Identification of vulnerability-generating mechanisms, key determinants, and indicators

Following a literature review we specified several mechanisms through which rapid coastal urbanization, i.e. rapid urban population increase, typically generates vulnerability.

Coastal cities are already disproportionately exposed and sensitive to cyclones [19,31] and floods [19,32]. Sections of most of the largest African coastal cities are currently at risk of flooding [33–35]. Examples of cities that are subject to increasing exposure and sensitivity to climate extremes include Dhaka (storm surges) [36], Sorsogon City, Philippines (taifuns) [37], Mumbai, Rio de Janeiro, and Shanghai (floods) [21]. The areas at risk are expected to significantly increase in the near term by urban land expansion [14] which is more rapid in low elevation coastal zones (defined as the contiguous area along the coast that is less than 10 metres above sea level) than in other places [4]. It causes large-scale land cover change in LICs and MICs [4,38], i.e. in countries that are generally experiencing much higher levels of urban expansion and population growth [39]. Unprecedented degradation and destruction of wetlands surrounding coastal urban agglomerations through increased demand for land [40] and subsequent encroachment [3,38,41] destroys important ecosystem functions and services for the urban and surrounding populations, in particular flood regulation [41–43]. This was shown in New Orleans, after the landfall of Hurricane Katrina in 2005 [44]. Depending on the specific ecosystem functions the wetlands provide for the urban population, their degradation has serious implications for urban livelihoods [45].

Rapid urbanization is increasing the vulnerabilities in urban populations by overstretching municipal management and planning capacity [6,10,11]. The rapid population growth from in-migration and internal population increase can overwhelm basic urban services, especially if municipal adaptive capacity is initially low. This overstretch occurs when new problems arise before long-standing ones have been dealt with [6]. Future climate change, for example, is expected to further stretch the management capacity in coastal areas [21,46,47] Overstretched management has been well documented in case studies, e.g. in Dhaka, Dar es Salaam [47] and Mumbai [21]. In the remainder of this article, the term, management’ comprises both management and planning.

Rapid population increases are often absorbed into the urban fabric through an increase in densely populated informal settlements [11]. The growth in informal settlements is largely driven by poverty and marginalisation of poor populations in and around megacities in developing countries [41,48] and is frequently underestimated [49]. Informal settlements have less capacity to deal with shocks, e.g. climate-related extremes [43,50,51] such as tropical cyclones and floods [52], and less capacity to deal with the ensuing negative impacts [53]. This particularly applies when poor management, and low building and infrastructure quality coincide with densely populated areas [11]. Marginalization of the poor has been observed inter alia in numerous coastal cities in India [54] and in Iliolo City in the Philippines [55]. Under inadequate urban management and unchecked growth, informal settlements commonly encroach more risk-prone areas where flood and cyclone exposure is high [56–58]. These areas are avoided by wealthier populations [11]. This leads to an increase in the number of vulnerable populations, population density, and low-quality buildings in floodplains. Informal settlement development in exposed areas has been observed in cities such as Lagos [34], Mumbai [59], Esmeraldas, Ecuador [60], and many other large African coastal cities [48].

Although future estimates of the additional number of people at risk from coastal flooding vary widely [61], all indicate a considerable increase due to surging populations in low-lying areas and to sea-level rise. This aspect becomes even more serious as the projections for future sea-level rise are adjusted upwards recently (e.g. [62]). The ongoing superimposition of sea-level rise on current flood levels increases the level of vulnerability in coastal cities to climate extremes [63]. This poses a major challenge for coastal management in terms of adapting to both rising storm surge levels and rising flood levels [8,64]. We summarize the major findings of the above-cited literature on urban socio-ecological vulnerability by seven key determinants and their interactions in Fig 1. This figure is also a template for all diagrams in Fig 2 in the discussion section, which each displays a typical and particularly problematic interaction of some of the key determinants.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Key determinants and their interactions, describing socio-ecological vulnerability under rapid coastal urbanization.

Arrows indicate the direction of influence.

https://doi.org/10.1371/journal.pone.0220936.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Six typical configurations of socio-ecological vulnerability of urban populations under rapid coastal urbanization.

