Geoscience Reference
In-Depth Information
Monitoring of the Soil Reservoir EXperiment, SMOSREX. Camporese et al. (
2009
) set up
synthetic soil profile assimilation experiments studying the effect of uncertainties,
ensemble size, bias and other factors with an EnKF. Because of the large impact of
parameters and forcings on soil moisture errors and biases, assimilation schemes have paid
increasing attention to including parameter estimation along with state updating, as, for
example, illustrated in Monsivais-Huerteroet et al. (
2010
). At present, EnKF filtering
experiments are being conducted at point-scales to further identify and address conceptual
problems with soil profile estimation, using surface observations (see, e.g., Han et al.
2012a
).
Another important conceptual problem with soil moisture assimilation, initially
addressed in a point-scale setting, is the direct assimilation of radiances or assimilation
using an observation operator. This is done to avoid inconsistencies between auxiliary
information that would be used in retrievals and that used in the land surface models.
Entekhabi et al. (
1994
) estimated 1-m-deep bare soil moisture profiles using synthetic
microwave brightness temperatures. This work was extended by Galantowicz et al. (
1999
)
using eight days of L-band brightness temperature (T
b
) data collected from a test plot in
Beltsville, USA. Pathmathevan et al. (
2003
) assimilated microwave observations with a
variational technique, but using a heuristic optimization, rather than an adjoint. Crosson
et al. (
2002
) tested T
b
assimilation at the point-scale with an EKF and showed that biases
could not be overcome through assimilation. Crow (
2003
) successfully assimilated T
b
for
soil moisture and showed improvements at the plot-scale, using either synthetic or real field
data. Crow analysed the EnKF performance in terms of the assumptions that underlie the
KF. Crow and Wood (
2003
) also used the EnKF at two sites within the Southern Great
Plains 1997 (SGP97) experimental domain and reported that T
b
data assimilation was able
to correct for rainfall errors. Wilker et al. (
2006
) highlighted the difficulty in mapping
heterogeneous soil moisture into T
b
using a forward operator and identified the repre-
sentativeness errors associated with these data. Similar to the above studies, Hoeben and
Troch (
2000
) used a KF including a forward backscatter model to explore the direct
assimilation of radar microwave signals to estimate soil moisture profiles.
Snow data assimilation has conceptual problems inherent to the cumulative and temporary
nature of this variable. Slater and Clark (
2006
) illustrated how a square root EnKF could
improve the snow state at in situ sites in Colorado during the accumulation and melt phase.
They also identified the temporal correlation in snowpacks and showed how it could limit the
efficiency of filtering if not accounted for properly. In a synthetic study, Liston and Hiemstra
(
2008
) proposed a technique to update snow retroactively, which would be useful for re-
analysis applications, if observations would only be available at the end of the snow season.
In situ snow data assimilation is performed operationally (see Sect.
4.7
below), usually with
simple assimilation techniques. An example where both the snow state and parameters were
estimated using an EnKF in a 1-D setting is given by Su et al. (
2011
).
A number of point- or single-grid-scale studies have tried to relate brightness temper-
ature data to snowpack characteristics (Durand et al.
2008
; Andreadis et al.
2008
), in
preparation for T
b
assimilation. Many of these studies highlight the large sensitivity of
snowpack estimates to model parameters (Davenport et al.
2012
), which makes both
forward simulation and inversion of T
b
observations for SWE estimation a difficult task.
4.6.2 Distributed Applications
The most obvious advantage of remotely sensed observations is the possibility of per-
forming
large-scale
and
spatially
distributed
assimilation.
It
should
be
recognized,
