The tropical Pacific atmosphere-ocean system (Box 1) exhibits variability over a broad range of timescales: the El Niño/Southern Oscillation (ENSO) dominates at interannual timescales (~2-7 years), and the trend from anthropogenic forcing at centennial. In the intermediate range, natural (internal) variations occur at quasidecadal and multidecadal timescales (7-70 years) 1 , encompassing the broadly termed tropical Pacific decadal variability (TPDV). TPDV represents the tropical expression of the Pacific decadal oscillation 2 in the North Pacific and the interdecadal Pacific oscillation 3 over the entire Pacific basin. Its positive phase is characterized by warm sea surface temperature anomalies (SSTAs) in the tropical Pacific and along the western coasts of the Americas, and by negative anomalies in the central and western midlatitudes of both hemispheres; the negative phase exhibits anomalies of the opposite sign.
TPDV has important climatic relevance. For example, it modulates ENSO characteristics 4,5 and some of its global impacts, including climatic variations over Antarctica 6 , Australian monsoon variability 7 and temperature and precipitation over the western USA 8 , making prediction of TPDV phases societally critical. TPDV is further linked to the rate of change of globally averaged surface temperature 9,10 as demonstrated by the decrease in globally averaged surface temperature trend during the cold TPDV phase in the first decade of the 2000s. Accordingly, understanding TPDV is integral to robustly separate the forced climate response from internally generated climate variability and thereby produce reliable projections of tropical Pacific and global climate 9 . Yet, some models appear to underestimate internally generated decadal variations [11][12][13] and might incorrectly simulate externally forced trends, introducing uncertainty in attribution analyses [14][15][16][17] . This ambiguity highlights the importance of a deepened understanding of internal low-frequency variability and prediction of decadal epochs in the tropical Pacific.
However, TPDV predictability currently remains elusive, largely related to complicated, and often competing, underlying mechanisms. For instance, TPDV could result as a residual of interannual ENSO variability 18 , or result from equatorial upwelling of subtropical temperature anomalies from the pycnocline (the vT ′ hypothesis, in which v indicates the time mean circulation and T ′ is the temperature anomaly) 19 , or changes in equatorial upwelling itself (the v T ′ hypoth- esis, in which v′ indicates the circulation anomaly and T is the time mean temperature) 20 . Moreover, these oceanic mechanisms could be driven by atmospheric forcing resulting from processes in the extratropical Pacific 21 , responding to equatorial SSTAs 22 or arising from interactions with the Atlantic and Indian Oceans 23,24 . Yet, no consensus exists on the effectiveness and relative importance of these processes.
In this Review, we critically elucidate the nature and relative importance of the mechanisms driving TPDV using evidence from observations, ocean reanalyses, dynamical models and paleoclimate proxies. We begin by describing salient features of TPDV in the context of the phase transition that occurred in the late 1990s. We follow with discussion of the leading oceanic and atmospheric processes relevant for TPDV, including as an ENSO residual; the vT ′ hypothesis; the v T ′ hypothesis; and extrat- ropical and tropical forcing and influences from other ocean basins. We end with recommendations for future research. Relative to other reviews that have considered internal and anthropogenically forced low-frequency variability 18 , focus here is on internal decadal variations.
The equatorial Pacific Ocean is often described as a system with a warmer and dynamically active upper layer and a colder and more quiescent bottom layer (see the figure, bottom). These two layers are separated by a region of sharp vertical density and temperature gradients, known as the pycnocline and thermocline, and are overlaid by a near-surface frictional layer -the Ekman layer.
The pycnocline links subtropical regions to the equator: subtropical waters can penetrate into the ocean interior at the latitudes where surfaces of constant density (isopycnals) meet the near-surface layer and then flow equatorward along those isopycnals. At the equator, these waters are brought to the upper layers by the upward vertical velocity (upwelling), and returned to higher latitudes by the flow in the surface Ekman layer (see the figure, bottom, black solid arrows), creating shallow overturning circulations in both hemispheres termed subtropical cells 20 Additionally, oceanic nonlinearities associated with strong El Niño events during a positive TPDV phase could induce strong negative feedbacks and cause a transition to negative TPDV 45 .
