Florida Hurricanes and Damage Costs - Tampa Bay...Hurricanes and Damage Costs 109 (hurricanes Andrew...
Transcript of Florida Hurricanes and Damage Costs - Tampa Bay...Hurricanes and Damage Costs 109 (hurricanes Andrew...
southeastern geographer, 49(2) 2009: pp. 108–131
Florida Hurricanes and Damage Costs
JILL MALMSTADTFlorida State University
KELSEY SCHEITLINFlorida State University
JAMES ELSNERFlorida State University
Florida has been visited by some of the most de-structive and devastating hurricanes on record inthe United States causing well over $450 billion indamage since the early 20th century. The value ofinsured property in Florida against windstormdamage is the highest in the nation and on therise. The frequency and severity of hurricanes af-fecting Florida are examined from the best set ofavailable data and the damages are related tocharacteristics of the storms at landfall. Resultsshow that normalized losses are increasing overtime consistent with small increases in hurricaneintensity and hurricane size. The best predictorof potential losses is minimum central pressure.Hurricane size alone or in combination with hur-ricane intensity does not improve on the simplerrelationship. An estimate of potential losses fromhurricanes can be obtained using a formula in-volving only a forecast of the minimum pressureat landfall. The ability to estimate potential lossesin Florida will increase the ability to estimatelosses in other areas of the United States, and willalso allow policy makers and insurance com-panies to provide relevant information to the con-cerned public.
key words: Florida, hurricanes, landfall,insurance, losses, trends, correlation
introduction
The hurricane is an awesome, yet deadlyand destructive natural phenomenon ofthe Earth’s occasionally tumultuous atmo-sphere. A hurricane is powered by the heatand moisture of the tropical oceans ratherthan thermal contrasts across latitudes asis the case for the more common extra-tropical cyclone. The result is a powerfulstorm, causing unprecedented amounts ofdeaths and economic loss. Dollar lossesfrom hurricanes are at the top of the list ofcatastrophic events ahead of tornadoesand terrorism. Not surprisingly, because ofits location relative to the warm waters ofthe North Atlantic (including the Gulf ofMexico and the Caribbean Sea), Florida ismore likely to get hit by a hurricane thanany other state in the union. On average atleast one hurricane strikes Florida everytwo years and a strong hurricane hits Flor-ida on average once every four years (aver-ages come from available data during1900–2007). Eight of the 10 most expen-sive hurricanes ever to make landfall inU.S. history have had at least some affecton Florida, causing in excess of $60 billion(constant 2005 dollars) in insured losses
Hurricanes and Damage Costs 109
(hurricanes Andrew 1992, Charley, Fran-ces, Ivan and Jeanne in 2004, and Katrina,Rita, and Wilma in 2005). For this reason,as well as the devastating impact thesestorms have on human lives, scientists havebeen tackling the issue of hurricanes inorder to further understand their charac-teristics and better predict their impendingimpact on the coastline ahead. This paperwill focus on economic loss in the state ofFlorida, as Florida represents a unique casestudy for hurricane science.
Interest in economic loss from hur-ricanes is not new, as it was discussedthroughout the 1960s (Demsetz 1962;Sugg 1967), but interest has become ele-vated in recent times due to the destruc-tive Atlantic hurricane seasons of 2004and 2005.
The most damage-causing characteris-tics of a hurricane are the high winds,storm surge and large waves, as they eachhave potential for total destruction of prop-erty and livelihoods (Williams and Duedall1997). This potential for damage has re-cently increased due, in part, to a notablerise in global atmospheric temperatures.
The issue of climate change bringsalong a challenge to hurricane research-ers, as attempts are made to try to quan-tify the impact of climate change on thefuture of hurricanes. While many otherfactors play a role in hurricane develop-ment and intensity, the increasing sea sur-face temperatures associated with climatechange provide an obvious increase of fuelfor these storms and a heightened causefor alarm. Elsner et al. (2008) found anincreasing trend in the strength of thestrongest hurricanes, especially the 90th
percentile, meaning, in short, the strong-est storms are getting stronger. Emanuel(2005) uses the observed increase in sea
surface temperature to explain the in-crease in power dissipation within the av-erage North Atlantic hurricane and foundthat tropical cyclones have become moredestructive within the last 30 years. Thecombination of increasing storm strengthand coastal develop ment should yield in-creasing economic loss due to hurricanes.Changnon (2003) believes that coastal de-velopment is the main reason for recentincreasing economic loss, as the increasein losses throughout the 1990s occurredwhen hurricane frequencies decreased;the data showing no shift due to globalwarming. That is not to say that weatherextremes do not cause notable increases ineconomic loss, as the active weather of1991–1994 and associated hurricanes, se-vere storms, floods, tornadoes, etc. causedmore economic loss due to weather eventsthan any other four-year period. Yet still,the largest increases occurred in areas withthe greatest population growth (Chang-non 1997).
Florida, like most of the coastal UnitedStates, has seen a building boom, and theincreasing population and wealth is forc-ing insurers and re-insurers to rethinktheir exposures. According to the U.S.Census Bureau, Florida has the highestpopulation growth among states affectedby hurricanes and is expected to gainabout 13 million residents between 2000and 2030. The Citizens Property InsuranceCorporation in Florida (aka, Citizens), setup by the State of Florida in 2002 to be theproperty insurer of last resort, is now thelargest provider of property insurance inthe state. Florida homeowners can buycoverage from Citizens if the rates for acomparable policy from a private insurerexceed by 15 percent Citizens’ rates.
According to the Insurance Informa-
110 jill malmstadt, kelsey scheitlin, and james elsner
tion Institute (Hartwig 2008), the value ofinsured coastal property in Florida ranksfirst in the nation and, as of 2007, exceeds$2 trillion with about 60 percent in com-mercial exposure and the rest in residen-tial exposure (AIR 2005). Florida home-owner insurers’ underwriting losses in2004 ($9.3 billion) and 2005 ($3.8 bil-lion) resulted in a four-year cumulativeloss of $6.7 billion, even after includingthe profitable years of 2006 ($3 billion)and 2007 ($3.4 billion), when there wereno hurricanes (Hartwig 2008). Since 1992,Florida insurers have experienced a netloss of $6.2 billion (Hartwig 2008).
Citizens’ total exposure to loss is highand growing, increasing from $154.6 bil-lion in 2002 to $434.3 billion during thefirst quarter of 2007 (Insurance Informa-tion Institute 2007). The number of pol-icies written by Citizens is also on the risewith the total reaching 1.35 million as ofJuly 31, 2007. If losses by Citizens exceedits claim-paying capacity in a single sea-son, the state is required to impose an as-sessment on other lines of insurance, in-cluding policies not written by Citizens.Loss assessments (collected from all in-sured property owners from the entirestate), general revenue appropriations,and the reinsurance market can be aug-mented with the issuance of catastrophebonds. Catastrophe bonds help alleviatethe risk of a catastrophic event by transfer-ring some of that risk to investors. In Julyof 2007, Citizens floated a catastrophebond worth nearly $1 billion, and in Julyof 2008, Berkshire Hathaway, Inc. agreedto buy $4 billion in bonds if Citizens incursat least $25 billion in losses. The state hasestimated the probability of this level ofdamage occurring annually in the state ofFlorida to be about 3.1 percent per year. In
2008, in exchange for taking on this risk,Florida will pay Berkshire Hathaway, Inc.$224 million for a guarantee that the statewill receive up to $4 billion if the damagethreshold is reached (Kaczor 2008). Largeinvestors are becoming increasingly in-terested in catastrophe bonds and otherinsurance-linked securities because of theirlow correlation to traditional financialmarket performance providing a better di-versification of investment portfolios.
