Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Explore the physical environment features in earthquake disaster area – A case study of 921 Earthquake in Taiwan Abstract Earthquake has been regarded as an infrequent and unpredictable, but fatal disaster. Conventional mitigation measures focused on structural engineering enhancement. However, weakness appeared by even larger earthquakes has outpaced the ability to mitigate the impacts to acceptable levels. Non-structural engineering measures way beyond just retrofitting have received attention. In fact, sensitive geologic environment might result in earthquake-induced ground damage. Hence, this study attempts to base on previous earthquake disaster to explore if there is any similarity in such physical environment by using spatial statistic analysis and principle component analysis (PCA). The results show that similar physical environment features including landslide prone areas and close to fault. In addition, some of high damage areas appear newly built buildings collapsed completely. Although the building construction process might be one of the factors, the continued approval in such sensitive environment might result in much serious fatalness in the future. Therefore, the retrofitting in sensitive physical environment is not only urgent but the avoidance new development might be important as well. Keywords: earthquake, spatial statistic analysis, principle component analysis, non-structural engineering measures 1. Introduction Asian region has been regarded as most frequently hit by natural disasters, especially for Asia is riddled with faults. Earthquakes are infrequent hazards but unpredictable feature that result in higher fatalness (Guha-Spair et al., 2010). In fact, earthquake don’t kill people, buildings do (Ambraseys, 2010). Therefore, conventional ways to mitigate earthquake disaster are to enhance buildings codes and structural engineering measures. However, weakness appeared in structural engineering measures such as risk of loss happened wherever development is allowed in hazardous areas, or the disaster beyond design standard might have much serious impacts on lives and properties. The threat posed by even larger earthquakes has outpaced the ability to mitigate the impacts to acceptable levels. People start to be aware that such disaster can be revised by humans but is not ultimately reducible to a human construction. Since the 1980s, non-structural measure such as land use, insurance, warning system way beyond retrofitting of seismic damages, seismic resistance, and better anti-seismic structures in both urban planning and architecture has received lot more attention in the worldwide (CENDIM et al., 2001; Coburn and Spence, 2002). Indeed, building is the major object to mitigate such serious disaster but should base on different perspective. When buildings located nearby geographical sensitive areas such as loose saturated sands and deep deposits of soft clays might result in earthquake-induced ground damages such as surface rupture, liquefaction, landslides, and land subsidence easily. In fact, many other studies started to discuss the potential relationship between earthquake hazard and some particular factors in earthquake prone countries such as Greece and New Zealand (Athanasopoulou et al., 2008; Ansal et al., 2011; Becker et al., 2013). Multiple land use regulations proposed on development including prohibition of development in high-hazard areas, low-density zoning to limit building intensity in hazardous areas, density bonus in return for reduced density in areas subject natural hazards, reduced property tax to encourage more open spaces, transfer of development rights (Burby and Dalton, 1994). However, it is difficult for zoning boards or governing bodies to identify such high-hazard areas for lacking of accurate earthquake forecast. Consequently, development continues in the potential path of intense ground shaking, and ground failures, and existing development in these areas remains at risk (Committee to Develop a Long-Term Research Agenda for the Network for Earthquake Engineering Simulation (NEES), 2003). Therefore, this study attempts to base on previous earthquake disaster to explore if there is any similarity in such physical environment and building feature. There are two kinds of methods are used in this study, spatial statistic analysis