LL2-15 — Ch15 Floods: lessons about early warning systems#

Late lessons from early warnings: science, precaution, innovation (EEA Report No 1/2013), Chapter 15, report pp. 347–368 (PDF pp. 349–370). The chapter text runs pp. 347–364. Table 15.1 is on p. 365 and the references on pp. 365–368.

Reading basis: I read the whole text extract in order, through the final marker (PDF 370 / report p. 368). I then checked every page from p. 347 to p. 365 visually in the PDF: the summary box, three panels, two boxes, Table 15.1 and the one photograph. The extraction was clean apart from section numbers (“15�1” = 15.1) and tab-spaced words. No figures or data were lost. The only image is a stock photo of flooded houses (p. 353), which is decorative.

Authors and standpoint#

Chapter author. The chapter is by Zbigniew W. Kundzewicz alone. It gives his name but no affiliation. Two footnotes (p. 347) say: - five reviewers commented on it: Keith Beven, Jim Hall, Irina Krasovskaia, Edmund Penning-Rowsell and Bellie Sivakumar; - the work was carried out within WATCH (Water and Global Change), an Integrated Project of the EU’s 6th Framework Programme.

Not stated in the chapter (from my background knowledge; verify before citing): Kundzewicz is a Polish hydrologist at the Polish Academy of Sciences (Poznań), linked to the Potsdam Institute for Climate Impact Research. He was a long-time IPCC author and edited Hydrological Sciences Journal.

The reference list itself shows where he stands: - he is first author of the IPCC AR4 Working Group II freshwater chapter (Kundzewicz et al., 2007); - he co-authored Milly et al. (2008), “Stationarity is dead”; - he co-authored the trend-detection studies on river flows (Kundzewicz et al., 2005; Svensson et al., 2005); - he wrote the analysis of the 1997 Polish flood (Kundzewicz et al., 1999).

So the Odra case (Box 15.1) is his own country’s disaster, which he studied first-hand. About 10 of the chapter’s ~77 references are his own work (about 13%).

Evident stance. The chapter reads like a review by a senior hydrologist and flood-risk scientist. It is not a polemic. - It argues for “living with floods”: a mix of structural and non-structural measures, zoning, early warning and participation. It is set against the older engineering goal of flood control. - It is sceptical of the idea that anything can make people completely safe. - It is careful about attributing floods to climate change. The body says no conclusive trend has yet been detected in river-flood data (p. 351), although the summary box is more assertive (see Limitations). - It is broadly positive about EU policy, especially the Floods Directive (Box 15.2). - The precautionary principle appears only once in the body (p. 361). “Precautionary” appears twice more: a Defra “precautionary allowance” (p. 355) and the summary’s description of Panel 15.3 (p. 347). In my judgement this makes it one of the least precaution-framed chapters in the report. That is a comparison with other chapters, and this audit did not check it against them.

Unusual scope, flagged by the chapter itself. The summary box states that “Unlike most other case studies presented in this report, this chapter focuses on flooding as a phenomenon and the requirements for effective early warning systems, rather than addressing a particular event” (p. 347). The hazard is natural, not a technology or substance. “Early warning” has two meanings here: 1. literal, operational flood warnings (short-term); 2. long-term warnings that flood frequency is changing, i.e. climate and land-use projections (p. 348).

For the project’s purposes, the chapter’s technology-governance content sits in three places: - structural flood defence (levees, dams, river “rectification”) treated as a technology with unintended consequences and lock-in; - warning systems treated as socio-technical chains; - how uncertainty in models and data is handled.

Panels (all three complement the chapter; none dissents from it). - Panel 15.1, Klement Tockner, “Managing river flood plains as ecosystems of global strategic importance” (pp. 349–350). An ecological critique of hydraulic engineering, with a Swiss policy example and a research agenda. His affiliation is not stated (background: freshwater ecologist, then at the Leibniz Institute of Freshwater Ecology and Inland Fisheries in Berlin). He argues for conservation and restoration, but pragmatically: he accepts “novel ecosystems” dominated by non-native species. He cites five works he wrote or co-wrote: Tockner et al. 2002, 2008 (cited three times) and 2009; Nakamura et al. 2006; and Sommerwerk et al. 2010. Tockner et al. 2002 is missing from the reference list. - Panel 15.2, Keith Beven, “Uncertainty in predicting floods (and other environmental variables)” (pp. 357–358). An epistemological argument about aleatory versus epistemic uncertainty. Beven also reviewed the chapter. He is a partisan in a live methodological dispute in hydrology: the “undermining the science?” exchange of 2006–2008, which appears in the references (Beven 2006a; Hall et al. 2007; Todini and Mantovan 2007; Sivakumar 2008; Andréassian et al. 2007; Montanari 2007). He acknowledges “a range of strongly held opinions” (p. 357) but argues his own side. About 8 of the references are his own work. Kundzewicz’s main text also draws on the same exchange: §15.3.3 (p. 359) cites Beven 2006a, Hall et al. 2007 and Sivakumar 2008 as “disagreeing in significant respects”, so the dispute is not confined to the panel. - Panel 15.3, Pier Vellinga and Jeroen Aerts, “Dealing with the risk of coastal flooding — experiences from European countries” (pp. 363–364). A review of coastal adaptation that promotes “climate proofing”, “superlevees”, “building with nature” and elevated development. The authors co-authored the “Climate Proofing the Netherlands” commentary they cite (Kabat et al., 2005). Their affiliation is not stated (background: VU University Amsterdam and Wageningen).

Epistemic community. The reviewers, panel authors and cited authors overlap heavily: - Beven is reviewer, panel author and heavily cited; - Hall is reviewer and cited four times; - Sivakumar is reviewer and cited; - Penning-Rowsell is reviewer and a co-author of the Foresight report (Evans et al., 2004).

This is normal for a specialist review. It does mean the chapter reflects one research community’s consensus and its internal debates, not an outside audit of flood management.

Section-by-section notes#

Summary box (p. 347)#

The unattributed summary makes five moves: 1. Problem getting worse. “Floods are an increasingly acute problem.” It asserts that “Intense precipitation has become more frequent and more intense”, and that human pressure and “flawed decisions about the location of human infrastructure” have raised both flood magnitudes and potential losses. 2. Scope. The chapter covers the phenomenon and the requirements for warning systems, not a single event (quoted above). 3. Reframing failure and success. “The occurrence of a flood need not be considered a ‘failure’ and, conversely, minimisation of losses may constitute a ‘success’.” There are lessons from every flood. 4. Paradigm claim. “Once we accept that no flood protection measures can guarantee complete safety, a general change of paradigm is needed to reduce human vulnerability to floods.” “Living with floods” is presented as “more sustainable than hopelessly striving to eradicate them.” 5. Chain claim. “Flood forecasting and warning systems fail because links in the chain perform poorly or fail completely. A single weak point in a system that otherwise contains excellent components may render the overall system performance unsatisfactory.” Success needs “integration of components and collaboration and coordination between multiple institutions.”

It then introduces the three panels. It says the third “reviews European experience with precautionary action.”

Epigraphs (p. 348)#

• Genesis 7:12, 19 (the Flood).
• George Berkeley: “We have first raised the dust and then complain we cannot see.” The quotation signals a theme: much of the difficulty is self-created, both by human modification of rivers and by human-made confusion. The chapter never develops the Berkeley line explicitly.

15.1 River floods and early warnings (pp. 348, 351–352)#

Definitions (p. 348). Flooding is “a potentially destructive abundance of water in a normally dry location”. The chapter sets aside urban, groundwater and coastal flooding (except in Panel 15.3) to focus on river (fluvial) floods from rain, snowmelt, ice, dam breaks, surges and obstructions.

How floods differ from the report’s other hazards (p. 348). Floods are “intermittent events”. This contrasts with hazards that “affect the environment continuously and can impose cumulative ‘pressures’”. My inference, not stated in the chapter: this matters analytically, because the attention dynamics (forgetting between events) differ from those of chronic exposures.

Floods as natural and historically beneficial (p. 348). People have settled floodplains for fertile soils, flat land, water supply and transport. The chapter says: “Riparian people have historically lived in harmony with nature, benefiting from benign floods”. This is unreferenced and somewhat romantic.

Two meanings of “early warning” (p. 348). - Short-term. A “flood warning” is a technical term: “specific timely information, based on a reliable forecast, that a high water level is expected at a particular location and time”, to allow emergency action such as strengthening dikes or evacuating. A “flood alert” comes earlier, is less specific, and aims at “raising vigilance”. A warning must come early enough, must fit catchment size, and “should also be expressed in a way that persuades people to take appropriate action”. - Long-term. A “prediction” of changed flood frequency relative to a reference period, for example 1961–1990 climate normals. Example: today’s 100-year flood becoming a 50-year flood. Such a warning “is (or should be) an important signal for decision-makers that the required level of protection is unlikely to be maintained in the future unless flood preparedness is improved.” The parenthesis “(or should be)” implies it often is not heeded.

Humans intensify flood risk (p. 348). “Flood risk has been greatly intensified by humans, who — to use the language of mechanics — have increased the load and decreased the resistance of the system.” Humans have raised flood magnitude for any given rainfall and amplified damage potential.

Loss statistics (pp. 348, 351). - Annual global economic losses from extreme weather, including floods, rose “ten-fold between the 1950s and 1990s in constant prices” (IPCC, 2001b). - Munich Re data (Kron, 2005): the number of “great flood disasters” (those needing international or interregional help) has grown, as have economic and insured losses. - Several river floods have each caused losses above USD 10 billion. - Individual events in less developed countries have killed more than 1,000 people (Kundzewicz et al., 2010a). - Highest losses: about USD 30 billion, China 1998. - Destructive floods are “commonplace” in many developing countries, especially in Asia (Bangladesh, China, India) and South America, but have hit “virtually all parts of the world, including Europe”. - Barredo (2007) finds rising numbers of flood disasters in Europe, with “high-impact floods … occurring more frequently”. - European flood damage in 2002 (above EUR 20 billion) was the highest of any year. - Note: the chapter uses unnormalised losses as evidence that the problem is “increasingly serious”. It lists exposure as one driver but does not separate hazard from exposure in these figures.

Three families of drivers (p. 351). 1. Socio-economic: more exposure from population and economic growth in flood-prone areas; land-use change (urbanisation, deforestation); “changing perceptions of risk”. 2. Terrestrial: land-cover change and river regulation (straightening, shortening, embankments). Draining wetlands, removing vegetation and expanding impermeable surfaces reduce storage, giving “a higher flood peak and a shorter time to peak”. 3. Climate: a warmer atmosphere holds more water, heavy precipitation becomes more frequent, and snow cover, seasonality and circulation change. These are “important, even if they may not always be detectable in the historical record.”

Encroachment (p. 351). - “Human encroachment into floodplains has increased exposure to floods.” - Encroachment “may increase as people become wealthier and technology or economic imperatives help populate more flood-prone areas.” This is a notable line: technology enables occupation of hazardous land. - “Many flawed decisions have increased the flood loss potential.” - Rhine: more than 10 million people live in areas at risk of extreme floods, with potential damage estimated at EUR 165 billion (EU, 2007). - In developing countries, “The hope of overcoming poverty drives poor people to migrate to informal settlements in endangered, flood-prone zones around mega-cities”, in zones “previously … left uninhabited on purpose because effective flood protection cannot be assured.”

Uncertainty about future flood risk and about trend detection (p. 351). - Flood risk is “likely to grow” in many places from anthropogenic and climatic factors, but “Quantifying flood statistics is difficult … and subject to high uncertainty.” - It quotes IPCC (2001a): “the analysis of extreme events in both observations and coupled models is underdeveloped”. - Models project increases in both the size and frequency of rain-caused floods. “Yet 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” (citing the author’s own Kundzewicz et al., 2005 and Svensson et al., 2005). - Detection is hard because of a “low signal-to-noise ratio”. A weak signal “(if any)” sits on strong natural variability and is confounded by land-use change. Wilby et al. (2008) “speculate that statistically robust trends are unlikely to be apparent for several decades.” - Land use matters, but less so in very intense rainfall, which produces high runoff in both urban and forested basins.

