The Increase in Global Ocean Heat Content and Favorable Conditions for Tropical Cyclone and CYCLOP Intensification: Accounting for El Niño

Abstract

The ocean heat content (“OHC”)—the heat energy within the ocean integrated to a reference depth—has physical drivers spanning spatial and temporal scales, including seasonality, the El Niño/Southern Oscillation (ENSO), and others. The present article investigates changes in the OHC100 during the period 1994–2020 using GLORYS12 monthly averaged ocean reanalysis. OHC100–ENSO correlation patterns are explored to glean insights about the oceanic mechanisms that facilitate the ENSO’s global teleconnections. After extracting known seasonality and ENSO signals using the Oceanic Niño Index (ONI), the OHC100 residual is analyzed to investigate multidecadal drivers of the OHC100. Lagged ENSO–OHC100 correlations (±12 months) reveal basin-scale oscillations in the sign of ENSO influence likely attributable to Rossby waves. The OHC100 is increasing globally (in total, 2.4 × 10²² J decade⁻¹), with the greatest increases near western boundary currents (WBCs). Some regions are decreasing, notably the Atlantic main development region (MDR) for tropical cyclones (TCs). Correlations and multidecadal variability in the OHC100 tendency (OHCT) and zonal and meridional advections of the OHC100 (ZAO and MAO) support the hypothesis that upper-ocean dynamics mediate ENSO teleconnections as well as exert independent control on OHC100 variability. Local increases in the OHC100 would support the observed TC rapid intensification irrespective of the ENSO phase as the TC-supporting region expands.

1. Introduction

It has long been established that because the Earth system is not in local thermal equilibrium, there is complex transport of heat from the equator to the poles that establishes global (systemic) thermal equilibrium [1]. Geologically recent changes in the Earth system have led to greater temperatures in the ocean [2,3] and atmosphere [4]. The ocean has a much higher heat capacity than the atmosphere [5], and 93% of the energy responsible for warming the Earth works to warm the ocean [6]. Hence, the ocean has a strong regulatory effect on the Earth system temperature, and its heat content is of considerable scientific and practical interest.

Thermodynamic ocean feedback to the atmosphere acts mainly via the ocean heat content (OHC) (an energy source) and sea surface temperature (SST) (which regulates the rate of heat exchange between the ocean and atmosphere). The OHC is well-studied, with several landmark papers quantifying the amount of heat within various depths of the ocean using observational datasets [7,8] as well as ocean reanalyses [9,10]. Global, gridded, physically consistent reanalyses are ideal for studying global patterns on multidecadal timescales. One major area of inquiry regarding the effects of multidecadal global ocean temperature change is in the frequency, distribution, and intensity of tropical cyclones (TCs). TCs are known to derive their energy source from the warm ocean, and the impact of the observed ocean warming on TC characteristics is complex and multifaceted. This is due to feedback mechanisms between the ocean and atmosphere acting across spatiotemporal scales, which may amplify or diminish TC characteristics circumstantially. The unexpectedly active TC seasons during the El Niño of 2023 highlight the challenge of predicting the outcome of these competing factors [11,12]. In contrast, recurrent mesoscale warm-core ocean eddies or marine heatwaves (MHWs) may amplify the OHC and drive TC intensification locally without the negative feedback associated with the ENSO.

The complexities of these processes are not sufficiently described by the SST alone [13,14]. Supporting evidence so far indicates that the thermal influence of the ocean on weather is limited to within about 300 m of the ocean surface [7], rather than depths of 300–2000 m [6,9] or whole ocean depth to which the OHC is sometimes calculated for analysis of Earth’s energy budget. MHWs are sometimes quantified using the heat content of the top 50 m [15] and arise both from heating of the ocean and OHC convergence. Mesoscale ocean eddies frequently have a strong core in the upper 100 m but not at the surface [16]. Physically, this means that only the near-surface ocean (about 100 m deep [17]) can interact with the atmosphere on spatiotemporal scales associated with weather. However, this relationship can be circumstantially perturbed by the ENSO, confounding attempts to study the long-term changes in the 100 m OHC (OHC100). Yet, there has been no attempt to remove the ENSO signal from the OHC100 to better understand the future implications for extreme weather potential. With these observations in mind, this study seeks to establish the OHC100’s relationship with the ENSO (i.e., correlations) and its multidecadal changes (i.e., non-ENSO residual) with commentary on implications for TCs. An additional goal is to leverage the global analysis to examine how implications for TCs may apply to dynamically similar extreme weather, including so-called “medicanes” and polar lows (PLs), which collectively have been dubbed Cyclones from Locally Originating Potential Intensity (CYCLOPs) [18].

