You are on page 1of 19

JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 105,NO.

C7, PAGES 16,803-16,821,JULY 15, 2000

An optimal definition for ocean mixed layer depth


A. Birol Kara
AdvancedSystemsGroup, SverdrupTechnology,Inc., StennisSpaceCenter, Mississippi

Peter A. Rochford and Harley E. Hurlburt


OceanographyDivision, Naval ResearchLaboratory, StennisSpaceCenter, Mississippi

Abstract. A new method is introduced for determining ocean isothermal layer


depth (ILD) from temperatureprofilesand oceanmixedlayerdepth (MLD) from
density profiles that can be applied in all regionsof the world's oceans. This
method can accommodatenot only in situ data but also climatological data sets
that typically have much lower vertical resolution. The sensitivityof the ILD and
MLD to the temperature differencecriteria usedin the surfacelayer depth definition
is discussed
by usingtemperatureand densitydata, respectively:(1) from 11 ocean
weatherstationsin the northeastPacificand (2) from the World OceanAtlas 199•.
Using these two data sets, a detailed statistical error analysisis presentedfor
the ILD and MLD estimation by season. MLD variations with location due to
temperatureand salinity are properlyaccountedfor in the definingdensity (Aat)
criterion. Overall, the optimal estimate of turbulent mixing penetration is obtained
usinga MLD definitionof AT=0.8øC, althoughin the northeastPacific regionthe
optimal MLD criterion is found to vary seasonally.The method is shownto produce
layer depthsthat are accurateto within 20 m or better in 85% or moreof the cases.
The MLD definition presentedin this investigation accurately representsthe depth
to which turbulent mixing has penetrated and would be a useful aid for validation
of one-dimensionalbulk mixed layer models and ocean general circulation models
with an embedded mixed layer.

1. Introduction the observations. One reason for this is that the depth
of turbulent mixing is sensitiveto the thermal and den-
The oceanmixed layer is generallyconsidereda quasi-
sity stratification at the base of the mixed layer and,
homogeneousregion in the upper ocean where there is
thereby, to the criteria chosento define MLD. The other
little variation in temperature or density with depth.
reasonis becausethe retreat of turbulent mixing to shal-
This definition is based on profiles from in situ data
lower depths proceeds much faster than the erosion of
that clearly reveal the presenceof approximately uni-
the stratification at the base,making a turbulent defini-
form vertical regionsof temperature and salinity start- tion of MLD inconsistent with an in situ data definition
ing at the ocean surfacesor at someshallow depth be- when there is a relaxation in winds. One-dimensional
low [e.g.,Roden,1979;PickardandEmery, 1990;Mon- models with high vertical resolution do quite well at re-
tereyand Levitus,1997]. Theseregionsof verticaluni- producingobservedMLDs from their modeledtempera-
formity owe their existenceto turbulent mixing gener-
ture and density profileswhen forced with observational
ated from the energyinput by the action of wind stress
and heat fluxes at the ocean surface. This turbulent
winds and heat fluxes [Kantha and Clayson,1994].
However,this is much more of a challengewhen imple-
origin is what has motivated many successfultheoret-
menting suchmixing models into ocean general circula-
ical descriptionsof the ocean mixed layer through dif-
tion models(OGCMs), as the latter are typically con-
ferent representationsof turbulence in numerical mod-
strained to using lower vertical resolution and higher
els [Welander,1981; Large et al., 1994; $kyllingstad "background" diapycnal diffusion, and they often use
et al., 1996]. climatologicalsurfaceforcing [e.g., Cherniawskyand
A commondifficulty encounteredis relating the mixed
Holloway,1991]. Moreover,an OGCM often usesver-
layer depth (MLD) predictedby turbulencemodelsto tical interpolation or a prognostic equation to predict
the MLD [e.g.,McCrearyet al., 1993;$terl andKatten-
Copyright2000 by the AmericanGeophysicalUnion. berg,1994].
Comparisonagainst a MLD climatology is frequently
Paper number 2000JC900072. desired when developing an OGCM with an embedded
0148-0227
/ O0/ 2000JC900072$09.O0 mixed layer as it helps to validate the implemented tur-
16,803
16,804 KARA ET AL.: OPTIMAL MLD DEFINITION

bulent physicsover ocean basins on monthly to annual is definedas the depth at which the in situ tempera-
timescales
[SterlandKattenberg,1994].A MLD defini- ture has decreasedto SST-AT, for some chosentem-
tion that seeksto representturbulent mixing as well as perature difference AT, while for the MLD it is the
possibleand yet is also robust enoughfor construction depth at which the calculated in situ density has in-
of a global MLD climatologyfrom existingin situ cli- creasedby Aat (at = p-1000kgm-3) fromthe surface
matologieswould be a useful aid for validating mixed value. With a gradientcriterion [e.g., Bathen,1972;
layer performance in OGCMs. The benefits of such a Lukasand Lindstrom,1991;Richardset al., 1995]the
MLD definition extend far beyond OGCM validation as ILD (MLD) is definedas the depth at whichthe gradi-
many MLD climatologieshave been usedto investigate ent OT/Oz (Oat/Oz)exceeds a specificvalue,assuming
a variety of oceanproperties. These includeinvestigat- that a sharp interface exists at the base of the surface
ing where penetrative solar radiant heat fluxes affect layer. Temperatureand salinity variationswith location
oceanmixedlayerheating[Ohlmannet el., 1996],de- are typically ignored in the MLD definitions.
termining regionsof barrier layers where salinity effects Noneof the studiesabove(seealsoTable 1) presented
impede vertical heat flux out of the base of the mixed a quantitative analysisjustifying a particular AT value
layer[Sprintalland Tornczak,
1992],andusingMLD cli- as the mostappropriatecriterion,and only You[1995]
matologiesas part of constraintsor inputs for an OGCM and MontereyandLevitus[1997]comparedtheirresults
in variousheat flux, sea surfacetemperature (SST), with differentapproaches.Given the sensitivityof tur-
and circulationstudies[Spell, 1991; Huang and Rus- bulent mixing to the thermal and density stratification
sel,1994]. Moreover,havingsuchvaluableMLD infor- at the baseof the mixed layer, askingwhetherthesecri-
mation would help modelersto couple atmosphereand teria are suitable seemsappropriate. Differencesin the
ocean models realistically and to understand the sea- criteria can lead to considerable differences in,,the ILD
sonalvariationsin the productivity of oceanecosystems and MLD, whichin turn, couldinfluencethe •tudy find-
[Obataet el., 1996]. MLD is alsoof particularimpor- ings, such as the phase and amplitude of the seasonal
tance not only becausethin mixed layers are more con- cycle of MLD. In that regard, one questionto investi-
duciveto surfacecoolingby wind-inducedupwellingbut gate is whetherthe definingcriteria for the ILD or MLD
more importantly becausethe mixed layer thicknessde- shouldincludethe influenceof seasonalchangesin the
termines the volume or mass over which the net surface thermoclineor pycnocline,respectively.
heat flux comesto be distributed[Chenet el., 1994]. In this paper we investigatethe sensitivityof the ILD
Several isothermal and MLD criteria that have been and MLD to variousfinite differencecriteria using a
used in the literature are given in Table 1. All the new method for determining such layer depths in the
definitions in the studies applied a simple criterion for ocean. We choseto developour method using a finite
determiningoceansurfacelayer depth from in situ data differencecriterionrather than a gradientcriterionfor
on the basisof changesin the propertieswith depth. two reasons:(1) the latter can only be usedwith pro-
They can be separatedinto two generalcategories.The filesof temperatureand salinitythat adequatelyresolve
first is to definean isothermallayer depth (ILD) from sharp gradients,and (2) a recent experimentalstudy
a temperature profile and assumethis to be the MLD has shown that the ILD based on difference criteria is
and the secondis to define a MLD from a density pro- more stable than the ILD based on gradient criteria
file. Either a property difference or a gradient crite- [Brainerdand Gregg,1995]. The methodintroduced
rion is used for this definition. For the former the ILD here is applicable for both the ILD and MLD defini-

