A GIS-Based Assessment of Flood Hazard through Track Records over the 1886–2022 Period in Greece

Abstract

This paper addresses the riverine flood events that have occurred in Greece over the last 136 years (i.e., during the 1886–2022 period), focusing, amongst others, on the case of urban floods. The flood record of various sites of the country has been collected and analyzed to determine their spatial and temporal distribution. Greece is a country where flood data and records are very scarce. Therefore, as there is not an integrated catalog of Greek floods spanning from the 19th century to recently, this is the first attempt to create an integrated catalog for Greece. The sources used include published papers, local and regional newspapers and public bodies (mainly the Ministry of Environment and Energy and the official websites of Greek municipalities). Additionally, the main factors responsible for their occurrence have been issued, regarding the country’s climatic, geological and geomorphological setting, as well as human interventions. In addition, the atmospheric circulation driving factors of floods are assessed via an unsupervised neural network approach (i.e., Self-Organizing Maps). Based on the results of this research, an online GIS-based database has been created, depicting the areas that have been struck by riverine floods in Greece. By clicking a flood event in the online database, one can view several characteristics, depending on data availability, such as duration and height of the rainfall that caused them and number of fatalities. Long-term trends of mean and extremes seasonal precipitation also linked to the spatial distribution of floods. Our analysis shows that urban floods are a very large portion of the overall flood record, and they mainly occur in the two large urban centers, Athens and Thessaloniki, as well as near large rivers such as Pineios. Autumn months and mainly November are the periods with higher flood hazards, based on past records and cyclonic atmospheric circulation constitutes the principal driving factor. Our results indicate that a flood catalog at national level is of fundamental importance, as it can provide valuable statistical insights regarding seasonality, spatial distribution of floods, etc., while it can also be used by stakeholders and researchers for flood management and flood risk analysis and modelling.

1. Introduction

A flood refers to the inundation of areas that are normally dry. Several flood types can be distinguished, such as riverine floods, coastal floods and floods that are owed to the rise of underground water [1,2,3]. Riverine floods, which are the subject of the present study, refer to the water flows that exceed the natural or artificial walls, levees, etc., of a channel and the consequent inundation of an area.

Floods constitute a very important natural process that plays a significant role in the hydrology of a drainage basin. Yet, as far as human activity is concerned, floods are considered as a natural disaster. In fact, floods are among the most frequent and costliest natural disasters, as well as being the most fatal ones, especially in Mediterranean countries [4,5,6,7,8,9]. They pose a very significant economic menace to local and regional communities [10,11]. The main reasons for this are the presence of many and small catchments due to the relief, as well as the rain events that often take place, which concern rapid and intense rainfalls within a short period of time. Additionally, a significant proportion of people reside within or in the vicinity of areas that are vulnerable to floods (e.g., channels or floodplains), as these regions are often very fertile, offering both cultivable areas and the necessary quantity of water [12].

Regarding Greece, its susceptibility to floods is higher compared to most European countries, which is owed to its relief (the presence of many small and mountainous catchments and, on the other hand, major floodplains), its Mediterranean climate (with an abundance in extreme weather phenomena), its lithological structure (presence of multiple impermeable lithologies), its limited forest cover, as well as the multiple human interventions [13]. Greece is among the European countries with the most flood events, as well as the most severe damages [14,15], which have shown an increasing trend over the last decades [16]. Flood data and records are very scarce, even today [16]. It is important to note that most Greek floods are flash floods [17]. And despite the flood protection measures and improvements to infrastructure, floods are becoming more and more frequent in Greece [18].

Floods are primarily caused by extreme precipitation events that are largely determined by atmospheric circulation patterns, leading to heavy rainfall. Their occurrence also depends on the state of the soil environment, the type of land cover and the topography. An area’s susceptibility to floods, besides meteorological conditions and phenomena, is monitored by the local geomorphological and physiographical regime, as well as the geological structure and land cover and management [2,14,19,20,21,22,23].

The amount and intensity of precipitation, especially in the context of extreme events, are extensively examined [24,25]. The IPCC report directly associates changes in extreme precipitation with the likelihood and severity of pluvial floods, which occur when the intensity of the rainfall surpasses the drainage capabilities of both natural and man-made systems [26]. According to the review of Madsen et al. [27], the Mann–Kendall test and Sen’s slope estimator have been widely applied to identify spatiotemporal precipitation trends in mean and extreme precipitation. Atmospheric circulation patterns are also crucial in understanding the synoptic-scale natural causes of floods [28,29,30,31]. In this context, an unsupervised clustering algorithm is used to associate extreme floods in the United States with atmospheric circulation patterns [32], identifying tropical cyclones as the primary drivers. In Europe, Brönnimann et al. [33] found that variations in both atmospheric circulation and moisture content influenced the multi-decadal variability in flooding. Weather patterns are also analyzed for their potential in medium-to-long-range flood forecasting in the United Kingdom, determining which ones are most closely linked to flooding [34].

Several methods exist for assessing flood hazards, depending on the type and characteristics of the floods, the geomorphological characteristics of the study area, the climate, as well as the data availability [1,14,35,36]. One of the most common, as well as reliable, of these is the geomorphological method, i.e., the study of all the characteristics that affect flood susceptibility, mainly focusing on the geomorphology [37,38]. Researchers also use hydrological and rainfall data when available [1,39]. In the field of developing flood risk management strategies, spatial statistics, and more specifically Hot Spot Analysis, can be valuable tools for understanding and mitigating flood risks [40,41].

Another method involves the study of historical floods that have struck an area. Historical flood data are as old as possible, whereas they are not only qualitive, but quantitative as well (to the extent to which such data can be obtained) [42,43]. This method is a comparative study, through which possible areas of high susceptibility to floods can be mapped and the flood frequency and intensity can also be estimated. A third method is the hydraulic one, that is, the usage of hydrological–hydraulic models with a view to quantitively studying a drainage basin’s hydrological regime (e.g., flood discharges) [44].

