1. Introduction

López and Holle (1996) have examined the long-term fluctuations in the number of lightning deaths and injuries from 1959 to 1990 for the contiguous United States. After taking into consideration the population increase, they found an overall trend amounting to a 30% reduction in casualties (deaths and nonfatal injuries together) during the period. They speculated that the trend resulted from improved forecasts and warnings, increased education efforts of the public, and socioeconomic changes.

Superimposed on the overall downward trend in casualties were fluctuations of one or two decades in duration. López and Holle (1996) also presented evidence that these oscillations are climatologically related. The patterns of those fluctuations were parallel to nationwide changes in thunderstorm-day frequencies, cyclone frequencies, and surface temperature values.

The data used by López and Holle (1996) covered only one-third of the century (32 yr) and thus it was not possible to detect long-term changes. In addition, the authors used information obtained only from Storm Data, a monthly publication of the National Oceanic and Atmospheric Administration (NOAA) that describes damaging and severe weather throughout the year. López et al. (1993) reported that Storm Data underestimated lightning deaths by at least 28% and underestimated lightning injuries requiring hospitalization by at least 42% in the state of Colorado during a 12-yr period.

It would be more desirable to use a longer and more reliable record of lightning casualties in the United States to study long-term trends and fluctuations and to see if the temporal patterns of these fluctuations correspond to similar changes in climatological parameters. The existence of such a record has recently come to our attention. It provides reliable lightning deaths statistics for 92 yr starting at the beginning of the century. We document this record and show the existence of an overall exponential reduction in the number of deaths and of residual fluctuations that match similar fluctuations in the number of thunderstorm days and surface temperature since 1900.

2. Sources of lightning-caused deaths data

Annual death statistics have been compiled and published by the federal government since 1900 with lightning as one of the causes of death. The Bureau of the Census established a national registration area for deaths that included states, the District of Columbia, and large cities having efficient systems for the registration of vital statistics. In 1900, there were 10 states and several cities included in this expanding area that annually reported copies of records filed in their vital statistics offices to the Bureau of the Census. The number of states in the registration area gradually increased and by 1930 all states except Texas were included. In 1935, all of the contiguous United States territory was covered. Table 1 shows the number of registration states reporting mortality data and the population represented by those states. For this study, only data from the contiguous states and the District of Columbia have been used. Data from isolated cities not included in a registration state were not used since it is difficult to obtain good estimates of the population covered. An annual report called Mortality Statistics (Bureau of the Census 1903–39) was published, covering data up to 1936. For data from 1937 to the present, the series has had the name Vital Statistics of the United States (Bureau of the Census 1940–46; Public Health Service 1947–1997). The reports for 1900 to 1944 were published by the Bureau of the Census and for 1945 to the present by the Public Health Service. At this time, the compilation is prepared by the National Center for Health Statistics, Centers for Disease Control and Prevention, and the Public Health Service of the U.S. Department of Health and Human Services.

3. Changes in yearly lightning deaths

The number of lightning deaths reported since 1900 by the Bureau of the Census and the Public Health Service is presented in Table 1. The number of deaths for the registration states and the District of Columbia (DC) until 1935, and for all of the contiguous United States and DC thereafter, are included. The data are portrayed graphically in Fig. 1 where the connected dots represent the yearly number of deaths. The yearly population of the reporting states and DC is also given in Table 1 and indicated by the solid line in Fig. 1. The dashed line represents the total population of the contiguous United States and DC.

During the first 20 yr of the record, the number of deaths increased from less than 100 to about 450 yr⁻¹. This sharp increase could be attributed in part to the increase in the population reporting to the Bureau of the Census and in part to the inclusion of more rural states in the registration area. However, considering the magnitude of the increase in the population compared to the magnitude of the increase in the number of deaths, and considering that in the early decades of the century the rural population was large even in more urbanized states, the increase in the number of deaths appears exceptionally large. During the 1920s and 1930s the number of deaths attributed to lightning was about 400 yr⁻¹, whereas recently it has been consistently less than 100 yr⁻¹. Since 1944, there has been a persistent drop in the number of deaths. This drop in the last half of the record is even more dramatic when the increase in population during the period, from less than 150 million to more than 250 million, is considered.

