Factors affecting temporal fluctuations in damaging storm activity in the United States based on insurance loss data

Meteorol. Appl. 6, 1 10 (1999) Factors affecting temporal fluctuations in damaging storm activity in the United States based on insurance loss data Stanley A Changnon, Department of Geography, University of Illinois at Urbana- Champaign, Champaign, Illinois, USA, and Changnon Climatologist at Mahomet, Illinois, USA Insured weather losses in the US reached record highs in the early 1990s, leading to major concerns in the insurance industry about the causes, including the possibility of climate change due to global warming. Several studies addressing the interpretation of the record high losses used historical insurance data sets, those for crop-hail losses and others based on the catastrophic events to the property insurance industry, both covering the 1949 1995 period. The past loss values were adjusted by insurance experts for shifting coverage, inflation, and evolving construction practices. The resulting adjusted values of crop loss and costly catastrophes to property both showed similar distributions for 1949 1995, with losses being high in the 1950s and again in the early 1990s. This distribution was found to have a weak relationship with extra-tropical cyclone activity in the US, but most of the recent increase in weather losses to insured property was found to be related to shifting societal factors that have put ever more property at risk in storm-prone areas, including coastal areas and large metropolitan areas. The low property and crop storm losses of the 1966 1985 period had created an incorrect perception of the weather risk in the insurance industry, which did not understand nor appreciate the existence of the decadal-scale fluctuations that exist in climate conditions. 1. Introduction Insured crop and property losses due to weather during 1991-1994 amounted to $34.5 billion, the highest values on record, and this amount is estimated to represent about 60% of the nation s total losses (Roth, 1996). This large increase in US insurance losses due to intense storms in the early 1990s created great concern amongst the crop, property, and reinsurance industry about the causes for these costly years (Insurance Services Office, 1994; Swiss Re, 1996). The latest Intergovernmental Panel on Climate Change (IPCC) assessment did not cite any trends in insurance data related to a global change in climate (Dlugolecki, 1996) but reported that many insurers believed that the worldwide losses were increasing owing to changes in climate. Speculation that the US storm-loss increase was related to climate change due to global warming began in 1993 (Linden, 1994; Leggett, 1993), whereas others believed the losses were due to shifts in natural conditions affecting climate (Glantz, 1995; Changnon et al., 1996a). The potential problems that a climate change resulting from any cause would create for the US insurance industry have been identified for crop insurance (Fosse & Changnon, 1993), for property insurance (Friedman, 1989; Lecomte, 1993), and for the reinsurance industry (Linden, 1994; Totsch, 1996). These problems related to possible shifts in the frequency or intensity of storminess, such as in thunderstorms and their damaging products, including hail, high winds, tornadoes, lightning, and heavy rains; in winter storms; and in hurricanes. This paper reports on findings from studies which assessed the temporal fluctuations in severe weather conditions based on insurance-derived data on losses. The first section addresses crop-hail insurance results and the second section addresses property losses recorded by the property-casualty insurance industry. 2. Crop-hail insurance Hail causes considerable damage to US crops and property, occasionally causes death to farm animals, but seldom causes loss of human life. Crop-hail losses in recent years nationally are estimated at $1.3 billion annually, representing between 1 and 2% of the annual crop value. Hail losses vary considerably regionally, representing, for example, 1 2% of the crop value in the Midwest, 5 6% of the crops produced in the High Plains, and much less elsewhere in the nation. The long-term trends in hail incidence and hail losses across the United States, and within states and discrete regions within the nation, were examined using 1

S A Changnon available data and results. Data on hail sufficient to be used in multi-decadal analyses comes from two sources: insurance records of hail loss and the records of hail of the US National Weather Service (NWS). 2.1. Crop-hail insurance data Crop-hail insurance data for each state and the nation have been systematically collected since 1948 by companies acting through a central association which has compiled the data and archived it (National Crop Insurance Services, 1996). These data have limitations, including the fact that not all farmers have purchased insurance coverage (hail insurance is estimated to cover 25 30% of all crop losses caused by hail). Second, crop-hail losses for a state or the nation shift with time owing to the amount of coverage (liability) and the crop value, as well as the temporal variations in hail occurrences (which are large). Fortunately, the industry devised an adjustment factor, loss cost, which is the amount of loss per year ($) divided by the annual amount of liability ($) multiplied by $100. The loss cost values for 1948 to 1996 for individual states and the nation