An Integrated PCA–AHP Method to Assess Urban Social Vulnerability to Sea Level Rise Risks in Tampa, Florida

Abstract

Coastal flooding risks are increasing with the rise of sea level due to global climate change. Such risk presents different groups of residents with diverse vulnerability. Using a coastal city as the study area, this study quantitatively and qualitatively examines social vulnerability across different communities at risk of sea level rising. This study develops a novel social vulnerability assessment method that integrates principal component analysis (PCA) and analytical hierarchical process (AHP), inputting the advantages of each into factor analysis for social vulnerability quantification. Twenty-four socioeconomic factors are first grouped into four different themes. PCA is then performed to identify major components within each theme. We apply AHP to generate the weighting scheme for each theme. Therefore, the AHP-derived weights to those PCA components according to each theme are used to obtain an overall social vulnerability index. The thematic map of PCA–AHP SVI showed that minority communities with low income, mobile home, and unemployed populations aggregated in the East Tampa shore (over 69% of the total minority), compared with the West Tampa (31% of the total minority). Our findings provide insightful spatial information of the social vulnerability index (SVI) that allows decision makers to make optimal efforts to mitigate damages and unexpected impacts on different communities, especially those marginalized minorities due to sea level rising risks.

1. Introduction

Sea level rise has become a major concern around the world, especially for coastal cities, due to climate change and global warming. The Climate Central recently reported climate change induced storms reaching four feet above the high-tide line could happen twice as often as now [1]. The recent IPCC (The Intergovernmental Panel on Climate Change) report projects that the global mean sea level will rise between 0.43 m under the RCP (Representative Concentration Pathway) 2.6 to 0.84 m under the RCP 8.5 by 2100 relative to 1985–2005 [2]. It is expected that the coastal areas will experience more frequent and severe high coastal tide flooding with the acceleration of sea level rise.

The understanding of social vulnerability in a coastal city is highly desired to help local governments to identify the communities that may need support before, during, and after extreme disasters. For example, before and during the occurrence of disasters, efficient plans could be made to organize resident evacuation, especially for those vulnerable populations who may need special help, such as minorities, the elderly, disabled people, those without vehicles, or underserved groups who do not speak English well. Additionally, the relevant organizations can prepare enough funding and potential needs to support the highly vulnerable communities for post-disaster recovery [3,4,5]. Hence, the social vulnerability index (SVI) is a critical quantitative measure to prioritize populations living in areas of high risk to sea level rise flooding.

In general, there are three types of SVI computation methods that have been widely accepted to measure social vulnerability to various natural disasters [3], including deductive, inductive, and hierarchical approaches, as shown in Figure 1a–c.

Deductive approaches, also known as theory-based approaches, usually only consider a small number of social factors for a study area and aggregate them to compute the overall social vulnerability index using the across-the-board sum. Hierarchical approaches divide social factors into multiple groups or themes. One commonly used approach is the CDC SVI method, which is composed of four themes including socioeconomic status, household composition and disability, minority status and language, and housing type and transportation. It first performs sub-indices analysis to compute social vulnerability scores for each individual theme by aggregating all its social factors, and then calculates the overall index by aggregating the scores of all the themes. Unlike both deductive approaches and hierarchical approaches, inductive approaches, also known as data-driven approaches, can consider a large number of social factors and perform factor analysis, such as principal component analysis (PCA), to find a small number of principle components (or themes). Then, for each component, its driving social factors can be identified using correlation analysis, and the overall index can be computed by summing all driving factors. The Cutter’s SoVI [4] serves as the foundation of the inductive approach, based on which many recent social vulnerability studies about various disasters have been performed. We summarize representative studies of these three types of approaches in

Table 1. Some representative studies of the traditional social vulnerability study approaches in the literature.

