No one objected to this at the ROPP-GG meeting on 20th June 2012. Need to update documentation accordingly.

Prepared mods for ROPP6.1. ropp_fm_bg2ro_1d, ropp_fm_bg2ro_2d, ropp_1dvar_refrac and ropp_1dvar_bangle now have a -check_qsat switch to switch on the check that q is not saturated, whether or not -use_logq is in effect. (Previously, there was no option to call check_qsat unless one was using --use_logq.) By default, no such check is made, ie supersaturation is allowed.

Details

Key codes: ropp_fm/model_ecmwf/ropp_fm_state2state_ecmwf.f90 and ropp_fm/model_meto/ropp_fm_state2state_meto.f90. Replace

  IF (x%use_logq) THEN
    x%shum = EXP(x%state(x%n_lev + 1: 2*x%n_lev)) / 1000.0_wp
    CALL check_qsat(x%shum, x%temp, x%pres)
  ELSE
    x%shum = x%state(x%n_lev + 1: 2*x%n_lev)
    WHERE (x%shum <= 0.0_wp)
      x%shum = 1.0e-9_wp
    ENDWHERE
  ENDIF

by

  IF (x%use_logq) THEN
    x%shum = EXP(x%state(x%n_lev + 1: 2*x%n_lev)) / 1000.0_wp
    IF (x%check_qsat) CALL check_qsat(x%shum, x%temp, x%pres)
  ELSE
    x%shum = x%state(x%n_lev + 1: 2*x%n_lev)
    IF (x%check_qsat) CALL check_qsat(x%shum, x%temp, x%pres)
    WHERE (x%shum <= 0.0_wp)
      x%shum = 1.0e-9_wp
    ENDWHERE
  ENDIF

I haven't bothered updating the TL or AD codes, because they didn't involve any calls to check_qsat anyway. This is probably wrong.

Results

ropp_fm

ropp6.0 -use_logq = ropp6.1 -use_logq -check_qsat to within rounding error of +/-1e-15.

ropp6.0 (no options) = ropp6.1 (no options) exactly. (Why the difference? -use_logq implies use of EXP(ALOG(q)), and perhaps these are treated slightly differently at ropp6.0 and 6.1?)

Impact of -check_qsat at 6.1: dq ~ 1e-3 g/kg at upper levels, whether or not -use_logq is on. Reasonable.

ropp_1dvar_refrac

use_logq=.false. by default. +/- check_qsat typically gives ~0.01K in temp; 0.10 g/kg in shum. Concentrated lower down? For some profiles it's exactly zero, which makes sense. Reasonable. (The cntl, ie without -check_qsat, seems to take more iterations in general.) use_logq=.true. ==> more similar convergence histories, but rather larger diffs in T and q.

ropp_1dvar_bangle use_logq=.false. by default. dshum ~ 0.01 g/kg (concentrated towards bottom); dtemp ~0.1K (all levs). Two profiles untouched. use_logq=.true. ==> rather larger changes in shum and temp, but still OK.

All this with minropp. Not bothering with LevMarq.

Asked Joe for some test data on 14/12/2013:

Hi Joe,

I've coded a fix for this.  Could you give me an example data set where the unmodified code causes trouble, so I can test it please?

Thanks,
Ian.

Joe replied on 11/1/2013:

Hi Ian
I have not forgot this ticket. Sorry for not responding. The thing is that I tried to
generate some examples in December. But I got confused because it turned out that the
ECMWF forecasted  profiles were cut of at RH = 100%. Previously I had some examples
with RH > 100 % in the ECMWF forecast. I need some time to sort it out.
Could it wait until February? I am on a field campaign (exceptionally) from 17-31 Jan.
joe

Didn't really have time to postpone ROPP6.1 any more, so closing ticket. Joe can reopen at ROPP7.0 if he's unhappy with the above solution.