LL2-15 hindsight check: Floods: lessons about early warning systems (Kundzewicz), Late lessons from early warnings: science, precaution, innovation (EEA Report No 1/2013), Ch. 15, pp. 347–368#

Checked 25–26 September 2026. The check covers what happened between publication (January 2013) and September 2026 that bears on the chapter’s claims, evidence, predictions and recommendations. Page numbers are report pages (PDF page minus 2). All web sources were accessed on 25 or 26 September 2026.

Author context. Zbigniew Kundzewicz stayed an active party to several of the disputes the chapter touches. - He led the review that later downgraded confidence in the kind of pan-European flood projections the chapter quotes (Kundzewicz, Krysanova, Dankers, Hirabayashi, Kanae, Hattermann et al., HSJ, online 29 Sept 2016). Several co-authors wrote the projections cited on p. 355 (Dankers; Hirabayashi). - He co-authored the 2015 defence of “stationarity is dead” against its critics (Milly et al., WRR, Sept 2015). The chapter’s statement that the stationarity assumption is “clearly incorrect” (p. 355) is therefore one side of a debate in which the author took part (Claim 6). - He was lead author of a 2014 review concluding that “it has not been possible to attribute rain-generated peak streamflow trends to anthropogenic climate change” (Kundzewicz et al., HSJ, online 20 Dec 2013).

None of this makes the chapter wrong. It does mean that some later “updates” come from the chapter’s own author, and in two cases (projections, detection) those updates are more cautious than the chapter.

Annex 3 note. Not applicable. Floods were not one of the 2001 cases, and the chapter is new in the 2013 report.

Access note. The web-search quota for this session had run out before this check began. Everything below was verified by fetching sources directly: - IPCC AR5 and AR6 chapter pages and PDFs; - OpenAlex, PubMed/PMC and Crossref for abstracts and full texts, until OpenAlex and (intermittently) Crossref rate-limited the session; - the EU Publications Office (Cellar), for Commission reports and EU legislation; - the European Court of Auditors; - the GOV.UK Content API and its published PDFs and data tables; - the Dutch Delta Programme; - Copernicus/NHESS open-access full texts; - the Spanish judiciary (CGPJ) news service for court documents on the October 2024 Valencia flood.

Several sources could not be reached: - EFAS news pages (rendered by script); - the Rhineland-Palatinate Landtag’s inquiry report on the 2021 flood (not located); - the German Federal Interior Ministry’s 2022 flood report (HTTP 400); - the archived Defra 2006 climate-change supplementary note (HTTP 405), so its figures are taken from the 2012 government guidance that reproduces them; - the UK Parliament EFRA committee report (403); - the CEH review of the 2015–16 UK floods (unreachable); - Swiss federal (BAFU, Fedlex) progress data and legal texts; - Wiley full texts (403).

Where no primary source could be reached, Wikipedia is cited and flagged as secondary.

Overview#

The chapter has aged well where it describes mechanisms, and less well where it quotes numbers or predicts the effect of a law.

What was vindicated or strengthened - Weakest-link failure of warning chains (pp. 347, 353, 359–360). The two deadliest European floods since publication reproduced the Vaison-la-Romaine and Odra patterns almost exactly, in wealthy states with good forecasts. - Germany, July 2021. Forecasts showed the extreme rainfall at least two days ahead (Mohr et al., NHESS, 6 Feb 2023), yet 29–35% of surveyed residents in the two worst-hit states received no warning (Thieken et al., NHESS, 3 Mar 2023). 190 people died, 134 of them in the Ahr valley (Rhein and Kreibich, NHESS, 10 Feb 2025). - Valencia, October 2024. The mobile alert went out at 20:11, after many people were already dead or trapped. The investigating judge described the official in charge of the emergency plan as having “important doubts about her own competences”, which “generated a paralysis at the decisive moments” (CGPJ, 6 Feb 2025, 24 Feb 2026). The court’s official death toll is 232 (CGPJ, 2 Sept 2026). - Short records and exceedance of design standards (pp. 351–352, 356). - Carlisle flooded again in December 2015, with more than 2,100 properties flooded in the city council area (National Flood Resilience Review, 8 Sept 2016). - The 2021 Ahr peaks “exceeded by far” the statistical 100-year levels but matched floods in 1804 and 1910 that had been left out of the hazard assessment (Ludwig et al., NHESS, 5 Apr 2023). - Three-quarters of the Ahr deaths occurred outside the mapped hazard zones (Rhein and Kreibich 2025). - Memory decay and the “hydro-illogical cycle” (pp. 360–361, 363). The mechanism gained independent empirical support: - flood memory in Czech settlement patterns fades within two generations (Fanta et al., Nat Commun, 7 Mar 2019); - US flood-insurance take-up spikes after floods and then “steadily declines to baseline” (Gallagher, AEJ Applied, July 2014); - English flood spending dipped after 2010/11 and jumped after the 2013–14 floods (Defra official statistics). - The summary’s claim that intense precipitation has become more frequent and intense (p. 347). IPCC AR5 said the same months later (Sept 2013). IPCC AR6 (2021) judged it likely, with a likely human contribution in Europe.

What was wrong, overstated or overtaken - The quantitative flood projections (p. 355) did not hold as numbers. - Kundzewicz and colleagues themselves concluded in 2016 that “it is naïve to expect availability of trustworthy quantitative projections of future flood hazard”. - IPCC AR6 has high confidence in increases only for Western and Central Europe, and medium confidence in decreases in the north, east and south. - The direction for Poland and northern Italy is contested. - “Robust trends unlikely for several decades” (p. 351) was too pessimistic. - Regional patterns of increasing and decreasing flood magnitudes across Europe were demonstrated by 2019 (Blöschl et al., Nature, 28 Aug 2019). - Changes in flood timing were demonstrated by 2017 (Blöschl et al., Science, 11 Aug 2017). - For the rare floods used in design (25- to 100-year events), significant trends remain no more common than chance (Hodgkins et al., J Hydrol, 2017). - AR6 still has low confidence in human influence on high river flows globally. - The Floods Directive’s promised risk reduction (Box 15.2, p. 361) remains unverified. - The European Court of Auditors found objectives “generally not quantified or time-bound” and funding “only partially identified and secured” (ECA SR 25/2018). - Eighteen years after adoption, the Commission’s own 2025 report says “it is thus difficult to conclude how effective flood risk management has been across the EU” (COM(2025) 2, 4 Feb 2025). - Detail errors in the design-allowance paragraph (p. 355). - The UK’s 2006 allowance was +20% for peak river flow from 2025 to 2115, not “after 2085” (the 2085 boundary belongs to the rainfall allowance). - Bavaria’s adopted climate factor is a uniform 1.15 for floods with return periods of 2–100 years, set in 2005 and not since updated. The “40–50%” figure for small floods does not appear in it. - Swiss restoration (Panel 15.1, p. 349). The legal commitment was real and was strengthened in 2011, but the adopted target (about 4,000 km by 2090) is about a third of the panel’s 11,000 km estimate, and delivery is “substantially slower than planned” (Buchecker et al., WSL, 30 Jan 2026).

Verdicts at a glance

How to read the lessons now. The chapter’s strongest material is its account of institutional mechanisms, which later events confirmed: - the gap between prediction and warning; - unclear legal authority to raise the alarm; - the failure of monitoring in the extremes it is meant to measure; - design standards built on short records; - memory decay after a disaster.

Its weakest material is quantitative: - the projection percentages; - the allowance figures; - the unreferenced causal claim that forecasts reduced deaths; - the untested prediction that a planning law would “considerably reduce” risk.

Claim-by-claim assessment#

Claim 1: Today’s 100-year flood becomes more frequent over more than 40% of Europe by 2071–2100, and a less-than-50-year flood over more than 30%. Increases in much of Poland, France, the UK, southern Sweden and northern Italy; decreases over most of Finland and European Russia (p. 355)#

Original claim. Citing Hirabayashi et al. (2008) and Dankers and Feyen (2008), aggregated in Kundzewicz et al. (2010b): “in aggregate terms, the control 100-year flood is projected to become more frequent over more than 40 % of Europe, and … over 30 % of Europe the mean recurrence interval of such floods is projected to decrease from 100 years to below 50 years by 2071–2100” (p. 355). The same page concedes that “no precise quantitative information can be delivered for long-term flood preparedness planning”.

Subsequent developments

Newer pan-European projections went further in the same direction: - Using an ensemble of the EURO-CORDEX regional climate projections under RCP8.5, the JRC projected that “on average, in Europe, flood peaks with return periods above 100 years are projected to double in frequency within 3 decades” (Alfieri et al., HESS, 11 May 2015).

The author and co-authors of the cited studies then downgraded confidence in the numbers: - Kundzewicz et al. (HSJ, online 29 Sept 2016) compared European flood projections and found discrepancies that “raise caution”. They wrote that “it is naïve to expect availability of trustworthy quantitative projections of future flood hazard”. They also concluded that “it does not seem possible to recommend which large-scale studies may be considered most credible in particular areas of Europe”. - The same author’s 2014 review, following IPCC SREX, found “only low confidence in numerical projections of changes in flood magnitude or frequency” (Kundzewicz et al. 2014).

