Introduction

1Since the International Decade for Natural Disaster Reduction -1990 to 1999- (UNDP, 2004), the issue of populations’ vulnerability has remained at the heart of the international community’s preoccupations (Kouassi et al., 2010; Lhomme, Laganier, Diab, Serre, 2013). Indeed, the recurrence of natural disasters and their impact have driven states to take action to help populations to adapt and reduce risks. Worldwide, “in 2015, natural disasters led to economic losses of 92 billion dollars, and average losses of disasters were estimated at over 300 billion dollars per year” (Hallegate, Vogt-Schilb, Rozenberg, 2016). According to CRED's EM-DAT database (CRED, 2020), flooding constitutes the primary risk of natural disasters in the world when one considers the number of events and individuals affected. To that end, many researchers throughout the world have taken on the study of natural disasters in order to understand the causes and to identify objectively the consequences and their inadequate management (Hostache, 2007; Karambiri, Tazen, Traore, Bologo/Traore, Coulibaly, 2015; Lhomme et al., 2013; Ntajal, Lamptey, Mahamadou, Nyarko, 2017; Pradhan, 2010; Van Westen, 2013). In most of the work done on risk disasters, the concept of vulnerability remains a major issue. These works generally show that vulnerability results from intrinsic characteristics of populations, their potential exposure to a hazard, their awareness linked to the exposure (Burton, Kates, White, 1993; Cutter, 1996), the capacity for adaptation which varies from one individual to another (Birkmann, 2007; Buckle, Marsh, Smale, 2001; Turner et al., 2003), the spatial occupancy (Cardona, 2004; Cutter, Boruff, Shirley, 2003) and the characteristics of the hazard (Wilhelmi, Morss, 2013). It comprises two dimensions: biophysical vulnerability related to hazard exposure and social vulnerability related to society (Cutter et al., 2003; Reghezza, 2006).

2Social vulnerability is a territorially rooted notion, central to risk analysis in the urban environment (Adger, 2006; Quenault, 2015). It denotes the incapacity of a society to anticipate a hazard, to cope with an emergency, to adapt its behaviour during a crisis and to reconstruct itself (Wisner, Blaikie, Cannon, Davis, 2003). It depends upon social factors such as demographic characteristics, discriminations, social inequality, imbalance of power (Becerra, 2012; Birkmann, Wisner, 2006) as well as institutional, economic, cultural, historical and political factors (Meschinet de Richemond, Reghezza, 2010). To these aspects, one can add health (Jonkman, Kelman, 2005; Ruin, 2007; Vinet, Boissier, Defossez, 2011), adaptive capacity (Lallau, Rousseau, 2009; Pelling, 2003), risk perception (Becerra et al., 2013; Bonnet, 2002; Kellens, Zaalberg, Neutens, Vanneuville, De Maeyer, 2011; Ruin, Lutoff, 2004), quality of life in the neighbourhoods and land tenure (Rufat, Tate, Burton, Maroof, 2015).

3Even if Africa is not the continent which has been the most affected by flooding globally, the impact and consequences of these disasters can prove to be devastating, putting it among the most vulnerable of continents. This is due to the considerable dependence of these countries on natural resources, to social and biophysical vulnerability and to a relatively limited adaptive capacity. These findings are all the more alarming in the countries of West Africa, which are already confronted with a generalized impoverishment and have, for two decades, been victims of flooding practically every year. This situation can be explained not only by extreme rainfall but by transformations in urban spaces, in their rhythm, form and governance (Hangnon, De Longueville, Ozer, 2015).

4Ouagadougou, the capital of Burkina Faso, is an example of a West African crossroads city whose economy has been badly affected by the consequences of natural disasters. It is an urban centre that has experienced an increase in the frequency of flooding, principally linked to rain runoff overloads and extreme storm events. The major floods of 2009, 2015, 2016 and 2018 caused considerable material damage and loss of human life (Pictures 1 and 2). They revealed certain risky collective behaviours in the post-disaster period, such as the resettlement of populations in flood-prone areas after the 2009 floods (Essone Nkoghe, 2012; Hangnon, Bonnet, Amalric, Nikiema, 2018). They also highlighted the city’s exposure and great vulnerability of its populations to flooding in particular.

