This post was updated 3/14/2026 mainly to bring some of the figures up to date, but I also fixed a minor bug in the R code. The data used was updated through 2025, when possible.
By Andy May
As described in my previous post, the ocean “mixed layer” is sandwiched between the very thin “skin” layer at the ocean surface and the deep ocean. The skin layer loses thermal energy (“heat”) to the atmosphere primarily through evaporation, gains thermal energy from the Sun during the day, and constantly attempts to come to thermal equilibrium with the atmosphere above it. During the day, with the Sun beating down on the ocean and calm clear conditions, the skin layer might be as thick as ten meters. At night or under windy conditions, it can be much less than a millimeter thick.
The mixed layer is a zone where turbulence, caused by surface currents and wind has so thoroughly mixed the water that temperature, density, and salinity are relatively constant throughout the layer. Originally, the “mixed layer depth,” the name for the base of the layer, was defined as the point where the temperature was 0.5°C different from the surface (Levitus, 1982). This was found to be an inadequate definition near the poles in the winter, because the temperature there, in certain areas, can be nearly constant to 2,000 meters, but the mixed layer turbulent zone isn’t that deep (Holte & Talley, 2008). Two of the areas that create problems when picking the mixed layer depth with the 0.5°C criteria are the North Atlantic, between Iceland and Scotland and in the Southern Ocean southwest of Chile. These areas are noted with light blue boxes in Figure 1.
Figure 1. The North Atlantic and Southern Ocean areas where defining the mixed layer depth is difficult because of downwelling surface water during the local winter.
The regions shown in Figure 1 are areas where significant downwelling of surface water to the deep ocean takes place, these are not the only areas where this happens, but these areas often contain nearly constant temperature profiles for the upper 1,000 meters, or even deeper. Figure 2 shows the average July temperature profile for the Southern Ocean area in Figure 1.
Figure 2. A Southern Ocean July average temperature profile from the blue region shown in Figure 1. The data used to make the profile is from more than 12 years, centered on 2008. Data from NOAA MIMOC.
As explained by James Holte and Lynne Talley (Holte & Talley, 2008), the deep convection in this part of the Southern Ocean has distorted the temperature profile to such an extent that a simple temperature cutoff cannot be used to set the mixed layer depth. Numerous solutions to the problem have been proposed over the years and these are listed and discussed in their article. Their proposed methodology is used by Sunke Schmidtko to define the mixed layer in the NOAA MIMOC dataset discussed below (Schmidtko, Johnson, & Lyman). The Holte and Talley method is complicated, as are many of the other solutions. It does not seem that a generally accepted methodology for defining the mixed layer has been found to date.
Depending upon location and season, the mixed layer depth changes. It is thickest in the local winter in the higher latitudes. There it can extend to 400 meters below the surface or farther using the Holte and Talley logic, and much deeper using the 0.5°C temperature cutoff. Figure 3 shows a map of the mixed layer depth in January.
However, the Northern Hemisphere mixed layer thickness thins during the northern summer months and it thickens in the Southern Hemisphere, especially in the Southern Ocean surrounding Antarctica, as shown in Figure 4.
The thicker mixed layer zones always occur in the local winter and reach their peak near 60° latitude as seen in Figure 5.
Figure 5. Average mixed layer thickness by latitude and month. The thickest mixed layer is reached in the Southern Hemisphere at about 55 degrees south. In the Northern Hemisphere the peak is reached around 60 degrees north. The mixed layer depth in this plot is computed using the methodology developed by Holte and Talley. Data from NOAA MIMOC.
The significance of the Mixed Layer
In our previous post, we emphasized that the mixed layer is in thermal contact with the atmosphere, with a small delay of a few weeks. It also has about 22 times the heat capacity of the atmosphere, which smooths out the radical changes in atmospheric temperature caused by weather events. Thus, when looking at climate, which is much longer term than weather, observing the trend in mixed layer temperature seems ideal. In Figure 6 we compare the yearly global average mixed layer temperature from MIMOC to three SST datasets. MIMOC uses Argo floats and other deepwater data to build monthly datasets of the mixed layer temperature and depth. These values are not from a specific year, but many years are pulled together to give a very high-resolution (0.5 x 0.5 deg) picture of the modern mixed layer (Schmidtko et al., 2013). Figure 6 compares this temperature to the global sea-surface temperature (SST) estimates from the Hadley Climatic Research Unit (HadSST 4.2), ICOADS version 3, and NOAA’s ERSST version 5. The values plotted are not anomalies but actual temperature in degrees C. The values averaged in figure 6 are corrected by different algorithms and gridded. The ICOADS values are the closest to the measured values, with minimal corrections and represents what they call a “simple mean.” I produced a latitude-weighted average of the respective global grids to account for the cell area difference from the equator to the poles.