Past efforts to take stock of changing upper-ocean OHC [6,7,9] have generally conducted the analysis without disaggregating the large, well-known ENSO signal from residual non-ENSO changes. Johnson & Lyman [6] examined the global patterns in the observed 700 m OHC for the period 1993–2019 and found that 56% of the global upper-ocean area is significantly warming and 3% is significantly cooling. Local values of the change over 27 years ranged from −8 to 7 Wm⁻², with an absolute value an order of magnitude greater than the global average, 0.60 Wm⁻² [6]. The global heating they calculated was equivalent to 0.42 Wm⁻² over the Earth’s entire surface during the study period [6]. This evaluation of historical changes demonstrates the dominant ocean warming pattern but does not address the degree of command that the ENSO exerts versus unquantified effects. Cheng et al. [9] examined the OHC100 and associated heat fluxes under varying ENSO conditions, confirming a strong diabatic negative OHC tendency (that is, flux of heat stored in the ocean to the atmosphere) during the positive ENSO phase, as well as vertical and horizontal redistribution of the OHC100 in their study domain equatorward of 60°. Xu et al. [19] addressed global change in ocean temperature from the standpoint of MHW occurrence and found that MHWs evident in the SST increased almost worldwide due to local effects rather than the ENSO during 1958–2017 in observations and CMIP6 historical simulations. Understanding the changes in the residual is as essential for anticipating future changes to the global system as changes in the ENSO. Since temporal variability is much greater at local than global scales, the more granular the assessment of these changes, the better the representation of dynamic regions like western boundary currents (WBCs) and shallow coastal regions. Deciphering the relative contributions of the ENSO and residuals to the OHC100 can support conclusions about future impacts on TCs.

On larger spatiotemporal scales, the ENSO is associated with a strong redistribution of the OHC, either through a direct influence, as in the Pacific Ocean [20], or via teleconnections globally [21]. Cooling of the ocean and atmosphere occurs globally during and after El Niño events [9], although the exact mechanism [be it the “delayed oscillator”, “recharge–discharge oscillator”, “Western Pacific oscillator”, or “advective–reflective” paradigm] and timescale by which the ENSO impacts the OHC are not fully understood. The delayed oscillator model formalizes the way in which Rossby waves and returning Kelvin waves could be involved in generating low-frequency oscillations associated with the ENSO [22]. The recharge–discharge oscillator model emphasizes the importance of building and dissipating OHC [22], a process largely mediated by the OHC tendency (OHCT) and meridional advection of OHC (MAO). The advective–reflective oscillator model was inspired by observations of reflecting equatorial Rossby and Kelvin waves and their associated zonal advection of OHC (ZAO) impacts [22]. Naturally, these factors are all represented in the unified ENSO oscillator model, of which the former models are all special cases. A substantial fraction of Earth’s ocean and atmosphere variability is explainable by correlation with the ENSO—even at timescales shorter and longer than the well-known 2–7 year ENSO cycle timeframe [23]. Additionally, the ENSO is asymmetric with respect to its positive (El Niño, warmer Pacific) and negative (La Niña, cooler Pacific) phases [24,25,26]. The broadband, asymmetric, nonstationary [27,28] nature of the ENSO contributes to the fundamental mystery of how it interacts with drivers of long-term background changes (henceforth “residuals”). Global changes in temperature may be classified as “Niño-like” (warming) or “Niña-like” (cooling) if they follow spatiotemporal patterns similar to the ENSO [29]. It has been proposed that historical external forcing, if it were to continue, would lead to more Niño-like conditions [30]. However, local drivers of change may not merely add to the ENSO-like conditions imposed by the global changes. For instance, local changes in temperature or heat fluxes may lead to changing local support for extreme weather but may not have other features associated with large-scale dynamics such as impacts on vertical shear of horizontal wind. ENSO impacts may occur with a lag an order of magnitude larger than the timescale of extreme weather (e.g., months earlier or later) and be facilitated by teleconnections [31,32]. This complex interaction may determine whether changes in the ENSO and residuals have positive or negative impacts on features of the global Earth system, including extreme weather such as TCs.

