Geoscience Reference
In-Depth Information
presence of liquid water in the snowpack and multiple other factors. To address the coarse-
scale issue, De Lannoy et al. (
2010
) proposed several 3-D filter options to disaggregate
SWE data and propagate data from observed swaths to unobserved regions. These tech-
niques showed great benefit in a synthetic data study. When using real AMSR-E retrievals
(De Lannoy et al.
2012
), and with bias mitigation through re-scaling added to the system,
the assimilation analyses were affected by a lack of a realistic interannual signal in the
retrievals.
To address the problems with SWE retrieval accuracy, the potential of direct radiance
assimilation has been investigated (Durand and Margulis
2006
; Andreadis et al.
2008
;
Durand et al.
2009
; DeChant and Moradkhani
2010
). However, these efforts rely on a good
description of the snowpack in the land surface model, which is not always available for
large-scale applications. To address this, Forman et al. (
2013
) developed an artificial neural
network as a computationally attractive forward model in readiness for large-scale radiance
assimilation. In preparation for the future SMAP mission, freeze-thaw assimilation (Bateni
et al.
2013
) has been investigated, because of its importance in understanding the carbon
cycle.
The above studies update either snow or soil moisture separately. A major challenge for
land data assimilation is making use of total water storage (TWS) observations from
GRACE, which include soil moisture, snow and other water components at a very coarse
scale (Table
1
). Total water storage can be decomposed into soil and snow components and
disaggregated to finer scales (Zaitchik et al.
2008
; Su et al.
2010
; Forman et al.
2012
;
Li et al.
2012
; Reichle et al.
2013
).
4.7 Towards Operational Land Data Assimilation
Land surface processes and their initialization are of crucial importance to address the
challenge of seamless prediction from weather to seasonal and climate timescales (Palmer
et al.
2008
). It is well established that high skill in short- and medium-range forecasts of
temperature and humidity over land requires proper initialization of soil moisture (Beljaars
et al.
1996
; Douville et al.
2000
; Mahfouf et al.
2000
; Drusch and Viterbo
2007
; van den
Hurk et al.
2008
). A similar impact from soil moisture has been established for seasonal
forecasts (Koster et al.
2004a
,
b
,
2011
; Weisheimer et al.
2011
). Initialization of snow
conditions also has a significant impact on forecast accuracy at weather timescales
(Brasnett
1999
; Drusch et al.
2004
). Operational land data assimilation has initially focused
on ingesting precipitation observations (e.g., Saha et al.
2010
; Reichle et al.
2011
), but
improved snow and soil moisture state updates are now emerging, as documented, for
example, for the ECMWF Integrated Forecasting System by de Rosnay et al. (
2012a
).
An unprecedented operational land data assimilation product will be provided by the
Global Modeling and Assimilation Office (NASA GMAO) in the form of a level 4 satellite-
based soil moisture product (Reichle et al.
2012
; De Lannoy et al.
2013
). The assimilation
of SMAP brightness temperatures into the Goddard Earth Observing System land surface
model will yield a global root zone soil moisture product.
5 Conceptual Problems and Key Challenges
To summarize, the conceptual problems in our understanding of the hydrological cycle
over land can be grouped by observing, modelling and data assimilation systems. These are
outlined below.
