3.2. Fatality Rate at the National and Regional Level
An overview of the fatality rate shows that, at the “national” level, F ranges from 0.314 (Greece) to 0.975 (Southern France). The mean F of the entire period and all the regions is 0.49 (Table 2
). This means that, between 1980 and 2018, in the Mediterranean Basin, an average of one person per million died every two years because of flooding. F is roughly decreasing from the western towards the eastern part of the Mediterranean Sea, as the rate is close to 1 in France and under 0.5 in Greece or Turkey. However, this apparent opposition hides huge local contrasts.
The study area gathers 153 NUTS 3 administrative units, 128 of which exhibited flood fatalities during the study period. The F ranges from 0 to 16.5 in Bayburt (Turkey). The median value of F = 0.27, and this raises to F = 0.41 if we exclude F = 0. The map in Figure 5
shows higher values of F in the areas of the western Mediterranean Basin (Southern France, Catalonia, and Balearic Islands) included in the analysed database. In Southern France and Catalonia, F is rather homogeneous (standard deviation = 1.02) and high (mean F = 1.14). In Greece, standard deviation of F at NUTS 3 level is 0.66 and the mean is F = 0.36, which illustrates a wider dispersion of F.
In Greece, the national F rate is rather low, but F values are very contrasted among NUTS 3 regions. Western Attica (Greece) is the region with the highest fatality rate (4.22), while most of the inland regions are not deadly zones. We assume that, in general, Greece is less hit by deadly storms than southern France, for example (lower frequency of huge rainfall intensity). However, at the local level, urbanization or bad land use planning can explain contrasts between Attica and other Greek regions [17
]. We might consider those results with caution because the size of administrative NUTS 3 units varies between countries. For instance, the "West Attica" unit (Greece) measures 1005 km2
and suffered from the floods of November 2017 in Mandra that caused 24 fatalities. Its population is 164,086 inhabitants, unlike the French NUTS 3 populations range from 291,000 to 1,827,719 for a 5387 km2
3.5. Spatial Discrepancies in Seasonality
The west–east decrease of F at the regional level is coupled with a seasonal evolution. A different monthly distribution of fatalities is observed between the western and eastern Mediterranean Basin (Figure 11
, Table 3
). In the western basin, deaths are concentrated in September and October. September is the deadliest period in Southern France and Calabria, and October in the Balearic Islands. In Catalonia, the most fatal month is November. In Greece, the mortality peak is also in November, but a significant number of flood fatalities are also observed in winter, the season during which flood fatalities in Catalonia are very rare. These results are consistent with the preliminary study shown about floods’ distribution in the Mediterranean [47
The monthly distribution of fatalities follows the frequency of huge events in the Mediterranean Basin [47
]. The Mediterranean Sea acts as a source of heat and moisture that, in conjunction with the surrounding complex topography, plays an important role in the intensification of precipitation events [48
]. Subsequently, the steep small catchments in the area favor the rapid concentration times, resulting in runoff that can rapidly induce devastating floods [49
]. The western part of the Mediterranean is affected by storms and cyclonic activity earlier in autumn (September and October), while the storminess and cyclonic activity over the eastern Mediterranean increases later on in November and in winter [50
]. In Southern France, data confirm that intense rainfall events are more frequent in September–October, which is consistent with monthly mortality (http://pluiesextremes.meteo.fr/france-metropole/IMG/sipex_pdf/saison_mois_plus.pdf
). A fine-grained analysis could show differences within the regions themselves. Catalonia is mainly affected by flash floods produced by local convective activity that is developed on the sea or forced by littoral and prelittoral mountains, where short and torrential streams are born. As more than 60% of the population lives on the coast, and this percentage increases in summer due to tourism, local impacts are strong, producing some fatalities usually owing to imprudent behaviors [30
]. Usually, summer events are produced by local and heavy precipitation that, with some exceptions, gives place to one or two fatalities. On the contrary, floods that trigger in autumn and winter are associated with organised long-lasting convective events or long periods of stratiform precipitation forced by synoptic conditions, in which embedded convection can be developed. Then, it can produce catastrophic and more extended events with a greater number of fatalities. During autumn, the warmer Mediterranean Sea favours instability that can be organised by a mesoscale low in the Catalan-Balear Sea. Then, Mesoscale Convective Systems can affect a more extended region, with total precipitation above 200 mm/24 h and more probabilities of having deaths. In some occasions, this kind of systems arrives to France (i.e., September 1992 and November 1999) [52
Regarding the Balearic Islands, the maximum yearly precipitation occurs in fall, with slight differences between the Islands owing to its geographical characteristics. Those rainfall events are related with the cyclogenetic activity in the western Mediterranean, which is common in autumn. Amounts can easily reach 100 mm/24 h and even 200 mm or more, and some rainy days account for more than half of the monthly precipitation total. Moreover, fall is the season with the largest number of floods events and flood-related fatalities as a result of those rainfall totals.