The interpretation of TPDV as an ENSO residual also involves subsurface anomalies. Western Pacific heat content exhibits a decadal modulation, with reduced heat content during periods of positive TPDV (such as 1976-1999) (Fig. 2b,d) and enhanced heat content during negative TPDV (such as 1999-2014) (Fig. 2b,d). These low-frequency variations are punctuated by the heat content changes associated with the recharge-discharge activity of individual ENSO events (Fig. 2b), which are the dominant signal in the eastern Pacific (Fig. 2c). The decadal modulation of tropical Pacific heat content could thus be interpreted as the low-frequency envelope of interannual ENSO variations.
However, ENSO characteristics also depend on the mean state 46,47 . Indeed, the warm phase of TPDV, characterized by weaker trade winds and a deeper thermocline in the eastern equatorial Pacific, favours more frequent and stronger eastern Pacific El Niño events (Fig. 2a), whereas negative TPDV phases are characterized by weaker central Pacific El Niño events (Fig. 2a). The mean state influence on ENSO was also highlighted by dynamic model prediction experiments in which the ENSO evolution and predictive skill 48,49 were highly dependent on the initial background conditions. The decadal modulation of ENSO, as captured in climate models by the second empirical orthogonal function of decadal SSTAs 4,50,51 , is significantly lag-correlated with TPDV 51 , with a large intermodel dependence 50 . ENSO decadal modulation appears to lead the opposite phase of TPDV by about 2 years, suggesting its possible role as precursor of TPDV phase transitions 51 . However, TPDV also leads the same phase of ENSO decadal modulation by 2 years with a higher correlation 51 , indicating that ENSO modulation by TPDV might be more prominent than the influence of ENSO activity on TPDV. Additionally, empirical models trained on observations indicate that tropical-extratropical interactions are key to the existence of TPDV, implying that TPDV cannot simply arise from processes occurring within the tropics as in the case of ENSO residuals 52 .
In addition to the possibility of being an ENSO residual, equatorward advection of temperature anomalies within the pycnocline (the vT ′ hypothesis) has been put forward as a driver of TPDV 19 (Fig. 3a). Two mechanisms by which subtropical signals reach the equator have been proposed: spiciness anomalies advected as passive tracer by the mean circulation and non-compensated temperature anomalies propagating as planetary (Rossby) waves, as discussed now.
Spiciness anomalies describe temperature anomalies with a density compensating salinity signal 53 ; they do not affect density and propagate along isopycnals as a passive tracer 54 (Fig. 3a). These warm-salty or cold-fresh anomalies are predominantly generated in the eastern subtropical Pacific 55,56 through shifts in spiciness gradients induced by wind-forced anomalous ocean currents 54 , or buoyancy-forced penetrative mixing 56 . Spiciness anomalies are subsequently advected by the subsurface branches of the subtropical cells (STCs) towards the equator. Despite some decay [57][58][59] , observations support the generation and propagation of spiciness anomalies from the eastern subtropics to the western tropical Pacific. However, whether these anomalies are advected all the way to the equator is much less clear, and the feasibility of a western boundary pathway is uncertain owing to the complexity of low-latitude western boundary currents (LLWBCs) and high mixing and water mass transformation 60 . A Lagrangian modelling approach indicates that spiciness anomalies reach the eastern equatorial band 61 , with clear dominance of southern hemisphere pathways. At large spatial scales, theoretical arguments suggest that pycnocline advection might result in a frequency spectrum of spiciness anomalies reaching the equator with enhanced power in the decadal range 62 .
Such decadal-scale spiciness anomalies might drive TPDV. Specifically, coupled model experiments suggest that equatorward-advected spiciness anomalies are upwelled to the surface where they rearrange equatorial SSTs, winds and the slope of the pycnocline 63 , in turn, inducing atmospherically forced off-equatorial spiciness anomalies of opposite sign, resulting in a 10-year cycle 54,63 . A heat budget analysis of the modelled equatorial Pacific mixed layer further confirms this influence of spiciness anomalies on TPDV 59 , although with a small magnitude relative to other heat budget terms leaving the efficiency of this mechanism unclear.