Picture 1: The city centre of Ouagadougou flooded on 1^st September 2009

Agrandir Original (jpeg, 96k)

credit: Ahmed OUOBA/ AFP

Picture 2: Collapsed house in the undeveloped area of Ouagadougou during the 2015 floods

Agrandir Original (jpeg, 128k)

credit: http://news.aouaga.com

5Today, although flood risk reduction measures have been implemented after the major floods mentioned above, Ouagadougou is still subject to flooding. The populations also still live in areas declared flood-prone, unbuildable or likely to be flooded and are still strongly impacted by these phenomena. To understand this situation, which incidentally is not specific to the city of Ouagadougou, several authors have worked on the concept of vulnerability and the means to measure it at the local or global level (Adger, Brooks, Kelly, Bentham, Eriksen, 2004; Balica, Wright, van der Meulen, 2012; Cutter et al., 2003; Flanagan, Gregory, Hallisey, Heitgerd, Lewis, 2011; Karambiri et al., 2015; Tazen et al., 2019). Most of these authors agree that social factors have a strong influence on risk management and coping skills and have therefore addressed vulnerability from a social perspective in their studies, either fully or partially. For instance, the study by Karambiri et al. (2015) assessed the vulnerability of Ouagadougou by determining, among other things, the flood social vulnerability index (FVI_social). This index, inspired by the work of Balica et al. (2012), is based on the aggregation of social indicators from censuses or reports by adding or multiplying them in some way. Another social vulnerability index that is widely used in the literature is the SoVI index developed by Cutter (2003). This index is also based on the aggregation of variables derived from censuses but this aggregation process is performed using principal component analysis.

6All these studies have the common feature of seeking to measure the absolute vulnerability of a given territory by aggregating several variables using various techniques and relying essentially on data from censuses or reports. Overall, this type of approach using vulnerability indexes raises a number of problems. A major limitation noted by Rufat (2013) is that in these studies “vulnerability indicators are added together or multiplied in different ways by different authors without taking their diversity and interaction into account”. Moreover, these indexes strongly depend on the initial choice of variables (Tate, 2012). Also, the latter are subject to collinearity bias (Rufat, 2013). The result is an arbitrary, subjective and incomplete presentation of vulnerability levels with indexes that tend to oversimplify the complexity of the concept and erase its heterogeneity at the local level. Because of all these limitations and the validity problems noted by some authors (Rufat, Tate, Emrich, Antolini, 2019; Schmidtlein, Deutsch, Piegorsch, Cutter, 2008; Spielman et al., 2020), this approach to vulnerability remains problematic for operational use by managers.

7So, how can we assess the vulnerability of populations in this context in order to better adapt prevention measures? To answer this question, Rufat (2013) proposed shifting the focus from measuring absolute vulnerability, which would be too complex to understand, to relative vulnerability.

8Relative vulnerability is related to understanding the processes that make or break vulnerability in a given locality and is more operational than the absolute measure of vulnerability. In other words, analysing relative vulnerability would serve to answer certain operational questions. Which are the most vulnerable populations in a given locality, and why are they vulnerable? Are there aggravating factors related to where they live? Although these questions are central to flood management in the city of Ouagadougou, they have not been given much thought in the city, as the application of relative vulnerability analysis techniques has focused on other regions. It is to compensate for this shortcoming that our study was envisaged.

Presentation of the study zone

9The city of Ouagadougou, located in the heart of West Africa, was chosen as the study area for the implementation of the Rufat approach because of its high exposure to flooding due to its characteristics. Comprising 55 sectors at an average altitude of 310 m, the urban municipality of Ouagadougou extends over 518 km² and has a flat, gently sloping terrain (Picture 3). The morphology of the town is monotonous with gentle slopes of 0.6 to 1% (Kêdowidé, Sedogo, Cissé, 2010). These features lead either to rainwater stagnation or ingress according to the nature of the ground.