The sampling effect and short records (pp. 351–352). “There is always potential for floods that are higher than those recorded in the past, even under stationary conditions”. This is a sampling effect: the longer the record, the more likely a record-breaking event somewhere with high damage potential. Defences are typically built for a 100-year return period, “perhaps with some additional safety margin”. But that design discharge “is generally estimated from records gathered over an inadequately short interval and is therefore uncertain.”

Physical extremes (p. 352). Record point rainfall: 1,340 mm in 12 hours, 1,825 mm in one day, 3,847 mm in eight days (WMO, 1986). Similar rainfall over or just upstream of a large city “could be expected to be utter destruction.”

Benefits of floods (p. 352). - Nile fertility. - Groundwater recharge. - Floods “may strengthen community solidarity and can enhance economic activity (related to flood preparedness and recovery)”. My comment: the second point is questionable, because recovery spending is not a net gain. The chapter offers no qualification. - “Any benefits that could be lost also need to be taken into account when considering flood prevention measures and options.” This is an explicit trade-off principle: suppressing a hazard also suppresses its benefits.

Panel 15.1 (Tockner): floodplains as strategic ecosystems (pp. 349–350)#

Floodplain ecology. Floodplains sit at the lowest point of the landscape and so “integrate and accumulate upstream and catchment processes.” In a natural state they are “disturbance-dominated ecosystems”, once comparable in diversity to tropical rainforests or coral reefs. That diversity “has, and continues to be, deeply degraded, a fact that has so far captured remarkably little attention at the policy level.”

The paradox. “Paradoxically, a good deal of this damage is done by hydrological engineering for flood defences”. Meanwhile other actions “simultaneously remove natural flood-mitigation features” (p. 349). The protective technology degrades the natural system that also provided protection.

Value. - Floodplains cover about 2% of land but provide about 25% of all continental ecosystem services, more than any other continental ecosystem type (Costanza et al., 1997; Oppermann et al., 2008 [the reference list has Opperman et al., 2009]; Tockner et al., 2008). - Services include flood regulation, water supply and groundwater recharge, waste treatment (nitrogen removal of 0.5–2.6 kg N per ha per day), fishing, grazing and more. - About half of the population of Europe and Japan lives on (former) floodplains, and about 80% of Bangladesh is deltaic floodplain.

Switzerland (p. 349). - Tockner calls Switzerland’s restoration commitment a “remarkable intention (probably unique in Europe)”, made “To reduce these costs”. So the panel presents it as an outlier, not a European norm. - Flood costs rose from 1970, totalling more than CHF 15 billion over 35 years (BUWAL and BWG, 2003). - Switzerland had “the highest global expenditure per capita on traditional hydraulic engineered flood control measures (CHF 45 billion between 1970 and 2005). This was unaffordable.” - The government “fundamentally altered its river management strategy”: future flood control is “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 returned to dynamic floodplain, about 11,000 km of streams and rivers restored, and 50,000 barriers removed. The current restoration rate is 15–28 km a year (Armin Peter, pers. comm.; EAWAG, 2006). - My arithmetic, not the panel’s: at that rate, 11,000 km would take roughly 400–730 years. - More than 95% of former floodplains (in Switzerland, from context) have been converted to industrial, agricultural or urban use.

Global threat (pp. 349–350). - In Europe, North America and Japan, more than 90% of former floodplains are “functionally extinct” or converted. - Many are now “novel or emerging ecosystems”. Up to 80% of benthic invertebrates on the Upper Danube, Rhine and Elbe are non-native (Sommerwerk et al., 2010). - 28,000 km of canals and navigable rivers connect 24 countries into a biological “super-catchment” of 2.45 million km², which is homogenising European freshwater life. - The relative causes of biotic turnover “remains largely unknown”.

Research agenda (p. 350). Tockner calls for a comprehensive overview of floodplains to set priorities, and for research on novel ecosystems: how they form, their services and resilience, and new indicators of biodiversity change.

Tulla’s Upper Rhine “rectification” (p. 350), the key historical technology case in the chapter. - “More than 150 years ago, Tulla’s regulation of the Upper Rhine was a technical masterpiece and one of the largest construction projects worldwide during the nineteenth century.” - The river was shortened from 345 to 273 km, 2,200 islands were removed (more than 100,000 ha between Basel and Strasbourg alone), and 240 km of major dikes were built (Blackbourn, 2006). - The floodplain went “from a fishery and waterfowl paradise to a productive agricultural area.” - Lost baseline. Reconstructing native biodiversity is difficult because the first systematic inventory of bottom-dwelling invertebrates was compiled “decades after the end of the ‘rectification’” (Lauterborn, 1905; the reference list misprints this as 2005). - Legacy biodiversity. The rich biodiversity that survives in remnant channels and riparian forests “is most likely a legacy of past hydrogeomorphic processes” (p. 350). Tockner says “most likely”. The implication, which he does not draw out, is that present richness may reflect past conditions and understate ongoing loss. - Risk shifted downstream. “the reduction in natural retention areas has led to an increase in flood peaks in the downstream sections of the Rhine valley.”

Governance implications (p. 350). - Without reference conditions, conservation targets are hard to set and the Water Framework Directive, which “strongly depends on defined reference standards”, is hard to implement. - Even the best-protected floodplains (Austria’s Alluvial Zone National Park on the Danube, the Odra National Park in Germany, the Danube Delta) “have been irreversibly modified”. - “Traditional conservation and restoration strategies will simply not maintain” them. Adaptive management must treat them “as coupled ecological-social-technological systems”, with “Active manipulation” for multiple services (Dufour and Piegay, 2009). - The EU should reconcile “the partly contradictory requirements” of the Water Framework, Habitats and Floods Directives, and set up a European competence network on large rivers.

15.2 Flood forecasting and warning (pp. 352–355)#

Components of an effective system (p. 352). 1. detecting the risk of a flood-triggering situation; 2. quantitative flood forecasting with adequate lead time; 3. a reliable warning message; 4. issuing and disseminating it to communities at risk; 5. adequate action by communities; 6. “post-hoc audits to learn lessons and improve the system for the future.”

These are non-structural measures that “can save lives and reduce material losses and human suffering.”

Historical note (p. 352). Noah is cited as the oldest “early warning”, and floods were long seen as “divine punishment”. A 1523 Augsburg forecast of a February 1524 flood, based on “a peculiar conjunction of planets in the constellation Pisces”, “turned out to be false” (Brázdil et al., 2005): a confident, ungrounded forecast.

15.2.1 Flood forecasts (p. 352). - A forecast gives timing, location, intensity and how the flood will evolve, using monitoring data and models. - Lead time depends on catchment size. Flash-flood and urban catchments respond in minutes to hours, so “observations of rising river water levels may come too late”, and radar (“which brings its own issues”) or rainfall forecasts are needed. Large rivers may give weeks. - Improvement efforts centre on data assimilation and adaptive forecasting (Young, 2002).

15.2.2 Flood warnings (pp. 352–353). - Warnings contain more than forecasts: recommendations or orders to evacuate or flood-proof (Smith and Ward, 1998). - Nigg (1995) gives two functions: assessment (from detection to a risk message for the locality) and dissemination. - A warning converts “scientific forecasts into lay language”. Communities should be “persuaded to take specific remedial action in time.” “Such warnings should be ‘populist’ in tone and communication but their design is actually a skilled task requiring careful forethought and design” (p. 353). - Alerts versus warnings. Warnings “have much shorter lead times that must be accurate to maintain public confidence.” - Vaison-la-Romaine, 1992 (p. 353). Météo-France issued two accurate alerts 12 and 24 hours ahead, “but because the local authorities did not know which catchment would be affected no warning was issued and many people died.” The chain failed at the translation from alert to localised warning. The death toll is not given. - EFAS, run by the European Commission’s Joint Research Centre, “now produces flood alerts for the whole of Europe” (JRC, 2011). - Forecasts ahead of warnings. “It is often noted that forecasts have advanced markedly, while progress in warnings has lagged behind”, despite advances such as automatic phone and text services and internet forecasts (p. 353). No source is given for “often noted”. - Nigg’s three requirements for a warning message: its basis must be credible; it must explain how far a specific area is at risk; and it must tell people what they can do (p. 353). - Speed of reaction is essential. Quality indicators include penetration (the share of those who need the information who actually get it) and satisfaction (pp. 353–354).

15.2.3 Flood warning errors (p. 354). - Two error types: (1) a warning is issued but the risk does not materialise; (2) no warning is issued but the risk and disaster occur. - Crucial distinction: the first type “does not include situations where the risk has materialised but the disaster has not.” - Netherlands, January 1995. A warning led to “massive evacuation”. The levees held, but the warning and evacuation “were justified and perceived by the population as the right decision. The risk of dike failure was high.” - Słubice (Poland), summer 1997. The town was evacuated but not inundated, thanks to major dike strengthening “and dike breaches upstream in Germany, which reduced the load”. One place’s defence failure relieved another. The chapter mentions this without comment. - “officials often hesitate to issue warnings due to fear of error, especially when warning systems are just developing or when there is still a great deal of uncertainty about the occurrence of the future event” (Nigg, 1995).

15.2.4 Credibility and efficiency (p. 354). - “The more personal the manner in which a message is delivered, the greater its credibility.” - In the UK, people in risk zones can sign up for phone, text or email warnings, but take-up “has not been high”, 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.” The chapter states the financial motive explicitly. The label chosen ignorance is mine. Strictly, the chapter describes reluctance to acknowledge the risk formally, not proof that people are unaware of it. The chapter also says “partly”, and it gives no data. - Efficiency factors: the credibility of the source (“which may vary according to the recipient”), accessibility of the channel, usefulness or redundancy of the information, and “the communication of a system’s resistance to floods.” Wording should be something the public “can understand and empathise with”, via channels the public “finds credible”.

15.2.5 Long-term early warnings (pp. 354–355). - Recent work stresses a second type of early warning when projections show “considerable changes in the anticipated risk”. Such warnings are “necessary to upgrade defence strategies”, for example the “time-consuming and resource-intensive task of strengthening levees” (p. 354). The lead time of the response (years to decades) drives the need for early warning. - Land-cover change and urbanisation raise risk “regardless of any change in climate”. - Very wet days contribute more to annual precipitation (Trenberth et al., 2007). - Floods from heavy rain are projected to increase and floods from snowmelt to decrease. However, snow cover may increase in cold areas where winter precipitation rises (p. 354). - Projections (pp. 354–355). Hirabayashi et al. (2008) and Dankers and Feyen (2008): - today’s 100-year floods become more frequent by 2071–2100 in much of Poland, France, the UK, southern Sweden and northern Italy, and less frequent over most of Finland and European Russia; - in aggregate, the control 100-year flood “is projected to become more frequent over more than 40 % of Europe”, and over 30% of Europe its recurrence interval falls “from 100 years to below 50 years by 2071–2100” (Kundzewicz et al., 2010b). - Stationarity (p. 355). “Systems continue to be designed and operated assuming stationarity (the past is the key to the future).” Because this “is clearly incorrect” (Milly et al., 2008), “existing design procedures need to be revised”. Otherwise systems will be “under- or over-designed and either not serve their purpose adequately or be overly costly (with excessive safety margins)”. This acknowledges that precaution has costs too. - Acting without precise numbers (p. 355). Detection difficulties and projection uncertainty “mean that no precise quantitative information can be delivered for long-term flood preparedness planning. Nevertheless”, water managers in Germany, the Netherlands and the UK “have begun to take account of these early climate change warnings explicitly in flood protection design codes”: - Bavaria: design values assume 40–50% more small and medium flood discharge by 2050 and about 15% more for 100-year floods. - UK (Defra, 2006): a “precautionary allowance” for peak river flow of +10% up to 2025 and +20% “after 2085”. The latter date should be checked against Defra’s guidance; I recall the bands as 2025–2115 but have not verified this. - Netherlands: a “climate change factor” in new plans. Rhine design discharge is to rise from 15,000 to 16,000 m³/s by 2015 and to 18,000 m³/s in the longer term (Kundzewicz et al., 2007). - This is a clear example of acting on projections before statistically robust detection. The chapter reports it approvingly but does not evaluate it.

