PROGRAM AND PLANS - Branch Policy on Selected Subjects

May 1, 1973


Subject:  PROGRAM AND PLANS - Branch Policy on Selected Subjects

Our recommendations on certain technical aspects of surface-water
hydrology change with time in response to the better understanding
provided by additional data and further analytical developments.
Major changes are generally incorporated in TWRI's, but some
changes evolve gradually and are not formally disseminated.  This
memorandum: describes Branch policy on several subjects.

1.  Extension of frequency curves

To discourage indiscriminate extensions, Surface Water Branch
Technical Memorandums 69.11 and 70.05 state that we should not
publish frequency characteristics for recurrence intervals greater
than twice the period of record (2N).  At recent Branch meetings,
it was agreed that the 2N rule should be modified.  The following
paragraph describes a revised policy on limits of frequency curve
definition and supersedes previous directives on this subject.

Listed below are the recommended minimum years of record needed to
define events of selected recurrence interval.

Recurrence interval -     10    25     50     100
Minimum years of record - 10    15     20      25

Magnitudes of floods for these recurrence intervals will have
about equal reliability, according to Hardison's paper, "Accuracy
of Streamflow Characteristics" (Prof. Paper 650-D, p. 210-212).
These extension limits are intended as a restriction for
publication of general information on flow characteristics at
gaged sites.

Greater extensions than those shown are permitted for certain
specified uses.  For example, it is permissible to estimate a 50-
year flood discharge from 10 or 15 years of record for inclusion
in a regional regression on basin characteristics, but the value
should not be published.  Also, greater extensions are permissible
when a value is needed for a hydrologic investigation; for
example, a 500-year peak may be estimated and used for a flood
insurance study.  However, a discharge characteristic which is
based upon a longer extension than described above and which is to
become public information, should be coordinated with other
concerned Federal agencies to identify and resolve potential
conflicts.  Minutes of the 72-3 meeting (June 6, 1972) of the
Hydrology Committee, Water Resources Council, noted that this
coordination was the responsibility of the investigating agency
and should be accomplished early in the investigation.

As described in Water Resources Council Bulletin 15, the log-
Pearson Type III analysis is the standard method for interpreting
a flood peak record (but not a low-flow or mean-flow record).
However, a graphical curve should be used if the LP Type III curve
does not provide a reasonable fit to the observed record or if any
significant historical data are available.

2.  Flood characteristics at gaged sites

Flood characteristics at a gaged site could be defined either by
probability analysis of the gage record or by a regional relation.
The characteristic to use and to publish should be the more
reliable of the two or perhaps a weighted average with the weights
based on an equivalent years of record concept.  This decision may
include some subjective judgment as well as the indices of
reliability, and no specific guidelines are provided.  The purpose
in discussing this problem without furnishing decision guidelines
is to rescind a policy of 10 or more years ago that recommended
use of characteristics estimated from regional relations at gage

3.  Regression analyses

The statement that 20 stations is a minimum sample that should be
used in a regression study (SWB Tech. Memo 70.05, p. Ei) was not
meant to apply to all hydrologic studies by regression methods.
Obviously, a regression using a lesser sample with only one or two
independent variables might provide a very useful relation.  The
adequacy of a sample of data probably should be judged as much
upon its representativeness of the population as upon the number
in the sample.

The standard error of a regression is only an index of the
reliability of estimates formed from the relation.  Standard
errors for linear   log-log relations should be reported to no
more than two significant figures in percent.  For relatively
small standard errors (40% or less), the positive and negative
deviations may be averaged and reported as a single value.  For
larger standard errors the positive and negative errors should be
reported as two separate values.  Standard errors should be
reported for defined relations, but in a designer's manual there
is no need for detailed explanations of the concepts and uses of
standard errors.

The most practical relation from a series of step-forward or
step-backward regression relations defined to estimate a flow
characteristic, should be selected on the basis of reliability and
simplicity of use.  A small reduction in the standard error of
estimate by inclusion of an additional independent variable in a
regression relation may not indicate a comparable increase in
reliability of estimates.  The decrease in standard error of
estimate must be weighed against the difficulty in defining an
additional independent variable.

4.  Adherence to guidelines

The Preface to the earlier TWRI's contains an excellent statement
on adherence to guidelines.

"Judgment must be used in deciding how closely to adhere to
instructions.  Instructions which include information on
preparation of data for use in a computer program must be closely
followed.  Instructions on methods of analysis are generally less
binding and are not to be utilized to the extent of inhibiting
initiative or stifling progressive development.  However, before
using a technique which differs substantially from one which has
been recommended, it should be discussed with the office of the
appropriate Branch Chief."

Walter Hofmann
Chief, Surface Water Branch

WRD Distribution:  A, B, S, FO