122 Keith Haines
(v) It should alIow an independent assimilation scheme for temperature profile
data (presumably available much more infrequently) to have a lasting impact on
the ocean model.
Other altimeter assimilation methods which share some of these properties
include, Oschlies and Willebrand (1996), Drakopoulos (1997) and Gavart and De
Mey (1997). In the section which folIows we show some results from applying this
scheme in a twin experiment formulation.
7.3 Convergence in twin-experiment assimilation experiments
A number of twin experiment papers have appeared in the literature showing
results from using the scheme described above, Cooper and Haines (1996), Drakopoulos (1997), Fox and Haines (1996), Fox et al. (2000). We will show some
results from using a Global high resolution ocean model, OCCAM (Ocean Circulation and Climate Advanced Model), which is a free surface model with l/4° horizontal resolution and 36 vertical levels. Performing assimilation with this model
takes the assimilation scheme into new areas such as the tropics and the weakly
stratified southem oceans where it has not been tested before. The OCCAM model
was spun up for eight years with monthly varying SST and SSS relaxation and
ECMWF wind stresses, and the ninth year was used as the 'true' circulation, and
the fields from the start of year 10 were used to initialise the assimilation run.
Assimilation of sea surf ace height data was performed at 15 day intervals, with the
first data being taken from day 1, year 9 ofthe truth run. This corresponds to model
January Ist. Fields in the model were altered according to equation (2) and velocities were updated geostrophicalIy, except within 3° ofthe equator where no velocity update was performed. It should be noted that the full height change fie1d is
assimilated, corresponding to the hypothetical case of perfect data, with known
Geoid and zero observational error.
To track the convergence of the assimilation run towards the truth run, RMS
errors (differences between values in the assimilation run and the model truth) of
the density, temperature, salinity and velocity fields were calculated for various
regions and depth ranges. Fig. 7.3 shows globally averaged RMS errors in alI these
quantities for both the upper water column (top 1000m) and below. Fig. 7.4 shows
the same quantities for the Tropical Pacific only and Fig.7.5 is for the Antarctic
Circumpolar current region. Errors in sea surface height before each assimilation
step are reduced by more than 50 % after one assimilation and the folIowing 15
days of model run, then continue to drop with maximum reductions of 80 % (and
still increasing) achieved after 10 assimilation steps. Velocity errors undergo the
largest reduction, particularly in the near-surface region. This is due to near-surface
currents being strongly tied to the sea surf ace height gradients via geostrophy. The
deeper currents are improved less in percentage terms by assimilation but interestingly convergence often continues during model runs. The improvement of deep
sub-thermocline currents is an almost unique feature of this assimilation method,
(v) It should alIow an independent assimilation scheme for temperature profile
data (presumably available much more infrequently) to have a lasting impact on
the ocean model.
Other altimeter assimilation methods which share some of these properties
include, Oschlies and Willebrand (1996), Drakopoulos (1997) and Gavart and De
Mey (1997). In the section which folIows we show some results from applying this
scheme in a twin experiment formulation.
7.3 Convergence in twin-experiment assimilation experiments
A number of twin experiment papers have appeared in the literature showing
results from using the scheme described above, Cooper and Haines (1996), Drakopoulos (1997), Fox and Haines (1996), Fox et al. (2000). We will show some
results from using a Global high resolution ocean model, OCCAM (Ocean Circulation and Climate Advanced Model), which is a free surface model with l/4° horizontal resolution and 36 vertical levels. Performing assimilation with this model
takes the assimilation scheme into new areas such as the tropics and the weakly
stratified southem oceans where it has not been tested before. The OCCAM model
was spun up for eight years with monthly varying SST and SSS relaxation and
ECMWF wind stresses, and the ninth year was used as the 'true' circulation, and
the fields from the start of year 10 were used to initialise the assimilation run.
Assimilation of sea surf ace height data was performed at 15 day intervals, with the
first data being taken from day 1, year 9 ofthe truth run. This corresponds to model
January Ist. Fields in the model were altered according to equation (2) and velocities were updated geostrophicalIy, except within 3° ofthe equator where no velocity update was performed. It should be noted that the full height change fie1d is
assimilated, corresponding to the hypothetical case of perfect data, with known
Geoid and zero observational error.
To track the convergence of the assimilation run towards the truth run, RMS
errors (differences between values in the assimilation run and the model truth) of
the density, temperature, salinity and velocity fields were calculated for various
regions and depth ranges. Fig. 7.3 shows globally averaged RMS errors in alI these
quantities for both the upper water column (top 1000m) and below. Fig. 7.4 shows
the same quantities for the Tropical Pacific only and Fig.7.5 is for the Antarctic
Circumpolar current region. Errors in sea surface height before each assimilation
step are reduced by more than 50 % after one assimilation and the folIowing 15
days of model run, then continue to drop with maximum reductions of 80 % (and
still increasing) achieved after 10 assimilation steps. Velocity errors undergo the
largest reduction, particularly in the near-surface region. This is due to near-surface
currents being strongly tied to the sea surf ace height gradients via geostrophy. The
deeper currents are improved less in percentage terms by assimilation but interestingly convergence often continues during model runs. The improvement of deep
sub-thermocline currents is an almost unique feature of this assimilation method,