12.12.2012 Views

chapter - Atmospheric and Oceanic Science

chapter - Atmospheric and Oceanic Science

chapter - Atmospheric and Oceanic Science

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Statistical analysis of extreme events in a non-stationary context<br />

into the future, estimating how often extreme events will occur in the future will<br />

remain a very difficult problem.<br />

Where hydrologic regimes are changing, a different approach to quantifying<br />

the probability of occurrence of extreme events is required, which avoids reference<br />

to the long-term frequency of occurrence. Recall that, in the first paragraph of this<br />

section, an event with return period T years was defined as the event which may<br />

occur in any one year with probability 1/T; under changing hydrological regime,<br />

this probability is no longer constant, <strong>and</strong> to generalize the concept of return period<br />

it is necessary to describe how the probability is changing. Clarke (2003) has<br />

suggested the following series of steps by which statements about the future frequency<br />

of occurrence of hydrological events become possible, where time-trends<br />

are found to exist in records. Step 1: identify the extreme event which, if it<br />

occurred, would influence the choice of decision. This might be, for example, a particularly<br />

intense rainfall over a duration that would lead to severe flooding. Call the<br />

magnitude of this event xcrit; several values of xcrit may be explored. Step 2:<br />

Assuming that the trend in regime exhibited in the available hydrologic records<br />

continues into the future at the rate hitherto observed, determine the probability distribution<br />

of the time to first occurrence of xcrit, the event selected at Step 1. Step 3:<br />

determine the probability of the extreme event xcrit occurring in the next t years,<br />

where t extends up to an acceptable (but limited: see discussion below) planning<br />

horizon. Clarke (2003) gave expressions for two cases: first, where trends have<br />

been detected in annual maximum rainfall intensity represented by a Gumbel distribution<br />

or, more generally, by a GEV distribution, with time-variant means; <strong>and</strong><br />

second, where trends have been detected in rainfall records consisting of pairs of<br />

values (t, xt), where xt is the magnitude of rainfall intensity exceeding some 'threshold'<br />

value xthresh, <strong>and</strong> t is the time at which xt > xcrit > xthresh occurs. Thus, Clarke's suggestion,<br />

in the presence of trend, is to replace the concept of 'event with return period<br />

T years', by the concept 'the probability that a critical event, suitably defined,<br />

will occur at least once during the forthcoming limited period of S years, assuming<br />

that the observed trend in the record continues over this limited period at the same<br />

rate as that recently observed'.<br />

In conclusion, changes in hydrological regime - whether as a consequence of<br />

climate change or of change in l<strong>and</strong>-use - require the concept of return period to be<br />

redefined. If further evidence in support of climate change accumulates, this will<br />

have important consequences for the many kinds of civil engineering project which<br />

have long been designed according to principles based on the return level of events<br />

with T-year return period, estimated from data sequences that are realizations of<br />

stationary processes.<br />

214

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!