This paper provides a climatology ofhurricanes and hurricane losses in Florida.It is hypothesized that hurricane intensityis a predictor of total economic loss dueto hurricane landfalls. The purpose is tobuild a foundation for assessing the likeli-hood of future hurricane losses. The strat-egy is to graph and tabulate the historicalrecord of hurricane strikes and their asso-ciated damage costs, and examine how thestatistics of occurrence, intensity, and sizeare related to losses. Although others haveexamined damage losses from hurricanes(Katz 2002; Pielke et al. 2008; Jagger etal. 2008), this work is the first to lookat the problem focusing exclusively onFlorida.
The paper begins with a brief descrip-tion of the data sets followed by an exam-ination of Florida’s hurricane statisticsfrom the period 1900–2007. Inter-annual,seasonal, and intra-seasonal variability ofvarious hurricane characteristics are ex-amined first. Then the distributions andtemporal variations of the direct damagecosts associated with Florida hurricanesare considered. To bring together geophys-ical and economic issues, relationships be-tween losses and hurricane characteristicsare examined. It is found that historicallosses correlate best with minimum cen-tral pressure alone.
Hurricanes and Damage Costs 111
florida hurricaneand loss data
This study relies on two principal sourcesof data. For hurricanes affecting Florida,we use the list of historical hurricanes em-ployed to evaluate a risk model developedby the Florida Commission on HurricaneLoss Projection Methodology (FCHLPM)and supported by a research grant from theFlorida Office of Insurance Regulation(FOIR). A hurricane is a tropical cyclonewith maximum sustained wind speeds of atleast 33 m/s (64 kt/74 mph). The data setlargely conforms to the U.S. National Hur-ricane Center’s HURDAT storm archive(Ho et al. 1987; Landsea et al. 2004), but in-cludes storms only for the state of Florida.This dataset is available online at http://www.aoml.noaa.gov/hrd/lossmodel/AllFL.html.
The focus here is on hurricanes that di-rectly strike Florida. A direct strike (or hit)is one in which all or part of the hurri-cane’s eye wall reaches the coast. For thiswork, the Florida coast is defined as theboundary of the sea with the mainland,including all barrier islands surroundingFlorida. For the purposes of this study, hur-ricanes that approach Florida, but wherethe eye wall remains out at sea (e.g., Hur-ricane Elena in 1985) are not considereddirectly striking Florida. A direct strike in-cludes landfalling hurricanes and thosethat hit the islands making up the FloridaKeys.
A hurricane can make more than onedirect hit on the state. This occurs for in-stance when it first strikes the peninsulathen moves out over the eastern Gulf ofMexico before striking the panhandle re-gion (e.g., Storm #3 in 1903). Since a hur-ricane weakens over land, the intensity of
the hurricane at second landfall is typ-ically less than at first landfall. That said,most of the descriptive statistics presentedin this study are based on characteristics atthe time of first landfall, which for our pur-poses, is defined as the first direct striketo the mainland or a direct hit to the Flor-ida Keys only if the hurricane makes noother landfall in the state. If the hurri-cane makes landfall in the state more thanonce, the landfall characteristics at max-imum intensity are used because this com-parison is between loss data and hurricanecharacteristics, and the majority of lossescome from the strike of greatest intensity.
For losses incurred by hurricanes thatdirectly strike Florida, normalized damagedata are taken from the work of Pielke et al.(2008). There are two normalization pro-cedures presented in Pielke et al. (2008),both of which are estimates of the damagethat would have occurred if historic hur-ricanes struck in the year 2005. One pro-cedure allows for changes in inflation,wealth, and population, and the other pro-cedure allows for inflation, wealth, and anadditional factor that represents a changein the number of housing units that exceedpopulation growth between the year of theloss and 2005. The methodology producesa longitudinally consistent estimate of eco-nomic damage from past tropical cyclones.Losses caused by each storm event are ag-gregated from around the entire state anddo not necessarily cluster around the loca-tion of landfall. However, it is well knownthat the amount of damage experienced ishighly dependent upon where the stormmakes landfall in terms of buildings, in-frastructure, population, and so forth. Al-though this research focuses on aggre-gated loss values, it is useful because itprovides a general understanding of the
112 jill malmstadt, kelsey scheitlin, and james elsner
1900 1940 1980
0
1
2
3
4
Year
Ann
ual C
ount
(190
0!20
07) a
0 2 4Hurricane Events
Num
ber o
f Yea
rs
0
10
20
30
40
50
60b
Figure 1. Florida’s annual hurricane occurrence (1900–2007). (a) Time series of annual Floridahurricane counts. Only storms that made a direct strike on Florida with hurricane-force winds and
that have available economic loss values are included. (b) Distribution of annual Florida counts. Thereare a total of 67 known Florida hurricanes in the 108-year period.
economic losses experienced throughoutFlorida from 1900 to 2007.
hurricane statistics
The analysis begins with an examina-tion of the frequency of Florida hurricanes.Here the record starts with the 1900 sea-son and ends after the 2007 season. Notethat these economic loss data are referredto as losses and damage costs interchange-ably throughout this study. A Florida hur-ricane is a tropical cyclone that makes atleast one direct strike on the state as a hur-ricane. A hurricane that makes more thanone landfall in the state of Florida (e.g.Storm #3 in 1903) is considered andcounted as one, single Florida hurricane.
Figure 1 shows the time series and dis-tribution of annual Florida hurricanes overthe 108-year period. There are a total of 67Florida hurricanes. There are 62 yearswithout a Florida hurricane and one year(2004) with 4 different hurricanes affect-
ing the state. There are more years withoutFlorida hurricanes during the second halfof the 20th century (Elsner et al. 2004).Approximately 16 percent of the yearshave more than one hurricane event. Theaverage annual number of Florida hur-ricanes is 0.62 hur/yr with a variance of0.72 (hur/yr)2. Assuming that hurricaneoccurrence in Florida follows a Poisson dis-tribution similar to the climatological rec-ord for the rest of the United States af-fected by hurricanes (Elsner and Jagger2006), this translates into a 46 percentchance that Florida will be hit by at leastone hurricane each year.
Florida’s hurricane season runs fromthe beginning of June through the end ofNovember (even though storms occasion-ally occur outside this season), but themost active months are September fol-lowed by October (Figure 2). In fact, twiceas many hurricanes have hit Florida in Oc-tober than in August. Collectively themonths of June, July, and November ac-
Hurricanes and Damage Costs 113
Mon
thly
Cou
nt (1
900!
2007
)
05
1015202530
J J A S O N D
a
150 200 250 300 350
0.0
0.2
0.4
0.6
0.8
1.0
Julian Day
Cum
ulat
ive
Dis
tribu
tion
!!!!! !!!
!!!!!!!
!!!!!!!!!!!!!!!!!!!
!!!!!!!!!!!! !!
Jun. 1 Sep. 8 Dec. 17b
Figure 2. Florida’s intra-seasonal hurricane occurrence. (a) Monthly counts and (b) cumulativedistribution function (CDF) of Florida hurricanes.
count for about 16 percent of all Floridahurricanes while August, September, andOctober account for the remaining 84 per-cent. Yet the monthly distribution does nottell the entire story; another way to look atintra-seasonal activity is with the cumula-tive distribution function (CDF). The CDFsuggests a division of the season into fourdistinct periods. The periods are markedby a nearly straight line on the CDF indi-cating a constant probability of observinga hurricane during the period, but the pe-riods do not have equal lengths. The earlyperiod runs from 1 June through 31 July.The early mid period (with a steeper slopeon the CDF) runs from 1 August throughabout 5 September. The mid-season pe-riod, featuring the highest probability ofobserving a Florida hurricane, runs from 6September through 25 October, althoughthere is a slight break in activity duringlate September and October (representedby a small line just prior to Julian Day 300in Figure 2). The late period runs from26 October through the end of November.