Table 1. Commonly Used ILD and MLD Criteria

Temperature-Based Criterion Density-Based Criterion


Author AT Author Aat

Thompson [1976] SST-0.2øC Miller [1976] 0.125at


Lamb[1984] SST-1.0øC Levitus[1982] 0.125at
Price et al. [1986] SST-0.5øC Lewiset al. [1990] 0.13at
Kelly and Qiu [1995] SST-0.5øC Spall[1991] 0.125at
Martin [1985] SST-0.1øC Sprintalland Tomczak[1992] 0.5(OatlOT)
Wagner[1996] SST-1.0øC HuangandRussell[1994] 0.125at
Obataet al. [1996] SST-0.5øC Ohlmannet al. [1996] 0.5(OatlOT)
MontereyandLevitus[1997] SST-0.5øC Montereyand Levitus[1997] 0.5(OatlOT)

Temperature-and density-based
criteria(ILD and MLD, respectively)usedfor determining
the oceansurfacelayer depth. SST-0.2øC, for example,indicatesthat the layer depth is defined
as the depth where the temperature is 0.2øC lessthan the SST. Note that most use 0.125at for
the MLD definitionbecauseit corresponds to the water masscharacteristicsof SubtropicalMode
Water in the North Atlantic as explainedby Levitus[1982].
KARA ET AL.: OPTIMAL MLD DEFINITION 16,805

tions at all latitudes and can accomodate the presence in this study is the monthly climatologyconstructed
of all casesof verticalstratificationin the ocean(e.g., from oceanweatherstation (OWS) observations in the
fossillayers,inversionlayers,and dicothermallayers). North Pacificfrom 1959to 1990 [Tabataand Weichsel-
We presenta thoroughanalysisof the errorsassociated baumer,1992; Tabataand Peart, 1992],hereinafterre-
with estimating layer depths using different criteria to ferred to as the OWS data. The 11 stations used in this
determine the optimal definition. study(Figure1) arelocatedat a rangeof oceandepths,
Section 2 describesthe data sets used in this study, as given in Table 2. Monthly meansof OWS temper-
and section3 explainsthe detailed methodologyfor es- ature and salinity are averagedat each standard level
timating the ILD and MLD. Section4 verifiesthe con- over a 31 year period. The standard levels used in the
sistencyof our methodologyby comparinglayer depths OWS data set are all at 10 m increments from the sur-
obtained from the two data sets: a monthly climatology face to a depth of 1000 m. We use these two data sets
from the WorldOceanAtlas 199• [Levituset al., 1994; to confirmthat inferring the MLD from Levitus data
LevitusandBoyer,1994](hereinafterreferredto asthe usingour methodologyyields valuesthat are consistent
Levitusdata) and a monthlyclimatologyfrom 11 ocean with thoseobtainedusinglong-term monthly time se-
weather station observations.Section5 investigatesthe ries of higher vertical resolution. The OWS data set
sensitivityof the ILD (MLD) to temperaturedifferences is chosenfor this purposebecauseof its easyavailabil-
in the temperature(density)criteria. Section6 presents ity, reliability, and widespreaduse for one-dimensional
the results of the analysis. mixed layer studies.
For both data sets used in this study the density
2. Data is calculatedusing temperature and salinity valuesat
given depths basedon the standard United Nations Ed-
Monthlyaveragedtemperatureand densityprofiles ucational,Scientific,and CulturalOrganization(UN-
are used from two different data sources. One of themESCO)equationof statewith nopressure dependence,
is the monthlyclimatologyfrom the Levitusdata. The i.e., zeropressure[Milleroet al., 1980;Millero andPois-
Levitusdata containuniformlygriddedmonthlycli- son, 1981]. The inclusionof pressureeffectsincreases
matologiesof temperatureand salinity at a latitude- the densitygradientsufficiently rapidlywith depththat
longitudegrid resolutionof 1ø x 1ø, with 19 standard it producesa markedlyshallower MLD that is strongly
depth levels to 1000 m. The vertical resolution de- inconsistent with the MLDs inferred from the corre-
creaseswith depthby 0, 10, 20, 30, 50, 75, 100,125,and spondingtemperatureand salinity profiles. This is il-
150m, every50 m to a depthof 300m, andthenevery lustratedwith the Levitusand OWS temperature,den-
100 m to a depth of 1000 m. The other data sourceused sity, and salinity profilesto 300 m depth at station J

58N

57N

56N

55N AJaska
Oyre . ..... ..
54N !
53N
West
Wind
Drift'
52N

51N

50N
K

49N

48N

47N
.

46N California
'current
'•'•,•
45N
14'7W 14•4W 14'1W 15'8W 15'5W 15'2.
W 12•9W 12'6W 125W

Figure1. Thelocations
of 11OWSsin thenortheastPacific
Oceanlabeled
alphabetically
from
A through
K. Majoroceanic
features
include
theCalifornia
Current,
thewestwinddrift,andthe
Alaskagyre[seeLynnandSimpson,
1987].
16,806 KARA ET AL.' OPTIMAL MLD DEFINITION

Table 2. North Pacific Ocean Weather Stations

OWS Latitude Longitude Depth, Distance,


m km

I A 48ø39.0'N 126ø40.0'W 1300 87


2 B 48ø44.6•N 127ø40.0•W 2500 161
3 C 48ø49.0•N 128ø40.0'W 2440 233
4 D 48ø58.2•N 130ø40.0'W 3300 380
5 E 49ø07.4•
N 132ø40.0•W 3275 526
6 F 49ø17.0•N 134ø40.0•W 3550 672
7 G 49ø26.0'N 136ø40.0•W 3775 817
8 H 49ø34.0•N 138ø40.0•W 3890 961
9 I 49ø42.0•N 140ø40.0'W 3880 1106
10 J 49ø50.2•N 142ø40.0•W 3910 1250
11 K 50ø00.0'N 145ø00.0•W 4200 1420

Listed are the maximum depths and offshore distancesof


the North Pacific OWSs usedin this study. Station K located
at (50øN, 145øW) is also known as "OceanWeather Station
Papa," and observations from this station have been used in
many studiesto validatemodels.Valuesgivenare adoptedfrom
Taharaand Brown[1994].

in February(Figure2). The temperature,density,and clearly show the mixed layer formed due to turbulent
salinity valuesin both data setsare in very goodagree- mixingfrom windsand surfaceheating/cooling. From
ment with eachother and corroboratethe reliability of the profilesit is evidentthat onemust usea pressure-
these two data sources. The temperatures and salin- independentequation of state if one is to be consistent
ities in the upper regionsof the profiles are, in gen- with the temperatureand salinity profilesin determin-
eral, at much the same values as at the surface and ing a MLD. (Note here that the pressure-dependent

Temperature
(øC) Density
(%) Salinity
(ø/oo)
5.0 5.6 6.2 6.8 25.5 25.9 26.3 26.7 32.5 33.0 33.5 34.0

50 50

lOO
[]øl"-•x-
xx••• i•-. ................. lOO
..................................
................... ..................
.........................
150 ........................... •'? ............................................... X .......................................... • ................. 150

200
.................................. ............................
2OO

250 250
•/ • Observation • Observation• • Observation

300 :/ evius evius [ evius [ 300
Figure 2. Profiles
oftemperature,
density
at andsalinityto 300m depthat stationJ in February.
Standardverticallevelsfor the Levitusdata are shownwith solidcircles,whilethe onesfor the
OWS data are shownwith opensquares.Note that the vertical resolutionfor OWS data is 10 m.
Densityprofilesfor both data setswereobtainedby usingthe pressure-independent equationof
state as explainedin the text. The pressure-dependent
densityprofilefor Levitusdata is shown
with a dashed line.
KARA ET AL.: OPTIMAL MLD DEFINITION 16,807

