Array out of bounds error in Thompson-aerosol aware scheme

[csmill]

In pull request #1834, the allocated middle dimensions of the lookup tables were changes from 9 to 1 for mp==8 and 28 in order to save memory when hail wasn't included (for example, the array called tmr_racg). However, On lines 2527 and onwards in module_mp_thompson.F onward, these arrays are accessed using a value of 5, which comes from idx_bg(k) which is set by idx_bg1 which is 5 when graupel is not included, leading to an out of bounds error.

This issue luckily seems not to crash the program when it is run normally, but as far as I can tell, it cannot be run in debug mode (configured with the -D option) without stopping.

Changing line 79 in module_mp_thompson.F to

      INTEGER, PARAMETER :: idx_bg1 = 1 ! index for rhog when mp=8 or 28

instead of = 5, prevents the model from crashing in debug mode.

It's fairly difficult for me to tell whether this "fix" has any unintended implications for the graupel-free versions of Thompson mp, aside from changing the value of rho_g.

I'm happy to give more details on my WRF config if needed.

[dudhia]

Thanks for reporting this. How about setting NRHG1=5 as a solution? I will cc @mefrediani on this.

[csmill]

Your suggestion works for me.

If you wanted to avoid increasing the memory usage, it might also be possible to allocate the tables so that they have only a dimension of 1 but so that the index is still 5 to prevent the memory overflow error. For example, something like this (not fully tested, but it runs without crashing with option 28):

!..Allocate space for lookup tables (J. Michalakes 2009Jun08).

      if (is_aerosol_aware) then
         ! Normal array with 9 columns starting at "1"
         NRHG_start = 1
         NRHG_end = NRHG ! = 9
      else
         ! Only use one column to reduce memory usage
         ! However, the column must have the address "5" to prevent out-of-bounds error
         NRHG_start = 5
         NRHG_end = 5
      endif



      if (.NOT. ALLOCATED(tcg_racg) ) then
         ALLOCATE(tcg_racg(ntb_g1,ntb_g,NRHG_start:NRHG_end,ntb_r1,ntb_r))
         micro_init = .TRUE.
      endif
                                                                    ! v-- change the allocate statement here
      if (.NOT. ALLOCATED(tmr_racg)) ALLOCATE(tmr_racg(ntb_g1,ntb_g,NRHG_start:NRHG_end,ntb_r1,ntb_r))
      if (.NOT. ALLOCATED(tcr_gacr)) ALLOCATE(tcr_gacr(ntb_g1,ntb_g,NRHG_start:NRHG_end,ntb_r1,ntb_r))
      ! if (.NOT. ALLOCATED(tmg_gacr)) ALLOCATE(tmg_gacr(ntb_g1,ntb_g,NRHG_start:NRHG_end,ntb_r1,ntb_r))
      if (.NOT. ALLOCATED(tnr_racg)) ALLOCATE(tnr_racg(ntb_g1,ntb_g,NRHG_start:NRHG_end,ntb_r1,ntb_r))
      if (.NOT. ALLOCATED(tnr_gacr)) ALLOCATE(tnr_gacr(ntb_g1,ntb_g,NRHG_start:NRHG_end,ntb_r1,ntb_r))

NRHG_start and NRHG_end are new integer variables.

I'm not sure how important the memory savings are versus the extra complexity and potential for confusion with using a non-1 starting index, but I thought I would add this suggestion just in case it's helpful.