Fatalities associated with the weather in the Czech Republic , 2000 – 2019

This paper presents an analysis of fatalities attributable to weather conditions in the Czech Republic during the 2000–2019 period. The database of fatalities deployed contains information extracted from Právo, a leading daily newspaper, and Novinky.cz, its internet equivalent, supplemented by a number of other documentary sources. The analysis is 10 performed for floods, windstorms, convective storms, rain, snow, glaze ice, frost, heat, and fog. For each of them, the associated fatalities are investigated in terms of annual frequencies, trends, annual variation, spatial distribution, cause, type, place, and time, as well as the sex, age, and behaviour of casualties. There were 1164 weather-related fatalities during the 2000–2019 study period, exhibiting a statistically significant falling trend. Those attributable to frost (31 %) predominated, followed by glaze ice, rain and snow. Fatalities were at their maximum in January and December and at their minimum in 15 April and September. Fatalities arising out of vehicle accidents (48 %) predominated in terms of structure, followed by freezing or hypothermia (30 %). Most deaths occurred during the night. Adults (65 %) and males (72 %) accounted for the majority of fatalities, while indirect fatalities were more frequent than direct ones (55 % to 45 %). Hazardous behaviour accounted for 76 %. According to the database of the Czech Statistical Office, deaths caused by exposure to excessive natural cold are markedly predominant among five selected groups of weather-related fatalities and their numbers exhibit a 20 statistically significant rise during 2000–2019. Police yearbooks of the fatalities arising out of vehicle accidents indicate significantly decreasing trends in the frequency of inclement weather patterns associated with fatal accidents, as well as a decrease in their percentage in annual numbers of fatalities. The discussion of results includes the problems of data uncertainty, comparison of different data sources, and the broader context.


Introduction 25
Natural disasters are accompanied not only by extensive material damage, but also by great loss of human life, facts easily derived from data held by re-insurance agencies (e.g. Munich RE, 2018;Swiss Re, 2019;Willis Re, 2019). It is therefore hardly surprising that this situation was also reflected in The Sendai Framework for Disaster Risk Reduction 2015-2030 (SFDRR), adopted at the Third UN World Conference on Disaster Risk Reduction in Sendai (Japan) on 18 March 2015. The Framework suggested targets and priorities intended "to prevent new and reduce existing disaster risks" (UNODRR, 2020;30 https://doi.org/10.5194/nhess-2021-14 Preprint. Discussion started: 21 January 2021 c Author(s) 2021. CC BY 4.0 License.
In the Czech Republic, just as on the international scale, studies of fatalities associated with heat waves are the most frequent 65 (e.g. Kyselý and Kříž, 2008;Kyselý and Plavcová, 2012;Knobová et al., 2014;Hanzlíková et al., 2015;Urban et al., 2017;Arsenović et al., 2019). However, attention has also been devoted to the fatal effects of cold spells. For example, Kyselý et al. (2009) and Plavcová and Urban (2020) analysed the impacts of compound winter extremes upon mortality rates. Brázdová (2012) worked on selected floods in the Czech Republic in order to develop a simple model for estimation of flood-fatalities. Czech flood fatality data also appeared in the EUFF database and were worked upon by Petrucci et al. 70 (2019a). Brázdil et al. (2019b) analysed the potential of documentary data in the study of weather-related fatalities and presented preliminary results for the 1981-2018 period.
The first two decades of the 21st century make up the period that experienced the most profound temperature increases world-wide since records began, inclusive of the Czech Republic (e.g. Zahradníček et al., 2020). The general increase in climatic and weather hazards raises the question as to whether this situation has also been reflected in the number of fatalities 75 associated with weather phenomena. The current paper consists of an investigation and analysis of variability and existing trends in weather-related fatalities over the territory of the Czech Republic in the 2000-2019 period with respect to a selection of influencing factors. The work is based on its own mortality database created from newspapers and in comparison with other official/administrative sources of information. Section 2 describes the basic data sources used for analysis, while the methodology appears in Section 3. The results in Section 4 describe weather-related fatalities, considered as a whole and 80 for individual weather phenomena, as well as addressing those selected from official databases. The results are further discussed with respect to data uncertainty, comparison of different fatality information sources and the broader (central) European context in Section 5. A summary of the most important results appears in Section 6.