Flood frequency analysis has been used in several flood hazard assessment models [45,46,47] because it allows for sites of high susceptibility to floods to be mapped [48,49,50] or, correspondingly, the probability of flood occurrence in specific areas to be calculated based on the statistical analysis of historical floods [46,51,52,53]. Flood records can supplement other datasets (for example, palaeoflood records and instrumental measures), both as events and regarding their socio-economic impacts [54]. It is thus very important to estimate the frequency in which an area is struck by flood events to assess its susceptibility to the phenomenon.

Several track records of flood events have been conducted in various regions globally. Gaume et al. [55], for example, created a dataset for flash floods that struck Europe between 1946 and 2007. They also used several criteria for defining which of the recorded floods would be catalogued as flash floods, in cases when the data available were scarce, which concerned the rainfall characteristics and the duration of the event. They obtained their data based on rainfall and discharge measurements, scientific publications, technical reports and local studies from several European countries.

Hall et al. [56] also created a database on European floods for the previous 40–50 years. For the raw data, they mainly used hydrological data (mean discharges, peak discharges, etc.) as primary sources. In cases where such data were unavailable, they used water levels. Gourley et al. [57] created a database for floods in the United States, using USGS streamflow observations, reports by the National Weather Service and testimonies from locals, which were obtained through questionnaires. Hilker et al. [58] also created a track record of flood and landslide events in Switzerland, where the phenomena started being systematically recorded since 1972. Their database extends into the 1972–2007 period and is mainly based on newspapers, also using official websites as an additional source. All of the above cases are examples of flood record databases, which are very important sources of multiple kinds of data, contributing to a significant extent to our understanding of the phenomenon [59].

Zêzere et al. [60] created a database for hydrogeomorphic hazards (including floods) in Portugal for the 1865–2010 period. They collected their data based on newspapers, both local and regional. Santos et al. [61] created a record of floods that occurred in Northern Portugal during the 1871–2011 period. Their record was based on data from Zêzere et al. [60], as well as supplementary data from newspapers.

Corella et al. [62] created a dataset of floods for northern Spain, based on different sources, each covering a different period. More specifically, they used hydrological measures and data from the last 20 years, palaeoflood analysis data from the last 1400 years, and recorded floods from the last 700 years. They were able to study the spatial and temporal trend of Spanish floods and noted the difficulty in comparing the data from different sources.

Barriendos et al. [63] conducted a statistical analysis in France and Spain for the floods of the 1600–1950 and 1300–1980 periods, respectively. They used old documents; data from municipal, governmental and parish/monastery archives; as well as private documents (e.g., from noble families, memoirs, etc.).

Regarding Greece, the only attempt to record past floods at a national scale was the study of Diakakis et al. [18]. They used flood records from official governmental agents (the Civil Protection, the Earthquake Rehabilitation Center, etc.), digital and in-print newspapers and scientific articles. They recorded a total of 545 flood events for the period of 1881–2010, with which they constructed a database. The events they have mapped concerned cases where the rivers or streams overflowed their beds. For an event to be included in the database, they set minimum damages of infrastructure or agriculture, injuries or fatalities as a criterium.

Our work is a track record of riverine floods, including major river floods, flash floods and urban floods, in the Greek territory during the last 136 years (1886 to 2022). The flood record has the form of an online geodatabase and is also depicted in this paper through mapping. The results have been analyzed to determine the floods’ spatial and temporal distribution and principal flood drivers are assessed. No flood catalogs exist in Greece containing events from the 19th century to recently, except for that of Diakakis et al. [18]. The latter, however, only contains the primary events, where impacts to the anthropogenic environment were reported (casualties, injuries or damages). Many events that occurred in areas where no infrastructure or cultivations could be affected have thus not been included. Therefore, this is a first attempt to include all flood events that could be recorded in a single database, regardless of damages. Our results indicate that a catalog of floods at the national level can prove very useful in their statistical analysis, their spatiotemporal character and their tendency over the last decades.

2. Regional Setting

Greece is a relatively small European country (with a surface of 131,957 km²), located on the southern tip of the Balkan Peninsula and is surrounded by the Mediterranean Sea. Geologically, it is situated at the collision zone between the European and African tectonic plates and is therefore highly structurally complex [64,65]. It is thus composed of various lithologies, including carbonates (limestones, dolomites, dolomitic limestones, brecciated limestones, etc.), clastic sediments (silts, clays, sands, gravels, etc.) and sedimentary rocks (shales, sandstones, conglomerates, breccia, etc.), volcanic rocks (lavas, tuffs, etc.) and plutonic rocks (granites, peridotites, etc.), which are in many sites metamorphosed. Additionally, they have various ages, depending on the area, ranging from the Paleozoic to the Quaternary. Therefore, Greece is lithologically very complex. As a result, each area has a different susceptibility to floods where geology is concerned.

What is more, due to its geological configuration in relationship to the tectonic plates of Africa and Eurasia [66,67], Greece is characterized by a manifold relief. The tectonic activity is intense in the territory, resulting in significant relief differentiations within small distances [66,68,69,70]. More specifically, Greece is characterized by a significant number of plain areas, which are interrupted by massifs. The mountainous areas are typically characterized by steep to very steep slopes and well-developed drainage systems.

One typical feature of the Greek landscape is the frequent shift between mountainous and thigh inclined areas and plane areas, often taking place in relatively short distances. Therefore, the morphology of Greece is twofold (Figure 1), regarding flood events. On the one hand, it includes many areas which are flat, either being floodplains of major rivers or deltaic plains of major rivers. These flat areas are usually moderately to densely populated. These areas are typical sites where riverine inundation owing to major rivers occur. On the other hand, the other part of Greece, including the islands, hosts many small mountainous catchments, which are the regions where flash floods mainly occur, usually after short and intense rainfalls [7,14,71,72,73].

According to Köppen and Geiger’s classification [74,75], the climate of the country is characterized as Mediterranean, with dry and warm summers and mild and wet winters. According to their classification and rainfall data from the Hellenic Meteorological Service [76], rainfalls mainly occur during the wet season, that is, between October and March. Intense and short storms and rainfalls are a typical part of the Greek weather [77]. Microclimate differentiations do exist in the country from area to area, even in small distances, according to the geographical setting and the relief [14].