provide a useful measure to analyse the temporal fluctuations of loss (Fosse, 1996). The national annual values of insured crop-hail losses appear in Figure 1, along with the amount of liability (Changnon et al., 1996a). Losses have grown with time, ranging from $15 million in 1948 to $129 million in 1974, then increasing rapidly to $265 million in 1980, and approaching $400 million in the early 1990s. Liability also climbed steadily in this 48-year period. Figure 1 also presents the adjusted loss cost values, the appropriate data to examine the trends in crop-hail losses. Relatively high values existed in the 1950s and early 1960s, and again in the early 1990s. After the peak centred at 1962 1963, values declined and remained relatively stable until the recent three-year high in 1992 1994. The long-term average loss cost for the US is $2.55. The highest three-year loss costs since 1947 were: $3.38 in 1961 1963, $3.27 in 1954 1956, and $3.25 in 1992 1994. From a risk and rating standpoint, the recent peak was preceded by an 11-year period (1981 1991) with relatively low loss cost values (Figure 1). Since rates are changed every three to five years, the length of this low loss period led to a considerable lowering of insurance rates. The relatively low rates in place when the three years (1992 1994) of high losses occurred effectively acted to over-emphasise for insurers the magnitude of the recent losses. But, when put in a 48-year time frame, the highs in the early 1990s rate third highest. No statistically significant long-term trend of decrease or increase in the crop-hail loss costs was found (Fosse, 1996). In essence, insurers had lowered rates too much and without consideration of the types of climateinduced fluctuation in hail activity. Major regional differences were found in crop-hail loss temporal distributions, a not unexpected outcome since hail is notoriously variable in both space and time (Changnon, 1977). As shown in Figure 2, upward trends exist in recent years in the loss cost values of the six states comprising the Northern High Plains, and since about 1970 in certain East Coast states (Virginia and North Carolina). Conversely, trends in loss costs in the Eastern Midwest and other adjacent states show continuing decreases over the past 30 years. 2.2. Hail-day data Changnon & Fosse (1981) described how climatic hail data were used by the crop insurance industry, illustrating how past fluctuations were a key issue since the Figure 1. The time distribution of national crop-hail annual loss costs (losses divided by liability times 100), losses, and liability for the 1948 1996 period (Changnon et al., 1996a). 2

Causes of fluctuations in insured storm losses Figure 2. Median annual loss cost values for crop-hail insurance in groups of states and for the 1948 1994 period (Changnon et al., 1996a). establishment of insurance rates rests on past experience. These applications included use of hail-day data from the NWS as a check on the insurance industry s historical hail loss records. The incidences of days with hail have been recorded at the 210 first-order weather stations across the nation since 1900, and these data offer an independent evaluation of the time distribution of hail for comparison with the crop-loss findings. Research has shown that the areal extent of extreme frequencies of hail days in a given year relate well to crop-hail insurance losses (Changnon & Changnon, 1997a). Temporal variations of hail days had been examined for stations in three states in differing hail climates. The results for four stations (distributed west east) in Nebraska appear in Figure 3 (Changnon & Changnon, 1996). All but Omaha showed low early (1921 1940) values. Then, all stations, including Omaha, had high hail values during 1951 1980, followed by low values for 1981 1990. The values for the 12 Texas stations showed two trends. Stations in north-west Texas and the Panhandle, areas with extensively insured crops, showed sizeable increases in hail days with time, peaking in the 1990s (part of the High Plains which had high losses, Figure 2). Stations in eastern and southern Texas showed declines in hail incidence with time (Changnon et al., 1996b). Hail values at Illinois 6 stations all displayed similar time distributions, with the highest values during 1955 1970 (when crop-hail loss values were highest), and then slowly declining hail-day frequencies from 1970 to the present (Changnon, 1995). As shown in Figure 2, Illinois and other Eastern Midwest states have experienced continuing declines in crop-hail loss cost values from the mid-1960s to the present. Key findings revealed in the hail-day values for Nebraska, Texas, and Illinois, and in their values of loss Figure 3. Decadal values of the number of hail days, expressed as a percent of the long-term average (1901 1990), for four weather stations in Nebraska, distributed from the far west (Kimball) to the far east (Omaha) (Changnon & Changnon, 1996). cost, is that their fluctuations agreed, and they also revealed that temporal fluctuations in hail incidence and damage vary considerably spatially across the United States. Years with major hail losses result from a few days with massive storm outbreaks (Changnon, 1977), and these conditions occur randomly in selected groups of years. A study of hail-days over the 1901 1980 period in the northern Great Plains showed major five- to ten-year fluctuations in hail frequency but no evidence of long-term trends up or down (Changnon, 1984). 