As a first approach, the risk of flooding can be understood through the “Source-Pathway-Receptor-Consequence (SPRC)” conceptual model (Gouldby and Samuels, 2005). This is a simple conceptual model for representing processes that lead to a particular consequence (FLOODsite, 2009a). For a risk to arise, there must be:

• A hazard that consists of a “source” or initiator event (e.g., in the marine domain, a cyclone or a tsunami, and in the land domain, coastal river flooding, runoff, rising water tables);

The SPRC model considers risk to be the result of the exposure of a vulnerable stake to a hazard. Therefore, flood risk can be defined in at least two alternative ways, the former being more suited to qualitative approaches and the latter to quantitative approaches (FLOODsite, 2009b).

The following formula reflects the first definition:

1) risk = hazard (flood) * exposure * vulnerability (of the society or the area).

This first definition is often preferred by urban planners, who usually consider the hazard as a given and utilize spatial planning and crisis management as the means to adapt to that given. The Figure 1 illustrates the application of this concept in the case of the impact assessment of coastal flooding on the Sangatte municipality (France). The cartographic representations of the hazard and the vulnerable exposed stakes give, by superposition, a representation of the risk.

However, in an attempt to quantify risk, and considering that the word « risk » suggests a probability of occurrence, a second definition may be sought. To do this:

• The two terms « exposure » and « vulnerability » are substituted by « consequences », with the consequences being generally more quantifiable (for example, in number of fatalities and economic damage) than the previous two terms;

This yields the second definition:

2) risk = probability (of the flood) * consequences.

The second definition is preferred by engineers, who usually strive for a reduction of the probability of flooding with protection means, and hence need to be able to calculate risk. Thomas et al. (2016) illustrated this approach. These authors examined the Dungeness and Romney Marsh coastal zone (United Kingdom), a region of high value in terms of habitat and energy assets. The objective was to redesign aging coastal defences in order to protect against a 0.5% annual probability event of flooding and to reduce consequences of storm surges to an acceptable level. Considering two parameters—the joint significant wave height (Hs) and water level (WL)—, Thomas et al. (2016) associated a storm impact model (simulating the overwashing discharge at the defence crest) with a flood inundation model (simulating the inundation extent and volume for the Hs and WL combination). Thus, they explored which combinations of wave and water levels generated the greatest threat, as shown by the Figure 2.

Klijn et al. (2015) showed that the two competing definitions can be reconciled and thus the dialogue between levee engineers and planners in charge of town planning can be facilitated.

Flood risk management has changed significantly in recent decades: until the end of the 20th century, the main objective was with strong engineering means, to keep flood off the land. Focus has been on reducing the probability that communities would experience floods, generally through investing in such measures as flood embankments, channel enlargement, pumping schemes, and sea walls. From the 2000s, the main objective became the reduction of risks for people and property. Therefore, measures to reduce the vulnerability of society were considered with the same attention as water management measures (FLOODsite, 2009a).

Figure 3 shows how, following the SPRC model, risk can be reduced by addressing one or more of the three risk factors: vulnerability, exposure, and/or hazard. The reduction of these three risk factors can be achieved through different policy and action choices over time until limits to adaptation might be reached (IPCC, 2019). Diverse measures can be implemented (general land use planning, dedicated protection systems, application of urban planning rules, flood forecasting and warning systems), but a “residual risk” will always remain.

In support of these strategies, the use of numerical models is a well-established technique for the understanding and representation of weather-marine phenomena (Daniel et al., 2007; Tiberi‐Wadier et al., 2021) and tsunamis (Sugawara, 2021), predicting their impacts on coastal zones and warn. In the fluvial field, studies using hydrodynamic models can also be utilised for flood management in general (Mehta et al.,2021), and in particular to make decision relating to the operation of dams, the relocation of people living in floodplain areas (Yadav and Mangukiya, 2021) and the developments of water courses and wetlands (Patel et al., 2018; Mehta and Yadav, 2020; Kumar and Mehta, 2021).

Flood risk management has proven successful at reducing the threat of some flooding hazards. It is a useful approach for assessing risks and guiding decisions on implementing protection measures (Disse et al., 2020). Nevertheless, in the changing context of the 21st century, marked by the increasing artificialization of soils and the multiplication of uses, the degradation of ecosystems and the services they provide (Millenn. Ecosyst. Assess., 2003; Millenn. Ecosyst. Assess., 2005; Campbell et al., 2009), flood risk management strategies based on the SPRC model or other simple models may be insufficient. To halt the ongoing processes in many territories, the scope of disaster risk reduction strategies needs to be broadened to address both natural and man-made hazards and all associated environmental, technological and biological risks (UNISDR, 2015).

To define these new strategies, the conceptual framework had to evolve. The concept of resilience corresponds to a new paradigm in the field of risk management and land use planning.

According to the International Levee Handbook (CIRIA MEDDE USACE, 2013), a levee system is defined as “a set of structures or protective elements, presenting an overall hydraulic coherence to ensure the effective protection of a previously identified group of stakes”.

