FLOOD00408

• County: Statewide
• Title: Techniques of Water-Resources Investigations of the US Geological Survey Regional Analyses of Streamflow Characterictics
• Date: 1/3/1997
• Prepared By: USGS
• Floodplain - Doc Type: Educational/Technical/Reference Information

TECHNIQUES OF WATER-RESOURCES INVESTIGATIONS

8
20,000
10,000
0
Z
0
u
w
OIl
'"
w
'"
....
w 1000
w
~
U
'"
::l
U
Z
0
0
0 100
~
~
'"
<( 016
w
,.
I
0

20X

20x

0"

\70

EXPLANATION

40

Parameter i$ mean annual
runoff, in inches
)( Based on 9 or more years
of record
. Based on less than 9
yeon of record

.,'

10 ·
1

.

\6x

:;:
~
~

010

o
~
w
>
..
:>
u
~
o
::: 2

s

x

~
Z
o
~

Q
~
:: 0.5
9
~

o

,

~
o
z 0.2
o S
~
..
>
w
Q

10 20
MEAN ANNUAL
RUNOF F
IN INCHES

10 100
DRAINAGE AREA, IN SQUARE MILES

1000

.

4. Graphical regression of lO-year floods defined from short records, Kootenai River bosin, Montana.

than about 0.8. However, under certain condi-
tions improvement can be obtained at smaller
values of the correlation coefficient (Matalas
and Jacobs, 1964).
The improvement of a frequency curve by
correlation with a longer record is obviously
desirable if one is concerned only with charac-
teristics at that site. But use of such a modified
frequency curve in a regional analysis may
not improve the result over that obtained by
use of the unmodified curve, the reason being
that the flood experience at the two sites may
not be independent. Because the regionaliza-
tion process attempts to average that part of
the variability due to random occurrences of
weather, a substantial extension of a fre-
quency curve on the basis of another one al-
ready included in the analysis will tend to
duplicate the experience at the site of the
longer record. This duplication may bias the
result of the regional analysis. As an extreme

example, consider a region containing 20 rec-
ords of 10 years and one record of 50 years,
and assume that each of the short records is
adequately correlated with the concurrent
part of the long record to justify extension to
50 years. If the 20 records were all for the
same 10-year period, the upper part of the
regionalized curve would be nearly identical
to the individual curve for the 50-year period
of record (given the same basin characteris-
tics). Now suppose that given the same data,
a regionalization based on unextended fre-
quency curves was made. This would still lean
heavily on the long record for the higher re-
currence intervals and might not be too much
different at 20 years and below because of
the high correlation among all records. So
which result is the better? There is little basis
for a decision, but because the difference in
results likely would be small, we probably
would select the method usiug the unextended

.