Physical Risk Indicators

Due to their frequency and high damage potentials, hurricane risks and impacts are at the forefront of coastal vulnerability research, especially hurricane-landfall probabilities in the Atlantic Ocean, Caribbean, and Gulf of Mexico (Gray, Klotzbach, and Landsea, 2003). Other risk indicators include the maximum exceedence probabilities for category-1 hurricane winds at the county level (Jagger, Elsner, and Niu, 2001) and susceptibility to inundation from hurricane storm surge (Mercado, 1994; National Hurricane Center, 2002). All of these indices are entirely based on the physical characteristics of the natural hazard with little or no attempt to address the human dimension of vulnerability.

While hurricane exposure dominates the coastal vulnerability literature, beach erosion is also important. Monitoring changes in littoral profiles to develop signatures of erosion is one approach used to quantify coastal erosion (Kana, 2003). Bryan et al. (2001) used GIS-based spatial models to determine regional vulnerability to inundation and erosion along portions of the south Australian coast. Zhang, Douglas, and Leatherman (2001) developed an erosion-potential index based on storm tides, wave energy, and storm duration from Nor'easters for much of the Atlantic Coast. Erosion hazards were the focus for a national study of historic erosion rates along US coastlines (H. John Heinz III Center, 2000a). This Heinz Center study included both physical and social attributes in the spatial delineation of properties, historic erosion rates, and future projections of economic losses. Unfortunately, the study did not consider changes in the built environment or changes in the rate of erosion as it made its forecast, but the report does take an important step toward identifying and measuring human-environment interactions, a necessary prerequisite for understanding place-based vulnerability.

Recently, the US Geological Survey (USGS) produced the Coastal Classification Atlas, which identified areas along the southwestern Florida coast that were susceptible to high wave action from coastal storms (Morton and Peterson, 2003). This study categorized the shoreline's geomorphic structures and ranked the density of development. Designed as a baseline for future vulnerability assessments, this study was limited in geographic coverage, but it does provide an initial prototype for integrating human and physical systems in the understanding of place-based vulnerability.

Another component of coastal ecosystem vulnerability is based on environmental degradation indicators, such as environmental sensitivity analyses (Jensen et al., 1993, 1998; Noronha et al., 2003), GAP analysis, and state Heritage Program rankings of ecological vulnerability (H. John Heinz III Center, 2000b). Gulf Coast states are working with the US Minerals Management Service to determine those on-shore regions most susceptible to degradation from oil pollution (H. John Heinz III Center, 2000b). Unfortunately, these studies are focused entirely on ecosystems and habitat degradation with no linkage to social vulnerability.

Recent interest in the impacts of global climate change resulted in a plethora of research on coastal vulnerability to sea-level rise. Among the first regional analyses of climate-change impacts were those conducted by Gornitz, Beaty, and Daniels (1997) and Gornitz and White (1992, 1994), who developed an index of vulnerability to sea-level rise along US coasts. The original study compared coastal segments based on seven physical parameters, including maximum wave height, shoreline erosion, and accretion rates, and landform susceptibility to inundation. The Geological Survey of Canada (GSC) and the USGS enhanced the initial Gornitz studies, creating an improved Coastal Vulnerability Index (CVI) (Hammer-Klose and Thieler, 2001; Shaw et al., 1998; Thieler and Hammer-Klose, 1999, 2000a, 2000b). The USGS baseline data have far fewer data gaps than its predecessors and provide a better metric of physical vulnerability at the county level.

Social Vulnerability and the Coast

Many of the social and economic characteristics that influence the vulnerability of individuals and communities along the coast are known at a conceptual level. However, this knowledge has not been translated into empirically based assessments of the socioeconomic vulnerability of coastal communities (H. John Heinz III Center, 2000b, 2002). Recently, the social vulnerability index (SoVI) provided a comparative spatial assessment of human-induced vulnerability to environmental hazards (Cutter, Boruff, and Shirley, 2003). This index explained around 76% of the variation in socioeconomic vulnerability in US counties. While this method was designed for all counties in the US, a subset of coastal counties (not including those in the Great Lakes, Alaska, and Hawaii) produced comparable results (80% of the variation in social vulnerability explained) (Boruff, Cutter, and Emrich, 2002).

