Flood Fatalities in Europe, 1980–2018: Variability, Features, and Lessons to Learn

Floods are still a significant threat to people, despite of the considerable developments in forecasting, management, defensive, and rescue works. In the near future, climate and societal changes as both urbanization of flood prone areas and individual dangerous behaviors could increase flood fatalities. This paper analyzes flood mortality in eight countries using a 39-year database (1980–2018) named EUFF (EUropean Flood Fatalities), which was built using documentary sources. The narratives of fatalities were investigated and standardized in the database reporting the details of the events. The entire dataset shows a stable trend on flood fatalities, despite the existence of individual increasing (Greece, Italy, and South France) and decreasing (Turkey and Catalonia) trends. The 2466 fatalities were mainly males, aged between 30–49 years and the majority of them happened outdoor. Most often people were dragged by water/mud when travelling by motor vehicles. Some cases of hazardous behaviors, such as fording rivers, were also detected. The primary cause of death was drowning, followed by heart attack. This work contributes to understand the human–flood interaction that caused fatalities. The changes in society’s vulnerability highlighted throughout this study contribute to manage future risks, to improve people protection actions, and to reduce risk behaviors.


Materials and Methods
The present research is the second phase of a project that started in 2017 aiming to create MEFF (MEditerranean Flood Fatalities) database, including flood fatalities that occurred on a 36-year period  in five study areas located in the Mediterranean area [18]. Regions analyzed in MEFF were the following: (1) Calabria (Italy); (2) Languedoc-Roussillon, and Provence-Alpes-Cote d'Azur (France); (3) Catalonia (Spain); (4) Balearic Islands (Spain); and (5) Greece.
The focus was on flood fatalities, defined as people who lost their life due to floods. The methodological approach was based on the systematic collection of fatal events descriptions from documentary sources and the disaggregation and systematization of all the available information in the fields of MEFF. Data analysis, on one hand supplied a series of results [18], and on the other hand, suggested further clues to be investigated. Thus, to enlarge the database, we searched for new study areas where inventories of flood damage were available. We identified four new countries (Table 1): (1) Czech Republic, (2) Israel, (3) Portugal, and (4) Turkey, that provided data on flood fatalities. We sent to each one of the new partners an empty template of MEFF to fill FF for their study area in the 1980-2018 period. Simultaneously, the original study areas in MEFF were extended by fatalities to 2018 and the Calabria was substituted by entire Italy. By these activities, the new EUFF (EUropean Flood Fatalities) database was created (see Section 2.2).

Study Areas and Information Sources
Data collection and analysis of FF was carried out for nine Study Areas (SA) located in eight countries. The entire area further analyzed is named total area (TOT-A) (  Table 2 shows the area of each SA and their general demographic data. Turkey is the largest (52.2% of TOT-A surface) and most populated (41.4% of the TOT-A population). Average population density is 182 inh/km 2 : the largest value pertains to Israel (378.1 inh/km 2 ) and the lowest to Greece (81.6 inh/km 2 ). The average age of population is around 41 years: the highest value pertains to Italy (48 years) and the lowest to Israel (31 years). On average, 48.9% of the population are males, and 51.1% females: the highest percentage of female pertains to Portugal (52.7%), while the lowest value (50.3%) pertains to both Israel and Balearic Islands.  [28] and by Vinet and Boissier [29].

8) Catalonia (CAT)
Catalonia (Spain) is a region in the northeast Iberian Peninsula. The most significant topographic features are the Pyrenees (over 2500 m a.s.l.), the Littoral range and the Pre-Littoral system, rising to higher than 500 m and 1200 m a.s.l., respectively. There are two wet seasons (autumn and spring) and two dry seasons (winter and summer). The mean annual precipitation can vary from 400 mm, in Central Depression, to 1200 mm, in the Pyrenees. CAT represents 2.1% of the area and 3.8% of population of TOT-A, with the population density equal to 235 inhabitants per km 2 , 50.9% of females. However, the population is mainly concentrated along the coast. For instance, Barcelona (with a density of 15,866.95 inh/km²) and two surrounding municipalities, concentrate 46% of the population of Catalonia. The mean age of population is 42 years. Data comes from the INUNGAMA database [30] that contains all the flood events that have produced socioeconomic impact between 1981 and nowadays. This database contains information such as the date, the counties and municipalities affected, the main rivers or basins involved, the impacts produced, and the type of flood. Information regarding victims has been complemented with newspaper data, mainly La Vanguardia, and official reports.