However, this definition does not explicitly include natural formations, the essential role of which must be stressed in protecting against coastal flooding (IPCC, 2019). Moreover, the term “levee system” refers only to one of the two essential functions of protection systems; that of containing water by obstructing its passage. But a protection system must perform a second function; that of draining the water that accumulates inside the system by facilitating its flow outwards. Therefore, a broader and more explicit definition will be adopted for coastal flood protection systems, namely:

A set of anthropogenic structures and natural formations, presenting a global coherence from the hydraulic point of view to ensure the effective protection of a previously identified group of stakes against coastal flooding of marine, fluvial, pluvial, or phreatic origin.

Figure 5 shows schematically how these elements can be positioned in a coastal zone.

Anthropogenic or natural structures with a recognized protective role generally perform this function only in relation to other structures (e.g., salt marshes, mangroves and other wetlands). Mangroves, salt marshes, and coral reefs occur along about 40–50% of the world’s coastlines (Burke et al., 2011; Giri et al., 2011; McOwen et al., 2017; IPCC, 2019). These ecosystems provide important ecosystemic services including coastal protection by wave attenuation and shoreline stabilisation. A global meta-analysis of 69 studies demonstrated that, on average, these ecosystems together reduced wave heights between 35 and 71% at the delimited locations considered (Narayan et al., 2016), with coral reefs, salt marshes, mangroves, and seagrass/kelp beds reducing wave heights by 54–81%, 62–79%, 25–37% and 25–45% respectively. Moreover, sandy beaches and dunes occur at all latitudes covering approximately 34% of the world’s coastlines (Hardisty, 1994). The protection offered by these natural formations can be considered close to 100% for a robust dune belt in the absence of a major disturbance disrupting the hydrosedimentary cell functioning.

Protection systems are therefore an integrated set of technical and natural elements, which do not provide protection separately but in combination. Moreover, the regular action of managers of natural or anthropogenic structures (as well as other actors involved for example in regulation or financing) is a necessary condition for the proper functioning of these systems. Protection systems can therefore be considered as socio-ecological systems (SES). Such a system is defined by the IPCC (2019) as:

An integrated system that includes human societies and ecosystems, in which humans are part of nature. The functions of such a system arise from the interactions and interdependence of the social and ecological subsystems. The system’s structure is characterised by reciprocal feedbacks, emphasising that humans must be seen as a part of, not apart from, nature (Berkes and Folke, 1998; Arctic Council, 2016).

According to (Millenn. Ecosyst. Assess., 2005), SESs produce a « bundle » of ecosystem services (ES), categorized as provisioning (e.g., freshwater, crops, and meat), regulating (e.g., flood and climate regulation), and cultural services (e.g., recreation and spiritual values).

SESs should be considered as Complex Adaptive Systems (CAS): a system of interconnected components characterised by emergent behavior, self-organisation, adaptation, and substantial uncertainties about system behavior (Biggs et al., 2012).

Building on this critical idea, Bigg et al. (2012) defines resilience as the capacity of an SES to sustain a desired set of ecosystem services in the face of disturbance and ongoing evolution and change.

Studying coastal flood protection systems requires reframing them within the context of more comprehensive systems, considering not only the events against which they provide protection, but also changes and disturbances, slow or rapid, localized or diffuse, to which they are exposed, and more generally, all the constraints arising from the natural, the built and the social environment, at the level of the territory or at higher levels (such as the region or the Earth). In summary, these elements can be described as follows.

In the current day and age, the coast is a space which is both particularly exposed to natural phenomena and yet remains attractive to people. It is also subject to the spread of urbanisation (Lichter et al., 2011; Jonkman et al., 2013; Güneralp et al., 2015; Neumann et al., 2015; Muis et al., 2016) in the context of increasing international trade and globalisation. These changes may lead to situations of local concern in terms of fitness for habitation and exposure to natural disasters (Hallegatte et al., 2013; Pycroft et al., 2016; Tiggeloven et al., 2020), in particular engendered by marine submersions.

Marine submersions are caused by storms or hurricanes (Harris, 1963; Graumann, 2006; Heurtefeux et al., 2007; Dietrich et al., 2010; Bertin et al., 2012; Mazas and Kergadallan, 2014; Emanuel, 2017; Emerton et al., 2020) or by earthquakes or other underwater terrain movements causing tsunamis (Margaritondo, 2005; Intergovernmental Oceanographic Commission, 2012; Archana and Twinkle, 2021; Schreurs, 2021). These events have been documented in the history of humanity as the origin of many major disasters, with arguably the deadliest events having been caused by phenomena of geophysical origin (Flather, 2001).

The phenomena at the origin of marine submersions are complex in both their mode of generation (weather-marine phenomena or tsunamis) and in their propagation to the coast, including their last phase of interaction with the coasts. The Figure 6 summarises the hydraulic processes in relation to the morphological evolutions that take place on the time scale of an event or on longer time scales.

The risk of coastal flooding is also generated by other phenomena that occur separately or jointly, such as the overflow of rivers, the accumulation of rainwater, and a rising water table. This situation requires special arrangements to be made in the design of protection systems: drainage of water in vulnerable areas must complement the defences against sea and river waters.