The use of a quantitatively derived social vulnerability index, such as SoVI, is important for two reasons. First, the method provides a useful tool for comparing the spatial variability in socioeconomic vulnerability using a single value derived from multivariate characteristics. Second, SoVI can be linked (statistically and spatially) to more physically based indices in calculating the overall vulnerability of a specific place. Not only does this index make a significant contribution to the methods and metrics used in vulnerability science, but it also provides important comparative information for policy makers and emergency managers.

Index Construction

To create an index of social vulnerability for the 213 US coastal counties, we replicated the methods first developed by Cutter, Boruff, and Shirley (2003). Of the 42 socioeconomic variables used in the original study, only 39 were available in the 2000 US Census (Table 1). The socioeconomic variables were placed in a principal components analysis (PCA). Using the varimax rotation option, 10 factors with eigenvalues greater than 0.95 were extracted. These factors explain 82% of the variance among US coastal counties (Table 2). All factors were scaled so that positive values indicate higher levels of vulnerability, while negative values decrease vulnerability. In those instances where the effect was ambiguous (e.g., age where elderly would load positively and younger children negatively), the absolute value was used. Once extracted and scaled, the factors were placed in an additive model (making no a priori assumption about the relative importance of each) to produce the overall coastal social vulnerability score (CSoVI).

Determining the physical vulnerability of US coastlines to environmental hazards followed the procedures used by the USGS (Thieler and Hammer-Klose, 1999, 2000a, 2000b). The USGS formulated the spatial representation of data for each coastline through hand-digitized line segments. We utilized the same six variables for each line segment (Table 3). These segments were then broken down to conform to county coastline extents, thus permitting analyses at the county level.

To calculate the CVI, each variable was ranked on an ordinal scale between one and five following the USGS methodology (Thieler and Hammer-Klose, 1999, 2000a, 2000b). CVI was then computed as the square root of the product of all ranked variables for each line segment divided by n (6),

The USGS data, however, are based on characteristics applicable only to the coastlines for which they were derived. Therefore, each individual variable is not directly comparable with the same variable on another coast. To create a regionally comparable index, the USGS then reranked CVI on an ordinal scale from one to four (Thieler and Hammer-Klose 1999, 2000a, 2000b). The reranked CVI was used in our analysis to test for regional differences in physical vulnerability.

The overall place vulnerability index (or PVI) is an additive model derived by summing the CVI (physical) and CSoVI (social) scores for each county. Due to the range of values for CVI and CSoVI, z scores were first calculated for each index as a means for creating comparable scales. To visually represent the extremes of the data, the PVI for all coastal counties was mapped into three categories (low, medium, and high) using the standard deviations from the mean as the classification scheme. This same procedure was used in mapping the constituent parts, physical vulnerability (CVI) and social vulnerability (CSoVI).

Determining Relative Importance

To determine the dominant influences on coastal vulnerability, socioeconomic or physical characteristics, two procedures were used. First, to test the degree of physical and socioeconomic influences on place vulnerability (PVI), a standard linear regression was performed with PVI as the dependent variable and all physical and socioeconomic variables as independents. Using a dependent variable composed of the independent variables violates statistical assumptions, but in this instance, the independent variables were not used to model the dependent variable, rather, the procedure was used to determine the relative influence of each on the aggregate vulnerability index (PVI) using the standardized beta coefficients.

The second procedure involved the examination of the physical and social characteristics of the least and most vulnerable counties. First, the standardized beta coefficients for all the independent variables were grouped into physical and socioeconomic categories and then summed. The average standardized beta coefficient for each group (physical and socioeconomic) was computed to see which group of variables explained the most variance in the model. The mean standardized beta coefficient was used due to the difference in the total number of variables for each category.

Regional Variability and Comparisons

An analysis of variance (ANOVA) tested for regional differences in the overall place vulnerability (PVI), social vulnerability (CSoVI), and physical vulnerability (CVI). Using a frequency histogram for each region, the normality of the data (as a whole and for each coast) was examined. Results of the ANOVA show that there are significant differences (at the 95% confidence level) for each of the indices (PVI, CSoVI, and CVI) between the Atlantic, Pacific, and Gulf Coasts (Table 4). These differences are not surprising given the diversity of physical characteristics of the coasts described earlier.

