Geoscience Reference
In-Depth Information
4.6 Data Assimilation Research Applications
Table
4
shows a selection of studies using a variety of observation types to improve the
land surface state or the state in a hydraulic, vegetation or snow model coupled to it.
Because of its success in highly nonlinear land surface modelling (Reichle
2008
), the
EnKF has gained a lot of attention. Therefore, state estimation studies using an EnKF or
EnKS (Ensemble Kalman Smoother, where the time integration is done forwards and
backwards) are organized separately from those that use any other assimilation technique
(e.g., variational, optimal interpolation). Also shown are a few examples on parameter
estimation in land surface or forward models. While this review focuses on state estima-
tion, parameter estimation and forcing correction are of utmost importance in land surface
models. Land surface models are not chaotic and thus benefit less from state estimation
than atmospheric or oceanic applications. By contrast, parameters and forcings determine
the major part of the land surface model uncertainty, and great advances can be expected
from combining state, bias, parameter and forcing estimation (Moradkhani et al.
2005b
;De
Lannoy et al.
2006
; Vrugt et al.
2012
). Here, we discuss a number of soil moisture and
snow-related studies done mainly for state updating, with particular attention to the con-
ceptual problems they address. Examples on evapotranspiration, surface or skin temper-
ature, LAI (leaf area index), discharge and water stage assimilation are also provided in
Table
4
, but not discussed in detail.
4.6.1 Single-column Applications
To explore the possibilities and limitations of assimilation schemes, numerous studies have
first explored single point-scale or grid cell-scale applications. For soil moisture assimi-
lation, conceptual problems include the propagation of information from the surface to the
entire soil profile; the optimization of assimilation techniques and update frequencies; and
the identification of an allowable level of uncertainty in surface observations to be useful in
a data assimilation scheme, mostly in view of satellite sensor design.
Georgakakos and Baumer (
1996
) performed a sensitivity study to document the impact
of observation noise on Kalman filter (KF) results. Calvet et al. (
1998
) and Wingeron et al.
(
1999
) assimilated surface soil moisture data from a soil profile in the highly instrumented
field site of the Monitoring the Usable soil Reservoir EXperiment (MUREX) in France to
update root zone soil moisture using variational approaches and investigated the impor-
tance of assimilation windows and observation frequencies. Similarly, Li and Islam (
1999
)
studied the effect of assimilation frequency while directly inserting gravimetric mea-
surements as surrogates for remote sensing data, and Aubert et al. (
2003
) suggested that a
1-week soil moisture update is sufficient. Walker et al. (
2001a
) showed in a synthetic
profile study that the KF was superior to direct insertion. In a subsequent study with real
data from the Nerrigundah catchment in Australia, Walker et al. (
2001b
) articulated the
idea that soil moisture assimilation can solve issues with errors in forcings or initial
conditions, but not errors caused by problems in the physics of the soil model.
De Lannoy et al. (
2007a
) used an EnKF to study vertical information propagation, and
the effect of assimilation depth and frequency for an extensive set of soil profiles in an
USDA field in Beltsville, USA. This study highlighted the effect of bias propagation
through the profile and the need for bias estimation, a conceptual problem that was
addressed with a two-stage forecast and bias filter (De Lannoy et al.
2007a
,
b
). At the same
time, Sabater et al. (
2007
) studied the concept of propagating surface observations to
deeper model layers using different types of filtering, using ground data from the Surface