An alternative mechanism within the vT ′ hypothesis is the propagation of temperature anomalies via Rossby waves. Oceanic Rossby waves cause isopycnal displacements that appear as temperature anomalies over time-mean isopycnal surfaces. These anomalies reach the equatorial thermocline via the western boundary and propagate eastward along the equator as equatorial Kelvin waves, altering equatorial SSTs. As such, Rossby wave activity has been related to decadal subsurface temperature anomalies in the tropical Pacific with maxima around 10°-15° N and 10°-14° S (refs. 26,64-68) (Fig. 1c).
However, the origin of the decadal timescale remains unclear given that the Rossby wave transit time at key latitudes (10°-15° N and 10°-14° S) is only 2-3 years. Several hypotheses have been put forward as an explanation. First, the latitudes of Rossby wave maxima coincide with areas of high zonal coherence of the wind forcing, which might be efficient in exciting large amplitude waves at decadal timescales 65 . Second, these latitudes coincide with the equatorward boundaries of the subtropical gyres, where instability processes can energize
The interior wind-driven zonal circulation is connected in the western Pacific to the equatorward flowing low-latitude western boundary currents, which are an important conduit for the redistribution of subtropical water to the western equatorial Pacific 170 and then into the tropical current system, including the EUC and the Indonesian Throughflow. 161 (shading) and decadal SST anomalies regressed onto the tropical Pacific decadal variability (TPDV) index (contours). b, Differences of linearly detrended sea-level pressure (SLP; shading) and vector wind anomalies 162 (arrows). c, Differences of linearly detrended sea surface height (SSH) anomalies 163 (shading) and decadal SSH anomalies regressed on the TPDV index (contours). d, Differences of un-detrended SSH anomalies 163 . e, Differences of detrended temperature anomalies zonally averaged between the western ocean boundary and the dateline; contours indicate the time mean (1979-2017) 15, 20 and 25 °C isotherms. f, As in panel e, but for temperature values averaged from the dateline to the eastern ocean boundary. In all panels, differences represent 1999-2014 minus 1984-1999, and regressions are calculated over 1958-2020. In panels a and c, solid contours represent positive anomalies and dashed contours represent negative anomalies, drawn at 0.1 °C intervals for SST and 1 cm for SSH. TPDV is associated with basinwide SST, SLP and wind anomalies and involves a reorganization of heat content in the tropics. planetary waves originating in the eastern midlatitudes of both hemispheres with longer transit times in the decadal range 66 . Finally, the timescale of the ocean response to anomalous winds can be extended to the decadal range by the slow eastward propagation of equatorial signals owing to the coupling of oceanic waves with local winds 67 . More generally, decadal timescales cannot be expected to coincide with the transit time of one single wave, but result from the collective effect of multiple waves generated over relatively broad latitude bands at different times, leading to a longer adjustment timescale.
Both mechanisms (advection and planetary wave activity) are likely to contribute to the equatorward propagation of temperature anomalies, with the impact of the South Pacific seemingly dominating spiciness propagation [67][68][69] owing to its larger and more direct equatorward transport [70][71][72][73] . In the North Pacific, the presence of the Intertropical Convergence Zone alters the depth of the pycnocline and creates a potential vorticity barrier 73 that limits the interior equatorward flow 61,73 (Fig. 3a). Global Climate Model sensitivity experiments, in which oceanic temperature and salinity anomalies are blocked from reaching the equator in both hemispheres, indicate that the southern vT ′ process acts as a delayed negative feedback for bi-decadal (12-25 years) variability, whereas oceanic wave adjustment has a dominant influence in the decadal range (9-12 years) 69 . The role of decadal anomalies from the South Pacific is also illustrated by their influence on the evolution of El Niño events during the first decade of the 2000 (ref. 74); cold anomalies in the southwestern tropical Pacific related to the negative TPDV phase during 1999-2014 might have impacted the development of El Niño events 74 , possibly leading to the unexpected termination of El Niño in 2014 (ref. 75). Rossby wave activity is also prominent in the North Pacific and provides an important contribution to decadal variability of the equatorial Pacific thermocline 65,66 through the western boundary pathway.