9) Balearic Islands (BAL)
The Balearic Islands archipelago (Spain) is situated off the eastern coast of Spain. It consists from five islands (Mallorca, Menorca, Eivissa, Formentera, and Cabrera) with adjacent islets. Precipitation expresses a clear Mediterranean pattern, with a maximum during autumn and a minimum in summer. Mean annual totals range from 1000 mm, in Mallorca, to 300 mm, in the southern islands of Eivissa and Formentera. BAL, with 4492 km 2 , accounts only to 0.3% of the area and to 0.6% of population for TOT-A. However, the population density achieves 258 inhabitants per km 2 , 50.3% of which are females. Data were obtained from a PhD thesis [31], and complemented by research in regional newspapers, such as Diario de Mallorca and Ultima Hora, and by data gathered for the implementation of the Flood Prevention Plan.   (1) Czech Republic (CZE)

EUFF Database
The Czech Republic (until 31 December 1992 the western part of Czechoslovakia) is located in Central Europe. It has an indented morphology represented by lowlands, highlands and mountains with altitudes between the lowest point at the north-west in Hřensko (115 m a.s.l.), and the highest point in Sněžka Mount (1603 m a.s.l.). The annual precipitation has a maximum in summer and a minimum in winter, and totals fluctuate between 400 mm and 1450 mm. CZE represents 5.3% of the surface, and 5.3% of population of TOT-A, with a population density of 133.8 inhabitants per km 2 . The average age of population is 43 years and 50.9% of the population is made of females. Data of flood victims comes from historical-climatological database of the Institute of Geography, Faculty of Science, Masaryk University in Brno, collected from different documentary evidence. Newspaper information was dominant in the 1980-2018 period, partly complemented by professional papers describing outstanding events and notes of observers at meteorological stations of the national network.

(2) Israel (ISR)
In this SA, four physiographic regions can be distinguished: (i) The coastal plain, with elevation from 10-20 m to about 100 m a.s.l., that extends from the Mediterranean Sea to the foothills. (ii) The central mountain belt, including the Galilee, Samaria and Judea mountains, with elevations between 500 m and 1000 m a.s.l. (iii) The Rift Valley area, a linear depression trending north-south. The Southern Negev desert covers almost half of the country, bordering in the south to the Red Sea. Climate varies from arid to semi-arid and humid. ISR represents 1.5% of the area and 4.2% of population of TOT-A, with a population density of 378.1 inhabitants per km 2 . The mean age of population is equal to 31 years. Females represent 50.3% of the population. Moshe Inbar conducts a database for natural hazards in Israel at the Department of Geography and Environmental studies, in the University of Haifa, for the period between 1948 and present days. The natural hazards include earthquakes, floods, landslides, droughts, and forest fires. Data source for human victims are newspapers and radio news. For the present work, only flood fatalities were extracted from the mentioned database.

(3) Italy (ITA)
The territory of Italy consists of a peninsula and 2 main islands located in the middle of the Mediterranean basin. More than 3 4 of the territory is formed by mountains and hills. The highest  [19][20][21].

(4) Turkey (TUR)
Turkey has a very complex topography, with mostly W-E oriented mountains. Mountains block the moist air flow towards inlands, resulting in a dry climate in the interior, and moist and mild climate in the south, west, and north of the country. Eastern part has relatively higher altitudes (up to 5137 m), with severe winters, while the southeast area has more semi-arid climate characteristics. Average annual precipitation is 574 mm; internal regions have only 250 mm, while the figure in northeast coast exceeds 2000 mm. With 783,562 km 2 , TUR covers 52.2% of TOT-A, and accounts for 41.4% of the population of TOT-A. However, the population density is only 104.7 inhabitants per km 2 . The average age of population is 32 years and females include 50.8% of the population. The flood fatalities data comes from the Turkish Severe Weather Database, which is built (and continuously updated) using official hazardous weather records, newspaper archives, voluntary reports, and other sources. The database includes tornadoes, severe hail, damaging winds, floods, lightning fatalities, and injuries. Further details regarding the database are discussed in several papers [22][23][24]. For the present work, only flood fatalities were extracted from the mentioned database.

(5) Greece (GRE)
Greece is mostly a mountainous country (about 80% of the territory) with the highest Mount Olympus (2917 m a.s.l.). It also has a complex land-water distribution with numerous islands forming a coastline in the length of 13,676 km. It is characterized by a Mediterranean climate. Mean annual totals up to 400-600 mm are observed over the eastern part of continental Greece, while the islands of the Aegean Sea are much drier. GRE represents 8.8% of the area and 5.4% of population of TOT-A, with the population density achieving only 81.6 inhabitants per km 2 . The mean age of population is equal to 45 years and females create 50.8% of the population. Data were obtained by the database of the National Observatory of Athens on high-impact weather events in Greece [25], enriched with details on victims gathered by the newspapers Rizospastis and Ethnos, reliable media websites, and the community of amateur meteorologists.