15.3 Flood preparedness systems: the science and its use (pp. 355–356, 359)#

• “Complete flood safety is impossible in low-lying areas adjacent to rivers.” Risk can be “considerably restricted” by a “site-specific mix of measures” (p. 355).
• Research funders are listed as including “the insurance and reinsurance industry”. The chapter calls research “indispensable to optimise preparedness systems” and does not discuss the funders’ interests.
• FLOODsite (p. 355). An EU FP6 integrated project (Klijn, 2009) treating risk as “a combination of hazard sources, pathways and consequences”. The chapter says the Floods Directive “explicitly refers to FLOODsite as the largest ever EU flood research project”. I doubt the legal text of Directive 2007/60/EC names FLOODsite; this needs checking. The project linked rainfall-runoff models to real-time radar and satellite data for flash floods and assessed the uncertainty of radar rainfall estimates.

15.3.1 Protect, adapt, retreat (pp. 355–356). - UK Foresight Future Flooding (Evans et al., 2004; Hall et al., 2005). “if existing policies are continued the flood risks in 2080 could increase substantially”, ranging from little increase to “a 20-fold increase” depending on scenario. The choice is between investing in sustainable approaches and “learning to live with increased flooding”. “An integrated portfolio of responses can reduce future risk by an order of magnitude”, although effectiveness “depends very much on the scenario”. - European strategy: “‘protect as far as technically possible and affordable’ and otherwise ‘adapt and accommodate’”, also called “living with floods” (Germany, Netherlands) and “making space for water” (Netherlands, UK). Retreat is the fallback (p. 356). - Measures follow flood frequency (Kron, 2005): natural measures for frequent floods, engineering for rare floods, and “residual risk management is essential for very rare floods”, including “those beyond the limits of past experience” (p. 356).

15.3.2 Structural and non-structural measures (pp. 356, 359). - Structural (“hard”) measures: dikes, dams, reservoirs, diversions, floodways, channel capacity. Their dimensions rest on probability theory for a “design flood”, for example a 100-year flood, “(although this is difficult to determine in practice)”. - Value judgement. A longer return period gives better protection at greater cost; “This raises various value judgements, such as whether to design dikes to withstand a 100-year flood or perhaps a 1 000-year flood.” “Moreover, it is misleading to expect complete flood protection or total certainty of outcomes” (p. 356). - Levee effect. Dikes “protect well against small- and medium-size floods”, but when a disastrous flood breaks them, “losses in a levee-protected landscape can be higher than in the absence of a levee”. The reasons given are “the false feeling of security” levees create and the “high damage potential in apparently (but not completely) safe areas.” “A dike designed for a 100-year flood is likely to fail if a 1 000-year flood occurs” (p. 356). No data or reference support the claim here; it is a well-known idea in the hazards literature that the chapter does not cite. - Carlisle, 2005 (p. 356). The flood came while plans for new defences were out for consultation. The planned defences met the UK 100-year standard and “if built they would have been breached”, because the event exceeded a 150-year return period. The defences were redesigned for 200 years. Beven’s panel dates it to January 2005 (p. 358). - High standards elsewhere. Japanese super-dikes 300–500 m wide protect major cities (Kundzewicz and Takeuchi, 1999). The Dutch Flood Defence Act of 1996 set design return periods of 1,250 years for middle and upper rivers and 2,000 years for lower reaches (Pilarczyk, 2007). - Non-structural (“soft”) measures: source control and watershed management, laws, zoning, economic instruments, forecast-warning systems, risk assessment, awareness-raising and databases. Source control (“catching water where it falls”) counters the effects of urbanisation; restoring storage in floodplains, polders and washlands matters. - Insurance and aid (pp. 356, 359). Insurance spreads risk “over many people and over a longer time”; aid compensates for uninsurable losses. UK insurers use flood-risk maps to price premiums. Post-flood aid “is essential to restore the livelihoods of survivors.” The chapter does not raise moral hazard. - Retreat (p. 359). Despite “encouraging examples” such as the US after 1993, “the permanent evacuation of floodplains is virtually unthinkable in most countries.” Bangladesh, “the most flood-prone country on earth”, has a growing population, depends on floods for soil fertility, and had more than two-thirds of its area flooded in 1998. New embankments would take scarce land. Its options are to reinforce existing defences and improve non-structural measures, including forecasting and warning. “optimum strategies for flood protection must be site-specific.”

15.3.3 Uncertainty in flood risk assessment (p. 359). - “there is much uncertainty in the hydrological studies that underpin flood risk assessment and management (e.g. in determining a 100-year flood).” - Beven (2006a), Hall et al. (2007) and Sivakumar (2008) judged uncertainty analysis in hydrology “highly unsatisfactory”. “Although disagreeing in significant respects, all called for the promotion of uncertainty analysis”. It “should not be an add-on element — an afterthought of little importance. This is easier said than done, however, as there is considerable uncertainty regarding uncertainty estimation.” - Remedy: “greater rigour and consistency in analysing and reporting uncertainties.” - Data scarcity. “There is an overwhelming scarcity of homogeneous long-term observation records.” - Instruments fail in extremes. Measuring extreme flows is hard because “rating curves are not available for the high flow range, gauges are destroyed by flood waves or observers are evacuated”, so indirect estimation is needed. The measurement system fails exactly when its data matter most. - Uncertainty in future projections “is very high … and grows the further we look into the future”. Near-term (the 2020s) is dominated by climate-model uncertainty, later periods by emission scenarios. - Scale mismatch. Coarse climate models versus drainage basins require downscaling. Much finer data are needed “for the ‘point’ scale of a locality (e.g. a small riparian town), which is the level at which costly adaptation is undertaken.” Decisions are made at a finer scale than the science can inform.

Panel 15.2 (Beven): uncertainty in predicting floods (pp. 357–358)#

Framing (p. 357). - The US IFMRC called the 1993 Mississippi rainfall “without precedent” (Galloway, 1995). The UK Pitt Report (2007) called summer 2007 “exceptional” and noted demand for better warnings. - The demand “appears to pose a relatively simple problem for hydrological science.” Beven sets aside the organisational and social problems to focus on the science, and generalises: this type of prediction is “increasingly demanded” and extends “all the way up to predictions about the future climate”. - “In fact, many of these apparently simple science problems turn out to be complex and fraught with uncertainties that are difficult to evaluate.”

Sources of uncertainty (p. 357). - Measurement limits on the pattern and intensity of rainfall and on the discharge–level relationship in extreme flows. - How much rain reaches the river and when, partly because of “a highly non-linear relationship between the wetness of a catchment before an event and runoff generation processes”. UK examples preceded by prior wetting: summer 2007, Lynmouth 1952, Boscastle 2004, Carlisle 2005, Cumbria 2009. - Models “can only approximate”. - Uncertainty in flood frequency, in future change, and in the effects of land management.

The debate (p. 357). Uncertainty estimation has been a major research topic for two decades. There is “still significant debate”, with “a range of strongly held opinions and some interesting debates. The failure to reach more general consensus over this period may appear surprising but there are some fundamental issues involved, which apply much more generally in environmental modelling”.

Aleatory versus epistemic (pp. 357–358). - Aleatory uncertainty comes from natural variability. Epistemic uncertainty comes from “the limitations of our knowledge about the system”. “Unknown unknowns” cannot be treated explicitly (Sivakumar, 2008). - It is “commonly assumed” that most uncertainty can be treated as aleatory, which allows “the full panoply of statistical theory”. Beven (2006b) argues this holds only in “ideal cases”. - “Treating the sources of uncertainty as if they were aleatory in non-ideal cases will produce over-confidence in the model predictions because of the time-variable nature of epistemic uncertainties” (p. 357). - “There is no general theory of how to handle epistemic uncertainties”. Fuzzy-set and possibility frameworks add flexibility but “more subjectivity”. - Probability theory is “logically consistent and objective — but if and only if the assumptions about the nature of the errors can be shown to be consistently valid and this is rarely the case.” “small departures from the ideal can lead to overconfidence” (Beven et al., 2008) (p. 358). - Key empirical claim: “In real applications, most sources of uncertainty have both aleatory and epistemic components, but it will often be the epistemic component that dominates” (p. 358).

Applied to flood inputs (p. 358). - Rain gauges, radar and numerical weather prediction (NWP) are all subject to epistemic error. - NWP “is useful for identifying potential flood events but does not (yet) give generally reliable rainfall forecasts”. Yet such forecasts are essential for flash-flood warning, “since there is no other way to provide warnings to the public with sufficient lead time.” - Gauge networks may miss convective cells and orographic effects. - Radar does not measure rainfall directly, and its reflectivity–rainfall relationship “will vary from event to event”.

For scientists and decision-makers (p. 358). - “There is no question that it is better to estimate some form of uncertainty … than to ignore the uncertainties.” - For the scientist, the problem is epistemic uncertainty. For the decision-maker, it is how to interpret uncertainty estimates “and under what circumstances not to rely on model predictions when making a decision.” - Flood-specific problems: conveying warnings when uncertainty can produce “one or a succession of false alarms”, and explaining (citing Kundzewicz) “to the public (and even to some decision-makers) that there is a finite possibility that even a new flood defence scheme may be breached … (or, as in Carlisle in January 2005, before the new design has been built).” - Uncertainty estimates matter if they change decisions. They are easier for users grounded in probability, but “can be misleading if they lead to overconfidence”. - “Essentially, science lacks a theory of the true information content of data that is subject to epistemic uncertainties (Beven, 2008).”

Practical route (p. 358). - “involve potential users of model predictions earlier in the prediction process”, so that decisions about evaluating model performance and handling uncertainty are made jointly. - The Environment Agency of England and Wales “is starting to implement this approach within the framework of ‘guidelines for good practice’”. - He notes that stakeholder involvement in prediction “has been advocated for some time (e.g. Stirling, 1999; EEA, 2001)”. EEA 2001 is the first Late Lessons report, so this is a self-reference within the project. The guidelines should give “a framework for agreeing on reasonable assumptions” about both types of uncertainty.

Assessment. The panel is the chapter’s most general and most transferable passage. It is also, by Beven’s own admission, one side of an unresolved dispute. The panel does not set out the opposing positions. It cites Hall et al. 2007, Todini and Mantovan 2007 and Montanari 2007 only as sources showing that uncertainty estimates are easier to understand for users grounded in probability (p. 358). My inference, from reference titles and background knowledge (to verify): those authors favour more formal statistical and Bayesian treatments and contest his framing. Also from background knowledge (to verify): the dispute centred on Beven’s GLUE method and its critics’ charge of statistical “incoherence”, which is the subject of Beven et al. (2008), “So just why would a modeller choose to be incoherent?”.

15.4 Lessons from floods (pp. 359–362)#

• First lesson. “there are always lessons to be learned, from every flood” (p. 359). Learning means understanding “the reasons for a system’s failure or inadequate performance, and identifying weak points using an holistic perspective.”
• The chain (p. 359). Any link can fail: observation, forecast (“grossly in error”), message, communication or response. “A single weak point … may render the overall system performance unsatisfactory.”
• Institutions (pp. 359–360). Components “must be adequately integrated but responsibility for them may reside with different agencies.” In emergencies, “it may become evident that distribution of roles of agencies is unclear and possibly redundant.” “The institutional framework is a key socio-economic determinant of a nation’s vulnerability against natural disasters” (Raschky, 2008).
• Learning from experience (p. 360). When large floods hit the same place twice in a short period (the Rhine at Cologne, December 1993 and January 1995), “losses during the second flood are typically far lower than those during the first” (Munich Re, 1997; Kundzewicz and Takeuchi, 1999). The first flood teaches homeowners, farmers, water professionals, legislators, zoning officers and administrators at every level.
• The hydro-illogical cycle (pp. 360–361). “memories can fade quickly after a flood”:
• a destructive flood generates enthusiasm and heavy spending, and authorities “elaborate ambitious plans and launch works but lessons are soon forgotten”;
• “After some time without flooding, willingness to pay for flood preparedness decreases sharply and projects are downscaled or suspended”;
• the next deluge starts a new cycle.

