Prediction of seasonal runoff in ungauged basins - Runoff Prediction in Ungauged Basins: Synthesis across Processes, Places and Scales - page 121

J F M A M

D

J

N

250 km

0 200 mm/month

N O S A J

Figure 6.17. Mean monthly runoff in France estimated from 872 gauging stations for the period 1981 - 2000. From Sauquet et al.( 2008 ).

criteria allowed the construction of a ow-regime map of

France, as illustrated in Figure 6.18 . The predictive accuracy
of geostatistical interpolation is presented in the comparative
assessment in Section 6.5 .

(e.g., Alley and Burns, 1983 ). The data can be stratified
by month, season, precipitation volume, or other climatic
discriminators. For example, Dey and Goswami ( 1984 )
found that runoff from Himalayan rivers was linearly cor-
related during the snowmelt period.
Runoff records may also be reconstructed based on
drainage area ratios, regression-based estimates of
monthly means and standard deviations using basin char-
acteristics, and linear and logarithmic correlation of con-
current runoff records (Hirsch, 1979 , 1982 ). Vogel and
Stedinger ( 1985 ) developed improved estimators of the
mean and variance for short-record gauges and applied
them for peak annual floods and monthly runoff. Longer
concurrent monthly records greatly enhanced the attract-
iveness of record augmentation procedures for estimating
the statistics of runoff occurring in each month because
the parameters of the model relating the runoff at the two
sites can be estimated more precisely than for floods.
However, the relationship between the concurrent runoff
at the two sites will, in general, exhibit some variation
from month to month so that this increased precision is
achieved by introducing some bias into the analysis,
which is relevant for time series reconstruction but not
for regime runoff transfer.
Solow and Gorelick ( 1986 ) applied co-kriging to esti-
mation of missing monthly runoff values in three records

6.3.4 Runoff estimation from short records
Short-term measurements can provide meaningful infor-
mation on seasonal runoff characteristics at a site. This is
true in regions with stable regime patterns. The stability of
a regime (see Section 6.2.1 ) is directly linked to the time of
observation required to estimate the long-term monthly
mean runoff (Rosenberg, 1979 ; Pfaundler et al., 2006 ).
The more stable a regime is, the shorter the observation
period needed to achieve a given accuracy. In regions with
a high degree of inter-annual variability, i.e., low stability,
the long-term mean flow regime does not describe the
conditions in individual years well. In this case, the mean
curve may be an artefact of averaging rather than a repre-
sentation of the hydrological dynamics ( Figure 6.9 ).
To extend a short runoff record, a relationship between
the short record and a longer (available) record must be
established. These relationships can be based on the cor-
relation between monthly runoff (Brown, 1961 ; Wright,
1976 ) determined through simple or multiple regression
methods on linear or logarithmically transformed data