A rich body of literature has emerged aiming to determine the magnitude and predictability of TC characteristics and their oceanic and atmospheric drivers, including the ENSO. This literature has largely addressed the SST, the OHC, ocean circulation, and the frequency, duration, intensity, and size of TCs. In each of these aspects, ocean temperature is a dominant controlling factor, with global and local drivers. Kortum et al. [33] investigated the relative roles of SST-forced changes and weather variability in driving shifting Atlantic TC tracks and frequency. They made a case for a substantial “weather”-driven component to changes in TC distribution and noted that internal climate variability (i.e., the ENSO) exerted little control, emphasizing a need to understand local factors influencing TC genesis location [33]. That paper was motivated by observations that the TC frequency has increased and the geographic distribution in the North Atlantic shifted during 1970–2021; however, Chand et al. [34] found that TC frequency has decreased globally and regionally relative to the pre-industrial era (during the 20th century), consistent with the argument that large-scale warming weakens atmospheric circulation and suppresses TC formation. Feng [35] analyzed Western North Pacific TC translation speed during 1980–2023 and found that 80 km decade-1 poleward migration of TCs increased the basin-wide translation speed by 5% (due to the beta effect), but regionally, an 18% slowdown occurred. The TC slowdown and poleward migration are linked to the same underlying SST warming and feedback to atmospheric steering flow and increasingly favorable conditions for TC maintenance [35]. Li et al. [36] examined rapid intensification (RI) in coastal regions (as opposed to the deep ocean where historical TC intensity studies have mainly focused) and found that RI events significantly increased in frequency and moved landward during 1980–2020, driven by intensification-favorable ocean conditions. Relative humidity and vertical wind shear were more favorable near the coast than in the open ocean, whereas TC potential intensity (PI) increased uniformly across most of the global ocean and was a dominant factor for RI [36]. Studies of PI [13,37,38,39,40,41,42,43,44,45,46,47]—a theoretical measure that prognosticates the strongest wind speed a TC could attain under given atmospheric and oceanic thermal conditions—have called attention to the controlling role of ocean dynamics in TC intensity without fully addressing the importance of ENSO influences on the OHC. Current methods of calculating PI focus on the SST, although it is known that subsurface ocean structure can impact TC intensity immensely [48,49,50,51,52,53,54,55] and that depth-integrated temperature metrics, like the OHC, can be better predictors than the SST [13,14]. Estimation of PI from thermodynamic principles is desirable due to challenges measuring and forecasting actual intensity changes. These research efforts collectively call for deeper exploration of how the OHC that interacts with the atmosphere to directly or indirectly influence TC geographic distribution, genesis, and intensification.

2. Materials and Methods

2.1. GLORYS12 Monthly Average Ocean Reanalysis and ENSO Index

The ocean reanalysis GLORYS12 provides geophysical variables on mean-monthly scales. GLORYS12 is a global eddy-resolving physical ocean and sea ice analysis having 1/12° × 1/12° horizontal resolution and covering the period 1993–present (of which we use the period 1994–2020) [56]. GLORYS12 is regridded bilinearly upscale onto a 0.5° × 0.5° monthly averaged grid. The Oceanic Niño Index 3.4 (ONI) was obtained from the National Weather Service website [57]. The ONI will be referred to as the “ENSO” for the sake of simplicity since it represents the SST anomaly in the Niño 3.4 region on which the prevalent ENSO indices are based. GLORYS12, like other reanalyses, has inherent biases with respect to its representation of, e.g., ocean temperature [58]. Nevertheless, because of the fair agreement with other reanalyses and the fact that fixed biases are unimpactful on residuals calculated for the present analysis, GLORYS12 serves as a suitable proxy for a global gridded observational dataset.

2.2. Justifying OHC100

Disaggregation of the ENSO and residual influences on the OHC100 is essential for identifying how the OHC100 has changed and may evolve in the future. The strong influence of the ENSO on the TCHP illustrates this well and supports the choice of the OHC100 for quantifying the ocean heat content in this study. The TCHP has been a popular metric for the OHC historically and long regarded as a natural metric to study ocean impacts on TCs. The justification for this metric has been the observation that TCs tend to intensify over waters warmer than 26 °C [59,60,61]. However, more recent research has revealed that this threshold is a non-mandatory condition for TC maintenance and intensification [62]. Despite this shortcoming, the TCHP has a stronger relationship with TC intensity than the SST does [14]. The area of the TCHP region increased on average by over 22,000 km² annually during the period 1994–2020 (Figure 1a), concurrent with a prevalently positive change in TCHP magnitude in the same period (Figure 1b). The mid-latitudes to the poles experience less intense solar radiation and therefore a cooler climatological ocean temperature and a smaller ≥26 °C area. Waters warmer than 26 °C expanded from the tropics towards the poles during the 1994–2020 study period. Periods of expansion and contraction largely coincide with transitions in sign of the ENSO phase, superimposed on a long-term change associated with the residual. Figure S1 illustrates how the TCHP correlation with the ENSO is still strong at the edges of the TCHP-defined region post removal since the region is often larger (smaller) in positive (negative) ENSO phases. Figure S2 shows how the frequency with which the TCHP exceeds zero varies with latitude. The effects of the ENSO can be treated consistently for a fixed OHC integration depth and no threshold temperature, as with the OHC100.

2.3. OHC100, Its Time Tendency, and Horizontal Advections

The OHC100, the total OHC above the 100 m isobath, is integrated from the base vertical level closest to 100 m to the surface—in the deep ocean regions of the GLORYS12 reanalysis, this is typically 112 m depth, and in shallow coastal regions, it can be shallower than 100 m. OHC100 advection that acts on the ocean–atmosphere heat budget is calculated by combining the OHC100 with ocean velocity fields and presented in its zonal (ZAO) and meridional (MAO) components. ZAO and MAO convergence can lead to accumulation of heat and impact air–sea interactions. The OHC100 tendency (OHCT) is calculated as a centered difference in the OHC100 with respect to time. All three of these OHC100 budget terms (collectively “budget terms” henceforth) have units of heat flux (Watts per square meter). It is important to keep in mind that the OHCT is positive (negative) when flux is into (out of) the upper 100 m of the ocean by convention. OHCT fluxes are generally between the ocean and atmosphere, although it is possible that some component of the flux is across the base of the upper 100 m region for which it is calculated. A positive OHCT can be regarded conceptually as the “temporary storage” of heat in the ocean and a negative OHCT as flux to the atmosphere available to drive atmospheric dynamics such as TCs. The OHCT has the opposite impact on heat available to the atmosphere compared to the ZAO and MAO of the same sign.

