BICONJ - Balance operator, error in reference free surface

[uranoscopus]

of Hi everybody,
When trying to run I4DVAR for an area of Western Med (with an Island in the middle of the domain),assimilating just ADT-AVISO (corrected in respect of the bias with model by subtacting and adding respectively their averages and model averages for zeta), I have the following symptoms:

-In log appears the following error message:
BICONJ - Balance operator, error in reference free surface: -1.45E-01
That does'nt crash the model, but is at least suspect!

-I have to lower time step under 60 secs (3 km resolution grid) to avoid blowing up after few hundreds of steps of the starting forward-nonlinear run. The nonlinear model with no DA was stable and runned 1 year with no problems.

-Even with a timestep of 45 secs the model blows up after lets say 5000 steps. What can be seen in log is crazy maximum velocities just before crashing, so probably the blows up is due to strange slopes in zeta determining such crazy velocities....and this remind me the first point. i.e. the BICONJ error.

Can anyone help me to get some light on this?
Where the reference surface is set?
Thanks in advance,
Antonio

[uranoscopus]

Update:
I was able to reduce such value of error in BICONJ by increasing number of iterations of Nbico in s4dvar.in but it seems not affect the result, so maybe it is not the responsible.
I have to explore in some other direction.
What is strange to me is that the nonlinear is stable, while the first run of the I4DVAR (that I think is a forward nonlinear run,...I'm correct?) promptly gives high unreal velocities, even if with extremely small DT i'm able to avoid the blowup.
So....



Last Update: ok I found that is the nonlinear to be unstable. I dunno why as I preserved the same configuration, I just changed roms version (adopting new *.in template) and I'm running in parallel while till now I runned serial.