Methods	Representative Studies	Descriptions
Deductive Approach	Cutter et al., 2000 [8]	Population, housing units, gender, race, income, and mobile homes are considered to measure the index to chemical release disasters.
	Montz and Evans, 2001 [9]	Gender, age, education, family structure, length of residence, occupation, and tenure are considered to measure the index to flash flood disasters.
	Wu et al., 2002 [10]	Population, housing units, gender, race, age, income, and house structure are considered to measure the index to sea level risk flooding.
	Collins et al., 2009 [11]	Population density, gender, age, disability, income, education, and local government revenue are considered to measure the index to flooding- and transportation-related hazardous material disasters.
Hierarchical Approach	Vincent, 2004 [12]	Five themes, including economic well-being and stability, demographic structure, institutional stability, global interconnectivity, and natural resource dependence, are considered to measure the index to climate-change-induced water availability issues in Africa.
	Chakraborty et al., 2005 [13]	Three themes, including population and structure, differential access to resources, and population with special evacuation needs, are considered to measure the index to hurricane hazards.
	Flanagan et al., 2011 [14]	Four themes, including socioeconomic status, household composition and disability, minority status and language, and housing type and transportation, are considered to measure the index to all general natural disasters, also known as CDC SVI.
	Mustafa et al., 2011 [15]	Three themes, including material, institutional, and attitudinal, are considered at the household and community levels to measure the vulnerabilities and capacities index (VCI).
Inductive Approach	Cutter et al., 2003 [4]	One of the most popular methods, called the Social Vulnerability Index (SoVI), is proposed in this study, which applies factor analysis to choose major components and correlation analysis to select driving factors of each component. A z-score method is used for standardization of the final scores.
	Finch et al., 2010 [16]	The SoVI was used to understand disaster disparities and differential recovery after the Hurricane Katrina in New Orleans.
	Schmidtlein et al., 2011 [17]	The SoVI was used to measure social vulnerability for earthquake losses in Charleston, South Carolina.
	Roncancio et al., 2020 [18]	The SoVI was used to measure pre-existing social vulnerability as a first step in national disaster risk reduction and climate change adaptation planning in Colombia.
	Jackson et al., 2021 [19]	The SoVI was used to measure social vulnerability to COVID-19 diseases in United States at the county level.

.

The current deductive approach is usually the simplest one, but since only a small set of social factors are considered, the social vulnerability analysis might neglect some important social groups and the results could be very inaccurate. Recent comprehensive comparative studies [3,5] have concluded that the hierarchical approaches are the most accurate, while the inductive approaches are the most precise. However, the current hierarchical approaches (e.g., CDC SVI) use a fixed structure of social factors and do not consider a particular type of disaster or geographic area, while the current inductive approaches (e.g., SoVI) group social factors into components or themes using factor analysis, and there are not any thematical connections in themes or their driving factors, i.e., some factors could be shared by multiple themes. Motivated by this, this paper aims to develop a new social vulnerability index approach that leverages the advantages of these two types of approaches, i.e., hierarchical modeling and inductive analysis. In addition, as pointed out by many researchers (e.g., [6]), subjectively weighting schemes among different factors or themes is inevitable in social vulnerability analysis, but the existing-factor-analysis-based approaches can only find weights from the data.

To address the above-mentioned knowledge gap, our approach further integrates the analytic hierarchy process (AHP) method [7] to allow weighting different themes based on experts’ input. For this reason, our proposed new method is named PCA–AHP integrated SVI, as shown in Figure 1d. Moreover, while some researchers have recently studied spatial demographic characteristics of urban communities which suffer various levels of flooding risks [7], there is limited research that explores the social vulnerability in coastal flooding due to the sea level rise under the increasing threats of global warming. This paper at the same time attempts to fill the knowledge gap of the spatial vulnerability of residents in a coastal city to sea level rise flooding, especially those underserved communities and marginalized groups. In short, our PCA–AHP integrated SVI method first organizes social factors into different themes, then applies PCA to each theme to generate major components which are further aggregated with weights of their eigenvalues for theme social vulnerability. Lastly, we leverage AHP as a weight scheme to quantify the weights of those themes, which then are applied to the summary of an updated social vulnerability index. In summary, this paper has the following contributions: (i). a new SVI modeling method is proposed to integrate experts’ input to prioritize social themes for vulnerability analysis; (ii). Underserved communities and marginalized groups are identified in a coastal city to sea level rise flooding, and (iii). The results show that the new PCA–AHP SVI method outperforms the existing approach. Below, this study introduces the study area in Section 2, data in Section 3, methods in Section 4, results and discussions in Section 5, and conclusion in Section 6.

2. Study Area

In this research, we chose the bay area of the Tampa City as our study area, which is located midway down Florida’s west coast, near the Gulf of Mexico (Figure 2). Tampa, the biggest city in the Tampa Bay area, has the largest open-water estuary in Florida with a total of about 400 square miles. Originating northeast of the city and flowing into the Hillsborough Bay, the Hillsborough River divides the downtown of Tampa into two geographies, the East Tampa and the West Tampa.

The 2020 U.S. Census shows that the population of the city of Tampa was 335,709 in total, including a white population of 211,217, African American 87,872, and Hispanic 77,472 etc. The city of Tampa has been attracting many new residents every year to settle down [20]. Tampa is of great interest and importance to understand social vulnerability, which is critical for coastal planning and assessing climate mitigation and urban sustainability.

