An integrated socio-environmental framework for mapping hazard-specific vulnerability and exposure in urban areas

ABSTRACT Hazards act upon vulnerability and exposure to create disaster risk. Despite the growth of disaster risk assessments, the number of approaches that develop vulnerability and exposure studies is still small when compared with hazard modelling. In fact, limited studies have considered the relationship between vulnerability and exposure variables and how this can change future management actions on a local scale. This paper addresses this gap by proposing an integrated framework with a combination of social and environmental sciences to map hazard-specific vulnerability and exposure in urban areas. Subjective (e.g. Participatory Approach) and objective methods (e.g. Shannon Entropy and Fuzzy Theory) were integrated into a pixel-by-pixel framework for enhancing the flooding management in Campina Grande, Brazil. The results express the spatial distribution of flood vulnerability and exposure and assess key issues for flood management in different vulnerability categories. Challenges for the integration of socio-environmental approaches in water resources studies are discussed.


Introduction
Through the rise of research in social and environmental sciences, there is a growing search for enhanced frameworks to mitigate disasters risk (Kunapo et al. 2018) and to increase resilience (Ciullo et al. 2017). Disasters are defined mainly in relation to its impacts (Kelman 2020); however, those impacts can vary drastically depending on the local context (Frigerio et al. 2016).
In 2007, the Intergovernmental Panel on Climate Change (IPCC) used a definition in which the impact of a disaster is a relation between exposure, sensitivity and adaptive capacity (IPCC 2007). In this report, the impact is defined as the 'vulnerability' of the system, 'sensitivity' is the effect of variations on the system, 'adaptive capacity' as the ability of the system to adjust to climate-related stimuli, and 'exposure' is expressed with climatic variations (IPCC 2007). In other words, reducing vulnerability was limited to addressing the impacts of the hazard, characterised as the interaction between exposure, sensitivity, and adaptive capacity (Kc, Shepherd, and Gaither 2015).
There is a significant interest in the concept of vulnerability in the literature. For Cutter, Boruff, and Shirley (2003), the concept of place vulnerabilities integrates biophysical and social vulnerabilities (i.e. social inequalities). According to Pescaroli and Alexander (2019) and Ghajari et al. (2017) vulnerability can be divided into multisectoral categories, namely physical, social, economic, environmental, psychological, structural, and institutional. In 2012, the IPCC replaced the vulnerability definition with the risk concept, as a function of hazard, exposure and vulnerability (IPCC 2012b). Vulnerability is described as the attributes of a system in danger of a hazard and the exposure as the location of elements that may be impacted by the hazard (Sharma and Ravindranath 2019). In practice, this indicates those specific characteristics were already in place before the hazard occurrence. In other words, the new IPCC concept shows that the devastating effects of a disaster depend on the local vulnerability of an exposed society.
However, the overload and 'similarity' between vulnerability assessment approaches are seen as barriers for application in a system (Sharma and Ravindranath 2019). Even though there is a need to consider the vulnerability as one element of the disaster risk, which makes possible to reduce the vulnerability before and after the hazard's occurrence (IPCC 2014) and as a strategy to increase resilience (Golz, Schinke, and Naumann 2014), many applications still consider vulnerability only as the impacts of a disaster (Yang et al. 2018;Weis et al. 2016;Kc, Shepherd, and Gaither 2015) which can confuse the policymakers and reduce the applicability in real cases.

Challenges of vulnerability and exposure assessments
In recent years, there has been a search for identifying ways for a better representation of hazard, vulnerability, and exposure, along with spatial science (e.g. GIS: Geographic Information Systems). Recent studies include terms as 'integrated' (Weis et al. 2016), 'hybrid' (Roodposhti et al. 2016), 'multicriteria decision analysis' (MCDA) and 'system-thinking' approaches (Gomez Martin et al. 2020). However, most frameworks are not applicable to different areas and different hazards. This is due to geographical differences, human interactions and lack of data (Robinson et al. 2019), governance arrangements (Driessen et al. 2018), the involvement of stakeholders and dynamism of cities (Ciullo et al. 2017). For Cutter et al. (2008), since losses can vary geographically, over time, and among different social groups, the vulnerability also varies over time and space, which provide barriers for the assessment in different areas (Pescaroli and Alexander 2019).
