Bending angle 1DVar fails with ducting profiles

[frae]

The bending angle 1DVar does a check on the provided sigmas in ropp_fm/common/ropp_fm_roprof2obs.f90:155

  y%cov_ok = .true.

  if (associated(y%cov%d)) deallocate(y%cov%d)  ;  call callocate(y%cov%d, n*(n+1)/2)

  do i = 1, n
     if (ro_data%Lev1b%bangle_sigma(i) > 0.0_wp) then
        y%cov%d(i + i*(i-1)/2) = ro_data%Lev1b%bangle_sigma(i)**2 ! pp, uplo = 'U'
     else
        y%cov_ok = .false.
     end if
  end do

With ducting profiles neither the bending angle nor the sigma is defined, thus cov_ok is set to false here and the 1DVar exists with an error. Alhough this does not only apply to ducting; I guess any multifile where not all profiles cover all levels will cause trouble here.

[Huw Lewis]

To be investigated with v1.2 testing.

[Huw Lewis]

I think this requires a user to ensure all bangle_sigma values are 'valid'. e.g. in file IT-1DVAR-02 there are invalid bending angle values, but the corresponding bangle_sigma has the same value as that used for the first valid data point. The cov_ok flag is used to decide whether it is suitable to construct the error covariance matrix. The impact of missing bangle data is addressed using the weights in computing the cost function.

There is no neat coding solution to solve this issue. The User Guide for v1.2 specifies that valid sigma values are required for all observation levels if using one of the 'variable sigma' covariance methods. See [1565]. Ticket closed.