IPCC AR6 (Working Group I, 2021) gives the current mainstream assessment: - Regional chapter (Ch. 12, Europe): - “There is high confidence of river floods increasing in Western and Central Europe (WCE) and medium confidence of a decrease in Northern (NEU), Eastern (EEU) and southern Europe (MED) for mid- and end-century under RCP8.5 and low confidence under RCP2.6.” - The projected change in the 100-year peak flow is roughly +10% by mid-century (+18% by end of century) in WCE, and −5% (−11%) in NEU, based on Di Sante et al. (2021) (AR6 WGI Ch. 12). - The chapter also reports JRC results showing “Europe is one of the regions where the largest increases in flood risk may occur, with only few countries in Eastern Europe showing a decrease (Poland, Lithuania, Belarus)”. - Extremes chapter (Ch. 11): - Global hydrological models give “decreasing frequency or magnitude … for central and eastern Europe and the Mediterranean (high confidence)”. - There is “low confidence in projected regional changes” overall (AR6 WGI Ch. 11). - The two AR6 chapters sit uneasily together on central Europe. The regional chapter is the more specific European assessment.

Observations since 2013 point the same broad way: - Floods increased in north-western Europe and decreased in southern and eastern Europe over 1960–2010. The observed changes are “broadly consistent with climate model projections for the next century” (Blöschl et al., Nature, 28 Aug 2019).

Region by region against the chapter: - UK and France: consistent. Both fall in the AR6 region with high-confidence increases, and in Blöschl’s north-western increase. - Finland and European Russia: consistent. AR6 gives medium-confidence decreases in the north and east, linked to reduced snowmelt floods. - Poland: contested. - The JRC projections reported in AR6 show a decrease, and Blöschl et al. (2019) observed decreasing floods in eastern Europe. - A bias-corrected catchment study projected increases in flood indices for most Polish catchments studied (Osuch et al., SERRA, online Aug 2016). - Northern Italy and southern Sweden: unresolved in the sources retrieved.

Verdict: partly held up. The direction of change for Atlantic and western-central Europe, and the northern and eastern decreases, match later assessments. Later JRC work projects even larger increases than the chapter implied. But the headline percentages rest on two early studies whose authors later said such numbers cannot be trusted at regional scale. AR6 assigns only low confidence to regional flood projections in general, and medium to high confidence only by broad region. Poland, one of the chapter’s named examples of increase, is now more often projected to see decreases.

Implication for weight. Use the chapter for direction and for its own caveat (“no precise quantitative information”), not for the 40%/30% figures. For current numbers cite AR6 Ch. 12. The author’s own 2016 reversal is a useful, fair illustration of how the confidence in projections changed after the event.

Claim 2: No conclusive, widespread climate-change trend in observed high river flows; Wilby et al. (2008) speculate robust trends are unlikely to appear for several decades (p. 351). The unattributed summary goes further: “Intense precipitation has become more frequent and more intense” (p. 347)#

Original claim. “No conclusive and ubiquitous climate change trend in the flood behaviour has been found, based on the global data on high river flows observed so far” (p. 351). Because the climate signal is weak against natural variability and land-use change, “Wilby et al. (2008) speculate that statistically robust trends are unlikely to be apparent for several decades” (p. 351). The summary asserts that intense precipitation “has become more frequent and more intense” (p. 347).

Subsequent developments

Heavy precipitation: the summary was in line with the IPCC then and since: - IPCC AR5 (SPM approved September 2013): “The frequency or intensity of heavy precipitation events has likely increased in North America and Europe” (AR5 WGI SPM). - IPCC AR6 (2021): heavy precipitation “has likely increased on the continental scale” over Europe, and “it is likely that human-induced climate change has contributed” there. Trends are detected with high confidence for northern Europe and the Alps, and attributed with high confidence in northern Europe (AR6 WGI Ch. 11, Executive Summary; Ch. 12, 12.4.5).

River floods: detection came in stages: - 2013–2014, the chapter’s position restated. The author’s 2014 review: “It has not been possible to attribute rain-generated peak streamflow trends to anthropogenic climate change over the past several decades” (Kundzewicz et al. 2014). A European state-of-the-art review called for new approaches using long records and flood-rich periods (Hall et al., HESS, 30 July 2014). - 2017: timing. Using 4,262 stations: “no consistent large-scale climate change signal in observed flood magnitudes has been identified so far”. But the authors found “clear patterns of change in flood timing”, for example earlier snowmelt floods in north-eastern Europe (Blöschl et al., Science, 11 Aug 2017). - 2017: major floods. Across more than 1,200 minimally altered catchments in North America and Europe, “the number of significant trends in major-flood occurrence … was approximately the number expected due to chance alone”. The changes “were dominated by multidecadal variability rather than by long-term trends” (Hodgkins et al., J Hydrol, 2017). - 2019: magnitude, regionally. The study reports “clear regional patterns of both increases and decreases in observed river flood discharges in the past five decades in Europe, which are manifestations of a changing climate”. Trends range from about +11% to −23% per decade (Blöschl et al. 2019). - 2020: historical context. The past three decades were “among the most flood-rich periods in Europe in the past 500 years”, unusually warm and with more summer floods (Blöschl et al., Nature, 22 July 2020). - 2021: global attribution. Using 7,250 gauges for 1971–2010, the observed trends were reproduced only when anthropogenic forcing was included (Gudmundsson et al., Science, 12 Mar 2021). - AR6 (2021) remained cautious. It found “low confidence about peak flow trends over past decades on the global scale”, though with increases in north-western Europe. It also found “low confidence in the human influence on the changes in high river flows on the global scale” (Ch. 11, 11.5). The European chapter states “high confidence of an observed increasing trend of river floods in Western and Central Europe” (Ch. 12). - Event attribution became routine. - For July 2021, climate change had increased the likelihood of the heavy rainfall by a factor of 1.2–9 and its intensity by 3–19% over the wider western European region (Tradowsky et al., Climatic Change, 29 June 2023). - A storyline study attributes a 21% increase in the 6-hour rainfall rate of the October 2024 Valencia storm to present-day warming (Calvo-Sancho et al., Nat Commun, 17 Feb 2026).

Verdict: partly held up. - The p. 351 statement was an accurate summary of 2013 evidence. - The summary’s precipitation claim was consistent with AR5 and was strengthened by AR6. It was not in fact inconsistent with the body, because heavy rainfall and river floods are different variables. - Wilby et al.’s “several decades” was too pessimistic for regional patterns in annual flood peaks and for timing, which emerged within four to six years of publication. - It was about right for the rare floods that matter for design, and for formal attribution of river-flood trends at global scale. AR6 still rates both low confidence.

Implication for weight. The chapter’s underlying lesson gained support: waiting for statistically robust detection would have meant decades of delay, so act on early warnings (pp. 351, 355). Bavaria (2005) and the UK (2006) adopted climate allowances 13–14 years before regional detection (2019), and the detection that followed confirmed the direction in north-western Europe. The claim about low signal-to-noise should be cited with the Hodgkins et al. qualification: detection depends on which floods you look at.

Claim 3: The EU Floods Directive (2007/60/EC) is “probably the most advanced flood protection and preparedness legislation worldwide”, and its implementation “should considerably reduce flood risk throughout the 27 EU Member States”; codifying preparedness “helps overcome the hydro-illogical cycle” (Box 15.2, p. 361)#

Original claim. The directive’s mandatory assessment, mapping and planning “are expected to result in an unprecedented multinational upgrading of preparedness systems” (Box 15.2, p. 361). The chapter adds that the directive leaves Member States flexibility over the level of protection and the timetables for measures.

Subsequent developments

The European Court of Auditors (Special Report 25/2018, adopted 19 Sept 2018; PDF): - Positives. The directive “had positive effects overall”: better coordination between the Commission and Member States, “standardisation of flood risk assessment and management”, and raised awareness. - Weaknesses: - objectives in the flood risk management plans (FRMPs) “generally not quantified or time-bound”; - “sources of financing … only partially identified and secured”; - ranking procedures that “did not allocate money in accordance with the priorities established”; - Member States visited “could not factor in the impact of climate change” and “generally used historical data”; - where Member States relied on private flood insurance, “coverage remained low”; - spatial-planning rules “more to do”; - two-thirds of visited plans did not focus on green infrastructure.

The Commission’s fitness check of EU water law (SWD(2019) 439, 10 Dec 2019; EUR-Lex): - “There is no information yet on the state of implementation of measures included in the first FRMPs or on the degree of flood risk reduction that has been achieved.” - A stakeholder focus group concluded “it is still too early to know whether the Directive has been entirely successful as it is somewhat dependent on the occurrence of flood events to test the modelling and measures employed”.

