decomp_cpools_vr_col is NaN

[samrabin]

Today on the software engineering standup, I mentioned a problem I was having in SoilBiogeochemPrecisionControl() where the run would crash with a floating-point error on the following line:

if (abs(cs%decomp_cpools_vr_col(c,j,k)) < ccrit) then

It was suggested that I check to see whether the referenced member of that array was NaN or Inf, and indeed it is NaN.

Bill noticed that the decomp_cpools_vr_col array isn't filled with spval, in SoilBiogeochemCarbonStateType → InitHistory(), for c12 as it is for c13 and c14. I tried the following modification:

diff --git a/src/soilbiogeochem/SoilBiogeochemCarbonStateType.F90 b/src/soilbiogeochem/SoilBiogeochemCarbonStateType.F90
index 7ef4265a4..0e5147a74 100644
--- a/src/soilbiogeochem/SoilBiogeochemCarbonStateType.F90
+++ b/src/soilbiogeochem/SoilBiogeochemCarbonStateType.F90
@@ -158,6 +158,7 @@ contains
        end if

        this%decomp_cpools_col(begc:endc,:) = spval
+       this%decomp_cpools_vr_col(begc:endc,:,:) = spval
        do l  = 1, ndecomp_pools
           if ( nlevdecomp_full > 1 ) then
              data2dptr => this%decomp_cpools_vr_col(:,1:nlevsoi,l)

but unfortunately the same crash happens.

My next course of action is going to be combing the code for all places where decomp_cpools_vr_col is modified, adding an isnan() check after each. Unless, that is, anyone has any ideas? Thanks in advance.

(Also: The array initialization idea I mentioned didn't pan out either.)

 

[samrabin]

I'm not sure exactly of the causal chain here, but I solved this by setting a minimum gddmaturity (heat unit index required for crop maturity, growing degree-days) to 1.0. Note that it only occurred in my experimental branch having to do with crop calendars, so this minimum only applies for runs that don't use the default CLM crop calendars/cultivars.