Documentary data 85
Newspapers and their websites are the most important source of documentary information for more recent times. They report not only upon political, socio-economic and social matters, but they also reflect considerable public interest in disasters, weather phenomena and associated damage and fatalities. This study gathers information from the print edition of the daily newspaper Právo and Novinky.cz, its internet equivalent. As well as its usual, wider coverage of national and international news, it also appears in several different editions that dedicate space to particular regions of the Czech Republic, thus 90 providing a highly useful source of fatality information. In addition to systematic reading of the newspaper, the research team employed the internet, entering a set of 52 key words (e.g. casualty, died, killed, black ice, flood, windstorm, lightning, frost, heat-wave) and 34 set phrases (e.g. wet road, slippery road, frost casualty, cold casualty, bad weather, bad visibility) to monitor Právo and Novinky.cz for further fatality events.
Individual newspaper reports differ in their style and approach to detail in descriptions of events resulting in fatalities. 95 Although some cases are made immediately obvious within the headlines (e.g. "Lightning kills woman", "Icy road leads to (ii) A woman died on 25 June 2008 during a thunderstorm in Svitavy (Právo, 27 June 2008, p. 1, 5): "According to the 105 police, a woman was hurrying home with a group of friends on the evening of Wednesday [25 June]. As the first drops of rain started to fall, their group of six took shelter under the small roof of a garden restaurant in a town stadium. At that moment, the wind gusted upon an avenue of poplar trees with such force that 10-m branches were torn from it. One such branch struck a 47-year-old woman standing only 10 m from safety. It hit her neck.
[…] The emergency services took her [to hospital] within ten minutes but she succumbed to injuries of the spine." 110 (iii) A woman died in an accident on a snow-covered road in the Bruntál region on 20 April 2017 (Novinky.cz, 2020): "April snowfall in the Bruntál region has claimed its first road victim. Two private cars crashed. A 58-year-old female passenger died and a further three people were injured. The road was covered in snow at the time of the accident. According to the police, a Mitsubishi car got into a series of skids on the snowy road and crossed into the opposite lane, where it collided with a Škoda Fabia. The female victim was a passenger in the Fabia. She was taken to Ostrava hospital, but died shortly 115 thereafter." The basic set of fatality data from Právo and Novinky.cz was further verified, and sometimes supplemented, by reports from further documentary sources, such as other national or local newspapers, professional reports/papers or specialist articles, either published in print or available online.

Data from the Czech Statistical Office 120
Mortality yearbooks for the Czech Republic, arranged according to cause of death, sex and age are published by the Czech Statistical Office (CSO). They contain detailed summary data, specified with respect to various additional facts concerning any given year for the entire Czech Republic as well as its administrative units (CSO, 2020a(CSO, , 2020b. Using the Office codes for cause of death employed on death certificates, the study herein considered: W00 -fall on ice or snow; X30 -exposure to excessive natural heat; X31 -exposure to excessive natural cold; X32 -exposure to solar radiation; X33 -lightning 125 casualty; X36 -avalanche, landslide or other earth movement casualty; X37 -natural catastrophic storm casualty; X38flood (inundation) casualty; X39 -exposure to other and/or non-specified natural forces. Detailed CSO information about each death was gathered, including age, sex, level of education, place of permanent residence, date and code of death, supplemented from 2010 onwards by place of death. https://doi.org/10.5194/nhess-2021-14 Preprint. Discussion started: 21 January 2021 c Author(s) 2021. CC BY 4.0 License. sometimes with ice jams on the rivers (mixed floods) on the one hand, and flash floods arising from cloudbursts or torrential rains during thunderstorms on the other.
(ii) Windstorm: Strong winds resulting from large horizontal gradients of air pressure, lasting from a few hours to some days, are considered windstorms. 165 (iii) Convective storm: This includes phenomena associated with the development of cumulonimbus cloud, such as very strong wind (e.g. squall, tornado, downburst), lightning strike, downpour, and hail.
(iv) Rain: This includes, in particular, rain and wet street communications surfaces/tracks.
(v) Snow: This includes, in particular, cases of snow calamity and avalanche.