In addition to upwelling of anomalous temperatures from the pycnocline, anomalous upwelling can also drive TPDV via changes in the transport of the STCs (Fig. 3b). Specifically, an increase in the equatorward mass transport of the STCs enhances equatorial upwelling, bringing colder pycnocline waters closer to the surface and cooling SSTs; reduced STC transport has the opposite effect 20 . At interannual timescales, these changes encapsulate the recharge-discharge of the equatorial upper-ocean heat content, underpinning ENSO evolution 76 . The relationship at decadal timescales suggests that similar underlying dynamics might be important at lower frequencies 77 . This hypothesis has been tested in simple models 20,[78][79][80] , observations 25,81 , ocean general circulation models [82][83][84][85] and ocean reanalyses 26,86 . For example, using binned observations of interior transport (zonally averaged pycnocline flow east of the LLWBCs at 9° N and 9° S) as a proxy for STC strength 25 reveals a decline in equatorward subsurface mass convergence after the mid-1970s, concurrent with the tropical Pacific warming associated with the 1976-1977 climate shift 2,25 (Fig. 4a,b). Ocean reanalyses and ocean models forced by observationally constrained surface fields confirm that increased interior equatorward mass convergence is associated with colder equatorial SSTs and vice versa, with high correlations at both interannual and decadal timescales 26,82,86 (Fig. 4c,d). Many climate models also show correlations between transport convergence and SST anomalies that are comparable with those obtained from ocean reanalyses, although some exhibit much weaker relationships 87,88 (Fig. 4e). In addition, transport variability is generally weaker in the models than in observations for the same SST variability 87,88 (Fig. 4f), suggesting a higher sensitivity of modelled SSTs to STC variability.
Besides the interior pycnocline transport, variability of the LLWBCs and ITF can also affect the equatorial mass convergence and equatorial upwelling. Anomalies in the LLWBC transport are of opposite sign to the interior transport anomalies 82,[89][90][91] , potentially leading to a partial compensation of interior mass convergence. However, given the complexity of the LLWBCs, and the sparsity of in situ observations in these regions, it is unclear what fraction of their anomalous transport recirculates in the western Pacific, exits the Pacific through the ITF or acts to alter the equatorial mass balance. The strength of the ITF has been shown to contribute to the mass and heat balance of the equatorial Pacific 92 at interannual timescales 93,94 , and it could likely influence the equatorial Pacific also at decadal timescales, suggesting a potential oceanic pathway for the Indian Ocean influence on TPDV.
The location of winds that are most influential on STC decadal variations is key to understanding their role in TPDV. Wind variations in subtropical regions could control STC transport and remotely affect equatorial SSTs 20,80 . However, meridional transport changes at each latitude appear to be established by westward-propagating oceanic Rossby waves, as part of the tropical adjustment to varying winds, and be largely controlled by the local wind forcing 26 , although influences from the 12°-20° latitude band might also have a role at decadal timescales 22,85,95,96 .
The possible origin and nature of these winds are discussed next.
Potential drivers of TPDV are not restricted to the equatorial region. The North Pacific Meridional Mode and South Pacific Meridional Mode 97,98 (NPMM and SPMM, respectively) (Fig. 5a,b) are important in this regard, reflecting SST patterns produced via off-equatorial turbulent heat fluxes and maintained through the wind-evaporation-SST feedback 99 .
These modes are important factors influencing equatorial dynamics (for example, through excitation of deep convection near the Intertropical Convergence Zone and corresponding equatorial wind anomalies 100 , and heat recharge-discharge in the equatorial pycnocline through meridional flows induced by wind stress curl anomalies -tropical wind charging [101][102][103] ) and thereby ENSO development 97,100,[104][105][106] .