Picture 3: Ouagadougou seen from the top of a building: a city with flat terrain

Agrandir Original (png, 587k)

Credit: DA M. Laure Carolle

10The city of Ouagadougou is drained by four principal canals resulting from development of topographical depressions (Figure 1). The hydrographic network, moreover, is made up of watercourses operating intermittently, and of four intra-urban dams. Three of these dams are inter-connected and numbered 1 to 3 from upstream to downstream. They discharge their water into the River Massili crossing the gazetted Bagr-Wéogo forest, the natural outlet of the city’s rainwater.

Figure 1: Presentation of the study area

Agrandir Original (jpeg, 396k)

11Several works based on the analysis of press-reported stories have reported an increase in the frequency of floods over the past 30 years in Burkina Faso in general (Taylor, Traoré/Bologo, Tazen, 2019) and in the capital city of Ouagadougou in particular (Hangnon et al., 2018). Tazen (2019) has noted, for example, an average of five floods per year in the city since the 2000s, with significant impacts on people and property.

12The problem of flooding is therefore a major issue for the city of Ouagadougou, particularly for the most vulnerable populations. It is important for decision-makers, beyond the aspects related to the hazard, to understand the processes of social vulnerability in order to provide solutions adapted to the context of Ouagadougou. This is the goal of our article.

Material and methods

14Census or report data are usually incomplete or on a scale that is inappropriate for urban studies, especially in West Africa. Moreover, when they are available, these data are generally from different sources that may have collected and published them at different times, which may compromise the selection of variables. Indeed, these variables may not correspond to social indicators from comparable periods.

16The questions were asked of a single representative of the household, namely the head of the family or another member of the household of over 18 years of age living for at least six months in the household. Responses from women were encouraged. During the survey, nine respondents did not meet the selection criteria or did not agree to the survey and were therefore excluded from the analyses. These individuals were not included in the 2137 respondents who all met the selection criteria.

18The number of households to be surveyed in each zone was set to best approximate the ratio of area in each zone to group membership (Table 1), to be spatially representative.

Table 1: Identification of the survey areas

Figure 2: Presentation of the study zones according to the four zones of the survey

Agrandir Original (jpeg, 372k)

19Analysing social vulnerability requires taking into account its multidimensional basis. With this in mind, the survey questionnaire was developed taking into account the main aspects that can influence social vulnerability. These were about:

• -inhabitants’ socioeconomic characteristics (household’s asset for wealth level, education, etc.)

• -knowledge about the flood phenomena (causes, awareness of exposure to risk, etc.)

• -experience of floods, damage and ability to recover

• -knowledge of prevention and safety regulations

• -potential behaviours when a flood occurs

20In total, 84 variables, mainly qualitative, considered as drivers for social vulnerability, were gathered to establish socio-spatial vulnerability profiles. The data processing consisted of a statistical analysis in several stages:

Calculation of a wealth index: principal component analysis (PCA)

21The wealth index is a composite measure of a household’s standard. It was used to understand the socio-economic status of households through a number of variables related to the goods they possess (number of television sets, refrigerators etc.). This is an interesting driver that has a definite impact on the level of social vulnerability of a territory (Fischer, Chhatre, 2016). The method proposed by the World Food Programme (2017) to establish this index consists, with SPSS software, in assigning a standardized weight or score to each selected quantitative variable through a principal component analysis (PCA). Individuals are then classified according to their score. The sample is finally divided into population quintiles. Each quintile corresponds with an economic well-being level from 1 (for the poorest) to 5 (for the richest) (INSD, 2015). We thus obtained a qualitative variable with 5 modalities which can be used thereafter to determine the profiles of vulnerability.