Despite the growth of mapping approaches, key uncertainties remain as challenges. First, the discussion of an appropriate method for indicators selection is still seen as a barrier (Malczewski and Rinner 2015). The choice of indicators and the quality of available data requires a deep understanding of the complex system (Frigerio et al. 2016). For Boroushaki (2017), another challenge is the assignment of criteria weights. Due to the complexity of systems, all the criteria do not have equal influence in a disaster (Perera et al. 2019). For vulnerability and exposure mapping assessments, many studies consider equal weighting (Hazarika et al. 2018) or either subjective or objective (Birgani and Yazdandoost 2018) methods for weights calculation.
In the last few years, the subjective method has been gaining importance in mapping approaches. It can be practised as a way to engage different stakeholders in the decision-making process (Assumpção et al. 2018). Also, collaboration strategies also enables stakeholders to select indicators (Song and Chung 2017) with consensus agreement (de Brito, Evers, and Höllermann 2017). However, some authors argue that the decision-maker may be unable to quantify weights preferences (Roodposhti et al. 2016), which can overestimate or underestimate the impacts. For those situations, other methods, such as the entropy weighting (Boroushaki 2017), artificial neural network (Kia et al. 2011) and deterministic analysis, can be used. To deal with the inherent uncertainty, the fuzzy theory (Kanani-Sadat et al. 2019) is widely used as a value scaling procedure (e.g. standardisation) sensitive to the spatial and temporal extent of the data. In summary, both entropy and fuzzy theory handle the associated 'vagueness' of data values using a statistical variation and represent weights and scale according to the information in the dataset (Hong et al. 2018).
In this regard, this paper proposes a novel framework, here termed 'integrated framework', to obtain vulnerability and exposure mappings of urban areas in the context of flooding. The framework was built upon the paradigm change definition of IPCC (2007IPCC ( , 2014, in which the vulnerability can be assessed before, during and after the hazard, not as the impact (risk definition) but as a range of attributes that can contribute to the vulnerability of places (Cutter, Boruff, and Shirley 2003). We suggest that vulnerability and exposure indicators must represent hazardspecific attributes (Sharma and Ravindranath 2019), expressing the strengths and weaknesses on a temporal and spatial scale but with both objective and subjective methods for assessment. Specifically, our paper aims to answer two specific questions: (i) How can social and environmental tools be integrated towards vulnerability and exposure mapping assessments? (ii) How can the relationship between vulnerability and exposure be tackled on a spatial scale?
An integrated framework was developed by combining tools for decision analysis in environmental science (e.g. Shannon Entropy and Fuzzy Theory) and social science (e.g. Participatory Approach) to map flood vulnerability and exposure. Flooding is considered the most frequent among natural disasters, driven mainly by climate change and rapid urbanisation inducing changes in watershed hydrology (Hammond et al. 2018;Kunapo et al. 2018). In this paper, the final flooding risk will not be obtained yet, since the paper's focus is to find reliable vulnerability and exposure assessments. This paper begins by presenting the study case, the socioenvironmental conceptualisation, and the integrated framework. After that, results express each disaster variable and the validation with a historical and participatory approach. Discussions highlight key aspects generating the flood vulnerability with interactions between social, institutional, and structural vulnerabilities. Finally, limitations and next steps of the socio-environmental framework are presented along with the conclusions of mapping flood vulnerability and exposure.

Study case
The study case for this research is Campina Grande -Paraíba, Brazil. According to the Brazilian Institute of Geography and Statistics (IBGE), the city's population was estimated as 409,731 inhabitants in 2019. The city is part of the Northeast region of Brazil, known as 'semiarid region' (Figure 1(a)), with long water scarcity periods (ANA 2018). Although the city has a dry climate, it also registers flooding cases. Campina Grande is currently monitored by the National Centre for Monitoring and Alert of Natural Disasters of Brazil (CEMADEN) and the Geological Survey of Brazil (CPRM).
