Turbulent Dissipation During a Westerly WindBurst: A Comparison of Large Eddy Simulation Resultsand Microstructure Measurements
E.D. Skyllingstad, W.D.Smyth, J.N. Moum, H. Wijesekera

College ofOceanic and Atmospheric Sciences

Oregon StateUniversity



Introduction
A large eddy simulation (LES) model of the upper ocean was initialized withmeasured profiles of temperature, salinity, and horizontal currents. Themeasurements were made at 2S; 156E on 31 December, 1992. Surface forcingconsisted of strong, steady winds and a diurnally varying heat flux (Fig. 1). The model was thenrun for 24 hours of physical time, forced continuously at the surface bymeasured fluxes of heat, salt and momentum. The model domain covered anarea of 384x384m in the horizontal, with periodic boundary conditions. Themodel bottom was at 96m, where a condition of outgoing radiation was imposed.Grid resolution was 1.5m in each direction (Fig. 2).The turbulence parameters generated by the model were then compared withequivalent quantities estimated from concurrent microstructure profiles.


Model Results
Typical flowfields generated by the model are shown in Fig. 3 and Fig. 4. The verticalvelocity at 5m depth (Fig. 3a) is dominated by Langmuircells driven by the strong wind. Within the mixed later, larger scale turbulentmotions are visible (Figs. 3b, 3c). At the baseof the mixed layer (Fig. 3d), wave like motionsare visible. Often, these take the form of overturning
Kelvin Helmholtz waves (Fig. 4b, left).

The main advantage of LES for the study of ocean turbulence is the verydetailed and comprehensive flowfield data that the model delivers. Fromfields like those illustrated in figures 3 and 4, we can gain a very clearpicture of the dynamics of mixing. The caveat, of course, is that the modelmay not be an accurate representation of real ocean physics. Comparisonslike those described below are needed in order to gain a full understandingof the strengths and limitations of ocean LES models.

LES Data Comparisons
The objective of this study was to compare high order turbulence statisticsoutput by the LES model with the same parameters derived from concurrentmicrostructure measurements. Quantities compared included the turbulentkinetic energy dissipation rate, , the temperature variance dissipationrate, , the mixed layer depth and the vertical turbulent fluxes of heat andsalt at the base of the mixed layer. Comparisons were performed mainly forthe night time, during which mixed layer turbulence was nearly stationary.

Comparison results are shown in Figs.5 and 6. The model showed significant skillat predicting measured values of (Figs. 5a, 6a) and (Figs. 5b, 6b), except for two major problems:

(1) Within the thermocline, stable stratification causes turbulentlength scales to contract to values below the resolution limit, invalidatingthe assumptions upon which the subgrid model is based. This diagnosis issupported by comparison of the Ozmidov scale and the buoyancy scale (Fig. 7). Agreement between these scales indicatesthat the model is representing subgrid effects accurately. Below the mixedlayer, the values of these length scales drop below the resolution limit. Deeper than this, the two scales diverge, indicating that modeledturbulence is too weak.

(2) The second problem occurred in the later stages of the simulation.In the model, turbulence near the mixed layer base decayed as the shearwhich sustained it was gradually mixed away. In the real ocean, large scalewave effects maintained that shear and thus kept turbulence levels roughlyconstant during the same interval.

Comparison of the intermittency factors for and was almost alwaysvery close.

Near the mixed layer base, the model showed significant skill at predictingvertical turbulent fluxes of heat and salt (Fig. 8),despite the fact that turbulent length scales are only marginally resolvedin that region of the water column. We believe that this is because fluxesare carried by the largest eddies, while dissipation occurs at much smallerscales. The hourly averaged fluxes agreed to within a root mean square discrepancyof , which is small compared to the uncertainty in the observational estimates.The 24 hour mean heat flux was predicted to within ~20% by the model, andmost of the error arose during the final eight hours of the simulation,during which model drift was evident.


Summary
A summary of the model verification is presented in Fig. 9 as a schematic overlaid on a time depth sectionof the Ozmidov length scale. Here, we imagine the depth time range of thesimulation as divided into five regions (all boundaries between regionsare approximate).

[i] Comparison was not attempted in the upper 10m because theobservational data is contaminated by ship wake.

[ii] Turbulence levels were unrealistically low in the early stagesof the simulation because of the time required for turbulence to spin up.This time is shortest near the surface because of strong wind forcing.

[iii] Below the mixed layer base, turbulence length scales contractedto unresolvable size due to stable stratification. This is indicated byOzmidov scales less than 3m (blue shading).

[iv] In the later stages of the simulation, the modeled mean profilesdrifted away from observed forms due to the absence in the model of largescale forcing effects which acted to sustain the shear near the mixed layerbase.

[v] In the remaining time depth region, labeled "optimalcomparison", both model and measurements are considered valid, andcorrespondence was very good.


Conclusions
The ocean LES model used here can be improved substantially by modificationsdesigned to mitigate the two major problems described above.

(1) Insufficient resolution in stable stratification: The simplestsolution to this problem is to increase resolution, particularly in thevertical. A more elegant solution is to adopt a subgrid model like thatproposed by Canuto and Minotti (JAS, 1993), which is designed explicitlyfor stably stratified flow.

(2) Model drift: This problem can be dealt with by adding extraforcing terms to the equations of motion which represent large scale effects(e.g. pressure gradients) that are not otherwise present in the model. Inthe case described here, this could best be done by assimilating interiorocean data. The horizontally averaged flow properties would be kept on trackby continuous relaxation towards the measurements, while waves and turbulencewould still be governed by the model.


Any Questions? e-mail e-mail me.