22They are calculated to establish the correlations between variables to determine the significant variables, with the least possible redundancy, in order to establish the profiles. The analysis of these correlations and their strength made it possible to eliminate the super-linked variables in order to make the factor analysis more robust (Habib Bawa, 2018; Ritschard, Zighed, Nicoloyannis, 2001; Yabi, Sossou, Akindele, Balogoun, Ogouwale, 2019). According to Rufat et al. (2019), “this process limited collinearity, prevented implicit weighting, strengthened statistical power and preserved a balance between the different dimensions of vulnerability”. This stage led to the adoption of 13 variables for the remainder of analyses (Table 2).

Table 2: Variables selected after correlation analysis

*Idk: I don’t know

Looking for profile characteristics: Multiple correspondence analysis (MCA)

23Since all the selected variables were qualitative, we opted for a multiple correspondence analysis as factor analysis before the clustering method. MCA is a descriptive statistical technique which also aims to summarize and visualize a table of data containing more than two qualitative variables (Renisio, Sinthon, 2014). The objective is to identify groups of people with a similar profile and to show the associations between variables and modalities. To take into account the sensitivity of the MCA to low numbers (less than 5%), we grouped the rare modalities as recommended by Husson et al. (2010). The factor coordinates obtained were thus used as quantitative variables for the clustering method.

24Identification of clusters: Hierarchical Agglomerative Clustering (HAC)

25It was conducted using the R software employing the FactoMineR package in accordance with Ward’s method and Euclidean distance matrix. It consists in dividing the populations into different clusters, based on the MCA results. The advantage of this technique is that the largest possible group of similar individuals can be grouped within the same cluster (cross-cluster homogeneity) and to define the most dissimilar cluster (cross-cluster heterogeneity) (Rérolle, Faisant, Telmon, Saint-Martin, 2015). Along with the inspection of the dendrogram and the analysing of statistics of inertia gain, we also determined the number of clusters retained using the Ratkowsky and Dunn indices. This number is the one that maximizes each of the indices. These indices are ranked by several studies as high performers for determining optimal number of clusters (Charrad, Ghazzali, Boiteau, Niknafs, 2014; Tian, Lemos, 2018). According to these metrics, eight clusters seem to best capture the clustering structure in the data, since they give the highest desired values (Table 3).

Table 3: Quantitative indices to determine optimal number of clusters

Results

Representativeness of the sample

26The random sampling approach resulted in an overall representativeness of the urban population for different socio-demographic variables, as shown in Table 4. Indeed, the verification of the representativeness of our sample is a prerequisite for validating the results, and therefore for interpreting them. Thus, we observe that the ratios of socio-economic variables such as gender, educational level and occupation are close to the ratios found in censuses conducted by the National Institute of Statistics and Demography of Burkina Faso. We can therefore conclude that our sample is somewhat socially representative.

Table 4: comparison of socio-demographic characteristics of respondents with census data

Socio-economic profile of populations and spatial distribution of individuals

27The socio-economic profiles determined by the PCA are presented in figure 3. These profiles are representative of Ouagadougou and show the spatial distribution of the level of economic well-being in the city.

Figure 3: Distribution of the population surveyed according to the wealth index

Agrandir Original (jpeg, 544k)

28Determining the wealth index of the people surveyed confirms that the majority of less well-off populations live in the outlying areas (Figure 3). Their presence can be explained by the level of land speculation in Ouagadougou and the difficulty for households to have access to parcels and dwelling places in the city-centre because of high prices.

29Two types of proprietors rub shoulders in the informal settlements (Robineau, 2014). On the one hand, there are families in precarious financial situations settling in an informal zone (which has not been divided into lots) because it is easier to acquire a parcel there for a lower price and these areas offer the advantage of access to the city’s amenities (e.g. work etc.). These families’ houses are generally built using precarious materials (Picture 4). On the other hand, middle-class families generally with lower resources than families in the city centre, however, often acquire an “alibi parcel” in the hope that one day they can benefit from a lot in a subdivided zone. In these cases, houses are built with better amenities (Picture 5).

