Integrated Models in Hydrology and Meteorology - Transport in the Atmosphere-Vegetation-Soil Continuum - page 344

normally provided by the atmospheric model are taken from observations. These
forcings include precipitation, incoming longwave and shortwave radiation, wind
speed, temperature and humidity. Furthermore, certain parameters are based on obser-
vations, such as albedo and roughness length. After the LSM has been run in this way,
observed luxes are used as a validation of the luxes produced by the model (e.g.,
sensible and latent heat lux, net radiation). Whereas developers of models usually
use their own data sets for a irst validation, coordinated validation exercises are also
used to intercompare the skills of different LSMs for a common set of observations.
Examples of these activities are PILPS (Project for Intercomparison of Land-surface
Parametrization Schemes: Henderson-Sellers et al., 1995 ), GSWP (Global Soil Wet-
ness Project: Dirmeyer et al., 2006 ) and initiatives linked to a certain region such as
West Africa (Boone et al. 2009 ) or surface types like cities (Grimmond et al., 2011 ).
Off-line testing does not provide insight in the effect that a certain LSM has on the
overall behaviour of the model. Therefor on-line tests are also needed, which might
reveal effects on, for example, precipitation, cloud formation, soil moisture depletion
and runoff (e.g., Balsamo et al. 2009 ). For on-line tests not only surface luxes can be
used for validation, but, for example, screen-level temperature, humidity, wind speed
and precipitation as well. For operational weather models these variables happen to
be important variables to assess the skill of a model.
In the operational use of LSMs observations play a role as well. In the context
of weather forecasting, each forecast needs to start with a correct initial state of the
model. For LSMs this initial state may comprise soil temperatures and moisture con-
tent at various depths as well as the amount of snow cover and intercepted water.
Without additional information, this initial state could be carried over from a previous
forecast, but any error in that forecast will remain, or even amplify, in the new fore-
cast. Therefore, similarly to what is done for the atmospheric part of weather models,
data assimilation can be used to correct the initial state of the LSM and bring it as
close to reality as possible. One of the methods that is used is based on the fact that
screen level air temperature and humidity relect in part the energy partitioning in the
surface energy balance: a low Bowen ratio will lead to low temperatures and high
humidity contents, whereas a high Bowen ratio will give warm, dry air. In turn, the
Bowen ratio is strongly inluenced by the amount of available soil moisture. Thus, an
error in the soil moisture content will lead to an error in the screen level temperature
and humidity and can hence be detected by comparing the forecast of T and RH with
the observed values for the same moment. Based on the discrepancy between the
two, the soil moisture can be adjusted (see, e.g., Giard and Bazile, 2000 ; Drush and
Viterbo, 2007 ).
Furthermore, directly observed variables such as snow cover or top-soil soil mois-
ture content can be used to correct the initial model state (see, e.g., Mahfouf, 2010 ).