of each study are contained in Appendices A through J.
A Water Resouces
B Sea Level Rise
E Aquatic Resources
F Air Quality
The goal of the Effects Report was to try to give a sense of the possible direction of changes from a global warming as well as a sense of the magnitude. Specifically, we examined the following issues:
The four regions chosen for the studies were California, the Great Lakes, the Southeast, and the Great Plains. Many studies focused on impacts in a single region, while others examined potential impacts on a national scale.
SCENARIOS USED FOR THE EFFECTS REPORT STUDIES
The Effects Report studies used several scenarios to examine the sensitivities of various systems to changes in climate. The scenarios used are plausible sets of circumstances although none of them should be considered to be predictions of regional climate change. The most common scenario used was the doubled CO2 scenario (2XCO2), which examined the effects of climate under a doubling of atmospheric carbon dioxide concentrations. This doubling is estimated to raise average global temperatures by 1.5 to 4.5deg.C by the latter half of the 21st century. Transient scenarios, which estimate how climate may change over time in response to a steady increase in greenhouse gases, were also used. In addition, analog scenarios of past warm periods, such as the 1930s, were used.
The scenarios combined average monthly climate change estimates for regional grid boxes from General Circulation Models (GCMs) with 1951-80 climate observations from sites in the respective grid boxes. GCMs are dynamic models that simulate the physical processes of the atmosphere and oceans to estimate global climate under different conditions, such as increasing concentrations of greenhouse gases (e.g., 2XCO2).
The scenarios and GCMs used in the studies have certain limitations. The scenarios used for the studies assume that temporal and spatial variability do not change from current conditions. The first of two major limitations related to the GCMs is their low spatial resolution. GCMs use rather large grid boxes where climate is averaged for the whole grid box, while in fact climate may be quite variable within a grid box. The second limitation is the simplified way that GCMs treat physical factors such as clouds, oceans, albedo, and land surface hydrology. Because of these limitations, GCMs often disagree with each other on estimates of regional climate change (as well as the magnitude of global changes) and should not be considered to be predictions.
To obtain a range of scenarios, EPA asked the researchers to use output from the following GCMs:
Figure 1 shows the temperature change from current climate to a climate with a doubling of CO2 levels, as modeled by the three GCMs. The figure includes the GCM estimates for the four regions. Precipitation changes are shown in Figure 2. Note the disagreement in the GCM estimates concerning the direction of change of regional and seasonal precipitation and the agreement concerning increasing temperatures.
Two transient scenarios from the GISS model were also used, and the average decadal temperature changes are shown in Figure 3.
EPA specified that researchers were to use three doubled CO2 scenarios, two transient scenarios, and an analog scenario in their studies. Many researchers, however, did not have sufficient time or resources to use all of the scenarios. EPA asked the researchers to run the scenarios in the following order, going as far through the list as time and resources allowed:
ABOUT THESE APPENDICES
The studies contained in these appendices appear in the form that the researchers submitted them to EPA. These reports do not necessarily reflect the official position of the U.S. Environmental Protection Agency. Mention of trade names does not constitute an endorsement.
THE IMPACT OF CO2 AND TRACE GAS-INDUCED CLIMATE CHANGES UPON HUMAN MORTALITY
Laurence S. Kalkstein Center for Climatic Research University of Delaware Newark, DE 19716
Contract No. CR81430101
The objective of this study is to estimate changes in human mortality attributed to potential changes in climate. The major result is an estimation of the number of deaths attributed to the increased incidence of extreme weather episodes predicted by numerous climate change models.
The evaluation covers 15 cities around the country, and daily mortality data for 11 summer and winter seasons were extracted and standardized in a manner that facilitates intercity comparisons. The mortality totals were divided into total and elderly categories, and separate evaluations were developed for all causes of death and those causes considered to be "weather-related."
A summary of the results follows:
Global warming could have an enormous impact upon human mortality through the 21st century. If the population does not fully acclimatize, over 7000 deaths attributed to the increasingly harsh weather can occur in the metropolitan areas of our 15-city sample. This figure is more startling when it is considered that these numbers correspond to average summer conditions. An analog of the very hot summer of 1980 occurring in the 21st century will no doubt increase weather-induced mortality to a much higher number than 7000.
Although fully acclimatized predictions are more modest, some general increases are still expected even under these conditions. If it is assumed that people partially acclimatize (possibly the most realistic scenario), the increases in mortality are larger, and weather-related deaths may increase by four to five times over present levels during the summer.
The impact of inadvertent climatic changes upon the human population has long been the subject of speculation. Although a large body of literature is devoted to the impact of variable climate upon the socioeconomic sector, very little has been done to estimate how predicted changes in climate might affect the health of the general population.
The objective of this study is to estimate changes in human mortality attributed to predicted changes in climate due to increased concentrations of CO2 and other trace gases in the atmosphere. The result will be an estimation of the number of deaths attributed to the increased incidence of extreme weather episodes predicted by various climate change models.
The research presented here addresses two important priorities in climate/health studies. First, an assessment of CO2/trace gas-induced climatic changes upon humans represents a timely and necessary aspect of air pollution-health analyses. The additional evaluation of climatic impact, rather than air pollution concentration alone, represents an important addition to any pollution/health research. Second, it is imperative that government agencies understand the implications of long-term climatic change in an attempt to develop mitigation policies. If the environmental impacts of inadvertent climatic change are quantified, regulatory action may be implemented with greater efficiency.
