Prediction of runoff hydrographs in ungauged basins - Runoff Prediction in Ungauged Basins: Synthesis across Processes, Places and Scales - page 261

Soil moisture and groundwater storage
Observed soil moisture data are another kind of dynamic
data that may help improve model parameters over a-priori
estimates in ungauged basins. Obtaining representative
soil moisture data within an ungauged basin is difficult
mainly because of scale issues ( Western and Blöschl,
1999 ; Western et al., 2001b , 2003 ). Ground measurements
may be representative of the entire root zone but are
usually limited to a few spots in a catchment (Grayson
et al., 1997 ). Spaceborne estimates of soil moisture can be
retrieved for large areas (e.g., Wagner et al., 2003 ) but may
be limited by shallow penetration depths, which are much
smaller than the root depth represented in many
hydrological models. Numerous methods have been
developed for dealing with this incompatibility, including
downscaling schemes and multilayer soil hydrological
models (Houser etal., 2000 ; Walker et al., 2001 ; Schuur-
mans and Troch, 2003 ) but challenges remain. There is a
multitude of soil moisture products that can be used to
constrain the model parameters in ungauged basins, such
as ERS scatterometer data ( Wagner et al. , 2007 ). The soil
moisture data may also help to update the simulated soil
moisture states themselves, which may improve the runoff
predictions.
Soil moisture can be assimilated into the hydrological
models by a range of techniques such as variants of the
ensemble Kalman filter (e.g., Moradkhani et al., 2005 ;
Komma et al., 2008 ; Crow and Ryu, 2009 ). A number of
synthetic experiments demonstrated the usefulness of soil
moisture data (e.g., Crow and Ryu, 2009 ). Meier et al.
( 2011 ) assimilated ERS scatterometer data into a model
for three subcatchments of the Zambezi River basin and
found significant improvements of the runoff predictions
over cases where no soil moisture data were used. Parajka
et al.( 2006 , 2009b ) constrained regionalised a-priori infor-
mation about the model parameters with soil moisture from
the ERS scatterometer data for 320 Austrian catchments
treated as ungauged. These data improved the runoff pre-
dictions in lowland agricultural catchments characterised by
low vegetation and small topographical variability, but this
was not the case in Alpine catchments. The value of using
satellite soil moisture data for constraining model param-
eters apparently depends on the surface characteristics.
A review of the potential of remotely sensed soil moisture
for runoff predictions is given in Bronstert et al.( 2012 ).
Important dynamic data on the subsurface are ground-
water levels. In physics-based models that use Darcy

for runoff simulations may depend on how spatially repre-
sentative they are, i.e., how heterogeneous the aquifer is.
Although at a larger scale, satellite data from GRACE
(Gravity Recovery and Climate Experiment) may also
have potential to constrain the parameters of runoff models
as they capture the total water storage in a catchment
(Güntner, 2008 ; Klees et al., 2008 ).

Evaporation
A major source of uncertainty in predicting runoff in
ungauged basins is the evaporation. Constraining model
parameters by estimates of evaporation is therefore attract-
ive. Winsemius et al.( 2008 ) constrained land surface
related parameter distributions of a conceptual semi-
distributed hydrological model by time series of satellite-
based evaporation. They estimated evaporation by the sur-
face energy balance algorithm for land (SEBAL) ( Allen
et al., 2007 ) and applied the approach to the ungauged
Luangwa River basin in Zambia. Remote sensing not only
provided the satellite data information on the largest
outgoing water balance term, evaporation, but also on
the depletion of soil moisture during the dry season.
They distributed the model parameters to which evapor-
ation is sensitive on the basis of dominant land cover
characteristics within the catchment, and conditioned them
on satellite evaporation estimates by Monte Carlo sam-
pling. The parameters so constrained were spatially clus-
tered and consistent with hydrological landscape units:
wetland dominated areas, forested areas and highlands,
which allowed a useful hydrological interpretation. The
method clearly improved the parameter estimates beyond
only using a-priori information. In a somewhat similar
study, Li et al.( 2009 ) calculated actual evaporation dir-
ectly using the Penman Monteith equation, with the sur-

face conductance in the Penman

-

Monteith equation
estimated from remotely sensed (MODIS) LAI. These data
assisted in estimating the parameters of a daily runoff
model in ungauged basins. Their results indicated that the
use of LAI data improved both the runoff model efficiency
during the calibration period as well as the daily runoff
prediction in ungauged catchments. The study concluded
that further improvements in model structure may help to
improve the efficiency of

-

the remotely sensed data

integration.

Water level and inundation patterns
Water level data are another valuable piece of information
for estimating model parameters in ungauged basins. Sun
et al. (2011) used satellite radar altimetric observations of
river water levels at the basin outlet to calibrate a hydro-
logical model, as a surrogate of runoff data. They coupled
the hydrological model with a hydraulic model describing
the relationship between runoff and water stage. The

s law,
the use of groundwater level data is of course a standard
procedure (e.g., Refsgaard, 1997 , 2001 ; see Figure 4.11 ).
For conceptual models, Kuczera and Mroczkowski ( 1998 ),
for example, suggested that, in their study, groundwater
level data did not constrain the parameters of a runoff
model much. The usefulness of groundwater level data

'