WATER & SOIL - These are not the droids you are looking for.

alysis of the country's rainfall intensity data. As this revision
used data for twice the number of stations used by
Robertson, more accurate estimates of 1024 for individual
catchments should be possible using Tomlinson's results.
Note that an areal reduction factor should not be applied
to an 1224 estimate. Further, the rainfall stations used in
the estimation of 1224 should not necessarily be the nearest
but should be the ones that record weather patterns that are
of most relevance to the catchment.
In the Southern Alps, Tomlinson's (1980) maps of rainfall
intensity assume that the intensity increases with altitude.
This increase was not considered in the estimates of
1224 used to obtain the regional mean annual flood equations.
Thus the use of the estimates of rainfall intensity in
the equations may lead to overestimates of Q for catchments
running into the Southern Alps. Therefore, when
calculating 1224, poinr estimates of intensity should be
averaged for the raingauges which receive rainfall typical of
that for the middle and lower parts of the catchment.
MARAIN The mean annual rainfall for a catchment may
be estimated directly from rainfall records for stations
within, or near to, the catchment. Where only short rainfall
records exist, or where there are none, estimates of
MARAIN should be obtained from the l:500 000 isohyetal
maps of l94l-1970 annual rainfall normals published by
the NZ Meteorological Service.
When there is at least one year of flood record available,
we suggest that both the available record and the regional
equation be used to obtain separate estimates of Q. These
estimates can then be combined to form a weighted average
estimate of Q , with the weighting of the Q value estimated
from the record being based on the length ofthe record relative
to Nr. Hence, for example, if N : 3 and N, : 4' the
weighting factors for the estimates taken from the record
and regional equation should be
3/'7 lì.e.
(ii) N > Nu
NN
N+Nu
N*N,
When the flood record length exceeds N' Q may be estimated
as the arithmetic mean of the annual series. It could
also be estimated f¡om a partial duration series (NERC
1975, pp. 185-213) when N is less than l0 years. In the case
where an outlier or historical flood peak Q."* occurs in an
annual series such that Q*"*/Q.e¿ ) 3, it is suggested that
Q be estimated graphically from a probability plot of the
annual series as the flood peak with return period T : 2.33
years.
more flexible frequency curves, it was found in the evaluation
tests (Appendix A) that a two-parameter distribution
gives a good approximation to a three-parameter one up to
T : 100 and that it can give more sensible results, even
though it may not produce quite as good a fit to the annual
senes.
(iii) N > 20
V/ith a flood record of 20 or more years in length, both
two- and three-parameter distributions should be fitted to
tbe annual series for the estimation of Qr ' A visual inspection
of the goodness-of-fit of the distributions to the series
should then be made and a distribution chosen for estimating
Qr . When it is difficult to decide between two or more
fitted distributions, the Q.¡ estimate should be determined
by averaging the estimates given by these distributions.
In applying frequency analysis methods to estimate a design
flood peak Qt, care should be taken to ensure that
they are not grossly extrapolated. For example, if N < 20
and the two-parameter EVI distribution is fitted to the annual
series, its extrapolation past T : 100 years for catchments
in some of the eastern regions, e.g' South Canterbury,
may lead to an under-estimation of Q1.
It is recommended that the fitted distribution should not
be extrapolated beyond a return period T : 5N' This limit
is less stringent than those often recommended in other
tests (e.g. a limit of T : 2N is suggested (ICE 1975) for the
NERC (1975) study) and extrapolation beyond it is unwise
on the basis of present evidence. The safest course of action
when T exceeds the extrapolation limit is to use the regional
curve, or the appropriate generalised curve when T exceeds
the upper limit of the regional curve.
Finally, when the record length N is sufficiently great to
warrant the performing of a frequency analysis, the resulting
Q1 estimates should be compared with that using the regional
(or generalised) curve, with Q being calculated from
the annual series. A, decision must subsequently be made as
to which estimate to accept for design. This may involve
taking a weighted average of the estimate from the frequency
analyses of the site data and the estimate obtained
using the regional curve, and this procedure is suggested
when T > 2N. In making this decision on the final Q1
value it should be lemembered that variations are inherent
in all flood records, especially small ones, so that the trend
in the probability plot should not be over-emphasised, even
though it may depart significantly from the regional one.
Instead, the emphasis should be placed on the regional
curve, for it represents the trend of all the flood peak data
for the region and its construction involved the averaging
out of the variations in the individual flood records.
5.3.3 Estimation of 01
(i) N