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

Late lessons from early warnings II (EEA 2013), pp. 347–368 (PDF 349–370).

Author. Zbigniew W. Kundzewicz, writing alone. He is a hydrologist and IPCC author, though the chapter states no affiliation. According to a footnote, the work was “carried out within” the EU FP6 WATCH project. Five reviewers are named, including Beven, Hall and Sivakumar.

Panels. Three, all of which complement the chapter; none dissents: - Tockner on floodplain ecology (pp. 349–350); - Beven on predictive uncertainty (pp. 357–358). Beven also reviewed the chapter; - Vellinga and Aerts on coastal flooding (pp. 363–364).

What kind of chapter this is. By its own account it covers “flooding as a phenomenon”, not a single case (p. 347). “Early warning” means two things here: operational flood warnings, and long-term warnings that flood frequency is changing (p. 348). The precautionary principle is mentioned once (p. 361).

Core argument. - Floods cannot be prevented and no defence guarantees safety. A flood therefore “need not be considered a ‘failure’”, and minimising losses “may constitute a ‘success’” (pp. 347, 359). - Humans have “increased the load and decreased the resistance” of the system (p. 348). - The chapter argues for moving from flood control to “living with floods”: a site-specific mix of defences, zoning, forecasting, warning, insurance and participation, plus plans for floods that exceed design standards (pp. 356, 362, 364).

Key evidence. - Warning chains break at weak links. - Vaison-la-Romaine, 1992: accurate alerts came 12–24 hours ahead, but no local warning was issued because officials did not know which catchment would flood (p. 353). - Odra, 1997 (Box 15.1, p. 360): the flood hit during a legal transition. Local authorities had no legal power to declare alerts, and civil defence was “geared to act in the event of a war”. 189,000 telecom links were cut and the flood information office itself flooded. The second flood wave was handled far better. - Short records and stationarity. Design floods are estimated from records that are too short, and record-breaking floods are expected anyway (pp. 351–352). Codes still assume “the past is the key to the future” (p. 355). Carlisle’s 2005 flood exceeded the planned 100-year defences before they were built (p. 356). - The levee effect. Defences “can generate” a “false feeling of security”, so losses when they fail “can be higher than in the absence of a levee” (p. 356). The chapter asserts this without data or citation. - Lock-in. - The 19th-century US choice to confine the Mississippi with levees “has remained influential on flood protection policy in the US and elsewhere”. The 1993 flood prompted a further “paradigm shift” towards buyouts and relocation, but “this response is not universal” (pp. 361–362). - Tulla’s Rhine “technical masterpiece” raised flood peaks downstream and erased a biodiversity baseline that was never recorded (p. 350). - Switzerland found its CHF 45 bn of engineering spending “unaffordable” and turned to ecological restoration (p. 349). - Uncertainty. There is “considerable uncertainty regarding uncertainty estimation” (p. 359). Beven argues that treating gaps in knowledge as random variability “will produce over-confidence”. Decision-makers need to know when “not to rely on model predictions” (pp. 357–358). - Acting before detection. No “conclusive and ubiquitous” climate-change trend has been found in global data on observed high river flows (p. 351). Yet Bavaria, the UK and the Netherlands wrote climate allowances into their design codes on the strength of projections (p. 355).

The authors’ lessons. - From the evidence: weakest-link failure; institutions shape vulnerability; experience cuts later losses; memories fade in a “hydro-illogical cycle” because flood return periods outlast political horizons (pp. 359–361). - Advocacy: “living with floods”; protection set at an “agreed safety level”; participation; writing preparedness into law, with the Floods Directive praised as “probably the most advanced” (p. 361); keeping options open for future generations (p. 362).

Main mechanisms. - Prediction improves faster than the capacity to warn and respond. - Officials fear false alarms. - People at risk avoid warnings to protect house prices and insurance (p. 354). - Monitoring fails in the extremes it is meant to measure. - Risk is displaced downstream. - Protection encourages people to build in exposed places (asserted, p. 356). - Attention rises after disasters and then decays. - Costs fall unevenly: on floodplain residents versus the nation, on poor migrants and on future generations.

Transferable insights (the full list of 23 is in the notes). 1. A warning system is only as strong as its weakest link, and the links belong to different institutions (pp. 347, 353, 359–360). Moderate. 2. The ability to predict outruns the ability to act on predictions (p. 353). Suggestive. 3. Who has legal authority to raise the alarm matters as much as who knows (p. 360). Suggestive–moderate. 4. Fear of false alarms suppresses warnings. Precaution followed by “no disaster” is not necessarily an error if the risk was real (p. 354). Moderate for the hesitation mechanism; suggestive for public acceptance, which is asserted only for the Netherlands in 1995. 5. People at risk may choose not to know when knowing carries a private penalty (p. 354). Suggestive. 6. Monitoring fails in the extremes it exists to observe (pp. 359–360). Moderate. 7. Standards based on short records and on the past give false precision (pp. 351–352, 355–356). Strong on the statistics; moderate on the claim that “stationarity is dead”. 8. Treating gaps in knowledge as random variability breeds overconfidence (pp. 357–358). Moderate. Contested within hydrology. 9. When the signal is weak against the noise, waiting for proof builds in decades of delay (pp. 351, 355). Moderate. 10. Protection can raise exposure (p. 356). Suggestive on the chapter’s own evidence: the mechanism is asserted, with no data or citation. The wider literature is said to be stronger, but that is not cited here (verify). 11. Early choices lock in trajectories for a century, and reversal comes through disaster or cost (pp. 349, 361). Moderate. 12. Local fixes export risk elsewhere and destroy natural capacities that did the same job (pp. 349–350). Moderate. 13. Transform first and record later, and the baseline is lost (p. 350). Moderate. 14. Attention follows disasters and then decays; writing preparedness into law is the proposed counter (pp. 360–361, 363). Moderate for the cycle, although the Dutch 1,000-year record and the ~50-year memory are unreferenced. Asserted for the remedy. 15. The acceptable level of protection is a value choice, not a technical fact (pp. 356, 362). Moderate.

Main caveats. - The chapter is a review built on illustrative cases, and many of its claims are unreferenced. Table 15.1 lists disasters, not warnings, so the delay between warning and action cannot be measured. - There are internal tensions. - The summary’s precipitation claim is supported in the body (p. 354), but its implied link to rising floods outruns the body’s finding of no “conclusive and ubiquitous” trend in observed river flows (pp. 347, 351). - Panel 15.3 calls superlevees “unbreachable” (p. 364), which sits uneasily with the chapter’s core lesson. - The main text is optimistic about forecasting (pp. 353, 364), while Beven cautions that rainfall forecasts are not yet generally reliable (p. 358). - There are also citation errors. - Losses are not normalised for growth in exposure and wealth. - The debate over whether “stationarity is dead” is not acknowledged. - Political economy and moral hazard in insurance and aid are absent. - Earlier warnings are omitted, such as Gilbert White’s 1942 work (my addition; verify). - The praise for the Floods Directive was an untested prediction in 2013. - In the chapter’s favour, it is fair to engineering (p. 364), names the costs of over-design (p. 355), and guards against hindsight by allowing for what could not have been known at the time (p. 362).