Flash Flood Risk Assessment for Girne Region, Northern Cyprus

Girne region is in the northern part of Northern Cyprus which is environmentally fragile and susceptible to natural disasters. Flash floods are a major problem in the region due to the heavy and torrential rainfalls in its urban environment. Therefore, this study aims to assess the flash flood risk for the Girne region, using the Geographic Information System (GIS). A mitigation flood risk plan is proposed and applied to the case study of the Girne region. The flood risk matrix is proposed based on the occurrence probability of the flood and the associated inundation depth. The risk matrix criterion was classified according to the degree of risks as high, moderate, and low. Five thematic maps affecting flood occurrences, including slope, elevation, land use, peak discharge, and flow accumulation, were classified to generate flood hazard maps. The results of the estimation of the magnitude of the inundation areas that can assess the degree of damage and its economic aspects are presented graphically. The developed flood risk matrix tool is a quantitative tool to assess damage and is essential for decision-makers. KeywordsFlash flood, Girne, Northern Cyprus, GIS, flood risk matrix


INTRODUCTION
Floods may occur suddenly and be accompanied by other hazards such as landslides, mudflows, and damages to infrastructure [1,2]. They occur due to the high intensity of rainfall within a shorter period [3]. The high-velocity runoff in small basins and the rapid rise of water and sediment transport have made flash floods extremely dangerous to infrastructure and human life. Flashfloods and used landslides have become major disasters worldwide [4]. In general, flash floods are the combined result of weather and geomorphological conditions [5][6][7]. Urban flooding is an essential issue globally due to its frequent occurrence that leads to damage to infrastructure and loss of lives. Flash floods in an urban area are recognized by the relatively short period of ephemeral water flow, which rarely lasts more than a day [8]. Thus, understanding the nature and the characteristics of flooding is one of the main goals of flood risk management. At present, flood maps are a potentially valuable tool for improving this understanding, but some drawbacks regarding the use of flood maps to mitigate flood risk and identify prone areas have been highlighted in the literature [9]. Anyhow, flood maps help to identify the most sensitive zones for civil protective actions, assess damages and make valid urban planning [8].
In Northern Cyprus, the coastal regions Girne (Kyrenia), Gazimağusa (Famagusta), and Güzelyurt (Morphou) are highly susceptible to urban flooding. During the recent years, heavy rain and thunderstorms affected over 3000 people, inflicting damages to the road network and the partial closure of a freeway in the Girne region [10][11][12][13]. This study aims to present a flood risk mitigation plan and a flood risk matrix technique for assessing risks in the urban Girne region of Northern Cyprus. The study focuses on the evaluation of the magnitude of the inundation areas and can assess the degree of damage from flash floods in the region for different frequencies of HEC-HMS hydrologic and 2D HEC-RAS hydraulic modeling techniques in order to delineate the inundation area at different degrees of flood hazards. To the best of our knowledge, there are no detailed studies in the Girne region regarding flash flood risk assessment.

II.
MATERIALS AND METHODS The proposed methodology aims to present a framework developed for flood modeling on regional scale combining GIS with a rainfall-runoff model (HEC-HMS) and a hydrologic model (HEC-RAS). Figure 1 illustrates the methodology used in this study. The flowchart of the proposed methodology.

A. Study Area
Girne region is located in the northwest of Northern Cyprus at 35.2992° N latitude and 33.2363° E longitude as shown in Figure 2  It is surrounded by the Five Fingers mountain (Beşparmak), where elevations range from -2 to 1026m above sea level (Figure 2(b)), which has a major impact on defining the weather on the island. Slopes range from 0 % to 20% as shown in Figure 2(c). Rainfall distribution over the country varies considerably among regions. The minimum and maximum average values of rainfall in Mesaoria are 300 and 400mm whereas they are 400 and 450mm in the Karpas Peninsula area. Snow is rarely occurring on the upper hills of the Besparmak Mountains in the North and is usually present at the peaks of the Troodos Mountains in South Cyprus throughout the year. The maximum snowfall over the Northern part of Cyprus in the last ten years (1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002) was measured to be 15cm on Besparmak, Selvili Tepe, and Katara hills. Additionally, temperatures ranged from 21Ԩ along the coast to 15Ԩ on the top of the Girne. Maximum temperature can reach 45Ԩ in the Mesaoria Plain. Moreover, there is no perennial stream in the Northern part of Cyprus. Twenty-eight streams located in the Northern part of Cyprus discharge about 27MCM of water annually [14]. Additionally, 10 ephemeral streams originating from the Troodos Mountain in the Southern part of Cyprus, discharge about 43MCM of water yearly [14]. Figure 3 shows the extracted channel networks and the flow connections from QGIS including dam locations. B. Extracted Digital Elevetion Model (DEM) and Hydrologic Analysis using QGIS and SAGA PLUG Digital Elevation Model (DEM) data can be downloaded directly and are free for use from USGS earth explorer (https://earthexplorer.usgs.gov). The extracted 3D models of the earth's surface have cells that are represented as raster images in a pixel grid format. Each cell indicates the covered area's average elevation value. The model was improved by filling sinks and depressions locations with the Wang Liu algorithm. Catchments, Sub-Basins (SBs), flow accumulation, drainage sites, and flow directions will be used as input for runoff simulation after implementing spatial and hydrological analysis tools such as Quantum GIS (QGIS) and saga plug.

