Detecting Flash Flood Susceptible Areas Using MultiCriteria Decision Making Model: A Case Study of Thai Nguyen Province, Vietnam45244

Detecting Flash Flood Susceptible Areas Using Multi-Criteria Decision Making Model: A Case Study of Thai Nguyen Province, Vietnam Duong Thi Loi 1* Phạm Anh Tuan 2 , Nguyen Van Manh 34

Detecting Flash Flood Susceptible Areas Using

Multi-Criteria Decision Making Model: A Case Study

of Thai Nguyen Province, Vietnam

Duong Thi Loi (1)(*) Phạm Anh Tuan (2) , Nguyen Van Manh (3)(4)

(1) Hanoi National University of Education, Hanoi, Vietnam

(2) Tay Bac University, Son La, Vietnam

(3) National Cheng Kung University, Tainan, Taiwan

(4) VAST Institute of Geography, Hanoi, Vietnam

*Correspondence: duongloi1710@gmail.com

Abstract: In this study, Flash Flood Potential Index (FFPI) integrated with Geographic Information

System (GIS) to determine the susceptible areas based on the pre-event characteristics in the study area Five different physical characteristics that relate to flash flood potentials such as slope, land use land cover, soil texture, forest canopy density, and drainage density were selected to calculate the index maps Each index was classified from 1 to 10 by identifying their influence levels with the presence or absence of flash flood As a result, the most susceptible areas were given value at 10 while the least susceptible areas were assigned a value at 1 These indices were then mapped and integrated into a weighted linear model Analytic Hierarchy Process (AHP) was used to determine the weighted correlation among elements based on their importance to this phenomenon Weighted Flash Flood

Potential Index (WFFPI) was generated based on the individual indices from Slope index, Soil index,

Land use cover index, Forest Canopy Density index, and Drainage Density index The final results described visually the spatial distribution of flash flood potential in the study area Accordingly, the susceptible areas with this phenomenon were divided into four levels including very high, high,

moderate and low

Keywords: Flash flood; Flash Flood Potential Index (FFPI); GIS; run-off

1 Introduction

Floods are considered as one of the most common types of natural disasters in Vietnam Like many countries in the world, the number of flood events has increased significantly in the last few decades and has caused many negative effects on the environment and society (Gioti et al, 2013) In which flash flood is the top weather-related killer (Jeffrey, 2013) Flash floods are defined as rapid-onset hydrologic events of short-duration, hence forecasting is difficult Moreover, in almost the developing countries, the serious shortage of flood warning system causes could limit the abilities in flood prevention The complex variations in climate, land use and other anthropogenic interventions also lead the changes in flood risk and make more complicated the problem (Nektarios et al, 2011) Along with the development of science technology, the understanding of flash flood causes has improved in recent years, but most of these researchers concentrate simulating the process of flash floods based on the recorded data from the meteorological stations (Evangelia et al, 2011) In addition, the need for specific detail data of flash flood from the stations for such models is a big challenge for many developing countries where there is a shortage of warning monitoring systems, especially in mountainous areas Flash floods are known as a natural phenomenon, however, they are directly affected by on-site hydrologic

factors such as soil component, slope and land cover These factors could promote or inhibit the runoff process, main cause of the flash flood Besides, physical factors are often stable and less variable Therefore, assessing the flash flood potential areas based on the pre-event characteristics is a reliable basis for disaster research and flash flood prediction It contributes importantly to natural disaster prevention and environmental protection

One of the previous methods for estimating the run-off potential is used flash flood potential index (FFPI) This index was made and developed by Greg Smith (2003) for the Colorado river basin, in the National Weather Service (USA) To calculate the FFPI values

at Colorado River basin, four input factors (slope, vegetation, soil type, and land use) were selected as parameters and processed in GIS environment Some authors modified the original Smith version of the FFPI for implementation and applied in other study areas, typically Brewster (2009), Kruzdlo (2010), Ceru (2012), Zogg and Deitsch (2013) These studies improved by considering and changing of the weighing input factors in the final equation of FFPI Obviously, the selection of input factors and weight estimation keep a crucial role in this methodology

