Adapting densely populated deltas to the combined impacts of climate change and socioeconomic developments presents a major challenge for their sustainable development in the 21st century. Decisions for the adaptations require an overview of cost and benefits and the number of stakeholders involved, which can be used in stakeholder discussions. Therefore, we quantified the trade-offs of common measures to compensate for an increase in discharge and sea level rise on the basis of relevant, but inexhaustive, quantitative variables. We modeled the largest delta distributary of the Rhine River with adaptation scenarios driven by (1) the choice of seven measures, (2) the areas owned by the two largest stakeholders (LS) versus all stakeholders (AS) based on a priori stakeholder preferences, and (3) the ecological or hydraulic design principle. We evaluated measures by their efficiency in flood hazard reduction, potential biodiversity, number of stakeholders as a proxy for governance complexity, and measure implementation cost. We found that only floodplain lowering over the whole study area can offset the altered hydrodynamic boundary conditions; for all other measures, additional dike raising is required. LS areas comprise low hanging fruits for water level lowering due to the governance simplicity and hydraulic efficiency. Natural management of meadows (AS), after roughness smoothing and floodplain lowering, represents the optimum combination between potential biodiversity and flood hazard lowering, as it combines a high potential biodiversity with a relatively low hydrodynamic roughness. With this concept, we step up to a multidisciplinary, quantitative multi-parametric, and multi-objective optimization and support the negotiations among stakeholders in the decision-making process.
The World Economic Forum ranked extreme weather events, natural disasters, and failure of climate-change mitigation and adaptation in the top five risks in terms of likelihood as well as in terms of impact (WEF, 2018). Between 1995 and 2015, floods made up 43 % of the global occurrences of disasters within the category of extreme weather events (Wahlstrom and Guha-Sapir, 2015). For the future, Alfieri et al. (2016) showed that rising global temperatures will further increase the frequency and magnitudes of alluvial floods globally. In addition, coastal flood hazards are expected to increase due to sea level rise and changing storm wave and storm surge characteristics (Pardaens et al., 2011; de Winter and Ruessink, 2017) with high adaptation costs for coastal flood damage (Hinkel et al., 2014). Coastal deltas are particularly prone to flooding due to the possible coincidence of peak river discharges and storm surges. But even without a storm surge, the increased backwater effect due to higher sea levels affects water levels in delta distributaries during alluvial flood events. The ongoing urbanization in many deltas, combined with the associated land subsidence, further increases the exposure to floods (Giosan et al., 2014; Tessler et al., 2015). Flood protection measures (interventions) therefore need careful integration in the spatial planning of transport infrastructure and cities with a temporal horizon of 2100 and beyond. At the conceptual level, a strong point has been made for adaptation planning and nature-based solutions (Brown et al., 2014; Cheong et al., 2013), but the tools are missing to quickly apply these concepts at the delta scale.
Measures for flood hazard reductions should also take nature restoration into account. Rivers and deltas potentially have a high biodiversity because of the periodic flooding (Tockner and Stanford, 2002; Ward et al., 1999). However, land use change and population growth have degraded the biodiversity of floodplain habitats, especially in North America and Europe (Vorosmarty et al., 2010). River restoration efforts over the last three decades have tried to reverse degraded fluvial ecosystems (Bernhardt et al., 2005; Buijse et al., 2002), which were supported by the Clean Water Act in the USA and the Water Framework Directive in the EU (Hering et al., 2010). The combined efforts of projects with a joint objective of lowering the flood hazard and increasing biodiversity proved possible, although the biodiversity of protected and endangered species is still far below its potential (Straatsma et al., 2017).
River management has to combine varying objectives: flood safety is the primary goal in densely populated deltas, and navigation is often second. River restoration is increasingly included in industrialized countries to improve the biodiversity. A key challenge for environmental management is the number and diversity of the actors and sectors involved in the decision process because each has its own perceptions, interests, and resources (Robinson et al., 2011). Differences are fueled by the short temporal horizon for political decisions and the long-term effects of climate change but also by the perceived necessity of landscaping (i.e., terraforming; reconfiguration of the channel-floodplain geomorphology in US terms) measures to climate-proof the delta and societal resentment against large measures. Given the multitude of objectives and options for spatial planning, the need for decision support systems (DSSs) has long been recognized. Ideally, a DSS provides rational input, which could remove emotional objectives against specific adaptations, but we recognize that the personal threat experience also strongly drives actions to limit the flood risk (Grothmann and Reusswig, 2006). Newman et al. (2017) reviewed 101 DSSs for natural hazards, of which 19 focused on fluvial flood risk. Two included studies in the Netherlands: Hübner et al. (2009) developed the “Nature-oriented flood damage prevention”, targeting regional water systems, whereas Schielen and Gijsbers (2003) created “DSS-large rivers”, which was oriented towards the national to continental scale. Both DSSs required the manual implementation of landscaping measures in the accompanying geographic information system (GIS). To the best of our knowledge, none of the DSSs for fluvial flooding listed by Newman et al. (2017) enabled the semi-automatic planning of measures at the spatial scale of the river reach. A semi-automatic system was presented by Straatsma and Kleinhans (2018), who used a rule-based system for the positioning and parameterization of measures. Coupling a DSS with semi-automated planning of mitigation measures could have additional value in the exploratory planning phase to provide all stakeholders with the efficiency of measures with respect to flood hazard reduction, costs, and biodiversity. No quantitative methodology exists that can combine the planning and parameterization of measures with a multidisciplinary evaluation.