The Commission’s 7th implementation report, covering the second FRMPs for 2022–2027 (COM(2025) 2, 4 Feb 2025; EUR-Lex): - Progress. - “Notable progress in flood risk management has been achieved throughout the EU since the introduction of the 2007 Floods Directive.” - All Member States considered climate change in their second preliminary flood risk assessments, against about half in the first round. - Nearly all Member States now publish hazard maps through online map viewers. - But: - “The plans include progress on implementing the measures rather than the progress towards targets … It is thus difficult to conclude how effective flood risk management has been across the EU.” - Only 21 Member States reported in time to be assessed. - Legal links between spatial planning and flood risk management appeared in only 8 of the 21. - The Commission again recommends that objectives “be specific, have a deadline where possible and be linked to quantitative progress indicators”.

Outcomes context (not a test of the directive): - EU losses from weather- and climate-related extremes were EUR 822 billion over 1980–2024, 25% of it in 2021–2024. Floods account for 47% of the total (EEA indicator, 14 Oct 2025). - The deadliest floods since publication struck Member States that had FRMPs in place: Germany in 2021 and Spain in 2024. - Without a counterfactual, none of this shows the directive failed. It shows that its effect on risk is unmeasured.

Related legislation that codified warning, not planning: - The European Electronic Communications Code requires Member States with public warning systems to ensure that, by 21 June 2022, warnings are transmitted to mobile users (Directive (EU) 2018/1972, Art. 110; EUR-Lex). Claim 4 describes how this worked in Valencia.

“Most advanced worldwide”: No later assessment retrieved compares the directive with other jurisdictions, so this remains an untested comparative judgement.

Verdict: partly held up. - The process predictions were met across the EU: assessment, mapping, planning, cross-border coordination, and taking climate change into account. - The substantive prediction, that implementation “should considerably reduce flood risk”, has not been verified. The Commission itself could not verify it in 2019 or 2025, and the ECA identified design features (unquantified objectives, unsecured funding) that make it hard to verify. - The claim that writing preparedness into law “helps overcome the hydro-illogical cycle” is also unverified. A legal duty to plan did not secure the funding the plans needed.

Implication for weight. Treat Box 15.2 as advocacy made at the start of implementation. It is useful as an example of how a process-based law is judged on process because it does not set measurable outcome targets. The later record supports a narrower lesson: codification harmonises and makes risk visible, but without quantified objectives and ring-fenced funding it cannot be shown to reduce risk.

Claim 4: Forecasts have “advanced markedly” while warnings have “lagged behind”; warning systems fail at their weakest link, including legal authority to warn and coordination between institutions (Vaison-la-Romaine 1992; Odra 1997) (pp. 347, 353, 359–360)#

Original claim. At Vaison-la-Romaine, Météo-France issued accurate alerts 12 and 24 hours ahead, “but because the local authorities did not know which catchment would be affected no warning was issued” (p. 353). On the Odra, “low-level authorities were not entitled to announce a flood alert”; responsibilities were “ambiguous and conflictual”; civil defence was “geared to act in the event of a war”; and 189,000 telecommunication links were cut (Box 15.1, p. 360). “It is often noted that forecasts have advanced markedly, while progress in warnings has lagged behind” (p. 353).

Subsequent developments

Warning chains can improve, as Germany showed between 2002 and 2013: - Households reporting no warning fell from 27% (2002) to 5% (2013), and companies from 45% to 3%. - The authors warned that “the challenge is to continuously maintain and advance an integrated early warning and emergency response system even without the occurrence of extreme floods” (Kreibich et al., NHESS, 29 Nov 2017).

Germany, July 2021 (peer-reviewed analyses): - The forecasts were good. - The DWD’s regional model “show[ed] the potential for a widespread heavy precipitation event … as early as 12 July 00:00 UTC”. High 24-hour totals “within the range of the observations” were predicted “more than 2 d … ahead of the event”. The convection-permitting model “captured the magnitude of the event right from … [its] first initialization”. - ECMWF’s extreme forecast index flagged the event from 10–11 July. - Yet “the population was informed too late or insufficiently about the extent of the imminent disaster” (Mohr et al., NHESS, 6 Feb 2023). - EFAS and the DWD had both forecast the triggering low-pressure system “since Sunday 11 July 2021” (Thieken et al., NHESS, 3 Mar 2023). - The warnings did not reach people or convey the danger (Thieken et al. 2023, survey of 1,315 residents): - 35% of respondents in North Rhine-Westphalia and 29% in Rhineland-Palatinate “did not receive any warning”; - of those warned, 85% “did not expect very severe flooding”; - in Ahrweiler, only about 18% of residents had subscribed to the KATWARN app; - the app’s 19:35 warning of more than 5 m at Altenahr “considerably exceeded the 100-year flood level of around 3.7 m”. - Legal authority and cost sat with the wrong actor. Thieken et al. report that “a state of emergency was declared too late and … evacuations of heavily affected settlement areas were initiated too late”. In most German states this declaration “is the responsibility of the district administrator, since in most cases the district also has to bear the incurred costs”. There is “no mandatory training in disaster management for district administrators, who are elected politicians”. Saxony, by contrast, has “an automatic declaration of a state of emergency if flood forecasts exceed the highest warning level”. - Timing and cost in lives. In Altenahr, the state of emergency and a 50 m evacuation order came at 23:09, “although the floodwater was already dangerously high”. The main causes of the 134 Ahr deaths were “the extreme severity of the flood and its underestimation by the population and authorities, as well as inadequate early warning and evacuation” (Rhein and Kreibich, NHESS, 10 Feb 2025). - Monitoring failed in the extreme. The Altenahr gauge “was completely destroyed during the flood” and its record had to be reconstructed (Mohr et al. 2023). This is a direct replay of the flooded Wrocław information office (p. 360). - Communications failed. Power and telecommunications broke down (Thieken et al., citing Koks et al. 2022). - Official follow-up (secondary source: German Wikipedia, “Hochwasser in West- und Mitteleuropa 2021”; the primary documents were not reached): - Koblenz prosecutors investigated the Ahrweiler district administrator for negligent homicide by omission and announced in April 2024 that no charges would be brought; - the Rhineland-Palatinate parliamentary inquiry published a final report of more than 2,100 pages on 2 August 2024; - in July 2025 the state interior ministry’s preliminary disciplinary findings alleged serious breaches of duty. - Remedies. Germany introduced cell-broadcast alerts in February 2023 (Thieken et al. 2023) and funded replacement of sirens dismantled after the Cold War (secondary, same Wikipedia article). The Cold War detail echoes the Odra civil defence “geared to act in the event of a war”.

Valencia, 29 October 2024 (primary court documents from the Spanish judiciary, CGPJ): - The alert came too late for many victims. On 6 February 2025 the investigating court at Catarroja recorded that victims were unaware of “la situación de altísimo riesgo” (the extremely high-risk situation) they faced from the overflowing Poyo ravine. - Many elderly victims died on the ground floors of their homes. - Others were in the streets or fetching cars from garages “cuando el mensaje de alerta … se envió a los teléfonos móviles a las 20.11 horas” (when the alert message was sent to mobile phones at 20:11). - The court asked the regional government who decided to send the alert. It also asked the river basin authority (CHJ) for Poyo flow data between 16:13 and 18:42 and how that data was transmitted (CGPJ, 6 Feb 2025). - Competence rested with the region. On 19 June 2025 the Valencia Provincial Court upheld the investigation of the former regional minister responsible for emergencies. It found she “contaba o podía contar con la información existente sobre los caudales” (had or could have had the existing information on the flows). It noted “la atribución legal del mando único de la emergencia a la Administración autonómica” (the legal assignment of the emergency’s single command to the regional administration) (CGPJ, 19 June 2025). - Doubt about competence produced paralysis. The investigating judge’s reasoned statement of 24 February 2026 (more than 100 pages) said: - the director of the emergency plan “tenía en un principio importantes dudas sobre sus propias competencias” (initially had important doubts about her own competences); - “esa indecisión, ese desconocimiento de aquello que podía acordar o no, generó una parálisis en los momentos decisivos” (that indecision, that ignorance of what she could or could not decide, generated a paralysis at the decisive moments); - there was a “flagrante falta de coordinación y cooperación entre las distintas Consellerias” (a flagrant lack of coordination and cooperation between the regional ministries) (CGPJ, 24 Feb 2026). - The courts separated system failure from individual guilt. On 16 March 2026 the regional High Court unanimously declined to investigate the former regional president. It found he did not hold the specific legal duty required for homicide by omission, and it stressed that it could judge only criminal, not “moral o político”, responsibility. The case against the former regional minister and her secretary for emergencies continues (CGPJ, 16 Mar 2026). - Official toll: 232 deaths (CGPJ, 2 Sept 2026). - The national forecast came early (secondary source: English Wikipedia, “2024 Spanish floods”, citing AEMET’s timeline). The national meteorological agency issued red warnings at 07:31–07:36 that morning, about 12.5 hours before the mobile alert. The phone-alert system mandated by EU law (Claim 3) existed; the decision to use it was the weak link.

The Commission now states the chain explicitly. “Indispensable elements of flood risk management are reliable forecasting and early warning systems to promptly activate civil protection measures” (COM(2025) 2).

Verdict: strengthened. Both events reproduce the chapter’s cases closely: - accurate regional forecasts without timely local warnings (Vaison-la-Romaine); - confusion over who had legal authority to act, and poor coordination between institutions (Odra); - monitoring and communications failing in the extreme (Wrocław).

