the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 18 Jun 2026
Split incentives in property-level flood adaptation across households, insurers and government in the Netherlands
Abstract. Flood adaptation can be organized across multiple spatial scales, from national flood defences to individual buildings. Property-level risk estimates drive household adaptation decisions. Yet for the Dutch housing stock, how flood risk concentrates, distributes across stakeholders, shifts under methodological uncertainty, and translates into adaptation budgets remains unquantified. For 12,992 embanked Dutch residential properties, we combine precipitation and flood defence failures across two hazard and two vulnerability approaches for 2025 and 2050. Flood damage is sharply concentrated, with the top 5 % of properties carry more than half of expected annual damage. The stakeholder split also shifts over time. In 2025 government bears 56 % of damages, insurers 32 % and households 12 %; by 2050 this becomes 13 %, 72 % and 15 %. The three-stakeholder coalition budget exceeds the household-only budget by an order of magnitude, so co-financing across stakeholders could help close the split-incentive gap. These stakeholder patterns hold across hazard and vulnerability methods, yet per-property budgets vary by an order of magnitude. These findings connect two debates usually held apart, on the credibility of property-level flood risk estimates and on how to finance adaptation under a split-incentive. Both bear on how such estimates should inform flood adaptation decisions.
Competing interests: Cees Oerlemans is affiliated with HKV, and Matthijs Kok was previously affiliated with HKV. The study uses the MijnWaterRisicoProfiel API, developed by HKV, and evaluates the BREACH method, developed by researchers holding both academic and HKV affiliations. HKV had no role in the study design, analysis, interpretation, or decision to publish, and the analysis was conducted independently. The remaining authors declare no competing interests.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.- Preprint
(994 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on egusphere-2026-3260', Anonymous Referee #1, 10 Aug 2026
-
RC2: 'Comment on egusphere-2026-3260', Anonymous Referee #2, 17 Aug 2026
The manuscript assumes universal insurance coverage for the sampled properties, with insurers covering 90% of structural and 80% of contents damage for precipitation and regional flooding, while the government is assigned 90% of damages from primary-defence failures. However, the authors acknowledge that approximately 10% of households have no flood coverage and that Wts compensation is discretionary; they also note that only approximately 60% of damage was covered by insurance and government compensation combined during the 2021 Limburg floods. These assumptions are especially important because the manuscript's main conclusion is that the household-only adaptation budget is approximately an order of magnitude lower than the three-stakeholder coalition budget. Since households are assigned only 10–20% of much of the damage by construction, part of this result follows directly from the allocation assumptions. The robustness analysis varies hazard and vulnerability models but keeps these institutional parameters fixed. I strongly recommend adding an institutional sensitivity analysis varying at least insurance take-up, reimbursement rates, deductibles, and Wts compensation/activation. The authors should then demonstrate whether the magnitude of the split-incentive remains similar under plausible alternative institutional configurations.
The manuscript deliberately couples the standard primary-defence method with the upper-bound precipitation interpretation and BREACH with the lower-bound interpretation, while the basis scenario combines an arithmetic average of the defence methods with the precipitation midpoint. The stated reason is to reduce the complete analysis from 27 combinations to nine. This coupling is problematic because the two uncertainty sources are conceptually independent. Pairing high-with-high and low-with-low creates constructed corner cases and prevents the reader from determining whether uncertainty originates from defence-failure probability, precipitation-depth discretisation, or their interaction. This is particularly important because the manuscript subsequently concludes that hazard uncertainty exceeds vulnerability uncertainty in 2025. I recommend evaluating the complete factorial combination. With approximately 13,000 properties, 27 combinations do not appear excessive relative to the importance of the uncertainty analysis. If the authors retain the reduced design, a much stronger statistical and conceptual justification is required, and the resulting scenarios should be described explicitly as bounding scenarios rather than alternative methods of equal standing.
The national pluvial maps contain the classes 0.05–0.10, 0.10–0.15, 0.15–0.20, 0.20–0.30, and >0.30 m. The manuscript then states that lower-bound, midpoint, and upper-bound values were evaluated for each class, while simultaneously acknowledging that the fifth class is unbounded. It is mathematically impossible to define a midpoint or upper bound for an unbounded class without an additional assumption. The exact numerical values assigned to the >0.30 m category must therefore be reported and justified. This issue may be particularly influential because precipitation dominates the upper-risk tail in the 2050 scenario.
The study excludes all 5 km × 5 km cells containing fewer than 300 dwellings before sampling. This procedure could systematically underrepresent rural and low-density areas, and these areas may differ in building characteristics, elevation, flood exposure, and availability of collective drainage infrastructure. Provincial proportionality alone does not demonstrate representativeness within provinces. Please report the number and proportion of national dwellings excluded by the 300-dwelling threshold, compare the sampled and national housing stocks on relevant characteristics, and explain whether sampling weights are required. Otherwise, the wording “representative sample” should be qualified.