30Many “poor” households, therefore, settle on the outskirts of town while hoping to be able to benefit from a parcel when subdivisions take place (Boyer, Delaunay, 2009). They understand the interest of moving from an informal settlement to a property in a subdivided zone, as the value of the parcel rises considerably during the subdivision, according to the neighbourhood, multiplying by two if not by three the price per square metre. Due to land-use speculation, it is not rare to see households sell parcels allocated to them on completion of the housing development, moving from one informal home to another in the hope of earning some money (Delaunay, Boyer, 2017).

Picture 4: House built of precarious materials in the informal settlement area of Somgandé.

Agrandir Original (jpeg, 64k)

Picture 5 : House built in the informal settlement of Zagtouli.

Agrandir Original (jpeg, 212k)

31In addition, the economic level of households, related to the locality of residence of households, has an impact on the damage they are likely to suffer in the event of a flood, but also on their ability to cope and their attitudes towards risk. The constructions using precarious materials due to low resources in the non-settled areas and the absence of urban planning make populations more vulnerable to the risk. Picture 6 highlights the disorganization in the informal sectors with narrow lanes that do not respect any norms, which makes access by vehicle almost impossible and complicates evacuation in the event of a disaster.

Picture 6: Spatial disorganization in undeveloped areas

Agrandir Original (jpeg, 268k)

33Thus, the economic well-being of households has an impact on the spatial occupation of the city but also, as will be shown in more detail, on its social vulnerability.

Underlying factors of vulnerability

34The multiple component factor analysis (MCA) resulted in the extraction by using the Kaizer criterion of 10 dimensions, which explained a total of 54.93% of the variance in the data set. The first and the second dimension explained 8.77% and 6.77% variability of the data, respectively. The details of eigenvalue and percentage variance explained by 10 dimensions using MCA output are provided in the supplementary information (Appendix B).

35MCA was used in our study to understand the distribution of individuals and variables (Appendix C). Figure 4 shows the weight of different variables and categories on the first factorial plane. The first dimension is differentiated on the basis of categories of wealth index (Wealth_index= 0.48) and level of education (education= 0.42). It seemed to explain the social level of residents. Thus, this first axis produced by the analysis structures the population of the survey and distinguishes:

• poor individuals (low wealth_index =<3), who have never been to school and who do not know if their zone is at risk of flooding or not,

• and individuals who do not present these characteristics;

36The second dimension opposes two aspects:

• the modalities at the bottom of the graph indicate a doubt about awareness of the instructions (on the right) and the precautions (on the left) and on perceptions of risk in the residential zone;

• and the modalities at the top of the graph, which correspond to the representation of those who have a clear position as to their knowledge of the guidelines and the risk in the residential zone.

37This dimension is discriminated on the basis of representations (percep_zone=0.18) and knowledge of precautions (warnings=0.33) and civil protection instructions (cp_instruc= 0.31).

Figure 4: Profile of the people surveyed according to their responses on axes 1 and 2 of the MCA

Agrandir Original (jpeg, 52k)

38The factor score generated for the 10 dimensions was used as input data for cluster analysis by Hierarchical Agglomerative Clustering.

Profiles of the most vulnerable populations in the city

39The HAC undertaken following the MCA makes it possible to highlight eight clusters with a certain diversity within each group (Figure 5).

Figure 5: Dendrogram of individuals resulting from Hierarchical Agglomerative Clustering of Ouagadougou’s inhabitants

Agrandir Original (jpeg, 52k)

40A scale of vulnerability from one to eight, based on the interpretation of these clusters, is constructed, with level 1 the lowest level of vulnerability and 8 the highest. This scale depends on an interpretation of statistical results obtained. It is therefore at least partially subjective, despite all the efforts made. Thus, the scales revealed by this method are not comparable from one study to another, even if the approach is reproducible.

41Figure 6 shows the distribution of the eight levels of vulnerability and summarizes the main factors that differentiate all these clusters of residents on the first factor plane. It shows that the groups are organized according to their social level as well as knowledge and representations. Details of each profile are mentioned in the different boxes.

• 9 Figure 6 is inspired by the results of the factor map presented in appendix E and obtained by hiera (...)