3. Property insurance 3.1. Catastrophe losses Weather-caused insured property losses began to increase in the late 1980s and reached record highs during 1991 1995. These loss values are based on estimated losses for major storms since the property insurance industry has not kept detailed loss records for individual losses due to severe damaging weather. However, since 1949, the industry, through its centralised Property Claim Services of the American Insurance Services Group, has identified and recorded each catastrophe, defined from 1949 to 1982 as a storm situation producing $1 million or more loss to property (not crops), and beginning in 1983 as $5 million or more in losses, a shift chosen to crudely adjust for growing inflation. For each catastrophe, insurance loss experts estimated the amount of dollar loss and identified the weather conditions causing the damage and the states where the losses occurred. Analyses of raw catastrophe loss data show everincreasing losses and number of storm events, as illus- 3

S A Changnon A major US insurance company systematically made major adjustments to the catastrophe database (done for each year) to perform internal assessments of shifting risks and to plan for rate adjustments. Their adjustment procedure included modifications of each event s value for the ever-changing dollar value, for changes in property density location, and for changing costs and types of construction. This adjusted catastrophe database offered an opportunity to meaningfully examine the temporal trends in catastrophes (Changnon & Changnon, 1992a). The adjustments made by the insurance company compare favourably to those made by other insurance experts. For example, Clark (1991) calculated that the adjustments of losses for inflation, population, and percentage of property insured for two 1954 hurricanes required multiplication of their losses by a factor of 23 to match 1991 loss levels. This approximates the insurance company s adjustment factor of 22 for one 1954 hurricane and 27 for the other storm (the difference in these two values was due to regional differences in property development since 1954). Clark s adjustment of the losses of Hurricane Betsy in 1965 to 1991 loss levels led to an estimate of $8 billion, compared to $7.6 billion using the insurance company s adjusted values. Pielke & Landsea (1998) also devised a loss normalisation scheme for adjusting historical hurricane losses to a 1995 base. This scheme was based on adjustments for changing inflation, wealth, and population; for example, a 1938 hurricane causing a loss of $306 million (1938 dollars) was adjusted to $16.6 billion in 1995 dollars. Figure 4. The amount of insured loss from catastrophes for 1962 1992, as shown in an insurance document (Insurance Services Office, 1994), are the shaded portions of the bars for 1962 1971, 1972 1981, 1982 1991 and 1992. Changes to the values for 1962 1971 and 1972 1981, done to adjust for changing factors including inflation, insurance coverage, and construction practices, are shown as angular hatching and raise the two ten-year totals considerably. Also shown are the adjusted decadal values (angular hatching) for 1950 1959 and 1952 1961, both added to the original published figure to show that large decadal values existed before 1961. Hurricane Andrew s large losses in 1992 were presented in the original published illustration to indicate the enormity of its loss in relation to those of the three preceding decades. trated by the three decadal loss bars (1962 1971, 1972 1981, and 1982 1991) in Figure 4 (Insurance Services Office, 1994). However, adjustments for everchanging dollar values, population, and insurance coverage have not been included in these catastrophe values. Further, this example, based on a corporate illustration (Figure 4), did not include the catastrophe losses for the pre-1962 period, an omission that helps create the impression of an ever-increasing weather loss situation with time. 4 These insurance company adjustments create a modified, or adjusted, historical catastrophe loss database that allows a more correct assessment of the temporal magnitude and variability of loss over time. The differences between the unadjusted and adjusted values are significant. For example, the adjusted catastrophe values are shown in Figure 4, raising the value of 1962 1971 from $10 billion (current dollars) to $24 billion (adjusted). Furthermore, addition on this graph of values for two earlier decades not shown in its original published form, 1950 1959 with $28.7 billion in adjusted losses and 1952 1961 with $22.3 billion (adjusted), helps to illustrate more properly the historical distribution of catastrophic losses. The adjusted 1950 1959 value is only 8% lower than the $31 billion in 1982 1991. This example reveals the importance of examining catastrophic losses based on values adjusted for changing conditions affecting loss. Otherwise, presentations can overemphasise the magnitude of the recent losses. Hereafter in the text and figures, all monetary values shown are the adjusted values, but if current year (unadjusted) values are shown, they are labelled. A study focusing on the catastrophes causing greater than $100 million in insured losses during 1950 1989 Figure 5. Pentadal frequencies of national catastrophes causing losses at three levels: >$100 million, >$200 million, and >$1 billion (all values adjusted) during 1950 1989 (Changnon & Changnon, 1992b).