3. Materials and Methods

In order to create a track record of riverine floods, including major river floods, flash floods and urban floods, in Greece, a database has been developed, composed of the sites that have been struck by flood events during the 1886–2022 period. This database contains general geographical information about each site, such as geographical coordinates, regional unit (formerly regional unit), municipality, etc. Given that multiple events could be described as “floods” in historic documents, our database contains any events in which the corresponding rivers, streams or torrents overflowed their beds and covered areas that would normally be dry, regardless of reported damages.

Multiple sources have been utilized to complete our database, such as published papers, public bodies, research programs, etc. A significant portion of the data presented in this paper has been retrieved from the Ministry of Environment and Energy [78] and the National Observatory of Athens [79]. Additionally, a significant number of the records have been made through the official websites of several Greek municipalities. The information acquired was confirmed through publications in scientific articles. The main papers utilized were [16,18,55,80,81,82].

Newspaper articles have also been used as a primary source for the recording of several events. Given that, in the case of floods, as well as any other natural phenomenon that can act at a local scale, regional or state newspapers may not contain a full record of the events, for example in a rural or uninhabited area, local newspapers have also been utilized. The most common of the newspapers (in print and/or online articles) utilized include I Efimerida, Armlife, Press-GR, Xalazi, Ecozen, Eleftheria Online, In-Gr, Naftemporiki, Parapolitika, Proto Thema, Protagon, Real News, Ta Nea, To Vima, Voreia and Rizospastis. And the local newspapers used (both in print and online) include Aegean News, Hania News, Eviaportal, Lesvospost, Neakriti, Thraki and Halkidiki News. In any case, events whose source of information was the newspapers were cross-checked with other sources (ministry, municipality archives or scientific papers).

Upon collection of the flood data, the geographical distribution of the floods was analyzed. The geographical analysis was conducted using the GIS software ArcGIS Pro by ESRI. It was also used to create the database, by digitizing (when necessary) and importing each event into a single database, as well as for the creation of the maps presented. Consequently, the online platform (ArcGIS Online) was utilized to develop an online map, which is freely accessible at the following link: https://evelpidou.maps.arcgis.com/apps/instant/sidebar/index.html?appid=56c91b63ed974c62a9833c81d365f261 (published 15 September 2023). It contains the database created in the framework of this paper and is open to the public. The database contains the flood events and the corresponding prefecture. Additionally, depending on the sources and the available data, some events also contain the duration and height of the rainfall that caused them, as well as the number of deaths they caused. Finally, the database contains the data source(s) for each and every flood event individually.

The flood records were used for spatial analysis using the Getis-Ord Gi* statistic tool which, in the environment of ArcGIS Pro, results in a Hot Spot spatial aggregation analysis. In the Getis-Ord Gi*, the statistical results of z-scores (standard deviations) and p-values (probability) indicate where geographic clustering of characteristics with high or low values occurs. The Getis-Ord Gi* operates by examining each characteristic in light of its surrounding features. A flood cluster feature has a high value and is surrounded by additional flood cluster features that have high values and consequently provides a statistically significant hot spot. A statistically significant z-score is produced when the local sum for a feature and its neighbors deviates significantly from the predicted local sum and deviates by an amount that is too great to be the product of random chance.

The Getis-Ord Gi* calculus is given as:

$$G_i^{*}=\frac{{\sum }_{j=1}^n{w}_{i,j}{x}_j-\overline{X}{\sum }_{j=1}^n{w}_{i,j}}{S\sqrt{\frac{\left[n{\sum }_{j=1}^n{w}_{i,j}^2-{({\sum }_{j=1}^n{w}_{i,j}^{})}^2\right]}{n-1}}}$$

(1)

where x_j is the attribute value for the j (floods), $w_{i,j}$ is the spatial weight between features i and j, n is equal to the total number of features (floods) and

$$\overline{\mathrm{X}}=\frac{{\sum }_{j=1}^n{x}_j}{\mathrm{n}}$$

(2)

$$S=\sqrt{\frac{{\sum }_{j=1}^n{x}_j}{\mathrm{n}}-{\left(\overline{X}\right)}^2}$$

(3)

The G_i^* static is a z-score, and no further calculations were required.

The analysis of the driving factors of flood events includes initially the analysis of annual and seasonal precipitation trends for the 1960–2020 period using precipitation data from the ERA5 global atmospheric reanalysis product of the European Center for Medium-Range Weather Forecasts (ECMWF) [83]. The hourly precipitation data were first aggregated to derive daily totals. These daily totals were then further accumulated to produce monthly figures. Finally, these monthly values were combined to calculate both seasonal and annual precipitation totals. Given that climatic data form the post-1960 period are very scarce and inaccurate, at least in Greece, no correlation between post-1960 flood events and corresponding climatic data was made.

The annual and seasonal trend analysis is performed using the Mann–Kendall Test and Sen’s slope estimator. The Mann–Kendall test is a non-parametric test used to detect monotonic trends and in our analysis the trends are assessed at the 99% confidence level in the daily precipitation time series. The 99% high confidence level is selected, as in our case, a high degree of certainty is required to reject the null hypothesis. The Mann–Kendall test and Sen’s slope estimator are commonly used together in environmental and hydrological studies [84,85]. The Mann–Kendall test is first applied to determine if a trend exists. If a trend is detected, Sen’s slope estimator is then used to quantify the rate of this trend. For a given time series the Mann–Kendall test statistic S is computed based on the pairwise comparisons of all data points. The test determines if the value of the test statistic is significantly different from what would be expected under the null hypothesis of no trend. The S sign provides the direction of the trend (positive for upward and negative for downward) [86,87]. The Sen’s Slope Estimator calculates all possible slopes between pairs of data points in the time series and the median of all slopes is the Sen’s slope. A positive slope indicates an increasing trend, while a negative slope indicates a decreasing trend [88]. The Sen’s slope estimator was chosen because of its robustness to outliers.

