Recent changes in global mean surface temperature (GMST) or sea surface temperature (SST) change have been heavily debated in the past ten years, because of an occurrence of a period (1998–2013) with a slower increase in both the GMST and SST compared to the long-term trend (i.e., since 1951) or than the previous decade (1983–1997) [1
] (Figure 1
). This phenomenon has been termed as a so-called “global warming hiatus” or “surface temperature hiatus” in many literature references [2
]. The use of these terms immediately raises a question. Does it indicate the slowdown of the human-induced global warming driven by the accumulation of greenhouse gases (GHG) in the climate system since industrial revolution, which is well established in the climate community [9
It is important to articulate a clear meaning to the “hiatus” description. In the scientific community, it generally describes a change in the range of surface warming that is outside the bounds of natural fluctuations superimposed upon a long-term trend. The requirement of statistical significance allows research tools to be used to determine whether or not some slowdown has occurred. Over the past few years, many studies have looked at this very issue. For instance, Karl et al. [11
], discovered artifacts in temperature measurements that, when accounted for, showed a significant increase in warming rate. Among the artifacts were differences in ship-based and buoy-based sensors and accounting for these differences. In addition, historical changes in the SST, in particular changes to measurement techniques circa 1940s were quantified and addressed. Finally, Karl et al. [11
] used an updated land-temperature dataset. The end result of these corrections was that over the preceding 17 years, GMST had increased without a halt, and in fact, had increased at a steady rate.
A series of other studies applied multiple and independent statistical tests to the GMST record and demonstrated, conclusively, that there is no statistically significant decrease in the rate of surface warming and that the rate of warming within the past two decades is not unusual in a continuously warming world [12
]. The importance of differing definitions of a “hiatus” was discussed in Medhaug et al. [16
] where contradictory conclusions can arise in the literature based on the definition. Consequently, elucidating the various different definitions of “pause” or “hiatus” is important.
Among the common definitions of pause/hiatus are: (1) a statistically significant change in the rate of global warming, as measured by changes to the heat balance of the planet; (2) a statistically significant change in the surface temperature record; (3) a non-statistically significant change in the rate of GMST change; and (4) Divergence between GMST predictions (from climate modes) and actual GMST measurements. Unfortunately, these definitions are often conflated and their separate identities must be maintained.
So, has there been a pause in global warming? The answer would be mistakenly “yes” only if one defines the “global warming” only by GMST changes (definition 3 above). Fundamentally, the global warming is caused by the Earth’s energy imbalance (EEI). During equilibrium climate states, the amount of incoming solar radiation is balanced by the outgoing radiation at the top of the atmosphere (TOA). However, increases in GHGs in the air trap more energy within the climate system, thus creating an imbalance of the Earth’s energy [17
]. More heat available in the climate system is manifested in many ways including increasing the GMST/SST [19
], increasing the ocean/land interior temperatures [20
], raising the sea level [21
], melting the ice sheets and permafrost [22
], altering the hydrological cycle [23
], changing the atmospheric and oceanic circulation [24
], supporting stronger tropical cycles with heavier rainfall [25
], among other “symptoms” of the global warming [9
Due to its large heat capacity and huge volume, the ocean has a greater capability to store heat than the other components of the earth system (i.e., land, atmosphere). Actually, more than 90% of the EEI has been stored in the ocean [26
]. This heat imbalance is manifested by the increase of ocean heat content (OHC). Therefore, from the energy point of view, the most fundamental metric for global warming is EEI and OHC [17
]. Also OHC is a robust metric for global warming since it is much less impacted by the natural fluctuations of the climate system than is GMST on inter-annual scales [27
Many studies have already shown that there is no slowdown in OHC records based on observational datasets [29
], reanalysis products [31
], and model simulations [32
]. Instead, as presented in Figure 1
, the rate of OHC increase is larger during the 1998–2013 period (0.98 ± 0.16 × 1022
J/yr, ~0.61 W m−2
averaged over the Earth’s surface) than the 1983–1997 period (0.70 ± 0.17 × 1022
J/yr, ~0.43 W m−2
) and 1955–1997 period (0.30 ± 0.07 × 1022
J/yr, ~0.19 W m−2
), consistent with the increase of GHG concentration in the atmosphere [10
] (The 95% confidence interval for the linear trend is provided according to Foster and Rahmstorf [34
]). Therefore, “global warming hiatus” to indicate a decrease in warming is misleading and has been misinterpreted; there is no slowdown in the global warming nor any decrease in the energy imbalance of the planet. It would be better to name the slowdown in the rate of GMST/SST during the 1983–1997 period as “surface warming slowdown” (SWS hereafter), which will be used in this study. And it is imperative to establish multiple climate indicators besides of GMST/SST as collected in last IPCC report [9
