A synthesis framework for runoff prediction in ungauged basins - Runoff Prediction in Ungauged Basins: Synthesis across Processes, Places and Scales - page 23

declustering property of geostatistics, i.e., the ability to
give less weight to observations that are close to each other
because they are correlated, so contain less information
about the random variable. For clarity, in Chapters 5 and
6 simple methods based on spatial proximity are also
discussed within the geostatistics section, since there are
practical similarities in the procedures of mapping with
the geostatistical method, even though, strictly speaking,
no random variables are involved.
Estimation from short runoff records: While this topic is
about runoff predictions in ungauged basins, there may be
instances where a short runoff record is indeed available.
The record may be too short to estimate the runoff signa-
tures to a level of accuracy that is sufficient for the problem
at hand. However, together with information from other
catchments in the region and the use of regionalisation
methods, it may be possible to exploit the information that
is contained in the short runoff records. Runoff information
from a neighbouring catchment is usually used to account
for the temporal variability in the runoff signatures in the
poorly gauged catchment of interest as a result of the run-
off records in that catchment being too short.

directly in terms of probabilities, often in an analytical
way, which makes for a clear model structure. However,
the model parameters may not be easy to identify in
ungauged basins.
Methods based on continuous rainfall

runoff models:
All the signatures (annual runoff, seasonal runoff, flow
duration curve, low flows, floods) in ungauged basins
can be estimated in a straightforward way if runoff hydro-
graphs are available in that catchment over a sufficiently
long period. One way of estimating these signatures there-
fore is first to estimate hydrographs in ungauged basins
and then to extract the signatures from them. If the focus is
on a particular signature, special considerations in rainfall

-

-
runoff modelling may apply, e.g., one may strive to repre-
sent runoff

model, if one is interested in low flows in ungauged basins.
This method hinges on the accuracy of runoff modelling in
ungauged basins, which often justifies the use of alterna-
tive methods.
Methods that exploit proxy data: While no runoff data
are available in ungauged basins, there may be other data
available that may contain useful information about the
runoff signatures. This method strives to make use of such
data as flood marks, vegetation patterns, and a range of
remote sensing products on hydrological variables such as
snow and soil moisture.

low flows particularly well by the rainfall -

2.3.2 Process-based methods of predictions in
ungauged basins
A second type of methods for estimating runoff in
ungauged basins is process-based methods. Process-based
methods are normally based on some combination of bal-
ance equations of mass, momentum and energy. Most of
them are deterministic methods, i.e., without random elem-
ents. However, there are some combinations of process-
based methods with statistical methods. The model struc-
ture, in most instances, is assumed a priori, based on a
conceptual understanding of the hydrological processes
operating at the catchment scale. For the case of predicting
runoff hydrographs, several methods include models that
are based on an understanding of hydrological processes
obtained at the laboratory scale. Examples are models that
use the Richards equations for estimating infiltration and
subsurface water movement. Model parameters for the first
type of conceptual models are usually inferred from par-
ameters that have been found by calibration to runoff in
neighbouring catchments. Model parameters for the second
type of models that are based on laboratory-scale
governing equations are usually inferred from field data
and similarity assumptions. In this topic, the process-based
methods have been assembled into three groups:
Derived distribution methods: In this type of approach,
the runoff signatures (such as floods) are estimated from
precipitation signatures (such as rainfall statistics). The
appealing feature of the derived distribution approach is
that

2.4 Assessment of predictions in ungauged basins
2.4.1 Comparative assessment as a means of synthesis
In the comparative hydrology approach, the idea is to learn
from the similarities and differences between catchments
in different places, and to interpret these in terms of under-
lying climate human controls. In a quantitative

science such as hydrology, learning comes from hypoth-
esis testing, and the hypotheses in the context of PUB are
runoff predictions in ungauged basins. Testing the predic-
tions against independent data demonstrates that the under-
standing of the system is real. Assessing the predictions of
runoff is therefore a scientific exercise and we can learn
from the performance of such predictions. A comparative
assessment provides a much wider richness of insights than
testing a model at a single place. One place has only one
history, whereas many places have multiple histories and
hence can contribute much to our understanding.
Assessing how well the runoff predictions perform is a
particularly important and interesting exercise because the
predictive uncertainties tend to be large relative to the
magnitude of the runoff to be predicted. The uncertainties
are due to many reasons. Hydrological processes have
enormous spatio-temporal variability, which is difficult to
capture. Runoff data are only collected at a few points in

- landscape -

the rainfall - runoff relationship can be formulated