Figure 6. The MIMOC global average is shown as a red box. It is compared to HadSST version 4.2 in orange, NOAA’s ICOADS SST in green, and NOAA’s ERSST in blue. NOAA’s MIMOC mixed layer temperature is placed at 2012 arbitrarily as it has no date. Data is from the respective agencies. The gray line is the number of HadSST4.2 observations (right scale). All averages are area weighted by latitude. The ICOADS trend is about three degrees per century, the ERSST trend is two degrees per century and the HadSST trend is one degree per century.
Decent mixed layer coverage is only available since 2004, so the years before then are suspect. The Hadley CRU temperatures are a little higher than the ERSST temperatures probably because of different sample areas. HadSST places nulls in cells with no measurements, especially under sea ice covered areas. ERSST assumes a temperature under the ice and uses extrapolation into areas with no data in order to get a complete grid. Thus, ERSST has more low temperature values in its grid than HadSST which lowers their global average temperature.
NOAA’s ICOADS (International Comprehensive Ocean-Atmosphere Data Set) has an increasing SST trend of around three degrees per century, higher than the trends for HadSST (about one degree per century) and ERSST (about 2 degrees per century). The gray line is the number of HadSST 4.2 SST observations to help judge the data quality for each year. The 25-year period in figure 6 is too short to be called “climate,” but is the period with the best mixed layer data.
The NOAA MIMOC mixed layer temperatures are much lower than the HadSST temperatures, probably due to sampling. Sea surface temperatures are not really the temperature at the surface, their nominal depth is 20 cm, which is below the skin layer most of the time, so these SSTs are normally from the mixed layer. Why do these three popular SST estimates all have different trends since 2000, when the data is the best? I don’t know the answer, they probably agree within the margin of error, which is large since the trends differ by a factor of three.
Sea surface and mixed layer temperatures should not need to be turned into anomalies unless they are compared to terrestrial temperatures. They are all taken at approximately the same elevation and in the same medium. All the datasets are global, with similar input data. All are gridded to reduce the impact of uneven data distribution. The grids all cover the oceans, but they handle observations differently and have different grid cell sizes. The MIMOC multiyear data set is complete and has a 0.5×0.5-degree cell size that is about 55km x 55km at the equator, HadSST 4.2 has a 5 x 5-degree cell which is 556km x 556km at the equator. ERSST is in between with a 2×2 degree cell. Both ERSST and MIMOC use extrapolation and assumptions to fill in their grids completely.
In Figure 7 we have plotted the HadSST and ERSST anomalies. How did they get these anomalies from the measured temperatures in Figure 6? Both anomalies are referenced to the respective average from 1961-1990, which is the native HadSST 4.2 reference.
Figure 7. HadSST version 4.2 and ERSST V5 temperature anomalies. After finalizing the corrections and converting to anomalies, difference between the two global SST estimates is reduced to about a tenth of a degree and the two warming trends are about the same.
Unlike weather stations on land, the ARGO floats and ships that make the basic measurements used in HadSST and ERSST move, there are very few fixed ocean buoys. The grid values averaged in figure 6 are mostly from a constantly changing set of instruments and devices. The same is true in figure 7 and the 1961-1990 reference temperature. It is unlikely that simply converting the temperature measurements to anomalies can account for the changes between figures 6 and 7.
The HadSST record is maintained by John Kennedy and his colleagues at the MET Hadley Climatic Research Unit (Kennedy, Rayner, Atkinson, & Killick, 2019). They note that their record is different from ERSST and admit it is due to the difference in corrections and adjustments to the raw data. Kennedy mentions that SST “critically contributes to the characterization of Earth’s climate.” I agree. Kennedy also writes that:
“One of the largest sources of uncertainty in estimates of global temperature change is that associated with the correction of systematic errors in sea-surface temperature (SST) measurements. Despite recent work to quantify and reduce these errors throughout the historical record, differences between analyses remain larger than can be explained by the estimated uncertainties.” (Kennedy, Rayner, Atkinson, & Killick, 2019)
One glance at Figure 6 verifies this statement is correct. Most of Kennedy’s 90-page paper catalogues the difficulties of building an accurate SST record. He notes that even subtle changes in the way SST measurements are made can lead to systemic errors of up to one degree, and this is the entire estimated 20th century global warming. We do not believe the SST record from 1850 to 2005 is worth much. The ambiguous data sources (mainly ships and buckets to WWII and ship intake temperatures after) and the imprecise corrections swamp any potential climate signal. The data is much better since 2005, but Figure 6 shows wide differences in compilations from different agencies. Next, we review the agency definitions of the variables plotted in Figure 6.
