The role of local atmospheric forcing on the modulation of the ocean mixed layer depth in reanalyses and a coupled single column ocean model
- 365 Downloads
The role of local atmospheric forcing on the ocean mixed layer depth (MLD) over the global oceans is studied using ocean reanalysis data products and a single-column ocean model coupled to an atmospheric general circulation model. The focus of this study is on how the annual mean and the seasonal cycle of the MLD relate to various forcing characteristics in different parts of the world’s oceans, and how anomalous variations in the monthly mean MLD relate to anomalous atmospheric forcings. By analysing both ocean reanalysis data and the single-column ocean model, regions with different dominant forcings and different mean and variability characteristics of the MLD can be identified. Many of the global oceans’ MLD characteristics appear to be directly linked to the different atmospheric forcing characteristics at different locations. Here, heating and wind-stress are identified as the main drivers; in some, mostly coastal, regions the atmospheric salinity forcing also contributes. The annual mean MLD is more closely related to the annual mean wind-stress and the MLD seasonality is more closely related to the seasonality in heating. The single-column ocean model, however, also points out that the MLD characteristics over most global ocean regions, and in particular in the tropics and subtropics, cannot be maintained by local atmospheric forcings only, but are also a result of ocean dynamics that are not simulated in a single-column ocean model. Thus, lateral ocean dynamics are essential in correctly simulating observed MLD.
KeywordsOcean mixed layer depth Atmospheric forcings Coupled single column ocean model Annual mean Seasonal variability Flux correction
The processes occurring in the upper ocean are predominantly forced by the atmosphere. Heating and cooling at daily to seasonal scales, wind-stress, evaporation and precipitation are the dominant physical processes that act on and interact with the upper layer of the ocean. Vigorous turbulent mixing near the surface by these forces results in a layer of homogeneous temperature and salinity (and thus density); its depth is defined as the ocean mixed layer depth (MLD).
Understanding the physics of the MLD is important for climate dynamics, because the thickness of the mixed layer (ML) modulates its heat capacity and hence its ability to store excess heat from the atmosphere (Godfrey and Lindstrom 1989; Maykut and McPhee 1995; Swenson and Hansen 1999; Peter et al. 2006; Dong et al. 2007; Montégut et al. 2007). In addition, MLD variations over the seasonal cycle or anomalies from it are one of the main processes for exchanging heat between the atmosphere and the deeper oceans. Proper quantification of the ML heat budget is important as it governs the evolution of the sea surface temperature (SST) (Chen et al. 1994; Alexander et al. 2000; Dommenget and Latif 2002; Qui et al. 2004; Dong et al. 2008). In a study using a simple stochastic model, Dommenget and Latif (2008) showed that the low frequency (e.g., decadal) variability in SST is a result of the interaction of the heat stored in the layers below the MLD and the heat in the mixed layer. It thus suggests that decadal SST variability strongly depends on how the MLD varies and interacts with the subsurface layers. According to them the sensitivity of SST to MLD variability maximizes in the midlatitudes. Therefore, seasonal to interannual, and maybe even decadal, variability in SST can be simulated in terms of the local air–sea interactions and proper representation of MLD variability. Mixed-layer dynamics are also important for the ocean’s biological productivity (Fasham 1995; Polovina et al. 1995; Narvekar and Kumar 2006) and acts as a medium for exchange of trace gases between the ocean and the atmosphere (Takahashi et al. 1997; Bates 2001; Sabine et al. 2004).