and principle component analysis (PCA). Location-specific characteristics such as landslide prone area, close to fault, and others are crucial to such large magnitude disaster area (Sklenicka et al., 2013). The application of spatial statistic analysis is to depict and explain how those damage buildings distributed in the space based on similar neighboring values (Tobler, 1970; Goodchild, 1986). Afterwards, PCA is then applied to categorize the similar features in different kinds of clustered patterns. The incorporation of both spatial statistic analyses and PCA into land use planning policy can provide more accurate information to allocate or prevent development in certain areas. Taiwan locates on the Pacific Ring of Fire where a seismically active zone is (the frequent convergence of the Philippine Sea plate and the Eurasian Plate), and geologists have identified forty-two active faults. Currently, a fault zone area 15 meters on each side of fault trance has been executed in Chelungpu Fault only in Taiwan. An improved understanding of the similarities of the physical environment in earthquake disaster area is able to help for recognizing earthquake prone areas for further building retrofitting and land allocation in the future. Therefore, this study attempts to apply spatial statistic analyses and principle component analysis (PCA) to explore spatial distribution pattern and similar physical environment and building features. In the next section discusses the conceptual model, variable definition, and methods. The following sections are the result of spatial statistic analyses on damage buildings and principle component. This paper concludes with a comparative study of physical environment and building feature in previous earthquake disaster area. 2. Research design and methodology 2.1 Conceptual model After 921 Chi-Chi Earthquake, over 2,400 people were killed, 10,000 people injured, and 100,000 buildings damage. Due to the epicenter, fault dislocation and ground deformation, huge live and property losses were aggregated in the central Taiwan, and 5,213 damage buildings are used in this study. The application of two spatial statistic analyses is to probe into if there is any significant cluster pattern on particular distance. Afterwards, Principle component analysis (PCA) will then be applied to categorize particular features of damage buildings. (See Fig.1) 921 Earthquake Disaster Buildings damage: 5,213 Spatial Distribution Pattern Cluster Distance Cluster Pattern Buffer High-High High-Low Physical Environment Low-High Low-Low Building Features Soil Landslide Height Age Fault mudslide Material Use Structure Façade Vegetation Fig. 1 Conceptual model 2.2 Variable definition 2.2.1 Physical environment The variables selected in physical environment are to assess the natural environmental conditions, and the variables include landslide risk area, land cover, soil type, fault, river, and debris flow stream. In each category, multiple variables are selected based on this study including high/moderate/others landslide risk, bare land, vegetation lane, colluvium, alluvium, debris flow stream, fault, and river. (See Table 1) Table 1 Variables in physical environment Variables Definition The variables to forecast landslide include slope, earthquake, geology, and ground structure etc. The landslide risk area can be Landslide risk area divided into three categories: high, moderate, and others. Bare land includes falling rock, slide, and Bare land flow etc. Land cover Vegetation Vegetation land includes grass, crops, and land trees etc. Colluvium is composed of a heterogeneous Soil type Colluvium range of rock types and sediments. Source Central Geological Survey, MOEA National Land Surveying and Mapping Center Taiwan Agricultural Alluvium Debris flow stream Alluvium refers to loose, unconsolidated soil or sediments, and the composition is ranging from silt, clay, and sand etc. A monitor database of landslide, complex landslide, debris flows and soil erosion etc. Fault Forty-two active fault traces have been identified. River River is a flowing watercourse toward ocean, sea, and lake etc. Research Institute, COA Soil and Water Conservation Bureau, COA Central Geological Survey, MOEA Water Resources Agency, MOEA 2.2.2 Building features Architecture and Building Research Institute (1999) conducted a report particular on damage building