The concept came from Donald Wilhite in the drought context in the mid-1980s; no reference is given. The chapter calls it “a general principle, valid across different political and economic systems.” Cause: “The return period of a destructive flood is usually much greater than the political horizon of decision-makers and the electorate”, which is set by terms of office and electoral cycles. “Of course, the hydro-illogical cycle is also at odds with the precautionary principle” (p. 361). This is the chapter’s only explicit mention of the principle. - Remedy by law (p. 361). “In some countries, codifying preparedness in legislation helps overcome the hydro-illogical cycle”, with the Floods Directive as the leading example. This is presented as a mechanism, but no evidence is given that codification has worked. - Mississippi lock-in (p. 361). In the mid-19th century (Williams, 1994) Congress considered two options: (a) using large parts of the floodplain for storage and overflow, or (b) embanking the river “in a single channel isolated from its floodplain”. It chose (b). “the decision has remained influential on flood protection policy in the US and elsewhere, leading to the transformation of rivers and reduction of wetlands worldwide.” - Table 15.1 dates the adoption of a levee-based policy to 1861, “following a report by the US Army Corps of Engineers” (Smith and Ward, 1998). From background: this is the Humphreys–Abbot report. An earlier report by Charles Ellet (1852) had advocated reservoirs and outlets; this is my addition, not in the chapter. - In 1936 the federal government “assumed primary responsibility for flood damage reduction” and, over half a century, “embarked on a multibillion dollar programme of structural defences” (Galloway, 1999). - 1993 as paradigm shift (pp. 361–362). The great mid-west flood “proved that structural ‘hard’ defences cannot guarantee absolute protection.” IFMRC recommended funding the purchase of land and structures “from willing sellers”, and “many vulnerable families have been relocated”. “However, this response is not universal. In most countries, people who suffer in a flood rebuild their houses (possibly more robustly) and their livelihoods in the place devastated by the flood … The hazard may not have decreased, however”. - Lessons from failed policies (p. 362). - Some protection infrastructure “limits options for future generations and introduces disturbances in ecosystems.” - The Environment Agency (1998) says schemes should protect the present generation but also “avoid as far as possible committing future generations to inappropriate options for defence”. - Renaturalisation “is now actually likely to come about”. - Some large reservoirs requiring flooded land or displaced people “certainly did not match the principles of sustainable development” (Kundzewicz, 1999). Decommissioning studies are under way, and some reservoirs have been decommissioned in France (Kernansquillec, Maisons-Rouges, St Etienne du Vigan). - One-sided decisions (p. 362). Major decisions such as building a large dam are often “based on one-sided arguments with important aspects neglected”. This bias “can result from a lack of knowledge and understanding at the time … as well as the evolution of value judgements over time.” This is fair-minded: it separates culpable one-sidedness from what could not have been known and from values that have shifted since. It partly guards against hindsight bias.

Box 15.1: Lessons learned from the 1997 Odra flood (p. 360)#

• The July 1997 flood came “after a long flood-free period” and exposed weaknesses in Poland’s preparedness, “in particular the deficiencies of relevant legislation” (Kundzewicz et al., 1999).
• Legal transition. “the previous regime’s laws were essentially abandoned and many new acts were passed during a short time. The distribution of responsibilities was ambiguous and conflictual”.
• Authority to warn. “low-level authorities were not entitled to announce a flood alert or the alarm status.” Such decisions had to come from provincial anti-flood committees “and were delayed as a result. Local authorities typically resorted to common-sense decisions, without waiting for instructions from above.”
• Information flow. Hydrometeorological stations reported upward to regional branches and gave information to local authorities only “on request”. “Some forecasts proved to be of low accuracy.”
• Roles and costs. Responsibilities and cost coverage for the army, police and fire brigades were undefined. “Polish civil defence was geared to act in the event of a war, rather than to deal with a peacetime emergency.” The institution was designed for a different threat.
• Infrastructure failure. About 189,000 telecommunication links were cut. Mobile phones were more reliable but limited.
• Lead time available but hard to exploit. Advance warning was available for the middle and lower Odra once the flood began in the headwaters in the Czech Republic and Poland. “In principle”, Brandenburg “had ten days before the arrival of the floodwater”. “Even so, detailed forecasts were difficult to obtain” because observations at several gauges were interrupted and the flood information office in Wrocław was itself flooded. The box does not say how Brandenburg used the lead time, so it gives no basis for calling the time wasted.
• Improvements. Weather radar network, automated observation and transmission, upgraded warning centres with telecoms that work without mains power, modernised warning, better regional, interregional and international information flows, and better forecast models. “Considerable investments were made”.
• Fast learning. “When a second flood wave occurred in July 1997 the preparedness and flood management were far better than during the first crest when the nation had largely been taken by surprise.”
• Table 15.1: the 1997 floods (Odra/Oder and Vistula) caused 110 deaths and damage “in the range of billions of dollars — a substantial portion of the GDP of the countries in economic transition” (p. 365).

Box 15.2: EU Floods Directive (p. 361)#

• Background: “Between 1998 and 2004, Europe suffered over 100 major destructive floods”, including the record August 2002 floods on the Danube and Elbe. “In response to these floods and projections of growing flood risks”, the EU adopted Directive 2007/60/EC “in April 2007”. The reference list dates the Directive to 23 October 2007, so the chapter is inconsistent. From background: the Parliament vote was in April and formal adoption in October.
• Scope: river, flash, urban, sewer and coastal floods. Objective: reduce and manage risks to health, environment, infrastructure and property. Background, to verify: I understand the Directive lets Member States exclude sewer flooding.
• Member States must produce, for each river basin or management unit: 1. a preliminary flood risk assessment (maps, past floods, processes and their sensitivity to change, development plans, likelihood of future floods including climate change and land-use trends, and estimated consequences); 2. hazard and risk maps for high probability (“a 10-year return period”), medium probability (100-year) and low probability (extreme events); 3. flood risk management plans “aimed at achieving the required levels of protection”.
• Background, to verify: the Directive text specifies “medium probability (likely return period ≥ 100 years)” and “high probability, where appropriate”. The 10-year figure appears to be the chapter’s own gloss.
• Flexibility. Member States decide “the level of protection required, the measures needed … and the timetables”. The chapter does not discuss whether this flexibility weakens the Directive.
• Evaluation (a prediction). “The directive is probably the most advanced flood protection and preparedness legislation worldwide and implementation should considerably reduce flood risk throughout the 27 EU Member States.” It is expected to produce “an unprecedented multinational upgrading of preparedness systems.” It quotes the Directive’s own claim that “the potential future damage to be expected if no action is taken distinctly outweighs the costs”. This is endorsement of a policy too new in 2013 to have been evaluated.

15.5 Conclusions: living with floods (pp. 362, 364)#

• “Floods are natural events that cannot be avoided.” Protection should be “up to an agreed safety level, this being a compromise between the desired level of safety and the accepted willingness to pay for protection.” Acceptable risk is presented as a negotiated social choice.
• The paradigm change is restated: “living with floods … seems more sustainable than hopelessly striving to eradicate them” (Kundzewicz and Takeuchi, 1999). “There must be action plans for events exceeding the design flood (i.e. when defences are bound to fail) and in this context early warning can save lives.”
• Myths (p. 362). “Misconceptions and myths about floods and flood protection are deeply rooted in society”, among the public, politicians and decision-makers alike. “Some people naively believe” that floods come at regular intervals, taking “return period” literally rather than as an average, and “that embankments offer perfect safety.” No evidence is cited.
• “Each tick of the clock marks the passage of time since the last flood and the countdown to the next.”
• Portfolio of measures (p. 362):
• rigorous zoning, “leaving floodplains with low-value infrastructure”;
• strengthening existing defences;
• monitoring;
• forecasting;
• warnings;
• evacuation;
• relief and recovery;
• insurance;
• capacity-building;
• “enhancing a participatory approach, including consultation on the preparedness strategy and the level of flood protection”, plus household flood-proofing for new and existing buildings.

The zoning item as printed reads “using regulations to develop flood hazard areas”. This is probably a slip for regulating or restricting development. - Participation and conflict (p. 362). “Only informed stakeholders can make rational decisions and agree on an acceptable flood protection strategy”. There may be “conflicting interests between those living in floodplains and demanding efficient and very costly protection, and the rest of the nation.” This names a distributional conflict but does not analyse it. - Awareness (p. 362). “Many past fatalities could have been avoided with greater awareness.” In the US, “most flood fatalities involve vehicles whose drivers underestimate the danger.” Unreferenced. - Structural measures still needed (p. 364). “Despite their shortcomings, hard structural flood protection measures, such as dams and levees, will be needed to safeguard existing developments, in particular in urban areas.” The ideal is a mix, with non-structural measures “normally conforming better to the spirit of sustainable development.” The chapter is fair here and does not dismiss engineering. - Anticipation (p. 364). “Mitigating flood risks requires a change from reaction to anticipation.” The challenge is better forecasting across time horizons, from quantitative precipitation forecasts for flash floods to longer-term forecasts for large basins. “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.” Unreferenced. - Encroachment as the root cause (p. 364). Smith and Ward (1998) identify floodplain development “as the major factor increasing flood risk”. “Floods constitute a danger to life and property only when humans encroach into flood-prone areas and become vulnerable.” Retreat is the remedy where land is already developed, but “In many countries the strategy of retreat is unpalatable, however, favoured by neither the broader population nor decision-makers.” In Bangladesh “it is simply not an option.” - Modern risk management (p. 364). Prevention, mitigation and preparedness, then pre-planned action during and after events. A “risk-based approach” analyses “probability and consequences of flooding across the full range of severity” and manages “all possible events”. Managers “should continuously acquire and update evidence about long-term changes in flood risk.”

Panel 15.3 (Vellinga and Aerts): coastal flooding (pp. 363–364)#

Scale (p. 363). - Coastal storm surges and floods are “the most frequent and costly extreme weather events occurring in Europe, representing 69 % of the overall natural catastrophic losses (CEA, 2007)”. The reference list has CEA 2008. The figure looks as if it may combine storms and floods generally; verify. - Xynthia (2010): France hardest hit, “51 casualties”, damage above EUR 1.5 billion. The cited AIR source’s title gives insured losses of USD 2–4.1 billion. Sea walls around the Île de Ré were damaged or destroyed. - The 1953 North Sea storm killed more than 2,000 people in the UK and the Netherlands. Surges also hit Hamburg, Bremen and the Baltic, the Rhône, Po and Danube deltas, and Venice (“Aqua Alta”).

Rising risk (p. 363). - Sea level has risen about 10 cm per century over the last two to three centuries and is expected to rise faster (IPCC, 2007). - Land subsidence from soil type and water extraction adds to this. - More people and economic activity sit in high-risk areas (Bouwer et al., 2007).

Surprise and forgetting (p. 363). - People are “usually taken by surprise by coastal floods due to their ‘low probability and high impact’ character.” Vulnerability rises “very gradually” with sea level and development. “There is never an acute reason for strengthening the coastal protection system or for elevating settlements until the area and its population are hit by a major flood”. - Dutch records show “people have been taken by surprise about once every hundred years for the last 1 000 years.” “Apparently after some 50 years the flood disaster tends to disappear from the collective memory”, and protection is then “insufficiently maintained”. - “The Netherlands and the United Kingdom only began investing in large-scale storm surge barriers after the devastating storm in 1953.” The challenge now is “to anticipate climate change and accelerated sea-level rise.”

Measures (pp. 363–364). - Surge barriers (London, the Netherlands, St Petersburg, Venice), expensive but built for roughly 1-in-1,000-year conditions. - Closing estuaries harms biodiversity and water quality, hence “building with nature”: beach nourishment and marsh development (Day et al., 2007). - UK and French building codes and zoning. - French community-based prevention plans, encouraged by better insurance terms (Letermy/Letremy, 2009): an insurance incentive for prevention. - Hamburg and Rotterdam are building outside the main defences but raising buildings. Hamburg is raising roads and bridges to 7.5 m above sea level for emergency access (Aerts et al., 2009).