This happened in two of Europe’s richest regions, a decade after the chapter, and with EFAS and national forecasting operating. The 2002–2013 German improvement shows the lesson is not fatalistic: chains can be strengthened, but they also decay.

Implication for weight. This is among the best-supported lessons in the chapter and can carry substantial weight. When citing it, add three post-2013 refinements: 1. Who pays for acting on a warning, as well as who is legally entitled to act, shapes the decision to warn (the German district-cost rule). 2. Automatic triggers are one design response (Saxony). 3. Mandated dissemination technology does not remove the human decision bottleneck (Valencia’s alert system).

Claim 5: Improvements in the lead time and accuracy of forecasts have “already” reduced flood fatalities in many countries (p. 364)#

Original claim. “Thanks to improvements in the advance time and accuracy of forecasts, it has already been possible to reduce the number of flood fatalities in many countries” (p. 364). The claim is unreferenced.

Subsequent developments

The trend exists in Europe: - Across 37 countries from 1870 to 2016, “after correcting for changes in flood exposure, there has been … a substantial decrease in flood fatalities” (Paprotny et al., Nat Commun, May 2018). - The decline is about 1.4% a year since 1870 and about 4.3–4.7% a year since 1950. - Deaths from flash floods fell more than deaths from river floods. - The long-run trend is statistically significant mainly for 1950–2016. - A 2024 follow-up using 1950–2020 data finds “a strong decline in flood vulnerability over time … for all three indicators of relative losses, suggesting improved flood adaptation” (Paprotny et al., Nat Hazards, 28 Nov 2024). - Globally, “rising per-capita income coincided with a global decline in vulnerability between 1980 and 2010” (Jongman et al., PNAS, Apr 2015).

The cause is not established: - Paprotny et al. (2018) list possible explanations: “vast improvements in communication and transportation”, and “the establishment of meteorological and hydrological agencies, which allowed for continuous observation and forecasting …, improved early warning and disaster preparedness”. - They do not test these explanations, they assume constant vulnerability when normalising, and they report “large underreporting of smaller floods”. - For EFAS, the modelled monetary benefit is “of the order of 400 Euro for every 1 Euro invested” (Pappenberger et al., Env Sci Policy, 2015). That figure concerns avoided damage, not deaths.

Counter-evidence since 2013: - The two worst European flood death tolls in decades, 190 in Germany in 2021 and 232 in Valencia in 2024, occurred where forecasts were good (Claim 4). - The 2021 deaths were concentrated among people over 60 (78%), and 75% occurred outside mapped hazard zones (Rhein and Kreibich 2025). - A new Euro-Mediterranean database records 3,737 flood deaths from 1980 to 2024 across 16 territories (FFEM-DB 2.0, Petrucci et al., Sci Data, 9 June 2026). It notes that death rates in 2010–2020 “differed markedly across local contexts”, rising up to 15-fold in highly vulnerable regions. The data descriptor does not report a trend.

Verdict: partly held up. Long-run declines in normalised flood deaths are real and well documented. Forecasting is a plausible contributor. But the chapter states a causal claim that no later study retrieved here has demonstrated, and the 2021 and 2024 disasters show forecast skill does not translate into fewer deaths without warning and response. That is the chapter’s own Claim 4.

Implication for weight. Cite Paprotny et al. for the trend and treat the causal attribution to forecasting as plausible but unproven. The p. 364 claim sits in tension with p. 353. The later record resolves the tension in favour of p. 353: forecasting is necessary, not sufficient.

Claim 6: Design based on an unchanging climate is “clearly incorrect”; Bavaria, the UK and the Netherlands have written climate allowances into design (p. 355)#

Original claim. - “Systems continue to be designed and operated assuming stationarity … Since this assumption is clearly incorrect … existing design procedures need to be revised” (p. 355, citing Milly et al. 2008). - Bavaria: design values take account of “a 40–50 % increase in small and medium flood discharges by 2050, and an increase of around 15 % in 100-year floods”. - UK: Defra (2006) includes “10 % up to 2025 and 20 % after 2085”. - Netherlands: by 2015 measures “should increase the design discharge from 15 000 to 16 000 m3/s and this should increase further to 18 000 m3/s in the longer term” (p. 355).

Subsequent developments

UK: the chapter misreports the 2006 allowance, and the regime has since been revised twice. - The 2006 figures. The 2012 government guidance reproducing Defra’s October 2006 FCDPAG3 supplementary note gives: - peak river flow: +10% for 1990–2025 and +20% for 2025–2115; - peak rainfall intensity: +5%, +10%, +20% and +30% for 1990–2025, 2025–2055, 2055–2085 and 2085–2115 (Technical Guidance to the NPPF, Table 5, 27 Mar 2012). - The chapter’s “20% after 2085” therefore mixes the flow and rainfall rows. The 20% flow allowance applied from 2025. - The current regime. The Environment Agency now issues peak-flow allowances by management catchment, derived from UKCP18, in central, higher-central and upper-end percentiles. These apply to projects from 20 July 2021. The guidance defines “non-stationarity” and points to Environment Agency guidance on non-stationary flood-frequency estimation (EA guidance, first published 22 July 2020, updated 17 May 2022). - Stress tests. The 2016 National Flood Resilience Review stress-tested flood maps against “plausible” extreme rainfall with 20–30% uplifts (NFRR, 8 Sept 2016).

Bavaria and Baden-Württemberg: allowances adopted early but rarely updated (Macdonald et al., J Flood Risk Manag, 9 Feb 2026; figures read from the authors’ preprint): - Bavaria uses “a uniform climate change factor of 1.15 … for all return periods between 2 and 100 years” (1.075 for the 200-year flood), derived in 2005. - Baden-Württemberg uses factors of 1.25–1.75 for the 2-year flood and 1.15–1.25 for the 100-year flood, varying by region. - “No legally binding update cycle is implemented for the climate change factors in the two German federal states.” - The chapter’s “40–50% increase in small and medium flood discharges” therefore does not match Bavaria’s adopted design factor. It may reflect projections behind the factor, or Baden-Württemberg’s graded factors. The 15% figure for the 100-year flood is correct.

Netherlands: a move to risk-based legal standards, with implementation lagging: - New standards. Legally binding flood-protection standards for all primary defences came into force on 1 January 2017. They are expressed as failure probabilities from 1/300 to 1/100,000 a year, and every levee must meet them by 2050 (Macdonald et al. 2026). - Climate-adjusted design values. These are used for the Rhine and Meuse. At the Lobith gauge, the discharge factors for the worst climate scenario in 2085 range from 1.4 (the 1-in-10-year flood) to 1.15 (the 1-in-10,000-year flood). Guidelines are revisited roughly every seven years. - The chapter’s discharge targets. The earlier design-discharge framing that the chapter’s 16,000 and 18,000 m³/s figures belong to has been replaced by these risk-based standards. The Room for the River programme ran 2006–2015 and aimed at 16,000 m³/s (secondary: Wikipedia). The 18,000 m³/s long-term target was not verified in any primary source retrieved. - Progress (Delta Programme 2025, 17 Sept 2024): - “Sixty-two percent of the primary flood defences do not yet comply with the standards that have been applicable since 2017”, which the programme says “is as expected” because the standards target 2050; - about 2,000 km of dike upgrades are needed; - “between the start of the HWBP in 2014 and year-end 2023, 219 kilometres of dikes were declared safe, of which 53 kilometres were actually strengthened”; - “implementation is currently lagging behind schedule”; - a “renewed Room for the River programme” is in preparation.

How widely allowances have been adopted: - Of 20 national and regional authorities in Germany, the Netherlands, Belgium and Luxembourg, “all … have adaptation plans … almost all regions have developed future flood projections but only three incorporate them into climate-adjusted design values”: the Netherlands, Bavaria and Baden-Württemberg. Saxony argues “the spread of projections is too large to derive design values” (Macdonald et al. 2026). - The ECA (2018) found that Member States visited “generally used historical data”. - The Commission (2025) still has to urge “ensuring that flood prevention measures are dimensioned on future flood conditions”.

The stationarity dispute, which the chapter does not acknowledge, continued: - Critics: - “Stationarity is immortal!”, arguing that any credible model needs a stationary random component and that “an unnecessary resort to nonstationarity may imply a reduction of predictive capabilities” (Montanari and Koutsoyiannis, WRR, Nov 2014); - nonstationarity should be invoked only when change “can be predicted in deterministic terms”, and past records should be “stationarized” to represent future conditions (Koutsoyiannis and Montanari, HSJ, online Sept 2014); - nonstationary frequency models can add so much uncertainty that “stationary models … should be retained as more theoretically coherent and reliable options” unless a deterministic mechanism is identified (Serinaldi and Kilsby, Adv Water Resour, 2015). - Defence. The reply “On Critiques of ‘Stationarity is Dead’” was co-authored by Kundzewicz (Milly et al., WRR, Sept 2015). - Practice. Practice has settled between the camps: projections are used as multiplicative allowances on historically estimated design values (“stationarizing” in Koutsoyiannis’s terms), with percentiles, stress tests and adaptive pathways. Fully nonstationary statistics are rarely used.

