Introduction - Runoff Prediction in Ungauged Basins: Synthesis across Processes, Places and Scales - page 9

laboratory to the global scale. The goal in this topic is to
predict runoff at the catchment scale. This necessarily
involves integration across spatial scales in some way.
There are two strategies for approaching this scale problem
to assist with making runoff predictions. The first is the
upward or mechanistic approach. It is strongly based on
laboratory experiments and involves an upscaling to the
catchment scale, often by spatially explicit, distributed
modelling approaches. While causal controls can be ana-
lysed very well by this approach, it is difficult to represent
all catchment-scale process interactions. The second is the
downward or holistic approach. It is strongly based on
behaviour observed at the catchment scale, often based
on lumped statistical methods or conceptual rainfall

runoff
models. While these types of approaches do have the
ability to capture process interactions at

-

Figure 1.7. Reconciliation of downward and upward approaches -
break or transition from a reliance on averaging and parameterisation
of lower-level features to a culture of discovery and explanation of
emergent phenomena at the higher level. From Sivapalan ( 2005 ).

the catchment

scale -

identifying causality may be very difficult. The two
approaches typically deal with scale issues in different
ways. For example, in the upward mechanistic approach
of rainfall

- if this information is represented in the data

ways. In the topic we are therefore inclusive of all these
approaches. In our considered view these approaches
need to be compared both in terms of their characteristics
and in terms of their performance when applied to real
catchments around the world. As the methods have dif-
ferent strengths and weaknesses, the choice of method
is an interesting and important issue, and there will be
many cases where it may be wise to use methods that
combine the strengths of both approaches and minimise
their weaknesses.

runoff modelling, measured point rainfall is
distributed across the catchments and then routed explicitly
across hillslopes and along the stream network to
obtain flood peaks of catchments of different sizes. In
contrast, in the downward statistical approach of regional
flood frequency analysis a scaling relationship is usually
established between flood peak and catchment area
that embodies all processes in a holistic way. In both
approaches, in practice, model parameters are usually cali-
brated in some way in order to reduce bias. Biases tend to
change significantly with location but tend not to change
much over the time scales we are interested in because
much of the bias is related to unknown subsurface charac-
teristics. Calibration therefore has the potential to increase
the accuracy of runoff predictions. However, it is clear that
if one calibrates parameters to compensate for the real
uncertainty, this is likely to be a

-

1.6 How to read the topic and what to get out of it
How is the synthesis across processes, places and scales
reflected in the organisation of the topic? Synthesis along
these three axes has in fact been the guiding principle in the
structuring of the topic.
Synthesis across processes is reflected in the way that
the topic is organised around runoff signatures. Each of the
Chapters 5 to 10 deals with one runoff signature

, which may
jeopardise the physical realism of the model and therefore
its predictive capability in ungauged basins.
This topic takes a view that transcends any particular
approach. We do not assume the primacy of either of the
two prediction approaches. Of course, if hydrological
information were to be available everywhere, and all the
time,

' quick fix '

from
annual runoff to runoff hydrographs. The commonality of
structure of each chapter acknowledges that signatures
have common causes in the way catchments function
hydrologically. The signatures are simply different mani-
festations of the same spectrum of catchment processes, so
a coherent and consistent treatment contributes to a syn-
thesis across processes.
Synthesis across places is reflected in the way that
hydrological similarity is one of the recurring themes of
the topic, also reflecting one of the key PUB themes (i.e.,
catchment similarity and classification). Hydrological
similarity appears explicitly in Chapters 5 to 10 as a
vehicle to advance understanding and predictions, through

-

the upward approach would be preferred. But it

never is. In fact this is the raison d être proffered for

the downward approach. We view runoff processes as
space-time patterns of hydrological variability. Any
approach is an approximate representation of this variabil-
ity. The organisation of catchments into a stream network
leaves an imprint on runoff response, turning them into
organised entities, and the aim of runoff predictions is to
connect the process to the pattern. Each of the two
approaches connects process and pattern in different

'