The observation error of satellite sensors

[weiwilliam]

For satellite sensors, GSI reads the satinfo file to assign the clear-sky variance (varch) for each channel of available sensors.
Recently, I found one thing that confused me in setuprad.f90.

The tnoise was assigned with varch twice at lines 556 and 844 of setuprad.f90

GSI/src/gsi/setuprad.f90

Line 556 in cf80b5c

tnoise(jc)=varch(j)

GSI/src/gsi/setuprad.f90

Line 844 in cf80b5c

tnoise(jc)=varch(j)

Later, tnoise is assigned to error0 at lines 1288 before the observation error inflation for all-sky.

GSI/src/gsi/setuprad.f90

Line 1288 in cf80b5c

error0(i) = tnoise(i)

Then, varinv is derived at lines 1305 with squared error0

GSI/src/gsi/setuprad.f90

Lines 1300 to 1307 in cf80b5c

	do i=1,nchanl
	mm=ich(i)
	channel_passive=iuse_rad(ich(i))==-1 .or. iuse_rad(ich(i))==0
	if(tnoise(i) < 1.e4_r_kind .or. (channel_passive .and. rad_diagsave) &
	.or. (passive_bc .and. channel_passive))then
	varinv(i) = varinv(i)/error0(i)**2
	errf(i) = error0(i)
	else

And the output in diagnostic files for "Inverse_Observation_Error" is derived at lines 2579.

GSI/src/gsi/setuprad.f90

Lines 2579 to 2580 in cf80b5c

	errinv = sqrt(varinv0(ich_diag(i)))
	call nc_diag_metadata("Inverse_Observation_Error", sngl(errinv) )

My confusion is here, if the varch is variance (defined in code and user guide), why it is squared to calculate inverse variance (varinv) at lines 1305?

If the "error" column in satinfo is standard deviation, then everything is good.
If that is variance, it means the observation error is squared twice for clear-sky radiances (no matter all-sky or not).
Hope if someone can help to clarify this, maybe some legacy changes I'm not aware of.
Thanks.

[weiwilliam]

I got answer from elsewhere.
It's standard deviation, which means everything is good now.
Thanks.