Even though flooding episodes are seen quite frequently, the city does not clearly define flood-prone areas. Data from CPRM specify 10 'risk areas' across the city, which refers not only to flooding but also to landslides and other disasters (Figure 1(b)). However, flooding cases are seen spatially dispersed in different areas of the city (Figure 1(b)), not only in the 'risk areas', which suggests there is a need for developing more accurate information for effective management (Alves et al. 2018). The middle-sized city (IBGE 2018) lacks in having sufficient flooding preparedness strategies for the population (Alves et al. 2020; and has a weak integration of urban planning and water resources management (de Araújo Grangeiro, Ribeiro, and de Miranda 2019).

The integrated socio-environmental approach
The methodology was constructed with basis in the disaster risk definition as the relationship between vulnerability, exposure and hazard (IPCC 2012b) ( Figure 2). The vulnerability is considered as the 'manifestation in a series of categories that do not develop independently but interact on many different time and space scales' (Pescaroli and Alexander 2019). Vulnerability is expressed as a function of several criteria, which has weaknesses (e.g. sensitivity) and strengths (e.g. capacity) that influence the conditions and the abilities of a society respond to harm in both temporal and spatial scales (UNDRR 2019). Figure 2 shows that vulnerability is determined by attributes that affect the consequences of a hazard. In this study, the sensitivity represents the weaknesses that can worsen the impacts of the disaster in the analysis. Capacity refers to the ability of societies and communities to prepare for and respond to current and future climate impacts (IPCC 2014). The system is also characterised by the elements located in hazard-prone areas, termed as 'exposed elements'. According to IPCC (2012a), exposure refers to the  presence of a vulnerable system at a location that could be adversely affected and can be represented by people, livelihoods, and assets.
In this context, we argue that both system attributes and exposed elements are directly related to a specific event; hence, we call them 'hazard-specific components' (Sharma and Ravindranath 2019). Due to the mixed temporal and spatial scales of hazards of different nature (i.e. drought, floods, landslides), several conditions can create vulnerability and exposure (Frigerio et al. 2016). In our approach, the vulnerability and exposure show an anticipatory state, or 'pre-existing state', concerning the hazard and will produce impacts, which can be increased or decreased by strategies, including coping capacity, risk perception, adaptation, and mitigation measures ( Figure 2). Therefore, the concept of vulnerability is a starting situation of the affected population before any interventions are undertaken (Climent-Gil, Aledo, and Vallejos-Romero 2018), which means that mapping vulnerabilities is a prerequisite for the proposition and implementation of strategies (Caldas et al. 2018).
Considering the discussion of social and environmental impacts in the disaster risk reduction (DRR) (i.e. see more details in Fuchs, Kuhlicke, and Meyer (2011)), we merged the definition of each disaster variable ( Figure 2) in a socio-environmental framework detailed in Figure 3. The methodology reflects a combination of social and environmental phases (SS and ES), which are detailed below.

Data collection
Initially, we contacted policymakers and specialists from Campina Grande for data collection. At this stage, the Civil Defence Agency, responsible for managing flooding in the city, provided data to describe 101 flood cases across the city from 2004 to 2011. These points (Figure 1(b)) do not necessarily represent all the flooding areas of Campina Grande but show areas that experienced flooding and people reported officially to the Civil Defence.
Simultaneously, international, local research and official data sources of Brazil (e.g. IBGE, CEMADEN, CPRM) were considered for preparing a preliminary list of vulnerability and exposure indicators.

The participatory approach
Subsequently, we developed a place-based citizen science project, called PLANEJEEE Project: To Plan Extreme Events ('Planeje Eventos Extremos' in Portuguese) from May to June of 2019 with specialists, policymakers, and citizens. The participatory approach was built upon the review of other citizen science studies (Ajibade and McBean 2014;Eitzel et al. 2017;Duan et al. 2018), GIS-MCDA approaches in similar fields, and the planning legislation of Campina Grande. Mixed qualitative and quantitative methods, namely focus groups, workshop and questionnaires, were held with stakeholders. The engagement opportunities aimed to discuss current challenges and strategies to mitigate floods and water shortage, the promotion of critical reflection from the participants (Groulx et al. 2017) and aspects to enhance risk communication (Cheung and Feldman 2019). Details of the PLANEJEEE Project are detailed below.