The implementation of large-scale measures requires the alignment of governance at national, provincial, and municipal level and the involvement, compensation, or expropriation of different land owners. Land owners and actors involved in river management have diverse perspectives on, and incentives for, implementing river management measures based on socioeconomic, cultural, and land use characteristics (Rosenberg and Margerum, 2008; Verbrugge et al., 2017). New adaptation measures could therefore be implemented faster when fewer stakeholders are involved, provided they can agree about the type of measure. Therefore, it is necessary to understand land owner characteristics and their motivating factors to indicate the potential for implementing large-scale measures (Rosenberg and Margerum, 2008). Our objectives were to (1) quantify multifaceted trade-offs between landscaping measures to adapt a large delta distributary to sea level rise and increased river discharge while honoring ecological value and societal stakes and (2) include government complexity by positioning the measures in areas owned by the two largest stakeholders versus all stakeholders based on a priori preferences. The measures were parameterized based on nature restoration principles or maximizing flood conveyance capacity.
The decision on different possible interventions requires a multidisciplinary evaluation. Scientists can help to support this evaluation by transforming their data, models, and tools into quantities that can be used to objectively evaluate the different interventions. Here we show an example of how using advanced DSSs, that include cost estimates, as well as physical and ecological quantities, can help to move towards an evidence-based decision based on multidisciplinary performance metrics. We assessed the ability of 17 measures to compensate for increased discharge and rising sea levels in the Waal River in the Netherlands while improving the potential for biodiversity.
The river Waal, a river reach 85 km long between the Pannerden bifurcation near the Dutch–German border and
Gorinchem, upstream of significant tidal influence. The center of the study area is located at 51
The study area is located in the Rhine delta (Fig. 1) in the Netherlands
and comprises the main channel and embanked floodplains of the Waal River.
The Waal is the main distributary of the Rhine River and is affected by
expected changes in peak discharge as well as sea level rise. The three main
concerns here are flood risk in view of global change, navigability and
ecosystem functioning. The study area spans a river reach 85 km long, with
an average water surface gradient of 0.10 m km
Characteristics of nine land owner types involved in the maintenance of floodplains (based on Fliervoet and Van den Born, 2017). The remaining area is owned by foundations (66) and churches (20).
Flood risk management involves a number of public and private stakeholders (Table 1). It traditionally focused on the construction and maintenance of embankments (artificial levees), but recently the link with sustainable spatial planning has gained attention (Jong and Brink, 2017). Governmental responsibilities are divided over four levels, i.e., national, provincial, regional (water boards), and local (municipalities). The two largest land owners are Public Works and Water Management (PWWM; Rijkswaterstaat in Dutch), which owns 2889 ha, mostly consisting of the main channel and groyne fields, and the State Forestry Service (SFS; Staatsbosbeheer in Dutch), which owns 2813 ha that is mostly situated in the floodplains (Table 1). PWWM is the national water authority, and SFS is the national nature conservation organization; both are governmental organizations. The total number of owners is 1233 between the embankments and 5512 within an additional 50 m buffer (Table 1) based on the cadastral database, which highlights the complexity of implementing area-wide measures.
Overview of attributes of floodplain sections (Fig. 1). The sections are labeled by the mean river kilometer, followed by the
“-r” for river right and “-l” for river left:
The floodplain consists of 94 individually labeled areas on the left river bank (south) and the right bank, to which we will refer as floodplain sections (Fig. 1, grey shading). The area per stakeholder type differs strongly over the sections (Fig. 2a) as well as the number of owners (Fig. 2b). For example, the section at river kilometer (rkm) 870 on river left (Fig. 2; 870-l) is called the Millingerwaard. It has a total surface area of 721 ha including the main channel, is largely owned by the State Forestry Service (428 ha), and has a total of 17 different owners, of which 12 are private citizens. Section 885_l (city of Nijmegen) contained the largest number of owners in the embanked area: 633 in total.
The rule-based planning and evaluation of measures required detailed input data (Table 2). Here, we describe the modeling tools plus their input data and the choices made within the hydrodynamic and landscaping scenarios.
Overview of input datasets used for intervention planning and evaluation.
Here, we briefly describe the existing tools, RiverScape, Delft3D Flexible Mesh, and BIOSAFE, and elaborate on the cost assessment and the land ownership of all stakeholder groups. Reproduction, application, and extension of this work can also be conducted with other hydraulic models and other codes for biodiversity, while RiverScape has novel capabilities (Straatsma and Kleinhans, 2018).
