4. Proposed Mechanisms
A prerequisite for a decadal slowdown in global-mean
warming must be a reduction of radiative heating available to the surface. This could come about from either a reduction of radiative forcing from above or from an enhanced uptake by the oceans below. In the first category, it has been proposed that a reduction of radiative forcing can arise by a decrease in stratospheric water vapor [74
], an increase in background stratospheric volcanic aerosols [75
], by 17 small volcano eruptions since 1999 [76
], increasing coal-burning in China [77
], the indirect effect of time-varying anthropogenic aerosols [28
], or a low solar minimum [78
]. These forcing reductions likely contributed no more than 20% of the slowdown [58
]. Schmidt et al. [79
] showed that a combination of a revised aerosol parameters applied to the recent moderate volcano eruptions and anthropogenic pollution from emergent China accounted for half of the discrepancy between CMIP5 models and observation, with the solar minimum and different phasing of ENSO one seventh. Asian pollution from coal burning however contain black carbon as well as sulphate aerosols, and Kühn et al. [80
] found that the warming effect of the former offset the cooling due to the latter, yielding very little global or regional effect.
The 30-year mid-20th century hiatus was also thought to be caused by aerosols from pollution in an emergent economy after the war [28
]. Because of the uncertainty associated with the aerosol parameters and response, it is difficult to assess the efficacy of the proposals. However, aerosol cooling of the surface of the ocean from above should destabilize the ocean’s vertical stratification, and there should be a subsurface signature but none was found. Zhang et al. [29
] pointed out that the subsurface observations in the Atlantic showed warming, in contrast to the modeled subsurface aerosol cooling. Chen and Tung [60
] found subsurface signature of salinity and heat content variations during the mid-20th century hiatus to be that of an internal ocean variability of increased heat uptake, similar to the current one.
Kosaka and Xie [38
] focused on the cold eastern equatorial Pacific as the main driver for the global warming slowdown. Their results demonstrated dramatically that the slowdown in global warming could be simulated when the observed SST in this cold-tongue region is specified in a climate model that does not by itself produce the slowdown. What caused the eastern Pacific to have more of the cold La Niña-type than the warm El Niño-type of events in observation—but not in the original model—was not pursued. While it is well known that the interannual part of ENSO influences the interannual variation of the global-mean surface temperature through atmospheric teleconnections [82
], the influence of the multidecadal variability in the Pacific on the global mean is much more subtle, as will be discussed in Section 5
on model “nudging” experiments. England et al. [39
] proposed that the intensification of the trade wind in the tropical Pacific is the cause for the global warming hiatus, by creating the La Niña-type pattern in the eastern Pacific SST. They attributed this observed 20-year intensification of the Pacific trade winds to the negative phase of the Interdecadal Pacific Oscillation (IPO) [84
], though the coauthors later casted doubt on that assertion [39
]. Dai et al. [85
] concluded that internal climate variability, “mainly through IPO, was largely responsible for the recent slowdown”. In a Perspective
, Clement and DiNezio [86
] gave a short review of the literature up to that time, arriving at a tentative conclusion that the Pacific is the “driver” of the observed decadal change in the current period, and much of the evidence in the papers reviewed rested ultimately on the IPO and its kin, Pacific Decadal Oscillation [87
Meehl and Hu [88
] gave the only mechanistic explanation of the IPO as wind-stress forced Rossby waves at 20N and 25S crossing the Pacific, with a decadal transit time, setting the timescale of the portion of the IPO spectrum in the 20–30 year period, but not the 60–70 year period now alluded to as explaining the multidecadal variability in the global-mean temperature seen in Figure 2
. We will discuss the IPO later in Section 6
and address whether or not it is “ENSO-like”.
Intensifying trade wind is not expected to subduct appreciable heat in the tropical Pacific, or for that matter, in the Pacific as a whole (see Figure 3
and Figure 5
inset). Huang [89
] investigated the adjustment of OHC in response to the wind-driven circulation in the form of “heaving” of the isopycnal surfaces and suggested that the adiabatic movement of warm water along such sloping surfaces as a mechanism for vertical redistribution of heat in the oceans. However, he concluded: “heat content variability inferred from our model is at least one order of magnitude smaller than the mean warming rate inferred from observations”.
