2.1 The resilience concept and modelling approaches

As the concept of resilience is multidisciplinary, its definition and application as a risk management strategy are extremely complex. In order to move towards its operationalization, it is necessary to build an analysis model to address the concept. Several studies have attempted to build analysis models to define indicators or a specific baseline (Bakkensen et al., 2017; Fox-Lent et al., 2015).

The 100 Resilient Cities (100RC) consortium was launched by the Rockefeller Foundation in 2013. The purpose of the 100 Resilient Cities consortium is to help cities around the world become more resilient to the physical, social and economic challenges of the 21st century. 100RC supports the adoption and integration of a vision of resilience that includes not only disasters – earthquakes, fires, floods, etc. – but also the tensions that weaken the urban area on a daily or cyclical basis. Resilience is defined as the ability of individuals, communities, institutions, businesses and urban systems to survive, adapt and evolve, regardless of the types of chronic stresses or shocks they may encounter. A holistic approach is advocated. 100RC has built a framework defining the characteristics of urban resilience (Fig. 1).

Figure 1100 Resilient Cities framework (taken from 100 Resilient Cities, 2015).

The definition of resilience via these indicators allows identifying criteria for resilience in a territory or within a population. It allows launching discussion around an initially fuzzy concept. However, it does not allow visualizing criteria or resilience potentials at the local level (100 Resilient Cities, 2015). Mapping is non-existent, and the absence of tangible data makes it difficult for local populations and actors to appropriate the concept, understand it and reproduce it.

Another study focused on identifying resilience capacities applied to urban networks. It led to the creation of the DS3 (spatial decision support system) model (Serre, 2018). Three resilience capacities were defined to study resilience (Serre, 2018), namely resistance, absorption and recovery (Fig. 2).

Figure 2DS3 Model (taken from Serre, 2016).

The resistance capacity is necessary to determine the material damage of the networks. It is a given that the more damaged a network is, the slower and more difficult it will be to return it to effective service. The results of the damage analysis make it possible to measure this damage and determine the interdependencies between the various components of the networks.

Absorption capacity represents the alternatives available to the network following a failure. The idea is to highlight solutions to maintain service continuity despite floods, operating in degraded mode.

Finally, the recovery capacity represents the time required to retrofit the networks until reaching a full level of service.

The DS3 model can be used to identify factors that would lead to increased urban resilience, highlighting the importance of urban networks and critical infrastructures. This technical approach focuses mainly on urban networks. However, cities comprise many factors, such as social dynamics, urban interactions and technical components, leading to additional indicators that must be monitored (Serre and Heinzlef, 2018).

A third study conducted by Cutter (Cutter et al., 2010) identified six indicators to measure resilience – social, economic, community, institutional, infrastructural and environmental. Each indicator is divided into sub-variables such as education, age, language proficiency, employment rate, immigration rate, access to food, disaster training, social stability, access to health, access to energy and so on. Each variable has a positive or negative effect on community resilience. Calculated using quantitative data, this method makes it possible to quantify and map resilience at the national level and more specifically at the county level in the United States. While this method greatly facilitates comparison across a large number of variables, the disadvantage is that the final score is not an absolute measure of community resilience for a single location but rather a relative value against which multiple locations can be compared. For this reason, the proposed work is done at the US scale (Fig. 3) and not at a finer scale or for a single year, not being a comparative work over several years.

Figure 3Disaster resilience value in USA (Cutter et al., 2014).

These three approaches attempt to address the biases of conceptualization and modelling resilience. But, in the first approach of the concept and data visualization, there is nothing evident about how the results should be processed and explored. In the second approach, the exploration of the results is visible through the application of the methodology, notably in a case study on Hamburg (Serre et al., 2016). Nonetheless, the exploration and analysis of the data are not accessible to the public concerned, limiting their understanding and appropriation of the method. Moreover, this approach analyses the territory only through urban networks and not with the other components that shape it. The third approach proposes a measurement and mapping of resilience, but the scale of analysis selected does not allow for decision-making by local stakeholders.