The following is a specific list of items addressed by this research:
1. Fifteen cities will be evaluated (Table 1), and present-day analogs that duplicate their predicted future climate regimes will be developed.
2. Weather/mortality relationships for the elderly (greater than 65 years of age) and the total population will be determined.
3. Estimates of future mortality assuming climatic warming will be presented. These estimates will be based on future weather scenarios as predicted by three Goddard Institute For Space Sciences (GISS) Global Circulation Models (GCMs).
There are several unavoidable limitations to this study. First, this evaluation concentrates on urban mortality variations. The scarcity of rural mortality data will curtail our development of an accurate rural mortality assessment procedure. Unforeseen future population changes and/or regional population shifts might represent a second limitation. Third, analog cities, used to approximate the future climate of our target cities, may be very different from the target cities in terms of architectural or structural makeup. Thus the microclimate within the dwellings of analog and associated target cities may vary. Fourth, the numerous interrelationships between weather, pollution, social factors, morbidity, and mortality are tremendously complex, leading to sharp disagreements among scholars involving the differential impacts of weather on human health. This has historically discouraged the development of deterministic weather/mortality models, which leads to increased difficulty with interpretation of results.
Although no previous study has attempted to predict the impact of future weather changes on mortality, considerable work relating to present climate/mortality relationships has been reported (White and Hertz-Picciotto, 1985; Munn, 1986; Kalkstein and Valimont, 1987). For example, studies at the Centers for Disease Control (CDC) have identified a number of factors that may inhibit the onset of heat stroke, including the increased use of air conditioning, consumption of fluids, and living in well-shaded residences (Kilbourne et al., 1982). Some researchers have found that many causes of deaths other than heat stroke increase during extreme weather (Applegate et al., 1981; Jones et al., 1982). In addition, mortality attributed to weather seems to vary considerably with age, sex, and race, although there is disagreement among researchers in defining the most susceptible population group (Oechsli and Buechley, 1970; Bridger et al., 1976; Lye and Kamal, 1977).
The impact of cold weather is less dramatic than that of hot weather, although mortality increases have been noted during extreme cold waves (CDC, 1982; Fitzgerald and Jessop, 1982; Gallow et al., 1984; Kalkstein, 1984). Hypothermia is a major contributor to weather-related mortality in winter, but many other causes of death also increase including influenza, pneumonia, accidents, carbon monoxide poisoning, and house fires (National Center for Health Statistics, 1978).
A frequent criticism of these studies points to certain cultural adjustments through time that may have an impact on weather/mortality relationships, such as the lessened exposure of people to extreme weather owing to the increased use of air conditioning. Surprisingly, several studies indicate that these cultural adjustments may have a minimal impact. Ellis and Nelson (1978) have noted that during the past 30 years, mortality during heat waves in New York City has not changed significantly despite the increased use of air conditioning. Analysis by Marmor (1975) supports this finding, and his study covering a 22-year period implied that air conditioning may be decreasing excess mortality during initial summer hot spells only. Thus it is possible that people do not require direct exposure to hostile external environments to be negatively affected by these environments. The knowledge that external unpleasant conditions exist might be sufficient to contribute to negative reactions (Ulrich, 1984).
One of the major questions that must be addressed when evaluating the impact of long-term changes in weather on human health involves the importance of acclimatization which represents the increased ability of humans to withstand stressful conditions with repeated exposure. Several studies have evaluated acclimatization as a factor contributing to heat-related deaths. Gover (1938) reported that excess mortality during a second heat wave in any year will be slight in comparison to excess mortality during the first, even if the second heat wave is unusually extreme. Two possible explanations for this phenomenon are provided. First, the weak and susceptible members of the population die in the early heat waves of summer, thus lowering the population of susceptible people who would have died during subsequent heat waves. Second, those who survive early heat waves become physiologically or behaviorally acclimatized and hence deal more effectively with later heat waves (Marmor, 1975). Findings at the recent Williamsburg Conference on Susceptibility to Inhaled Pollutants support the second idea. Reactive subjects responded only at the beginning of the ozone season each year (spring and summer), and were generally not affected by exposure later in the year (fall). Rotton (1983) suggests that geographical acclimatization is also significant, and people moving from a cool to a subtropical climate will adapt rather quickly, often within two weeks. However, the population must still make behavioral and cultural adjustments (Ellis, 1972). Further support for geographical acclimatization is provided by Kalkstein et al. (1986), who noted that mortality increased dramatically during heat waves in northern cities, but no mortality increase was observed in southern cities even under the hottest conditions.
This report will describe the methodology used in the development of weather-related mortality predictions into the future. In addition, results of the empirical evaluation will be presented and interpreted, and an evaluation of the potential socioeconomic implications will be attempted.
A very detailed mortality data base is presently available from the National Center for Health Statistics (NCHS), which contains records for every person who has died in this country from 1964 to the present (NCHS, 1978). The data contain information such as cause of death, place of death, age, date of death, sex, and race. These data were extracted for the standard metropolitan statistical areas (SMSAs) of all the cities incorporated in this study for 11 years: 1964-66, 1972-78, and 1980 (during intervening years, a sizable amount of information was missing from many records). The number of deaths each day for each SMSA was tabulated and divided into total deaths and elderly deaths. Thus, the relative sensitivities for both categories could be determined, as weather probably exerts a differential influence upon mortality between categories.