3. Data

In this study, we first used the high-resolution LIDAR-based accurate digital elevation model (DEM) data to identify flooding zones for a given sea level rise. Then, from the flood census blocks, we leveraged their demographic census data to evaluate social vulnerability to the flooding.

Four datasets have been used in this study, including a LiDAR-based digital elevation model (DEM) dataset used to determine high-risk sea level rise flood areas with a coastline dataset, a parcel-level property GIS dataset which provides information about values and units of properties which are located in the high-risk flood areas, and a demographic census dataset which includes measurements of various social factors at the tract level. In particular, the LiDAR-based DEM dataset is part of a series of DEMs built by the National Oceanic and Atmospheric Administration Office (NOAA). The DEMs are processed to be “hydro flattened”, i.e., water elevations are less than or equal to 0 m. It has a 3-m spatial resolution (cell size) with a vertical accuracy of 10 cm. The parcel-level property GIS data used in this study are collected for the city of Tampa from the Florida Department of Revenue (i.e., https://floridarevenue.com/property/, (accessed on 9 October 2021)) with a local horizontal accuracy of 5 feet at a 95% confidence interval. It contains various information of each individual properties, including assessed value, land use type, property size, number of units, and so on, which are collected yearly by property appraisers and tax collectors in local governments. Following our previous study [21], we first used the DEM dataset and the coastline dataset to identify all possible flood areas whose elevations are below a given sea level and are also connected to the coastline (see Methods section for details), and we used the property GIS dataset to summarize the total value of properties and lands which are expected to be flooded for a given sea level. Similar to our previous study, only subsets of data for the coastal areas of Tampa were extracted from their whole datasets.

Additionally, this study used the demographic data which are part of the American Community Survey (ACS), 2015-2019 (5-year) dataset. These data are at the level of census tracts that are small subdivisions of counties. Each census tract generally has the number of residents between 1500 and 8000, with a typical size of 4000 people. While mapping these data in a spatial map can effectively show geographic patterns of various potential vulnerability across different tracts to the disaster, this study chose 24 variables as socioeconomic factors to measure social vulnerability for possible sea level rise flood areas. These selected social factors fall into four different themes, including socioeconomic status, race and gender, household composition, and family special needs. Understanding social vulnerability to sea level rise flooding can greatly help disaster mitigation, disaster preparedness, disaster response, and disaster recovery.

4. Methods

To estimate the future socioeconomic impact of sea level rise for the population of the study area, this study first identified areas with high flooding risks, i.e., those areas whose elevations are below a given sea level. Then, we calculated the social vulnerability index over these locations using the proposed new PCA–AHP SVI method to assess the populations’ vulnerability.

4.1. Flooding Area Identification

We first followed our previous study [21] to extract all possible areas to be flooded in the target area. Here, we briefly summarized this flooding area identification procedure using an ArcGIS tool, and interested readers are referred to our previous study [21]. The following procedures were performed, including a raster calculation procedure and a reclassification procedure which identified all areas whose elevations are below a given sea level, and a spatial intersection procedure which only kept those areas which are connected to the coastline by intersecting a coastline layer. In particular, we had:

i.

Raster calculation: this aimed to produce a new raster image P which only included those areas below a given sea level h by comparing them with a DEM raster image D:

$$p=\left\{\begin{array}{c}0\\ 1\\ No Data\end{array}|\begin{array}{c}\mathrm{if} d>h\\ \mathrm{if} d\le h\\ d=No Data\end{array}\right\}$$

(1)

where d and p are the variables for the input DEM image D and the output raster image P, respectively.

ii.

Pixel reclassification: this aimed to reclassify the raster image P so that only flooding areas were kept using the following equation:

$$g=\left\{\begin{array}{c}1\\ No Data\end{array}|\begin{array}{c}p=1 \left(i.e., d\le h\right)\\ p=0 \mathrm{or} p=No Data \left(i.e., d>h \mathrm{or} d=No Data\right)\end{array}\right\}$$

(2)

where g is the new cell value in the output raster image G. It shows that all unflooded areas are assigned as No Data, and only those areas below the given sea level h remain.

iii.