(6) Portugal (POR)
Portugal is located in the southwest of the Iberia Peninsula. The elevation ranges from 0 m a.s.l., near the coast, to 1993 m a.s.l. in the Central Mountain range. The climate is controlled by the transition between the Mediterranean and the Atlantic conditions. The mean annual precipitation is around 900 mm, ranging from less than 500 mm in northeast and south, to more than 2000 mm in the northwest mountains. Rainfall amounts tend to increase with increasing latitude, elevation and proximity to the Atlantic Ocean. POR accounts for 6.1% of the area and 5.2% of population of TOT-A, with a population density of 111.2 inhabitants per km 2 . The mean age of population is equals to 46 years and the percentage of female population is 52.7%. The Disaster Database [26], based on a systematical collection of floods and landslides that have caused human damages in Portugal referred to in newspapers, is the main data source. Details about flood victims were further complemented with media websites and published papers about flood fatalities [3,27].

(7) South France (SFR)
The SFR includes the former region Languedoc-Roussillon (now part of Occitan region) and Provence-Alpes-Cote d'Azur regions. Ponds and deltas are typical of the lowland in the western part (Languedoc). The relief is steeper and valleys deeper in the eastern part (Provence). The coastal plains are surrounded by 2500 m high summits in the Pyrenees and the Alps, and 1500 m in the Cévennes. The summer drought and the intense rainfall in autumn are the key features of the climate. The rainfall concentrates over the period September-December (50% of annual total). Winter is drier with cold continental winds. SFR represents 3.6% of the area and 3.6% of population of TOT-A, with a population density ranging from 78.1 inh/km 2 (Languedoc-Roussillon, Midi-Pyrénées) to 156 inh/km 2 (Provence-Alpes-Cote d'Azur) with an average of 117 inhabitants per km 2 . The mean age of population is equal to 42 years. Females represent 51.9% of the population. Data were collected by the department of geography of the University Paul Valéry Montpellier 3 (UMR GRED laboratory), starting from documentary sources and newspapers and complemented through post flood surveys published by the PhD of L. Boissier [28] and by Vinet and Boissier [29].

(8) Catalonia (CAT)
Catalonia (Spain) is a region in the northeast Iberian Peninsula. The most significant topographic features are the Pyrenees (over 2500 m a.s.l.), the Littoral range and the Pre-Littoral system, rising to higher than 500 m and 1200 m a.s.l., respectively. There are two wet seasons (autumn and spring) and two dry seasons (winter and summer). The mean annual precipitation can vary from 400 mm, in Central Depression, to 1200 mm, in the Pyrenees. CAT represents 2.1% of the area and 3.8% of population of TOT-A, with the population density equal to 235 inhabitants per km 2 , 50.9% of females. However, the population is mainly concentrated along the coast. For instance, Barcelona (with a density of 15,866.95 inh/km 2 ) and two surrounding municipalities, concentrate 46% of the population of Catalonia. The mean age of population is 42 years. Data comes from the INUNGAMA database [30] that contains all the flood events that have produced socioeconomic impact between 1981 and nowadays. This database contains information such as the date, the counties and municipalities affected, the main rivers or basins involved, the impacts produced, and the type of flood. Information regarding victims has been complemented with newspaper data, mainly La Vanguardia, and official reports.

(9) Balearic Islands (BAL)
The Balearic Islands archipelago (Spain) is situated off the eastern coast of Spain. It consists from five islands (Mallorca, Menorca, Eivissa, Formentera, and Cabrera) with adjacent islets. Precipitation expresses a clear Mediterranean pattern, with a maximum during autumn and a minimum in summer. Mean annual totals range from 1000 mm, in Mallorca, to 300 mm, in the southern islands of Eivissa and Formentera. BAL, with 4492 km 2 , accounts only to 0.3% of the area and to 0.6% of population for TOT-A. However, the population density achieves 258 inhabitants per km 2 , 50.3% of which are females. Data were obtained from a PhD thesis [31], and complemented by research in regional newspapers, such as Diario de Mallorca and Ultima Hora, and by data gathered for the implementation of the Flood Prevention Plan.