A number of studies have examined the ML dynamics and demonstrated the variation in MLD at different spatial and temporal scales (Monterey and Levitus 1997; Kara et al. 2000; Cronin and Kessler 2002; Kara et al. 2003; Montégut et al. 2004; Halkides and Lee 2009). Most of these studies were limited to either a small region or at specific locations where in situ data is available, and then extrapolated over a wider domain using some specific techniques. Since the advent of Argo data, reanalysis methods have produced a much more reliable source of representations of the state of the ocean at global scale. It has been shown that in summer the MLD is dominated by entrainment due to wind-induced mixing, whereas in winter surface buoyancy forcing is the main driver (Alexander et al. 2000). These theories represent generalised concepts of MLD variability in the global ocean. However, the relative role of these forcings varies in time and space. This uneven distribution of MLD forcing on the seasonal time scale is not understood in detail over the global ocean. Lorbacher et al. (2006) and Carton et al. (2008) examined the variability in MLD at global scale from observations, but these are sparse over the Southern Ocean. Numerous attempts have been made to understand the ocean surface boundary layer and its sensitivity to atmospheric forcing (e.g., Adamec and Elsberry 1984; Large and Crawford 1995; Kantha and Clayson 1994), but again they have mostly been limited to one location or in a small ocean basin.
The focus of the present study is to examine the mean, seasonal cycle and variability of the MLD over the global ocean and how it relates to the different atmospheric forcings and the upper-ocean stratification. The aim is to understand the driving mechanisms of the MLD in different regions and to what extent these can be understood by the local air–sea interactions. In this we will focus on open ocean regions and do not discuss sea ice regions or shallow coastal ocean regions. Therefore, the MLD and the atmospheric forcing terms will be analysed in ocean reanalysis data and in a single-column ML ocean model coupled to an atmospheric general circulation model (AGCM). The latter model simulation allows insight as to what extent the MLD characteristics can be simulated just by local air–sea interactions. It also allows us to diagnose the limitations of local air–sea interactions assumption by analysing the flux correction terms used in the model to maintain a density profile close to the observed mean profile.
The paper is organized as follows: a detailed description of the reanalysis data used, the coupled model and the method to estimate the MLD is presented in the following section. Section 3 focuses on the observed characteristics of MLD estimated from reanalysis data. In Sect. 4 most of the analysis is repeated for the single-column model simulation. Finally, the results are summarised and discussed in Sect. 5.
2 Data, model and methods
Reanalyses and data from a single-column ocean model are used to understand the characteristics of the global ocean MLD and its relation to local atmospheric forcing. Details of the datasets and methods used in this study are described below.
2.1 Reanalysis data sets
The characteristics of the MLD presented here are partly based on vertical profiles of temperature and salinity from the German contribution to Estimating the Circulation and Climate of the Ocean system (GECCO2) for the period 1948–2011. The data has a spatial resolution of 1 × 1 degrees in longitude and latitude and comprises 50 vertical levels. The synthesis used model assimilation with available hydrographic and satellite data. A complete description of this data is given by Köhl (2014). Two other data assimilation products are also analysed in addition to GECCO2: The Global Ocean Data Assimilation System (GODAS) data (Behringer and Xue 2004) implemented at National Centers for Environmental Prediction (NCEP) for the period 1980–2013 and the Simple Ocean Data Assimilation (SODA v2.2.4) data for the period 1950–2010 (Carton and Giese 2008). In the following analysis we will focus all analyses on GECCO2, because it provides data for a longer time period than the NCEP-GODAS dataset and compared to the SODA dataset the results appear to be slightly less noisy. The reanalyses are a combination of model dynamics and sparse observational data, therefore one can expect biases between the products that arise from different assimilation procedures (Kröger et al. 2012). Overall the results we present in this study are qualitatively the same in all three datasets. Any significant differences are pointed out in the text.
2.2 The coupled single column ocean model (ACCESS-KPP)
A 50-year simulation of a coupled atmosphere–ocean single-column model is used to study the MLD variability over the global ocean. The atmospheric component of the coupled model consists of the Australian Community Climate and Earth-System Simulator (ACCESS) version 1.3 atmospheric GCM, which is similar to the UK Met Office’s Global Atmosphere version 1.0 with the addition of some modifications included by the Centre for Australian Weather and Climate Research (CAWCR). A horizontal resolution of 3.75° longitude by 2.5° latitude (referred to as N48) and 38 vertical levels are applied in this model setup. A detailed description of the atmospheric model is given in Bi et al. (2013).