Spatial intersection: However, these flooding areas identified in Step 2 may include areas whose elevations are lower than h but are not connected to the ocean. Examples of such outliers include lakes, ponds, and valleys, which are not flooded due to the sea level rise and should be removed. This can be done by selecting only those flooding areas that spatially intersect with the coastlines provided by the NOAA. We first converted the raster data G to the polygon feature class consisting of a set of polygon features S = $\left\{s_1,s_2,\dots s_N\right\}$, each of which denotes a possible flooding area whose elevation is lower than the sea level. We supposed that N polygons were identified in total. Given a coastline L, we checked if each of N polygons intersects with L or not, and only kept those intersects, using the following equation:

$$S=\left\{\begin{array}{c}S-\left\{s_i\right\} \\ S\end{array}|\begin{array}{c}\mathrm{if} s_i \cap L=\varnothing , \mathrm{removed} \\ \mathrm{otherwise} \end{array}\right\}, (i=1,2,\dots , N)$$

(3)

With the final list of flooding areas S, we could further identify all current properties (e.g., houses, buildings, and other infrastructure) in flooding areas by spatially overlapping the property map with flooding areas. We leveraged the property tax dataset, which provided house-level GIS information. In this property map, each property was represented by a polygon, and only those properties which were overlapped with one flooding area in S were identified as flooding properties.

4.2. An Integrated PCA–AHP SVI Method

While CDC SVI has been well recognized as an effective hierarchical metric to measure the vulnerability of social groups when facing various disasters, it still suffers from the following major issues: first, for each theme, only a small number of social factors are considered, indicating that many other social factors are neglected; secondly, all of its social factors are summed with same weights [14], indicating that all social factors have the same importance or contribution to the theme score; similarly, all themes contribute the same to the overall social vulnerability score [14]. These issues may lead to an underestimated social vulnerability analysis using the overall CDC SVI score. To address the above issues in CDC SVI, in this section, we present a new SVI metric by integrating PCA and AHP methods with the goal of being able to include more and diverse social factors and apply weighted importance to the overall SVI, i.e., different weights for individual themes. The architecture of the proposed PCA–AHP SVI method is illustrated in Figure 1d, and here we detail PCA and AHP methods individually in the following subsections.

4.2.1. PCA for Factor Analysis

As shown in Figure 1d, similar to CDC SVI, we adopted a hierarchical modeling method which groups all selected social factors into different themes. While this approach is quite general and can consider as many social factors as needed for different applications, in this study, 24 social factors or variables were chosen according to the data available in the census data as shown in

Table 2. Social vulnerability variables for coastal Tampa areas in Florida, USA.

Concept	No.	Variable name	Description
Socioeconomic Status	1	V_EMPLOYMENT	Percentage of population 16 years and over not in labor force
	2	V_POVERTY	Percentage of families with income below the poverty level
	3	V_FOOD	Percentage of families who receive food stamps
	4	V_INSURANCE	Percentage of population without health insurance
	5	V_NODIPLOMA	Percentage of population 25 years and over without diploma
	6	V_HIGHSCHOOL	Percentage of population 25 years and over regular high school diploma
	7	V_NOSCHOOL	Percentage of population 25 years and over with no schooling completed
Race and Gender	8	V_BLACK	Percentage of Black population
	9	V_ASIAN	Percentage of Asian population
	10	V_ISLANDER	Percentage of native Hawaiian and other Pacific Islander population
	11	V_NATIVES	Percentage of American Indian and Alaska native
	12	V_OTHERS	Percentage of all other races
	13	V_FEMALE	Percentage of female population
Household Composition	14	V_SINGLE	Percentage of families with single parent
	15	V_SINGLEEMPLY	Percentage of families with single parent who is also employed
	16	V_FESENIOR	Percentage of female population that are over 65 years old
	17	V_FEEMPLY	Percentage of females who are employed
	18	V_SENIOR	Percentage of population 65 years or older
	19	V_YONG	Percentage of population under 14 years
Family Special Needs	20	V_DISABILITY	Percentage of families with disabilities
	21	V_LANGUAGE	Percentage of population speaking English less than well
	22	V_NOVHICLE	Percentage of families with no vehicle
	23	V_MOBILEHOME	Percentage of mobile houses
	24	V_MOBILEPOP	Percentage of population living in mobile houses

.

The communities that have a higher vulnerability are usually associated with higher percentages of populations of underrepresented groups, particularly minorities, families below poverty, those who are unemployed, single-parents, low income, low education, or need special needs (e.g., disability services, language supports, and transportation). These 24 independent variables for each census tract can be grouped into four themes: socioeconomic status, race and gender, household composition, and family special needs. Since each theme is comprises a number of social factors, data redundancy may occur, and it is necessary to reduce the data to allow for a robust and consistent social vulnerability analysis over time with the change of global warming. Therefore, there is a need to perform factor analysis, and here the PCA method is adopted to decompose the original data into a smaller number of components or dimensions.

