118
RAINER BLECK
consistency problem by advecting only one variable, S, and diagnose
T from S and the coordinate value ρ. (This strategy clearly works only
in isopycnic layers. MICOM’s nonisopycnic slab mixed layer requires a
second prognostic thermodynamic tracer aside from S.)
The pitfalls of diagnosing T from S and ρ are obvious and have
prompted some MICOM users working on polar ocean circulation problems to switch from S to T as prognostic variable (e.g., Holland and
Jenkins, 2001). Treating T as a diagnostic variable not subject to an explicitly enforced conservation law is also problematic in climate models
used for predicting secular temperature changes in the ocean-atmosphere
system. But the alternative, advecting T and diagnosing S everywhere
in a global model, has its own drawbacks because of the strong correlation between ρ and T in the stratified low- to mid-latitude upper ocean
which makes salinity a relatively poorly constrained diagnostic variable
there.
Dispensing of a conservation equation for S may also be more detrimental to dynamic stability than dispensing of one for T because of the
somewhat stronger control exerted by the atmosphere on the oceanic T
field. This is to say that spurious salinity transients are harder to control
in a model (in the absence of artificial restoring boundary conditions,
that is) because of the lack of a natural restoring process on salinity
akin to thermal relaxation. Given that salinity is more likely to act
dynamically as a “loose cannon”, one can argue that, globally speaking,
S conservation is more important than T conservation in situations where
a choice must be made between the two.
Since there is no guarantee that ρ is spatially uniform in any given
HYCOM coordinate layer, HYCOM must everywhere carry two prognostic thermodynamic tracers. The strategy adopted in the production
version is to treat both T and S as prognostic variables and delegate the
coordinate maintenance task to the grid generator. Unfortunately, this
choice is not optimal in all respects.
The real ocean has a tendency toward “density compensation”, meaning that T, S fields evolve in a manner which minimizes the dynamic
effects of T, S contrasts on the buoyancy field. This is to say that salinity
fronts are often accompanied by compensating temperature fronts. The
main problem with advecting T, S in a numerical model (any model, not
just HYCOM) is that numerical shortcomings of the transport algorithm
can and will destroy the spatial coherence of T, S fronts. In HYCOM,
this will lead to localized ρ anomalies which the grid generator, in an
attempt to restore target density, will convert into undulations in the
layer thickness field. This adjustment in turn causes additional vertical
dispersion.
RAINER BLECK
consistency problem by advecting only one variable, S, and diagnose
T from S and the coordinate value ρ. (This strategy clearly works only
in isopycnic layers. MICOM’s nonisopycnic slab mixed layer requires a
second prognostic thermodynamic tracer aside from S.)
The pitfalls of diagnosing T from S and ρ are obvious and have
prompted some MICOM users working on polar ocean circulation problems to switch from S to T as prognostic variable (e.g., Holland and
Jenkins, 2001). Treating T as a diagnostic variable not subject to an explicitly enforced conservation law is also problematic in climate models
used for predicting secular temperature changes in the ocean-atmosphere
system. But the alternative, advecting T and diagnosing S everywhere
in a global model, has its own drawbacks because of the strong correlation between ρ and T in the stratified low- to mid-latitude upper ocean
which makes salinity a relatively poorly constrained diagnostic variable
there.
Dispensing of a conservation equation for S may also be more detrimental to dynamic stability than dispensing of one for T because of the
somewhat stronger control exerted by the atmosphere on the oceanic T
field. This is to say that spurious salinity transients are harder to control
in a model (in the absence of artificial restoring boundary conditions,
that is) because of the lack of a natural restoring process on salinity
akin to thermal relaxation. Given that salinity is more likely to act
dynamically as a “loose cannon”, one can argue that, globally speaking,
S conservation is more important than T conservation in situations where
a choice must be made between the two.
Since there is no guarantee that ρ is spatially uniform in any given
HYCOM coordinate layer, HYCOM must everywhere carry two prognostic thermodynamic tracers. The strategy adopted in the production
version is to treat both T and S as prognostic variables and delegate the
coordinate maintenance task to the grid generator. Unfortunately, this
choice is not optimal in all respects.
The real ocean has a tendency toward “density compensation”, meaning that T, S fields evolve in a manner which minimizes the dynamic
effects of T, S contrasts on the buoyancy field. This is to say that salinity
fronts are often accompanied by compensating temperature fronts. The
main problem with advecting T, S in a numerical model (any model, not
just HYCOM) is that numerical shortcomings of the transport algorithm
can and will destroy the spatial coherence of T, S fronts. In HYCOM,
this will lead to localized ρ anomalies which the grid generator, in an
attempt to restore target density, will convert into undulations in the
layer thickness field. This adjustment in turn causes additional vertical
dispersion.