The ocean component of the coupled model consists of a single-column first-order nonlocal K-Profile Parameterisation (KPP) ML model, which uses the vertical mixing scheme of Large et al. (1994). It comprises 40 vertical levels with the layer thickness increasing exponentially from the surface to 1000 m deep. The single-column ocean model is coupled to the atmospheric model at every ocean grid point of the atmospheric model, and at the time step of both atmosphere and oceanic models (30 min), following Klingaman and Woolnough (2014) and Hirons et al. (2015). The KPP model is very adaptable and flexible; it can be applied to AGCM of any resolution. The atmospheric model provides the surface heat flux, wind-stress and freshwater flux (evaporation minus precipitation or E–P) to the ocean at each time step. The ML model does not represent processes such as horizontal advection or upwelling from below the limited vertical domain. Mixing in the interior (layer bellow the surface) is governed by shear instability, which is modelled as a function of the local gradient Richardson number. A boundary layer depth is determined at each grid point, based on the critical value of the turbulent processes parameterised by a bulk Richardson number. Vertical diffusivity coefficients due to turbulent shear are estimated in the diagnosed boundary layer depth. Mixing is strongly enhanced in the boundary layer in both convective and wind driven situations enabling boundary layer properties to penetrate well into the thermocline. A detailed description of the first order nonlocal KPP fundamentals is given in Large et al. (1994). The ocean model is embedded with varying sea-ice concentration at the higher latitudes. The sea ice variability is simulated by a simple thermodynamical model of melting and freezing of sea ice by local forcings. However, regions with sea ice will not be discussed in this study. A flux correction is applied to the ocean temperature and salinity tendency in order to reduce the climate drift, which is described in the following sub-section.
2.3 Flux corrections
The ACCESS-KPP model computes the diffusivity profiles at each grid point and then estimates the temperature and salinity profiles. Since ocean dynamics (mainly advection) are absent in the model the temperature and salinity drifts away from the observed reference values for a longer run (many decades). So it is necessary to add flux corrections in order to prevent climate drift. The flux correction values are computed such that the mean seasonal cycle of temperature and salinity closely follow the climatology from WOA09 (Antonov et al. 2010; Locarnini et al. 2010). The temperature and salinity flux corrections are obtained in a number of iterations with the ACCESS-KPP model, where temperature and salinity are free to evolve. In each iteration biases between the model variables and the climatological values are then computed and added as an additional forcing to the model’s tendency equation in the next iteration. This process is repeated until the biases between the model variables and the climatological reference values are sufficiently small. The flux corrections vary at each grid point and for each month of year but are state independent and remain the same from 1 year to the next. Characteristics of the flux correction terms are discussed in the analysis section (see Sect. 4.4).
2.4 Method for MLD estimation
A wide range of criteria exists for defining MLD from the vertical profiles of temperature, salinity or density. Generally, these criteria fall into two categories: the difference criteria and the gradient criteria. The difference criteria define the MLD as the depth where the oceanic property has changed by a critical value from a reference depth near to the surface (Kara et al. 2000; Montégut et al. 2004). The gradient criteria detect the shallowest depth where the vertical gradient of the oceanic tracers exceeds a given value (Brainerd and Gregg 1995; Lorbacher et al. 2006). The difference criterion largely depends on the difference value chosen and produces misrepresentation of the MLD in regions of weakly stratified surface layers. The gradient criteria also depend on the threshold value chosen for the derivative, but since the gradient is expected to be large at the base of the MLD, it gives more accurate estimations of MLD. Lorbacher et al. (2006) further developed the gradient criteria by adding the standard deviation threshold to assume the shallowest extreme curvature. As a first guess, the nearest MLD is assumed at the local maximum/minimum of the second derivative of the gradient and where the gradient exceeds 0.25 of gradient maximum of the profile. As a secondary boundary condition, 30-m standard deviation (>0.02) is included to distinguish the near-homogeneous region and a standard deviation of gradient over the upper level is then set (0.004 km−1 for high and 0.002 km−1 for low resolution). A more accurate estimation of MLD is provided by an exponential interpolation method (for thick layers), and hence the Lorbacher et al. (2006) criteria is used to estimate the ML from the density profiles for all the data in this study. The results in this paper also hold when difference criteria are used for estimating MLD. We checked with difference in density of 0.03 kg/m3 and variable density criteria (Montégut et al. 2004) that results shallower or deeper ML than that obtained by Lorbacher et al. (2006). However, the spatial pattern remain fairly similar for all the criteria.