Based on Fig 1 (key determinants of vulnerability and how they influence each other) each diagram is a typical and particularly problematic configuration. Red boxes and arrows signify particularly problematic situations and influences based on indicator values, grey boxes signify less problematic situations and influences. Green boxes and arrows refer to entry points for vulnerability reduction. The number of each diagram refers to a corresponding subsection in the Discussion section. The colored boxes in the lower right hand corners indicate which urban vulnerability profiles each diagram applies to. LDCs stands for Least Developed Countries.

https://doi.org/10.1371/journal.pone.0220936.g002

As depicted in Fig 1 rapid urban population increase typically increases the exposure and sensitivity of urban populations to climatic extreme events under overstretched or unwise urban management. The increase in exposure is also driven by unmanaged urban expansion into risk prone areas, and by sea-level rise. The increase in exposure and sensitivity is also driven by the marginalization of poor populations into risk prone low-lying areas and dense informal settlement development, and by degradation of flood regulating ecosystems such as wetlands through unchecked urban expansion, when urban management cannot effectively regulate development.

We identified 11 indicators with global coverage to render quantitative information on the key determinants (Table 1, [102]). We did not impose a hypothesized predefined relationship between the indicators, following Kok et al. [25]. Instead, we (inductively) explored the structure in the data-space to obtain insights into typical vulnerability patterns. In Table 1 each indicator is attributed to one of the three vulnerability components commonly used in frameworks for vulnerability analysis in the context of global environmental change: exposure, sensitivity and adaptive capacity [65–67]. Thereby we understand ‘damage’, e.g. from floods, as ‘exposure times sensitivity’.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Indicators and indicator datasets used, including their assignment to vulnerability components, the key determinants they are related to, and the level of spatial data aggregation.

https://doi.org/10.1371/journal.pone.0220936.t001

The reasons for choosing each dataset, and details on data resolution and data treatment, are provided in the Supporting Information. A spatial resolution of 0.5°lon x 0.5°lat (30 arc minutes), which is compatible with the integrated assessment model IMAGE [77], was chosen for the indicators: The analysis presented in this paper focused on eliciting the most recent patterns of vulnerability based on a common time frame for all the data required. This is extensively discussed in the section “Discussion of the method”. However, there is a clear interest in how the patterns evolve over time, especially under future climate change. Integrated assessment models such as IMAGE can provide consistent data for future projections for this analysis, taking into account different assumptions and uncertainties. Using data projections from the IMAGE model Lüdeke et al. [30] already analyzed projected changes in patterns of smallholder vulnerability to global environmental change in global drylands.

Many of the datasets used for the indicators were on a much finer scale, such as 3 arc seconds for [76], 2.5 arc minutes for [75], and 5 arc minutes for [68]. These datasets were aggregated to 30 arc minutes for the analysis.

Identification of vulnerability profiles and their spatial distribution

We applied the established cluster analysis method k-means to integrate the 11 datasets and to identify typical combinations in the data structure [26,30]. This method is suitable for large datasets and generates rather compact clusters [78] which fits our objective to identify typical vulnerability profiles, represented by the cluster center. The optimal number of clusters in the data set was identified by applying a consistency measure (see, e.g., 15). After multiple repetition of the clustering process and the corresponding allocation of grid cells to clusters, this measure analyzes the stability of the results depending on the cluster number. If the pre-given number of clusters fits the underlying structure in the analyzed data set the results become more stable (see Supporting Information for further details on the cluster analysis and consistency measure (including S1 Fig)).

The Fraiman measure was employed to investigate the importance of each indicator in shaping the clusters. With this measure we were able to determine how sensitive the cluster partition was to each single indicator by fixing it at its mean-value (so-called ‘blinding’ of indicators), and comparing ‘blinded’ runs of the clustering with the runs that included all indicators without blinding [79].