Risks on most coastlines will be greatly aggravated by sea level variations. In particular, rising sea levels, by increasing water depths, promote the propagation of waves to the coast and contribute greatly to the increase in the frequency of extreme events as well as to the weakening of coastlines, especially by erosion. Sea level variations occur for the following reasons, summarized by the IPCC (2019):

• Changes in ocean volume as a result of a change in the mass of water in the ocean (e.g., due to melting of glaciers and ice sheets);

These processes are represented schematically in the Figure 7.

Among the causes of sea level change identified by the IPCC (2019), it is possible to distinguish between changes in the masses or volume of the oceans, and changes of the land with respect to the sea surface. In the first case, a sea level change is defined as « eustatic »; otherwise, it is defined as « relative » (Rovere et al., 2016).

This leads to two essential definitions (IPCC, 2019):

• The Global Mean Sea Level (GMSL) can be defined as an average of the eustatic sea level at a global scale. The following definition can therefore be adopted:

Mean distance between the sea surface and the center of the Earth.

• The Relative Sea Level (RSL) can be defined as the following:

Sea level measured by a tide gauge with respect to the land upon which it is situated.

Climate change effects (described in the IPCC 2007; IPCC, 2012; IPCC, 2014; IPCC, 2019; IPCC, 2021), and in particular the RSL rise, is on most coastlines causing a worsening of coastal risks (not only does this include the risk of coastal flooding, but also the risks of erosion, shoreline retreat, and dune migration). However, at present, attributing impacts to sea level rise remains difficult in most regions (IPCC, 2019), since impacts were exacerbated by human-induced non-climatic drivers, such as land use change, land subsidence caused by groundwater and/or hydrocarbon extraction (Dixon et al., 2006; Nicholls et al., 2008; Erkens et al., 2015; Sarah and Soebowo, 2018; Tessler et al., 2018), disturbance of sedimentary movement in rivers and along coasts (Syvitski and Saito, 2007; Syvitski, 2008; Vinh et al., 2014; Tessler et al., 2015; Welch et al., 2017; Dunn et al., 2018; Ouillon, 2018; Tessler et al., 2018; Thi Ha et al., 2018), coastal erosion (Doody 2013; Pontee, 2013; Luijendijk et al., 2018; Ndour et al., 2018) and coastal habitat degradation (Dahl and Stedman, 2013; Mediterranean Wetlands Observatory, 2014; Murray et al., 2014; Gardner et al., 2015; Kummu et al., 2016; Ramsar, 2016; Li et al., 2018; Sippo et al., 2018).

In the coming decades, the frequency of extreme sea levels will vary significantly across coastlines (Buchanan et al., 2016). In particular, many coastal areas in the lower latitudes may expect amplification factors of 100 or larger by mid-century, regardless of the scenario. By the end of the century and in particular under RCP8.5, such amplification factors are widespread along the global coastlines (Vousdoukas et al., 2018).

Depending on the coastline, the risk of coastal flooding can vary. On a global scale, there are three cases of particular concern (IPCC, 2019):

• Deltas: the long-term sustainability of populated deltas is often more affected by large-scale engineering projects than by sea-level rise. For deltas, accelerated soil compaction associated with petroleum and groundwater mining can exceed natural subsidence rates by an order of magnitude. Consequences include shoreline erosion, threatened mangrove swamps and wetlands, increased salinisation of cultivated land, and hundreds of millions of humans that are put at risk;

In order to favour systemic approaches, it is important to consider that a protection system belongs to a larger system. For this reason, the system of protection will be considered as a component of the territory to which it provides protection, making it possible to pay particular attention to the methods dealing with the relations between this system and other elements of the territory. The bibliographical analysis will therefore cross-examine the points of view of specialists in protection structures and specialists in land use planning.

With regard to land planning, account will be taken of the fact that urban systems contain assets of high value and complex and interdependent infrastructure networks (e.g., power supplies, communications, water, transport, etc.). The infrastructure networks are critical for the continuity of economic activities as well as for the population’s basic living needs. The availability of these resources is also required for fast and effective recovery after flood disasters. Damage to critical infrastructure assets during flooding can result in significant secondary consequences which may be more serious than the direct damage caused by the flood. For example, the destructive power and widespread disruption to infrastructure caused by Sandy in 2012 to the east coast of North America and the Caribbean has led this hurricane to be ranked as one of the costliest storm events for insurers (FloodProBE, 2013).

At the shoreline, marine floodings are generated by tsunamis (the effects of the earthquake may also be felt at the submersion site) or occurs during the passage of atmospheric depressions, which are often associated with high winds and waves breaking inland of the intertidal zone. If interest is first on the coastal hydraulic phenomena, the hazard may also occur, not only by overflowing and overtopping from the sea, but also through coastal river flooding, runoff, rising water tables … However, the flooding can generate other phenomena: humid ground or powerful currents that may cause mudslides, tree falls, cliff failures, cavity collapses, or erosion impacting building foundations or local infrastructure … The adoption of a systemic approach to risk in the context of observed and projected global changes is required to comprehend, as openly as possible, the conditions for the emergence of risk. In this type of approach, the hazard should no longer be considered in a restricted manner resulting only from hydro-meteorological phenomena, but rather as emerging from a combination of phenomena, potentially of various natures which develop across a range of timescales. The definition of « hazard » in the Hyōgo Framework for Action 2005–2015 (UNISDR, 2015) is therefore more appropriate:

« A potentially damaging physical event, phenomenon or human activity that may cause the loss of life or injury, property damage, social and economic disruption, or environmental degradation. Hazards can include latent conditions that may represent future threats and can have different origins: natural (geological, hydrometeorological and biological) or induced by human processes (environmental degradation and technological hazards) ».

