Introduction

Given significant population increases in tropical cyclone (TC)-prone coastal regions of the continental United States (CONUS), it is crucial to have a publicly available fatalities database that allows for a better understanding of past fatalities directly related to TCs. A clearer picture of how fatality trends have evolved over time may, in turn, enhance TC risk communication in vulnerable regions. Despite the significance of an accurate assessment of potential TC-induced fatalities, limited information exists1. Two papers assessing historic CONUS TC direct fatalities are Rappaport2 and Rappaport3. Rappaport2 reported on direct fatalities between 1970–1999 that identified among other things “the date, cause, and location of each fatality.” Based upon an analysis of 65 Atlantic TCs that affected the CONUS during this timeframe, he estimated a total of 600 direct fatalities and concluded that freshwater flooding was the primary cause of fatalities during this time interval. In 2014, this data was updated to include the years from 1963 to 2012. This study demonstrated that water was still the dominant cause of direct death, but that storm surge was the largest individual fatality driver (~49%), with freshwater flood being the second largest driver (~27%)3.

Unfortunately, the raw data sets described in the above publications are not publicly available. Today, the only free, publicly available TC fatalities dataset updated to 2024 is housed with the Billion-Dollar Weather and Climate Disasters4 dataset. However, this dataset does not provide a complete list of fatalities, and is limited to only storms that have caused more than 1 billion dollars in damage. Furthermore, the fatalities are a total summary of both “direct” and “indirect” fatalities, and are not broken down by cause of death/peril (e.g., storm surge, freshwater flood, and tornado). In addition, databases like EM-DAT5, an Emergency Events Database maintained by the Center for Research on the Epidemiology of Disasters (CRED), and SHELDUS6, a county-level hazard loss database housed at Arizona State University, also report on deaths per disaster event, but like NCEI, these fatalities are not broken down by cause of death.

The primary goal of this study is to provide a transparent, publicly-available dataset for direct CONUS TC fatalities between 1963–2024. Throughout the remainder of this paper, all fatality counts and figures refer exclusively to direct deaths associated with TCs. Our dataset is likely incomplete and imperfect, but in making it public, the authors hope it can be improved upon and added to over time. We examined 767 TCs (i.e., tropical storms and hurricanes) that formed in the Atlantic Basin from 1963 to 2024, of which 181 had direct CONUS fatalities. Specifically, our dataset includes:

  • Name and year of TC landfall location, Saffir-Simpson Hurricane Wind Scale category, maximum winds and landfall pressure, if available

  • Cause of direct fatality, if available

  • State in which direct fatality occurred, if available

  • Source(s) of information, and citation for each storm

Results

Direct fatalities

Our database shows that 2642 people died in the CONUS, or its coastal waters, from TCs during the period 1963–2024 (Fig. 1). This number likely represents a modest underestimate of the total, because it does not fully account for people who were reported as missing but actually died. We found that one out of every four Atlantic basin TCs caused loss of life in the CONUS, and that on average, three TCs per year caused CONUS fatalities. The total equates to ~43 deaths per year on average, but as also noted in Rappaport (2014), the number of deaths varies greatly by year. Between 1963–2024, four storms were responsible for >100 direct fatalities: Katrina (2005), Camille (1969), Helene (2024), and Agnes (1972). Between 2016–2024 seven years had annual fatalities >30 individuals, with 6 individual storms having >20 direct fatalities: Matthew (2016), Harvey (2017), Florence (2018), Ida (2021), Ian (2022), and Helene (2024).

Fig. 1: Annual CONUS direct TC fatalities between 1963–2024.
Fig. 1: Annual CONUS direct TC fatalities between 1963–2024.
Full size image

Colors indicate cause of death broken down by freshwater flood, offshore, storm surge, surf, tornado, wind, and unknown.