equation of state can be used if one considersthe dif- indicative of regionsof extremely deep convectivemix-
ference in density for a parcel of seawater relative toing associatedwith the global overturning circulation.
the backgroundvalue. If one usesthe potential tem- The implementation of the criteria used here requires
perature, this gives the exact density differencefor wa-
a multiple-step procedure that is separately applied
ter transportedto a given depth.) As we have found, when determining an ILD or MLD. A schematic dia-
this can be easily overlookedwhen usingthe UNESCO gram(Figure3) showsthe determinationof ILD (MLD)
equation of state for the first time in a mixed layer when usingthe Levitus data accordingto a temperature-
model. Note that an incompressibleequation of state is based(density-based)criterion. We first describethe
consistentwith the incompressibilityassumptioninher- procedure for determining an ILD.
ent within the majority of one-dimensionalmixed layer 1. The temperature at 10 m depth is chosenas the ini-
modelsand OGCMs and that defininga MLD that can tial referencetemperaturevalue (Tref) for determining
be usedfor OGCM validationis oneof the goalsof this the ILD. This depth is chosento eliminate any possible
study. bias in the profile data due to "skin effects" at the ocean
surface[Fairallet al., 1996].In the majorityof casesfor
the Levitus data the temperature at 10 m is very close
3. Methodology to the surface temperature value. While this reference
The method given here is able to accommodatethe depth imposesa minimum value of 10 m for the ILD, we
wide variety of temperature profiles that occur within note that OGCMs typically limit their minimum MLD
the global ocean. This includes temperature inversions to 10 m or more[e.g.,Cherniawsky andHolloway,1991;
that occur at high latitudes, a subsurfacemixed layer McCrearyet al., 1993;$chopfand Loughe,1995].
underlying a surfacethermal inversion, multiple fossil 2. A search is then made of the temperature pro-
layers beneath the surface mixed layer, a dicothermal file data for a uniform temperature region. We define a
layer, as well as the typical temperature profiles with uniform"well-mixed"temperatureregionas any pair of
strong and weak thermoclinesfound in the middle and temperaturevalues(Tn and Tn+l) at adjacentdepths
low latitudes[e.g., Brainerdand Gregg,1995]. For a (h• and hn+l) in the profile that differ by lessthan
discussionthat defines and explains the formation of one-tenth the temperature differencecriteria AT defin-
some of these various characteristics we refer the reader ing the ILD (e.g.,AT=0.2 ø, 0.5ø, 0.8ø, and 1.0øC),i.e.,
to SprintallandRoernrnich
[1999]. differenceslessthan or equal to 0.1 AT. For the exam-
Here the criteria for defining a MLD are developed ple profiles shownin Figure 3, the standard levels hn
through subjectiveanalysisof temperature and density and hn+l correspondto 100 and 125 m, respectively.
profilesfrom the Levitus data with the view that the 3. If a uniformtemperatureregionis found,the value
mixed layer is the regionjust below the ocean surface of referencetemperatureTrefis updatedto the temper-
where the temperature or density is approximately uni- ature value T• at the shallowerdepth h• of the pair
form. The samemethodologyis later applied to OWS of profilepoints. This is donefor everyoccurrenceof a
data. This goes beyond the work of previous authors pair of pointsoccurringwithin the first uniformtemper-
(Table 1), who usedonly a fixed decreasein tempera- ature region so that the referencetemperature is that at
ture or increase in density from a value at a reference the baseof the well-mixed region. The ILD will then be
depth near the surface. The greater complexity of our the depth at whichthe temperaturehas changedby an
algorithmwasnecessaryin orderto obtain a MLD that absolute value of AT from this reference value. For ref-
is consistentwith what one would infer from inspection erencepurposeswe shallrefer to this latter temperature
of the profilesin many regionsof the world ocean. The as the base temperature Tb.
simpler criteria used in previous studies were found to 4. Temperaturechangeswith depth of either signare
fail in many casesin the presenceof fossillayers, inver- used in determining ILD. Thus the value of the base
sionlayers,and dicothermallayers. TheseyieldedMLD temperature is given by
values that differed by more than 20 m from those ob-
Tn < Tn+l ,
tained usingour methodology,oftenreachingdifferences
as large as hundredsof meters. The criteria we apply Tb -- Tref Tre
+f- AT
AT >_T+i.
for the ILD and MLD become similar to those of other
authors(Table 1) for thosecaseswherethere is no sub- If found, the depth of Tb is determinedby linear in-
surfaceregion of uniform temperature and density,for terpolationwithin the depthinterval (h•, hn+l). This
example, a stable thermocline. depth definesthe ILD for the applied temperature cri-
From an examination of the global MLD fields we teria AT.
obtained using our methodology,we find no need to 5. If no depth range (h• and hn+l) is found such
imposea maximumdepthfor the mixedlayer[e.g.,Lev- that (T• andT•+i) containsT•, then the profiledata is
itus, 1982]as reasonable valuesare obtainedover99% searchedagain,startingfrom the 10 m referencedepth,
of the world oceanarea. The remaining1% of the cases for a temperature changeof AT from the 10 m ref-
are consequences of highly uniform vertical profilesoc- erencetemperature. This can be a positive or nega-
curring at high southernlatitudes, and these could be tive changeaccordingto the temperature variation with
16,808 KARA ET AL.: OPTIMAL MLD DEFINITION

Temperature
(øC) Density((•t)
6.25 6.50 6.75 7.00 7.25 25.25 25.50 25.75 26.00 26.25
I I I

lO -lO
href T (Jref href-
20 20
30 - 30

50 - 50

(Jn hn
75 -75

100
ha %+1 hn+l- lOO

AT

125
hn+l rn+l - 125

150 - 15o

Figure 3. A schematicillustrationof the ILD (h•;(T)) and MLD (h•;(at)) determinationusing


the temperatureand densityprofilesat the locationof stationJ (Figure 2) from the Levitusdata
in February.For easeof notationwe usethe samesymbolsfor the standardlevels(hn and hn+•)
whendescribingthe procedurefor both criteria. The depth at whichthe ILD is foundis shown
with an open circle on the temperatureprofile, and the temperature at this level is denotedas
the base temperature Tb.

depth. This occursat high latitudesfor two general where $ is the salinity and the pressureP is set to zero
cases:(1) whenthere is a largetemperatureinversion [MilleroandPoisson,1981;Milleroet al., 1980]. For
at the surfaceand the temperature at depth never de- our example(Figure3) the ILD (i.e., h•;(T)) is found
creasesto as low a valueand (2) whenthe temperature betweenthe 100 and 125 m standard levels, while the
remains almost constant to the bottom of the ocean. MLD (i.e., hL(rrt)) is foundbetweenthe 75 and 100 m
In both casesthe ILD is set to the depth of the ocean standard levels for the same location. This is a more
bottom if no depth is found at which the temperature carefultreatment of Arrt in a density-baseddefinitionof
has changedby AT. MLD than has been considered in the literature to date
Note that this method does not use temperature gra- (Table1) asit takesfull accountof densitychanges due
dientsaspart of its criteriafor determiningthe ILD for to temperature and salinity variations with location.
the reasonsgivenin section1. Reliableapplicationof For eachstationA throughK (seeFigure1) the ILD
suchcriteria requiressufficientlyhigh resolutionin the and MLD are obtainedusingthe Levitusdata according
profile data to determineaccuratelythe temperature to a temperature-or density-based criterion, respec-
gradients.With climatological data setssuchas Levi- tively. The same proceduresare applied to the OWS
tus, whichhaveonly 19 standardlevelsdistributedover temperature(density)profilesat the samelocationsto
a 1000m depth,suchverticalresolutionis not available. determinethe ILD (MLD). The higher 10 m resolution
To have a robust algorithm, we have thereforeadopted of the OWS data enablesthe layer depths to be more
a simple approachbasedon a AT change. accurately obtained. The OWS data also exhibit a shal-
The MLD determined from density followsthe same lower and sharper thermocline because no horizontal
procedureas for temperature but with a density vari- averagingof the temperature and salinity profiles was
ation determined from the correspondingtemperature done as for the Levitus data. For simplicitythe ILD
changeAT in the equationof state obtained from the Levitus data set will henceforth be
represented by hL(T), and the onefrom the OWS data
Aat - at(T + AT, S, P) - at(T, S, P), (1) setwill be representedby ho(T). Similarly,hL(rrt) and
KARA ET AL.- OPTIMAL MLD DEFINITION 16,809

ho(r)
h(T)
JanMarMay Jul SepNov JanMarMay Jul SepNov JanMarMay Jul SepNov

o ho((5,),

JanMarMay Jul SepNov JanMarMay Jul SepNov JanMarMay Jul SepNov


I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I

25- "•" !" •["•'";," ;'"•" a' T"i'"•'"!'-

50- J...
75-
- -•.......... 75 •
. . r . . 2 . . . r . . .•. . . r . . .•...... .•....... ',.......

