Geography Reference
In-Depth Information
Figure 5.20. Observed mean annual
evaporation plotted versus modelled
predictions, for 262 catchments in
Australia. Catchments are labelled by
seasonality of rainfall. The 1:1 line is
plotted for reference. From Potter
et al.(
2005
).
1800
Summer-dominant rainfall
Non-seasonal
Winter-dominant rainfall
1500
1200
900
600
300
0
0
300
600
900
1200
1500
1800
Predicted mean annual evaporation (mm/yr)
than an hour. To compute a time series of annual runoff in
a PUB setting requires a time series of precipitation and
relevant forcing variables for the target catchment, along
with appropriate model parameters. Assuming that a rea-
sonable predicted time series of runoff output can be
obtained from the model, the mean, inter-annual variability
and auto-correlation of annual runoff can be computed
from the statistics of this time series. Other annual series-
based analyses include trend analysis (Chiew and McMa-
hon,
1993
; Salas,
1993
; Milly et al.,
2008
), runs analysis
(Yevjevich,
1967
; Saldarriaga and Yevjevich,
1970
; Sen,
1976
; Hisdal et al.,
2001
; Peel et al.,
2004a
,
2005
), estab-
lishing the probability distribution function of the annual
runoff (Vogel and Wilson,
1996
; McMahon et al.,
2007b
),
and analysing the time series structure of the data that will
allow more sophisticated analysis such as stochastic data
generation to be carried out (Matalas,
1967
; Stedinger and
Taylor,
1982a
,
b
; Hipel and McLeod,
1994
; Thyer et al.,
2002
).
The difficulty in applying continuous models is two-
fold: (i) identifying an appropriate model structure and
(ii) obtaining the necessary parameter set(s) that allow the
model to produce plausible runoff values. Parameter
regionalisation to ungauged catchments is often con-
founded by poor parameter identification at gauged catch-
ments. There are numerous sources of parameter
uncertainty, including errors in input data, errors in model
structure and errors of the calibration data. Even the choice
of the objective function and optimisation technique for
calibration contribute to uncertainty (see Peel and Blöschl,
2011
, and references therein). These issues are discussed in
more detail within the PUB framework in
Chapter 10
and
are not addressed here. There are few practical methods
and only limited guidance for objectively assessing model
structure for ungauged catchments (e.g., when using
lumped conceptual models, more arid catchments gener-
ally require more complex models). Once a model struc-
ture is chosen, there are many methods for parameter
estimation (see Section 10.4). Thus, the benefits of making
high resolution temporal predictions trade off to some
extent against the challenges imposed by managing uncer-
tainty in these predictions.
5.4.3 Proxy data on annual runoff processes
Tree ring chronology and paleoclimatology
Proxy data allow analysts to extend time series of annual
runoff to periods prior to runoff observations. A statistical
relationship, usually regression, is established between
observed runoff and one or more proxy data series, which
is then used to synthesise a time series of annual runoff
driven by the long proxy records. An abundance of litera-
ture exists where tree ring chronology or other paleocli-
mate proxy records are used to develop satisfactory
relationships with observed annual runoff data. The
NOAA Satellite and Information Service (NOAA Paleocli-
matology,
2011
) lists runoff reconstructions for several US
rivers, the Selenge River in Mongolia, and the Burdekin
River and other Queensland rivers in Australia. Examples
of recent runoff reconstructions outside the USA include
the Churchill River in northern Saskatchewan, Canada
(1840
2002) (Beriault and Sauchyn,
2006
), four rivers in
coastal Queensland, Australia (Lough,
2007
), the Murray
River
-
1988) (Gallant and Gergis,
2011
), the Yellow River in western China (Gou et al.,
in Australia (1783
-
Search WWH ::
Custom Search