Katrina (2005) outpaces all events, having approximately twice as many deaths (520 deaths) in comparison to the second highest death toll from Camille (1969; 288 deaths) (Table 1). Agnes (1972; 125 deaths) and more recently Helene (2024; 176 deaths) also caused greater than 100 deaths. Betsy (1965; 93 deaths), Sandy (2012; 74 deaths), Harvey (2017; 68 deaths), Ian (2022; 66 deaths), Floyd (1999; 56 deaths) and Ida (2021; 55 deaths) all resulted in more than 50 fatalities. While all of these storms made landfall in the CONUS as hurricanes, both Tropical Storm Allison (2001; 41 deaths) and Tropical Storm Amelia (1978; 33 deaths) caused >30 deaths, demonstrating that tropical storms can produce significant fatalities. We also note that while Agnes made landfall in Florida as a hurricane, >100 of Agnes’s deaths occurred in the mid-Atlantic due to freshwater flooding, associated with a secondary landfall of Agnes as a tropical storm. A variety of different mechanisms can drive catastrophic inland flooding, including slow translation speed, interaction with elevated topography, interaction with other low-pressure systems and predecessor rain events, as well as other factors.

Table 1 Top 20 deadliest CONUS TCs from 1963 to 2024
Full size table

Peril-driven fatalities

The causes of direct death for 2450 out of 2642 were classified using the National Weather Service established groupings: storm surge, freshwater flood (rain), surf (here we include rip current to align with Rappaport2), offshore (rough seas/boating), wind (including tree fall), and tornado. Three of the top five deadliest storms had fatalities dominated by storm surge, with Katrina (2005) claiming the greatest number of lives lost due to storm surge (516) during the research period (1963–2024) (Table 1 and Fig. 2). Camille (1969) and Betsy (1965) also caused significant surge-driven loss of life (118 and 81 respectively). Thirteen of the top twenty storms were dominated by rainfall-driven loss of life. When we look at the incidence of direct deaths from freshwater flood across all storms, we find that 10% of storms were responsible for >1 rainfall death, while surge was responsible for >1 death in only 3% of storms between 1963–2024 (Fig. 2). These results indicate that death from TC rainfall was more common than from surge, but that some single event TCs between 1963–2024 caused significant loss of life from storm surge-related drowning (5 individual TCs each caused >40 direct fatalities).

Fig. 2: Direct fatalities by peril by pentad from 1963 to 2024, where the bar colors denote the cause of death.
Fig. 2: Direct fatalities by peril by pentad from 1963 to 2024, where the bar colors denote the cause of death.
Full size image

The red asterisk at 1963–1964 denotes that the period is shorter (i.e., 2 years) than the other periods displayed.

Between 1963–2024, 80% of fatalities were from water perils (freshwater flood, surge, surf and offshore) (Fig. 3), while ~13% of deaths were due to wind and tornado. The remaining ~7% of deaths were unknown. When comparing the 1963–2012 dataset representing the established baseline of TC mortality patterns reported on in Rappaport3 to our 1963–2024 extended dataset, we note some similarities and some differences (Fig. 3). Rappaport3 found that 88% of deaths between 1963 and 2012 were from water perils (relative to 80% from 1963 to 2024). Rappaport3 found that storm surge was responsible for 49% of fatalities, while our study finds that storm surge was responsible for 33%. Rainfall-related deaths comprise 27% of fatalities in Rappaport3, while our study finds that 36% of fatalities are rainfall-related.

Fig. 3
Fig. 3
Full size image

Direct fatalities from 1963 to 2024 broken down by cause. The specific percentage of total direct fatalities by cause is also provided.

A discussion article from the National Hurricane Center7 notes that 57% of direct fatalities were from freshwater flooding between 2013 and 2022. Overall percentages like these in our dataset, and other datasets, are highly influenced by individual events that resulted in significant fatalities, such as Katrina (2005), Camille (1969), and Helene (2024). To understand changes through time, we look at the cause of death on a pentad level. In general, rainfall fatalities tend to dominate the record in all decades, except for the 2000s, which were dominated by Katrina’s (2005) storm surge loss of life (516 surge fatalities) (Fig. 2).