Citizens were selected based on the Civil Defence flood dataset and by suggestions from the residents themselves with 172 households interviewed. Specialists and policymakers were asked to join a workshop and focus groups, according to their research field (for specialists) or position in the city council (e.g. planning, urban services, engineering, health, education, traffic, GIS, science and technology), water companies (e.g. AESA and CAGEPA) and to the society (e.g. Civil Defence, CONCIDADE, NGO). We aimed to engage with different individuals that support city management. In summary, 27 people attended the workshop and focus groups with 22 survey answers (n total = 199).

• The selection of vulnerability and exposure mapping indicators:
Within this approach, the selection of the indicators was made in two phases, referred to as 'social phase (SS)' and 'environmental phase (ES)' in the framework ( Figure 3). The indicators were selected into four stages: (i) selection according to the flooding causation (questionnaires to all stakeholders), (ii) selection according to the social context (survey with residents), (iii) discussion of key challenges and solutions (workshop with specialists and policymakers) and (iv) comparison with previous studies.
The preliminary list of indicators (i.e. referred in section 3.1) of vulnerability and exposure mappings was used to prepare the questionnaires for the stakeholders' collaboration. Surveys were developed with a 5-point Likert scale (i.e. 1 -less importance to 5 -more importance). If the respondents were unsure, they could opt for the 'I don't know' option. Empirical statistical analysis tools (mean -M and standard deviation -SD) were used to examine the questionnaire answers in Python. Although there is a consensus of indicators that may influence vulnerability to disasters, particularly in the social context (Cutter et al. 2008), our intention with the PLANEJEEE Project was to find indicators that would characterise the city in the mappings according to the view of stakeholders. This is based upon on the assumption that they have knowledge by living experiences of the city exposed to floods (Hardoy, Gencer, and Winograd 2019). Our intention with the indicator's choice is not to discard other criteria but to provide mappings according to the city's pre-existing context.
In this way, to investigate reasons for the vulnerability in the city, the stakeholders were asked what the flooding causations are ( Table 1). The options encompassed four main categories of vulnerability suggested by Ghajari et al. (2017) and Pescaroli and Alexander (2019) regarding issues related to 1 -Households (social and structural vulnerability); 2 -Conditions of the drainage system (structural vulnerability); 3 -Interventions in the city (structural vulnerability); 4 -Legislation (institutional vulnerability). Table 1 shows that all stakeholders scored the options b, c and d with the highest scores (M: from 4 to 5). Also, the SDs of these options (b, c, and d) are smaller than 1, which represents a good consistency of answers. In general, stakeholders consider issues related to social, structural, and institutional vulnerabilities as the main causes of flooding.
To investigate the social context, we evaluated specific issues through citizens' participation. From the 172 respondents, 94.8% of the residents faced the previous flooding in the city, and 75.46% had flooding inside their property (direct experience). Approximately 38% of the respondents had more than 55 years old, and 53% of the households had children living in the property. About the income, preliminary results indicated that 80% of the interviewed citizens receive less than two minimum wages monthly. Residents were asked what the limitations for applying flood reduction measures are, in which 'money' was considered the primary constraint. Approximately 36.6% and 57% scored the option as 4 (high) and 5 (very high), respectively. The mean for this question was 4.5 (between 4 and 5 Likert scales) and SD 0.72. These multisectoral issues were also discussed in the PLANEJEEE workshop with specialists and policymakers. Participants were divided into four groups with severe flooding cases to discuss challenges and possible solutions for both flooding and water shortage. Maps with the vulnerability and exposure indicators were provided to facilitate the spatial visualisation of the 'weaknesses' and 'strengths' of the city. Stakeholders discussed the importance of other indicators like elevation and damages of important assets in case of flooding, which were included in the list of indicators. A summary of discussions is expressed in the Table X1 in the supplementary materials.
Lastly, the final choice of indicators took into consideration studies in similar fields ( Table 2). The vulnerability indicators represent the 'current state', which can vary according to time and space (IPCC 2014). The exposure refers to the density of vulnerable residents and the distance to critical infrastructure, which will have considerable damage if they are exposed to the hazard. Table 2 provides a brief explanation of each indicator.