In addition to ethnic and racial disparities, economic differences exist between the regions. The average per capita income of Pacific Coast counties is twice that of both the Atlantic and Gulf coastal counties. Not surprisingly, the median value of owner-occupied homes in Pacific Coast counties is twice that of Gulf Coast counties and one and one half times that of Atlantic Coast counties. The density of commercial establishments on the Atlantic Coast is twice that of the Pacific Coast and 10 times that of the Gulf Coast counties. Finally, the earning density (a measure of county wealth derived from earnings in all industries standardized by square miles) of Atlantic coastal counties is twice that of Pacific Coast counties and 20 times the earning density of counties along the Gulf Coast.

Atlantic Coast

Mapping the CVI for the Atlantic Coast shows high index values are clustered in the mid-Atlantic counties, particularly in North Carolina and Virginia (Figure 1). For the most part, these counties have a gentle coastal slope, less than a two-foot mean tidal range, are eroding at a rate of more than 1 m per year, and are categorized as a barrier island, sand beach, salt marsh, mud flat, or delta. Mathews County, Virginia, is a good example of a county categorized as highly vulnerable based on physical characteristics (Table 5). Those counties exhibiting low levels of physical vulnerability tend to group along the North Atlantic coastline in the New England states. On average, low physically vulnerable counties experience more accretion than erosion, have steeper slopes, are less affected by sea-level rise, and are categorized as having rocky, cliffed coastlines with glacial and alluvial deposits. Cumberland, Maine, is an example of a coastal county with a low physical vulnerability score.

From a socioeconomic perspective, the same geographic pattern appears, with high levels of social vulnerability clustered in counties along the mid-Atlantic Coast and low levels of social vulnerability along the North Atlantic shoreline (Figure 1). Counties with high levels of socioeconomic vulnerability have large percentages of persons in poverty, those receiving social security benefits, African Americans, and female-headed households. Conversely, those counties with low levels of social vulnerability have little poverty and unemployment, few mobile homes, and low levels of international immigrants. New York (Manhattan borough) has the highest social vulnerability score, while Poquoson County, Virginia, has the lowest (Table 5).

North Atlantic coastal counties have the lowest levels of place vulnerability (Figure 1) of all the regions. The most vulnerable counties again are grouped in the mid-Atlantic region—along the North Carolina, Virginia, and Maryland coastlines. The major exceptions to this are metropolitan New York counties and the south Florida counties of Miami-Dade, Indian River, St. Lucie, and Martin, each having relatively moderate levels of physical vulnerability but high enough levels of socioeconomic vulnerability that, when combined, produce a high PVI value. Northampton, Virginia (on the Delmarva peninsula), and Cumberland, Maine (Portland), have the highest and lowest levels of overall vulnerability, respectively (Table 5).

Gulf Coast

Gulf Coast counties with the highest levels of physical vulnerability are clustered along the Louisiana and Mississippi coasts (Figure 2). These counties are characterized by barrier islands, sand beaches, salt marshes, mud flats, and deltas, with less than a meter tidal range, little slope, and high sealevel rise and erosion rates. Counties with low levels of physical vulnerability have coastlines with greater slopes and lower erosion rates. These counties are concentrated along the western portion of the Florida panhandle. Plaquemines County, Louisiana, has the greatest physical vulnerability, while Hillsborough (Tampa Bay area) has the least (Table 5).

The highest levels of socioeconomic vulnerability are found in counties along the southern Texas coast and along the northwestern portion of Florida's west coast (Figure 2). There was an interesting split among the factors that contributed to high levels of social vulnerability in each of these two areas. In Texas, high CSoVI values are due to large percentages of Hispanic persons, international immigrants, and high levels of poverty and unemployment (e.g., Cameron County, Texas) (Table 5). Along the Florida coast, high CSoVI values are attributed to large elderly populations, numerous mobile homes, and large numbers of Social Security-benefit recipients. For those counties with low levels of socioeconomic vulnerability (found sporadically along the northern coastline of the Gulf of Mexico, such as Jefferson Parish, Louisiana), there is no dominant single indicator.