The objective of this work is therefore to model and operationalize resilience as comprehensively and exhaustively as possible. The aim is to analyse it at the local level in order to advise stakeholders and lead to decision-making that integrates resilience strategies into risk management.

2.2 Geovisualization techniques: added values in risk management processes

For many years, risk mapping was one of the main methods used to analyse, represent and examine the multiple characteristics of risks and risk management strategies (Barroca and Serre, 2018). However, new methods have been introduced, such as geographic information systems (GISs) and scientific information visualization (Kraak, 2003). A GIS gives access to voluminous and heterogeneous tools like databases and graphic applications to establish interactions between data and maps. These interactions can be visualized through an interface used to explore the characteristics of data. The adaptation of scientific visualization to mapping was initially called “geographic visualization” and then “geovisualization” (Maceachren and Kraak, 1997). Geovisualization is defined as “the set of visualization tools that allow interactive exploration of geolocated data in order to build knowledge without assumptions a priori” (Maceachren and Kraak, 1997). Geovisualization includes fields such as scientific visualization, mapping, image processing, knowledge extraction and GISs.

Therefore, geovisualization is a synthesis approach applied to GIS techniques that integrates practices such as mapping, visualization, data and image analysis by analysing geospatial data (MacEachren and Kraak, 2001). This methodology offers the possibility of representing multidimensional, voluminous and heterogeneous data. More specifically, geovisualization is mainly adapted to the representation and analysis of georeferenced data. The mapping exercise is divided into several objectives: explore, analyse, synthesize and present. Geovisual tools must be adapted to these different uses. The different tools currently available can be differentiated by three criteria. The first is the audience, which can range from the “general public” with little knowledge of geovisualization issues to experts with good knowledge of the subject. The second is the degree of interactivity offered by the geovisualization tool. The last criterion is knowledge of the data, which varies from the domain of the known to the domain of the unknown (MacEachren and Kraak, 1997). The four uses of geovisualization can therefore be placed inside a cube. Each axis of this cube ($x,y,z$) represents one of the three criteria previously mentioned (audience, interactivity, data relations).

This data representation (Donolo, 2014) – also called “virtual science” – allows constructing, reconstructing, representing and interpreting scientific issues (Yasobant et al., 2015). The ability to represent spatiotemporal data in different forms provides better understanding of the different phenomena involved, resulting in either better dissemination of the information or better decision-making. This methodology allows exploring hypotheses, sharing arguments, developing solutions and, most importantly, building common knowledge around the same issue (MacEachren, 1997).

Consequently, these characteristics and advantages make geovisualization an interesting methodology for studying risk management. Crisis management is, indeed, a concrete example of where it is useful to use visual, map-based tools to integrate, assess and apply multisource geospatial information and data (MacEachren et al., 2004). Indeed, in a context of climate change and related uncertainties, modelling or simulating disasters such as floods is becoming increasingly complex. Current techniques are limited in the face of the complexity of floods, particularly because of the multiple reasons, sources and causes of disasters (Leskens et al., 2014; Löwe et al., 2018), as they are essentially used to model urban planning projects or response strategies to cope with the increase in the occurrence of such events. Many studies have used geovisualization to analyse the complexity of flood risks, whether to analyse flooding from the perspective of risk, for instance expected damage (Meyer et al., 2009; Ward et al., 2011); hazard, such as duration, velocity, water depth, etc. (Schumann et al., 2009); and management strategy (de Moel et al., 2015) at the national (Burby, 2001), regional (Elmer et al., 2012; Gaslikova et al., 2011; Vorogushyn et al., 2012) and local (Aerts et al., 2013; Apel et al., 2009; Gerl et al., 2014) levels and even on the built scale (Fig. 4), with, for example, FReTs (flood resilience technologies; Schinke et al., 2016; Golz et al., 2015). Geovisualization techniques make it possible to aggregate different types of raw data (e.g. underground dynamics, urban structure, building vulnerability); transform them by joining these data (Fig. 4), calculating the damage rate based on these raw data; and then produce a final result, translated into a dynamic, understandable and accessible map.