Certain causes of death, deemed "weather-related," were factored out and evaluated separately using two procedures (Table 2). First, weather-related causes of death were subjectively identified after consultation with Dr. Melvyn Tockman, an epidemiologist from Johns Hopkins University, and Dr. Steven Parnes, head, Division of Otolaryngology, Albany Medical College. The medical experts examined a listing containing approximately 10,000 causes of death (Department of Health and Human Services, 1980) and identified the causes they considered to be directly or indirectly influenced by weather. Second, a more objective method to isolate weather-related causes was attempted by correlating annual fluctuations for every cause of death for the entire nation with population-weighted values of monthly mean temperature and precipitation. The population-weighted procedure for developing weather variables is used commonly be the National Oceanic and Atmospheric Administration (NOAA) to estimate national impacts of weather on society; refer to Warren and LeDuc (1981) for computational details.
There is conflicting evidence in the literature about the validity of factoring out weather-related causes of death. Many researchers continue to utilize total mortality figures in their analyses, as deaths from a surprisingly large number of causes appear to escalate with more extreme weather (Applegate et al., 1981; Jones et al. 1982). In an attempt to circumvent this apparent disagreement among researchers, weather-related and all causes categories ware evaluated separately in this study for the total and elderly mortality categories.
Although they are probably less meaningful for inter-regional comparison than standardized values, understandardized mortality values provide a better estimate of the total magnitude of weather's influence upon mortality. The unstandardized values are computed for each city by multiplying the death rate by the true population of the city's SMSA (using 1980 census data).
DETERMINATION OF WEATHER/MORTALITY RELATIONSHIPS
The procedural framework used in this study for developing and interpreting climate/mortality relationships is outlined in Figure 1, which should supplement the discussion in this subchapter. Prior to an evaluation of the effects of future warming on mortality, it is necessary to define the historical relationships between weather and mortality. This section first describes the procedure to develop historical relationships, and then outlines the procedure to evaluate future climate/mortality relationships.
Weather has been demonstrated to have some impact on daily mortality (Figure 2). During the heat wave of late July 1980 in New York City, deaths rose to over 50 percent above normal on the day with the highest maximum temperature (Kalkstein et al., 1986). Deaths among the elderly showed similar increases. In this study, daily changes in mortality were compared to 12 different weather elements which might have some influence on death rates (Table 3). One of these elements, a "time" variable (TIME), was also incorporated, which evaluated the intra-seasonal timing of the weather event. For example, it is hypothesized that a heat wave in August might have less of an influence than a similar heat wave in June, as the population would be unaccustomed to the June event. Thus, TIME simply assigns each day a number (e.g., June 1 is 1, June 2 is 2, July 1 is 31) representing its position in the summer (or winter) season.
Initial observations of daily deaths versus maximum temperature suggest that, in summer, weather has an impact on only the warmest 10-20 percent of the days; however, the relationship on those very warm days is impressive (Figure 3). Somewhat similar findings were uncovered for winter, and for certain SMSAs, the coldest 10 percent of days exhibited good weather/mortality relationships. Data were analyzed for the total and elderly mortality categories for each SMSA during summer and winter to compare the maximum temperature on the day of the deaths, as well as one, two, and three days prior to the day of the deaths to determine if a lag time exists between weather and the mortality response.
A unique aspect of this study involves the determination of a "threshold temperature," which represents that temperature beyond which mortality significantly increases (Kalkstein and Davis, 1985). The threshold temperature is calculated objectively by measuring the dissimilarity of mortality rates above and below a given temperature (refer to Kalkstein, in press, for a more detailed discussion). The threshold temperature for total deaths in New York City, for example, is 92deg.F (Figure 3), and mortality increases dramatically at temperatures above this level. This procedure can be repeated for winter, where the threshold temperature represents the temperature below which mortality increases.
Once the threshold has been established, a procedure named "all regression" is used to determine which combination of weather elements (listed in Table 3) produces the best models ("best" is defined as possessing the highest R value) for days beyond the threshold temperature (Draper and Smith, 1981). The next step involves choosing which regression model (for each combination of weather elements) best represents the historical relationships for that city. Complete multiple linear regressions were run for each model, which included regression diagnostics such as residuals plots and variance inflation factors (VIF) (SAS Institute, 1985). A high VIF indicates that two or more collinear independent variables are included in the model. When this is the case, one of the collinear variables is omitted from the model; the remaining variable explains a greater amount of the variance in mortality than the omitted variable. Because of this collinearity problem, CDH and maximum temperature are virtually never included in the same regression. The final selected model must meet the following criteria:
An "adjusted" R statistic was used in this study, which accounts for degrees of freedom in the regression model (Draper and Smith, 1981). Regression models possessing low degrees of freedom often exhibit inflated R values, and the use of an adjusted R statistic minimizes this problem. These guidelines insured some degree of quality control in the regression modeling, as thousands of regressions were computed to determine the best models.
Applying algorithms to weather scenarios
With historical relationships established, the next step is an attempt to estimate changes in mortality which might occur with predicted climatic warming. This study utilizes three GCM transient runs provided by NCAR/EPA (Jenne, 1987), and future predictions of climate have been developed for the cities in this study. The three runs are GISS transient A1 (covering a 17-year period 30 years after the base period), GISS transient A2 (covering a 17-year period 60 years after the base period), and GISS 2XCO2.
As described earlier, the base period for mortality includes a 17-year period extending from 1964 to 1980 (note that only 11 years of mortality data were available through this period). Thus the GISS transient A1 scenario will estimate mortality for the period 1994-2010. The GISS transient A2 scenario will estimate mortality for the period 2024-2040. The GISS 2XCO2 scenario will assume double CO2 conditions occurring during the base period 1964-1980. New mortality estimates for each city were created for each scenario by using the algorithms developed from the historical data.
