Bug in ropp_fm refractivity calculation when called with both -new_op and -comp

[cburrows]

Stig Syndergaard noticed that the refractivity forward model resulted in background refractivity values being too large when both the -new_op and -comp options were used together. From his original email:

We are currently working on getting our software (mostly ROPP) ready for our reprocessing. One issue there is the generation of reference ECMWF model profiles using ropp_fm_bg2ro_1d. We were planning to run ropp_fm_bg2ro_1d with both -new_op and -comp options. But when comparing to our Metop NRT stats, where we use neither options for historical reasons, we found a larger refractivity bias than expected. It seems to be only a problem when both options are used together.

The attached plots shows results of comparing mean departures from ECMWF for one day. In all plots the red curve is the stats from a reprocessing experimental run (called exp.16) where we used -new_op and -comp to obtain the ECMWF profiles, and the black curve is the NRT stats where we used neither options. These curves are the same in all plots, and can be used as reference against the green curve. The green curve then shows different results when generating the ECMWF profiles for the NRT data using ropp_fm_bg2ro_1d with different options.

refrac_0:90_MET_1dayset_exp4.png: using ropp_fm_bg2ro_1d --no-ranchk -f -d -l You see a difference in bias above 10 km of about 0.1% between exp.16 and the NRT stats. Don't mind the large biases below 8 km in NRT, they are not the focus here.

refrac_0:90_MET_1dayset_exp4a.png: using ropp_fm_bg2ro_1d --no-ranchk -f -d -l -new_op Pretty much the same, indicating that new_op alone seems to work fine and doesn't do much. There are 137 ECMWF levels, so I think this is expected. Would you agree?

refrac_0:90_MET_1dayset_exp4b.png: using ropp_fm_bg2ro_1d --no-ranchk -f -d -l -comp Here the bias became about 0.05% larger at high altitudes, and smaller in the troposphere. At high altitudes (in thin air) I think this matches well with the use of kappa1_comp which is 77.643 as opposed to 77.6 for kappa. And -comp then does its thing in the more denser troposphere. We think this is as expected. Do you agree? (actually, while writing this I'm thinking that it should become smaller at high altitudes because I'm plotting O-B ... hmmm, I don't know enough about the expected effect of -comp to be sure).

refrac_0:90_MET_1dayset_exp4c.png: using ropp_fm_bg2ro_1d --no-ranchk -f -d -l -new_op -comp Here the bias in the NRT experiment became similar to the one in exp.16, which was also with -new_op -comp. So that is consistent. But it doesn't seem right given that new_op alone did almost nothing, and that -comp alone gave a bias to the other side (at high altitudes). Do you agree?

[cburrows]

refrac_090_MET_1dayset_exp4a.png

[cburrows]

refrac_090_MET_1dayset_exp4b.png

[cburrows]

refrac_090_MET_1dayset_exp4c.png

[cburrows]

refrac_090_MET_1dayset_exp4.png

[cburrows]

This bias was attributed to a bug in ropp_fm/refrac_1d/ropp_fm_refrac_1d_new.f90 (and the corresponding TL and AD) where the geopotential height computed with compressibility options was not being retained for the interpolation of the model variables onto the observation levels. The refractivity coefficient with compressibility was, however, being applied. If correctly set, these two adjustments cancel to a good extent, so if either is neglected, a bias will exist.

A patched subroutine was provided to Stig who verified that the fix worked as expected (see refrac_090_MET_1dayset_exp4d.png).

The corresponding TL and AD routines have also been adjusted similarly and the 'make tests' indicate that these are correct.

[cburrows]

refrac_090_MET_1dayset_exp4d.png

[Ian Culverwell]

Chris's changes (basically replacing x%geop by z_geop in the new refrac routines) test out OK with buildpack_all (inc sunf95 on $LOCALDATA and ifort on Cray). We also get errors in t_fascod_4 (FM FASCOD -comp -new_op) if we use the old ropp_fm/data/FASCOD_non_ideal_new_op.nc.

These changes have been committed to the ROPP90_prototype at r5143.

[Ian Culverwell]

fix_diff.png

[Ian Culverwell]

Impact of change on refracs calculated by ropp_fm_bg2ro_1d -comp -new_op on the profiles in ropp_fm/data/FASCOD_non_ideal_new_op.nc is:

0.1% is similar to what Stig saw.

[Ian Culverwell]

Corresponding changes have been made to trunk at r5144.

[cburrows]

Test folder ran successfully. Closing ticket.