However, the NPMM and SPMM are also involved in the development of TPDV. For instance, 'Atm-Slab' models (atmospheric models coupled to slab ocean models) 107,108 exhibit a frequency spectrum reddening of weather and climate variability at decadal timescales through a sequence of extratropical-to-tropical influences (ENSO precursors to ENSO development) and tropical-to-extratropical feedbacks (ENSO teleconnections) 107 , as supported by observations 109 . Indeed, model experiments 110 indicate that ENSO teleconnections from the central equatorial Pacific reinforce the NPMM and increase its persistence, resulting in the decadal NPMM variations detected in century-long coral time series from the northeastern subtropical Pacific 111 . Additionally, tropical wind anomalies associated with the Meridional Modes may induce meridional pycnocline flow (as with the Tropical Wind Charging mechanism), providing the atmospheric forcing needed to alter the strength of the STCs and produce equatorial SST anomalies. Sensitivity experiments with simple dynamic models also indicate that extratropical stochastic wind forcing produces low-frequency changes in the equatorial thermocline and multiyear ENSO variations 112 .
The impact of the SPMM and NPMM on TPDV is not equal. The influence of the SPMM is thought to dominate 108,[113][114][115][116] . For instance, idealized nudging of oceanic variability to climatological values over 30 o S-10 o S caused a ~30% reduction in decadal-scale SST variability in the equatorial Pacific 114 . However, new evidence is emerging for a mode of variability that links the North Pacific with the Central Equatorial Pacific via the NPMM (and thus termed NP-CP mode) at decadal timescales 45,[117][118][119] . This mode involves SST anomalies typical of the NPMM and includes an SSH component with a pattern similar to that typical of decadal differences 118 (Fig. 1c), implying an important role for ocean dynamical processes in TPDV. Thus, both hemispheres can potentially provide the atmospheric forcing for TPDV, but the question of which hemisphere dominates remains outstanding.
As for extratropical forcing, wind responses to tropical decadal SST anomalies might also be important in driving TPDV. Specifically, SST anomalies in the central equatorial Pacific, where decadal anomalies are more prominent, excite atmospheric Rossby waves, whose subtropical component weakens the subtropical trade winds in both hemispheres 110,120 (Fig. 5a,b). These equatorially forced subtropical wind anomalies then reinforce the equatorial anomaly through thermodynamic (for example, triggering deep convection) 100 or dynamic (for example, through changes in equatorward mass transport induced by the anomalous winds) 26 processes. Accordingly, a feedback loop between equatorial and off-equatorial regions is created, reddening the power spectra and contributing to the meridionally broader SST anomaly pattern found at decadal timescales 18,27 .
Low-frequency equatorial SST anomalies also alter the Walker and Hadley circulations, influencing TPDV. In particular, warming along the Pacific equator, mimicking climate change conditions, intensifies the ascending branch of the Hadley circulation, in turn, enhancing off-equatorial trade winds [121][122][123] . The resulting ocean circulation adjustment leads to strengthened STCs and cooling of the equatorial Pacific at a later time 121 -a delayed negative feedback to the original equatorial SST anomalies 22,123 . Cold decadal conditions in the tropical Pacific have the opposite effect: a weaker Hadley cell, weaker trade winds, weaker STCs and a warmer equatorial Pacific. This feedback loop between equatorial SST anomalies and off-equatorial wind variations supports the view of TPDV as a tropical-extratropical-coupled cyclic mode of variability. However, the ability to robustly detect these links in the relatively short and noisy observational record challenges interpretation.
Besides the aforementioned TPDV mechanisms internal to the Pacific, decadal SST variability in the Indian and Atlantic Oceans also has the potential to generate variability in the Pacific 124 via atmospheric teleconnections. These teleconnections occur through a series of atmospheric and oceanic responses to the initial SST, reflecting a Gill-type response 125 , as supported by idealized numerical model experiments 37,[126][127][128][129][130] (Fig. 5c,d); anomalous atmospheric convection and diabatic heating overlying the initial SST; near-surface zonal wind convergence into the convective region and zonal wind divergence aloft; an eastward-propagating equatorial Kelvin wave emanating away from this heat source and westward-propagating Rossby waves to the north and south of the heat source; and descending motion throughout the rest of the tropics. Alternate Atlantic to Pacific pathways have also been proposed to occur via the midlatitudes along a curved pathway through the North Pacific to the western equatorial Pacific 131,132 or through the tropics owing to SLP difference and induced surface wind changes across the Panama Isthmus [133][134][135] . Similarly, the linkages between the Indian and Pacific Ocean can occur via wind changes across the Maritime Continent 37 or through stationary extratropical wave trains 136 .
