This is a short reference of possible causes for the kf-eta-related errors:

WOULD GO OFF TOP: KF_ETA_PARA ...

There are many suggested solutions for these but I always manage to find some new way to produce these crashes.

Commonly suggested causes/solutions

I’ve found these solutions listed on various web-pages

  • Move domain boundaries (try to avoid steep topography)
  • Decrease time-step (if this is an isolated event in a longer run, try to increase the time-step again after the next restart interval)
  • Turn on w_damping (tries to reduce vertical winds)
  • Change PBL-scheme (I’ve never tried this)

Less commonly suggested causes/solutions

  • Change microphysics (I’ve never tried this)
  • Turn off w_damping (?!) might work in some cases
  • Disable optimizations at compile-time (set FCOPTIM=O1 or = O0 in configure.wrf)
  • Buggy metgrid tables (check met_em*-files for possible errors)

Causes/solutions that worked for me

  • Bad input data: pressure. E.g. when forcing data is on sigma levels but METGRID.EXE didn’t read the PRES*-files(!). Then met_em*-files will not have the correct pressure data. At the lowest model layer everything looks normal but all layers above that simply have the inverse level number as pressure; this causes all kinds of weird behaviour in WRF, especially for the vertical wind.
  • Two-way nesting Can cause weird feedback patterns near domain boundaries. I’ve usually found the cause of the problem to be something else but if there are no other obvious causes, it could be worth testing to spin-up the inner domains before turning on the 2-way nesting…

Finding the root of the problem

  • Look for unrealistic values the input data in e.g. ncview.
  • Decrease the output time-step to about a minute and see if anything strange happens with before the crash.
  • Does the error always occur in the same point? If so, look at input/output values in that point

Further reading