To study the hazard implies first analyzing the different physical components, their possible combinations and the chain of events. In a global approach, Gill and Malamud (2014) present a broad overview of the interaction relationships between 21 natural hazards. The Figure 8 represents a selection of 16 of them, drawn from six hazard groups: geophysical, hydrological, shallow Earth, atmospheric, biophysical, and space hazards. As a necessary complement, Liu et al. (2016) provide a systematic hazard interaction classification based on the geophysical environment that allows for the consideration of all possible interactions (independent, mutual exclusion, parallel and series) between different hazards, and for the calculation of the probability and magnitude of multiple interacting natural hazards occurring together.

Increasingly, multi-hazard risk assessments are undertaken at the coast (IPCC, 2019). For example, Kunte et al. (2014) characterised the vulnerabiliy of coastal lands in India to erosion, submersion, and inundation, and demonstrated that the inclusion of socio-economic parameters should be considered in the overall assessment of vulnerability. Van Hattum et al. (2020) also stressed the importance of socio-economic parameters and noted that poverty in particular magnifies the impact of floods as poor people are often highly exposed to floods, and are also more vulnerable and less resilient.

To comprehensively understand risk, Hagenlocher et al. (2018) have come to the conclusion that the definition of adaptation strategies and measures (including ecosystem-based options) requires spatially explicit information on the exposure, vulnerabilities, and risks associated with different combinations of hazards in an integrative manner, not only focusing on societal aspects or ecosystems alone, but also taking into account interactions with other systems, such as neighbouring territories. In particular, ecological dimension should be considered more systematically in risk assessments. Addressing this gap, the authors presented an innovative assessment concept and indicator-based methodology that is sensitive to the specific (multi-)hazard setting in a given territory (Figure 9). These modular sets of vulnerability indicators adapt flexibly to the hazard situation.

It should be noted, however, that this methodology is based on quantitative indicators and cannot report cascading effects and feedbacks. In addition, it ignores the hazards of biological origin, for example, the likelihood that floods degrade health conditions, in particular, by the deterioration of drinking water and sanitation systems. Analyses which widen the spectrum of investigations to include the emergence of risks should therefore be considered, but these analyses are conditional on their application in a smaller geographical area, allowing better consideration of local specificities. Thus, Cerema (2019) established, for the city of Semarang (Indonesia), a more open model of systemic functioning, with emphasis on the importance of domino effects and feedbacks.

Finally, it should be stressed that the multi-risk approach should not only apply at the level of the territories, but should also apply directly to protective structures since they are sensitive to a wide range of factors: meteorological, hydraulic, and morphological actions, as well as biological and anthropogenic actions. For example, the effects of prolonged periods of drought should be carefully considered, particularly in the context of climate change: levees are vulnerable to drought if the structure or its subsoil contains peat or highly organic clay in a zone/level below which the groundwater table may fall during drought. Dehydration of peat in levees causes shrinkage and a further loss of weight causes reduced stability. This may ultimately result in breaching due to horizontal sliding (CIRIA MEDDE USACE, 2013). An example of failure due to sliding occurred in Wilnis (Supplementary Figure S1), the Netherlands, in 2003. Bottema et al. (2020) described in detail the drought effects and drought management practices for Dutch and English levees.

Whether one considers several phenomena in a risk analysis, or after preliminary investigations, one chooses to focus on one of them, it is always necessary:

• Spatially, to limit the territory in which the study is being carried out and within that territory, if necessary, carry out new spatial subdivisions (in particular if analytical approaches are to be considered);

This framing can progress only by iteration with investigations about the context and the phenomena likely to contribute to the manifestation of risk.

Regarding the risk of urban flooding, considering that this hazard cannot be managed in isolation at the city scale, and responses to potential flood impacts are complicated by interlinked political, socio-economic, and environmental changes, Zevenbergen et al. (2008) suggested to develop a framework in which spatio-temporal relations are further defined and investigated, in order to « provide clarity regarding both the feedback loops that cause vulnerability, as well as those that build resilience, and how they interact across differing spatial scales ».

To achieve this objective, cities can be considered as an intermediate-level SES between a higher-level SES (in this example, the watershed) and lower-level SESs, (buildings, infrastructure, and networks). The Figure 10 is a schematic depiction of the propagation of a flood wave through a territory in the case of a failure or overtopping of the protection system. In addition to its particular role in flood risk management, a coastal flood protection system could be considered in this scheme in the same category as buildings and other networks. A protection structure or a natural formation could therefore appear on the figure in place of a building.