When measuring the impact of warming on future mortality, the question of acclimatization must be considered. Will people within each city respond to heat as they do today? Or will their reactions be similar to those of people who presently live in hotter climates? There is much disagreement in the literature concerning human acclimatization to changing weather. Some research indicates that acclimatization responses are very rapid (Marmor, 1975; Rotton, 1983), others think it is a much slower process (Kalkstein and Davis, 1985; Ellis, 1972), and a few imply that virtually no acclimatization occurs at all (Steadman, 1979). It is obvious that the full range of possibilities must be examined. First, the historical algorithms for each city that were developed from the previously described multiple regression procedure were applied to the three future weather scenarios. The mortality increases estimated from this procedure imply no acclimatization, as an assumption is made that people will respond to heat in the future in much the same way that they do today. Second, analog cities were established for each city evaluated to account for full acclimatization. For example, the use of one of the GISS scenarios to predict future weather in New York City will produce a regime which will approximate another city's present weather in the U.S. Thus using the GISS transient A2 scenario to predict New York City's summer weather for the period 2024-2040 yields a weather regime approximating that of Kansas City, MO, today. Since Kansas City residents are fully acclimatized to this regime, the weather/mortality algorithm developed for Kansas City can be utilized for New York City to account for full acclimatization if New York's weather approximates that predicted by GISS transient A2.
One potential problem that arises from the utilization of weather analogs involves the possible difference in racial and socioeconomic composition between the evaluated city and its analog. The utilization of standard city death totals minimizes this problem.
Selection of analog cities
Present-day analogs to account for full acclimatization were selected for each city evaluated in this study for the three GISS transient scenarios for summer and winter, and mortality models were created for them using the procedure described earlier. Each analog was selected from a pool of almost 50 cities around the country (Table 4), representing virtually every weather regime found in the continental United States. This large sample size of cities permits an inter-regional evaluation of human weather/mortality responses on a scale larger than ever before.
The analog cities for summer were determined by comparing three weather variables (mean maximum temperatures, mean minimum temperatures, and the mean number of days with maximum temperatures over 90deg.F) for the three summer months (June, July, and August) . This process was repeated using mean maximum temperatures, mean minimum temperatures, and the mean number of days with maximum temperatures below 32deg.F for the three winter months (December, January, and February) to determine winter analog cities. The statistical procedures used to determine the closest analog are described in detail within another manuscript (Kalkstein, in press), and these techniques produced analogs which were very close to the estimated future climate of the target cities.
Figure 4 illustrates the hypothetical differences expected in mortality with full, partial, and no acclimatization. It is probable that the acclimatized models (based on warmer city analogs) will show smaller increases in mortality than the unacclimatized models since residents have already adapted to the increased warmth. Thus, for warming scenarios of seven or more degrees, the differences in predicted deaths between full and no acclimatization situations may be very large (area hatched between lines 1 and 2). It is obviously necessary to consider a situation where partial acclimatization will be a likely result. It is possible that people will fully acclimatize to the increased warmth, but even if the population is capable of full behavioral acclimatization, it will take many years for the physical structure of the city to conform to the hotter climate (eg. total air conditioning of dwellings, construction of new structures with heating/cooling systems capable of meeting the demands of the new climate). It is improbable that people will not acclimatize at all to the increased warmth, as a majority of previous research rejects the notion that the population cannot acclimatize at least partially to changing weather conditions. For example, although it appears probable that people might adapt very well to the predicted warming, the urban structures will not be changed within the next 70 years to reflect the type of architecture best suited for the warmer climate. Today most poor inner city southerners reside in small single family dwellings which have adequate ventilation and often renective aluminum roofs. However, their counterparts in large northeastern and midwestern cities live in row homes constructed of brick and possessing black roofs which readily absorb solar radiation. These structures become much holter during extreme weather, and it is doubtful that the architectural makeup of these northern urban areas will change quickly enough to adapt to the predicted increasing warmth. Thus, a third estimate reflecting partial acclimatization (reflecting full acclimatization among the population but little change in the urban infrastructure) is included in Figure 4, with values intermediate between the full and no acclimatization possibilities.
In certain cases, it is possible that no extra deaths will be predicted for full acclimatization, as residents are conditioned to hot weather. For example, in Jacksonville, Florida, heat waves appear to produce no extra deaths (Figure 5); the relationship is so poor that it is virtually impossible to determine a threshold temperature.
Threshold temperatures were established for each of the 15 cities for total deaths and deaths among the elderly for the summer and winter seasons (Table 5). The threshold temperatures varied predictably between cities, and in the summer the southern and southwestern cities demonstrated the highest threshold temperatures. Similar findings were uncovered for threshold temperatures in the winter. Although there was considerable variation in threshold temperatures between cities, very little variation was detected between the two death categories within cities. For example, the summer threshold temperature in Atlanta was 94deg.F for the total and elderly death categories. It does not appear that any particular age group exhibits a distinctively high or low threshold temperature.
Very little lag time was noted between the weather mechanism and associated mortality response for all cities in summer (Table 5). In most cases, the mortality response occurred on the same day as the responsible weather mechanism (lag time = 0 days), although one-day lag responses were detected in some of the models. In winter, however, longer lag times were often noted, and for some cities the mortality response occurred three days after the responsible weather mechanism.