Figure 4Links between resilience modelling and geovisualization techniques – at the building scale (Schinke et al., 2012).

To summarize, geovisualization helps in exploring data, using visual geospatial representations to imagine hypotheses, solve problems and co-construct scientific knowledge (Kraak, 2003). Therefore, geovisualization methodology improves territorial knowledge and leads to tools such as decision support systems by making possible dialogues between users and stakeholders and promoting collaborative approaches. In the field of risk management, it is essential to defuse subjects of tension in order to present a risky situation objectively.

2.3 Making urban resilience operational through geovisualization techniques

Although several methods can be used to model risk characteristics, such as hydraulic modelling (Ernst et al., 2010) and geomorphological parameters (Bathrellos et al., 2012), it is quite difficult to model such fuzzy concepts like resilience and vulnerability, despite the common use of the latter in risk management. While the implementation of resilience policies and the design of resilient cities are desirable, assessing resilience and implementing it are complex. Several researchers have examined the difficulty of defining, implementing and evaluating urban resilience, usually through a geovisualization approach.

Cutter and Finch (2008) presented SoVI (Social Vulnerability Index), a tool providing a county-level (USA scale) comparative metric of social vulnerability to natural hazards based on socioeconomic and demographic profiles. The aim of SoVI is to illustrate the geographic patterns of the USA by defining social vulnerability as the sensitivity of a population to natural hazards and its ability to respond to and recover from them. Using several maps and in view of improving emergency management, SoVI identifies which areas of the American territory are more or less vulnerable and why. Geovisualization techniques improve understanding of the concept of vulnerability and help urban managers to localize vulnerable areas and variables.

Figure 5The ViewExposed interface (Opach and Rød, 2013).

Based on SoVI, a Norwegian study examined the vulnerability of territories to climate change (Opach and Rød, 2013). To avoid an increase in local and national vulnerability, the researchers built a ViewExposed tool (Fig. 5), whose objective was to inform local authorities about the most vulnerable areas of Norwegian territory and the causes of this vulnerability.

The ViewExposed tool focuses on Norwegian municipalities' exposure to natural hazards and the capacity of local populations to resist them. This interface tool was designed for professionals, local elected officials and local residents. It is the result of collaboration between scientists and local experts via workshops. Although focused on the concept of vulnerability, this tool also integrates the response of local managers and actors to natural disasters, corresponding to the first step in resilience integration.

Another team of researchers and insurers developed a tool to help people in the Nordic countries to protect themselves and prepare for climate risks. The main target users are private landowners, but this tool can also be used by land-use planners and property managers. The tool, VisAdapt (Johansson et al., 2016), is intended as a guide on how to prepare for climate events liable to affect individual homes. It is very simple to use, so every citizen can employ it. The obvious interest of this tool is that it allows addressing local inhabitants directly by proposing solutions to adapt to natural risks linked to climate change.

These tools have the merit of proposing operational instruments to obtain a clearer idea of the vulnerability concept involved. The main scale is above all the national scale, which, despite major advances in the visualization and knowledge of vulnerable zones, does not always lead to decision-making by local actors and managers, since the scale is sometimes too broad for action. Beyond the spatial scale, the choice of data and tools for processing, analysis and visualization has not been designed for non-expert audiences. Data are not always freely accessible, nor are the processing and representation tools. Some tools are intended only for professionals, while others point to the need to open the results to a wider audience. In addition, the data are not accessible and downloadable, which makes the methodology difficult to adopt and reapply in other territories outside the scope of expertise of the research team. The limits are therefore divided between the choice of spatial scale, the free and accessible nature of data and tools, and the non-integration of local actors and thus the assurance of their understanding of the tools and concepts used. In addition, this research focuses on the “vulnerability” prism of risk management. While we defend the fact that these two concepts are linked and inseparable (Provitolo, 2012) in the apprehension of climate disruption (Heinzlef, 2019), the difficult definition of resilience and its operationalization is noteworthy. When vulnerability is defined as the propensity of a territory and a population to suffer damage, resilience focuses on the strategies and means to prepare territories and populations for the increase in risks and their damage in order to limit the negative impacts. Resilience is therefore more complex to quantify, operationalize and visualize. Here, we intend to overcome these limitations by proposing the approach we developed.

