Introduction

Between 13 and 20 July 2021, the Dutch Province of Limburg faced an acute high-water event caused by two days of heavy rainfall with 160–180 mm of rain falling on 13 and 14 July (Appendix A, Table A3, E1). This event became known as the 2021 summer flood. The flood was extreme because of both the timing—a flood in the summer low-flow season—and the large amount of precipitation over a large area [1]. The flood was almost immediately connected to the underlying, more creeping crisis of climate change [2, 3]. In the Netherlands, the event caused €383 million worth of damage, and the flood affected over 4,400 citizens, but did not lead to casualties like in Belgium and Germany [4, 5].

After the event, multiple evaluations were organised and requested by various policy actors. The Dutch national government evaluated the crisis response, the effectiveness of physical flood risk measures, and programmes under execution. The Ministry of Infrastructure and Water Management (see Appendix A, Table A3, E7) set up a special governmental evaluation committee in 2022. This committee concluded in its final report that impacts of extreme events cannot be fully prevented, especially as climate change is expected to lead to extreme weather events becoming more frequent and intense.

For ensuring long-term climate robustness, it is important that the crisis evaluation studies conducted in the aftermath of a crisis do not only evaluate the effectiveness of crisis management responses and damage compensation, but also draw lessons to prepare for more extreme weather events due to climate change. Existing studies in the field of crisis and disaster management have since long been concerned with evaluating how well governments have been responding to crises e.g., [6,7,8]. There is also valuable literature on disaster preparedness e.g., [9,10,11], including literature on individual preparedness, risk perception and community resilience e.g., [12,13,14]. In addition, an increased awareness of climate change among scientists has stimulated scholars to develop adaptive water management approaches for dealing with floods. These studies tend to focus on collaborative learning and designing strategies with more flexibility to be able to accommodate future extremes e.g., [15,16,17].

However, to understand the combination of lessons drawn after crises for crisis management, disaster prevention, and the anticipation of future events, we need to take a closer look at the breadth and depth of the lessons drawn from shocks. The breadth of learning concerns the different ways of handling and preparing for flood events (such as, spatial planning measures and crisis response systems), whereas the depth of the learning is about whether lessons drawn involve more incremental up to more paradigmatic and transformative changes.

Therefore, this study aims to address this gap by analysing the depth and breadth of lessons drawn from an acute crisis to understand how these lessons contribute to becoming more climate robust in the future. We do so by combining the multilayer safety concept for water management with literature on policy learning and forward-looking policymaking. The concept of multilayer safety helps to understand the breadth of learning by distinguishing lessons about flood prevention, flood impact reduction, and crisis management [18]. Policy learning helps to analyse the depth and scope of learning by dividing learning into single-, double-, and triple-loop learning [19,20,21].

The guiding research question is: What long-term lessons for flood risk management have been drawn from an acute flood crisis?

Research sub-questions are used to present the results in a clear and structured way:

  1. 1.

    What were the flood protection policies before the 2021 summer flood and how do these policies connect to the different layers of flood safety?

  2. 2.

    What single-, double-, triple-loop lessons from the 2021 summer flood have been drawn for the different layers of flood safety?

  3. 3.

    To what extent are lessons drawn from the 2021 summer flood forward looking?

In the next section, we introduce this study’s main concepts and analytical framework. In Sect. "Methods", we describe our methods for data collection and analysis. Section "Results" presents the main results for each sub-question. We end with a discussion and conclusion section, in which we also reflect on the main implications for theory and practice.

Analytical framework

To be able to answer our research questions, we introduce an analytical framework consisting of policy learning (depth and scope of learning), various ways of managing flood risk (breadth of learning), and the extent to which lessons on flood safety are forward looking (do lessons anticipate the future?).

Policy learning

To understand learning after water crises, we use literature from policy studies that distinguishes between three types of policy learning: single-loop, double-loop, and triple-loop learning [19,20,21]. Although the concept of social learning has been criticised early on for lacking in conceptual clarity in comparison to other concepts [22], it is still a valuable body of work to inform our understanding of the three different types or degrees of learning.

Single-loop learning or incremental learning, also called first-order change, consists of improving the actions that are part of the usual repertoire. In the social learning literature, this is understood as ‘incremental improvement of action strategies and daily routines without questioning the underlying assumptions’ [20]. Single-loop learning is associated with incremental change, optimising existing solutions and routinised decision-making [21].

Double-loop learning [23] or second-order change, involves revisiting assumptions and rethinking the frames that underlie the actions taken. In the social learning literature, it is defined as ‘revisiting … assumptions (e.g., about cause–effect relationships) within a value-normative framework’ [20]. In double-loop learning, policies and policy instruments are adjusted but remain within the existing policy paradigm [21]. Compared with single-loop learning, double-loop learning is more in-depth, because it challenges existing frames of reference, and broader in scope, because policies and policy instruments are adjusted [24].

Triple-loop learning, or third-order change, is learning for transformational change by rethinking the systemic context in which frames for actions are formed. In the social learning literature, it is understood as requiring ‘re-examining the underlying ideological and value system’ [20]. Triple-loop learning involves a paradigm shift with radical changes in the underlying policy discourse [21]. Triple-loop learning is very difficult and likely to entail policy failure, disagreements among experts, and conflict over authority. With its transformational ambitions, triple-loop learning is both the most in-depth form of learning, because it is about radical change, and the form of learning with the widest scope, because it implies system-wide change [25].

