1. Introduction
Records of tornado occurrence are currently spread among individual countries and regional research groups, with varying quality, standards, and temporal coverage. Thus, it is difficult to draw conclusions about the worldwide spatial or temporal distribution of tornado frequency or strength. Efforts to estimate a global climatology have been made, notably by Fujita (1973) (Fig. 1) and Goliger and Milford (1998), but relied on fragmented and sometimes contradictory sources. The more recent European Severe Weather Database (ESWD), published by the European Severe Storms Laboratory (Dotzek et al. 2009), has been used to estimate tornado climatologies for Europe (Grieser and Haines 2020), but a similar resource that allows for a holistic worldwide view does not exist. Most analyses of tornado activity are performed only for smaller regions or individual countries (e.g., Hosen and Jubayer 2016; Chernokulsky et al. 2020). There is a need to move toward a globally unified record of tornado occurrence to standardize record-keeping and better inform studies of tornado climatology.
Estimated tornado occurrence in a 4-yr period by Fujita (1973).
Citation: Bulletin of the American Meteorological Society 105, 7; 10.1175/BAMS-D-23-0123.1
At the time of writing, tornado data prior to 1950 in the United States, where tornadoes are by far the most frequent (Fujita 1973), were not available in a digital format. Instead, they are contained in Significant Tornadoes, 1680–1991, a print volume by Grazulis (1993b) documenting thousands of significant or deadly tornadoes. This collection describes the geographic location of each tornado with textual descriptions (e.g., state, county, and town) rather than geographic coordinates, making it difficult to perform any spatial analyses.
Additionally, the current tools for examining tornado climatology along with relevant environmental data are difficult to use or incomplete and out of date. Consequently, there are no easily accessible resources that researchers or the general public can use to inform themselves about historical tornado occurrence.
Using datasets from a number of sources, we provide a global, digitized tornado database containing over 100 000 tornadoes in all tornadically active regions of the world. This is the most comprehensive such database to date. The database also includes a digitized and georeferenced version (covering 1680–1949) of the data in Grazulis (1993b), which can be much more easily analyzed. This expands the amount of digital data available for historical and climatological research. Detailed digital data for the United States were previously only available back to 1950 [as the “OneTor” dataset, from the Storm Prediction Center (2022), used in, e.g., Brooks et al. (2003); Barrett and Gensini (2013); Tippett et al. (2015); Agee and Taylor (2019)]. During writing, a similar digitized dataset derived from Grazulis (1993b), covering 1880–1989, was published by Fricker et al. (2023). The dataset described in this paper was developed independently, though the methodologies used are similar.
The unified data structure we develop supports detailed geometries, and a script was written to take in paths or polygons in Keyhole Markup Language (KML) form (a primary format used by Google Earth Pro and GIS software) and add them to the database. This means that printed high-resolution track maps, for example, in Fujita (1976), could be traced in Google Earth Pro and exported into the database, which makes them easier to access and work with. An online interface for the data was also built, with complex filtering capabilities as well as fifth major global reanalysis produced by ECMWF (ERA5) and Twentieth Century Reanalysis (20CR) layers for thousands of selected cases. This tool can be used to map and gain insight into historic tornado events, as already demonstrated in Schumacher (2022), where it is used to visualize a 1922 tornado outbreak described in Grazulis (1993b).
In this paper, we use the compiled data to investigate the global distribution of tornado occurrence and record-keeping, as well as the interdependence between the two. We present the compiled database as-is, acknowledging known biases existing therein. As the true global tornado climatology is fundamentally intertwined with the historical context surrounding its observation, we analyze trends in our database with both in mind.
2. Methodology
a. Data sources.
Generally, national or multinational tornado datasets are either maintained by governmental organizations (e.g., in the United States, Japan, or New Zealand) or assembled by individual research groups (e.g., in Argentina, Bangladesh, or Mexico). The ESWD, the primary dataset for Europe, is composed of many such regional datasets (e.g., Chernokulsky et al. 2020). These datasets vary in their accuracy, structure, and precision, and even more so in their size (Table 1). Their limitations are discussed in more detail in section 3a.
Major datasets used, with region, year of first record, year of last record, and number of tornadoes recorded. Some sources overlap.
Salient attributes we include (though not available for all tornadoes in all datasets) are intensity, intensity scale [e.g., the Fujita (F) or enhanced Fujita (EF) scales], date and time, path width, pathlength, fatalities, and injuries. We make an effort to retain attributes unique to constituent datasets, for example, the “time_accuracy” attribute present in the ESWD.
Information not included in any major available dataset was obtained from other reliable sources [such as government outlets, reports, or scientific case studies, e.g., De Coning and Adam (2000)], which were used both to add new tornadoes and fill in missing data for existing ones. Doing so is quite time-consuming, as a vast number of different sources are involved, and the effort to find and parse them is continuing. To regulate the process, every tornado in the database was given a mandatory “Source” attribute which either referred to one of our major datasets or another source (usually a web link). We did not infer tornado intensities from damage descriptions; if no intensity rating was explicitly given in a reliable source, the corresponding tornado was rated “(E)F-Unknown [(E)FU].”.
b. Merging modern U.S. data.
