Most 14 version of "landmod.f90"
--------------------------------
   1594       dnpp(:)            = 0.
   1595       dgpp(:)            = 0.
   1596       dgppl(:)           = 0.
   1597       dgppw(:)           = 0.
   1598       dvwmax(:)          = dwmax(:)
   1599
   1600       where (dls(:) > 0. .and. dglac(:) == 0.)

The bug is a missing line at 1599 like:

              dvfor(:) = dforest(:)

The purpose was to work with local copies of the arrays
"dwmax" and "dforest". So the original arrays would be 
unaffected if NBIOME was set to 0 (no interaction).
The bug had the effect that "dvfor" started with the
initialized values of zero after every restart,
because "dforest" was read from the restart file
but not copied to "dvfor".

The current version Most15 with the corresponding code
fragment is now:


Most15 version of "simba.f90"
-----------------------------
    113 ! Following plasim arrays are used but not modified
    114 ! -------------------------------------------------
    115 ! dwatc(:) = soil wetness [m]
    116 ! dgppl(:) = GPP light limited
    117 ! dswfl(:) = short wave radiation [W/m2]
    118 ! devap(:) = surface evaporation (negative)
    119
    120 ! Make local copies of some plasim arrays
    121 ! They are copied back if NBIOME is set to 1 (interactive)
    122
    123 zforest(:) = dforest(:) ! Forest cover (0.0 - 1.0)
    124 zwmax(:)   = dwmax(:)   ! Bucket depth
    125
    126 dnpp(:)   = 0.0  ! Net   primary production [kg C m-2 s-1]
    127 dgpp(:)   = 0.0  ! Gross primary production [kg C m-2 s-1]
    128 dgppw(:)  = 0.0  ! GPP water limited
    129
    130 where (dls(:) > 0.0 .and. dglac(:) == 0.0) ! land cell with no glacier


"dvwmax" was renamed to "zwmax" and
"dvfor" to "zforest" following the coding rules for the Planet Simulator.
This and the new comments should clarify the roles of these arrays.