A data acquisition framework for runoff prediction in ungauged basins - Runoff Prediction in Ungauged Basins: Synthesis across Processes, Places and Scales - page 29

3 A data acquisition framework
for runoff prediction in
ungauged basins

Contributors: B. L. McGlynn,* G. Blöschl, M. Borga,
H. Bormann, R. Hurkmans, J. Komma, L. Nandagiri,
R. Uijlenhoet and T. Wagener

3.1 Why do we need data?
Most river basins around the world are ungauged; indeed,
only a few are gauged. Therefore, when runoff is required
at any ungauged river or catchment, it is estimated through
some kind of extrapolation from a gauged site to that
ungauged site, which is not straightforward. This is the
whole raison d

what models we should choose, and it can help us interpret,
condition and reject the predictions made by a model.
Hence, data are more than just input to a model. The value
of data becomes paramount when one begins to accept the
notion that catchments are complex systems, reflecting the
co-evolution of climate, soils, topography and vegetation,
and the patterns one sees in the landscape structure and the
runoff response (e.g., signatures) are emergent patterns, and
reflect more than the mere balance equations that are embed-
ded in many of today's process-based models. Therefore,
there is value and much to be learned from the combination
of runoff, climate and catchment data, a learning process that
we have called

etre of the PUB initiative. One way or
another, this extrapolation requires data of many kinds.
Extrapolation from gauged to ungauged catchments
requires a model of some kind, be it statistical, process-
based or a combination thereof. Implementation of models
needs data

'

all types of models need data to implement at
the ungauged location; indeed, all models gain legitimacy
from data as part of the validation process. Normally, and
certainly throughout this topic, we consider data of three
different kinds: runoff data (in gauged locations), climate
(input) data and catchment characteristics data.
Statistical models attempt to build statistical (e.g., regres-
sion) relationships between runoff at gauged locations and
associated climate and catchment data, which can then be
extrapolated for predictions in ungauged basins with the use
of local climate and catchment data. Process-based models
do the same, except that they benefit from the use of univer-
sal balance laws (mass balance, momentum balance etc.),
but they too need all three kinds of data (runoff, climate and
catchment) at gauged locations for calibration/validation/
conditioning, and climate and catchment data at

-

. Data will be the
ultimate source of the understanding that is embedded in all
the models because, when understood properly, they reflect
the co-evolution that is common to all catchments.
We can thus summarise the need for data in three categor-
ies: (i) data needed to read and understand the landscape in a
hydrological context; (ii) data needed to develop regression
relationships that will be used in statistical models; and (iii)
data needed for process-based models, such as climatic
forcing and parameter values, data to assist with model
development (inference from rainfall

' reading the landscape '

runoff data), and to
calibrate or validate models developed elsewhere.
The starting point for any PUB study therefore has to
include an assessment of available data and the information
that can be derived from this data. This activity includes the
need for catchment interpretation based on the available
database and the time frame of the study. Depending on
the application and the resources available, runoff predic-
tion in ungauged basins is generally based on different data
acquisition strategies, ranging from global data sets of typ-
ically low resolution to local and regional data sources of
varying availability and accuracy followed by field obser-
vations/assessment of local system characteristics. If the
resources are available, the most accurate runoff predictions
can be obtained by utilising very local data describing the
specific characteristics and behaviour of the system. The
input data requirement will be dependent on the nature of

-

the

ungauged locations where predictions are needed.
However, one should not be fooled into thinking that the
data are just inputs to a model, in the sense of

' '

.
Data have hydrological context, and contain hydrological
content. The data relating to runoff, climate and catchment
collected from any one place, interpreted by a trained hydrolo-
gist, and informed by prior knowledge from outside the place,
can reveal a lot about the hydrology of the place; it can inform

grist to the mill