(vi) Glaze ice: This includes ice-patches or glaze-ice cover on streets, roads and communications. 170 (vii) Frost: This includes severe frosts occurring as a part of cold spells as a cause of death, as well as fatalities involving bodies of water insufficiently frozen for the activity undertaken on ice.
(viii) Heat: This includes extremes of high temperature occurring in the course of heat waves.
(ix) Fog: This includes cases of significantly decreased horizontal visibility due to fog.
With respect to the limited length of the series of weather-related fatalities (20 years) compiled, and due to deviations from 175 normality in data distribution within certain categories, trend analysis of fatalities was based on two approaches. The first employed a simple regression model based on least-squares estimate and evaluation of the statistical significance of slopes, based on the t-test. The second -nonparametric -approach evaluated the significance of the trend by means of the Mann-Kendall (Kendall, 1975) test and Sen's method for assessment of the magnitude of the trend. In the latter approach, the data need not conform to any particular distribution. These two different methods were considered sufficient to provide a robust 180 estimate of trends.
Fatalities in each group (i)-(ix) were analysed in detail in terms of their annual numbers, with linear trend estimated from least squares (statistical significance was set at the level of 0.05), of annual variations, spatial distribution and structure according to cause of death, type of death, place of death, time of death, age, sex and behaviour (Figs. 1-9). The same was performed for all groups together (Fig. 10) and separately for fatalities in vehicle accidents (Fig. 11). Some of these 185 characteristics were considered in analyses based on the official demographic databases of the CSO (Fig. 12) and/or the police database of vehicle accidents (Fig. 13).
Numbers of fatalities were presented at monthly, seasonal (winter -DJF, spring -MAM, summer -JJA, autumn -SON), half-year (winter -October to March, summer -April to September), and annual levels. in eastern Moravia (Fig. 1c). While 83.0 % of casualties drowned, 10.7 % died due to health problems (e.g. heart failure during rescue work) (Fig. 1d). A total of 82.1 % of fatalities were evaluated as "direct" (Fig. 1e). As might well be expected, 70.5 % of the fatalities took place in running water or close it; 16.1 % died in collapsing buildings (Fig. 1f). Despite the fact that the times of death were not specified for over half the fatalities, a local maximum appeared in the evening (Fig. 1g). In the demographic structure of fatalities, adults (58.0 %) and males (73.2 %) clearly predominated (Fig. 1hi). Non-hazardous 205 behaviour among flood fatalities was more prevalent than hazardous (43.8 % to 38.4 %) (Fig. 1j).   but statistically insignificant trend was evident. In terms of annual distribution, nearly half the fatalities (49.8 %) occurred in the summer months, with the highest proportion in July (21.5 %). Proportions in the months of the other three seasons did not rise above 10 % (Fig. 4b). Rain-related fatalities were distributed over the whole Czech Republic, with a higher concentration in some of the smaller regions and lower frequency near borders, for example, north-western, south-western and southern Bohemia and south-western Moravia (Fig. 4c). All these fatalities were classified as indirect consequences of 260 vehicle accidents. They tended to occur in the afternoon (26.3 %) (Fig. 4d). In terms of structure, adults made up 68.3 % of fatalities (age not reported for 20 %) and males 65.9 % (Fig. 4ef). A total of 97.6 % the fatalities were classified as arising out of hazardous behaviour on the part of the victims, or of the driver(s) responsible for the accident.  A falling linear trend reached 4.5 fatalities/10 years and was statistically significant. As is to be expected, snow-related fatalities occurred only in the months of the winter half-year (with the exception of a single fatality in April), with maxima in January (34.3 %) and February (29.2 %); the percentages for November and December were identical, at 12.4 % (Fig. 5b). In https://doi.org/10.5194/nhess-2021-14 Preprint. Discussion started: 21 January 2021 c Author(s) 2021. CC BY 4.0 License. terms of spatial distribution, snow-related fatalities tended to concentrate into a large number of smaller areas, while in some 275 larger regions no localities with casualties were recorded at all (Fig. 5c). Vehicle accidents were involved in a total of 84.7 % of the fatalities (Fig. 5d), thus 94.9 % of them were classified as "indirect" (Fig. 5e). The percentage of people dying on the roads achieved 81.0 % (Fig. 5f); more than 5 % occurred in the built-up areas and in open countryside (avalanche casualties).