Participatory-fuzzy-entropy integration
The Shannon Entropy method was adopted to compute indicators' weights. This method starts by computing a decision-matrix for the set of indicators (step 1) where a certain quantity of information can be used to find appropriate weights to each indicator (Boroushaki 2017). The data-driven method is considered as a measure of uncertainty (Birgani and Yazdandoost 2018).
In this paper, we developed a pixel-by-pixel analysis, by coupling GIS and Python, in which all the points of the surface are computed to find the weights. The final raster-matrix represents 670,364 cells with 10 m × 10 m analysed in relation to the criteria to map vulnerability and exposure. Results show the diversity degree for each criterion, where larger values denote that more diverse information is contained in a set of criterion values (Boroushaki 2017). The greater the entropy index, the greater the influence of the mapping criterion (Roodposhti et al. 2016). A Python script was developed for the weight's calculation according to the following steps: • Step 1: Calculate the normalised value (r ij ) of each cell (x ij ) to each j-th criterion in the decision-matrix: Step 2: Entropy (E j ) is calculated as a set of values of j-th criterion for m pixels: where the constant k (k ¼ 1 1 ln m ) ensures that remains between 0 and 1. • Step 3: Diversification degree (d j ) implying uncertainty is calculated for each j-th criterion as: Step 4: The final weight of j-th criterion is calculated based on the following equation: where w j is the weight of j-th criterion without consideration of stakeholders' preferences. The final weights for each criterion can be seen in Table 3.
In mapping analyses, a major contribution can be seen with fuzzy set theory and fuzzy membership functions (FMFs) to deal with vague data, e.g. Roodposhti et al. (2016). FMFs represent the degree of membership value concerning a particular indicator of interest. The fuzzy theory is a method used to minimise inherent uncertainty from data and improve the results' reliability (Gheshlaghi and Feizizadeh 2017). There is no optimal method for choosing the types of fuzzy functions (Roodposhti et al. 2016). This study used the linear FMFs, that transforms the input values linearly on the 0 and 1 scale, with 0 being assigned to the lowest input value and 1 to the largest input value. All inbetween values received some membership value based on a linear scale, with the larger input values being assigned a greater possibility, closer to 1.
The linear FMFs were applied to express the direction of analysis to each indicator (Table 3) within the spatial tools of ArcGIS Pro (ESRI). For example, the pixels with more 'imperviousness' increase the flood vulnerability. Each indicator was mapped with the 'fuzzified' functions and represent layers for the vulnerability and exposure mapping (Figure 1(a-n) in the Supplementary Materials). Along with the participatory approach to select the indicators, the integration between fuzzy theory and entropy is made by proposing an equation to obtain the final vulnerability and exposure. The final map will be a sum of weighted and 'fuzzified' indicators (Roodposhti et al. 2016). The final maps follow the equation: where DV is the disaster variable (vulnerability and exposure) of flooding hazard in each pixel, w j stands for the weight of each criterion and f j for the fuzzy standardised criterion. The final maps are presented in Figure 4(a-c). Each disaster variable was classified in a five-range susceptibility according to geometric intervals and natural breaks of the dataset, from 'very low' to 'very high'.

Results
The flood vulnerability of Campina Grande is shown in Figure 4 (c). Overall, the final map reveals areas that are more susceptible to flood, according to a combination of physical, social, structural, environmental, and institutional vulnerabilities. The mapping expresses the different levels of flood vulnerability and highlights specific locations with fewer conditions to deal with the extreme rainfall event. The anticipatory strategy to deal with floods is following the 'pre-existing' state detailed previously (see section 3). In summary, the 'very high' and 'high' vulnerabilities correspond to 15.80% and 25% of the city area, respectively. The     (2013) ' þ ' indicates an increasing function. ' À ' indicates a decreasing function 'moderate' vulnerabilities are seen spread throughout the city (34.20% of the city area). The least vulnerable ('very low' and 'low') areas represent approximately 25% of the city. The vulnerability map was validated in two stages. First, we compared the flood vulnerability levels according to mixedsource information datasets. Datasets were built with the flooding complaints from 2004 to 2011 provided by the Civil Defence (n = 101) and with the residents (n = 123) that confirmed to have direct experience (DE) with flooding in the participatory approach. The 224 points represent 'known' locations with flooding (Figure 4(c)).