In the Atlantic, the multidecadal variability in the SST is understood from modeling studies to be caused by the variations of subsurface AMOC. “Water hosing” experiments of Zhang et al. [25
] suggest that freshening of subpolar North Atlantic waters can lead to a slowdown of AMOC and a cooling of the surface temperature under preindustrial conditions. In the presence of top-of-atmosphere radiative imbalance, Chen and Tung [60
] showed using subsurface ocean data, including salinity and ocean heat content, that as the AMOC sped up during 1999–2005, it subducts heat in the subpolar latitudes of the North Atlantic. That this is a period of increasing overturning in the Atlantic was previously calculated by Willis [68
] using satellite altimetry data available since 1993. The proposed mechanism is as follows: As AMOC speeds up, it brings the warmer and more saline subtropical surface water to the sub-polar latitudes of the Atlantic, where it loses part of its heat to the cold atmosphere and sinks due to its saltiness. The heat released by the warm water to the atmosphere melts glacier ice over Greenland and the surrounding areas bounding the North Atlantic, gradually leading to a freshening of the North Atlantic water that eventually slows the sinking. As AMOC slows, its northward transport of heat slows and the freshwater outflow from glacier ice melt is reduced. Salinity and hence density of the seawater build up slowly over decades, until it is dense enough to initiate another speeding up of the AMOC. In this way the AMOC alternately speeds up and slows, taking approximately 60–70-years for a full cycle, leading to warmer and colder AMO at the sea surface with the same multi-decadal variations. A more detailed discussion can be found in Sarachik et al. [90
]. Most recent CMIP5 models tend to have 20–30-year oscillations [91
], with the exception of MPI and an older version of GFDL model.
compactly shows the OHC changes in the four ocean basins. The size of the different ocean basins is taken into account. Figure 6
a,b show that before 2005, the North Atlantic subducted a large amount of heat through deep convection down to at least 1500 m in the subpolar Atlantic, caused by a sped-up AMOC [60
]. After 2005, AMOC slowed and the deep convection reduced. There is cooling in the subpolar Atlantic, but warming in the subtropical Atlantic due to the northward displacement of the Gulf Stream as AMOC slowed [92
]. Figure 6
c,d are meridional integral of the Pacific and Indian Oceans as a function of longitude. There was a continuous trend of trade wind intensification over both periods, creating a La Niña mean state in the eastern tropical Pacific. The intensifying trade wind blows more warm surface water from the eastern Pacific to the western Pacific, and a little over the Indonesian Through Flow into the upper layer of the Indian Ocean, which can be seen clearly in this figure. Its magnitude of warming is relatively small and it is shallow and located near the Indonesian Through Flow opening to the Pacific. This figure adds observational confirmation of the possible Pacific origin of the Indian Ocean warming [47
]. Indian Ocean’s shallow warming is a secondary player in the heat sequestration in the energy budget below 200 m, consistent with Figure 5
and Table 1
. In the Pacific, the heat sequestered in the western Pacific is large, but the change is mostly balanced by the cooling in the eastern Pacific above 300 m in the two periods. When summed over the basin, Pacific also plays a minor role in the energy budget, as shown in Figure 5
and Table 1
. In the Southern Ocean, the heat storage increase is large after 2005 when integrated over the circumpolar band of the ocean. The larger magnitude is due to the larger area being integrated over in the circumpolar band in our way of accounting for the total contribution to the global OHC. Locally the warming is not particularly intense in the Southern Oceans. The increase appears to be a continuation of the warming from 1990s, but the data before 2004 was not adequate.
5. Model “Nudging” Results and Their Interpretations
The unexpected slowdown in global warming offers lessons that can help future climate model developments. Fyfe and Gillett [51
] noted the severe discrepancy that existed in the eastern Pacific, with the observed SST much colder than any of the model runs. Kosaka and Xie [38
] attempted to remedy this deficiency in the GFDL model by specifying from observation the colder eastern equatorial Pacific SST. The model surface air temperature (SAT) over this region was initially warmer than observed. Model “SST is restored to the observed evolution by a Newtonian cooling over the deep tropical eastern Pacific”. Thus a numerical heat sink was introduced, absorbing whatever energy is needed to bring the model SST in that region back to the observed cold temperature. The atmosphere always tends to move heat from other (warmer) regions into the cold eastern Pacific to be absorbed, eventually yielding a global hiatus with just enough heat transported to this region to be absorbed, and the model SAT becomes compatible with the specified observed SST, meaning that the model SAT becomes close to the observed SAT. One may be able to assess whether the Pacific heat sink in the model is consistent with the observed OHC in the Pacific to provide a crucial check on the reasonableness of the proposed mechanism. However, the model diagnosed heat flux in the ocean was “overridden” in the nudging experiment of Kosaka and Xie [38
], and so the amount of heat uptake cannot be compared with observation. But such information is available in another model (see later).