Relationship between seasonal accumulated cyclone energy (ACE) and CONUS TC fatalities

Accumulated cyclone energy (ACE) is an integrated metric accounting for TC intensity and duration and is often used as a way to evaluate seasonal TC activity, including serving as the metric for which NOAA classifies TC seasons. It is calculated as the sum of the squares of the estimated 6-hourly maximum sustained wind speed (in units of knots) for all systems while they are at least tropical storm strength8. The relationship between seasonal ACE and CONUS TC fatalities reveals a significant positive correlation, indicating that years with higher ACE tend to produce higher fatalities (Fig. 4). A negative binomial model is used to fit the fatality counts (Fig. 4a). A logarithmic link function is then used to fit the relationship between seasonal ACE and the number of fatalities. Note that the negative binomial model fitted after excluding direct fatalities from Katrina (2005) still shows an increasing trend with statistical significance; however, it is not as pronounced as the model with Katrina (Supplemental Fig. 1). We also note that pentads with increased direct fatalities tend to be associated with active periods for CONUS landfalling named storms, CONUS landfalling hurricanes and basinwide ACE (Supplemental Fig. 2). The interannual correlation between CONUS landfalling hurricanes and basinwide ACE between 1963 and 2024 is 0.58, while the correlation between ranked CONUS direct fatalities (to reduce the impact of the strong outlier 2005 season) and basinwide ACE is 0.57 (Supplemental Fig. 3).

Fig. 4: Relationship between seasonal ACE and direct fatalities.
Fig. 4: Relationship between seasonal ACE and direct fatalities.
Full size image

a Fitted response and 95% CI band of non-zero fatality counts from a negative binomial model, and b Box and whisker analysis of the approximate top 25% (15 years), middle 50% (32 years) and bottom 25% (15 years) of seasonal ACE years and corresponding fatalities between 1963 and 2024. The horizontal line denotes the median value, the box denotes the range from the first quartile (Q1) to the third quartile (Q3) (e.g., the interquartile range (IQR)), while the whiskers are defined to span from Q1 − (1.5* IQR) to Q3 + (1.5 * IQR).

While ACE is significantly related to the number of fatalities, substantial scatter around the trend line demonstrates that ACE alone is an incomplete predictor of TC mortality. Notably, several high-ACE seasons resulted in relatively few fatalities, while some lower ACE years produced disproportionately high death tolls, suggesting that factors beyond overall seasonal energy, such as storm track, landfall location, population exposure, and preparedness, also play critical roles in determining fatality outcomes. The box-and-whisker analysis (Fig. 4b) further illustrates this disconnect by stratifying fatalities by Atlantic seasonal ACE. The top 25% of ACE seasons (>127 ACE; red) show the highest median fatality count (45 deaths) and the greatest variability, with an interquartile range spanning from 24 to 73 deaths. The middle 50% ACE seasons (63–127 ACE; orange) demonstrate lower median fatalities (23 deaths) with moderate variability, while the bottom 25% of ACE seasons (<63 ACE; blue) consistently produce the fewest fatalities, with a median of 4 deaths and minimal spread.

Genesis, track landfall and storm lifetime

The genesis, track and landfall locations for the top 10 deadliest storms are shown in Fig. 5. There is no consistent preferred track or genesis location for these TCs. Several storms formed in the central tropical Atlantic (Floyd (1999), Harvey (2017) and Betsy (1965)) while others formed in the Caribbean (Camille (1969), Ian (2022), Ida (2021), Helene (2024), Sandy (2012), Agnes (1972)) and consequently tended to be shorter-lived (Table 1). Katrina (2005), the deadliest storm since 1963, formed over the Bahamas. We note that of the top 20 deadliest TCs (Table 1), 9 were TCs that formed less than four days before landfall. We also note that large loss-of-life storms are not all associated with landfall in densely populated regions. While Katrina caused high mortality in densely populated New Orleans, the regions with the most deaths associated with Camille and Helene were not in densely populated regions. For example, approximately half of the fatalities associated with Camille occurred in Nelson County, Virginia, a sparsely populated, rural, mountainous county. Similarly, Helene’s deaths were spread across numerous inland rural and mountainous counties in North Carolina, South Carolina, Georgia, and Florida9. Buncombe County, North Carolina (population ~270,000 in 2020), saw the highest fatalities during Helene10.