According to conventional reasoning, rainfall is often considered as the most important factor for forecasting floods, however, what happens to the rain when it is combined with the conditions on the ground can sometimes be more important In addition,

a flash flood can even occur with the drought condition and ground is not saturated (Christopher et al, 2010) Therefore, in some cases, runoff production processes were selected to study instead of rainfall characteristics In this paper, the modified-FFPI model was used to assess the risk of a flash flood The geographic factors at study area have a major impact on the timing of runoff, amount of infiltration, and severity of flash flooding The major factors selected as the index to assess the flash flood potential areas include slope, soil, land use, forest canopy density, and drainage density Each of these indices contains the value from 1 to 10 corresponding to the probability of flash flood from least to most and then added in a weighted linear model to create the FFPI map Weights are evaluated based

on their influence on the runoff - a direct expression of flash floods With the support of ArcGIS and ENVI software, digital maps and a satellite image are collected to build the database for studying Moreover, GIS is also a useful tool for integrating multiple indices of influence for flash flood hazard susceptibility mapping (Chen et al, 2016) This research focuses on identifying the flash flood potential areas in the study area and it can be used as

a useful material for helping decision-makers and research about the natural disaster in the local area

2 Methodology

2.1 Study area

Thai Nguyen is a province located in the Northeast region of Vietnam The geographical coordinate stretches across from 20020’ to 22025’North and 105025’ to 106016’ East Thai Nguyen shares its border with six provinces including Bac Can, Vinh Phuc, Tuyen Quang, Lang Son, Bac Giang, and Hanoi and its terrain has many mountain ranges running

from the north to the south The study area is covered by 3541.5 sq km (Fig 1) Thai Nguyen

is considered as the gateway for socio-economic exchange between the Northeast Area and the Red River Delta However, Thai Nguyen has been affected by flash floods in recent years with serious damages to humans and properties From 1994 till now, there are about four floods each year in the average and affected areas around 10 - 40 sq.km (Nguyen et al 2009) Thai Nguyen province becomes one of the most hard-hit with many deaths and reported injuries caused by flash floods in the north of Vietnam

Figure 1 Thai Nguyen administrative map

2.2 Data collection

In this research, the spatial data such as administrative, hydrology, soil and land use cover data were collected from different sources to calculate the index maps corresponding with input geographic factors More specifically, topographic map and soil map in 2010 provided by the Thai Nguyen Department of Natural Resources and Environment were used to calculate slope index, soil texture index and drainage density index The land use land cover map in 2010 was also provided by Thai Nguyen Department of Natural Resources and Environment, but the attributes were updated based on fieldwork and compared to Landsat 8 taken in 2016 This satellite data was also used to calculate Forest Canopy Density Although the data derived from different sources and time, this does not affect much to the accuracy of result Because both topography and soil texture are little changed elements, especially being topography, a relatively inherent factor Therefore, there

is almost no difference in these factors in a short time Moreover, land use land cover is the element that easily changes in a short time under human activities, so this data was updated

to 2016 from the remote sensing data and information from fieldwork to ensure the update

of research results

Five factors were considered to select including slope, soil, land use cover, forest canopy density, and drainage density Each index map was consequently given FFPI values (from 1 to 10) according to each data type The value of 10 means the highest potential for a flash flood to occur and 1 means the lowest (Ziyue et al 2015) To derive flash flood potential index maps, all maps were overlaid and weighted based on the AHP (Analytic Hierarchy Process) method Based on the outcome, the FFPI values were classified and mapped The final result was divided into four classes consisted of very high, high, moderate and low potential flash flood Low potential flash flood areas had the value from 0 to 2.5 Similarly, moderate potential areas, high potential areas, and very high potential areas were given the value from 2.5 to 5, from 5 to 7.5 and more than 7.5, respectively The methodology was illustrated in the flowchart (Figure 2)