This effect on heat absorption in the eastern Pacific is not as prominent in a model setup where the observed atmospheric field is specified, as in Drijfhout et al. [93
], because the SAT over the cold SST is also cold. Instead of specifying the wind field only in the tropical Pacific, Drijfhout et al. [93
] specified the meteorological field throughout the globe, and found that it drove an increase in ocean heat uptake of 0.5 W·m−2
during 2001–2009 over 1992–2000. The Southern Ocean (south of 35S) and the subpolar North Atlantic were dominantly responsible for such an increase, and the tropical Pacific less so. Watanabe et al. [94
] specified the wind stress from observation for 30S–30N over the world oceans, including the tropical Atlantic, not just over the Pacific. Zonal mean subsurface temperature of the world oceans, which was allowed to adjust, was shown by Watanabe et al. [94
] and compared favorably with observation, pointing to the importance of high latitude oceans, especially the North Atlantic, in deep heat sequestration. The high-latitude response in the model is remotely forced by the nudged tropical Atlantic winds, and is absent in the experiment of England et al. [39
], where only the Pacific winds are nudged. Inferred regions of heat sink from “nudging” experiments are seen to be sensitive to the region of “nudging”. Douville et al. [95
] suggested that the nudging or pacemaker experiment should include not just the tropical Pacific but also the North Atlantic. In addition Douville et al. [95
] found that most climate models underestimate the potential role of North Atlantic ocean while overestimate the influence of ENSO on global-mean surface temperature (GMST). In particular “none of the CMIP5 models is able to capture the observed persistent positive correlations between AMV and GMST, which might explain why the role of North Atlantic multi-decadal variability has been so far much less tested than the role of the tropical Pacific.”
Trenberth et al. [96
] showed that a somewhat better simulation is obtained over Eurasia if the negative latent heating implied by the cold (and dry) eastern and central Pacific SST is put aloft instead of at the surface, using an atmosphere-only model.
Meehl et al. [55
] were the first to show in a coupled model without nudging that periods of global-warming hiatus occur even though the model top-of-the atmosphere radiative imbalance happens to be constant. They attributed the internal ocean variability to the IPO, the AMOC and the Antarctic Bottom Water. Guemas et al. [97
] in a retrospective prediction of the surface hiatus of the past decade attributed the onset to an increase in ocean heat uptake in the Pacific and Atlantic. Meehl et al. [98
] suggested that if model’s IPO is made to coincide with observation, CMIP5 models could possibly predict the slowdown.
6. IPO as an Interdecadal Modulation of ENSO
The IPO came to the public’s attention recently as the role of internal multidecadal climate variability in influencing the global surface temperature is increasingly recognized in the scientific literature. Two appealing aspects of the IPO/PDO have often been alluded to: (i) the phase of the IPO appears to coincide with the “climate regime shifts” in the global-mean surface temperature, so that by implication the former causes the latter, and (ii) its spatial pattern is “ENSO-like” and therefore it implicates the tropical Pacific [39
]. It turns out that neither of these attributes is robust.
As a low-frequency modulation of ENSO, IPO is obtained by performing Empirical Orthogonal Function (EOF) decomposition on the low-passed SST data using a Fourier filter. A reasonable 15-year low passed filter, reasonable for getting the “interdecadal oscillation”, is sufficient in filtering out the high variance, high-frequency ENSO in the tropical Pacific, leaving very little ENSO-related signature. This is consistent with the historical definition of IPO by Folland et al. [84
]. A 13.3-year low-pass filter was used to “remove ENSO-related variability”. As a consequence, their IPO does not have the ENSO-related variability in the eastern tropical Pacific. They used 84 years of SST data, from 1911–1995. The degrees of freedom is extremely small after low passed by a 13.3 year filter [99
]. Later Parker et al. [100
] extended the record backward by 20 year to 1891, used an 11-year low-pass filter and redefined the second EOF of the low-pass global SST as the IPO. Its spatial pattern has what appears to be a more prominent ENSO-related variability in the tropical eastern Pacific. Tung et al. [101
] reexamined Parker et al.’s calculation, removing the poorer quality data prior to 1911. The “ENSO-related” feature of Parker et al.’s IPO largely disappears, and remains absent when 20 more years of recent data are added to the end of the time series. Because of the closeness of the eigenvalues in the EOF analysis of decadally filtered SST, the second and third EOFs form what North et al. [102
] called “effectively degenerate multiplet” within the large sampling error of the finite sample. Some authors used the second EOF as the IPO and others chose the third EOF. The second EOF looks more like the AMO while the third more like the PDO. They are in effect not distinct because of the degeneracy.