125 - - 125 •'

150 - - 150•t
ii": :: ] Station
A I : :: ,: :, ,: I Station
E I: :: :, :: :: ]
, i i : i i i i i i i i i , i i i i
stationr I'i ::
i i i i i i i

Figure 4. ILDs (ho(T) andhL,(T))andMLDs (ho(at) andh•;(at)) at stationsA, E, andK for


the months from January to December. A AT interval of 0.8øC is used for the ILDs and MLDs.
Note that the axis labeling is for every other month.

ho(at) will denotethe MLD obtainedfrom the Levitus This is likely due to the high continentalrainfall known
and OWS profiles,respectively. to occur in this region.
Figure 4 showsthe monthly ILD and MLD valuesob- A comparison of monthly layer depths when using
tained for stations A, E, and K when using a AT value differentAT values(Figure 5) showsthat the ILD and
of 0.8øC. In general,ho(T) and hœ(T) agreequite well MLD both increasewith larger AT, as expected. Vary-
with each other for all stations, and the same is true ing AT causesgreater changesin the MLD than in the
for ho(rrt) and hœ(rrt). However,the MLD are shal- ILD becauseof the additional effectsof salinity. This
lower than the ILD for both data sets. In particular, is especiallytrue for AT values of 0.2ø and 0.5øC. We
for stationA the ho(T) and hœ(T) valuesare quite dif- also note that the MLD are onceagain shallowerthan
ferentfrom their ho(rrt) and hœ(rrt)counterparts.The the ILD for the same AT value.
correspondingdifferencesare minor for stations E and
K, which are farther offshore.The larger differencebe-
tween ILD and MLD at station A is because of offshore 4. Comparison of the Layer Depths
advection of fresh water, which causesthis station clos- Here we examinelayer depthsobtainedfrom two dif-
est to the coast(Table 2) to havea shallowerhalocline. ferent data sets for the same AT value to determine
16,810 KARA ET AL.' OPTIMAL MLD DEFINITION

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
0

lOO
100•
150

200
| ! \!"•oi i ! 1!! 200
il Temperature-based
criterion •
250 250

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

50- 50 •'
-i

100- ..... o- -o-•.......... • ......... •.......... •......... •.......... •......... •........ -q--o
lOO ,•
,

•.o øc::
150- ..... a..................... !.......... •.......... a.......... •.......... 1.......... •.......... .•.......... •.......... 4.......... •..... -150 =

200- -200
Density-basedcriterion
MLD (hœ({Jt))
250 - 250

Figure 5. Layerdepths(ILD and MLD) by monthat stationH obtainedusingthe Levitus


data(hœ(T)andhœ(at))fora rangeof AT values:
0.2ø,0.5ø,0.8ø,and1.0øC.

the differencewhen using the ILD as the MLD. We larger than hL(at) in winter and spring,while the re-
have used AT values of 0.2ø, 0.5ø, 0.8ø and 1.0øC to verseis true overallfor the h¸(T) and hL(T) values.In
obtain the ILD and MLD at each station by month for general,the mean valuesfor the given AT interval and
both criteria. For the oceanweatherstations(seeFig- state variablecriterion (T or at) showcloseagreement
for the two data sets.
ure 1) this provides11 ILD (MLD) for eachmonthfor
the Levitus and OWS data, separately,for a given AT For each station we also examine the difference be-
valueand temperature-(density-)basedcriterion. tween the ILD and MLD by month and present the
To provide sufficiently accurate statistics, the meanresultsgroupedby season(Figure 7). An interesting
of the ILD and MLD for each AT trend is found when examining the number of cases
value are calcu-
lated by season.We follow Levituset al. [1994]in our wherethe h•-ho differenceis positiveor negative(Ta-
definition of seasons: January, February, and March ble 3). The h(T) layer depthsobtainedfrom Levitus
(winter); April, May, and June (spring);July, August, are typically greater than those obtained from OWS in
and September(summer);and October,November,and all seasons,especiallyin winter. In fact, no casesexist
December(fall). The mean valuesfor the ILD and where hsu(T)< ho(T) in winter, spring,and summer
MLD (Figure 6) are very closeto each other in sum- when usinga AT=0.5øC. Note that differencesbetween
mer and fall, indicating that they producecomparable hsu(at)and ho(at) in springare quite smallrelativeto
layer depthsin theseseasons.In winter and spring,in- those in winter and fall and that the scatter in winter
creasinglylarger differencesoccurwith larger AT value, and fall is larger, as will be discussedin more detail in
with a differenceof up to 80 m resulting in winter for section5. In summerand fall we againnoticethe h• (T)
AT=I.0øC. The mean values of ho(rrt) are always valuesare usuallygreaterthan the ho(T) values,while
KARA ET AL.' OPTIMAL MLD DEFINITION 16,811

Mean Meanhœ(tJt)
Mean
ho(T) Mean
ho(Ort)
150 150
Winter Winter
• 100 100

• 50
o
[ t ! i i
,.-/ i
/ l i
5O

0
150
15o
Spring Spring
lOO 100

5o 5O

0
150 150
Summer Summer
100 100

50 5O

0 0
150 150
Fall Fall
100 100

5O
50
0.2øC 0.5øC 0.8øC 1.0øC 0.2øC 0.5øC 0.8øC 1.t2
Temperature
Difference(AT) Temperature
Difference(AT)
Figure6. Mean
layer
depths
(ILDandMLD)obtained
using
theLevitus
(hœ(T)
and
andOWSobservations
(ho(T)andho(at))when
using
AT values
of0.2ø,0.5ø,0.8ø,and1.0øC.
Thenumber
of cases
for eachAT category
in a givenseason
is 33. Themeanvalues
obtained
from the ILD and MLD are shown separately.

the reverseis true for hœ(at)and ho(crt). The reason ho(T)) and (hœ(crt)andho(crt)) are comparedunder
hœ(T) > ho(T) overall theverticalresolutionthe same AT criteria for each seasonusing severaldif-
isbecause
of the Levitus data is coarserthan that of the OWS data ferent statistical measures. Several measures are used

at greaterdepthsandbecause averag- becauseevaluating the performanceof estimating the


of the horizontal
the Levitusdata. Thisresultsin a ILD and MLD in arriving at the best choicefor AT is
ing in constructing
broaderanddeeperthermoclinein Levitus.This biasis important. We subsequentlyinvestigatethe differences
largerin winterandspringwhenthe hL(T) valuesare that arise when usingan ILD versusMLD definitionfor
large.Notethat the variabilityin the numberof cases a given AT in order to assessthe relative error intro-
wherehL(at)-ho(crt)ispositive ornegativeismuchless duced by the former relative to the latter.
thanfor the hL(T)-ho(T) cases in eachseason(leftto We considervarious statistical measurestogether to
rightin Figure7). Thisreveals that the density-basedmeasure the strength of the relationship between the
criterion is a more reliable indicator of MLD as it pro- pairs hL(T) and ho(T) or hL(rrt) and ho(rrt). The
vides a consistent trend between different data sources preferred indicator of the strength of the statististical
that is relativelyinsensitive
to the AT valueused. relationship between the layer depths from the Levi-
tus data (hL) and the onesfrom the OWS data in the
5. Sensitivity to Temperature northeastPacificOcean(ho) is the coefficientof deter-
Differences
mination(r2). The reasonis that the correlation
coeffi-
cient r may givea misleadingimpressionof a closerrela-
In this sectionwe undertake a sensitivitystudy to de- tionshipbetweenhL andho than actuallyexists[Neter
termine the best AT value to be used for the ILD and et al., 1988]. Furthermore,
r 2 possesses
a meaningful
MLD. For this study the pairs of values(hœ(T) and operational interpretation as a measureof the propor-
16,812 KARA ET AL.- OPTIMAL MLD DEFINITION

o h(D - ho(D
+ - ho(O,)
AT=0.2
ø, 0.5ø 0.8o, o, 0.5ø
1.0øC AT=0.2 0 8ø 1.0øC
- 30
20 •
10 •*
0 n
10 •
20 •'
-50• + :: I Winter[

- 30
20
10 œ
0
10 m
20 •

30
20 •
10 o
., 0 --
10 m
-10 -
20 •
-20 -

0 33 66 99 132

Cases(StationA throughStationK)
Figure 7. The scatterdiagramof differences
in the ILD (hœ(T)-ho(T))andMLD (hœ(at)-
ho(at)) between
LevitusandOWSobservations. Panels(a)-(d) showthenumberof cases
when
hL > ho or whenhL _<ho. The symbolson the barsdenotecorrespondence
with the data in
panels(e)-(h). The y axesare differentfor eachseason.