Verdict: partly held up. - The chapter’s recommendation was taken up in the UK and the Netherlands, both of which strengthened it after 2013, and it has survived there. The ECA and the Commission endorse it. - Adoption beyond these pioneers remains patchy. - Bavaria’s 2005 factor has not been updated in 20 years. - The Dutch programme is behind schedule and underfunded (Claim 8). - Two of the chapter’s three illustrative figures are misstated. - “Clearly incorrect” overstates a position that remained contested in the hydrological literature, and the author was a party to that dispute.

Implication for weight. The lesson “do not design on the assumption that the past will repeat” deserves weight. The strong version, “stationarity is dead”, should be cited as contested. Check the chapter’s allowance figures before reuse.

Claim 7: After the January 2005 flood (return period above 150 years), Carlisle’s planned 100-year defences were redesigned to a 200-year standard (p. 356)#

Original claim. The 2005 flood occurred while plans for 100-year defences were out for consultation; the defences “would have been breached because the 2005 event had a return period in excess of 150 years. After the event, the defences were therefore redesigned to address a 200-year return period” (p. 356). The chapter also records the Cumbria floods of 2009 (p. 358), and notes that design can be breached “by the next flood event (or, as in Carlisle in January 2005, before the new design has been implemented)” (p. 359).

Subsequent developments

Carlisle flooded again in December 2015 (Storm Desmond): - Scale. “More than 2,100 properties were flooded in the area of Carlisle City Council, and over 400 properties were flooded in the Eden catchment upstream”. Flood depths in Carlisle exceeded 2 m. The review lists major Carlisle floods in 1771, 1822, 1856, 1925, 1931, 1968, 2005 and 2015 (National Flood Resilience Review, 8 Sept 2016). - Record flows. The government’s Cumbria plan records that “the Eden experienced the highest flow levels, recorded on any river in England; following the highest day of rainfall recorded” (Cumbria Flood Action Plan overview, 30 June 2016). Its Carlisle table puts the impact at “approximately 1,900 homes and businesses” (Carlisle community action table). The difference from the review’s 2,100 presumably reflects the area counted. - A new scheme. In July 2019 the Environment Agency began a GBP 25 million Carlisle scheme “increasing protection from the previous scheme completed in 2005 to a 0.5% chance of flooding in any one year”, which is the 200-year standard (EA, 31 July 2019). Phase 2 followed in 2020 (EA, 31 July 2020). - An unresolved discrepancy. The EA’s 2019 wording implies that the defences in place in 2015 did not provide the 200-year standard the chapter says they were redesigned to. It also dates the “previous scheme” to 2005. I could not retrieve a primary document (such as the Cumbria County Council Section 19 report) that settles what standard the post-2005 defences were built to, or whether they were overtopped in 2015.

The general lesson gained independent support: - Historical records. Adding historical flood estimates since 1800 to Eden gauge data from 1967 cut the 95% confidence intervals for floods rarer than about 75 years by roughly half (Parkes and Demeritt, J Hydrol, July 2016). This is a direct remedy for the chapter’s “records too short” diagnosis (pp. 351–352). - The Ahr, 2021. Peak discharges “exceeded by far the statistical 100-year return levels” but were “comparable to the reconstructed major historical events of 1804 and 1910, which were not included in the flood risk assessment so far” (Ludwig et al. 2023). 75% of the Ahr deaths occurred outside officially mapped hazard zones (Rhein and Kreibich 2025). - A global sample. Across 45 paired flood and drought events, risk management “generally reduces the impacts … but faces difficulties in reducing the impacts of unprecedented events”. “If the second event was much more hazardous than the first, its impact was almost always higher”, because events “exceeded the design levels of levees and reservoirs” (Kreibich et al., Nature, 3 Aug 2022).

Verdict: strengthened. The belief in the reader’s note is confirmed: Carlisle flooded again in December 2015, a decade after the event the chapter describes, on a record flow for any English river. The pattern the chapter drew from Carlisle recurred on a larger scale in 2021: design estimates from short records were exceeded by events the historical record already contained.

Implication for weight. Strong support for the lesson that standards based on short records will be exceeded. Do not repeat the chapter’s specific statement that the post-2005 defences were built to 200 years without checking the Carlisle Section 19 report.

Claim 8: The “hydro-illogical cycle” (heavy investment after a disaster, then forgetting and cuts) is “a general principle, valid across different political and economic systems”; the Dutch coastal record shows a collective memory of about 50 years and surprise floods roughly once a century (pp. 360–361, 363)#

Original claim. After a destructive flood “heavy expenditure follows … After some time without flooding, willingness to pay for flood preparedness decreases sharply and projects are downscaled or suspended”. This is presented as “a general principle, valid across different political and economic systems”, because return periods exceed political horizons (pp. 360–361). Vellinga and Aerts: “after some 50 years the flood disaster tends to disappear from the collective memory and flood protection measures are insufficiently maintained” (Panel 15.3, p. 363).

Subsequent developments

Empirical evidence of memory decay: - Settlement location. In 1,293 Czech settlements founded over nine centuries, “for a period of one generation after each flood, new settlements appeared in safer places. However, respect for floods waned in the second generation”. The authors conclude that flood memory “fades away already within two generations” (Fanta et al., Nat Commun, 7 Mar 2019). This is consistent in order of magnitude with the Dutch ~50-year claim. - Insurance take-up. US flood-insurance take-up “spikes the year after a flood and then steadily declines to baseline”, which fits a learning model “that allows for forgetting” (Gallagher, AEJ Applied, July 2014). - Warning sign-ups. Sign-ups in England spike during events. During Storm Babet (October 2023) the warning service “recorded its busiest day ever – more than 5,500 new users registered” (EA, NaFRA 2024, 17 Dec 2024). - Paired floods. Where a second flood caused less damage, this was “mainly due to significant reductions in vulnerability”, and “there remains the challenge to stimulate measures … in periods in which extreme events do not occur” (Kreibich et al., Earth’s Future, 26 July 2017). - Reactive management. In both Germany and England, flood risk management “is still a reactive, event-driven process” (Surminski and Thieken, Earth’s Future, Aug 2017).

Spending series, England (official statistics, real terms at 2024/25 prices; Defra FCERM funding data tables, updated 15 July 2026): - Total FCERM expenditure:

• Reading the series. The rise after 2007/08, the dip after 2010/11 and the spike in 2014/15 follow the summer 2007 floods, the 2010 fiscal consolidation and the winter 2013–14 floods. The timing matches the cycle, but that causal reading is mine; the statistics do not state causes. After 2015/16 the series rises fairly steadily under six-year capital programmes, which suggests multi-year commitments can damp the cycle.
• Maintenance lags. The share of high-consequence flood assets at or above target condition fell from 97.9% (Q4 2018/19) to 92.1% (Q2 2024/25), then 93.0% (Q4 2025/26) (same tables; there is a series break in 2018).
• The NAO’s findings (Resilience to flooding, 15 Nov 2023):
• the doubled GBP 5.2 billion programme “is forecast to deliver protection to far fewer properties by 2027 than was promised”;
• “for the lack of £34 million in annual maintenance funding for 2022-23, more than 200,000 properties are at increased risk”;
• government “has not set a target for the level of flood resilience it expects to achieve”.

Netherlands (Delta Programme 2025): - The Delta Fund, which is legally secured, holds about EUR 27.5 billion for 2015–2050 against estimated needs of about EUR 41 billion. The programme calls the shortfall of “over € 13 billion … too large”. - The six storm-surge barriers “still comply with the standard at present but a substantial and growing backlog of major maintenance has developed” (DP2025). This bears directly on Panel 15.3’s “insufficiently maintained”. - The ~50-year memory figure itself could not be tested against post-2013 data.

Individual cases: - Germany. Households reporting no warning fell from 27% (2002) to 5% (2013), then 29–35% reported no warning in 2021 (Kreibich et al. 2017; Thieken et al. 2023). The 2021 flood was a different type (a flash flood in small catchments), so the comparison is indicative only. - Valencia (secondary source: Wikipedia, “2024 Spanish floods”): - a regional emergency unit created after the 2019 Vega Baja floods was closed by the new regional government after the 2023 election as a “superfluous expense”; - a 2008 programme of about EUR 200 million to protect the southern towns was largely never implemented.

Verdict: partly held up. The mechanism is well supported by independent evidence: memory decay over one to two generations, post-disaster spikes, and chronic under-maintenance. But the chapter calls it a “general principle, valid across different political and economic systems”, and that overstates it. England since 2015 and the Netherlands show that multi-year, legally backed funding can smooth capital spending. What resists smoothing is maintenance and delivery: deteriorating asset condition, the Dutch maintenance backlog and funding gap, and English programme under-delivery.

Implication for weight. Use the cycle as a robust tendency, not a law. The refined lesson is that institutional commitments can hold capital investment steady, but the cycle reappears in maintenance, staffing and readiness for rare events, which are the least visible parts of the system.