The validation was performed by extracting the vulnerability mapping pixel values with the 'Sample tool' within ArcGIS Pro (ESRI). Since the flooding in the 224 points was confirmed by residents and authorities, our assumption was that it represents areas with a relationship (strong or weak) between vulnerability and exposure that conveyed in the flooding occurrence (Hazarika et al. 2018). This is based on the dynamic character of disaster, in which the characteristics in place on the instant the hazard takes place will define the intensity of the impacts (Pescaroli and Alexander 2019). The sample analysis showed that 196 points with flood complaints and residents (DE) were classified with the 'moderate and high' vulnerability to flooding, which validates the mapping in approximately 90% of the dataset. It is important to mention that the other 28 points represent areas with 'very low' and 'low' classifications of vulnerability but still can turn into a disaster if in contact with exposure and extreme precipitation. Additionally, the mapping indicators (Tables 2 and 3) and final outputs (Figure 4(a-c)) were discussed and approved by authorities of Campina Grande in meetings held in January 2021.

Discussion
Due to the nature of flood events and the relatively reduced time for preparation when the rainfall occurs, actions for flood management are mostly applied after it is transformed into impact (Pescaroli and Alexander 2019). Although disasters are not preventable, we recognise the importance of analysing vulnerability assessments and hazards mitigation as essential for reducing impacts (Frigerio et al. 2016). In this study, we consider these issues as influences that need to be fully understood regarding their patterns of vulnerabilities and exposures to reduce the risk (Pescaroli and Alexander 2019). In other words, we suggest that interactions between the multiple components of vulnerability provide insights on how to improve disaster management, including the amelioration of the structure of urban drainage system (Sivapalan, Savenije, and Blöschl 2012) and the reduction of social and institutional vulnerabilities (Cunico and Lohmann 2017;Marchezini et al. 2017) before the hazard occurrence.
In this context, using the case of Campina Grande -Brazil, we discuss how the interactions between social, structural, and institutional vulnerabilities converge to generate the overall vulnerability and exposure. Since most research focuses only on vulnerability assessments as the disaster impacts but pays less attention to the reasons behind vulnerabilities and approaches for alleviating these issues (Ghajari et al. 2017), we consider these results to contribute to the socio-environmental discussion of how to mitigate flooding with a vulnerability perspective.

Aspects generating the flood vulnerability
When analysing the vulnerability of the system, this approach considers the interrelationship between datasets characterised by the following situations: (i) the increase of sensitivity and the decrease of capacity will generate more vulnerability, and (ii) the intersection of vulnerable, exposed and hazard-prone areas will culminate in the disaster occurrence. The analysis is based on crosstabulated pixel-by-pixel information of vulnerability indicators according to Pearson correlation. Figure 5 shows correlations from −1 to 1, indicating negative and positive correlations, respectively. In this phase, our intention was not to state causality between datasets but to evaluate how the indicators typically move together.
Pixels with 'Open Sewage -OS' and 'Accum. Garbage -AG' indicators are positively correlated (+0.68), which indicates that households have increasing and simultaneous issues with drainage capacity. This result was confirmed in the PLANEJEEE project, in which the low maintenance and design of network were underlined as the main causations to the flooding in Campina Grande (Table 1). Overall, structural vulnerabilities are suggested as flood causations in Brazil, particularly related to the drainage capacity (Sarmento Buarque et al. 2020;Goncalves et al. 2018). Cities with large geographical differences regarding urbanisation and climate are susceptible to floods, especially in areas with poor risk communication, social inequalities and lack of capacity (Marchezini et al. 2017).