Climate proofing (p. 364). - Projections are “inherently uncertain”. Levees and barriers “are currently designed according to deterministic principles, using relatively short historic records and extrapolation methods”, but extremes are rare and data sparse. “Hence, new probabilistic methods are needed that provide engineers with a range of possible scenarios”. This terminology differs from the chapter’s “probability theory” description of design (p. 356); both describe extrapolating a design event from short records. - “Superlevees” can host urban development (Vellinga et al., 2009). “Such unbreachable dikes were developed in Japan near Tokyo, to protect against the potentially very high seas caused by tsunamis.” - Two tensions. (a) The word “unbreachable” sits uneasily with the chapter’s central lesson that no defence guarantees safety (pp. 347, 356, 362). It may be meant narrowly, as resistant to breach rather than to overtopping, but the panel does not say so. (b) The main text presents the Japanese super-dikes as flood protection for “major cities in Japan” in a chapter about river floods, with no mention of tsunamis (p. 356). From background, to verify: they were built mainly along rivers against overtopping and breach. - Dutch water boards are considering superlevees. - Sand is now placed on the foreshore. Planners, developers and insurers “are actively engaged in reconsidering current regulations” (Burby, 2001).

Table 15.1: Early warnings and actions (p. 365)#

Observations. - The table carries the report’s standard title, “Early warnings and actions”, but it mainly lists disasters and policy or technology milestones, not warnings and responses. - The forecasting entries (1854, 1970s) are about warning technology, in the chapter’s operational sense of early warning. - No entry records a warning about the harms of the structural flood-control paradigm itself. The closest is the 1994 IFMRC recommendation, which came after the 1993 disaster, not before it. - The table therefore gives little help in measuring the delay between warning and action.

References (pp. 365–368)#

About 77 entries. Citation and consistency problems noted: - Lauterborn 1905 in the text versus 2005 in the references; - Tockner et al. 2002 cited but not listed; - “Oppermann et al., 2008” in the text versus “Opperman … 2009” in the references; - CEA 2007 in the text versus CEA 2008 in the references; - “Letermy” versus “Letremy”; - Andréassian et al. (2007) listed but not cited; - Wilhite (the hydro-illogical cycle) cited but not listed; - the Floods Directive dated April 2007 in Box 15.2 versus 23 October 2007 in the references; - Kabat et al. (2005) page range misprinted (“238–284”); - editor misspelled in Nigg (1995) (“Hotlick-Jones”); - further minor slips visible in the list itself: “Robsinson” (Tockner et al. 2009 editors), and “Neill, R.V.” and “van der Belt” in Costanza et al. 1997. From background, to verify: these should read O’Neill and van den Belt, Brázdil et al. in Hydrol. Sci. J. 51(5) is probably 2006 rather than 2005, and the Faulkner et al. Ambio volume “(16/7)” is probably 36(7).

None of these changes the substance, but they indicate light editing.

Case timeline (composite; this is a thematic chapter, not a single case)#

The chapter has no single warning-to-response arc. I have reconstructed several sub-arcs from the text and Table 15.1.

A. The structural “flood control” paradigm and its reversal (US, with parallels in Europe).

• Lag: about 130 years from the 1861 choice to the 1993–94 official recognition in the US. The chapter does not present this as “warning ignored”, and it identifies no earlier warning that was dismissed. From background, not in the chapter: Ellet (1852) proposed reservoirs and outlets, and Gilbert F. White’s Human Adjustment to Floods (1942) argued against relying on structural works alone and described the effect of development moving in behind defences. So credible warnings existed decades earlier than the chapter’s narrative suggests. Verify before use.
• What drove the change: disaster (1993) and cost (Switzerland), not warnings.

B. Short-term warning systems: failures and learning.

• Lag: learning within weeks in the Odra case (between flood crests). The time from Vaison-la-Romaine to EFAS is not stated in the chapter.

C. Coastal flooding: the surprise cycle (Netherlands and UK). - For a millennium, the Dutch were “taken by surprise about once every hundred years”, with a collective memory of about 50 years (p. 363). - 1953: more than 2,000 dead, after which surge barriers were built. Investment came only after disaster (p. 363). - 1996: Dutch Flood Defence Act sets 1,250- and 2,000-year river standards (p. 356).

D. Long-term (climate) early warnings: acting before detection. - IPCC (2001a): analysis of extremes “underdeveloped” (p. 351). - 2003–2010: projection studies (Hall et al.; Hirabayashi et al.; Dankers and Feyen; Kundzewicz et al.) warn of more frequent floods over large parts of Europe (pp. 354–355). - 2005: Kundzewicz et al. and Svensson et al. find no conclusive global trend in observed high flows (p. 351). - 2006: Defra precautionary allowances (p. 355). - 2008: Milly et al. declare stationarity “dead”; Wilby et al. expect robust trends to take decades to appear (pp. 351, 355). - By 2013: Bavaria, the UK and the Netherlands build climate factors into design codes (p. 355). - Lag: short. Design-code adjustments followed the projection-based warnings within a few years (Defra 2006; Bavaria and the Netherlands by 2013). They came before statistical detection of trends in observed river flows. The warning here is the projection, not a detected trend, so the lag between warning and action is not negative. What is distinctive is acting before confirmation. My judgement, not checked against other chapters in this audit: that is unusual in this report.

E. European policy response. - 1998–2004: more than 100 major destructive floods. August 2002: more than EUR 20 billion. April/October 2007: Floods Directive (p. 361). - The EU response came about 5 years after the 2002 floods. Implementation is in national hands with flexible timetables.

Harms and costs cited across the chapter. - Global losses up ten-fold, 1950s to 1990s. - China 1998: about USD 30 billion and more than 3,600 deaths. - China 1931: 145,000 to 3.7 million deaths. - Europe 2002: more than EUR 20 billion. - Rhine: EUR 165 billion potential damage and more than 10 million people at risk. - Odra/Vistula 1997: 110 deaths. - 1953 North Sea: more than 2,000 deaths. - Xynthia 2010: 51 deaths, more than EUR 1.5 billion. - Switzerland: CHF 15 billion in flood costs and CHF 45 billion in engineering spending. - Ecological: more than 90% of former floodplains in Europe, North America and Japan functionally extinct or converted to cropland and urban areas.

The authors’ own lessons and conclusions#

Lessons derived from evidence in the chapter: 1. No defence guarantees safety; beyond-design events will happen. Evidence: the sampling effect and short records (pp. 351–352), the 1993 US flood (p. 361), Carlisle 2005 (p. 356), the levee effect (p. 356). Hence action plans for events that exceed design (p. 362). 2. Warning systems fail at their weakest link. The chapter lists technical and human links alike: observation, forecast, message, communication and response (p. 359). Its two illustrative cases put the weak link mainly in institutions: local authorities in Vaison-la-Romaine 1992 (p. 353), and legal authority and roles in the Odra 1997 (Box 15.1, p. 360). The Odra case also involved technical failures: gauges, telecoms and low-accuracy forecasts. The chapter does not claim that the weak link is usually human rather than technical. Stated generally on pp. 347 and 359. 3. Institutional design is a key determinant of vulnerability. Evidence: Odra and Raschky (2008) (pp. 359–360). 4. Experience reduces losses in a repeat event. Evidence: Cologne 1993/1995, the Odra second crest (p. 360). 5. Memory fades, producing the hydro-illogical cycle. Asserted as a general principle (pp. 360–361). Panel 15.3 supplies the Dutch 1,000-year record (p. 363). 6. Design assumptions of stationarity are wrong and codes need revising. Evidence: Milly et al. (2008) and the projection studies (p. 355). 7. Uncertainty analysis is inadequate and should be central, not an afterthought. Evidence: the critiques cited (p. 359) and Beven’s panel (pp. 357–358). 8. Past big-infrastructure decisions were often one-sided, because of knowledge limits and changing values (p. 362). Evidence: examples of reservoirs and decommissioning, but no systematic study. 9. Structural defences have ecological costs and can raise downstream risk. Evidence: Panel 15.1 on Tulla and the Rhine and on Switzerland (pp. 349–350).

Recommendations and advocacy (normative): 1. Adopt “living with floods” and accommodation in planning (pp. 347, 356, 362). 2. Set protection at an “agreed safety level” that balances safety against willingness to pay (p. 362). 3. Use a site-specific portfolio: zoning, defences, monitoring, forecasting, warnings, evacuation, relief, insurance, capacity-building, participation and household flood-proofing (pp. 355, 359, 362). 4. Codify preparedness in legislation to break the attention cycle; praise for the Floods Directive (p. 361). 5. Move “from reaction to anticipation”; improve forecasting across time horizons (p. 364). 6. Base management on risk across the full range of severity and keep updating evidence on long-term change (p. 364). 7. Keep options open for future generations; renaturalise and decommission where appropriate (p. 362). 8. Inform stakeholders and involve them in setting protection levels (p. 362). 9. Beven: bring the users of predictions into the modelling process early (p. 358). 10. Tockner: conserve and restore floodplains as strategic resources; manage them adaptively as coupled social-ecological-technological systems; reconcile the EU directives; create a competence network (pp. 349–350). 11. Vellinga and Aerts: climate proofing, probabilistic design, superlevees, building with nature, insurance-linked prevention plans (pp. 363–364).

Predictions embedded in the conclusions: - the Floods Directive “should considerably reduce flood risk” (p. 361); - flood risk “is likely to grow” (p. 351); - the 100-year flood becomes more frequent over more than 40% of Europe by 2071–2100 (p. 355); - robust trends are unlikely to be detectable for decades (Wilby et al., p. 351); - renaturalisation is “now actually likely to come about” (p. 362).

Mechanisms and dynamics#

1. How warnings arise and how they fail to become action. - The chapter’s clearest contribution is its anatomy of a warning system as a chain: detection, forecast, message, dissemination, response, audit (pp. 352, 359). - Failures cluster at the interfaces between organisations. In Vaison-la-Romaine, the forecaster’s accurate but spatially vague alert did not match the local authority’s need for a catchment-specific trigger (p. 353). In the Odra, legal authority to raise the alarm sat at provincial level while local authorities had the knowledge and urgency (p. 360). - An accurate signal can be wasted because the next link cannot act on it in the form it arrives. - “forecasts have advanced markedly, while progress in warnings has lagged behind” (p. 353). Investment and professional prestige flow to prediction science more than to the social machinery of warning. This is my inference; the chapter only notes the gap.

2. Fear of error and the asymmetry of blame. - Officials “hesitate to issue warnings due to fear of error”, especially with new systems or high uncertainty (p. 354). - The chapter’s reframing is important: when an evacuation is followed by no disaster because the defences held, that is not a false alarm (Netherlands 1995, Słubice 1997; p. 354). - For the Netherlands 1995 only, the chapter says the evacuation was “perceived by the population as the right decision” (asserted, not evidenced). It says nothing about how Słubice residents perceived their evacuation. - Beven adds that a “succession of false alarms” erodes the meaning of warnings (p. 358). - The dynamic: officials weigh the visible cost of a false alarm against the diffuse cost of a missed warning. My inference, not a claim the chapter makes: read together, Nigg’s point about official hesitation and the Dutch 1995 example suggest the public may be more tolerant of precautionary action than officials fear. The chapter gives no data on this.

3. Chosen ignorance by the people at risk. - UK residents avoid signing up for warnings because acknowledging risk could lower house prices or make them uninsurable (p. 354). - Here the recipients of a warning have financial reasons not to receive it. Being identified as exposed carries a private penalty (property value), while the benefit (warning) is uncertain. - Insurers’ flood-risk maps set premiums (p. 359), which strengthens the incentive. The chapter does not connect these two points.

4. The measurement system fails in the extreme it exists to measure. - Gauges are destroyed, observers evacuated, and rating curves are unavailable for high flows (p. 359). - In the Odra, gauges stopped and the flood information office in Wrocław was itself flooded (p. 360); 189,000 telecom links were cut. - Consequences: (a) warnings fail at peak need; (b) the historical record of extremes is systematically thin, which feeds the short-record problem in design (pp. 352, 359).

