Adaptation is cost-effective to offset rising river ood risk in Europe

8 River flooding in Europe could rise to unprecedented levels due to global warming and continued 9 development in flood-prone areas. Here we appraise the potential of four key adaptation strategies to 10 mitigate flood risk across Europe based on detailed flood risk modelling and cost-benefit analysis. We find 11 that reducing flood peaks using retention areas is economically the most attractive option. In a scenario 12 without climate mitigation, they can lower projected flood losses in Europe by the end of the century from 13 42 to 7.5 €billion/year and population exposed by 81%, or achieve a risk level comparab le to today. This 14 would require an investment of 2.9 €billion/year over 2020 - 2100, with a return of 4€ for each 1€ invested. 15 The risk-reduction potential of economically-optimised strengthening of dykes is somewhat lower with 16 71% for a comparable annual investment. These measures avoid floods to happen and their cost- 17 effectiveness increases with the level of global warming. Implementing building-based flood proofing 18 measures and relocating people and assets are less cost-effective but can reduce impacts in localized 19 areas. 20


Main Text 21
Introduction 22 River floods are a major cause of damage in Europe (Wallemacq and House, 2018). Absolute losses have 23 generally increased over time mainly due to human encroachment and economic development on flood-24 prone land that resulted in a strong rise in exposure and loss of natural storage capacity (Bouwer, 2011;25 Merz et al. increasing dyke height (Ward et al., 2017). However, to find the most effective strategy, limit potential 45 negative environmental effects and avoid maladaptation, it is essential to consider a range of measures 46 (Jongman, 2018). Among those, nature-based solutions have recently gained attention as more 47 Damage reduction measures applied at building scale have a low environmental impact, are relatively 160 easy to implement and can be adapted to changing conditions (Akadiri et al. 2012). However, they cannot 161 prevent other types of flood damage, such as to transport infrastructure (Bubeck et al., 2019) or 162 agriculture (Tapia-Silva et al., 2011). As a result the average reduction in damage attainable in Europe is 163 12%, which is considerably lower than hazard-reduction measures. More importantly, because floods are 164 not avoided population exposure is not reduced (Figure 1c), even though the degree to which people are 165 affected is lowered. 166 Flood proofing is more effective (reduction in damage above 10%) in Eastern Europe countries such as 167 Estonia, Hungary and Czech Republic, because of lower protection standards and the presence of hotspots 168 of exposed assets (Table 1). Damage reduction ratios above European average are also attainable in 169 Sweden, Belgium and United Kingdom, due to the high economic exposure. Conversely, BCR values are 170 below one in the Netherlands, where country-scale high protection standards make flood proofing less 171 likely to be used (Table 1). Overall, findings suggest that flood proofing of buildings is effective for 172 protecting areas frequently exposed to low or moderate floods and with high concentration of exposed The cost-benefit analysis shows that it is the least cost-effective measure among all the adaptation 181 measures considered here ( Figure 1). The implementation across the areas with a positive benefit-cost balance would lead to an overall reduction in flood damages of just 2%, corresponding to a BCR of 2.3, 183 while each year only 4700 (1%) less people would be exposed to floods. This is because relocation is 184 economically convenient in a minority of NUTS2 regions concentrated in UK, Spain and around the Baltic 185 region (Figure 2). Costs of relocation are high as they include the demolishing of existing buildings, the 186 acquisition of new land and the construction of new infrastructures. Indeed, relocated people are 187 generally offered a partial compensation for their properties by the local government (Kick et al., 2011), 188 thus suggesting that financial incentives are necessary to promote relocation measures. In regions with 189 BCR > 1, long term flood economic damage may become comparable to and greater than the value of new 190 land and buildings, because of either low protection standards or concentration of high value assets. 191 These findings suggest that relocation can be cost-effective in localized areas, as well as for sensitive or 192 critical buildings and infrastructures frequently exposed to floods. 