1 Introduction

Empirical studies of residential flood vulnerability have generally concentrated on either the geophysical characteristics or the socio-economic characteristics of a given region, rather than a combination of the two. Geophysical characteristics mainly considered include drainage, rainfall, topography, elevation, location in flood prone areas, and lack of green space (Hughes 1982; Penning-Rowsell 1996; Campana and Tucci 2001; Nyarko 2002; Afeku 2005; Tsheko 2003; Xeflide and Ophori 2007; Zheng et al. 2008; Karley 2009). Other studies have considered land use, land cover classification, and distance to a major river (Wassman et al. 2004; Dewan et al. 2007; Forkuo 2011; Fosu et al. 2012).

Socio-demographic variables such as a household’s ability to evacuate and recover after floods, frequency of experiences of flooding, living in the same area for a period, adequate sanitation services, proper city planning, social networks, poverty, and overpopulation have been identified as having effects on flood vulnerability. Individual characteristics, including gender, age, and number of children, have also been identified as influencing flood vulnerability (Chan and Parker 1996; Mustapha 1998; Wong and Zhao 2001; Andjeilkovic 2001; Kolsky and Butler 2002; Bruijn 2004; Afeku 2005; Chakraboty et al. 2005; Shaw et al. 2005; Lebel et al. 2006; Brouwer et al. 2007; Fekete 2009; Nguyen and James 2013).

Furthermore, assessments of socio-demographic characteristics and flood vulnerability have largely ignored materials used in housing structures, e.g. wall and floor materials. However, these construction materials could influence flood vulnerability. In addition, studies using subjective assessments of flood vulnerability are not very common. However, due to the fact that people’s perceptions influence their risk behaviour, and therefore their vulnerability, understanding perceptions about a phenomenon is very significant for the design of effective communication, mitigation as well as coping and adaptation strategies. (Bankoff et al. 2004; Heitz et al. 2009; Ho et al. 2008). This study intends to fill this knowledge gap. It is important to study floods particularly in this setting because empirical downscaling of GCM simulations by Hewitson and Crane (Christiansen et al. 2007) suggests wetter trends along the Gulf of Guinea, where Accra, the metropolitan capital of Ghana, is located, and heavy rains are the most common cause of floods (Few 2003).

1.1 Vulnerability to climate change

Vulnerability to climate change is the degree to which geophysical, biological, and socio-economic systems are susceptible to, and unable to cope with, adverse impacts of climate change (Schneider et al. 2007, p. 783). Assessing vulnerability is a complex endeavour since it cannot be easily quantified, and the physical and social processes that shape local conditions as well as the ability to cope are very dynamic (O’Brien et al. 2004). Although most of the assessments of vulnerability have been objective quantifiable ones, some subjective assessments of vulnerability have also been used in the literature. Objective assessments have mainly been at the country level and considered gross product and per capita income. This method, however, fails to capture vulnerability at the individual or household level. Other assessments have estimated the cost of damage caused by the natural hazard under study (Bankoff et al. 2004). Furthermore, some indices and metrics to assess vulnerability have also been developed (Benson 2004; Kamanou and Morduch 2004). Assets, income, and well-being/capabilities have also been used to assess vulnerability at the household level (Albert et al. 2007). Finally, quite a number of studies on objective assessments of vulnerability have mapped vulnerability (Haddad 2005; Brooks et al. 2005), including spatial mapping (Kasperson et al. 2005; Schroter et al. 2005).

Subjective assessments of vulnerability, on the other hand, are not so common in the literature (Baker 1991; Norris et al. 1999; Helweg-Larsen 1999; Botzen et a1. 2009; Peacock et al. 2005). Such assessments are important because people’s perceptions influence their risk behaviour, and therefore their vulnerability, which is useful for local policy formulation (Bankoff et al. 2004). This study uses a subjective assessment of vulnerability.

1.2 Conceptual framework

As shown in Fig. 1, the study conceptualises that after a heavy rainfall event, it is the geophysical characteristics (mean elevation, slope, proximity to the lagoon, proximity to the sea, and drain length) of a household, coupled with the socio-demographic characteristics (presence of females, presence of males, household size, wealth quintile, and locality) including structure materials (main wall material and main floor material) used in constructing the house, that predispose heads to perceive their households as vulnerable to experiencing floods.