Learning about multiple layers for flood safety

To understand the lessons drawn about flood safety more specifically, we use the multilayer safety concept to categorise the lessons in policy evaluations after the 2021 summer flood in Limburg.

Different layers can be distinguished as part of multilayer safety [18, 26]. These layers consist of a combination of water management approaches to deal with floods: flood prevention with defensive measures (layer 1), flood impact reduction by applying spatial planning measures (layer 2), and crisis and disaster management measures (layer 3). The concept was introduced in 2009 in the Netherlands with the National Water Plan (Appendix A, Table A1, P22) and fits the European Union’s Flood Directive (2007/60/EC).

Multilayer safety focuses on both threats and opportunities. Types of possible measures within each layer include:

  • Flood prevention (layer 1): reinforcing existing dikes and installing new (types of) dikes and barriers to reduce the failure of existing flood protection measures.

  • Spatial planning (layer 2): creating areas for water retention and storage, emergency inundation areas, preventing building in high-risk areas, constructing elevated buildings, protecting critical infrastructure. We treat all references with regard to creating room for the river as part of layer 2, because this leads mostly to changes in functions and has objectives other than flood safety objectives only (such as spatial quality).

  • Crisis management (layer 3): improving crisis and evacuation plans and training, developing evacuation routes and shelters, crisis communication.

Research and policy reports show that layer 1—the defensive flood prevention approach—still prevails, as it is seen as most effective according to general public perception [18, 27].

Elements of forward-looking learning

To be able to understand the extent to which the policy lessons about the different levels of multilayer safety will help to improve long-term climate resilience and also anticipate potential future crises instead of focusing solely on evaluating the crisis response to the 2021 summer flood, we want to analyse the extent to which policy learning after the flood was forward looking.

Pot et al. [28] developed a framework for understanding and assessing forward-looking decisions about water infrastructure and water management. In this article, we apply this framework, for the first time, to understand and analyse the extent to which lessons drawn from an acute crisis qualify as forward looking. With that, we use the framework for a different step in the policy process than the framework has been designed for: it was designed for decision-making, and we now apply it to evaluation [29].

Three elements help to assess the extent to which policy lessons can be characterised as forward looking:

  1. 1.

    Consideration of long-term and future developments in crisis evaluations: does the evaluation go beyond the extreme weather and acute crisis of the flood event by looking into the future and discussing the underlying creeping crisis of climate change [30] and other long-term crises, e.g., related to planetary boundaries [31]? The exact meaning of long term differs per actor and policy problem and is therefore left open. This article does differentiate between looking backwards and formulating actions purely related to the crisis response of the 2021 summer flood versus looking forward and formulating lessons for flood safety to prepare for future events. Lessons analysed do not involve the crisis response itself, but lessons that anticipate climate change and other developments to cope with future events [32].

  2. 2.

    Recommendation of robust and/or flexible solutions: robustness means choosing solutions that can withstand shocks and remain functional under a broad range of possible future circumstances, including hydroclimatic extremes. To test for robustness, qualitative and quantitative stress tests with one or a large number of possible extreme weather events can be especially helpful to assess whether the water management system can cope with extreme circumstances [33, 34]. Flexibility means having a plan to change solutions over time when conditions change or new insights emerge [35, 36]. Flexibility also implies experimentation with monitoring and evaluating implemented solutions to adjust them if circumstances change [37].

  3. 3.

    Anticipation of different possible futures: do the evaluation studies explicitly promote the development or use of long-term visions and vision-making exercises to develop more desired futures for the regional water management and governance system (backcasting) and/or recommend scenario (forecasting) approaches to explore multiple possible future circumstances [38, 39]?

Methods

This section describes the case of the 2021 flood event in the Netherlands and then discusses how we collected and analysed crisis evaluation studies and interviews.

Case description

The summer floods of 2021, affecting many areas in western and central Europe, formed the focal event for this study [1]. The geographical area central to the research is the lower part of the Meuse basin, located in the Dutch Province of Limburg. We focus on the Dutch part of the Meuse because the Netherlands is a frontrunner in water management and especially flood risk management, and did not see casualties after the flood. Moreover, the flood did mainly occur in the Meuse tributaries, as opposed to earlier floods of the river Meuse. For other countries, it could be interesting to understand how the country evaluates the shock and tries to build further on their approaches such as room for the river [15], especially in the face of climate change.

In Limburg, the geographical boundaries of the province and the regional water authority coincide. The Meuse is a rain-fed river, originating in northern France, with seasonally fluctuating discharges depending on precipitation and evapotranspiration [40]. The Meuse responds quickly to precipitation, making the basin sensitive to flooding and drought [41]. To illustrate, extremes in discharge range from 3300 m3/s during flash floods (summer flood of 2021) to effectively no discharge (25 m3/s) [42].

Various public organisations at all governmental levels are involved in governing the Meuse and its tributaries. The Meuse itself is categorised as a primary water system, meaning that it is governed through national programmes and institutions. The National Water Authority (Rijkswaterstaat) is responsible for maintenance of the river’s primary water system [43] in terms of the protection levels determined by the Ministry of Infrastructure and Water Management. Long-term flood safety is also one of the Delta Decisions within the Dutch Delta Programme (Appendix A, Table A1, P30). The Delta Programme is institutionalised in a Delta Act (as part of the Water Act) and led by the Delta Commissioner, whose task is to draft and direct the Delta Programme and to inform the Minister of Infrastructure and Water Management on its progress. The Water Act also provides for a Delta Fund to finance the Delta Programme’s activities (see Article 7.22 of the Dutch Water Act).

