Victims crossing overflowing watercourses with vehicles in Spain

Heavy rainfall causes many watercourses to overflow. In these circumstances, crossing by car even on a road, can be extremely dangerous; however, dozens of drivers are swept away every year in their vehicles. This paper analyses this type of accident in Spain between 2008 and 2018, recording the date, location, number of victims, age and gender, and rainfall during the event. The results show that 125 accidents occurred with 200 victims including 45 fatalities. Most accidents occurred in E, S and SE Spain, where the rainfall irregularity is greater, during December, October and March, although fatalities were concentrated in September and October. Among the victims male drivers dominated, with an average age of 52 years. The main cause of these accidents was the drivers' behaviour due to: underestimating risk, overconfidence, overvaluation of their driving skills, an excess of trust in the authorities, ignorance about vehicle drag and buoyancy risks, and, social pressure. To reduce these risks, it is necessary to increase adaptation and protection measures on roads, but above all, a change in drivers' behaviour to stop them trying to cross‐flooded rivers.


| INTRODUCTION
Roads are linear elements that intersect watercourses in their layout. These crossings can be done through bridges, drainage works or fords, depending on the importance of both the road and the watercourse. Main roads usually have wide river crossing works, which guarantee traffic safety during intense rainy episodes. However, many secondary and unpaved roads have smaller drainage works and bridges, and even cross non-permanent streams by fords.
Rainfall distribution in the Mediterranean region is very irregular, with high concentrations in a few days (García-Barrón, Aguilar-Alba, Morales, & Sousa, 2017), which favours rivers overflow. There are a large number of non-permanent watercourses where water only circulates some days a year, during intense rain episodes, but in those situations the flow can be torrential. Many of these watercourses are crossed, directly or through minor drainage works, by local roads; they are passable almost always, except when there is heavy rain. In addition, during extraordinary episodes river levels rise and can exceed drainage works and bridges, flooding the roads, even the main ones.
Climate change is affecting this region intensely, producing a reduction in average rainfall (11% between 1961 and 2006), although irregularly, with a significant reduction in February, June and March and an increase in May, August, September and October (Río, Herrero, Fraile, & Penas, 2011). There are discrepancies about changes in the intensity and frequency of extreme rainfall, due to the high uncertainty: van den Besselaar, Klein, and Buishand (2012) and the fifth IPCC report (Christensen et al., 2013;Hartmann et al., 2013) indicated that the irregularity in the region will increase, and Barrera-Escoda, Gonçalves, Guerreiro, Cunillera, and Baldasano (2014) found an increase of the frequency of heavy precipitation events during the 2021-2050 period in Catalonia (Spain), but Valdés-Abellán, Pardo, and Tenza-Abril (2017) point to a decreased frequency of heavy precipitation events in the last 20 years.
Also important are changes in land use, with great influence on runoff; at basin scale these changes are more important than those associated with climate (Whitfield, 2012). Consequently, river overflowing at road crossings is common throughout the Mediterranean region, and increases predictably in the future.
The aim of this paper is to analyse accidents crossing overflowing rivers by car between 2008 and 2018 in Spain. Based on the results, the most usual risk profile by gender and age has been established, proposing mitigation and adaptation measures, with an special attention to drivers' behavioural adaptation.

| METHODS
The study area is Spain, and the period 2008-2018. Accidents crossing overflowing rivers on vehicles during this period have been analysed. An accident is considered to occur when a vehicle that was crossing a watercourse, by its natural channel or in an area where as a result of a flood was overflowed over a track or road, is trapped, immobilised, submerged or dragged, affecting its occupants, who are forced to escape, be rescued or die. Accidents that occurred when vehicles fell into rivers during accidents not associated with rain or river overflow (e.g., collisions in which vehicles fell into the water or distractions from drivers that caused them to rush into waterways), or when the vehicles were dragged by the water while they were parked have been excluded.
A database was developed, which included the information found on each accident. The established fields were: accident date; accident location (place, province and region); driver data (gender, age and damage suffered -died, rescued or escaped themselves-); passenger 1 to 4 data (gender, age and damage suffered); media information (name and publication date); and rainfall in the location during the accident date.
The information sources are news from national or regional newspapers, radio or TV channels, or major news agencies; no reference outside these media has been considered.