In terms of improving the understanding of flood hazards, the impact of large-scale atmospheric circulation is examined. To this end, an environment-to-circulation approach is applied using the Self-Organizing Map (SOM) neural network. The approach is applied in various contexts in the climate science [89,90]. The SOM is two-layer neural network that uses an unsupervised learning algorithm with the scope to reduce the complexity of multi-dimensional data into a two-dimensional map, making it easier to identify patterns and relationships. The SOM in this work is used as a cluster analysis tool for a comprehensive representation of the atmospheric circulation types on a topological map. An SOM consists of a grid of nodes (neurons), typically organized in a two-dimensional map. During training, the network undergoes iterative adjustments to its weights to better match the input data’s characteristics. The map “self-organizes” in a way that reveals the clusters in the data. The applied framework includes the following steps: (a) definition of the spatial and temporal scales; (b) standardization of the spatiotemporal time series and data reduction using principal components analysis; (c) classification and assignment of cases into atmospheric circulation patterns using the SOM algorithm; and (d) visualization of atmospheric circulation patterns and assembly of the catalog of the patterns that are associated with the flood events. The domain of application is Europe for all flood events included in the database from 1960 to 2020, and the method is applied for the daily fields of the mean sea level pressure (MSLP), extracted from the ERA5 reanalysis database. The optimal number of atmospheric patterns that are associated with floods is determined using the Davies–Bouldin index, a metric for evaluating clustering algorithms [91]. The optimum configuration implies that clusters are more compact (i.e., members of a cluster are closer to each other) and well-separated from other clusters.

Figure 2 shows a flow chart of our methodology.

4. Results

A total of 1994 flood events have been recorded during the 1886–2022 period in the present study. The developed GIS-based database is also accessible via this link: https://evelpidou.maps.arcgis.com/apps/instant/sidebar/index.html?appid=56c91b63ed974c62a9833c81d365f261 (published 15 September 2023). This section addresses their spatial and temporal distribution, based on statistical analysis.

4.1. Seasonality of Flood Events

The seasonal analysis of flood frequency (Figure 3) reveals that the majority of floods are observed during the autumn (51.5%) and winter (27.3%) months. In contrast, only 11.1% and 10.1% of flood events occurred in spring and summer, respectively. With regard to monthly frequencies, the highest number of floods is observed from October to January. November has the highest occurrence, with 412 floods (20.7% of the total records), followed by October, with 381 events (19.1%). The fewest events are recorded in April, with only 35 floods (1.8% of the total floods) (Figure 4).

Regarding the yearly distribution of flood events (Figure 5a), it is notable that, prior to 1960, there were fewer than ten floods per year. This could be due to the incomplete documentation of older flood events. For the 20th century, where records are more reliable, there is not a consistent trend, as many fluctuations are evident. The period from 2000 to approximately 2015 experienced the highest number of flood events per year. In the past 7 years, the number of severe flood events has decreased, largely due to the implementation of protective measures. It is important to note that flood events in the 19th century are characterized by a significant level of uncertainty concerning flood intensity, damages, etc. Additionally, flood records at the beginning of the 20th century are sparse due to wars and conflicts.

Focusing on the analysis of the 10-year running mean (i.e., decade-long moving average), we detect and analyze the long-term flood frequency trends. The analysis provides the critical periods that long-term trends are identified along with the sign of the increase/decrease. According to Figure 5, two common subperiods are defined (1980–2007 and 2008–2022) as the critical breakpoint, and the change in trends is observed in 2007 for both yearly (Figure 5a) and seasonal time series (Figure 5b–e). In detail, increasing trends are observed from 1980 to 2007, while decreasing trends are recorded from 2008 to 2022.

The trend analysis for flood events is conducted for the two subperiods using the Mann–Kendall test and Sen’s slope estimator. This test determines if there is a statistically significant increasing or decreasing trend in flood frequencies for the annual and seasonal time series data. If a significant trend is identified, Sen’s slope estimator is then employed to quantify the magnitude of these trends. The analysis results (

Table 1. Sen slope coefficients for annual and seasonal flood events during 1980–2007 and 2008–2022. Trends identified by the Mann–Kendall test are significant at the 99% confidence level.

Frequency	1980–2007	2008–2022
Annual	2.89	−4.17
Winter	0.55	−1.50
Spring	0.25	−0.17
Summer	0.25	0.00
Autumn	1.25	−1.67

) show statistically significant increasing trends from 1980 to 2007. In contrast, the period from 2008 to 2022 exhibits statistically significant decreasing trends. There is a gradual increase in annual flood events (2.89 events/year) from 1980 to 2007, while a more pronounced decrease is observed from 2008 to 2022 (4.17 events/year). This trend pattern is consistent during the winter and autumn periods, which, as previously mentioned, experience more flood events.

4.2. Geographical Distribution of Flood Events

The spatial distribution of the recorded floods is presented in Figure 6. Each region has been symbolized based on the total number of flood records during the 1886–2022 period (Figure 7). Five categories have been defined, namely a total of 0 to 10, 11 to 30, 31 to 80, 81 to 100 and 101 to 240 flood events (respectively, 0 to 0.07, 0.08 to 0.22, 0.23 to 0.44, 0.45 to 0.59, 0.60 to 0.74 and 0.74 to 1.76 flood events per year).

Table 2. Number and frequency of flood events in the 13 regions of Greece.

Region	Number of Floods	Annual Flood Frequency
West Macedonia	24	0.18
Northern Aegean	57	0.42
Ionian islands	64	0.47
Southern Aegean	68	0.50
West Greece	99	0.73
Peloponnese	123	0.90
Crete	117	0.86
Epirus	127	0.93
Central Greece	202	1.49
Attica	210	1.54
Thessaly	223	1.64
East Macedonia and Thrace	267	1.96
Central Macedonia	413	3.04
Total	1994	14,66

shows the number of flood events per Greek region for the 1886–2022 period, as well as the mean annual number of floods. The area of Macedonia seems to have been struck by the largest number of flood events, since it is a very large area. Central Macedonia is the region with the highest number of floods over the 1886–2022 period (413 events). The Eastern Macedonia and Thrace region shows the second largest number, measuring 267 flood events. These are followed by Thessaly, Attica and Central Greece (223, 210 and 202 events, respectively). It is noteworthy that West Macedonia only measures 24 recorded events. This is owed to the highly mountainous relief of the region and the absence of significant plain areas.