5. Knowledge production and the handling of uncertainty. - Short records and the sampling effect. The “100-year flood” is estimated from records too short to define it (pp. 352, 356, 359). Record-breaking floods are expected even without change (pp. 351–352). - Stationarity as an embedded assumption. Design codes encode “the past is the key to the future” (p. 355). The chapter calls it “clearly incorrect”. - Misclassified uncertainty. Treating uncertainty that comes from limited knowledge as if it were random variability (“aleatory”) lets analysts use standard statistics, but “will produce over-confidence” (p. 357). There is “no general theory” for the other kind (“epistemic”) uncertainty (p. 357), and “considerable uncertainty regarding uncertainty estimation” (p. 359). - Unknown unknowns are acknowledged as untreatable by definition (p. 357). - Scale mismatch. Science works at coarse scales, while decisions and spending happen at the “point” scale of a town (p. 359). - Signal and noise. A real trend can be undetectable for decades against natural variability and confounding land-use change (p. 351). This makes “wait for proof” a structurally slow policy. - “Unprecedented” framing. Post-event reports call events “without precedent” (1993) or “exceptional” (2007) (p. 357). Read against pp. 351–352, such framing is statistically expected rather than a sign of the truly unforeseeable. This is my synthesis; Beven does not argue it explicitly.

6. Mental models of engineers and decision-makers. - Control paradigm. The 19th-century choice to confine the Mississippi to a single channel (p. 361) and Tulla’s “technical masterpiece” on the Rhine (p. 350) show engineering confidence that the river could be mastered. The chapter’s language (“hopelessly striving to eradicate”, p. 362) marks this as the discarded model. - Design-standard thinking. The “design flood” becomes a threshold below which people feel safe. That produces a “false feeling of security” and heavy development behind defences, so exceedance is catastrophic (p. 356). - Literal reading of probabilistic terms. The chapter says “misconceptions and myths” are “deeply rooted” among the public, politicians and decision-makers. It then says “Some people naively believe” the “return period” is a schedule and embankments offer “perfect safety” (p. 362). No evidence is cited for how widespread these beliefs are. Probabilistic language is misread as deterministic reassurance. - Deterministic design from short records (Panel 15.3, p. 364). - Self-correcting reflection. The chapter recognises that one-sided past decisions reflect both “lack of knowledge … at the time” and “the evolution of value judgements over time” (p. 362). This is an important check on hindsight.

7. Lock-in and path dependence. - Policy lock-in. The mid-19th-century Mississippi choice “has remained influential on flood protection policy in the US and elsewhere” (p. 361). It was entrenched by the 1936 federal commitment and the 1944 Act (pp. 361, 365). - Physical and ecological lock-in. Dikes and river regulation enable agriculture and settlement on former floodplains. In Europe, North America and Japan, more than 90% of former floodplains are functionally extinct or converted (p. 349). The Upper Rhine has undergone “long-term and fundamental modification”, and even the best-protected floodplains “have been irreversibly modified” (p. 350). - Settlement lock-in. After floods, people rebuild in place (p. 362). Retreat is “unpalatable” to citizens and decision-makers (p. 364) and “virtually unthinkable in most countries” (p. 359). In Bangladesh it is not an option (pp. 359, 364). - Reversal is slow and costly. Swiss restoration at 15–28 km a year against an 11,000 km need (p. 349). Reservoir decommissioning is only beginning (p. 362). - Lost baseline. Once transformation precedes inventory, the reference state cannot be recovered (Lauterborn, p. 350). Legal tools that depend on reference conditions, such as the Water Framework Directive, are then weakened.

8. Risk displacement and system effects. - Upstream retention loss raises downstream peaks (Rhine, p. 350). - Dike breaches in Germany reduced the load on Słubice in Poland (p. 354). The spatial coupling runs both ways, and defences in one place shift water and risk to another. - Draining wetlands and paving surfaces raise and speed up flood peaks (p. 351). - Flood defences remove natural flood mitigation (p. 349), a self-undermining protective technology. - Runoff responds non-linearly to prior catchment wetness (p. 357), so small differences in antecedent conditions produce large differences in outcome.

9. Time horizons, memory and attention (the hydro-illogical cycle). - The key structural mismatch: “The return period of a destructive flood is usually much greater than the political horizon of decision-makers and the electorate” (pp. 360–361). - Disaster opens a window for spending, forgetting closes it, and projects are “downscaled or suspended” (p. 360). - In the coastal record the cycle runs on about a 50-year memory (p. 363): “There is never an acute reason” to strengthen protection while vulnerability grows slowly (p. 363). - Slowly rising risk (sea level, exposure) never triggers action on its own; only discrete disasters do. - The proposed counter is legal codification (p. 361), which binds future budgets and procedures to present knowledge.

10. Learning. - Learning is fast and real when events are close together: Cologne 1993/1995 and the Odra crests (p. 360). - Box 15.1 shows crisis-driven institutional upgrading: radar, automation, resilient telecoms, international data flows. - The same passage shows the limit: learning is event-driven and decays (p. 360). - The chapter’s call for “post-hoc audits” as a formal link in the chain (p. 352) is an institutional attempt to make learning systematic rather than episodic.

11. Costs, benefits and who bears them. - Protected versus taxpayers: “conflicting interests between those living in floodplains and demanding efficient and very costly protection, and the rest of the nation” (p. 362). - The poor: migrants move into flood-prone informal settlements that had been left empty on purpose (p. 351). - Future generations: infrastructure “limits options for future generations” (p. 362). - Displaced people: reservoirs required “the displacement of many people” (p. 362). - Ecosystems and those who depend on them: fisheries, waterfowl, and drinking-water and nutrient services are lost (pp. 349–350). - Insurance and aid spread losses over people and time (pp. 356, 359). The chapter does not examine whether insurance or aid encourages further floodplain occupation. - Benefits of the hazard itself: fertility, groundwater, ecosystem services. Suppressing the hazard removes them (pp. 348, 352). - Costs of over-precaution: “overly costly (with excessive safety margins)” (p. 355); 1,000-year protection is “far more costly” (p. 356). - Unaffordability as a driver of change: Switzerland (p. 349).

12. Governance and institutions. - Fragmented responsibility across agencies (p. 359). - Legal transitions create ambiguity (p. 360). - An institution designed for one threat (war) was misapplied to another (flood) (p. 360). - Centralised authority to warn delays action (p. 360). - EU framework law with national flexibility (p. 361). - Insurance-linked incentives for community prevention plans in France (p. 363). - Participatory setting of protection levels (p. 362). - Stakeholder co-design of model evaluation in the UK (p. 358). - Contradictions between EU directives (Water Framework, Habitats, Floods) (p. 350).

13. Framing and language. - “Flood control” versus “flood management” and “living with floods” / “making space for water” (pp. 356, 361, 362). The shift in slogan signals a shift in mental model. - “Hopelessly striving to eradicate” (pp. 347, 362) is rhetorically loaded against the control paradigm. - “Technical masterpiece” (p. 350): Tockner acknowledges the engineering achievement before showing its costs, a pattern of admired interventions with hidden side effects. - “Hydro-illogical cycle” (p. 360): memorable naming of an attention pathology. - “Misconceptions and myths”, “naively believe” (p. 362): a deficit-model tone towards the public and decision-makers. - “Unbreachable” (Panel 15.3, p. 364): the reassurance language the chapter warns against, used by a panel. - “Without precedent” and “exceptional” (p. 357): post-hoc framing that can suggest unforeseeability. - “Populist in tone” but skilled in design (p. 353): warnings must be simple on the surface and carefully engineered underneath. - “Failure” and “success” reframed (p. 347): a flood is not a failure, minimised losses are a success. This moves the evaluation criterion from preventing the event to limiting its consequences.

14. Innovation. - Technical: - telegraph forecasting (1854); - radar and satellite (1970s); - data assimilation and adaptive forecasting; - hydrodynamic routing with GIS visualisation; - EFAS; - automatic phone and text warnings; - FLOODsite’s radar-linked flash-flood forecasting (pp. 352–355, 365). - Engineering: superlevees, surge barriers, foreshore nourishment, elevated districts (pp. 356, 363–364). - Institutional: the Floods Directive; Swiss law tying flood control to ecological integrity; buyouts; French insurance-linked prevention plans; the UK Environment Agency’s uncertainty guidelines (pp. 349, 358, 361, 363). - The chapter’s narrative is that innovation in prediction outpaced innovation in warning and response (p. 353). It also says technology can enable encroachment into hazardous areas (p. 351).

15. The precautionary principle and the burden of proof. - Precaution appears in four forms: - (a) design allowances for projected change, adopted without statistically detected trends (p. 355); - (b) evacuation when risk is high, even though the disaster may not come (p. 354); - (c) keeping options open for future generations (p. 362); - (d) the hydro-illogical cycle as the antithesis of precaution (p. 361). - The chapter does not demand proof of trend before adaptation, which implicitly lowers the burden of proof for protective action. - It also warns of over-design costs (p. 355). This is a balanced, cost-aware precaution, not a precautionary absolute.

Transferable insights (technology-neutral)#

1. A warning system is only as strong as its weakest link, and the links usually belong to different institutions. Technical excellence in one part (detection, prediction) cannot make up for a failure in translation, authority, communication or response. Evidence: pp. 347, 352, 359–360; Vaison-la-Romaine (p. 353); Odra (p. 360). Strength: moderate. The mechanism is clear and consistent with the wider hazards literature, but the chapter supports it with two illustrative cases, not a systematic comparison.

2. The ability to predict tends to advance faster than the institutional ability to act on predictions. Accurate but non-specific signals can go unused when the recipient needs a localised, actionable trigger. Evidence: p. 353 (the “often noted” gap; Vaison-la-Romaine). Strength: suggestive. One vivid case plus an unreferenced general observation.

3. Who is legally allowed to raise the alarm matters as much as who knows. Centralising authority to warn delays action. Periods of institutional transition create ambiguous responsibilities just when clarity is needed. Institutions built for one kind of threat may be misfit for another. Evidence: Box 15.1 (p. 360). Strength: suggestive to moderate. A single, well-documented case analysed by an author who studied it directly.

4. Fear of false alarms suppresses warnings, especially when systems are new and uncertainty is high. Precautionary action followed by “no disaster” is not necessarily an error if the risk was real. The public may accept this more readily than officials assume. That last point is my inference from the Dutch 1995 case; the chapter does not compare public and official attitudes. Evidence: p. 354 (Nigg; Netherlands 1995; Słubice 1997); p. 358 (Beven on repeated false alarms). Strength: moderate for the hesitation mechanism (a cited literature source); suggestive for public acceptance (asserted, no data).

5. People at risk may choose not to know when being identified as exposed carries a private financial penalty. Evidence: p. 354 (UK warning sign-up, house prices, insurability); p. 359 (risk maps used to price insurance). Strength: suggestive. One national example, the word “partly”, no quantification. The mechanism is plausible and important.

6. Monitoring systems tend to fail during the extremes they exist to observe, so data on the most consequential events are systematically sparse. Evidence: pp. 359–360 (gauges destroyed, observers evacuated, flood information office flooded, 189,000 telecom links cut). Strength: moderate. Concrete examples; the logic is general.

7. Safety standards built on short historical records and an assumption that the future resembles the past give false precision and can be overtaken by events or by change. Record-breaking events are statistically expected even without change. Evidence: pp. 351–352, 355–356, 359; Carlisle 2005 (p. 356). Strength: strong for the short-record and sampling point (standard statistics, illustrated). Moderate for “stationarity is clearly incorrect”, which the chapter states more categorically than the field’s debate warranted (see Limitations).

8. Treating uncertainty that stems from limited knowledge as if it were ordinary random variability produces systematically overconfident predictions. Decision-makers need guidance on when not to rely on model outputs. Evidence: Panel 15.2 (pp. 357–358); p. 359. Strength: moderate. The general point (unmodelled structural error means overconfidence) is widely accepted. Beven’s specific position is contested within hydrology, as he acknowledges.

