Geoscience Reference
In-Depth Information
Data assimilation (Kalnay
2003
) provides an intelligent method to fill in the observa-
tional gaps using a model or to steer models using observations. By intelligent, it is meant
an ''objective'' way which makes use of quantitative concepts (e.g., mathematical) for
combining imperfect information. By combining observational and model information,
data assimilation can be used to test the self-consistency and error characteristics of this
information (Talagrand
2010b
).
In this paper we focus on off-line land data assimilation, where the LSM is uncoupled
from an atmospheric model. By using an uncoupled LSM, it can be forced with more
observation-based forcings, rather than often inaccurate atmospheric analyses, and less
computational resources are needed. The uncoupled approach can be regarded as a first
step towards the land data assimilation goal of coupling an LSM to an atmospheric model
to improve predictions at weather, seasonal and climate timescales (Palmer et al.
2008
).
In this paper we discuss observations (Sect.
2
), models (Sect.
3
) and data assimilation
methods (Sect.
4
) used in the studies of the hydrological cycle and provide illustrative
examples, with a focus on soil moisture. We pay special attention to the conceptual
problems and key challenges associated with making use of observational and model
information of the land surface in data assimilation systems (Sect.
5
). We finish by pro-
viding conclusions (Sect.
6
).
2 Observations of the Hydrological Cycle
Observations of the hydrological cycle are commonly divided into conventional obser-
vations (e.g., in situ ground-based measurements such as screen-level relative humidity)
and remotely sensed observations (e.g., satellite or aircraft microwave observations). These
data sets are complementary: conventional observations have relatively high spatio-tem-
poral resolution (order metres and minutes) but only have local coverage, so have poor
representativity for a large area; satellite observations have relatively low spatio-temporal
resolution but have global coverage, so have good representativity for a large area. In situ
observations are typically used as ground truth for calibration and validation of remote
sensing products, and model and assimilation results.
Table
1
gives an overview of satellite sensors and missions that contribute to our current
understanding of the hydrological cycle or may potentially contribute to this understanding
in the near future. Depending on the observed wavelengths, the orbit altitude and design
details, there are large differences in horizontal, vertical and temporal resolution of each
observation type. For example, satellite-based observations of soil moisture are made using
passive and active microwave instruments. The horizontal resolution of these sensors
ranges from 50 to 10 km; the temporal resolution is about one observation every 2-3 days,
depending on the location on Earth. These instruments typically penetrate the first few
millimetres to centimetres of the soil: a few millimetres for the X-band (8-12 GHz, e.g.,
Advanced Microwave Sounding Radiometer for EOS, AMSR-E; Njoku and Chan
2006
);
*1 cm for the C-band (4-8 GHz, e.g., AMSR-E; Advanced SCATterometer, ASCAT;
Bartalis et al.
2007
); and *5 cm for the L-band (1-2 GHz, e.g., Soil Moisture Ocean
Salinity, SMOS; Kerr et al.
2010
). An immediate conceptual problem is to estimate soil
moisture of actual interest in the root zone (1 m) at a finer resolution. For this, observa-
tional information needs to be transferred from the surface layer to the root zone (e.g.,
Calvet et al.
1998
; Sabater et al.
2007
; De Lannoy et al.
2007a
; Draper et al.
2012
) and
downscaled from the coarse scale to finer scales (Reichle et al.
2001a
; Pan et al.
2009
;