after 921 Chi-Chi Earthquake. The survey included building height, building year, construction material, building use, anti-earthquake structure, and façade pattern. (See Table 2) Table 2 Variables is building features Variables Definition According to Building Code and (1) ≤ 3 (floors) Building Regulations, the construction structure (2) 4-6 (floors) height is related to the height. (3) ≥ 7 (floors) Building (1) Before 1974 Different building time refers to year different period of the improvement in (2) 1975-1982 earthquake resistance requirement in (3) 1983-1989 Building Code and Regulations. (4) 1990-1997 (5) After 1997 Construction (1) Reinforced concrete (RC) The Manual of Structure Repairment Material and enhancement has explained how (2) Steel frame/ Steel RC different material should be improved. (3) Brick/Wood/others Building use (1) Residential The use of building is related to the capacity it is. Generally, public use (2) Mixed use such as hospital and school etc. might (3) Public use have more people in the daytime. (4) Industrial use Residential has relatively lower people but in the nighttime. Anti-earthquake structure Anti-earthquake structures include special moment – resisting frame, diagonal bracing frame, shear wall, and brick wall. Façade pattern Façade patterns include Arcade, high ceiling, arcade without pillar, and second floor setback. Resource: Architecture and Building Research Institute, 1999 2.3 Methods 2.3.1 Spatial autocorrelation analysis Spatial autocorrelation analysis often applies Moran’s I to comparing factors of neighboring areal units, and similar values among neighboring units indicates a strong positive spatial autocorrelation, and vice versa. Basically, the concept of spatial autocorrelation analysis is based on Tobler’s (1970) statement that everything is related but near things are more closely related (his “First Law of Geography”). Moran’s I can be defined as 𝐼= 𝑛 ∑ ∑ 𝑤𝑖𝑗 (𝑥𝑖 −𝑥̅ )(𝑥𝑗 −𝑥̅ ) 𝑊 ∑(𝑥𝑖 −𝑥̅ )2 …………(1) where 𝑥𝑖 and 𝑥𝑗 are the values of variables in areal unit i and unit j; 𝑥̅ is the mean value of variables in all spatial units; 𝑤𝑖𝑗 is the spatial weights matrix; (𝑥𝑖 − 𝑥̅ )(𝑥𝑗 − 𝑥̅ ) is cross –product of the variances between neighboring values and the overall mean; W is the sum of all elements of the spatial weights matrix. The value of Moran’s I ranged from -1 to 1. -1 indicates an extremely negative spatial autocorrelation while 1 is extremely positive autocorrelation. In order to detect the spatial autocorrelation, it should be compared to the expected value of Moran’s I: 𝐸(𝐼) = −(1)/(𝑛 − 1) …………(2) E(I) is always negative for E(I) is inversely related to the areal units. I > E(I) indicates a clustered pattern for similar features in adjacent areal units; I ≅ E(I) indicates random pattern for no particular patterns or similarity; I < E(I) indicates a dispersed pattern for different features in adjacent areal units. 2.3.2 Principal component analysis (PCA) Principal component analysis (PCA) is a transformation process from correlated variables to uncorrelated variables through orthogonal linear transformation. It is a useful approach to explore patterns within multivariate data set (Sun et al., 2009; Abdi and Williams, 2010; Shi, 2013). Principle components will then be designed according to variance. Normally, the first principle component represents the largest variance, and the variance will be decreased afterwards. The formula of PCA is as follows: 𝑈1 = 𝑙11 𝑥1 + 𝑙12 𝑥2 + ⋯ + 𝑙1𝑝 𝑥𝑝 𝑈2 = 𝑙21 𝑥1 + 𝑙22 𝑥2 + ⋯ + 𝑙2𝑝 𝑥𝑝 …………(3) … {𝑈𝑚 = 𝑙𝑚1 𝑥1 + 𝑙𝑚2 𝑥2 + ⋯ + 𝑙𝑚𝑝 𝑥𝑝 where n refers to spatial units, refers to the number of variables, 𝑥𝑝 refers to the original variables, and 𝑈𝑚 refers to principle components. 𝑈1 , 𝑈2 , …, 𝑈𝑚 𝑚 ≤ 𝑝 are linear combinations of 𝑥𝑝 . Generally, there is a ranking of principle components based on the eigenvalues. According to the mathematical transformation result, there will be p variables account for the total variability. The number of principle components is subjective to the study itself and how much interpretation can be achieved (Srivastava, 2002). SPSS (version 17.0) is used to reorganize multi correlated variables according to varimax rotation, Kaiser criteria (eigenvalues>1), and stepwise exclusion approach. In the end, the model will then be tested significance based on Kaiser-Meyer-Olkin (KMO) and Barlett’s tests. 