EM 1110-2-1100 (Part II)
30 Apr 02
Parameters in Extremal Distribution Functions
Figure II-8-3. Probability distribution functions for long-term statistics
(b) The preferred approach to data selection is to take the maximum value from each event to create a
partial duration series of extreme values. Typically, the events are storms, ranging from small, weak events
to the most severe storms of record. Small events can be difficult to identify. Further, they are of little
interest if an adequate number of bigger storms is available in the record. Often the partial duration series
will be censored to exclude data values less than some threshold value (Figure II-8-4). Thus the extremal
analysis can focus on a smaller series representing truly significant events. The threshold is often chosen so
that the number of data values in the series is greater than the number of years of record (generally 1-3 times
the number of years of record).
(c) Another accepted approach is to take the maximum value from each year of record to form an annual
maximum series. A record length on the order of 20 years or more (yielding at least 20 data values) is needed
for this approach. In coastal engineering, record lengths are often shorter than this requirement. Another
drawback to the annual maximum series arises from the possibility of multiple severe storms occurring in the
same year, as in El Nino/Southern Oscillation events, which sporadically distort winter storm seasons along
the U.S. Pacific Coast (see Part II-2). Only the maximum event in each year is considered. Other storms,
which may be bigger than the maximum event in many other years, will be ignored.
(2) Estimating parameters in extremal distribution functions.
(a) Generally an extreme data value series (either partial duration series or annual maximum series) is
treated as a sample from a process that follows one of the extremal distribution functions presented in
Hydrodynamic Analysis and Design Conditions