The OneTor dataset is compiled by the Storm Prediction Center (SPC) from tornado data kept by the National Centers for Environmental Information (NCEI) and is a widely used and comparatively accurate tornado dataset. However, it does not include individual county segments of tornado tracks, meaning that paths of multicounty tornadoes are slightly more well-resolved in the NCEI dataset. It also does not include the textual event narratives present in the NCEI dataset. For these reasons, the NCEI dataset was merged with OneTor using several Python scripts, which link NCEI path segments and then match them with OneTor tracks.
Information from the Damage Assessment Toolkit (DAT), a National Weather Service tool for collection and dissemination of storm survey information (NOAA 2018), was also included, via a similar process. The DAT contains detailed surveys of tornado paths and polygons representing damage swaths of varying intensities (as well as individual damage points, which we do not use) and primarily covers the period from 2011 to the present. The DAT was used only for path information for existing tornadoes in NCEI/OneTor. If the DAT has no track information for a tornado, NCEI is used. In general, if only the start and end points are available, they are connected to create the track.
The NCEI and DAT merges were done for two reasons: First, to increase the amount of geographic detail in our database, and second, to create connections between OneTor tornadoes and their NCEI textual narratives. For any other information (e.g., fatalities, width, and timing), OneTor took precedence over NCEI, which took precedence over the DAT. Both the NCEI and DAT merges were followed by quality control to ensure that information was properly merged, which was done by visualizing the data with the online interface (section 3f) and inspecting it for errors introduced by NCEI or DAT, which were then manually corrected. However, for the sake of simplicity, we interpreted the original OneTor dataset as-is and did not attempt to correct errors or biases therein, though this is an area where future work would be valuable. Section 3a discusses some of these biases, and a thorough description of anomalies specific to United States tornado data is available in Edwards et al. (2022).
c. Digitization of Grazulis data.
Geospatial data on tornado tracks in Grazulis (1993b) are in the form of distances and directions relative to towns or other landmarks (e.g., “Moved ENE [east-northeast] from south of Yorkville to near Oswego”). The pathlength and affected states and counties are also listed for each tornado (e.g., 7 miles, Kendall County, Illinois, for the same case).
A group including several of the authors and other contributors familiar with geographic information systems (GIS) and the history of tornadoes used this information to reconstruct tornado paths. They were instructed to use Google Earth software to draw a line based on the location descriptions and given direction of motion and then ensure that it intersected the given counties and was roughly the given length. For example, the sample descriptors above would be used to construct a 7-mile-long path heading from south of Yorkville east-northeast to near Oswego while remaining within Kendall County. Because exact distances to Yorkville and Oswego are not given, there is some uncertainty in the exact start and end locations, but the county and pathlength values act as constraints. The resultant start and end coordinates, along with given attributes such as fatalities, injuries, or path width, were recorded in a system of spreadsheets which was then compiled into our database.
In some cases, the counties affected and/or pathlength in a Grazulis entry were inconsistent with the description of impacted towns. For these cases, a hierarchy of importance for each piece of geographic information was established, as follows:
- 1)Towns directly impacted
- 2)Start and end location
- 3)Distances to towns mentioned
- 4)Counties affected
- 5)Pathlength
Elements higher on the list were favored at the expense of lower ones. The locations of towns directly impacted by the tornado were considered with the highest priority, as they would be the clearest in the historical record (i.e., most likely to be reported in newspapers). Overall, the pathlength was considered with lowest priority, as it is far less likely to be reported in the news. Consequently, and importantly, pathlengths in our database differ from those in Grazulis when they are not consistent with listed impacted towns.
When there was not enough information in Grazulis to establish an estimated path, the tornado was recorded as a single point. On rare occasions, there was no more geographic information included than the county impacted, in which case the approximate center of that county was recorded as the tornado location. Additionally, as the Grazulis dataset extends back more than three centuries, some descriptions included places which no longer existed, and for these cases, external resources such as historic maps were consulted in order to determine their location at the time of the event.
3. Data discussion
a. Limitations.
Many of the biases in the constituent datasets used are also present in the combined database. These limit the usability of the database for drawing certain climatological conclusions or at least require further data filtering work in order to do so. While key limitations are described here, any specific information pertaining to a particular dataset is available through the sources in Table 1.
1) Tornado numbers.