tion of the.variability in the observationsaccountedfor stein[1989].Thesestatisticalmeasures


aregivenby the
by a linearmodelin the forecasts[Murphy,1988].For relations
example,a 0.90r 2 valueshowsthat 90%of the variance
in theobservations
isreproduced
bytheestimates.
The r2 _ [$a$a•ao
L$ao '
(2)
mean squareerror (mse) and skill score(SS) are also
usedto verifyour resultssincer 2 ignoresboth'condi- mse
- (a•_ao)=+s•,•
tional and unconditionalbias (Bcondand Buncond,-re-
spectively),whichare basicaspectsof assessing
perfor- + $•o - 2r $• &o, (a)
mance[Murphy,1995].Unconditional
bias(alsocalled ss - ," - [• - (&/S•o)]•
systematicbias) is a measureof the differencebetween Bcond
the means of two data sets, while conditional bias is
a measureof the relative amplitudeof the variability -[(h-• - ao)/S•o]•, (4)
in the two data setsas explainedby Murphy and Ep- Buncond
KARA ET AL.' OPTIMAL MLD DEFINITION 16,813

Table 3. Comparison
of Levitusand OWS LayerDepths

AT Criterion Winter Spring Summer Fall

0.2øC ILD 24 (73%) 16 (48%) 18 (55%) 15 (45%)


MLD 10 (30%) 14 (42%) 11 (33%) 8 (24%)
0.5øC ILD 33 (100%) 33 (100%) 33 (100%) 26 (79%)
MLD 10 (30%) 11 (33%) 13 (39%) 9 (27%)
0.8øC ILD 29 (88%) 17 (52%) 21 (64%) 16 (48%)
MLD 10 (30%) 11 (33%) 14 (42%) 9 (27%)
1.0øC ILD 29 (88%) 18 (55%) 22 (67%) 16 (48%)
MLD 10 (30%) 10 (30%) 14 (42%) 11 (33%)

The numberof casesby seasonwhen hL > ho for the ILD and


MLD for givenAT values.Foreachcategory
the percentage
is givenin
parentheses.

whereShLho
2 is the covariance
betweenhœand ho'' ShL
2 computedSS and r 2 valuesfor the winter ILD and
and$•o arethesample
variances
of thehL andho MLD valuesbetweenthe Levitusand OWS data (Ta-
values,respectively;and ShEand $ho are the standard ble 4) showthat the skill scoreand linear association
deviationsof hœand ho values,respectively.The mean are greaterfor the MLD. The negativeskill scoresreveal
values for hL and ho are representedby hL and ho, that hœis not a skillfulmeasureof ho for the givencri-
respectively. teria. The statisticalmeasuresgivenin Table 4 indicate
that use of a 0.8 ø or 1.0øC value for AT for the winter
5.1. Determination of the Best AT Value
MLD yields the best agreementbetween hœ and ho.
For winter a lm'gescatterexistsbetweenthe hœand Furthermore,the use of 1.0øC is superiorto 0.8øC be-
ho ILD (Figure8) for all AT values.In contrast,the causethe mseand Buncond valuesare larger for the lat-
MLD agreementis quite goodfor AT _>0.5øC. The ter case. However,the differencesbetweenusingthese

(.h(ot)oh(T)

0 50 100150200 0 50 100150200 0 50 100150200 0 50 100150200

ho (m) ho (m) ho (m) ho (m)

160-[
....-:0-'-2•øC---[
.........
•..r...--.•..........
zX.'..r.._=.•
120.............................................
' z•r=•.0ø•
!........
:•
/•o
............

..........
Q-•
.......
-:........ :........ :..... c9-/-%-•

8o....... .......
'!........................ ......
,:.......
........
0
0 40 80 120160 0 40 80 120160 0 40 80 120160 0 40 80 120160

ho (m) ho (m) ho (m) ho (m)


Figure 8. Prediction of (top) winterand (bottom)springILDs and MLDs at the OWSs
A throughK (ho) (seeFigure1) versusthosefromLevitus(hœ)whenusingAT inter-
vals of 0.2ø, 0.5ø, 0.8ø, and 1.0øC. The numberof cases(h(T) and h(crt))is 33 (i.e.,
11 stationsx 3 months/season)
for eachAT category.Note herethat we useLevituswinter
(i.e.,January,
February,
andMarch)andLevitusspring(April,May,andJune)to describe
seasons. Axis scales are different for each season.
16,814 KARA ET AL.' OPTIMAL MLD DEFINITION

Table 4. SeasonalComparisonsof ILD and MLD

AT Criterion SS r2 rmse Bcond Buncond


Winter
0.2 øC ILD -0.67 0.26 17 0.7514 0.1766
MLD -0.86 0.35 26 0.6446 0.5625
0.5øC ILD -0.34 0.64 18 0.0306 0.9503
MLD O.59 O.83 13 0.1432 0.0947
0.8 øC ILD -0.84 0.49 23 7 0.0618 1.2672
MLD O.7O O.86 115 0.1241 0.0423
1.0øC ILD 0.40 0.55 35 8 0.0003 0.1514
MLD 0.75 0.88 103 0.1110 0.0275
Spring
0.2øC ILD 0.86 0.88 6.2 0.0245 0.0006
MLD 0.53 0.77 10.0 0.2244 0.0150
0.5øC ILD 0.67 0.72 16.8 0.0327 0.0186
MLD O.89 O.9O 7.7 0.0011 0.0107
0.8øC ILD 0.87 0.90 17.1 0.0185 0.0094
MLD 0.91 0.92 8.1 0.0006 0.0159
1.0øC ILD 0.94 0.94 13.9 0.0004 0.0001
MLD 0.90 0.92 8.6 0.0008 0.0181
Summer
0.2øC ILD 0.79 0.81 1.6 0.0165 0.0033
MLD 0.81 O.85 1.5 0.0193 0.0216
0.5øC ILD 0.54 0.59 2.6 0.0319 0.0237
MLD 0.75 0.79 1.9 0.0357 0.0111
0.8øC ILD 0.61 0.71 2.3 0.0465 0.0494
MLD 0.70 0.77 2.2 0.0729 0.0006
1.0øC ILD 0.62 0.71 2.2 0.0353 0.0494
MLD 0.71 0.79 2.1 0.0812 0.0006
Fall
0.2øC ILD 0.72 0.77 8.2 0.0445 0.0137
MLD 0.62 0.81 9.2 0.0484 0.1394
0.5øC ILD 0.70 0.72 9.5 0.0243 0.0016
MLD O.83 O.9O 6.5 0.0231 0.0526
0.8øC ILD 0.87 0.89 6.0 0.0125 0.0001
MLD O.85 O.9O 6.1 0.0141 0.0352
1.0 øC ILD 0.77 0.84 7.9 0.0548 0.0711
MLD 0.79 0.82 7.1 0.0180 0.0338

Comparisons between the ILD and MLD from the OWS


observations
(ha) and from the Levitusdata (ha) obtainedusing
the methodologyin section3 for the 11 stationsby season.Both
the ILD and MLD are included for the same AT. All r values are
statisticallysignificantat 95% confidenceinterval basedon the
Student's t test. Note that a SS of 1 is equivalent to a perfect
match between 2 data sets, while a SS< 0 is an unacceptable
match. The rmse is in meters and the number of cases(n) is 33
for each category.

two AT wouldbe negligible.In general,the MLD gives in the summer,often leadingto the developmentof a
much more consistent hœand ho values than the ILD in seasonalthermoclinewith a shallowmixed layer and a
winter. For the spring, again, a large scatter exists be- stablewatercolumn.The highestSS and r 2 valuesoc-
tweenhL and ho whenusingthe ILD with AT=0.8 and cur for the 0.2øCcriterion(Table4), and the relatively
1.0øC (Figure 8) but only for large layer depths. The small bias and root mean squareerror (rmse) values
SSandr 2 values(Table4) confirm that the hœ(crt)val- confirmthat the best agreementbetweenhœ(T) and
uesare in very goodagreementwith ho(crt)whenusing ho(T), as well as betweenhœ(crt)and ho(crt),occurs
AT-0.5 ø and 0.8øC. Analysisof SS for MLD indicates for this AT value. Use of the density-basedcriterion
that goodagreementexistsbetweenhœ(crt)and ho(crt) is always better than the temperature-basedcriterion
for all AT values. A MLD obtained using AT=0.5 ø or for definingthe mixed layer becausethe SS and r 2 val-
0.8øC is optimal becauseboth of them yield very large ues for the MLD are always larger than for the ILD
SS values(0.89 and 0.91, respectively). for any AT value. The ILD and MLD comparisons for
The summerILD and MLD are very shallowand typ- fall showmore scatter when usingAT=0.2 ø and 0.5øC
icallyhavedepthsof m 25 m (Figure9). This is not sur- than for the larger AT values(Figure9). A compari-
prising becausethe upper oceantemperature increases son of the layer depths between the Levitus and OWS
KARA ET AL' OPTIMAL MLD DEFINITION 16,815