Causes of fluctuations in insured storm losses Figure 6. Regional frequencies of property catastrophes causing >$100 million in losses (adjusted) for five-year periods from 1950 to 1994 (Changnon & Changnon, 1992a). showed a U-shaped distribution (Figure 5) with major highs in recent years (Changnon & Changnon, 1992b). However, distributions shown by catastrophes at the costlier loss levels, >$200 million and >$1 billion in losses, were different from the $100 million distribution. Both showed a slight decrease from 1950 to 1989. The study also found major differences in the temporal shifts across large regions of the US. Figure 6 presents graphs showing that the catastrophic storm distributions in the central and northern areas were U-shaped, whereas those in the southern areas were flat until abrupt increases began in the 1970s. This increase agrees with the rapid increase in population in the South and South-east and particularly in their coastal areas (Culliton et al., 1990). Figure 7 shows that the regional fluctuations in >$100 million catastrophes sorted by type of storm (Changnon & Changnon, 1992a). The thunderstormrelated events in most US areas had the U-shaped distribution seen for all catastrophes (Figure 5), revealing that thunderstorm-caused catastrophes controlled the national totals. They also show that the major recent increases in thunderstorms occurred in the South and Central areas. The hurricane distributions (Figure 7(b)) vary between regions but with no distinct up or down trends. Winter storms causing >$100 million have become prevalent in recent years, peaking in 1980 1984, and then decreasing. Figure 8 shows the time distributions for 1949 1994 of the intensity (defined as the annual amount of catastrophic loss divided by number of catastrophes per year) and the number of catastrophes based on the 189 catastrophes causing >$100 million in losses (Changnon et al., 1996a). Both expressions of these 5

S A Changnon Figure 7. Regional frequencies of catastrophes causing >$100 million in losses (adjusted) based on three conditions: (a) those from thunderstorm-related losses, (b) those for hurricane losses, and (c) those for winter storm losses during the 1950 1994 period (Changnon et al., 1996a). Figure 8. Five-year average values of the number of catastrophes causing >$100 million in losses (adjusted), their intensities (losses divided by frequencies of catastrophes), and the number of extra-tropical cyclones in the US, for 1950 1994 (Changnon et al., 1996a). costly events had a U-shaped distribution. Figure 8 also shows the temporal distribution of extra-tropical cyclones in the US, and it has some resemblance to the catastrophe distributions. However, the annual cyclone frequency explained only 16% of the variation in the number of >$100 million catastrophes and 28% of their intensity variations. As a test of the possible effect 6 expected under global warming, annual mean temperatures for 1949 1994 were compared with catastrophe frequencies. Temperature values explained 39% of the variations found in the number of catastrophes causing $10 million to $100 million (Changnon et al., 1996a), but this may well be a spurious relationship.