Many tornado events go unreported or unrecorded, especially in areas or time periods with less infrastructure or lower population density (Aguirre et al. 1994; Anderson et al. 2007). This is reflected, for example, in the Grazulis dataset, where the annual number of tornadoes increases sharply around 1875. This is primarily due to the fact that tornado information in the United States began to be compiled systematically for the first time by John Park Finley in 1879 (Galway 1985). The more general increase throughout the time period is due to a combination of better access to more recent tornado records (i.e., newspapers from 1900 being more readily available than those from 1700) and population expansion that allowed more tornadoes to be observed. Similar patterns are present in most other datasets we include [e.g., the ESWD—Dotzek et al. (2009)], but they are unlikely to be physical trends, and instead reflect societal growth over time. Underreporting of the total tornado number can also occur when tornado families are judged to be individual tornadoes (Doswell and Burgess 1988).
The effects of underreporting are not felt equally across the intensity distribution. Since tornado intensity scales are based on damage, intense tornadoes are much more likely to be documented (even when there exists no robust system of tornado observation) because their damage is more evident. For this reason, apparent (fictitious) increases over time in tornado activity are much more pronounced for subsignificant events (Fujita 1987; Brooks and Doswell 2001; Dotzek et al. 2009).
In historical reconstructions of tornado events, where the best available sources are newspaper reports written by journalists rather than meteorologists, severe wind events such as microbursts can be misidentified as tornadoes. Reports of tornadic activity are also sometimes sensationalized (Bradford 1999). Still, the underreporting effects previously discussed are generally of much larger magnitude, meaning that any observed tornado count for a given region, especially in the premodern era, is very likely an underestimate.
The compiled database contains just over 100 000 tornadoes. We are aware of a dataset for China, described in Chen et al. (2018), containing 4763 tornadoes, but were not able to access it (and we plan to incorporate it if it becomes available). An updated tornado climatology for Australia that adds more than a thousand events is also in development (Allen and Allen 2016). Otherwise, we are not aware of additional available and large datasets (>1000 novel events), so much of the remaining information is likely scattered in newspaper archives and other historical accounts. Thus, we believe the 100 000 mark provides a reasonable lower bound for the total number of tornadoes that have been recorded (and whose record still survives) in human history. However, given reasons mentioned prior, the actual number of tornadoes that occurred throughout the world during the documented period is likely far greater.
2) Intensity.
Even when all tornadoes are accounted for, there can be biases in the intensity distribution stemming from the fact that determining it is far from straightforward. The Fujita scale estimates intensity based on damage and was not well-codified until its enhancement in 2007 (Edwards et al. 2013), with further improvements planned (Marshall et al. 2022). Internationally, new intensity scales such as the Japanese enhanced Fujita scale (Okuda et al. 2015) or the Tornado and Storm Research Organization (TORRO) scale (Elsom and Meaden 1982) have been developed. These scales are intended to be convertible to the F scale but contain guidelines that better reflect the landscape and construction practices of the region. Translating between these scales can introduce bias, as was documented in the transition from the F to EF scale (Edwards et al. 2013, 2021). Currently, tornadoes in our database have ratings on one of the F or EF scales, though this would change when tornadoes are added whose ratings are only available on another scale.
A perhaps larger source of variation in recorded intensity distributions is the methodology and personnel used. For example, U.S. tornadoes in the period from 1950 to the mid-1970s were retroactively rated based on news accounts (Edwards et al. 2021), resulting in an inflation in the number of significant tornadoes during that time (Agee and Childs 2014; Grazulis 1993a). Generally, tornadoes for which an intensity rating is unassigned are given (E)F-Unknown, or (E)FU; in modern U.S. data, this occurs when a tornado is confirmed (e.g., visually), but no damage is found (NOAA 2021). In other cases, tornadoes may be rated (E)FU because they were not assessed by professional surveyors, and in some datasets [such as León-Cruz et al. (2022) and National Institute of Water and Atmospheric Research (NIWA 2021)], the tornadoes in this category are the majority. Conversely, other datasets, such as the Australian dataset, or the NCEI/SPC until 2016, rate tornadoes that did not cause observable damage (E)F0. This results in a shift in the overall intensity distribution.
3) Casualties.
Ashley (2007) notes that information on tornado fatalities is probably the most complete part of the post-1960 record, since fatalities are the most significant impact of tornadoes. Still, statistics of tornado casualties suffer from even more of a population bias than overall tornado counts, as casualties can only occur when tornadoes directly impact populated areas (Agee and Taylor 2019). Furthermore, the increase in tornado reports around 1875 coincides with an increase in the number of killer tornadoes, meaning there were likely unreported tornado deaths occurring before that time. Racial bias can also influence totals, for example, in the 1840 Natchez, Mississippi tornado, where possible deaths of slaves were not counted (Brooks and Doswell 2002).
It is also sometimes nontrivial to establish whether a casualty was a direct result of the tornado itself—for example, in cases of heart attacks, electrocutions, or house fires following tornado damage (Brooks and Doswell 2002). Conventionally, only direct casualties are considered, but some groups take a more holistic view when reporting totals. Injury counts also suffer from all of the biases listed above and are in general less likely to be recorded.