For a « system of protection » SES, additional guidance is provided by research on:

• The assessment of levee performance and the development of geographic information systems for their maintenance (Serre et al., 2009; Michel, 2018);

Understanding how phenomena interact with one another, and their significance for people and ecosystems, requires a holistic approach that looks at human and natural dynamics together. The concepts of resilience and SES contribute to provide this holistic view (Arctic Council, 2016).

The key starting point of this process is the integration of different kinds of knowledge and a variety of experiences in which scientists, the public, and decision-makers in policy and practice collaborate to generate not only scientifically reliable, but also context-appropriate, socially robust and actionable knowledge (Weichselgartner and Kasperson, 2010), while always keeping in mind that knowledge integration is necessary but not sufficient: cooperation between actors must continue with resilience-building programmes (Weichselgartner and Kelman, 2014).

Increased interest in resilience management has driven the development of standardized tools for:

• Assessments that focus on deepening understanding of system dynamics; and

Measuring and monitoring a narrow set of indicators or reducing resilience to a single unit of measurement may block the deeper understanding of system dynamics needed to apply resilience thinking and inform management actions (Quinlan et al., 2016). This observation only reflects the aforementioned problem: analytical approaches may prove inappropriate to apprehend complex dynamic systems. Observation of these systems should necessarily involve systemic approaches. To manage the systemic approach of a theme (« risks to a coastal area » is a possible theme, for example) or a territory, an observatory may be a pertinent tool for providing information, but also for disseminating this information. An observatory can be defined as an observation system set up by one or more organisations to monitor the evolution of a phenomenon, an area or a portion of a territory in time and space; most observatories take the form of computer applications in which data are aggregated and presented in the form of tables, maps, or statistical indicators. The objectives, the users, and the temporalities considered by the observatory should be precisely defined (Bourlier et al., 2020).

Risk assessment methods at the level of a territory focus on urban environments, where flood protection stakes are most prevalent. The scale of the urban district is especially important because it is at this scale that cities carry out their adaptation, transformation and/or development operations (Serre et al., 2018). These approaches are selected for their overall operational character, as well as for the links established with flood protection systems. However, there are other models or approaches (Barroca et al., 2013; The Rockefeller Foundation, 2015) aimed at the analysis or assessment of urban resilience from a more general point of view.

The methods used to study an urban system consist of a double characterisation of its structural and functional aspects. The structural aspect corresponds to the spatial organisation of the system whereas through functional analysis, it is essential to characterise the temporal phenomena, such as flows, exchanges, and feedback. The understanding of risk at the scale of a territory can therefore be engaged by successively defining the limits of the system, its environment, its components, and finally by associating functions (Serre et al., 2018). In an urban system, this method shows the interdependence of the networks and critical infrastructures of the studied area, and hence, the domino effects and impact chains (Bourlier et al., 2020).

The Figure 11 shows the application of a functional analysis at the neighbourhood level. It indicates contact and normal operational flow relationships between a neighbourhood’s individual components and its external environment factors.

The predominance of the systems approach is marked by the focus of the study, which is mainly on the relationships between the elements, and avoids analysing the resilience of a territory through a sectoral approach based on specializations (e.g., flood protection, energy and transport). It should be noted that analytical approaches can be efficient only if the limitations of the system—determining what is internal and what is external—have been well delineated and the internal organisation of the system has been well described. Risk analysis can be continued by studying failure mechanisms for each function (Balsells et al., 2012). In applying these methods, the relations within the territory in question (the neighbourhood) and with the external environment of that territory should be well identified, particularly in regards to the structures performing a protection function. The interdependence plays the role of a risk diffusion factor. And as Lhomme et al. (2013) stated, networks then act as vectors of risk propagation. These analysis principles have been implemented by several authors, for example, by studying the effects of Hurricane Katrina (Heinzlef et al., 2019) and New Orleans’ recovery (Balsells et al., 2012) after this event.

Although for Serre et al. (2018), a neighbourhood is considered a social, economic, and physical/technical system, the above-mentioned authors generally placed more emphasis on the physical/technical dimension. In addition, other authors address:

• The human aspect of the management of the urban system. For example, Zevenbergen et al. (2015) presented a pragmatic screening procedure, referred to as a « Quick Scan methodology ». Its purpose is « to provide guidance for network operators and decision makers on identifying and rating those critical infrastructure networks and hot spot buildings that may be at risk from flooding, and assessing where intervention will be most feasible and cost beneficial—the so-called quick wins »;

The multiple natural and anthropogenic actions to which the levees are exposed, as well as the heterogeneity of the structures and their very long length, requires the consideration of significant margins of uncertainty of the structure’s behaviour. The protection provided by levee systems is therefore always limited, and must be considered as such: when natural phenomena exceed a certain intensity, the structures allow water to pass through, either by overflowing or wave overtopping, or through a breach in the event of an accidental rupture. In either case, the risk may increase, in particular by an acceleration of hydraulic phenomena. Since the positive effect of reducing flood frequency is more immediately apparent than its potential negative effects, levee systems often produce a false sense of security (CIRIA MEDDE USACE, 2013).