Typical landscaping measures implemented in this paper (figure after Middelkoop and Van Haselen, 1999).
In current river management practice, managers propose measures in the embanked floodplains together with landscape architects, engineers, policy advisors, and local stakeholders. Based on a sketch of the intervention, a GIS specialist translates the position of the measure and the parameterization in terms of land cover and terrain height into layers of spatial data, for example with HEC-GeoRAS (Ackerman, 2011) or Baseline (Scholten and Stout, 2014). Both steps are time-consuming, and, therefore, often only a few scenarios are developed (Nardini and Pavan, 2012). For this study, we used RiverScape, a software tool for the rule-based positioning and parameterization of flood hazard reduction measures. A detailed description of the tool is given by Straatsma and Kleinhans (2018). In brief, given a set of raster layers describing the hydrodynamic conditions at design discharge, the geometry, and the land cover, this tool proposes the location of seven different types of measures (Fig. 3), based on a set of rules for positioning and parameterization. The new bathymetry and the height of minor embankments (minor levee for the prevention of summer flooding of floodplain agricultural fields) and groynes (wing dikes, or spur dikes) are controlled by user-specified reference heights expressed as water levels at the river axis with a specific exceedance probability. The new land cover is given as an ecotope and roughness class. RiverScape was extended with a masking option to enable the application of the measures over arbitrary areas, in this case, the areas owned by specific stakeholder groups. Each measure is defined by seven raster layers (area, bathymetry, ecotopes, trachytopes (roughness class), groyne height, minor embankment height, and main embankment height), together with the user settings. The generation of measures takes less than 2 min, which makes it a fast option for scenario development.
RiverScape was coupled to a calibrated 2-D hydrodynamic model. We used Delft3D Flexible Mesh (DFM), the open-source hydrodynamic model that is developed and maintained by Deltares (2016). The computational core of DFM solves the shallow water equations based on the finite-volume method on an unstructured grid (Kernkamp et al., 2011; UGRID Conventions, 2016). The computational mesh of the study area consisted of 71 000 active cells between the main embankments. The DFM input files consisted of bathymetry, trachytopes, fixed weirs for groyne height and minor embankment height, thin dams for buildings and bridge pillars, and dry areas for embankment relocation. These files were updated with the RiverScape measure definition of each intervention following procedures described by Straatsma and Kleinhans (2018). The boundary conditions are specified by the upstream discharge and the downstream water level (Fig. 1). Only stationary discharges were simulated.
We applied the BIOSAFE model (De Nooij et al., 2004; Lenders et al.,
2001; Straatsma et al., 2017) to evaluate the measures on the potential
biodiversity for protected and endangered species that are representative of
the fluvial environment. The BIOSAFE conceptual model comprises a set of
links between riverine species and legal and policy documents for species
protection on the one hand and links between species and ecotopes on the
other hand. These two sets create a link between the legal domain and
ecotopes via species. BIOSAFE calculates scores of potential biodiversity
for seven taxonomic groups. The scores represent potential species presence
based on habitat requirements, which were weighed by (1) the number of legal
and policy documents assuming equal importance for all documents and (2) the
surface area of associated ecotopes, and that are normalized by the area
under consideration, enabling the comparison of the scores over floodplain
sections with different sizes. In this study we calculated the
River restoration projects are costly, but costs are often reported in aggregated form over the whole project (Bernhardt et al., 2005). Ayres et al. (2014) compiled the available evidence of the cost of river restoration and showed that the costs varied strongly for a single type of measure but also that only a few of the cost estimates contained information on different cost items. They proposed a cost typology, which distinguished nonrecurring costs (planning, transaction, land acquisition, and other construction/investment costs) from recurring costs (annual maintenance and monitoring costs). In this study, we limited the scope to the nonrecurring costs required for the implementation of the measures. The data for the cost appraisal (Appendix A; S. Prins, unpublished data) should be interpreted as indicative, as neither the building contractor nor the client that orders the measure wants to share their valuation tables for fear of losing their position during negotiations. Recurring costs are generally an order of magnitude lower and depend on the land management after the implementation of the measure.
The preprocessing for the cost evaluation consisted of the information extraction from several sources to obtain the spatial distribution of the cost items. The so-called cost maps were subsequently overlaid with the measure definition. We used (1) the BAG-2015 database, a cadastral database for building locations and building types; (2) Top10vector, a vector-formatted geodatabase containing the roads' location, type, and width; (3) the ecotope map of 2012 for a simplified land cover; (4) the Baseline geodatabase for the position and length of groynes and minor and main embankments; and (5) the soil pollution map (Table 2 and references therein). Polluted soil is expected only in the top 1 m of the soil (Middelkoop, 2002) from sediment deposition in the floodplains. The ecotope map was used for the cost of roughness smoothing. Units of river kilometer were used to calculate the cost of dike raising. To calculate the nonrecurring costs, we determined the capital expenditures (CAPEX) for each measure. We are aware that additional operational expenditures (OPEX) increase the total cost over the lifetime of the measure and that different trade-offs could be found depending on the temporal horizon, but this is beyond the scope of this study. For each cost item, we mapped the unit cost in Euro per unit; the standard deviation is around 15 % of the unit price (Appendix A). The spatial distribution of the costs of smoothing, road removal, minor embankment removal, building acquisition and demolition, and land acquisition (Fig. 2d) indicates that the acquisition cost of land and buildings dominates the overall cost of measures.
