Process realism: flow paths and storage - Runoff Prediction in Ungauged Basins: Synthesis across Processes, Places and Scales - page 53

4 Process realism: flow paths
and storage

Contributors: D. Tetzlaff,* G. Al-Rawas, G. Blöschl, S. K.
Carey, Ying Fan, M. Hrachowitz, R. Kirnbauer, G. Jewitt,
H. Laudon, K. J. McGuire, T. Sayama, C. Soulsby, E. Zehe
and T. Wagener

4.1 Predictions: right for the right reasons
Runoff estimates in ungauged basins will always be based
on the application of some type of model that represents the
processes operating in the catchment. The level of detail at
which these processes are represented varies with the model
type, ranging from statistical models that represent the pro-
cesses in a basic way to detailed process-based models that
attempt to capture the key aspects of the bio-physical catch-
ment structure and their dynamic controls on storage, flow
paths and flow processes. One of the key characteristics one
would like a model to have is process realism, i.e., the model
should be a close representation of the real-world hydro-
logical processes (see Section 2.6). The main reason for the
need for process realism is that the PUB problem is essen-
tially an extrapolation problem, and extrapolations tend to
be more reliable if the processes are represented in a faithful
way. The extrapolation can be either in space, focused on
estimating key runoff signatures from similar neighbouring
catchments, or an extrapolation with the use of basic climate
and catchment data but without the ability to calibrate the
model against runoff data. Calibration against runoff data is
not an option in ungauged basins.
The model can still fit the data well, however, even if it
is not realistic in the sense described above. For example, a
regression model of low flows on the basis of catchment
characteristics may fit the data in the region well, but if the
coefficients do not truly represent the main process con-
trols on low flows in the region (e.g., geology, precipita-
tion), chances are that the model will not perform well
in an ungauged catchment. In other words, if the model
lacks realism and is simply a statistical best fit to the
data, the biases and uncertainties may turn out to be large.
A realistic model, in contrast, can be extrapolated more
reliably to ungauged basins. As Kleme š

reflect, even if only in a simplified form, the essential
features of the physical prototype. '

A crucial aspect one expects of a realistic model is to
represent the subsurface well (Beven, 2000 ). The subsur-
face structure of the catchment determines the time-
invariant controls on the gradients of hydraulic potential
that drive subsurface water flows, their dependence on
internal states and the patterns of flow resistance. Gradients
and resistances together determine the spectrum of subsur-
face flow velocities and flow paths. Catchments have
evolved through the interaction of several landscape pro-
cesses, climate and vegetation, and the non-linearity of the
processes and process interactions involved has invariably
produced an intricate pattern of surface and subsurface
flow paths tortuous and inter-connected flow paths oper-

ating at a multitude of scales. While there are a number of
geophysical measurement methods available to character-
ise the subsurface structure that governs these flow paths,
they tend to be time consuming to perform and the level of
detail with which these can be resolved at the catchment
scale remains limited. Understanding flow paths and stor-
age at the catchment scale is thus a major challenge. This
chapter therefore focuses on the realism of flow paths and
storage representations at the catchment scale for the pur-
pose of estimating runoff in ungauged catchments.
As rain falls on the ground, part of the water is inter-
cepted by vegetation (and the soil, including leaf litter)
and evaporates directly, part of the water infiltrates into
the soil, and the remaining part of the water runs off on the
surface. This partitioning occurs at multiple scales and can
be conceptualised in terms of several hillslope runoff
mechanisms that together link up to the catchment scale.
These include infiltration excess overland flow when the
rainfall intensity is larger than the ability of the soil to
infiltrate; saturation excess overland flow when the soil is
saturated due to prior rainfall and/or upwelling ground-
water; and subsurface flow through interconnected net-
works of vertical and lateral preferential flow paths.
These preferential flow paths reduce flow resistance along

-

( 1986a , p.178S)

put it: For a good mathematical model it is not enough to

work well. It must work well for the right reasons. It must

'