The prefectures with the highest number of floods over the 1886–2022 period are Attica, Thessaloniki, Achaia, Messinia, Serres, Evros and the prefectures of Thessaly, Crete and the Dodecanese. More specifically, the prefectures where the highest number of flood events occurred during the studied period include Attica, Thessaloniki, Phthiotis and Evros. Of these, the Attica and Thessaloniki prefectures have been struck by the highest number of floods (211 and 122, respectively), predominantly urban ones, due to high urbanization. Phtiotis has been struck by more floods than Thessaloniki (129), mainly due to the presence of the Spercheios river. The Magnesia, Serres and Karditsa prefectures follow, having high flood frequency.

Τhe regional units of Attica, Phthiotis and Thessaloniki have been struck by the highest numbers of floods (210, 129 and 122, respectively), predominantly urban ones, due to high urbanization. The Phtiotis regional unit has been struck by more floods than Thessaloniki, mainly due to the presence of the Spercheios river and its floodplain and deltaic plain (Figure 8). Figure 9 shows the regional units with the most flood events.

4.3. The Most Severe Floods in Greece

One of the most destructive flood events was that of 4 June 1907 in Trikala and was owed to the Litheos river, crossing the city [92]. The event lasted for 7 h. It was estimated to have caused more than 100 fatalities, as well as extensive damages. Most of the fatalities mentioned in the previous subsection are owed to this single event.

Another severe flood was that of Athens on 6 November 1961, which resulted in 44 fatalities and more than 200 injuries. It followed an intense 10 h duration rainfall and was caused by the Kifisos river. Another severe event was on 20–21 October 1994 on Rhodes Island. It was caused by several rivers flowing in its eastern part. In the same time, Podonifits stream in Athens inundated the city, resulting in many damages [93]. A total of 17 fatalities resulted from these two events.

Another severe event was that of 14–15 November 2017 in Mandra, Attica, owed to the torrents of St. Catherine and Soures [94,95]. It was the result of a 150–160 mm rainfall event of 7 h duration. Crucial was the role of the settlement’s location, on the two streams with no planning standards. Finally, a flash flood struck central Euboea after the occurrence of a prolonged and intense storm between 5 and 9 August 2020. On 9 August, the flood occurred [96].

4.4. Climatic Drivers of Floods

In Figure 10, the mean trends of precipitation at the yearly and seasonal scales are presented for the 1960–2020 period. With respect to the yearly analysis, a clear statistically significant decrease is observed for the western part of Greece, while mixed trends are reported for the eastern part (e.g., positive trends for Attica). An important finding regarding the seasonal analysis is that for the autumn period (Figure 10e), where most of the flood events are reported, the spatial distribution of trends exhibits a clear shift towards positive values, mainly in northern Greece, the northern Aegean and in eastern-central Greece.

Focusing on extreme precipitation, the 95th percentiles are calculated on a seasonal basis from the daily precipitation records, and a trend analysis is performed for the 1981–2020 period. Days with precipitation levels exceeding the 95th percentile are categorized as days with heavy rainfall. The results are presented in Figure 11 and highlight (in yellow) the high-risk areas that are prone to flood hazards due to meteorological phenomena on a seasonal basis. In detail, special focus should be given to urban regions within these areas such as cities located in western and northern Greece. During summer, it should be noted that most mainland parts of Greece exhibit increasing trends. Extreme precipitation events can be attributed partially to the warmer conditions that lead to an increase in the ability of the atmosphere to hold more moisture.

The analysis of the atmospheric circulation drivers of the recorded flood events from 1960 to 2020 is presented in Figure 12 for the MSLP variable. The optimum configuration of the SOM map was selected using the Davies–Bouldin index, a metric that is used to evaluate clustering algorithms [91]. In our case, the selected SOM is a 4 × 3 topology, as presented in Figure 13. The seasonal analysis of the relative frequencies of occurrence for each type reveals a clear seasonal pattern of the SOM map. The wintertime patterns are organized at the bottom row of the map (types 2, 3 and 4), the single summer type in the upper corner (type 9) and the remaining types are primarily observed during the autumn. The critical factor in terms of atmospheric circulation is the exact location of the depressions with respect to Greece. In more detail, for types 3, 4, 7 and 8, the low pressure is located in the north of Greece, while the low pressure for the 1, 2, 5, 6 and 10 types is in the east, and in combination with an anticyclonic circulation, a strong southern flow is observed in the region.

5. Discussion

Greece is one of Europe’s nations most frequently struck by flood events, either flash floods or typical riverine ones. The reason for this is its lithological variety, its climate (frequent high-intensity and short-duration rainfalls and storms) and its relief (significant relief and altitude differentiations within short distances), as well as high urbanization [14].

Research in Greece regarding floods and their causes and metrics is incomplete, among other reasons, due to the lack of hydrological measures. Additionally, mountainous catchments, which are responsible for a significant number of floods in Greece [97], are scarcely monitored [14,98]. This, in association with the high urbanization rates, could be the main reason for the high number of fatalities in Attica compared to most other regional units.

Regarding the spatial distribution of floods in Greece, it is worth mentioning that the central mountain range, which runs in a NNW-SSE direction, separates continental Greece into two zones. Western Greece is generally characterized by a greater annual rain height than eastern Greece [82]. As a result, eastern Greece is characterized by more frequent extreme weather phenomena, which justifies the fact that, in our database, eastern Greece has generally been struck by more floods than western Greece, thus confirming that flash floods mainly occur after intense and short rainfall/storm events rather than prolonged rainfalls of low intensity [7,14,71,72]. Annual precipitation has been reduced in Greece as a result of the climate crisis [95], but this has led to an increase in the intensity of rainfall events [99,100,101,102]. This is also established from the results of our analysis according to the findings of the mean and 95th percentile precipitation trends.

Regarding the geographical distribution, we observe that many regional units that are drained by major rivers have been struck by a significant number of floods. Regarding the geographical distribution, we observe that many prefectures that are drained by major rivers have been struck by a significant number of floods, such as Larissa (Pineios river), Serres (Strymonas river), Evros (Evros river), etc.; this is in accordance with a previous study [18]. The regional unit of Phthiotis is drained by the Spercheios river. These rivers have been known to cause frequent floods, many of which cause several damages to infrastructure [103,104,105,106,107]. Additionally, several regional units that are mountainous, have been struck by a significant number of floods. This shows that in Greece, floods have two aspects, riverine floods, owing to major river inundations, and flash floods, occurring in small mountainous catchments, and both are frequent.

