402
MICHAEL J. BELL ET AL.
(1997) and ice thickness distribution scheme of Lipscomb et al. (2001) are
being trialled.
2.4
Assimilation methods
Data assimilation is based on a new version of the analysis correction
(a/c) scheme. The a/c scheme was originally devised by Lorenc et al. (1991)
assimilation. Analysis steps are performed once per day. Each observation
makes its full impact on the model on the day it arrives and on subsequent
days is taken into account by giving additional weight to the model at the
observation’s location. Each analysis step consists of a number of iterations.
On each iteration the observations are separated into groups which are easily
related (thermal profiles, saline profiles, surface temperature, surface
height). For each group of observations (e.g. the temperature profile data),
increments are calculated first for the directly related model variables (e.g.
the temperature fields). These increment fields are then used to calculate
increments for less directly related model variables (e.g. the velocity fields)
using hydrostatic and geostrophic balance relationships, water property
conservation or statistical relationships. These balancing increments make
the analysis multivariate. Increments are also made to the observations
(Bratseth 1986) so that the iterations converge towards the statistically
optimal analysis. The univariate components of the model error covariance
are specified as the sum of two 3D error covariances, one describing the
ocean mesoscale, the other large scales including atmospheric synoptic
scales (Martin et al. 2002). These and the observation error covariances are
estimated from statistics of observation minus model values obtained from
hindcast assimilations. Altimeter data are assimilated by displacement of
dynamical balance near the equator (see section 3.1) and analyses performed
with large correlation scales are used to attempt to remove large-scale biases
in the AVHRR surface temperature data.
3.
Trouble-shooting, assessments of impact and
developments
3.1
Trouble-shooting
Bell et al. (2004) report a serious problem encountered assimilating
thermal profile data into the global FOAM configuration in the equatorial
Pacific region where the TAO moorings provide good observational
and implemented for FOAM by Bell et al. (2000a). The new version (Bell
et al. 2003) provides a sub-optimal approximation to a variant of 4D variational
isopycnal surfaces (an extension of the Cooper & Haines 1996 scheme).
A pressure correction technique (Bell et al. 2004) is employed to improve the
MICHAEL J. BELL ET AL.
(1997) and ice thickness distribution scheme of Lipscomb et al. (2001) are
being trialled.
2.4
Assimilation methods
Data assimilation is based on a new version of the analysis correction
(a/c) scheme. The a/c scheme was originally devised by Lorenc et al. (1991)
assimilation. Analysis steps are performed once per day. Each observation
makes its full impact on the model on the day it arrives and on subsequent
days is taken into account by giving additional weight to the model at the
observation’s location. Each analysis step consists of a number of iterations.
On each iteration the observations are separated into groups which are easily
related (thermal profiles, saline profiles, surface temperature, surface
height). For each group of observations (e.g. the temperature profile data),
increments are calculated first for the directly related model variables (e.g.
the temperature fields). These increment fields are then used to calculate
increments for less directly related model variables (e.g. the velocity fields)
using hydrostatic and geostrophic balance relationships, water property
conservation or statistical relationships. These balancing increments make
the analysis multivariate. Increments are also made to the observations
(Bratseth 1986) so that the iterations converge towards the statistically
optimal analysis. The univariate components of the model error covariance
are specified as the sum of two 3D error covariances, one describing the
ocean mesoscale, the other large scales including atmospheric synoptic
scales (Martin et al. 2002). These and the observation error covariances are
estimated from statistics of observation minus model values obtained from
hindcast assimilations. Altimeter data are assimilated by displacement of
dynamical balance near the equator (see section 3.1) and analyses performed
with large correlation scales are used to attempt to remove large-scale biases
in the AVHRR surface temperature data.
3.
Trouble-shooting, assessments of impact and
developments
3.1
Trouble-shooting
Bell et al. (2004) report a serious problem encountered assimilating
thermal profile data into the global FOAM configuration in the equatorial
Pacific region where the TAO moorings provide good observational
and implemented for FOAM by Bell et al. (2000a). The new version (Bell
et al. 2003) provides a sub-optimal approximation to a variant of 4D variational
isopycnal surfaces (an extension of the Cooper & Haines 1996 scheme).
A pressure correction technique (Bell et al. 2004) is employed to improve the