Causes of fluctuations in insured storm losses Figure 9. The time distributions, based on catastrophes that caused losses between $10 million and $100 million (adjusted values), for five-year periods of the number of catastrophes, the amount of loss from these catastrophes, and the US population (Changnon et al., 1997). The temporal distributions of 707 catastrophes (based on those causing $10 million to $100 million) and their adjusted losses during 1949 1994 were examined (Changnon et al., 1997). Figure 9 shows ever-increasing values of frequency and amount of loss from 1949 to 1994 (Changnon et al., 1996a). Notable is the close relationship between these upward trends and US population, indicating that the adjusted catastrophic values had not totally handled shifts in demographic risk. Figure 10 presents a normalisation of the catastrophe losses, achieved by dividing annual adjusted losses by annual population values for all 896 catastrophes since 1949 that caused >$10 million in losses (adjusted). This has an essentially flat distribution with five peaks (1950, 1954, 1965, 1989, and 1992) all resulting from major hurricanes in those five years. This result does not suggest a long-term increase in total catastrophes and indicates that the increase in the $10 million to $100 million catastrophes was an artefact based on inadequate societal adjustment factors (Changnon & Changnon, 1997b). Results of these studies of the nation s property catastrophe data reveal that the time-related increase in catastrophic events and their losses is largely a function of the increased target-at-risk, as indicated by population as a surrogate measure of the property at risk. The insurance company s adjustments in the catastrophe data for shifting property-at-risk did not capture all the changes affecting risk, such as growth of property density by location and the changing value of personal property (Changnon & Changnon, 1997b). Furthermore, regional analyses showed that the greatest relative increases in >$100 million catastrophes occurred in the South-east and South (Figure 5), where the rate of population growth since 1950 has been the nation s highest (Changnon et al., 1996a). Demographic changes have included the movement of people to more dangerous locations such as the coastal areas on the East Coast and Gulf Coast. For example, from 1970 to 1990 the south-eastern Atlantic coastal areas had a 75% increase in population density, far outpacing the national increase of 20% (Roth, 1996). Also, the value of property was escalating, and from 1980 to 1993 the value of insured residential exposures increased 166% and commercial exposures rose 193% (Insurance Research Council, 1995). 3.2. Property-hail losses In recent years several major hail-caused property losses occurred in cities like Denver with current year values of $650 million in hail damage in July 1990, Orlando with $85 million in April 1992, Wichita with $215 million in June 1992, Oklahoma City with $200 million in April 1992, Dallas with $227 million in April 1995, and Ft. Worth with $300 million in May 1995 (Changnon, 1997). These large urban loss values Figure 10. Annual catastrophe losses divided by US population for the 1949 1994 period based on all catastrophes causing >$10 million in losses (Changnon et al., 1997). 7

S A Changnon suggest a developing trend. Annual insured crop-hail losses during 1990-1995s were less than $400 million (current values), and these large big-city annual property losses were well in excess of the crop-hail losses in 1990, 1992, and 1995. Twenty years ago the average annual crop-hail losses were eight times those of property (Changnon, 1977), but the high property hail losses of 1990 1995 suggest that the ratio of crop to property losses has drastically shifted. This further indicates that property losses have been increasing with time, owing largely to the nation s ever-increasing property at risk, which is related to ever-increasing sizes of the nation s metropolitan areas. 3.3. Fluctuations in storm frequencies Available information on the temporal behaviour of various severe storm conditions was assessed to discern how they related to the fluctuations in insured losses during the 1949 1995 period. Gabriel & Changnon (1990) analysed US historical thunderstorm data and found a peak in the 1940s and 1950s, a decline in frequencies during the 1960 1980 period, a brief increase during the mid-1980s, and then a further decline into the 1990s. Analyses of the annual frequencies during 1953-1989 of tornado-day values, a measure of tornadoes least influenced by shifts in population and storm warnings, showed a flat time trend since 1953 (Grazulis, 1991). Landsea et al. (1996) analysed hurricane activity during the 