1. Introduction
The population density in coastal zones is significantly higher than other areas [
1,
2] due to sufficient resources for human activities and civilization of early days [
3]. For example, maritime transport connects the whole world, and the port is an important part. Most of the world-class urban agglomerations are located in coastal zones, such as the Atlantic coast of the northeastern United States, the Pacific coast of Japan, the urban clusters of northwestern Europe, and the Yangtze River Delta of China. The coastal migration trend will continue [
4] as a consequence of accelerating development and utilization [
5]. Approximately 50% of the population is predicted to settle in the coastal areas by 2030 [
6], thus, they are not only the precious land resources of coastal countries, but also the link of foreign trade and cultural exchange. Coastal areas are highly productive at a global scale and occupy an important position in the ecosystem. For example, coastal wetlands provide numerous important functions, such as waste assimilation, nursery areas for fisheries and mariculture, flood protection, and nature conservation [
7]. However, coasts are easily subjected to natural hazards such as tropical cyclones, terrestrial floods, storm surges, and tsunamis. [
8,
9]. Most coastal zones worldwide are affected by climate change, especially global sea level rise [
10,
11], which influences the frequency and intensity of storm surge [
12]. China’s coast has played an important role in various aspects, namely, economic, political, and military [
13]. Considering the high concentration of assets and population, the vulnerability of China’s coastal zones should be evaluated, which is now regarded as a requirement for the effective development of emergency management capability [
14].
In 1974, White [
15] first defined vulnerability as
“the degree to which a system, sub-system, or component which is likely to experience harm due to exposure to a hazard, either a perturbation or stress”. Thereafter, vulnerability has been one of the conceptually written terms rooting in geography and natural hazard research [
16], which now relates to multiple natural impacts and social effects, such as salinity incursion, drought, bushfire, flooding and inundation, erosion and sedimentation, poverty, and land use change [
3,
17,
18,
19]. Vulnerability has been differently conceptualized and represented among different studies because of the various social contexts and stress effects [
20,
21,
22,
23,
24]. With regard to environmental hazards, numerous researchers distinguish biophysical and social vulnerabilities as two dimensions [
3,
25,
26,
27,
28,
29,
30,
31,
32,
33,
34].
Biophysical vulnerability focuses on hazards’ magnitude and duration, topography and land cover that influence the potential for harm [
23,
35,
36]. Such vulnerability arises from the approach based on assessments of hazardous events and their effects; the role of human systems in mediating the risk is downplayed or neglected [
37].
On the contrary, social vulnerability constructs on human community attributes, such as living standards, economic status, public infrastructure, institution and demographics, and historical process [
38,
39,
40,
41,
42]. These factors influence the society’s capability to prepare for, respond to, and recover from natural disasters [
3,
43,
44].
Social vulnerability is a pre-event state of being [
25,
45], indicating the sensitivity and susceptibility of a system to natural hazards. Such vulnerability is more complex than the biophysical one. However, socially created vulnerability can be reduced through policy implementation and human activities, thereby attracting an increasing number of researchers. In China, Chen et al. [
46] measured social vulnerability to natural hazards in the Yangtze River Delta region. Ge et al. and Su et al. [
47,
48] categorized social vulnerability patterns in China’s coastal zone. Zhou et al. [
49] indicated the spatiotemporal change of social vulnerability in China from 1980 to 2010. However, nearly all social vulnerability assessments are based on administrative boundaries and are inaccurate because the actual distribution is fairly heterogeneous across these spatial units. In the present work, we refer to the model proposed by Cutter et al. [
29], which shows that social vulnerability consists of two parts: potential exposure (PE) and social resilience (SR) [
50,
51]. The final explicit raster-level social vulnerability index (SoVI) [
47,
52] can be obtained by combining the potential exposure index (PEI) of 250 m resolution with the county-level social resilience index (SRI). Accordingly, the spatial delineation of social vulnerability is significantly improved and hotspots of high social vulnerability can be identified within administrative units.
“Potential exposure” can be understood as the degree to which a system will be adversely affected by natural hazards. Although all people living in hazardous areas are vulnerable [
14], we use the spatial distribution of population as the index for PE [
3,
53,
54,
55]. Given the same degree of exposure to hazards, regions with several inhabitants will be more socially vulnerable than those with low densities [
53]. Thus, precise population distribution is crucial in PE mapping. In China, census data are accurate first-hand statistics that can be used to explore population distribution and have been adopted by most studies [
46,
47,
48]. However, census data provide no detailed information because people are nonuniformly distributed within administrative boundaries but gather in the city center [
55,
56]. The emergence of remote sensing and geographical information system (GIS) techniques provides us with a new way to obtain explicit population distribution data [
56,
57]. We combined multi-source remote sensing data (GHSL, NDVI, NTL, land use and DEM) in the random forest (RF) model to disaggregate the county-level census data to 250 × 250 m grids [
58,
59]. As such, accurate population maps were produced, which is conducive to the PE assessment.