Collectively, these changes influence TPDV. They alter the global Walker circulation on intraseasonal through multidecadal time scales 23,24,126,128,[137][138][139][140] , spreading the diabatically generated tropospheric temperature anomaly through the entire tropics (the weak temperature gradient approximation) 141,142 and increasing the vertical stability of the troposphere (the tropospheric temperature mechanism) 143 . All three of these processes can alter the Pacific trade winds leading to changes in central-eastern Pacific SSTs that can be further amplified owing to the tightly coupled nature of the atmosphere-ocean system in the tropical Pacific 128,137,144 .
Through these mechanisms, observations suggest that TPDV has responded to Atlantic and Indian Ocean SST forcing. For instance, Atlantic warming had a prominent role in the transition from a positive TPDV in the 1990s to a negative TPDV in the early 2000s 11,23,128,130,139 . This importance can be linked to the fact that the Atlantic-Rossby wave-induced wind anomalies modulate winds in the tropical Pacific, and this surface wind modulation is strongest in the central Pacific where the Rossby and Kelvin waves collide (Fig. 5c,d). This strong effect of the Atlantic on the Pacific is likely to have been relatively consistent from 1870 onwards, although its dominance might have been different in the past 124,131,145 . By contrast, the influence of the Indian Ocean in isolation is thought to be minor during the TPDV transition in the early 2000s 23,126 or more important in amplifying the Pacific response to Atlantic forcing 128 . Yet, the magnitude of the Pacific response to idealized Indian Ocean SST forcing is more prominent over longer periods, as during 1980-2010 and 1958-2010 (refs. 24,136).
However, there are limitations in understanding the influence of other ocean basins in driving TPDV. Uncertainties arise from discrepancies between some model results. For example, although interbasin interactions are thought to amplify TPDV, model simulations in which the Atlantic or Indian Ocean influence is removed instead suggest that TPDV is intensified in the absence of Atlantic or Indian Ocean coupling 146,147 . Also, the connection between the Atlantic and Pacific becomes less clear when partially coupled numerical experiments become more realistic 148 . These uncertainties indicate possible limitations of currently used partially coupled experiments 149 .
TPDV of 7-70 years is linked to coherent basin-scale SST and SLP anomalies, with global impacts. Despite a limited historical record of subsurface data, surface manifestations of TPDV are also associated with a reorganization of tropical Pacific upper-ocean heat content, most notably in the zonal direction, suggesting the involvement of ocean dynamical processes. Indeed, several mechanisms have been proposed to explain TPDV. Although it is plausible it might simply arise as a residual of random ENSO variations 18,38 , TPDV leads decadal ENSO modulation by a few years 51 . This lead-lag relationship suggests that ENSO decadal changes are likely a consequence of the slowly varying background conditions, not their cause. A strong relationship between decadal variations in the strength of the STCs and equatorial SSTAs provides support for the v T ′ hypothesis. However, the largest correla- tions occur at zero lag, making a causal relationship between STC transport and equatorial SST changes unlikely. Instead, concurrent STC and equatorial SST variations are part of the tropical pycnocline adjustment to varying wind forcing, mediated by Rossby wave activity 26 .
