Geoscience Reference
In-Depth Information
The major drawback of the above techniques is the underlying assumption that the
model states have a Gaussian distribution. The PF does not require a specific form for the
state distribution, but its major drawback is that distribution of particle weights quickly
becomes skewed, and a re-sampling algorithm needs to be applied.
The EnKF and PF are complementary. This complementarity makes a hybrid EnKF/PF
version highly attractive for systems that can exhibit nonlinear and non-Gaussian features,
an example being the land surface. For example, the EnKF could be used as an efficient
sampling tool to create an ensemble of particles with optimal characteristics with respect to
observations. The PF methodology could then be applied on that ensemble afterwards to
resolve nonlinearity and non-Gaussianity in the system. This method is getting increased
attention (see, e.g., Kotecha and Djuri´
2003
).
4.5 Example of a Land Data Assimilation System
For illustrative purposes, we describe the elements of the NILU SURFEX-EnKF land data
assimilation system (Lahoz et al.
2010b
). These elements are the following: (1) a data
assimilation scheme (mainly variants of the EnKF, but also variants of the PF, and the
EKF); (2) a land surface model (SURFEX model developed at M´t´o-France, Le Moigne
2009
); (3) observations; (4) the observation operator; and (5) error characteristics for the
model and the observations.
The SURFEX model used at NILU (and at M
´
t
´
o-France) can be run in uncoupled or
coupled mode. It includes the following elements:
• A soil and vegetation scheme: ISBA and ISBA-A-gs;
• A water surface scheme: COARE/ECUME (Coupled Ocean-Atmosphere Response
Experiment/Exchange Coefficients from Unified Multi-campaign Estimates) for the
sea; FLAKE for inland water;
• Urban and artificial areas: Town Energy Balance—TEB model;
• A surface boundary layer (SBL) scheme;
• Chemistry and aerosols;
• A land use database: ECOCLIMAP.
Figure
3
illustrates how SURFEX works. During a model time step, each surface grid
box receives from the atmosphere the following information: upper air temperature, spe-
cific humidity, horizontal wind components, pressure, total precipitation, long-wave
radiation, short-wave direct and diffuse radiation and, possibly, concentrations of chemical
species and dust. In return, SURFEX computes averaged fluxes of momentum, sensible and
latent heat, and, possibly, chemical species and dust fluxes. These fluxes are then sent back
to the atmosphere with the addition of radiative terms like surface temperature, surface
direct and diffuse albedo, and surface emissivity.
The above information transferred to the atmosphere from the land surface provides the
lower boundary conditions for the radiation and turbulent schemes in an atmospheric
model coupled to SURFEX or forced by SURFEX output. In SURFEX, each grid box is
made up of four adjacent surfaces: one for nature, one for urban areas, one for sea or ocean
and one for lake, identified by the global ECOCLIMAP land database. The SURFEX fluxes
are the average of the fluxes computed over nature, town, sea/ocean or lake, weighted by
their respective fraction.
The assimilation system at NILU is illustrated in Fig.
4
with reference to the EnKF. It
can assimilate the following data: (1) 2-m screen-level temperature (T
2m
) and 2-m screen-
level
relative
humidity
(RH
2m
)
provided,
for
example,
by
the
SYNOP/CANARI
