Tuning and Validating HYCOM’s KPP Mixed-Layer Alan J ...47N: Grid Aspect Ratio (SCPX/SCPY) 60S 40S...
Transcript of Tuning and Validating HYCOM’s KPP Mixed-Layer Alan J ...47N: Grid Aspect Ratio (SCPX/SCPY) 60S 40S...
Tuning and Validating HYCOM’s KPP Mixed-Layer
Alan J. Wallcraft and E. Joseph Metzger
Naval Research Laboratory
A. Birol Kara
COAPS, Florida State University
2003 Layered Ocean Model Users’ Workshop
February 10, 2003
� Long-term Global HYCOM Objective
Æ To depict and predict the 3D global oceanstate at fine resolution (0.08Æ on the equator,�7 km at mid-latitudes)
� Strategy
Æ Begin with 0.72Æ resolution global HYCOM andoptimize the KPP mixed layer performance (thesubject of this presentation)
Æ Before the end of FY03, start running anddevelopment of 0.24Æ global HYCOM
Æ Submitted DoD HPC Challenge proposal(FY04-06) to continue development at0.24Æ and start 0.08Æ model development
0.72 degree Global Domain
� Pan-Am Global Grid
Æ 0.72 degree equatorial Mercator 78S-47N
Æ Arctic bi-polar patch above 47N
� Low resolution global had patch at 59N� Can’t include Hudson Bay
Æ Double latitudinal resolution near the equator
Æ Halve latitudinal resolution in Antarctic
� Coastline at 50m isobath
Æ Closed Bering Strait
Æ No Sigma (terrain-following) vertical coordinate
� Same 26-layers as 0.08 degree Atlantic
� First Z-level 3 m, increases 1.125x up to 12 m
47N: SCPY (km)
60S
40S
20S
EQ
20N
40N
0E 20E 40E 60E80E 100E 120E 140E 160E 180W160W140W120W100W 80W 60W 40W 20W
-10 0 10 20 30 40 50 60 70 80 90
GLBa0.72ci 1.0
17. to 79.
47N: Grid Aspect Ratio (SCPX/SCPY)
60S
40S
20S
EQ
20N
40N
0E 20E 40E 60E80E 100E 120E 140E 160E 180W160W140W120W100W 80W 60W 40W 20W
0. 0.2 0.4 0.6 0.8 1. 1.2 1.4 1.6 1.8 2. 2.2 2.4
GLBa0.72ci 0.03
0.50 to 3.17
0.72 degree Global Standard Configuration
� KPP mixed layer
� Energy-Loan ice model
� Sigma-theta (some Sigma-2 runs, not discussed here)
� Horizontal diffusion chosen to suppress eddies
� Initialize from Levitus
� ECMWF Reanalysis monthly mean forcing
Æ Plus 6-hrly wind anomalies from sep94-sep95
� Annual mean of 10 largest rivers via precip bogus
� Strong relaxation to monthly Levitus SSS
Æ Necessary to prevent SSS drift
� Weak relaxation to monthly Levitus SST
Æ Parameterizes SST’s effect on longwaveradiation
Longwave Radiation and SST
� Longwave Radiation is sum of:
Æ Upward blackbody longwave radiation� ��� � ����� ������ ���� ��� �������
Æ Downward atmospheric longwave flux� Highly dependent on cloudiness� Unknown dependence on SST
(assume independent)
� If longwave was calculated using a SST of ���:
Æ ������ � ������� ��������������
Æ ������ � ������� ��
����� � ���
Æ ��
��� ����� ������ ���� ��� �������
� Or, approximately (in ����):
Æ ��
��� � ����� ����� ��
� So, ������������ is a good longwave correction
Æ Equivalent to 3.5 m reference thickness and30 day e-folding SST relaxation
Æ 10x weaker than typically SSS relaxation
� Ocean Model Intercomparison Project includes��������������� as a longwave correction
Variable Turbidity
� Penetrating solar (shortwave) radiation is importantfor accurate SST
� HYCOM parameterizes turbidity via a “two band”version of Jerlov water types
Æ One of 5 classes, red and blue bandsÆ Same everywhere in time and space
� NLOM used “single band” kPAR turbidity
Æ Photosynthetically Available RadiationÆ kPAR from SeaWIFS k490Æ monthly climatological fields
� Single band approach only OK for bulk mixed layer
� Added two band Jerlov-like kPAR scheme
Æ kPAR isn’t a good fit to JerlovÆ Much better than single class everywhere
� Need a better scheme:
Æ 3-band based on chlorophyll(Morel & Antoine, 1994)?
Æ Jerlov-like based on k490?