Despite its name, the main component of the IPO is actually the AMO, along with some PDO and ENSO. They are “mode-mixed” because of the low-pass filtering. See Figure 7
, showing the composition of the IPO. At decadal filtering commonly used, ENSO is almost entirely filtered out. There is a small trend (TR) and a small PDO component. The largest component is the AMO. Chen and Tung [7
] found that it is actually the AMO component in some of the IPO indices that appear to mimic the low frequency variation of the global-mean surface temperature. The PDO contributes very little to the global-mean surface temperature (see Figure 8
), because of its compensating warm and cold centers of action, as shown in more detail in Chen and Tung [7
Although IPCC referred to the 15-year period 1998–2012 as the “hiatus”, the trend to 2014 as reported by Karl et al. [2
] disqualifies the period to 2014 as a “hiatus”, but it can still be considered a “slowdown”. The appearance of the “warm blob” in the north Pacific [103
], possibly due to air-sea interaction, made 2014 a warm year, but that phenomenon did not last. Then came the extreme El Nino of 2015–2016, producing record surface temperature globally. So the period of slowdown probably ended in 2014.
In trying to understand this phenomenon the community, through intensive modeling and analysis efforts, seemed to have gained a better recognition of the role of multidecal internal variability in climate change and a clearer assessment of various factors of external forcing. The new Argo data allowed a glimpse of the global oceans below the surface about the possible pathways of heat uptake.
Multidecadal variations in the global surface temperature were previously noted [20
], and the negative phase of the recent cycle appears to account for the slowdown in warming in the 21st century. The spatial pattern of this variability contributing to the global mean was found to center in the North Atlantic, and much weaker in the Pacific [20
] (Figure 3
). There was indeed a cooling trend in the eastern tropical Pacific for 1997–2014, but there was a compensating warming trend in the western Pacific (Figure 4
), explainable by the trade wind blowing warm surface water from the east to the west Pacific.
Although many in the scientific literature attribute the current (and prior) episode of the warming slowdown to the IPO, significant mathematical issues exist with its definition [101
] due to its extremely low degrees of freedom. The ENSO-related spatial pattern often referred to as the IPO was found to be an artifact of strong sensitivity to sparse data prior to 1910. When that part of the data was excluded, the spatial pattern is no longer ENSO-like, along with the implication that IPO modulates ENSO. This was shown in Tung et al. [101
]. The time series of the IPO which appears to coincide with the multidecadal variation of the global-mean surface temperature turns out to be the AMO component contained in the IPO. Low-pass filtering of the SST in the definition of IPO mixes AMO with PDO.
“Nudging” experiments focused on the Pacific produced a slowdown in warming probably because of the presence of a numerical sink of heat introduced. That the numerical sink was introduced in the Pacific in these numerical experiments probably should not be interpreted as where the real sink should be located. When wind stress was specified over pan tropical oceans instead of just over the eastern tropical Pacific, large deep heat sequestration is found in high latitude oceans, especially the North Atlantic, instead of over the equatorial ocean. It is now known that “nudging” or “pacemaker” experiments are sensitive to experimental setup, and it was suggested that surface conditions in other ocean basins, including the tropical Atlantic, should be included.
In a somewhat related study on the transient climate sensitivity of CMIP5 coupled climate models to 4 × CO2
forcing, Kostov et al. [106
] found that “a substantial portion of the global ocean heat uptake occurs within a relative small region in the North Atlantic and that anomalous heat is advected to depth along the upper AMOC cell” and intermodal differences in response to the greenhouse forcing “can be understood in terms of variations in the depth of heat storage, which in turn reflects the depth and strength of the AMOC.” “Models with a deeper and stronger overturning circulation store more heat at intermediate depths, which delays the surface temperature response on multidecadal time scales”. This model result is consistent with the observational evidence presented here.
Observed subsurface ocean heat content data show that the major sinks of heat are in the North Atlantic and the Southern Ocean, accounting for a majority of the heat stored in the intermediate layers of the world’s oceans, although the debate continues regarding whether it is the Pacific-Indian oceans on one side or Atlantic-Southern Oceans on the other side that is mostly responsible for causing the warming slowdown. Regardless, the result so far favors the explanation of the warming slowdown as an internal variability of the ocean’s ability in storing the heat that otherwise would have warmed the surface more.