2.4. Quantifying Correlations, Residuals, and ENSO Impacts

The influences of known signals—the seasonal cycle and ENSO, which vary in space and time—can be disaggregated from the remaining short- and long-term residuals with standard signal processing techniques. While the seasonal cycle has some nuance due to its spatial variation, it is not the focus of the present study; it is removed by calculation and subtraction of the month-wise mean (e.g., mean of all January timesteps at each latitude and longitude—call this “deseasonalization”). Similarly, the ENSO contribution to any variable in question can be removed using its statistical relationship with the ONI. The ONI is calculated based on a lagged climatology and therefore possesses a bias towards El Niño conditions due to a warming climate [63], but it is the simplest reliable method to quantify the ENSO for the signal processing performed here.

Correlations with the ENSO and residuals for the OHC100, ZAO, MAO, and OHCT are computed for the study period 1994–2020 and examined as spatial maps. The convention is to call correlation at n months of lag “lag n correlation”. The lag 0 correlation illustrates the relationship between the time series of the ENSO and the OHC100, OHCT, ZAO, and MAO in their respective subplots during the month when the ENSO phase and observed data occur. This is in contrast with negative lag correlations (lag −12 to lag −1 correlations) for which data occur before the ENSO and positive lag correlations (lag +1 to +12 correlations) for which data occur after the ENSO. While a strong correlation need not suggest a causal relationship between the ENSO and a given variable, a strong correlation is often a consequence of causal links. It is reasonable to interpret low correlation as a non-causal relationship at the given lag—this does not preclude a strong correlation from occurring at a different lag and should not be generalized for all lags including those outside of the scope of this study. However, it should be noted that significant correlation is more difficult to establish at greater lag times since the number of degrees of freedom to work with decreases with increasing lag. With 324 months (27 years) of GLORYS12 monthly averaged ocean reanalysis, the number of degrees of freedom is 324 − 2 = 322, which for a two-tailed t-test with a significance level α = 0.05 (T ≈ 1.968) implies that, in general, r-statistics with magnitude greater than 1.968 ([1.9682 + 322]1/2) − 1 = 0.109 are significant on all correlation plots in these results. For the ±12 lag correlations, it is necessary to use 24 fewer months to properly align the ONI and data, in which case, the significant r-statistic magnitude is 0.113.

Penland and Matrosova [27] as well as Compo and Sardeshmukh [64] provided concise explanations of the difficulties surrounding ENSO removal and a critique of the regression method utilized here. The present study benefits from the perceived disadvantages of quantifying the ENSO with regression since the spatial correlation maps can be examined through the lens of how ENSO signals propagate in lag space under the assumption of linearity. Some components of the ENSO and the residual are intertwined (viz. nonlinear), with different frequency components of the ENSO influence peaking in correlation at different lags [23,24,27,28,64]. Hence, it must be kept in mind that the correlations should be read as “the linear correlation with Niño 3.4 region SST from lag 0 with the state at each spatial coordinate at the current lag”. It should also be kept in mind that the residual represents “the change of the component of the variable in question that is, conservatively speaking, not related to ENSO or seasonality”, knowing that the true change may be enhanced by nonlinear interactions with the ENSO. Our correlation analysis focuses on lags ±12, spanning the timescale for which significant correlations are known to exist (for instance, a couple of seasons later in the Atlantic [65]). We avoid the difficulties associated with empirical orthogonal function (EOF) analysis and Fourier frequency filtering, since we hypothesize that dynamic aspects of the ENSO (i.e., dipole and higher harmonic modes of ENSO variability that are necessary for observed signal spatial propagation to exist) and residuals are important for effects on TCs. The ENSO regression method for examining residuals has been justified in multiple publications [29,66].

3. Results

This section examines the OHC100’s relationship with the ENSO and the multidecadal changes in the non-ENSO residual. These research objectives introduced in Section 1 are framed with two research questions regarding ENSO-driven and residual-driven components of the OHC100 in Section 3.1 and Section 3.2. Detailed examination of the OHC100, OHCT, ZAO, and MAO correlations with the ENSO is organized by ocean basin within 3.1. These results are contextualized with implications for TCs in Section 4, which separately considers where the ENSO and residuals dominate in Section 4.1 and how these results impact TCs and extreme weather more broadly in Section 4.2.