3.1 A framework for defining resilience data?

To analyse urban territories including their complexity through the prism of resilience, it was necessary to define their issues and challenges; their dynamics, material and immaterial interactions; and their structures that impact the functioning of urban space. It is essential to understand the city as a system (Gardner, 2016). These urban systems, like any living organism, are complex and hierarchical. Some studies have explored the impact of rapid urbanization, leading to complex territorial responses and the lack of suitable reactions. In parallel, challenges can increase when the country's gross domestic product (GDP) decreases. Nevertheless, in some case studies, urbanization has been shown to have other results and is one of main elements to be taken into account when building response capacity to risks (Garschagen and Romero-Lankao, 2015). This response capacity can be determined by flood preparedness (Chinh et al., 2016), government implication and risk governance (Garschagen, 2015). Studying the city in the face of risks and its resilience capacity requires considering different spatial scales of interactions and challenges. Therefore, several questions must be asked to support the understanding of the concept of resilience and decision-making, including the following. Who is vulnerable and/or resilient? What? When? What elements could limit the impacts of a crisis like a flood event? Are they efficient before, during and after a flood?

These questions allowed us to establish three resilience indicators to study technical, urban and social resilience (Heinzlef et al., 2019). The methodological choice of using indicators was based on several arguments. The first one is that by defining and characterizing an abstract concept, indicators allow sensitizing both the scientific community and the public to complex subjects (Prior and Hagmann, 2014). In addition, resilience indicators can make an important contribution to assessing a community's needs and goals while helping it to develop resilience strategies (Cutter, 2016). These indicators, useful when creating a strategy, are also important for monitoring the decision-making process. Finally, an essential benefit of using such indicators is that they can act as driving factors for risk management by including the concept of resilience clearly and more holistically (Linkov et al., 2014). The objective is to analyse the different social, urban and technical components (Serre and Heinzlef, 2018) of the area concerned (Table 1). The indicators were designed after adapting the Baseline Resilience Indicators for Communities (BRIC) methodology (Cutter et al., 2008; Patil et al., 2008; Singh-Peterson et al., 2014).

Table 1Example of data selection, sources and references.

Download XLSX

3.2 Resilience processing

The advantages of using geovisualization techniques to remove barriers to resilience are as follows:

• analysing heterogeneous and geolocated data,

• supplying a visualization based on the most recent scientific advances,

• extracting, producing and sharing data with innovative layouts.

Figure 6Resilience characteristics (Serre and Heinlef, 2018).

To address these three pillars, we propose clarifying some of the resilience criteria defined above (Fig. 6) around three resilience indicators: social, technical and urban.

We argue that analysis of resilience at the local level is facilitated by using open-access data. On the other hand, data processing and analysis become more understandable for local actors when tools are chosen that highlight the visualization.

3.2.2 Method and tools for resilience assessment

After selecting the raw data, data were transformed and normalized with a theoretical orientation. In order to understand the frequency of each variable, each item of raw data has been transformed into percentages:

$$\begin{array}{}\text{(1)}& {\displaystyle }\text{Social Resilience}={\displaystyle \frac{\text{Variable}}{\text{IRIS Population}}},\text{(2)}& {\displaystyle }\text{Urban resilience}={\displaystyle \frac{\text{Variable}}{\text{IRIS AREA}}},\text{(3)}& {\displaystyle }\text{Technical resilience}={\displaystyle \frac{\text{Variable}}{\text{IRIS AREA}}}.\end{array}$$

Nevertheless the weighting is 1 for all the variables (Holand et al., 2011). This single weighting is explained by the willingness to avoid disparities between the variables (Fekete, 2009), since some of them are sensitive and subjective. Indeed, we have no theoretical references (Esty et al., 2005), and there is no practical experience (Fekete, 2009) on which to determine weights that are mostly subjective. Besides, applying such weights does not necessarily reflect decision makers' and urban planners' priorities and realities (Cutter et al., 2010). Nonetheless, since this approach puts forward a participatory and collaborative methodology, readjusting the weight of these variables with regard to managers' perceptions is entirely justified and in line with the current approach.

