This post takes a look at the default behavior of ICAR with regards to the calculation of the Brunt-Väisälä frequency 
and discusses an alternative method that is arguably better suited from a theoretical point of view. Note that the data and results here are not peer reviewed. However we have a manuscript in preparation that discusses aspects of this. I’ll add the corresponding links once they are available.
Linear interpolation as vertical extrapolation method in ICAR
So far when ICAR used a forcing dataset where the z-levels did not vary with time it did extrapolate some quantities downward from the lowest forcing level by keeping them constant. This could potentially lead to unwanted behavior as I’m going to show with an example in the following. I’ll refer to the ICAR version before the code modification as “ICAR old” and the version after the modification as “ICAR new”.