The 1st quartile, median, and 3rd quar-tile dates are 242, 260, and 284, respec-tively. This implies that only 25 percent ofthe Florida hurricane season is typicallycomplete by 31 August, half the season iscomplete by 18 September, and 75 percentof the season is over by 12 October. The 1st
quartile date, 31 August, falls into the sec-ond period (early-mid) established by theCDF. As expected, the median date, 18September, falls during the mid-seasonperiod established by the CDF, which isslightly more than a week after the mediandate for all Atlantic hurricanes (Elsner andKara 1999). The 3rd quartile date, 12 Octo-ber, also falls into this mid-season period.None of these dates fall into the late periodestablished by the CDF showing that, forthis sample of Florida hurricanes, the ma-jority of the season is over by 12 October.
When comparing this seasonality ofFlorida hurricanes to the remainder of theUnited States coastline, Florida stands asunique. Coasts that are affected by Atlantichurricanes extend from Texas to Maine,
114 jill malmstadt, kelsey scheitlin, and james elsner
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 22
Annual Sequence Number of a Florida Hurricane
Freq
uenc
y
0
2
4
6
8
10
Figure 3. Florida’s occurrence of hurricanes by sequence number. The number refersto the sequence of tropical storms and hurricanes within a season, where 1 is the
first named tropical cyclone of the season.
and have been regionalized into four areasfor comparison purposes. The first regionextends from Texas to Alabama, and thesecond region consists of only Florida. Thethird and fourth regions consist of the en-tire eastern coast beginning with Georgiaand ending at Maine, and are separated atthe state of Virginia. It is found that for allregions, the month of highest hurricaneoccurrence is September. The main differ-ence between Florida and the other re-gions is that Florida experiences hurri-canes during the entire span of the Atlantichurricane season with a median date of 18September, while the other regions experi-ence hurricane seasonality of a differentscale. Region 1 experiences hurricanes fromJune through October with a median dateof 30 August. Region 3 is affected by stormsfrom July through November and has a me-dian date of 4 September, and the northernmost region, region 4, experiences stormsduring the shortest amount of time thanany other region, July through September
with a median date of 11 September. Flor-ida, therefore, is unique in its susceptibilityto hurricanes because it experiences amuch longer season than other regions,and it experiences more storms later in theyear than any other region as its medianlandfall date is later than anywhere else.
It is interesting also to consider whichtropical storm of the season is most likelyto strike Florida. Figure 3 shows the dis-tribution of storm numbers associatedwith Florida hurricanes. Storm number re-fers to the sequence of tropical storms andhurricanes within a season, with the firstnamed tropical cyclone being storm one.The plot shows that historically the mostlikely hurricane to affect Florida is the 5th
tropical cyclone of the season followed, inlikelihood, by the 2nd tropical cyclone.Florida has seen every sequence numberthrough 15. The highest sequence number(22) was Hurricane Wilma in 2005. Therewere a record 27 tropical storms and hur-ricanes during this remarkable season.
Hurricanes and Damage Costs 115
1900 1920 1940 1960 1980 2000
900920940960980
Year
Min
. P (h
Pa)
a
Min. P (hPa)
Freq
uenc
y
900 940 980
0
5
10
15
b
1900 1920 1940 1960 1980 2000
80
100
120
140
Year
Max
. Spe
ed (k
t)
c
Max. Speed (kt)Fr
eque
ncy
60 80 100 120 140
0
5
10
15
d
Figure 4. Florida’s hurricane intensity at time of landfall (1900–2007). (a) Time series of minimumcentral pressure (hPa) and (b) distribution of minimum central pressure. (c) Time series of maximum
wind speed (kt) and (d) distribution of maximum wind speed.
The average intensity (as measured bythe minimum central pressure) of Floridahurricanes at the time of first landfall is 966hPa (millibars), and 90 kts as measured bythe maximum wind speed. The average in-tensity of Florida hurricanes at second land-fall is 981 hPa (75 kt). All five of the secondlandfalls (for storms striking Florida morethan once) occurred over the northwesternpart of the state. These five events have amean central pressure at first landfall of 970hPa showing that, for these five storms, theaverage difference between first and sec-ond landfall is +11 hPa. This increase in airpressure from first to second landfall in-dicates a decrease in intensity for thesestorms. The lowest pressure of any Floridahurricane is 892 hPa, which occurred withthe Labor Day hurricane of 1935 that de-molished the middle Florida Keys.
Figure 4 shows the time series and his-togram of minimum central pressures andwind speeds at landfall. If the hurricanemade more than one landfall (or Keyscrossing), the highest intensity (lowestpressure and highest wind speeds) is used.There appears to be no obvious long-termtrend in these variables for this sample ofFlorida hurricanes. The distributions areskewed (negatively for pressure and posi-tively for wind speed) as expected from aset of data representing a threshold pro-cess (only cyclones at hurricane intensityare considered).
Locations of hurricane landfalls areshown in Figure 5. The points delineatewhere the eye crossed the shore (or crossedthe Keys). Symbols signify intensity as de-termined by the maximum wind speedsand grouped by the Saffir/Simpson cate-
116 jill malmstadt, kelsey scheitlin, and james elsner
!88 !86 !84 !82 !80 !78
24
26
28
30
32
!
!
!
!!
!
!
!
!
!
!
!
!
!
!
!88 !86 !84 !82 !80 !78
24
26
28
30
32
Sarasota!!!
Cedar Key!!!
!!!Vero Beach!
!
64!83 kt83!96 kt96!114 kt114!135 kt> 135 kt
°°N
°°WFigure 5. Florida hurricane landfalls 1900–2007. The locations indicate the landfall (or Key crossing)
locations, including second landfalls. The symbols denote hurricane intensity as measured by themaximum wind speed and categorized by the Saffir-Simpson hurricane damage potential scale. Anotable lack of hurricane strikes have occurred along the peninsula north of Cedar Key. All of thestrongest hurricanes (Category 4 & 5) have occurred south of a line from Sarasota to Vero Beach.
gories. Landfalls are more common overthe southern half of the peninsula includ-ing the Keys and along the panhandle.There is a notable lack of hurricane strikesalong the northeast coast and around thewestern peninsula north of Cedar Key. Allof the strongest hurricanes (categories 4and 5, having wind speeds of 114 kt or
greater) have occurred south of a line fromSarasota to Vero Beach.
The size of hurricanes directly affectingFlorida varies from storm to storm. Figure6 shows the time series and distribution ofthe radius of maximum wind (RMW) atlandfall as an indication of hurricane size.Five of the 67 Florida hurricanes do not
Hurricanes and Damage Costs 117
!!
!
!
!
!
!
!
!
!!
!
!
!
!
!!!
!
!
!
!
!
!
!
!!
!
!!
!
!
!
!
!
!
!
! !
!
!
!
!
!
!!
!
!
!
!
!
!!
!
!
!
!
!
!
!
!
!!
!
1900 1940 1980
20
40
60
80
100
120
Year
Rad
ius
of M
ax. W
inds
(km
)a
Radius of Max. Winds (km)
Freq
uenc
y
0 20 60 100
0
5
10
15
20
b
Figure 6. Florida’s hurricane size at time of landfall (1900–2007). (a) Time series of radius ofmaximum wind (km) and (b) distribution of the radius.
have a value for RMW. The variation insize is quite large but there appears to bea modest trend toward larger hurricanes.The average size is 41 km (radius) with avariance of 434 km2. The distribution ispositively skewed with most hurricaneshaving an RMW between 20 and 60 km,and only a few greater than 80 km. Histor-ically, the smallest storm was HurricaneCharley (2004) at 8 km and the largestwas Hurricane Earl (1998) at 119 km.
damage statistics
The data on damage losses from hur-ricanes are taken from Pielke et al. (2008).The values represent the total estimatedeconomic damage amounts normalized to2005 dollars. The values are based on totaldamage estimates as opposed to insuredloss figures. Economic damage is the di-rect loss associated with a hurricane’s im-pact. It does not include losses due to busi-ness interruption or other macroeconomiceffects including increases in demand forconstruction materials and other house-
hold items. Total damage costs are twicethe estimated insured damage costs. De-tails and caveats of the normalization pro-cedure are provided in Pielke et al. (2008).The complete set of data used in this studyis provided as an Appendix. It should benoted that prior to 1940, 32 storms madelandfall somewhere on the United States’coastline with no reported damages, whereonly 8 such storms have occurred since1940 (Pielke et al. 2008). Some damagelikely occurred during all early 20th centurystorms but the lack of data probably resultsin an undercount of the overall economicloss from the storms affecting the UnitedStates prior to 1940, and, if at least onehurricane strikes Florida every two years,there is an undercount of overall damagein Florida prior to 1940 as well.