The adaptation budget is calculated using a fixed 10-year horizon and 2.8% real discount rate, producing an annuity factor of 8.62. The conclusion that adaptation becomes financially relevant only above approximately the 80th percentile depends directly on this factor. I recommend showing sensitivity to reasonable combinations of measure lifetime and discount rate, for example shorter and longer lifetimes. The comparison between adaptation budgets and engineering cost ranges should also ensure consistent price years and comparable adaptation configurations.The adaptation budget is calculated using a fixed 10-year horizon and 2.8% real discount rate, producing an annuity factor of 8.62. The conclusion that adaptation becomes financially relevant only above approximately the 80th percentile depends directly on this factor. I recommend showing sensitivity to reasonable combinations of measure lifetime and discount rate, for example shorter and longer lifetimes. The comparison between adaptation budgets and engineering cost ranges should also ensure consistent price years and comparable adaptation configurations. Relatedly, because the manuscript studies property-level decision-making, reporting only mean budgets within broad percentile bands may obscure considerable heterogeneity. The authors should consider reporting the percentage of individual properties for which the coalition budget exceeds representative cost thresholds, rather than relying primarily on band means. This would provide a more policy-relevant measure of adaptation viability.
The EAD formulation integrates damage only from zero probability to the probability of the most frequent modelled flooding scenario. For precipitation, the most frequent mapped design event has a 10-year return period. The analysis therefore appears to assume zero relevant property damage from events more frequent than the most frequent modelled scenario. This assumption should be stated explicitly and justified. If more frequent pluvial events can produce non-zero property damage, excluding this portion of the probability curve could underestimate EAD, particularly for insurer-covered precipitation losses. Table 3 defines the maximum adaptation configuration as 0.8 m dry-proofing with a 50% wet-proof reduction, and the results and Table 5 also refer to an 80 cm barrier with 50% residual damage reduction. However, the Figure 2 caption describes the maximum configuration as 50 cm dry-proofing with 50% contents protection. The authors need to confirm exactly which configuration was implemented in the calculations and make the method, figures, captions, and tables consistent. The distinction between reducing “contents damage” and reducing “residual damage” is also substantive rather than merely editorial. A related inconsistency concerns the 0.20 m floor-height correction. The methods state that all simulated water depths were reduced by 0.20 m, whereas the Results refer to a “0.20 m precipitation height correction.” Please clarify whether the correction applies to every hazard source or only precipitation. There are also several smaller presentation issues. For example, the hazard-method section states that the API combines “three flood sources” but then lists precipitation, regional water-system overflow, regional-defence breaches, primary-defence breaches, and high water in unembanked areas. The terminology and number of flood sources should be made consistent throughout the paper. In the abstract, “with the top 5% of properties carry more than half” should be corrected to “with the top 5% of properties carrying more than half.”Citation: https://doi.org/10.5194/egusphere-2026-3260-RC2
Data sets
Data and scripts underlying: "Split incentives in property-level flood adaptation across households, insurers and government in the Netherlands" Cees Oerlemans https://data.4tu.nl/file/aead25f3-a4fe-4284-b92b-d8112d4326b7/918ed047-c8e7-4b28-9e07-22aab48975b9
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 66 | 40 | 9 | 115 | 6 | 4 |
- HTML: 66
- PDF: 40
- XML: 9
- Total: 115
- BibTeX: 6
- EndNote: 4
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
"This paper examines the modelling of flood damage to properties in the Netherlands, disaggregating losses across three key stakeholders, households, insurers, and government, and investigating how uncertainties in hazard and vulnerability models influence the resulting damage estimates. By providing expected annual damage projections for these stakeholders at two future time points (2025 and 2050), the manuscript aims to inform property-level flood adaptation strategies and address the split incentive problem in risk mitigation. The work is well-organized and clearly articulated. I find the research questions highly relevant to future climate adaptation efforts in Europe. Notably, the authors demonstrate that per-property adaptation cost estimates, which are derived from established hazard and vulnerability models for the Netherlands, are significantly affected by modelling uncertainties. These findings underscore existing limitations in current modelling approaches, which hinder effective adaptation planning for present and future risks, and strongly supports the need for further research in this domain.
