Sir

The tsunami death toll from the earthquake of 26 December 2004 is a tragedy of such dimensions that it creates a tipping point in hazard perception and risk management, with existing procedures being thrown into a new and critical light (“Inadequate warning system left Asia at the mercy of tsunami” Nature 433, 3–6; 2005 10.1038/433003a).

In particular, an urgent review is needed of the process for issuing tsunami warnings. Developed by Earth scientists over four decades, it has insufficient input from risk specialists who understand the difficulties of emergency decision-making under uncertainty.

Tsunamis occur most frequently in the Pacific Ocean, so existing monitoring and warning systems have been developed for those conditions. Because earthquakes that could lead to a tsunami happen about once a year on average, concern over the economic cost of false alarms means that detecting a very large earthquake is not enough — additional data from pressure sensors or tide gauges are needed to confirm that a tsunami has indeed been generated.

Outside the Pacific Ocean, it would be impractical to insist on certainty before issuing a warning. Without adequate monitoring of seismic activity and sea conditions, such an approach may be ineffective. Furthermore, in developing nations with high population density, the cost–benefit balance between human safety and financial loss is tilted more heavily towards avoidance of mass casualties than in affluent Hawaii or Oregon.

An alternative approach is to establish evidence-based criteria for raising different levels of tsunami alert. This would reduce the chance of complete alert failure, at a cost of some additional false alerts.

An analogy might be made with gale warnings, whereby the possibility of a severe storm is sufficient for seafarers to be alerted some hours in advance, but not forcibly grounded.

A risk-informed warning system, based on global online tsunami databases (see http://www.ngdc.noaa.gov/seg/hazard/tsu.shtml and http://tsun.sscc.ru/tsulab/On_line_Cat.htm), and implemented via the Bayesian Belief Network formalism, could help decision-makers incorporate the intrinsic uncertainties in tsunami forecasting.

Belief Networks are probabilistic inference graphs providing a logical basis for reasoning under uncertainty, and aid decision-making by accounting jointly for uncertainties associated with accumulated experience (including expert judgement) and with processed statistical data. Such a scheme would provide a rational basis for tsunami alerts, and allow for a new low-alert level, akin to a gale warning, that would advise against sea and beach excursions during times of heightened risk after a major earthquake.

For coastal authorities, a low-level false alarm every few years would be a small price to pay for avoiding a catastrophe. Indeed, such warnings may prove cost-effective, if the return of tourism requires reassurance. And as with fire drills, the occasional alert would serve as a useful reminder to the local community of a latent threat.