1944 1994 period and found a general decline with time. Studies of flood-producing seven-day rainfall events during the 1921 1994 period in the central US found increases in some areas with decreases in others (Kunkel et al., 1993). Karl et al. (1995) found that one-day precipitation events of 5 cm have made an increasingly larger contribution to the nation s annual precipitation during the 1910 1993 period, but these 5-cm events do not equate to damaging events. These measures of various types of storm activity over time do not show major increases into the late 1980s and early 1990s like the catastrophic storm losses did. This further supports the conclusion that the insured property loss increases were largely a function of changes in society and not in weather conditions. 4. Summary and conclusions The studies of the insurance-derived measures of damaging severe storms revealed important findings for the nation and for various regions that collectively explained why insured losses in the early 1990s were so large. An important question relates to whether the severe weather across the US during recent years was the worst on record. There are two answers to this question. First, weather events of the 1991 1995 period were extreme. Property losses from catastrophes reached an all-time peak, and crop losses from hail in 8 1992 made it the worst loss year since records began in 1948. Second, after adjusting the storm losses since 1949 for changes in coverage, inflation, and societal factors, a different perspective concerning the recent large losses emerged. In most respects, the property-related losses for catastrophes causing losses >$100 million (adjusted) and for loss costs (a liability-adjusted expression for crop-damaging hailstorm losses) of the early 1990s were generally comparable to those of the early 1950s, and no long-term linear up trend was found in these values. However, temporal analyses of the propertybased catastrophes show major differences depending on the magnitude of loss. Catastrophes causing $10 $100 million (adjusted) losses, 80% of all catastrophes, showed steady increases in losses and storm frequency from 1949 to 1995. However, catastrophes causing >$100 million had a U-shaped distribution, being high in the 1950s and 1990s, but the most costly catastrophes, those causing >$1 billion, had a flat distribution over time. When annual losses of all catastrophes causing >$10 million in losses (adjusted) were divided by annual population values, the resulting temporal distribution for 1949 1994 was flat and marked by only five peaks, caused by major hurricanes. The distribution of crop-hail insurance cost values also did not have a long-term trend up or down, and recent national loss cost values ranked third highest for the 1948 1994 period. Societal factors and atmospheric conditions have been investigated as potential explanations for the temporal variations in catastrophes. Population (reflecting the changing property target and increasing sensitivity to damaging storms), frequency of extra-tropical cyclones, and annual temperature and precipitation values were found to explain 80% of the historical fluctuations in catastrophe frequencies and losses (Changnon et al., 1996a). Population growth explained much of the ever-increasing number of catastrophes and their volume of losses, particularly for the 707 events each causing less than $100 million in losses. The extra-tropical cyclone frequency in the eastern United States decreased from record high values in the 1950s until the 1970s and has since been increasing. The temporal distribution of these cyclones during 1949 1994 explained 28% of the temporal variations found in catastrophe intensities since 1949, and 43% of the variation in national crop-hail loss costs. Karl et al. (1995) presented a climate extreme index for the US which has a temporal behaviour similar to that of the cyclones. Data for various storm types, including tornadoes, thunderstorms, and hurricanes, did not show major upward shifts in the 1990s, further illustrating that the recent peaks in losses were a result of societal factors, not major changes in atmospheric conditions. Regionally, the insurance loss findings showed major differences in temporal fluctuations and trends. The

