Letter: Letters : Whether prediction

From David Dunkerley, Monash University

Victoria, Australia

The article “Far out forecasting” by Robert Matthews (Features, 12 October, p
37) seems to accept in a rather uncritical way the reality of statistical
forecasting of events that are rare, such as damaging floods.

For example, he reports that Dutch workers have applied extreme value theory
to estimate the height of sea wall needed to withstand “one-in-10 000-year”
floods.

Is this credible? The past 10 000 years spans the Holocene period of earth
history. During this time, there were major changes in the amount and
distribution of monsoon rains, related to slow changes in the planetary orbit.
As a result, lakes developed in what is now the hyperarid Sahara.

In the next 10 000 years, such orbital changes will, of course, continue. On
top of that are the potentially larger environmental changes wrought by people,
such as the enhanced greenhouse effect. One of the possible effects of this is
considerably increased storm intensity in some areas. The distribution of
changes over the Earth’s surface seems likely to be complex; it is certainly
ill-understood at present.

All of this seems to suggest that the distributions of extreme values of
climatic variables are unlikely to stand still in coming decades, let alone
millenniums. The application of statistical methods that do not account for
future environmental change runs the risk of being seriously erroneous.