To carry out news searches a general search engine has been used (Google). The searches have been done applying several filters, language (conducting the searches in Spanish, and referred to Spain), and date (carrying them out year by year for all the studied period). Keywords established to conduct the searches (originally in Spanish) were: 'sweep away'/'rescued'/'disappeared'/ 'drowned'; 'crossing'; 'river'/'stream'/'watercourse'; 'car'/ 'vehicle'/'motorcycle'. Different combinations of these terms have been done, repeating the searches until no new results were obtained. Each search with a combination of some of the indicated terms resulted in around 100 entries, so with all the possible combinations several hundreds of results per year were obtained, many of them repeated. The results were filtered on a case-by-case basis by reading the titles of the entries; the process was quick and simple, because most results lack a relationship with the studied topic, focusing only on those related to accidents.
The results related to accidents with vehicles crossing rivers were analysed in detail. Firstly, the information source (media) quality was analysed, to accept it or not; only information from the media indicated above has been considered, omitting web pages, videos or other sources whose reliability could not be proven. Secondly, the news reports were read, completing an entry in the database with the information obtained in each one. The data obtained for each entry were, at least, the date and location of the accident, the number of victims, and the name of the media and publication date. Depending on the event the amount of information obtained was more or less; the news about accidents with fatalities used to be more complete, including data on gender and age of the victims. Frequently the same event was published in different media, especially the most tragic; all of them were consulted, to complete the maximum number of fields in each database entry. The reference included was that of the media that provided more complete information, or more than one when they were complementary. In events with poor information complementary searches have been done, limiting the date and entering new terms, such as the location of the accident.
Voumard, Derron, and 6 Jaboyedoff (2018) applied a similar methodology to analyse natural hazard events affecting transport networks in Switzerland; Diakakis and Deligiannakis (2015) used press articles to identify flood fatalities in Greece; and Pereira, Zêzere, Quaresma, Santos, and Santos (2016) use also newspapers information to construct a database on flood and landslide consequences in Portugal. Other authors used data from social media to assess flood impacts or risks (e.g., Smith, Liang, James, & Lin, 2017for UK, or Fang, Hu, Shi, & Zhao, 2019. Once the search for information was completed, each database entry was completed including the rainfall record for the accident location and date. The information was obtained from the State Meteorological Agency databases (AEMET, 2019), consulting the meteorological station with daily rainfall records closest to the accident site, on the date on which it took place or the day before (to take into account concentration runoff times and possible delayed floods). When the searches did not provide results, due to a lack of data, the next nearest station was chosen, until results were obtained.

| RESULTS
Between 2008 and 2018, 125 accidents produced when trying to cross overflowing watercourses with vehicles were registered, 33 of them with fatalities ( Table 1). As a result, 200 people were involved in these accidents; 45 died, 137 were rescued and 18 escaped by themselves. The average number of victims per accident is 1.6, and the death rate per accident was 0.23. The victims were 92% Spanish and 8% foreign (most of them from the EU).
There has been an increase in the number of accidents over the period studied, although those with fatalities have increased with less intensity. The number of fatalities was irregular over the period (2018 recorded the highest number of victims of the decade), but the number of rescued victims was markedly higher in 2016 and 2018 ( Figure 1). Average values throughout the studied period were 11.4 accidents/year, 18.2 victims/year and 4.1 deaths/year; however, in 2018 these figures were 43, 64 and 12, respectively, 2.9-3.7 times higher than the average values.
Months with most accidents were December, October and March, while accidents with fatalities were mainly concentrated on September and October ( Figure 3). The average of the daily rainfall during the accidents that have taken place in each month has been calculated. Analysing these results in parallel to the number of accidents, a close relationship is observed, especially in accidents with fatalities ( Figure 3).
Average daily rainfall during accidents is 43.5 mm, with maximums of 259 mm (Malaga in 2018) and 241.6 mm (Almería in 2012). There are some accidents where rainfall is null or very low, which may have several explanations. Firstly, the network of meteorological stations with daily rainfall data is limited; the stations closest to the accidents may not have data on the date they occurred. Secondly, the river overflowing may be due to spatially remote precipitation, for example at the head of the basin; this means that even if it does not rain in a certain area, the watercourses may be overflowed. Finally, some accidents may be due to the river flow without the influence of exceptional rainfall; this could be the case of an accident in Córdoba in 2010, with scarce rainfall, but a large river with an important flow regularly, which the victim tried to cross.