While the time of day at which death took place remained unknown for 38.0 % of fatalities, the morning and afternoon proportions were 18.2 % and 17.5 % respectively (Fig. 5g). Despite the relatively high percentage of fatalities for which age 280 was unspecified (29.2 %), 51.1 % were categorised as "adult" (  glaze-ice fatalities occurred between October and April with a maximum in December (33.8 %), followed by January with 27.0 % (Fig. 6b). They occurred over the whole territory with higher concentrations in some areas, lower or none in others ( Fig. 6c). Vehicle accidents were involved in a total of 95.0 % of glaze-ice fatalities (Fig. 6d), i.e. indirect casualties. The place of death was a road for 87.8 % of fatalities, and 9.0 % took place on communications in built-up areas (Fig. 6e).
Although exact time of death remained unknown for nearly a fifth of the fatalities, 36.5 % occurred during the morning (   (Fig. 7a). However, the relevant falling linear trend of 5.8 fatalities/10 305 years was statistically insignificant. The highest percentage of frost-related fatalities appeared in January (34.7 %) and December (28.6 %). Beyond the months of the winter half-year, only one fatality was recorded, in April (0.6 %) (Fig. 7b).
Since the majority of deaths occurred among the homeless, fatalities were partly concentrated into large cities/towns such as Prague (79), Ostrava (24) and Brno (15) (Fig. 7c). While most succumbed to freezing or hypothermia (96.4 %), the remainder concerned cases in which the ice on ponds, reservoirs or rivers was insufficient for the activity undertaken and 310 people fell through it and drowned (Fig. 7d). A similar proportion (96.1 %) of fatalities were characterised as "direct" (Fig.   7e). Open spaces in built-up areas accounted for 48.3 % in place of death, followed by open countryside (18.6 %) (Fig. 7f).

Heat
Only 20 fatalities in the Czech Republic were attributed to heat (heat waves) during the 2000-2019 period, an average of one fatality a year. Nine such cases were recorded in 2006, while in seven of the years only one fatality occurred, and none at all in 10 of the years. This is reflected in a statistically insignificant falling linear trend (Fig. 8a). Fatalities appeared only from May to August, with a maximum in June at 45.0 % (Fig. 8b). The spatial distribution of such a low number of fatalities 325 reveals no features worthy of mention (Fig. 8c). Heart failure appears as the main cause of death, classified as "direct". A total of 40.0 % of fatalities occurred in built-up areas, but place of death was not specified in records for 35.0 % (Fig. 8d).
The time of day was not entered for 70.0 % of the fatalities (Fig. 8e). Despite the fact that deaths in the "adult" category remained the highest (45.0 %), the percentage of the elderly was significantly high (25.0 %), the same figure as the  While five fatalities occurred in 2014, no such event was recorded for nine of the years (Fig. 9a). The rising linear trend was statistically insignificant. While 77.8 % of the fatalities occurred in SON, with a maximum in September (33.3 %), the remainder were recorded in the winter months (Fig. 9b). The geographical distribution for such a small number of cases is of 340 a somewhat random character (Fig. 9c). Decreased visibility was evident in the 88.9 % of fatalities attributable to vehicle accidents; two fatalities (11.1 %) occurred in an aeroplane crash (Fig. 9d). These indirect casualties took place on roads (83.3 %), also partly in the countryside (aeroplane crash) and in built-up areas (Fig. 9e). Half the fatalities occurred during the morning (Fig. 9f). The predominance of adults (83.3 %), males (66.7 %) and hazardous behaviour (94.4 %) characterised other features of fog-related fatalities (Fig. 9ghi).   10a). A falling trend of 25.0 fatalities/10 years proved statistically significant at the significance level of 0.05. Almost a third of the fatalities (30.9 %) were taken up by frost-related cases, followed by glaze ice (19.1 %), rain (17.6 %) and snow (11.8 %). While floods were responsible for 9.6 % of fatalities, other weather factors stood at well below 5 %. These 355 proportions influence the annual distribution of fatalities, bringing the maxima to January (21.7 %) and December (18.0 %), while minimum fatalities was recorded in April (1.9 %) and September (2.7 %) (Fig. 10b)  Vehicle accidents are the most frequent causes of death (48.4 %), followed by those associated with cold spells leading to 370 freezing or hypothermia (29.8 %); drowning takes up 10.1 % of fatalities (Fig. 10d). Weather-related indirect fatalities are more frequent than those classified as direct (54.8 % to 45.2 % respectively) (Fig. 10e). The majority of fatalities occur on roads and communications (47.4 %), followed by open spaces in built-up areas (21.9 %). The areas around running water and reservoirs or in open countryside take their toll (10.0 % and 9.1 % respectively) (Fig. 10f). Available data indicates that most deaths occur by night, but 39.7 % of fatalities were not attributed to any particular part of the day (Fig. 10g). In the 375 structure of fatalities, the numbers of adults (64.6 %) and males (71.9 %) clearly predominate (Fig. 10hi). Over threequarters of fatalities (75.9 %) may be attributed to hazardous behaviour on the part of actual casualties or that of other people immediately responsible for their deaths (Fig. 10j).
