See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/317263996

A national scale flood hazard mapping

methodology: The case of Greece – Protection
and adaptation policy...

Article in Science of The Total Environment · May 2017

DOI: 10.1016/j.scitotenv.2017.05.197

CITATIONS READS

0 260

2 authors:

Nektarios Kourgialas George P Karatzas

N.AG.RE.F. - NATIONAL AGRICULTURAL RESE… Technical University of Crete
55 PUBLICATIONS 322 CITATIONS 213 PUBLICATIONS 1,191 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

1. “Groundwater Flow Simulation – Seawater Intrusion Inhibition”, 6th Framework Programme 2000-
06, “ΕΝΤΕΡ 2001”. View project

Multiobjective Optimization of Cosolvent Flushing of NAPLS from Contaminated Groundwater View

project

All content following this page was uploaded by Nektarios Kourgialas on 02 October 2017.

The user has requested enhancement of the downloaded file.

 1

7 A national scale flood hazard mapping methodology: The

8 case of Greece – protection and adaptation policy
9 approaches
10 Nektarios N. Kourgialas(a)*, George P. Karatzas(b)
11 (a) Hellenic Agricultural Organization (H.A.O.-DEMETER), National Agricultural Research Foundation
12 (N.AG.RE.F.) - Institute for Olive Tree, Subtropical Crops and Viticulture (Water Recourses, Irrigation &
13 Environmental Geoinformatics Lab.), Agrokipio, 73100 Chania, Greece.
14 (b) School of Environmental Engineering - Technical University of Crete, Polytechneioupolis, 73100 Chania,
15 Greece.
16 *corresponding author: kourgialas@nagref-cha.gr; nektarios.kourgialas@enveng.tuc.gr
17
18 Abstract
19 The present work introduces a national scale flood hazard assessment methodology, using multi-
20 criteria analysis and artificial neural networks (ANNs) techniques in a GIS environment. The
21 proposed methodology was applied in Greece, where flash floods are a relatively frequent
22 phenomenon and it has become more intense over the last decades, causing significant damages in
23 rural and urban sectors. In order the most prone flooding areas to be identified, seven factor-maps
24 (that are directly related to flood generation) were combined in a GIS environment. These factor-
25 maps are: a) the Flow accumulation (F), b) the Land use (L), c) the Altitude (A), b) the Slope (S), e)
26 the soil Erodibility (E), f) the Rainfall intensity (R), and g) the available water Capacity (C). The name
1
 1 to the proposed method is “FLASERC”. The flood hazard for each one of these factors is classified
2 into five categories: Very Low, Low, Moderate, High, and Very High. The above factors are combined
3 and processed using the appropriate ANN algorithm tool. For the ANN training process spatial
4 distribution of historical flooded points in Greece within the five different flood hazard categories
5 of the aforementioned seven factor-maps were combined. In this way, the overall flood hazard map
6 for Greece was determined. The final results are verified using additional historical flood events that
7 have occurred in Greece over the last 100 years. In addition, an overview of flood protection
8 measures and adaptation policy approaches were proposed for agricultural and urban areas located
9 at very high flood hazard areas.
10
11 Keywords: artificial neural networks, GIS, agricultural and urban flash floods, flood hazard,
12 mitigation measures
13
14 Highlights
15  A new integrated national scale flood hazard mapping method based on GIS and ANN techniques
16  This study provides, for the first time, a flood hazard map for entire Greece
17  The methodology was validated based on historical flood events
18  24 % of the total area of Greece is under very high flood hazard (50-year return period)
19  An overview of flood protection and adaptation strategies for agricultural and urban sectors

20
21 Introduction
22 Floods are considered to be among the most frequent and destructive types of natural disasters
23 worldwide, with significant consequences including: a) human and animal life losses, b) destructions
24 of infrastructures, communication networks, and agricultural/livestock buildings, c) loss of crops
25 and soils, d) transport of sediment loads and pollutants (Downton, 2001; Golian, 2010). Worldwide
26 floods, especially during the last decades, record significant high rates both in absolute number of
27 events, and in terms of financial losses. Specifically, it is estimated that floods cause about 40% of
28 the damages caused by all natural disasters (Ologunorisa and Abawua, 2005; Munich Re, 2016).
29 The different types of flooding are: river floods, flash floods, coastal flooding, urban floods,
30 and sewer flooding (Kourgialas and Karatzas, 2016). Flash flood in the Mediterranean areas is the
31 most common type of flood which occurred due to the small size of the river basins (short-time flow