Average annual rainfall in the provinces with most accidents and more fatalities is (in brackets the average number of rainfall days per year): Valencia 475 mm (43.9); Almería 200 mm (25.4); Murcia 297 mm (36.5); Mallorca (Balearic Islands) 411 mm (50.9); consequently, rainfall during accidents is, on average, 8.5-20.4% of annual rainfall, reaching 48.5% in Málaga in 2018, an even 120.8% in Almería in 2012, a daily event greater than the annual average rainfall. Regarding the maximum rainfall in 24 hr, the annual average for all the regions is about 35 mm, with a maximum of 51 mm in the east, although it is a very irregular parameter; the mean of the extreme maximums is 170 mm, with an exceptional value of  (Figure 4). Months with greater accident rate usually had also higher rainfall associated with them, especially on those with fatalities; accidents with fatalities in September and October had on average 70 mm of daily rainfall, a value higher than the usual.
The analysis of accidents and fatalities related to precipitation shows that the maximum accident rate occurs with reduced rainfall, although the number of fatalities is low ( Figure 5); when rainfall increases the number of victims also does, up to a limit of 30-60 mm, falling above Gender information was obtained for 60.5% of the victims, 73.6% of drivers and 38.7% of passengers (Table 2); it was available for all the deceased victims except one passenger (100% of drivers and 97.8% of total fatalities). Age was obtained for all the deceased, but only in 25% of surviving victims; information was available, therefore, for 42% of the victims, 44.2% of the drivers and 100% of the fatalities. Pereira et al. (2016) indicate that age is a factor frequently missing in reports about fatalities caused by floods in Portugal. Sharif, Jackson, Hossain, Bin-Shafique, and Zane (2010) got information on gender of 81% and on age of 58% of victims of motor vehicle-related flood accidents in Texas, and Salvati et al. (2018) data on gender of 79.3% and about the age of 78.1% of flood fatalities in Italy.
Average age of the victims was 48.2 years (Table 2); age increased from victims that had escaped, the younger (41.3), to that deceased, the elder (50.9). Driver's age (50.5) was higher than the average age of the victims, and passengers' lower (43.4). Gender of the victims ( Table 2) is 71.9% male and 28.1% female. The percentage of male victims increase for those escaped and rescued, and decreases for the deceased; male victims dominate, but the proportion of deaths among victims is higher in women. There is dominance of male drivers (84.8%) and female passengers (69%).
Average age of male drivers (52.2) was higher than female (39.8) (Table 3); taking into account only deceased drivers, average age increased in male (55.4) but decreased in female (39). In male drivers, the age increases from the victims that had escaped, the younger, F I G U R E 3 Accidents distribution by months and associated daily rainfall. Left: total accidents. Right: accidents with fatalities to those deceased, the elder, but in female not; however the total number of female drivers is significantly lower.
As a result, the archetypal driver that crossed an overflowing river in Spain was a male of around 52 years, dying in a third of cases, especially the older ones ( Figure 6).

| DISCUSSION
Most accidents crossing overflowing rivers occurred in E, S and SE Spain. In these regions rainfall irregularity favours non-permanent watercourses, which remain dry nearly all year round, but may have torrential flows during a few days a year, after intense rains that produce flash floods. Minor roads frequently cross these watercourses with fords or small drainages, useful almost always but extremely dangerous during heavy rains; this explains to a large extent the concentration of accidents in these regions. Flash floods have also been identified as an important cause of fatalities in Texas, Greece or Portugal (Papagiannaki, Lagouvardos, & Kotroni, 2013;Pereira, Diakakis, Deligiannakis, & Zêzere, 2017;Sharif et al., 2010). Gissing, Opper, Tofa, Coates, and McAneney (2019) highlight as a risk factor a small upstream catchment length, which may increase the rate of rise of floodwaters.
Months with most accidents in Spain are December, October and March, but those with more fatalities are September and October; most flood events in the North-Western Mediterranean occur between August and November (Llasat et al., 2014). Flood fatalities predominantly occur during autumn in Greece and during winter in Portugal (Pereira et al., 2017); in Spain fatalities occur predominantly in autumn, although December (winter) is the month with more accidents. Mediero, Santillán, Garrote, and Granados (2014) indicate that autumn floods showed decreasing trend in eastern Spain until 2009, but they have been important in the last decade; with the exception of the floods of October 2007, the number of people affected by floods between 2010 and 2013 is higher than between 1997 and 2007 (Gutiérrez, 2016).