We suggest that not only structural aspects are inherently corroborating for flood vulnerability but also institutional and social aspects. Since changes in land-uses contribute to a more significant frequency and intensity of floods by increasing surface runoff (Caldas et al. 2018), it is essential to inspect the imperviousness of land-use rates by management authorities. In the Brazilian context, a weak inspection of the legislation fulfilment is seen in different cities. The Master Plan of Campina Grande regulates the maximum imperviousness of 80% of the area in each lot; however, this threshold is often exceeded by residents without any consequence (Alves et al. 2020) (i.e. see Alves et al. (2020) for a complete legislation analysis).
Stakeholders of the PLANEJEEE Project also highlighted other issues, like the low implementation of legislation, including urban planning (i.e. the Master Plan) and the lack of regulation specific to drainage as causes for vulnerability. For example, even though there is a requirement of updating the Master Plan in every 10 years, the latest version of Campina Grande's Master Plan is from 2006. On the institutional level, other aspects like the poor collaboration between academia, citizens, and public/private administration were highlighted and reflects the disconnection of urban and water resources planning in the city (de Araújo Grangeiro, Ribeiro, and de Miranda 2019).
In an attempt to identify the conditions that make people or a place more vulnerable (Cutter et al. 2008), we emphasise the importance of analysing the social context of the city. This is based on the assumption that it reflects a 'potential of loss' (Cutter 1996) that in the context of disaster risk management is the most tangible manifestation of the social construction of risk (IPCC 2012b;Hazarika et al. 2018). The blue boxes on Figure 5 indicate the negative correlation between datasets. For example, the comparison between 'Imperviousness' ('institutional vulnerability') and 'Income' ('social vulnerability') shows that when the imperviousness increases, the percentage of people with fewer income decreases (−0.50). This result suggests that more imperviousness is found in locations where more people with higher incomes live. This result corroborates with Cutter, Boruff, and Shirley (2003), where it is shown that social processes interact with natural processes and the built environments to redistribute the risks and the impacts of the hazards. Our analysis supports the conclusion that people with more income tend to reduce perviousness and create more flooding, which can indicate a low-risk perception of residents.
In addition, the participation of residents in the socioenvironmental methodology allowed the conclusion that social aspects are also primary contributors for vulnerability since fewer individual and community resources for recovery are available in Campina Grande. Citizens detailed flood damages that are not only related to the duration of the flood event but also in the aftermath. Approximately 44% of the participants claimed to have lost assets in a flood event and 63% had to be temporarily moved to another location after the event. Besides, the residents mentioned problems with mud, animals (e.g. mice, snakes, and cockroaches) and structural losses that appeared after the runoff of waters. This shows that income is not only needed to mitigate and to cope with flooding but also to recover from the hazard. For Birkmann (2007), difficulties in recovering from the negative impacts of hazardous events also generate vulnerability, which makes coping and recovering part of its assessment. Therefore, we suggest that a combination of institutional and social strategies to provide better financial conditions and generate enhanced risk perception and coping capacity must be implemented in the city as resources to decrease flood vulnerability and increase resilience (Nguimalet 2018).
Finally, we suggest that reducing exposure of most affected groups is also crucial for minimising future risks. This factor is confirmed in the literature (Cutter et al. 2008) where demographic groups like the percentage of elderly and children's presence impose more difficulties on the community to cope with flooding (de Brito, Evers, and Höllermann 2017). For Fuchs, Kuhlicke, and Meyer (2011), exposure can be seen as the relationship of elements at risk to the hazard. Therefore, defining exposure is a bridging element between the natural and social scientific part of the risk. In other words, the exposed elements detailed in this study (Figure 4  (a,b)) are vital for management since it shows the density of people and distance to assets, which will be impacted by the hazard and vulnerability indicators. Hence, the spatialisation of exposure enables the assessment of exposed areas with a social view that can help managers and policymakers for the flood management.

Limitations and next steps of the socio-environmental approach
As the vulnerability is a relationship between a series of categories that are not independent but interact on many different time and space scales (Pescaroli and Alexander 2019), the dynamic aspect of vulnerability is key. In this context, the choice of indicators, including the assessment of weights and standardisation is still difficult. In this regard, our approach provided a pixel-by-pixel mapping, in which relationships were assessed and discussed between the indicators themselves and each disaster variable. We argue this strategy could be used to prioritise areas for reducing vulnerability before the flooding.