Many of these effects are stronger in countries with less developed infrastructure, such as Bangladesh, where estimates of death tolls are both high and uncertain. Most reported fatality counts for the 1989 Manikganj tornado, which is estimated to have caused a world-record 1300 deaths, are rounded to the nearest hundred (Cerveny et al. 2017). It should be noted here that in general, for all datasets, specific values indicating the precision of casualty information are not available, meaning that they must be inferred by the user.
4) Other attributes.
Path widths are difficult to measure and rather unreliable at high degrees of precision—for example, exhibiting unnatural, unexplained shifts over time in U.S. data for F3 tornadoes in the 1970s and 1980s (Brooks 2004). Complicating the climatology further, U.S. tornado surveys report mean path width prior to 1995 and maximum path width afterward (Agee and Childs 2014). Pathlength is more easily measured, but in older records can suffer from underestimates, through incomplete surveying of the entire track, or overestimates, through judgment of tornado families as individual tornadoes (Doswell and Burgess 1988).
Caution should be exercised when analyzing tornado paths and damage swaths (polygons) from the DAT. As there is no formally standardized procedure for generating them, they are drawn at the discernment of the surveyor based on damage points (NOAA 2018). This may result in difficulty when attempting to cross compare patterns at a high level of detail.
Some datasets recorded all tornado paths as single point coordinates (e.g., Mexico, Argentina, China), while others included paths of two or more points (e.g., Europe, Canada, Japan). Most datasets had a mix of both, with the proportions of each depending on the quality of observations. Furthermore, a few datasets, such as those for Argentina and South Africa, record only the town in which a tornado occurred, meaning that the coordinates listed are simply those of the town center.
Overall, each dataset we include employed a different methodology in its creation, resulting in some level of bias. With no single objective procedure for resolving these biases, we present the combined database as-is and recommend careful consideration of individual data collection methodologies in future analysis using it.
b. Grazulis dataset.
Now, focusing specifically on the Grazulis (1993b) dataset, various spatial and temporal trends are present. For reasons mentioned in the previous section, the yearly tornado count increases significantly after around 1875 (Fig. 2). The more general increase throughout the time period is also apparent, as a result of technological improvements and population growth. The westward expansion of population in the United States, in particular, is also evident in the data (Fig. 3). By 1900, there was sufficient settlement in the plains and south to “uncover” broadly the pattern of tornado distribution that is known today.
Annual tornado counts on a logarithmic scale from Grazulis (1993b). Colors indicate the assigned intensity rating on the F scale.
Citation: Bulletin of the American Meteorological Society 105, 7; 10.1175/BAMS-D-23-0123.1
Heat map of tornado occurrences per 10 000 km2 from Grazulis (1993b) from 1750 to 1949, split into 50-yr intervals. The bin size is 1°.
Citation: Bulletin of the American Meteorological Society 105, 7; 10.1175/BAMS-D-23-0123.1
c. Expanded U.S. tornado climatology.
As discussed in section 3a, fictitious increases over time in tornado activity are much more pronounced for subsignificant events due to population and observational effects. Thus, here we compare the Grazulis and OneTor datasets only for tornadoes rated (E)F3 or higher (Fig. 4).
Yearly tornadoes rated (E)F3 or higher in the United States. Pre-1950 from Grazulis (1993b) and post-1950 from Storm Prediction Center (2022), with a dashed line denoting the 1950 mark.
Citation: Bulletin of the American Meteorological Society 105, 7; 10.1175/BAMS-D-23-0123.1
A few features are evident: For one, intense tornado activity decreases during the Dust Bowl in the mid-to-late 1930s. The 1974 and 2011 “Super Outbreaks” are also clearly visible. The aforementioned increase in intense tornado frequencies during the period 1950–70, due to changes in rating methodology, is discernible when comparing both to later years and the Grazulis data.
The digitization of the Grazulis data also allows for a broadened perspective of tornado fatalities in the United States. Figure 5 depicts a general decrease in tornado fatalities as well as in their year-to-year variance over time. The combined effects of the Joplin, Missouri tornado and the Super Outbreak of 2011 cause that year to appear as an extreme outlier compared with the three decades prior. The contribution of the Tri-State tornado, in 1925, is also visible.
Yearly tornado fatalities in the United States, 1880–2021. Pre-1950 from Grazulis (1993b) and post-1950 from Storm Prediction Center (2022), with a dashed line denoting the 1950 mark and a trendline shown with a slope of −1.18 fatalities per year. A similar graph, running up to 2000, is present in Brooks and Doswell (2002).
Citation: Bulletin of the American Meteorological Society 105, 7; 10.1175/BAMS-D-23-0123.1
d. Global tornado fatalities.