PCA takes a table with N observations (rows), each of which has P variables (columns). Here, we assumed that one theme included P social factors, and the PCA method output P eigenvectors or components and their corresponding eigenvalues. Mathematically, PCA performed the transformation of the original data vector (e.g., one observation) to a new component vector. Since the eigenvalue could represent the total amount of variance to be explained by the component, these P components were usually sorted in a descending order by their eigenvalues, i.e., the first component had the largest eigenvalue. We only kept the first few components that could explain the majority of variance of the original data for each theme, for example, at least 85% variance used in this study. We supposed that the first $m_i$ components that could explain at least 85% variation of the original variables were selected for the i-th theme. Next, we computed scores for each theme by summing these components weighted by their corresponding eigenvalues:

$${\mathit{s}}_i={\displaystyle \sum }_{j=1}^{m_i}{r}_{ij}{\mathit{w}}_{ij},\hspace{1em}i=1,2,3,4$$

(4)

where ${\mathit{s}}_i$ denotes the score of each theme which is a vector of all N considered flood census blocks, ${\mathit{w}}_{ij}$ denotes the j-th eigenvector for the i-th theme, and $r_{ij}$ is the corresponding eigenvalue.

After this, we performed standardization to normalize these theme scores $s_i$ to obtain the theme indices, which ranged from 0 to 1. Here, we leveraged the percentile index which was also used in CDC SVI for standardization. In particular, we calculated a percentile rank for each census tract for each of these theme scores. The percentile rank of a score was the percentage of scores in a frequency distribution that were less than that score. The following formula was calculated for the percentile rank as its social vulnerability index:

Percentile Rank = (Rank − 1)/(N − 1)

(5)

It would be noticed that when ties occur, the smallest of the corresponding ranks will be assigned.

4.2.2. AHP

AHP for Theme Weighting

Once the four theme indices were computed, we further combined them to compute the overall social vulnerability index. Instead of applying the same weight to all themes, here we adopted the analytical hierarchical process (AHP) to assign weights to each theme. AHP is a commonly used approach to support multi-criteria decision making based on quantitative assessment [22]. A pairwise comparison among all relevant criteria is performed to arrive at a scale of preference, based on the best knowledge and experience of decision makers [23,24,25]. The pairwise comparison has the range from 1 to 9, indicating equal importance to extreme importance, as shown in

Table 3. Scale for pairwise comparison (according to Saaty and Vargas, 1991 [28]).

Intensity Importance	Description
1	Equal Importance
3	Moderate Importance
5	Strong Importance
7	Very Strong Importance
9	Extreme Importance
2, 4, 6, 8	Intermediate Values
Reciprocals	Inverse Comparison

. Firstly, the AHP builds pairwise comparisons for all 4 themes and creates a 4 × 4 matrix as shown in

Table 4. Comparison matrix for social vulnerability index computation.

Matrix	Socioeconomic Status	Race and Gender	Household Composition	Family Special Needs
Socioeconomic Status	1	1/3	3	1/2
Race and Gender	3	1	5	2
Household Composition	1/3	1/5	1	1/2
Family Special Needs	2	1/2	2	1
Total	6.33	2.03	11	4

. In the matrix, each row shows the value that is the result of the comparison of two social vulnerability themes. For example, the first row of Table 4 shows the comparison between the socioeconomic status and all other indicators. Here, the socioeconomic is more important than the household composition (with a weight of 3), and it is much less important than the race and gender indicator (with a weight of 1/5). However, the comparison was inversed when we compared the race and gender indicator with the socioeconomic status indicator in the second row. Next, we normalized this matrix by dividing each column by corresponding sums [26,27], as shown in

Table 5. Normalized matrix for social vulnerability index computation.

Matrix	Socioeconomic Status	Race and Gender	Household Composition	Family Special Needs	Sum	Mean	Weight
Socioeconomic Status	0.16	0.16	0.27	0.125	0.715	0.18	1.80
Race and Gender	0.47	0.49	0.46	0.50	1.92	0.48	4.80
Household Composition	0.05	0.10	0.09	0.125	0.365	0.09	0.92
Family Special Needs	0.32	0.25	0.18	0.25	0.99	0.25	2.48
Total	1.00	1.00	1.00	1.00	4.00	1.00	10

. Then, we computed the average value of each row multiplied by 10 as the weights of individual themes for the overall social vulnerability index computation. Lastly, the final social vulnerability index was computed by finding the percentile rank of the weighted score based on AHP.