Mapping place vulnerability results in a different pattern all together (Figure 2). Counties with high PVI values are located along the southern Texas coast and in Louisiana, while counties with low levels of vulnerability are located along the western portion of the Florida panhandle and around the Tampa Bay area. The counties with high PVI values along the Texas coast are due to high levels of socioeconomic vulnerability, while counties along Louisiana have high levels of physical vulnerability. Low PVI values along the western portion of the Florida panhandle are a function of low values on both the physical and social indicators.

Pacific Coast

Pacific Coast counties with high physical vulnerability are located in central Oregon and in the San Francisco Bay area (Figure 3). These counties have low, cliffed coastlines, cobble and sand beaches, alluvial plains, deltas, and estuaries. As expected, counties classified as highly vulnerable based on the physical indicators had a steeper slope and higher erosion rates. San Francisco County, California, is an example of the most physically vulnerable county in the Pacific Coast region, while Clallam, Washington, bordering the Strait of Juan de Fuca, has the lowest physical vulnerability (Table 5).

As with the Gulf of Mexico, there was a considerable variability in the social vulnerability of the Pacific Coast (Figure 3). Counties in the San Francisco Bay area with high levels of socioeconomic vulnerability have large percentages of Asian and Hispanic persons and increasing populations coupled with already dense residential and commercial development. Other counties on the Pacific Coast with high CSoVI values, such as Del Norte, California, have large Native American populations, large percentages of female-headed households, and relatively large numbers of Social Security-benefit recipients. Pacific counties with low levels of socioeconomic vulnerability have less ethnic diversity and lower levels of commercial and residential development. A good example is Lane, Oregon (the host county for the Oregon Dunes National Recreation Area).

The overall place vulnerability of Pacific Coast counties is greatest in the San Francisco Bay area (San Francisco and San Mateo counties) and is a function of socioeconomic factors (Figure 3). Counties with low levels of place vulnerability appear in a random geographic pattern (Table 5).

The first question posed in this research asked whether there was regional variation in the place vulnerability of US coastal counties. Results from the ANOVA and the spatial analysis found regional differences not only in place vulnerability but also in physical and socioeconomic vulnerability as well. These findings, especially in the case of physical differences, are consistent with the extant literature. Socioeconomic differences appear between each coast, but there are few studies to either support or refute this finding at the present time.

Social and Physical Influences of Vulnerability

Determining the greatest influence on the overall vulnerability of each region (physical or social characteristics) was the second objective of this article. To assess this, a simple regression analysis was performed using PVI as the dependent variable and the CVI and CSoVI as the independent variables. For all coasts, the average standardized beta co-efficient for physical variables was 0.239, while the average standardized beta coefficient for the socioeconomic variables was 0.046 (Table 6), initially suggesting physical characteristics are the greater determinant of overall vulnerability. This conclusion holds for the Atlantic and Pacific Coasts. For Gulf Coast counties, however, the average standardized beta coefficient for socioeconomic variables was larger than the average standardized beta coefficient for physical variables (0.409 and 0.360, respectively). This suggests that, in this region, social vulnerability is the greater driver of place vulnerability (Table 6).

When examining individual counties, the influences are more apparent. For example, it is the combination of both high levels of social and physical vulnerability that produce San Francisco's ranking (Table 7). Yet, in Cameron, Texas, the coastal vulnerability index is quite low but is overwhelmed by the social vulnerability component. In the case of Cumberland, Maine, low values on both physical and social indicators combine to produce a score representing the least vulnerable county in the United States. If only the top 10 most vulnerable counties are examined, social characteristics are as dominant as physical ones in determining vulnerability.

Individual Indicators of Vulnerability

When aggregated for all coasts, sea-level rise is the most important single indicator of place vulnerability, followed by slope, mean wave height, percent over 65, and the density of commercial development (Table 8). For the Atlantic Coast, density of commercial development is the single most important contributor in the model, with slope, sea-level rise, mean wave height, and percent 65 and over following closely behind (Table 8). In the case of the Gulf Coast, percent 65 and over has the greatest influence on place vulnerability, followed by birth rate, sea-level rise, mean wave height, and median age. For the Pacific Coast, percent Asian explains the greatest variation in the model followed by housing-unit density, erosion/accretion rate, percent renter, percent females in the labor force (Table 8).