- Email: [email protected]
A GIS-based approach for assessing social vulnerability to flood and debris flow hazards
Journal Pre-proof A GIS-based approach for assessing social vulnerability to flood and debris flow hazards Chien-Hao Sung, Shyue-Cherng Liaw PII:
S2212-4209(19)31379-2
DOI:
https://doi.org/10.1016/j.ijdrr.2020.101531
Reference:
IJDRR 101531
To appear in:
International Journal of Disaster Risk Reduction
Received Date: 3 October 2019 Revised Date:
16 January 2020
Accepted Date: 15 February 2020
Please cite this article as: C.-H. Sung, S.-C. Liaw, A GIS-based approach for assessing social vulnerability to flood and debris flow hazards, International Journal of Disaster Risk Reduction (2020), doi: https://doi.org/10.1016/j.ijdrr.2020.101531. This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. © 2020 Published by Elsevier Ltd.
1 2 3 4 5
A GIS-based Approach for Assessing Social Vulnerability to Flood and Debris Flow Hazards Chien-Hao Sung 1, Shyue-Cherng Liaw 1,* 1
Department of Geography, National Taiwan Normal University, Taipei 10610, Taiwan. * Correspondence: [email protected]; Tel.: +886-2-7734-1649
6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
Abstract: Owing to the escalating environmental hazards caused by climate change, the mitigation of disaster becomes extremely important. The investigation of social vulnerability is a prerequisite for formulating a mitigation plan to environmental hazards. This research applies a GIS-based approach with the Social Vulnerability Index (SoVI) to investigate and quantify the social vulnerability to environmental hazards in Yilan County, Taiwan. In order to construct the SoVI, the literature review was conducted, and 12 variables were selected. Through Principal Component Analysis (PCA), the 12 variables were reduced into four principal components. In order to explore the spatial pattern of SoVI, the spatial autocorrelation analysis was applied. The result showed that there were 26.5% of communities in Yilan County with a high level of SoVI, and most of these communities were mainly located in mountain areas. The unfavorable topography features cause the distributions in mountain areas. On the other hand, there were 37.3% communities with a related low level of SoVI, and these communities were located in plain areas. The inaccessibility caused by topography creates an incapability, resource-lacking environment and lead to a high value of SoVI. In addition, this research applied Geographically Weighted Regression (GWR) to validate SoVI, and the result of the R2 value was 0.769. Also, the standardized residuals showed no spatial autocorrelation, meaning the SoVI had the adequate explanatory ability. This research provided a set of valid indicators to explore the social vulnerability for decision-makers to formulate the mitigation plan of environmental hazards. Besides, SoVI is a suitable tool for visualizing and quantifying the potential loss to environmental hazards.
23 24
Keywords: Social vulnerability, SoVI, Spatial autocorrelation, Geographically weighted regression
25
1. Introduction
26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49
Since the late 20th century, the frequency of extreme climate events has increased dramatically, and also augments the intensity and damage to flood and debris flow [1-3]. As a result, environmental hazards reduction and mitigation have become a critical task for the government. Therefore, one of the most important requisite analysis is the potential loss caused by environmental hazards. Over the last two decades, vulnerability is becoming one of the essential terms for the analysis of potential losses caused by environmental hazards. Visualizing and quantifying vulnerability enable the decision-maker to recognize the vulnerable areas and forecast potential losses. According to [4,5], the vulnerability composes of social vulnerability and biophysical vulnerability and can be represented as potential losses to environmental hazards [4-7]. The social vulnerability can be regarded as the partial outcome of social inequalities [4]. In other words, social factors have shaped, influenced, or govern the ability of susceptible groups to respond when they face environmental hazards. Social factors include a lot of variables, such as gender, ethnicity, age, socioeconomic status, and education, etc [4-7]. General speaking, environmental hazards do not randomly affect society [8,9]. Usually, the groups that already marginalized by unfavorable socioeconomic status are the most likely to be jeopardized by environmental hazards [10]. When encountering environmental hazards, these groups are susceptible to the impacts [4,11-13]. Most of the researches apply the concept of social vulnerability to environmental hazards to quantify and investigate the susceptible groups [14-19]. On the other hand, biophysical vulnerability represents the geographic context of an area [5]. Based on the different locations, the biophysical vulnerability will also differ from place to place. For example, due to the topographic characteristics, the coastal areas are often affected by storm surges and floods. As a result, two of the most common factors affecting the biophysical vulnerability are the storm surge and flood in coastal areas under climate change [20-24]. Although the geographic context varies from place to place, undoubtedly, the concept of vulnerability to environmental hazards is suitable for around the world since it has been applied worldwide, such as Europe [9,25], America[4,5], Asia[26-29], Oceania[30], and Africa[31,32].
2 of 15
50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75
Both the social and biophysical vulnerabilities are related to the potential loss induced by environmental hazards. However, it does not mean that exposure to biophysical hazards will determinedly increase vulnerability [9,33]. Those who affected by the result of social inequalities are the most vulnerable groups since they are marginalized by the disadvantage socioeconomic status [10]. Therefore, social vulnerability is the prerequisite for assessing vulnerability to environmental hazards. In addition, the spatial scale is another crucial factor [8]. For example, the investigation can be conducted on a national, county, and sub-county scale. However, some scales are far inadequate for investigation. Because the result of the investigation on the national or regional scale is too general to offer a specific analysis, the majority of researches is conducted in the county or even sub-county scale [4,8]. The Yilan County is the administrative district with the largest plain area in eastern Taiwan. It is also the most susceptible area to environmental hazards due to the lack of topographic protection from the typhoons by the Pacific Ocean. In this decade, environmental hazards including floods and debris flow have seriously caused tremendous damage to Yilan County. Thus, a holistic investigation of social vulnerability is urgently required for the county government to formulate a thorough mitigation plan. So, the primary purpose of this research is to quantify social vulnerability to environmental hazards in Yilan County with the latest data. Despite the fact that social vulnerability has been applied in many various regions based on the different social and geographic contexts, the content of social vulnerability needs to be modified to fit the specific context. Therefore, this research aims to construct a set of Social Vulnerability Indicators (SoVI) based on the study area’s context. Comparing with other research, some SoVI variables from previous studies are not included in this study. One of the major reasons is owing to the data accessibility. Some variables are inaccessible since it is based on the Computer-Processed Personal Data Protection Law in Taiwan. In addition, some of the social contexts in other countries don't suitable in our study areas. For example, the variable of mobile homes is very important for environmental hazards mitigation [4]; however, there are no mobile homes in our study area. Moreover, this research applies the data with community-scale in order to investigate the variation of SoVI spatially. This research also uses spatial autocorrelation analysis to explore the distribution pattern.
76
2. Materials and Methods
77
2.1. Yilan County
78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95
The study area of this research is Yilan County, where is the most developed county in eastern Taiwan (Fig 1). Its’ population is about 460 thousand people. There are 12 townships with 233 communities. The total area of Yilan County is 2,144 km2, and the elevation is from sea level up to the highest 3,589 m. The mean elevation of Yilan County is 830 m. The southwestern part is a mountainous area, while the northeastern part is plain and the most populated area. Compared to other counties, Yilan County has freeway and railroad connections to the capital of Taiwan, Taipei. After the revision of the Agricultural Development Act, it is allowing the rapid and large scale of land development. Therefore, Yilan County has also undergone a significant industrial transformation and economic development within the last decade. Due to the location, accessibility, and topography, Yilan County undoubtedly has the best status of further development in eastern Taiwan. However, owing to the geographical location, Yilan County is also the first and foremost county that encounters the typhoon and tropical cyclone coming from the western Pacific Ocean. According to the Central Weather Bureau of Taiwan, annually, Taiwan will have to face at least 4 typhoons. Owing to the tremendous amount of precipitation brought by the typhoon and tropical cyclone, the floods become the most notorious natural hazard events in Yilan County. Moreover, because of the steep topography and fragile geology, the debris flow also frequently occurs in our study area. These natural characteristics had created a tremendous amount of floods and debris flow. Recently, due to climate change, the situation had become much more severe and created uncountable damage.
3 of 15
96 97 98
Fig. 1. The distribution of elevation, transportation path, major townships, and the most common path of the typhoon that affect study area.