Of the total of 1164 weather-related fatalities, 66 (5.7 %) were identified as non-Czechs, in the country on business, holiday or only in transit. Among these people, those from neighbouring Slovakia (17 fatalities (Fig. 11a). Among individual events, falling linear trends were statistically significant for glaze ice and snow. Annual variation exhibits two maxima, primarily in the winter 390 months, arising out of glaze ice and snow (January 19.2 %) and secondarily in the summer months in response to rain (July 7.8 %) (Fig. 11b). In the spatial distribution of fatalities, concentrations around main roads/highways, or certain parts of them, together with the main routes out of the country, are apparent (Fig. 11c). Using the vehicle accident casualties classified within "indirect deaths" and "hazardous behaviour" (96.8 %), 94 % of them died on roads and the remaining 6 % in built-up areas and the countryside. Although the times day at which death occurred are absent for 21.1 % of fatalities, they 395 were for the greater part recorded in the morning (24.7 %) (Fig. 11d). Adults made up 63.9 % of fatalities and numbers of deaths among the elderly were slightly higher than those for children (6.3 % and 5.3 % respectively) (Fig. 11e). There were more than twice as many male fatalities as female (64 % and 29 % respectively) (Fig. 11f).

Vehicle accidents
As reported in Section 2.3, the police yearbooks recording the accident rate in the Czech Republic facilitate the creation of 445 series of fatalities associated with vehicle accidents in relation to weather conditions. Such conditions are classified by the police as: fog, onset of rain and light rain, rain, snowfall, rime and glaze ice, gusty winds, and other inclement weather

Data uncertainty
The newly-created database of weather-related fatalities created for the purposes of this study and based on documentary 465 data suffers from certain data uncertainties. These have been previously mentioned in, for example, contributions addressing the use of documentary data in historical climatology (Brázdil et al., 2005(Brázdil et al., , 2010 and in historical hydrology (Brázdil et al., 2006). Newspaper reports have served as a vital source in the creation of databases of weather-related fatalities, to the point at which the approach has become quite common practice. They have been used, for example, for Switzerland (Hilker et al., https://doi.org/10.5194/nhess-2021-14 Preprint. Discussion started: 21 January 2021 c Author(s) 2021. CC BY 4.0 License. (Zêzere et al., 2014), southern France (Vinet et al., 2016), Calabria in southern Italy (Aceto et al., 2017;470 Petrucci et al., 2018) and Mallorca (Grimalt-Gelabert et al., 2020). The results of working with Czech newspaper data over a relatively extended 20-year period may be influenced by the profound changes in society, both in the media market and in internal changes in the actual newspaper employed. The publisher and editor-in-chief of any given newspaper decide strategy. They are subject to a wide and complex range of influences: how much space will be devoted to certain kinds of information, largely on the basis of the situation on the media market and perceived interest of target readers, and sometimes 475 the political orientation of the news source. Further, in the background lie personnel changes, reduction of regional editorial staff (largely in the light of digitization), different quantities of space given to regional and countrywide reporting, advertising space (influencing both space for other reports and arising out of reluctance to report matters that might offend advertisers), competition in reporting, and availability of regional/local news from other bodies such as the police or the Czech Press Agency. Reader fatigue is also important; certain kinds of fatal event became ever-more-familiar and reader 480 interest wanes. Moreover, the real number of fatalities may also be underestimated, particularly in situations involving the severely injured being taken to hospital (e.g. after a vehicle accident, falling trees/branches, hypothermia etc.). Only seldom, and if the follow-up is deemed in some-way "remarkable", additional information is later to be found, i.e. if injured people really died. All the above circumstances may be reflected in spatial and temporal non-homogeneity of fatality data derived from documentary evidence. It should therefore be borne in mind that the database created for the purpose of this research 485 tends to represent a somewhat lower estimate of weather-related fatalities.