Once the threshold temperatures were established, the "all regression" procedure was performed for days above the threshold to determine the weather elements having the greatest impact on present-day mortality in each city. A large number of statistically significant models were uncovered for both seasons and for both death categories. Many of the relationships were more impressive than expected, with R values frequently exceeding 0.250, especially in summer. During the warm season the most important weather variables proved to be CDH and TIME, which were directly and inversely related to mortality, respectively. The inverse TIME relationship suggests that the timing of the weather event is often as important as the magnitude. Hot weather occurring early in the season appears to have a more devastating impact than similar weather occurring in August, implying that acclimatization to hot weather might occur rapidly within a season. The importance of CDH and relative insignificance of maximum temperature are also noteworthy. This suggests that the intensity of the heat event might be of lesser importance than the duration of the event. Winter relationships appear to be substantially weaker than those in summer, and thermal variables (MAXT, MINT, HDH) appear to be much less important. Refer to Kalkstein, in press, for a more complete explanation.
Although similar numbers of statistically significant models were found for "all causes" and "weather-related" causes of death, the level of significance varied considerably between the two in both seasons. The proportion of statistically significant models with R values exceeding 0.250 was much higher for the all causes category in summer. When significant all causes and weather-related models were uncovered for the same category within a city, the all causes model usually possessed the higher R value. This is consistent with the findings of past studies on summer mortality as described previously, and the results of the all causes, rather than weather-related, models will be used exclusively here for the summer season. The winter season produced opposite results, and the "weather-related" category normally possessed a higher R value than its "all causes" counterpart; thus the former will be emphasized here for the winter season.
The historical evaluation of mortality produced very interesting geographical distributions, indicating that the impact of weather on mortality varies considerably on an inter-regional level. For example, people living in the northern part of the country appear much more susceptible to heat-related mortality than those in the South. For a full discussion of these inter-regional variations, refer to Kalkstein (in press).
The algorithms developed through the regression analyses were employed to estimate the number of deaths attributed to the weather for each of the 15 evaluated cities. In addition, the use of future weather scenarios assuming long-term climatic warming permitted application of the algorithms to predict future trends in mortality. Both unacclimatized (using the historical algorithm for a city to predict future mortality in that city) and acclimatized (using the algorithms for analog cities to predict future mortality) estimates will be presented, using the GISS warming scenarios described earlier. A statistically significant model at the 0.05 level or better was required if any estimate was attempted. If no statistically significant model was uncovered, it was assumed that weather has no impact on mortality, and a value of 0 deaths was assigned.
Analog cities were developed for summer and winter for each city to develop acclimatized predictions (Tables 6-7). In certain cases, the change in climate was so small (i.e., the GISS Transient A1 run) that a city could be an analog of itself (refer to the Dallas summer analog).
Initially, an estimate of present-day mortality attributed to weather was attempted by utilizing the created regressions to develop historical algorithms for the 15 cities (Table 8). Using the algorithms, mortality was estimated for every day having temperatures above the threshold in the 11-year sample. These mortality estimates were compared to the 11-year mean mortality for the month, and the difference between the estimate derived from the algorithm and the long-term monthly mean was considered to be the days weather-induced mortality. For example, assume that June 1, 1980, experienced temperatures above the threshold in Cincinnati. The historical algorithm estimated that 130 deaths occurred in Cincinnati on that day. The 11-year June mean mortality for Cincinnati is 115 deaths. Thus, 15 deaths were attributed to weather in Cincinnati on June 1.
The deaths attributed to weather were summed daily for all days above the threshold temperature for each month over the 11-year period. The average monthly mortality was extracted to represent the expected mortality attributed to weather during an average month in the evaluated city. Seasonal averages were derived from the monthly values.
For total deaths, it appears that New York experiences the greatest mortality totals, amounting to 320 deaths from all causes during an average summer attributed to weather. Chicago and Philadelphia ranked second and third, respectively, both averaging well over 100 standard city deaths during an average summer. Lowest values were found in New Orleans and Oklahoma City, where 0 deaths were attributed to the weather. During an average summer season over 1150 standard city deaths attributed to weather are estimated to occur in the SMSAs of the 15 cities. Of course this figure is much higher during extremely hot summers.
July is by far the most significant month, accounting for approximately two-thirds of all mortality in summer attributed to weather. Although August is generally much warmer than June in most of the cities, mortality for both months is quite similar, and in some cases, June's predicted mortality exceeds that of August. This reflects the importance of within-season acclimatization (Kalkstein, in press), indicating that early-season (June) heat waves generally exert a greater impact than late-season (August) heat waves of similar magnitude.
The evaluation of predicted future unacclimatized, acclimatized, and partially acclimatized mortality in summer based on the warming scenarios yielded interesting results (Table 9a, b, c). Virtually all cities exhibited an increase in mortality as the scenarios became warmer for both total and elderly deaths, and the total unacclimatized mortality estimate exceeded 7400 attributed to weather during an average summer under 2XCO2 conditions. This is almost seven times higher than the number of deaths attributed to weather under average summer conditions today. However, the magnitude of the increase varied significantly between cities. New York continued to be the city with the greatest number of deaths through most of the warming scenarios, with over 1700 deaths attributed to weather during an average summer under 2XCO2 conditions. Other cities with predicted rapid rises in mortality with future warming were Los Angeles, Memphis, Philadelphia, and New York. New Orleans and Oklahoma City, with statistically non-significant mortality models, will probably not be affected significantly by future warming.