99
2.2. SoVI Variable Selection
100 101 102 103 104 105 106 107 108 109 110
Based on the difference of context and the scale of data, the SoVI is a highly modifiable tool for quantifying social vulnerability to environmental hazards. The SoVI can be applied in various kinds of scale and selects different kinds of socioeconomic variables such as age, population density, gender, ethnicity, financial support, and education [4-7]. Age can represent the physical ability for movement or mobility. Elderly residents and infants have the possibility to increase the difficulty of evacuation [8,34]. Therefore, we take the population that is older than 65 years old or younger than 5 years old as the variables. Also, because of the convenient freeway, Yilan County and Taipei City are highly connected. Taipei City is the capital of Taiwan, which has more employment opportunities and pulls the population migration from Yilan County. Thus, solitary elderly become extremely common in our study area. The solitary elderly is exceptionally vulnerable to environmental hazards since the majority of them are unemployed and lack financial resource [34]. Thus, we select the solitary elderly as one of our variables. Moreover, population density can stand for
4 of 15
111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162
potential exposure to environmental hazards and can also be regarded as the requirement of resources during or after the natural disaster [6]. Thus, we take the population density as one of our variables. Considering gender issue, females would be physically more vulnerable than males and could also encounter hazards more difficult during the recovery phase due to unequal wages, employment, and family responsibilities [24]. According to the Report of the Women’s Living Condition Survey 2015 in Taiwan [35], which conducted by the Health and Welfare Department, 44.4% of the female population is unemployed between the age of 15 to 64. The reason caused this situation is mainly owing to “household tasks” and “taking care of children” [35]. Therefore, when facing an environmental hazard, the female might not have sufficient financial resources to cope with it. Undoubtedly, gender is one of the debatable variables, yet, in Taiwan's social context, and women certainly have a significant difficulty when facing natural hazards. Consequently, we select the female population as one of our variables. Ethnicity, one of the most common variables, can represent the language and culture barrier. For specific ethnicities, foreigners and indigenous peoples, they are considered as the disadvantaged group of the social network, education level, information source, and financial resource. They might lead to erroneous decision-making and catalyst potential loss [4,6,8]. Most of the foreign labors in Taiwan are manual labor and engaged in the most laborious physical job with a limited payment and education. Without a doubt, when facing natural hazards, they are surely the susceptible group. The indigenous population, in Taiwan’s context, is one of the most common underprivileged groups. According to the Economic Status Survey of Indigenous Peoples 2014 in Taiwan [36], which conducted by the Council of Indigenous Peoples, the gross annual income of an average indigenous family was 658,117 NTD while the gross annual income of an average family was 1,071,427 NTD. Moreover, based on the Annual Status Report of Indigenous Population and Health Condition 2015 in Taiwan [37], which also conducted by the Council of Indigenous Peoples, the life expectancy of the indigenous peoples is shorter by 8.3 years compared to a normal citizen. The reason caused this situation is the inequalities of health and medical resources. Furthermore, generally speaking, most of the indigenous peoples have lower education [38]. Therefore, indigenous people’s situation is extremely vulnerable when they are facing an environmental hazard. Thus, we select the foreign labors and indigenous population as the variables. Financial support and education status also play a significant role. Lack of financial resources and lower education levels could mean the potentially insufficient information, lacking mobility, and inadequate preparedness ability [4-7,24,34]. Consequently, we take the population without a high school diploma and low-income households as variables. Moreover, Taiwan is one of the areas that have the lowest birth rate in Asia. In other words, population aging is already a significant issue. Under this circumstance, a family will have more burden [24,34]. Thus, we use the dependency ratio and the aging index as the variables. We also adopt the physical and mental disorder as one of the variables since the majority of them will require extra resources to take care of. When facing natural hazards, undeniably, they are the disadvantaged groups and will become a burden for their families. The scale of data also plays an essential role. In order to construct a valid set of SoVI to explore the social vulnerability at the community level, we select 12 variables based on the literature review and data accessibility. These variables include Population Density, Female Population, Dependency Ratio, Solitary Elderly, Physical and Mental Disorder, Aging Index, Population Under 5 Years Old, Population Aged 65 Years and Above, Indigenous Population, Foreign Labors, Low-Income Households, and Population Without High School Diploma (Table 1). All the variables were community-scale data and obtained from the database of the National Geographic Information System (NGIS). We standardize the variables due to their different units. The standardization rescales the variables with a zero mean and a standard deviation of one. Then we apply the Principal Component Analysis (PCA) to compress the variables into a smaller set. Also, PCA can reduce multi-collinearity and redundant data. Based on the Kaiser-Meyer-Olkin (KMO) and Bartlett’s Test of Sphericity, we can identify the result of PCA, whether it is acceptable. In order to increase the interpretability of the factors, we apply the Varimax rotation with Kaiser Normalization, which is the most common approach to extracting and interpreting the components. Then, the factors with an eigenvalue larger than 1 are extracted. Finally, we classify SoVI into five categories based on the level of standard deviation (SD) including very high (> 1.5 SD), high (1.5 to 0.5 SD), medium (0.5 to -0.5 SD), low (-0.5 to -1.5 SD), and very low(< -1.5 SD).
5 of 15
Table 1 Variables and descriptive statistics Variables
Max
Min
Moran's I
p-value
Population Density
30024.25
0.34
0.634
< 0.05
Effects on vulnerability to hazards Positive
Female Population
3288
22
0.582
< 0.05
Positive
Dependency Ratio
56.32
26.97
0.420
< 0.05
Positive
Solitary Elderly
14
0
0.197
< 0.05
Positive
Physical and Mental Disorder
406
0
0.198
< 0.05
Positive
Aging Index
1200
26.26
0.107
< 0.05
Positive
Population Under 5 Years Old
263
0
0.433
< 0.05
Positive
Population Aged 65 Years and Above
1996
32
0.337
< 0.05
Positive
Indigenous Population
1242
0
0.650
< 0.05
Positive
Foreign Labors
581
0
0.244
< 0.05
Positive
Low-Income Households
110
0
0.488
< 0.05
Positive
Population Without High School Diploma
1845
30
0.706
< 0.05
Positive
163 164
2.3. Spatial Autocorrelation Analysis
165 166 167 168 169 170
This research applies global and local spatial autocorrelation to investigate spatial distribution of SoVI. Moran’s I is a global scale indicator for measuring spatial autocorrelation and identifying the spatial clusters along Yilan County. On the other hand, the Local Indicator of Spatial Autocorrelation (LISA) can measure the local association between variables and the spatially weighted value of the same variables. It can also visualize the sites of the cluster and outlier. The equations of Moran’s I and LISA are demonstrated as following [39-41]: Moran’s I =
∑ ∑ ij
LISA =
×
∑ ∑ ij( )( )
∑ ( )
( ) ∑ ij
∑ ( )
−
(1) (2)
171 172 173 174 175 176
Where N is the number of spatial units. Wij is the spatial weighted matrix. Xi and Xj represent the value of variables. is the mean value of variables. Through the Moran’s I, we can identify the spatial pattern of SoVI, whether it’s a cluster or not. Also, whether the cluster is statistically significant. Moreover, through LISA, we can visualize the distribution of the cluster. The result of LISA will present the spatial distribution of the cluster on a map.
177
2.4. Geographically Weighted Regression (GWR)
178 179 180 181 182 183 184 185 186
Geographically Weighted Regression (GWR) is a spatial multivariate regression. It can explore not only the relationship between SoVI and real environmental hazards but also the investigation of the effectiveness of SoVI. Compared to other spatial multivariate regression such as Ordinary Least Squares (OLS), Spatial Lag Model (SLM), and Spatial Error Model (SEM), GWR can solve the non-stationarity by separating study area into several sub-study areas through generating several kernels [42,43]. So, GWR can represent the authentic situation within the study area and has a better explanatory ability [43]. Based on the previous research [43,44], the variable bandwidth of kernel is relatively suitable for clustered data. As a result, the kernel type in this research is “adaptive” [42,43]. The equations of GWR are demonstrated as below [42-44].
6 of 15
GWR: () = () + () + () + ! ()! + ⋯ + # ()# # () = ( % ()) % ()
' () 0 () = & 0 0
0 ' () 0 0
'* () = ,
⋱ 0
0 0
0 (2) ../ 1 4 3
0 0 + 0 '* ()
(3) (4)
(5)
(6)
187 188 189 190 191
Where Xmi is the independent variables, while # () represents the coefficient. The coefficient is calculated by the matrix of independent variables (Xmi) and the matrix of weights specific to location u. W(u) is the weighted matrix in the model calculated by Wn(u). The dn(u) is the distance between the nth observation and the location of u. In addition, the h is the radius of the kernel.
192
3. Results and Discussion
193
3.1. PCA and SoVI
194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224
The global Moran’s I is applied to identify the spatial distribution of all variables. The Moran’s I and p-value show that all the variables have statistically significant clusters (Table 1). Moreover, we conduct the principal component analysis to reduce these variables. The KMO of PCA is 0.647. Also, Bartlett’s Test of Sphericity is highly significant (p-value <0.01). The result is acceptable. Based on the eigenvalue, PCA compresses these 12 variables into four principal components (Fig 2). These four components explain 77.0% of the total variables. The first component, population factor, includes the Population Without High School Diploma, Aging Index, Female Population, Solitary Elderly, Physical and Mental Disorder, and Population Aged 65 Years and Above variables. The second component, economic factor, composes of the Indigenous Population, Low-Income Households, and Population Under 5 Years Old variables. The third principal component, social factor, includes the Population Density and Dependency Ratio variables. The last component, labor factor, includes the Foreign Labors variable only. These four principal components can individually interpret the total variables as follows: population factor (28.6%), economic factor (24.0%), social factor (15.5%), and labor factor (8.9%). Fig 3 and Table 2 demonstrate the distribution of four components and the result of global spatial autocorrelation analysis. The Moran’s I shows that all factors are clustered. According to the p-value, the clusters are statistically significant. The majority of population component congests in the mountain areas, but some of them also distribute into coastal areas. Compared to other components, the distribution of the population component does not have a specific pattern. Some of them distribute into mountain areas while others distribute into coastal areas. On the other hand, most of the economic component has a conspicuous cluster in mountain areas. One of the most important reasons is that the level of development in mountain areas is not very well. Due to the unfavorable topographic feature and inaccessibility compared to plain areas, mountain areas have far fewer economic resources and opportunities. As a result, the majority of economic component assembles in mountain areas. The social component highly clusters in the plain areas because there are the most rapid development areas in Yilan County, especially for urban areas such as Yilan City and Luodong Town. Besides, some communities also have high values in the social component. Some of them are located in essential nodes of traffic networks such as the Chauyan, Tianf, and Tianshan communities. Finally, some communities with a high value of the labor component are located in the surrounding areas of Su'ao Harbor, such as Nanning and Nanxing communities. In Yilan County, foreign labors can be classified into two main categories, nursing labors, and fishery, and maritime labors. Fishery and maritime labors highly cluster in the surrounding areas of Su'ao Harbor. However, the distribution of nursing labors is far dispersed.
7 of 15
225 226
Fig. 2. The scree plot and eigenvalue of PCA. Based on the eigenvalue and scree pot, PCA generates four principal components.
227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244
Fig 4 (a) shows the spatial distribution of SoVI in the study area and result of global spatial autocorrelation analysis. Totally, there are 233 communities in Yilan County. Based on the analysis of SoVI, there are 20 communities (8.6%) with Very High levels of SoVI, while there are 42 communities (18.0%) with a High level of SoVI. Over a quarter of communities (26.6%) in Yilan County is classified as a high social vulnerability class. Most of them cluster in mountain areas. The reason for this special distribution is owing to the characteristics of mountain areas. The topographic feature and inaccessibility cause mountain areas becoming unsuitable areas for economic development. Also, some of the communities are also highly incapability. Fig 4 (b) shows the result of local spatial autocorrelation analysis. Most communities located in mountain areas are highly social vulnerable with the H-H cluster. However, most of the low SoVI communities aggregate in the center of plain areas with the L-L cluster. This reveals that the socioeconomic status in plain areas is significantly better than mountain areas. The flat topography and convenient accessibility are very suitable for economic development. Moreover, the revision of the Agricultural Development Act also accelerates the transformation of land use. The traffic connection between Yilan County and Taipei City is quickly increasing after the construction of the freeway, especially in plain areas. Therefore, compared to mountain areas, plain areas have a better socioeconomic status to reduce social vulnerability. Table 2 The global autocorrelation analysis of all factors
245
Component
Moran’s I
p-value
Population factor
0.335
< 0.05
Economic factor
0.584
< 0.05
Social factor
0.594
< 0.05
Labor factor
0.224
< 0.05
8 of 15
246 247 248 249
Fig. 3. Spatial distribution of the four components derived from PCA and the result of global spatial autocorrelation analysis: (a) population factor; (b) economic factor; (c) social factor; (d) labor factor.