The calculation of the cost per measure comprised the overlay of the cost maps with the measure definition. For side channel recreation and floodplain lowering, the unit costs of earthwork per cubic meter were added. The volume of earthwork depended on the measure settings and the existing topography. Postprocessing was required to correct for the use of data from different sources. For example, the ecotope map does not contain road information, but the cost for road removal should be equal to zero for a smoothing measure. No land acquisition costs are assumed for roughness lowering.
We used the cadastral map (
The modeling tools of Sect. 3.1 enabled the exploration of different adaptation scenarios with respect to changing hydrodynamic boundary conditions and adaptation measures. We used 17 measures (Table 3) for each of the three hydrodynamic scenarios. Trade-offs between flood safety, implementation cost, potential biodiversity, and number of stakeholders were quantified for each hydrodynamic scenario. The trade-offs consisted of the measures that represented the optimal combination of two variables and were represented as a line. The optimal measures were extracted from the convex hull of the measures' scores in the attribute space. No attempt was made to select a single optimal measure by means of minimizing an objective function because such techniques require weighing factors for the four aspects, and these are currently unknown. The weighing factors can also change quickly due to changing public opinions and political will, and they differ between stakeholders. Nonetheless, we highlight specific interventions as optima in the multidimensional parameter space.
Overview of 17 landscaping scenarios: six measures, two locations based on stakeholders (all stakeholders and large estate owners), and two design principles (smooth and natural). All measures were evaluated for three sets of hydrodynamic boundary conditions (Q16_dh0.0, Q18_dh0.0, and Q18_dh1.8). The abbreviations are used in results' figures.
NA: not applicable.
The hydrodynamic boundary conditions were given by the upstream river
discharge and the downstream water level. The embankments have a flood
protection standard for a flood (Q16) with a statistical return period of
1250 years (Silva et al., 2004). However, Q16 did not include
the projected effects of climate change. The future design discharge of the
river Rhine is uncertain. In policy documents, 18 000 m
The options for landscaping measures for flood safety and river restoration by far exceed the two options to change hydrodynamic boundary conditions. The evaluation of all these options in terms of flood hazard, biodiversity, and costs is very time-consuming to carry out using preexisting software. The multitude of possible scenarios is driven by the responsible authorities and the number of stakeholders, their land ownership, and their preferred land use and legal permissions. At the same time, the perceived urgency to reduce the flood risk, the available budget, and political will also affect the choice for measures and the speed of implementation, although all stakeholders agree about flood safety as the number one priority. We simplified and summarized the stakeholders' preferences for specific measures based on their organizational objectives and responsibilities (Table 1) in order to derive a manageable set of scenarios for landscaping measures (Table 3). The stakeholders' preferences (Table 4) were based on Table 1, literature (Fliervoet and van den Born, 2017; Fliervoet et al., 2013), and expert judgement.
Simplified stakeholder preferences for different measures.
Legend is as follows:
The directorate for Public Works and Water Management (PWWM) is the largest landowner (Table 1). They are the responsible authority for the flood protection objectives together with the water boards. Therefore, both are in favor of measures that improve the flood protection levels, especially on their own lands, such as roughness smoothing, floodplain lowering, and groyne lowering (Table 4). Although many side channels were constructed in the Room for the River program to realize flood protection and nature objectives, the PWWM does not have a clear preference for side channels because of high maintenance costs and increased sedimentation in the navigation channel (Van Vuren et al., 2015).
The State Forestry Service aims to develop more natural (unregulated) river systems by giving more room to natural erosion and sedimentation processes, in line with the vision of “self-regulating nature” (Stanford et al., 1996; Ward et al., 2001). This vision became a source of information for the Dutch ecological rehabilitation programs of the Rhine branches and Meuse River (Buijs, 2009), and it addresses measures which create a more dynamic floodplain environment, such as the construction of a side channel or lowering of the floodplains (Table 4).
The provincial governments are responsible for maintaining and developing nature in the floodplains since the decentralization in 2014 from the Ministry of Economic Affairs. They plan and implement EU Natura 2000 objectives, based on European legislation, and they allocate subsidies for nature conservation. This may require changes in land use, which are in turn regulated by the municipalities. Although the provinces are the nature authority, they are in favor of measures that have multiple objectives, such as constructing a side channel or lowering a floodplain. Both measures have the opportunity to reconcile the objectives of flood protection and restoring nature in the floodplains.