C. The Wang Liu Method
Virtually, all DEMs contain flat areas or depression pixels that may be actual landscape representations of artifacts. These features must be handled before any further hydrological analysis can be conducted [15]. Various algorithms have been developed for correcting these aspects, differing in how they handle the nature of the depressions. The Wang Liu method was utilized to fill sinks and depression areas in the extracted DEM, and preserve the downward slope along the flow path, in addition, to enhancing hydrological analysis for surface elevations [15].

D. Rainfall Data and Return Period Analysis
The data have been collected from the Meteorological Department located in Lefkoşa, Northern Cyprus for a period from 1995 to 2015. Figure 4 shows the total amount of rainfall each year. The maximum and minimum values of total rainfall were recorded in 2012 and 2013 with a value of 593.86mm and 182.79mm respectively. Moreover, the estimation of the return period of high intensity of rainfall is particularly significant for many engineering elements including the construction of contour drains, dams, road culverts, airfield drainage, storm sewers in metropolitan areas, and flood or sediment control dams in water catchments. The measurement of the likely time interval between the occurrence of one event and the occurrence of a subsequent event of equal or larger magnitude is the return period. When a variable (X) equal to or higher than x occurs once every T years on average, then the probability of occurrence ܲ is shown in (1) [16].
Several approaches have been presented to establish the plotting position, which refers to the probability value assigned to each piece of data to be plotted. Equation (2) is the most plotted position formula [16]: where m is the rank value in a list ordered by descending magnitude, n is the total number of values to be plotted, and b is a parameter that varies by formula (b=0.5 for Hazen, b=0.3 for Chegodayev, b=0 for Weibull, b=3/88 for Blom, b=1/3 for Tukey, and b = 0.44 for Gringoten) [16].
E. Assessment of Land Use, Soil Characteristics, and Curve Number (CN) Surface runoff is the excessive rainfall after evaporation and infiltration have been excluded. Based on exhaustive rainfall/runoff data variety of land cover conditions and soil characteristics (moistures, structures, textures, and compactions), the method results in the establishment of a set of curves called Curve Number (CN). This empirical parameter is used in hydrology for predicting direct runoff or infiltration from rainfall excess. In general, soil groups are classified into four classes according to the Natural Resources Conservation Service (NRCS), which helps to identify land use classes and CN [17]. In this study, a virtual map ( Figure 5) was prepared by utilizing spatial analysis tools in QGIS using Landsat-8 images, which were uploaded from USGS earth explorer dated on 20 th of February 2020.

F. Rainfall-Runoff Simulation Model
Runoff simulation was done using HEC_HMS based on previous studies [18,19] that utilized the software to assess flood impact and predict flows. The main elements to run simulation are: sub-basins, time-series data, control specifications, and metrological data. SCS CN was utilized to calculate the hydraulic loss rate. The 56 sub-basins were extracted using QGIS as shown in Figure 6.

G. Loss Method
Many methods have been utilized to simulate infiltration losses such as exponential, deficit and constant, initial and constant, Green and Ampt, Smith Parlange, and Soil Moisture Accounting-SMA SCS CN. The nominated method in this study is SCS CN which is used to determine the hydrologic loss rate. Equation (3) is used to calculate the runoff of the hydrographic basin: where ܳ is the surface runoff amount, ܲ is the rainfall amount, and ܵ is the potential maximum retention amount, which can be calculated by using the below function: where CN is the curve number.