3. Spatial distribution of damage buildings Chi-Chi earthquake occurred in 1999 and generated 30 seconds of extremely strong shaking. The epicenter was at the middle of Taiwan. The Chelungpu Fault, an active fault trace, breaks along the western edge of the central mountains. The north-south trending fault ruptured for over 80 kilometers. Tectonic warping, or folding, associated with the faulting caused additional upward ground deformations of 6 to 7 meters, particularly in the northern reached of the rupture. Near the fault trace and to the east of the rupture zone had higher ground motions than to the west of the rupture zone. In addition, Chi-Chi earthquake has caused further extensive ground deformations such as landslides and liquefaction. The Chelungpu Fault generated more than 1,800 landslides throughout the central mountain region. Two significant landslides, one has swept away an entire village, and another one has dammed up a river and become an artificial lake. Due to soil liquefaction, ground subsidence contributed to some bridge damage, and ground settlement caused 300 houses damage. Overall, there were 2,400 people killed, 10,700 people injured, over 8,500 buildings destroyed, 6,200 buildings seriously damaged, 100,00 people become homeless, and economic loss estimated to $10 to $12 billion. (See Fig.2) Fig. 2 Study area There was total 5,213 buildings damage during the earthquake in the center of Taiwan. The lower buildings got serious damage for it is popular to live in detached or attached houses. Relatively, the less high-rise buildings the less damage in this area. Older buildings, RC buildings and other material (such as brick, wood, corrugated steel shack) had much serious damage for the previous building technique might not strong enough to against earthquake. Residential is the most popular land use type so no doubt it had the highest amount damage. Most buildings did not have anti-earthquake structure or even they had the resisting frame was still not strong enough. Arcades is the building with poorly designed columns that failed at the first floor easily. (See Table 3) Table 3 Building features of damage buildings Height (floors) 1-3 4-6 >7 Serious damage Moderate damage Slight damage 3,194 794 739 203 94 65 44 38 42 Before 1974 1,814 230 165 Building Year 1974198319901982 1989 1997 579 343 264 272 152 147 240 151 157 After 1997 124 56 72 Table 3 Building features of damage buildings (cont.) Serious damage Moderate damage RC 1,782 550 Material SRC Others 16 1,388 9 313 Residential 2,980 Mixed 247 766 99 Use Public Use 101 37 Industrial 37 9 Slight damage 502 5 299 732 65 30 6 Table 3 Building features of damage buildings (cont.) Serious damage Moderate damage Slight damage Anti-earthquake structure Resisting Diagonal Shear Brick Frame Frame Wall Wall 353 2 77 1,552 Arcade 478 High Ceiling 39 Facade No Pillar Arcade 46 2F Setback 16 156 1 36 507 253 22 21 4 152 0 30 491 211 18 11 6 In this study, the first step is to explore how damage buildings distributed in the space. The application of the incremental autocorrelation analysis might help to determine the cluster pattern at an appropriate scale: z-score returned. In the result, there are multiple peaks happened represented clustering are most pronounced, and the first z-score returned is 600 meter. (See Fig.3) 600 meter will then be applied to buffer range. The incremental autocorrelation analysis is global statistic for it is a summary measure of the entire study area. However, the magnitude of spatial autocorrelation might not necessarily in consistency over the region. The application of local indicators of spatial autocorrelation (LISA) is able to capture the spatial heterogeneity of spatial autocorrelation (Anselin, 1995). The local Moran’s I captures how neighboring values are associated with each other. Fig. 3 Spatial autocorrelation by distance Fig. 4 Result of local indicators of spatial association The result of LISA shows that 470 spatial units are High-High (high values of damage clustered), 346 spatial units are High-Low (high values of damage surrounded by low values of damage), 152 spatial units are Low-High (low values of damage surrounded by high values of damage), and 833 spatial units are Low-Low (low values of damage clustered). Other damage buildings are distributed randomly in the study area. (See Fig. 4 and Table 4) Table 4 The spatial distribution of damage buildings according to LISA High-High High-Low Low-High Low-Low Serious damage 470 346 0 0 Moderate damage 0 0 46 262 Slight damage 0 0 106 571 4. Results Principle component analysis is used to categorize variables in physical environment and building feature. The results show that Kaiser-Meyer-Olkin (KMO) and Barlett’s tests are significant in High-High, High-Low, and Low-Low category but except Low-High category. 