There are two sets of damage estimatesbased on slightly different normalizationprocedures provided in Pielke et al. (2008).The two approaches are the methodologyused by Pielke and Landsea (1998), whichadjusts for inflation, wealth, and popula-tion updated to 2005, and the methodol-
118 jill malmstadt, kelsey scheitlin, and james elsner
ogy used by Collins and Lowe (2001),which adjusts for inflation, wealth, andhousing units updated to 2005 (Pielke etal. 2008). Pielke et al. (2008) have takenthe methodologies given by Pielke andLandsea (1998) and Collins and Lowe(2001) and have slightly adjusted theirmethodologies to be appropriate for 2005dollars. Here we focus on the data set fromthe Collins and Lowe methodology, butnote that both data sets are quite similar. Inboth cases, researchers have estimated to-tal dollar value of damage that historicstorms would have caused had they oc-curred in 2005—given all the growth anddevelopment that has taken place sincethese historical storms occurred. The Col-lins and Lowe methodology produces atemporally consistent estimate of eco-nomic damage from past tropical cyclonesaffecting the U.S. Gulf and Atlantic coasts.Results presented in this study are not sen-sitive to the choice of data set. The Collinsand Lowe methodology is used for thisstudy, as opposed to that of Pielke andLandsea, because the housing unit variableincluded in Collins and Lowe is more rele-vant when dealing with economic loss thanpopulation statistics. The Collins and Lowe(2001) values, adjusted to 2005 dollars inPielke et al. (2008), are presented in ourAppendix under the Damage column.
Table 1 lists the top ten all-time hur-ricane loss events in Florida since 1900.The damage amount (loss) is in billions ofU.S. dollars. Fourteen of the 67 Floridahurricanes do not have an estimated lossvalue for unknown reasons. Topping thelist is the Great Miami Hurricane of 1926with an estimated total damage to Floridaof $129 billion. Again, this dollar figurerepresents an estimate of the total damageif the same cyclone were to have hit in
Table 1. Top ten loss events from Floridahurricanes (1900–2007). Damage amount is inbillions of U.S. dollars, normalized to the dollar
value of 2005. Loss values come from theadjustments made to Collins/Lowe (2001)
presented in Pielke et al. (2008).
Rank Storm Year RegionLoss
($bn)
1 Great Miami 1926 SE 129.02 Andrew 1992 SE 52.33 Storm # 11 1944 SW 35.64 Lake Okeechobee 1928 SE 31.85 Donna 1960 SW 28.96 Wilma 2005 SW 20.67 Charlie 2004 SW 16.38 Ivan 2004 NW 15.59 Storm # 2 1949 SE 13.5
10 Storm # 4 1947 SE 11.6
2005. Hurricane Andrew, which hit south-east Florida in 1992, comes in second witha damage tag of $52.3 billion if it wouldhave hit in 2005. Note that 3 of the top tencostliest Florida hurricanes occurred in2004 and 2005.
The total normalized losses for the 53Florida hurricanes (for which contempo-rary damage estimates are available) wouldbe $459 billion if these storms occurred in2005. Eighty percent of this total is from thetop 11 (21 percent) storm event losses. Themedian loss amount is $2.21 billion, but the95th percentile value is $33.3 billion. Figure7 shows the time series and histogram ofhurricane damage losses in the state of Flor-ida since 1900. The distribution is highlyskewed with many relatively small lossesand few very large losses. The Great MiamiHurricane of 1926 is clearly the worst lossevent (normalized) in Florida since 1900.There are two years with total losses of less
Hurricanes and Damage Costs 119
1900 1920 1940 1960 1980 20000
20406080
100120
Year
Dam
age
(200
5 $U
S b
n) a
Damage (2005 $US bn)
Freq
uenc
y
0 50 100 150
0
10
20
30
40
b
1900 1920 1940 1960 1980 2000
789
1011
Year
Log
Dam
age
(200
5 $U
S b
n)
c
Log Damage (2005 $US bn)Fr
eque
ncy
6 7 8 9 10 11
02468
1012
d
Figure 7. Florida’s hurricane damage losses (1900–2007). (a) Time series and (b) distribution oflosses by event and (c) time series and (d) distribution of the logarithm (base 10) of losses. Some years
are without loss events.
than $15 million. The Insurance Service Of-fice (ISO), a private corporation that pro-vides information about risk assessment,defines a catastrophe as an event that causesmore than $25 million in insured ($50 mil-lion total) losses and causes a major disrup-tion (Insurance Information Institute2008). Of the Florida hurricanes that haveavailable economic loss values, 96 percentof the events were above this $50 millionthreshold. There are only two storms of the53 in this sample that do not have lossesexceeding this amount, and they are Flor-ence (1953) and Floyd (1987).
A good way to examine skewed distri-butions is to use logarithms. Figure 7cshows the time series and histogram aftertaking the logarithm (base 10) of each an-nual loss amount. In this figure, a value ofnine indicates a billion dollar loss, a value
of 10 indicates a $10 billion loss, and avalue of 11 indicates a $100 billion loss.Consistent with the modest increase in sizefor hurricanes affecting Florida seen in theprevious section, there appears to be aslightly increasing trend in the upper andlower quartile amounts of normalizeddamage since 1900, although theseslightly increasing trends are not statis-tically significant.
To estimate the annual probability ofyearly losses exceeding specified amountsthe normalized data are fit to a model. Themodel consists of the generalized Paretodistribution (GPD) to describe the behav-ior of extreme losses and the Poisson dis-tribution to specify the rate of loss yearsabove a given threshold level (Jagger andElsner 2006). Here the threshold value isset at $250 million as a compromise be-
120 jill malmstadt, kelsey scheitlin, and james elsner
Annual Probability
Lo
ss E
xce
ed
an
ce
(b
n)
$0.1
$1
$10
$100
0% 5% 10% 15% 20%
Figure 8. A model for Florida’s hurricane damage losses. The solid curve is based on using a generalizedPareto distribution for describing the magnitude of yearly total loss amount and a Poisson distribution
for the number of years exceeding a threshold loss amount of $250K. The dashed lines are the upperand lower 95 percent confidence limits on the loss model. The small boxes are empirical estimates ofthe loss amount and the large box corresponds to a total loss of $25 bn. The empirical estimates are
based on 1-e[-a*(rank-0.5)/M], where a is the number of years with losses (M) divided by the total numberof years (N), and rank is the order of losses, with a rank of 1 being the greatest loss.
tween being low enough to retain enoughyears to estimate the parameters of theGPD, but high enough so that the yearlyloss amount (exceedance) follows a GPD(Jagger and Elsner 2006). The model spe-cifies exceedance loss levels as a functionof annual probabilities.