The integrated participatory-fuzzy-entropy approach considered uncertainty error possibilities since conception (Sharma and Ravindranath 2019) until applying the framework (Malczewski and Rinner 2015). We aimed to increase the collaboration of stakeholders, in all SES phases, from indicators to the mapping results and validation. This appears to have great importance in real-life applications since there is a need to select indicators that the stakeholders can understand for later use (Parker et al. 2019). This is based on the conclusion presented by Fuchs, Kuhlicke, and Meyer (2011) where the importance of clearly describing and defining the components of risk and/or vulnerability is considered essential for the management.
However, a limitation of our work is that the lack of datasets could constrain it. Without datasets, the objective phase of the framework would not be possible. So, it can only be applied in areas with representative data. Similar to vulnerability, the risk is complex and dynamic, which requires regular re-assessment (Peduzzi 2019). Using our framework, the reassessment can be facilitated by the classes of indicators and by the pixel-by-pixel analysis, in which the stakeholder will have the complete analysis of the behaviour in each variable (sensitivity, capacity and exposure). Further research must apply this methodology considering future changes in the datasets.
The use of mixed-source information is being significantly used in flooding studies in the last years (Sarmento Buarque et al. 2020) as a low-cost tool for low monitoring areas. In this context, we developed a validation approach based on the confirmation of flood cases in which approximately 90% of the points were validated. In addition, to evaluate the acceptability of policymakers, the indicators and mappings were presented and discussed with stakeholders. However, a more specific approach with the collection of more recent flooding cases and flood levels may be implemented in the future.
In this work, flooding hazard itself was not considered in the mapping. Future results will express the impacts generated by the interrelationship between hazard, vulnerability, and exposure. The risk areas will be analysed in the system to locate adaptation and mitigation strategies for enhancing the system and reducing flooding. For further steps, it is important to link the vulnerability and exposure maps with social and institutional vulnerabilities, in a broader context with integration with other elements of capacity, such as risk perception, to the proposal and placement of solutions for reducing the flooding in cities with context similar to Campina Grande.

Conclusions
Even though disaster risk reduction research tends to focus mostly on hazard modelling (Peduzzi 2019) and in larger scales (Parker et al. 2019), this work detailed and quantified vulnerability and exposure in local scale. This work stands out in multidisciplinary research in water management since, until today, less effort has been made for addressing disaster risk variables beyond hazards modelling (Peduzzi 2019).
Our work's novelty is also shown by approaching risk components with a more holistic framework, where GIS is used as a geographic bridge between social and physical sciences (Lund 2015). Understanding vulnerability and exposure is extremely important for reducing the impacts of disasters in complex urban systems. In our work, vulnerability and exposure are expressed according to the new paradigm approach of IPCC (IPCC 2014(IPCC , 2012b. The combination of sensitivity, capacity and exposure is expressed with multiple indicators according to vulnerability categories (physical, environmental, social, structural and institutional) that create overall flood vulnerabilities (Cinner et al. 2018). Our results express the need to consider each multisectoral category of vulnerability as an important step for managing disaster risk reduction.
Our approach relies on inputs from households, policymakers, local experts and pre-existing datasets, where it is possible to prioritise indicators and areas that require more intervention and support from the city administration. The involvement of relevant stakeholders from different levels and sectors provided valuable input and datasets for the assessment and can improve co-ownership and acceptance of the results (Hardoy, Gencer, and Winograd 2019). The vulnerability and exposure are already conceptually complex, so, our objective was to characterise the system according to views from a multidisciplinary group of stakeholders (Hazarika et al. 2018), including the residents.
The disaster variable maps are essential for managers and policymakers to manage disaster risk, including the selection and proposition of solutions strategies (Caldas et al. 2018) and the selection of 'hotspots' areas for DRR. In this sense, the integrated framework can be a tool where specific issues in both social and environmental perspectives can be directly tackled to generate a future reduction of sensitivity and exposure as well as the improvement of capacity rates. Finally, the hazard-specific approach provides an opportunity to produce in-depth knowledge of how disasters are created, in a local scale, and can be input for disaster risk management before, during and after the extreme event.