Greece is characterised by a Mediterranean type climate. Papagiannaki et al. [53
] show that autumn, and particularly November, shows the highest frequency of damaging rainfall events in 2005–2014 in Attica. During spring and summer, rainfall amounts are small over the major part of the country, with the exception of the mountainous areas of western and northern Greece, where thunderstorms are frequent [54
]. Especially during summer, the eastern part of the country as well as the Aegean Sea are influenced by strong and dry northern winds, named etesians [55
]. During autumn and winter, rainfall amounts are quite high over the western part of continental Greece and over the western part of Crete island, where they may reach or exceed 2000 mm/year. During the same period, yearly accumulations up to 400–600 mm are observed on average over the eastern part of continental Greece, while the islands of the Aegean Sea are much drier. Flash floods over Greece are associated with heavy precipitation produced by intense and sometimes explosive cyclone activity [56
], but also by less intense cyclones with relatively long-lasting embedded mesoscale convective systems that interact with the complex topography of the area [58
In Turkey, the monthly distribution of fatalities is scattered between the spring, summer, and fall seasons. To identify the different profile of mortality among the Turkish regions, we isolated the Mediterranean part of the country from the rest. On the Turkish Mediterranean coast, the month with the highest number of deaths is November (Figure 11
, Figure 12
and Figure 13
), which is consistent with what happens in Greece. In eastern Turkey, the maximum number of flood fatalities is in July. Deaths are the result of summer storms that can produce flash floods and are related to the diurnal cycle of convective activity. In eastern Turkey, 20% of annual precipitation falls in summer, while this percentage is only 5% on the western coast [59
]. The thunderstorm activity, tornadoes, severe hail, and even lightning related fatalities all indicate a peak in late summer [50
]. This distribution is consistent with the flood events distribution (including flash floods and surface water floods) [25
This west–eastern gradient is corroborated by the density of fatal events (we call a “fatal event” a flash flood event with at least one fatality). The 1809 deaths in the MEFF v2.0 DB correspond to 455 events, which means that, on average, each of the fatal events caused four victims. However, the density of fatal events differs from one region to another (Table 4
). There is again a clear west–eastern gradient, with densities over 1 in the Occidental Mediterranean Basin (Catalonia, Balearic Islands, Southern France) and less than 1 in the eastern part (Calabria, Greece, Turkey).
At the scale of the six study areas, the correlation between the density of fatal events and the population density is strong (R2
= 0.72) (Figure 14
). This is a logical conclusion because the higher the population density, the more population is exposed. Conversely, the average number of deaths per event is not related to the population density (R2
= 0.066) (Figure 15
). Just because a region is densely populated does not mean that the average number of deaths per event will be high. It means that other variables such as prevention, disaster preparedness, or flood warning influence the average death toll of flood events.