4.1 Physical environment For the High-High category, the PCA of twelve indicators extracted four components that explained 69% of the variance and 0.678 of the KMO value in the data. The indicators “high landslide risk”, “moderate landslide risk” and “colluvial” show high positive correlation in HH_PC1 and explained 32% of the variance. HH_PC1 is renamed “serious landslide prone area.” The second principle component HH_PC2 explains 14% of the variance with the indicators “fault distance” and “mudslide stream distance.” HH_PC2 is renamed “close to fault and mudflow.” The indicators “vegetation” is highly positive in HH_PC2 and explained 12% of the variance. HH_PC3 is renamed “soft ground surface.” The fourth principal component HH_PC4 explains 12% of the variance with the indicators “uncovered” and “mudslide stream distance.” HH_PC4 is renamed “mudslide prone area.” (See Fig. 5) (a) HH_PC1: Serious landslide prone area (b) HH_PC2: Close to fault and mudflow (c) HH_PC3: Vegetated ground surface (d) HH_PC4: Mudslide prone area Fig. 5 Principle components of physical environment in High-High category For the High-Low category, the PCA of twelve indicators extracted three components that explained 74% of the variance and 0.669 of the KMO value in the data. The first principle component shows high positive correlation with “high landslide risk”, “moderate landslide risk”, “low landslide risk”, “colluvial” and “mudslide stream distance” explained 40% of the variance. HL_PC1 is renamed “serious landslide prone area”. The indicators “fault distance” and “river distance” show high positive correlation in HH_PC2 and explained 21% of the variance. HH_PC2 is renamed “close to fault and river.” HH_PC3 explained 13% of the variance with the indicators “moderate landslide risk” and “alluvial.” HL_PC3 is renamed “moderate landslide prone area.” (See Fig. 6) (a) HL_PC1: Serious landslide prone area (b) HL_PC2: close to fault and river (c) HL_PC3: Moderate landslide prone area Fig. 6 Principle components of physical environment in High-Low category For the Low-Low category, the PCA of twelve indicators extracted three components that explained 67% of the variance and 0.610 of the KMO value in the data. LL_PC1 explains 32% of the variance with the indicators “high landslide risk”, “moderate landslide risk”, “low landslide risk”, “colluvial” ,and “mudslide stream distance.” LL_PC1 is renamed “serious landslide prone area.” The second principle component explains 20% of the variance with the indicators “high landslide risk”, “low landslide risk” and “vegetation.” LL_PC2 is renamed “moderate landslide prone area.” The indicators “vegetation” and “alluvial” are positive correlated and explains 15% of the variance. LL_PC3 is renamed “vegetated ground surface.” (See Fig.7) (a) LL_PC1: Serious landslide prone area (b) LL_PC2: Moderate landslide prone area (c) LL_PC3: Vegetated ground surface Fig. 7 Principle components of physical environment in Low-Low category 4.2 Building feature For the High-High category, the PCA of six indicators extracted two components that explained 47% of the variance and 0.640 of the KMO value in the data. HH_PC1 represents 29% of the total variance, and the indicators “year” and “anti-earthquake structure” show a high positive significance. In order to investigate how “year” affect High-High category, the PCA of five indicators extracted two components that explained 54% of the variance and 0.210 of the KMO value in the data. Although the KMO value is too low, the two components reveal the building feature in “1974-1982” and “1983-1989.” HH_PC2 represents 17% of the total variance, and the indicator “material” is highly positive. In order to investigate how “material” affect High-High category, the PCA of three indicators extracted one component that explained 61% of total variance and 0.448 of the KMO value. Although the KMO value is not significant enough, the component reveal the building feature