193 Past flood events suggest that flood relocation primarily occurs after catastrophic events for which the 194 reconstruction costs are of the same magnitude as buying a new property (López-Carr and Marter- Kenyon 195 2015). There is also a low social acceptance of relocation measures as people feel uncomfortable with 196 losing ancestral lands and properties as well as breaking long-standing ties with their communities and 197 other networks. On the other hand, relocation is the most robust long-term solution as flood risk is 198 avoided through a removal of exposure, and the land that has become available after relocation can be 199 used for the retention of flood peaks. We focused our analyses on adaptation scenarios based on the application of a single type of measure. 203 The outcomes suggest that 'hybrid' strategies, with different measures working in synergy and optimised 204 at the level of river basins are likely to be the best strategies to maximise local benefits and minimise 205 drawbacks of each measure, in line with recent findings (Du et al., 2020). For instance, it is advisable to 206 use dykes to protect against frequent low-magnitude events, and retention systems to mitigate extreme 207 flood peaks. Foreseeing backup risk-reduction measures, such as flood proofing of buildings, helps 208 minimising impacts when hazard-protection measures fail or are not sufficient to prevent flooding. 209 Integrating physical risk reduction measures with financial instruments such as insurance would further 210 reduce overall impacts on the economy and society (Kron et al., 2019). 211 The adoption of adaptation strategies should not be alternative to risk-informed land use planning. In past 212 decades, urban areas expanded considerably in flood-prone areas under increasing population pressure 213 and due to the benefits associated with settling close to river courses (Kummu et al. 2011), a trend that 214 has not slowed down even in recent years (Mård, et al. 2018). Our projections show that socioeconomic 215 growth and urban expansion will increase economic losses by more than 70% across Europe in 2100 (Table  216   S4). As such, taking into account flood risk in planning could be an effective way to reduce future flood 217 impacts. 218 The cost-benefit analysis does not include social, environmental and cultural aspects, which would require 219 more complex multi-criteria analyses (e.g. using the concept of social vulnerability as in Kind et al., 2020). 220 The inclusion of these aspects would likely improve the cost-effectiveness of nature-based solutions such 221 as retention areas, as highlighted in previous studies (EEA, 2017). 222 Local cost-effectiveness of measures can deviate strongly from those presented herein due to site-specific 223 characteristics. The present analysis is therefore not meant to replace detailed analyses at local and 224 regional scale, which are necessary for an effective and reliable design and implementation of adaptation

Climate projections 238
Projections of river streamflow with global warming are based on an ensemble of 11 bias-corrected 239 regional climate projections from EURO-CORDEX (Table S1)  to as "base", was used a reference. We consider future climate scenarios corresponding to an increase in 242 global average temperature of 1.5, 2 and 3°C above preindustrial temperature. The 1.5°C and 2°C warming 243 scenarios are explicitly considered in the Paris Agreement, while a 3°C global warming is a more realistic 244 scenario to expect by the end of the 21st century if adequate mitigation strategies are not taken. We 245 evaluate each warming scenario assuming stabilized climate from the time indicated in Supplementary 246 suggest that the effect of pathway to global warming levels is small compared to the models' variability, 251 except for strongly not time-invariant variables such as sea level rise. 252

Flood hazard and risk projections 253
We used the climate projections to generate daily streamflow simulations with Lisflood, a distributed, 254 physically based hydrological model, run at 5km grid resolution (Burek et  the Ageing Report assumes that two out of the three determinants of economic growth, technical 283 progress and capital accumulation, would reach a steady state (with constant growth rates) by the year 284 2060. That was assumed as well for the following decades. The third contributor to growth (the labour 285 input) was assumed to evolve in a proportional way with respect to population (i.e. same growth rate). 286 That means ignoring possible changes in the labour market conditions, such as the employment rate. 287 Population projections for 2061-2100 are taken from the latest United Nations demographic report 288 (medium variant), and they are explicitly considered in the computation of the economic growth figures 289 (more details can be found in Ciscar et al., 2017). 290 We represent river flood risk as expected annual economic damage (EAD) and expected annual population 291 exposed (EAPE), following the approach described by Rojas et al. (2013). For the baseline scenario, EAD 292 and EAPE are calculated by constructing impact-probability curves based on the six return periods 293 considered by flood hazard maps and taking into account local protection levels. Changes in future flood 294 impacts are derived considering the flood frequency shift for the six reference events (i.e. magnitudes 295 corresponding to a return period of 10, 20, 50, 100, 200 and 500 years under the baseline scenario) and 296 for protection levels. All economic risk estimates in this work are expressed in €2015 values. 