In the global distribution of recorded tornado fatalities, Bangladesh (and surrounding portions of India) drastically stand out (Fig. 6). Bangladesh accounts for 8325 of the recorded fatalities in our database, India accounts for 1473, the United States accounts for 21 548, and the rest of the world combined accounts for 3824. As a major caveat, however, fatality or injury information was unavailable for South Africa and for the major Chinese datasets used. Casualty information was also unavailable for our Argentina dataset, although we augmented it for significant events with information from other sources (see section 2a). Otherwise, almost all recorded global tornado fatalities are included, highlighting regions of high tornado activity. Anomalous events, such as the 1987 Edmonton, Canada, tornado and an 80-fatality tornado near Veliky Novgorod, Russia, in 1572, stand out.
Global distribution of total (not annual) recorded tornado fatalities over all time per 10 000 km2, using a logarithmic color scale. The bin size is 2°. Fatality information is not available for South African and most Chinese tornadoes.
Citation: Bulletin of the American Meteorological Society 105, 7; 10.1175/BAMS-D-23-0123.1
e. Global frequency distribution.
Having compiled the major datasets available throughout the world, we attempt here to estimate a global tornado frequency distribution. To do this, we subjectively estimate the time period of best coverage for each dataset. Most countries have records that roughly resemble Fig. 2, where tornado counts gradually increase over time and subsequently begin to plateau. These “plateau” periods were assumed to be those for which observations were most complete and thus were selected for annual frequency analysis. For example, data for Mexico span 2007–20 and yearly tornado count stabilizes at around 2015, so the range from 2015 to 2020 (inclusive) is used to generate a two-dimensional spatial histogram where total tornado counts within each grid cell are divided by six to obtain an annual frequency. This is repeated for each dataset, and the histograms are superposed to create Fig. 7. It is acknowledged that certain datasets will have shorter temporal periods used, resulting in greater uncertainty in the generated annual histograms. At least 6 years of data are used for every dataset, and estimated periods of best coverage for each dataset are listed in the appendix (Table A1).
Estimated global distribution of annual tornado frequency per 10 000 km2. The bin size is 2°.
Citation: Bulletin of the American Meteorological Society 105, 7; 10.1175/BAMS-D-23-0123.1
Seasonal trends in this tornado frequency estimate are depicted in Fig. 8. The northward shift in activity during summer in the United States is evident, in addition to increased activity across the Northern Hemisphere during that season. Interestingly, the northern Mediterranean coast experiences a maximum in tornado frequency during the cold season. And as expected, the Southern Hemisphere is more tornadically active during its spring and summer (SON-DJF).
Estimated global distribution of annual tornado frequency per 10 000 km2, split by season. The bin size is 2°.
Citation: Bulletin of the American Meteorological Society 105, 7; 10.1175/BAMS-D-23-0123.1
f. Online interface.
An online user interface (https://tornadoarchive.com/explorer/) for exploring the compiled data was built. It includes an advanced temporal filter which allows selection of specific days or months of the year across some range of years. Complex filtering by tornado attributes such as intensity, length, width, fatalities, injuries, country, or state is also possible (Fig. 9).
Screenshot of the online interface showing the 21 Mar 1932 tornado outbreak, with 2-m dewpoint, MSLP, and 10-m winds from 20CR, and tornado track information from Grazulis (1993b). At bottom is an hourly tornado histogram. Above that is a domain selection dropdown, seven buttons to toggle visibility on/off for each (E)F-rating, and a filter query box.
Citation: Bulletin of the American Meteorological Society 105, 7; 10.1175/BAMS-D-23-0123.1
The interface allows reanalysis data to be displayed for certain dates with substantial tornado activity. Five displays are possible: 2-m temperature/MSLP/10-m winds, 2-m dewpoint/MSLP/10-m winds, 500-mb wind/height, 850-mb wind/height, and SBCAPE/hodographs (1 mb = 1 hPa). These parameter combinations were chosen because they are most commonly used to diagnose severe storm environments.
The interface is primarily intended as an informative and educational tool for both the public and tornado researchers. The former may use it to diagnose the historical patterns of tornado occurrence in a local region or learn about tornado climatology in other regions worldwide, while the latter may use it to find potential subjects for case studies or recognize correlations among specific tornado attributes that warrant further investigation.
4. Conclusions
This paper describes a global tornado database consisting of information compiled from 19 different sources, which we plan to periodically update as more sources become available. For the first time, it provides a central hub for data that were previously dispersed in a variety of online and print publications.
The database includes a digitized, georeferenced version of all tornadoes through the year 1949 contained in Significant Tornadoes (Grazulis 1993b), which expands the amount of digital data available for historical and climatological research. The database also fully supports detailed path geometries, allowing easier and more interactive access to geospatial data about tornado damage patterns.
The overall database has many potential uses: its global reach allows for worldwide estimates of tornado climatology as well as detailed international studies of tornado history, including intercomparisons of tornado observation and documentation methodologies. Such work can help highlight regions of the world in which modern tornado surveying capabilities are most lacking. Furthermore, this paper itself can serve as a starting point for researchers assembling global tornado datasets in the future.