Thus, these wave-mediated adjustment processes, encompassing the non-compensated component of the vT ′ hypothesis, emerge as a robust feature of TPDV. For instance, Rossby wave activity alters pycnocline depth and manifests itself as temperature anomalies that propagate on mean isopycnals, contributing to TPDV given their transit times and interaction with the forcing, the latter including preferential response to the larger spatial and temporal scales of the winds 65 . Propagation of salinity-compensated temperature anomalies (spiciness anomalies) is also supported in modelling contexts as a potential mechanism 61,63 . Yet, limited observational evidence of anomalies reaching the equatorial region, in addition to a small modelled influence 59 , call into question the magnitude of the compensated vT ′ component. The atmospheric response to decadal SSTAs in the equatorial Pacific, internal atmospheric variability in the extratropical Pacific and atmospheric influences from the Atlantic and Indian Oceans are all potentially important drivers of the aforementioned oceanic processes. 163 , and the dashed blue line indicates a density ridge in the 5°-10° N latitude band known as the 'potential vorticity barrier' 73 . The lower panel depicts zonally averaged isopycnal depths (from 23 kg m -3 to 25.5 kg m -3 with a spacing of 0.5 kg m -3 ; solid lines: 23, 24 and 25 kg m -3 ; dashed lines: 23.5, 24.5 and 25.5 kg m -3 ), and the flow of equatorward spiciness anomalies along isopycnal surfaces. b, As in panel a, but for the v T ′ mechanism, schematically illustrated with mean (black arrows) and anomalous (red arrows) flows, which reveal how flow along isopycnal surfaces connects the subtropics to the tropics. Both vT ′ and v T
′ mechanisms are proposed as potential contributors to tropical Pacific decadal variability. EUC, equatorial undercurrent; LLWBC, low-latitude western boundary current.
However, it is clear that many questions still remain about the nature of TPDV. There are some similarities between TPDV and ENSO, but while ENSO is an ocean-atmosphere coupled phenomenon, whose growth and phase transitions rely on coupled feedbacks, it is not clear whether the same is true for TPDV. Although there are indications that low-frequency equatorial heating 121 or individual ENSO events 43 induce off-equatorial winds favourable for a TPDV phase reversal, there is still uncertainty about the origin and nature of the winds involved. Internally generated wind anomalies in the subtropical-tropical regions create equatorial SST anomalies 21 , which then reinforce the subtropical wind anomalies through atmospheric teleconnections, increasing their persistence to enhance Fig. 4 | Assessment of the v T ′ hypothesis. a, Observed mean zonally integrated interior meridional pycnocline transports at 9° N and 9° S, computed over 1956-1965, 1970-1977, 1980-1989 and 1990-1999. b, Observed mean meridional transport convergence across 9° N and 9° S (purple), computed as the difference between Southern and Northern Hemisphere transports, and sea surface temperature (SST) anomalies averaged over the central and eastern equatorial Pacific (black line; 9° N-9° S, 90° W-180° W). Error bars represent one standard error. c, Reanalysis interior meridional transport convergence anomalies 165 (seasonal cycle removed) across 9.5° N and 9.5° S in the Pacific (black), and SST anomalies averaged over 9.5° N-9.5°S, 90° W-180° W (red). Meridional velocity anomalies used to compute transports, and SST anomalies are linearly detrended. The value in the top left indicates correlation at zero lag between the time series. d, Same as panel c but for 7-year low-pass-filtered anomalies. Values indicate mean decadal transport convergence and SST anomalies between vertical dashed lines. e, Correlations between transport convergence at 9° N and 9° S and equatorial SST anomalies in 4 ocean reanalyses 163,164,166,167 lower-frequency variability 110 . Decadal timescale SST anomalies in the Atlantic and Indian Oceans also induce wind anomalies in the tropical Pacific conducive to the development of SST anomalies of the opposite sign 23,127,128,130 . However, the extent to which wind forcing from the extratropics or from other ocean basins might itself be the result of forcing from the tropical Pacific is not clearly understood. Furthermore, the relative magnitude of these various sources of wind variability in forcing TPDV is not known. A further uncertainty is related to whether the wind variations arise from deterministic processes operating on decadal timescales, or whether the decadal timescale processes in the Pacific are simply the result of stochastic white noise forcing that the ocean integrates through its inertia to produce a red noise spectral response. A full understanding of TPDV requires all these outstanding uncertainties be resolved.
Properly designed coupled model sensitivity experiments, in which SSTs are prescribed in certain regions, could be used to isolate the contribution of the different regional sources of wind anomalies. As these experiments can be affected by model biases and might be difficult to interpret 149 , they should be complemented by analyses of multivariate empirical models 150 , which are trained on observations and allow a cleaner decoupling of feedbacks among different variables and regions [151][152][153] . In addition, simple ocean models that capture Rossby wave dynamics 26,154 can help to assess the role of different aspects of the winds, including location and spectral characteristics, in reproducing key features of TPDV.