Flood safety measures are implemented through the Hoogwaterbeschermingsprogramma (HWBP) (Flood Protection Programme), a programme coordinated by the Delta Programme and executed by the 21 regional water authorities and the National Water Authority. Primary dike enforcements fall under the HWBP. The HWBP also relates to dikes categorised as part of the secondary or regional water system, such as tributaries of the Meuse. The management of these regional water systems and HWBP execution for secondary or regional water systems fall under the responsibility of regional water authorities. However, it is the province that decides on the flood protection level for non-primary dikes. Lastly, for local waters, also managed by the regional water authorities, there is no relevant national safety classification.

Data collection

The data collected consisted of policy documents (P), interviews (I), and evaluation studies (E) (see Appendix A, Tables A1, A2, and A3).

For research sub-question 1 policy documents (n = 42) from 1975 to 2021 were collected to create an historical overview of the policy programmes and physical measures implemented in the case study area before the 2021 flood. These documents were retrieved from public databases such as: officiëlebekendmakingen.nl (for national-level developments on flood policy), the National Water Authority’s website (for specific information on the Meuse Works), and iBabs.nl (for agendas and meeting notes of provinces and regional authorities). Furthermore, scientific publications on Dutch flood risk governance e.g., [44, 45] served as the starting point for the identification of the most significant historical developments, informing further data collection. By way of a member check, the historical overview of previous and existing flood-related policy programmes and the list of evaluation studies were shared with three interviewees to verify correctness and completeness. The results were confirmed, and the member-check contributions were processed where applicable.

For research sub-questions 2 and 3, evaluation studies were collected, and interviews were conducted. Evaluation studies (n = 11) were collected to analyse how the crisis was evaluated, the official lessons drawn for long-term climate resilience soon after the acute shock, and the extent to which such lessons are forward looking. The evaluations, produced between September 2021 and December 2022, were collected primarily through an open search using several keywords, including evaluation, high water, summer flood, and crisis, and were verified with the use of interviews to ensure the inclusion of all evaluations in the aftermath of the 2021 summer flood for the Limburg area.

Interviews (n = 7) were held between February and May 2022 with representatives of all relevant public organisations with formal responsibilities for flood risk management (see Appendix A, Table A2). These organisations include the Delta Programme, the National Water Authority (Rijkswaterstaat), the Ministry of Infrastructure and Water Management, Limburg Province, and the Limburg Regional Water Authority. The interviewees were selected based on their connection to the distinct geographical case study area (Limburg), on their topical expertise (flood risk management), and their involvement around the event (summer 2021). This resulted in a clear group of potential interviewees. The interviews aimed to explore how the actors viewed the impact of the crisis on policy agendas and existing flood governance structures. Therefore, reoccurring questions revolved around how the interviewee would reflect on the summer flood from the perspective of their organisation and for flood risk management in general, what kind of interventions or solutions would be needed (ranging from technical to administrative and political), potential for doing this integrally, and which implications this could have for future flood risk governance. Moreover, we explored the type of forward-looking lessons drawn by interviewees.

Data analysis

For research sub-question 1, the document analysis served as a basis to create an historical overview of the management and governance developments regarding flood safety in the Netherlands and specifically for the Meuse basin in Limburg in the recent past. Through a qualitative analysis, the documents were categorised in terms of year and main contents in relation to flood risk management, creating an overview of the presence of safety layers 1, 2, and 3 in these policy documents over time.

For research sub-questions 2 and 3, evaluation studies and interview data (consisting of interview notes and transcripts) were coded in Atlas.ti (version 23), a programme used for qualitative data analysis. The coding procedure for both datasets involved a combination of deductive theory-based coding for the depth and scope of learning and forward-looking elements and inductive, topic-based coding for multilayer safety (see Appendix C). It is important to note that we inductively added two layers to the multilayer safety concept in our data analysis after reviewing the data. The extension of the concept from three to five layers after the 2021 summer flood is a policy lesson in itself, yet we also wanted to understand the extent to which these new layers were already addressed in more detail as part of the policy evaluations. The layers that were added are layer 0 (water awareness) and layer 4 (climate-robust recovery). For sub-question 2, all elements of the evaluation studies and the interviews that dealt with the subject of multilayer safety were identified and inductively coded by topic. Coding for multilayer safety aspects continued until no new codes emerged from the data, after which the codes were reorganised by topic and merged where possible. Hereafter, the material was revised using the new codes. The lessons about multilayer safety were then analysed based on the depth and scope of policy learning, by labelling lessons as single-, double-, or triple-loop learning. This was done deductively following the theory on the types of policy learning. Moreover, our understanding of Dutch flood risk management policy developments prior to the 2021 summer flood allowed us to categorise a lesson as single-, double-, or triple-loop through expert judgement.

For sub-question 3, to analyse the extent to which the lessons on multilayer safety were also forward looking, the theoretical framework with the described characteristics of forward-looking learning was applied, and deductive codes were derived from this framework. A final analytical step comprised highlighting the co-occurrence of levels of policy learning and forward-looking characteristics, providing insight into the extent to which lessons drawn from the 2021 summer flood in Limburg were forward looking.

Results

Development of flood protection policy in the Netherlands

In this section, we answer sub-question 1: What were the flood protection policies before the 2021 summer flood and how do these policies connect to the different layers of flood safety? The 2021 summer flood in Limburg central to this study was the first major flooding event after the floods of 1993 and 1995, which affected this same region. In this section, we provide an overview of main flood protection policies until 2021 and how they address the different layers of flood safety. A more detailed description of major events and the content of the policy programmes in Dutch water management can be found in Appendix B.

