122
RAINER BLECK
The dashed lines in Fig. 3 show two choices of χ which roughly optimize ρ –χ orthogonality at high and low temperatures. The corresponding λ values are -0.13 psu/ ◦ C and -0.26 psu/ ◦ C, respectively. (Note that
χ is expressed here in salinity units.)
The advantages of replacing ρ, S advection by ρ, χ advection are difficult to quantify, mainly because the complexities of the wind- and
thermohaline-forced ocean circulation make it hard to identify numerical solutions that are clearly impacted by errors resulting from advecting
S and treating T as a diagnostic variable.
Fig. 4 illustrates the extent to which two HYCOM solutions, one based
on ρ, S advection and one on ρ, χ advection using λ = -0.26 psu/ ◦ C, diverge during a global 100-year coarse-mesh global simulation forced by
monthly climatology. In this figure, the gap between isopycnal layer
interfaces in the two simulations is colored red or blue depending on
whether the interfaces in the χ -based solution are at a greater or shallower depth, respectively, than the corresponding interfaces in the reference solution based on S advection (λ = 0). Fig. 4 shows that the use of
spiciness leads to a very slight density increase at low-to mid-latitudes
(blue coloration), while density is seen to decrease to a somewhat larger
degree at high latitudes (red coloration).
Differences between a model run based on λ = -0.13 psu/ ◦ C and the
S-advecting reference run are roughly half as large as those shown in
Fig. 4.
It is virtually impossible to judge whether the ρ, χ -based solution
represents an improvement over the ρ, S-based one. We are able to state,
however, that the switch from S to χ advection induces much smaller
changes (at least in this particular experiment) than what is typically
seen in 100-year experiments when surface forcing fields or aspects of
model physics are changed. This is a positive result, because it indicates
that the increased robustness of the algorithm for diagnosing T from
the two prognostic mass field variables is not achieved at the price of
encountering new potentially harmful model sensitivities.
Perhaps the best evidence that advecting χ instead of S leads to improved model performance is seen in a multi-century integration of a
coupled ocean-atmosphere model consisting of HYCOM and the GISS
atmospheric general circulation model. An early experiment based on
the traditional ρ, S-advecting version of HYCOM showed incidents of
anomalous ice growth occurring roughly once per century (grey curve in
Fig. 5). Detailed diagnostics of these events showed that the ice growth
took place in a region of the Arctic Ocean where slight errors in diagnosing T from ρ and S were compounded by a convective instability
feedback which spread the error conditions over a large depth range and
RAINER BLECK
The dashed lines in Fig. 3 show two choices of χ which roughly optimize ρ –χ orthogonality at high and low temperatures. The corresponding λ values are -0.13 psu/ ◦ C and -0.26 psu/ ◦ C, respectively. (Note that
χ is expressed here in salinity units.)
The advantages of replacing ρ, S advection by ρ, χ advection are difficult to quantify, mainly because the complexities of the wind- and
thermohaline-forced ocean circulation make it hard to identify numerical solutions that are clearly impacted by errors resulting from advecting
S and treating T as a diagnostic variable.
Fig. 4 illustrates the extent to which two HYCOM solutions, one based
on ρ, S advection and one on ρ, χ advection using λ = -0.26 psu/ ◦ C, diverge during a global 100-year coarse-mesh global simulation forced by
monthly climatology. In this figure, the gap between isopycnal layer
interfaces in the two simulations is colored red or blue depending on
whether the interfaces in the χ -based solution are at a greater or shallower depth, respectively, than the corresponding interfaces in the reference solution based on S advection (λ = 0). Fig. 4 shows that the use of
spiciness leads to a very slight density increase at low-to mid-latitudes
(blue coloration), while density is seen to decrease to a somewhat larger
degree at high latitudes (red coloration).
Differences between a model run based on λ = -0.13 psu/ ◦ C and the
S-advecting reference run are roughly half as large as those shown in
Fig. 4.
It is virtually impossible to judge whether the ρ, χ -based solution
represents an improvement over the ρ, S-based one. We are able to state,
however, that the switch from S to χ advection induces much smaller
changes (at least in this particular experiment) than what is typically
seen in 100-year experiments when surface forcing fields or aspects of
model physics are changed. This is a positive result, because it indicates
that the increased robustness of the algorithm for diagnosing T from
the two prognostic mass field variables is not achieved at the price of
encountering new potentially harmful model sensitivities.
Perhaps the best evidence that advecting χ instead of S leads to improved model performance is seen in a multi-century integration of a
coupled ocean-atmosphere model consisting of HYCOM and the GISS
atmospheric general circulation model. An early experiment based on
the traditional ρ, S-advecting version of HYCOM showed incidents of
anomalous ice growth occurring roughly once per century (grey curve in
Fig. 5). Detailed diagnostics of these events showed that the ice growth
took place in a region of the Arctic Ocean where slight errors in diagnosing T from ρ and S were compounded by a convective instability
feedback which spread the error conditions over a large depth range and