3 Observed MLD
- Extra-tropical-seasonal Regions where the standard deviation of the seasonal cycle is larger than 60 % of the annual mean MLD (Fig. 3a). The criteria (60 %) cover the regions where significant seasonal variability occurs. These are regions with a very shallow warm-season MLD and a substantially deeper cold-season MLD. This is mostly found in a band along 30°–40° with some regions extending further to higher latitudes. The most extreme seasonality is seen in the Southern Ocean in the Indian and Pacific sections, which are within the Antarctic Circumpolar Current (ACC) (Sallée et al. 2006; Dong et al. 2008). The far northern Atlantic is also a region of extreme seasonality.
Constant-deep Regions with an annual mean MLD > 30 m and a standard deviation of the seasonal cycle that is <60 % of the annual mean MLD (Fig. 3b). The 30 m selection criteria roughly mark half-standard-deviation away from the global mean MLD. These are mostly the off-equatorial tropical and subtropical regions, but also large fractions of the Southern Ocean and parts of the far northern Pacific.
Constant-shallow Regions with an annual mean MLD < 30 m and a standard deviation of the seasonal cycle that is <60 % of the annual mean MLD (Fig. 3c). These mark mostly coastal regions and in particular upwelling regions along major continents and equatorial regions, but also coastlines in the north-western North Atlantic and west of Australia. Most of the North Indian Ocean falls in this shallow regime. But in NCEP-GODAS much of the North Indian Ocean falls in the constant-deep regime. It should be noted that MLD in NCEP-GODAS (annual mean is ~52 m) is deeper than the other two reanalyses data (~41 m in GECCO2 and SODA). The difference could be associated with different model assimilation procedures (Kröger et al. 2012).
The following analysis focuses on how these different mean MLD regimes relate to the local atmospheric forcing and the stratification of the upper ocean.
3.1 The annual mean MLD and its relationship with local forcing
The annual mean net heat flux (NHF) in the higher latitudes (>40°) is mostly negative, which implies that the ocean is losing heat (Fig. 4a). Heat loss at the surface makes the upper ocean statically unstable, which tends to deepen the ML. The upwelling regions along the coastlines and the equator and in the tropical Indian Ocean are marked by annual mean heat gain, which tends to support shallow MLD. Overall there is only a moderate match between the annual mean NHF forcing and the MLD (spatial correlation of −0.3). Most of the constant-shallow MLD regime seems to be associated with annual mean heat gain. Also many of the extra-tropical regions with deeper annual mean MLD are associated with corresponding heat loss. However, many regions with shallower than global mean MLD are associated with annual mean heat loss, which does not support the observed shallow MLD (e.g. coastal regions in the western North Pacific and Atlantic and southern subtropical regions in the Indian and Pacific Oceans). Also regions with deeper than global mean MLD are associated with annual mean heat gain, which does not support the observed deep ML (e.g. off-equatorial tropical Pacific and parts of the Southern Ocean).