EUFF Database
The limitations associated with documentary sources are widely addressed in literature [21,32,33]. The most important ones are as follows: 1.
Data completeness depends on the scale: international news usually report only catastrophic events, while local media also mention events of smaller severity; 2.
Data completeness and quality varie with time, and strongly increase in recent decades, due to news websites; 3.
Details available can differ from one country to another (i.e., due to privacy laws, newspapers in some countries do not report the names of victims).
Because EUFF is a database of FF obtained from documentary sources, it can be affected by incompleteness, which is difficult to quantify. It is impossible to "validate" data in such a kind of database, because independent ancillary information does not exist: all available information is important for database compilation. The construction of the database is a sort of "artisanship work", where all the data found are exploited [18]. If new information becomes available, it is crosschecked with the previous data, and the database is updated accordingly.
Narratives of fatal events gathered from documentary sources were disaggregated in database fields describing victim's profile and the circumstances of the deaths. EUFF follows the data organization already tested in published papers [18,21,23,34]. Each row contains data about a single fatality, organized in fields clustered in six sections, the detailed description of which is available in [18] (2019). MEFF is publicly available but, due to privacy issues, names and surnames of FF are not included (https://data.mendeley.com/datasets/rh9mx7fh7b/1).
EUFF database contains 2466 FF that occurred between 1980 and 2018 in the 9 study areas. The main features of MEFF and EUFF databases are compared in Table 1. The fields of EUFF are very similar to those in MEFF and are reported in (Table 3).

Methods
As for data analysis in this paper, it is not focused on testing any existing hypothesis, but on information collection and explanation. This qualitative research method can be assimilated to the Grounded Theory Approach, a method of research accepted throughout the social sciences and nursing. This method is described as the "discovery of emerging patterns in data" with the aim to generate theory from the research situation in the field, as it is [35].
All the data discussed are available in the tables, both as numbers and as percentage of the total data available, in order to highlight their significance. We discuss the analyses performed using the If we neglect for a while that the number of "FF" represents the number of people who lost their life, it can be argued that the number of "data" are not sufficient to perform complex statistical analyses, for this reason we performed simply descriptive statistical elaborations. Particularly, we present the assessed trend of #FF for TOT-A and SA, and we express it by using the slope angle of the trend line.
Using the large amount of data collected, we assessed seasonality of both events and fatalities and their relative trends. Moreover, we assess the trend of the number of fatalities per event, which represents, to a certain extent, the severity of the event with respect to people.
To compare the number of FF among the different SA, we introduced the Flood Impact Index (FII) that represents a normalization of the number of victims to the surface and population of the SA. It is defined as follows: The ratio #FF/Inhabitants × 100,000 represents the flood mortality on the population of the SA, while #FF/Area (km 2 ) × 1000 actually represents the spatial density of flood fatalities in the SA.

Spatiotemporal Analysis of Fatalities
Between 1980 and 2018, 812 floods killed 2466 people in TOT-A (Table A1). The highest numbers of FF were recorded in TUR (50.4%), followed by ITA (16.5%) and SFR (11.1%). On average, each event killed 3 people (AV#FF/EV), but this figure reaches the maximum in TUR (3.8) and the minimum in CZE (2.1) ( Figure 2). information collection and explanation. This qualitative research method can be assimilated to the Grounded Theory Approach, a method of research accepted throughout the social sciences and nursing. This method is described as the "discovery of emerging patterns in data" with the aim to generate theory from the research situation in the field, as it is [35].
All the data discussed are available in the tables, both as numbers and as percentage of the total data available, in order to highlight their significance. We discuss the analyses performed using the whole dataset and compare them with working hypotheses available in literature and elaborations at the scale of study areas.
If we neglect for a while that the number of "FF" represents the number of people who lost their life, it can be argued that the number of "data" are not sufficient to perform complex statistical analyses, for this reason we performed simply descriptive statistical elaborations. Particularly, we present the assessed trend of #FF for TOT-A and SA, and we express it by using the slope angle of the trend line.
Using the large amount of data collected, we assessed seasonality of both events and fatalities and their relative trends. Moreover, we assess the trend of the number of fatalities per event, which represents, to a certain extent, the severity of the event with respect to people.
To compare the number of FF among the different SA, we introduced the Flood Impact Index (FII) that represents a normalization of the number of victims to the surface and population of the SA. It is defined as follows: The ratio #FF/Inhabitants × 100,000 represents the flood mortality on the population of the SA, while #FF/Area (km 2 ) × 1000 actually represents the spatial density of flood fatalities in the SA.