297 We evaluated the overall reliability of the data and models composing the risk modelling framework 298 (Supplementary Information). Most of the models and datasets have been validated to some extent 299 against observed or higher resolution data in past research studies. We also compared modelled annual 300 average economic losses against reported losses retrieved from numerous sources. We find that in a 301 number of countries (such as Czech Republic, Germany, Italy, and United Kingdom) the difference 302 between modelled and reported losses is within 50%. These countries account respectively for more than 303 50% and 70% of overall modelled and reported losses. Losses are overestimated by more than 100% in 304 suitable for application within a pan-European framework (e.g. the cost to increase the height of one 314 linear kilometre of dyke by one meter). We also compiled information to clarify the link between 315 implementation costs and impact reduction (e.g. damage reduction factors reported for specific flood-316 proofing measures). We decided to include in the adaptation analysis only measures for which we found 317 sufficient information on quantitative costs (especially unit costs) and performance estimates. Table S2 provides a description of the four adaptation measures considered in this study, while Table 2 summarizes  319 the unit costs derived from the database of adaptation measures. 320

Modelling of the adaptation measures 321
Strengthening of dyke systems 322 We model the increasing of dyke height along the river network following the approach proposed by Ward  (Table 2). 331

Retention areas 332
The design and modelling of retention areas requires the development of an algorithm to allocate storage 333 areas within each river basin, based on the available storage capacity and the required level of protection. 334 We first calculate the maximum storage capacity in floodplains along the river considering agricultural 335 (excluding permanent crops, e.g. orchards, vineyards) and semi-natural (e.g. permanent grassland, 336 wetlands, excluding forests) areas within the 1-in-500-year floodplain, derived from the refined CORINE 337 Land Cover (Rosina et al., 2018). Then, we calculate flood volumes that can be accommodated by present-338 day protection standards and the flood volumes that need to be stored in each future scenario along the 339 river network. Flood volumes are estimated for each point of the river network using synthetic 340 hydrographs calculated with the Lisflood hydrological model, following the approach by Alfieri et al. (2014). Finally, the required storage volumes are calculated iteratively along the river network (i.e. design 342 minus present volumes) starting from the most upstream reaches. The iterative procedure is designed to 343 calculate the reduction of flood volumes along the river network given by upstream storage. In other 344 words, part of the flood volumes stored in a section of a river basin is subtracted from the flood volumes 345 in all downstream branches. The iterative procedure is executed separately for each design level of 346 protection, and assuming a constant return period of flood peaks in the entire river network (e.g. assuming 347 to protect the entire river basin against 1-in-100-year discharge). Implementation costs are calculated 348 based on the overall flood volume to store (Table 2). 349 In this work, we assume that the implementation of flood proofing measures can reduce overall damage 356 to exposed buildings by a specific fraction (e.g. 10%, 30% etc.), which is taken as design criterion. Using 357 the available database of adaptation measures, we relate damage reduction ratios with implementation 358 costs, by averaging data from all case studies in which flood proofing measures were applied. In other 359 words, "the analysis considers a standard/average flood-proofing implementation, based on available 360 literature information. Given the scale of application, we assume that damage reduction and costs can be 361 linearly correlated, because the measures can be applied over an increasing number of buildings. Note 362 that we excluded building elevation measures from the analysis because they are often not feasible for 363 existing buildings, and because their cost is comparable to relocation measures. We further assume that 364 infrastructural and agricultural damages cannot be reduced through flood-proofing measures of the built-365 up area, meaning that potentially, on average, 90% of the expected annual damage can be reduced. 