We hope that the digitized Grazulis dataset will open possibilities for more broad climatological studies (perhaps using the improved sample size to examine correlations between tornadic activity and features such as teleconnections or drought) and research on events prior to 1950. General analyses reveal the effects of westward American expansion on the tornado record, a decrease in tornadic activity during the Dust Bowl, and a pronounced reduction in tornado fatalities in the modern era.
However, any analyses using the database will require careful consideration of biases therein, many of which we have discussed. Indeed, we find that the constituent datasets used contain substantial variation in completeness and precision, especially outside the United States. As such, we emphasize the need for a more standardized international repository of tornado occurrence data and hope that our work provides a first step in that direction.
In addition, we present an online user interface that enables users to explore the database through an interactive map and advanced filtering options. It also allows for the display of reanalysis data for selected dates based on a threshold for the number and intensity of tornadoes that day. As the interface allows efficient filtering and visualization of the database, it is a useful quality control tool. In general, it serves as an informative and educational tool for both the public and tornado researchers, providing insight into tornado occurrence, climatology, and potential subjects for further investigation.
Acknowledgments.
We would like to acknowledge the following contributors for their help in data processing (especially digitization of the Grazulis data): Braxton Banks, Caitlin Beeson, Jonathan Beeson, Jonny Glessner, Anthony Haberman, James Hyde, Luciano Maldonado, Casey Shanaberger, Stephen Szulborski, Raven Vasquez, and Nick Wilkes. We also thank Spencer Dant and Jim Tang for their help in software matters. Finally, we thank J. Poterjoy and three anonymous reviewers with whose comments this manuscript greatly benefited from.
Data availability statement.
All data used in this paper, except for data sourced from the ESWD, are available at https://github.com/malcolmmaas/tornadoarchive-pub. ESWD data are subject to additional terms and conditions and can be obtained upon request: https://www.eswd.eu.
APPENDIX Global Frequency Distribution Supplemental Table
Table A1 lists the start and end years of the periods in each major dataset subjectively determined to have maximum data coverage, as described in section 3e.
Periods of maximum coverage for each major dataset.
References
Agee, E., and S. Childs, 2014: Adjustments in tornado counts, F-scale intensity, and path width for assessing significant tornado destruction. J. Appl. Meteor. Climatol., 53, 1494–1505, https://doi.org/10.1175/JAMC-D-13-0235.1.
Agee, E., and L. Taylor, 2019: Historical analysis of U.S. tornado fatalities (1808–2017): Population, science, and technology. Wea. Climate Soc., 11, 355–368, https://doi.org/10.1175/WCAS-D-18-0078.1.
Aguirre, B. E., R. Saenz, J. Edmiston, N. Yang, D. Stuart, and E. Agramonte, 1994: Population and the detection of weak tornadoes. Int. J. Mass Emerg. Disasters, 12, 261–277, https://doi.org/10.1177/028072709401200301.
Allen, J. T., and E. R. Allen, 2016: The tornado climatology of Australia 1795-2014. Columbia University, 1 pp., https://iri.columbia.edu/∼tippett/SCC2/PostersArchive/2016SCCALLEN_TornadoClimPoster.pdf.
Anderson, C. J., C. K. Wikle, Q. Zhou, and J. A. Royle, 2007: Population influences on tornado reports in the United States. Wea. Forecasting, 22, 571–579, https://doi.org/10.1175/WAF997.1.
Ashley, W. S., 2007: Spatial and temporal analysis of tornado fatalities in the United States: 1880–2005. Wea. Forecasting, 22, 1214–1228, https://doi.org/10.1175/2007WAF2007004.1.
Balbi, M., and P. Barbieri, 2018: Enfoque científico del riesgo—Evaluación del potencial de tornados en la república Argentina. Tech. Rep., 116 pp., https://doi.org/10.13140/RG.2.2.35047.80801.
Barrett, B. S., and V. A. Gensini, 2013: Variability of central United States April–May tornado day likelihood by phase of the Madden-Julian Oscillation. Geophys. Res. Lett., 40, 2790–2795, https://doi.org/10.1002/grl.50522.
Bertoni, B. G. L., 2013: Estudos de caso de tornados, uso e ocupação do solo, topografia e análise meteorológica destes eventos atmosféricos no sul e sudeste do Brasil. M.S. thesis, Dept. of Environmental Science, University of São Paulo, 127 pp., https://doi.org/10.11606/D.90.2013.tde-02052013-160126.
Bikos, D., J. Finch, and J. L. Case, 2016: The environment associated with significant tornadoes in Bangladesh. Atmos. Res., 167, 183–195, https://doi.org/10.1016/j.atmosres.2015.08.002.
Bradford, M., 1999: Historical roots of modern tornado forecasts and warnings. Wea. Forecasting, 14, 484–491, https://doi.org/10.1175/1520-0434(1999)014<0484:HROMTF>2.0.CO;2.