9 of 15
250 251 252
Fig. 4. The distribution and cluster map of SoVI (a) the spatial distribution of SoVI and result of global spatial autocorrelation analysis; (b) the result of LISA.
253
3.2. GWR
254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276
The variables applied to construct SoVI are highly collinearity. The raw variables cannot take as the independent variables. PCA is one of the most common approaches to deal with collinearity. PCA can compress similar variables into a principal component. Through this approach, PCA can eliminate the collinearity within the variables. The independent variables we used are the principal component generated by PCA, including (a) population factor; (b) economic factor; (c) social factor; and (d) labor factor. In order to validate SoVI, we apply GWR to conduct this analysis. The dependent variable applied in this research is the combination of flood and debris flow events shown in Fig 5. These events represent the most common environmental hazards in Yilan County. The occurrence of environmental hazards is selected due to the scale of data. NGIS provides the detail occurrence of environmental hazards in Yilan County, and all the data is recorded on the community scale. Moreover, based on the literature review, the SoVI performs well for the occurrence of environmental hazards [45]. The potential loss is highly related to the occurrence of environmental hazards. Therefore, we select the occurrence of flood and debris flow events as the dependent variables. The kernel type we applied is “Adaptive.” The bandwidth method is “BANDWIDTH PARAMETER”. Based on the results of GWR in Table 3, the R2 is 0.769, and the Adjusted R2 is 0.577 for the whole county scale. This value is relatively high and seems acceptable. Fig 6 shows the standardized residual of GWR. About 142 (60.9%) of the communities show the insignificant residual (-0.5SD to 0.5SD). In order to ensure the result of the model is not biased, we apply the global spatial autocorrelation analysis to examine the standardized residual of GWR. If the result of GWR is not biased, the spatial autocorrelation analysis of the standardized residual will show no significant spatial autocorrelation. The Moran’s I is 0.02, and the p-value is 0.526. The result of the global spatial autocorrelation shows the distribution of standardized residuals is random. In other words, the SoVI is not biased. The R2 and the Moran’s I of the standardized residual reveals that SoVI is a valid indicator in this study, and the goodness of fit for this model is statistically adequate.
10 of 15
277
278 279 280
Fig. 5. The spatial distribution of standardized flood and debris flow events. Table 3 The results of GWR Neighbors
R2
Adjusted R2
30
0.769
0.577
AICc
StdResid Moran’s I
p-value
594.713
0.02
0.526
StdResid: Standardized residual
281
11 of 15
282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303
Fig. 6. The distribution of GWR’s standardized residual.
Fig 7 shows the coefficient of four components since GWR separates Yilan County into several sub-areas through creating many kernels with different bandwidth. The coefficient of component is highly variable in spatial distributions. We categorize the coefficient by the natural break. The GWR could be used to explore the spatial relationship between communities directly affected by environmental hazards and localities with higher or lower SoVI values [46]. Fig 7a shows the coefficient of the first component, the population factor. There are 77 (33.0%) of the communities show a positive correlation to environmental hazards. The urban areas such as Yilan City and Luodong Town show a significant positive correlation. Also, the near mountain areas, Sanxing, Yuanshan, and Datong Townships, show a positive correlation, while the southwestern mountain areas and northern coastal areas show a negative correlation to environmental hazards. Fig 7b shows the coefficient of the second component, the economic factor. There are 114 (48.9%) of the communities show a positive correlation to environmental hazards. Especially for the populated urban areas, where located at the center of the study areas such as Yilan City and Luodong Town, show a significantly positive correlation to the environmental hazards. However, the coastal areas of Wujie and Zhuangwei Townships show a negative correlation to environmental hazards. Fig 7c shows the coefficient of the third component, the social factor. In General, most communities show a negative correlation. Only 25 (10.7%) of the communities show a positive correlation in relative dispersed distribution. Fig 7d shows the coefficient of the last component, the labor factor. Totally, 144 (61.8%) of the communities in our study area show a positive correlation. In Only coastal areas, the northern and middle parts show a negative correlation. In mountain areas, only 4 communities show a negative correlation.
12 of 15
304 305 306
Fig. 7. Coefficient of the four components: (a) population factor; (b) economic factor; (c) social factor; and (d) labor factor.
307 308
4. Conclusions
309 310 311 312 313 314 315
In conclusion, this research provides not only a holistic exploration but also a validation of social vulnerability to environmental hazards in Yilan through SoVI and spatial analysis. Social vulnerability is a crucial concept for disaster prevention and damage prediction. The visualization of SoVI is essential for decision-makers to formulate a thorough emergency plan. Moreover, this research also validates the effectiveness of the concept of social vulnerability to environmental hazards. Irrefutably, SoVI is a simple, adaptable, flexible, and widely used tool for assessing, quantifying, and visualizing social vulnerability to environmental hazards. Nevertheless, validation is still important. As a result, we apply the GWR to
13 of 15
316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345
investigate the effectiveness of the SoVI. The result shows that the SoVI in this research is a valid and unbiased indicator. Owing to the different context in various places, we should modify the variables of SoVI to adapt to the context in Taiwan. Totally, there are 12 variables selected. Based on SoVI and spatial autocorrelation analysis, we quantify, visualize, and analyze the distribution of social vulnerability to environmental hazards in Yilan County. Results show that most of the communities in mountain areas have a higher level of SoVI. In other words, susceptible groups cluster in mountain areas mostly. When encountering environmental hazards, the potential loss in mountain areas is relatively higher. The topography of our study area is relatively complex, including mountain and plain areas. These features allow us to take account of the interaction between the social fabric and natural characteristics. We found the topography characteristics and accessibility are the most significant factor. In mountain areas, the inaccessibility is triggered by topography characteristics and leads to a low level of development. The mountain areas have far fewer economic resources and opportunities. These factors lead the mountain areas with high social vulnerability. Previous researches also show that social vulnerability to environmental hazards in mountain areas is higher than in plain areas. Therefore, for the decision-maker, the main goal of hazard mitigation in mountain areas should reduce the level of social vulnerability in order to minimize potential loss. Moreover, the decision-maker can also pre-evacuate the residents in mountain areas when forecasting serious typhoon coming to avoid potential damage. On the other hand, as for the plain areas, the decision-maker should focus on the dense population areas, the urban areas. The high population density also means a large potential exposure to environmental hazards. Although the residents in plain areas have relatively better socioeconomic status and lower social vulnerability than mountain areas, it only means the potential loss to environmental hazards is relatively insignificant compared to mountain areas. When encountering environmental hazards, the plain areas still have a chance to face the damage brought by natural hazards. Thus, for the plain areas, formulating a plan or project for emergency response and evacuation is also very critical. The measurement of social vulnerability is essential and shouldn't be ceased in the future works, especially under the setting of climate change. We should take the measurement of social vulnerability as the basis, and focus on the ability to coevolve with the natural hazards especially for those areas with higher social vulnerability. In order to cope with the potential stronger environmental hazards, the ability to be more resilient should be the focus in future work.
346
References
347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367
[1] Intergovernmental Panel on Climate Change (IPCC), Climate Change 2001: Impacts, Adaptation, and Vulnerability. Contribution of Working Group II to the Third Assessment Report of the Intergovernmental Panel on Climate Change; J.J. McCarthy, O.F. Canziani, N.A. Leary, D.J. Dokken, K.S. White, Eds, Cambridge University Press. Cambridge, UK https://www.ipcc.ch/report/ar3/wg2/. [2] Intergovernmental Panel on Climate Change (IPCC), Climate Change 2007: Synthesis Report. Contribution of Working Groups I, II and III to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change; The Core Writing Team, R.K. Pachauri, A. Reisinger, Eds, IPCC. Geneva, Switzerland https://www.ipcc.ch/report/ar4/syr/. [3] Intergovernmental Panel on Climate Change (IPCC), Climate Change 2014: Synthesis Report. Contribution of Working Groups I, II and III to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change; The Core Writing Team, R.K. Pachauri, L.A. Meyer, Eds, IPCC. Geneva, Switzerland https://archive.ipcc.ch/report/ar5/syr/. [4] S.L. Cutter, B.J. Boruff, W.L. Shirley, Social vulnerability to environmental hazards, Soc. Sci. Q. 84(2) (2003) 242-261. https://doi.org/10.1111/1540-6237.8402002. [5] S.L. Cutter, Vulnerability to environmental hazards, Prog. Hum. Geog. 20(4) (1996) 529-539 https://doi.org/10.1177%2F030913259602000407. [6] W.N. Adger, N. Brooks, M. Kelly, G. Bentham, M. Agnew, S. Eriksenet, New indicators of vulnerability and adaptive capacity. Tyndall Centre for Climate Change Research, Norwich, UK, 2004 http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.112.2300&rep=rep1&type=pdf. [7] W.N. Adger, Vulnerability, Global. Environ. Chang. 16(3) (2006) 268-281 https://doi.org/10.1016/j.gloenvcha.2006.02.006.