Regarding the seasonal distribution of floods, many studies have shown that the autumn months are those with the highest number of events [14,18,108,109]. This is owed to the fact that during that period, rainfall events have started, saturating the soil and the geological formations, whereas the infiltrated water has not yet been absorbed by the plants or has not yet infiltrated further, respectively. Consequently, in the following rainfalls, less water has the capacity to infiltrate the soil and surface runoff is increased.

A typical example is the flash flood in Herakleion (Crete) on 13 January 1994. The flood was owed to a storm that hit the drainage basin of the Giofyros stream. However, in the previous days, rainfall of light intensity had preceded, which caused a partial saturation of the soil. This contributed to the intensification of the flood event, as the water infiltration was hindered [110].

A critical factor, especially during winter and autumn, is the atmospheric circulation and specifically the location of the low-pressure centers with respect to Greece. A conclusion that can be drawn from this is that autumn and winter months seem to be the ones when most flood events occur, with November bearing the most floods compared to the other months.

5.1. Urbanization and Land Use Changes and Flood Frequency and Impacts

Over the recent years, a significant proliferation of urbanization and touristic development has characterized the Greek territory, as well as many other countries globally. Many flood incidents have been recorded since the 19th century in Greece. Older events are usually scarcely mentioned in the literature and/or with inaccuracies and incomplete data. Nowadays, several flood protection measures have been applied in various regions of Greece. Therefore, a decrease in the impacts of floods has been observed in low-land areas [14]. However, an increase in the number of urban flood events can be observed in the 21th century compared to the previous centuries, despite the application of protective measures [18,82,111]. This often results in multiple losses and severe infrastructure damages [14]. This can be deduced by our data, as, since 1994, there has been a very significant increase in the number of floods. Of course, this could be just a result of poor and limited flood recording prior to this period. In any case, our data show a decreasing trend in the number of floods after 2010 compared to the 1994–2010 period.

This is owed to the constantly increasing urbanization. Multiple human interventions increase the flood frequency and impacts, such as constructions on or near channel beds, barriers, forest fires and deforestations, as well as inefficient flood protection measures [112]. Urbanization and the consequent structures cause alterations in the course of water, thus leading to frequent flood issues. A typical characteristic of urbanization is the habitation and constructions in areas that are prone to floods, thus resulting to a further augmentation of the flood risk [14].

In our database, we can observe that, especially after 1900, most floods have occurred in the urban fabric, and particularly where construction does not follow specific standards (the most typical locations being Athens and Thessaloniki). Based on the spatial distribution of the recorded floods, we have created a confidence map, showing the possibility for a flood to occur, based on the number of recorded floods in each area (Figure 14). It is worth mentioning that the highest confidence (≥90%) is found around urban centers, indicating the high risk of these areas. It is noticeable that not all large centers show such a high confidence. In several cases, flood protection measures, when existent, are not based on an actual environmental or hazard analysis study [113]. Poor maintenance and development planning is another fact that renders the large urban centers more prone-flood [113].

Urbanization has had a great impact on the environment and the evolution of natural processes [114]. Athens poses the best example of rapid and intense urbanization in Greece. Back in 1892, only 3% of the Athenian drainage basin was urbanized [115]. Nowadays, almost the entire plain of Athens has become part of the city. The rest of Attica has also been almost entirely urbanized, bearing many towns and suburbs [116]. Attica hosts approximately 40% of the Greek population. The drainage network of Athens is quite well-developed. Until the late 19th century, urban development was minimal and the natural environment was in harmony with the human environment. At the end of the 19th century, the torrents and streams of Attica started being entrenched for the construction of both sewer and road networks [117]. More specifically, initially, streams and torrents were used as open sewer systems. The rapid and intense urbanization of Athens did not allow for independent sewer networks to be constructed. Subsequently, they were used for the construction of an underground sewer system. The surface of the sewer systems, i.e., the former streams, were used for the construction of major roads. After 1934, when the above activities were conducted, there is an outbreak, starting in 1961 with 13 flood events. In 1945, streams in Athens had a total length of 1280 km, which reaches 434 km nowadays, which is more than a 66% reduction. All of the above are consistent with our findings: up until 1934, only one flood event per year is recorded for some years.

Regarding land use changes, Pineios is a very typical example. The floodplain of the Pineios river (Thessaly) has been cultivated since the Neolithic period [118,119]. At the same time, pasturing has also been practiced since ancient times [120]. Settlements started developing at least during the Pre-Classical period [121]. Up until the beginning of the 20th century, cultivations were mostly limited within the floodplain, in flat regions. In the early 1920s, the local population was increased due to the arrival of Greek immigrants from Asia Minor. Cultivations started taking place in low hills, leading to the limitation of the natural vegetation [122]. It is worth noting that the population of the city of Larissa, located in the center of the Pineios floodplain, has shown more than a 400% increase in its population since 1940; from 1940 to 1960, it was 71%; up to 1980, the increase was 121%; up to 2000, it was 203%; and by 2011, it was 350% [123]. This population trend has resulted in multiple land use changes in the area. Besides the intensification of agriculture [122], multiple interventions have taken place, including the dredging of stream beds and diversions. According to our analysis (Figure 15), except for two peaks in 1987 and 1994, the total number of floods shows a more or less stable tendency before 1000 and a tendency to increase after 2000, which is consistent with the rise in population and cultivation activities and confirms that increased human interventions were responsible for a significant number of floods.

5.2. Major Factors Causing Flood Events in Greece

By combining the above results and observations, we can enumerate the major factors causing and/or affecting the floods in Greece (at state scale). First, the Mediterranean climate is characterized by mild winters [74,75], with few rainfalls, but many individual extreme events (extreme precipitation, storms and, more recently, Mediterranean hurricanes). Rainfall is generally concentrated in the period from October to March [77], which explains the fact that the highest numbers and intensities of flood events are found between October and January [13,16,18]. Of course, this is not the case for areas at high altitudes. Bard et al. [124] recorded flood events in the Alps for the 1961–2005 period and found that a significant number of these events are owed to snowmelt and thus occur predominantly in spring months. Moreover, annual precipitation has been reduced in Greece as a result of the climate crisis [95], but this has led to an increase in the intensity of rainfall events [99,100,101,102].