As noted above, the highest accident rate occurred with reduced rainfall, increasing with rainfall, decreasing later and increasing again (Figure 4). This seems to be related to a mismatch between the actual and perceived risk. With scarce rainfall risk perception is low, leading to greater drivers' imprudence, although with not much fatalities; as indicated by Hamilton et al. (2016) when water depth is low drivers believe they are capable of crossing it. When the rainfall increases the risk does too, but perception does not do it in parallel, so the number of victims increases. Higher rainfall produces an increased sense of risk, and greater prudence, until very high values are reached, with which the risk becomes extreme, even with awareness about it.

F I G U R E 4 Rainfall in 24 hr
F I G U R E 5 Accidents and fatalities related to rainfall 2017, Salvati et al., 2018 andPetrucci et al., 2018 in Italy;or Vinet, 2017 in Europe). Attending only to drivers, male percentage is even higher, 84.8% of total and 86.2% of the deceased (Table 2); deceased male drivers in Greece were 85.6% (Diakakis & Deligiannakis, 2013), 75.8% in Europe (Jonkman & Kelman, 2005) and 71.2% in Australia (FitzGerald et al., 2010). This over-representation shows a different degree of exposure between males and females .
In the 2018 census of drivers of Spain (Figure 7; DGT, 2019) driving licences of passenger cars (the most commonly involved in the analysed accidents) were 54.3% male and 45.7% female; however, gender differences begin to increase from an age of 45, so that at 74 years 83% of driving licences belong to men. This disproportion may explain the higher male accident rate in elderly drivers, but not below 45 years, although the car use rate in Spain is usually somewhat higher in men. Aparicio, Arenas, Mira, Páez, and Furones (2017) compare the involvement of men and women in traffic accidents in Spain, concluding that the ratio men/women died is around 3.4:1, and that female drivers have more respectful behaviour with traffic regulations, fewer infractions and less risk taking. Many researches about risk perception conclude that females perceived more risks than males (Albentosa, Stephens, & Sullman, 2018;Deffenbacher, Lynch, Filetti, Dahlen, & Oetting, 2003;Diakakis & Deligiannakis, 2013Lancaster & Ward, 2002;Sharif et al., 2010;Siegrist et al., 2005;SIRC, 2004).
Average age of victims was 48.2 years, increasing to 50.9 on the deceased. Results vary according to authors and regions: 51 years in France (Boissier, 2013); 20-60 years in Europe and Texas (Jonkman & Kelman, 2005;Sharif et al., 2010); 10-29 and over 60-70 years in the US and Australia (Ashley & Ashley, 2008;FitzGerald et al., 2010); 18-35 for people crossing flooded areas on foot in Iran (Shabanikiya,   (Aceto et al., 2017). Average age of drivers is 50.5 years, increasing to 53 years on the deceased; this result is consistent with the obtained for Greece, mainly 40-69 years (Diakakis & Deligiannakis, 2013), but different from the obtained in the US, mainly 18-35 years (Drobot et al., 2007). The average age in affected and deceased male drivers (52.2 and 55.6, respectively) is higher than in female (39.8 and 39.3).
There is a demonstrated tendency of young people to have riskier driving behaviour (Amponsah-Tawiah & Mensah, 2016;Cassarino & Murphy, 2018;Kinnear, Kelly, Stradling, & Thomson, 2009;Lancaster & Ward, 2002;Scott-Parker, Watson, King, & Hyde, 2012;Siegrist et al., 2005;Taubman, Skvirsky, Greenbury, & Prato, 2018). Middle-aged and older drivers are more conservative than younger drivers (Ahmed & Ghasemzadeh, 2018), so accident liabilities reduce with age due to increased maturity and driving experience (Maycock, 2002). However, the results of our research show that the most affected are precisely middle-aged drivers. Drivers learn by receiving feedback from driving (Martinussen, Møller, & Prato, 2014), but most drivers have never crossed an overflowed watercourse, so they lack that feedback; in addition, experience of mature drivers is counterproductive, giving them a false sense of security.
Among the registered victims there are 10 minors, 37% of the occupants and 12% of the total victims of known age. The average age of the minors ranges from 1 to 15 years, with an average value of 7.1 years; 4 of them died (mean age 5.3 years) and 6 were rescued (mean age 8.3 years). It is a remarkable percentage, because it shows that the presence of toddlers in the vehicle does not seem to be a sufficient reason to discourage drivers from crossing flooded areas, which reinforces the idea of risk undervaluation.
