To determine the value of the adjustable parameters of an ocean model
required to optimally fit the observations, an adaptive inverse method
is developed and applied to a sea surface temperature (SST) model of
the tropical Atlantic. The best-fit calculation is performed by minimi
zing a misfit between observed and simulated data, which depends on th
e observational and the modeling errors. An adaptive procedure is desi
gned in which the model being tuned is also used to construct a model
of the observational errors. This is done by performing the optimizati
on on the mean seasonal cycle and using the SST anomalies obtained for
different years and plausible forcing fields as additional informatio
n to construct a sample estimate of the observational error covariance
matrix. Assuming idealized modeling errors, the procedure is applied
to the SST model of Blumenthal and Cane, yielding refined estimates fo
r several models and heat flux parameters. The simulation of the mean
annual SST is improved, but not the simulation of seasonal and interan
nual variability. The model-observation discrepancies remain too large
to be solely attributed to atmospheric and oceanic data uncertainties
and are linked to the model's rudimentary geometry and its incorrect
representation of SST cooling by upwelling. The existence of larger mo
del deficiencies than was originally assumed in the model errors is co
nfirmed by a statistical test of the correctness of the assumptions in
the inverse calculation.