Claim 9: Switzerland made flood control legally conditional on improving floodplain ecology; about 11,000 km of rivers and 22,000 ha to restore and 50,000 barriers to remove, at a current rate of 15–28 km a year (Panel 15.1, Tockner, p. 349)#

Original claim. Switzerland, having spent CHF 45 billion on engineered flood control between 1970 and 2005 (“unaffordable”), changed strategy: “future flood control measures are now required by law to be linked to a concurrent improvement of the integrity of river floodplain ecosystems”. Estimated needs: “about 22 000 ha of cultivated land … about 11 000 km of streams and rivers need to be restored. The current restoration rate ranges between 15 and 28 km per year; and 50 000 barriers must be removed” (Armin Peter, personal communication; EAWAG 2006) (p. 349).

Subsequent developments - The law. The revised Waters Protection Act came into force in 2011. It created “a legal obligation to make waters more natural again and to restore engineered waters in their natural functions”, and requires cantons to prepare strategic revitalisation plans (Buchecker et al., WSL, 30 Jan 2026, German; my translation). - The adopted target is “to revitalise around 4,000 km of waters by 2090”, about 50 km a year. - Progress. “However, implementation is proceeding substantially more slowly than planned” (the same report, citing Thomas and Renner 2021). - Context from the same report. About 14,000 km of Swiss watercourses have been engineered, and up to 90% of original floodplain areas destroyed. - What could not be verified. I could not retrieve FOEN progress statistics, the legal texts from Fedlex, or the status of barrier remediation and hydropower remediation. The claim that flood control is legally conditional on ecological improvement (grounded in the Hydraulic Engineering Act and the Waters Protection Act) was therefore not checked against the statutes.

Verdict: partly held up. The core claim, that Switzerland put river restoration into law as a reversal of its engineering lock-in, held and was reinforced by the 2011 revision. The panel’s quantitative figures were needs estimates from a personal communication. The adopted legal target is about a third of the 11,000 km figure and spread over 80 years, and the later literature reports delivery well behind plan.

Implication for weight. Good illustration that lock-in can be reversed by law, and of how slowly and at what reduced scale reversal proceeds. Do not cite the 11,000 km, 22,000 ha or 50,000-barrier figures as Swiss policy.

Claim 10: UK flood-warning sign-up is low partly because residents fear lower house prices or losing insurance (p. 354); numerical weather prediction “does not (yet) give generally reliable rainfall forecasts” for flash-flood warning (Beven, Panel 15.2, p. 358)#

Original claim. In the UK “people in flood risk zones can sign up to receive warnings … but the take-up has not been high. This is partly because many people do not want to recognise that they are in a flood risk zone because of a fear of reduced house prices or the inability to acquire insurance” (p. 354). Beven: numerical weather prediction “is useful for identifying potential flood events but does not (yet) give generally reliable rainfall forecasts”, which are “essential … especially for flash floods in small basins” (p. 358).

Subsequent developments

(a) Warning uptake and insurance: - Registrations (EA annual reports): - over 1.3 million landline and mobile customers (EA, 2011–2017 summary); - over 1.58 million properties by March 2024 (EA FCERM report 2023–24); - over 1.5 million by March 2025 (EA FCERM report 2024–25). - These figures include “phone numbers registered in areas at risk from flooding that are automatically opted-in”. Part of the uptake problem was therefore met by automatic enrolment rather than by persuasion. - Properties at risk. About 6.3 million properties in England are at risk from all sources, rising to about 8 million by mid-century (NaFRA 2024). The previous assessment put 2.6 million at risk from rivers and the sea, the main scope of the warning service. The EA reports I retrieved give no official take-up percentage. - Uptake after a flood. After the 2015 flood, Carlisle had “an uptake of 79% of at risk properties registered” (Cumbria plan, Carlisle table). Uptake is high where memory is fresh (Claim 8). - Broadcast alerts. UK Emergency Alerts, a cell-broadcast system, has been “operationally live since March 2023” (EA 2023–24 report). It bypasses sign-up altogether. - Insurance. Flood Re, a reinsurance scheme making flood cover “more widely available and affordable”, launched in April 2016 and is due to run until 2039 (Defra research page, 24 Feb 2021; Flood Re). This addresses the “inability to acquire insurance” motive the chapter names. - The motive itself. I found no post-2013 study that directly tests whether fear of house-price or insurance effects deters registration.

(b) Rainfall forecasting for flash floods: - Progress, 2021. For the July 2021 event, forecasts predicted high rainfall totals within the range of the observations more than two days ahead, and the convection-permitting model captured the magnitude from its first run. But “the precise prediction of the rainfall totals for the affected areas [was] only possible a few hours in advance” (Mohr et al. 2023). - The limit that remains. A review of surface-water flood forecasting found that it “presents a unique challenge due to the high uncertainties around predicting the location, timing, and impact of what are typically localized events”. It also found that, with convection-permitting and ensemble models, “it is now theoretically feasible to develop operational surface water forecasting systems” (Speight et al., WIREs Water, 25 Feb 2021). - Valencia, 2024. Rainfall of 771.8 mm in 16 hours at Turís and a national one-hour record of 184.6 mm (Calvo-Sancho et al. 2026) show the tail that forecasts and warnings must handle. - A gap in this check. Because the bibliographic APIs rate-limited the session, I did not retrieve a systematic verification study of European precipitation-forecast skill trends.

Verdict: partly held up. - Warnings: take-up remained incomplete. Policy moved to automatic opt-in and cell broadcast, and Flood Re addressed the insurance side. The chapter’s causal explanation (house prices, insurance) was plausible but remains untested. - Forecasting: Beven’s hedge “(yet)” was well judged. Rainfall forecasting improved enough to flag the 2021 event days ahead, but reliable location-specific totals for small, fast catchments are still available only hours ahead.

Implication for weight. Use the take-up point as a hypothesis about private penalties for acknowledging risk. The strongest later evidence is indirect: policymakers designed around voluntary sign-up rather than relying on it. Beven’s point stands as an accurate statement of a persistent limit, now narrowed.

Cross-cutting observations for using this section as a lens#

1. Prediction outran action, and still does. Two decades after Vaison-la-Romaine, the deadliest floods in Germany (2021) and Spain (2024) came with good forecasts and late or absent local warnings (Claim 4). In technology-neutral terms: improving the upstream signal does not improve outcomes if the downstream decision, dissemination and response links stay weak. This is the chapter’s most transferable and best-confirmed mechanism (pp. 347, 353, 359–360).
2. Authority and cost to act are as important as knowledge. The German district that must declare an emergency also pays for it. In Valencia the official with legal command doubted her own competence and “paralysis” followed. Automatic triggers are one institutional answer: Saxony declares an emergency automatically when forecasts pass the top warning level (Claim 4). This extends p. 360 with an explicit cost-bearing dimension.
3. Mandated tools do not remove the decision bottleneck. EU law required mobile public-warning systems by June 2022, and Valencia had one. It was used at 20:11. Codifying the channel is not codifying the decision (Claims 3 and 4).
4. A law that mandates process gets evaluated on process. The Floods Directive produced maps and plans everywhere. Because objectives were rarely quantified or time-bound, neither the Commission nor the ECA could say, 11 to 18 years later, whether risk had fallen (Claim 3). A lens question: does a governance instrument define measurable outcomes, and who funds delivery?
5. Learning is real, and it decays. German warning performance improved between 2002 and 2013, then failed in 2021. Memory fades within one to two generations. Insurance take-up returns to baseline. Maintenance falls behind even when capital budgets rise (Claim 8). Institutional commitments can hold capital spending steady; the cycle reappears in maintenance and preparedness for rare events.
6. Records are short, and history is information. The Ahr floods of 1804 and 1910 and the Eden floods since 1800 lay outside the instrumental records used for design. Including them changes the risk picture substantially (Claim 7). Three-quarters of the 2021 Ahr deaths occurred outside mapped hazard zones.
7. Unprecedented events defeat systems built for precedented ones. Across 45 paired events, management reduced impacts unless the second event exceeded design levels (Kreibich et al. 2022; Claim 7). This is the empirical core of pp. 351–352 and 356, and the reason for planning explicitly for floods beyond the design standard (pp. 356, 362).
8. Act before detection, but hold projections loosely. Climate allowances adopted in 2005–06 preceded detection (2017–2019), and the detection confirmed their direction (Claims 2 and 6). The projections cited as justification were later judged untrustworthy in their details by their own authors (Claim 1). The durable lesson is to use projections for direction, stress tests and adaptive pathways, not as point targets.
9. Adoption of the right idea is slow and patchy. Only three of 20 regions studied use climate-adjusted design values. Bavaria’s factor has not been revised since 2005. Dutch dike upgrades and Swiss restoration run behind schedule (Claims 6 and 9). Knowing what to do is not the bottleneck; money, delivery capacity and update cycles are.
10. Positionality and dissent. Kundzewicz was a party to the stationarity dispute (Milly et al. 2015) and later revised his own confidence in projections (Kundzewicz et al. 2016/2017). All three panels complement rather than challenge the chapter. The main contested scientific points (stationarity; detection versus attribution) are not aired in it.
11. Slips to correct when citing:
    • The UK 2006 allowance was +20% peak flow for 2025–2115, not “after 2085” (p. 355).
    • Bavaria’s adopted factor is a uniform 1.15, not graded 40–50% for small floods (p. 355).
    • The status of Carlisle’s post-2005 design standard is unresolved (p. 356; EA 2019 wording conflicts).
    • The Swiss restoration figures are needs estimates, not adopted targets (p. 349).
    • “Probably the most advanced” (p. 361) is an untested comparative claim.
    • The p. 364 causal claim about fatalities is unreferenced, and sits in tension with p. 353.