3.1. How Do ENSO Correlations of the OHC100, ZAO, MAO, and OHCT Change from ±12 Months Lag Globally at the Mesoscale?

Figure 2 depicts lag 0 ENSO correlation with the OHC100, OHCT, ZAO, and MAO. Strong (|r| > 0.25) and significant (|r| > ~0.1, see Section 2.4) correlations occur for all four variables. The OHC100 (Figure 2a) has significant correlations nearly globally, whereas the OHCT (Figure 2b), ZAO (Figure 2c), and MAO (Figure 2d) are significant exclusively within 20° latitude of the equator. None of the variables exhibited features identifiable as WBCs, which appears to contradict the previous literature finding strong ENSO signals in WBC strength [67]. Features likely associated with propagating equatorial Rossby waves (dispersive, propagate westward along the thermocline) and Kelvin waves (non-dispersive, propagate eastward along the thermocline about three times faster than Rossby waves) were apparent. Evidence of equatorial features agrees well with the proposed ENSO mechanisms.

Transition zones between positive and negative OHC100-ENSO lag 0 correlations (Figure 2a) are a notable feature that has been evident in previous studies [9,68]. These transition zones must occur due to the continuous nature of the r-statistic, under circumstances where two variables can have a changing sign of correlation. However, their coherence is not enforced in any way—they could very well have been incoherent transitions from one sign to the other, but this is not the case. Approximately zero correlation occurs in a strip usually oriented north–south with sub-basin-scale east–west meanders, a few hundred kilometers across; the approximate positions of these features are marked with solid lines (Figure 2a). The transition zones remain coherent and evolve at the various lags studied (Figure 3). Lags ±1, ±3, ±6, and ±12 are shown in Figure 3, and all lags showing the temporal continuity of these patterns are available in the Supplementary Material (Figures S3–S7). The transition zones are sharper around lag 0 than at ±12 lag (Figure 3) but remain evident across the lags examined (Figure S3). Considering transition zones one by one in the Pacific, Indian, and Atlantic Oceans, some patterns of changing position, sign, and intensity in the lag space emerge.

In the Pacific Ocean, the peak positive correlation of the OHC100 and ENSO moves from Northeast and West Pacific regions at lag −12 to the equator at lag 0, then reverses direction, precessing around the continental and equatorial margins of the North Pacific. The eastward propagation from lag −8 to lag 0 and poleward redistribution at positive lags described by Trenberth [69] is evident in Figure 3. This is consistent with the adiabatic horizontal redistribution of the OHC by the recharge–discharge mechanism. A negative correlation in the West Pacific grows from −12 to the peak at +3, concurrent with the eastward movement of a region of positive correlation in the South Pacific. The Indian Ocean has positive correlations generally when the equatorial Pacific does (and vice versa), although the correlations are weaker. Lag +12 shows weak correlations of both signs at all locations; this is likely due to the myriad known dynamical factors controlling the Indian Ocean OHC100. The tropical Indian Ocean SST is known to show basin-wide warming about six months after positive ENSO, called the Indian Ocean basin (IOB) mode of the Indian Ocean SST interannual variability [70]. The triggering of downwelling Rossby waves by anomalous easterlies [71] is a likely mechanism by which correlations propagate at positive lag times. There is also evidence that meaningful correlations with the ENSO can persist until lag +12 in the Indian Ocean: Rossby waves can lead to weakening of the southwest monsoon a year later in the summer [72] and, hence, positive wind–evaporation–SST feedback [73], warming the whole basin. The basin-wide oscillation is evident in the SST lag correlation with the ENSO (Figure S4). The IOB persists from boreal winter to the following summer despite dissipation of the positive-ENSO-related SST anomalies in the central and Eastern Pacific [26]. The Atlantic Ocean signals are weaker than in the other basins but are usually negatively correlated at negative lags and positively correlated at positive lags. The gradual increase in correlation with the ENSO at positive lags in the Atlantic is consistent with the documented anomalous post-positive-ENSO Atlantic warming and post-negative-ENSO Atlantic cooling [74].

The unified oscillator model of the ENSO predicts that the physical processes that regulate the ENSO are linked to the ZAO, MAO, and OHCT. The OHCT lagged correlations with the ENSO (Figure 4) exhibit negative correlations with ENSO lock-step, with OHC100 positive correlations on the same timescale. This matches the association of the OHCT with the off-equatorial discharge of accumulated OHC to the atmosphere, although the OHCT still has strong ENSO correlations well outside of the tropics (Figure S5). ZAO and MAO relationships with the ENSO (Figures S6 and S7) are principally linked to equatorial wave processes and only weakly correlated at higher latitudes. One might imagine that OHC100 correlations should be a weighted average of the correlations of the mediating oceanic processes involved. However, it seems more likely that in regions where OHC100 correlation has a high R magnitude, but the mediating oceanic processes do not, ENSO impacts are mediated by atmospheric teleconnections that cannot be adequately resolved with ocean reanalysis alone. For instance, in the equatorial Pacific, it is known that surface wind mediates zonal transport directly or indirectly by triggering wave dynamics [75], but in off-equatorial regions, the direct impacts of surface-wind-mediated processes that modulate the OHC100 may not manifest as a coherent transport signal. In the equatorial Pacific, this may be related to the Madden–Julian Oscillation (MJO), an atmospheric disturbance that propagates eastward from the equatorial Indian Ocean to the Pacific Ocean on 30–90-day timescales. This manifests as strong correlations between the ENSO and zonal and meridional ocean transport between the equatorial Pacific Ocean and Indian Ocean (Figure S10).