2
 1 concentration), the geomorphology (high slopes or/and low permeability geological formations),
2 and the intense rainfalls (Diakakis et al., 2012). These characteristics make flash flood a common
3 phenomenon not only to low-lying or adjacent to rivers areas, but also to mountainous plateaus.
4 Flash floods are characterized as one of the most disastrous hazards in terms of mortality and
5 infrastructures (Georgakakos, 2006). Recently, characteristic examples in Mediterranean countries
6 where flash floods caused significant economical damages and/or human losses can be seen in Italy
7 (Moramarco et al., 2005; Molinari et al. 2014), Greece (Kourgialas and Karatzas, 2014a;
8 Papagiannaki et al., 2015), France (Gaume et al., 2004; Delrieu et al., 2005), and Spain (Llasat et al.,
9 2013).
10 Flash floods are strongly influenced by meteorological conditions and different land uses,
11 thus flash flood events, in the upcoming decades, are going to be highly influenced due to climate
12 and land use changes (Posthumus et al., 2008; Muis et al., 2015; Kourgialas et al., 2015). Roudier et
13 al., 2016 investigate the hydrological impacts of a +2°C global warming on extreme hydrological
14 events (floods and droughts) in Europe. The results of this research show that flood magnitudes are
15 expected to increase notably in areas south of 60oN, except in the areas of Poland, Bulgaria, and
16 south of Spain, while north of 60oN floods will decrease, with the exception of some coastal areas
17 in Norway and southern Sweden where floods probably will be increased. An integrated analysis of
18 floods along with droughts shows that the impact of global warming will be more extreme in Spain,
19 Portugal, Greece, France, and Albania. Specifically, flooding events will be expected to be more
20 frequent in recent years especially in the Mediterranean region. Thus, all the necessary measures
21 to avoid or minimize the consequences of the oncoming floods should be taken by the
22 Mediterranean countries.
23 Many countries especially in Southeast Asia and Africa face a severe increase in flood risk (Di
24 Baldassarre et al., 2010a; Winsemius et al., 2016). Flood risk maps are generated using multi-
25 parametric approaches that combine physical (hazard) and socio-economic (vulnerability) factors.
26 Nevertheless, existing global flood risk projections fail to accurate predict the dynamics of socio-
27 economic development or/and climate change (Ward et al., 2015; Winsemius et al., 2016). The
28 knowledge of flood hazard is also essential to adapt any strategy for minimizing the flood risk, which
29 in turn can reduce the losses of human life and damages in urban and rural sectors (Pappenberger
30 et al., 2013; Sampson et al., 2015). Flood hazards can be defined as threatening events, or the
31 probability of occurrence of a potentially damaging phenomena within a given time period and area
32 (Di Baldassarre et al., 2010b). However, in recent years, flood hazard maps are still lacking in many
3
 1 countries. This is mainly due the limited availability of adequate data for flood hazard studies such
2 as hydrologic observations, historical flood events, and topographical surveys of
3 channels/floodplains (Samela et al., 2017). All these data are rarer in ungauged river basins of many
4 developing countries of Asia, Africa, or South America. Based on this, for producing large scale flood
5 hazard maps, several researchers have addressed the prediction of flood hazard at ungauged sites
6 by using regionalization methods, simplified hydrological routing schemes, methods based on
7 geomorphology for generating relationships between flood and floodplain, and using high
8 resolution satellite images for quantifying changes in surface water (Nardi et al., 2006; Dodov et al.,
9 2006; Manfreda et al., 2011; Pekel et al., 2016; Samela et al., 2017).
10 Based on the Floods Directive 2007/60/EC, each member of the European Union should
11 design flood hazard maps using different levels of hazard (probabilities of flooding). This
12 classification would make the information of flood hazard more obvious to the local authorities and
13 easily understandable to the public, providing at the same time valuable spatial information
14 regarding the degree of flood hazard and the priorities regarding the planning of the flood
15 protection measures (Yannopoulos et al., 2015). At the present time, reporting the progress of EU
16 members on national scale flood mapping this information is missing from Bulgaria, Greece, Malta
17 and Portugal. Thus, one of the aims of this study is to cover this knowledge gap providing an easy
18 to be applied methodology for assessing flood hazard at national scale.
19 Hydrological/hydraulic models have been widely used in many flood mapping studies
20 (Chatterjee et al., 2008; Ballesteros et al., 2011; Brocca et al., 2011; Kourgialas and Karatzas, 2014a;
21 Papaioannou et al., 2016). In many cases, these models are inadequate for ungauged or/and
22 national scale areas, such as Greece, where the majority of the country’s territory consists of small
23 ungauged or poorly gauged river basins. Valuable tools to overcome these hydrological/hydraulic
24 modeling limitations are the combination of artificial neural network (ANN) and GIS techniques.
25 ANNs are commonly used data-driven methods that have proved capable of prediction of water
26 resource variables with great accuracy (Besaw et al., 2010). ANNs display mapping effectiveness to
27 generalize reliable data in parallel with high speed. Based on these capabilities, ANNs are designed
28 for pattern recognition and classification applications, which make them very suitable for flood
29 forecasting (Toth et al., 2002; Rietjes and de Vos, 2008; Sulafa, 2014). Their drawbacks are related
30 to their black box approach (no physical meaning of the concerning parameters), and extrapolation
31 capacity (He et al., 2014). The ANN disadvantages can be overcome combining them with GIS
32 techniques. GIS are particularly useful in flood hazard mapping as it can incorporate both the spatial
4
 1 and physical dimension of the floods (Yahaya et al., 2010; Wang et al., 2011; Chau et al., 2013;
2 Kazakis et al., 2016). Up to now, flood mapping using ANNs and GIS have been applied in various
3 small scale case studies (Islam and Sado, 2002; Masoud et al., 2012; Song et al., 2013).
4 The main objective of this work is to propose an integrated and easy to apply ANNs and GIS
5 modeling approach, able to provide reliable large scale flood hazard maps in data scarce regions.
6 This modeling approach combines different factor-maps which capture all the aspects of the
7 hydrological cycle that contribute in flood generation. The proposed methodology shows an
8 advance over existing methods, since: a)it integrates seven different factors, that are directly related
9 to the flood development which are flow accumulation, land use, altitude, slope, soil erodibility,
10 rainfall intensity, and available water capacity - into a unified GIS model that incorporates all
11 information necessary to reduce the level of uncertainty of the flood hazard, b) it consolidates ANNs
12 techniques in order to face an important problem in a GIS multi-criteria analysis. This problem is
13 related to the determination of the weights of the involved factors, which in most cases are subject
14 to the subjective estimation of the decision-maker, c) it is effectively adapted to complex
15 geomorphological large scale environments, such as the country Greece, providing, for the first
16 time, a flood hazard map for the entire country, d) it is validated for different historical flood events
17 took place in Greece the last 100 years, and e) it provides an overview of flood protection measures
18 and policy approaches for agricultural and urban areas located at very high flood hazard areas
19
20 Study area
21 Greece is located in the eastern part of the Mediterranean zone covering an area of about 132.000
22 km2. Greece has the longest coastline in Europe, exceeding 15.000 km along the Mediterranean Sea.
23 The geological formations in Greece mainly consist of karstic (carbonate rocks) and porous (alluvial
24 and/or neogene deposits) formations (Daskalaki and Voudouris, 2008). Greece has 54 prefectures
25 (the country's main administrative unit), and 14 Water Districts (Dokou et al., 2015), which are : 01)
26 West Peloponnese, 02) North Peloponnese, 03) East Peloponnese, 04) West Central Greece, 05)
27 Epirus, 06) Attica, 07) East Central Greece, 08) Thessaly, 09) West Macedonia, 10) Central
28 Macedonia, 11) East Macedonia, 12) Thrace, 13) Crete, and 14) Aegean Islands. All these Water
29 Districts are presented in the map of Figure 1a. An extended part of Greece is mountainous covering
30 about 80% of the country’s area. The population of Greece reaches 11 million inhabitants mainly
31 distributed at coastal regions and on the islands. The population magnitude in the main urban

5
 1 centers is classified in the following order: Athens (Water District of Attica), Thessaloniki (Water
2 District of Central Macedonia), Patras (Water District of North Peloponnese) and Herakleion ( of
3 Crete). A significant part of Greece, mainly lowland areas, is covered by intensively agricultural areas
4 such as tree crops (olive, citrus, peaches, apples, vineyards etc) and annual crops (cereals, maize,
5 cotton, potatoes, tomatoes etc), (Ministry of Agriculture of Greece, 2010). Greece is characterized
6 by small to medium size river basins, with ephemeral streams to be the majority of drainage
7 networks. The special geomorphology of these small river basins, which exhibit intense topography
8 with thin and impenetrable soil, as well as the lack of large river networks makes flash flooding to
9 be the most common type of flood inundation caused by intensive rainfalls. Moreover, river flow
10 hydrological data are scarce since most of the catchments are ungauged or limited monitoring,
11 therefore, the available database of historical flood events is mainly reported than recorded data
12 (Kourgialas et al., 2012).
13 In the present study, based on different sources of information, an extensive database of
14 historical flooding events of the last 100 years was developed (Diakakis et al., 2012; Karymbalis et
15 al., 2012; Diakakis and Deligiannakis, 2013; Bathrellos et al., 2016; Kourgialas and Karatzas, 2016;
16 Ministry of Environmental and Energy of Greece, 2016; Diakakis 2017). Based on this, six hundred
17 (600) flood events, that caused extensive damages in rural and urban sectors, with six hundred
18 nighty (690) human casualties have been reported/recorded across the country (Diakakis and
19 Deligiannakis, 2017). These data show that urban environments tend to have a higher flood
20 recurrence rates than rural areas. Figure1b represents the distribution of floods that occurred
21 across Greece during the last 100 years, expressed as the number of events per 100 km2 in each
22 prefecture (Diakakis et al., 2012).
23

24 Figure 1
25 Methodology
26 In order to determine the flood hazard areas in Greece, five different flood Hazard Levels (HL) were
27 considered (very high, high, moderate, low and very low). Each level describes the probability to
28 have a flood event within a hydrological year as following: 2% probability (50-year return period) -
29 Very high flood hazard level, 1% probability (100-year return period) - High flood hazard level, 0,5%
30 probability (200-year return period) - Moderate flood hazard level, 0,2% probability (500-year
31 return period) - Low flood hazard level, <0,2% probability (> 500-year return period) - Very low flood