Group I: Fastest population increase and pronounced poverty under overstretched management

The red and black profiles in Fig 4 are characterized by extreme (i.e. severe) manifestations of problems: the most rapidly increasing coastal populations go together with the highest slum-population levels of all profiles (i.e. the highest averages for these two indicators). The co-occurrence of the lowest increase rates in urban expansion hint at relatively dense informal settlement development. Judging by the lowest values for indicators of adaptive capacity—government effectiveness and income–urban management in this group is severely overstretched by such rapid growth and urban poverty. This indicates that the urban areas characterized by these profiles are in an extremely challenging position to reduce vulnerability. The two profiles significantly differ in cyclone occurrence and–less significantly—in damages from climate extremes.

Group II: Rapid population increase and most rapid expansion intensify high flood damages under moderate adaptive capacity

The three group II profiles (grey, yellow, purple) display the highest flood occurrences and damages, the highest totals of low-lying settlement, the fastest urban expansion rates, and relatively high wetland prevalence. In mechanistic terms, this indicates common urban expansion into flood-prone low-lying areas. The product of expansion rate and wetland prevalence is similar for the three profiles, indicating a high group-typical endangerment of the remaining natural flood protection. The three profiles subsumed under group II differ significantly in cyclone occurrence and damages, and differ to some extent in slum population levels and income.

Group III: Few and less severe problems under slower population increase and high adaptive capacity

Group III profiles are characterized by the highest average income and government effectiveness, relatively low marginalization, and low occurrence of climate extremes. In mechanistic terms, the dark and light blue profiles are under the least pressure, and most favorably positioned to further reduce the few discernible problems: Urban management does not appear to be overstretched, and is subject to handling the “slowest” population increase rates. The two profiles subsumed under group III differ significantly in flood sensitivity, and less significantly in adaptive capacity.

4.1. Severely overstretched urban management (Group I and Bluefields profiles)

Vulnerabilities to global change can outpace vulnerability reduction in rapidly growing LICs and MICs [7,9]. If overstretched management–indicated by ineffective government under low average income under a multitude of severe problems—is a proxy of this risk, then our study suggests that vulnerability reduction is being, or risks being, outpaced in over 50 (60%) out of 84 countries exhibiting rapid coastal urbanization (Fig 2, 2.1.). The three profiles that show the clearest signs of management overstretch (Sittwe, Maumere, and Bluefields profiles) account for over a third (36%) of all rapidly urbanizing coasts, and 145m people (5% of the global urban population) in 57 countries.

4.2. Vulnerability-generating mechanisms are dramatically stacked against coastal urban areas in LDCs (Group I profiles)

Inspection of all profiles shows that income and government effectiveness decrease with the rate of urban population growth, while urban poverty increases. The urban areas in group 1 (red Maumere and black Sittwe profile) exhibit both characteristics in extremis: They are simultaneously subject to the lowest income, most ineffective governments, most prevalent poverty, and world’s most rapid coastal population growth rates (Fig 2, 2.2.). This makes is clear that vulnerability-generating mechanisms are dramatically stacked against the fastest-growing and most impoverished coastal urban areas in the world, which at the same time have the least adaptive capacity to reduce vulnerability. This applies to many LDCs along Western and Eastern African tropical coasts, Yemen, Pakistan, Myanmar, the Eastern Indonesian Archipelago, Solomon Islands, and Vanuatu (Maumere profile). In the Sittwe profile the situation is additionally aggravated by high levels of vulnerability to climatic extreme events: This means that urban areas in Madagascar, northern Mozambique, Small Island Developing States, Myanmar, and parts of the Philippines are subject to a “worst case reality” of well documented drivers of coastal urban vulnerability. In these areas hydro-meteorological climate extremes are not stopping rapid urban population increase. Klose and Webersik [93,94] have shown that this is the case for population development and tropical storms in Haiti.

4.3. LDCs are the locus of the fastest but least expansive urbanization (Group I profiles)

Unsurprisingly, urban areas in all rapidly urbanizing coastal fringes are expanding. However, coastal urban population growth is most rapid where increase rates in urban area are slowest, i.e. in the red Maumere and black Sittwe profiles. The severe levels of slum population in these profiles suggest that the swell of inhabitants is largely absorbed on limited area by existing informal settlements and the erection of new ones (Fig 2, 2.3.). According to literature urban areas are generally increasing at twice their population growth rates [39], and population density is generally stagnating or in decline [4,39]. On the other hand, none of the 30 representative cities Angel et al.’s findings are based on are located in the Maumere or Sittwe profile. Therefore, further investigation of population density change in these urban areas could nuance the ratio between area and population growth rates on the extreme end of the urban spectrum, i.e. for the fastest growing cities with high slum population levels.