14 of 15
368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418
[8] T.G. Frazier, C.M. Thompson, R.J. Dezzani, A framework for the development of the SERV model: A Spatially Explicit Resilience-Vulnerability model, Appl. Geogr. 52 (2014) 158-172 https://doi.org/10.1016/j.apgeog.2014.04.004. [9] I. Frigerio, S. Ventura, D. Strigaro, M. Mattavelli, M.D. Amicis, S. Mugnano, M. Boffi, A GIS-based approach to identify the spatial variability of social vulnerability to seismic hazard in Italy, Appl. Geogr. 74 (2016) 12-22 https://doi.org/10.1016/j.apgeog.2016.06.014. [10] J.I. Uitto, The geography of disaster vulnerability in megacities: A theoretical framework, Appl. Geogr. 20(1) (1998) 7-16 https://doi.org/10.1016/S0143-6228(97)00041-6. [11] B. Wisner, P. Blaikie, T. Cannon, I. Davis, At Risk: Natural Hazards, People's Vulnerability and Disasters, 2nd ed, Routledge, Oxfordshire, UK, 2003 https://www.preventionweb.net/files/670_72351.pdf. [12] J. Birkmann, Risk and vulnerability indicators at different scales: Applicability, usefulness and policy implications, Environ. Hazards. 7(1) (2006) 20–31 https://doi.org/10.1016/j.envhaz.2007.04.002. [13] A. Fothergill, E.G.M Maestas, J.D Darlington, Race, ethnicity, and disasters in the United States: a review of the literature, Disasters. 23(2) (1999) 156-173 https://doi.org/10.1111/1467-7717.00111. [14] L. Rygel, D. O'Sullivan, B. Yarnal, A method for constructing a social vulnerability index: An application to hurricane storm surges in a developed country, Mitig. Adapt. Strat. Gl. 11(3) (2006) 741-764 https://doi.org/10.1007/s11027-006-0265-6. [15] S.L. Cutter, C. Finch, Temporal and spatial changes in social vulnerability to natural hazards, Proc Natl Acad Sci Usa. 105 (7) (2008) 2301-2306 https://doi.org/10.1073/pnas.0710375105. [16] A. Fekete, Validation of a social vulnerability index in context to river-floods in Germany, Nat. Hazard. Earth. Sys. 9(2) (2009) 393-403 https://doi.org/10.5194/nhess-9-393-2009. [17] N.J. Wood, C.G. Burton, S.L. Cutter, Community variations in social vulnerability to Cascadia-related tsunamis in the U.S. Pacific Northwest, Nat. Hazards. 52(2) (2010) 369-389 https://doi.org/10.1007/s11069-009-9376-1. [18] E. Tate, Social vulnerability indices: a comparative assessment using uncertainty and sensitivity analysis, Nat. Hazards. 63(2) (2012) 325-347 https://doi.org/10.1007/s11069-012-0152-2. [19] S. Khan, Vulnerability assessments and their planning implications: a case study of the Hutt Valley, New Zealand, Nat. Hazards. 64(2) (2012) 1587-1607 https://doi.org/10.1007/s11069-012-0327-x. [20] S.F. Balica, N.G. Wright, F. van der Meulen, A flood vulnerability index for coastal cities and its use in assessing climate change impacts, Nat. Hazards. 64(1) (2012) 73–105 https://doi.org/10.1007/s11069-012-0234-1. [21] Q. Dinh, S. Balica, L. Popescu, A. Jonoski, Climate change impact on flood hazard, vulnerability and risk of the Long Xuyen Quadrangle in the Mekong Delta, Int. J. River. Basin. Manag. 10(1) (2012) 103–120 https://doi.org/10.1080/15715124.2012.663383. [22] M.K.A. Kablan, K. Dongo, M. Coulibaly, Assessment of social vulnerability to flood in urban Côte d’Ivoire using the MOVE framework, Water. 9(4) (2017) 292-311 https://doi.org/10.3390/w9040292. [23] T. Morimoto, Spatial analysis of social vulnerability to floods based on the MOVE framework and information entropy method: Case study of Katsushika Ward, Tokyo, Sustainability. 11(2) (2019) 529-548 https://doi.org/10.3390/su11020529. [24] L.S. Hung, C. Wang, B. Yarnal, Vulnerability of families and households to natural hazards: A case study of storm surge flooding in Sarasota County, Florida, Appl. Geogr. 76 (2016) 184-197 https://doi.org/10.1016/j.apgeog.2016.09.021 [25] I.Frigerio, M.D. Amicis, Mapping social vulnerability to natural hazards in Italy: A suitable tool for risk mitigation strategies, Environ. Sci. Policy. 63 (2016) 187-196 https://doi.org/10.1016/j.envsci.2016.06.001 [26] J.S. Lee, H.I. Choi, Influence of Construction Schemes for a Non-compensatory Composite Indicator on Flood Vulnerability Assessments in the Korean Peninsula, Int. J. Civil. Eng. 17(8) (2018) 1–9 https://doi.org/10.1007/s40999-018-0382-0 [27] Y.J. Lee, Social vulnerability indicators as a sustainable planning tool, Environ. Impact. Asses. 44 (2014) 31-42 https://doi.org/10.1016/j.eiar.2013.08.002 [28] L.F. Chang, S.L. Huang, Assessing urban flooding vulnerability with an emergy approach, Landscape, Urban. Plan. 143 (2015) 11-24 https://doi.org/10.1016/j.landurbplan.2015.06.004
15 of 15
419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467
[29] T.H. Siagian, P. Purhadi, S. Suhartono, H. Ritonga, Social vulnerability to natural hazards in Indonesia: driving factors and policy implications, Nat. Hazards. 70(2) (2014) 1603-1617 https://doi.org/10.1007/s11069-013-0888-3 [30] S. Khan, Vulnerability assessments and their planning implications: a case study of the Hutt Valley, New Zealand, Nat. Hazards. 64(2) (2012) 1587-1607 https://doi.org/10.1007/s11069-012-0327-x [31] E. Mavhura, B. Manyena, A.E. Collins, An approach for measuring social vulnerability in context: The case of flood hazards in Muzarabani district, Zimbabwe, Geoforum. 84 (2017) 103-117. [32] M. Dzoga, D. Simatele, C. Munga, Assessment of ecological vulnerability to climate variability on coastal fishing communities: A study of Ungwana Bay and Lower Tana Estuary, Kenya, Ocean. Coast. Manag. 163, (2018) 437–444 https://doi.org/10.1016/j.ocecoaman.2018.07.015 [33] B. Jones, J. Andrey, Vulnerability index construction: methodological choices and their influence on identifying vulnerable neighbourhoods, IJEM. 4(2) (2007) 369-395 https:/doi.org/10.1504/IJEM.2007.013994 [34] C. Wang, B.Yarnal, The vulnerability of the elderly to hurricane hazards in Sarasota, Florida. Nat. Hazards. 63(2) (2012) 349-373 https://doi.org/10.1007/s11069-012-0151-3 [35] Ministry of Health and Welfare, R.O.C. (Taiwan), Report of the Women’s Living Condition Survey 2015, https://www.mohw.gov.tw/dl-30768-11e9c325-3c6c-4079-8870-124fa8c5e10d.html [36] Council of Indigenous Peoples, R.O.C. (Taiwan), Economic Status Survey of Indigenous Peoples in Taiwan 2014, https://www.apc.gov.tw/portal/getfile?source=2D838540F5D6F659FAFB9859EF31AC3B381A272F479D65 D98D902DFAAFC2E154084F20DD5F0F37964575BA3D2CADEAE87E08C4C4F83F6C7D3B91B9DF716 59F0C&filename=D0572478D51EFA60DD66604754BE10B807F619060818CEA0D0636733C6861689. [37] Council of Indigenous Peoples, R.O.C. (Taiwan), Annual report of Population and Health Condition of Indigenous Peoples 2015. https://www.apc.gov.tw/portal/getfile?source=2D838540F5D6F659FAFB9859EF31AC3B381A272F479D65 D98D902DFAAFC2E1547AEE5B785E323204555BE00A12C2A71E41B16B2642EE8A363B91B9DF71659 F0C&filename=A656B8F1D8AB64973D3CD606F117241350E32B03FBDD42DA88CAD01538D4794C8E 54517BBFE816C6AC1C5A2019A292E815E6A529B9D3427D. [38] Y.C. Chang, C.L. Yang, An Analysis of the Indigenous Peoples, Directorate General of Budget, Accounting and Statistics, Executive Yuan, R.O.C. (Taiwan), https://www.dgbas.gov.tw/public/Attachment/56115154ZX6CXRZE.pdf. [39] P.A.P Moran, The interpretation of statistical maps, J. R. Stat. Soc. B. 10 (1948) 243-251 https://doi.org/10.1111/j.2517-6161.1948.tb00012.x [40] P.A.P Moran, Notes on continuous stochastic phenomena, Biometrika. 37 (1950) 17-23 https://doi.org/10.1093/biomet/37.1-2.17 [41] L. Anselin, Local indicators of spatial association-Lisa, Geogr Anal. 27(2) (1997) 93-11 https://doi.org/10.1111/j.1538-4632.1995.tb00338.x [42] A.S. Fotheringham, M.E. Charlton, C. Brunsdon, Geographically weighted regression: a natural evolution of the expansion method for spatial data analysis, Environ. Plann. A. 30(11) (1998) 1905-1927 https://doi.org/10.1068%2Fa301905 [43] D.K. Yoon, J.E. Kang, S.D. Brody, A measurement of community disaster resilience in Korea, J. Environ. Plann. Man. 59(3) (2016) 436-460 https://doi.org/10.1080/09640568.2015.1016142 [44] A.S. Fotheringham, C. Brunsdon, M. Charlton, Geographically Weighted Regression: The Analysis of Spatially Varying Relationship, Wiley. New York, USA, (2003)https://www.wiley.com/en-us/Geographically+Weighted+Regression%3A+The+Analysis+of+Spatially +Varying+Relationships+-p-9780470855256 [45] L.A. Bakkensen, C.F. Lent, L.K. Read, I. Linkov, Validating resilience and vulnerability indices in the context of natural disasters, Risk. Anal. 37(5) (2016) 982-1004 https://doi.org/10.1111/risa.12677 [46] I. Török, Qualitative assessment of social vulnerability to flood hazards in Romania, Sustainability 10, (2018) 3780 https://doi:10.3390/su10103780
A GIS-based Approach for Assessing Social Vulnerability to Flood and Debris Flow Hazards Chien-Hao Sung 1, Shyue-Cherng Liaw 1,* 1 Department of Geography, National Taiwan Normal University, Taipei 10610, Taiwan. * Correspondence: [email protected]; Tel.: +886-2-7734-1649
Conflicts of Interest: The authors declare no conflict of interest. Shyue-Cherng Liaw Prof. of Geography, NTNU, Taiwan
S2212-4209(19)31379-2
DOI:
https://doi.org/10.1016/j.ijdrr.2020.101531
Reference:
IJDRR 101531
To appear in:
International Journal of Disaster Risk Reduction
Received Date: 3 October 2019 Revised Date:
16 January 2020
Accepted Date: 15 February 2020
Please cite this article as: C.-H. Sung, S.-C. Liaw, A GIS-based approach for assessing social vulnerability to flood and debris flow hazards, International Journal of Disaster Risk Reduction (2020), doi: https://doi.org/10.1016/j.ijdrr.2020.101531. This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. © 2020 Published by Elsevier Ltd.