6
1 hazard level. The final map of flood hazard was created using the combination of seven (7) individual
2 thematic maps directly related to flooding. The thematic maps were related to: the Flow
3 accumulation (F), b) the Land use (L), c) the altitude (A), b) the Slope (S), e) the soil Erodibility (E), f)
4 the Rainfall intensity (R), and g) the available water Capacity (C).
5 The factors (F), (A), (S), (E), (R), and (C) have numerical values, while the (L) factor is
6 expressed in a descriptive form. The effect of each factor to flood hazard level is mapped based on
7 five different (HL) from very high to very low. For the factors with numerical values the Jenk’s
8 Natural Breaks method was used to indicate the five different flood hazard levels. For the (L) factor
9 the classification depends on the influence of the land use on the flood process. For instance, forest
10 areas with high density land cover indicate a very low flood hazard level, whereas burnt forest areas
11 indicate areas a very high flood hazard level (Martin-Vide et al., 1999). The above mentioned factors
12 as well as the (HL) are shown in Table 1 and they were selected based on literature review and
13 quantification of experts’ opinions such as hydrogeologists, hydro-agronomists, and
14 environmentalist (Eimers et al., 2000; Chenini et al., 2010; Yahaya et al., 2010; Kourgialas and
15 Karatzas, 2011; Kazakis et al., 2015).
16 Each of the aforementioned maps/factors were georeferenced to the EGSA’87 (Greek
17 Coordinate System - EPSG Projection 2100). The produced raster maps have a geometric resolution
18 equal to 20m. Field measurements and GIS techniques were used to determine and digitalize the
19 above seven maps/factors. Specifically, in order to create the Digital Elevation Model (DEM) for
20 Greece topographic maps (1:50 000 and 1:5000) were used. Using DEM, the slope map (degree) was
21 produced applying the 3D Analyst tool in a GIS environment. Generally, slope and altitude are
22 inversely proportional to the appearance of flood events (Kourgialas and Karatzas, 2016), (Table 1).
23 Also, the Arc Hydro - flow accumulation algorithm was used for the creation of the flow
24 accumulation map (ESRI, 2008). Flow accumulation map specifies the number of pixel-cells that
25 hydrologically add to each raster cell. In this map, the lower the accumulation of runoff in a pixel-
26 cell the lower the flood hazard (Table 1). The original data related to the soil water capacity and soil
27 erodibility of Greece were obtained from the European Soil Bureau Network
28 (http://eusoils.jrc.ec.europa.eu/wrb/). Soil erodibility is a crucial factor that affects the flooding
29 flow hydrographs as well as the consequent damages. Specifically, soil erodibility is an average
30 annual value of the soil profile reaction to the processes of soil disconnection and transport by
31 surface flow and raindrop (Panagos et al., 2012; Kourgialas et al., 2016). Decomposition rate
32 depends on soil consistency; the less coherent the soil the more easily it can be carried away by the
7
 1 surface flow which transfers significant amounts of sediment load leading to siltation of water ways.
2 The large sediment yield is a common characteristic of flash floods causing extensive destructions
3 (Jinren and Yingkui, 2003). Moreover, the higher the soil erodibility is, the larger the peak outflow
4 rate is (Chang and Zhang, 2010). Based on the above, soil erodibility is strongly related to the flood
5 hazard and areas with increased values are more prone to flooding. Soil water capacity is associated
6 directly with the ability of soil to absorb water that controls the amount of surface runoff. Thus,
7 high values of soil water capacity can reduce the flood hazard, while soil water capacity values close
8 to zero can lead to an increased surface runoff and a greater flood hazard (Cazemier et al., 2001),
9 (Table 1).
10 In order to create the land use map of Greece remote sensing methods (satellite images)
11 and the Corine land cover database (European Environmental Agency - EEA,
12 http://www.eea.europa.eu) were used. The land use factor is associated with the vegetation cover
13 that controls both the amount of precipitation and the time that it takes to reach the soil surface
14 contributing to surface runoff. Thus, dense vegetation cover with deep and extent root zones can
15 reduce flood and water erosion hazard, while areas without vegetation can lead to an increased
16 surface runoff and have a greater hazard of flooding (Posthumus et al., 2008), (Table 1). For the
17 Rainfall intensity (R) factor monthly rainfall values from 407 rainfall stations were used for the time
18 period 1960-2014, (www.meteorologia.gr). The spatial distribution of these stations is shown in
19 Figure 1c. Based on these data, the Modified Fournier Index was employed to create the map of the
20 Rainfall intensity in Greece (Morgan, 2005):
12
p2
MFI = ∑
21 1 P , (1)

22 where:
12
23 MFI: the Modified Fournier Index, ∑: the 12-month summation, p: the average monthly rainfall,
1

24 and P: the average annual rainfall.

25 MFI indicator indicates the sum of the average monthly rainfall intensity at a station. Next, in a GIS
26 environment, using the spline method, the MFI value of each station was interpolated. Spline is
27 considered to be the most appropriate interpolation method to depict smoothly varying surfaces of
28 phenomena such as rainfall (Goovaerts, 2000; Lloyd, 2005; Kourgialas and Karatzas, 2014b).
29 Monthly rainfall data could be considered as an acceptable mapping approach for rainfall intensity

8
 1 factor, since: a) this factor does not require exact values of rainfall intensity but classifies it into five
2 categories of hazard, b) this approach was justified by a large number of meteorological stations,
3 and c) in Mediterranean ephemeral streams the surface runoff has seasonal character (Camarasa
4 Belmonte and Segura Beltran, 2001).
5
6 ANN technique for the creation of final flood hazard map
7 The above seven factors/maps do not have the same weight of influence on flood generation, thus
8 a correlation approach of all factors was applied to create the final flood hazard map for Greece.
9 This analysis was performed using ANN techniques of the matlab toolbox. ANNs have been
10 confirmed as a particularly classified modeling method for solving non-linear problems such as the
11 classification of flood hazard level (Dandy and Maier, 2000; Kumar et al., 2004). ANN consists of
12 simple processing structures named neurons, which are connected to a system network by a
13 number of weights. Each neuron calculates the network's output by taking a number of inputs,
14 weighting them, summing them up, adding a bias, and using the result as the argument for a singular
15 valued transfer function (Rietjes and de Vos, 2008).
16
17 Input-output variables / Performance evaluation
18 As it was mentioned above the historical flood events used in this study are referred to the last 100
19 years (very high or high flood hazard areas). However, in order for ANN to be trained in areas of
20 moderate to very low flood hazard, a random distribution of 200 theoretical flooded points, located
21 in the moderate, low and very low flood frequency category map of Greece, as described in Figure
22 1b (Diakakis et al., 2012), were considered. Based on the above, an integrated database of 800 flood
23 points was created from 600 historical and 200 theoretical flooded points located in all flood hazard
24 areas. For ANN training purposes, 70% of the above historical and theoretical flooded points
25 distributed in the entire territory of Greece were used, while a dataset of the remaining 30% flooded
26 locations were used for the purpose of testing. To convey the connection between historical-
27 theoretical flooded points and the seven involving factors maps the “Extract by Mask” tool in GIS
28 environment was used. According to this tool, the cells of each of the seven factor maps (raster files)
29 that correspond to the historical-theoretical flooded points (mask data) were extracted. In this way,
30 the extracted points were identified and exported as a database D1 which indicates the coordinates
31 and the five different flood hazard levels, described as numerical values, for each point and factor
32 map. This database D1 is used to train the ANN in matlab environment. The simulation performance
9
 1 of the ANN model was evaluated based on the Root Mean Square Error (RMSE), the coefficient of
2 determination (R2), and the Nash-Sutcliffe model efficiency coefficient (NSC). In a GIS environment,
3 each of the seven factor maps that describe the five levels of flood hazard is converted into a point
4 shapefile (32236 points per factor). Each point contains the information of the coordinates as well
5 as the relevant flood hazard. Subsequently, a second database D2 of the seven point shapefiles is
6 exported in a excel sheet to be used as the input (sample) to the already trained ANN model. A final
7 output database D3, from the 32236 points, was created for the entire territory of Greece. Next,
8 database D3 was introduced in the GIS environment and by choosing the appropriate spatial
9 interpolation method the final flood hazard map of Greece was created (raster format with 100 x
10 100 m resolution). For the spatial interpolation, Inverse Distance Weighting (IDW), Spline, and
11 Kriging methods were tested. The choice of the most reliable spatial interpolation method was
12 based on the statistical metric of the total error. The final flood hazard map was verified based on
13 the spatial distribution of the 600 flood events that have occurred across Greece the last 100 years.
14
15 ANN architecture
16 A feed forward neural network, trained with the Levenberg-Marquardt method (Rumelhart et al.,
17 1994), was used to estimate the final flood hazard map. In addition, one hidden layer was employed.
18 This choice was done based on the literature review according to which for most ANN applications
19 one hidden layer is sufficient and it offers adequate generalization performance and also reduces
20 the extra computational time of a second layer (Kourgialas et al., 2015; Rumelhart et al., 1994). The
21 trial and error method was used to determine the optimal number of neurons, in order the network
22 to have enough degrees of freedom and at the same time to be capable of generalization. (Teschl
23 and Randeu, 2006). Networks with 10-150 neurons were tested. The network with the smallest
24 number of neuron that gave the best combination of errors was chosen as the optimal model. The
25 sum of training, testing and validation errors was used as the performance indicator. The proposed
26 ANN modeling approach is schematically represented in Figure 2.

27 Table 1
28 Figure 2
29

30 Results
10
 1 Table 1 presents the classification of the seven factors at the five hazard levels. These seven maps
2 developed by the classification method are presented in Figures 3a and 3b. The combination of the
3 seven maps (flow accumulation, land use, altitude, slope, soil erodibility, rainfall intensity, and
4 available water capacity) was applied based on the proposed ANN methodology. Using the ANN
5 modeling approach the final map of flood hazard areas in Greece was generated by combining the
6 above mentioned factors.