The annual mean wind-stress forcing (Fig. 4b) shows a slightly better match to the annual mean MLD than NHF (spatial correlation of 0.4). In particular, higher-latitude regions with deep annual mean ML are associated with strong mean wind-stress and equatorial regions with shallow annual mean MLD are associated with weak mean wind-stress. However, there are also large regions where the annual mean MLD is deep despite fairly moderate annual mean wind-stress (e.g. latitude bands around 30° in both hemispheres) and regions where the annual mean MLD is shallow despite fairly strong annual mean wind-stress (e.g. subtropical South Indian Ocean and the Arabian Sea). One can also notice that for most regions NHF and wind-stress have similar influences on the MLD, with both either supporting deeper ML (e.g. most extra-tropical regions) or supporting shallower ML (e.g. equatorial regions). Notable exceptions here are parts of the subtropical regions, parts of the Southern Ocean, the far northern Pacific, the western North Atlantic and the Arabian Sea.
The precipitation and evaporation (E–P) plays in general a minor role in mixing processes in the upper ocean by changing the density structure (Dong et al. 2009), which is also highlighted here by the mismatch of the annual mean E–P (Fig. 4d) with the annual mean MLD for most regions (spatial correlation of 0.0). The mean E–P forcing is also mostly unrelated to the NHF and wind-stress forcings. In some regions, however, where both NHF and wind-stress forcings do not match the MLD, the E–P forcing does indeed seem to match the MLD. These appear to be mostly smaller and coastal regions (e.g. coastal regions of Greenland and some parts of the western boundary of the North Pacific and Atlantic).
Remarkable are the shallower MLD regions in the subtropical Indian and western Pacific Oceans. Here, all three atmospheric forcings would support deeper annual mean MLD. The shallower MLDs here seem to match the stronger upper-ocean stratifications, displayed in Fig. 4c (spatial correlation of −0.5) in terms of the density gradient between the surface and 100 m depth. Although, upper-ocean stratification is partly a reflection of the atmospheric forcings, it is also influenced by oceanic processes. Since the annual mean atmospheric forcings seem to mismatch the mean MLD in these regions, it suggests that oceanic processes independent of the local atmospheric forcing contribute to the stratification of the upper ocean and thus a shallower MLD.
3.2 The MLD seasonal cycle and its relationship with local forcing
3.3 Anomaly variability and its relationship with local forcing
Next, it is interesting to regard the MLD variability beyond the seasonal cycle. The second column of Fig. 1 shows the standard deviation (stdv) of monthly mean MLD anomalies estimated for each season. The patterns of strong and weak MLD stdv match very well the patterns of deep and shallow mean MLD (first column in Fig. 1). Thus, MLD variability is stronger when the mean MLD is deep and MLD variability is weak when the mean MLD is shallow. This pattern even holds when we consider the coefficient of variance (CV; the ratio of stdv over the mean) as shown in the last column of Fig. 1. The equatorial Pacific and Atlantic, however, mark regions where the MLD variability is strong even though the mean MLD is fairly shallow, which is consistent with the study by Lorbacher et al. (2006). In these regions the concept of MLD may not be so useful, as they are more dominated by lateral ocean wave dynamics in the thermocline. It is also remarkable to note that a CV > 100 % is observed in the spring and winter seasons in the higher latitudes of North Atlantic and Pacific and in the Southern Ocean. This suggests a substantial amount of MLD variability. For most other regions the CV values are larger than 30 %, indicating significant MLD variability for most regions in most seasons. Keerthi et al. (2012) have also shown large CV values (>40 %) in the western and eastern equatorial Indian Ocean during boreal summer.
The cross-correlations of MLD with both NHF and wind-stress are mostly on a similar level, with some more widespread and clearer correlation for the NHF with MLD. Cross-correlations of MLD with E–P are in general weaker than those with NHF and wind-stress, but are mostly positive, which is consistent with increased evaporation (salinity forcing) leading to deeper MLs (see SFig. 1). This relationship is strongest in fall and winter seasons in both hemispheres.
4 The ACCESS-KPP model
We now take a look at the MLD and its relation to the atmospheric forcings in the ACCESS-KPP model with a single-column ocean ML model. Thus, in this model the simulated MLD is purely a result of local atmospheric forcing, vertical mixing and a heat and salinity flux correction, which is independent of the background state, to prevent the model from drifting away from the observed mean density profiles.
