全球变化知识资源中心

Global ocean sampling of water-column temperature in the twentieth century was spatially and temporally sparse⁵, characterized by strong coverage biases towards the Northern Hemisphere, towards the continental coastlines, and seasonally towards summer. Roughly half a million temperature/salinity profiles to at least 1,000 m were collected by research vessels, mostly in the past 50 years. Additional lower accuracy and shallower temperature-only data have been obtained from commercial and naval vessels. These help to mitigate the coverage deficiencies but raise additional concerns regarding measurement bias errors⁶.

Today the Argo Program² provides systematic coverage of global ocean temperature/salinity from 0–2,000 m using 3,500 autonomous profiling floats spaced about every 3° of latitude and longitude, each providing a temperature/salinity profile every 10 days. Profiling float technology⁷ allows data to be collected without a ship by long-lived free-drifting instruments. Argo has collected 1.2 million temperature/salinity profiles and continues to provide 10,000 profiles per month, with far greater spatial and temporal homogeneity than that achieved historically. Previous investigations of ocean heat content⁵ have combined Argo and historical data of variable quality, and these studies have been impacted by coverage and measurement bias issues. Here we estimate ocean heat gain over the 2006–2013 period for which Argo coverage is global (Methods), and through the exclusive use of Argo data with uniformly high quality.

Argo’s ocean temperature data set is invaluable for estimating the net radiation balance of the Earth. The deduced excess of downward over outgoing radiation⁸ driving global warming is too small to measure directly as radiative fluxes⁹. About 93% of this net planetary energy increase is stored in the oceans¹, a result of the large heat capacity of sea water relative to air, the ocean’s dominance of the planet’s surface area, and the ocean’s ability to transport excess heat away from the surface into deep waters. Using historical ocean temperature data together with modern Argo data the increasing heat content of the upper ocean has been estimated to be in the range 0.3 to 0.6 W m⁻², averaged over the area of the Earth, for periods ranging from the past 135 years¹⁰ to the past 50 (refs 11, 12, 13, 14), 20 (ref. 15), or 8–12 (refs 16, 17, 18) years. These estimates of the rate of ocean heat gain are remarkably similar given the disparate time spans and the potentially large errors due to poor coverage in historical data sets^{9, 18, 19}. It should be noted that errors in these earlier studies are almost as large as the signal.

Although heat gain is measured by the vertically integrated temperature change through the water column, sea surface temperature (SST) is also of interest because it sets the temperature of the base of the marine atmosphere. Global mean SST has increased by about 0.1 °C decade⁻¹ since 1951 (ref. 20) but has no significant trend for the period 1998–2013. Explanations for the recent ‘pause’ in SST warming include La Niña-like cooling in the eastern equatorial Pacific²¹, strengthening of the Pacific trade winds²², and tropical latent heat anomalies together with extratropical atmospheric teleconnections²³. However, it is heat gain and not SST that reflects the planetary energy imbalance and thus the warming rate of the climate system. The high variability of the SST record serves to emphasize that it is a poor indicator of the steadier subsurface-ocean and climate warming signal.

As Argo profiles are randomly distributed, spatial and temporal gridding is required. To demonstrate the robust nature of the signals, three contrasting statistical methods of estimating global heat content patterns from raw Argo profiles are used. An optimal interpolation²⁴ (OI) and a robust parametric fit²⁵ (RPF) are applied to temperature profiles and a reduced space optimal interpolation²⁶ (RSOI) is applied to depth-integrated heat content estimates. Further details are provided in the Methods. In each case, we report anomalies from a mean of data from January 2006 to December 2013.

Global mean SST anomalies from 1998 to 2013 in the NOAA (National Oceanic and Atmospheric Administration) OI SST product²⁷ (Fig. 1) illustrate the large interannual variability that characterizes SST and marine atmospheric temperature. There is no significant trend in the time series for either 1998–2013 (the recent ‘pause’) or 2006–2013 periods. The globally averaged temperature anomaly at 5 m depth from the Argo OI (ref. 24; not used in the NOAA SST product) tracks the SST product closely. As the gridded Argo data set does not include high latitudes, marginal seas, and continental shelves, a more direct comparison of Argo near-surface temperature with the NOAA SST product is made by masking the latter to exclude these same regions (Fig. 1). Differences among the time series show the weak sensitivity in this global metric to Argo’s lack of observations in some regions, such as the Indonesian seas, and to undersampling in others (also Supplementary Figs 1 and 2).

5-m Argo OI temperature (black), NOAA OI SST v2 (ref. 27) masked to the same area as the Argo OI (solid red), NOAA OI SST v2 without the Argo mask (dashed red). All figures are 12-month running means unless otherwise noted.

The 95% confidence intervals reported in the text and figures measure how well the Argo heat content, 12-month running mean time series, is fitted by a linear trend. Apart from the goodness of fit, three sources of error in the interpolated Argo data set are considered: Argo’s less than global spatial domain; Argo’s undersampling of some regions within that domain; and systematic pressure errors in Argo measurements.