It would be important to consider what factors are responsible for the shape of this lightning death distribution during the century. To start, the effect of the increase in the population can be taken into account by normalizing the death series by dividing by the population of the reporting states. This has been done in Fig. 2, where the number of deaths per million people are presented. The actual numbers are also given in Table 1. In the earlier part of the series, the year-to-year fluctuations are relatively large, probably due to the random inclusion of states with different demographic and climatic conditions. However, after 1925 the fluctuations are consistently much smaller and regular, and decrease as the number of deaths per million people decreases.

There is, however, a large drop between 1944 and 1945. The number of deaths in the 5 yr before that time had been an average of 390 yr⁻¹, whereas in the next 5 yr starting with 1945 they dropped to an average of 268 yr⁻¹ (2.88 and 1.85 in the number of deaths per million people). That drop coincides with a change in the responsibility for collecting mortality data from the Census Bureau to the Public Health Service and a new tendency in the classification of causes of death. The new methodology emphasized coding for the underlying medical antecedent cause of death. This new coding practice would have made it very difficult to include indirectly caused lightning deaths, for example, deaths from burns received in house fires started by lightning strikes. After a careful examination of the history of mortality data collection in the United States, and the changes in reporting and coding practices, it has been concluded that the drop was caused artificially by changes in classification methodology.

The dramatic decline in reported lightning deaths since 1940 has been noticed by several researchers. Weigel (1976) attributed the drop to changes in coding procedure when indirect lightning deaths such as those resulting from fires started by lightning were coded to other causes rather than lightning. Based on the testimony of a former National Weather Service climatologist, he assigns the change in coding practices to 1953. Mogil et al. (1977) also noticed the decline and Weigel’s explanation but pointed out that the decline began prior to 1953. Duclos and Sanderson (1990) compared Public Health Service data for 1972–84 with medical examiner reports from North Carolina. They found that 8 out of 64 medical examiner reports that had lightning as a contributing cause did not appear in the Public Health Service database. Recently, Brigham (1995) commented on the absence of secondary lightning deaths in Public Health Service statistics and estimated that if lightning-related fire deaths were coded as lightning, the total number of lightning-related deaths would increase by about 20% yr⁻¹.

To make the two parts of the normalized series compatible, a polynomial curve was fit to the section from 1900 to 1944 and another to the section from 1945 to 1991. The difference between the extrapolated values to the middle of 1944 from both curves was 0.45 deaths per million people. This difference is equivalent to 17.4% of the extrapolated deaths according to the earlier series. The normalized values of the series from 1945 to 1991 were adjusted upward assuming that the reported values were 17.4% lower than what the true values should have been. In other words, it was assumed that of all possible yearly lightning deaths, 17.4% were indirect and therefore did not appear in the reported data. The resulting series is shown in Fig. 3. The adjusted normalized series indicates an overall decrease during the century in the number of deaths per million people from a high of more than 6 to a low of about 0.4. The dashed curve represents an exponential function fitted to the normalized death values. The correlation coefficient of the fit is 0.88.

4. Demographic changes and lightning deaths

It is important to consider the reason for the exponential decrease in the number of normalized deaths due to lightning. In examining detailed accounts of lightning deaths and injuries before the turn of the century (Kretzer 1895), it was noticed that a large number of deaths occurred in a rural setting where people were more exposed to the lightning threat. Also, it is well known that the number of people in rural areas has dramatically decreased with time. Figure 4 shows the percent of the population living in rural areas at the time of each decennial census since 1890. The percent of the rural population has decreased from a maximum of about 60% in 1900 to about 25% in 1990. An exponential curve has also been fitted to this demographic data and is shown in the figure as a dashed curve. The correlation coefficient was 0.99 in this case. The only significant departures from an exponential decay in the rural population are a slowing of the trend in the 1930s and early 1940s, probably as a result of the Great Depression, and a compensating acceleration of the trend in the 1950s and 1960s in the postwar era.