Fig. 5: TC tracks and landfall locations for the top 10 deadliest TCs.
Fig. 5: TC tracks and landfall locations for the top 10 deadliest TCs.
Full size image

Track colors denote Saffir–Simpson Hurricane Wind Scale category, while the triangle size denotes the number of fatalities. The map was created using ArcGIS® software by Esri. ArcGIS® and ArcMap™ are the intellectual property of Esri and are used herein under license. Copyright © Esri. All rights reserved. For more information about Esri® software, please visit www.esri.com. Sources: ESRI. TomTom. Garmin, FAO, NOAA, USGS, © OpenStreetMap contributors, and the GIS User Community. Atlantic tropical storm and hurricane track data were downloaded from the National Hurricane Center best track database (HURDAT233).

Discussion

Previous studies of United States TC fatalities have suggested that TC-induced mortality has generally declined over time since 19501,2,11. When we analyze direct fatalities aggregated by year, we find no significant temporal trend (r² < 0.01, Supplemental Fig. 4) between 1963 and 2024, indicating that CONUS hurricane fatalities have not shown a consistent increase or decrease over this 60-year period. However, we do note that decades with the highest fatalities are the 1960s, the 2000s (due largely to Katrina) and the 2020s, even though this decade has several years yet to go (Fig. 2). Of particular interest is that recent storm activity has resulted in considerable wind fatalities, with the 2020s yielding the highest wind fatalities of any decade in this study. At least half of the wind fatalities (65 fatalities) in the 2020s were due to Helene (2024). Of the 65 direct wind fatalities, 61 were caused by falling trees during the storm. Even though the immediate cause of the tree-related deaths was high wind, rainfall likely played an important role in saturating the soil and weakening the root anchoring systems, subsequently making trees more prone to uprooting or breaking during strong winds. The sheer volume of rainfall associated with Helene caused catastrophic flooding despite moderate drought conditions across the Southern Appalachians prior to its landfall12.

As previous studies have noted11, an overall decline in fatalities or an insignificant trend through time is meaningful considering the rapid population growth in high TC-risk areas such as Florida and Texas. The lack of a trend may indicate the relative success of TC forecasting, emergency management, and increased rigor of building codes in offsetting an increasing exposure. For example, Ian severely impacted Fort Myers in 2022, producing a ~15-foot storm surge13. The growth in Lee County, FL, alone between 1960 and 2020 was 72% (54,539 to 760,822 residents). Sixty-six deaths were reported, with 44 from storm surge. These fatalities likely would have been much higher without improvements in storm surge modeling, operational issuance of Storm Surge Watches and Warnings (starting in 2017) and pre-landfall mandatory evacuations. These watches and warnings provide clearer communication of surge threats with recent enhanced evacuation compliance and evacuation infrastructure and planning (post-Katrina, 2005).

We found that above-normal Atlantic ACE seasons from 1963 to 2024 (>126.1 ACE per NOAA’s definition) increase the odds of a high fatality year (≥30 fatalities) to 70% (Fig. 4). Seasons that are not classified as above-normal ACE seasons (<126.1 ACE) had only a 21% chance of ≥30 fatalities. The presence of outliers in the top ACE quartile, including at least one season exceeding 90 fatalities, underscores that the most energetically active seasons carry elevated risk for catastrophic mortality events, though high ACE does not guarantee high fatalities.

While not in our dataset, given that the National Hurricane Center has not completed its post-season reports, the 2025 season is a good example of an above-normal ACE season with relatively few CONUS TC fatalities. Preliminary, the 2025 seasonal ACE was 133, above the climatological average of 123 and considered active, with 13 named storms and 4 major hurricanes in the Atlantic basin. However, only one tropical storm, and no hurricanes, made landfall in the CONUS. Initial reports indicate 6 rainfall deaths from Tropical Storm Chantal, and two rip current deaths from Tropical Storm Erin but no other TC-related deaths in the CONUS in 2025.