Resilience reveals the capability of a social system to resist and recover from a disturbance within the normal confines of daily life [
60,
61]. Thus, the more resilient a society is, the less vulnerable it will be [
62]. This study regards counties as the evaluation units for the following reasons: (1) the most accurate socioeconomic data that we can obtain are those at the county level and (2) disaster management and recovery measures are implemented at the county level [
49]. Considering that SR is related to multiple socioeconomic characteristics, a large amount of measurable variables were aggregated using principal component analysis (PCA) to establish the SRI. This method was advocated by Cutter et al. in 2003 [
29] and became the primary procedure in social vulnerability research.
Most recent studies on the vulnerability of China’s society are concentrated in a certain year, ignoring its development trend over time. On this basis, the present work addressed the last critique of social vulnerability assessment by providing a time series of changes in vulnerability patterns. The dynamic nature of social vulnerability is illustrated by using a replication of the baseline indicators for China’s coastal zone over two different time periods (2000 and 2010). We used the global Moran’s I spatial autocorrelation tool in ArcGIS 10.2 to test for significant clustering in SoVI for 2000 and 2010. Subsequently, the local indicators of spatial association (LISA) cluster map based on local Moran’s I was used to explore the distribution of hot and cold spots [
49,
60,
63]. Then, the percent change of SoVI was calculated, and local Moran’s I was used to identify regions with high temporal change [
64].
In this study, we focused on China’s coastal zone as a case study. This study aimed to improve SoVI spatial delineation and investigate the SoVI spatiotemporal change from 2000 to 2010. The research contents are as follows: (1) construction of PEI and SRI distribution maps, (2) aggregation of PEI and SRI to obtain SoVI and calculation of the percent change, and (3) use of global and local Moran’s I to identify a local cluster on the basis of the abovementioned results for locating the areas with SoVI spatiotemporal change.
2. Study Area
The study area (
Figure 1) consisted of 292 county-level units, including counties, county-level cities, and city districts, which are in the third level of Chinese administrative hierarchy [
48]. Hong Kong, Macao, and Taiwan were excluded because of their distinct political and economic environment. Mainland China, which geographically extends from 108°20′59″ E to 124°20′56″ E and from 18°15′16″ N to 39°59′56″ N, has a coastline of over 18,000 km crossing three climate zones: tropical, sub-tropical, and temperate [
47]. The coastal zone scope is a long narrow belt area alongshore, stretching from the Liaodong Bay in the north to the Gulf of Tonkin in the south [
47]. With the Hangzhou Bay as the boundary, the northern coastal zone is dominated by the plain coast (sand shore), whereas the southern part is dominated by the mountain coast (rock bank) [
13]. The coastal zone is easily attacked by natural hazards, such as storm surge, tsunami, and earthquake, because of its complex climatic and geological conditions. For example, 345 tropical cyclones hit the Chinese coastal zone between 1961 and 2008 [
65], posing a great threat to the safety of people’s lives and property. China Statistical Yearbook 2011 indicates that the coastal region contributes approximately 18% of the total population and 35% of the national GDP compared with nearly 4% occupation of land area [
47,
48]. The developed but highly vulnerable area can be effectively managed on the basis of the social vulnerability measurement to combat natural disaster risks, which was the purpose of this case study.
5. Discussion
The spatial distribution maps and transfer matrixes demonstrate that most counties are considered “very low” or “low” in terms of PEI and SoVI in 2000 and 2010. Without evident regional shift, the clusters with a maximum value of PE and social vulnerability are located in Tianjin, Yangtze River Delta, and Pearl River Delta where the SR is categorized as high level. Here, the driving factor was PE. Most counties belong to “low” and “medium” levels of SRI. Therefore, the SR of the study area still has a large room for improvement. In previous studies on China’s coastal zone, Ge et al. [
48] proposed that counties with high adaptive capacity and low sensitivity were concentrated in Yangtze and Pearl River Deltas. Su et al. [
47] identified Tianjin, Shanghai, Guangzhou, Shenzhen, and Dongguan as extreme groups of high adaptability. Guangzhou, Tianjin, and Shanghai were less sensitive to hazards. These results are consistent with the SRI spatial pattern proposed in the present paper.
The temporal analysis suggests the following: (1) The spatial correlation in 2010 was slightly stronger than that in 2000. Such variation may be the result of accelerated urbanization: the rapid development of urban regions leads to the concentration of population and resources. (2) The percent change of SoVI from 2000 to 2010 indicates an upward trend in social vulnerability and is emphasized in Yangtze River Delta. The identification regarding highly vulnerable and rapidly changing regions has practical significance for protection and risk mitigation.
The dominant factors contributing the most to SR can be discovered through PCA because of high explained variance [
49]. Accordingly, stakeholders are provided with a benchmark reference to enhance SR and reduce social vulnerability. Occupational and economic structures are the top two drivers that represent approximately 50% of variance in 2000 and 2010. Occupation and economy are closely related: if a large number of people work in tertiary industries, then the economy will greatly advance [
83]. The tertiary industry incudes transportation, warehousing, and postal services; information transmission, software, and information technology services, wholesale and retail, financial industry, leasing and business services, and scientific research and technical services [
84]. Shanghai, Tianjin, Guangzhou, and Zhejiang chartered by high GDP are prosperous regions of the tertiary industry in China’s coastal zone [
85], the SR of which is also at a high level (
Figure 5). Therefore, a society’s capability to resist disruptive events can be strengthened by accelerating industrial transformation and developing production and service industry.