This has led to an extensive literature on:

• Modelling and assessing the risk of levee failure (e.g., Stanczak and Oumeraci, 2012; Wu and Li, 2017; Zhang et al., 2017). The effect of transitions, whether in the geometric design of the structure or in materials used, should not be overlooked in assessments as shown for example by van Bergeijk et al. (2019) and Simm et al. (2019);

Chapter 8 of the International Levee Handbook (CIRIA MEDDE USACE, 2013) reviews good engineering practices for structural dimensioning by combining hydraulic and geotechnical approaches. It can also be used to analyse the risk of failure.

Nevertheless, the problem of performance limits and the increase in risk in the case of an event exceeding the nominal operating conditions should be tackled within a broader framework than that of a single structure with only the protection function. On this topic, Marijnissen et al. (2021) assessed the safety of a double-levee system (Supplementary Figure S2).

In the more general case of a system with a front and a rear defense (e.g., a storm surge barrier and levees), the front defense can improve the reliability of the rear defense by reducing the load on this rear defense (Dupuits et al., 2017). Other studies also show the effectiveness of hybrid solutions combining land-based structure (for example a dike) with the effect of a natural formation such as coastal wetlands and mangroves (Vuik et al., 2018; Vuik et al., 2019). Similarly, the construction of brushwood dams is often a relevant solution: at exposed coasts, strong waves and currents may impede the settling of fine sediments and the establishment of salt marsh vegetation or mangroves (Winterwerp et al., 2013). Construction of a system with brushwood dams (Figure 12) creates shelter, facilitates sedimentation, and prevents erosion. Combining these dams with drainage ditches improves consolidation of the settled sediment. This method has successfully been applied for centuries, and has led to artificial salt marshes along 450 km of the Dutch, German, and Danish Wadden Sea coastline (Bakker et al., 2002; Hofstede, 2003).

Risk assessment approaches for protection systems are hybrid in nature. The diversity of natural structures and formations, as well as the diversity of the environments in which these components are placed, require the adoption of systemic approaches. However, the objectives of performance and rationalization of the resources employed, also force the adoption of analytical approaches. In particular, this is done in order to evaluate their reaction to various demands, or to determine quantitatively the human, financial, and material requirements for the management and adaptation of the protection system. Framing analytical approaches through systemic approaches is an ongoing challenge for managers of protection systems.

The structure performance assessment is crucial since this process provides input for the evaluation of the residual risk in a leveed area. In addition, the results of the evaluation of the structures will determine, along with the results of the risk attribution at the scale of the system, the priorities for action: if the risk attributed to a structure is high, then the requirements on its structural state will also be high, which will give it a better ranking in the intervention priorities of the levee manager. The main difficulty therefore lies in evaluating the performance of each component of the system. To meet this challenge, the concept of the fragility curve was explored. In the context of flood risk assessment, a fragility curve represents the probability of breach of the defence given a set of loading conditions and therefore represents (along with the probability of overflowing and wave overtopping) part of the likelihood of water flowing from source to receptor in the SPRC model. This analytical approach can be either deterministic or probabilistic, as shown in the Supplementary Figure S3. In the deterministic design, the assumption is made that the probability of failure is zero until the design load event is reached, at which point the probability switches to 1. In reality, as the load rises above the design condition, the probability of failure rises and only after considerable extra load has been applied does it actually approach 1. Similarly, there is a small but significant probability of failure at conditions below the design loading. This is what probabilistic approaches represent (Simm et al., 2008).

While these modes of representation are interesting for understanding the complexity of levee behaviour, the associated methods are difficult to apply due to the geotechnical heterogeneity, uncertainties about hydraulic loads, —even more so on coastlines exposed to waves—, and the multiple failure modes. In fact, since levees commonly extend over several tens or even hundreds of kilometres, their management can benefit from computer assistance in terms of data organisation, management, and representation (spatial representation and evolution over time), but will always require expertise for their diagnosis. For example, the system designed by Serre et al. (2009) has two components:

• A geographic information system (GIS) application that allows collecting data necessary for managing a levee portfolio;

This type of operating mode led Hathout et al. (2019) to question the validity of human judgment. These authors presented approaches of elicitation, calibration and aggregation of expert opinions to evaluate the reliability of river levees. Applying the proposed weighting and aggregation procedure reduces variability in the probability of failure estimate and provides a better probabilistic distribution of aggregated expert opinion than that of individual expert opinions. The results allowed for the identification of a trend of aggregated expert opinions indicating over or lack of confidence.

This review of the state-of-the-art shows that it is dangerous to seek « automation » of risk assessment at the level of a structure and, a fortiori, at the level of a protection system. The methods of evaluation which combine human expertise and the advantages of data management by computer means should indeed be sought. Since each structure is unique, monitoring its behaviour over time is essential, especially during major events through feedback. However, it should be considered in the application of this principle that in a future affected by multiple changes, decisions based solely on past events will no longer necessarily be relevant and that more anticipation will be required in the application of risk analysis methods and also in the definition of adaptation strategies.