9. When a real change is small relative to natural variability, statistical confirmation may lag by decades, so waiting for proof builds in delay. Some institutions adjusted design standards on projections before detection. Evidence: p. 351 (low signal-to-noise; Wilby et al.); p. 355 (Bavaria, UK, Netherlands allowances). Strength: moderate. The detection-lag argument is sound and the examples of action are real. The chapter does not evaluate whether the allowances were well calibrated.

10. Protection can raise exposure. Defences that eliminate frequent small losses encourage development and complacency behind them, so rare large failures become more damaging. Evidence: p. 356 (levee effect); p. 351 (encroachment enabled by wealth and technology); p. 362 (belief in “perfect safety”). Strength: suggestive, on the chapter’s own evidence. The chapter asserts the effect with a clear mechanism and hedges it (“can be higher”), but gives no data or citation. From background, to verify: the wider hazards literature, which the chapter does not cite, supports it more strongly. Do not cite this chapter as the evidence for it.

11. Early technical and policy choices can lock in a trajectory for a century or more, and reversal usually comes through disaster or unaffordability rather than foresight. Evidence: p. 361 (Mississippi choice “remained influential”; 1993 shift); p. 349 (Swiss costs “unaffordable”); p. 350 (Upper Rhine). Strength: moderate. The historical examples are documented. The claim of worldwide influence is asserted, and the chapter does not trace the political economy that sustained the lock-in.

12. Interventions that “solve” a problem locally can export it elsewhere and destroy natural capacities that performed the same function. Evidence: p. 350 (Rhine downstream peaks); p. 349 (flood defences removing natural flood mitigation); p. 351 (drainage and impermeable surfaces). Strength: moderate. Mechanism well described, with one well-documented historical case.

13. Transform first and record later, and you lose the baseline needed to judge the damage or guide restoration. Laws that depend on reference conditions are then hobbled. Evidence: p. 350 (Lauterborn inventory after rectification; Water Framework Directive reference standards). Strength: moderate. One clear case with a stated regulatory consequence.

14. Attention and investment follow disasters, then decay (the hydro-illogical cycle), because hazard return periods exceed political and electoral horizons. Slowly growing risks never provide “an acute reason” to act. Evidence: pp. 360–361; Panel 15.3 (p. 363: Dutch 1,000-year record, ~50-year memory, barriers only after 1953). Strength: moderate. The pattern is widely described and is illustrated consistently: the Odra flood struck “after a long flood-free period” (p. 360), and the Netherlands and UK invested in surge barriers only after 1953 (p. 363). But the supporting material is not referenced: - the Dutch “once every hundred years for the last 1 000 years” record has no citation; - the ~50-year memory is hedged with “Apparently”; - Wilhite’s concept is not referenced; - no spending time series is given.

The claim that it is “a general principle, valid across different political and economic systems” is asserted.

15. Binding preparedness into law is proposed as a counter to attention cycles. Evidence: p. 361 (Floods Directive). Strength: asserted. In 2013 there was no evidence of the Directive’s effect, and its national flexibility may weaken the binding effect.

16. Direct experience produces fast learning when events recur close together, but it is event-driven and fades. Formal post-event audits are a way to make learning systematic. Evidence: p. 360 (Cologne 1993/1995; Odra second crest); p. 352 (post-hoc audits). Strength: moderate. Loss data are cited (Munich Re) and the Odra case is documented.

17. The acceptable level of protection is a value judgement and a negotiated trade-off between safety and willingness to pay, not a technical fact. Technical design standards embed and hide these choices. Evidence: pp. 356, 362. Strength: moderate. Well argued. Contrasting standards show that the choice varies: 100 years (UK), 200 years (Carlisle redesign), and 1,250 and 2,000 years (Netherlands) (p. 356). The Floods Directive also leaves “the level of protection required” to Member States (p. 361).

18. Protection has distributional consequences. Those protected, those who pay, those displaced, the poor who move into hazard zones and future generations are different groups. Retreat is politically unpalatable even where rational. Evidence: pp. 351, 359, 362, 364. Strength: moderate as a description; the chapter names the conflicts but does not analyse them.

19. Hindsight review of large, hard-to-reverse projects often finds one-sided arguments, partly from genuine ignorance at the time and partly from values that have changed since. Designing to avoid “committing future generations to inappropriate options” is a guard. Evidence: p. 362 (dams, reservoir decommissioning, Environment Agency 1998). Strength: asserted, but reasoned and self-aware about hindsight.

20. Suppressing a hazard also suppresses its benefits, which must be counted. Evidence: pp. 348, 349, 352 (fertility, groundwater, ecosystem services worth ~25% of continental services from ~2% of land). Strength: moderate for ecosystem services (cited valuations, though Costanza-type valuations are themselves contested). Weak for the “community solidarity and economic activity” claim.

21. Probabilistic risk language is routinely misread as a timetable or a guarantee, by publics and decision-makers alike. Good warnings must be credible, specific to place, actionable and, ideally, personal. Evidence: p. 362 (return-period myths); pp. 353–354 (Nigg’s criteria; personal delivery). Strength: moderate for the communication criteria (drawn from the literature). The claim about “myths” is asserted without evidence.

22. Bringing the users of predictions into the modelling process early, to agree how performance and uncertainty will be judged, is a practical response when no settled theory of uncertainty exists. Evidence: p. 358 (UK Environment Agency guidelines, Stirling 1999, EEA 2001). Strength: suggestive. Advocated and only just being implemented in 2013.

23. Portfolios of different measures, matched to local conditions and to the full range of event severity, outperform reliance on a single mode. Residual-risk plans are needed for beyond-design events. Evidence: p. 355 (Foresight: an integrated portfolio can cut risk by an order of magnitude, scenario-dependent); pp. 356, 359, 362, 364. Strength: moderate. Rests on scenario modelling and expert consensus rather than observed outcomes.

Limitations, contestation and bias check#

Nature of the chapter. - This is a textbook-style review of a natural hazard, not a documented case of warnings ignored. - Evidence is mostly illustrative cases (Vaison-la-Romaine, Odra, Carlisle, Cologne, Netherlands 1995) and secondary compilations (IPCC, Munich Re, Smith and Ward). - Several strong claims are unreferenced: - the hydro-illogical cycle’s universality, and Wilhite himself; - public acceptance of the 1995 evacuation; - “often noted” forecast–warning gap; - US vehicle fatalities; - forecast improvements having reduced fatalities; - riparian “harmony with nature”; - the “myths” held by publics and politicians; - Panel 15.3’s Dutch record of surprise “about once every hundred years for the last 1 000 years” and the ~50-year collective memory (p. 363). - Table 15.1 lists disasters and milestones, not warnings and responses, so the chapter gives little basis for measuring the delay between warning and action.

Internal inconsistencies. - Summary versus body on trends: a tension, not a contradiction. - The summary asserts as fact that “Intense precipitation has become more frequent and more intense” (p. 347). The body does support that precipitation claim: very wet days contribute more to annual precipitation (Trenberth et al., 2007; p. 354). - What the body does not support is the summary’s implied link to rising floods. The body finds no “conclusive and ubiquitous” climate-change trend in observed high river flows, calls the signal weak “(if any)”, and expects robust trends perhaps decades away (p. 351). - The summary’s “increasingly acute problem” can still be justified by exposure and losses, not by hazard trends. A reader of the summary alone would overestimate the evidence on flood trends. - Panel 15.3 “unbreachable dikes” (p. 364) contradicts the chapter’s core lesson that no defence guarantees safety (pp. 347, 356, 362). Its explanation of Japanese superlevees as tsunami protection differs from the main text’s account of river super-dikes (p. 356). - Box 15.2 dates the Directive to “April 2007”; the references say 23 October 2007. - Design practice: different emphasis. The main text says structural dimensions are “based on probability theory” to withstand a design flood (p. 356), for river defences. Panel 15.3 says coastal levees and barriers are “designed according to deterministic principles” (p. 364). These are different contexts, and both can be true: a single design event is set by statistical frequency analysis, and the structure is then designed deterministically to that one value. The difference is one of wording and emphasis, not a factual contradiction. - The chapter says the Floods Directive “explicitly refers to FLOODsite” (p. 355); this needs checking against the legal text. - Box 15.2’s “10-year return period” for high-probability maps is probably a gloss, not the Directive’s wording (verify). - Multiple citation errors (listed under References above).

Contestation acknowledged or not. - Stationarity. The chapter says stationarity “is clearly incorrect” (p. 355). The only source it cites for this is Milly et al. (2008), which Kundzewicz co-authored. From background, verify: this was contested before 2013 (Lins and Cohn, 2011, “Stationarity: wanted dead or alive?”) and after (Montanari and Koutsoyiannis, 2014, “Stationarity is immortal!”). The chapter does not acknowledge the debate. The substantive disagreement is partly semantic, about whether non-stationary models are justified by the evidence, as opposed to whether the world changes. - Uncertainty methodology. The chapter and Beven acknowledge a live dispute (pp. 357, 359). Beven’s panel argues his side, and the opposing view (formal statistical and Bayesian treatment of errors) appears only as citations. A reader should treat “treating uncertainty as aleatory will produce over-confidence” as a respected but contested position. - Loss trends. The chapter uses unnormalised loss growth (“ten-fold”, p. 348) to show floods are “increasingly serious”. It lists exposure as a driver (p. 351) and Panel 15.3 cites Bouwer et al. (2007). But it does not tell the reader that normalisation studies (Bouwer, Pielke and others) largely attributed loss growth to exposure and wealth, not hazard. Background, to verify: Bouwer (2011) found no trend after normalisation. - Ecosystem service valuations (Costanza et al., 1997: the ~25% figure) are influential but methodologically contested. The panel presents the figure without caveat. - The 69% coastal share of European catastrophe losses (p. 363) looks doubtful as stated and should be checked.

Advocacy versus analysis. - “Living with floods” is a normative position. The chapter argues for it well, but “hopelessly striving to eradicate” (pp. 347, 362) caricatures the alternative somewhat. - The chapter is fair in conceding that structural measures “will be needed” (p. 364) and that retreat is often not feasible (pp. 359, 364). - The praise for the Floods Directive (p. 361) is an unevaluated prediction from an EU-funded author in an EU-agency report. The Directive’s flexibility on protection levels and timetables is noted but not examined as a weakness. - Tockner’s panel is advocacy for floodplain conservation research, calling for a competence network and a research agenda. This is legitimate, but it also aligns with his own research field. - Vellinga and Aerts promote concepts they helped originate (climate proofing).

What is missing (fairly, given scope). - Political economy. There is almost no analysis of the interests that sustained the structural paradigm: engineering agencies, construction, agricultural and development interests on reclaimed floodplains, local politics of protection spending. The chapter attributes past errors to knowledge gaps and changing values (p. 362), not to interests. This contrasts with most Late Lessons chapters. It may reflect the chapter’s genre, or an under-analysis. - Insurance and aid as moral hazard. Subsidised insurance or reliable disaster aid can encourage floodplain occupation. The chapter treats both only as risk-spreading goods (pp. 356, 359). From background: this is a large literature, for example on the US National Flood Insurance Program. - Social vulnerability. Beyond brief mentions of poor migrants and Bangladesh, the chapter does not analyse who dies or loses most within societies (age, income, disability, tenure). - Earlier warnings about the structural paradigm. Gilbert White’s 1942 work and the levee-effect literature are not cited, so the chapter understates how early the key insight was available. - Evaluation evidence. No data are given on the actual performance of warning systems (penetration rates, lead-time gains, lives saved), despite the chapter naming penetration as a quality indicator (p. 354).

Hindsight bias. - The chapter is less exposed than most because it is not a blame narrative. It explicitly allows that past decisions reflected “a lack of knowledge and understanding at the time” and “the evolution of value judgements over time” (p. 362). - The “technical masterpiece” framing of Tulla (p. 350) also credits the original achievement. Judging 19th-century river engineering by 21st-century ecological values is still a form of hindsight, and the chapter says so implicitly.