4.1 The annual MLD and its relationship with local forcing
The relationship between the atmospheric forcings and the MLD in the ACCESS-KPP model is also similar to the observed (compare Figs. 4 and 10). Again the wind-stress forcing tends to be more strongly related to the annual mean MLD than the heat flux forcing and again the E–P forcing has no relationship with the annual mean MLD.
4.2 The seasonal cycle of MLD and its relationship with local forcing
As stated above, the model is able to reproduce the observed seasonal changes reasonably well (compare Figs. 2b and 12b): it essentially captures the zonal bands of the strong extra-tropical-seasonal MLD regime in the midlatitudes. Seasonal amplitude is weaker in the tropics, subtropics and also in parts of the Southern Ocean. As already indicated in the previous section, the ACCESS-KPP model tends to overestimate the seasonality in the eastern North Pacific and the western coastal region in the North Atlantic. The model is good at representing the mean summer MLD, but overestimates MLD during fall and spring. This however, cannot be due to stronger seasonality in the atmospheric forcings, as they appear to be similar or weaker than observed. This suggests that the ACCESS-KPP model is missing some ocean dynamics necessary to capture these features.
The seasonal cycle of the atmospheric forcings is well captured by the ACCESS-KPP model and again the relationship between the heating and wind-stress forcings and the MLD are mostly as observed (compare Figs. 5 and 11). Thus, the seasonality in the NHF is the main forcing for the seasonal cycle of the MLD in ACCESS-KPP model.
4.3 Anomaly variability and its relationship with local forcing
The overall strength and seasonal distribution of the anomaly variability of the MLD in the ACCESS-KPP model is very similar to the observed (compare Figs. 1 and 9). The stdv of the MLD is mostly strong where the mean ML is deep, but it follows the mean MLD even more closely than observed. Thus, the coefficient of variance is more homogenous in the ACCESS-KPP model.
As in the observations the effect of the anomalous wind stress forcings is stronger in the warmer, shallower MLD seasons. The lag-1 correlation of the heat flux is also stronger in the cold, deep ML seasons as in the observations. The E–P also shows higher correlations with MLD (see SFig. 1 and 2) than observed, but still much weaker than the correlations with heat flux and wind stress, suggesting that the E–P forcing is not as important.
4.4 Missing dynamics and the role of flux corrections
5 Summary and discussion
In this study, the spatial and temporal relationship between the local atmospheric forcing and the mean, seasonal cycle and variability of the MLD are analysed using a recent ocean reanalysis product (GECCO2) and a single-column ocean model (KPP) coupled to an atmospheric GCM (ACCESS1.3) over the global ocean. The aim of this study was to understand the driving mechanisms of the MLD in different regions and to what extent these can be understood in terms of local air–sea interactions. The focus was on the MLD characteristics of the annual mean, the relative seasonal cycle strength and the variability anomalous from the seasonal cycle. To further simplify the fairly complex characteristics three main regimes were introduced based on the characteristics of the annual mean and the relative seasonal cycle strength of the MLD.
The annual mean MLD over most open ocean regions (away from sea ice and shallow coastal regions) follows the wind and heat forcing. Regions with stronger mean wind forcing tend to have larger annual mean MLD and regions with annual mean net heating tend to have shallower annual mean MLD. The annual mean wind forcing strength appears to be most strongly related to the annual mean MLD and stronger than the net heat forcing. Both of these forcing, however, show spatial correlations less than ~0.5 with the global pattern of the annual mean MLD, indicating that the relationship is more complex. The large flux correction needed to prevent model drift signifies that ocean dynamics are also important for simulating MLD.
