blowup with PE-NaN

edit

I am running a simulation which blows up at a certain number of timesteps (after ~6 months of simulation) after the first initiation.

The logfile show PE and TotalE as NaN. Nothing wrong with KE. Saving the output 1 step before the problem show nothing wrong or abnormal values for zeta,ubar,vbar etc.

So far I have tried: -increasing/decreasing barotropic/baroclinic timesteps -different forcing dataset -different BC dataset - define/undef QCORRECTION The blowup is in the exact same time-step

Any ideas what could cause this problem?