Away from the equator and in the Atlantic Ocean, the transition zones are better explained by changes in ocean temperature, likely induced by atmospheric changes attributable to the ENSO (as opposed to, e.g., changes in the intensity of western boundary currents, which showed no significant correlation with the ENSO). The substantial OHCT correlation at higher latitudes could be an indicator of atmospheric mediation. However, there are significant nontrivial differences between OHC100 correlations and SST correlations at the transition zones (Figure S8), confirming the importance of subsurface processes. Decoupling at even greater depths is expected to be greater, which agrees with results that suggested the OHC2000 leading the ENSO by a variable number of months between 1980 and 1999 [9]. Heat accumulates in the W. tropical Pacific during La Niña [76]. At negative lags, the W. tropical Pacific becomes increasingly polarized, with a sharp gradient from negative (west) to positive (east) correlation with the ENSO arising. Simultaneously, the positive correlation intensifies and grows eastward, with the strong positive core eventually touching S. America at lag −1 and beyond. This is consistent with observed patterns [28,76,77,78]. The general reversal of patterns between negative and positive phases that sends heat back to the tropics resembles the diabatic heating impacts of the ENSO recharge phase, about which there are few publications [79,80] despite it being an important aspect of frameworks explaining ENSO evolution.

3.2. What Does the Geographic Distribution of Residuals of These Variables Reveal About the Oceanic and Atmospheric Drivers of Change?

The OHC100 residual (Figure 5) is positive nearly everywhere with exceptions in the MDR, Agulhas Current, Niño 3.4 region, south of Greenland, around Antarctica, and a portion of the Kuroshio Current (Figure 5a). Skewness towards positive residuals and spatial patterns are similar to those in Johnson & Lyman [6] calculated for 1993–2019 and 700 m depth but are significant over a greater area in this study—possibly due to the careful treatment of the ENSO signal. The greatest warming is poleward of 30°; however, the equatorial region equatorward of 10° is warming everywhere except the Niño 3.4 region and a small piece of the MDR. In the Niño 3.4 region, the residual is expected to be relatively small since the ONI is defined using this region, and therefore, the ENSO should explain nearly all variability observed in this region. The residual there is nearly zero. Integrating the OHC100 residual over the area of Earth’s ocean yields a net warming rate of about 2.4 × 10²² J decade⁻¹, which is comparable with the findings from 1993 to 2021 for the upper 2000 m presented in Su et al. [8] calculated from other ocean models and remote sensing observations. When combined with the observation from Levitus et al. [7] that the OHC below 300 m does not substantially impact the global OHC variability, this indicates that most of the global OHC residual is represented in the OHC100. However, this should not minimize the impact of warming at greater depths, which is within an order of magnitude of the surface.

Figure 4. Selected lag correlations between OHCT and ENSO at ±12-month lags.

Figure 4. Selected lag correlations between OHCT and ENSO at ±12-month lags.

Figure 5. Residuals of OHC100 (a), OHCT (b), ZAO (c), and MAO (d). The solid horizontal line marks the equator; dashed and dotted horizontal lines mark 10° and 30° north and south of the equator, respectively. Note the different units and scale of the OHC100, a measure of heat, compared to the other three fluxes. Also note the factor of five difference between the larger ZAO and MAO scales and the OHCT scale.

Figure 5. Residuals of OHC100 (a), OHCT (b), ZAO (c), and MAO (d). The solid horizontal line marks the equator; dashed and dotted horizontal lines mark 10° and 30° north and south of the equator, respectively. Note the different units and scale of the OHC100, a measure of heat, compared to the other three fluxes. Also note the factor of five difference between the larger ZAO and MAO scales and the OHCT scale.

