Cloud ceiling failure in in-line post

[WenMeng-NOAA]

Cloud ceiling failure in in-line post

[WenMeng-NOAA]

Per Fanglin's suggestion, a new total cloud fraction is initialized at individual grid instead of a box of
grids as:
DO L=LM,1,-1
DO J=JSTA,JEND
DO I=1,IM
FRAC =CFR(I,J,L) !- 3D cloud fraction (from radiation)
PCLDBASE=PMID(I,J,L)
IF (PCLDBASE>=PTOP_LOW) THEN
CFRACL(I,J)=MAX(CFRACL(I,J),FRAC)
ELSE IF (PCLDBASE>=PTOP_MID) THEN
CFRACM(I,J)=MAX(CFRACM(I,J),FRAC)
ELSE
CFRACH(I,J)=MAX(CFRACH(I,J),FRAC)
ENDIF
TCLD(I,J)=MAX(TCLD(I,J),FRAC)
ENDDO ! I
ENDDO ! J
ENDDO ! L

This algothrims will avoid calling AllGETHERV.

[WenMeng-NOAA]

Some explanation from from Fanglin:
Brad Ferrier developed a scheme to compute instantaneous Low/Mid/High cloud fraction in the POST for NAM. Because NAM only has binary cloud fraction (1 vs 0) at each model grid, he used the average over a few model grids over a 14 by 14-km box to compute "fractional" cloud cover.

Wen applied this method to compute L/M/H clouds for GFS.v16. It worked in offline post; however, it caused the model to hang in inline post. After long hours of debugging, Wen found an MPI gathering call in this section of code caused the model to hang. The computation needs points cross the boundary within each task.

Unlike NAM, GFS produces true fractional cloud. We decided to get rid of this complicated approach, and instead use GFS predicted 3-D cloud fraction to derive L/M/H clouds. This "NAM" approach is ad hoc in nature since the corresponding L/M/H cloud-base pressures are nevertheless not areal averages ( PCLDBASE=PMID(I,J,L) !-- Using PCLDBASE variable for convenience).

[WenMeng-NOAA]

On 06/19/2020, the branch release/gfs_v16 was committed at @35daf5 with PR #142. A new tag
upp_gfsv16_release.v1.0.9 was created for gfs v16 parallel run and installing post/8.0.9.