7 Figures 3a, 3b
8 Regarding the results of the ANN, different networks were developed, trained, validated and
9 compared. Among the networks with a neuron number between 10 and 150 that were tested, the
10 one with the smallest number of neurons that gave the best combination of training, testing and
11 validation errors was the one with the 120 neurons. The simulation performance of the ANN model
12 was evaluated on the basis of the following goodness of fit metrics: the Root Mean Square Error
13 (RMSE), the coefficient of determination (R2), and the Nash-Sutcliffe model efficiency coefficient
14 (NSC). Thus, the following values were obtained: for the training period, RMSE = 0.37, R2 = 0.98, and
15 NSC= 0.97 and for the testing period, RMSE = 0.41, R2 = 0.98, and NSC= 0.97. For the validation
16 period the RMSE index was equal to 0.52, the R2 = 0.96, and the NSC= 0.95. The above statistical
17 metrics indicate a very good performance of the outputs of the ANN modeling approach. These
18 outputs (database D3) describe the spatial distribution of the 32236 points covering the entire
19 territory of Greece where each point represents a corresponding flood hazard level. Also, as
20 mentioned earlier, the database D3 is introduced into GIS and based on the appropriate spatial
21 interpolation method (Inverse Distance Weighting, Spline, and Kriging) the final flood hazard map
22 for Greece was created. For the creation of the final flood hazard map the appropriate interpolation
23 method was verified and selected based on the historical flooding points in Greece (last 100 years).
24 For the total 600 recorded/reported historical flooded points, 510 have a 50-year return period,
25 while the rest of them have a 100-year return period. Specifically, in Figure 4 a,b the locations of
26 the historical flooded points, occurred within a 50-year return period, are marked with black
27 symbols, while historical flooded points occurred within a 100-year return period are marked with
28 blue symbols. According to this figure, the majority of the historical flooded points are located in
29 the western part of Greece. Table 2 shows the number and the percentage error of flooding points
30 that are located at very high and high flood hazard areas of the final map for each of the three
31 interpolation methods. According to this table the Spline interpolation method shows the best

11
 1 results (total error of 2.17%) as almost all the recorded/reported historical floods with a 50-year
2 return period (502 of 510 points, 98.4%) occurred at the very high flood hazard areas with only eight
3 of them to be in the areas of high flood hazard. On the other hand, the majority of
4 recorded/reported historical floods with a 100-year return period (85 of 90 points, 94.4%) were
5 located within the identified areas of high flood hazard, with only five of them to be in the areas of
6 moderate flood hazard. The good agreement between historical flood events and the final flood
7 hazard map appears to be not only at the low and semi-mountainous lands but also at the mountain
8 plateaus areas of Greece.
9 All of the aforementioned information validates the spatial reliability of the proposed
10 methodology for the territory of Greece. According to the verified final flood hazard map (Figure 5),
11 an area of 31040 km2 (24%) of the Greek territory is subject to very high flood hazard and 19686
12 km2 (15%) is under high flood hazard. In addition, 26103 km2 (20%), 27695 km2 (21%) and 25647
13 km2 (20%) can be characterized as areas of moderate, low and very low flood hazard, respectively.
14 In addition, regarding the land uses, 856 km2 (about 37.22%) of the artificial areas of Greece is under
15 very high flood hazard. Regarding the agricultural land, arable area of about 7027 km2 (39.04% of
16 the total arable area in Greece), permanent crops of about 854 km2 (8.99% of the total permanent
17 area in Greece) is under very high flood hazard. Furthermore, pasture areas of about 2377 km2
18 (31.66% of the total pasture area in Greece) are under very high flood hazard level.
19 Based on the spatial distribution of flood hazard in Greece (Figure 5), very high flood hazard
20 areas are equally distributed between the western and eastern part of Greece. Additionally, flood
21 occurrence appears to be more intense in the lowlands and urban areas of Greece. For agricultural
22 intensification areas such as the western part of Peloponnese, the central Macedonia, and Thessaly
23 the final map shows extended areas characterized by very high flood hazard (Figure 5). This can be
24 justified by taking into consideration that in farmlands more runoff is generated and discharged
25 more rapidly due to agricultural intensification (O’Connell et al., 2007; Morris et al., 2010). In
26 addition, many of the above cultivated areas are low land valleys, where the groundwater over-
27 exploitation has often led to a land subsidence, which in fact increases significantly the flood hazard
28 level. Furthermore, very high flood hazard areas are observed at extended urbanization areas such
29 as the Water District of Attica.
30 Table 3 shows the flood hazard areas and the corresponding percentage for each of the 14
31 Water Districts of Greece. According to this table the Water Districts of Attica has the highest
32 percentage of the area (about 33%) which is subject to very high flood hazard, followed by the Water
12
1 District of West Peloponnese with a percentage of very high flood hazard area to be equal to 31%.
2 The Water Districts with the lowest percentages of very high flood hazard appear to be the areas of
3 West (13%) and East Macedonia (11%).
4 In a GIS environment, the above classification results and the flood hazard maps (including
5 map coordinates) can be produced by local authorities. Thus, maps of high flood hazard areas with
6 the relative information could be distributed to the citizens, and farmers, to raise awareness. As
7 mentioned above, Greece is considered as one of the most important cultivation regions in the
8 Mediterranean zone. Based on the final flood hazard map, 3500 km2 (29.3%) of the agricultural
9 areas are located in very high flood hazard regions. Thus, a significant percentage of the agricultural
10 sector in Greece is under very high or high flood hazard. Furthermore, higher than 30% of the
11 artificial land use in Greece is under very high flood hazard.
12 For the current work, the validation of the modeling results with historical flood events
13 strengthens the accuracy of the final results. The limits of the proposed methodology are: (a) the
14 necessity of detailed meteorological, topographical, as well as soil-land use knowledge of the study
15 area, and (b) extended database of flood data for a long historical period. However, the above limits
16 minimize the uncertainty in the prediction of the flood hazard. Moreover, in this study seven factors
17 are used for capturing the information related to the generation of floods in order to reduce the
18 uncertainty and to improve the utility of hazard modeling for decision making. While the uncertainty
19 of flood hazard results, that derives from the historical flood data adaptability, for some input
20 factors such as land use can be significant, for the other six factors is usually low, provided that the
21 map information is accurate and up-to-date.
22