4.4. High flood vulnerability is typical, yet a result of different combinations of causes (Sittwe and Group II profiles)

In the past, flood vulnerability in urban areas has been largely driven by socio-economic mechanisms: Noted causes include marginalization of the poor, ecosystem degradation, and poorly governed rapid urbanization [19,51,54,95]. By indicating each of these causes, as well as flood occurrence, damages, and adaptive capacity, this study shows two things: that severe values of these causes are typically combined to generate high flood vulnerability, and in which coastal fringes they co-occur.

For one, the noted causes co-occur in extremis in the impoverished black Sittwe profile in LDCs, as discussed in 4.2 (Fig 2, 2.4a). These causes translate into large damages from floods, cyclones, or both. The causes also co-occur in all profiles of group 2 (grey Brus La Laguna, yellow Cebu, and purple Bluefields profiles). Here, however, high flood vulnerability is better explained by severe manifestations of different causes, namely the most rapid urban expansion rate and highest total of low-lying settlements (Fig 2, 2.4b). This co-occurrence supports the finding that expansion is more rapid in the low elevation coastal zone [4]. In these profiles, we conclude that flood exposure is driven by large-scale flood-prone expansion into low-lying areas and rapid population growth. Flood sensitivity is driven by massive loss of wetlands (Fig 2, 2.4.). On the one hand, this causal combination can lead to particularly high flood vulnerability under low government effectiveness (Bluefields profile). On the other hand, the same causal combination leads to lower flood vulnerability under more effective government (Cebu profile). This is noteworthy, because in this profile occurrence of and damages from floods and cyclones are extremely high, and its low-lying areas along the subtropical coasts in Asia and Central America, the majority of The Philippines, China, Vietnam, Bangladesh and India (Bay of Bengal) are heavily populated.

4.5. Patterns of most rapid coastal development are in direct conflict with current and future climate adaptation (Bluefields, Muisne and Group I profiles)

Urban areas with the most rapid growth rates for population and urban expansion, respectively, are precisely the urban profiles other vulnerability-generating mechanisms are also most severe in. Based on this finding, it can be concluded that the spatial patterns of the most rapid growth rates for population and urban expansion circa the year 2000 are in direct conflict with vulnerability-reduction efforts to climatic extreme events (Fig 2, 2.4.). This conflict can be expected to worsen under potential future increases in exposure and sensitivity to flooding and cyclones. We suggest that urban areas are particularly sensitive to future increases if their profiles exhibit a) very low current (circa the year 2000) exposure and adaptive capacity (Fig 2, 2.5a), or b) high sensitivity (damage/occurrence ratio) (Fig 2, 2.5b).

Case a) applies to the 51m people in 41 countries (18% of the rapidly urbanizing coastal fringe) constituting the impoverished Maumere profile. Increases in exposure to flooding, e.g. due to climate change, may cause particularly severe managerial problems: The very low occurrence level circa the year 2000 and very low adaptive capacity suggest high damages once exposure increases. This situation is not acting as an early warning signal for potential future increase in damages because of sea-level rise, storm surges, and pluvial and fluvial flooding. Case b) applies to 196m people in 58 countries (38.5% of the rapidly urbanizing coastal fringe) in profiles with low and high income levels—the Bluefields profile, Sittwe and Muisne profiles. Currently, low to moderate occurrence in these profiles already translates into similar damages observed under much higher occurrence in other profiles. This suggests a low capacity for climate-related disaster prevention or post-disaster management. This deficiency is particularly concerning, because ongoing sea-level rise is co-occurring with high values of rapid growth, urban poverty, and low-lying settlement.