2009), Portugal
All the above serves to highlight the vital role played by critical evaluation in the use of documentary sources, especially with reference to fatality data. Inclusion in the database herein gave preference to reports containing more detailed information concerning a given fatality, particularly those that provided name (sex), age, place and the specific cause of death. The team has remained aware of the drawbacks of employing only information summarising the total number of 490 fatalities during any given event or period. This is also a tendency typical of the reporting of disastrous natural events, in which descriptions of material damage often take precedence over more personal matters, such as detailed descriptions of place, time and cause of fatalities. Reporting fatalities without the necessary details may result either in underestimation of real numbers on the one hand, or even exaggeration of them on the other.
Other types of bias may also appear in official databases, such as those of the CSO. Determination of cause of death on a 495 death certificate is based on some degree of subjective perception on the part of the doctor filling it out. Even the most experienced health workers are forced to select from a broad scale of "official" definitions, in which certain categories may be understood differently by individual doctors (e.g. excessive natural heat, excessive solar radiation, and non-specified natural forces). While the database herein includes all weather-related fatalities that occurred in the Czech Republic, CSO collects only data concerning Czech citizens, excluding the deaths of non-Czechs on Czech territory and including deaths of 500 Czechs that take place beyond the borders. The integration and cross-checking of data between our database and that of the CSO is complicated by the fact that information about place of death has only appeared in the latter since 2010. https://doi.org/10.5194/nhess-2021-14 Preprint. Discussion started: 21 January 2021 c Author(s) 2021. CC BY 4.0 License.

Weather-related fatalities in different databases
Weather-related fatalities in our database may be discussed in relation to those of CSO and police vehicle accident reports, although they are not fully comparable. If frost-related fatalities (Fig. 7) are considered against "fatalities due to excessive 505 natural cold" according to the CSO (Fig. 11, X31), the CSO fatality figure is seven times higher, with a statistically significant rising linear trend (in contrast to the falling and insignificant trend in our database). Both series agree upon a maximum of fatalities in 2010 and during January and December in annual variation. While the CSO database gives fatalities in every month of the year, no casualty was identified from May to September in our database. Both databases show that the highest percentage of fatalities occurs among males and adults, but they differ more widely in percentages of 510 the elderly (32.5 % for CSO and only 10.3 % in our database, but with 21.4 % of fatalities of unknown age in the latter).
The above two fatality datasets may be compared in terms of selected characteristics of DJF severity in the Czech Republic, as calculated from 268 homogenised temperature series. These series included mean DJF temperature, mean minimum temperature Tmin, numbers of frosty days with Tmin ≤ −0.1 °C, and numbers of days with Tmin ≤ −5.0 °C and Tmin ≤ −10.0 °C).