ßh(•t) oh(T
)•

30 i it- 65øc
!
I ! io

• 20 .......
i....
• '-.-::
.......
4........
15

10
•0 •5 20 25 30 •0 •5 20 25 30 •0 •5 20 25 30 10 15 20 25 30

ho(m) ho(m) ho(m) ho (m)

/•r= b.2
ø Xr= b.5
ø
80 ' - --•...... •...... • -

E 60 ......
•........
•---:
• 40 20
........
i
0 20 40 60 80 0 20 40 60 80 0 20 40 60 80 0 20 40 60 80
ho(m) ho(m) ho(m) ho (m)
Figure 0. ThesameasFigure8 butfor Levitussummer
(i.e.,July,August,andSeptember)
andLevitusfall (i.e.,October,
November,
andDecember).

lOO 100
.......Winter tr.." -_
90
.......
Winter
Spring
Summer
Spring
Summer
90

8o Fall Fall - 8O

70 i - 70

6o 6O
50% :' 50%
5o
ß

5O

40 40
't
30 - 30
: /

20 - 20

lO
.,,,-:'
I 10

_ .

I I I I
,.--. I I I I
-60-40-20 0 20 4C 60 -60-40-20 0 20 40 60
h•(T)-ho(T) (m) hœ((jt)-ho((Jt)
(m)
Figure 10. Percentage cumulativefrequencyof (left) ILD (hL(T)-ho(T)) and
(right)MLD (hL(at)-ho(at))differences
between
the Levitusdataandthe OWSdata. Re-
sultsare shownfor eachseasonseparately.The horizontalline represents
the fiftiethpercentile,
whichcorresponds to the medianlayerdepthdifference
for eachseason.
16,816 KARA ET AL.- OPTIMAL MLD DEFINITION

Temperature
Density
Density Density Density(•,) Density(•,)
24.5 25.5 26.5 24.5 25.5 26.5 24.5 25.5 26.5 24.5 25.5 26.5

25 25

5O 5O

75 75

100 100 •

125 125 •

150 150

175 175

200 • • , , 200
5 8 11 14 5 8 11 14 5 8 11 14 5 8 11 14

Temperature
(øC) Temperature
(øC) Temperature
(øC) Temperature
(øC)
Figure 11. Temperatureand densityprofilesconstructed from the monthlyLevitusdata at
station H for the monthsof February,May, August,and November.The layer depth obtained
from the temperature-based criterionILD (hL(T)) is markedwith a solidline on the temperature
profiles,andtheoneobtainedfromthe density-based criterionMLD (hL(at)) is alsomarkedwith
a solidline on the densityprofile. Note that in both criteriaa AT valueof 0.8øCis used.

data basedon r 2 valuesalonesuggests that agreement our method for determininglayer depth producesval-
is always poorer for the ILD than for the MLD. These uesthat agreeto within 20 m or better on averagefor
r 2 valuesreflectthe relativelycloseagreementbetween comparabletemperatureand densityprofiles.
hœ(at) and ho(at). An examinationof all the Bcond
and Buncondvalues along with the rmse and SS values 5.2. Relative Error of the ILD as the MLD
indicates that the best overall agreement between the '
Levitus and OWS data are obtainedwhen usinga MLD Using OWS H as an example,clearly the use of an
with a AT=0.8øC criterion. ILD (h(T)) asa MLD (h(at)) canresultin largeerrors
Given•hat oneof our ultimategoalsis to construct when estimatingwinter MLD. For example,the error is
a MLD climatology for use in model validation, quan- ashighas87 m for the caseshownin Figure11. The dif-
titatively determining the inherent uncertainty in the ferences
for springand fall haverelativelysmallvalues
ILD and MLD obtained using our methodology is im- of 12 and 23 m, respectively,and almost no difference
portant. To determine this, we perform a final analysis in August. The latter is due to the shallow seasonal
for each seasonby combining together by seasonthe thermoclineand pycnoclineand becauseof generally
ILD and MLD differences between the two data sets weak turbulent mixing in the summermonths. To de-
(hœ- ho) for all valuesof AT. Figure 10 showsthe termineif this is a generaltrend when usingthe same
separate cumulative distribution of the ILD and MLD AT values,we extendour analysisto all 11 OWSs and
differencesfor each season. In winter, m 45% of the examinethe differences betweenthe ILD and MLD by
differencesfor the ILD (hœ(T)-ho(T)) are between- seasonusing both the OWS and Levitus data. The scat-
20 and 20 m, while 85% fall within this range for the ter is foundto be verylargefor the winterMLD using
MLD(hœ(at)-ho(at)).
Forthisseason
themedian
val- eitherthe OWS or Levitusdata (Figure12). The large
uesare 21 m (-7 m) for the ILD (MLD). In springand scatter is especiallyapparent for the AT value of 1.0øC
fall the same -20 < hœ- ho < 20 m interval yields when using the Levitus data. In most of the casesthe
percentagevaluesthat are even higher (94 and 98%, 150-200m depthsobtainedfor the ILD are deeperby
respectively)than in winter. These resultsverify that 20-100 m, than thoseof the MLD. For springthe ILD
KARA ET AL.' OPTIMAL MLD DEFINITION 16,817

+ 0.2øC
o 0.5øC
ß 0.8øC
• 1.0øC

200[ ' 200

lOO lOO

50 ++2 .• 50
0 •'; 'i i '; I 0
0 50 100 150 200 0 50 100 150 200
ho(T) (m) h•(T) (m)

160 ........... i......


/............ - 160
-
120

120..............................
80 ...........
;.
...................
:,................. ............. 80

........ •-................ ' ....•'""r


.......... •.........40
•2.........

0 4'0 80 120 160 0 4'0 8'0 120 160


ho(T) (m) h•(T) (m)
Figure 12. Scatterplotof the ILD (h(T))versus the MLD (h(at)) from OWS and Levitusdata
in winter and spring,separately.The OWS and Levitus layer depth are denotedby the O and L
subscripts,respectively.The analysisis shownfor AT valuesof 0.2ø, 0.5ø, 0.8ø, and 1.0øC. The
numberof cases(n) is 33 for eachcategory(11 stationsx 3 months/season).
Note herethat we
useLevituswinter (January,February,and March) and Levitusspring(April, May, and June)
to describe seasons.

is almostequalto the MLD for 0.2øC (Figure 12) and data, the largest differencebetweenthe ILD (h•;(T))
is in closeagreement for the 0.5øC criterion in all but and the MLD (h•;(at)) is only 6 m. Figure 13 shows
a few cases. On the other hand, an ILD estimate for that the layer depths from Levitus often do not agree
the spring MLD using 0.8ø and 1.0øC is still largely an well with those obtained from OWS. This is because
underestimate. While the differences between ILD and
the OWS temperature and density profiles have much
MLD for the Levitus data are larger than for the OWS higher vertical resolution, resulting in better estimates
of the actual MLD over the region.
data, especiallyin winter, this differenceis quite small.
In summer the prediction of the ILD as the MLD does To indicate statistically the agreement between the
not yield anysignificantover-estimation/underestimat- ILD and MLD criteria for both data sets, we use the
ion for both data setswhen consideringall different AT r 2, SS and Bcondand Buncond values.The one excep-
values(Figure 13). Although a little more scatter oc- tion is that we only showthe total dimensionlessbias as
curs in the summer layer depth when using the Levitus the sum of the conditional and unconditional biases for
16,818 KARA ET AL.: OPTIMAL MLD DEFINITION

+ 0.2øC
o 0.5øC
ß 0.8øC
• i.0øC

Summer
(OWS)
I• I Summer
: :
(Levitus)
: :
I • 32
32"

E 24 .i24
: 'O ',
•16

8
i 16
8

0 0
0 8 16 24 32 0 8 16 24 32

ho(T) (m) h,,(T) (m)

Fall(OWS) Fall (Levitus) 'i


•.xl80
...........
i---r----
..... ..........
i......
60

,%40
.......... i ..........
::
' '•'---:- .......... •..........
..........
ii..........
•'.*"•"•ii-'""•'"
!'a : a:
:I..........
•'40
'" .......
'---•--,
, o + :.....
•....
o..........
::..........
20
: : :i: :