Spatiotemporal Analysis of Fatalities
Between 1980 and 2018, 812 floods killed 2466 people in TOT-A (Table A1). The highest numbers of FF were recorded in TUR (50.4%), followed by ITA (16.5%) and SFR (11.1%). On average, each event killed 3 people (AV#FF/EV), but this figure reaches the maximum in TUR (3.8) and the minimum in CZE (2.1) ( Figure 2).  In TOT-A, the average number of fatalities per year (AV#FF/Y) is 63.2: the relatively highest portion of fatalities corresponds to TUR (31.9) and the lowest to BAL (0.5). The value of the ratio #FF/Area (km 2 ) × 1000 is 1.64, and reaches the highest values in BAL (4.66) and CAT (3.11). The ratio #FF/Population × 100,000 is 1.24, and it shows the highest value for SFR (3.77) and the lowest for ISR, ITA and POR (0.65).
During the 1980-2018 period, the general trend of FF seems quite stable and the number of fatalities per event slightly decreases, even though the situation is different for individual SA (Figure 3). Looking on SA linear trends, #FF is decreasing for TUR and CAT and increasing for GRE, ITA, and SFR. The number of fatalities per event (#FF/#EV) decreases in TUR and increases in GRE, CZE, and SFR. These graphs show that the annual amount of FF is very high in TUR (more than 50 in 6 years of the reference period), while there are other study areas where the maximum annual amount of FF did not surpass 5 fatalities (e.g., CZE and BAL), even if these values must be reported to the size and population of the SA. Water 2019, 11, x FOR PEER REVIEW 11 of 28   Figure 4 shows the monthly numbers of fatalities (#FF), and monthly numbers of events (#EV). In TOT-A, 560 (69%) of events occurred between June and November, causing totally 1894 fatalities (77%). The events exhibit the maximum in October (122 events causing 360 fatalities, i.e., 15% of FF) and a secondary maximum in July (103 EV, causing 387 FF, i.e., 16% of FF). November was the most hazardous month in terms of fatalities (402, 16%). The peak in November reflects the general trend of TOT-A, while the one in July (387, 16%) is strongly affected by the high number of FF recorded in TUR (307, 12%). Monthly distribution of FF is highly affected by the climate features and monthly distribution of rainfall in each SA.    By rescaling our results using the Flood Impact Index (Table 4) it must be noted that the largest impact pertain to SFR, where the number of FF is very high, if normalized to both population and extent of the territory (19.12). The second value of FII pertains to BAL (8.42), followed by CAT (4.13). By splitting the study period in decades, and looking at the line slope for #FF of the different SA, further 'qualitative' details can be noticed (

Basic Features of Fatalities
In this section, we summarize the main relationships between the variables characterizing the fatal events, referring to the further quoted appendices. Figure 6 presents the relationships between gender and the other variables, while Figure 7 presents the relationships between age of FF and the other variables.

Basic Features of Fatalities
In this section, we summarize the main relationships between the variables characterizing the fatal events, referring to the further quoted appendices. Figure 6 presents the relationships between gender and the other variables, while Figure 7 presents the relationships between age of FF and the other variables. Water 2019, 11, x FOR PEER REVIEW 13 of 28 Figure 6. Relationships between gender and place, activity, condition, dynamic, protective, and hazardous behavior, for the victims of known gender (1936, 78.5% of FF). Unknown percentages are in red for females and blue for males.


Light conditions. There is evidence that dark conditions worsen the situation, causing a slightly greater percentage of FF in TOT-A, as can be inferred by the higher value of FF in dark conditions in SAs where data on this variable are almost complete, as in GRE, POR, and ITA (Table A1).  Gender. For TOT-A, fatalities were mainly males (46.9%). Looking at the SA, males represent the larger percentage of FF (Table A1). The largest difference between genders pertains to CZE (71.1%   • Light conditions. There is evidence that dark conditions worsen the situation, causing a slightly greater percentage of FF in TOT-A, as can be inferred by the higher value of FF in dark conditions in SAs where data on this variable are almost complete, as in GRE, POR, and ITA (Table A1).

•
Age. The majority of fatalities were in the age of 30 ÷ 49 years (15%), followed by 50 ÷ 64 years (13%) and 65-84 (12.9%) (Table A1). Females were more numerous than males in the following classes: <15 years, 15-29 years, 65-84 years, and over 85 years (Figure 8).     (Table A3). If we analyze motor vehicles altogether (bus, car, caravan, tractor and van), the percentage of cases rises until 19.8%. Condition slightly differs among males and females, but fatality by car was the most frequent for both genders. In all but the over 85-year fatalities, the majority of victims were by car and secondly standing. • Activity. Traveling/to home/to work was the most frequent activity in which victims were involved at the moment of fatal events, both in TOT-A (445, 18%) and in all SA, except CZE, POR, and SFR (Table A3). This is true for both males and females. Males were working and females were doing recreational activities. Stronger differences pertain rescuing someone, more frequent among males than females, and hunting and fishing, detected only in cases of male fatalities. Traveling/to home/to work was found to differ among the age categories. Fatalities for people over 85-years occurred while either sleeping or doing recreational activities. • Dynamic. Dragged by water/mud was the most frequent dynamic of fatal events in TOT-A (1231, 49.9%) ( Table A4). The victims under 85 years were more commonly dragged by water/mud, while people above 85 years were very likely to be blocked in a flooded room.

