Prediction of low flows in ungauged basins - Runoff Prediction in Ungauged Basins: Synthesis across Processes, Places and Scales - page 179

in every region. Eng and Milly ( 2007 ) found a baseflow
recession time constant to improve the regional regression
model in the eastern USA. The trade-offs between data
length and the performance of these techniques have not
yet been fully explored.

1.40

0.60

0.30

8.4 Process-based methods of predicting low flows
in ungauged basins
While statistical methods exploit only static information,
process-based methods additionally exploit dynamic infor-
mation of the hydrograph and explicitly represent the
dynamics of low flow processes. For low flow prediction,
these methods focus on the recession parts of the hydro-
graph. The methods can be formulated either in the prob-
ability space (derived distribution methods) or as
simulation models of the whole runoff hydrograph ( Chap-
ter 10 ). This section discusses some specific aspects of
process-based models with regard to low flow prediction
in ungauged basins. One advantage of process-based
models is that they can explicitly account for any changes
in the precipitation regime and the catchment response
characteristics. Also, they may be able to represent local
peculiarities of catchments such as abstractions in more
detail than statistical methods.

0.10

0.06

Q 7,20

0.03

0.5 1.4 2.8 5.7 14.2 28.3

Runoff − Penn Creek (m 3 /s)

Figure 8.12. Estimation of low flow index Q 7,20 at Lost Creek, USA,
using the baseflow correlation method. Runoff records from Lost
Creek (subject site) are plotted versus the long record data for Penn
Creek (donor gauge), and a regression line is used to transfer the
low flow with a return period of 20 years from the donor site to
the subject site. From Riggs ( 1985 ).

that the relationship between annual minimum flows is
similar to the relationship between instantaneous base-
flows. Zhang and Kroll ( 2007a ) found the assumptions
generally to be reasonable for the USA. Stedinger and
Thomas ( 1985 ) examined the performance of baseflow
correlation with 20 pairs of runoff sites, and Reilly and
Kroll ( 2003 ) expanded this analysis to over 1300 runoff
sites in the USA. They found this method to perform well
when baseflow measurements of independent low flow
events and donor sites located within 200 km were used.
The method is improved as the number of baseflow
measurements is increased, although some levelling off
of performance was observed with more than 15 baseflow
measurements. When only five baseflow measurements
are available, the use of multiple donor sites can signifi-
cantly improve the performance of this method (Zhang
and Kroll, 2007a , b ).
Augmented regression refers to the inclusion of runoff-
derived indices in regional regression models. With this
technique, a small number of runoff measurements are
applied to estimate catchment indices that are difficult to
obtain by other means. For low runoff statistics, these
indices often relate to the hydrogeology of the catchment.
Vogel and Kroll ( 1992 ) employed estimators of the base-
flow recession constant in a physically derived regression
model to improve low runoff statistics in Massachusetts.
Kroll et al.( 2004 ) examined the use of baseflow recession
constants and the baseflow index in regional regressions
throughout the USA. They found that the inclusion of these
hydrogeological indices improved the low flow predictions

8.4.1 Derived distribution methods
In the derived distribution approach, the low flow indices
are obtained by combining the statistical characteristics of
precipitation with those of the catchment response.
Gottschalk and Perzyna ( 1989 ) incorporated runoff pro-
cesses in a distribution function of low flow in terms of
baseflow recession. The derived distribution function con-
tains four parameters of which two are determined from a
traditional recession analysis of low flow periods. The
other two are derived from a statistical analysis of
the maximum length of periods when pre-

cipitation is less than an assumed threshold value.
The distribution function with the same parameters can
be applied to calculate mean low flow for different dur-
ations. They tested the method for summer low flows in
gauged catchments in southern and western Norway.
Gottschalk et al.( 1997 ) extended this body of work to
deduce expressions for a family of low flow distributions
related to linear and non-linear recession models. The
distributions included the Weibull distribution. The
approach was shown to be promising in regions where it
is difficult to fit a regional distribution to the sample data
and to estimate low flow distributions in ungauged basins
(Pacheco et al., 2006 ). Specifically, Pacheco et al.( 2006 )
extended the previous work of Gottschalk et al. (1997) to

' dry weather '