Extreme events over time - Extreme Events: A Physical Reconstruction and Risk Assessment - page 264

forrandomness, many of the mechanisms that influence tropical cyclone occur-
rence and abundance do display serial correlation over the long term. These
mechanisms include the SOI, PDO and SST (Figs. 10.2--10.4). If these boundary
mechanisms are not random then it is also unlikely that tropical cyclones will
occur in a truly random fashion over time periods longer than the short histor-
ical record. Given that the preservation potential of the tropical cyclone palaeo-
record in this region results in only the most extreme events being registered
over time, it is logical that these events most likely occurred during phases
of increased occurrence probability represented by the crests of the probability
curve (Fig. 10.5). At these times these extreme events would have been equivalent
to the1in166yeareventor p = 0.006 (annual exceedence probability, AEP). If an
event of this magnitude were to occur during a period represented by a trough
in the probability curve then it could be regarded as a 1 in 500 year event or
p = 0.002. In this sense, these latter events are not outliers but events whose
probability varies with time.
This same approach can be adopted for other kinds of hazards such as
river floods. Phases of enhanced flooding in the Burdekin River, North Queens-
land, based upon the long-term coral luminescence record (see Chapter 3,
Fig. 3.11), have a period of approximately 150 years. In this case large-magnitude
floods occurring during the enhanced phase of flooding, or cycle peak, would
have a probability of occurrence (AEP) of 0.0225 and during troughs an AEP
of 0.0075. Statistical analysis of this flood record using the KPSS test for sta-
tionarity (Kwiatkowski et al ., 1992)shows that it displays non-stationarity. Two
approaches could be taken in determining the AEP of given magnitude floods
in this situation. The first is as mentioned above where a mean AEP (0.015) of
large-magnitude floods could be assigned to cover the entire record. Or alterna-
tively, separate probability cycles could be developed for the enhanced phases
and lesser magnitude phases each with their own mean AEP. The Burdekin River
coral luminescence flood record is only 350 years in length and ideally a longer
record might be better suited to the latter approach for estimating probability.
These approaches to assessing temporally variable probabilities for hazard risk
assessment can only be regarded as preliminary at this stage. To be more realistic
they will requireuncertainty margins and a reasonable idea of the position of
thepresent day in the probability cycle. The causes of the cycles are also unlikely
to be as straightforward as those producing clusters of earthquake events. Cycles
in the probability of occurrence of atmospherically generated hazards are likely
to be a function of the interaction of many cycles, both within one variable
and between different variables. At times the interaction of these cycles will
tend to enhance the probability due to some amplification of the necessary
conditions required for development of the hazard in question. For example,