Remarkable are a few regions, in which neither of the atmospheric forcings seem to match the annual mean or seasonal cycle characteristics of the MLD. The shallower MLD regions in the subtropical Indian and western Pacific Oceans do not seem to relate to any of the three atmospheric forcing. The western boundaries of the North Pacific and Atlantic have very strong seasonal cycle in NHF, but not in the MLD. Also, the very strong seasonality in the Indian and Pacific Oceans’ ACC regions are not directly related to the seasonality in either of the atmospheric forcings. Since the atmospheric forcings seem to mismatch the MLD characteristics (in the reanalysis datasets) in these regions and the single-column ocean model is also not able to simulate these characteristics, it suggests that oceanic processes independent of the local atmospheric forcing contribute to the stratification of the upper ocean and thus the MLD in significant ways.
The annual mean E–P forcing is for most regions not related to the annual mean MLD, suggesting it is not a dominating forcing. However, in some regions, where both annual mean NHF and wind-stress tendencies do not match the MLD, the E–P forcing does indeed seem to play a role for the annual mean MLD (e.g. coastal regions of Greenland and some parts of the western boundary of the North Pacific and Atlantic).
The relative seasonal cycle strength of the MLD is also mostly related to the relative seasonal cycle strength of the net heating and wind stress, but here the relationship with the net heating is stronger than with the wind forcing. Thus, the relative seasonal cycle strength of the MLD mostly follows the strength of the seasonal cycle in the net heating.
Extra-tropical-seasonal This regime of the MLD in the midlatitudes is primarily a result of the strong seasonal cycle in the heating. The seasonal cycle in heating is indeed strongest in the midlatitudes and matches the strongest relative seasonal cycle strength of the MLD fairly well. In this regime, however, the atmospheric storm track positions also play a major role as well as oceanic fronts and strong boundary currents that are not simulated in the model.
Constant-deep The regime, where the ML is fairly deep (>30 m) throughout the year with no strong seasonal cycle, is mostly found in the subtropics and some high latitudes. These are regions with no strong annual mean heating nor cooling, no strong seasonal cycle in heating and the winds tend to be slightly less than the global mean.
Constant-shallow Shallow MLDs without strong seasonal cycles are mostly found in coastal regions, equatorial regions and the tropical Indian Ocean. Some of these regions are fairly well related to the local forcings, such as coastal and some equatorial upwelling regions. However, much of this constant-shallow MLD regime is not well matched with the local forcings and is not well simulated in the ACCESS-KPP simulation, suggesting that local forcings and processes are not sufficient to explain these MLD characteristics. In particular, in the higher latitudes coastal regions, the central ocean basin equatorial regions and in the Indian Ocean the constant-shallow MLD regimes are not matched to local forcings or processes.
The overall strength of the MLD variability is in general proportional to the mean MLD, with stronger variability associated with a deeper mean MLD (e.g. cold season). The MLD variability appears to be quite significant with values of 30 % to more than 100 % of the seasonal mean MLD values.
Most of this is well captured by the ACCESS-KPP model, but in some regions this is not well simulated by the model, including the equatorial Pacific. In higher latitudes the model does not simulate as strong MLD variability relative to the seasonal mean MLD as observed.
MLD variability is mostly negatively correlated with heat flux variability and positively with wind stress variability. In the mid to higher latitudes the relationship to heating tends to be stronger in the cold (deep ML) seasons and the relationship to wind forcing tends to be stronger in the shallower MLD (warm) seasons. In seasons with deeper mean MLD the relationship to the atmospheric forcings tends to become stronger when the atmospheric forcings lead by about 1 month, which seems to be consistent with a larger inertia of the MLD when it is deeper.
The ACCESS-KPP model is consistent with these observed relationships between the local atmospheric forcings and the MLD, but the relationships are significantly stronger in the model simulations. This may to some part reflect the simplification of the model to only local vertical mixing processes and neglecting other ocean dynamics, which leads to an overestimate of the role of local forcings on the MLD variability. But it may also point to errors in the reanalysis datasets that lead to inconsistencies between atmospheric forcings and the ocean state that artificially degrade the relationship between local forcing and the MLD.