Annual SeaWiFS KPAR
60S
40S
20S
EQ
20N
40N
80E 100E 120E 140E 160E 180W160W140W120W100W 80W 60W 40W 20W 0E 20E 40E 60E
0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
GLBa0.72ci 0.002
0.040 to 0.200
SST Metrics I
� Run for 5 years and form monthly means
Æ May not be long enoughÆ Length of run vs confidence in resultsÆ Takes two days on 64 IBM POWER4 cpus
� Compare monthly SST to Reynolds and Smith(R&S) climatology
Æ Monthly anomaliesÆ Annual mean differenceÆ RMS differenceÆ Correlation CoefficientÆ Skill Score� Correlation squared - Unconditional Bias
- Conditional Bias� Maximum is 1, but minium is -infinity� Measure of error w.r.t. seasonal cycle
(i.e. w.r.t. standard deviation)� Use a minimum of 1 degC for standard
deviation� Still get poor skill scores near equator
SST Metrics II
� Purpose of comparison is to find “good enough”configuration
Æ Assume that “skill” on climatological forcing ismaintained on interannual forcing
Æ Is monthly thermal climatological forcing enough?
Æ NLOM experience suggests that this is OK, butcan’t be certain until we run interannually withHYCOM
� Targets:
Æ Annual mean error 0.5 degC
Æ Correlation Coefficient � 0.6
Æ Skill Score � 0.3
� Use zonal averages to reduce amount of data
Æ Average not necessarily best statistic
� A few large negative skill scores candominate the average
Æ Same targets as for full field
Simulation History
� Expt 1.5:
Æ Standard KPP
Æ Jerlov Ia everywhere
Æ No rivers
Æ No SST relaxation
� Expt 1.8:
Æ Monthly kPAR turbidity
Æ Annual rivers
� Expt 3.3:
Æ “Longwave” SST relaxation
� Expt 3.9:
Æ Modified KPP shear instability
� Expt 4.2:
Æ Twin of 3.3 (same winds)
Æ FNMOC thermal forcing (average 98-01)
-1
-0.50
0.51
-80
-60
-40
-20
020
4060
80
metric
latit
ude
ZO
NA
L S
ST
ME
TR
ICS
(III
), H
YC
OM
VS
Rey
nold
s&S
mith
SS
01.
5S
S 0
1.8
-1
-0.50
0.51
-80
-60
-40
-20
020
4060
80
degrees C
latit
ude
ZO
NA
L S
ST
ME
AN
ER
RO
R, H
YC
OM
VS
Rey
nold
s&S
mith
mea
n 03
.9m
ean
03.3
mea
n 01
.8
-1
-0.50
0.51
-80
-60
-40
-20
020
4060
80
metric
latit
ude
ZO
NA
L S
ST
ME
TR
ICS
(III
), H
YC
OM
VS
Rey
nold
s&S
mith
R 0
3.9
R 0
3.3
R 0
1.8
-1
-0.50
0.51
-80
-60
-40
-20
020
4060
80
metric
latit
ude
ZO
NA
L S
ST
ME
TR
ICS
(III
), H
YC
OM
VS
Rey
nold
s&S
mith
SS
03.
9S
S 0
3.3
SS
01.
8
03.3 vs R&S SST: Mean Error
60S
40S
20S
EQ
20N
40N
60N
60N
20E 40E 60E 80E 100E120E140E160E180W160W140W120W100W80W 60W 40W 20W 0E 20E 40E
-2.5 -2. -1.5 -1. -.5 0. 0.5 1. 1.5 2. 2.5
GLBa0.72ci 0.2
-5.0 to 8.7
04.2 vs R&S SST: Mean Error
60S
40S
20S
EQ
20N
40N
60N
60N
20E 40E 60E 80E 100E120E140E160E180W160W140W120W100W80W 60W 40W 20W 0E 20E 40E
-2.5 -2. -1.5 -1. -.5 0. 0.5 1. 1.5 2. 2.5
GLBa0.72ci 0.2
-4.9 to 8.1
03.3 vs R&S SST: Correlation
60S
40S
20S
EQ
20N
40N
60N
60N
20E 40E 60E 80E 100E120E140E160E180W160W140W120W100W80W 60W 40W 20W 0E 20E 40E
0. 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.
GLBa0.72ci 0.1
-0.8 to 1.0
03.3 vs R&S SST: Skill Score
60S
40S
20S
EQ
20N
40N
60N
60N
20E 40E 60E 80E 100E120E140E160E180W160W140W120W100W80W 60W 40W 20W 0E 20E 40E
-1. -.8 -.6 -.4 -.2 0. 0.2 0.4 0.6 0.8 1.
GLBa0.72ci 0.1
-20.1 to 1.0
04.2 vs R&S SST: Skill Score
60S
40S
20S
EQ
20N
40N
60N
60N
20E 40E 60E 80E 100E120E140E160E180W160W140W120W100W80W 60W 40W 20W 0E 20E 40E
-1. -.8 -.6 -.4 -.2 0. 0.2 0.4 0.6 0.8 1.
GLBa0.72ci 0.1
-21.1 to 1.0
Conclusions
� KPP is performing well in HYCOM
� Western Equatorial Atlantic a major trouble spot
Æ Modifying KPP shear instability helps
Æ More needs to be done
� In general, most of the SST error is in the annualmean
� Not yet clear how much is due to forcing and howmuch due to KPP