Fig. 1
figure 1

Conceptual framework showing geophysical and socio-demographic characteristics of households and perceived vulnerability to flooding

Full size image

First, it is expected that heads of households in houses located in areas with lower elevation will have higher flood vulnerability perceptions. This is based on the assumption that water flows from higher to lower elevations. Second, heads of households in houses located on gentler slopes are expected to perceive their households as more vulnerable to floods. This is due to the fact that gentler slopes drain flood water slower than steeper slopes. Third, heads in houses with closer proximity to the lagoon are more likely to perceive their households as vulnerable to floods as a result of the higher risk of direct inundation by the lagoon during floods. Fourth, heads in houses closer to the sea would have higher perceptions of flood vulnerability because flood water continually moves towards the sea. Fifth, drains carry away flood water, and therefore, heads in houses located in enumeration areas without drains are expected to perceive their households as more vulnerable to floods.

Regarding socio-demographic characteristics, although females are usually considered vulnerable during disasters, some studies have highlighted the important role of women and their social networks in household disaster preparation and management. Hence, heads in households with higher proportions of females are expected to perceive their households as less vulnerable to floods. The number of persons living in a household could be related to the economic fortunes of the household. Thus, heads in households with more members are expected to be economically disadvantaged compared to those with fewer members and therefore have higher perceptions of flood vulnerability. Similarly, households in the poorest wealth quintile are expected to have higher perceptions of flood vulnerability. Furthermore, the locality in which a household is can determine perceptions of flood vulnerability. This may be due to a combination of geophysical and socio-demographic characteristics. It is hypothesised that household heads in Agbogbloshie will perceive their households as most vulnerable to floods compared to household heads in James Town and Ussher Town.

Finally, the effects of flooding on materials used to construct a structure include general displacement, collapse, and overturning (Hughes 1982). Flooding has different effects on different structure building materials. Walls made of inorganic material such as soil brick, soft brick, and stone rubble have low shear strength. They fail easily under impact and dynamic loads. When the wall material is made of organic material, damage occurs very easily, mostly because they are lightweight. The response of bricks to floods differs considerably, mainly due to how well they are constructed. Well fired bricks have been noticed to withstand floods for centuries, but those not well made suffer from softening, disintegration, and cracking. Flooding does not usually adversely affect concrete: mass, reinforced, or precast.

1.3 Study area

Accra the capital of Ghana had a total population of 1,848,614 in 2010 (GSS 2012). The study communities are Ga Mashie (James Town and Ussher Town), and Agbogbloshie (Fig. 2). James Town was colonised by the British and Ussher Town by the Dutch. Ga Mashie consists of the nuclei of Ga people, both around the beach and extending inland to other communities such as Adedenkpo, Ayalolo, and a highly heterogeneous migrant settlement, Old Fadama, adjoining this traditional enclave (Henry and Fayorsey 2002). Agbogbloshie, on the other hand, lies in the suburbs of Accra and has one of the largest markets in the Accra Metropolitan area. It is a densely populated, permanent community with some residents tracing their roots back to the early 1960s. The community is very close to the major drain in Accra as well as the Korle Lagoon. The major stream draining central Accra is the Odaw River which enters the Korle Lagoon and consequently the Gulf of Guinea. Most of the areas around the Odaw River are built up, with many of the streams being channelised (Masiyandima et al. 2003).

Fig. 2
figure 2

Map of Accra showing the study areas. Source: Centre for Remote Sensing and Geographical Information Systems, University of Ghana, Legon

Full size image

Accra has expanded spatially; however, this has not been accompanied with requisite management and planning. This exposes many of the dwellers of the city to flood disasters (Afeku 2005). Along with the spatial expansion, which Andjeilkovic (2001) described as the primary cause of urban flooding, there have been severe rainstorms preceded by long-lasting moderate rainfall which saturate the soil (Afeku 2005). The absence of adequate drainage and sanitation services has led to the widespread and severe nature of floods in Accra (Andjeilkovic 2001; Afeku 2005). For instance, surcharges, resulting from the clogging of drains and street inlets by silt and garbage, and inadequate street cleaning practices, have been identified as the major contributory factors to flooding in Accra (Kolsky and Butler 2002; Afeku 2005).