The decreasing OHC100 in the Irminger Sea (south of Greenland) likely reflects increasing loss of heat due to interaction with buoyant surface glacial meltwater—and not necessarily just from the southeastern fjords, because the Fram Strait delivers water to the region from virtually all of Greenland’s coast. Some component may also be attributable to the observed Atlantic Meridional Overturning Circulation (AMOC) weakening since 2015 [6]. Correspondingly, decreases near Antarctica are likely related to meltwater entering the Southern Ocean. To our knowledge, the strong negative OHC100 residual near 10° N off the west coast of Africa has not been identified to date. This region is strongly impacted by Saharan dust, which impacts radiative forcing over the Atlantic Ocean [81] and leads to cooler SSTs [82,83]. The shape of the region of decreasing OHC100 in the MDR matches the typical warm dry region circumscribed by a north–south-oriented warm atmospheric ridge that results from Saharan dust [84]. If this similarity in shape of the OHC100 residual and the region impacted by dust is due to aerosol forcing, Saharan dust effects may have increased in intensity or frequency during the study period. This aligns with the finding of Miller & Ramseyer [85] that Saharan air layer outbreaks became 10% more common during 1981–2020. Figure S9 shows the SST residual globally for comparison with the OHC100 residual; in the MDR, the SST has a weaker signature than the OHC100, which may have led to this feature being overlooked in previous studies. An additional factor that could have obscured this oversight is the fact that significant ENSO impacts on the SST and OHC100 are in some regions, including the MDR, not in-phase and can even have opposite signs (Figure S8). The quantified decoupling between the correlations of the OHC100 and SST with the ENSO can be explained by the lead of shallow OHC100 ahead of SST development documented by Meinen & McPhaden [77]—however, the results here have a broader geographic scope than the central equatorial Pacific, and in other regions, the SST may lead the OHC100.

The MDR decreasing OHC100 region is not the only mesoscale feature apparent in the OHC100 residual: unlike the ENSO correlations in Figure 3, Figure 4 and Figure 5, WBCs and eddy-like features are apparent and dominate the areas with the largest residuals in addition to the equatorial wave-like features. Changes in basin-scale circulation have already been observed, with both a poleward shift in global subtropical WBCs and warming of WBCs 2–3 times the global mean [86]. These features are even more apparent in the OHC budget term residuals (Figure 5b,d). In all three budget terms, the heat flux residuals are most pronounced at WBCs, the equator, and the ITCZ mean position. The residuals are weak at the highest latitudes where the ocean is frequently ice-covered.

It may be useful to think of the OHCT residual (Figure 5b) as the “acceleration of OHC100” during the 27-year period in question. The OHCT residual is more negative (positive) with increasing (decreasing) flux to the atmosphere. All areas with a negative OHC100 residual (Figure 5a) exhibit a positive OHCT residual (Figure 5b), but the opposite is not universally true. In the Pacific, the band of locally elevated OHCT residual between 5 and 10° N coincides with a positive OHC100 residual. The tropical Indian and Atlantic Oceans both exhibit an east–west basin-scale asymmetry with a positive OHCT near Africa and a negative OHCT elsewhere. At the poleward extensions of the WBCs, there is typically a greater occurrence of a negative OHCT residual than near the tropical origins of the currents; the warming tropical ocean transports heat poleward via WBCs, which would lead to increased fluxes via the OHCT to the atmosphere at mid–high latitudes.

Not all areas with large changes in the ZAO and MAO residuals (Figure 5c,d) have similarly strong residuals in the OHC100: this should be interpreted as changes in ocean circulation impacting the transport of the OHC100 more than the change in the OHC100 itself. A nuance of advection residuals is that the sign convention, positive for east in the ZAO and positive for north in the MAO, leads to the interpretation of increasing residuals as “more eastward” and “more northward” transport as well as “increasing magnitude”. That is, a negative residual may arise from a weakening or change in direction of flow. However, the direction of the monthly average flow does not generally undergo reversals, so residuals should be interpreted as mostly attributable to changing magnitude. The greatest impact of flow reversal is likely in eddy-rich regions where changing mesoscale variability may lead to a dynamic ocean feature translating in space. As the ocean warms, there is more OHC100 to transport. The ZAO and MAO can attain values of around 5000 W m⁻² yr⁻¹ in the areas of the most intense increase. One possible explanation for the mesoscale structure of the residual maps is that there are variations in mesoscale dynamics on timescales similar to the length of the 27-year dataset. Thus, mesoscale eddy processes may transport long-term signals on a multidecadal timescale, localizing rather than diffusing changes in the system. Near the equator, the wave-like structure may indicate amplification of the OHC signal that the waves carry. These patterns can be attributed to two factors: increasing tropical OHC100 that is pumped poleward by the global thermohaline circulation and movement of the western boundary currents. The former is a robust signal based on the global change in the OHC100, whereas the latter is sensitive to the selection of the analysis period, since the position of western boundary currents has substantial annual-to-multidecadal variability [87].