Choices in water management: flood prevention or spatial adaptation?

The start of the Delta Works, an immense coastal protection project initiated after the North Sea floods of 1953, led to a reflection on the strength and safety of the Dutch river dikes (Appendix A, Table A1, P1). Actions to reinforce part of these dikes were taken in the years that followed. However, with these dike reinforcements, the cultural and historical landscapes were greatly impacted, leading increasingly to societal turmoil (P1). In 1975, the Minister of Transport and Water Management commissioned a study to review and, if necessary, revise the river-dike safety standards in the Netherlands, thereby exploring the consideration of flood safety against other, societal costs and discussing possible alternatives to traditional dike reinforcement (P1). This study put riverine flood risk management high on the national agenda. The advice as issued in 1977 on securing safety standards of 1/1250 was repeated in 1992, stating that riverbed widening is no feasible alternative to having dike reinforcements because of its great implications regarding alterations to the landscape (P7).

However, continuous soil subsidence and changing climatic conditions would require dikes to become ever stronger in the future. As this was recognised as an impossible challenge, answers were sought in a combination of technology and spatial planning endeavours (P3). Approaches involving spatial planning solutions were gaining ground and became more concrete when the winning idea of a national future river landscape design competition presented a combination of flood safety together with nature and agricultural functions (P4) (layer 2). This plan inspired the later Room for the River policy guideline (P11).

Flood risk management in Limburg, an enduring policy discussion

Spatial planning solutions for flood safety gained traction after the Meuse floods of 1993 and 1995. In December 1993, after a period of persistent rainfall, the Meuse flooded. Approximately 8000 people were evacuated, leaving 250 million Dutch guilders’ worth of damage (P8). A Special Committee on the Meuse Flood was installed by the national government in 1994, marking the start of Limburg’s inclusion in the national planning for, and financing of, flood protection measures. The Committee argued that riverbed widening should be implemented where possible (layer 2) and dikes constructed (layer 1) where widening alone would not be sufficient to reach a safety standard of 1/250 (P8). Riverbed widening, alongside nature development, would especially take place along the Grensmaas. Its activities were financed via an innovative collaboration: a consortium of public and private partners would extract gravel in return for nature development afterwards (P9).

Soon after, in January 1995, the Meuse basin was affected again by a big flooding event, also occurring in the rivers Rhine, Waal, and IJssel. In reaction to these high water levels, the government started the Delta Plan Major Rivers (Deltaplan Grote Rivieren) to accelerate dike reinforcements and to speed up the implementation of the advised measures along the unbanked areas of the Meuse (P10). For Limburg, it meant the construction of emergency dikes (safety standard of 1/50) in the absence of embankments to quickly achieve a higher security level and to simultaneously start riverbed widening (P10). At the same time, the Ministry of Transport and Water and the Ministry of Housing, Spatial Planning, and Environment stated that dike reinforcements are not sustainable in the long term. The emergency dikes were needed in the short term, but the goal was to make them redundant as a result of spatial planning measures (layer 2) (P11). This 1996 policy guideline became more concrete with the Spatial Planning Key Decision on the Room for the River Programme in 2006, describing measures to be taken between 2006 and 2015 (P18). The Room for the River Programme was designed for the Rhine and the downstream stretch of the Meuse, and thus not as such implemented in Limburg Province. A similar programme, yet following a different financial model, was initiated in 1997 for the Meuse in Limburg—the Meuse Works—split into three sub-programmes: the Meuse Route, Grensmaas, and Zandmaas (P12). The Meuse Works aimed to create more room for the river and to develop nature areas, as well as to reinforce the emergency dikes of 1995 (P12).

Although the Meuse Works was celebrated for its functioning during the most recent Meuse floods in 2021, the programme was also critiqued. Limburg’s distinctive landscape characteristics make river widening projects and dike reinforcements societally challenging. The national discussions of 1975/1977 on safety standards affecting natural and cultural-historical values are still lively in Limburg, questioning how to balance landscape conservation with flood safety. Limburg Province’s request to reconsider the safety standards for 22 of 45 dike trajectories is exemplary (P39). The question was raised as to whether the current standard of 1:100, the national minimum standard for primary dikes, is not too stringent for these 22 trajectories. Limburg Province argued that, considering evacuation fractions (layer 3) and the assumed damage caused by flooding, a standard of 1:30 would be sufficient for these 22 trajectories, as meeting a 1:100 standard would greatly impact the landscape, accompanied by high monetary costs (P39). However, the Delta Commissioner advised adherence to the 1:100 norm, pointing out that a debate on the height of dikes turned into a debate on safety standards and that there were alternative, innovative ways to meet the standards (layer 1) (P42). The urgency of a well-functioning flood safety system was underscored only a month after the Delta Commissioner’s advice, with the Meuse summer floods of 2021.

Multilayer safety in policy developments

Following from the above historical overview of the developments around riverine flood safety and safety standards, it is clear that traditional flood prevention in the form of dikes (layer 1) alternate and co-develop with spatial adaptation for flood safety (layer 2). On the other hand, layer 3, crisis management, is under-represented. According to Bosoni et al. [18], this could be explained by a limited integration of crisis management with the other two layers. Figure 1 shows how the various layers of flood safety were covered in water management policies between 1975 and 2021.