The rural population curve is shown in Fig. 5 superimposed on the adjusted normalized lightning-death curve. A remarkable agreement in terms of the overall trends can be clearly seen. So it appears that the secular exponential decrease in the population-adjusted lightning deaths is closely related to the exponential reduction of the rural population. The long-term decrease in lightning deaths has been previously noticed by several authors and a decrease in the rural population has been hypothesized as a factor, together with improvements in electrical systems, medical treatment, education, and meteorological warnings (Weigel 1976; Mogil et al. 1977; Duclos and Sanderson 1990; López et al. 1995; Brigham 1995; López and Holle 1996). Many of these other factors, however, are also related to the decrease in rural population due to emigration to cities or enhanced urbanization of rural areas, thus the strong agreement between the trend in lightning deaths and rural population.

5. Decadal fluctuations in lightning deaths

A closer inspection of the normalized deaths series in Fig. 3 reveals that, superimposed on the overall exponential decrease trend, there are suggestions of oscillations of 2 to 3 decades in duration. These oscillations can be considered more carefully by removing the exponential trend of the series. Figure 6 shows the number of deaths per million people for each year as departures from the value given by the exponential curve. There are three major oscillations that decrease in amplitude with time (1900–20, 1920–50, and 1950–80). The obvious question is, what caused these oscillations? López and Holle (1996) have suggested that similar fluctuations in lightning casualties from 1950 to 1990 are related to climatic factors. It would be desirable to directly compare the residual lightning death series of Fig. 6 with a century-long series of lightning density. Unfortunately, nationwide lightning flash data exist only for a few years (Orville 1994) and proxies must be used to investigate a climatic influence.

6. Lightning deaths and thunder-day frequencies

The next closest meteorological observation containing lightning activity information is the recording of the presence of thunder during the day. Long series of the yearly number of days with thunder exist for several stations in the United States. It should be kept in mind, however, that the number of days with thunder during a year is not necessarily a direct indication of the lightning activity during that year, as days with thunderstorms producing few lightning flashes are given the same weight as days with many flashes. Also, it is questionable how representative of the lightning activity over a broad region are the records of audible detection of thunder at a point. Compounding the problem is the fact that the number of stations recording thunder days is very small and variable, especially over the western states.

Nevertheless, Changnon (1985) investigated the secular variations in thunder-day frequencies during the period 1901–80 throughout North America and showed that long-term changes in thunder-day frequencies provide an indication of long-term climatic fluctuations. Using the records from 86 weather stations in the United States and Canada, he showed strong agreement between regional and continental thunder trends and other meteorological trends indicative of changes in circulation conditions. Furthermore, he established that the temporal variations in thunder-day frequencies were largely caused by temporal variations in synoptic weather conditions that produce thunderstorms.

Figure 7 shows the time series of mean 5-yr frequencies of thunder days obtained from Changnon (1985) (dashed line) and the normalized lightning deaths departures from Fig. 6 (solid line). The lightning death data are shown as mean 5-yr frequencies to match the thunderstorm series. The thunder-day series shows a marked increase in the first two decades of the century, with a moderate drop in the 1920s and an increase to the maximum in the series in the 1930s. After 1940, there was a drop in the number of days with thunder to 1970. From the minimum at the end of that period, the number rose considerably during the first 5 yr of the 1970s, dropping again during the next 5 yr.

The lightning death departures follow closely the same oscillations in thunder days, except that after 1960, the amplitude of the death departure oscillations are much smaller. The lightning death curve shows a small but gradual increase that is matched by the average trend in thunder days in the 1960s and 1970s, although the thunder-day trend is much higher and the pentadal values oscillate considerably around the mean trend.