With continuous warming of our climate, why was there a non-statistically significant slowing of the GMST/SST increase during the 1998–2013 period? The leading hypothesis is that this SWS was regulated by the natural variability generated near the air-sea interface, such as the Interdecadal Pacific Oscillation (IPO) or Pacific Decadal Oscillation (PDO) [35
] in the Pacific Ocean and the Atlantic Multi-decadal Oscillation (AMO) [37
] in the Atlantic Ocean. Some of the previous model-based studies reported that the tropical Pacific was the main “pacemaker” as the stronger trade winds in the central and eastern Pacific increased the cold water upwelling in the tropical eastern Pacific, which cools the local sea surface, and increases the warm water penetration into the ocean subsurface as the subtropical cell is strengthened [2
]. The GMST/SST cooling induced by the tropical Pacific cooling offsets the warming effects in the other oceans, leading to weak warming rates [8
]. However, other studies reported the importance of Atlantic and Southern Oceans [39
], arguing that the sea surface slowdown is caused by heat transported to deeper layers driven by the change of the Atlantic Meridional Overturning Circulation. A detailed overview of the potential drivers of the SWS can be found in Liu et al. [40
Many of these hypotheses relate surface changes (GMST/SST) to the ocean subsurface changes (OHC), as increased ocean heating driven by GHG must be sequestered in the ocean interior if it does not appear in the near surface layers (represented by GMST/SST). And then an outstanding question arises: where is the heat? What is the relationship between GMST/SST and OHC on decadal scales? This question is yet to be fully answered and a vigorous scientific debate has emerged. Palmer and McNeall [41
] suggested that global OHC changes on a decadal scale is out of phase with GMST in the simulations of the fifth phase of the Coupled Model Inter-comparison Project (CMIP5). And many other studies tried to identify local OHC “hot spots” and hypothesized their role in global GMST/SST changes [32
]. However, contradictory conclusions were drawn by these studies, highlighting different ocean basins for SWS. The contradictions in previous OHC-related studies is tied to two issues:
Uncertainties in OHC products, which are not fully accounted for in previous studies related to SWS. They were substantial differences among OHC datasets [26
], which is one reason for the debate. Recently, progress has been made to understand the error in OHC estimates and improve the OHC record [29
]. This progress will be discussed in Section 3
and will allow for a better identification of the OHC change during the SWS period.
On a decadal scale, natural variability in OHC records are mixed with forced changes by GHGs (manifested by a long-term warming trend in OHC), aerosols, ozone, volcanoes (manifested by a several-years decrease in OHC records) etc. [32
]. Therefore, one has to separate the natural variability related to SWS from other changes such as a long-term anthropogenic warming signals. One method is to use climate models [32
], but the short-coming of this approach is the model error at the ocean subsurface [54
], resulting in some inconsistency among model-based studies [43
]. This study used a simple method accounting for the forced changes (will be introduced in Section 2
) the results will be shown in Section 3
This study is organized as follows: data and methods will be introduced in Section 2
. In the results section (Section 3
), we will first revisit the improvements to OHC estimates and show the impact of OHC errors in investigating the surface warming hiatus (Section 3.1
). Then we will quantify OHC changes based on improved ocean observations during the SWS period (1998–2013) and compare them with the 1982–1997 reference period. This decadal scale OHC change will also be compared with the long-term ocean warming (i.e., OHC change for 1955–2017). These comparisons will highlight the distinct pattern of the decadal ocean heat redistribution from the long-term trend (Section 3.2
). In Section 3.3
, potential links between the key modes of climate variability and OHC on decadal scales are investigated. Implications for the drivers of the ocean heat redistribution during the SWS period will be given. The conclusion and discussion are presented in the final Section 4
2. Data and Method
An observational ocean temperature product from the Institute of Atmospheric Physics (IAP) [29
] is used in this study. The IAP product has advantages in both instrumental bias correction (necessary to ensure high-quality observations) and mapping method (to provide a homogenous product with complete global ocean coverage). The instrumental bias refers to systematic errors in Expendable Bathythermograph (XBT) data [58
]. It was clear that the major uncertainty in OHC record comes from two sources: bias corrections for XBT data and different choices made in the mapping method [48
]. The impacts of these two advances on the SWS discussion will be discussed in Section 3.1
. Other errors are either less understood (i.e., quality control for the data) or dependence on the performance of the mapping method (i.e., choice of climatology to calculate the anomaly field).