The Consistency Check of AHP

Consistency of the weighted matrix using AHP needed to be checked before its use in the weighted social vulnerability index in Table 5. This could confirm the reliability of the weighting scheme to find the overall social vulnerability index. Following previous AHP works [22,28], we used the following equation to check the consistency:

$$CR=\frac{CI}{RI}$$

(6)

where CR denotes the consistency ratio, CI denotes the consistency index, and RI denotes the random index for social vulnerability indicator (

Table 6. Random index (RI) for social vulnerability indicators.

N	1	2	3	4	5	6
RI	0	0	0.58	0.90	1.12	1.24

).

The Consistency Index (CI) can be computed using the following equation:

$$CI=\frac{{\lambda }_{max}-n}{n-1}$$

(7)

where ${\lambda }_{max}$ is the largest eigenvalue of the matrix, which can be calculated as follows: First, the column total of each theme in Table 4 are multiplied by the mean value of corresponding themes in Table 5; then, these values are summed to get the value of ${\lambda }_{max}$. In addition, n is the number of themes that are considered.

According to the theory of AHP, the weights are consistent if the value of CR is less than 0.10, also indicating the reliability of the matrix. In this study, we had RI = 0.9, n = 4, ${\lambda }_{max}$ = 4.1178, CI = 0.0393, and CR = 0.0436. Since CR was less than 0.10, it suggests that the weights are consistent, and we can rely on the weighting matrix. Then, we had the final social vulnerability index model after analyzing each theme separately and assigning the weights to each theme using AHP. Below, we obtained the following final model to calculate the overall social vulnerability score for each census block:

The Overall Social Vulnerability Score = 1.80 × Socioeconomic Status + 4.80 × Race and Gender + 0.92 × Household Composition + 2.48 × Family Special Needs

5. Results and Discussions

5.1. Flooding Properties with Sea Level Rise

It has been observed that the global sea level has risen about 21–24 cm since 1880 with the increasing global temperature, and the trend has been accelerating for the past decades. In 2020, the global mean sea level was 0.9 cm above the 1993 average. The most recent IPCC report (AR6) concluded that it is virtually certain that the global mean sea level will continue to rise over the 21st century after running a set of global climate models. It is likely that the global sea level will increase by 0.3 cm under the low greenhouse gas pathway, and 2.5 m under a pathway with high emissions by 2100. Hence, we analyzed the spatial distribution of flooding properties with sea levels ranging from 0.1 to 2.5 m. Figure 3a–d shows the summarized flooding properties in terms of the total amount of properties, their current property values, land values, and land areas, respectively. It shows a linear increase in the number of flooding properties and total land values with the rise of sea level, while dramatic increases are observed for the total property values at the sea level of 1.2 m, and for the total land areas at the sea level of 2.1 m.

5.2. PCA–AHP Integrated Social Vulnerability Index

This section summarizes the results of social vulnerability analysis over the flood areas if the sea level rises by 1 m, a possible sea level rise by 2100 which is projected by IPCC under the RCP 8.5.

5.2.1. Theme Components of Social Vulnerability

For each theme, we performed PCA to identify the major components that can explain at least 85% of the variance, as shown in Figure 4, Figure 5, Figure 6 and Figure 7.

In particular, we kept the first three components for the theme of socioeconomic status that could explain 88% of the total theme variance, two components for the theme of race and gender that could explain 90% of the total theme variance, three components for the theme of household composition that could explain 93% of the total theme variance, and three components for the theme of family special needs that could explain 85% of the total theme variance. These components were further summed with the weights of their corresponding eigenvalues to compute the social vulnerability score for each theme. Using the percentile rank, the theme’s social vulnerability index was calculated, which ranged from zero to one. An index of one indicated the highest vulnerability and an index of zero indicated the lowest vulnerability. We mapped four theme SVI indices in Figure 8, Figure 9, Figure 10 and Figure 11 based on four categories: 0.00–0.25 (low vulnerability), 0.25–0.5 (intermediate low vulnerability), 0.50–0.75 (intermediate high vulnerability), and 0.75–1.00 (high vulnerability). These maps clearly visualize how these social themes are spatially distributed across flooding areas.