in “reinforced concrete.” (See Fig. 8) (a) HH_PC1: Structure feature (b) HH_PC2: Material Fig. 8 Principle components of building feature in High-High category For the High-Low category, the PCA of six indicators extracted two components that explained 45% of the variance and 0.636 of the KMO value in the data. HL_PC1 represents 27% of the total variance, and the indicators “year” and “anti-earthquake structure” show a high positive significance. Another PCA is applied to the indicator “year,” and two components that explained 53% of the variance and 0.127 of the KMO value. Although the KMO value is too low, the two components reveal the building feature in “before 1974” and “after 1997.” HL_PC2 represents 17% of the total variance, and the indicator “use” is highly positive. The PCA of four indicators extracted one component that explained 45% of total variance and 0.232 of the KMO value. Still, the KMO value is not significant enough, but the component reveal that building feature in “mixed use.” (See Fig. 9) (a) HL_PC1: Structure feature (b) HL_PC2: Use type Fig. 9 Principle components of building feature in High-Low category For the Low-Low category, the PCA of six indicators extracted two components that 41% of the variance and 0.581 of the KMO value in the data. LL_PC1 represents 19% of the total variance, and the indicator “Year” is highly positive. In addition, the PCA of five indicators extracted two components that explained 47% of total variance and 0.105 KMO value in the data, and the two components reveal the building feature in “1994-1997” and “after 1997.” LL_PC2 represents 22% of the total variance, and the indicators “material” and “anti-earthquake structure” are moderately positive. The PCA of three indicators extracted one component that explained 64% of variance and 0.442 KMO value, and the component reveal the building feature in “other material (brick, wood, corrugated steel shack).” (See Fig. 10) (a) LL_PC1: Building year (b) LL_PC2: Structure feature Fig. 10 Principle components of building feature in Low-Low category 5. Comparative study of physical environment and building feature According to the PCA results, there are similar physical environment features including “serious landslide prone area”, “moderate landslide prone area”, “close to fault/river/debris flow steam”, and “vegetated ground area”. In “serious landslide prone area”, the High-High area locates in the eastern area and aggregated significantly in the northern and southern part. Comparing to “serious landslide prone area,” “moderate landslide prone area” is located in the western area. In addition, “Low-Low” area is located in the periphery of “High-Low” area. Due to building damage in “High-Low” area is quite serious (see Table 4), both “serious landslide prone area” in High-High area and “moderate landslide prone area” in High-Low area appear similar building damage feature. Although the newest buildings are not as many as the oldest category, the serious damage indicates that the poor design or inappropriate development might be the main reason. Therefore, both “serious landslide prone area” in High-High areas and moderate landslide prone area“ in High-Low areas should be implemented building investigation on both oldest and newest buildings. In addition, the future development should require geology investigation and structure engineering report before approving building permission. (See Fig. 11 and 12) Fig. 11 Serious landslide prone area Fig. 12 Moderate landslide prone area The north-south trending faults ruptured over 80 kilometers, and both the hanging wall of the fault moved westward and upward by 1 to 2 meters and serious ground deformation of 6 to 7 meters impacted seriously in the east of the fault (Chi-Chi Reconnaissance Team, 2000). Due to the serious surface rupture and ground deformation, the east becomes much sensitive than the west. Therefore, “vegetated ground surface” is way beyond serious damage in High-High area than in Low-Low area. In addition, the distance to the fault indeed results in serious building damage, and the distance might be varied. Currently, the fault zone area in 15 meters of each side of fault trace might not be enough, and it is necessary to have a complete review on adequate fault zone area in the future. Fig. 13 Close to fault or river Fig. 14 Vegetated ground surface In the end, this study