Our database exhibits a closer relationship between fatalities and temperature characteristics than that of CSO. Statistically 515 significant Pearson correlation coefficients lie between 0.63 (number of frosty days) and 0.87 (number of severe frost days with Tmin ≤ −10.0 °C); the situation is opposite for the CSO database -between 0.53 (number of severe frost days with Tmin ≤ −10.0 °C) and 0.61 (number of frosty days). This reflects the higher degree of press attention paid to fatalities, particularly among the homeless, during severe cold spells, than is the case of less extreme temperatures over the whole DJF period.  Heat-related fatalities are deeply underestimated in both our database and that of the CSO (X30 category) (20 and 30 545 fatalities respectively). The two independent series return statistically insignificant linear trends. The study herein did not give particular attention to heat-wave-related mortality, since there exist a plethora of such Czech analyses (among them Kyselý and Kříž, 2008;Kyselý and Plavcová, 2012;Knobová et al., 2014;Hanzlíková et al., 2015;Urban et al., 2017;Arsenović et al., 2019). Different data and other approaches to analysis have been utilized. Taking  However, the linear trends for the series in both databases exhibit statistically significant decreases. Although the categories 555 of weather events are not the same, the differences in their corresponding percentages are very large. Compared with official police data, our database overestimates percentages heavily for glaze ice (37.5 % to 8.4 %) and snow (20.6 % to 10.4 %) and underestimates for rain (36.2 % to 35.7 % plus 25.2 % for "onset of rain and light rain" police category), fog (2.8 % to 11.1 %) and other inclement weather (2.8 % to 9.2 %). The above discrepancies in our data compared with police data may be explained by the nature of public information concerning vehicle accidents. The greater part of vehicle accidents reported 560 in the press only seldom provide details of ambient weather patterns, perhaps in the event of snow or glaze-ice calamities, or when major highways are closed by weather (and consequent accidents). From reports of the type that "driver did not adapt speed to road conditions" or "the reason for the accident is subject to further investigation", it is impossible to derive weather information. Similar interpretation difficulties exist for vehicle accidents described as "skid on slippery road", which we interpreted as glaze ice during the winter months. Of course, not every vehicle accident with casualties was reported in the 565 press, although the total numbers of fatalities generated by an extreme event, or at any particular weekend, were often mentioned without necessary details.
Despite all these uncertainties, our database contrasts with other, official, databases in providing more detailed information about the circumstances surrounding fatalities and permits a more complete overview of the causes and consequences of fatal events. 570

The broader context
Beyond the work herein, in which our figures for weather-related fatalities are considered against the databases kept by the In other European countries, Badoux et al. (2016), addressing the 1946-2015 period for Switzerland, found an average of 14.7 fatalities a year. Terrain that includes the Alps inevitably dictated that the highest percentage of fatalities was attributable to snow avalanches (37 %). This was followed by lightning strikes (16 %), floods (12 %), windstorms (10 %), rockfalls (8 %), landslides (7 %) and "other processes" (9 %). Antonescu and Cărbunaru (2018)  Republic -exhibited quite stable linear trends. In more detail, the structure of their total of 2466 flood fatalities detected features similar to those disclosed in our analysis (see Fig. 1): a prevalence of male fatalities aged 30-49 years, the majority of deaths outdoors, drowning as a primary cause of death, followed by indirect deaths arising out of heart failure. Casualties were most frequent in vehicles carried away by water or mud. Paprotny et al. (2018), in an analysis of floods that did damage 600 in 37 European countries from 1870 onwards, found a substantial decrease in flood fatalities, despite increases in annually inundated areas and the numbers of people affected. Sharma et al. (2020) investigated over 4000 winter drowning events resulting from falling through ice in the course of a range of activities, covering 10 Northern Hemisphere countries. Children and adults aged up to 39 years were at the highest risk. They maintained that the potential for this type of accident was rising with warmer winters. Our database documents 605 only 14 such fatalities during 2000-2019, occurring especially when crossing or skating on insufficiently frozen water bodies of natural or anthropogenic origin or as a result of hazardous behaviour on the part of children (29 %), adults (36 %) or the elderly (21 %). Age was not specified for two of the casualties. https://doi.org/10.5194/nhess-2021-14 Preprint. Discussion started: 21 January 2021 c Author(s) 2021. CC BY 4.0 License.
Summarising the results of trend analysis for weather-related fatalities in the Czech Republic during the 2000-2019 period, statistically significant falling linear trends were revealed by both methods of linear trend calculation for convective storms, 610 glaze ice, snow and all weather-related fatalities, as well as for windstorms according to the Mann-Kendall test. The remaining groups of events returned statistically insignificant trends. In the CSO database, statistically significant rising fatality trends emerged for categories X31 (excessive natural cold) and X30+X32 (excessive natural heat and solar radiation) according to both methods applied. The trend lines appearing in Figs. 1-13 show lower values of slope for Sen's estimate, since this approach is less sensitive to outliers, providing reliable results even in situations where one extremely high value 615 may influence the slope of the trend line in the classical least-square regression method (Gilbert, 1987). In this sense, our estimate of significant trends provides quite consistent results. These trends may be compared with those in other contributions. For example, Analitis et al. (2008), addressing 15 European cities in 1990-2000, maintain that a 1 °C decrease in October-March temperatures contributed to a 1.35 % increase in the daily number of total natural deaths, with comparable or higher increases in cerebrovascular (1.25 %), cardiovascular (1.72 %) and respiratory (3.30 %) deaths. Plavcová and 620 Urban (2020), using mortality data for the Czech Republic for 1982-2017, recorded that sudden rises in minimum temperature and drops in pressure had a generally significant impact on excess mortality, by 3.7 % and 1.4 % respectively.