It is noteworthy the presence of 8% of foreign victims, a value near the 10% of foreigners dead in floods in Catalonia, NE Spain, indicated by Nakamura and Llasat (2017). Spain is the second tourist destination in the world, with 81.8 million of international tourists in 2017 (UNWTO, 2018), which largely justifies this high percentage. In addition, drivers have a particular perception of risk in each country (Nordfjaern, Jørgensen, & Rundmo, 2011;Şimşeko glu, Nordfjaern, & Rundmo, 2012), and even between regions of the same country (Lu, Zhang, Peng, & Sertajur, 2014); affected foreigners might not be familiar with the road environment (Yannis, Golias, & Papadimitriou, 2007), and especially with the irregularity of rainfall in the Mediterranean region. Hamilton et al. (2016) analyse drivers' willingness to drive through flooded waterways, including attitudinal, social and efficacy beliefs; drivers' perception of risk, and if they think they will not experience negative consequences have great influence, but also their social environment and companions (e.g., partner or friends).
The main cause of river-crossing accidents seems to be overconfidence and overvaluation of driving skills, which reduce the perception of risk (Diakakis & Deligiannakis, 2013;Horswill et al., 2002;Ruin et al., 2007;Siegrist et al., 2005;Terpstra, 2011), underestimating the existing danger and the capacity of the flowing water to drag vehicles (FitzGerald et al., 2010). Flood risk perception is low among the population in Spain (Nakamura & Llasat, 2017), and also in some regions drivers are accustomed to crossing dry watercourses, and this habit leads to an excess of confidence, fatal in cases of overflow. A vehicle exposed to flooding, after losing stability, becomes buoyant and may be washed away (Martínez-Gomariz, Gómez, Russo, & Djordjevi c, 2017;Shah, Mustaffa, Martínez-Gomariz, Kim, et al., 2019). However, drivers perceive their cars as heavy and stable; water forces, which cause vehicle sliding and buoyancy, are not evident. Vehicle stability on flooded roads depends on water depth and velocity; several studies analyse these factors (Martínez-Gomariz et al., 2017;Martínez-Gomariz, Gómez, Russo, & Djordjevi c, 2018;Shah, Mustaffa, Martínez-Gomariz, Yusof, & Al-Qadami, 2019;Shah, Mustaffa, Martínez-Gomariz, Yusof, & Al-Qadami, 2019;Shah, Mustaffa, Martínez-Gomariz, Kim, et al., 2019;Smith et al., 2019;Bocanegra, Vallés-Morán, & Francés, 2020). Horizontal water force may produce vehicle sliding, and with a depth around 0.60 m water floating may occur (Kramer, Terheiden, & Wieprecht, 2016); the depth increases buoyancy and reduces the force required to move the vehicles (Smith et al., 2019). Shah et al. (2019) indicate that a small passenger car (weight ≤800 kg) that progresses slowly along a flat flooded road remains stable if the product of water velocity and depth is less than 0.70 m 2 /s. Consequently, when depth increases lower water velocity is required to drag a car, or lower depths were required at high flows ; Teo, Xia, Falconer, and Lin (2012) indicate that deep water at low speeds can cause as much damage as shallow water at high speeds. Martínez-Gomariz et al. (2018) analyse stability criteria for vehicles exposed to flooding, concluding that the most safety is AR&R (Shand, Cox, Blacka, & Smith, 2011); in this criterion the limiting water depths for stability are, depending on vehicle size, 0.3-0.5 m in still water and 0.10-0.20 in high velocity flow. In addition, when rivers overflow there is usually little or no visibility of what is below the water surface, so even shallow waters with moderate flow can make vehicles unstable (Taylor et al., 2019). If the vehicle is surrounded by calm waters, the occupants can usually leave, although with difficulties, but if it is dragged by a rapid flow the chances of escape are greatly reduced .
Drivers frequently think that their displacements are essential, so when weighing the desire to move and the risk assumed the latter is underestimated; the first impulse is to cross the river and not to turn around. Sometimes the urgency is certain, but in the vast majority of cases this reckless behaviour is not justified.
The characteristics of the road also seem to influence the accident rate. Gissing et al. (2019) indicate as usual characteristics associated with accidents in Australia the absence of roadside barricades and lighting, dipping road grades, lack of curb and guttering and difficulties to turn around.