Sources#

IPCC assessments - IPCC (2013). Climate Change 2013: The Physical Science Basis, Summary for Policymakers (AR5 WGI). https://www.ipcc.ch/site/assets/uploads/2018/02/WG1AR5_SPM_FINAL.pdf - Seneviratne, S.I., Zhang, X. et al. (2021). Chapter 11: Weather and Climate Extreme Events in a Changing Climate. In: Climate Change 2021: The Physical Science Basis (AR6 WGI). https://www.ipcc.ch/report/ar6/wg1/chapter/chapter-11/ - Ranasinghe, R. et al. (2021). Chapter 12: Climate Change Information for Regional Impact and for Risk Assessment (AR6 WGI), section 12.4.5 Europe. https://www.ipcc.ch/report/ar6/wg1/chapter/chapter-12/

Flood trends, projections and attribution (peer-reviewed) - Alfieri, L., Burek, P., Feyen, L., Forzieri, G. (11 May 2015). Global warming increases the frequency of river floods in Europe. HESS 19:2247. https://doi.org/10.5194/hess-19-2247-2015 - Blöschl, G. et al. (11 Aug 2017). Changing climate shifts timing of European floods. Science. https://doi.org/10.1126/science.aan2506 - Blöschl, G. et al. (28 Aug 2019). Changing climate both increases and decreases European river floods. Nature 573:108–111. https://doi.org/10.1038/s41586-019-1495-6 (abstract via PubMed PMID 31462777) - Blöschl, G. et al. (22 July 2020). Current European flood-rich period exceptional compared with past 500 years. Nature. https://doi.org/10.1038/s41586-020-2478-3 - Calvo-Sancho, C. et al. (17 Feb 2026). Human-induced climate change amplification on storm dynamics in Valencia’s 2024 catastrophic flash flood. Nat Commun. https://doi.org/10.1038/s41467-026-68929-9 - Gudmundsson, L. et al. (12 Mar 2021). Globally observed trends in mean and extreme river flow attributed to climate change. Science. https://doi.org/10.1126/science.aba3996 - Hall, J. et al. (30 July 2014). Understanding flood regime changes in Europe: a state-of-the-art assessment. HESS 18:2735. https://doi.org/10.5194/hess-18-2735-2014 - Hodgkins, G.A. et al. (2017). Climate-driven variability in the occurrence of major floods across North America and Europe. J Hydrol. https://doi.org/10.1016/j.jhydrol.2017.07.027 - Kundzewicz, Z.W. et al. (online 20 Dec 2013; 2014). Flood risk and climate change: global and regional perspectives. HSJ 59(1):1–28. https://doi.org/10.1080/02626667.2013.857411 - Kundzewicz, Z.W., Krysanova, V., Dankers, R., Hirabayashi, Y., Kanae, S., Hattermann, F.F. et al. (online 29 Sept 2016; 2017). Differences in flood hazard projections in Europe: their causes and consequences for decision making. HSJ 62(1). https://doi.org/10.1080/02626667.2016.1241398 - Ludwig, P. et al. (5 Apr 2023). A multi-disciplinary analysis of the exceptional flood event of July 2021 in central Europe, Part 2: Historical context and relation to climate change. NHESS 23:1287. https://doi.org/10.5194/nhess-23-1287-2023 - Osuch, M. et al. (online Aug 2016). Projected changes in flood indices in selected catchments in Poland in the 21st century. SERRA. https://doi.org/10.1007/s00477-016-1296-5 - Tradowsky, J.S. et al. (29 June 2023). Attribution of the heavy rainfall events leading to severe flooding in Western Europe during July 2021. Climatic Change. https://doi.org/10.1007/s10584-023-03502-7

Warnings, fatalities and event analyses (peer-reviewed) - Jongman, B. et al. (Apr 2015). Declining vulnerability to river floods and the global benefits of adaptation. PNAS. https://doi.org/10.1073/pnas.1414439112 - Kreibich, H., Müller, M., Schröter, K., Thieken, A.H. (29 Nov 2017). New insights into flood warning reception and emergency response by affected parties. NHESS 17:2075. https://doi.org/10.5194/nhess-17-2075-2017 - Kreibich, H. et al. (26 July 2017). Adaptation to flood risk: results of international paired flood event studies. Earth’s Future. https://doi.org/10.1002/2017EF000606 - Kreibich, H. et al. (3 Aug 2022). The challenge of unprecedented floods and droughts in risk management. Nature. https://doi.org/10.1038/s41586-022-04917-5 - Mohr, S. et al. (6 Feb 2023). A multi-disciplinary analysis of the exceptional flood event of July 2021 in central Europe, Part 1: Event description and analysis. NHESS 23:525. https://doi.org/10.5194/nhess-23-525-2023 - Pappenberger, F. et al. (2015). The monetary benefit of early flood warnings in Europe. Env Sci Policy. https://doi.org/10.1016/j.envsci.2015.04.016 - Paprotny, D., Sebastian, A., Morales-Nápoles, O., Jonkman, S.N. (May 2018). Trends in flood losses in Europe over the past 150 years. Nat Commun. https://doi.org/10.1038/s41467-018-04253-1 (full text via PMC5974183) - Paprotny, D., ‘t Hart, C.M.P., Morales-Nápoles, O. (28 Nov 2024). Evolution of flood protection levels and flood vulnerability in Europe since 1950 estimated with vine-copula models. Nat Hazards. https://doi.org/10.1007/s11069-024-07039-5 - Paprotny, D., Terefenko, P., Śledziowski, J. (6 Nov 2024). HANZE v2.1: an improved database of flood impacts in Europe from 1870 to 2020. ESSD 16:5145. https://doi.org/10.5194/essd-16-5145-2024 - Petrucci, O. et al. (9 June 2026). FFEM-DB 2.0: an updated and extended spatio-temporal dataset of flood fatalities occurred from 1980 to 2024 in the Euro-Mediterranean region. Sci Data 13:1331. https://doi.org/10.1038/s41597-026-07511-w (full text via PMC13590483) - Rhein, B., Kreibich, H. (10 Feb 2025). Causes of the exceptionally high number of fatalities in the Ahr valley, Germany, during the 2021 flood. NHESS 25:581. https://doi.org/10.5194/nhess-25-581-2025 - Speight, L.J., Cranston, M.D., White, C.J., Kelly, L. (25 Feb 2021). Operational and emerging capabilities for surface water flood forecasting. WIREs Water. https://doi.org/10.1002/wat2.1517 (abstract only) - Thieken, A.H. et al. (3 Mar 2023). Performance of the flood warning system in Germany in July 2021: insights from affected residents. NHESS 23:973. https://doi.org/10.5194/nhess-23-973-2023

Design standards, stationarity and memory (peer-reviewed and institutional research) - Buchecker, M. et al. (30 Jan 2026). Bevölkerungseinbindung in der Revitalisierungsplanung von Fliessgewässern. WSL. https://doi.org/10.55419/wsl:42459 (abstract only) - Fanta, V., Šálek, M., Sklenicka, P. (7 Mar 2019). How long do floods throughout the millennium remain in the collective memory? Nat Commun 10:1105. https://doi.org/10.1038/s41467-019-09102-3 - Gallagher, J. (July 2014). Learning about an infrequent event: evidence from flood insurance take-up in the United States. AEJ: Applied Economics. https://doi.org/10.1257/app.6.3.206 - Koutsoyiannis, D., Montanari, A. (online Sept 2014; 2015). Negligent killing of scientific concepts: the stationarity case. HSJ. https://doi.org/10.1080/02626667.2014.959959 - Macdonald, E., Merz, B., Aerts, J.C.J.H., Dewals, B., Kwadijk, J., Slager, K. et al. (9 Feb 2026). Future climate change assessment in flood risk management: a synthesis of practices in Germany and the BeNeLux countries. J Flood Risk Manag. https://doi.org/10.1111/jfr3.70173 (read from the authors’ preprint: https://orbi.uliege.be/bitstream/2268/341234/1/MacDonald_2026_Preprint.pdf) - Milly, P.C.D., Betancourt, J., Falkenmark, M., Hirsch, R.M., Kundzewicz, Z.W., Lettenmaier, D.P. et al. (Sept 2015). On critiques of “Stationarity is dead: whither water management?”. WRR. https://doi.org/10.1002/2015WR017408 (abstract only) - Montanari, A., Koutsoyiannis, D. (Nov 2014). Modeling and mitigating natural hazards: stationarity is immortal! WRR. https://doi.org/10.1002/2014WR016092 - Parkes, B., Demeritt, D. (July 2016). Defining the hundred year flood: a Bayesian approach for using historic data to reduce uncertainty in flood frequency estimates. J Hydrol. https://doi.org/10.1016/j.jhydrol.2016.07.025 - Serinaldi, F., Kilsby, C.G. (2015). Stationarity is undead: uncertainty dominates the distribution of extremes. Adv Water Resour. https://doi.org/10.1016/j.advwatres.2014.12.013 - Surminski, S., Thieken, A.H. (Aug 2017). Promoting flood risk reduction: the role of insurance in Germany and England. Earth’s Future. https://doi.org/10.1002/2017EF000587