The same analysis was done in the mesh of NUTS 3 and grid, but the results are not relevant (correlation coefficient close to 0 regardless of the type of relationship)
3.6. Time Evolution of Mortality
The annual number of flood fatalities in the MEFF v2.0 DB area shows significant temporal variations (Figure 16
). The 1980s were less deadly in almost all of the countries (30 deaths annually on average over the study area). The year 1988 inaugurates a difficult period, with numerous catastrophic events, with on average 64 deaths per year between 1988 and 1998. In France, this high level of flood-related mortality increased flood-awareness, and culminated with the setup of the Plan de Prévention des Risques, in 1995, after the particularly dramatic floods in 1992–1994 [63
]. Then, since the early 2000s, mortality has remained rather low (on average 40 deaths/year).
The moving average of F at a five-year time-step (Figure 17
) reflects the trends identified with row data, that is, the rise of mortality, particularly in the 1990s (F = 0.7). Since the last twenty years, there has been a slight decline in the fatality rate (below 0.5 since 2000). However, during the last period 2015–2018, F has risen to 0.51 because of severe events in Southern France, Balearic Islands, Greece, Calabria, and Turkey. During this four-year period, we count 222 flood fatalities. This means that, despite that the trend of loss of life is generally decreasing, the fight against flood risk is not won yet.
At the country level, the mortality shows a decreasing trend in the western Mediterranean Sea, in particular in Catalonia, where the fatality rate decreased from 0.63 to 0.26 (Figure 18
and Figure 19
). Catalonia underwent deadly events in the 1980s and 1990s (1982, 1987, 1988, 1994, and so on), but since 2000, it seems to be less affected. Although very local convective events seem to increase in part of this region, they are usually not associated with deaths [30
]. Then, this trend could be associated with a hazard change, but also with the improvement of flood prevention and management measures [47
]. However, the case of Southern France seems very unstable, owing to recurrent fatal events (1992, 2002, 2010, 2015, 2018).
Conversely, flood-related mortality remains high in the eastern Mediterranean Basin, especially in Greece (1990, 1994, 2007, 2017, and so on), where the fatality rate has been constantly increasing since 1995 (F(1995) = 0.19 and F(2018) = 0.92), while the population is rather stable over the period (Figure 19
3.7. Relationship with Extreme Rainfall
Eventually, another key question is the relationship of mortality with potential drivers of mortality. Does mortality reflect the geography of intense rainfall? Alternatively, are human factors such as density of population, vulnerability, or prevention measures more decisive? That is a key question that the present work has not addressed. Nevertheless, one of the first explanatory factors that should be examined is the frequency and intensity of severe rainfall events. The western–eastern gradient of F in the Mediterranean (Table 2
, Figure 6
) could be linked to a decrease in the frequency and intensity of severe rainfall between the western and eastern parts of the basin. Precipitation data provide evidence for this relationship (Figure 20
). The thresholds of intense precipitation are higher in the western part of the basin: Catalonia, Southern France, and Liguria. A map of extreme daily precipitation has been produced with the EOBS dataset version 19.0e [64
]. The EOBS dataset is a gridded interpolation of observed daily precipitation data and covers the period 1950 to 2018 with a resolution of 25 km. Only the grid points with at least 20 years of complete data were considered for the analysis. The annual maximum daily precipitation was extracted for each grid point and fitted with a general extreme value (GEV) distribution to compute the 20-year return period [65
]. The precipitation map corresponding to a 20-year return period is plotted in Figure 20
. The areas with the highest precipitation rates are located on the Ligurian coast in northern Italy, Slovenia, the Cevennes mountainous range in Southern France, and the Catalonian coast of northeast Spain. However, it must be noted that the EOBS dataset suffers from an uneven spatial coverage of the underlying meteorological stations used for the interpolation, resulting in spatial inhomogeneity of the estimated precipitation, as noted by Hofstra et al. [66
]. Consequently, the results for North African countries, the Balkans, and Eastern Europe must be interpreted with care, as the station density in these areas is low. Nevertheless, the highest frequency of intense daily precipitation in Southern France could be an explanation of the high fatality rate in this region.
Other researches show that the specific peak discharges evaluated after huge floods are higher in western Mediterranean regions, especially on the coast between Catalonia and Liguria (Italy) [67