The definition of strategies for the adaptation and a fortiori the transformation of SESs can be based on the establishment of scenarios. Scenario planning is a systemic method for thinking creatively about possible complex and uncertain futures. The central idea of scenario planning is to consider a variety of possible futures that include many of the important uncertainties in the system rather than to focus on the accurate prediction of a single outcome (Peterson et al., 2003). Scenarios can help overcome fundamental problems about transformation by organising thinking in a coherent way. With so many changes happening simultaneously, the very complexity of the situation can become a barrier to understanding and action. The scenarios can organise information about transformation in a comprehensible way that facilitates discussion and action (Walker et al., 2004).

The definition of adaptation strategies largely depends on the dynamic nature of the SESs, which requires considering several time horizons and assessing uncertainties. The study of several scenarios helps to achieve these two objectives. To this end, it is possible to use general data, although transposition across territories is often difficult:

• For climate change scenarios, the use of climate model projections (Moss, 2008; Moss et al., 2010; van Vuuren et al., 2011) is recommended. However, the further down the chain of effects, the larger the uncertainties about the degree of an effect (Haasnoot, 2013), and sometimes even about the direction. Moreover, when climate changes beyond the range historically observed, unforeseen shifts may occur which complicate the downscaling of global climate scenarios to the national or regional level (Klijn et al., 2015);

The use of scenarios may also involve analytical techniques, including:

• The policy tipping points (Kwadijk et al., 2010). A tipping point is reached if the magnitude of change is such that the current management strategy can no longer meet its objectives. Beyond the tipping point, an alternative adaptive strategy is needed. By applying this approach, the following basic questions of decision makers are answered: what are the first issues that we will face as a result of climate change and when can we expect this. In the west of the Netherlands, the results show, for instance, that climate change and the RSL rise are more likely to cause a threat to the fresh water supply than flooding; and

Addressing the major challenge of climate change adaptation and in order to take into account the Sustainable Development and Poverty Eradication Goals, it is also possible to use the concept of Climate-resilient development pathway (CRDP), represented by the Supplementary Figure S5 (Kainuma, 2018; Roy et al., 2018).

CRDPs are increasingly being explored as an approach for combining scientific assessments, stakeholder participation, and forward-looking development planning, acknowledging that pursuing CRDP is not only a technical challenge of risk management but also a social and political process (Roy et al., 2018). This implies an organisation of the process by the institutions in charge of the governance of the territories. CRDPs aim to establish narratives of hope and opportunity that can extend beyond risk reduction and coping—in particular integrating the fight against climate change (Amundsen, 2018)—and are critical in driving motivation, creative thinking and behavioural changes in response to climate change (Myers et al., 2012; Prescott and Logan, 2018). The definition of the scenarios should be based on general strategies for enhancing social-ecological resilience. These principles, on which CRDPs are based, were the subject of an integrative presentation by Bigg et al. (2012) on the basis of a review of the scientific literature (Supplementary Figure S6).

According to Bigg et al. (2012), the principles of resilience management fall into two levels:

• Principles that relate to generic SES properties to be managed; and

The implications for coastal flood protection systems of these principles are presented below. Issues related to the adaptation of governance are presented in this section. Issues related to adaptation strategies will be addressed in the next section.

A territory subject to coastal flooding should be considered a CAS and the issue of protection against this risk can, as such, be dealt with in conjunction with other issues including freshwater management. In the context of adaptation to the effects of climate change, this is particularly relevant when one of the main threats to coastal territories is the salinisation of freshwater bodies and soils. « Understanding an SES as a CAS » involves a paradigm shift. This paradigm shift in particular implies a change in the mode of governance. The Supplementary Figure S7 shows how the evolution of mental perceptions and representations on this subject of fresh water, through the transition from an analytical to a systemic way of thinking, may result in a change in governance and improved flood resilience. The standard perception of freshwater management leads to inflexible, and often ineffective, command-and-control approaches, whereas accounting for ecological complexity opens avenues to adaptability, which builds better resilience to events such as flooding (Moberg and Galaz, 2005). One such pathway may be to assess the role of wetlands in flood mitigation and to anticipate across several scales (including those of the region and of the territory) increases in capacity that may be required in response to climate change (e.g., changes in precipitation patterns and changes in thermal regulation by glaciers) and land-use changes (e.g., increases in runoff caused by deforestation and urbanization).

The framework established by Bigg et al. (2012) calls for governance that intends to bridge two approaches:

• Adaptive management, which focuses on “learning by doing” as a strategy to deal with uncertainty in CASs; and

These two approaches are often grouped under the same term « adaptive co-management » since the research of Ruitenbeek and Cartier (2001). Adaptive co-management combines « the iterative learning dimension of adaptive management and the linkage dimension of collaborative management in which rights and responsibilities are jointly shared » (Armitage et al., 2007).

The participatory nature of adaptive co-management enables sharing and reflecting on experiences, ideas, and values with others, which builds trust and relationships and facilitates social learning as well as collective action (Olsson et al., 2004). Participatory learning processes can help actors learn about each other’s mental models, which builds social capital, in turn supporting institutional change and conflict resolution (Biggs et al., 2011).