Brooks, H. E., 2004: On the relationship of tornado path length and width to intensity. Wea. Forecasting, 19, 310–319, https://doi.org/10.1175/1520-0434(2004)019<0310:OTROTP>2.0.CO;2.
Brooks, H. E., and C. A. Doswell, 2001: Some aspects of the international climatology of tornadoes by damage classification. Atmos. Res., 56, 191–201, https://doi.org/10.1016/S0169-8095(00)00098-3.
Brooks, H. E., and C. A. Doswell, 2002: Deaths in the 3 May 1999 Oklahoma City tornado from a historical perspective. Wea. Forecasting, 17, 354–361, https://doi.org/10.1175/1520-0434(2002)017<0354:DITMOC>2.0.CO;2.
Brooks, H. E., C. A. Doswell, and M. P. Kay, 2003: Climatological estimates of local daily tornado probability for the United States. Wea. Forecasting, 18, 626–640, https://doi.org/10.1175/1520-0434(2003)018<0626:CEOLDT>2.0.CO;2.
Bureau of Meteorology, 2022: Severe storms archive. Accessed 12 January 2022, http://www.bom.gov.au/australia/stormarchive/.
Cerveny, R. S., and Coauthors, 2017: WMO assessment of weather and climate mortality extremes: Lightning, tropical cyclones, tornadoes, and hail. Wea. Climate Soc., 9, 487–497, https://doi.org/10.1175/WCAS-D-16-0120.1.
Chen, J., and Coauthors, 2018: Tornado climatology of China. Int. J. Climatol., 38, 2478–2489, https://doi.org/10.1002/joc.5369.
Chernokulsky, A., and Coauthors, 2020: Tornadoes in northern Eurasia: From the middle age to the information era. Mon. Wea. Rev., 148, 3081–3110, https://doi.org/10.1175/MWR-D-19-0251.1.
De Coning, E., and B. Adam, 2000: The tornadic thunderstorm events during the 1998-1999 South African summer. Water SA, 26, 361–376.
Doswell, C. A., and D. W. Burgess, 1988: On some issues of United States tornado climatology. Mon. Wea. Rev., 116, 495–501, https://doi.org/10.1175/1520-0493(1988)116<0495:OSIOUS>2.0.CO;2.
Dotzek, N., P. Groenemeijer, B. Feuerstein, and A. M. Holzer, 2009: Overview of ESSL’s severe convective storms research using the European Severe Weather Database ESWD. Atmos. Res., 93, 575–586, https://doi.org/10.1016/j.atmosres.2008.10.020.
Edwards, R., J. G. LaDue, J. T. Ferree, K. Scharfenberg, C. Maier, and W. L. Coulbourne, 2013: Tornado intensity estimation: Past, present, and future. Bull. Amer. Meteor. Soc., 94, 641–653, https://doi.org/10.1175/BAMS-D-11-00006.1.
Edwards, R., H. E. Brooks, and H. Cohn, 2021: Changes in tornado climatology accompanying the enhanced Fujita scale. J. Appl. Meteor. Climatol., 60, 1465–1482, https://doi.org/10.1175/JAMC-D-21-0058.1.
Edwards, R., M. S. Elliott, P. T. Marsh, and D. A. Speheger, 2022: Errors, oddities and artifacts in U.S. tornado data, 1995–2021. Proc. 30th Conf. on Severe Local Storms, Santa Fe, NM, Amer. Meteor. Soc., 8.3B, https://www.spc.noaa.gov/publications/edwards/oddities.pdf.
Elsom, D. M., and G. T. Meaden, 1982: Suppression and dissipation of weak tornadoes in metropolitan areas: A case study of greater London. Mon. Wea. Rev., 110, 745–756, https://doi.org/10.1175/1520-0493(1982)110<0745:SADOWT>2.0.CO;2.
Environment and Climate Change Canada, 2021: Canadian national tornado database: Verified tracks (1980-2009). Accessed 21 January 2022, https://open.canada.ca/data/en/dataset/65658050-7a80-4da3-9a09-da137c203a34.
Fricker, T., T. Wells, J. T. Allemeier, and A. Baxter, 2023: Digitizing and geocoding historical records to improve an understanding of tornado climatology. Prof. Geogr., 75, 750–762, https://doi.org/10.1080/00330124.2023.2169176.
Fujita, T. T., 1973: Tornadoes around the world. Weatherwise, 26, 56–83, https://doi.org/10.1080/00431672.1973.9931633.
Fujita, T. T., 1976: Old and new concept of tornadoes. Dept. of Geophysical Sciences, University of Chicago, 12 pp.
Fujita, T. T., 1987: U.S. tornadoes. Part 1: 70-year statistics. Tech. Rep., 133 pp., https://ntrs.nasa.gov/api/citations/19870016253/downloads/19870016253.pdf.