Estimates of future acclimatized mortality indicate that predicted warming might lessen weather-induced mortality in certain cities if acclimatization proceeds rapidly. In Atlanta, Detroit, Los Angeles, Memphis, New Orleans, and St. Louis virtually no weather-induced deaths were predicted if the residents of these cities acclimatize. This is particularly surprising for St. Louis, where unacclimatized future mortality estimates and historical estimates were very high. In approximately half of the cities mortality totals declined as the scenarios became warmer, and in cities where increases did occur, they were much smaller than those uncovered under no acclimatization. The 15-city acclimatized totals showed a significant drop in mortality under conditions predicted in the early 21st century (GISS Transient A1), followed by a relatively modest rise (when compared to unacclimatized results) for the mid 21st century (GISS Transient A2). A more significant rise in acclimatized mortality is noted under 2XCO2 conditions, with values approximately double todays weather-induced mortality.
Estimates of future total mortality assuming partial acclimatization were calculated by computing the mortality totals exactly halfway between full and no acclimatization values. With partial acclimatization about 3800 standard city deaths are predicted for the 15-city sample under 2XCO2 conditions. This represents a substantial increase over present-day mortality estimates attributed to weather, and although these values should be viewed cautiously, indications are that mortality will increase substantially if partial acclimatization takes place.
Mortality estimates for the 65 and older category were very similar to the total mortality estimates. Substantial increases in unacclimatized mortality were noted in virtually all cities. Partial acclimatized values were less than unacclimatized, but a substantial increase of almost 2400 deaths above present conditions was noted for the 15-city sample under 2XCO2 conditions. Much like the results from the total death category, acclimatized mortality predictions for the 65 and older group were much smaller than the unacclimatized values.
The great disparity between acclimatized and unacclimatized predictions is troubling but not surprising. If people do not acclimatize to predicted warming, weather-induced mortality will rise at a very rapid rate because (1) the number of days exceeding the threshold temperature will increase, providing a larger group of total days when weather-induced mortality will be a factor; and (2) since CDH and maximum temperature are directly related to mortality in almost all the summer models, mortality will necessarily increase as the magnitudes of these weather variables increase. If acclimatization is complete, increasing warmth will produce a much smaller rise in mortality, which parallels the response of people today who reside in hot climates. In almost all of the southern cities in our 15-city sample, present-day weather-induced mortality was estimated to be lower than in the northern cities. These southern cities represent analogs of expected climate in the northern cities, and the lower present-day mortality rates indicate that southerners have indeed acclimatized to the frequent hot weather episodes that occur.
The following winter mortality predictions are for weather-related causes of death only. Mortality predictions were also constructed for all causes of death, but these models generally possessed lower R2 values than their weather-related counterparts. Thus a decision was made to use the weather-related predictions to ensure a more accurate result.
Present-day estimates of winter mortality indicate that weather-induced deaths are much less important in winter than in summer (Table 10). The city with the greatest predicted number of winter deaths, New York, averages only 56 standard city deaths per season. About half of the cities in the 15-city sample produce 10 or less weather-induced winter deaths during an average season. In some cases, statistically significant models were developed for winter for these cities, but the relationships were relatively weak, yielding mortality predictions that approached 0.
For total deaths, the four top ranking cities are all northern or midwestern locations. New York, St. Louis, Chicago, and Kansas City account for 170 of the 296 total weather-induced deaths which occur in the 15-city sample during a typical winter. The occasional severe cold waves encountered here obviously have a significant impact. Surprisingly, Minneapolis mortality totals are quite low compared to Midwestern cities located farther south. It is possible that residents of Minneapolis are so conditioned to winter cold that they are behaviorally adapted and take the proper precautionary steps to avoid cold weather impacts. The five cities which record the lowest number of weather-induced winter deaths are located in milder climates. In a previous study Los Angeles and San Francisco never exhibited winter conditions severe enough to produce weather-induced deaths (Kalkstein and Davis, 1985). However, the other southern cities recording few deaths do experience occasional cold waves, especially Atlanta, Memphis, and Oklahoma City. Perhaps the cold episodes are of too short a duration or are not extreme enough to create problems.
The sensitivity of the elderly to winter-induced mortality is surprisingly low, and only 157 elderly deaths attributed to weather are predicted during an average winter. Once again, the cities with the greatest predicted elderly mortality are located in northern or midwestern locations, while a majority of the low ranking cities are found in milder climates.
These results seem to counter some of the findings in the summer models regarding the relative impact of weather. During summer, many of the cities exhibiting the highest weather-induced mortality totals were located in the northern United States, where extreme heat occurs less often. There was a clear suggestion that people react to weather in a relative, rather than absolute, fashion in summer, as cities with low mortality estimates were found in some of the hottest regions in the country. During winter, the findings suggest that this relative response is less pronounced, as many of the cities with the highest winter mortality totals are located in regions with relatively severe winter climates, such as Chicago, Kansas City, New York, and St. Louis. Thus, the impact of weather on mortality in winter appears to be more absolute.
In both the total and elderly predictions for winter, February exhibits a surprisingly low number of winter weather-induced deaths. This occurs although February is normally colder than December; once again, an element of within-season acclimatization appears to be present and people are more sensitive to early cold waves than to later cold waves of similar (or greater) magnitude. Nevertheless, the TIME variable is less important in winter than in summer, and within-season acclimatization is less pronounced in winter.