4.6. Building on government effectiveness and experience provides entry points for vulnerability reduction (Cebu profile)

Generally speaking, it is critical for LICs and MICs is to reduce risks from climate change to urban areas through mindful and integrated strategies for urban land-use planning, channeling settlement, and climate adaptation [96]. In this context, Huq et al. [51] indicate that pro-poor urban climate adaptation policies are needed. These policies are particularly effective under the involvement of households, communities, but also various levels of government [8]. Measures of upgrading informal settlements through collective organization have shown potential to reduce vulnerability to extreme climate events in multiple African and Asian cities [90].

Ideally, jointly enhancing a) the adaptive capacity of flood-sensitive slum populations and b) the effectiveness of government-led reduction of exposure and sensitivity would benefit all flood-vulnerable profiles (Group II and Muisne profiles) (Fig 2, 2.6). A conceivable strategy for reducing the causes of severe exposures is diverting extremely rapid urban expansion from heavily populated, and remaining unpopulated, risk-prone low-lying areas. Likewise, a conceivable strategy for reducing a cause of severe exposures is to protect the limited remaining wetlands from further encroachment. Another effective response is the community-based initiative “Homeless People’s Federation of the Philippines” [90,97]. If pronounced social disparities indicate poor top-down organizational abilities to implement such measures, such as in the Muisne profile, or governments lack resources or political will, collective organization in the affected communities may be a more promising way to reduce vulnerability [21].

4.7. Discussion of the method

This study provides the first approximation of the recent situation of coastal vulnerability of urban populations under rapid urbanization. The data sets for the identification of this global typology for coastal urban vulnerability under rapid urbanization are available at [98]. Recently the opportunity for a follow-up study, which employs the most recent data, has emerged.

We conducted our analysis between 2010 and 2015 with the aim of providing the most recent approximation of the current situation. At the time of this analysis, the most recent year for which all of the georeferenced indicator datasets were available was the year approx. 2000. For this common year of reference we used the best available datasets. The common dataset reference to the year 2000 also avoided retro-causality between indicator datasets.

We recognize that new data has become available in recent years for most indicator datasets. For example, the 2009 Global Assessment Report on Disaster Risk Reduction has since superseded the data from Dilley et al. [75] on floods and tropical cyclones by offering a higher resolution and improved modeling of risk frequency and severity [99].Further datasets became available after this analysis was completed. For example, a newer version of the Gridded Population of the World (GPW) v.4 dataset was published in 2016 [100]. However, the global data wetlands prevalence dataset has not been updated to a more recent time frame.

Our results can be interpreted as a missing baseline for commonalities and differences in vulnerability between rapidly urbanizing coastal areas based on the most recent and complete available data at the time of analysis between 2010 and 2015. Overall, we did not redo the analysis with the more recent datasets, because our complete indicator set in a common time frame (2000) supplies a useful near-recent baseline in its own right: First, a baseline to compare to does not exist yet. Second, now that more recent data has become available, the opportunity can be seized to use the same methodology with the latest data. Comparing these results with our baseline would elicit how vulnerability patterns change over time. Follow-up studies can be also based on projected indicator data from an integrated assessment model such as IMAGE. In proof of this concept, Lüdeke et al. [30] have already analyzed changes in the typology of smallholder vulnerability to global environmental change in global drylands using projected data from IMAGE.

The analysis has a few limitations. The first limitation is the specificity of an indicator, which can show a large spread around its cluster center value (i.e. stretched central quartiles, see Fig 5). This means that the cluster membership of a grid cell may not say much about each specific indicator. In this case it is important that the cluster center value of such an indicator does not play a decisive role in the interpretation of the cluster or profile group it belongs to. The second limitation is that our results do not distinctively address specific urban agglomerations, or cities, according to their administrative boundaries, but rather grid cells. The reason for this design is clear: It is not feasible to unambiguously assign all comparable subnational data to rapidly urbanizing coastal cities globally. The third limitation is the use of national-level data when subnational data was either unavailable (e.g. for government effectiveness), or unfeasible (for GDP per cap), for capturing local or regional vulnerability-generating phenomena in urban areas. At the same time, national level GDP per cap differentiates between LICs, MICs, and HICs. This can help indicate the resources a national economy can mobilize for vulnerability reduction, e.g. for climate change adaptation [90]. We ruled out using the subnational dataset of ‘‘gross cell product” per cap [101], because numerous coastal countries with rapid growth lack the data.