Fig. 1
figure 1

Historical overview of the presence of layer 1: flood prevention (in blue), layer 2: spatial adaptation (in green), and layer 3: crisis management (in red). A light-shaded box indicates the presence of one policy document that emphasises the use of the respective layer; a dark-shaded box indicates the presence of two or more policy documents that prioritise the use of that layer. The vertical lines indicate an extreme event (yellow for droughts in 1976 and 2018, red for floods in 1993, 1995, and 2021)

Full size image

Policy learning for multilayer safety

This section answers research sub-question 2: What single-, double-, triple-loop lessons from the 2021 summer flood have been drawn for the different levels of flood safety? To answer this, we analysed the depth and scope of lessons drawn from the 2021 summer flood about the multiple layers of flood safety. The analysis revealed lessons about adding layers and integrating the multiple layers of flood safety. We first discuss these and then discuss the lessons drawn for each layer specifically.

Lessons for integrating safety layers in flood risk management

The use of the multilayer safety concept itself has developed in two ways as a result of the 2021 summer flood: the first policy lesson was to add two more layers to this concept for improving long-term flood safety and the second lesson was to recommend the application of the multiple layers of safety also beyond the primary flood defense system.

First, in the final evaluation report of the 2021 summer flood, the following two layers were proposed as necessary to add to the existing layers for flood safety [46]:

  • Water awareness (which is indicated as layer 0, a sort of basis for the other layers to build on): means that citizens, companies, and governments are more aware of water risks and are able to take mitigation measures. An example is the development of a label for houses that indicates their vulnerability to water risks and flooding in particular.

  • Climate-robust recovery (layer 4): means that the recovery and restoration of damage after water-related crises should happen in a way that strengthens long-term robustness to deal with future floods. In other words, structural improvements are applied in the post-disaster phase. This is also referred to as building back better [47, 48].

Second, the flood showed that applying the multilayer safety concept to the primary water system is not sufficient. That is, the safety-layer approach should also be applied to the regional water system with the goal of working towards an integrated, systemic approach (Appendix A, Table A2, I7). This approach then also allows for the consideration of other challenges such as drought, heat stress, biodiversity, and nitrogen (Appendix A, Table A3, E6).

Different types of measures such as river widening and HWBP dike reinforcements are already often addressed integrally at project level. At system level, this is not yet the case. The Beleidstafel endorses the efforts of the Integrated River Management programme to secure and strengthen this integrality at system level. (E7)

Thus, translating the ambition of enlarging the multilayer safety concept to a systemic scale would also require constant attention on connecting the HWBP policy to river widening approaches, beyond single projects. In the following section, we will analyse the depth and breadth of policy learning for each of the five layers of the multilayer safety concept.

Lessons for different layers of safety

Table 1 shows the lessons drawn for each separate layer of flood safety. Triple-loop learning occurred only in layer 2—spatial adaptation—with the suggestion of a need for a systemic leap towards thinking differently about spatial planning in secondary river systems. For the other lessons, no triple-loop learning was identified.

Table 1 Overview of types of learning per safety layer
Full size table

We briefly elaborate upon some of the lessons for each flood safety layer to provide some more context.

In layer 0, the focus of the lessons is primarily on policy instruments needed to instil a common sense of water and climate risk in society, providing graspable information, i.e. through water labels as illustrated by the following quote.

Similar to an energy label, a water label can indicate the risk of water nuisance or flooding of a house. This allows citizens to make choices in this and also to prepare themselves properly in in case of an increased risk. (E7)

As the emergence of layer 0 in the multilayer safety model is a lesson in itself, it is evident that most lessons are classified as single-loop learning. Water awareness is considered a precondition for generating climate adaptive responses and therefore crucial to work on (Appendix A, Table A3, E6). Moreover, the 2021 summer floods created momentum for a societal discussion on what is to be considered an unacceptable risk and what is residual risk, generating conversations about a risk approach in which society should be actively involved (Appendix A, Table A2, I1).

The focus of the lessons in layer 1 is on prevention through dike reinforcement, following the HWBP policy. It was stressed that the HWBP should be expanded to areas currently not included in the HWBP programming, yet experiencing the impacts of the summer flood (Appendix A, Table A2, I7). Another lesson for strengthening flood prevention was to be more transparent about the safety standards for secondary water systems, to be able to test them for future climatological conditions, like as for primary water systems (E6).

Furthermore, the summer flood highlighted the need to alter how dike trajectories are prioritised for their reinforcement. What such a change could look like is illustrated by the following quote.

The determination of the urgency of measures within the HWBP is currently based only on distance to standard. The Limburg dikes have a relatively low standard, which means that the absolute distance to the standard is by definition small and the calculated strength is lower. As a result, the actual probability of flooding is high. In the HWBP programming, there is room to deviate from prioritisation, but in practice this proves difficult. Therefore, it may be desirable not only to prioritise based on distance to standard but also to take into account the probability of being affected by flooding. (E7)

This reflects the characteristics of double-loop learning, as it requires a revisiting of assumptions on the criteria for prioritisation.

Central lessons for layer 2, spatial adaptation, are about river widening strategies in the regional water systems. The Meuse Works proved effective along the primary water system, the Meuse itself. However, the 2021 flood impacted mostly the area of the Meuse’s tributaries. This explains the presence of double-loop learning: there is a shift in perception of what could still be considered water nuisance or whether it should be perceived as a flood safety issue. Water flow rates in secondary systems can become so high that ‘it is actually a bit insane that we have been defining it as a kind of water nuisance issue, whereas it is obviously just unsafe’. (I5).