H. Transform Method
In previous studies, 7 methods have been employed to calculate the excess of precipitation to runoff: Mod-Clark, SCS unit hydrograph, user-specified graph, user-specified unit hydrograph), Snyder unit hydrograph, Clark unit hydrograph, and kinematic wave. The nominated method in this study was SCS unit hydrograph-based lag time calculated for each subbasin. Lag time (ܶlag) represents the length of time between the peak discharges of the centroid of precipitation mass and the peak discharges of the resulting hydrograph it depends on the time of concentration (ܶC) as explained in (5) and the time of concentration (ܶC) calculated by (6) [20]: where ‫ܮ‬ represents the length of the main channel in km, ‫ܪ‬ represents the difference between the mean basin elevation and the outlet elevation in m.

I. Weighting of Parameters Which Influence Flood Events
Based on the literature review and the Delphi approach [21,22], 5 parameters were considered and rated to assess flood risk: Flow accumulation (F), Peak discharge (P), Elevation (E), Land use (L), and Slope (using ArcGIS). Spatial tool analysis was utilized to develop numerical thematic maps for flood hazard maps that were georeferenced to EPSG Projection 6312 Cyprus coordinate system. Parameter rating was done by applying Jenk's Natural Breaks method. The selected method classified Risk Levels (RLs) to 5 categories and assigned numerical values from 1 to 5: catastrophic RL = 5, severe RL = 4, major RL = 3, minor RL = 2, low RL = 1. Considering that the nominated parameters have different influences on flood generation. Correlation analysis was applied taking into consideration the influence of each factor on the other factors. Thus, two kinds of influence were utilized: major influence was assigned 1 point for the parameter bearing a direct influence on the remaining parameters, minor influence was assigned 1/2 point for first parameter bearing an indirect influence on the other parameters. The calculated rate is summarized in Table I.

III. RESULTS AND DISCUSSION
Based on the extracted factors, hazard and inundation prone areas with a high risk of flooding can be identified and to be used for risk matrix identification and quantitative analysis. Five thematic maps were developed and classified based on the selected method.

A. Catchment Area and Flow Delineation
The catchment consists of 56 sub-catchments with a total area of about 640km 2 . The geometric and morphological parameters of the sub-catchment were used as main inputs to simulate runoff by utilizing the HEC-HMS model to calculate an excess of precipitation and peak discharge. Also, by utilizing the SCS method, ܶC and ܶlag were calculated. As shown in Figure 7, the highest values for ܶC, and ܶlag were found to be 229.39min and 137.64min respectively for subbasin 10 and the minimum values were recorded for sub-basin 25 with a value of 48.25min and 28.90min respectively. This variation occurred due to the variance of elevation difference, stream length, and other morphological parameters influencing the estimation of the time of concentration.

B. Land Use -Land Cover and Curve Number Estimation
Remote sensing techniques were employed to classify Land Use and Land Cover (LULC). Girne LULC was extracted from the high-resolution Landsat-8 rasters, by utilizing spatial tool analysis in ArcGIS and maximum likelihood classifications based on NVDI. Five categories (road, woods, urban areas, vegetation, and pasture) were employed to calculate the CN in Figure 8. The maximum value of CN was 86.64 for sub-basin 1 which consists of 85% pasture area and the minimum value

C. Return Period Analysis
Six different methods were employed to calculate the design period for the annual maximum rainfall for the Girne region. As a result, the return design period of infrastructure hydraulic design is less than or equal to the recorded period data. The estimation of quintiles using empirical distribution functions or plotting point methods was followed. The return period of the 40.14mm event was 23 and 44 years, based on plotting points applied by the Hazen and Weibull methods respectively. The rainfall depth is estimated and summarized in Table II.