Causes of fluctuations in insured storm losses crop-hail losses in the High Plains and South-east have peaked in recent years, but since 1980 losses in the Midwest have been on a steady decline. The property catastrophe findings suggest losses have grown rapidly in the southern coastal sections where population has grown rapidly. However, in the northern sections of the nation, the long-term distribution of storm catastrophes shows a different temporal distribution, having been high in the 1950s early 1960s, low in the 1970s, and high again in the late 1980s 1990s. This illustrates that climate variations differ across an area as large as the United States and that national values tend to average out many sharp regional differences. In calibrating the sizeable insurance industry reactions to the large losses of the 1990s, it is also important to take into consideration preceding conditions. Property catastrophe losses in the preceding years of the 1980s were less, and those of 1965 1980 were the lowest of the past 47 years. The high crop-hail losses of 1992 1994 were also preceded by a run of 11 years with exceptionally low losses. The contrasts between low values for 10 20 years and then the high values of the 1990s produced an exaggerated impression of the magnitude of the recent losses. Decision makers in the property-casualty insurance industry had become complacent and did not understand the sizeable shifts in the weather threat posed by the decadal scale fluctuations that exist in the storm climate of the US (Roth, 1996). A well-known aspect of the US climate is that extreme events tend to cluster in groups of years and that there are typically wide swings in the frequency and magnitude of extreme weather events. Storm events in the 1990s, when appropriately adjusted for inflation, population shifts, and other factors affecting insured loss, were extreme but not unique. The property catastrophe results reveal that shifts over the past 50 years have been driven largely by population increases, lifestyle changes, including increased property values, and shifting demographic patterns in various areas of the nation, and to a lesser degree by atmospheric fluctuations. Results indicate that insurers need to gather better data on property losses, to continually adjust losses for societal shifts over time, to monitor and anticipate more effectively societal trends affecting vulnerability to weather damage, and to closely monitor climate fluctuations in various weather conditions capable of producing damage (Changnon et al., 1997). Comparison of the adjusted measures of insurance losses reveal that those of the early 1990s are not unlike past extremes. Shifts due to anthropogenic-induced climate change appear unlikely. The Intergovernmental Panel on Climate Change (IPCC) latest findings agree. Their conclusion about changes in extreme weather events states, Overall, there is no evidence that extreme weather events, or climate variability, has increased in a global sense through the 20th Century. There is some evidence of regional scale changes, some to greater variability, some toward lower variability (Nicholls et al., 1996). In addressing the individual storm hazards, the IPCC indicated: (a) no trend in hurricanes crossing the US coast since 1900; (b) the evidence of changes in extra-tropical synoptic cyclones was inconclusive; (c) there was a trend in recent decades to more 5-cm rain days in the US; (d) nothing had been reported about changes in hail; and (e) no evidence of consistent increases in tornadoes, thunderstorms, or dust storms. In summary, the recent fluctuations in damaging storm activity in the US are considered to reflect natural variations in the climate. These results for the United States may also be indicative of why weather losses globally have also increased rapidly in recent years. Acknowledgements The advice and comments of E. R. Fosse, former head of the Crop-Hail Insurance Actuarial Association and director of the Federal Crop Insurance Corporation, and the comments of E. L. Lecomte, former director of the Insurance Institute for Property Loss Reduction, are gratefully acknowledged. Portions of this research were supported by grants from the University Corporation for Atmospheric Research and the Electric Power Research Institute. References Changnon, S. A. (1977). The scales of hail. J. Appl. Meteorol., 16: 626 648. Changnon, S. A. (1984). Temporal and spatial variations in hail in upper Great Plains and Midwest. J. Climate & Appl. Meteorol., 23: 1531 1541. Changnon, S. A. (1995). Temporal fluctuations of hail in Illinois. Miscellaneous Publication 167, Illinois State Water Survey, Champaign, IL, 19 pp. Changnon, S. A. (1997). Trends in hail in the United States. In Proceedings of Workshop on the Social and Economic Impacts of Weather, National Center for Atmospheric Research, Boulder, CO, 19 34. Changnon, D. & Changnon, S. A. (1997a). Surrogate data to estimate crop-hail loss. J. Appl. Meteorol., 26: 456 465. Changnon, D. & Changnon, S. A. (1997b). Evaluation of weather catastrophe data for use in climate change investigations. Climatic Change, 27: in press. Changnon, S. A. & Fosse, E. R. (1981). Impacts and use of climatological information in the hail insurance industry. In Proceedings of Climate and Risk Conference, Mitre Corp., Washington, DC, 28 pp. Changnon, S. A. & Changnon, J. M. (1992a). Temporal fluctuations in weather disasters: 1950 1989. Climatic Change, 22: 191 208. Changnon, S. A. & Changnon, J. M. (1992b). Storm catastrophes in the United States. Natural Hazards, 6: 93 108. 9