Figure 8 shows the model as a curve ona semi-log plot; the higher the annual loss,the lower the probability of occurrence.The model indicates a 5 percent chance oflosses exceeding about $19.6 billion on anannual basis and a 10 percent chance oflosses exceeding $5.8 billion. Finally, Flor-ida can expect a storm to produce at least$1 billion in damage once every five years(a probability of 20 percent in any givenyear). According to the model, a loss of atleast $25 billion occurs with an annual
probability of about 2.1 percent, which is apercentage point below the state’s esti-mate of 3.1 percent mentioned in theIntroduction. Although this difference isnot statistically significant it shows thatthe state of Florida estimates their cata-strophic losses ($25 billion +) to occur abit more often than this model suggests.
trends and associations
As seen in the previous section, thereappears to be increasing trends in the sizeand intensity of Florida hurricanes and innormalized damage costs. To examine theseobservations in more detail, trends are com-puted and examined using ordinary leastsquares regression and quantile regression(Elsner et al. 2008). Ordinary regression is a
Hurricanes and Damage Costs 121
1900 1940 1980920930940950960970980990
Year
Min
. P (h
Pa)
a
1900 1940 1980
80
100
120
140
Year
Max
. Spe
ed (k
t)
b
1900 1940 1980
20406080
100120
Year
Rad
ius
of M
ax. W
ind
(km
)
c
1900 1940 1980
789
1011
Year
Log
Dam
age
(200
5 $U
S b
n)
d
Figure 9. Florida’s hurricane trends (1900–2007). (a) minimum central pressure, (b) maximum windspeed, (c) radius of maximum wind, and (d) logarithm of damage costs (losses). The thick line is the
trend in mean value. The upper, thin line is the trend in the upper quartile values for wind speed,RMW, and damage cost and the lower quartile for central pressure. The lower, thin line is the trend in
lower quartile values for wind speed, RMW, and damage cost, and the upper quartile for centralpressure. The trend values and standard errors are given in Table 2.
model for the conditional mean, where themean is conditional on the value of the ex-planatory variable. Quantile regression ex-tends ordinary least-squares regression toquantiles of the response variable. Quan-tiles are points taken at regular intervalsfrom the cumulative distribution function ofa random variable. The quantiles mark a setof ordered data into equal-sized data sub-sets. Thus, quantile regression is a model forthe conditional quantiles. For trend analysisthe explanatory variable is year. Relation-ships between hurricane characteristics andlosses are also examined.
Figure 9 shows the median and upperand lower quartile trends in hurricane in-tensity and size at landfall, and the corre-
sponding damage costs. Downward trendsare found in the mean and lower quartileof minimum central pressures, and up-ward trends are found in the mean andupper quartile of maximum wind speeds,both showing an increase in the strong-est storms over time. An upward trend isfound in the upper quartile of the size ofFlorida hurricanes. This indicates that onaverage and for the strongest cyclones,Florida hurricanes are getting more pow-erful over time. Trend values along withtheir associated standard errors are givenin Table 2. The relatively large standarderrors on the trends indicate the increasesin trend values shown in Table 2 are notstatistically significant against the null hy-
122 jill malmstadt, kelsey scheitlin, and james elsner
Table 2. Trend statistics and standard error(1900–2007). The lower quartile of the pressure
trend corresponds to the upper quartile ofdamage cost trend.
Minimum Central Pressure (hPa/yr)
Trend S.E.upper quartile 0.000 0.0367mean –0.064 0.0687lower quartile –0.083 0.0918
Maximum Wind Speed (hPa/yr)
Trend S.E.upper quartile +0.091 0.1403mean +0.094 0.0787lower quartile 0.000 0.1225
Radius to Maximum Wind (km/yr)
Trend S.E.upper quartile +0.318 0.1038mean +0.093 0.0865lower quartile –0.204 0.0614
Log Damage Costs (/yr)
Trend S.E.upper quartile +0.0042 0.00398mean +0.0018 0.00408lower quartile +0.0056 0.00356
pothesis of no trend. However, the upwardtrends in the 25th and 75th percentiles ofdamage costs might be associated with theupward trends in the power characteris-tics of hurricanes as seen in the previoussection.
Figure 10 shows a scatter plot matrixalong with regression lines of damage costsas a function of hurricane characteristics.It is clear that there is a statistically signifi-cant relationship between the intensity ofthe hurricane at landfall and the amount ofdamage. This strong relationship is seen
using either minimum central pressure ormaximum wind speed as the indicator ofhurricane intensity. However, the rela-tionship between damage amount andhurricane size is much less clear. In fact,the weak negative relationship is counter-intuitive as the larger hurricanes are asso-ciated with somewhat less damage. Thissomewhat puzzling observation can be ex-plained by the fact that hurricane intensityis inversely related to hurricane size for thisset of hurricanes. Thus the larger hur-ricanes tend to be weaker and thus causeless damage.
It has been suggested that estimationsof potential losses from hurricanes com-bine intensity and size characteristics(Kantha 2006). The Carvill Hurricane In-dex (CHI), discussed in Kantha (2006),determines a numerical measure of thepotential for damage from a particularhurricane event. On the Chicago Mercan-tile Exchange, the CHI is used as the basisfor trading hurricane futures and for trad-ing options about how best to mitigate thestorms, and captures this idea using thefollowing equation:
CHI = (v/vo)3 + 1.5 (r/ro) (v/vo)2 (1)
where v is the maximum wind speed (kt),vo is the threshold hurricane-wind speed(64 kt), r is the radius of threshold hurri-cane-wind speed or greater (km), and ro isthe threshold radius (97 km). To obtain r,a form of the Rankine vortex equation isused to obtain the radial decay of thewinds from their maximum value (Hol-land 1980). The equation is given by:
r = rmax (v/vo)1.5 (2)
The CHI is computed from the set ofFlorida landfalling characteristics. As ex-pected, the relationship between damage
Hurricanes and Damage Costs 123
920 940 960 980
789
1011
a
Minimum Central Pressure (hPa)
Log
Dam
age
(200
5 $U
S b
n)
80 100 120 140
789
1011
b
Maximum Wind Speed (kt)
Log
Dam
age
(200
5 $U
S b
n)
20 40 60 80 100 120
789
1011
c
Radius to Maximum Winds (km)
Log
Dam
age
(200
5 $U
S b
n)
5 10 15 20
789
1011
d
Carvill Hurricane Index
Log
Dam
age
(200
5 $U
S b
n)
Figure 10. Scatterplot matrix of damage costs and hurricane characteristics. Logarithm of damage costas a function of (a) minimum central pressure, (b) maximum wind speed, (c) radius to maximum
wind, and (d) Carvill Hurricane Index. The thick line is the trend in mean value, the thin lines are the95 percent confidence limits on the trend.
losses and the CHI is positive and signifi-cant. However the relationship does notappear to be stronger than either of theintensity estimates alone. The strong linkbetween hurricane intensity and damagecost coupled with the rather weak linkwith hurricane size indicates that theSaffir-Simpson hurricane scale, which isbased solely on wind speed, is, in largemeasure, an adequate measure of poten-tial damage amount, at least in Florida.Yet based on the somewhat better correla-tion of losses with minimum central pres-sure (see Table 3), we argue that centralpressure be used as a single variable forpotential loss estimation.
For instance, regressing the logarithm(base 10) of losses onto the minimum cen
tral pressure, an equation representing aloss index for Florida, called the Floridahurricane loss index (FHLI), is defined bythe following equation:
FHLI = 1040.912-0.0329pmin (3)
where pmin is the minimum central pres-sure in units of hPa forecast at landfall.Values of FHLI are damage estimates ex-pressed in dollar amounts. For multiplelandfalls the lowest minimum pressure isused. This model (which is not applicablefor hurricanes that hit only the FloridaKeys) explains only 40 percent of the vari-ability in the logarithm of Florida lossamounts but compares favorably with theCHI, which explains less than 28 percentof the losses. Table 4 shows resulting ex-
124 jill malmstadt, kelsey scheitlin, and james elsner
Table 3. Correlation of hurricane characteristicsat landfall with damage costs (losses) based on53 Florida hurricanes. Correlation coefficient r
and the associated 95 percent confidence intervalon that correlation under the null hypothesis of
zero correlation.
r 95% Confidence Interval
P min –0.59 (–0.74, –0.38)W max +0.52 (+0.29, +0.70)RMW –0.13 (–0.39, +0.14)CHI +0.53 (+0.30, +0.70)
pected economic loss from the FHLI basedon the pressure categorization associatedwith the Saffir-Simpson Scale (Kantha2006). The expected losses do not reflectfuture changes in wealth and inflation, northe expected increases in coastal develop-ment. It is important to note that these lossindex values will be highly dependentupon where the storm makes landfall andthe amount of development and popula-tion in the affected area.