Figure 69: Distribution of the surveyed populations into eight clusters on the 1st factor plane

Agrandir Original (jpeg, 408k)

42The specific characteristics of each cluster are analysed in the sub-sections below.

43Type 1 - vulnerability level 1. The first group (n=18.20%) with the lowest level of vulnerability is the affluent households (wealth_index = 5), whose representatives have mostly a high level of education and a good knowledge of flood precautionary measures. They are convinced that their residential area is not subject to flooding or prohibited for construction probably because they have very rarely been flooded.

44Type 2 - vulnerability level 2. The second group (n=6.46%) concerns mostly middle class households (wealth_index= 3). Their household-representatives are aware of the precautions to take, do not have risk behaviour but are ignorant of the causes of the floods.

46Type 4 – vulnerability level 4. The fourth group (n=17.69%) concerns household-representatives who claim to be aware of the precautions to take but do not know the civil protection guidelines. They have risk behaviour (wanting to return home during a flood) but they know about the causes of flooding. They think furthermore that the state is responsible for floods.

47Type 5 – vulnerability level 5. The fifth group (12.72%) concerns households whose representatives are mainly illiterate, and have doubts about their knowledge (precaution, perception of the area, civil protection guidelines). They do not have risk behaviour (going to see the flood) but do not think about clearing gutters to allow water to run off.

48Type 6 - vulnerability level 6. The sixth group (6.27%) concerns poor households with household-representatives possessing a low level of education. They are aware of the precautions to take but not the civil protection instructions nor the cause of floods. They do not have risk behaviour.

49Type 7 - vulnerability level 7. The seventh group (10.81%) concerns the household-representatives with a low level of education who do not know the civil protection guidelines nor the precautions to take and have risk behaviour. They do not want to be warned in the event of a flood.

50Type 8 - vulnerability level 8. The eighth group (17.74%) concerns household-representatives with a low level of education, who have several risk behaviours (going to see the flood, not staying where the flood took them by surprise) despite some knowledge of the causes of flooding. They have no knowledge, moreover, of the civil protection guidelines.

Spatial distribution of vulnerability profiles

51Figure 7 shows the spatial location of respondents according to their vulnerability profile. This enables identification of settlements of the less vulnerable populations in the central zone of Ouagadougou, in particular the neighbourhoods of Koulouba, Larlé, Ouidi, Gounghin, Zogona-Wemtenga and Dassasgho. A wealth index and a high level of education as well as knowledge of the precautions to take contribute to a low vulnerability in this part of the city (Appendix D).

Figure 7: Vulnerability profile of Ouagadougou populations

Agrandir Original (png, 872k)

52The sectors near the intra-urban dams (Dapoya, Tanghin, Somgandé) have themselves a low socio-spatial vulnerability index. This can be explained by the victim relocation measures that were undertaken by the municipality (Hangnon et al. 2018). As well, the effectiveness of the awareness campaign undertaken by the civil protection services in these neighbourhoods, and the acquisition of a certain risk culture, could also have contributed to improving the profile of the populations in this zone.

53However, the most vulnerable zone is located in the outskirts of Ouagadougou, principally in the western zone in the neighbourhoods of Rimkieta, Bassinko, Marcoussi, Nonghin and Bissighin. A low level of education, risk behaviour as well as a lack of knowledge of the safety guidelines characterize the individuals in this group (Appendix C). Furthermore, the zones of Rimkieta-west and Balkuy are zones where no flood-linked regulations are applied, despite strong social vulnerability. These sectors, in fact, have been seriously affected by flooding in the last 20 years (Arcens Somé, 2012). These are zones situated near the shallows with a great biophysical vulnerability. A certain discrepancy exists, therefore, between the effective vulnerability and the flood regulations in these zones because the topography and social aspects are not criteria which are taken into account when delineating the zones at risk.

54Vulnerability of the populations in the Rimkieta zone is all the greater as it constitutes, together with the neighbourhoods of Marcoussi and Zongo, places with the lowest wealth index (see Figure 3). Their inhabitants will consequently find it much harder to recover from the impact of a flood since the means at their disposal are not substantial enough.