The water boards are responsible for, and the owners of, the dikes and minor embankments. They prefer the following measures: minor embankment lowering and dike raising. Private land owners (especially farmers) and companies are often against measures such as minor embankment lowering and floodplain lowering, due to the increased flooding frequency of their land, which negatively affects their activities. The sand, gravel, and clay mining industries are often in favor of measures that lead to the excavation of soil in the floodplains. Finally, the province foundations (“Geldersch and Brabants Landschap”) are aiming at the conservation of nature and cultural heritage, such as braid hedges (woven growing hedges used as parcel delineation) or fortresses in the floodplains. These foundations perceive minor embankment lowering and floodplain lowering negatively and roughness smoothing positively where a “park-like” landscape is concerned (Fliervoet et al., 2013).
Stakeholder involvement and their land ownership determined the locations where the measures were positioned. The first option for the location of measures was the combination of the large stakeholders (LS) that own large areas, being PWWM and SFS, based on their a priori preferences of measures. They own 56 % of the whole study area together (Table 1) and 31 % of the groyne field plus floodplain area. They agree with floodplain lowering as a possible measure, disagree with roughness smoothing, and do not disagree with the other measures (Table 4). SFS opposes roughness smoothing because it opposes their vision of self-regulating nature, except when it is performed within the context of cyclic floodplain rejuvenation. Given the large area owned by these two stakeholders only, they can relatively easily implement the different measures on their own property, even though they pursue different ultimate objectives. The second option was that the measures could be implemented on the property of all stakeholders (AS). This means that all 1233 stakeholders would need to endorse the measure or be compensated. Given the different objectives of the stakeholders, this can only be realized after a major disaster when the urgency is high. Although this is not the current state in the Netherlands, we still include it because of the changing hydrodynamic boundary conditions over time. This gave two location scenarios: LS and AS.
Six adaptation measures were implemented in the groyne field and the existing floodplains (Fig. 3); measures in the main channel and the areas protected by the main embankments were not considered. The design principle of the measure affected the choices made within RiverScape with respect to the new land cover and the cross-sectional shape of new side channels. The first option, labeled as “smooth”, was to optimize the conveyance capacity of the floodplain, whereas the second option, labeled as “natural”, included ecological qualities as favored by SFS and the provincial government. No difference between smooth and natural designs was implemented for minor embankment lowering, groyne lowering, and dike raising because the land use is assumed to remain identical.
We parameterized the measures in RiverScape (Straatsma and
Kleinhans, 2018) with the following settings. Firstly, roughness lowering
(smoothing) was applied over 100 % of the location (LS or AS) and
resulted in production meadow (ecotope UG-2 and trachytope 1201) for the
smooth scenario and natural grassland (ecotope UG-1 and trachytope 1202) in
the natural scenario. Production meadow has a slightly lower roughness than
natural grassland, with Chézy coefficients of 38 and 35 m
In order to quantify the trade-offs, we will first describe individual components to gain a detailed understanding. The hydrodynamic evaluation is given most attention because flood safety represents the top priority for all stakeholders.
The measures were positioned in the areas owned by the large stakeholders (PWWM and SFS; Fig. 4), or over the whole area. The fraction of the surface area owned by the large stakeholders varied strongly between the floodplain sections. The main channel and the groyne field are completely owned by the PWWM, whereas SFS owns the majority of specific floodplains sections, such as around 900-l and 928-r (Fig. 1).
Location of the two largest stakeholder, the state (Public Works and Water Management) and the State Forestry Service (SFS). The other public and private stakeholders own the remaining areas in white. The location of the side channels represents the smooth option for the whole area (AS). Water flows right to left. Measures are implemented over the whole reach, independent of the reference water level being exceeded.
The different hydrodynamic boundary conditions (Fig. 5a, b) led to spatial variation in the water levels. Q18_dh0.0 gives a 0.33 m increase in water levels at the downstream boundary, which increased rapidly in the upstream direction to a maximum of 0.76 m at rkm 880. Conversely, the Q18_dh1.8 scenario gave the highest water level increase downstream: 2.12 m. In the upstream direction, the differences decline due to the reduced impact of the backwater effect further upstream, with a minimum increase of 0.76 m at rkm 868, the model boundary on the upstream end. Two measures give lower water levels than in the reference situation in the Q18_dh0.0 and Q18_dh1.8 scenario, which is an over-dimensioning of the measures.