23 Figure 4
24 Figure 5
25 Table 2
26 Table 3
27
28 Discussion

13
 1 Based on the above results, it is very important to highlight flood protection and adaptation
2 approaches for agricultural and urban sectors such that to minimize the consequence of flood
3 damages under different human activities and climate change conditions.
4 An appropriate flood protection plan should start from the mountainous regions (elevations
5 > 800m). In these regions, the kinetic energy of flow is increasing, while the water volume is
6 collected and promoted to the low regions. Thus, technical measures such as inhibitory dams and
7 tanks can be established for the interception of an oncoming flood event. Specifically, at locations
8 where the kinetic energy of the flow is increased the above small hydraulic structures can contribute
9 to the time desynchronization of the water accumulation, as well as to the decrease of the released
10 kinetic energy (Borrows and Bruin, 2006; Kourgialas and Karatzas, 2011). Moreover, in mountainous
11 areas protection measures due to deforestation should be considered, since they can prevent
12 downstream flood phenomena (Evrard et al., 2007).
13 In the intermediate regions (elevations from 800 – 200m) as well as for low regions
14 (elevations < 200m), common practices that lead to interception of floods, include the cleaning of
15 watercourse from deposits and the protection of banks from erosion. Moreover, stone walls could
16 be constructed along the river banks where overbank flows and erosion phenomena have occurred
17 in the past. In general, these regions are mainly suitable for interventions that ensure the natural
18 operation of the flood plain and protect very high flood hazard areas (Förster et al., 2005). Also, in
19 these regions intensive and irresponsible human activities in urban and rural sector tend to magnify
20 the hazard of flood; therefore the above mentioned operations are of high importance. Agricultural
21 land can act as a pathway to flood generation leading to significant damages in the urban and rural
22 sector. In the intermediate and low land areas, many studies highlight the link between agricultural
23 land management and flood generation at farm and sub-catchment scale (<10km2) (O’Connell et al.,
24 2007; Environment Agency, 2008). Specifically, land management practices can have greater impact
25 on the delay of the flood peaks, especially for small and medium events increasing the flood warning
26 times (Salazar et al., 2009). As a result, good agricultural land practices can retain water in the
27 landscape contributing to flood hazard alleviation. Taking into consideration that Greece has many
28 agricultural areas mainly located at the intermediate and low lands this evidence could be very
29 important for an appropriate flood management plan at national scale. Based on this, the present
30 study highlights and identifies best land management practices which enhance the infiltration
31 capacity and the degree of flow connectivity for runoff control. Infiltration capacity can be increased
32 based on practices which reduce soil compaction and improve soil structure, for instance: grazing
14
 1 management, seasonal removal of livestock to avoid poaching of soils, avoidance of field machinery
2 and field operations under wet conditions, soil improvement measures such as conservation tillage,
3 and improvement field drainage system. On the other hand, practices which control runoff by
4 influencing the rate at which water from fields discharges into watercourses (degree of flow
5 connectivity) include contour ploughing, retention ponds, artificial banding, hedgerows, stonewalls,
6 field margins, buffer strips and woodlands (Wheater, et al., 2009; Carroll et al., 2004; Morris et al.,
7 2010).
8 Apart from these pre-flood measures, at river basin scale, it is vital to establish an adequate
9 flood warning system. This system should consist of optimal located telemetric hydro-
10 meteorological stations. The data of these stations will be used as input to a real time hydrological-
11 hydraulic simulation system (Brocca et al., 2011; Kourgialas et al., 2012). Based on this, the local
12 authorities can maximize: a) the potential warning time (time between the beginning of rainfall and
13 the threshold exceedance), b) the response-warning time which includes the recognition of the
14 threat and the decision making approach, and c) the mitigation time which will moderate the flood
15 disasters and it will reduce their effects. To increase the mitigation time, actions such as the removal
16 of destroyed structures from the flooded regions, campaigns and briefing for the public, and flood-
17 preventing to existing buildings could be considered. Many researchers emphasize that the above
18 flood mitigation actions are more beneficial in the urban sector than in the agriculture (Thieken et
19 al., 2007; Morris et al., 2010). Specifically, in high flood hazard agriculture areas adapted strategies
20 could be followed by farmers as a suitable solution for reducing the damages. These adaptations
21 could include strategies related to mitigation flooding at downstream urban regions by making
22 space for flood water at intermediate agricultural areas (rural floodplains) and/or switching
23 economic beneficial crops that are more tolerant to climate change effects and especially to
24 flooding (Blackwell and Maltby, 2006).
25 Many crops, during critical growth periods, are sensitive to anaerobic soil conditions caused
26 by the excess water covering the land during a flood event. Thus, in case of a flood event, a yield
27 reduction or a death of sensitive to flood crops can be observed (Morris and Hess, 1988). The impact
28 of flooding on agriculture varies noticeably according to tolerance of the particular crop to water
29 excess , the duration, the frequency, the depth, and the seasonality of the event. At farm scale, the
30 flood consequences can affect the profitability of the farm business, however at national scale the
31 economical impact of flooding depends on whether crops lost in one area can be replaced in terms
32 of production elsewhere in the country. Based on this, in low flood hazard areas, an adaptation
15
 1 strategy could exchange less valuable crops for more valuable ones and take up unused land.
2 Additionally, at areas where floods are frequently observed (very high flood hazard areas) the land
3 use may be limited to low productivity or/and flood-tolerant species. In Greece agriculture occupies
4 a large proportion of the landscape. Therefore, an adopting agricultural approach at very high flood
5 hazard areas could be the effectively cultivated flood-tolerant species according the specific
6 climatological conditions of each area. At the same time, this would ensure a satisfactory and
7 surplus income (Morris et al., 2010; Kourgialas and Karatzas, 2016).
8 In the Mediterranean area flash floods take place mainly in winter, spring and autumn. Based
9 on this and taking into consideration whether different mature plants are or not in the growing
10 season (when the plants are in the growing season they tend to be more flood sensitive), the order
11 of the most common Mediterranean cultivated flood tolerance plants is quince, pear, apple, plum,
12 the order for moderately flood-tolerant species is citrus, cherry, apricot, peach, and almond, while
13 for non –flood tolerant plants the order is olive, vineyard, avocado, and annual crops (Schaffer et
14 al., 2006; Posthumus et al., 2008; Kourgialas and Karatzas, 2016). Greece is mainly dominated by
15 olive, citrus, vineyard, avocado, and annual crops. These cultivations are generally sensitive to
16 flooding (not tolerant to wet or flooded soil conditions). Thus, for a national adaptation strategy in
17 the agricultural sector and by taking into consideration the specific climatological condition as well
18 as the farmers’ income, pear and apple trees could be a prospective horticulture for very high flood
19 hazard areas of semi-mountainous and mountainous regions in Greece and quince for low lands,
20 respectively (Kourgialas and Karatzas, 2016). For annual crops located at very high flood hazard
21 areas it is recommended to be replaced by summer growing period varieties of species (complete
22 their life cycle between floods) or/and flood tolerance grassland vegetation (grazing and energy
23 grasses, herbs etc.). Grassland plants survive flash flood wave because of their small size and thin
24 flexible leaves (Day et al., 1980). Based on literature, some annual forage legumes such as Trifolium
25 michelianum, distributed in the Mediterranean basin, are flood-tolerant plants (Striker and Colmer,
26 2016). Specifically, Trifolium michelianum flourishes in sandy to clayey soils, requires precipitation
27 between 350–750 mm/year and ensures dry mass yield between 5-8 t/ha/year. The important role
28 of forage legumes consists also in their contribution to the nitrogen (N) economy of grazing lands,
29 increasing the production of herbage and the quality of grazing livestock (Porqueddu et al., 2016).
30 The above species should also be selected based on the specific climatologically conditions of each
31 area and the income that its crop can ensure to farmers.
32
16
 1
2 Conclusion
3 For many EU members, such as Greece, a national scale flood hazard mapping and a
4 protection/adaptation strategy approach do not exist. In this paper, the spatial flood hazard in
5 Greece is identified using GIS and ANNs techniques. Moreover, this work highlights the importance
6 of ANNs in producing flood hazard maps especially at large/national scale. The flood hazard areas
7 were determined based on the classification of the flood hazard in five categories ranging from very
8 high to very low. Thus, urban and agricultural areas that are most susceptible to flooding were
9 determined. These areas were identified based on the combination of seven flood hazard maps-
10 factors directly related to the creation of a flood. The map-factors used for the present study were:
11 the flow accumulation, the land use, the altitude, the slope, the soil erodibility, the rainfall intensity,
12 and the available water capacity. Using historical flood events and by combining the above
13 mentioned factors with ANN and GIS modeling approaches the final flood hazard map of Greece
14 was created. The ANN simulation performance was evaluated on the basis of various statistical
15 metrics, while the spatial reliability of the proposed methodology was verified by historical flood
16 events occurred the last 100 years. GIS results show that 24 % of the total area of Greece is under
17 very high flood hazard, while a very high percentage of artificial areas, about 37.22%, are under very
18 high flood hazard. In addition, based on the final flood hazard map 29.30% of the agricultural areas
19 are located in very high flood hazard regions. Specifically, in Greece 39.04% of the total arable,
20 8.99% of the total permanent, and 31.66% of the total pasture crops areas are under very high flood
21 hazard. Furthermore, regarding the Water Districts of Greece, the Water District of Attica appear to
22 have the highest percentage (33%) of very high flood hazard, followed by the Water District of West
23 Peloponnese with a percentage of 31%, while at the last position is the Water District of east
24 Macedonia with a percentage of 11%. Based on these findings and taking into consideration that
25 Greece probably will be strongly affected by the climate change, this study investigates and suggests
26 flood protection and adaptation approaches for rural and urban sectors minimizing the
27 consequence of flood damages under different human activities and climate change conditions.
28 Specifically, an overview of flood protection measures for very high flood hazard areas located in
29 low, semi-mountainous and mountainous regions was presented. Moreover, this study highlights
30 the importance of the agricultural sector to mitigate flooding: a) at downstream urban regions by
31 making space for flood water at intermediate agricultural areas, b) by switching to economic

