In order to best interpret observations gathered
during this effort, we are
developing an ocean model which can put
the observations into a dynamical
context. The model must be made capable of reproducing
the observations, and the physics must be accurate enough to not
require ad-hoc forcing in compensation.
Models
frequently
use unphysical forcing to bring the model into agreement with the data
at each timestep, so the time series of model state is not consistent
with the model physics and realistic forcing.
Our hypothesis is that our model will be
good enough so that the model
output can match the observations, except for the errors in the
forcing, initial conditions and boundary conditions (F-IC-BC,
hereinafter) which are
themselves the result of limited and noisy data.
If we can bring the model into agreement with the data by adjusting
these parameters within their error bars, we have obtained a
dynamically consistent model evolution, which can be used to analyze
the physics controlling the annual and interannual variability.
The procedure of fitting the model to the data by adjusting these
parameters is also a rigorous way of testing the model skill. A model
with poor physics should be unable to match an extensive dataset,
although achieving sufficient data density to disprove the model may
be difficult.
Mean and annual cycle modeling:
We have configured the OPYC model (Oberhuber, 1993, JPO; Miller et al., 1994,
Clim Dyn)
to the Pacific
Ocean region at nominally 1.5 degree resolution (with telescoping north-south
enhancement to 0.67 degrees at the equator) and with
ten isopycnal layers plus the bulk surface mixed layer.
We have tested the model seasonal cycle by forcing
it with climatological monthly mean surface forcing
of wind stress, heat fluxes and fresh-water fluxes.
Initial inspection of the model output revealed
the model does a reasonable job of representing the upper-ocean
structures of currents, sea level,
temperature and salinity throughout the Pacific.
For example, Figure 1 shows the good agreement between
model sea level annual harmonic
along with the observed harmonic from Topex.
However, we also recognize the imperfections
of the present seasonal cycle simulation.
For example, we noted
a problem with the mixed layers being too deep in the
subtropical gyre, a consequence of Ekman downwelling
acting on the bulk mixed layer during the deepening (fall/winter)
seasons. Discussions with J. Oberhuber resulted in
his designing a new mixing scheme that reduces this
deep mixed layer effect. The Ekman
downwelling also may be too large there which
may be alleviated by perturbed wind stress curl forcing.
An additional problem with unphysically large
vertically integrated north-south volume
transports is presently being addressed by evaluating
the mass budgets in each layer.
Various sensitivity tests of the model mixed-layer parameters
and mixing parameters on
the model mixed layer structure and current fields have been
executed to build insight into the model response characteristics.
Several long runs forced by anomalous forcing derived from
NCEP reanalyses (heat, momentum and fresh-water fluxes)
from 1958-1997
have been executed and are presently under analysis.
Preliminary results indicate that midlatitude forcing
functions derived from NCEP yield somewhat large model SST anomalies,
while tropical NCEP forcing is too weak during warm or cold events
to produce realistic tropical SST anomalies.
Cayan has recently completed a surface flux dataset
derived from COADS which we expect to give better results
in data-intensive regions.
Subsurface data provided by White will provide additional
verification.
Validation dataset:
We have completed interpolation machinery to convert model output to
to a format suitable for comparing with
observations that we wish to fit, including Topex altimetry, XBT
sections, float velocities and profiles, drifters velocities, and
archived hydrography. Using this mapping, we have
compared the open-loop (without optimization) model run output with
our datasets to evaluate the initial quality of the model,
and to search for problems which must be solved before the model can
be used to fit the observations. Preliminary comparisons with the
high-resolution XBT lines show encouraging levels of skill,
although significant
differences exist due to the lack of mesoscale variability in the
model.
Inverse method:
We have begun to test the inverse method (Bennett, 1992, Cambridge
University Press) by
running some sensitivity experiments.
The first set of sensitivity experiments involves
testing for non-linearity of the response
by adding a small-amplitude, large-scale
constant perturbation to one component of the forcing
(e.g., zonal wind stress) and comparing a multi-year
integration to one with an equal but oppositely signed perturbation.
Except in the tropical region where instability waves
are prevalent, the model response is largely linear in wind stress,
heat flux anf TKE input.
We are also using these multi-year runs to study the
spin-up times of the model.
We have laid the groundwork for the multiple sensitivity runs of the
model with a reduced state space of perturbations to the
climatological forcing. Preliminary inverses are now
being computed.