EU institutions and law - European Court of Auditors (adopted 19 Sept 2018). Special Report 25/2018: Floods Directive: progress in assessing risks, while planning and implementation need to improve. https://www.eca.europa.eu/Lists/ECADocuments/SR18_25/SR_FLOODS_EN.pdf - European Commission (10 Dec 2019). Fitness check of the Water Framework Directive, Groundwater Directive, Environmental Quality Standards Directive and Floods Directive, SWD(2019) 439 final (retrieved via Cellar). https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:52019SC0439 - European Commission (4 Feb 2025). Report on the implementation of the Water Framework Directive and the Floods Directive (third RBMPs, second FRMPs), COM(2025) 2 final (retrieved via Cellar). https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=COM:2025:2:FIN - European Commission. Water Framework Directive and Floods Directive implementation reports (index page). https://environment.ec.europa.eu/topics/water/water-framework-directive/implementation-reports_en - Directive (EU) 2018/1972, European Electronic Communications Code, Art. 110 (public warning system) (retrieved via Cellar). https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32018L1972 - European Environment Agency (14 Oct 2025). Economic losses from weather- and climate-related extremes in Europe (indicator). https://www.eea.europa.eu/en/analysis/indicators/economic-losses-from-climate-related - European Environment Agency (11 Mar 2024). European Climate Risk Assessment (EUCRA). https://www.eea.europa.eu/en/analysis/publications/european-climate-risk-assessment

United Kingdom (government and audit) - DCLG (27 Mar 2012). Technical Guidance to the National Planning Policy Framework, Table 5 (reproducing Defra FCDPAG3 supplementary note, Oct 2006). https://assets.publishing.service.gov.uk/media/5a79a6a6e5274a684690b1b3/2115548.pdf - Environment Agency (first published 22 July 2020; updated 20 July 2021, 30 Nov 2021, 17 May 2022). Flood and coastal risk projects, schemes and strategies: climate change allowances. https://www.gov.uk/guidance/flood-and-coastal-risk-projects-schemes-and-strategies-climate-change-allowances - HM Government (8 Sept 2016). National Flood Resilience Review. https://assets.publishing.service.gov.uk/media/5a7f973de5274a2e8ab4d1af/national-flood-resilience-review.pdf - Environment Agency / Cumbria Floods Partnership (30 June 2016). Cumbria Flood Action Plan (overview and Carlisle community action table). https://www.gov.uk/government/publications/cumbria-flood-action-plan ; https://assets.publishing.service.gov.uk/media/5a803c16ed915d74e33f93d7/cumbria-flood-plan-overview.pdf ; https://assets.publishing.service.gov.uk/media/5a7573a7e5274a1622e21e3b/cumbria-flood-plan-community-carlisle.pdf - Environment Agency (31 July 2019). £25m Flood Risk Management Scheme for Carlisle gets underway. https://www.gov.uk/government/news/25m-flood-risk-management-scheme-for-carlisle-gets-underway - Environment Agency (31 July 2020). Online consultation on next phase of the £25m Carlisle flood risk management scheme. https://www.gov.uk/government/news/the-environment-agency-offers-online-consultation-on-next-phase-of-the-25m-carlisle-flood-risk-management-scheme - Defra (updated 15 July 2026). Flood and Coastal Erosion Risk Management in England: Central Government Funding and Performance (official statistics and ODS data tables). https://www.gov.uk/government/statistics/flood-and-coastal-erosion-risk-management-in-england-central-government-funding-and-performance ; https://assets.publishing.service.gov.uk/media/6a55fcfaa6586e258d371e55/FCERM_Funding_and_Performance_Official_Statistics_Data_Tables_July_2026.ods - National Audit Office (15 Nov 2023). Resilience to flooding. https://www.nao.org.uk/reports/resilience-to-flooding/ (summary page only) - Environment Agency. Managing flood and coastal erosion risks in England: 1 April 2011 to 31 March 2017. https://assets.publishing.service.gov.uk/media/5aba2d0e40f0b67d64e218dd/1_April_2011_to_31_March_2017_managing_FCERM.pdf - Environment Agency. FCERM reports for 2023–24, 2024–25 and 2025–26 (the last published 30 June 2026). https://www.gov.uk/government/publications/flood-and-coastal-risk-management-national-report - Environment Agency (17 Dec 2024; updated 5 Aug 2026). National assessment of flood and coastal erosion risk in England 2024. https://www.gov.uk/government/publications/national-assessment-of-flood-and-coastal-erosion-risk-in-england-2024 - Defra (24 Feb 2021). Examining whether flood insurance is affordable and available (research page). https://www.gov.uk/flood-and-coastal-erosion-risk-management-research-reports/examining-whether-flood-insurance-is-affordable-and-available - Flood Re. About us. https://www.floodre.co.uk/about-us/

Netherlands - Delta Programme Commissioner (17 Sept 2024). Delta Programme 2025 (complete version, English). https://english.deltaprogramma.nl/documents/publications/2024/09/17/dp2025-complete-version

Spain: court documents on the 29 October 2024 flood (Consejo General del Poder Judicial, TSJ Comunidad Valenciana news) - 6 Feb 2025: court asks the regional government who decided to send the SMS alert, and the CHJ for Poyo flow data 16:13–18:42. https://www.poderjudicial.es/cgpj/es/Poder-Judicial/Tribunales-Superiores-de-Justicia/TSJ-Comunidad-Valenciana/Noticias-Judiciales-TSJ-Comunidad-Valenciana/El-juzgado-vuelve-a-requerir-a-la-Generalitat-para-que-informe-sobre-quien-decidio-enviar-el-SMS-de-alerta-por-la-Dana-y-pide-a-la-CHJ-datos-del-caudal-del-Poyo-entre-las-16-13-y-las-18-42-horas-del-29-de-octubre - 19 June 2025: Provincial Court upholds investigation of the former regional minister for justice and interior. https://www.poderjudicial.es/cgpj/es/Poder-Judicial/Tribunales-Superiores-de-Justicia/TSJ-Comunidad-Valenciana/Noticias-Judiciales-TSJ-Comunidad-Valenciana/La-Audiencia-de-Valencia-confirma-la-decision-de-la-juez-de-investigar-a-la-exconsellera-de-Justicia-e-Interior-por-la-gestion-de-la-Dana - 24 Feb 2026: investigating judge’s reasoned statement to the TSJCV concerning the former regional president. https://www.poderjudicial.es/cgpj/es/Poder-Judicial/Tribunales-Superiores-de-Justicia/TSJ-Comunidad-Valenciana/Noticias-Judiciales-TSJ-Comunidad-Valenciana/La-magistrada-que-instruye-la-causa-por-la-gestion-de-la-Dana-eleva-exposicion-razonada-al-TSJCV-contra-el-expresidente-de-la-Generalitat-Carlos-Mazon - 16 Mar 2026: TSJCV declines to investigate the former regional president. https://www.poderjudicial.es/cgpj/es/Poder-Judicial/Tribunales-Superiores-de-Justicia/TSJ-Comunidad-Valenciana/Noticias-Judiciales-TSJ-Comunidad-Valenciana/El-TSJCV-rechaza-investigar-al-expresidente-de-la-Generalitat-por-la-gestion-de-la-Dana-al-no-apreciar-en-la-exposicion-razonada-de-la-instructora–un-fundamento-solido-y-objetivo–de-que-los-hechos-descritos–revistan-caracter-de-delito- - 2 Sept 2026: official death toll raised to 232. https://www.poderjudicial.es/cgpj/es/Poder-Judicial/Tribunales-Superiores-de-Justicia/TSJ-Comunidad-Valenciana/Noticias-Judiciales-TSJ-Comunidad-Valenciana/La-jueza-eleva-a-232-la-cifra-oficial-de-victimas-mortales-causadas-por-la-Dana-de-Valencia

Secondary sources (used only where no primary source could be reached; flagged in text) - Wikipedia (English), “2024 Spanish floods” (accessed 25 Sept 2026): AEMET warning times, the regional emergency unit’s closure, the 2008 programme. https://en.wikipedia.org/wiki/2024_Spanish_floods - Wikipedia (German), “Hochwasser in West- und Mitteleuropa 2021” (accessed 25 Sept 2026): Koblenz prosecution outcome (April 2024), Rhineland-Palatinate inquiry final report (2 Aug 2024), disciplinary findings (July 2025), siren programme. https://de.wikipedia.org/wiki/Hochwasser_in_West-und_Mitteleuropa_2021 - Wikipedia (English), “Room for the River (Netherlands)” (accessed 25 Sept 2026): programme dates and 16,000 m³/s aim. https://en.wikipedia.org/wiki/Room_for_the_River(Netherlands)