5.2.2. The Overall SVI

The overall SVI can be computed from the percentile ranks of the four themes using an across-the-board sum method with weights computed by the AHP method. Figure 12a maps the overall SVI across the areas which are projected to be flooded when the sea level is raised 1 m. It can be observed that: First, the East Tampa shore is dominated by high overall SVIs, except for several areas, including the Kitchen Nature Preserve and the Fred and Idah Shultz Nature Preserve, with very small populations. Second, while most areas in the West Tampa shore have low overall SVIs, several areas including Bay Crest Park and the Bayside West have high social vulnerability mainly due to the high percentage of senior people who are vulnerable against flooding disasters, although they may have high income. Lastly, we observed that some themes play an important role in the calculation of the overall social vulnerability for some areas, particularly the race and gender theme, which has the largest weight toward the overall SVI. For example, we have seen that Port Tampa (the most southwest of the map) has a high overall social vulnerability because of the high social vulnerability for the theme of socioeconomic status and the theme of race and gender.

5.3. Comparisons with CDC SVI

In this subsection, we compared the overall social vulnerability index obtained by our integrated PCA–AHP SVI method with the traditional CDC SVI method. Since the CDC SVI measures the social vulnerability in a structure shown in Figure 1b using an across-the-board sum of the percentile ranks of all four themes where all themes have the same weight to the overall SVI, some important social groups could be underestimated in its overall CDC SVI results, particularly the minorities that are emphasized in our PCA–AHP SVI method. Here, we show the overall CDC SVI for the flooded areas in Figure 12b. Compared with our PCA–AHP SVI results shown in Figure 12a, it shows that the overall CDC SVI is generally lower than the overall SVI in flooded areas of the West Tampa shore. For example, Port Tampa (the most southwest of the map) has a very low overall CDC SVI but exhibits a high overall PCA–AHP SVI. The reason for such differences between CDC SVI and PCA–AHP SVI is that Port Tampa is composed of a large population of minority groups.

To better illustrate this, we created a map to show the distribution of minorities in the study area. In Figure 12c, the percentage of the minority population is categorized into four groups using the natural break method. Port Tampa has a very high minority population (as shown in the left figure of Figure 12c), but the CDC SVI (the left figure of Figure 12b) results in the lowest vulnerability values, which means CDC SVI significantly underestimates the highly aggregated minority communities in Port Tampa. However, our PCA–AHP method (the left figure of Figure 12a) estimates high SVI values for Port Tampa. Similarly, in the northwest of Bay Crest Park, the vulnerability is also significantly underestimated. On the other hand, the CDC SVI method significantly overestimated vulnerability in other Tampa coastal areas; for example, an obviously overestimated big community area is the central East Tampa shore district with a low percent of minorities, but the CDC SVI method resulted in the highest vulnerability values. In contrast, our PCA–AHP obtained reasonable lower SVI for this district. In CDC SVI, the minority is just one vulnerability indicator in the theme of minority status and language, and hence its contribution to the overall SVI could be easily canceled out by many other indicators; an example of this is the language indicator, given the fact that African Americans can speak English well. In our PCA–AHP SVI, race is reasonably assessed and weighted by experts as one of the main indicators to the calculation of the overall SVI. Interestingly, both CDC SVI and our PCA–AHP SVI share only one similar pattern of social vulnerability in the Northeast Tampa shore, where the percentage of minority populations is high, because the vulnerability of this area is also highly influenced by the population with a low socioeconomic status (i.e., Figure 8) and high special family needs (i.e., Figure 11). It further reveals that East Tampa consists of 69% minorities in all sea level rise flooding areas, while West Tampa has the remaining 31% minorities.

In summary, the CDC SVI applies the same structure to all US states and counties to the assessment of social vulnerability to various disasters. While it has the advantage of a “one-for-all” metric without the need of any modifications or refinements for natural or manmade disaster situations, the CDC SVI does not consider the specific characteristics of a disaster in a geographic area. In contrast, our integrated PCA–AHP SVI used in this paper leverages factor analysis and the AHP method to weight all proper social, economic, and demographic factors, which should be screened first before applying them to social vulnerability quantification. The proposed PCA–AHP SVI showed significant advantages over the CDC SVI in integrating experts’ understanding into factor weighting scheme, which optimizes the quantification of diverse factors’ contributions to the social vulnerability index. The PCA–AHP has significant potential to be applied to the social vulnerability assessment of many other disasters.

5.4. Discussions

In this research, we analyzed the knowledge gap of the spatial vulnerability of residents in a coastal city to the sea level rise flooding using our social vulnerability index. In particular, we evaluated 24 social factors from the aspects of social economic status, household structure, race, gender, age, employment, education, and special needs. Notably, high risk areas are usually dominated by poor groups who might be a minority with a low income, a language barrier, or a disability. These residents are usually not well prepared for the occurrence of any disastrous event. In this paper, we proposed an integrated PCA–AHP method to measure the social vulnerability index. The proposed PCA–AHP SVI method first applies factor analysis to identify the number of components to be considered for each theme, followed by an AHP method which finds the weights of individual themes to compute the overall SVI.