Data on rainfall in the study area from 1935 to 2005 shown in Fig. 3 indicate that there has been variability in the annual rainfall, but the linear trend line indicates a general reduction. This suggests that though rainfall has been generally reducing, it has continued to vary across the years.

Fig. 3
figure 3

Annual rainfall (in mm) for Accra—1935–2005. Source: Ghana Meteorological Agency

Full size image

Figure 4 shows that generally, the highest amount of monthly rainfall occurs in June and the lowest in December and January. The trend has not changed for the seven decades analysed. The peak rainfall season has always been in June and the second peak in October, giving a kind of bimodal rainfall pattern.

Fig. 4
figure 4

Mean monthly rainfall (in mm) for Accra by decades, 1935–2005. Source: Ghana Meteorological Agency

Full size image

1.4 Data and methods

Data for this study are from two sources. First, a digital map of the study area was obtained from the Survey and Mapping Division of the Lands Commission of Ghana. The digital map is used to calculate mean elevation, slope, proximity to the lagoon, sea, and drain length by area. Each housing structure is mapped using global positioning systems. The elevation of each house is calculated from the available spot heights and contours in closest proximity. Houses are classified into two, i.e. those with elevations of <500 m above sea level and those 500 m or higher above sea level. Slope is change in elevation divided by distance. Therefore, the distance between the highest and lowest spot height and/or contour that best represents the enumeration area is extracted from the map. The difference between the two elevations is then divided by the extracted distance. Houses are assigned slope figures per enumeration area as follows: <0.010°; 0.010°–0.013°; 0.014°–0.017°; and 0.018° or steeper.

The proximity of a house from the lagoon and the sea is determined by offsetting the boundaries of the high tide mark of the lagoon and the sea at 100-m intervals till 500 m. The crossover offset lines of the boundary lines for each interval are trimmed to achieve a closer to reality estimation. The 100 m offset at which a housing structure overlapped is deemed the proximity of the structure to the water body, i.e. lagoon or sea, and both the lagoon and sea are classified as <500 and 500 m or farther. Drain availability is analysed in terms of the length of the major drain in an enumeration area divided by the land area, and houses are assigned the various categories: 0.000 m (no drain), 0.001, 0.002, 0.003–0.010, 0.011–0.017, and 0.018 m and longer.

Second, the socio-demographic characteristics used in the analysis are derived from the EDULINK Round II Household Survey (collected data on poverty, livelihoods, health, and environmental factors) undertaken in 2011. It classifies the presence or absence of both males and females in households, and single member, double member, triple member, quadruple member, quintuple member, and sextuplet or more member households. The others are wealth quintile (poorest, poorer, middle, rich, and richest) and locality (Agbogbloshie, James Town, and Ussher Town). Wall material is classified into two, i.e. more permeable (mud, wood/timber, iron sheet, and cardboard sheet) and less permeable (brick, concrete, and block). Similarly, floor material is classified into two, namely more permeable (mud, dung, and sand) and less permeable (wood plank, polished wood, vinyl tile, and cement). Finally, the dependent variable, i.e. perception of flood vulnerability, is derived from responses from household heads to the question, “Do you perceive your household to be vulnerable to floods?” The responses are either in the affirmative or negative.

Two binary logistic regression models are used in the analysis because the dependent variable has a dichotomous response. The first model ascertains the association between geophysical characteristics of households on the one hand and perceived vulnerability to floods on the other. The second model controls for selected socio-demographic characteristics including materials used for house structure in the first model.

2 Results and discussion

2.1 Geophysical characteristics

Table 1 shows geophysical characteristics of the study communities. Agbogbloshie has the largest land area of 96,114 m and the longest drain (64 m). However, regarding availability of drain by land area, Agbogbloshie has the lowest mean figure of 0.001 m. In addition, Agbogbloshie has the lowest mean elevation (3.744 m), while Ussher Town has the lowest slope of 0.013 m. Regarding proximities to lagoon and sea, Agbogbloshie and Ussher Town are closest to the lagoon and sea, respectively, with figures of 440 and 388 m.