The flood also showed that dike strengthening along the Meuse tributaries alone will not be sufficient or satisfactory for citizens and other stakeholders. Triple-loop learning is reflected in the idea to introduce a Room for the River approach for secondary systems and small water bodies. Implementing this in brook systems requires a paradigm shift and systemic change (E7).

Until recently, we have had no integration regarding learning about what we do with Room for the River and what that would mean for Room for the Brooks, to put it that way. That changed as a result of the flood … and we want to learn as much as possible from what we have done for Room for the River and see how we can implement that in Room for the Brooks. (I3)

It is no surprise that layer 3, crisis management, is covered in policy evaluations after a flood event. However, our focus is on measures related to improving (future) flood safety and not on an evaluation of the crisis response to the 2021 summer flood. What we see is that most lessons can be classified as single-loop learning, centred around the subject of improving the preparedness of the (crisis) organisations (I6; E4), but also of citizens, as shown in the following quote.

It was striking that self-reliance turned out to be greater than previously thought and the initiatives and solutions were creative. Partly because of this, there is a broad desire to support these initiatives and solutions in a more structured way. (E11)

Linking to the overall lessons for integrating the safety layers, there is double-loop learning on including regional water systems in preparing for water crises to improve system-based and multidisciplinary decision-making (E2, E9, E10).

Lastly, layer 4 is very limitedly covered in the policy evaluations, as it is a new addition to the multilayer safety framework. The only way in which it is present is in the recommendation that climate-robust measures in the recovery phase should become the standard norm. This reflects a level of learning based on rethinking frames, as it requires acting differently upon repairing damage after an extreme event.

Forward-lookingness of learning

This section answers research sub-question 3: To what extent are lessons drawn from the 2021 summer flood forward looking? We analyse the lessons on the different safety layers, and Table 2 provides an overview of the forward-looking (FWL) elements found as part of these lessons. We briefly discuss each forward-looking aspect found in the evaluation studies and interviews conducted in the aftermath of the 2021 flood event.

Table 2 Forward-looking (FWL) aspects per safety-layer lesson
Full size table

First, a forward-looking problem understanding was present for each of the layers. Prominent problems in layer 2 and layer 4 were a combination of the expected extremes under climate change (Appendix A, Table A3, E6, E7) and spatial planning issues such as a lack of space for interventions (E6) and surface hardening (I6; E7). A problem understanding for layer 3 was the unavailability of centralised knowledge for managing the impact in the primary water system and the lack of information on the flood behaviour of regional water systems (I7; E7), complicating future crisis and security management. Whereas the problem understanding for layer 3 revolves around information management for safety measures, layer 0 centralises a changing climate as a main factor for an increased residual risk. Conversations on how to deal with these risks are emerging, illustrated by the following quote: “Upfront, with society, you have to be more explicit on what is an acceptable risk and what is not, so that everyone knows” (I1). Lastly, the current national prioritisation and budget allocation for dike reinforcements (layer 1) is flagged as potentially problematic for safeguarding areas along the Meuse in Limburg in the future (I5).

Second, robust and flexible solutions were found primarily as part of layers 0, 2, and 3. The use of stress tests especially was suggested as a robust solution to test for a variety of extreme scenarios. Think of using stress tests to assign water and climate labels (layer 0) (E7), to check for risk-related flooding impacts in the secondary system (layer 2) (I7, E7), or to visualise how to include regional waters in crisis preparations (layer 3) (E1, E7, E10). Whereas robust solutions are found mainly in layers 0 and 2, adaptive solutions are represented in layer 3. This makes sense, as layer 3—crisis management—is all about responding to crises. To make this response more forward looking, the solution lies in adaptiveness based on monitoring and real-time scenario changes (E6, E7, E9, E10, E11).

Third, visions, scenarios, and long-term objectives were recommended for layers 1, 2, 3, and 4. More specifically, using qualitative scenarios including climate narratives and impact mapping was recommended for layer 3 (I6, E4, E10), just like drafting quantitative scenarios for emergency responses based on pre-defined key decision moments (I6) in order to better understand and prepare for year-round collective flood responsiveness. For layers 2 and 4, visions are prominently present justifications for forward-looking actions. These visions are, for example, linked to viewing the water system on a different, larger scale including the regional water system and bordering areas in Belgium and Germany (I3, I5), to societal transitions such as the energy transition (I7, E7), or to set goals and ambitions such as devising water and soil steering principles in spatial planning or creating a climate-robust landscape (E6, E7).

Forward-looking learning for future flood risk management beyond multilayer safety

Our analysis revolved around multilayer safety, which we use to analyse the breadth of learning. However, additional lessons with relevant forward-looking aspects that go beyond multilayer safety emerged from the data. As these lessons are likely to impact how future flood risk management could look, we decided therefore to mention them briefly here too.

First, forward-looking lessons are drawn about how information and knowledge will have to be managed for longer-term anticipation of flood events. That is, the models and scenarios used to anticipate crises were considered not accurate or forward-looking enough. To illustrate, the inflow of the Meuse’s tributaries, just like the effects of the HWBP measures taken over the past years, were not accounted for in the models, resulting in an inaccurate and incomplete risk estimation (E2, E9). Moreover, the models were not able to reconstruct the extreme precipitation leading to the summer flood, as the hydraulic models were calibrated on the basis of the floods of 1993 and 1995 (E1). The learning is in how to prepare and validate models for the future (E1), for example by sharing models and monitoring data internationally (E6) and by increasingly including climate change impacts in modelling for hydraulic thresholds (E7).