D. Rainfall-Runoff Simulation
In this study, the selected region is within ungagged catchments, and no runoff records were available. A Synthetic Unit Hydrograph (SUH) was utilized to calculate direct runoff and excess rainfall by using HEC-HMS. The main parameters in the utilized model are hydrological data and sub-catchment physical features. Computed runoff characteristics in terms of Peak Discharge (PD) value and Runoff Volume (RV) are summarized in Table III for various design periods. For example, it is found that the maximum value of PD was 15.3cu.m/s for sub-basin 23 with a total area of 41.40km 2 , while the maximum RV was recorded for sub-basin 1 with a value of 381.68mm 3 for the selected design period with 2001 as the base year. It should be noted that the output from the model would be used as input for the hydraulic model. Table IV lists the classification of the nominated factors into hazard levels. The calculated percentage of each parameter was found by multiplying the proposed weight of effect with the rate. In general, the hazardous area cannot be estimated by considering the effect of each factor independently. The integration of all factors is necessary to obtain the overall map of flood-hazard areas. Since not all factors have the same degree of influence on the hazardous areas, the weighting approach was applied. The selected parameters have the most substantial impact on flood risk based on their rate and weight, taking into consideration the cross-pollination between parameters. In this study, the ratings for the hazard levels (expert judgment), in points, which are illustrated in Table IV are: Catastrophic = 5, Major =4, Moderate = 3, Minor = 2, Low = 1. It should be noted these observations are supported by previous works [23,24]. Moreover, the thematic maps were generated by utilizing GIS tools as shown in Figures 9 and 10. In this study, the high-risk areas are shown in blue colors such as Akdeniz, Grine, karakum, Geçitköy, Cape Kormakitis, and Karaoğlanoğlu with slopes less than 3.6% and elevations less than 129m. While the slope and higher elevation had minimal impact on the flood risk, the low-risk areas were shown in red color for the Five Fingers mountain (Beşparmak).   Additionally, a land-use/land cover map was extracted from the land sat-8 image. Land-use/land cover factor which influences the flood risk impact where the vegetation cover exists will reduce flood impact by controlling surface runoff when reaching the soil, while without vegetation cover in an urban area it will increase the flood risk as shown in Figure 11. This outcome is in accordance with the results of [23]. Peak discharge values were determined by utilizing runoff simulations and a thematic map (Figure 11) was generated by utilizing IDW (Inverse Distance Weighting) tool in GIS. Finally, the flow accumulation thematic map was extracted as shown in Figure 12. It is observed that sub-basins 4 and 2 have higher risk levels (i.e. blue color represents the high value of flow accumulation).  F. 2D Flood Mapping and Hydraulic Model using HEC-RAS HEC-RAS was utilized to generate the hydraulic 2D flood mapping based on excessive precipitation and inflow hydrographs in unsteady flow conditions. The selected design periods were used to calculate Water Surface Elevation (WSE) and flow depth. For the next 5 years, the flood map predicts a maximum depth of 10.25m downstream and a minimum depth of 0.002m at high elevations, classifying the risk as catastrophic (the 25-year map is shown in Figure 13). A bit of water spreading over the mainstream and channel networks is shown while in the 50-year map more stream spreading were seen. For the 100-( Figure 14) and 200-year maps, the flood was significant and dominant. Other researchers who analyzed the inundation depth can support these observations. For instance, authors in [25] found that the flow depth for the low-risk area is less than 0.1m at the high elevation while 3m and above for high-risk areas mainly in major down streams.

E. Parameter Weighting and Map Classifications
Flood hazard maps, which were generated from the combined 5 thematics for different design periods were utilized to identify prone areas in the Girne region. In the context, lowlands, such as the Geçitköy located in the sub-basin 23 and Akdeniz in the sub-basin 4, are most vulnerable to flood occurrence, and meanwhile, at the highest land areas, minor risk occurs at Five Fingers mountain, while catastrophic risk areas in blue color are spreading in major downstream areas in Geçitköy. As illustrated in Figures 15 and 16, the low-risk area decreased from 30.42km 2 to 13.45km 2 while the catastrophic event area increased from 0.006km 2 to 4.344km 2 . The differences are due to changes in land use, urban and human activities, and climate change.

G. Risk Flood Matrix
The risk matrix ( Figure 17) was established based on the likelihood of flow depth occurrence and its impact. A classified consequence (low, minor, major, severe, and catastrophic) is represented in a horizontal row, and the probability of occurrence of the return period in a vertical row. The degree of risk was determined by multiplying probability and impact, and the results were classified as low, medium, and high which are represented in green, yellow, and red, respectively.

H. Defining Strategies
The strategies are classified into two categories, namely vulnerable local community and structural strategies [26][27][28]. The vulnerable local community strategy aims to eliminate and mitigate risk [27] due to the lack of response at an emergency. The structural strategy aims to control risk by reducing the impact of the flood. Figure 18 summarizes

IV. CONCLUSIONS
The current paper intends to propose an established framework for a flood risk mitigation plan (Figure 19), which comprises data collected from weather forecasting in partnership with the water development department and crisis services. Hydrological analysis and risk analysis were conducted and measures were recommended. Flood parameters were illustrated in thematic maps in GIS using the IDW method. Peak discharge for various design periods was estimated using runoff simulation HEC-HMS. Flow depth was estimated using the hydrologic model HEC-RAS. Land uses were extracted from Landsat images-08. Elevation and slopes were extracted from a DEM. Flood hazard and inundation thematic maps revealed that lowlands are the most vulnerable to flooding, whereas higher land areas are less vulnerable to flood risk. The obtained results will assist planners and decisionmakers in identifying prone areas to take appropriate and timely steps during the pre-disaster period. Flood risk mitigation plan.