This impact was significantly exacerbated if the two events occurred simultaneously or when they were compounded by other extremes (heavy precipitation, snowfall, maximum temperature rise) and combinations thereof (14.4 %). Holle (2016) reported a large reduction in lightning fatality rates for western Europe and some other regions during recent years, 625 associating it with changes in society from the largely rural, agricultural to the primarily urban. Franzke and Torelló i Sentelles (2020), collating figures worldwide, found statistically significant increasing trends for heat-wave-and floodrelated fatalities.
Statistically significant falling trends also appear in the numbers of Czech fatalities associated with vehicle accidents taking place in bad weather conditions such as rain, snow, glaze ice, fog, etc. if conclusions are based on the authors' database and 630 that of the police in 2000-2019. Because the majority of these cases take place during the winter half-year, in which warmer winters with "better" weather for traffic may play some role, the influence of other factors may be higher. These could include such matters as safer cars, better roads and their closer maintenance, weather forecasting, general media warnings before particularly adverse weather, raising of public awareness of road safety, and improved emergency services and health systems, among other things. Trends in the Czech Republic are partly, for example, in agreement with Andrey (2010), who 635 disclosed a downward trend in the relative risk of casualty during rainfall and no significant change during snowfall in an analysis of 1984-2002 data concerning weather-related crash risk in automobile transport in 10 Canadian cities.
The question remains open as to the extent to which variability and trends in weather-related fatalities may be attributed to specific factors in the light of the data uncertainty discussed in Section 5.1. Climate variability in recent decades, as documented by many climatological papers (see e.g. Brázdil et al., 2017Brázdil et al., , 2019aBrázdil et al., , 2020Zahradníček et al., 2020), and socio-640 economic factors, as mentioned for weather-related car accidents in the previous paragraph, all have parts to play. Each individual fatality is a synergy of different, largely random circumstances, including hazardous behaviour in many cases. https://doi.org/10.5194/nhess-2021-14 Preprint. Discussion started: 21 January 2021 c Author(s) 2021. CC BY 4.0 License.
Although Franzke and Torelló i Sentelles (2020) found that socio-economic factors had no significant direct impact on statistically significant increasing trends in heat-wave-and flood-related fatalities worldwide, and argued for the significant influence of climate variability on the numbers of fatalities, the authors maintain that, at regional or national scales, socio-645 economic factors and hazardous behaviour may be of high importance. Of course, this in no way implies that the real and pressing effects of climate variability, as shown through the example of Fig. 14, should be overlooked.

Conclusion
The following conclusions may be drawn from this analysis of weather-related fatalities in the territory of the Czech (iii) The structure of weather-related fatalities indicates the highest percentages for males, adults, indirect types of death, vehicle accidents due to inclement weather conditions, and freezing or hypothermia, night or morning times of deaths and 660 hazardous behaviour on the part of casualties or other persons responsible for the fatal incident.
(iv) Fatalities derived from the database of Czech Statistical Office, ordered according to the type of weather-related death, exhibit statistically significant rising trends in the "excessive natural cold" and in the combined "excessive natural heat and solar radiation" categories. In terms of sex and age structures, males and adults predominate.
(v) Vehicle-accident fatalities during bad weather conditions, as extracted from police yearbooks, show statistically 665 significant falling trends for all and individual weather events. Rainy weather, with c. 61 %, predominates in the latter category, followed by fog, snowfall and glaze ice.
(vi) Trends in weather-related fatalities in the Czech Republic reflect, in addition to recent weather and climate changes, certain socio-economic factors and especially people's behaviour, which may be for c. 75 % characterised as hazardous.