The general relief of Greece is another driving factor. As mentioned above, Greece is made up of multiple mountainous catchments of small size and high inclinations, which are the most susceptible to floods, especially flash floods [125,126]. At the same time, it contains many plains of large rivers (such as Pineios, Strymonas and Evros). These rivers drain mountainous areas with very well-developed drainage networks. Often, the shift from mountainous to plain areas is sudden. As a result, these large rivers often inundate their floodplains. It is also worth mentioning that the mountain range of Pindos, which runs in a NNW-SSE direction, separates continental Greece into two zones. Western Greece is generally characterized by a greater annual rain height than eastern [82]. As a result, eastern Greece is characterized by more frequent extreme weather phenomena, which justifies the fact that in our database, eastern Greece has generally been struck by more floods than western Greece, thus confirming that flash floods mainly occur after intense and short rainfall/storm events rather than prolonged rainfalls of low intensity [7,14,71,72].

Another driving factor is the increased urbanization and land use change. This is evident, among others, by the fact that during the last few decades, floods have been increased in the major Greek urban centers. Multiple floods were partly caused by human interventions in the riverbeds, one of the most typical being that of Mandra (Attica) in 2017 [94].

5.3. Advantages of a Flood Catalog/Database at National Level

While several advances have been made regarding forecasting and early warning, damages from floods and especially flash floods seem to have increased in the last decades [127]. Flood datasets can play a very significant role in understanding, as well as predicting, future flood phenomena [57]. Floods, and especially flash floods, are difficult to monitor and record because they do not follow a specific pattern as to where, when or how they occur, but are instead characterized by spatiotemporal variety, among other reasons [57,128]. Because of this, in combination with their rapid and unannounced occurrence, they are very scarcely monitored and recorded compared to other weather-related phenomena [129,130]. In addition, the equipment necessary for collecting hydrological measurements and data is characterized by a high cost to obtain and maintain. Such equipment is thus only rarely used in such areas. Therefore, the available data from mountainous and small catchments regarding flood events are in most cases only fragmentary [55,131]. For these reasons, detailed catalogs of flood events are usually incomplete [57], as they do not cover all the individual cases. Modern tools, however, such as remote sensing, airborne data, etc., have rendered the gathering of flood data less demanding [132,133,134,135,136].

Based on the above, one can conclude that a complete catalog of flood events (flash floods, as well as typical riverine floods) at a national level can be very helpful in decision management and flood management, as well as flood risk modelling (cf. [16,80,137,138,139,140,141]). What is more, a statistical analysis of the catalog, like the one conducted in the present study, can give multiple pieces of information, such as seasonality and spatial distribution of flood events. Past flood records can provide valuable insights on the floods’ behavior in different areas/periods and, thus, contribute to the improvement of existing knowledge and technology for risk management [113]. Extreme flood events can also be reconstructed when using historic flood data [63].

Another important issue is that people often do not conceive of the potential impacts of a flood event. Several studies have shown that people often underestimate a flood event, or are unaware of its potential consequences, which results in increased fatalities [4,14,127,142,143,144,145,146]. This is also the case in Greece. Papagiannaki et al. [147] conducted a survey after an extreme flood event in Attica in October 2015. According to their results, preparedness was very limited, even after the event. Diakakis et al. [148] studied Atticans’ perceptions of floods and found that they showed limited knowledge on floods, as well as limited preparedness.

And this is not only the case of floods. Despite the frequency of such natural and social disasters, their severity and the damage and losses they cause, education on disasters at primary and secondary schools is unfortunately not given its proper value. This was confirmed by a survey for 2000 educators and students in Greece in 2020 [149,150]. According to the survey, 95% of the educators answered that they consider it possible for a rapid onset natural disaster to occur while they are at school. They also deem that building evacuation exercises do require effort (66%) and the recurrence of the procedure is necessary in order to be effective when needed (97%). Another 75% of educators stated that they know how to react if a rapid onset event takes place, and 57% think that they know everyone’s duty and where every object they may need is located if a natural disaster suddenly occurs. On the other hand, 70% of educators stated that they do not feel safe and secure in such a case. The majority (97%) desire an edification program to be included in the school courses and consider that an organized annual edification program, i.e., that takes place in all schools in the country, would be useful. More than half of the educators do not know how to offer first aid in case of a potential injury, while 35% are ignorant of the means of communication in order to offer first aid. Regarding students, the feeling of safety and security reaches 67%. In total, 87% of the students wish to participate in activities and games aiming to educate them on rapid onset natural disasters. These results were obtained by students and educators in Greece, but other European countries would arguably show similar results as well.

A direct result of this is that stakeholders in Greece (and other countries) often do not take the necessary measures and/or do not follow the appropriate practices in order to mitigate flood hazards. In a survey in Heracleion, Crete, Anogeiannaki [151] questioned the local stakeholders’ perceptions on extreme floods and the effects of climate change. She found out that many stakeholders do not intend to take the necessary protective measures, despite previous flood events. In a more general study, Papakonstantinou [152] found that there was, on the other hand, limited trust towards the state regarding protection from natural disasters. Diakakis et al. [148] confirm this lack of trust. It is interesting to mention that stakeholders in Athens consider earthquakes as the more severe and most important natural disaster in Athens, followed by forest fires and place floods in the third position [148]. As a contradiction to this belief, it is worth noting that floods are much more important, not only in terms of numbers, but in terms of damages and fatalities as well [153,154,155].

It is thus evident that, in any case, an integrated catalog of flood events at a national level could arguably contribute to raising people’s awareness regarding the severity of the phenomenon, as well as other characteristics (mainly spatial and temporal distribution, as well as relationship to urbanization). Actions need to be taken, besides those aiming to increase protection against floods and reduction of potential impacts, in order to increase people’s awareness and knowledge on the factors causing floods and mitigation strategies [149,150]. These actions need to be taken for all citizens and address the importance of prevention as opposed to rehabilitation [148].

