Tuning and Validation of Ocean Mixed Layer Models

1
Tuning and Validation of Ocean Mixed Layer Models

• David Acreman

2
In partnership to provide world-class ocean
forecasting and research
3
Overview

• The FOAM system
• The ocean mixed layer
• Kraus-Turner and KPP models
• Model performance and tuning at OWS Papa
• Model performance and tuning vs Argo data
• Effect of tuning in a global model

4
Forecasting the open ocean the FOAM system
FOAM Forecasting Ocean Assimilation Model

• Operational real-time deep-ocean forecasting
  system
• Daily analyses and forecasts out to 6 days
• Low resolution global to high resolution nested
  configurations
• Relocatable system deployable in a few weeks
• Hindcast capability (back to 1997)

5
The Mixed Layer (1)

• Surface layer of the ocean where temperature,
  salinity and density are near uniform due to
  turbulent mixing.
• Mixed layer deepens due to wind mixing and
  convection.
• Mixed layer shallows when winds are low and solar
  heating restores stratification.
• The depth of the mixed layer shows seasonal
  variability (deepens in autumn, shallows in
  spring).

6
The Mixed Layer (2)

• Mixed layer depth is an important output from
  FOAM
• Properties of the mixed layer affect
  ocean-atmosphere fluxes.
• Mixed layer depth also influences biological
  processes.

7
Mixed Layer Depth diagnostic
Use the Optimal mixed layer depth definition of
Kara et al. Search for a density difference which
corresponds to a temperature difference of 0.8 C
at the reference depth.
Figure from Kara et al, 2000, JGR, 105 (C7), 16803
8
Annual cycle of mixed layer depth from 1 degree
global FOAM
9
The Kraus-Turner Model

• The Met Office ocean model uses a bulk mixed
  layer model, based on Kraus and Turner (1967), to
  mix tracers.
• The model assumes a well mixed surface layer and
  uses a TKE budget to calculate mixed layer depth.
  
• A 1D configuration was used to validate and tune
  the model.

10
K-Profile Parameterisation of Large et al

• More sophisticated than KT.
• Doesnt assumed well mixed surface layer.
• Models turbulent fluxes as diffusion terms.
• Based on atmospheric boundary layer models.

11
Ocean Weather Station Papa

• Frequently used for validation and tuning of 1D
  mixed layer models
• Located in N.E. Pacific at 50N, 145W
• Ran Kraus-Turner and KPP models for one year
  starting in March 1961 (same as Large et al 1994)
• Used vertical resolutions of 0.5m, 2m, 5 and 10m
• Forcing fluxes calculated using bulk formulae
  (met data courtesy of Paul Martin)

12
Performance at OWS Papa (0.5m resolution)
13
Performance at OWS Papa (2m resolution)
14
Performance at OWS Papa (5m resolution)
15
Performance at OWS Papa (10m resolution)
16
Tuning the Kraus-Turner Model

• KT model based on a TKE budget.
• Sources of TKE are wind mixing and convection.
• Generation of TKE due to wind mixing given by
  W??u3
• 15 of PE released by convection is converted to
  TKE.
• TKE reduced by work done in overturning stable
  stratification and by dissipation.
• Dissipation represented by exponential decay with
  depth TKE exp (z/?).
• The free parameters ? and ? can be tuned to
  improve performance (currently ?0.7, ?100m in
  FOAM).

17
Tuning at OWS Papa

• Ran many model realisations with different values
  of ? and ? parameters
• Calculated mean and RMS errors in mixed layer
  depth
• Plotted errors vs. ? and ? parameters
• Tuned at 10m, 2m and 0.5m vertical resolutions

18
OWS Papa Tuning Results (10m resolution)
Mean errors
RMS errors
Minimum RMS errors with ?0.775, ?40m
19
OWS Papa Tuning Results (2m resolution)
Mean errors
RMS errors
Minimum RMS errors with ?1.275, ?30m
20
OWS Papa Tuning Results (0.5m resolution)
Mean errors
RMS errors
Minimum RMS errors with ?1.225, ?30m
21
Performance at OWS Papa (0.5m resolution)
22
Temperature and temperature error from tuned OWS
Papa K-T model
23
Model tuning using Argo data

• Argo floats are autonomous profiling floats which
  record temperature and salinity profiles
  approximately every 10 days.
• A large number of annual cycles are available for
  model tuning.

24
Kraus-Turner Model Tuning using Argo

• Forcing from Met Office NWP fluxes.
• Initial conditions from Levitus climatology.
• Temperature and salinity profiles assimilated
  over 10 day window.
• Vertical model levels based on operational FOAM
  system (10m near surface).
• Calculate mean and RMS errors, excluding cases
  with significant advection.
• Average over sample of 218 floats.
• Run KT model using different values of ? and ?.

25
Tuning results all floats
RMS errors
Mean errors
Smallest RMS errors with ?1.5, ?40m
26
Tuning results assimilation of one profile only
Mean errors
RMS errors
Smallest RMS errors with ?1.1, ?40m
27
Case study Argo float Q4900131

• Location 46N, 134W.
• Forcing from Met Office NWP fluxes.
• Initial conditions from float temperature and
  salinity profiles.
• No assimilation of data.
• Compare three different models Kraus-Turner,
  Large and GOTM.
• Run models at high vertical resolution (0.5m) and
  study annual cycle.

28
Case study Argo float Q4900131 (2)

• K-T model uses ?0.7, ?100m.
• GOTM version 3.2
• GOTM results courtesy of Chris Jeffery (NOC).

29
Case study Argo float Q4900131 (3)

• KT model uses l1.5, d40m.

30
New parameters in global FOAM

• Ran 1 year hindcast using global 1 degree FOAM
• Kraus-Turner parameters were changed to ?1.5,
  ?40m
• Plotted difference in mixed layer depth between
  models with old and new parameters

31
Difference in mixed layer depth
32
Conclusions

• The Kraus-Turner model can give a good
  representation of mixed layer depths when tuned.
• Optimum parameters for the Kraus-Turner scheme
  are ?1.5, ?40m with assimilation.
• Without ongoing assimilation the optimum value of
  ? is reduced.
• The Large et al KPP scheme tends to give mixed
  layers which are too shallow particularly at low
  vertical resolutions.