FM shows NaN values in Bending Angle

[frae]

One of the ECMWF input files to FM leads to a NaN value in the bending angle results.

I'll generate the ticket and then attach the netcdf input file to ecmwf2ro_1d.

[frae]

File is too large for attachement, TRAC complains. Find it at:

​ftp://sat.physik.uni-bremen.de/incoming/axel/nanprofile.nc

[marq]

Could you please chack if this is still a problem, or if it has been fixed in the meantime?

Thanks!

[frae]

ECMWF Profile leading to NaN in Bending Angle

[frae]

Profile leading to NaN in BA

[frae]

Okay, just checked the relevant profile again and it still leads to NaN in the bending angle. It is a particular nasty profile as you can see from the plot attached (actually 2 identical plots because svn.marquardt.sc showed trac errors when first uploading the file, so now you got 2 of them...).

[Huw Lewis]

On testing ROPP v1.1 using NAG complier, we came across a problem with Abel transform attempting to compute square root (x(i+1) - a) where a > x(i+1). This is thought to be because the testing for the lowest usable background level was not exiting once the first instance of increasing x was found (working from the top of the profile). Tests show that other compilers would produce bending angle = NAN in this case, consistent with the initial error.

​https://trac.grassaf.org/ropp/browser/ropp_src/trunk/ropp_fm/bangle_1d/ropp_fm_abel.f90 has now been fixed to identify the lowest usable background level as the first instance (countine from model top) where x is not monotonically increasing with height.

[(none)]

Milestone 1.1 deleted