Our PCA–AHP results also show that most of the highly vulnerable residents who are usually associated with a low income, mobile home, unemployed, and minority residents live in the East Tampa shore, compared with the residents living in the West Tampa shore. Although all these flooding areas with high and low social vulnerability are near to the coast, we find that the high social vulnerability areas are typically surrounded by wetlands or industrial fields that have long distances to the beach, but conversely, the low social vulnerability communities are in the area with open water resource or easy access to the well-developed beach.

Compared with the existing CDC SVI method, the advantage of our PCA–AHP SVI method is that it can integrate experts’ understanding and decisions to prioritize social themes for a particular type of disaster in a specific geographic area. The CDC SVI method is a hierarchical method which defines a hierarchy of social factors. Following the hierarchical structure of social factors, researchers compute the overall social vulnerability index for any target area, and thus the CDC SVI method can be called a “one-for-all” metric for all US states or regions and for any disasters. In contrast, our new SVI method allows us to weigh factors and themes differently based on the characteristics of the data and/or the target area. Given this fact, the PCA–AHP SVI method obtains significantly different results for most areas in West Tampa from the CDC SVI method, particularly for those areas with a high percentage of minorities, while the CDC SVI method underestimates their social vulnerability because it mainly applies the same weight of all factors to the SVI calculation, which could cancel out the contribution of the minority factor to vulnerability measurement. In contrast, our proposed PCA–AHP SVI method not only simplifies data input using PCA but also optimizes the most important step of weighing factors in SVI calculation using the AHP approach. Methodologically, the PCA–AHP SVI method could significantly improve the SVI estimations, which was proved in urban communities’ vulnerability to sea level rise in Tampa City as demonstrated in this study.

This study further provided fine geographic information of populations and buildings that will face flood risk with the rise of sea level. The findings of this study help us to understand which areas along the shoreline are more vulnerable to floods under different sea levels, where most populations and properties could face flood exposure, what their values could be, and how the vulnerability spatially varies across different social groups and locations. Although some areas may have a small number of populations, they should not be neglected or even be paid more attention as they may be more vulnerable than others due to their low incomes, limited access to the public resources and information, and/or special needs for flooding evacuation and mitigation. The locations and values of those buildings to be flooded and populations to be impacted could be good information for local governments to allocate optimal financial budgets and resources for sea level rise flooding mitigation. Hence, this study makes it possible to make early planning and take optimal actions to reduce potential damages and impacts on social, economic, and environmental aspects, because both population and buildings are identified, quantified, and mapped. This integrated PCA–AHP GIS analytics method for coastal flooding exposure assessment and mapping of the social groups of populations and their vulnerabilities can be applied to other coastal cities against sea level rise.

However, there are some limitations of this study, which we plan to address in our future research. Firstly, we used current geographic and social data to reveal impacts of sea level rise which will occur in the future. It is necessary to understand how geographic features (e.g., shorelines [29] and flooding areas [30]) and social groups change over time. Secondly, this paper only selected one study area for method validation, and more extensive comparative studies in other regions with the same or different climates are required.

6. Conclusions

Coastal flooding has become a growing threat to residents who live on the coast due to the rise of sea level with global warming. Such a threat will also be presented to an increasing population under rapid coastal urbanization. There is a high need to understand the areas to be flooded and the different levels of vulnerability of different groups of residents who live in these areas. This paper presented a new social vulnerability measurement method, which leverages both factor analysis and AHP method to quantitatively and qualitatively map social vulnerability across different groups. Our results revealed that populations living on the East Tampa shore are generally much more vulnerable than ones living on the West Tampa shore. While the East Tampa shore is mostly surrounded by many wetlands or industrial fields, where there are more populations with low incomes, mobile homes and no transportation, the West Tampa shore is urbanized very well with easy access to open water, in which more populations with high incomes live. However, there are several areas to be flooded in the West Tampa shore with high minority populations which are vulnerable to flooding disasters.

While only one coastal city has been studied in this paper, the proposed method can be easily applied to other areas and/or regions which face increasing sea level rise flooding risks due to climate change. As a part of our future work, we plan to perform studies on other urbanized coastal regions with the same or different climates using the proposed PCA–AHP method. Moreover, this paper does not consider the evolution of coastline over time which may have impacts on the identification of sea level rise flooding areas, and we plan to integrate coastline extraction methods using satellite images in future studies.