The effects of the measures differed strongly between the type of measure and the location (Fig. 6, left column). We compared the simulated water level with the reference situation (Q16_dh0.0, without measures) for all hydrodynamic and landscaping scenarios to see to what extent the measures can lower the flood hazard or mitigate the changing hydrodynamic conditions. Groyne lowering mainly affected the upstream area, with a maximum lowering of 0.06 m. Minor embankment lowering gave a maximum lowering of 0.07 and 0.11 m for large stakeholders (LS) and all stakeholders (AS), respectively. Roughness lowering was more effective, with a maximum of 0.21 m for AS_smooth and a minimum of 0.09 m for LS_natural. LS_smooth and AS_natural both reach 0.14 m water level lowering. The largest effects and the largest differences between the measure parameterizations were observed for side channels and floodplain lowering. Side channels showed a sequence of backwater curves from the individual measures, with a maximum lowering of 0.38 m for AS_smooth and 0.34 m for AS_natural (Fig. 6) in the upper part of the Waal, which is due to the combined effects of all channels and the higher water surface slope. Floodplain lowering gave the maximum water level reductions, which ranged between 0.62 m for lowering_LS_natural and 1.37 m for lowering_AS_smooth.
Water level changes (dh) at the river axis relative to the reference situation (Q16_dh0.0). The water levels of Q18_dh0.0 and Q18_dh1.8 are shown as grey lines. Water flows left to right, with the upper and lower area highlighted as grey vertical bands; see text for statistics. Measure (msr) type is indicated on the right. The grey lines in the middle and right column represent the effects of boundary conditions only.
The water level lowering per measure was almost equal between the
hydrodynamic scenarios. The correlation coefficient between water level
lowering exceeded 0.999 between Q16_dh0.0, Q18_dh0.0, and Q18_ dh1.8. The similarity in lowering can be
observed in Fig. 6 by comparing the differences in water level between the
reference (
The measure efficiency in compensating the changing hydrodynamic conditions (Fig. 6 middle and right columns) showed that only floodplain lowering was able to lower water levels below the reference situation. Lowering was more efficient upstream, with the zero crossing at rkm 933 and 920 for AS_smooth in Q18_dh0.0 and Q18_dh1.8, respectively (Fig. 6). The other measures did not lower the water levels below the reference. All measures were more efficient in water level lowering in the upper reach. On average, the lowering was 2.2 times larger upstream (rkm 868–894) than downstream (rkm 919–945), ranging from 1.1 times for smoothing_AS_natural to 4.4 times for minemblowering_LS. Floodplain lowering at the locations of all stakeholders could compensate for the additional discharge (Q18_dh0.0) both upstream and downstream, but for Q18_dh1.8 this measure only suffices upstream (Fig. 6 bottom right panel).
The overall changes in
Changes in
The
The implementation costs varied strongly between the measures, ranging from EUR 1.7 billion for lowering_AS to EUR 2 million for floodplain smoothing (Fig. 8). The largest costs are inferred by the acquisition of buildings and land, and the costs of forest removal and mowing for floodplain smoothing are very low. The costs of raising the dikes represent a 1 m increase in dike height. The fraction of the costs for building removal is 9 % for the LS case and 35 % for the AS case, indicating that the LS have relatively few built-up areas.
Comparison of costs between scenarios in millions of euro (M euro).
The combination of the water level lowering, changes in
Scatterplot matrix of the main criteria for intervention planning. The grey areas indicate the optimum combination of criteria for each panel. Similar colors represent one type of measure, and shading represents the difference in location. The two design principles are visualized, with a rectangle for natural and a triangle for smooth. The diamond represents the reference situation.
We considered a low number of stakeholders favorable for fast implementation
of a measure. The number of stakeholders involved in measures has an optimum
in water level lowering for dike raising (
The trade-offs between
Cost-effectiveness measures in terms of water level lowering consisted of dike raising and smoothing_AS_smooth (Fig. 9c), which are the two traditional methods of flood risk prevention in the Netherlands. Floodplain lowering_LS touches the grey optimum line and keeps an intermediate position. Lowering_AS is almost 3 times more expensive than dike raising and does not lower the water levels much in the lower reaches (Fig. 6). All roughness smoothing measures have low implementation costs, but it should be noted that they will differ in productivity after implementation.
Improving the
The flood protection structures and the land use in the delta are driven and constrained by three main needs: flood risk, socioeconomics, and ecology, as protected by national and European law. In this paper, we quantified the trade-offs between 17 landscaping measures to adapt a large delta distributary to increased flood hazards from sea level rise and increased river discharge, and we showed the effects of governance complexity using land ownership and stakeholder preferences as a proxy. Our methodology suits the early stages of the planning process as it provides an overview of possible measures to adapt to climate change and the associated capital expenditures, plus the hydrodynamic and ecologic effects. Compared to other DSSs targeted at intervention planning (Hübner et al., 2009; Schielen and Gijsbers, 2003), we added the option for automatic positioning and parameterization of measures over arbitrary areas, costs, and the number of stakeholders. Our modular structure enables more evaluation criteria to be added. In the final stages of intervention planning, additional weighting of interventions is required in practice using a multi-criteria analysis. Changing the weights will alter the trade-offs between the evaluation parameters. For example, the single objective of flood hazard reduction would rank embankment raising, floodplain lowering, side channels, and roughness smoothing as top priorities, whereas conversion to natural grassland would be favored from the river restoration perspective of protecting threatened and endangered species.