Following this process, it was necessary to determine the normalization. Normalization allows adjusting a series of values (typically representing a set of measurements) according to a transformation function to make them comparable with certain specific reference points. We proceeded with a min–max standardization (Casadio Tarabusi and Guarini, 2013) to obtain a positive resilience impact variable (Eq. 4) and a negative resilience impact variable (Eq. 5), where each variable is decomposed into an identical range between zero (worst rank) and 1 (best rank), to create indicators with similar measurement scales, and to compare them:

$$\begin{array}{}\text{(4)}& {\displaystyle \frac{x-\min \left(x\right)}{max\left(x\right)-\min \left(x\right)}},\text{(5)}& {\displaystyle }\mathrm{1}-{\displaystyle \frac{x-min\left(x\right)}{max\left(x\right)-min\left(x\right)}}.\end{array}$$

The choice of processing tools was influenced by the availability of open-source tools in order to uphold transparency and collaborative approaches as well as the availability of such tools to all stakeholders. To create the computer script, we used a tool, the Feature Manipulation Engine (FME), to extract, transform and load raw data (extraction–transformation–loading). Its interface allows visualizing each step of the processing, from loading raw data (INSEE files) to choosing variables, while integrating the resilience formula to finally obtain the results. Although this tool has a cost, it is nevertheless used by GIS practitioners at a large scale, nationally and internationally.

Figure 7Details of the social resilience assessment process.

Download

Several steps (Fig. 7) were necessary to set up the computer script, integrate the input data, create a geometry, generate the processing and forecast an overall resilience value. The output data are in SpatiaLite format (sl3 format), which is a spatial extension of SQLite and provides vector geodatabase capacity. This format can be understood by many processing, visualization and mapping software applications, including QGIS.

Once the computer processing was completed, the visualization and analysis work was done via a GIS, namely the QGIS software (Fig. 8). It allows the automatic spatialization of data according to data variables or variables resulting from relationships between objects and finally the use of graphical tools to visualize and differentiate data (sizes, colours, distances).

Figure 8QGIS interaction architecture for resilience assessment.

Download

The map is therefore a decision-making tool in the sense that it represents and filters a mass of data and makes them accessible and comprehensible. But the production of a map cannot in itself be considered a spatial decision support system: its value first depends on the consistency and reliability of the information collected upstream, then on its structuring and effective readability.

4.1 Avignon flood issues

Avignon, the chief administrative centre of Vaucluse, is faced with flood risk due to its proximity to the confluence of the Rhône and Durance rivers (Fig. 9). The island of Barthelasse, the largest river island in Europe, is the area of Avignon most affected by the Rhône's floods. It serves as a buffer between the city and the Rhône and serves to absorb floods. The few existing dikes protect the island from low floods, but it is still floodable, as shown by the 2 m floods in 1993, 1994, 2002 and 2003.

Figure 9Extreme flood scenario in Avignon (Heinzlef et al., 2019), inspired by © DREAL PACA.

Therefore, a spatial decision support system that integrates resilience into practice would be helpful for a flood-prone community. It was developed in partnership with the Avignon city council GIS department. This collaboration took place at several levels, namely in the involvement of local actors in the study, in the data exchange process and in the choice of processing tools, to ensure and improve the reusability of the methodology once the study had been completed. The final choice of resilience variables, data processing and their final visualization was made in constant collaboration with the city's technical services to ensure that data and their analyses were shared and understood.

4.2 Resilience to flood assessment in Avignon: a few results

The city was divided into a local scale – IRIS scale – to visualize which areas are resilient or not. The indicators – social (Fig. 10), urban and technical – and each variable (Fig. 11) included in the model can be visualized. Therefore, it is easier to perceive which variables improve resilient capacities and which areas have developed these variables or not. As each indicator is independent of each other, it is easier for politicians and managers to work on variables with low levels of resilience and identify areas to be redeveloped and/or reintegrated into urban dynamics.