Galway, J. G., 1985: J.P. Finley: The first severe storms forecaster. Bull. Amer. Meteor. Soc., 66, 1389–1395, https://doi.org/10.1175/1520-0477(1985)066<1389:JFTFSS>2.0.CO;2.
Goliger, A., and R. Milford, 1998: A review of worldwide occurrence of tornadoes. J. Wind Eng. Ind. Aerodyn., 74–76, 111–121, https://doi.org/10.1016/S0167-6105(98)00009-9.
Grazulis, T. P., 1993a: A 110-year perspective of significant tornadoes. The Tornado: Its Structure, Dynamics, Prediction, and Hazards, Geophys. Monogr., Vol. 79, Amer. Geophys. Union, 467–474.
Grazulis, T. P., 1993b: Significant Tornadoes, 1680-1991. Environmental Films, 1326 pp., https://books.google.com/books?id=A8NbPwAACAAJ.
Grieser, J., and P. Haines, 2020: Tornado risk climatology in Europe. Atmosphere, 11, 768, https://doi.org/10.3390/atmos11070768.
Hosen, M. S., and A. Jubayer, 2016: Chronological history and destruction pattern of tornados in Bangladesh. Amer. J. Environ. Prot., 5, 71–81, https://doi.org/10.11648/j.ajep.20160504.11.
Japan Meteorological Agency, 2022: Gust case list. Accessed 21 January 2022, https://www.data.jma.go.jp/obd/stats/data/bosai/tornado/list.html.
León-Cruz, J. F., L. F. Pineda-Martínez, and N. Carbajal, 2022: Tornado climatology and potentially severe convective environments in Mexico. Climate Res., 87, 147–165, https://doi.org/10.3354/cr01692.
Marshall, T. P., T. M. Brown-Giammanco, S. Krautwurst, and N. L. de Toledo, 2022: On the current revision of the enhanced Fujita (EF) scale. 30th Conf. on Severe Local Storms, Santa Fe, NM, Amer. Meteor. Soc., 11.1A, https://ams.confex.com/ams/30SLS/meetingapp.cgi/Paper/407207.
Muller, Z., 2022: The complete history of tornadoes in South Africa. Accessed 12 February 2022, https://afriwx.co.za/resources/history-tornadoes-south-africa/.
National Weather Service, 2022: Damage assessment toolkit. Accessed 27 February 2022, https://apps.dat.noaa.gov/stormdamage/damageviewer/.
NCEI, 2022: Storm Events Database. Accessed 26 March 2022, https://www.ncdc.noaa.gov/stormevents/.
NIWA, 2021: Historic weather events catalogue. Accessed 12 January 2022, https://hwe.niwa.co.nz/.
NOAA, 2018: Post-storm data acquisition. National Weather Service Instruction 10-1604, 12 pp., www.nws.noaa.gov/directives/sym/pd01016004curr.pdf.
NOAA, 2021: Storm data preparation. National Weather Service Instruction 10-1605, 110 pp., https://www.nws.noaa.gov/directives/sym/pd01016005curr.pdf.
Okuda, Y., and Coauthors, 2015: Overview of the development of the Japanese version of the improved Fujita scale. 2015 Japan Society for Wind Engineering Annual Research Presentation, Japan Wind Engineering Society, 117–118, https://doi.org/10.14887/jaweam.2015.0_117.
Schumacher, R., 2022: The Colorado killer tornadoes of November 4, 1922. Authorea, 14 pp., https://doi.org/10.22541/au.166742612.24202428/v3.
Sills, D. M. L., and Coauthors, 2020: The Northern Tornadoes Project: Uncovering Canada’s true tornado climatology. Bull. Amer. Meteor. Soc., 101, E2113–E2132, https://doi.org/10.1175/BAMS-D-20-0012.1.
Storm Prediction Center, 2022: Severe weather database. Accessed 25 January 2022, https://www.spc.noaa.gov/wcm/#data.
Tippett, M. K., J. T. Allen, V. A. Gensini, and H. E. Brooks, 2015: Climate and hazardous convective weather. Curr. Climate Change Rep., 1, 60–73, https://doi.org/10.1007/s40641-015-0006-6.
Vicencio, J., and Coauthors, 2021: The Chilean tornado outbreak of May 2019: Synoptic, mesoscale, and historical contexts. Bull. Amer. Meteor. Soc., 102, E611–E634, https://doi.org/10.1175/BAMS-D-19-0218.1.
Zhang, Y., G. Fang, W. Pang, J. Cao, S. Cao, and Y. Ge, 2022: Tornado database in China mainland from 2003 to 2019. 4TU.ResearchData, accessed 12 July 2022, https://doi.org/10.4121/19801633.v1.
Zhou, R., Z. Meng, and L. Bai, 2020: Tornado database in China (2007–2016). Peking University Open Research Data Platform, accessed 6 February 2022, https://doi.org/10.18170/DVN/QKQHTG.