The IAP product uses a new XBT correction scheme which has been recommended by the XBT scientific community [46
]. A mapping method uses the data only near the analyzed grid within an assumed area to perform the reconstruction at individual grid cells, with the size of the area defined by the influencing radius. The covariance defines the correlation between the analyzed grid with the adjacent locations, which are used in the reconstruction. The IAP mapping method uses the Ensemble Optimal Interpolation (EnOI) framework with first-guess and covariance from a number of CMIP5 simulations. Models provide more reliable covariance than the traditional parameterization, which always assumes a Gaussian distribution (e.g., in Levitus et al. [60
] and Ishii et al. [61
]). Use of models allows the choice of a larger “influencing radius” than previous methods, and ensures a near-global fractional coverage (defined as the fraction of total ocean area obtained by the mapping method).
For comparison, we also use two other gridded ocean temperature products constructed based on in-situ ocean observations and different mapping methods and XBT correction schemes. The first is Ishii data with two different versions: an old version first published in 2003 [61
] (Ishii-old) and an improved version released in 2017 [47
] (Ishii-new). Ishii-old data was used in many studies related to the “hiatus” [32
] and has been discussed in follow-on communications [63
]. Therefore, it is worthwhile to examine its quality in the present analysis. EN4 data from Met Office was also used in [62
]. We will provide some analyses on these datasets in our study, but a more comprehensive comparison among these datasets can be found in Wang et al. [45
] and a careful evaluation on IAP-mapping can be found in Cheng et al. [29
Differences between the linear trends of the SWS period (1998–2013) and the reference period (1982–1997) are used to examine the ocean heat content redistribution related to SWS (this method is named “Trend-Diff method”). As discussed before, there is a substantial global warming (GHGs forced) signal in the OHC records which should be removed or reduced in order to show the natural variability. Using the trend differences reduces the impact of long-term anthropogenic warming, assuming the long-term warming is constant in time (especially during the 1982–2013 period). However, with more GHGs accumulating in the atmosphere over time, there should be an acceleration of long-term warming (nonlinear and non-constant), as shown in climate models [33
]. Another way to consider the long term trend is to use the global mean time series as a primary indicator of the non-constant component, and remove it from each grid point prior to carrying out an analysis, as proposed in Trenberth and Shea [66
] (this method is named “Glb-Trend-Remove method”). Both methods are tested in this study and show similar geographical patterns. Consequently, the result based on the first method is always shown.
The results could be sensitive to the choice of time-window, so we also tested the results by using several other choices (i.e., 1998–2012, 1999–2013, 2000–2013), showing that they do not impact the OHC pattern. All these methods and time-window choices impact the quantification of the OHC changes in different ocean basins, but the key conclusion remains unchanged.
Motivated by the debate related to the “global warming hiatus”, this study first suggests that it is a mistake to deny the “global warming” according to a slowing rate increase of GMST or SST over 13 years, despite the inability to find any statistically significant change in trends of these metrics. On the other hand, EEI and global OHC are fundamental metrics, as they indicate the radiative imbalance driven by GHG and they are less impacted by natural variability. Based on global OHC records, there is no indication of any pause in “global warming”, and in fact, the ocean warming is accelerating. Therefore, the term of “surface warming slowdown” was used in this study to clearly indicate any changing trend was unassociated with global warming. The scientific question then turns to: why is there a slowdown in GMST/SST with the continuous global warming and how are the GMST/SST changes related to ocean heat uptake?
On this basis, we reviewed the progress in OHC measurements and show the estimation error in traditional datasets could impact the identification of the ocean heat uptake. In this effort, a new dataset (IAP) was available for use. The most notable improvements in the new data are improved corrections for XBT bias, which significantly improvs the data quality; and a better mapping method, which effectively reduces the “conservative error” in previous dataset.
Based on the IAP data, complete 3-D temperature changes in the SWS period (related to the 1982–1997 period) in the ocean interiors are provided, with several “hot spots” and “cold spots” being identified. The decadal-scale ocean heat redistribution pattern is distinct from the long-term ocean heat uptake, confirming that natural variability dominates the decadal scale local OHC changes.
To give some implications on the driver of the ocean heat redistribution, we further perform a regression analysis to identify the OHC pattern related to three different climate models: ENSO, IPO, and AMO. The results indicate that none of them can solely explain the formation of the OHC patterns in the SWS period. Nevertheless, they can be important in different locations. The implication is that there might be no single answer to the occurrence of the SWS, instead, it is a combination of different phenomenon and different climate variability. Several model-based analyses also suggest that there are different flavors of the SWS period [8
]. More analyses are required to investigate climate sensitivity and ocean heat uptake efficiency on decadal scales [98