0. o
0 2'0 4'0 60 80 o 20 40 60 80

ho(T) (m) h,.(T)(m)


Figure13.Thesame
asFigure
12butforsummer
(July,
August,
andSeptember)
andfall(Oc-
tober,November,
andDecember):

simplicity.Fromthe smallr 2 values(< 0.2) for winter negativeSS valuesagainindicatethat the thermocline
(Figure14), no linearrelationshipexistsbetween the and pycnoclinedo not occurat the samedepth. The
ILD and MLD valuesduring this season. Under the best agreementbetweenthe isothermaland isopycnal
standardStudent'st test [e.g., Wilks,1995]an abso- layersoccursfor summerfor both data sets. The val-
lute valueof at leasto.aa(0.11)is needed for r (r2) uesof r 2 and SS are the highestfor this season,and
to bestatistically
significant
at 95ø76
forn=33. Ther2 the bias valuesare alsonegligible.In fall the biasesare
valuesare always< 0.11 in winter for both data sets modestlylarger than thosefor summer.
and for all AT values(Figures14a and 14d), except Finally, we note that casesmay exist where deter-
AT=0.2øC for OWS. The winter SS have large nega- miningthe ILD and MLD for the sameAT criterionis
tive values,with the largestnegativevalue occurring desirable,eventhoughthe MLD yieldsthe true depthof
for AT=I.0øC (Figures14b and 14e). This indicates the mixed layer. A particularexampleis in determining
that the thermoclineand pycnoclineoccurat different wherebarrierlayersexist (ILD >MLD) or possiblesub-
depthsduringwinterfor the regionof the North Pa- ductionregionsexist (ILD<MLD), asthesehaveimpli-
cificOceanoccupiedby the OWSs. Eventhoughthe cationsfor the oceanheat budget. Casesmay also exist
r 2 valuesarelargefor OWS andLevitusin spring,the where knowing how to obtain an optimal MLD via a
SS valuesare negativefor AT=0.8 ø and 1.0øC. These suitable AT definition for the ILD is useful. For exam-
KARA ET AL.' OPTIMAL MLD DEFINITION 16,819

[ o.øc o.søco.8øc
I -oøc
I
1.0 - 1.0
0.8 -0.8
0.6 0.6
0.4 0.4
0.2 0.2
0.0 i i i 0.0
Winter Spring Summer Fall Winter Spring Summer Fall

1.0
•...(.b.)...O....W._.S.
..................................................... 1.0 ca
0.0 .......................
o.o •..

-4.0 -4.0
-{5.0 -15.0 •'
i i i i

Winter Spring Summer Fall Winter Spring Summer Fall

. (..f)
._L..e..v..i.
tus...................................8.0
8.0.._O...W__..S.
...................................................
't
2.0
4.0•.m,
200

0.0 , , , i i i 0.0
Winter Spring Summer Fall Winter Spring Summer Fall

Figure 14. Coefficient


of determination
(r2), skill score(SS) and bias (unitlessconditional
plus unconditionalbias) valuesfor the layer depth obtainedfrom the ILD versusthe MLD.
The statisticsare illustratedfor layerdepthfrom OWS data (ho(T) versusho(at)) and Levitus
data (ht,(T) versusht,(at)), separately.Note that the metricsare shownfor eachAT valuein
each season.

pie, this information would be especiallyimportant for They are found to be consistentwith each other to
determining the MLD when no salinity measurements within 20 m or better on average. Our statistical anal-
are available,a situation that commonlyoccursin many ysis indicates that overall, the optimal estimate of the
in situ samplingsover most of the global ocean. MLD is obtained using a density-basedcriterion with
AT=0.8øC. Here optimal is defined as the MLD defi-
nition that provides the most consistentagreement be-
6. Summary and Conclusions tween the MLD values of the two data sets. For the
In this paper we introduced a new method for ac- northeast Pacific region the optimal criterion for MLD
curately determining the ocean isothermal layer depth is found to vary seasonally:winter (h(at), AT=0.8 ø
(ILD) and mixed layer depth (MLD) from tempera- or 1.0øC), spring (h(at), AT=0.5 ø or 0.8øC), sum-
ture and density profilesthat can be applied through- mer (h(T), AT = 0.2øC),andfall (h(at), AT = 0.8øC).
out the world ocean. It differs from earlier approaches This seasonaldependencemerely reflects the seasonal
in that here the ILD (MLD) is definedas an abso- changesin the stratification of the upper oceanthat are
lute changein AT (Aat) with respectto an approx- producedby turbulent mixing. During the nonsummer
imately uniform region of temperature(density)just months the surface forcing of winds, moderate heat-
below the ocean surface. The method has been vali-
ing, and strongcoolingproducesa weakerthermocline
and pycnocline,whereasa much sharper stratification
dated by applying it to an observationaldata set of the
northeastPacific (OWS) and a globalmonthly clima- at shallowerdepth developsunder the strong surface
tology (Levitus) and comparingthe resultingMLDs. heating of summer. The seasonaldependenceof the
16,820 KARA ET AL.: OPTIMAL MLD DEFINITION