Examination of float distributions during early years of Argo implementation (Supplementary Fig. 1) shows only a few floats in the Southern Hemisphere by 2003, and large coverage gaps remaining in 2004 and 2005. Coverage-related errors are assessed using satellite sea surface height (SSH) in relation to Argo steric height. SSH changes are due to ocean mass variation as well as to steric change. First, a time series of SSH anomaly³⁸ averaged over the OI domain (Supplementary Fig. 2, red) is similar to that averaged over the global altimetric SSH domain (dashed red). The trends of these time series from 2006–2012, 0.25 ± 0.09 cm yr⁻¹ and 0.29 ± 0.08 cm yr⁻¹, are not significantly different, nor are those from the SSH trends over the RPF and RSOI domains. The area of the global ocean (3.6 × 10¹⁴ m²) is about 20% greater than the area of the OI, RPF and RSOI domains (3.0, 3.0 and 3.1 × 10¹⁴ m², respectively, Supplementary Fig. 3). As the SSH increase is similar in gridded and non-gridded regions, the heat content integrals for the OI, RPF and RSOI domains are estimated for the global ocean using the ratio of areas (1.2).

With regard to undersampling, particularly while the Argo array was being implemented, a linear regression estimate of de-trended steric height onto de-trended SSH captures 85% of the variance of interannual SSH from 2006 to 2012 (Supplementary Fig. 2). During 2004–2005 the regression estimate based on steric height is visually poorer. It is concluded that Argo coverage was adequate for estimation of global averages in the period beginning in 2006. Examination of the year-to-year distributions of active Argo floats (Supplementary Fig. 1), and the gaps in the array, reinforces this conclusion.

Drift in Argo pressure sensors³⁹ is measured by sampling atmospheric pressure when the float is on the sea surface, and correcting the profile for this drift. One widely used model of Argo float (APEX) had firmware that truncated negative pressure drifts in a substantial number of floats. Effort by Argo data managers has resulted in labelling of cycles whose pressure offset cannot be corrected owing to this truncation. These cycles are excluded from our analysis (Supplementary Fig. 1) to prevent a time-dependent pressure bias and a resulting bias in heat gain. Of the 1.2 million Argo profiles collected so far, 300,000 were excluded from the OI analysis, including 62,000 for uncorrectable pressure drift, 188,000 for other data quality checks²⁴, and 48,000 for location outside the OI gridded region. Similar quality checks apply to the data used in RPF and RSOI estimates. For assessment of global change it is essential that only high-quality data be included.

In the OI (ref. 24) estimate of global ocean heat content anomaly, the temperature anomaly on a 1° latitude/longitude grid with 58 pressure levels is integrated over the constant volume of the gridded ocean (Supplementary Fig. 3), and multiplied by a mean value of density and specific heat capacity. Values from July 2005 to July 2014 are smoothed with a 12-month running mean to form a 2006–2013 series. The RPF (ref. 25) comprises a mean, a trend, seasonal variability and Southern Oscillation Index response, but constrained to Argo data from January 2006 to December 2013. Temperature trends are converted to heat content using a constant salinity of 34.5.

For the RSOI method²⁶, heat content trends are directly estimated using empirical orthogonal functions derived from gridded satellite SSH. Monthly and 1° grid bin-averaged ocean heat content anomalies are mapped using the weighted least-squares procedure previously applied to sea level⁴⁰ and ocean heat content estimation¹². The mapping is applied to Argo steric and ocean heat content anomalies, defined relative to climatologies based solely on Argo observations spanning the periods 2004–2012 and 2006–2012 (ref. 41), with the reduced space covariance derived from dense satellite altimeter sea level anomalies. Reconstructions are performed for all nominal depth levels taking into account bathymetry, and recombined together to extend solutions to shallower and marginal seas.

Argo data are archived by Argo’s Global Data Assembly Center as monthly full data set snapshots with a digital object identifier (DOI) assigned to each.

For the OI analysis, this includes Argo data to July 2014: http://dx.doi.org/10.12770/a9acdd12-9d1b-4961-87ae-b935c4bdcb89.

For the RSOI and RPF analyses, this includes Argo data to December 2013: http://dx.doi.org/10.12770/c1f5c810-a4b8-4f12-9531-3452fbaeb206.

Corrected online 05 February 2015

In the version of this Letter originally published, in the paragraph beginning 'The large interannual variability…' the third sentence should have read: 'The opposing anomalies in the 0–100 and 100–500 m layers are related to El Niño/Southern Oscillation (ENSO) variability in the depth and slope of the equatorial Pacific thermocline³'. This error has been corrected in all versions of the Letter.

1. Rhein, M. et al. in Climate Change 2013: The Physical Science Basis (eds Stocker, T. F. et al.) 264–265 (IPCC, Cambridge Univ. Press, 2013).
2. Gould, J. et al. Argo profiling floats bring new era of in situ ocean observations. Eos Trans. AGU 85, 185–191 (2004).
   • Article
3. Roemmich, D. & Gilson, J. The global ocean imprint of ENSO. Geophys. Res. Lett. 38, L13606 (2011).
   • Article
4. Shindell, D. T. Inhomogeneous forcing and transient climate sensitivity. Nature Clim. Change 4, 274–277 (2014).
   • CAS
   • ISI
   • Article
5. Abraham, J. P. et al. A review of global ocean temperature observations: Implications for ocean heat content estimates and climate change. Rev. Geophys. 51, 450–483 (2013).
   • ISI
   • Article
6. Wijffels, S. E. et al. Changing expendable bathythermograph fall rates and their impact on estimates of thermosteric sea level rise. J. Clim. 21, 5657–5672 (2008).
   • ISI
   • Article
   URL: