Outcomes of synthesis - Runoff Prediction in Ungauged Basins: Synthesis across Processes, Places and Scales - page 369

elevation on predictive performance is more complex than
for aridity and air temperature. In many regions, predictive
performance improves with increasing elevation because
elevation is usually a surrogate for humidity (higher eleva-
tion corresponds to lower temperature and less energy and
therefore higher humidity). In cold regions, the occurrence
of snow is associated with high elevations. Both higher
humidity and snow contribute to an improvement of pre-
diction performance with elevation. However, there are
other regions where the predictive performance in fact
decreases with elevation, which may be related to an
increase in aridity with higher elevation. Elevation there-
fore shows a complex pattern, which is more difficult to
generalise because of varying co-dependencies among
processes.
Generally, if the individual regions from the L2 assess-
ments in Chapters 5

water has to travel. This is discussed in Chapter 4 in terms
of transit time distributions.
Landscape characteristics, behaviour and dominant
processes change with increasing catchment area. For
example, headwater catchments tend to be steep (and so
landslides are more common here), whereas flatland catch-
ments are larger and tend to be dominated by groundwater
aquifers, wide floodplains, frequent inundations etc. The
two therefore exhibit very different behaviour patterns.
With increasing catchment area new processes thus
become dominant, in a manner that depends on climate.
For example (see Chapter 2 ), in arid climates streams

become types (lose water to the underlying aquifer

system, and/or to evaporation), while in humid climates
streams tend to be

' losing '

types (are fed by water from
adjacent aquifers). Such changes and transitions make
catchment area a holistic similarity index (of the Darwinian
kind) for runoff signatures. So, in the Darwinian context,
scale or size can be viewed as a similarity parameter by
itself, since it reflects the legacy of co-evolution of natural
landscapes.
Another quality of scale is that data availability tends
to increase with catchment size. Smaller catchments tend
to be largely ungauged (see Chapter 3 , Figure 3.1 ); as
catchments get larger as one moves downstream, they are
more likely to contain rainfall and runoff gauging sta-
tions, which contribute to improved model performance.
Chapter 3 has proposed a hierarchical data collection
approach that can exploit the trade-off between scale,
data availability and costs. Global data sets provide more
generalised information at lower costs to the individual
user, whereas dedicated local measurements provide
detailed information at high costs over small spatial
scales. In this hierarchical data collection approach, one
begins at the global scale and, depending on resources
available, zooms in to different levels of detail at con-
secutively finer scales. When zooming in, a hierarchy of
controls from climate to local catchment and anthropo-
genic effects become evident, and can therefore be
deciphered.
On the other hand, distributed process-based models
need large amounts of data, of the kind that generally tend
to become less available with increasing scale. In fact,
data such as soils and vegetation characteristics used for
larger scale modelling tend to stem from regional data-
bases compiled from both ground data and remote sensing
data.

' gaining '

10 are examined, there is a tendency
for better predictability when the runoff seasonality is
strong and not so variable between years. This applies
not only to predictions of seasonal runoff, but to other
signatures as well, most notably the runoff hydrographs.
In some parts of the world, seasonality is strong and
predictable: this includes countries experiencing Mediter-
ranean climate, or cold regions with a significant snowfall/
snowmelt component. In both situations predictability
tends to be higher across all signatures. On the other hand,
in monsoon regions of the world (e.g., Asia, East Africa
and northern Australia), where monsoons introduce strong
seasonality, there is considerable inter-annual variability in
the timing and magnitude of precipitation, which may
actually reduce the predictability.

-

12.2.3 Synthesis across scales
Generic findings from chapters (non-assessment)
Hydrological processes occur at all scales, from micro-
scopic water flow in soil pores to global-scale interactions
of soil moisture and climate. The goal in this topic is
to predict catchment-scale runoff, which involves integra-
tion across spatial scales in some way. In the Newtonian
context (see Chapter 2 for a detailed explanation),

'
stands for scaling from higher to lower spatial resolutions
(upscaling) and vice versa (downscaling). Most of our
understanding of processes is at relatively small scales
(i.e., relative to the scales at which individual model elem-
ents are constructed), and yet predictions are needed at
large (catchment) scales. The way we normally achieve
such upscaling is to divide the landscape into different
(subgrid) units, and use spatially distributed modelling
methods (or other upscaling schemes that account for
catena effects and other organised heterogeneity) to gener-
ate predictions at the larger scales. Scale also has an effect
on the timing of runoff, since it determines the distance

' scale

Performance as a function of catchment size
The comparative assessment of runoff prediction perform-
ance in ungauged basins in this topic focused on the role of
catchment scale as a similarity parameter. Again, the aim
was to understand whether there are general patterns in