Table 1 Geophysical characteristics by enumeration area
Full size table

2.2 Univariate analysis

After presenting mean values of geophysical characteristics of enumeration areas in Table 1, Table 2 provides proportions of geophysical and socio-demographic characteristics including structure materials of houses and perceptions of flood vulnerability by households. Almost 48 % of houses are located below 10 m, one in three houses is located on a slope <0.010°, one in ten houses is <500 m to the lagoon, and half of the houses are <500 m to the sea. Thus, more houses in the study sample are located closer to the sea than to the lagoon. Table 2 further shows that almost eight out of ten houses are located in areas without any drains.

Table 2 Percentage distribution of geophysical characteristics and socio-demographic characteristics including structure materials of houses and perception of flood vulnerability
Full size table

Table 2 also presents results of socio-demographic characteristics including structure materials of houses. It shows that 63.6 % of households have females, 62.4 % have males, and a third of households are single member. Households in the sample were purposively selected to obtain equal proportions for the different categories of wealth quintiles. As a result, 20.0 % each of the households are in the poorest, poorer, middle, rich, and richest wealth quintile categories. Furthermore, 15.6, 29.2, and 55.2 % of households are in Agbogbloshie, James Town, and Ussher Town, respectively, and while 22.8 % of walls of houses are made up of more permeable materials, 7.6 % of floors of houses are made up of more permeable materials. Finally, while 20.3 % of household heads perceived that their households are vulnerable to floods, 79.7 % of heads do not perceive that their households are vulnerable to floods.

2.3 Bivariate analysis

This section of the paper considers the association between geophysical characteristics and socio-demographic characteristics including structure materials of houses, on the one hand, and perception of vulnerability to floods, on the other.

The result in Table 3 shows that at the bivariate level, all the five geophysical characteristics are significantly associated with flood vulnerability perceptions. This shows the important role geophysical characteristics play in flood vulnerability as stated by other studies (Dewan et al. 2007; Zheng et al. 2008; Karley 2009; Forkuo 2011; Fosu et al. 2012). First, as expected, almost double (27.2 vs 14.1 %) the proportion of household heads living in houses located at an elevation of below 10 m perceive that their households are vulnerable to floods, compared to heads in houses located at an elevation of 10 m or higher. Second, slope did not show any clear cut pattern.

Table 3 Bivariate analysis showing geophysical characteristics and socio-demographic characteristics including structure materials of houses and perception of flood vulnerability
Full size table

Third, while a higher proportion (56.9 %) of heads in households closer to the lagoon (<500 m) perceive their households as vulnerable to floods, in the case of proximity to the sea, a reverse scenario is shown. Thus, a lower proportion (15.4 %) of heads in households closer to the sea (<500 m) has flood vulnerability perceptions. This could be attributed to the fact that the lagoon, and not the sea, is perceived by community members as the source of most of the flooding events in the study communities. Fourth, about 18.0 % of household heads in enumeration areas with no drains have flood vulnerability perceptions.

With regard to socio-demographic characteristics including structure materials of houses, the presence of females and males in households and main wall and floor materials of houses are significantly associated with flood vulnerability perceptions. Thus, 17.4 and 25.6 % of heads in households with and without a female, respectively, perceive their households as vulnerable to floods. In addition, 23.1 and 15.8 % of heads in households with and without a male, respectively, perceive their households as vulnerable to floods. Furthermore, the more permeable the construction materials for wall and floor, the more heads perceived their households as vulnerable to floods.

The three other socio-demographic characteristics that are statistically not significant are household size, wealth quintile, and locality. Single-member (23.1 %) and double-member (23.1 %) household heads have the highest flood vulnerability perceptions. The analysis of wealth quintile in Table 3 shows that heads of households in the middle quintile (29.6 %) have the highest flood vulnerability perceptions. Furthermore, heads in households in Agbogbloshie have the highest flood vulnerability perceptions (41.3 %). This can be attributed to the geophysical characteristics of the localities presented in Table 1 which show that Agbogbloshie has the shortest drain length by area, the lowest mean elevation, and is closest to the lagoon which, according to household heads, is responsible for most of the flooding in the study area.