S A Changnon Changnon, S. A. & Changnon, D. (1996). Hail in Nebraska. National Crop Insurance Services, Overland Park, KS, 22 pp. Changnon, S. A., Changnon, D., Fosse, E. R., Hoganson, D., Roth, R. J. & Totsch, J. (1996a). Impacts and Responses of the Weather Insurance Industry to Recent Weather Extremes. Changnon Climatologist, Mahomet, IL, 232 pp. Changnon, S. A., Changnon, D. & Changnon, J. (1996b). Hail in Texas. National Crop Insurance Services, Overland Park, KS, 27 pp. Changnon, S. A., Changnon, D., Fosse, E. R., Hoganson, D., Roth, R. J. & Totsch J. (1997). Effects of recent extremes on the insurance industry: major implications for the atmospheric sciences. Bull. Am. Meteorol. Soc., 78: 425 435. Clark, K. M. (September 30, 1991). Catastrophe Reinsurers Need More Better Data. National Underwriter, Property & Casualty Edition, p. 48. Culliton, T. J., Warren, M., Goodspeed, T. R., Remer, D., Blackwell, C. & McDonough, J. J. (1990). Fifty Years of Population Change along the Nation s Coasts, 1960 2010. National Oceanic and Atmospheric Administration, Rockville, MD, 41 pp. Dlugolecki, A. F. (1996). Financial services. In Climate Change in 1995: Impacts, Adaptation and Mitigation of Climate Change: Scientific and Technical Analyses, Intergovernmental Panel on Climate Change, Cambridge University Press, Cambridge, UK, Chapter 17: 539 560. Fosse, E. R. (1996). Impacts and responses to the 1991 1994 weather: the crop insurance industry. In Impacts and Responses of the Weather Insurance Industry to Recent Weather Extremes, Changnon Climatologist, Mahomet, IL, 75 100. Fosse, E. R. & Changnon, S. A. (1993). Potential impacts of shifts in climate on the crop insurance industry. Bull. Am. Meteorol. Soc., 74: 1703 1708. Friedman, D. G. (1989). Implications of Climate Change for the Insurance Industry. Travelers Insurance Co., Hartford, 23 pp. Gabriel, K. R. & Changnon, S. A. (1990). Temporal fluctuations in thunderstorm days in the US Climatic Change, 15: 455 477. Glantz, M. H. (April 1995). The science and politics of global warming. Insurance Specialist, 2:10 12. Grazulis, T. P. (1991). Significant Tornadoes 1880 1989. Vol. 1: Discussion and Analysis. Environmental Films, St. Johnsbury, VT, 526 pp. Insurance Research Council (1995). Coastal Exposure and Community Protection. Wheaton, IL, 48 pp. Insurance Services Office, Inc. (1994). The Impact of Catastrophes on Property Insurance. ISO Insurance Series, New York, NY, 50 pp. Karl, T., Knight, R., Easterling, D. & Quayle, R. (1995). Trends in US climate during the twentieth century. Consequences, Spring: 3 12. Kunkel, K. E., Changnon, S. & Shealy, R. (1993). Spatial and temporal characteristics of extreme precipitation events in the Midwest. Mon. Wea. Rev., 121: 858 866. Landsea, C. W., Nicholls, N., Gray, W. M. & Avila, L. (1996). Downward trends in the frequency of intense Atlantic hurricanes during the past five decades. Geophys. Res. Letters, 23: 1697 1700. Lecomte, E. L. (1993). Remarks. Climate Change and the Insurance Industry: The Next Generation, College of Insurance, New York, NY, 1 6. Leggett, J. (1993). Securing profitability in an adverse climate. Climate Change and the Insurance Industry: The Next Generation, College of Insurance, New York, NY: 1 13. Linden, E. (1994). Burned by warming. Time, 14 March, p. 79. National Crop Insurance Services (1996). Crop Hail Insurance Statistics, United States. Overland Park, KS, 48 pp. Nicholls, N., Gruza, G., Jouzel, J., Karl, T., Ogallo, L. & Parker, D. (1996). Observed climate variability and change. Chapter 3 in Climate Change 1995: The Science of Climate Change, IPCC, Cambridge Univ. Press, Cambridge, UK, 137 192. Pielke, R. A. Jr. & Landsea, C. W. (1998). Normalized hurricane damages in the US. Wea & Forecasting, 16, in press. Roth, R. J. Jr. (1996). The property casualty insurance industry and the weather of 1991 1994. Impacts and Responses of the Weather Insurance Industry to Recent Weather Extremes, Changnon Climatologist, Mahomet, IL, 101 132. Swiss Re (1996). Natural catastrophes and major losses in 1995: decrease compared to previous year, but continually high level of losses since 1989. Sigma, 2: 1 50. Totsch, J. M. (1996). Impacts and responses of the property catastrophe reinsurance market. Impacts and Responses of the Weather Insurance Industry to Recent Weather Extremes, Changnon Climatologist, Mahomet, IL, 133 152. 10