Estimates of future winter mortality based on the warming scenarios were developed, producing unacclimatized, partially acclimatized, and acclimatized values for total deaths and elderly deaths. The results were very different than those uncovered in summer (Table 11a, b, c). For total deaths assuming no acclimatization, a decline in weather-induced mortality was noted for most of the cities in the 15-city sample as the scenarios became warmer. Of the 15 cities, 10 registered five deaths or less under 2XCO2 conditions. These results differ markedly from summer, where unacclimatized total mortality showed a dramatic rise as the scenarios became warmer. The winter results suggest that a warmer climate will correspond with a lesser number of days below the threshold temperature, and the potential for weather-induced mortality will decrease correspondingly.
If full acclimatization is assumed, the impact of future warming on mortality also creates a pronounced drop. The lack of acclimatized mortality with increased warmth is consistent with previous winter findings, and mortality in many of the southern cities (which constitute the analogs for the warmer weather scenarios) is not significantly affected by winter weather. Values for partial acclimatization are similar to those for full acclimatization, and very few deaths are predicted under 2XCO2 conditions. Elderly response with climatic warming is very similar to the total population response.
The objective of this study was to estimate changes in human mortality attributed to predicted changes in climate due to increased concentrations of CO2 and other trace gases in the atmosphere. The major result was an estimation of the number of deaths attributed to the increased incidence of extreme weather episodes predicted by numerous climate change models.
The evaluation covered 15 cities around the country, and daily mortality data for 11 summer and winter seasons were extracted. The mortality totals were divided into total and elderly mortality categories, and separate evaluations were developed for all causes of death and those causes considered to be "weather-related."
A summary of the results follows.
1. Predictions of weather-induced mortality occurring during summer were attempted for the 15 cities exhibiting significant weather/mortality relationships. It is estimated that approximately 1150 deaths occur during an average summer season in the SMSAs of the 15 cities. New York City, Chicago, and Philadelphia ranked first, second, and third, respectively, and each city averaged well over 100 standard city deaths per summer. The five highest ranking cities were all found in the Midwest or Northeast. The five lowest ranking cities were found in the South, with New Orleans and Oklahoma City experiencing virtually no deaths attributed to
2. Predicted future unacclimatized mortality in summer rose rapidly as the scenarios become warmer. The total mortality estimate exceeded 7400 deaths attributed to weather during an average summer under 2XCO2 conditions. The magnitude of the increase varied significantly between cities. New York exhibited the greatest number of deaths throughout all the warming scenarios. Other cities with predicted rapid rises in mortality with future warming were Los Angeles, Philadelphia, and St. Louis. New Orleans and Oklahoma City were cities least affected by predicted warming.
3. Predicted future acclimatized mortality in summer indicated that warming might lessen weather-induced mortality in about half of the 15-city sample if acclimatization is complete. A more modest rise in mortality is predicted for the 15-city sample if full acclimatization occurs; however, weather-induced mortality is predicted to double over present levels if acclimatization is complete. Acclimatized mortality estimates for the 65+ age category were very similar to the total mortality estimates. By the design of the model, the relatively small rise in acclimatized mortality paralleled the response of people today who reside in hot climates. Southern cities represented analogs of expected climate in northern cities, and these cities exhibited fewer numbers of weather-induced deaths in summer. Estimates of future mortality assuming partial acclimatization (values midway between full and no acclimatization) were also developed, and these indicated sizable increases in mortality as the weather became warmer.
4. Present-day estimates of winter mortality indicated that weather-induced deaths were much less important in winter than in summer. The results differed from those uncovered in summer regarding the relative impact of weather. Most of the estimated winter deaths occurred in regions with relatively severe winter climates, while the smallest numbers of deaths were found in mild weather cities. It appeared that the impact of weather in winter is more absolute, while the impact of weather in summer tends to be relative.
5. Winter unacclimatized, partially acclimatized, and acclimatized predictions indicated that sharp drops in mortality are expected if the weather becomes warmer. The unacclimatized results differed from those uncovered in summer, when dramatic rises in mortality were predicted. The unacclimatized drop in winter may be related to fewer numbers of days below the threshold.
It is quite obvious that predicted warming could have an enormous impact upon human health through the 21st century. If the population does not acclimatize, over 7000 deaths attributable to the increasingly harsh weather can be expected in the metropolitan areas of our 15-city sample. This figure is more startling when it is considered that these numbers correspond to average summer conditions. An analog of the very hot summer of 1980 occurring in the 21st century will no doubt increase weather-induced mortality to a much higher number than 7000.
Although fully acclimatized predictions are more modest, some general increases are still expected even under these conditions. If it is assumed that people will partially acclimatize (possibly the most realistic scenario), the increases in mortality are more impressive, and weather-related deaths will increase by four to five times over present levels. Thus it appears that specific policy decisions are necessary to prepare for a significant rise in human mortality if the warming scenarios come to pass.
1 Although the information in this report has been funded wholly or partly by the U.S. Environmental Protection Agency under cooperative agreement number CR81430101 at the Center for Climatic Research, University of Delaware, it does not necessarily reflect the Agency's views, and no official endorsement should be inferred from it.
Applegate, W.B., Runyan, J.W., Jr., Brasfield, L., Williams, M.L., Konigsberg, C., and Fouche, C. Analysis of the 1980 heat wave in Memphis. Journal of the American Geriatrics Society. 29:337-342, 1981.
Bridger, C A., Ellis, F.P., and Taylor, H.L. Mortality in St. Louis, Missouri, during heat waves in 1936, 1953, 1954, 1955, and 1906. Environmental Research. 12:38-48, 1976.