overlays “serious landslide prone area” in both High-High areas and High-Low areas and “close to fault” latest built environment. The results show that there are large amount of existing buildings located in these earthquake prone areas in the northern area. In order to prevent next serious earthquake hit this area, it is necessary to implement building investigation and a retrofit program is also urgent. In addition, the future development should be more deliberate in such area. (See Fig. 15) Fig. 15 Built area in 2010 6. Conclusion Earthquake is infrequent but unpredictable disaster, and even larger magnitude has outpaced the ability of human beings to mitigate. Land use regulation and zoning have been discussed quite a while as nonstructural engineering measures, and they have been implemented mostly in fault zone area to prevent the surface rupture disaster. However, earthquake induced disaster are more than surface rupture such as liquefaction, landslide, land subsidence and so on. The results in this study show that there are similar physical environment features and way beyond the fault itself. Besides, the newest structure followed the newest Building Code and Regulations but defeated in the end. The continued approval development in such sensitive geological or earthquake prone areas might result in another fatalness disaster in the future. Reference Abdi, H., and Williams, L. J. (2010). Principal component analysis. Wiley Interdiscip. Rev. Comput. Stat., 2(4), 433-459. Ambraseys, N. N. (2010). A note on transparency and loss of life arising from earthquakes. Journal of Seismology and Earthquake Engineering, 12: 83-88. Ansal, A., Tönük, G., and Kurtulus, A. (2011). Seismic microzonation and earthquake scenarios for urban sustainability, in: Iai, S. (ed.) Geotechnics and Earthquake Geotechnics Towards Global Sustainability Geotechnical, Geological and Earthquake Engineering, Springer Science+Business Media, 151-168. Athanasopoulou, E., Despoiniadou, V., and Dritsos, S. (2008). The impact of earthquakes on the city of Aigio in Greece. Urban planning as a factor in mitigating seismic damage. 2008 Seismic Engineering Conference Commemorating the 1908 Messina and Reggio Calabria Earthquake. Becker, J. S., Beban, J., Saunders, W. S. A., Van Dissen, R., and King, A. (2013). Land use planning and policy for earthquake in the Wellington Region, New Zealad (2001-2011). Australasian Journal of Disaster and Trauma Studies, 2013(1), 3-16. Burby, R. J., and Dalton, L. C. (1994). Plans can matter! The role of land use plans and state planning mandates in limiting the development of hazardous areas. Administration Review, 54(3), 229-238. Center for Disaster Management (CENDIM), Bogazici University and the Center for Hazards and Risk Research (CHRR), and Columbia University. (2001). Urban risk management or national disasters: A research planning workshop. Chi-Chi Reconnaissance Team. (2000). Event Report Chi-Chi, Taiwan Earthquake. Risk Management Solution Inc. Coburn, A., and Spence, R. (2002). Earthquake protection. John Wiley & Sons, Wussex. Committee to Develop a Long-Term Research Agenda for the Network for Earthquake Engineering Simulation (NEES). (2003). Preventing Earthquake Disasters: The Grand Challenge in Earthquake Engineering: A Research Agenda for the Network for Earthquake Engineering Simulation (NEES). National Academies Press. Guha-Sapir, D., Vos, F., Below, R., and Ponserre, S. (2011). Annual Disaster Statistical Review 2010: The Numbers and Trends. Brussels: CRED Goodchild, M. F. (1986). Spatial Autocorrelation. CATMOG 47. Norwich: GeoBooks, University of East Angolia. Architecture and Building Research Institute. (1999). 921 Chi-Chi Earthquake Report on Damage Buildings. National Center for Research on Earthquake Engineering. Shi, Y. (2013). Population vulnerability assessment based on scenario simulation of rainstorm-induced waterlogging: a case study of Xuhui District, Shanghai City. Nat. Hazards, 1189-1203. Srivastava, M. S. (2002). Methods of Multivariate Statistics. Wiley-Interscience, New York, N. Y. Sun, A. L., Shi, C., and Shi, Y. (2009). The preliminary inquiry of flood vulnerability space changes in coastal provinces and autonomous regions. Environ. Sci. Manag., 34(3), 36-40. Tobler, W. R. (1970). A computer movie simulating urban growth in the Detroit region. Economic Geography, 46(supplement), 234-240.