A data acquisition framework for runoff prediction in ungauged basins - Runoff Prediction in Ungauged Basins: Synthesis across Processes, Places and Scales - page 36

relationship in the rating curve has been derived from spot
measurements at a location with a (reasonably) fixed cross-
sectional geometry at different flow conditions or has been
pre-calibrated for a particular flow control structure.
Many studies have estimated the magnitude and impact
of rating curve uncertainty on runoff data and hydrological
modelling (Clarke et al., 2000 ; Peterson-

the PUB study, meteorological data at different temporal
and spatial scales are useful for predictions of various
runoff signatures (low flow,

flood forecasting, design

value).
Statistical methods for runoff predictions often require
catchment precipitation data. Analogously to the use of
runoff data, this is for identifying pooling groups as well
as for statistical predictive methods. For example, catch-
ment precipitation estimates (for instance, the mean annual
precipitation) are sometimes used as an auxiliary variable
in regionalisation methods ( Chapters 8 and 9 ).
Process-based methods for predicting runoff are always
driven by meteorological data as model forcing. Actual soil
moisture conditions directly affect runoff generation pro-
cesses and therefore are also important for flood and low
flow prediction. Soil moisture is usually simulated as an
internal model state in hydrological models, while the main
emphasis lies in runoff simulation. Soil moisture data
provide useful information to simulate the temporal and
spatial soil moisture dynamics in a more realistic way.

verleir, 2004 ;
Di Baldassarre and Montanari, 2009 ; Liu et al., 2009 ;
McMillan et al., 2010 ). Major sources of uncertainty
include data scarcity at high or low flow conditions, flow
outside the control structure during high flow conditions or
changes to the channel geomorphology. In some cases the
rating curve and the data points it was calibrated to might
be available and uncertainty can reasonably be estimated.
Uncertainty will likely be larger when ephemeral streams
are considered, due to the difficulty of measuring runoff in
such streams (Blasch et al., 2002 ; Adams et al., 2006 ). The
consideration of uncertainty in the runoff estimates (histor-
ical or spot gauging) can be useful to account appropriately
for the value of data available and avoid over-conditioning.

3.3.3 How valuable are runoff data for PUB?
Transferring hydrological information (e.g., model param-
eters, hydrological indices, runoff values) from neighbour-
ing gauged to ungauged catchments has been widely
investigated in recent decades as a method for runoff
prediction in ungauged basins (Merz and Blöschl, 2004 ;
Oudin et al., 2008 ). These works showed that use of data
from the nearby donors generally, though not always,
improves the quality of the runoff predictions. As runoff
propagation through branching networks provides a funda-
mental constraint to the distance metric, upstream and
downstream catchments would have to be treated differ-
ently from neighbouring catchments that do not share a
subcatchment. Also, climate plays a role in the predictive
power of data transfer. Patil and Stieglitz ( 2011 ) showed
that high runoff similarity among nearby catchments (and,
therefore, good predictability at ungauged catchments) is
more likely in humid runoff-dominated regions than in dry
evaporation-dominated regions.

3.4.2 Precipitation
Information about precipitation at different temporal and
spatial scales is essential for many PUB applications. It is
used as an auxiliary variable in statistical analysis (runoff
regression) or as input for hydrological rainfall

runoff
models. Precipitation data are available at the global scale
as a modelled and remotely sensed product down to the
point scale at rain gauges. The temporal scale varies from
minutes (at rain gauges) to monthly mean values for the
global products.
Global precipitation data: Many precipitation databases
are available globally. Global precipitation data are usually
a combined product from rain gauges, weather radar,
numerical weather prediction models and estimates from
remote sensing (Cheema and Bastiaanssen, 2012 ).
Examples are the Climate Research Unit database, CMAP
(CPC Merged Analysis of Precipitation) and WORLD-
CLIM (Hijmans et al., 2005 ). The last has a very high
spatial resolution (1 km or 30 arc-seconds), but contains
only monthly climatologies. Reanalysis data are the output
of numerical weather prediction models, which are condi-
tioned on available actual observations using data assimi-
lation routines. The NCEP/NCAR reanalysis data are
available from 1948 at a spatial resolution of approxi-
mately 210 km (Kistler et al., 2001 ; Kanamitsu et al.,
2002 ) and at a 32 km resolution for North America from
1979 (Mesinger et al., 2006 ). ECMWF offers three major
reanalysis products: ERA15 (1978

-

3.4 Meteorological data and water balance
components
3.4.1 What meteorological data and water balance
components are needed for PUB?
Appropriate meteorological inputs (precipitation, air tem-
perature, evaporation, snow cover) are needed to estimate
the required runoff response in ungauged basins either
based on rainfall

- 94, c. 120 km spatial

resolution), ERA40 (1957 - 2001, 100 km), ERA-Interim

runoff models or transferred information
from gauged catchments. Depending on the objectives of

- (1989 present, 80 km) (Simmons et al., 2007 ). Since 1997,

the Tropical Rainfall Measuring Mission (TRMM)

-