17
 1 beneficial crops that are more tolerant to flooding, and c) by relocating the production to low flood
2 hazard areas by displacing less valuable crops or taking up unused land. All these adaptation policy
3 approaches can be incorporated into a national scale flood decision support system, ensuring
4 the sustainability of agriculture and the protection of civilians.
5
6 References
7 Ballesteros, C.J.A., Eguibar, M., Bodoque, M.J., Diez-Herrero, A., Stoffel, M., Gutierrez-Perez, I.,
8 2011. Estimating flash flood discharge in an ungauged mountain catchment with 2D hydraulic
9 models and dendrogeomorphic palaeostage indicators. Hydrol. Process. 25(6), 970-979.
10 Bathrellos, G.D., Karymbalis, E., Skilodimou, H.D., Gaki-Papanastassiou, K., Baltas E.A., 2016. Urban
11 flood hazard assessment in the basin of Athens Metropolitan city, Greece. Environ. Earth Sci. 75,
12 319. doi:10.1007/s12665-015-5157-1
13 Blackwell, M.S.A., Maltby, E., (eds) 2006. Ecoflood guidelines. How to use floodplains for flood risk
14 reduction. Project report EUR 22001. European Communities. Belgium.
15 Borrows, P., Bruin, D. de., 2006. The management of riverine flood risk. Irrigation and Drainage 55,
16 151-157.
17 Brocca, L., Melone, F., Moramarco, T., 2011. Distributed rainfall-runoff modelling for flood
18 frequency estimation and flood forecasting. Hydrol. Process. 25, 2801–2813.
19 Camarasa Belmonte, A.M., Segura Beltran, F., 2001. Flood events in Mediterranean ephemeral
20 streams (ramblas) in Valencia region, Spain. Catena 45(3), 229–249.
21 Carroll, Z.L., Bird, S.B., Emmett, B.A., Reynolds, B., Sinclair, F.L., 2004. Can tree shelterbelts on
22 agricultural land reduce flood risk? Soil Use and Management 20,357-359.
23 Cazemier D.R., Lagacherie P., Martin-Clouaire R., 2001. Geoderma A possibility theory approach for
24 estimating available water capacity from imprecise information contained in soil databases.
25 Geoderma 103,113–132.
26 CEC (Commission of European Communities), 2007. Directive 2007/60/WE of the European
27 Parliament and of the Council 23 October 2007 on the Assessment and Management of Flood
28 Risk. Brussels: Official Journal of the European Communities 6.11.2007 L 288
29 Chang, D. S., Zhang, L. M., 2010. Simulation of the erosion process of landslide dams due to
30 overtopping considering variations in soil erodibility along depth. Nat. Hazards Earth Syst. Sci.,
31 10, 933-946, doi:10.5194/nhess-10-933-2010.
32 Chatterjee, C., Förster, S., Bronstert, A., 2008. Comparison of hydrodynamic models of different
33 complexities to model floods with emergency storage areas. Hydrol. Process. 22(24), 4695–4709.
34 Chau, Vu. Ngoc., Holland, J., Cassells, S., Tuohy, M., 2013. Using GIS to map impacts upon agriculture
35 from extreme floods in Vietnam. Applied Geography 41, 65-74.
36 Chenini, I., Mammou, A.B., May, M.E.L., 2010. Groundwater recharge zone mapping using GIS-based
37 multi-criteria analysis: a case study in central Tunisia (Maknassy Basin). Water Resources
38 Management 24, 921–939.
39 Dandy, C. G., Maier, R. H., 2000. Neural networks for the prediction and forecasting of water
40 resources variables: a review of modelling issues and applications. Environmental Modelling and
41 Software 15,101–124.
42 Day, RT., Keddy, P.A., McNeill, J., Carleton, T., 1988. Fertility and disturbance gradients: a summary
43 model for riverine marsh vegetation. Ecology 69, 1044-1054.

18
 1 Delrieu, G., et al., 2005. The catastrophic flash-flood event of 8–9 September 2002 in the Grad
2 region, France: a first case study for the Cevennes-Vivarais Mediterranean hydro-meteorological
3 observatory. J. Hydrometeorol 6(1), 34–52. doi:10.1175/JHM-400.1
4 Di Baldassarre, G., Montanari, A., Lins, H., Koutsoyiannis, D., Brandimarte, L., Blöschl, G., 2010.
5 Flood fatalities in Africa: From diagnosis to mitigation. Geophysical Research Letters 37(22),
6 L22402.
7 Di Baldassarre, G., Schumann, G., Bates, P., Freer, J., Beven, K., 2010. Floodplain mapping: a critical
8 discussion on deterministic and probabilistic approaches. Hydrological Sciences Journal 55(3),
9 364-376.
10 Diakakis, M., 2017. Flood seasonality in Greece and its comparison to seasonal distribution of
11 flooding in selected areas across southern Europe. Journal of Flood Risk Management 10(1), 30-
12 41.
13 Diakakis, M., Deligiannakis, G., 2013. Changes in flood mortality during the last 50 years in Greece.
14 Bulletin of the Geological Society of Greece, vol. XLVII 2013 Proceedings of the 13th International
15 Congress, Chania, Sept. 2013 XLVII, No 3 1397 - 1406
16 Diakakis, M., Deligiannakis, G., 2017. Flood fatalities in Greece: 1970-2010. Journal of Flood Risk
17 Management 10(1), 115-123.
18 Diakakis, Μ., Mavroulis, S., Deligiannakis, G., 2012. Floods in Greece, a statistical and spatial
19 approach. Nat Hazards 62, 485-500.
20 Dodov, B.A., Foufoula-Georgiou, E., 2006. Floodplain morphometry extraction from a high-
21 resolution digital elevation model: a simple algorithm for regional analysis studies. Geosci.
22 Remote Sens. Lett. IEEE 3 (3), 410-413.
23 Dokou, Z., Kourgialas, N.N., Karatzas, G.P., 2015. Assessing groundwater quality in Greece based on
24 spatial and temporal analysis. Environmental Monitoring and Assessment, 187 (12)774, 1-18.
25 Downton, M.W., Pielke, R., 2001. Discretion without accountability: politics, flood damage, and
26 climate. Nat. Hazards Rev. 2(4), 157–166.
27 Eimers, J. L., Weaver, J. C., Terziotti, S., Midgette, R. W., 2000. Methods of rating unsaturated zone
28 and watershed characteristics of public water supplies in North Carolina. Raleigh, NC: Water
29 Resour. Invest. Report 99- 4283.
30 Environment Agency, 2008. Delivery of Making Space for Water- HA6 Catchment Scale Land-Use
31 Management & HA7 Land Management Practices. Environment Agency, Bristol.
32 ESRI, 2008. ArcView 9.2 User Manuals. Environmental System Research Institute, 380 New York
33 Street, Redlands, CA, 92373 USA.
34 Evrard, O., Persoons, E., Vandaele, K., Wesemael, B. van., 2007 Effectiveness of erosion mitigation
35 measures to prevent muddy floods: a case study in the Belgian loam belt. Agriculture, Ecosystems
36 and Environment 118,149-158.
37 Förster, S., Kneis, D., Gocht, M. and Bronstert, A. 2005. Flood risk reduction by the use of retention
38 areas at the Elbe River. Int. J. River Basin Management 3(1), 21-29.
39 Gaume, E., Livetb, M., Desbordesc, M., Villeneuve, J.P., 2004. Hydrological analysis of the river Aude,
40 France, flash flood on 12 and 13 November 1999. J. Hydrol. 286,135–154.
41 doi:10.1016/j.jhydrol.2003.09.015
42 Georgakakos, K. P., 2006. Analytical results for operational flash flood guidance. J. Hydrol. 317, 81–
43 103.
44 Golian, S., Saghafian, B., Maknoon, R., 2010. Derivation of Probabilistic Thresholds of Spatially
45 Distributed Rainfall for Flood Forecasting. Water Resour. Manage. doi: 10.1007/s11269-010-
46 9619-7.