1 2 3 4 5
A GIS-based Approach for Assessing Social Vulnerability to Flood and Debris Flow Hazards Chien-Hao Sung 1, Shyue-Cherng Liaw 1,* 1
Department of Geography, National Taiwan Normal University, Taipei 10610, Taiwan. * Correspondence: [email protected]; Tel.: +886-2-7734-1649
6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
Abstract: Owing to the escalating environmental hazards caused by climate change, the mitigation of disaster becomes extremely important. The investigation of social vulnerability is a prerequisite for formulating a mitigation plan to environmental hazards. This research applies a GIS-based approach with the Social Vulnerability Index (SoVI) to investigate and quantify the social vulnerability to environmental hazards in Yilan County, Taiwan. In order to construct the SoVI, the literature review was conducted, and 12 variables were selected. Through Principal Component Analysis (PCA), the 12 variables were reduced into four principal components. In order to explore the spatial pattern of SoVI, the spatial autocorrelation analysis was applied. The result showed that there were 26.5% of communities in Yilan County with a high level of SoVI, and most of these communities were mainly located in mountain areas. The unfavorable topography features cause the distributions in mountain areas. On the other hand, there were 37.3% communities with a related low level of SoVI, and these communities were located in plain areas. The inaccessibility caused by topography creates an incapability, resource-lacking environment and lead to a high value of SoVI. In addition, this research applied Geographically Weighted Regression (GWR) to validate SoVI, and the result of the R2 value was 0.769. Also, the standardized residuals showed no spatial autocorrelation, meaning the SoVI had the adequate explanatory ability. This research provided a set of valid indicators to explore the social vulnerability for decision-makers to formulate the mitigation plan of environmental hazards. Besides, SoVI is a suitable tool for visualizing and quantifying the potential loss to environmental hazards.
23 24
Keywords: Social vulnerability, SoVI, Spatial autocorrelation, Geographically weighted regression
25
1. Introduction
26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49
Since the late 20th century, the frequency of extreme climate events has increased dramatically, and also augments the intensity and damage to flood and debris flow [1-3]. As a result, environmental hazards reduction and mitigation have become a critical task for the government. Therefore, one of the most important requisite analysis is the potential loss caused by environmental hazards. Over the last two decades, vulnerability is becoming one of the essential terms for the analysis of potential losses caused by environmental hazards. Visualizing and quantifying vulnerability enable the decision-maker to recognize the vulnerable areas and forecast potential losses. According to [4,5], the vulnerability composes of social vulnerability and biophysical vulnerability and can be represented as potential losses to environmental hazards [4-7]. The social vulnerability can be regarded as the partial outcome of social inequalities [4]. In other words, social factors have shaped, influenced, or govern the ability of susceptible groups to respond when they face environmental hazards. Social factors include a lot of variables, such as gender, ethnicity, age, socioeconomic status, and education, etc [4-7]. General speaking, environmental hazards do not randomly affect society [8,9]. Usually, the groups that already marginalized by unfavorable socioeconomic status are the most likely to be jeopardized by environmental hazards [10]. When encountering environmental hazards, these groups are susceptible to the impacts [4,11-13]. Most of the researches apply the concept of social vulnerability to environmental hazards to quantify and investigate the susceptible groups [14-19]. On the other hand, biophysical vulnerability represents the geographic context of an area [5]. Based on the different locations, the biophysical vulnerability will also differ from place to place. For example, due to the topographic characteristics, the coastal areas are often affected by storm surges and floods. As a result, two of the most common factors affecting the biophysical vulnerability are the storm surge and flood in coastal areas under climate change [20-24]. Although the geographic context varies from place to place, undoubtedly, the concept of vulnerability to environmental hazards is suitable for around the world since it has been applied worldwide, such as Europe [9,25], America[4,5], Asia[26-29], Oceania[30], and Africa[31,32].
2 of 15
50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75
Both the social and biophysical vulnerabilities are related to the potential loss induced by environmental hazards. However, it does not mean that exposure to biophysical hazards will determinedly increase vulnerability [9,33]. Those who affected by the result of social inequalities are the most vulnerable groups since they are marginalized by the disadvantage socioeconomic status [10]. Therefore, social vulnerability is the prerequisite for assessing vulnerability to environmental hazards. In addition, the spatial scale is another crucial factor [8]. For example, the investigation can be conducted on a national, county, and sub-county scale. However, some scales are far inadequate for investigation. Because the result of the investigation on the national or regional scale is too general to offer a specific analysis, the majority of researches is conducted in the county or even sub-county scale [4,8]. The Yilan County is the administrative district with the largest plain area in eastern Taiwan. It is also the most susceptible area to environmental hazards due to the lack of topographic protection from the typhoons by the Pacific Ocean. In this decade, environmental hazards including floods and debris flow have seriously caused tremendous damage to Yilan County. Thus, a holistic investigation of social vulnerability is urgently required for the county government to formulate a thorough mitigation plan. So, the primary purpose of this research is to quantify social vulnerability to environmental hazards in Yilan County with the latest data. Despite the fact that social vulnerability has been applied in many various regions based on the different social and geographic contexts, the content of social vulnerability needs to be modified to fit the specific context. Therefore, this research aims to construct a set of Social Vulnerability Indicators (SoVI) based on the study area’s context. Comparing with other research, some SoVI variables from previous studies are not included in this study. One of the major reasons is owing to the data accessibility. Some variables are inaccessible since it is based on the Computer-Processed Personal Data Protection Law in Taiwan. In addition, some of the social contexts in other countries don't suitable in our study areas. For example, the variable of mobile homes is very important for environmental hazards mitigation [4]; however, there are no mobile homes in our study area. Moreover, this research applies the data with community-scale in order to investigate the variation of SoVI spatially. This research also uses spatial autocorrelation analysis to explore the distribution pattern.
76
2. Materials and Methods
77
2.1. Yilan County
78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95
The study area of this research is Yilan County, where is the most developed county in eastern Taiwan (Fig 1). Its’ population is about 460 thousand people. There are 12 townships with 233 communities. The total area of Yilan County is 2,144 km2, and the elevation is from sea level up to the highest 3,589 m. The mean elevation of Yilan County is 830 m. The southwestern part is a mountainous area, while the northeastern part is plain and the most populated area. Compared to other counties, Yilan County has freeway and railroad connections to the capital of Taiwan, Taipei. After the revision of the Agricultural Development Act, it is allowing the rapid and large scale of land development. Therefore, Yilan County has also undergone a significant industrial transformation and economic development within the last decade. Due to the location, accessibility, and topography, Yilan County undoubtedly has the best status of further development in eastern Taiwan. However, owing to the geographical location, Yilan County is also the first and foremost county that encounters the typhoon and tropical cyclone coming from the western Pacific Ocean. According to the Central Weather Bureau of Taiwan, annually, Taiwan will have to face at least 4 typhoons. Owing to the tremendous amount of precipitation brought by the typhoon and tropical cyclone, the floods become the most notorious natural hazard events in Yilan County. Moreover, because of the steep topography and fragile geology, the debris flow also frequently occurs in our study area. These natural characteristics had created a tremendous amount of floods and debris flow. Recently, due to climate change, the situation had become much more severe and created uncountable damage.
3 of 15
96 97 98
Fig. 1. The distribution of elevation, transportation path, major townships, and the most common path of the typhoon that affect study area.