optimal MLD criteria is therefore not surprising, and Fairall, C. W., E. F. Bradley, J. S. Godfrey, G. A. Wick,
the optimal criteria may vary with other regionsof the J. B. Edson, and G. S. Young, Cool-skin and warm-layer
world ocean. However, if one accepts the universality effectson sea surfacetemperature, J. Geophys.Res., 101,
1295-1308, 1996.
of turbulence theory, and in particular the dependence Huang, R. X., and S. Russell, Ventilation of the subtropical
of buoyant fluxes on thermal stratification, then one North Pacific, J. Phys. Oceanogr.,2•, 2589-2605, 1994.
would expect the overall criterion given aboveto apply. Kantha, L. H., and C. A. Clayson,An improvedmixed layer
The work presentedhere providesresearcherswith some model for geophysicalapplications, J. Geophys.Res., 99,
25,235-25,266, 1994.
guidanceas to the best choiceof criterion for defining
Kelly, K. A., and B. Qiu, Heat flux estimates for the west-
the MLD.
ern North Atlantic, I, Assimilation of satellite data into
The detailed comparisonbetween MLDs inferred from a mixed layer model, J. Phys. Oceanogr.,25, 2344-2360,
1995.
t•voindependent
datasetshasshownthat the inherent
variability of the MLD for any definition only allows for Lamb, P. J., On the mixed layer climatology of the north
an accuracyof 20 rn in 85% of the cases. This implies and tropical Atlantic, Tellus, Set. A, 36, 292-305, 1984.
Large, W. G., J. C. McWilliams, and S.C. Doney, Oceanic
that one can only expect the ocean MLDs from mod- vertical mixing: A review and a model with nonlocal
els and climatologyto agree within the same order of boundary layer parameterization, Rev. Geophys.,32, 363-
accuracy. This is an important limit of uncertainty for 403, 1994.
developersof OGCMs to keep in mind when validating Levitus, S., Climatological atlas of the world ocean, NOAA
their modelsagainstMLD climatologies.Constructinga Prof. Pap. 13, 173 pp., U.S. Govt. Print. Off., Washing-
ton, D.C., 1982.
MLD field using a AT=0.8øC criterion applied to den-
Levitus, S., R. Burgett, and T. P. Boyer, World OceanAtlas
sity profilesmay providea more meaningfuldata set for 199•, vol. 3, Salinity, NOAA Atlas NESDIS 3, 99 pp., U.S.
validation of OGCMs with embeddedmixed layers. For Govt. Print. Off., Washington, D.C., 1994.
this reasonwe have used our optimal definition to con- Levitus, S., and T. P. Boyer, World OceanAtlas 1994, vol. 4,
struct global MLD climatologiesfrom annual, seasonal, Temperature,
NOAA AtlasNESDIS J, 117pp., U.S. Govt.
and monthly global climatologiesof Levitus tempera- Print. Off., Washington, D.C., 1994.
Lewis,M. R., M. Cart, G. Feldman,W. Esaias,and C. Mc-
ture and salinity. We refer to these layer depth clima- Clain, Influenceof penetratingsolarradiationon the heat
tologiesas the Naval ResearchLaboratory Mixed Layer budgetof the equatorialPacificOcean,Nature,3• 7, 543-
Depth (NMLD) climatology. 544, 1990.
Lukas,R., and E. Lindstrom,The mixedlayerof the western
Acknowledgments. The authors would like to acknowl- equatorialPacificOcean,J. Geophys. Res.,96, 3343-3357,
1991.
edgethe contributionsof severalpeopleto this manuscript:
first, R. Brown of the North Pacific Marine ScienceOrgani- Lynn,R. J., andJ. J. Simpson, The CaliforniaCurrentSys-
zation Technical committee and H. Freeland and F. Whit- tem: The seasonalvariability of its physical characteris-
ney of the Institute of OceanSciences,Department of Fish- tics, J. Geophys.Res., 92, 12,947-12,996, 1987.
eriesand Oceansin Canada,for providing11 OWSs (line P) Martin, P. J., Simulationof the mixedlayerat OWS Novem-
data, without which this work would not have been possi- ber and Papa with severalmodels,J. Geophys.Res., 90,
ble, and second,A. J. Wallcraft of the Naval ResearchLab- 903-916, 1985.
oratory (NRL), StennisSpaceCenter, and J. B. Elsnet of McCreary, J.P., P. K. Kundu,andR. L. Molinari,A numeri-
the Department of Meteorology, Florida State University, cal investigation of dynamics,thermodynamics andmixed
for their discussionsand valuable comments. Our special layer processes in the Indian Ocean,Prog. Oceanogr., 31,
thanks go to the reviewersfor their suggestions,which im- 181-244, 1993.
proved the quality of this paper. This work was funded Miller, J. R., The salinity effect in a mixed layer ocean
by the OIfice of Naval Research (ONR) and is a contri- model, J. Phys. Oceanogr.,6, 29-35, 1976.
bution of the Basin-Scale Prediction System project under Millero, F. J., and A. Poisson,Internationalone-atmosphere
program element 602435N and the Dynamics of Coupled equationof stateof seawater,DeepSeaRes.,Part A, 28,
Air-Ocean Models Study under program element 61153N. 625-629, 1981.
SverdrupTechnologyInc. is funded under subcontractfrom Millero, F. J., C.-T. Chen,A. Bradshaw,and K. Schleicher,
the NRL. This is contributionNRL/7323-99-0027 and has A new high pressureequationof state for seawater,Deep
been approved for public release. Sea Res., Part A, 27, 255-264, 1980.
Monterey,G., and S. Levitus,SeasonalVariabilityof Mixed
Layer Depthfor the World Ocean,NOAA Atlas NES-
References DIS 1,i, 100 pp., U.S. Govt. Print. Off., Washington,
D.C., 1997.
Bathen, K. H., On the seasonalchangesin the depth of the
mixed layer in the North Pacific Ocean, J. Geophys.Res., Murphy, A. H., Skill scoresbasedon the mean squareerror
77, 7138-7150, 1972. and their relationshipsto the correlationcoeificient,Mon.
Brainerd, K. E., and M. C. Gregg, Surfacemixed and mixing Weather Rev., 116, 2417-2424, 1988.
layer depths, Deep Sea. Res., Part A, 9, 1521-1543, 1995. Murphy,A. H., The coeificients of correlationand determi-
Chen, D., A. J. Busalacchi,and L. M. Rothstein, The roles nation as measuresof performancein forecastverification.
of vertical mixing, solar radiation, and wind stressin a Weather Forecasting,l O, 681-688, 1995.
model simulation of the sea surface temperature seasonal Murphy,A. H., andE. S. Epstein,Skillscores andcorrela-
cycle in the tropical Pacific Ocean, J. Geophys.Res., 99, tion coeificientsin modelverification,Mon. WeatherRev.,
20,345-20,359, 1994. 117, 572-581, 1989.
Cherniawsky,J., and G. Holloway, An upper-oceangeneral Neter, J., W. Wasserman, and G. A. Whitmore,Applied
circulation model for the North Pacific: Preliminary ex- Statistics,1006pp., Allyn and Bacon,NeedhamHeights,
periments, Atmos. Ocean, 29, 737-784, 1991. Mass., 1988.
KARA ET AL.: OPTIMAL MLD DEFINITION 16,821

Obata, A., J. Ishizaka, and M. Endoh, Global verification Tabata, S., and W. E. Weichselbaumer, An update statistics
of critical depth theory for phytoplankton bloom with cli- of oceanographic
data basedon hydrographic/STDcasts
matological in situ temperature and satellite ocean color made at stations 1 through 6 along line P during January
data, Y. Geophys.Res., 101, 20,657-20,667, 1996. 1959 through September1990, Can. Data Rep. Hydrogr.
Ohlmann, J. C., D. A. Siegel,and C. Gautier, Ocean mixed Ocean $ci., No. 108, 317 pp., 1992.
layerradiant heatingand solarpenetration: A global anal- Tabata, S., and R. M. Brown,Hydrographic/CTD observa-
ysis, Y. Climate, 9, 2265-2280, 1996. tions made during the Ocean Climate Monitoring Study,
Pickard, G. L., and W. J. Emery, Descriptive Physical 1981-1991: A summary of operational phase of study,
Oceanography, 320 pp., Pergamon,Tarrytown, N.Y., 1990. Can. Data Rep. Hydrogr. Ocean $ci., 136, 42 pp., 1994.
Price, J. F., R. A. Weller, and R. Pinkel, Diurnal cycling: Tabata, S., and J. L. Peart, Statistics of oceanographic
Observations and models of the upper ocean responsedi- data basedon hydrographic/STDcastsmade at stations
urnal heating, cooling,and wind mixing, J. Geophys.Res., 7 through 12 along line P during January 1959 through
91, 8411-8427, 1986. September 1990, Can. Data Rep. Hydrogr. Ocean Sci.,
Richards, K. J., M. E. Inall, and N. C. Wells, The diurnal No. 109, 343 pp., 1992.
mixed layer and upper ocean heat budget in the west- Thompson,R. O. R. Y., Climatologicalmodelsof the surface
ern equatorial Pacific, J. Geophys.Res., 100, 6865-6879, mixed layer of the ocean, J. Phys. Oceanogr., 6, 496-503,
1995. 1976.
Roden, G.I., The depth variability of meridional gradients Wagner, R. G., Decadal scaletrends in mechanismscontrol-
of temperature, salinity and soundvelocity in the western ling meridional sea surfacetemperature gradients in the
North Pacific, J. Phys. Oceanogr., 9, 756-767, 1979. tropical Atlantic, J. Geophys.Res., 101, 16,683-16,694,
Schopf,P.S., and A. Loughe, A reduced-gravityisopycnal 1996.
ocean model: Hindcasts of E1 Nifio, Mon. Weather Rev., Welander, P., Mixed layers and fronts in simple ocean cir-
123, 2839-2863, 1995. culation models, J. Phys. Oceanogr., 11, 148-152, 1981.
Skyllingstad, E. D., T. Paluszkiewicz,D. W. Denbo, and Wilks, D. S., Statistical Methods in the Atmospheric Sci-
W. D. Smyth, Non linear vertical mixing processesin the ences,467 pp., Academic, San Diego, Calif., 1995.
ocean: Modeling and parameterization, Physica D, 98, You, Y., Salinity variability and its role in the barrier layer
574-593, 1996. formation during TOGA-COARE, J. Phys. Oceanogr.,
Spall, M. A., A diagnosticstudy of the wind- and buoyancy- 25, 2778-2807, 1995.
driven North Atlantic circulation, J. Geophys. Res., 96,
18,509-18,518, 1991.
Sprintall, J., and D. Roemmich, Characterizing the struc-
ture of the surfacelayer in the Pacific Ocean, J. Geophys.
H. E. HurlburtandP. A. Rochford,
OceanographyDivi-
Res., 10•, 23,297-23,311, 1999.
sion,NavalResearchLaboratory,
StennisSpaceCenter,MS
Sprintall, J., and M. Tomczak, Evidence of the barrier layer 39529.(hurlbutt
@nrlssc.navy.
mil;rochford@nrlssc.navy.
mil)
in the surfacelayer of tropics, J. Geophys.Res., 97, 7305-
A. B. Kara, Advanced SystemsGroup, Sverdrup
7316, 1992.
Technology,Inc., Stennis Space Center, MS 39529.
(kara@nrlssc.navy.
mil)
Sterl, A., and A. Kattenberg, Embedding a mixed layer
model into an oceangeneralcirculationmodel of the At-
lantic: The importance of surfacemixing for heat flux and (ReceivedJuly 19, 1999; revisedMarch 1, 2000;
temperature, J. Geophys.Res., 99, 14,139-14,157, 1994. acceptedMarch 17, 2000.)

You might also like