Figure 10Social resilience indicator – multi-scenario, Avignon scale (IRIS-scale analysis). Panel (a) identifies the most resilient areas (greener) according to the social resilience characteristics before a crisis, panel (b) identifies the most resilient areas according to the social characteristics during a crisis and panel (c) identifies the most resilient areas according to the social characteristics after a crisis – © Open Database License, “ODbL” 1.0.

The difference between the three maps in Fig. 10 is explained by the different scenarios considered before, during and after a crisis. Not every variable is included in every scenario. For example, age variables are important before (preparation, knowledge of risk, etc.), during (understanding of the situation, ability to move, etc.) and after the reconstruction process. On the contrary, whether or not individuals have a job does not play a role during the crisis but is decisive afterwards in order to rebuild and relaunch an activity.

Figure 11Population between 25- and 39-year-old-scenario before crisis, at the Avignon scale (IRIS-scale analysis). The map identifies the value of the population variable 25–39 years old according to the total IRIS population before a crisis – © Open Database License, “ODbL” 1.0.

Figure 11 shows the location of individuals aged 25 to 39 years, with a segment of the population potentially more resilient, before, during and after. Indeed, they can have a risk culture beforehand, act and survive during, and restart an activity after the disruptive event. They are more prevalent in the city centre and in the south and south-east. This is mainly due to the location of the two universities, in the city centre and outside the city walls, which favours student accommodation and low-cost housing.

Table 2Variable values – scenario before crisis.

Download XLSX

Table 2 presents the value of each variable according to the pre-crisis scenario for the IRIS of Barthelasse and Courtine (Table 2) after min–max standardization. These values illustrate the representativeness of each variable in the territory and make it possible to understand the social and spatial dynamics at the IRIS scale for Barthelasse and Courtine. This detailed analysis, carried out on a variable-by-variable basis, allows engaging in a discussion with local actors in an attempt to reintegrate neighbourhoods at the margins of territorial functioning in order to work on the integration of urban resilience in the face of daily territorial stresses and when confronted by a more exceptional event such as a flood.

Regarding the urban resilience indicator, INSEE data are available from 2009 to 2013, thus making it possible to perform a multi-date analysis over several years (Fig. 12) and gain understanding of urban evolutions and resilience trends. For instance, certain elements have evolved, such as the proportion of tourist accommodation and surgical and hospital activities, thereby increasing resilience capacities. Moreover, the advantage of using open data allows temporal as well as spatial scales to evolve, and the indicators can therefore be tested on other municipalities on the national territory.

Figure 12Comparative analysis between 2009 and 2013 for the urban indicator, post-crisis scenario, at the Avignon scale (IRIS-scale analysis). The map on the left identifies the most resilient areas according to the urban characteristics after a crisis in 2009; the map on the right identifies the most resilient areas according to the urban characteristics after a crisis in 2013 – © Open Database License, “ODbL” 1.0.

After reflection on the visibility of the results, information sharing, and a neutral and collaborative approach, we are considering making our work accessible to inhabitants by developing a website to continue the risk communication process on flood risks and strengthen the geovisualization process. This website, which is currently subject to reflection, was developed with the creation of interactive maps accessible via a weblink (http://umap.openstreetmap.fr/fr/map/rg_avantcrise_353189#12/43.9311/4.8914, last access: 9 April 2020) and a QR code (Fig. 13).

Figure 13Weblink global resilience – pre-crisis scenario map – QR code – © UMAP-Open Database License, “ODbL” 1.0.

Concerning the contribution of geovisualization used to promoter dialogue on the issue of territorial resilience, workshops were organized to develop interaction around the maps produced and the database provided and accessible to stakeholders. These workshops provided an opportunity for scientific experts and critical infrastructure managers as well as decision makers in risk management strategies to exchange views in order to support the reflection process and foster long-term collaboration. These maps and this new database allowed the actors to extract new knowledge from the decision support tool, especially theoretical knowledge provided by the maps and consistent with the database. This knowledge is both current and part of a long-term construction, since the data evolve as a function of INSEE production.