On « coastal territory » SESs, the utilisation of an expanded circle of actors can be sought in different ways, for example:

• In the definition of town planning rules, a consultation phase may involve all stakeholders in the decision-making process. Perherin (2017) described in detail how the debate between the actors can be organised around the production of hazard maps carried out as part of the elaboration of coastal risk prevention plans;

The evolution of modes of governance is a necessary condition for change when a profound transformation is envisaged. This is supported by three studies realised in China (Jiang et al., 2018), in Australia (Hurlimann et al., 2014) and in Italia (Vitale et al., 2020). In the three cases, it appears that the change in governance should also be accompanied by institutional changes, and administrative reorganisation. The governance implications of the adaptive co-management principles should therefore be well understood methodologically. Olsson et al. (2004) showed that the self-organising process of adaptive co-management development, facilitated by rules and incentives of higher levels, has the potential to expand desirable stability domains of a region and make SESs more resistant to change. The European Flood Risk Directive (Directive 2007/60/EC) is in line with this principle by encouraging the participation of local stakeholders in the development of flood risk management plans. Its application has indeed led to an institutional and administrative restructuring of governance, by developing more integrated visions of land use planning by combining the issues of risk prevention and management of aquatic environments. The refocusing of institutions on the geographical perimeters of emerging risk (e.g., river basins) is another notable advance (Cerema, 2020).

While a review of the modes of governance can be initiated by a top-down process as is the case for the implementation of the European directive, which can lead to real progress, this observation should not, however, obscure the fact that all regions of the world do not benefit from a national or supra-national organisation that encourages changes in the modes of governance. To initiate a change in governance, Termeer et al. (2017) proposed a methodological framework to clarify how governance can be improved through bottom-up initiatives and how this change can be initiated. In complement, Cundill and Fabricius (2010) identified system attributes and key variables that could form the basis for monitoring the governance dimension of adaptive co-management.

Building on previous research (Wong, 2014), the IPCC report (2019) describes, in addition to the option of no response, five main modes of adaptation to mean and extreme sea level rise: advance, protect, retreat, accommodate, and ecosystem-based adaptation. These options are shown in the Supplementary Figure S9.

The IPCC report (2019) in Chapter 4 provides a comprehensive analysis for each of these options based on six criteria: observed responses, projected responses, cost of responses, effectiveness of responses, co-benefits and drawbacks of responses, governance challenges (or barriers), and economic efficiency.

Nevertheless, the typology of adaptation options with reference to sea level rise is based on strong implicit assumptions that are not justified in many cases. Thus, the options described are only applicable if the following assumptions are, either completely or partially, verified:

• The coastal strip must have a significant slope, and therefore the possibility of reducing the exposure of the stakes by providing a zone for retreat at a short distance from the coast. The reality of territories is often quite different: the deltas are very flat areas (Syvitski and Saito, 2007) and many small islands do not have an area allowing for retreat;

The main interest of the approach focused on RSL rise is to provide spatio-temporal benchmarks in the definition of protection strategies. This parameter is also used in global studies to identify, under several scenarios, the exposed territories, the associated demographics, and economic issues in order to determine the benefits that could be expected from effective protection. Nevertheless, the results of these studies appear to be fragile when it comes to assessing the real possibilities for adaptation of territories and their protection systems. This fragility naturally increases over longer-term projections (Hallegatte et al., 2013; Tiggeloven et al., 2020).

As for the technical aspects, it is important to avoid approaching coastal flood protection from an economic perspective with overly simplified approaches. As coastal flood protection systems consist of a set of structures such as levees (sometimes layered in several rows), breakwaters, seawalls, which are possibly supplemented by storm barriers and natural features, the economic assessment should be based on:

• The length of the structures and not the length of the coastline. Where possible, consideration should also be given to the height of the structures as proposed by Lenk et al. (2017) on coastal dikes;

The studies of Lenk et al. (2017) and Aerts (2018) provided interesting theoretical contributions to cost assessment. However, the results obtained remain fragmented, and the uncertainties on unit costs are far too high for a construction manager to utilise for cost estimates for a particular project (a rapid estimate on the basis of the unit costs appearing in the current public work contracts would be more accurate). The main reason is that the initial data used for these studies comes from bibliographic reviews, with data of multiple international origins. In reality, estimates can only be made on the basis of costs observed currently in the given context (which therefore presupposes the availability of recent national or regional studies), because the availability of resources and the general economic context are decisive.

The international state-of-the-art on this issue should therefore progress, not by seeking cost estimates, but by supplementing and refining the methodology for observing costs on a territory.

In conclusion, decisions on a territory cannot be made by considering the various issues separately and by seeking solutions of only modifying the coastline, as might be suggested by the simplified schemes presenting adaptation options. On the contrary, the issues should be considered globally and should take into account the physical, human, and biological dynamics of the territory envisaged in its three spatial dimensions. The use of methods that make it possible to understand these developments is therefore necessary to give meaning to adaptation decision-making. Ultimately, if simplified approaches are not devoid of interest, they are analytical approaches that should be supplemented with a comprehensive systemic approach.