366 Cost of flood-proofing measures are usually available at building scale. These were translated in unit costs 367 related to building surface (€/m2) using building area (if available), or assuming a standard building area 368 of 100m 2 where no information is available (Table 2). We assume that the same costs apply to all building 369 types, even though literature studies usually focus on residential buildings). We calculate implementation 370 costs as a function of the total built-up area located within the 1-in-500-year flood extent, and the damage 371 reduction ratio required. The built-up area is derived from the Global Human Settlement maps for Europe 372 (Florczyk et al., 2019). Note that we assume that population exposed is not reduced by this adaptation 373 strategy, as building-based measures do not prevent floods from occurring 374 Relocation 375 Relocation measures are designed assuming that a fraction of the exposed buildings and population 376 located in flood-prone areas are moved to a flood-safe area. We consider for relocation all built-up areas 377 located within the 1-in-500-year flood extent, for consistency with the approach adopted for all the other 378 measures. Additional tests run considering only built-up areas more frequently exposed (e.g. located 379 within the 1-in-50-year flood extent) did not show significant changes at European and country scale in 380 terms of cost-benefit analysis. We do not make any assumption about the place of destination of relocated 381 assets and people, as such decision would be highly subjective, nor do we consider possible costs for 382 resettlement (e.g. realization of infrastructure networks). We assume that implementation costs increase 383 linearly with exposure reduction, and that the exposure reduction for buildings can be used to determine 384 the reduction in population exposed (e.g. relocating 20% of buildings implies the relocation of 20% of local 385 population). 386 The evaluation of each adaptation strategy is performed using a cost-benefit analysis that optimises the 388 benefits (avoided economic damages) and the costs of implementation and maintenance over the In accordance to the literature, we assume that maintenance costs amount to 1% of total construction 396 costs (Ward et al., 2017; Aerts, 2018). Similar as the implementation cost, we assume that the effect of 397 the measures applied (protection level for dykes strengthening and retention areas, or damage reduction 398 rate for flood proofing and relocation) increases linearly from 2020 (no effect) to the design value in 2050, 399 and then remains constant. Implementation costs are calculated differently for each adaptation measure 400 as described in the Section "Modelling approach for adaptation measures". 401 For each adaptation measure we simulate different design options (e.g. raising dykes over a river stretch 402 by different height increases corresponding to a range of design return periods). For dikes strengthening 403 and building of retention areas, the optimal design level for each strategy was considered to be the one 404 providing the maximum net present value (NPV) at NUTS2 level, defined as the sum of investment costs 405 (that are negative) and economic benefits (avoided economic losses, positive) over the lifetime of the 406 project. For relocation and flood-proofing of buildings, NPV is calculated by aggregating costs and benefits 407 at 5km resolution, which corresponds to the grid used to aggregate flood impacts and derive future river 408 flow projections (Alfieri et al., 2015;Mentaschi et al., 2020). 409 Future costs and benefits are discounted to present-day values using a 5% discount rate for EU countries 410 eligible for the EU Cohesion Fund and 3% for other Member States and the UK, following the European 411 Commission's guidelines on infrastructure investments (EC, 2014). The cost-benefit analysis is applied for 412 the three warming scenarios in order to understand the performance of the adaptation options for 413 different levels of global warming. As an indication of the performance we also present the Benefit-to-414 Cost Ratio (BCR), which is the ratio of the total discounted benefits to costs. We calculate BCR values for 415 NUTS2 administrative regions, as well as countries and the EU + UK. For relocation and flood-proofing of 416 buildings, aggregation of results at NUTS2 and country level is done taking into account only 5km areas 417 with positive NPV. We further present benefits of adaptation in terms of the reduction in population 418 exposed to flooding. 419 Note that we could not quantify the environmental costs and benefits of the available adaptation 420 measures. However, we provide a qualitative assessment of these factors in the discussion of results. 421 Moreover, the reduction in population exposed was not monetized in the cost/benefit analysis, due to 422 the lack of accurate information on impacts (both physical and social) and sensitivity issues in attributing 423 economic value to human lives.