All in all, our research outlines the significance of a record on flood events at national level. While there exist several catalogs at European scale, they cannot be used at the national level. For example, Gaume et al. [55] have listed a significant number of European floods. They also contain a small number of Greek floods. However, their time span only covers the period from 1946 and up to 2007. Additionally, they address only floods whose magnitude exceeded a specific threshold. Hall et al. [56] also created a catalog of European floods, but only for the past 40–50 years. Their database does not extend to our period of study. As a result, such studies provide integrated catalogs for the case of Europe, but are inadequate for the scale of Greece (or, respectively, another state). In order to construct a full catalog of European floods, the best method would be to create smaller databases at a national level for many individual states, which should then be combined. The present study is one such case for an individual state (Greece).

5.4. Major Limitations in the Current Study

In this section, it is important to note the primary limitations that occurred in our study, which are generally present in any attempt to record historic events based on historic sources.

Initially, in the Greek region, the start of the recording of flood events is very recent. Thus, a systematic record only exists for the past 30 years or so [18]. There are, however, records in official reports and the Greek press that can provide (and have provided in this study) an adequate flood record for previous periods. In such cases, where instrumental data are very scarce, the analysis of historic floods can be of high importance, even when studying the floods’ qualitative characteristics [156,157].

A major problem is the fact that, still, our understanding of flood tendency in small mountainous catchments is limited [158]. Such catchments are the ones where flash floods predominantly occur [7,14,71,73]. On the other hand, they occur in a very small time frame, which, in combination with their remoteness and inaccessibility, prevents their monitoring [129,130]. Thus, in most cases, and especially in order documents, a flood event might be “overseen” unless it affects an inhabited or cultivated area [18]. Given the scarce monitoring, we assume that many flood events that occurred in uninhabited and uncultivated catchments were never recorded and are thus not included in our database.

Another issue is that each flood event differs from any other in terms of flood type, causes, geomorphic characteristics, etc. This poses a challenge in comparing flood events in totally different regions and extracting statistical trends [124]. What is more, many floods have occurred in basins that have been affected by human interventions. And in most cases, it is difficult to quantify the effects of human activities on flood events compared to natural characteristics [124].

Another issue is the fact that our analysis concerns data that were collected through different sources, for different regions, from various periods, of different flood types and with different degrees of errors and accuracy. The comparison of heterogenous data is always a challenge [62,159].

According to Barriendos et al. [63], old documents can provide information regarding hydrological and climatic parameters for a flood event, but need to be taken into account in a hierarchy. They also mention that, in such cases, criteria need to be applied in order to ensure the best data quality and accuracy possible. They mention the importance of cross-checking a flood event’s characteristics using different sources (e.g., documents) in order to determine its geographical distribution.

Previous authors [160,161,162,163,164] have set a number of criteria when taking old documents into account. More specifically, a document needs to be public, so that it can be analyzed at a reasonable timetable. It needs to publish flood events regularly, to ensure that no events have been omitted. Also, the source must be reliable, to ensure that no misinterpretations are made of the flood characteristics. The author(s) need to be contemporary with the corresponding event, and the document itself needs to be the original one, without modifications. Finally, the information needs to be objective and non-partial. Barriendos et al. [63] additionally mention that the date and circumstances of writing must also be considered.

6. Conclusions

A total of 1994 floods have been recorded in Greece over the last 136 years (1886 to 2022). They have been statistically analyzed, and they have been depicted in corresponding maps, as regards both their temporal (seasonal, annual, etc.) and geographical distributions. From our study, the following conclusions have been drawn:

• There were more records during the 21st century than in previous centuries, due to scarce recording and measuring, which was partly owed to the conflicts and wars.
• Taking this into account, the flood number seems to have increased from approximately 1886 to 2010, possibly as a result of climate change (smaller number of rainfalls, but more intense and frequent extreme rainfall/storm events). Between 2000 and 2015, the flood record reached its peak, and between 2015 and 2022, a slight decrease was observed.
• Regarding rural regional units, the eastern part of the Greek mainland and the Aegean islands show a higher flood frequency compared to the western mainland and the Ionian islands, even though western Greece has a higher annual precipitation. This has been attributed to the important role of extreme events, which seem to be more frequent in eastern Greece, generating flash floods.
• Both mountainous areas and regional units drained by major rivers were found to be inundated relatively frequently, which indicates that, in Greece, both flash floods in mountainous catchments and typical floods owed to major rivers are important, and both can potentially cause severe damages.
• Urban floods make up a very large portion of the overall flood record, and they mainly occur in the two large urban centers, Athens and Thessaloniki, due to the continuously increasing urbanization and human interventions, in association with ineffective or inexistent flood protection measures.
• Most floods occurred during the autumn and winter months and predominantly between October and January. November has the most flood events, followed by October. During the spring and summer months, floods are less common, with April bearing the smallest number of events.
• High-resolution analysis of the driving factors of atmospheric phenomena can contribute to the development of early warning systems for civil protection. For instance, a critical factor, especially during winter and autumn, is the location of the low-pressure centers with respect to Greece.
• The mean trends of precipitation at the yearly and seasonal scales for the 1960–2020 period demonstrate a clear statistically significant decrease in the western part of Greece. Mixed trends are recognized for the eastern part (e.g., positive trends for Attica). For the autumn period, when most of the flood events are reported, the spatial distribution of trends exhibit a clear shift towards positive values mainly in northern Greece, the northern Aegean and eastern-central Greece.
• The trend analysis for the 1981–2020 period of the 95th percentiles—as calculated on a seasonal basis from the daily precipitation records—was performed, showing that the majority of high-risk areas for floods are located in western and central Greece.

Our study emphasizes the importance of the creation of an integrated flood catalog, addressing both flash floods and typical riverine floods, as well as both floods of high and lower intensity and/or spatial distribution, at the national level. Flood hazard analysis can be more accurate and effective if flood frequency and historic flood events are taken into account. And while a catalog at a continental or international level may prove very effective, it cannot be utilized for individual cases at a state level. The creation of such a database can prove very useful in statistical analyses regarding floods’ spatiotemporal distributions, connections with urbanization and/or land use changes, as well as their tendency in the last decades.