•
Protective behaviors were detected in 130 FF (5.3%) (Table A4). Even if the protective attempts were unsuccessful, we identified mainly attempts to get out of car (62, 2.5%) and moving to safer place (44, 1.8%). Among people for whom we have corresponding information, females were more numerous than males. Concerning the classes of age, the majority of people showing protective behavior were in the class 30-49 years, surprisingly followed by children (0-14 years).

•
Hazardous behaviors were detected in 262 FF (10.6%), and the majority of cases concerns males (Table A5). This may be associated with the lower level of flood-risk perception attributed to males, according to recent research targeting Greek population [36]. People exhibiting hazardous behavior were first between 30 and 49 years and secondly between 50 and 64 years. The majority was either fording rivers (95, 3.9%), as frequently reported in literature [37,38] or staying on river banks (40, 1.6%) and staying on bridges during floods (25, 1%). Fording rivers were the most frequent hazardous behavior among both females and males. Almost all the types of hazardous behaviors were detected in people in the ages between 30 and 64 years. After 30 years, trying to save cars and belongings were often detected, while in people older than 50 years, also refuse warning and refuse evacuation were detected. • Cause of death. Six types of clinical causes of death were reported, although in different environments the list must be updated. In TOT-A, the cause of death was mainly drowning (1693 FF, 68.7%) (Table A5). Collapse/heart attack caused 237 (9.6%) fatalities. Surprisingly, the victims killed by collapse/hearth attack seem common in all the classes of ages, and not restricted to elderly fatalities, as could be expected (Figure 9).
Water 2019, 11, x FOR PEER REVIEW 16 of 28 Figure 9. Number of fatalities (on y-axis) sorted per age (x-axis) and cause of death for TOT-A.

Discussion
Data collected in the individual SA present different levels of completeness concerning variables describing fatal events. Then, the reliability of data for each variable is measurable using the percentages of cases known. Based on these percentages, results obtained for each variable can be considered either reliable or purely indicative, thus requiring further investigation.

Data Numerousness
Simplifying the approach presented by [18], we defined data reliability as high, medium, and low.
a. Variables of low reliability (data available between 0% and 30% of the total):


Protective and Hazardous behaviors were reported in 5.3% and 10.6% of the cases, respectively, thus they are not strictly representative of all what really happened. Data about hazardous behavior were not gathered in ISR and BAL, because the data sources did not reported details on this, while in POR were detected for 47.8% of its fatalities.  Disability: probably due to privacy reasons, the narratives of the events are not very explicit about this factor. In addition, we were not sure that this information was correctly reported, even in the events more accurately described. The cases declared of disability are only 54 (2.2%).
b. Variables of medium reliability (data available between 60% and 30% of the total):  Activity is available only in 32.0% of the cases, and data are most abundant for GRE (71.2%) ( Figure 10

Discussion
Data collected in the individual SA present different levels of completeness concerning variables describing fatal events. Then, the reliability of data for each variable is measurable using the percentages of cases known. Based on these percentages, results obtained for each variable can be considered either reliable or purely indicative, thus requiring further investigation.

Data Numerousness
Simplifying the approach presented by [18], we defined data reliability as high, medium, and low. a.
Variables of low reliability (data available between 0% and 30% of the total): • Protective and Hazardous behaviors were reported in 5.3% and 10.6% of the cases, respectively, thus they are not strictly representative of all what really happened. Data about hazardous behavior were not gathered in ISR and BAL, because the data sources did not reported details on this, while in POR were detected for 47.8% of its fatalities. • Disability: probably due to privacy reasons, the narratives of the events are not very explicit about this factor. In addition, we were not sure that this information was correctly reported, even in the events more accurately described. The cases declared of disability are only 54 (2.2%).
b. Variables of medium reliability (data available between 60% and 30% of the total): • Activity is available only in 32.0% of the cases, and data are most abundant for GRE (71.2%) ( Figure 10).  Despite data uncertainty, that must be taken in account, EUFF database significantly contributes to fill a gap in information on FF on a large scale, in areas where similar databases are not available, and especially covering such a long study period. Particularly, it fills the gap of data on floods causing a number of victims under the threshold to be included in international databases (for EM-DAT, i.e., is 10 people, https://www.emdat.be/). In contrast to international databases that follow a multihazard approach, the EUFF only contains FF, not aggregated with fatalities caused by other hazards as lightning, wind, and landslides. This drive to a stricter analysis and affordable results, based on features detected during floods with fatalities. For each victim, the narrative of the event is separated in elementary inputs (fatality age, gender, activity, place, behavior, etc.), and this allows to analyze and assess the relative weight of different features in the fatal events. Despite data uncertainty, that must be taken in account, EUFF database significantly contributes to fill a gap in information on FF on a large scale, in areas where similar databases are not available, and especially covering such a long study period. Particularly, it fills the gap of data on floods causing a number of victims under the threshold to be included in international databases (for EM-DAT, i.e., is 10 people, https://www.emdat.be/). In contrast to international databases that follow a multi-hazard approach, the EUFF only contains FF, not aggregated with fatalities caused by other hazards as lightning, wind, and landslides. This drive to a stricter analysis and affordable results, based on features detected during floods with fatalities. For each victim, the narrative of the event is separated in elementary inputs (fatality age, gender, activity, place, behavior, etc.), and this allows to analyze and assess the relative weight of different features in the fatal events.