Of course, the actual amount of dam-age a hurricane inflicts will also depend tosome extent on its forward speed and therate at which the wind subsides over land.Neither of these characteristics are consid-ered here, but have been analyzed else-where. Huang et al. (2001) considers eco-nomic loss as a function of the wind decayrate, and Watson and Johnson (2004)look at forward speed as one of the param-eters of their hurricane loss estimationmodels. These characteristics could be in-cluded in this model in a future study totry and increase its ability to explain thevariability in the logarithm of Florida lossamounts.
summary
More hurricanes strike Florida thananywhere else in the United States. Rec-ords of Florida hurricanes have recentlybeen updated and are reliable back to1900. This study examines various statis-tics of hurricanes affecting the state overthe period 1900–2007 and their associ-ated damage costs.
It is shown that the annual count ofFlorida hurricanes is consistent with a ran-dom Poisson process with a mean of 0.62hurricanes per year that translates to anannual probability of 46 percent for atleast one hurricane. Florida differs fromother regions of the United States in termsof hurricane seasonality because it is af-fected by storms throughout the entire At-lantic hurricane season, and it experiencesstorms later into the year than any otherarea of the United States’ coastline.
Although the variability in the amountof damage is quite large from one hurri-cane to the next, normalized losses are in-creasing over time, which is consistentwith the slight increases in hurricane in-tensity and hurricane size. The model pro-vided shows that on an annual basis, wecan expect a 10 percent chance of lossesexceeding $5.8 billion and a 5 percentchance of losses exceeding $19.6 billion.In addition, each year Florida has a 20 per-cent chance of experiencing at least $1 bil-lion in hurricane related losses; in otherwords, the State can plan on at least $1billion in losses once every five years.
Of the hurricane landfall characteris-tics examined here, the best predictor ofpotential losses is minimum central pres-sure. Hurricane size by itself or in com-bination with hurricane intensity does not
Hurricanes and Damage Costs 125
Table 4. Expected loss computed using the Florida Hurricane Loss Index based on categorical pressurevalues presented by Kantha (2006). Approximate exponent values are x where FLHI = 10x. Expected
loss is given in US dollars, normalized to 2005 dollar amounts. These approximate losses do not reflectfuture changes in wealth, inflation, and property, and are highly reliant on where the storm actually
makes landfall in terms of development and population.
CategoryPmin
ValuesApproximate
Exponent Values (x)Approximate Expected Loss
(Normalized 2005 $US)
1 989–980 8.40–8.69 $250 million–$499 million2 979–965 8.70–9.17 $500 million–$1.49 billion3 964–945 9.18–9.90 $1.50 billion–$7.99 billion4 944–920 9.91–10.69 $8.00 billion–$49.99 billion5 [920 ?10.70 ?$50.00 billion
improve on the simpler relationship. Anestimate of potential losses from hur-ricanes can be obtained by a formula in-volving only an estimate of the minimumpressure at landfall. Expected economicdamage costs are computed using theFHLI and categorized to provide a scalesimilar to the Saffir-Simpson for economicloss based on minimum central pressure.
In one sense Florida has been ratherfortunate. Although the 2004 and 2005seasons featured 7 Florida hurricanes,there are more years during the secondhalf of the record without a Florida hur-ricane. Moreover, despite some large lossessince 1950, Florida has not seen a repeat,in terms of losses, of the Great Miami Hur-ricane (of 1926).
Florida, along with other coastal states,is in a race to retrofit and harden its in-frastructure before another major storm oc-curs. Over the past 20 years alone, Hur-ricane Andrew (1992) almost made a directhit on downtown Miami, Hurricane Floyd(1987) made a last minute turn away fromthe state, and Hurricane Charley (2004)veered east and away from the Tampa Bay
area just hours before landfall. Had any ofthese storms made direct strikes on urbanareas, they could have caused losses largerthan anything Florida has experienced todate.
Recently, researchers have made im-provements in understanding and predict-ing hurricane intensity (Jones et. al 2006,Davis et. al 2008), hurricane tracks (Bar-ret et al. 2006), and seasonal hurricaneactivity (Elsner and Jagger 2006). In com-bination with this paper, better under-standing of hurricane activity and result-ing damage can better prepare coastalcommunities with what to expect witheach approaching season, allowing for in-formed decisions by their citizens, policymakers and insurance agencies about thefuture of Florida’s hurricane seasons andthe proper way to mitigate.
appe
ndi
xYe
arR
egio
nSe
qN
ame
Mo
Da
Lat
Lon
Wm
axPm
inR
MW
Tim
eSN
BRD
amag
eCo
de
1903
FLSE
3St
orm
39
1126
.180
.175
976
8023
0039
75.
2 bi
llion
119
03FL
NW
3St
orm
39
1330
.185
.680
975
NA
2100
397
NA
219
04FL
SE3
Stor
m3
1017
25.3
80.3
7098
5N
A70
040
7N
A1
1906
FLSW
2St
orm
26
1725
.280
.875
979
4870
041
6N
A1
1906
FLN
W6
Stor
m6
927
30.4
88.7
9595
880
1200
420
NA
019
06FL
SW8
Stor
m8
1018
25.1
80.8
105
953
6411
0042
214
2 m
illio
n1
1909
FLSW
10St
orm
1010
1124
.781
100
957
4018
0045
043
3 m
illio
n1
1910
FLSW
5St
orm
510
1826
.582
9595
523
600
456
814
mill
ion
119
11FL
NW
2St
orm
28
1130
.387
.670
985
NA
2200
458
286
mill
ion
119
12FL
NW
4St
orm
49
1430
.488
.465
990
4880
046
6N
A0
1915
FLN
W4
Stor
m4
94
30.1
85.4
8097
5N
A10
0048
0N
A1
1916
FLN
W13
Stor
m13
1018
30.3
87.4
100
974
3514
0049
4N
A1
1916
FLSW
14St
orm
1411
1524
.582
7098
548
1800
495
NA
119
17FL
NW
3St
orm
39
2930
.486
.785
966
6130
049
8N
A1
1919
FLSW
2St
orm
29
1024
.481
.711
592
927
400
505
720
mill
ion
119
21FL
SW6
Tam
paBa
y10
2528
82.8
9095
234
1900
516
3.2
billi
on1
1924
FLN
W4
Stor
m4
915
30.2
85.7
6599
048
1500
531
NA
119
24FL
SW7
Stor
m7
1021
25.9
81.4
8097
535
300
534
NA
119
25FL
SW2
Stor
m2
121
27.2
82.5
6599
2N
A43
053
7N
A1
1926
FLN
E1
Stor
m1
728
28.3
80.6
7596
026
600
538
3.6
billi
on1
1926
FLSE
6G
rtM
iam
i9
1825
.680
.311
593
535
1200
543
129
billi
on1
1928
FLSE
1St
orm
18
827
.480
.380
977
4860
055
6N
A1
1928
FLSE
4La
ke9
1727
.180
.111
593
551
600
559
31.8
bill
ion
119
29FL
SE2
Stor
m2
928
25.1
80.7
8594
851
1800
563
256
mill
ion
119
29FL
NW
2St
orm
29
3029
.785
.365
988
NA
1700
563
NA
2
1933
FLSE
5St
orm
57
3027
.480
.370
985
4820
0059
1N
A1
1933
FLSE
12St
orm
129
426
.980
.111
594
824
400
598
1.4
billi
on1
1935
FLSW
2La
borD
ay9
324
.980
.714
089
211
130
620
NA
319
35FL
NW
2La
borD
ay9
429
.783
.475
985
3919
0062
03.