Figure 2 Methodological flow chart

2.3 Data processing

 Identify the slope index

The slope was generated from the Digital Elevation Model (DEM) with 30 m cells, obtained by interpolating from the Thai Nguyen topographic map provided by Thai Nguyen Department of Natural Resources and Environment of scale 1: 50.000 Many scientists indicated that the variation of the slope is one of the important factors affecting the timing of runoff and amount of infiltration and infiltration rate decreases with the increase of slope angle by the experimental projects such as GreenHill et al, 1983; Fox et al,

1997 and Akbarimehr et al, 2012 In general, the rain with high intensive at slope exceeding

30 percent leads to extremely quick runoff and rapid response in local creeks and stream (Jeffrey, 2013; Wenbin et al 2015) Slope values were classified from 1 to 10 Percent slope greater than 30% were given a 10 on the scale and the slope values in the range from 0% to 30% were divided from 1 to 9 (Smith, 2003) The classification of the slope was described in table 1 The processing was done with the support of ArcGIS software and the result was shown in figure 3

Table 1 Classification of Slope by FFPI value

 Identify the soil texture index

Soil map at scale 1: 250 000 provided by Thai Nguyen Department of Natural Resources and Environment was used to carry out the soil texture index The basis for dividing the soil map by FFPI value is based on the soil texture which is determined by the proportions be weight of clay, silt, and sand A soil texture containing more sand will have better infiltration but it is hard to create the surface runoff, while more clay soils will restrict the infiltration but it can promote the runoff (Jeffrey, 2013) According to the soil map, Thai Nguyen province was covered by five soil texture types, such as Sand, Sandy Clay Loam, Silt Loam, Clay Loam, Water body, and Rock Mountain In addition, the soil textures with different thickness affect significantly to infiltration Accordingly, a soil texture with thick sand layer will infiltrate better than others, but it will be the opposite of thick clay soils In the study area, soil textures were distributed with different thicknesses and divided into three levels: from 0 to 70 centimeter (cm), from 70 to 100 cm and above 100cm As a result, the higher the infiltration is, the smaller the FFPI value is Based on the real data on soil texture integrated with classification method by Smith 2010, the FFPI values corresponding soil textures were shown in table 2 The combination of soil texture and soil layer thickness was applied to build the soil index and it was described in figure 4

Figure 3 Slope map (a) and slope

Table 2 Classification of Soil texture by FFPI value

Soil texture Soil layer thickness (centimeter) FFPI

Sandy Clay Loam

From 0 to 70 4 From 70 to 100 3 Above 100 2

Silt Loam

From 0 to 70 5 From 70 to 100 6 Above 100 7

Clay Loam

From 0 to 70 8 From 70 to 100 9 Above 100 10

 Identify the land use land cover index

The spatial data of land use land cover map 2010 provided by the Thai Nguyen Department of Natural Resources and Environment was used to generate the land use land cover index map Accordingly, the study area was divided into 10 types of land use land cover (Fig 5a) The FFPI values were given for the different types of land use land cover based on the storage ability on the foliage and roof of vegetation (Pei-Jun et al, 2007; Ebrahimian et al, 2009; Lincoln et al, 2016) and classified in table 3 The residential lands, especially the urban areas have a high flash flood potential due to the domination of impermeable surfaces and compacted soils as well as the shortage of natural vegetation This speeds up the runoff and causes the flood In the opposite manner, in areas with better

Figure 4 Soil texture map (a) and Soil texture index (b)

storage capacities such as forest, paddy, the risk of flash floods will be limited The result

of land use land cover index was mapped and shown in Figure 5b

Table 3 Classification of Land use land cover by FFPI value

Figure 5 Land use land cover map (a) and Land use land cover index (b)

Forest canopy density (FCD) is considered as an important index to identify the flood potential because it can intercept rainfall, slowing its fall to the ground Therefore, it has a positive impact on preventing floods Besides, it also helps in regulating the interchange of heat, water vapor, and atmospheric gases - the main factors lead to weather variation FCD index was calculated based on the reflective value from bands of Landsat 8 OLI (Operational Land Imager) Landsat 8 taken in October 2016 and provided by United States Geological Survey (USGS) was selected and processed in this paper In the first step, bands of Landsat images were converted from DN (Digital Number) to TOA (Top of Atmosphere) reflectance The purpose of this work is to eliminate the negative effects of the atmosphere on image quality This process has been done by using equation (1)

