solar_irradiance_data: itsi may be incorrectly declared as integer

[jimmielin]

This issue was flagged by claude-opus:4.8[1m]: it noticed that the following variable is declared as integer and may be silently truncated as a result; it should be declared as real(kind_phys):

atmospheric_physics/schemes/radiation_utils/solar_irradiance_data.F90

Line 324 in 2180f3c

integer, allocatable :: itsi(:)

...

atmospheric_physics/schemes/radiation_utils/solar_irradiance_data.F90

Lines 366 to 372 in 2180f3c

	if (has_tsi .and. (.not. do_spectral_scaling)) then
	call file_reader%get_var('tsi', itsi, errmsg, errflg, (/index/), (/nt/))
	if (errflg /= 0) then
	errmsg = subname // errmsg
	return
	end if
	if ( any(itsi(:nt) < 0._kind_phys) ) then

the above line 372 shows that it might have been intended to be a real.
...

atmospheric_physics/schemes/radiation_utils/solar_irradiance_data.F90

Lines 397 to 398 in 2180f3c

	if (has_tsi .and. (.not.do_spectral_scaling)) then
	sol_tsi = itsi(1) + delt*( itsi(2) - itsi(1) )

The same variable with similar logic appears to be declared as real(r8) in CAM:
https://github.com/ESCOMP/CAM/blob/ec42689850c06f0d4472e826b30ad69728ba42cf/src/chemistry/utils/solar_irrad_data.F90#L41

  real(r8) :: itsi(2)
...
       if (has_tsi .and. (.not.do_spctrl_scaling)) then
          ierr = pio_get_var( file_id, tsi_vid, (/index/), (/nt/), itsi )
...

    if (has_tsi .and. (.not.do_spctrl_scaling)) then
       sol_tsi = itsi(1) + delt*( itsi(2) - itsi(1) )
    endif

Tagging @peverwhee

The original message written by Claude was:

Problem: itsi is declared integer, allocatable :: itsi(:) but holds total solar irradiance (tsi, ~1361 W m⁻²). The generic get_var binds to the integer overload (get_var_int_1d) and truncates the file's real TSI to whole numbers. The interpolation sol_tsi = itsi(1) + delt*(itsi(2)-itsi(1)) is then performed in integer arithmetic, and the validity check any(itsi(:nt) < 0._kind_phys) compares integers to a real literal. This degrades the prescribed solar constant to ~1 W/m² quantization and discards sub-W/m² solar-cycle variability — the exact signal this data path exists to provide.

[adamrher]

Well that's concerning. I think the global mean incoming solar is the solar constant divided by 4, so at most this would introduce a 0.25 W/m2 error to the TOA budget.

[jimmielin]

I think this module is just used in CAM-SIMA so it should be fine (for CAM) and the actual impact is very limited since CAM-SIMA RRTMGP runs with no differences against the CAM snapshot. Maybe the snapshot run does not 'exercise' itsi as much? Not sure why it wouldn't show up as answer differences when making the CCPP conversion (is itsi read from snapshot?)