Met Office Hadley Centre HadSST 4 data set.
This data was read from a HadSST NetCDF file. NetCDF files are the way most climate data are delivered, I’ve explained how to read them with R (a high quality free statistical program) in a previous post. The variable read from the HadSST file was labeled “tos,” it is a 5-degree latitude and longitude grid defined as “sea water temperature.” The documentation says it is the ensemble-median sea-surface temperature from HadSST v4.2. The reference given is the paper by John Kennedy already mentioned (Kennedy, Rayner, Atkinson, & Killick, 2019). HadSST uses data from ICOADS release 3, supplemented by drifting buoy data from the Copernicus Marine Environment Monitoring Service (CMEMS). Kennedy mentions the difference between his dataset and ERSST v5 seen clearly in Figure 6. The Hadley Centre SST is corrected to a depth of 20 cm, thus the vast majority of the values measured are from the mixed layer.
NOAA ERSST v5
In Figure 6 we plot yearly global averages of the ERSST v5 NetCDF variable “sst,” which is defined as the “Extended reconstructed sea surface temperature.” They note that the actual measurement depth varies from 0.2 to 10 m, but all measurements are corrected to the optimum buoy measurement depth of 20 cm, precisely the same reference depth as HadSST. Also, like HadSST, ERSST takes its data from ICOADS release 3 and utilizes Argo float and drifting buoys between 0 and 5 meters to compute SST. This makes sense, since ERSST agrees with the University of Hamburg dataset and NOAA’s MIMOC (mixed layer) datasets, which also rely heavily on Argo float data. As discussed above, SST (at 20 cm) and the mixed layer temperature should agree closely with one another almost all the time. The ERSST anomalies plotted in Figure 7 are computed from sst using the HadSST reference period of 1961-1990. The basic reference to ERSST v5 is a paper by Boyin Huang and colleagues (Huang, et al., 2017).
Like Kennedy, Boyin Huang directly addresses the differences between ERSST and HadSST. Huang believes that the differences are due to the different corrections to the raw data applied by the Hadley Centre and NOAA.
NOAA MIMOC
The global average NOAA MIMOC mixed layer “conservative temperature” is plotted in Figure 6 as a box that is very near the ERSST line. It is plotted as one point in 2012 because the MIMOC dataset uses Argo and buoy data over more than 12 years centered on about that year. The global average temperature of all that data is 18.1°C from 0 to 5 meters depth. Conservative temperature is not the same as SST. SST is measured, conservative temperature is computed such that it is consistent with the heat content of the water in the mixed layer and takes into account the water salinity and density. However, we would expect SST to be very close to the conservative temperature. Since the conservative temperature more accurately characterizes the heat content of the mixed layer, it is more useful than SST for climate studies. The primary reference for this dataset is the already mentioned paper by Schmidtko et al.
NOAA ICOADS
The NOAA ICOADS line in Figure 6 is downloaded from the KNMI Climate Explorer and labeled “sst.” The description is: “Sea Surface Temperature Monthly Mean at Surface.” ICOADS version 3 data is used in all the other agency datasets described here, but the organization does not do a lot of analysis. By their own admission they provide a few “simple gridded monthly summary products.” Their line is shown in Figure 6 for reference, but it is not a serious analysis and probably should be ignored. It does help show how imprecise the data is and how dependent the ERSST and HadSST results are on their corrections to the data.
Conclusions
The total temperature spread shown in Figure 6 is nearly 3°C in places and yet these agencies are starting with essentially the same data. This is not an attempt to characterize the SST and mixed layer temperature one-hundred years ago, these are attempts to tell us the average ocean surface and mixed layer temperature today. I have no idea whether HadSST’s or ERSST’s temperatures are correct but the difference between them (about 0.6°C) betrays a large uncertainty even in modern times with the best equipment.
I lean toward the ERSST temperatures since it is infilled using reasonable assumptions and it matches the MIMOC global mean which uses data from multiple years. The difference in actual temperature is easier to explain since the sampled areas are different, but the two warming trends are different also. I suppose that since the ERSST complete grid is covering more of the polar regions than HadSST, and they warm more rapidly, could explain at least part of the difference in trends. We have better data available since about 2004, but obviously no agreed method of analyzing it. When it comes to global warming the best answer is we don’t know.
Given that oceans cover 71% of the Earth’s surface and contain 99% of the heat capacity, the differences in Figure 6 are large. These discrepancies must be resolved, if we are ever to detect climate change, human or natural.
I processed an enormous amount of data to make this post, I think I did it correctly, but I do make mistakes. For those that want to check my work, you can find my R code here.
You can download the bibliography here.
14 thoughts on “The Ocean Mixed Layer, SST, and Climate Change”