While overall the local forcing perspective provides a fairly good representation of most of the MLD characteristics in the global oceans, it does have some limitations as mentioned above. This is in particular highlighted by the fact that the ACCESS-KPP model does require significant flux correction in temperature (most importantly) and also in salinity. These are important to maintain upper-ocean stratification close to observations. Without such terms in the ACCESS-KPP model, the upper-ocean stratification in most subtropical regions would collapse after about 10 years and the MLD would deepen to the base of the ACCESS-KPP model (1000 m). Thus, with the flux correction terms, the coupled single-column model allows to study the upper ocean–atmosphere interaction with higher near-surface vertical resolution incorporated with a better vertical mixing. This coupled framework is computationally less expensive and allows identification of the role of atmospheric forcings for the upper-ocean processes.
The analysis has shown that the MLD characteristics arise from complex interactions between the local forcings, ocean stratifications and potentially lateral ocean dynamics. Most of the results discussed here result from comparisons of the overall statistics of MLD and local forcings in observations and in the ACCESS-KPP simulation. To further untangle the interactions and the relative contribution of different forcings and different oceanic processes it would require sensitivity experiments with the ACCESS-KPP or similar model simulations, in which forcings or elements of the forcings are turned ‘off’ or in which processes are altered or turned ‘off’. However, this is beyond this study and is left for future analyses.
The authors would like to thank Australian National Computational Infrastructure, in Canberra, for providing computational platform for simulation of the ACCESS-KPP coupled model. The ARC Climate System Science (CE110001028) supported this study. Nicholas Klingaman was funded by the National Centre for Atmospheric Science-Climate, a collaborative centre of the Natural Environment Research Council, under agreement R8/H12/83/001.
- Antonov JI, Seidov D, Boyer TP, Locarnini RA, Mishonov AV, Garcia HE, Baranova OK, Zweng MM, Johnson DR (2010) World ocean atlas 2009, vol 2, salinity. In: Levitus S (ed) NOAA Atlas NESDIS 69, U.S. Government Printing Office, WashingtonGoogle Scholar
- Behringer DW, Xue Y (2004) Evaluation of the global ocean data assimilation system at NCEP: the Pacific Ocean. In: Eighth symposium on integrated observing and assimilation systems for atmosphere, oceans, and land surface. Washington State Convention and Trade Center, Seattle, pp 11–15Google Scholar
- Bi D, Dix M, Marsland SJ et al (2013) The ACCESS coupled model: description, control climate and evaluation. Aust Meteorol Oceanogr J 63:41–64Google Scholar
- Dong S, Sprintall J, Gille S, Talley L (2008) Southern Ocean mixed layer depth from Argo float profiles. J Geophys Res Oceans. doi: 10.1029/2006JC004051
- Kantha L, Clayson C (2000) Small scale processes in geophysical fluid flows. International geophysics series, vol 67. Academic Press, New York, pp 157–160Google Scholar
- Locarnini RA, Mishonov AV, Antonov JI, Boyer TP, Garcia HE, Baranova OK, Zweng MM, Johnson DR (2010) World ocean atlas 2009, volume 1, temperature. In: Levitus S (ed) NOAA Atlas NESDIS 68, U.S. Government Printing Office, WashingtonGoogle Scholar
- Lorbacher K, Dommenget D, Niiler P, Köhl A (2006) Ocean mixed layer depth: a subsurface proxy of ocean atmosphere variability. J Geophys Res Oceans. doi: 10.1029/2003JC002157
- Montégut C, Madec G, Fischer A, et al (2004) Mixed layer depth over the global ocean: an examination of profile data and a profile based climatology. J Geophys Res Oceans. doi: 10.1029/2004JC002378
- Monterey G, Levitus S (1997) Seasonal variability of mixed layer depth for the World Ocean, NOAA Atlas NESDIS 14:100 p. Natl. Oceanic Atmos. Admin. Silver Spring, MdGoogle Scholar