7. Lightning deaths and surface temperatures

The thunder-day series provided information on climatic changes at the storm scale. A variable that can indicate long-term climatic fluctuations over the United States as a whole is the yearly average surface temperature. A comparison of this variable with the yearly lightning death departures could clarify the relationship between lightning deaths and climatic fluctuations at the national scale.

An analysis of global surface air temperatures indicates an average warming during the last 100 yr (Hansen and Lebedeff 1988). The decades of 1940–70, however, had declining temperatures, especially in the Northern Hemisphere (Jones et al. 1986). Since the mid-1970s, there has again been a warming trend. Heim et al. (1993) have produced average monthly and annual temperatures values for the United States, weighted by the areas of the state climatic divisions for the period 1931–91. Each division represents a region within a state that is climatologically quasihomogeneous or, in some cases, a semihomogeneous drainage basin. The NOAA Climate Diagnostic Center (CDC) has extended this dataset to cover the rest of this century and the last part of the previous century.

Figure 8 shows the temperature series derived from the CDC data for the contiguous 48 states and DC, for the period 1900–95, during the months of June–August (JJA) when most of the lightning deaths occurred (thin line). The normalized death departure series has been superimposed (dotted broad line). The temperature series indicates an increase of 2°–2.5°C from the beginning of the century to 1935. This was followed by a rapid drop of about 2°C from 1935 to the middle of the 1960s, then a gradual increase during the rest of the series.

These general trends are again very well matched by the lightning death departures. The early warming trend corresponds to an increase of 4 deaths per million people during roughly the same period, the cooling from 1935 to the middle of the 1960s corresponds to a drop of almost 2 deaths per million, and the warming in the last part of the series is matched by a much slower but steady increase in the death rate. In general, the two series show good correspondence on the major longer-term features. Even some of the year-to-year fluctuations are well correlated in time, although not necessarily in amplitude. The slow rate of increase in the death rate during the last 2 decades, compared to the temperature series, could be due to the assumption of a uniform percentage rate of indirect lightning deaths.

8. Summary and conclusions

Long-term fluctuations in the number of lightning deaths from 1900 to 1991 have been examined by analyzing lightning death data compiled by the Bureau of the Census and the Public Health Service for the contiguous United States and the District of Columbia. The effect of the varying population during the period was removed by dividing the yearly values of reported deaths by the corresponding yearly population of the reporting states. The resulting population-normalized series revealed the presence of an overall exponential trend in the number of deaths per million people producing a decrease from more than 6 deaths per million to less than 0.5 during the 92 yr of data. This exponential trend is also present in the decrease of the rural U.S. population for the period. The two datasets agree very well and this suggests that the downward trend in deaths resulted to a large extent from a migration of people from rural to urban areas, or from the progressive urbanization of rural areas. Other factors that can reduce the lightning risk are improvements in electrical systems and fire resistance of houses, medical treatment, education, and meteorological warnings. Many of these, however, are also related to the decrease in rural population and may account for the good agreement between the trends of lightning deaths and rural population. In some cases, however, such improvements with time might have been independent of the increase in urban population and may account for some of the minor discrepancies between the two series.

Superimposed on the overall downward trend there were fluctuations of 2 or 3 decades in duration. These fluctuations have been isolated by removing the exponential decrease trend from the lightning deaths series. There are three major oscillations that decrease in amplitude with time throughout the period. From a comparison of the series of lightning death departures from the exponential trend with two different climatological variables, these oscillations appear to be climate related. The patterns of these fluctuations were parallel to nationwide changes in thunder-day frequencies and average surface temperature values.

Since the lightning death data are of a unique nature and have not been used in their entirety before, some references to the history of mortality data collection in the United States have been presented. There have been changes in the coverage of the mortality data collection and the classification procedure that need to be taken into account to make the lightning death information uniform from year to year. The raw data used have also been provided in tabular form so that they are available for future studies of the lightning hazard.