Another problem is that social communication about flood risk is insufficient (Nakamura & Llasat, 2017). The media only pay attention to floods when they have already occurred, and especially when there are fatalities or major damages, but there is no real social awareness of this risk; even public agencies give permits for buildings or infrastructures in flood prone areas. Many fatalities during floods are related to motor vehicles (Boissier, 2013;Diakakis & Deligiannakis, 2015;Jonkman & Vrijling, 2008;Petrucci et al., 2019;Salvati et al., 2018;Sharif et al., 2010;Smith et al., 2019;Vinet, 2017), but there is no enough social awareness about that. Pereira et al. (2017) indicate for Greece and Portugal that although fatalities inside buildings have been reduced, vehicle-related deaths have been rising, and our results for Spain also found an increase in the number of accidents.
It is economically non-viable to build bridges or big drainages in all watercourse crossings. For example, in E, S and SE Spain, the regions with more accidents, there are a lot of watercourses with riverbeds that can exceed 100 m width, where water flows only a few days a year; a bridge over them is a large-scale work, reasonable on main roads, but not on minor roads. Consequently, the presence of dangerous crossing points is unavoidable.
It is desirable that the authorities signalise dangerous crossings, and close them during intense rains, but in practice, it is impossible to do throughout the country; it can be done on major roads but not on all local and dirt roads. Signalling is only effective if it is placed just during the flooding, which is impossible at all risk points in the event of a flash flood. In addition, signalling does not guarantee that drivers do not cross flooded sections; several of the recorded accidents occurred on roads closed to traffic. Similarly, most people drowned in US after hurricane Floyd had received severe weather warnings (Yale et al., 2003).
Environmental education through traditional knowledge transfer, reinforcement of the belief in the risk of climate change or influencing on the perception of risk does not seem to be effective to achieve adaptive behaviours (Choonm, Ong, & Tan, 2018;Grothmann & Reusswig, 2006;Li, Juhsáz-Horváth, Harrison, Pintér, & Rounsevell, 2017); even flood victims do not perceive climate change as a personal and direct risk (Whitmarsh, 2008). Current campaigns are ineffective and mainly reactive, so it is necessary to change the methods, including new information that, if relevant and persuasive, can change behaviours (Bamberg, Ajzen, & Schmidt, 2003).

| RECOMMENDATIONS
Intense rainfall episodes may produce watercourses overflowing; crossing by car in these circumstances, even on a road, is extremely dangerous. During these rainfall events it is important the response of the authorities, but also the behaviour of the drivers; however, many drivers, most mature men, try to cross-overflowing rivers and flooded sections. These behaviours may be due to underestimation of risks, overconfidence, overvaluation of driving skills, prepotency, an excess of trust in the authorities, ignorance about vehicle drag and buoyancy risks and social pressures.
It is necessary to raise awareness among the authorities and citizens, establish adequate protocols for action and, above all, a behavioural adaptation of the drivers. Some possible actions are: -Improve road safety redesigning river crossings and installing warning signs (Diakakis & Deligiannakis, 2015), reinforcing the role of the provincial institutions (Sánchez, Escribano, & Tejada, 2018), increasing funding to reduce the vulnerability of the road network against hazards (Voumard et al., 2018) and foreseen diversion routes in the most critical points (Jenelius, 2009;Voumard et al., 2018).
-Promote citizens' awareness about hazards, especially about flood risk, to encourage a self-protective behaviour (Grothmann & Reusswig, 2006;Nakamura & Llasat, 2017;Sharif et al., 2010), with special attention to the most prone regions. It is possible to implement advertising campaigns, similar to those carried out to raise awareness about traffic accidents or dangers of drinkdriving.
-Reinforce warning campaigns to the population during rainfall events that can lead to flooding or to river overflowing, using media and social networks.
-Include hazard perception elements in the driving theory test (Sexton, 2010), especially about the danger of crossing overflowing rivers; drivers must internalise that they cannot drive through flooded areas (Jonkman & Kelman, 2005). Driving courses should emphasise on safety skills more than on driving skills, to avoid creating a false sense of self-confidence (Sümer, Özkan, & Lajunen, 2006).
-Education, starting from schoolchildren, should disassociate driving task from masculine roles (Albentosa et al., 2018); it is not necessary to take risks driving for being a man.
-Promote information to foreign residents and visitors about the risks associated with flooding, with the collaboration of embassies, town halls, travel agencies and car rental agencies.
In conclusion, it is necessary a greater effort of the authorities to better manage risks associated with river overflow, which will probably increase in the future, but also a change in drivers' behaviour, which should be promoted by all the authorities, not only traffic but also educational.

DATA AVAILABILITY STATEMENT
Data sharing is not applicable to this article as no new data were created or analyzed in this study.