Similar lessons are drawn for working with more forward-looking scenarios. Examples include the construction of storylines for future low-likelihood yet high-impact scenarios (E1), acting on the reasonable worst-case scenarios (I6, E1), or running more scenarios including climate extremes (E3, E7). Another example is to develop scenarios for regional systems, including the seven tributaries of the Meuse (E5, E9, E11). To translate the forward-looking models and scenarios into actions, a community of practice could be started to bundle knowledge and to efficiently address challenges (E7).

Second, there has been forward-looking learning regarding using stress tests. Stress tests are used to identify potential current and future climate vulnerabilities in a certain area. Next to flood risk, these tests also assess vulnerabilities to drought, heat stress, and water nuisance. Currently, there is no uniformity in how the stress tests are conducted and they thus cannot be used to formulate a regional or national image on, for example, flood safety (E6). The 2021 summer flood showed that the present stress tests do not match the needs and that additional, trans-regional stress tests are needed in addition to the conventional stress tests (E7). These trans-regional stress tests cover a larger area than the conventional stress tests with the aim of connecting better to the full system’s dynamics, allowing for understanding and testing cascade effects (E7). For example, the impacts of climate change should be viewed from a trans-regional perspective in order to inform spatial planning, especially in areas bordering Germany and Belgium (E7). Performing trans-regional stress tests could have policy implications, as reflected in the following quote:

To be truly prepared, it is important that climate adaptation goals, governance, and monitoring come together. Therefore, strengthening governance is urgent. (E7).

Discussion

Looking back to what happened after the 1993–1995 floods, an initial focus on dike strengthening later shifted towards the approach of giving more room to the river by restoring floodplains, a policy idea that had been around for quite some time already [49]. The ‘room for the river’ approach has since become firmly established in the Dutch water policy landscape, and the 2021 flood became an occasion for promoting that approach for tributaries was well. The 2021 flood also became the occasion for expanding the multilayer safety approach to include ‘water awareness’ and ‘climate-robust recovery’. Again, these ideas are not entirely new, but have often failed to get attention. The strong trust in the technical capacities of Dutch public water managers to prevent disasters is paradoxically not conducive to high (public) awareness about what might happen and how best to recover. The inclusion of new layers in the multilayer safety approach puts awareness and recovery more strongly on the flood risk management agenda. Below, we elaborate our reflections on this study’s major findings, implications and limitations into five separate points.

A first point of reflection is the changing narrative about the role of the secondary, regional water system in relation to flood safety. Prior to the 2021 summer flood, the primary system was managed for flood risk reduction for safety purposes, whereas the secondary system was associated mostly with limiting water nuisance and creating freshwater buffers to combat droughts. The 2021 summer flood, however, showed that the Meuse’s tributaries could also pose threats to human safety. This realisation generated a shift in flood safety perceptions: regional water systems play a role in managing flood risk. As a result, creating more room for brooks and small water bodies for flood protection purposes was suggested, thereby building onto strategies formulated for the primary water system. Moreover, the water retention measures implemented as part of a programme of Limburg’s Regional Water Authority (P37) could play a role in achieving higher flood safety in the regional system if safety was approached from a systems perspective. This resonates with what Collentine and Futter [50] state about the role of green infrastructure in flood management for water retention purposes. They argue that natural water management retention measures help to reduce the impacts of floods and droughts as they can both reduce flood peaks and retain water. Such measures are diverse and include making room for the river, afforestation, and restoring ponds and wetlands.

Second, the empirical findings show that there needs to be a stronger connection between, on the one hand, crisis management and the crisis response phase and, on the other hand, flood risk management and the anticipation of possible flood events. This also connects to literature about resilience, where responding to shocks and the recovery after shocks are linked to adaptive and transformative responses to flood hazards [51,52,53]. In this study, both the regional and the national water authority wrote in their evaluations after the 2021 summer flood that there was a need for coordination with the crisis management units of two safety regions, yet that clear communication lines were not always present and that the exact roles and strategies were not clearly defined during the crisis. This might have implications for future flood risk and crisis management structures in which planning for flood risk reduction could have a more central place in the tasks and capacities of actors working directly on crisis and security management at regional and local level.

Third, by adding layer 0 and layer 4, the expanded multilayered safety framework now consists of water awareness (layer 0), flood prevention (layer 1), spatial planning (layer 2), crisis management (layer 3), and climate-robust recovery (layer 4). This is a very interesting example of policy learning in itself, but also merits attention from a conceptual perspective. More attention for water awareness resonates with international discourses on community resilience [54], but its applicability to large parts of the Netherlands is limited because flood defences are organised mostly by public authorities and at regional level rather than at community level. Community resilience is therefore more an additional than a central new perspective in Dutch flood risk management. Reflection on the expanded framework also suggests that two dimensions are mixed together: (1) the temporal phases of the disaster risk reduction cycle, often captured as prevention, preparedness, response, and recovery [42] and (2) the three layers or entry points for flood safety measures: preventive flood defences (like dikes and storm surge barriers), spatial planning measures, and civil protection measures to rely on during crisis management. Climate-robust recovery (layer 4), for example, connects closely to the recovery phase in disaster risk reduction, and the idea of building back better [47, 48], but the type of measures taken for climate-robust recovery might go across layers 1–3. For conceptual clarity, it would be valuable to disentangle temporal phases from layers of measures.