The quantification of the trade-offs showed clear optima in the parameter
spaces between water level lowering, potential biodiversity, implementation
costs, and the number of stakeholders (Fig. 9). It confirmed the
cost-effectiveness of dike raising and roughness smoothing, which are the
measures that represent the traditional flood protection strategy. We showed
that flood safety can only be maintained by raising the dikes by 1 m
or excavating the floodplains over the entire area, as long as only measures
inside the embanked floodplains are considered (Fig. 6). Large-scale
embankment relocation can also lower water levels by a meter
(Straatsma and Kleinhans, 2018). None of these options are
politically accepted at the moment, given the recent completion of the
so-called Room for the River program. This program aimed at increasing the
design discharge from 15 000 to 16 000 m
The quantified trade-offs led to new insights in the potential effectiveness of the two largest stakeholders for water level lowering and biodiversity. Measures in the areas owned by the two largest stakeholders lowered the water levels more effectively per unit area because these stakeholders own the areas with the highest conveyance capacity. The parameter space between potential biodiversity and implementation cost gave a surprising quasi-horizontal trade-off, with the highest biodiversity scores for roughness smoothing and conversion to natural grassland at a low cost, which is followed by the reference situation. This highlights natural grassland as a good candidate for multi-objective optimization on biodiversity increase and flood hazard decrease because the difference in water level lowering was small. Such an intervention would not have shown up in the typical cost-effectiveness of measures. The addition of biodiversity scores to the evaluation enables simultaneous evaluation of measures for the EU Flood Directive and Water Framework Directive.
The owner-specific areas for measures served as a proxy for the complexity of implementation in terms of governance because more owners means longer implementation times. It created insight into the possible contributions of the stakeholders in large-scale interventions. Decision-making in integrated river management is more complex and dynamic in reality because of the number and diversity of stakeholders and sectors involved, each with their own views, interests, and resources (Mostert et al., 2007; Robinson et al., 2011). The outcome of stakeholder sessions, the preferred measure, will vary depending on the individuals involved in the debate, which possibly leads to a suboptimal solution. We determined the stakeholder preferences a priori (Table 4) and used their preferences and land ownership to position and parameterize the measures. Our findings suggest that despite varying views of stakeholders, it is useful to model stakeholder preferences based on factors such as their role, responsibility, and visions for floodplain management. This provides valuable knowledge for stakeholders prior to or during engagement in multi-stakeholder platforms from the perspective of social learning (Borowski, 2010). The scores per land owner group (LS and AS) provide the stakeholders with a better understanding of the possibilities and limitations of the solution space. It has the potential to accelerate the decision-making processes because the stakeholder preferences and their interdependence are concisely visualized and immediately apparent with our methodology. Mutual recognition of interdependence and a shared understanding of the possible solutions are essential elements in the decision-making process (Ansell and Gash, 2008).
To the best of our knowledge, stakeholder sessions have not been repeated to assess their variation in outcomes because they are time-consuming. Alternatively, the variation in stakeholder processes could be modeled using game theory, or agent-based models, but this is still in its infancy: Samsura et al. (2010) used game theory to extract the strategic decisions used by stakeholders, Strager and Rosenberger (2006) integrated GIS with stakeholders preferences in a spatial multicriteria analysis to identify high priority areas for land conservation, and Becu et al. (2003) created an agent-based system of a catchment in northern Thailand, including farmers' individual decisions. They attributed the agents with the availability of water, land, cash, and labor force and focussed on decisions made by farmers, instead of including the preferences made by other stakeholders. These studies generally focussed on catchment scale and require substantial adaptation before they can be applied in the management of large lowland rivers. Our a priori preferences could serve as input for these types of models, and their output could be used to drive the planning of measures.
In this paper, we used water level lowering as the starting point for the positioning of the measures. An alternative to the owner-specific areas of measure locations could be ecological or financial considerations. Ecological optimization would involve ranking the ecotopes according to their potential biodiversity. High-ranking ecotopes should be left untouched, whereas ecotopes with a low potential biodiversity are suitable candidates for river restoration measures. Additional weighting of the ranking could be the hydrodynamic roughness or specific taxonomic groups. Likewise, the locations with high economic value could be left untouched to lower the costs of implementation. All these choices could be implemented as alternatives in the rule-based positioning and parameterization of measures.
The limited capacity within the floodplain area to lower the flood hazard points to the need to create more space for the river and robust measures for additional discharge (Q18) and sea level rise (dh1.8). Our results are useful for developing an integrated river management plan because we provided large-scale boundaries for decision-making at the scale of a river reach. The results can help to argue in favor of establishing multi-stakeholder platforms, such as river basin organizations, collaborative watershed partnerships, stewardship councils (in Dutch Waardschap), and “collaborative superagencies” (Fliervoet and van den Born, 2017; Jaspers, 2003; Pratt Miles, 2013; Verbrugge et al., 2017). With such collaborative structures, the major problem of fragmentation in terms of the number of land owners (Table 1, Fig. 9) could potentially be overcome.