However, I found that the manuscript could benefit from several improvements; in the methods section, where some modelling detailed are omitted or not explained with enough details, in the results section, where inconsistencies in the phrasing and formulation makes the arguments sometimes hard to follow, and in the discussion section, where I found some aspects pertaining to the quantification of the uncertainties to be overlooked. For instance, most of the description of the hazard models used for this study are outsourced in the references. However, some of these references are in Dutch, which limits the research’s accessibility and reproducibility for non-Dutch speakers. This includes for instance all the details on the models’ spatial resolution and climate scenarios used for future risk assessment. In addition, I found the that some uncertainties that affect the modelling framework are overlooked in the discussion. For instance, the modelling framework assumes no intensification of extreme precipitation by 2050 in Europe, despite climate projections indicating increasing intensity of heavy precipitation in Europe [1]. This assumption may therefore lead to an underestimation of future pluvial flood risk. Overall, the discussion would benefit from a clearer statement of which uncertainties are quantified, and which are not.
For these reasons, I recommend this paper to be sent back to the authors for minor revisions. I thank the authors for their contribution and wish them a good continuation of their work.
Major comments:
I found some parts of the methodology not sufficiently described. In particular, I found that the three different hazard models used in the study (primary defence breaches, regional defence breaches, precipitation) are not described with a sufficient level of details for the reader to properly understand the implications of the different modelling choices, and to understand how the different modelling components interact with each other. For instance, there is no mention of the spatial structure of the models’ outputs; are these maps, or aggregate, at which resolution? Knowing the spatial resolution of the different hazard models would help the readers to link the hazard models to the exposures model, for which spatial resolution is explicitly mentioned. Furthermore, other details pertaining modelling choices of the hazard are only mentioned allusively (e.g. Climate change effects on coastal and fluvial flooding are incorporated through scenarios with elevated external water levels.), whereas the provided reference in Dutch does not allow non-Dutch speakers to look up clarifications for these modelling details (for instance what are the specific scenarios, and elevated water levels). The manuscript would clearly benefit a more detailed description of the different hazard models, where appropriate and where the provided references are not accessible to all.
Minor comments:
Line 99-100: we drew 12,992 dwellings, a ratio of one 100 sampled dwelling per 662 in the Dutch housing stock (Table 1). Why did you choose these numbers specifically? Please clarify.
Line 130: Climate change effects on coastal and fluvial flooding are incorporated through scenarios with elevated external water levels (Rikkert et al., 2025). Please add descriptions of the modelling of climate change effects as the provided reference is not available in English.
Line 133-134: Failure probabilities for regional defences were set at one-fifth of the applicable statutory safety standard, following STOWA (2020). Why do you choose a ratio of 1/5? Please clarify as the provided reference is not in English.
Line 145-146: You increase the frequencies of the flood events but do not increase their intensities. Is that a reasonable assumption given that future precipitation events are expected to become more intense with climate change? Please explain.
Line 168: It is unclear what the Van Ederen curves refer to; please name the different curves that you will describe when introducing the paragraph, in line 165: We used two sets of depth-damage curves that are in active use among Dutch practitioners; the SSM curves and the Van Ederen curves.
Line 173-174: […] the Van Ederen vulnerability scenario in this study differed from SSM only in the structure-damage component. How do the two curves differ? Please clarify.
Line 265: It is not clear to me what shallow events means here. Do you mean pluvial flooding? Please clarify.
Line 277: what are Lorenz curves? Please describe or provide a reference.
Line 306: Insurers received the largest share of avoided damages in the P50 - P90 and P99 - P100 bands; in the P90 - P99 bands government received the largest share, because primary defence breach still drove baseline EAD there. I do not see these specific bands on any of the plots or table, can you please clarify which figure or table you are referring to when describing the results?
Line 324: The corners bracket the plausible range. This sentence is unclear to me. To which corners are you referring to? If you are referring a figure, please mention it explicitly at the beginning of the sentence or paragraph.
Figure 3: I find this figure hard to read. If possible, consider other visualizations (e.g. bar plots).
Figure 4: four range-bars compare households (HH), insurers (INS), government (GOV) and the location total. Please check consistency between the figure’s labels and the legend. What is location total? Please also check for colour-blind friendliness.
Line 333-334: per-property EAD followed insurers > government > households in eight of nine method combinations. Can this be inferred from the figure? If not, clearly state that this is not inferable from the figure or consider alternative visualization.
Line 347: Below P95, the three-way coalition budget spreads remained small relative to plausible adaptation costs. I think it might be helpful here to remind what are “plausible adaptation costs”.
Line 386: a ten-year record captures mostly frequent events even with Limburg included. Do we now the estimated return period of an event such as Limburg? If yes, including this information could be helpful for contextualization.
Line 391-393: The three-stakeholder coalition budget remained seven to nine times the household-only budget in every band where adaptation is financially relevant. Do you mean here that the household-only budget equals the three-stakeholder budget most of the time? If not, please clarify or rephrase.
References