55What is more, Bassenko and Yagma constitute a new subdivided zone dating from the 2000s (Boyer, Delaunay, 2009) and displaced person sites from the 2009 flood (Essone Nkoghe, 2012). These displaced persons were not necessarily made aware of the precautions and behaviour to adopt, hence their great vulnerability.

Discussion

Natural flood-linked risks: the benefits of creating socio-spatial vulnerability profiles based on survey data

57However, all of these indexes cannot properly assess vulnerability because information is lost when attempting to synthesize it, despite its multidimensional nature, into a single index. Thus, they have been highly criticized in the scientific literature (Rakotoarisoa et al., 2018). Our typology approach, inspired by the work of Rufat (2013; 2019), which groups the population into several levels of vulnerability based on several specific factors, is therefore very appropriate.

58Many of the aforementioned works on vulnerability assessment using indexes actually focus on the exposure of a population, an asset or a locality to risk. They therefore focus on the quantifiable characteristics of societies in order to analyse the potential of damage. However, social representations are essential aspects to take into consideration in order to understand how a risk is likely to have impacts on the territory (D’Ercole, Metzger, 2009). Indeed, the extent of losses, although linked to quantifiable socio-economic characteristics, will not be the same depending on the attitude of the populations affected by a risk. Indeed, social vulnerability cannot be reduced to a simple analysis of exposure to risk.

59Vulnerability assessment therefore goes beyond exposure issues by analysing the occupancy of unsuitable buildings (Jonkman, Kelman, 2005; Vinet et al., 2011) and more generally land use, population behaviour before, during and after a flood (Ruin, 2007; Wilson, 2006), socio-economic conditions (financial and psychological capacities) as well as non-compliance with safety instructions. It is therefore necessary to determine different profiles that take into consideration the specificities of each area, taking into account both quantifiable socio-economic characteristics and psychosocial aspects, which incidentally remain undetectable by census data.

60These census data are often either outdated or at scales that do not allow for local vulnerability analysis. For instance, in the case of Burkina Faso, the reports presenting the results of the 2020 general population census only presented data at a scale greater than or equal to the municipality. Analysing vulnerability at lower levels (e.g. by neighbourhood) is almost impossible with this type of data. Moreover, they tend to ignore the particularities and heterogeneity in each municipality. Our approach based on survey data is therefore more interesting.

61Although there are limits to the profiles that we are putting in place, in particular in terms of replicability because of the costs of the surveys, it nevertheless makes it possible to mitigate certain deficiencies of indicators based only on quantifiable aspects of census data (Böhringer, Jochem, 2007; Jones, Andrey, 2007; Tate, 2012). For example, the selection of variables is not limited to available census data, but was thought out prior to the survey to take into account the holistic nature of vulnerability.

An original approach to determine vulnerability profiles and their underlying drivers

62Our study on vulnerability profiles in Ouagadougou is essentially based on a sociological approach. The study of social vulnerability profiles is an alternative approach to indices that quantifies the vulnerability of a territory by aggregating several variables into a single synthetic index. This method combines factor analysis and classification to determine spatially coherent vulnerability profiles. In our study, we opted for an hierarchical agglomerative clustering, which remains one of the most widely used methods in the literature. This clustering approach has been used to highlight population typologies in several studies with or without factor analysis as an input (Chang, Yip, Conger, Oulahen, Marteleira, 2018; Fischer, Chhatre, 2016; Rufat, 2013; Shukla, Agarwal, Gornott, Sachdeva, Joshi, 2019; Wood, Jones, Spielman, Schmidtlein, 2015).

63The value of this approach is that it indicates the level of vulnerability of each profile but also explains the reasons for that vulnerability. At a spatial scale, the method is an opportunity to explain why some local areas are more vulnerable than others and thus to adopt targeted measures to improve the situation. Spatially representing vulnerability profiles therefore has policy implications by highlighting priority areas for assistance and coping capacity. As our results show, many variables are likely to have an impact on social vulnerability. The MCA carried out before the clustering is a good way to make a sparing selection of these variables. Combined with classification methods, it also offers the advantage of determining profiles on the basis of qualitative data.