Our methods were limited to the implementation of the measures and the
effects on the peak water levels. Several extensions would create additional
value for decision support. Firstly, extending flood hazard to flood risk of
the protected land would provide insight into the costs of the measures in
relation to the avoided losses in the case of inundation of the protected land.
For this, the failure probability of the embankment should be assessed
(Marijnissen et al., 2019) as part of a full flood
risk assessment (Vrijling, 2001). Secondly, the altered flow
patterns from the measures will give a morphologic response over time in the
floodplain and in the main channel. Increased floodplain inundation affects
the sediment deposition, with a mean sedimentation rate of 0.13 mm d
Adapting large and densely populated deltas to changing hydrodynamic conditions is a daunting task, especially since both river restoration and socioeconomic developments compete for space, which prevents a single-objective solution. Careful spatial planning with stakeholder involvement should benefit from the integrated assessment of possible alternatives. We presented a rule-based method for the implementation and evaluation of landscaping measures, which was used to evaluate 17 scenarios based on the type of measure, the number of stakeholders involved, and ecological design principles. We found that (1) the traditional measures of flood hazard reduction in the Netherlands, dike raising and roughness lowering, represent the most cost-effective solutions; (2) using natural grassland as the land cover after roughness smoothing and floodplain lowering provides an optimum between improved biodiversity and water level lowering; (3) the two largest stakeholders could effectively lower flood levels as they own 31 % of the groyne field plus floodplain, but the water level lowering from measures in these locations accounted for 34 % to 54 % of the lowering due to measures in the whole study area; and (4) only floodplain lowering over the whole area can compensate for the increased discharge and sea level rise at the costs of EUR 1.74 billion and the involvement of 1200 stakeholders.
Our method and its application provide decision makers and local stakeholders with (1) a wide range of measures that either requires the two largest or all of the owners in the area and (2) a standardized quantification of the trade-offs between water level lowering, ecology, and implementation costs. No single measure ranked highest on all attributes, underlining the wickedness of the problem. Our approach contrasts with the detailed analyses carried out in typical practice of river management, which normally considers a single floodplain section at a time due to the governance complexity. For these sections detailed plans are made in cooperation with stakeholders. Our setup enables fast exploration of pathways at the scale of a whole river reach, which can be adjusted by changing the rules for positioning and parameterization of the measures. The method can be transported to other regions, such as the Elbe, Mississippi, and Mekong rivers, and upscaled to the entire delta to support sustainable land use planning. Extensions of the method are required to include morphological changes, recurring costs, timing of measures, cost–benefit assessment, and vegetation succession. We argue that our results provide a common ground for any stakeholder meeting, which increases mutual understanding. Application in real-life stakeholder sessions is required to prove this point. The benefit of our approach lies in the large scale of the measures and the multiple criteria used in the evaluation, which enable higher-quality and more transparent planning with long time horizons. It also shows the future challenges and normative choices that need to be made.
Flood hazard management is embedded in a larger framework of river management as guided by international legislation (e.g., EU Flood Directive and Water Framework Directive). To achieve societal impact and adjust to changing boundary conditions, we need evidence-based and quantitative trade-offs to weigh the different stakes in society. With our approach, we are moving away from the traditional hydraulics-only analyses and towards multidisciplinary, multi-parametric, multi-objective optimizations for supporting the negotiations among stakeholders in the decision-making process.
The data used in this paper are not publicly accessible due to limitations on the hydrodynamic model input data. However, the authors can be contacted by email (m.w.straatsma@uu.nl) for assistance in acquiring data access and acknowledgement.
Nonrecurring costs of measure implementation in unit prices. The unit prices represent the 2015 price level and exclude VAT and indirect costs for engineering, design, and unexpected costs. “kEUR” denotes EUR 1000.
The authors’ contributions varied between the following four categories. The contributions to conception and design were MWS 30 %, MGK 20 %, JMF 20 %, JAHK 15 %, and FB 15 %; contributions to data collection were MWS 80 % and JMF 20 %; contributions to data analysis and conclusions were MWS 70 %, MGK 20 %, and JMF 10 %; and the contributions to the paper preparation were MWS 60 %, MGK 10 %, JMF 10 %, JAHK 10 %, and FB 10 %.
The authors declare that they have no conflict of interest.
This article is part of the special issue “Flood risk assessment and management”. It is a result of the EGU General Assembly 2018, Vienna, Austria, 8–13 April 2018.
This research is part of the research programme RiverCare, supported by the domain Applied and Engineering Sciences (AES), which is part of the Netherlands Organization for Scientific Research (NWO), and which is partly funded by the Ministry of Economic Affairs under the Perspective Programme. We thank Mitja Brilly and Holger Sandberg plus the two anonymous reviewers for their feedback, which significantly improved the paper.
This research has been supported by the NWO-TTW (grant no. P12-14).
This paper was edited by Cristina Prieto and reviewed by Mitja Brilly and two anonymous referees.