19
 1 Goovaerts, P., 2000. Geostatistical approaches for incorporating elevation into the spatial
2 interpolation of rainfall. J. Hydrol. 228(1-2), 113–129.
3 He, Z., Zhao, W., Liu, H., 2014. Comparing the performance of empirical black-box models for river
4 flow forecasting in the Heihe River Basin, Northwestern China. Hydrological Processes 28, 1-7.
5 Islam, M.M., Sado, K., 2002. Development priority map for flood countermeasures by remote
6 sensing data with geographic information system. J. Hydrol. Eng. 9,346–355.
7 Jinren, R.N., YingkuiK, L., 2003. Approach to soil erosion assessment in terms of land-use structure
8 changes. J. Soil Water Conserv. 58 (3), 158–169.
9 Karymbalis, E., Katsafados, P., Chalkias, C., Gaki-Papanastassiou, K., 2012. An integrated study for
10 the evaluation of natural and anthropogenic causes of flooding in small catchments based on
11 geomorphological and meteorological data and modeling techniques: The case of the Xerias
12 torrent (Corinth, Greece). Zeitschrift für Geomorphologie 56 (1), 045-067.
13 Kourgialas, N.N., Dokou, Z., Karatzas, G.P., 2015a. Statistical analysis and ANN modeling for
14 predicting hydrological extremes under climate change scenarios. the example of a small
15 Mediterranean Agro-Watershed. J. Environ. Manage. 154, 86–101.
16 Kourgialas, N.N., Karatzas G.P., 2014. A hydro-sedimentary modeling system for flash flood
17 propagation and hazard estimation under different agricultural practices. Nat. Hazards Earth
18 Syst. Sci. 14, 625-634, doi:10.5194/nhess-14-625-2014.
19 Kourgialas, N.N., Karatzas, G.P., 2011. Flood management and a GIS modelling method to assess
20 flood-hazard areas—a case study. Hydrological Sciences Journal 56(2), 212–225.
21 doi:10.1080/02626667.2011.555836.
22 Kourgialas, N.N., Karatzas, G.P., 2015b. Groundwater Contamination Risk Assessment in Crete,
23 Greece using Numerical Tools within a GIS Framework. Hydrological Science Journal 60:1, 111–
24 132, doi: 10.1080/02626667.2014.885653.
25 Kourgialas, N.N., Karatzas, G.P., 2016. A flood risk decision making approach for Mediterranean tree
26 crops using GIS; climate change effects and flood-tolerant species. Environmental Science and
27 Policy 63, 132-142.
28 Kourgialas, N.N., Karatzas, G.P., Nikolaidis, N.P., 2012. Development of a thresholds approach for
29 real-time flash flood prediction in complex geomorphological river basins. Hydrological Processes
30 26, 1478 – 1494.
31 Kourgialas, N.N., Koubouris, G.C., Karatzas, G.P. Metzidakis I., 2016. Assessing water erosion in
32 Mediterranean tree crops using geoinformatic techniques and field measurements; the effect of
33 climate change. Natural Hazards, DOI: 10.1007/s11069-016-2354-5.
34 Kumar, D. N., Raju, K. S., Sathish, T., 2004. River Flow Forecasting Using Recurrent Neural Networks.
35 Water Resources Management 18,143– 161.
36 Llasat, M. C., et al., 2013. Towards a database on societal impact of Mediterranean floods within
37 the framework of the HYMEX project, Nat. Hazards Earth Syst. Sci., 13, 1337–1350,
38 doi:10.5194/nhess-13-1337-2013.
39 Lloyd, C. D., 2005. Assessing the effect of integrating elevation data into the estimation of monthly
40 precipitation in Great Britain. J. Hydrol. 308(1-4), 128–150.
41 Manfreda, S., Di Leo, M., Sole, A., 2011. Detection of flood prone areas using digital elevation
42 models. J. Hydrol. Eng. 16 (10), 781-790.
43 Martin-Vide, J.P., Ninerola, D., Bateman, A., Navarro, A., Velasco, E., 1999. Runoff and sediment
44 transport in a torrential ephemeral stream of the Mediterranean coast. Journal of Hydrology 225,
45 118–129.
46 Masoud B. K., et al., 2012. An artificial neural network model for flood simulation using GIS: Johor
47 River Basin, Malaysia. Environ Earth. Sci., 67,251-264.
20
 1 Ministry of Agriculture of Greece, 2010. The Greek Agriculture in Numbers: Basic Characteristics.
2 (http://www.minagric.gr/.).
3 Ministry of Environmental and Energy of Greece, 2016. Special secretariat for water:
4 Implementation of the Floods Directive. Available from:
5 http://84.205.246.56/Default.aspx?tabid=252&language=el-GR.
6 Molinari, D., 2014. Ex post damage assessment: an Italian experience. Nat. Hazards Earth Syst. Sci.
7 14(4), 901–916.
8 Moramarco, T., Melone, F., Singh, V.P., 2005. Assessment of flooding in urbanized ungauged basins:
9 A case study in the Upper Tiber area, Italy. Hydrological Processes, 19 (10), 1909-1924.
10 Morgan, R. P. C., 2005. Soil Erosion and Conservation. Oxford: Blackwell Publishing Ltd.
11 Morris, J., Hess, T.M., 1988. Agricultural flood alleviation benefit assessment. A case study. Journal
12 of Agricultural Economics 39,402-412.
13 Morris, J., Hess, T.M., Posthumus, H., 2010. "Agriculture's Role in Flood Adaptation and Mitigation
14 Policy Issues and Approaches", in Sustainable Management of Water Resources in Agriculture,
15 OECD Publishing, Paris. DOI: http://dx.doi.org/10.1787/9789264083578-9-en.
16 Muis, S., Güneralp, B., Jongman, B., Aerts, J.C.J.H., Ward, P. J., 2015. Flood risk and adaptation
17 strategies under climate change and urban expansion: A probabilistic analysis using global data.
18 Science of the Total Environment 538, 445-457.
19 Munich Re, 2016. Natural Catastrophes 2015. Annual Figures. Munich Re NatCat Service. Available
20 Online:http://www.munichre.com/site/corporate/get/params_E1254966961_Dattachment/11
21 30647/Munich-Re-Overview-Natural- catastrophes-2015.pdf.
22 Nardi, F., Vivoni, E. R., Grimaldi, S., 2006. Investigating a floodplain scaling relation using a
23 hydrogeomorphic delineation method. Water Resources Research 42, W09409.
24 O’Connell, E., Ewen, J., O’Donnell, G., Quinn, P., 2007. Is there a link between agricultural landuse
25 management and flooding? Hydrology and Earth Systems Science 11, 97-107.
26 Ologunorisa, T. E., Abawua, M. J., 2005. Flood Risk Assessment: A Review, J. Appl. Sci. Environ. Mgt.
27 9(1), 57 – 63.
28 Panagos, P., Meusburger, K., Alewell, C., Montanarella, L., 2012. Soil erodibility estimation using
29 LUCAS point survey data of Europe. Environ. Model Softw. 30,143–145.
30 Papagiannaki, K., Lagouvardos, K., Kotroni, V., Bezes, A., 2015. Flash flood occurrence and relation
31 to the rainfall hazard in a highly urbanized area. Nat. Hazards Earth Syst. Sci., 15, 1859-1871,
32 doi:10.5194/nhess-15-1859-2015.
33 Papaioannou, G., Loukas, A., Vasiliades, L., Aronica. G. T., 2016. Flood inundation mapping sensitivity
34 to riverine spatial resolution and modelling approach. Nat. Hazards 83(1), 117-132.
35 Pappenberger, F., Dutra, E., Wetterhall, F., Cloke, H., 2013. Deriving global flood hazard maps of
36 fluvial floods through a physical model cascade. Hydrology and Earth System Sciences 16, 4143-
37 4156.
38 Pekel, J. F., Cottam, A., Gorelick, N., Belward, A. S., 2016. High-resolution mapping of global surface
39 water and its long-term changes. Nature 540, 418-422.
40 Porqueddu, C., Ates, S., Louhaichi, M., Kyriazopoulos, A. P., Moreno, G., del Pozo, A., Ovalle, C.,
41 Ewing, M. A., Nichols, P. G. H., 2016. Grasslands in ‘Old World’ and ‘New World’ Mediterranean-
42 climate zones: past trends, current status and future research priorities. Grass Forage Sci., 71, 1–
43 35, doi:10.1111/gfs.12212.
44 Posthumus, H., Hewett, C.J.M., Morris, J., Quinn, P.F., 2008. Agricultural land use and flood risk
45 management: engaging with stakeholders in North Yorkshire. Agricultural Water Management
46 95,787-798.