99
2.2. SoVI Variable Selection
100 101 102 103 104 105 106 107 108 109 110
Based on the difference of context and the scale of data, the SoVI is a highly modifiable tool for quantifying social vulnerability to environmental hazards. The SoVI can be applied in various kinds of scale and selects different kinds of socioeconomic variables such as age, population density, gender, ethnicity, financial support, and education [4-7]. Age can represent the physical ability for movement or mobility. Elderly residents and infants have the possibility to increase the difficulty of evacuation [8,34]. Therefore, we take the population that is older than 65 years old or younger than 5 years old as the variables. Also, because of the convenient freeway, Yilan County and Taipei City are highly connected. Taipei City is the capital of Taiwan, which has more employment opportunities and pulls the population migration from Yilan County. Thus, solitary elderly become extremely common in our study area. The solitary elderly is exceptionally vulnerable to environmental hazards since the majority of them are unemployed and lack financial resource [34]. Thus, we select the solitary elderly as one of our variables. Moreover, population density can stand for
4 of 15
111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162
potential exposure to environmental hazards and can also be regarded as the requirement of resources during or after the natural disaster [6]. Thus, we take the population density as one of our variables. Considering gender issue, females would be physically more vulnerable than males and could also encounter hazards more difficult during the recovery phase due to unequal wages, employment, and family responsibilities [24]. According to the Report of the Women’s Living Condition Survey 2015 in Taiwan [35], which conducted by the Health and Welfare Department, 44.4% of the female population is unemployed between the age of 15 to 64. The reason caused this situation is mainly owing to “household tasks” and “taking care of children” [35]. Therefore, when facing an environmental hazard, the female might not have sufficient financial resources to cope with it. Undoubtedly, gender is one of the debatable variables, yet, in Taiwan's social context, and women certainly have a significant difficulty when facing natural hazards. Consequently, we select the female population as one of our variables. Ethnicity, one of the most common variables, can represent the language and culture barrier. For specific ethnicities, foreigners and indigenous peoples, they are considered as the disadvantaged group of the social network, education level, information source, and financial resource. They might lead to erroneous decision-making and catalyst potential loss [4,6,8]. Most of the foreign labors in Taiwan are manual labor and engaged in the most laborious physical job with a limited payment and education. Without a doubt, when facing natural hazards, they are surely the susceptible group. The indigenous population, in Taiwan’s context, is one of the most common underprivileged groups. According to the Economic Status Survey of Indigenous Peoples 2014 in Taiwan [36], which conducted by the Council of Indigenous Peoples, the gross annual income of an average indigenous family was 658,117 NTD while the gross annual income of an average family was 1,071,427 NTD. Moreover, based on the Annual Status Report of Indigenous Population and Health Condition 2015 in Taiwan [37], which also conducted by the Council of Indigenous Peoples, the life expectancy of the indigenous peoples is shorter by 8.3 years compared to a normal citizen. The reason caused this situation is the inequalities of health and medical resources. Furthermore, generally speaking, most of the indigenous peoples have lower education [38]. Therefore, indigenous people’s situation is extremely vulnerable when they are facing an environmental hazard. Thus, we select the foreign labors and indigenous population as the variables. Financial support and education status also play a significant role. Lack of financial resources and lower education levels could mean the potentially insufficient information, lacking mobility, and inadequate preparedness ability [4-7,24,34]. Consequently, we take the population without a high school diploma and low-income households as variables. Moreover, Taiwan is one of the areas that have the lowest birth rate in Asia. In other words, population aging is already a significant issue. Under this circumstance, a family will have more burden [24,34]. Thus, we use the dependency ratio and the aging index as the variables. We also adopt the physical and mental disorder as one of the variables since the majority of them will require extra resources to take care of. When facing natural hazards, undeniably, they are the disadvantaged groups and will become a burden for their families. The scale of data also plays an essential role. In order to construct a valid set of SoVI to explore the social vulnerability at the community level, we select 12 variables based on the literature review and data accessibility. These variables include Population Density, Female Population, Dependency Ratio, Solitary Elderly, Physical and Mental Disorder, Aging Index, Population Under 5 Years Old, Population Aged 65 Years and Above, Indigenous Population, Foreign Labors, Low-Income Households, and Population Without High School Diploma (Table 1). All the variables were community-scale data and obtained from the database of the National Geographic Information System (NGIS). We standardize the variables due to their different units. The standardization rescales the variables with a zero mean and a standard deviation of one. Then we apply the Principal Component Analysis (PCA) to compress the variables into a smaller set. Also, PCA can reduce multi-collinearity and redundant data. Based on the Kaiser-Meyer-Olkin (KMO) and Bartlett’s Test of Sphericity, we can identify the result of PCA, whether it is acceptable. In order to increase the interpretability of the factors, we apply the Varimax rotation with Kaiser Normalization, which is the most common approach to extracting and interpreting the components. Then, the factors with an eigenvalue larger than 1 are extracted. Finally, we classify SoVI into five categories based on the level of standard deviation (SD) including very high (> 1.5 SD), high (1.5 to 0.5 SD), medium (0.5 to -0.5 SD), low (-0.5 to -1.5 SD), and very low(< -1.5 SD).
5 of 15
Table 1 Variables and descriptive statistics Variables
Max
Min
Moran's I
p-value
Population Density
30024.25
0.34
0.634
< 0.05
Effects on vulnerability to hazards Positive
Female Population
3288
22
0.582
< 0.05
Positive
Dependency Ratio
56.32
26.97
0.420
< 0.05
Positive
Solitary Elderly
14
0
0.197
< 0.05
Positive
Physical and Mental Disorder
406
0
0.198
< 0.05
Positive
Aging Index
1200
26.26
0.107
< 0.05
Positive
Population Under 5 Years Old
263
0
0.433
< 0.05
Positive
Population Aged 65 Years and Above
1996
32
0.337
< 0.05
Positive
Indigenous Population
1242
0
0.650
< 0.05
Positive
Foreign Labors
581
0
0.244
< 0.05
Positive
Low-Income Households
110
0
0.488
< 0.05
Positive
Population Without High School Diploma
1845
30
0.706
< 0.05
Positive
163 164
2.3. Spatial Autocorrelation Analysis
165 166 167 168 169 170
This research applies global and local spatial autocorrelation to investigate spatial distribution of SoVI. Moran’s I is a global scale indicator for measuring spatial autocorrelation and identifying the spatial clusters along Yilan County. On the other hand, the Local Indicator of Spatial Autocorrelation (LISA) can measure the local association between variables and the spatially weighted value of the same variables. It can also visualize the sites of the cluster and outlier. The equations of Moran’s I and LISA are demonstrated as following [39-41]: Moran’s I =
∑ ∑ ij
LISA =
×
∑ ∑ ij( )( )
∑ ( )
( ) ∑ ij
∑ ( )
−
(1) (2)
171 172 173 174 175 176
Where N is the number of spatial units. Wij is the spatial weighted matrix. Xi and Xj represent the value of variables. is the mean value of variables. Through the Moran’s I, we can identify the spatial pattern of SoVI, whether it’s a cluster or not. Also, whether the cluster is statistically significant. Moreover, through LISA, we can visualize the distribution of the cluster. The result of LISA will present the spatial distribution of the cluster on a map.
177
2.4. Geographically Weighted Regression (GWR)
178 179 180 181 182 183 184 185 186
Geographically Weighted Regression (GWR) is a spatial multivariate regression. It can explore not only the relationship between SoVI and real environmental hazards but also the investigation of the effectiveness of SoVI. Compared to other spatial multivariate regression such as Ordinary Least Squares (OLS), Spatial Lag Model (SLM), and Spatial Error Model (SEM), GWR can solve the non-stationarity by separating study area into several sub-study areas through generating several kernels [42,43]. So, GWR can represent the authentic situation within the study area and has a better explanatory ability [43]. Based on the previous research [43,44], the variable bandwidth of kernel is relatively suitable for clustered data. As a result, the kernel type in this research is “adaptive” [42,43]. The equations of GWR are demonstrated as below [42-44].
6 of 15
GWR: () = () + () + () + ! ()! + ⋯ + # ()# # () = ( % ()) % ()
' () 0 () = & 0 0
0 ' () 0 0
'* () = ,
⋱ 0
0 0
0 (2) ../ 1 4 3
0 0 + 0 '* ()
(3) (4)
(5)
(6)
187 188 189 190 191
Where Xmi is the independent variables, while # () represents the coefficient. The coefficient is calculated by the matrix of independent variables (Xmi) and the matrix of weights specific to location u. W(u) is the weighted matrix in the model calculated by Wn(u). The dn(u) is the distance between the nth observation and the location of u. In addition, the h is the radius of the kernel.
192
3. Results and Discussion
193
3.1. PCA and SoVI
194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224
The global Moran’s I is applied to identify the spatial distribution of all variables. The Moran’s I and p-value show that all the variables have statistically significant clusters (Table 1). Moreover, we conduct the principal component analysis to reduce these variables. The KMO of PCA is 0.647. Also, Bartlett’s Test of Sphericity is highly significant (p-value <0.01). The result is acceptable. Based on the eigenvalue, PCA compresses these 12 variables into four principal components (Fig 2). These four components explain 77.0% of the total variables. The first component, population factor, includes the Population Without High School Diploma, Aging Index, Female Population, Solitary Elderly, Physical and Mental Disorder, and Population Aged 65 Years and Above variables. The second component, economic factor, composes of the Indigenous Population, Low-Income Households, and Population Under 5 Years Old variables. The third principal component, social factor, includes the Population Density and Dependency Ratio variables. The last component, labor factor, includes the Foreign Labors variable only. These four principal components can individually interpret the total variables as follows: population factor (28.6%), economic factor (24.0%), social factor (15.5%), and labor factor (8.9%). Fig 3 and Table 2 demonstrate the distribution of four components and the result of global spatial autocorrelation analysis. The Moran’s I shows that all factors are clustered. According to the p-value, the clusters are statistically significant. The majority of population component congests in the mountain areas, but some of them also distribute into coastal areas. Compared to other components, the distribution of the population component does not have a specific pattern. Some of them distribute into mountain areas while others distribute into coastal areas. On the other hand, most of the economic component has a conspicuous cluster in mountain areas. One of the most important reasons is that the level of development in mountain areas is not very well. Due to the unfavorable topographic feature and inaccessibility compared to plain areas, mountain areas have far fewer economic resources and opportunities. As a result, the majority of economic component assembles in mountain areas. The social component highly clusters in the plain areas because there are the most rapid development areas in Yilan County, especially for urban areas such as Yilan City and Luodong Town. Besides, some communities also have high values in the social component. Some of them are located in essential nodes of traffic networks such as the Chauyan, Tianf, and Tianshan communities. Finally, some communities with a high value of the labor component are located in the surrounding areas of Su'ao Harbor, such as Nanning and Nanxing communities. In Yilan County, foreign labors can be classified into two main categories, nursing labors, and fishery, and maritime labors. Fishery and maritime labors highly cluster in the surrounding areas of Su'ao Harbor. However, the distribution of nursing labors is far dispersed.
7 of 15
225 226
Fig. 2. The scree plot and eigenvalue of PCA. Based on the eigenvalue and scree pot, PCA generates four principal components.
227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244
Fig 4 (a) shows the spatial distribution of SoVI in the study area and result of global spatial autocorrelation analysis. Totally, there are 233 communities in Yilan County. Based on the analysis of SoVI, there are 20 communities (8.6%) with Very High levels of SoVI, while there are 42 communities (18.0%) with a High level of SoVI. Over a quarter of communities (26.6%) in Yilan County is classified as a high social vulnerability class. Most of them cluster in mountain areas. The reason for this special distribution is owing to the characteristics of mountain areas. The topographic feature and inaccessibility cause mountain areas becoming unsuitable areas for economic development. Also, some of the communities are also highly incapability. Fig 4 (b) shows the result of local spatial autocorrelation analysis. Most communities located in mountain areas are highly social vulnerable with the H-H cluster. However, most of the low SoVI communities aggregate in the center of plain areas with the L-L cluster. This reveals that the socioeconomic status in plain areas is significantly better than mountain areas. The flat topography and convenient accessibility are very suitable for economic development. Moreover, the revision of the Agricultural Development Act also accelerates the transformation of land use. The traffic connection between Yilan County and Taipei City is quickly increasing after the construction of the freeway, especially in plain areas. Therefore, compared to mountain areas, plain areas have a better socioeconomic status to reduce social vulnerability. Table 2 The global autocorrelation analysis of all factors
245
Component
Moran’s I
p-value
Population factor
0.335
< 0.05
Economic factor
0.584
< 0.05
Social factor
0.594
< 0.05
Labor factor
0.224
< 0.05
8 of 15
246 247 248 249
Fig. 3. Spatial distribution of the four components derived from PCA and the result of global spatial autocorrelation analysis: (a) population factor; (b) economic factor; (c) social factor; (d) labor factor.