Freshwater flood tends to dominate the record through time (Fig. 2), with 10% of all landfalling TCs over the last 61 years causing at least one freshwater flood death. Several recent catastrophic rainfall events (Harvey (2017; 65 rainfall deaths), Florence (2018; 17 rainfall deaths), Ida (2021; 49 rainfall deaths), and Helene (2024; 95 rainfall deaths)) give prominent examples of significant inland freshwater flooding fatalities. With recent research pointing toward a potential increase in TC-related rainfall in both observational14,15,16 and modeling studies17,18 this trend may be expected to continue in the future. As also previously noted, some of these rainfall storms tended to have shorter Atlantic lifetimes (<96 h between named storm and CONUS landfall) (Table 1), making these exceptional events more difficult to forecast and potential evacuations more difficult to implement. This has significant implications for risk assessment and emergency management strategies; however, these strategies may need to be tailored to be region-specific.

In 2017, Harvey made landfall along the Texas coastline. Although Harvey’s time between regeneration as a tropical storm and landfall was only 65 h, the storm was the most significant TC rainfall event in United States history, both in scope and peak rainfall amounts. The National Hurricane Center reported record daily rainfall accumulations from Hurricane Harvey, ~24 to 48 h after landfall in the Houston area, when the storm’s slowing motion and interaction with a stationary front produced the heaviest rain19. Even though the National Hurricane Center forecasts had warned of significant rainfall potential in advance of tropical storm formation, many residents were not prepared for this TC rainfall event. Jonkman et al.20 showed that most of Harvey’s drowning fatalities (65 fatalities) occurred outside of FEMA’s 100-year floodplain. The study concluded that a combination of social vulnerability indexing (SVI), including SVI’s household income, age and disability status, and percentages of racial minorities and flood susceptibility accurately hindcasted TC fatalities during Harvey21. Studies from Katrina (2005) combining SVIs and storm surge footprints found similar results, highlighting the need for targeted emergency management toward flooding risk, particularly in socially vulnerable regions22,23,24,25.

While coastal populations continue to grow, so too do inland communities. Several recent lethal inland flood TCs are a stark reminder that TC deaths can occur far inland from coastal zones and also far from the track of the storm center. In these cases, a conventional focus on storm intensity at landfall may miss a critical dimension of risk. Helene (2024) made landfall in the Florida panhandle region only 83 h after tropical storm formation. In just 12 h, Helene accelerated towards the northeast and then turned northwest over the Appalachian Mountains, causing tremendous flooding over the Southern Appalachians. Most of the stream gauges in the region experienced their largest historical annual maximum event peaks, and the devastating freshwater flooding resulted in 95 of the 176 deaths recorded. Recent research argues, however, that Helene was not unpredictable if one considers the historical records that show at least ten major flood events between 1924 and 2013 in this region, with the most severe events associated with TCs26. Moreover, there may be a mismatch between the perception of flood risk and the actual risk in this inland region. Data from Amorim et al.26 show that market penetration by the National Flood Insurance Program (NFIP) was incredibly low in the western portions of North Carolina where ~1.5% of the buildings in this area were covered for flood (in comparison to ~14% of the buildings in the eastern third of North Carolina), with NFIP uptake rate actually decreasing over the last decade.

This lack of flood insurance penetration in inland communities has been previously pointed out as a chronic issue pervasive across the United States27. This may point to a larger conversation around perception of TC risk in the region. Elliot28 noted that just 2167 of ~127,000 properties (~1.7%) in the hardest hit Buncombe County of North Carolina were in the official FEMA flood zones and subject to flood purchase requirements. Unfortunately for many Buncombe County residents, Helene’s floodwaters penetrated far beyond these flood zones. The concern here is that a mandatory purchase requirement may imply that if you are not required to have a policy, because you are outside of the official flood zone, then you must be “safe”. Elliott28 attributes the lack of flood insurance to the mismatches between the perception and the actual flood risk (as discussed above), in addition to flood insurance cost, the confusion surrounding what home insurance covers and limited private sector penetration.