64However, some simplifications may threaten the validity of the method. The analysis, for example, is based on the relatively subjective choice of the number of dimensions to be retained in the MCA, which can have a significant effect on the final result. This subjectivity is linked to the existence of various criteria (elbow rule, Kaizer criterion, analysis of variance, etc.) for the choice of dimensions.

65Although the employee classification method provides interesting results, it seems appropriate to consider a consolidation of the typology highlighted by the HAC. Indeed, many clustering methods have emerged for several years now (dynamic clouds method, fuzzy method, self-organizing map…). Some of them, such as the K-means method, have already been used in studies successfully and could be considered in our case to reinforce the coherence of clusters.

Urban policies, risk behaviours and social status: three key factors of social-spatial vulnerability

66Individuals showing a high level of social vulnerability can be found in particular on the urban fringes on the outskirts, essentially in non-subdivided housing areas.

67The most vulnerable populations, therefore, shown by our analysis, are the populations which have a tendency to adopt risk behaviours in going to see the flood or in not staying where they are when a flood occurs. Observable risk-taking behaviour could be justified by minimizing the impacts of the disaster and the belief in the effectiveness of coping strategies. These behaviours can largely be understood as “the thrill of the sight of a flood” (Wilson, 2006, p. 59), the desire to offer assistance in the event of people in difficulty or of not wanting to leave their goods unattended. Wilson (2006) talks of a “normality bias” for the loss of goods is often interpreted by the victims in terms of known consequences whereas the risk of death from a flood is unknown, difficult to predict and often minimized. Risk behaviour is seen in the same vein, therefore, as strategies implemented by populations to “cope” with risks. Colbeau (2002) shows that these strategies take account of cognitive and social aspects in conjunction with the perception of risk, of relativizing the situation, the desire to regain a certain control and the level of stress induced by danger.

69To reduce social vulnerability in Ouagadougou, policies to fight against poverty and flooding should be implemented as well as awareness systems which can be effectively disseminated to populations. Awareness has already been initiated by the state using the print media to disseminate the proper conduct to have in the event of flooding, but the question arises as to the effectiveness and adequacy of this written mode of raising awareness in a country where the rate of illiteracy is close to 75% (INSD, 2015).

Conclusion

70Preparing for, coping with, protecting against, and recovering from floods depends as much on an individual's intrinsic characteristics (socio-demographics, income, knowledge etc.) as their ability to correctly assess hazards, adopt appropriate behaviour, and cope with risk. In our study, the large amounts of data collected and the statistical techniques used allowed for a comprehensive analysis without preconceived ideas about the knowledge and spatial distribution of social vulnerability levels in Ouagadougou.

71Our study provides managers with a framework for understanding and assessing social vulnerability. However, although the implemented classification method is highly interesting, it can be improved, mainly through consolidation techniques for identified clusters.

72In view of the results obtained, it appears that for a more effective management of flood risks, the city of Ouagadougou should focus on certain measures. These include improving the living conditions of the population (fighting poverty and illiteracy) and raising their awareness to promote appropriate behaviour. Furthermore, improvement in housing conditions by means of better spatial organization of dwellings is an important line to define in urban planning documentation. Developing social housing in the city might reduce the pressure on informal settlements and limit land-use speculation.

73In our study, we have discussed the constraints of using census data for classification. However, the costs of implementing the urban survey we used (about 10 000 euros just for the survey) can be a significant barrier to identifying and characterizing vulnerability profiles. Reflection is therefore necessary to facilitate this type of evaluation on a regular basis and in several localities.

74If alternatives are developed to facilitate access to data, particularly qualitative data, the social vulnerability profiles method could be considered as a multi-scale technique for assessing and understanding vulnerability at both local and sub-regional levels.