9 of 15
250 251 252
Fig. 4. The distribution and cluster map of SoVI (a) the spatial distribution of SoVI and result of global spatial autocorrelation analysis; (b) the result of LISA.
253
3.2. GWR
254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276
The variables applied to construct SoVI are highly collinearity. The raw variables cannot take as the independent variables. PCA is one of the most common approaches to deal with collinearity. PCA can compress similar variables into a principal component. Through this approach, PCA can eliminate the collinearity within the variables. The independent variables we used are the principal component generated by PCA, including (a) population factor; (b) economic factor; (c) social factor; and (d) labor factor. In order to validate SoVI, we apply GWR to conduct this analysis. The dependent variable applied in this research is the combination of flood and debris flow events shown in Fig 5. These events represent the most common environmental hazards in Yilan County. The occurrence of environmental hazards is selected due to the scale of data. NGIS provides the detail occurrence of environmental hazards in Yilan County, and all the data is recorded on the community scale. Moreover, based on the literature review, the SoVI performs well for the occurrence of environmental hazards [45]. The potential loss is highly related to the occurrence of environmental hazards. Therefore, we select the occurrence of flood and debris flow events as the dependent variables. The kernel type we applied is “Adaptive.” The bandwidth method is “BANDWIDTH PARAMETER”. Based on the results of GWR in Table 3, the R2 is 0.769, and the Adjusted R2 is 0.577 for the whole county scale. This value is relatively high and seems acceptable. Fig 6 shows the standardized residual of GWR. About 142 (60.9%) of the communities show the insignificant residual (-0.5SD to 0.5SD). In order to ensure the result of the model is not biased, we apply the global spatial autocorrelation analysis to examine the standardized residual of GWR. If the result of GWR is not biased, the spatial autocorrelation analysis of the standardized residual will show no significant spatial autocorrelation. The Moran’s I is 0.02, and the p-value is 0.526. The result of the global spatial autocorrelation shows the distribution of standardized residuals is random. In other words, the SoVI is not biased. The R2 and the Moran’s I of the standardized residual reveals that SoVI is a valid indicator in this study, and the goodness of fit for this model is statistically adequate.
10 of 15
277
278 279 280
Fig. 5. The spatial distribution of standardized flood and debris flow events. Table 3 The results of GWR Neighbors
R2
Adjusted R2
30
0.769
0.577
AICc
StdResid Moran’s I
p-value
594.713
0.02
0.526
StdResid: Standardized residual
281
11 of 15
282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303
Fig. 6. The distribution of GWR’s standardized residual.
Fig 7 shows the coefficient of four components since GWR separates Yilan County into several sub-areas through creating many kernels with different bandwidth. The coefficient of component is highly variable in spatial distributions. We categorize the coefficient by the natural break. The GWR could be used to explore the spatial relationship between communities directly affected by environmental hazards and localities with higher or lower SoVI values [46]. Fig 7a shows the coefficient of the first component, the population factor. There are 77 (33.0%) of the communities show a positive correlation to environmental hazards. The urban areas such as Yilan City and Luodong Town show a significant positive correlation. Also, the near mountain areas, Sanxing, Yuanshan, and Datong Townships, show a positive correlation, while the southwestern mountain areas and northern coastal areas show a negative correlation to environmental hazards. Fig 7b shows the coefficient of the second component, the economic factor. There are 114 (48.9%) of the communities show a positive correlation to environmental hazards. Especially for the populated urban areas, where located at the center of the study areas such as Yilan City and Luodong Town, show a significantly positive correlation to the environmental hazards. However, the coastal areas of Wujie and Zhuangwei Townships show a negative correlation to environmental hazards. Fig 7c shows the coefficient of the third component, the social factor. In General, most communities show a negative correlation. Only 25 (10.7%) of the communities show a positive correlation in relative dispersed distribution. Fig 7d shows the coefficient of the last component, the labor factor. Totally, 144 (61.8%) of the communities in our study area show a positive correlation. In Only coastal areas, the northern and middle parts show a negative correlation. In mountain areas, only 4 communities show a negative correlation.
12 of 15
304 305 306
Fig. 7. Coefficient of the four components: (a) population factor; (b) economic factor; (c) social factor; and (d) labor factor.
307 308
4. Conclusions
309 310 311 312 313 314 315
In conclusion, this research provides not only a holistic exploration but also a validation of social vulnerability to environmental hazards in Yilan through SoVI and spatial analysis. Social vulnerability is a crucial concept for disaster prevention and damage prediction. The visualization of SoVI is essential for decision-makers to formulate a thorough emergency plan. Moreover, this research also validates the effectiveness of the concept of social vulnerability to environmental hazards. Irrefutably, SoVI is a simple, adaptable, flexible, and widely used tool for assessing, quantifying, and visualizing social vulnerability to environmental hazards. Nevertheless, validation is still important. As a result, we apply the GWR to
13 of 15
316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345
investigate the effectiveness of the SoVI. The result shows that the SoVI in this research is a valid and unbiased indicator. Owing to the different context in various places, we should modify the variables of SoVI to adapt to the context in Taiwan. Totally, there are 12 variables selected. Based on SoVI and spatial autocorrelation analysis, we quantify, visualize, and analyze the distribution of social vulnerability to environmental hazards in Yilan County. Results show that most of the communities in mountain areas have a higher level of SoVI. In other words, susceptible groups cluster in mountain areas mostly. When encountering environmental hazards, the potential loss in mountain areas is relatively higher. The topography of our study area is relatively complex, including mountain and plain areas. These features allow us to take account of the interaction between the social fabric and natural characteristics. We found the topography characteristics and accessibility are the most significant factor. In mountain areas, the inaccessibility is triggered by topography characteristics and leads to a low level of development. The mountain areas have far fewer economic resources and opportunities. These factors lead the mountain areas with high social vulnerability. Previous researches also show that social vulnerability to environmental hazards in mountain areas is higher than in plain areas. Therefore, for the decision-maker, the main goal of hazard mitigation in mountain areas should reduce the level of social vulnerability in order to minimize potential loss. Moreover, the decision-maker can also pre-evacuate the residents in mountain areas when forecasting serious typhoon coming to avoid potential damage. On the other hand, as for the plain areas, the decision-maker should focus on the dense population areas, the urban areas. The high population density also means a large potential exposure to environmental hazards. Although the residents in plain areas have relatively better socioeconomic status and lower social vulnerability than mountain areas, it only means the potential loss to environmental hazards is relatively insignificant compared to mountain areas. When encountering environmental hazards, the plain areas still have a chance to face the damage brought by natural hazards. Thus, for the plain areas, formulating a plan or project for emergency response and evacuation is also very critical. The measurement of social vulnerability is essential and shouldn't be ceased in the future works, especially under the setting of climate change. We should take the measurement of social vulnerability as the basis, and focus on the ability to coevolve with the natural hazards especially for those areas with higher social vulnerability. In order to cope with the potential stronger environmental hazards, the ability to be more resilient should be the focus in future work.
346
References
347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367
[1] Intergovernmental Panel on Climate Change (IPCC), Climate Change 2001: Impacts, Adaptation, and Vulnerability. Contribution of Working Group II to the Third Assessment Report of the Intergovernmental Panel on Climate Change; J.J. McCarthy, O.F. Canziani, N.A. Leary, D.J. Dokken, K.S. White, Eds, Cambridge University Press. Cambridge, UK https://www.ipcc.ch/report/ar3/wg2/. [2] Intergovernmental Panel on Climate Change (IPCC), Climate Change 2007: Synthesis Report. Contribution of Working Groups I, II and III to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change; The Core Writing Team, R.K. Pachauri, A. Reisinger, Eds, IPCC. Geneva, Switzerland https://www.ipcc.ch/report/ar4/syr/. [3] Intergovernmental Panel on Climate Change (IPCC), Climate Change 2014: Synthesis Report. Contribution of Working Groups I, II and III to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change; The Core Writing Team, R.K. Pachauri, L.A. Meyer, Eds, IPCC. Geneva, Switzerland https://archive.ipcc.ch/report/ar5/syr/. [4] S.L. Cutter, B.J. Boruff, W.L. Shirley, Social vulnerability to environmental hazards, Soc. Sci. Q. 84(2) (2003) 242-261. https://doi.org/10.1111/1540-6237.8402002. [5] S.L. Cutter, Vulnerability to environmental hazards, Prog. Hum. Geog. 20(4) (1996) 529-539 https://doi.org/10.1177%2F030913259602000407. [6] W.N. Adger, N. Brooks, M. Kelly, G. Bentham, M. Agnew, S. Eriksenet, New indicators of vulnerability and adaptive capacity. Tyndall Centre for Climate Change Research, Norwich, UK, 2004 http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.112.2300&rep=rep1&type=pdf. [7] W.N. Adger, Vulnerability, Global. Environ. Chang. 16(3) (2006) 268-281 https://doi.org/10.1016/j.gloenvcha.2006.02.006.