We stress that risk communication and public messaging should emphasize inland flood threats alongside coastal flood threats. As noted by Hong et al.29 there is a need for tailored, region-specific approaches to disaster preparedness and risk communication. The discrepancies between perceived and actual preparedness highlighted by Hong et al.29 and other studies can inform more effective strategies for enhancing community resilience in the face of increasing exposure in TC-prone regions. Infrastructure planning should include flood mitigation investments needed for inland areas previously considered outside primary TC risk zones.

Methods

Following Rappaport2, we report on deaths occurring as a direct result of the forces of TCs or their remnants. Direct deaths include those individuals who drowned in storm surge, rough seas, rip currents, and freshwater floods. Direct deaths also include casualties resulting from lightning and wind-related events (e.g., flying debris, collapsing structures), with a subset of statistics kept for fatalities specifically due to tornadoes. Deaths occurring “indirectly,” from such factors as heart attacks, housefires (e.g., from untended candles), electrocution from downed power lines, and vehicle accidents on wet roads, are not reported on in this paper. Because we are following Rappaport2 and Rappaport3, which focused on Atlantic TCs, we have not included fatalities from eastern North Pacific TCs. This may be included in a future study, but we suspect there may be challenges in collating fatalities from eastern North Pacific TCs (e.g., the change between the Redwood City National Weather Service Office to the National Hurricane Center between the 1987 and 1988 hurricane seasons, lack of Monthly Weather Review articles reviewing the 1963–1967 eastern North Pacific TC seasons as well as the general infrequency of eastern North Pacific TCs generating significant loss of life in the United States).

We follow closely Rappaport’s peril groupings3 with some slight changes. Like Rappaport3, we use the categories termed storm surge, freshwater flood (rain), surf (here we include rip current to align with Rappaport3), offshore (rough seas/boating) and wind (including also tree fall) and tornado. Not all of the direct death incidents cited could be completely or unambiguously documented, especially for those in the earlier years of the database (e.g., 1960s). Therefore, a direct death “Unknown” category is also implemented. This is a new addition, as previous studies did not note such a category, although Rappaport3 sometimes notes an “Other” category.

Data sources

To obtain the data, we used similar sources as reported in previous studies2,3, such as:

  1. 1.

    National Hurricane Center, Tropical Cyclone and Storm Reports and National Hurricane Center Preliminary Reports30,31

  2. 2.

    Tropical system summaries published in Monthly Weather Review31 NOAA Library32,33

  3. 3.

    National Hurricane Center “Storm Wallets”34

  4. 4.

    US Army Corps of Engineers Reports35

  5. 5.

    Peer-reviewed publications and books on specific storms (e.g., Katrina, Camille, etc.).

  6. 6.

    Coroner and Medical Examiner Reports

  7. 7.

    Newspapers

Methodological process

More specifically, as a starting point, we utilized the Atlantic tropical storms and hurricanes dataset downloaded from the National Hurricane Center best track database (HURDAT2; Landsea and Franklin 2013) that includes storm name, year, landfall date, landfall location, and landfalling wind and pressure. We chose not to include unnamed storms (systems that were identified as named storms after the season) but verified that unnamed storms were not associated with CONUS fatalities. We also included fatalities from non-landfalling storms in our database if there were direct CONUS fatalities associated with that non-landfalling storm. To assess the causes of fatalities, we focus on 1963–2024 due to the more detailed reporting that is available. Between 1963 and 2024, there were 767 named TCs that we examined.