1 bi
llion
119
35FL
SE6
Stor
m6
114
25.9
80.1
6597
319
1500
624
5.6
billi
on1
1936
FLN
W5
Stor
m5
731
30.4
86.6
8097
335
1500
629
126
mill
ion
119
39FL
SE2
Stor
m2
811
27.3
80.2
7098
548
1900
659
NA
119
39FL
NW
2St
orm
28
1329
.784
.970
985
NA
065
9N
A2
1941
FLSE
5St
orm
510
625
.480
.310
595
434
1030
675
362
mill
ion
119
41FL
NW
5St
orm
510
729
.884
.775
981
3490
067
5N
A2
1944
FLSW
11St
orm
1110
1926
.982
.465
962
4763
070
735
.6 b
illio
n1
1945
FLN
W1
Stor
m1
624
28.9
82.6
8097
548
1100
708
NA
119
45FL
SE9
Stor
m9
915
25.3
80.3
120
940
2322
0071
610
.1 b
illio
n1
1946
FLSW
5St
orm
510
827
.882
.765
989
4830
072
399
1 m
illio
n1
1947
FLSE
4St
orm
49
1726
.480
.113
594
748
1500
728
11.6
bill
ion
119
47FL
SW8
Stor
m8
1012
25.2
81.2
7598
024
200
732
540
mill
ion
119
48FL
SW7
Stor
m7
921
24.6
81.6
105
NA
NA
1400
740
NA
319
48FL
SW7
Stor
m7
922
25.6
81.2
100
963
130
740
3.6
billi
on1
1948
FLSW
8St
orm
810
524
.781
110
NA
2420
0074
1N
A3
1948
FLSE
8St
orm
810
525
.280
.411
097
729
2200
741
565
mill
ion
119
49FL
SE2
Stor
m2
827
26.8
80.1
130
954
420
744
13.5
bill
ion
119
50FL
NW
5Ea
sy9
528
.782
.610
595
827
1200
760
973
mill
ion
119
50FL
SE11
King
1018
25.8
80.2
9098
811
600
766
3.7
billi
on1
1953
FLN
W8
Flor
ence
926
30.3
86.2
8098
248
1700
793
14.3
mill
ion
119
56FL
NW
7Fl
ossy
924
30.3
86.5
8097
434
2300
829
711
mill
ion
119
60FL
SW5
Don
na9
1024
.880
.811
593
034
700
864
NA
3
appe
ndi
x (c
ontin
ued)
Year
Reg
ion
Seq
Nam
eM
oD
aLa
tLo
nW
max
Pmin
RM
WTi
me
SNBR
Dam
age
Code
1960
FLSW
5D
onna
910
25.9
81.6
120
938
3416
0086
428
.9 b
illio
n1
1964
FLSE
5Cl
eo8
2726
.180
.190
968
1310
0089
64.
7 bi
llion
119
64FL
NE
6D
ora
910
29.9
81.3
9596
163
500
897
6.6
billi
on1
1964
FLSW
11Is
bell
1014
25.8
81.3
110
964
1921
0090
262
4 m
illio
n1
1965
FLSE
3Be
tsy
98
25.1
80.4
110
952
3711
0090
64.
0 bi
llion
119
66FL
NW
1A
lma
69
29.9
84.4
8097
347
2000
910
81.3
mill
ion
119
66FL
SW9
Inez
104
2580
.575
984
2718
0091
813
1 m
illio
n1
1968
FLN
W8
Gla
dys
1019
28.8
82.6
7097
732
500
936
495
mill
ion
119
72FL
NW
2A
gnes
619
29.9
85.4
6598
337
2100
979
411
mill
ion
119
75FL
NW
5El
oise
923
30.4
86.2
110
955
2612
3010
082.
8 bi
llion
119
79FL
SE4
Dav
id9
326
.780
8597
250
1500
1044
2.2
billi
on1
1985
FLN
W5
Elen
a9
230
.489
.110
095
927
1300
1100
3.8
billi
on0
1985
FLN
W11
Kate
1121
3085
.585
967
1922
3011
061.
1 bi
llion
119
87FL
SW7
Floy
d10
1225
.280
.465
993
7622
0011
192.
6 m
illio
n1
1992
FLSE
2A
ndre
w8
2425
.580
.314
592
219
905
1166
52.3
bill
ion
119
95FL
NW
5Er
in8
227
.780
.475
985
5660
011
911.
4 bi
llion
119
95FL
NW
5Er
in8
330
.387
.185
974
2415
0011
91N
A2
1995
FLN
W15
Opa
l10
430
.387
.110
094
280
2200
1201
6.3
billi
on1
1998
FLN
W5
Earl
93
30.1
85.7
7098
711
960
012
3112
6 m
illio
n1
1998
FLSW
7G
eorg
es9
2524
.582
.290
975
1716
3012
331.
1 bi
llion
119
99FL
SW9
Iren
e10
1524
.681
.765
987
5613
0012
49N
A3
1999
FLSW
9Ir
ene
1015
25.2
81.2
6598
448
1900
1249
1.2
billi
on1
2004
FLSW
3Ch
arle
y8
1327
82.1
125
947
821
0013
1316
.3 b
illio
n1
2004
FLSE
6Fr
ance
s9
527
.280
.290
960
8460
013
169.
7 bi
llion
1
2004
FLN
W9
Ivan
916
30.2
87.8
9594
337
730
1319
15.5
bill
ion
120
04FL
SE10
Jean
ne9
2627
.380
.210
595
172
400
1320
7.5
billi
on1
2005
FLN
W4
Den
nis
710
30.4
8711
094
613
1930
1329
2.2
billi
on1
2005
FLSE
11Ka
trin
a8
2526
80.1
6598
219
2230
1336
NA
120
05FL
SW22
Wilm
a10
2425
.981
.611
095
074
1030
1347
20.6
bill
ion
1
Not
es19
06 S
torm
2 P
min
take
n fr
om th
e hu
rric
ane
earl
ier i
n th
e da
y19
06 S
torm
8 P
min
take
n fr
om th
e hu
rric
ane
earl
ier i
n th
e da
y19
11 S
torm
2 L
andf
all p
oint
in A
laba
ma
1916
Sto
rm 4
P m
in a
nd w
ind
spee
d fr
om h
urri
cane
3 h
ours
late
r19
26 S
torm
6 A
dditi
onal
loss
es fr
om F
LSW
& A
L 1.
08E+
1019
41 S
torm
5R
MW
take
n fr
om 2
nd la
ndfa
ll19
48 S
torm
7 P
min
take
n fr
om e
arlie
r alo
ng tr
ack
1960
Sto
rm 5
RM
W ta
ken
from
isla
nd p
ass
1985
Sto
rm 5
Los
ses f
rom
AL
& M
S ar
e in
clud
ed20
04 S
torm
6 R
MW
take
n fr
om w
hen
it w
as o
ver t
he B
aham
as20
04 S
torm
9 L
andf
all p
oint
in A
laba
ma
Dam
age
= C
ollin
s/Lo
we
(200
1) d
ata
adju
sted
to 2
005
dolla
rs p
rese
nted
in P
ielk
e et
al.