21
 1 Rietjes, T. M., de Vos, N. J., 2008. Multi objective training of artificial neural networks for rainfall-
2 runoff modelling. Water Resources Research 44 (8), DOI: 10.1029/2007WR006734.
3 Roudier, P., et al., 2016. Projections of future floods and hydrological droughts in Europe under a
4 +2°C global warming. Climatic Change 135(341), doi:10.1007/s10584-015-1570-4.
5 Rumelhart, D.E., Widrow, B., Lehr, M.A., 1994. Basic ideas in neural networks. Communications of
6 the ACM 37 (3), 87-92.
7 Salazar, S., et al., 2009. Efficiency of non-structural mitigation measures: “room for the river” and
8 “retaining water in the landscape”. In Flood Risk Management: Research and Practice. Samuels
9 et al. (eds). Taylor & Francis Group. London.
10 Samela, C. Troy, T.J., Manfreda, S., 2017. Geomorphic classifiers for flood-prone areas delineation
11 for data-scarce environments. Advances in Water Resources 102, 13-28.
12 Sampson, C. C., Smith, A. M., Bates, P. D., Neal, J. C., Alfieri, L., Freer, J. E., 2015. A high-resolution
13 global flood hazard model. Water Resources Research 51, 7358-7381.
14 Schaffer, B., Davies, F.S., Crane. J.H., 2006. Responses of subtropical and tropical fruit trees to
15 flooding in calcareous soil. HortScience 41, 549-555.
16 Song, L., Zhao, S., Liao, W., Wang, Z., 2013. Neural Network Application Based on GIS and Matlab to
17 Evaluation of Flood Risk. International Conference on Remote Sensing, Environment and
18 Transportation Engineering RSETE, 296-299.
19 Striker, G.G., Colmer, T, D., 2016. Flooding tolerance of forage legumes. Journal of Experimental
20 Botany. doi:10.1093/jxb/erw239.
21 Sulafa H.E., 2014. Artificial Neural Networks (ANNs) for flood forecasting at Dongola Station in the
22 River Nile, Sudan. Alexandria Engineering Journal 53 (3), 655-662.
23 Teschl, R., Randeu, W.L., 2006. A neural network model for short term river flow prediction. Nat.
24 Hazards Earth Syst. Sci. 6 (4), 629-635.
25 Thieken, A.H., Kreibich, H., Müller, M., Merz, B., 2007. Coping with floods: preparedness, response
26 and recovery of flood-affected residents in Germany in 2002. Hydrol. Sci. J. 52, 1016–1037.
27 Toth, E., Brath, A., Montanari, A., 2002. Neural networks and non-parametric methods for improving
28 real-time flood forecasting through conceptual hydrological models. Hydrology and Erath system
29 sciences 6(4), 627–640.
30 Wang, Y., Zhongwu, L., Zhenghong, T., Guangming, Z., 2011. A GIS-based spatial multi-criteria
31 approach for flood risk assessment in the Dongting Lake Region, Hunan, Central China. Water
32 Resour. Manag. 25 (13), 3465–3484.
33 Ward, P. J., Jongman, B., Salamon, P., et al., 2015. Usefulness and limitations of global flood risk
34 models. Nature Climate Change 5(8), 712-715.
35 Winsemius, H.C., et al., 2016. Global drivers of future river flood risk. Nature Climate Change 6, 381-
36 385.
37 Yahaya, S., Ahmad, N., Abdalla, R.F., 2010. Multicriteria Analysis for Flood Vulnerable Areas in
38 Hadejia-Jama’are River Basin, Nigeria. European Journal of Scientific Research 42(1), 71-83.
39 Yannopoulos, S., Eleftheriadou, E., Mpouri, S., Giannopoulou. I., 2015. Implementing the
40 Requirements of the European Flood Directive: the Case of Ungauged and Poorly Gauged
41 Watersheds. Environ. Process. 2(1), 191-207.
42
43
44
45 LIST OF FIGURES
46 Figure 1. a) Water Districts of Greece, b) Spatial distribution of flood events as expressed by
47 number of events per 100 km2 in each prefecture, c) Meteorological stations used in this study.

22
 1 Figure 2. Flow chart of the proposed methodology.
2 Figure 3a. Flood hazard maps for factors: Flow accumulation, Land use, Altitude, and Slope.
3
4 Figure 3b. Flood hazard maps for factors: soil Erodibility, Rainfall intensity, available water
5 Capacity.
6 Figure 4. a) Very High and High flood hazard areas for Greece, b) Validation process of the selected
7 flood hazard interpolation map (Spline method) with historical flooded points.
8 Figure 5. Final flood hazard map for Greece.
9
10 LIST OF TABLES
11
12 Table 1. Categorization of the factors affecting flood hazard areas on Greece.
flood Hazard Level (HL) flood Hazard Level (HL)
Factors Domain of effect Descriptive form Numeric values
713750 - 1074824 Very High 5
(F) 377867 - 713750 High 4
Flow accumulation 151147 - 377867 Moderate 3
(pixels) 41985 - 151147 Low 2
0 - 41985 Very Low 1
Urban & bare area Very High 5
(L) Scrub, herbaceous, annual crops, High 4
Permanent crops, Fruit trees Moderate 3
Land use
Agro-forestry areas, Pastures Low 2
Mixed forest Very Low 1
0 - 263 Very High 5
(A) 263 - 547 High 4
Altitude 547 - 930 Moderate 3
(m) 930 - 1.465 Low 2
1.465 - 2.458 Very Low 1
(S) 0 - 7.56 Very High 5
Slope 7.56 - 14.83 High 4
14.83 - 23.00 Moderate 3
(degree)
23.00 - 33.28 Low 2
33.28 - 77.16 Very Low 1
Very Strong Very High 5
(E) Strong High 4
Moderate Moderate 3
soil Erodibility
Weak Low 2
Very Weak Very Low 1
137 - 800 Very High 5
(R) 106 - 137 High 4
Rainfall intensity 79 - 106 Moderate 3
(MFI units) 65 - 79 Low 2
0 - 65 Very Low 1
(C) ~0 Very High 5
available water Capacity <100 High 4
100-140 Moderate 3
(mm/m)
140-190 Low 2
>190 Very Low 1
13
14
15 Table 2. Validation of the final flood hazard map based on different interpolation methods
Number of historical Number of historical
flooded points located at flooded points located at
Interpolation method Total Error (%)
Very High flood hazard zone High flood hazard zone

IDW 491 of 510 points 81 of 90 points 4.66

Spline 502 of 510 points 85 of 90 points 2.17

23
 Kriging 482 of 510 points 77 of 90 points 6.83
1

24
Table 3. Flood hazard areas and the corresponding percentages both for the whole area of Greece and for each Water District area.
flood Hazard Water District Water District Water District Water District
Area Area Area Area Area
(GR1)- West (GR2)- North (GR3)- East (GR4)- West
Level (Km2) %
Peloponnese
(Km2) %
Peloponnese
(Km2) %
Peloponnese
(Km2) %
Central Greece
(Km2) %
in Greece -flood hazard- -flood hazard- -flood hazard- -flood hazard-
very low 26247 20 very low 1096 15 very low 1358 19 very low 969 12 very low 4018 39
low 27796 21 low 1485 21 low 1498 21 low 2014 24 low 2203 21
moderate 26203 20 moderate 1210 17 moderate 1443 20 moderate 2164 26 moderate 1323 13
high 19785 15 high 1162 16 high 1257 17 high 1570 19 high 941 9
very high 31040 24 very high 2228 31 very high 1714 24 very high 1606 19 very high 1916 18
Water District Water District
Water District Area Water District Area Area Water District Area Area
(GR7)- East (GR9)- West
(GR5)- Epirus (Km2) % (GR6)- Attica (Km2) % (Km2) % (GR8)- Thessaly (Km2) % (Km2) %
Central Greece Macedonia
-Flood hazard- -flood hazard- -flood hazard-
-flood hazard- -flood hazard-
very low 2442 25 very low 321 10 very low 2316 19 very low 2461 19 very low 4572 34
low 2161 22 low 566 18 low 2536 21 low 2344 18 low 3631 27
moderate 1438 15 moderate 692 22 moderate 2508 21 moderate 2802 21 moderate 2407 18
high 1193 12 High 514 17 high 1904 16 high 1984 15 high 1436 11
very high 2675 27 very high 1012 33 very high 2785 23 very high 3484 27 very high 1529 11
Water District Water District Water District
Area Area Water District Area Water District Area Area
(GR10)- Central (GR11)- ) East (GR14)- Aegean
(Km2) % (Km2) % (GR12)- Thrace (Km2) % (GR13)- Crete (Km2) % (Km2) %
Macedonia Macedonia Islands
-flood hazard- -flood hazard-
-flood hazard- -flood hazard- -flood hazard-
very low 694 7 very low 1246 17 very low 2354 21 very low 823 10 very low 1577 16
low 1962 19 low 1759 24 low 2246 20 low 1690 21 low 1701 18
moderate 2593 26 moderate 2113 29 moderate 1713 16 moderate 1945 24 moderate 1853 19
high 1861 18 High 1255 17 high 1613 15 high 1525 19 high 1570 16
very high 2972 29 very high 930 13 very high 3084 28 very high 2226 27 very high 2879 30
1
2
3
4
5
 6

View publication stats