Broader Context and Regional Peculiarities
In EUFF, the majority of FF are males, as obtained in similar studies performed in Switzerland [39], in Europe, USA [40], and in Australia [37,38], showing that males are more exposed to flooding than females. This can depend on two factors: (i) males were more numerous than females in outdoor works, and, until recently, rescue services (e.g., fire fighters, police, and defense forces) consisted entirely of males; (ii) probably, males are most inclined towards risk taking behaviors [19]. Concerning the age of people (majority of adults in working age), it can be explained by similar reasons: people who daily reach the work place are more exposed to floods outdoor, while retired people spent more time at home. Thus, elderly people are more frequently trapped by flood in their home while adults and children are dragged outdoors [41,42].
Compared to EUFF, it appears that in other part of the world the percentage of fatalities by car is higher. The authors of [43], for example, detected 4586 flood fatalities in US between 1959 and 2005, 63% happened in vehicles, while in our database fatalities in vehicles were 19.8% of the total.
For Greece, the flood mortality rates are lower than the average value assessed for TOT-A. However, unlike the average and most EUFF regions, there seems to be quite a strong upward increasing trend in both annual deaths and #FF per event. This may be related to low levels of flood risk awareness and precautionary behaviors among Greek citizens [36,44].
The decrease of FF in Catalonia is due to different factors. Firstly, the frequency of catastrophic floods has decreased in the last decades [45] although this can be part of the natural variability, as has been found for the last 700 years [46]. Secondly, it depends on the significant improvement of risk awareness, preparedness, and emergency management [47] and urbanistic rules that forbid the creation of new urban settlements in flood prone areas. As an example, more than 815 people died in one single event that affected a small region in Catalonia in the nighttime, on 25 September 1962 [48]. A similar pluvial event on 10 June 2000 only killed three people.
In SFR, the evolution of FF is a bit erratic. After a little deadly 1980-1990 decade, the 1990s and beginning of the 2000s were marked by numerous and serious deadly events. This was probably related to an increasing of torrential rain, which affected high-populated sectors. As a result, there was the strengthening of flood prevention with the creation of plans for prevention of risks in 1995 and the enactment of the law risk of July 2003. Mortality was minimal between 2004 and 2009, but serious disasters (as in the departments of Var in 2010, Côte d'Azur in 2015, and Aude in 2018) recently worsen the human toll of floods (more than 10 people per year on average).
More generally, basing on the trend in the study period, SA can be divided in three main groups, according to its annual trends of FF: (i) a downward trend group (CAT, BAL, TUR, POR); an uptrend group (GRE, SFR, ITA); and a stable trend (ISR and CZE). Flood events that generated fatalities are very diverse in the SA without a specific grouping.
Data availability and corresponding features can be gathered in two distinct groups of SA. The first one can be defined as a data scarce context, where more than seven EUFF variables where not available for 50% of FF in a SA. In this group, we include TUR, ISR, SFR and CZE, where the variables that are mostly missing or are scarce correspond to activity, condition, light conditions, protective behavior, and hazard behavior. On contrary, there is a group of SA with a relatively higher amount of data availability (more than 50% of availability per FF variables), including ITA, BAL, CAT, GRE, and POR. In this group, the variables less detailed are protective behavior and hazardous behavior. Differences can be justified by different climatic characteristics among the SA, which range from Mediterranean climate to temperate and semi-arid climate, which controls the amount and intensity of annual and monthly rainfall distribution and the flood frequency. Additionally, several geomorphological and hydrological factors control the predisposing factors of floods on the field. Other aspects are less controlled by physical constraints, like for instance the hazardous behaviors taken by individuals or stakeholders. And last but not least, the human exposure of flood hazardous zones can be controlled by demographical and economical drivers, but also by the existence/inexistence of spatial planning to avoid hazardous zones.

