Sigma coordinate models currently suffer from the difficulty encountered in
specifying the horizontal pressure gradient. Most of me efforts to reduce
this truncation error lead to the loss of at least one integral constraint
in the pressure gradient scheme (energy, tracer balance, etc.). Starting fr
om a usual Arakawa-based scheme, it is shown that judicious adjustment of t
he sigma surfaces together with control of the bottom topography slope para
meter and an "optimal modification" of the measured initial pressure held l
ead to a drastic reduction in the truncation errors. The first experiment i
s based on a coastal academic domain in order to compare the results of the
proposed inverse method with published data. A few experiments are then co
nducted on the eastern pan of the Gulf of Lions in the Mediterranean using
data from the Suiviolion experiment.