Lλ = MLQcal + AL (1)

where, Lλ = TOA spectral radiance (Watts/( m2 * srad * μm))

ML = Band-specific multiplicative rescaling factor from the metadata

AL = Band-specific additive rescaling factor from the metadata

Qcal = Quantized and calibrated standard product pixel values (DN)

In the second step, FCD was generated based on the indices including advanced vegetation index (AVI), bare soil index (BI) and canopy shadow index (SI) These indices were calculated as equation (2), (3), (4) The last step, VD (Vegetation density) - which was derived from the combination of AVI and BI, and SSI - which was derived from SI by using

a linear transformation were used to calculate the FCD This process was carried out as equation (5) This process was done with the support of ArcGIS and ENVI software (Duong

et al, 2017)

3(B5 1)*(65536 B4)*(B5 B4)

Where, B5 is near infrared band and B4 is red band

100 100

* ) 2 5 ( ) 4 6

(

) 2 5 ( ) 4 6

(

















B B B B

B B B B

BI (3)

Where, B2 is green band, B4 is red band, B5 is near infrared band and B6 is shortwave infrared band

3 ( 65536 B2 ) * ( 65536 B3 ) * ( 65536 B4 )

Where, B2 is green band, B3 is blue band and B4 is red band

1 1

*  

 VD SSI

In this case, the values of FCD were given in value range from 1.44625% to 90.7236

% FCD Index was reclassified then by FFPI value from FCD map with the values range from 1 to 10 and mapped in figure 6 As a result, the high flash flood potential corresponds with the low FCD and vice versa (Table 4)

Table 4 Classification of Forest Canopy Density by FFPI value

 Identify the drainage density index

Drainage density was determined based on the total length of stream per basin area

by Roberte Horton 1945 DEM data 30 meters used first to determine the boundary of river basins by the interpolation method with the support of ArcGIS software According to the interpolation result, the study area was divided into five river basin, those are Cau river basin, the Cong river basin, Cho Chu, Nghing Tuong, and Du river basin In the next step, hydrographic system maps taken from Thai Nguyen Department of Natural Resources and Environment and river basins map were combined to take out the drainage density map As the result, the drainage density values corresponding with the Nghing Tuong river basin,

Du river basin, Cau river basin, Cong river basin, and Cho Chu river basin were 0.68, 0.82, 1.43, 2.0 and 2.75 respectively (figure 7) After that, this result was classified into FFPI value and mapped A drainage basin with a large number of tributaries has a higher stream density than a basin with very few streams (Pallard et al, 2009; Gregogy, 2010), therefore the FFPI value increased corresponding with increasing of drainage density

Figure 6 Forest Canopy Density (a) and Forest Canopy Density

Index (b)

 Generation of Weighted Flash flood potential index (WFFPI)

Analytic Hierarchy Process (AHP) was used for determining the weights of the individual index This method was developed by Thomas Saaty in 1990 and become one of the most famous methods for making multi-criteria decisions Saaty’s method describes the level of importance of parameters and their relationship on a scale of 1 to 9 After the computation of weights using Saaty’s pairwise comparison method, the Consistency Ratio (CR) in this case was 0.018992 The Slope was considered as the most important index and given the weight at 0.47 and the Drainage Density was determined as the least important index and weighted at 0.04 The Soil index, Land use land cover index and Forest Canopy Density index were weighted with 0.09; 0.3; and 0.1 respectively Therefore, Weighted Flash

Flood Potential Index (WFFPI) was generated based on the individual indices from Slope

index, Soil index, Land use land cover index, Forest Canopy Density index, and Drainage Density index and computed as given (6):

N

D F

S L

M WFFPI0.47( )0.3( )0.09( )0.1( )0.04( ) (6)

Where, WUGSI: Weighted Flash Flood Potential Index

M: Slope index

L: Land use land cover index

S: Soil texture index

F: Forest canopy density index

D: Drainage density index

N: Sum of weightings

Figure 7 Drainage Density (a) and Drainage Density Index (b)