Systematically, every National Hurricane Center Tropical Cyclone Report and Monthly Weather Review (MWR) Report was reviewed for fatality information for each storm. For storms that noted cause of death, and/or location of death, the information was included in the database. This process was replicated for each storm from 1963 to 2024. From our research, we found that starting in 1963, MWR began to break down fatalities by peril/cause of death (e.g., wind, surge and freshwater flood). For example, Beulah (1967) caused 10 deaths in Texas from floods and 5 deaths in Texas from tornadoes. However, other high-fatality TCs in the 1960s did not report the deaths by peril. For example, in 1965, Betsy was responsible for 189 deaths, but the causes of these fatalities were not reported in the Monthly Weather Review article. For Betsy, we relied on a U.S. Army Corps of Engineers report35 that detailed 81 Louisiana fatalities, most likely from drowning, and the Encyclopedia of TCs, Typhoons, and Cyclones36 documenting marine deaths. Similarly, for Hurricane Inez (48 deaths) no breakdown was found in the MWR article, so in this case we utilize the National Hurricane Center storm wallets which indicated 45 deaths from rough seas. We found Blanchard’s “Deadliest American Disasters and Large Loss of Life Events website”37 helpful in validating fatalities reporting for events with more than 10 fatalities.

CONUS named storm and hurricane landfalls from 1963 to 1970 and 1983 to 2024 were tabulated from the Atlantic and Oceanographic and Meteorological Laboratory’s landfalling named storm (https://www.aoml.noaa.gov/hrd/hurdat/uststorms.html) and landfalling hurricane tables (https://www.aoml.noaa.gov/hrd/hurdat/UShurrs_detailed.html). For 1971–1982, when the National Hurricane Center’s Best Track Reanalysis is incomplete, we took the intensity of the TC at its six-hourly position prior to landfall as its landfall intensity. Only the strongest CONUS landfall was included for each storm.

Data uncertainty

Several inherent challenges introduce uncertainty into TC fatality data analysis, which must be acknowledged when interpreting our findings. Large-scale catastrophic events often overwhelm local death investigation systems, resulting in incomplete or inconsistent documentation. Katrina (2005) exemplifies this challenge, where an initial official death toll of 1500 was reported in Monthly Weather Review: “The total number of fatalities directly related to the forces of Katrina is estimated to be about 1500 spread across four states, with about 1300 of these in Louisiana, about 200 in Mississippi, 6 in Florida, and one in Georgia. Especially for Louisiana and Mississippi, the number of direct fatalities is highly uncertain, and the true number will probably not ever be known. Several hundred persons are still reported missing in association with Katrina”8. An updated (4 January 2023) Hurricane Katrina National Hurricane Center TC report38 was later released with an estimated 520 direct fatalities following comprehensive epidemiological studies. Remote areas, resource-limited jurisdictions, and events causing widespread infrastructure damage present difficulties for systematic fatality documentation. TCs are often complex multi-hazard events that frequently produce simultaneous and overlapping hazards, including storm surge, extreme winds, inland flooding, tornadoes, and secondary effects such as infrastructure failure. This temporal and spatial overlap makes definitive attribution of individual fatalities to a single cause challenging or, in some cases, impossible. For example, a fatality occurring during combined surge and wind events may be classified based on investigator judgment rather than conclusive physical evidence. This ambiguity is particularly pronounced in coastal zones where surge and wind hazards coincide.

Methods for classifying and attributing TC-related fatalities have changed substantially over the study period. Earlier storms may have lacked systematic investigation protocols. The National Hurricane Center’s definition of direct versus indirect fatalities, the time window considered for storm-related deaths, and the rigor of causal determination have all evolved. Additionally, changes in coroner/medical examiner practices, improvements in forensic capabilities, and shifting standards for what constitutes a “TC-related” death introduce temporal inconsistencies in the dataset that may confound trend analysis.

As we noted earlier, fatality data are compiled from multiple sources, including the National Hurricane Center, Monthly Weather Review publications, individual state vital statistics, news reports, and academic studies. Discrepancies between these sources are common, and reconciliation often requires subjective judgment. We have attempted to use the most authoritative and recent compilations available, noting discrepancies where significant.