Although spiciness anomalies do not seem to substantially affect TPDV, current evidence is based on a limited number of analyses using just over two decades of observations from Argo floats and primarily conducted with ocean-only models. However, the expected concentration of variance at decadal timescales of spiciness anomalies arriving at the equator, and the resulting rearrangement of the tropical climate, suggests that spiciness anomalies could still be a potentially important driver of TPDV in the coupled setting. Thus, the role of spiciness should be further investigated in the context of coupled models. Availability of long time series from model simulations with realistic mixing parameterizations, achieved through either higher spatial resolution or improved model design, would be critical to more reliably assess the impact of spiciness on TPDV. Finally, a major limitation in our understanding of TPDV stems from the relatively short observational record, which does not allow a robust characterization of decadal variability, and a proper assessment of climate models fidelity in simulating it. More extensive investigations of TPDV using multicentury paleoclimate records could provide critical insights on the key features of TPDV and better constrain climate models evaluation at decadal timescales (Box 2). This Review has not addressed the question of how TPDV might change in response to external forcing. However, changes in the characteristics of TPDV as a result of anthropogenic forcing can be expected. Increasing surface temperatures will result in increased ocean stratification 155 , leading to faster Rossby wave propagation, shorter adjustment timescales and reduced growth and predictability of Pacific decadal variability 156 , which might lead to weaker, shorter timescale TPDV in the future 157 . The expected reduced influence of Atlantic variability on ENSO owing to increased tropospheric stability 158 can also reduce the influence of Atlantic decadal variability on TPDV. Yet, the wind-evaporation-SST feedback is projected to increase owing to warmer sea surface temperatures and increased evaporative response, leading to an enhanced impact of the NPMM on ENSO and possibly on TPDV 159,160 . These, and other possible processes and their interactions, need to be assessed in climate models to determine how TPDV might change in a warmer world.
The brevity of the instrumental record limits analyses of tropical Pacific decadal variability (TPDV) with instrumental observations. Paleoclimate proxies, particularly tropical corals and sclerosponges, provide opportunities to track the low-frequency variations of the tropical oceans over centuries. Over the most recent phase transitions of TPDV, corals have recorded associated changes in dynamically relevant fields, including sea surface temperature 171,172 , salinity [173][174][175] , westerly wind bursts 176 and upwelling 177,178 . Proxy records have provided evidence of interactions among different ocean basins at both interannual 179 and decadal 180 timescales. Proxy records from the Eastern Subtropical North Pacific, where sea surface temperature anomalies might reflect NPMM activity, illustrate high levels of decadal variability coherent with the Central Equatorial Pacific records, supporting the potential involvement of the NPMM in TPDV 174 .
Additionally, paleoclimate analyses provide a perspective into the range of TPDV found over centuries-millennia, which can be used to assess model simulations of TPDV. The figure compares TPDV across five different instrumental products 161,[181][182][183][184] , two generations of climate models (CMIP5, CMIP6, historical 185,186 and Past1000 (refs. 187,188) experiments) and three different sources of paleo data (coral δ 18 O from the central and eastern equatorial Pacific [189][190][191] , field reconstructions 192,193 and paleo data assimilation products 194,195 ) using violin plots 196 . The number of data sets used for each violin is indicated by N. TPDV is described in terms of the standard deviation of decadal variations (7-70 years) of the Niño3.4 index (annually averaged sea surface temperature anomalies in the 5° S-5° N, 170° W-120° W region).
Violin plots for each data set are based on decadal standard deviations of 100-year sliding windows allowing for 50 years overlap between segments. Individual dots represent the decadal standard deviation of each unique 100-year segment. The median and interquartile range of these values is indicated by the white dots and vertical lines, respectively, whereas the width of the violin plot for each standard deviation indicates the corresponding frequency of occurrence. Notably, the instrumental record does not cover the full range of decadal variability suggested by both paleoclimate proxy reconstructions and climate models, although the median standard deviation is very similar among products.
CSIRO Oceans and Atmosphere, Aspendale, Victoria, Australia.