Fourth, what are the pathways towards transformation after a flood? We have observed mostly single-loop and double-loop learning after the 2021 flood, and only limited triple-loop learning. Of course, one flood event is unlikely to single-handedly bring about transformational change [55], especially as there were no casualties in the Netherlands and the primary flood defences along the Meuse did not fail. The major impacts occurred in the Meuse’s tributaries, allowing the then Delta Commissioner to claim in the media that the flood safety measures implemented along the Meuse had worked and that the approach needed to be expanded along its tributaries [56]. Still, a focusing event like this can play a role in a longer-term transformational pathway towards in-depth and system-wide change. Single events are not sufficient but may be necessary for policy lessons to take hold. Lessons learnt can contribute to small wins that can accumulate into transformational change over time (e.g., climate insurance tools), and they can play a role in developing stricter flood safety standards or inform the development of plans for the large-scale restructuring of flood risk infrastructure along the river [14].

Finally, we have connected different types of policy learning with multilayer safety and have applied the forward-looking framework for the first time to a different cycle within the policy process (from decision-making to evaluation) and to a context of crisis versus a context of investment planning. This opens up possibilities for the future, as the different forward-looking elements can be used for policy evaluation as well as possibly for the design of policies to increase the long-term horizon of policies and solutions to be implemented after flood events.

Limitations and future research directions

This study has focused primarily on analysing the evaluation studies executed after the 2021 summer flood to see what lessons have been drawn for the Limburg area in the Netherlands. Two limitations arise from this: first, evaluation studies tend to focus on crisis management and evaluating the crisis management response and focus less on long-term flood safety measures. However, one of the evaluations did focus on the way in which existing flood safety measures had performed, and another larger study (Appendix A, Table A3, E7) had a wider, more visionary scope with the aim of also looking into the future. Secondly, the focus on evaluation after a shock limits the time horizon and tends to look backwards rather than forwards. We recommend future research that will investigate the actual implementation of policy lessons drawn after the 2021 summer flood. As the current research was conducted in a limited timespan (less than two years) after the flood, the new policy agendas had not yet been formulated and published, and measures had not been implemented. However, it will be interesting and important to analyse the implemented water management measures and newly created water policies in the aftermath of the 2021 summer flood and to consider the extent to which this flood will lead to changes in the Delta Programme for flood safety when the programme is revised in its regular cycle (planned for 2026). We also encountered some difficulties with distinguishing the three types of policy learning, which are broadly defined, and we have tried to operationalise them in the way in which we have applied them to our results for the domain of flood risk management (see Table 1). This operationalisation should be further tested in future research.

Another future research direction arises from the country scope of this study. This article focused on the Netherlands and impacts in Limburg Province. The Netherlands is an interesting case for other countries, as governance structures regarding water management are well developed. The observed patterns and recommendations could be informative for other contexts, and the analytical framework could well be applied to other geographical areas. It would be especially interesting to compare the German, Belgian, and Dutch response to the 2021 summer flood with the same analytical framework because in Belgium and Germany, the high-water event led to more than 220 fatalities and infrastructure damage costing multiple billions of Euros [6].

Conclusion

This article aimed to answer the question What long-term lessons for flood risk management have been drawn from an acute flood crisis? The results are based on an analysis of the policy evaluations of the 2021 summer flood in the south of the Netherlands. We first analysed the flood protection policies before the 2021 summer flood to understand what lessons were new. Based on an analysis of policy evaluations after the flood and interviews, we analysed what lessons were drawn for different layers of multilayer flood safety and assessed how policy lessons also anticipate future events and risks for the region.

We found that water and flood risk management policies before the 2021 summer flood focused mostly on flood prevention through dike strengthening and, from 1993 onwards, on spatial planning measures with the start of the Room for the River programme after the river floods and large evacuations of 1993 and 1995. The crisis management layer was not so well represented until 2021.

An important lesson drawn by the 2021 flood evaluation committee was to add two layers of flood safety to the multilayer safety approach: water awareness and climate-robust recovery. Societal water awareness is at the basis of a functioning multilayer safety approach according to the committee, and focuses on the awareness of non-governmental actors of different water and climate risks. Climate-robust recovery is introduced to emphasise that the type of recovery has implications for future flood safety and focuses on building back better after floods to improve long-term robustness. The adding of these two layers reflects a form of double-loop learning, by advocating a risk-based approach and stimulating private sector and citizens’ responsibilities.

Another important lesson, that reflects the deepest type of learning involving paradigmatic shifts, was to extend the application of the multilayer safety concept beyond the primary flood system to the secondary, tributary, system. The lesson drawn is to create more room for these tributaries, in parallel to the Dutch Room for the River programme that was implemented between 2006 and 2015.

Finally, more incremental types of learning by extending existing practices were also present. Examples are expanding dike-strengthening projects, providing clarity on safety standards, and improved information sharing and communication towards citizens before, during and after flood events.

To improve the long-term preparedness to deal with future extreme events, we also analysed the extent to which lessons are concerned with anticipating climate change. The ways to prepare for climate change focus strongly on improving stress tests, including more extreme weather events. Regions are also recommended to develop future visions to implement lessons for layer 2 (spatial adaptation) and layer 4 (climate-robust recovery); such visions can include changes to existing land use, locations for water retention, new ways of building, and changes with regard to where to build.

Future research is recommended to analyse the implementation of the recommendations from policy evaluations and to study the extent to which climate-robust recovery has been realised. Governments will also need to engage more with an integrated understanding of the different layers for multilayer safety to ensure that regions are able to prepare for future floods—which will become more frequent and intense as a consequence of climate change.