2.4 Multivariate analysis

The result of the first binary logistic regression model that ascertains the effects of geophysical characteristics and perceptions of flood vulnerability is shown in Table 4. Three of the variables, namely elevation, slope, and length of drain, are significantly associated with flood vulnerability perceptions. This further emphasises the significant role of geophysical characteristics in flood vulnerability. As expected, heads of households in houses located higher than 10 m are less likely to perceive their households as vulnerable to floods, compared to those located below 10 m. The analysis also shows that the steeper the gradient in slope of a house’s location, the less likely the perception of vulnerability to floods. In both scenarios of higher elevation and steeper slope, water run-off is quite rapid, which makes it impossible for water retention to subsequently cause floods. Regarding length of drain by area, household heads with drains in their communities are more likely to perceive their households as vulnerable to floods, compared to those with no drains. This is contrary to expectation and could be attributed to the fact that most of the drains in the study communities are choked with plastic and other waste materials (Codjoe et al. 2013).

Table 4 Binary logistic regression showing relationship between geophysical characteristics and socio-economic characteristics including structure materials of houses and perception of flood vulnerability
Full size table

In the second model, socio-demographic characteristics including structure materials of houses have been controlled for. Yet, three of the geophysical characteristics that are statistically significant in the first model, namely elevation, slope, and length of drain, maintained their statistical significance in the second model, further underscoring the importance of geophysical characteristics to flood vulnerability perceptions. Not only did the variables maintain their statistical significance, but also two out of three (elevation and slope) carried the expected signs. Thus, first, heads of households located in houses higher than 10 m and above are less likely to perceive their households as vulnerable to floods, compared to those located below 10 m. Second, the steeper the gradient in slope of a house’s location, the less likely the household head’s perception of vulnerability to floods. Third, household heads with drains in their communities are more likely to perceive their households as vulnerable to floods, compared to those without drains. This is contrary to expectation and some explanation as to how this is so has been provided above.

The result in the second model also shows that a fourth geophysical characteristic, proximity to the sea, was significantly associated with flood vulnerability perceptions. Thus, household heads whose houses are located farther away from the sea are less likely to perceive their households as vulnerable to floods, compared to those who are located closer to the sea. The conclusion that could be drawn at this point is that the introduction of socio-demographic characteristics does not alter the relationship between geophysical characteristics and perceptions of flood vulnerability and that it is the geophysical characteristics rather than socio-demographic characteristics that have strong associations with perceptions of flood vulnerability.

Two of the socio-demographic characteristics, i.e. wealth quintile and locality, are significantly associated with flood vulnerability perceptions. As far as wealth quintile is concerned, household heads in the rich and richest categories are less likely to perceive their households as vulnerable to floods, compared to household heads in the poorest quintile. Finally, household heads in James Town and Ussher Town have less flood vulnerability perceptions compared to household heads in Agbogbloshie. As already mentioned, this can be attributed to the geophysical attributes of Agbogbloshie, namely the shortest drain length by area, the lowest mean elevation, and the closest proximity to the lagoon.

3 Conclusions

Empirical analysis of the association between a combination of geophysical and socio-demographic characteristics, on the one hand, and perceptions of flood vulnerability, on the other, was undertaken in this study in Accra, the metropolitan capital of Ghana. This is important because empirical downscaling of GCM simulations by Hewitson and Crane (Christiansen et al. 2007) suggests wetter trends along the Gulf of Guinea, where Accra is located, and heavy rains are the most common cause of floods (Few 2003).

The study shows that even when a subjective assessment of vulnerability is undertaken, it is geophysical characteristics that have significant associations with perceptions of flood vulnerability. A central theme in this study is that geophysical characteristics influence perceived vulnerability to floods. It is people’s awareness and perceptions of these characteristics and their relevance which were expected to influence flood vulnerability perceptions. The question to ask is, “Are people aware of the significance of the geophysical environment around them?” Although people may not be aware of some aspects of their geophysical surroundings, the results indicate that they are aware of other aspects. For example, the fact that the presence of drains—often clogged and ineffective—actually increases perceived flood vulnerability does indicate that people’s interpretation of the geophysical characteristics is a key aspect of perceived flood vulnerability. People are also aware of the proximity of their houses to the sea as well as the elevation and slope of their housing structures, which influences their perceptions of the vulnerability of their households to floods. It must, however, be mentioned that the socio-demographic variables examined in the study are rather limited; thus, the influence of such variables in general could not have been properly assessed.