Centers for Disease Control. Exposure related hypothermia deaths--District of Columbia 1972-1982. In: Morbidity and Mortality Weekly Report. December, 1982. pp. 31-50.
Department of Health and Human Services. The International Classification of Diseases, 9th Revision. Clinical Modification, Volume 3: Procedures. Second edition, U.S. Dept. of Health and Human Services, September, 1980.
Draper, N., and Smith, H. Applied Regression Analysis. Second Edition. John Wiley and Sons, 1981. 673 pp.
Ellis, F.P. Mortality from heat illness and heat-aggravated illness in the United States. Environmental Research. 5:1-58, 1972.
Ellis, F.P., and Nelson, F. Mortality in the elderly in a heat wave in New York City, August, 1975. Environmental Research. 15:504-512, 1978.
Fitzgerald, F.T., and Jessop, C. Accidental hypothermia: A report of 22 cases and review of the literature. Advanced Internal Medicine. 27:127-150, 1982.
Gallow, D., Graham, T.E., and Pfeiffer, S. Comparative thermo-regulatory responses to acute cold in women of Asian or European descent. Human Biology. 56:19-34, 1984.
Gover, M. Mortality during periods of excessive temperatures. Public Health Reports. 53:112-1143, 1938.
Jenne, R. "GISS Transient Runs." NCAR/EPA Doc. 4, U.S. Environmental Protection Agency, Washington, DC, 1987, 3 pp.
Jones, T.S., Liang, Kilbourne, E.M., Grifrm, M.R., Patriarca, P A., Wassilak, G.G., Mullan,R.F., Herricck, R.F., Donnel, H.D., Jr., Choi, K., and Thacker, S.B. Morbidity and mortality associated with the July 1980 heat wave in St. Louis and Kansas City. Journal of the American Medical Associations. 247:3327-3330, 1982.
Kalkstein, L.S. The impact of winter weather on human mortality. In: Climate Impact Assessment: United States. U.S. Department of Commerce. December, 1984. pp. 21-23.
Kalkstein, L.S. The impact of CO2 and trace gas-induced climate changes upon human mortality. U.S. EPA Office of Research and Development Report, in press.
Kalkstein, L.S., and Davis, R.E. The development of a weather/mortality model for environmental impact assessment. In: Proceedings of the 7th Conference of Biometeorology and Aerobiology, 1985. pp. 334-336.
Kalkstein, L.S., Davis, R.E., Skindlov, J.A., and K.M. Valimont. The impact of human-induced climatic warming upon human mortality A New York City case study. Effects of Changes in Stratospheric Ozone and Global Climate, Volume 3: Climate Change. U.S. Environmental Protection Agency, October, 1986.
Kilbourne, E.M., Choi, K., Jones, T.S., Thacker, S.B., and the Field Investigation Team. Risk factors for heatstroke: A case-control study. Journal of the American Medical Association. 247:3332-3336, 1982.
Lilienfeld, A.M., and Lilienfeld, D.E. Chapter 4: Mortality statistics. Foundations of Epidemiology. Oxford University Press, 1980. pp. 51-65.
Lye, M., and Kamal, A. The effects of heat wave on mortality rates in elderly patients. The Lancet. 1:529-531, 1977.
Marmor, M. Heat wave mortality in New York City, 1949 to 1970. Archives of Environmental Health. 30:131-136, 1975.
Mausner, J.S., and Bahll, A.K. Chapter 7: Measures of morbidity and mortality. Epidemiology: An Introductory Text. W.B. Sanders Co., 1974.
Munn, R.E. The value of climatological information in assessments of the state of human health. Leningrad, U.S.S.R. In: Proceedings of the WMO/UNEP/WHO Symposium on Climate and Human Health. International Institute for Applied Systems Analysis, Luxenburg, Austria, 1986.
National Center for Health Statistics. Standardized Micro-Data Tape Transcripts. U.S. Department of Health, Education, and Welfare, Washington, DC, 1978.
Oechsli, F.W., and Buechley, R.W. Excess mortality associated with three Los Angeles September hot spells. Environmental Research. 3:277-284, 1970.
Rotton, J. Angry, sad, happy? Blame the weather. U.S. News and World Report. 95:52-53, 1983.
SAS Institute Inc. SAS User's Guide: Statistics; Version 5 Edition. SAS Institute Inc., 1985. 956 pp.
Steadham, R.G. The assessment of sultriness: Part I: A temperature-humidity index based on human physiology and clothing science. Journal of Applied Meteorology. 18:861-873, 1979.
Ulrich, R.S. View through a window may influence recovery from surgery. Science. 224:420-421, 1984.
Warren, H.E., and LeDuc, S.K. Impact of climate on energy sector in economic analysis. NOAA/Center for Environmental Assessment Services, Environmental Data and Information Service: United States. 1981. pp. 1431-1439.
White, M.R., and Hertz-Picciotto, I. Human health: Analysis of climate related to health. In: Characterization of Information Requirements for Studies of CO2 Effects: Water Resources, Agriculture, Fisheries, Forests, and Human Health. M.R. White, ed., DOE/ER-02326, CO2 Res. Div., U.S. Dept. of Energy, Washington, DC, 1985, pp. 172-205.
Willmot, C.F., Ackleson, S.G., Davis, R.E., Feddema, J.J., Klink, K.M., Legates, D.R., O'Donnell, J., and Rowe, C.M. Statistics for the evaluation and comparison of models. Journal of Geophysical Research. 90:8995-9005, 1985.