A data acquisition framework for runoff prediction in ungauged basins - Runoff Prediction in Ungauged Basins: Synthesis across Processes, Places and Scales - page 35

and cannot be adequately replaced by other data sources.
All PUB methods

through the GRDC, is the European Water Archive
(EWA) of EUROFRIEND, the European group of the
FRIEND (Flow Regimes from International Experimental
and Network Data) initiative. EWA also contains infor-
mation about smaller, relatively undisturbed catchments.
Data are stored from about 3700 gauging stations in 29
countries. However, most of the gauging stations are con-
centrated in Western Europe. A more regional database
than EWA, also hosted by GRDC, is the ARDB (Arctic
Runoff Data Base), which is actually a subset of GRDB.
The database currently holds river runoff time series data
from a total of 2405 gauging stations in the arctic region
with the earliest records from 1877 and an average time
series length of 33 years, with a range from 1 to 123 years.
There are 1024 stations featuring daily data, while 2193
stations only contain monthly data. At the GRDC website,
there are various other data sets and subsets that might be
suitable for specific purposes and data availability will
vary widely depending upon the country and region of
interest.
An emerging technology to gauge water levels remotely
is laser altimetry by satellites, for which spatial and tem-
poral densities are rapidly increasing through the launch
of more and more satellites (such as TOPEX/Poseidon,
Jason-1, ICESat etc.) (Lettenmaier and Famiglietti, 2006 ;
Alsdorf et al., 2007 ). Not only can laser altimetry poten-
tially be used to record lake and river levels (Höfle et al.,
2009 ), it can also be used to measure cross-sections (by
measuring the width at different levels) or to derive rating
curves (on the basis of slopes and cross-sectional infor-
mation). In the future it could also be used for real-time
flood forecasting or flood inundation modelling.
The availability of continuous and long-term data sets
on runoff varies dramatically throughout the world (Kund-
zewicz, 2007 ). The lack of runoff data everywhere and the
decline of existing gauging stations are of course the
reasons for PUB in the first place (Stockstad, 1999 ), des-
pite the fact that the practical value of runoff data is often
much larger than the cost of their monitoring (Cordery and
Cloke, 1992 ). The decline of networks also suggests that in
many cases there will be inactive gauges that nonetheless
will provide some indication of the dynamics of the system
during previous time periods and conditions (e.g., Winse-
mius et al., 2009 ). Also, availability of data may some-
times be an issue due to administrative barriers (Viglione
et al., 2010a ).
Regardless of how the runoff observations are obtained,
measurement uncertainty will always be present and can be
considerable. Assessment of data quality and estimation of
data uncertainty are therefore important steps in any mod-
elling exercise. Stream gauges typically take continuous
measurements of river stage, which are translated into
runoff values using a rating curve. The stage

- and it does not matter how appropriate

or innovative are only the second best option after the use

of observed runoff data. However, if the best option is not
available we have to think about alternative strategies to
gain insight into catchment runoff characteristics. To use
stream gauge time series data from neighbouring gauges is
a good strategy, because the data structure and sensitivity
are similar. Depending on the aim of the PUB study, infor-
mation about catchment runoff at different temporal scales
is useful for predictions in the field of various hydrological
aspects, such as low flow, flood forecasting and design
value estimation. The data needed can best be discussed
separately for statistical and process-based methods.
Statistical methods for predicting runoff signatures usu-
ally require runoff data in neighbouring catchments. This is
for identifying pooling groups as well as for the statistical
predictive methods. For example regression equations
between catchment characteristics and runoff in neighbour-
ing catchments are used to estimate runoff at the target
location based on the catchment characteristics in that
catchment.
Process-based methods for predicting runoff often need
runoff data in neighbouring catchments for estimating
model parameters through calibration that are then trans-
ferred in space, or to transfer runoff characteristics to act as
constraints. Most importantly, runoff data may be available
at upstream or downstream locations. If these are close, the
more elaborate methods in Chapters 5 to 10 may not be
needed and simply scaling of the observed runoff to the
target area by the ratio of the catchment areas may be more
straightforward and more reliable. Also, opportunistic
gauging or short-term measurements at the basin of interest
can provide very valuable insight and aid as predictive
constraint. Generally, stream gauge rich environments or
regions might lend themselves best to statistical
approaches to PUB since interpolation distances are mostly
short. On the other hand, stream gauge poor environments
might require more process-based approaches to PUB.
Ideally, both tracks can be taken to constrain likely
predictions.

-

3.3.2 What runoff data are there?
Although not fully globally available, an extensive data-
base is available (GRDB, Global Runoff Data Base) at the
Global Runoff Data Center (GRDC) containing runoff
records from about 7300 gauging stations from 156 coun-
tries, with an average record length of 38 years. GRDC
operates under the auspices of the World Meteorological
Organization (WMO) and also offers other data products,
such as freshwater fluxes into oceans along coastlines, and
river basin outlines. A second database, also available - discharge