Importance of the Study and Future Research
The importance of the paper depends on the significant input to the overview of situations leading to FF in different environmental and cultural frameworks. The impact of this work is in two points: (a) The presentation and exploitation of the absolutely new and unpublished database, which required strong efforts in terms of coordination and data homogenization; (b) The results of the paper give original insight into the knowledge of people-flood interaction, which could be used as a building block in increasing resilience campaigns. The results may support educational campaigns tailored to the features really detected and aiming to manage risk and improve people protection in forthcoming floods.
After a long work of data gathering and an intensive phase of their systematization, we created an important source of data, only partially exploited in the present paper. The research will then continue trying to enlarge the total area studied, firstly by adding further countries to the research group, and secondly extending Spanish and French regions to cover the entire countries. Planned activities for the forthcoming of the present research concern further elaboration, both at the scale of the study areas and considering the TOT-A, for the features deserving better understanding, such as the relationship between flood magnitude and number of victims. Further analyses of the historical series collected will highlights major changes in circumstances of fatal events throughout the years, thus supplying a picture of the current frequent situations in which people could be hurt in future events.
In addition, one of the biggest challenges of our research will be to involve indices of wellbeing and economic situation of the various countries involved in the research, to evaluate the hypotheses available in the literature linking these parameters to the vulnerability of people.

Conclusions
Between 1980 and 2018, 812 floods killed 2466 people in nine study areas located in Europe. Monthly distribution of both flood events and flood fatalities strictly depends on monthly distribution of rainfall in each study area. In TOT-A, 69% of events occurred between June and November, causing 77% of fatalities. The events exhibit the maximum in October (15%) and a secondary maximum in July (13%). November was the most hazardous month in terms of fatalities (16%).
As a whole, the number of fatalities per event slightly decreased, while the general trend of fatalities seems stable. An increasing trend is observed in Greece, Italy, and South France, especially from 2000 to 2018. This may be due to a combination, in these study areas, of intense floods and low ability of people to react, and can be affected by changes in population density in hazardous areas.
The highest numbers of fatalities were recorded in Turkey (50.4%), Italy (16.5%), and South France (11.1%). By normalizing fatalities to the number of inhabitants of each study area, it appears that South France shows the highest rate, meaning that in this area floods affect the largest percentage of inhabitants with respect to the other study areas. By calculating the flood impact index, taking into account both the number of inhabitants and the density of fatalities normalized to the surface of each study area, we obtained the impact of floods on human lives. This index assumes the highest value for South France followed by Balearic Islands.
The majority of victims were residents in the area of the event. Males were more numerous than females, especially in Czech Republic and Greece. Fatalities were mainly aged between 30 and 49 years, and between 50 years and 64 years. Females were more numerous than males in the age classes <15 years, 15-29 years, and over 65 years. Mortality does not increase with age, neither for males nor for females, while fatalities were more abundant in those parts of the population who were more involved in outdoor activities, related to both work and traveling.
Fatal events occurred more frequently outdoor, and particularly on the roads. The majority of victims, both males and females, were by car or other motor vehicles, traveling/to home/to work when they were dragged by water/mud. It seems that elderly are not particularly vulnerable: the few fatalities over 65 years (1.9%), oppositely to the other classes of age, were mainly killed indoor, blocked in a flooded room, when sleeping.
The primary cause of death was drowning. The second most frequent cause was collapse/heart attack, which was detected in all the classes of ages.

•
Protective behaviors, as attempts to get out of car and moving to safer place, were more frequent in female than in males, and mainly in the class 30-49 years, surprisingly followed 0-14 year-old victims.

•
Hazardous behaviors, such as fording rivers and staying on riverbanks or bridges, were more frequent in males, but fording rivers were also numerous among females. Fatalities in the ages between 30 and 64 years exhibited all the types of hazardous behaviors identified in this study. Fatalities over 30 also died trying to save cars and belongings, while victims over 50 also refused warning and evacuation. This information confirm the educational importance of this study in prevention of fatal events.
As all the research based on documentary data sources, incompleteness can affect data. The level of detail is strictly related to both age and severity of the events: low-severity events occurred several years ago can be scarcely documented, while more details can be found on severest events recently occurred.
EUFF database represents a unique source of data for the study of floods victims with a broad potential for further spatial and temporal extension for different use, even if some data uncertainty following from the use of documentary data has to be taken into account. The novelty of this study lies in the use of data describing flood fatalities in different countries, allowing to investigate local features governing behavioral choices in the flood-people interactions. Moreover, the study period is long enough to identify trends and perform statistical elaboration. Results can be used for the education of population, teaching to not underestimate the danger of floods and to avoid hazardous behaviors.
Future developments will try to enlarge database by adding further countries. Planned activities concern the analysis of: (i) Relationship between flood magnitude and number of victims; (ii) evolution of circumstances of fatal events throughout the years; and iii) relationships with indices of wellbeing and economic situation of the countries involved in the research, to highlight possible relationships of these parameters with people vulnerability. Resechers who like to contribute to this database can contact us.

Protective Behaviour
Climbing

Hazardous Behaviour
Check damage during the event