14 of 15
368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418
[8] T.G. Frazier, C.M. Thompson, R.J. Dezzani, A framework for the development of the SERV model: A Spatially Explicit Resilience-Vulnerability model, Appl. Geogr. 52 (2014) 158-172 https://doi.org/10.1016/j.apgeog.2014.04.004. [9] I. Frigerio, S. Ventura, D. Strigaro, M. Mattavelli, M.D. Amicis, S. Mugnano, M. Boffi, A GIS-based approach to identify the spatial variability of social vulnerability to seismic hazard in Italy, Appl. Geogr. 74 (2016) 12-22 https://doi.org/10.1016/j.apgeog.2016.06.014. [10] J.I. Uitto, The geography of disaster vulnerability in megacities: A theoretical framework, Appl. Geogr. 20(1) (1998) 7-16 https://doi.org/10.1016/S0143-6228(97)00041-6. [11] B. Wisner, P. Blaikie, T. Cannon, I. Davis, At Risk: Natural Hazards, People's Vulnerability and Disasters, 2nd ed, Routledge, Oxfordshire, UK, 2003 https://www.preventionweb.net/files/670_72351.pdf. [12] J. Birkmann, Risk and vulnerability indicators at different scales: Applicability, usefulness and policy implications, Environ. Hazards. 7(1) (2006) 20–31 https://doi.org/10.1016/j.envhaz.2007.04.002. [13] A. Fothergill, E.G.M Maestas, J.D Darlington, Race, ethnicity, and disasters in the United States: a review of the literature, Disasters. 23(2) (1999) 156-173 https://doi.org/10.1111/1467-7717.00111. [14] L. Rygel, D. O'Sullivan, B. Yarnal, A method for constructing a social vulnerability index: An application to hurricane storm surges in a developed country, Mitig. Adapt. Strat. Gl. 11(3) (2006) 741-764 https://doi.org/10.1007/s11027-006-0265-6. [15] S.L. Cutter, C. Finch, Temporal and spatial changes in social vulnerability to natural hazards, Proc Natl Acad Sci Usa. 105 (7) (2008) 2301-2306 https://doi.org/10.1073/pnas.0710375105. [16] A. Fekete, Validation of a social vulnerability index in context to river-floods in Germany, Nat. Hazard. Earth. Sys. 9(2) (2009) 393-403 https://doi.org/10.5194/nhess-9-393-2009. [17] N.J. Wood, C.G. Burton, S.L. Cutter, Community variations in social vulnerability to Cascadia-related tsunamis in the U.S. Pacific Northwest, Nat. Hazards. 52(2) (2010) 369-389 https://doi.org/10.1007/s11069-009-9376-1. [18] E. Tate, Social vulnerability indices: a comparative assessment using uncertainty and sensitivity analysis, Nat. Hazards. 63(2) (2012) 325-347 https://doi.org/10.1007/s11069-012-0152-2. [19] S. Khan, Vulnerability assessments and their planning implications: a case study of the Hutt Valley, New Zealand, Nat. Hazards. 64(2) (2012) 1587-1607 https://doi.org/10.1007/s11069-012-0327-x. [20] S.F. Balica, N.G. Wright, F. van der Meulen, A flood vulnerability index for coastal cities and its use in assessing climate change impacts, Nat. Hazards. 64(1) (2012) 73–105 https://doi.org/10.1007/s11069-012-0234-1. [21] Q. Dinh, S. Balica, L. Popescu, A. Jonoski, Climate change impact on flood hazard, vulnerability and risk of the Long Xuyen Quadrangle in the Mekong Delta, Int. J. River. Basin. Manag. 10(1) (2012) 103–120 https://doi.org/10.1080/15715124.2012.663383. [22] M.K.A. Kablan, K. Dongo, M. Coulibaly, Assessment of social vulnerability to flood in urban Côte d’Ivoire using the MOVE framework, Water. 9(4) (2017) 292-311 https://doi.org/10.3390/w9040292. [23] T. Morimoto, Spatial analysis of social vulnerability to floods based on the MOVE framework and information entropy method: Case study of Katsushika Ward, Tokyo, Sustainability. 11(2) (2019) 529-548 https://doi.org/10.3390/su11020529. [24] L.S. Hung, C. Wang, B. Yarnal, Vulnerability of families and households to natural hazards: A case study of storm surge flooding in Sarasota County, Florida, Appl. Geogr. 76 (2016) 184-197 https://doi.org/10.1016/j.apgeog.2016.09.021 [25] I.Frigerio, M.D. Amicis, Mapping social vulnerability to natural hazards in Italy: A suitable tool for risk mitigation strategies, Environ. Sci. Policy. 63 (2016) 187-196 https://doi.org/10.1016/j.envsci.2016.06.001 [26] J.S. Lee, H.I. Choi, Influence of Construction Schemes for a Non-compensatory Composite Indicator on Flood Vulnerability Assessments in the Korean Peninsula, Int. J. Civil. Eng. 17(8) (2018) 1–9 https://doi.org/10.1007/s40999-018-0382-0 [27] Y.J. Lee, Social vulnerability indicators as a sustainable planning tool, Environ. Impact. Asses. 44 (2014) 31-42 https://doi.org/10.1016/j.eiar.2013.08.002 [28] L.F. Chang, S.L. Huang, Assessing urban flooding vulnerability with an emergy approach, Landscape, Urban. Plan. 143 (2015) 11-24 https://doi.org/10.1016/j.landurbplan.2015.06.004
15 of 15
419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467
[29] T.H. Siagian, P. Purhadi, S. Suhartono, H. Ritonga, Social vulnerability to natural hazards in Indonesia: driving factors and policy implications, Nat. Hazards. 70(2) (2014) 1603-1617 https://doi.org/10.1007/s11069-013-0888-3 [30] S. Khan, Vulnerability assessments and their planning implications: a case study of the Hutt Valley, New Zealand, Nat. Hazards. 64(2) (2012) 1587-1607 https://doi.org/10.1007/s11069-012-0327-x [31] E. Mavhura, B. Manyena, A.E. Collins, An approach for measuring social vulnerability in context: The case of flood hazards in Muzarabani district, Zimbabwe, Geoforum. 84 (2017) 103-117. [32] M. Dzoga, D. Simatele, C. Munga, Assessment of ecological vulnerability to climate variability on coastal fishing communities: A study of Ungwana Bay and Lower Tana Estuary, Kenya, Ocean. Coast. Manag. 163, (2018) 437–444 https://doi.org/10.1016/j.ocecoaman.2018.07.015 [33] B. Jones, J. Andrey, Vulnerability index construction: methodological choices and their influence on identifying vulnerable neighbourhoods, IJEM. 4(2) (2007) 369-395 https:/doi.org/10.1504/IJEM.2007.013994 [34] C. Wang, B.Yarnal, The vulnerability of the elderly to hurricane hazards in Sarasota, Florida. Nat. Hazards. 63(2) (2012) 349-373 https://doi.org/10.1007/s11069-012-0151-3 [35] Ministry of Health and Welfare, R.O.C. (Taiwan), Report of the Women’s Living Condition Survey 2015, https://www.mohw.gov.tw/dl-30768-11e9c325-3c6c-4079-8870-124fa8c5e10d.html [36] Council of Indigenous Peoples, R.O.C. (Taiwan), Economic Status Survey of Indigenous Peoples in Taiwan 2014, https://www.apc.gov.tw/portal/getfile?source=2D838540F5D6F659FAFB9859EF31AC3B381A272F479D65 D98D902DFAAFC2E154084F20DD5F0F37964575BA3D2CADEAE87E08C4C4F83F6C7D3B91B9DF716 59F0C&filename=D0572478D51EFA60DD66604754BE10B807F619060818CEA0D0636733C6861689. [37] Council of Indigenous Peoples, R.O.C. (Taiwan), Annual report of Population and Health Condition of Indigenous Peoples 2015. https://www.apc.gov.tw/portal/getfile?source=2D838540F5D6F659FAFB9859EF31AC3B381A272F479D65 D98D902DFAAFC2E1547AEE5B785E323204555BE00A12C2A71E41B16B2642EE8A363B91B9DF71659 F0C&filename=A656B8F1D8AB64973D3CD606F117241350E32B03FBDD42DA88CAD01538D4794C8E 54517BBFE816C6AC1C5A2019A292E815E6A529B9D3427D. [38] Y.C. Chang, C.L. Yang, An Analysis of the Indigenous Peoples, Directorate General of Budget, Accounting and Statistics, Executive Yuan, R.O.C. (Taiwan), https://www.dgbas.gov.tw/public/Attachment/56115154ZX6CXRZE.pdf. [39] P.A.P Moran, The interpretation of statistical maps, J. R. Stat. Soc. B. 10 (1948) 243-251 https://doi.org/10.1111/j.2517-6161.1948.tb00012.x [40] P.A.P Moran, Notes on continuous stochastic phenomena, Biometrika. 37 (1950) 17-23 https://doi.org/10.1093/biomet/37.1-2.17 [41] L. Anselin, Local indicators of spatial association-Lisa, Geogr Anal. 27(2) (1997) 93-11 https://doi.org/10.1111/j.1538-4632.1995.tb00338.x [42] A.S. Fotheringham, M.E. Charlton, C. Brunsdon, Geographically weighted regression: a natural evolution of the expansion method for spatial data analysis, Environ. Plann. A. 30(11) (1998) 1905-1927 https://doi.org/10.1068%2Fa301905 [43] D.K. Yoon, J.E. Kang, S.D. Brody, A measurement of community disaster resilience in Korea, J. Environ. Plann. Man. 59(3) (2016) 436-460 https://doi.org/10.1080/09640568.2015.1016142 [44] A.S. Fotheringham, C. Brunsdon, M. Charlton, Geographically Weighted Regression: The Analysis of Spatially Varying Relationship, Wiley. New York, USA, (2003)https://www.wiley.com/en-us/Geographically+Weighted+Regression%3A+The+Analysis+of+Spatially +Varying+Relationships+-p-9780470855256 [45] L.A. Bakkensen, C.F. Lent, L.K. Read, I. Linkov, Validating resilience and vulnerability indices in the context of natural disasters, Risk. Anal. 37(5) (2016) 982-1004 https://doi.org/10.1111/risa.12677 [46] I. Török, Qualitative assessment of social vulnerability to flood hazards in Romania, Sustainability 10, (2018) 3780 https://doi:10.3390/su10103780
A GIS-based Approach for Assessing Social Vulnerability to Flood and Debris Flow Hazards Chien-Hao Sung 1, Shyue-Cherng Liaw 1,* 1 Department of Geography, National Taiwan Normal University, Taipei 10610, Taiwan. * Correspondence: [email protected]; Tel.: +886-2-7734-1649
Conflicts of Interest: The authors declare no conflict of interest. Shyue-Cherng Liaw Prof. of Geography, NTNU, Taiwan