(200
8) ro
unde
d to
1 si
gnifi
cant
dec
imal
poi
ntSN
BR=
Sto
rm se
quen
ce n
umbe
r as c
atal
ogue
d in
HU
RD
AT d
atas
etCo
de=
0: N
ot d
irec
t lan
dfal
l, hu
rric
ane
forc
e w
inds
may
hav
e be
en fe
lt so
mew
here
in th
e st
ate
1: F
irst
dir
ect l
andf
all,
Keys
hit
if th
is is
the
only
dir
ect h
it in
the
stat
e2:
Sec
ond
dire
ct la
ndfa
ll3:
Key
s hit
if th
e hu
rric
ane
mad
e la
ndfa
ll el
sew
here
in th
e st
ate
Uni
tsW
ind
spee
ds (
kt);
Pre
ssur
e (h
Pa);
RM
W (
stat
ute
mile
s)
130 jill malmstadt, kelsey scheitlin, and james elsner
acknowledgmentsThanks go to King Burch for his contribution
of background material on Florida’s insurance.Thanks also go to Thomas Jagger for his assistancewith the loss model. All computations were com-pleted using R Statistical Package (R Develop-ment Core Team 2007). This work is supported bythe National Science Foundation (ATM-0738172)and by the Florida Catastrophic Storm Risk Man-agement Center. Finally, the lead author wouldlike to thank Chris Meindl for his encouragementand editorial guidance.
referencesAIR Worldwide. 2005. The Coastline at Risk,
September 2005. Accessed at http://www.air-worldwide.com/—public/NewsData/000797/The—Coastline—at—Risk-AIR—Worldwide.pdf.
Barrett, B.S., L.M. Leslie, and B.H. Fiedler.2006. An Example of the Value of StrongClimatological Signals in Tropical CycloneTrack Forecasting: Hurricane Ivan (2004).Monthly Weather Review 134:1568–1577.
Changnon, S.A. 2003. Shifting economicimpacts from weather extremes in theUnited States: A result of societal changes,not global warming. Natural Hazards29:273–290.
Changnon, S.A., D. Changnon, E.R. Fosse,D. Hoganson, R.J. Roth, and J. Totsch. 1997.Effects of recent extremes on the insuranceindustry: Major implications for theatmospheric sciences. Bulletin of the AmericanMeteorological Society 78:425–435.
Collins, D.J., and S.P. Lowe. 2001. A macrovalidation dataset for U.S. hurricane models,Casualty Actuarial Society Forum. Arlington,VA: Casualty Actuarial Society. Accessed athttp://www.casact.org/pubs/forum/01wforum/01wf217.pdf.
Davis, C., W. Wang, S.S. Chen, Y. Chen,K. Corbosiero, M. DeMaria, J. Dudhia,
G. Holland, J. Klemp, J. Michalakes,H. Reeves, R. Rotunno, C. Snyder, andQ. Xiao. 2008. Prediction of LandfallingHurricanes with the Advanced HurricaneWRF Model. Monthly Weather Review136:1990–2005.
Demsetz, H. 1962. Economic gains from stormwarnings: Two Florida case studies,Memorandum RM-3168-NASA. SantaMonica, CA: The Rand Corporation. 43pp.
Elsner, J.B., and A.B. Kara. 1999. Hurricanes ofthe North Atlantic: Climate and Society. NewYork: Oxford University Press.
Elsner, J., J. Kossin, and T. Jagger. 2008. Theincreasing intensity of the strongest tropicalcyclones. Nature 455:92–95.
Elsner, J.B., and T.H. Jagger. 2006. Predictionmodels for annual U.S. hurricane counts.Journal of Climate 19:2935–2952.
Elsner, J.B., X.F. Niu, and T.H. Jagger. 2004.Detecting shifts in hurricane rates using aMarkov chain Monte Carlo approach.Journal of Climate 17:2652–2666.
Emanuel, K. 2005. Increasing destructivenessof tropical cyclones over the past 30 years.Nature 436:686–688.
Hartwig, R.P. 2008. Florida Property InsuranceFacts. New York: Insurance InformationInstitute. Accessed January 2008 at http://www.iii.org/media/research/floridafacts08.
Ho, F.P., J.C. Su, K.L. Hanevich, R.J. Smith, andF.P. Richards. 1987. Hurricane Climatologyfor the Atlantic and Gulf coasts of the UnitedStates, NOAA Technical Memorandum,NWS-38. Silver Spring, MD: NationalOceanic and Atmospheric Administration,National Weather Service. 195pp.
Holland, G.J. 1980. An analytic model of thewind and pressure profiles in hurricanes.Monthly Weather Review 108:1212–1218.
Huang, Z., D.V. Rosowsky, and P.R. Sparks. 2001.Hurricane simulation techniques for the
Hurricanes and Damage Costs 131
evaluation of wind-speeds and expectedinsurance losses. Journal of Wind Engineeringand Industrial Aerodynamics 89:605–617.
Insurance Information Institute. 2007. FloridaCitizens Property Insurance Corporation(CPIC): What It Is and How It Works.Accessed August 2007 at http://server.iii.org/yy—obj—data/binary/777211—1—0/FLCitizensbrief2007.pdf.
Insurance Information Institute. 2008. Factsand Statistics, Catastrophes: U.S. Accessed athttp://www.iii.org/media/facts/statsbyissue/catastrophes/.
Jagger, T.H., and J.B. Elsner. 2006. Climatologymodels for extreme hurricane winds nearthe United States. Journal of Climate19:3220–3226.
Jagger, T.H., J.B. Elsner, and M.A. Saunders.2008. Forecasting U.S. insured hurricanelosses. In Climate Extremes and Society,editors H.F. Diaz and R.J. Murnane, 189–208. New York: Cambridge University Press.
Jones, T.A., D. Cecil, and M. DeMaria. 2006.Passive-Microwave-Enhanced StatisticalHurricane Intensity Prediction Scheme.Weather and Forecasting 21:613–635.
Kaczor, Bill. 2008. Florida, Berkshire Hathawaystrike $224 million deal on insurance fund.Insurance Journal: Southeast News.Accessed July 2008 at http://www.insurancejournal.com/news/southeast/2008/07/31/92371.htm.
Kantha, L. 2006. Time to Replace the Saffir-Simpson Hurricane Scale?. Eos 87 (1):3.
Katz, R.W. 2002. Stochastic modeling ofhurricane damage. Journal of AppliedMeteorology 41:754–762.
Landsea, C.W., C. Anderson, N. Charles,G. Clark, J. Dunion, J. Fernandez-Partagas,P. Hungerford, C. Neumann, andM. Zimmer. 2004. The Atlantic hurricanedatabase re-analysis project:
Documentation for the 1851–1910alterations and additions to the HURDATdatabase. In Hurricanes and Typhoons: Past,Present, and Future, editors. R.J. Murnameand K-B Liu, 177–221. New York: ColumbiaUniversity Press.
Pielke, R.A, Jr., and C.W. Landsea. 1998.Normalized hurricane damages in theUnited States: 1925–1995. Weather andForecasting 13:621–631.
Pielke, R.A, Jr., J. Gratz, C.W. Landsea,D. Collins, M.A. Saunders, and R. Musulin.2008. Normalized hurricane damage in theUnited States: 1900–2005. Natural HazardsReview 9:29–42.
R Development Core Team. 2007. R: A languageand environment for statistical computing.Vienna, Austria: R Foundation for StatisticalComputing.
Sugg, A. 1967. Economic aspects of hurricanes.Monthly Weather Review 95:143–146.
Watson, C.C. and M.E. Johnson. 2004.Hurricane loss estimation models. Bulletinof the American Meteorological Society85:1713–1726.
Williams, J. and I. Duedall. 1997. FloridaHurricanes and Tropical Storms. Gainesville,FL: University Press of Florida.
jill malmstadt is a Master’s student in theDepartment of Geography at the Florida StateUniversity, Tallahassee, FL 32306. Email:[email protected] scheitlin is a Ph.D. student in theDepartment of Geography at the Florida StateUniversity. Email: [email protected]. james elsner is a Professor in theDepartment of Geography at the Florida StateUniversity. Email: [email protected]. Hisresearch interests include hurricanes, climate,and spatial statistics.