Social vulnerability is the product of the social inequality and historic patterns of social relations [
29,
62] that influence or shape the susceptibility and adaptive capacity of various groups to harm. Social inequality, which includes economic vitality, families and households, age, infrastructure, literacy, and disability, can be improved by socially based services such as housing, welfare, health, and education. However, historical factors, such as class, ethnicity, race, and gender, manifest as deeply embedded social structural barriers that are resistant to change. Although social disparities are inadvertently perpetuated, we can reduce it through transnational flows of capital/goods, information/ideas [
86], and implementation of new social policies.
Significant differences can be explored between urban and rural areas. The economically developed regions, such as Yangtze and Pearl River Deltas, and the metropolises, including Tianjin, Qingdao, Suzhou, Fuzhou, Xiamen, and Shantou, have a high SoVI. The percent change of SoVI in these places is almost at the maximum level. Such trend can be explained by the “urban restructuring” thesis [
63,
87,
88], which focused on the influences of capital and labor restructuring on urbanization. The rapidly growing economy and large demand for labor attract a sizeable migrant population pour into metropolises. Accordingly, the PE of those areas increases and the less developed regions may face hardship. The migrants become vulnerable because the presence of inequality makes it difficult for them to find good paying jobs or affordable housing [
87,
88] and participate in a political planning process for having their voice heard [
62]. Migrants tend to cluster in highly developed but vulnerable areas. Consequently, the vulnerability of such migrants in the region increases. In this context, the government must promote inter-regional coordinated development and pay great attention to the floating population referring to equality [
89]. The high-resolution distribution patterns exhibit that hot spots almost concentrate in the downtown of each administrative unit. Developing suburban and satellites can alleviate pressure in the city center and reduce vulnerability.
Our study still has certain limitations. Data availability is a crucial factor during variable selection, especially in the third level of Chinese administrative hierarchy. In this study, we mainly used the fifth (2000) and sixth (2010) national population census data, which are the latest and highly accurate data that we could obtain to simulate SR at a county level. Thirty-three variables were selected to calculate SR, but they were insufficient. Several indicators were excluded because of certain limitations, such as income and consumption level, infrastructure construction, individual health status, transportation and communication facilities, and disaster awareness. The limitations can also lead us to rely on readily available variables, which are inevitable to be overlapped; however, they are not the highly appropriate indicators for social vulnerability [
14,
49,
56].
The SoVI has been broadly applied to reflect multidimensionality, reduce complexity, and visualize results. However, whether the pre-event descriptions correspond to post-disaster outcomes is an ongoing issue. Few attempts to validate the SoVI metric were found, and extreme event losses have been regarded as a frequently proposed approach [
40,
90,
91,
92]. Cutter et al. [
40] suggested that SoVI in a post-event situation, such as Hurricane Katrina, can be validated by comparing the losses and recovery result of different affected regions with the previously calculated SoVI. Rufat et al. [
90] compared the outcomes of Hurricane Sandy with the SoVI constructed by four types of models. However, the correlation coefficients were low. The proposed method assumed that when facing the same grade pressure, the economic loss is great if the social vulnerability of a region is high, which is seldom the case [
40,
48], because natural disasters are geographically heterogeneous. Moreover, the poor individuals are sensitive, but they do not have much to lose.
Consistency is a prerequisite for scientific assessments [
54]; thus, the same indicators were selected in 2000 and 2010 in this study. However, the same parameters may not capture changes that happened in the decade [
8]—new and important factors may be affecting social vulnerability in 2010. We tried to make indicators comprehensive to remedy this recognized defect. However, deficiencies still persist due to data limitations. Subjectivity and arbitrariness are inevitable when selecting variables. Thus, the established evaluation index system should be revised and further calibrated when it is applied to other cases [
47].
6. Conclusions
This study developed a social vulnerability assessment metric for China’s coastal zone at the grid level. The quantitative indicator models of social vulnerability were based on two dimensions: PEI, which expresses the exposure degree of a system, and SRI, which refers to the system’s capability to absorb, adapt to, and recover from a disturbance. Multisensor remote sensing data, demographic and socioeconomic data, GIS techniques, RF model, and PCA were used to develop 2000 and 2010 SoVIs. The spatial distribution and LISA clustering maps of SoVI indicate that dissemination areas with high social vulnerability values are concentrated in Yangtze River Delta, Pearl River Delta, and eastern Guangzhou. Such areas are densely populated and economically prosperous. The maximum increase occurs in the Yangtze River Delta, which is categorized as the H–H cluster, denoting the region that deserves great attention.
The spatiotemporal comparison analysis of the SoVI value and the two dimensions that compose social vulnerability (PE and SR can be an effective tool for risk reduction planning. However, whether the indicator model of social vulnerability truly represents such abstract concept remains unaddressed. Thus, the validation method is worthy of in-depth research in the future.