Case selection. - The warning-failure examples (Vaison-la-Romaine, Odra) are well known. - The success examples (Netherlands 1995, Słubice, Odra’s second wave, Cologne) are also selected. - There is no attempt at a representative sample. European focus dominates, with brief US, China, Japan and Bangladesh references.

How the pro-precaution framing may shape conclusions. - Weakly. The chapter is organised around risk management, not the precautionary principle. The main normative commitments are “living with floods”, portfolio management and participation, which are not uniquely precautionary. - If anything, the report’s framing may have pulled the chapter towards the “early warning” metaphor even though its content is mostly about warning systems rather than warnings ignored. - The chapter’s own balance points are notable: it names the costs of over-design (p. 355) and says “Any benefits that could be lost also need to be taken into account” (p. 352).

Dissent within panels. - None of the panels dissents from the chapter. - The clearest tension is Panel 15.3’s “unbreachable” language (p. 364). - Beven’s panel supports and extends the chapter, and even cites Kundzewicz’s point about Carlisle (p. 358). - There is a milder difference of emphasis on forecasting: - the main text says forecasts “have advanced markedly” (p. 353), and that better lead time and accuracy have “already” reduced fatalities in many countries (p. 364; unreferenced); - Beven stresses that numerical weather prediction “does not (yet) give generally reliable rainfall forecasts”, even though these are essential for flash-flood warning (p. 358).

The two accounts are compatible (large basins versus flash floods), but a reader should not take the main text’s optimism as covering flash floods. - No industry, insurer, engineering-agency or regulator voice appears.

Notable quotes#

1. “A single weak point in a system that otherwise contains excellent components may render the overall system performance unsatisfactory.” (p. 347)
2. “We have first raised the dust and then complain we cannot see” (George Berkeley, epigraph, p. 348)
3. “Paradoxically, a good deal of this damage is done by hydrological engineering for flood defences” (Tockner, Panel 15.1, p. 349)
4. “It is often noted that forecasts have advanced markedly, while progress in warnings has lagged behind” (p. 353)
5. “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)
6. “Systems continue to be designed and operated assuming stationarity (the past is the key to the future).” (p. 355)
7. “losses in a levee-protected landscape can be higher than in the absence of a levee due to the false feeling of security that levees can generate” (p. 356)
8. “Treating the sources of uncertainty as if they were aleatory in non-ideal cases will produce over-confidence in the model predictions” (Beven, Panel 15.2, p. 357)
9. “The return period of a destructive flood is usually much greater than the political horizon of decision-makers and the electorate” (pp. 360–361)
10. “There is never an acute reason for strengthening the coastal protection system or for elevating settlements until the area and its population are hit by a major flood” (Vellinga and Aerts, Panel 15.3, p. 363)

Reserve quotes: - “Polish civil defence was geared to act in the event of a war, rather than to deal with a peacetime emergency.” (Box 15.1, p. 360) - “there is considerable uncertainty regarding uncertainty estimation” (p. 359) - “under what circumstances not to rely on model predictions when making a decision” (Beven, p. 358) - “major decisions, such as regarding the construction of a large dam, are based on one-sided arguments with important aspects neglected” (p. 362) - “avoid as far as possible committing future generations to inappropriate options for defence” (Environment Agency, 1998, quoted p. 362) - “Of course, the hydro-illogical cycle is also at odds with the precautionary principle.” (p. 361) - “Tulla’s regulation of the Upper Rhine was a technical masterpiece” (Tockner, p. 350) - “Apparently after some 50 years the flood disaster tends to disappear from the collective memory” (Vellinga and Aerts, p. 363)

Open questions#

1. Did the Floods Directive “considerably reduce flood risk”? (p. 361) What do the Commission’s implementation reports and the European Court of Auditors’ review say? I believe the Court of Auditors published a special report on the Directive around 2018; verify. Did national flexibility on protection levels and timetables dilute it?
2. Did the warning chain fail again after 2013 in the way the chapter describes? My background understanding is that in the July 2021 floods in Germany and Belgium, EFAS alerts were issued days ahead but local warnings and responses failed, with very high loss of life. If confirmed from primary sources (EFAS, German parliamentary inquiries), this is a strong test of insights 1–3. Retrieve primary documents rather than press accounts.
3. Carlisle again. The post-2005 defences were designed for a 200-year return period (p. 356). Background, to verify: Carlisle flooded again in Storm Desmond (December 2015), with defences overtopped. If so, that is a direct test of the short-record and exceedance lessons (insight 7).
4. Detection of flood trends. Wilby et al. (p. 351) expected robust trends to take decades. Background, to verify: Blöschl et al. (2017, Science; 2019, Nature) reported detectable changes in timing and regional magnitude in European floods. How does IPCC AR6 (2021) assess observed and projected river-flood changes? Did detection come sooner than the chapter expected?
5. Projections. Are the >40% / >30% European projections (p. 355) upheld, revised or reversed in later ensembles (for example JRC work around 2015–2020)?
6. Design allowances. Were the UK allowances (+10% / +20%, p. 355; check the “after 2085” date), the Bavarian climate factor and the Dutch Rhine design discharge (16,000 m³/s by 2015) implemented and later revised upward or downward? This bears on whether acting before detection was well calibrated.
7. Swiss restoration. Was the 11,000 km target (p. 349) adopted in law? Background: I recall a 2011 revision of the Swiss Waters Protection Act with a lower restoration target over about 80 years. What has the actual restoration rate been?
8. The “chosen ignorance” mechanism (p. 354). Is there quantitative evidence that property-value or insurance fears suppress uptake of flood warnings? How did later UK changes (automatic enrolment of landlines; the Flood Re scheme from 2016, per my background) alter the incentive?
9. Hydro-illogical cycle. Is there quantitative evidence (budget time series after major floods) for the spending-and-forgetting pattern the chapter asserts (p. 360)? Has legal codification actually damped it?
10. Levee effect. What is the empirical evidence, then and now, on development behind defences and on losses in protected versus unprotected areas? The chapter asserts it without data (p. 356).
11. Uncertainty debate. Did the aleatory–epistemic dispute in hydrology (Panel 15.2) converge after 2013? Did the Environment Agency’s “guidelines for good practice” with early user involvement (p. 358) become standard practice?
12. Numerical weather prediction. How far has rainfall forecasting for flash floods improved since Beven wrote that NWP “does not (yet) give generally reliable rainfall forecasts” (p. 358)?
13. Political economy gap. Can primary sources (US Congressional history of the 1861 and 1936 decisions; Army Corps history) show what interests sustained the structural paradigm, which the chapter leaves unanalysed?
14. Figures to verify from primary sources: - the 69% coastal share of catastrophe losses (CEA); - Xynthia casualties (51); - the Japanese superlevee purpose; - whether the Floods Directive text mentions FLOODsite; - the Directive’s return-period wording; - Vaison-la-Romaine death toll (not given in chapter).

Audit log#

Audited 2026-09-25 against the full text extract (PDF pp. 349–370), with Table 15.1 and the single image checked against the PDF. Nearly all quotations were checked verbatim and are correct. Page references, numbers, names and panel attributions were accurate apart from the items below.

• Authors: marked “one of the least precaution-framed chapters” as an unchecked comparison with other chapters.
• Panel 15.1 authors: corrected Tockner’s self-citations from “about four times” to five works (Tockner et al. 2002/2008/2009, Nakamura et al. 2006, Sommerwerk et al. 2010).
• Panel 15.2 authors: noted that the main text (§15.3.3, p. 359) also cites the Beven/Hall/Sivakumar dispute, not only the panel.
• §15.1: marked the “attention dynamics” point as my inference.
• §15.1: added the omitted Barredo point (“high-impact floods … occurring more frequently”) and the “commonplace in developing countries” line.
• §15.1 benefits: marked the recovery-spending criticism as my comment.
• Panel 15.1: added the omitted “remarkable intention (probably unique in Europe)” framing of the Swiss policy.
• Panel 15.1: added the omitted “legacy biodiversity” point (p. 350).
• §15.2.2: corrected a page reference (pp. 353–354).
• §15.2.4: made clear that “chosen ignorance” is the notes’ label. The chapter describes reluctance to acknowledge risk, “partly”, with no data.
• Panel 15.2: added the omitted key claim that the epistemic component “will often” dominate (p. 358).
• Panel 15.2 assessment: marked the identification of Hall, Todini, Mantovan and Montanari as opponents favouring formal statistics as inference from background. The panel cites them only on the interpretability of probabilistic estimates.
• §15.4: restored the hedge “it may become evident that” in the quote on unclear agency roles.
• Box 15.1: renamed “Lead time wasted” to “Lead time available but hard to exploit”. The source says Brandenburg “in principle” had ten days and gives no account of how the time was used.
• Panel 15.3: softened “contradicts” to “sits uneasily with” on “unbreachable”, and noted a possible narrow meaning (breach versus overtopping).
• Panel 15.3: made clear that the main text does not call the Japanese super-dikes river defences in so many words.
• Table 15.1 observations: removed the unchecked “unlike other chapters’ tables” claim and refined which entries relate to warnings.
• Timeline A: noted that the chapter does not itself link the mid-19th-century Congress choice (p. 361) to the 1861 Army Corps report (p. 365).
• Timeline B: replaced “10 days’ lead time … poorly used” with the source’s wording.
• Timeline D: corrected “Lag: negative”. The lag between warning (projections) and action was short. What happened before confirmation was action ahead of statistical detection.
• Harms list: restored “or converted” to the >90% floodplain figure.
• Authors’ lesson 2: removed the unsupported gloss that the weak link is “often a human or institutional one, not a technical one”, and noted the Odra technical failures.
• Mechanism 2: limited the public-acceptance claim to the Netherlands in 1995 (nothing is said about Słubice), and marked the “public more tolerant than officials fear” point as my inference.
• Mechanism 6: corrected the overgeneralisation to the source’s “Some people naively believe”.
• Mechanism 7: restored the >90% figure’s scope (“or converted”; Europe, North America, Japan) and used the panel’s own words for the Upper Rhine.
• Insight 4: marked the public-versus-official acceptance comparison as inference.
• Insight 10 (levee effect): downgraded from “moderate” to “suggestive” on the chapter’s own evidence (no data or citation). The stronger external literature is flagged as background.
• Insight 14: kept “moderate” but added that the Dutch 1,000-year record, the ~50-year memory (“Apparently”) and Wilhite are unreferenced.
• Insight 17: corrected the page references for the contrasting standards (all on p. 356; the Directive’s flexibility is on p. 361).
• Limitations: added the Panel 15.3 Dutch record to the list of unreferenced claims.
• Limitations: recast “summary versus body” as a tension, not an inconsistency. The body supports the precipitation claim (p. 354) but not the implied flood trend (p. 351).
• Limitations: recast “probability theory” versus “deterministic principles” as a difference of emphasis in different contexts, not a contradiction.
• Limitations: noted that the only source cited for “stationarity is clearly incorrect” is Milly et al. (2008), which the chapter author co-authored.
• Limitations: softened “serves his field’s interests” (Tockner) to “aligns with his own research field”.
• Dissent within panels: added the milder difference of emphasis between the main text’s forecasting optimism (pp. 353, 364) and Beven’s caution on rainfall forecasts (p. 358).
• References: added further visible slips (“Robsinson”, “Neill”, “van der Belt”), plus flagged, unverified background corrections for Brázdil and Faulkner.
• Digest: “funded by” changed to “carried out within” (WATCH footnote).
• Digest: restored the hedges on failure and success (“need not”, “may”) and corrected the pages to 347 and 359.
• Digest: levee effect now “can generate”, with no data noted.
• Digest: Mississippi now “has remained influential”, not “until 1993”.
• Digest: the no-trend claim is now “conclusive and ubiquitous” and global.
• Digest: insight 4 split rating; insight 10 downgraded to suggestive; insight 14 caveat added; caveats recast as tensions.
• Checked and left unchanged:
• no mentions of contemporary technologies or companies outside the source;
• background claims (affiliations, Floods Directive wording, Defra dates, stationarity debate, the post-2013 open questions) were already flagged “verify” and are left as flagged.