4. Discussion

4.1. Which Regions Are ENSO-Dominated, and Which Are Residual-Dominated?

Impacts of a changing upper-ocean heat budget on TCs occur across spatial and temporal scales. The ENSO has a global broadband signal that peaks on 3–7-year timescales, and long-term residuals with multidecadal timescales accumulate via changing dynamics that are most energetic at the mesoscale. These signals combine to dictate how the ocean drives TCs, controlling their thermal energy source and modulating the physical processes that shape their characteristics. The ENSO exerts its strongest influence on the tropical Pacific but ripples outward geographically and in lag space to have direct impacts and teleconnections globally. The results presented here support the various distinct ENSO models and the propagation mechanisms that they suggest, as well as their combination in the unified oscillator model. Wherever the magnitude of correlation with the ENSO approaches unity, the ENSO explains almost all variability, even on multidecadal timescales. Elsewhere, other factors that impact the residual are important. Disaggregating the influences of the ENSO and residuals means identifying where the residual is important enough to reduce the ENSO correlation—at all lags. Observing that the WBCs, high latitudes, and shallow coastal regions possess both an intense increasing OHC100 residual and a weak ENSO influence in general, these areas stand out as being controlled mainly by local non-ENSO processes. Over the course of years or decades, impacts on the order of ENSO-driven variations elsewhere may impact these regions. As for the other regions, the robust increasing residual over much of the world ocean would likely lead to more Niño-like conditions in the tropical Pacific, as in Vecchi & Soden [88]. The ENSO impacts the intensity and occurrence of extreme weather events including TCs [89], which is likely to intensify when the residual agrees in sign with the ENSO impact. This highlights the simultaneous importance of mesoscale features and the dominance of the ENSO in these areas.

4.2. What Are the Implications for TCs and Other Extreme Weather?

It is important to determine what effect changes in the OHC100 would have on the dynamics that control TC characteristics. For instance, vertical shear of horizontal wind (which is detrimental to TC formation and intensification) may only be impacted by large-scale changes in heating that occur on at least months-long timescales (ENSO-like) and could be unaffected by smaller or shorter-scale heat flux changes (local), or vice versa. Using increasing temperature as an example, because most of the world is experiencing this type of change, there are two possibilities. One is that warming will lead to more positive-ENSO-type conditions globally on average, and the diagnosed effects of the positive ENSO phase on TCs will prevail, with less supportive atmospheric conditions but greater energy to drive storms that survive. The other possibility is that the greater energy source will persist, but because the mechanism is not ENSO-like, the negative feedback that would reduce TC intensity might not kick in, leading to a unilateral increase in TC strength if the pattern continues. In the latter case, increased latitudinal shear of ocean temperature and increased stratification work to increase TC-induced cooling and dampen TC intensification [90,91]. Geographic variation in ENSO strength opens the possibility of a mixed response, with some regions experiencing intensified ENSO-like variability and others experiencing mainly effects of the residual. Increases in conditions supporting TCs in the central Pacific could, via increased TC activity, result in the increased transport of warm water to the cold tongue zone of the equatorial Pacific and promote El Niño conditions, as hypothesized by Fedorov [92]; in contrast, regions distantly impacted by the ENSO via teleconnections like the N. Atlantic could experience TC intensification simply by merit of local heat flux changes.

A deep warm subsurface layer can directly enhance heat fluxes to the atmosphere under TC conditions [93]. Positive residuals of the ZAO and MAO indicate increasingly favorable mean conditions for TC intensification. A warming ocean increases not only the heat available to TCs for intensification but also the temperature disparity between the surface and outflow level near the tropopause. This temperature disparity is a primary factor controlling TC intensity, so regions with an increasing OHC100 are likely to see an increasing PI as well. The ENSO modulates TC activity worldwide [94], and patterns in the ENSO lagged correlations may provide early indicators of TC activity when the supportive ENSO phase coincides with the TC season. For instance, OHC variability in the Eastern Pacific Ocean is valuable for TC forecasting in the region [95,96], and this region regularly shows high correlation with the ENSO. Similarly, correlations driven by teleconnections at greater lag times could provide some predictive capability on subseasonal timescales. Warming in inland seas such as the Mediterranean Sea may explain the recent increase in so-called “medicanes” during the study period [18,97]. Local maxima in residuals may also explain recent increases in TC frequency in the Bay of Bengal relative to the Arabian Sea [98].

Marine heatwaves (MHWs) are a possible mechanism by which changing ocean temperatures force extreme weather across scales. MHWs originate and evolve under oceanic and atmospheric forcing. MHWs are often associated with the ENSO [99], although there may be multiple mechanisms by which MHWs occur, only some of which are related to the ENSO. For those that are ENSO-related, it has been shown that MHW forecast skill can increase when models are initialized during El Niño [99] in areas that are highly correlated with the ENSO. The importance of local versus remote dynamics on MHWs is a current area of research [100]. MHWs can increase in frequency even if local SST variability does not increase [19], which can be attributed to the importance of subsurface processes like ZAO and MAO convergence. Changing ocean conditions, whether due to the ENSO or residual effects, may impact the prevalence or intensity of MHWs, with a cascade effect on the frequency or intensity of extreme weather. Increasing ZAO and MAO point to OHC100 convergence contributing to MHW intensity or occurrence; a positive OHCT residual suggests increasing flux into the ocean and may support the idea that atmospheric forcing contributes to increasing MHW longevity in these regions [101]. Changes in MHWs have been hypothesized to feed back to TC rapid intensification in some regions [102], especially in shallow regions vulnerable to coastal downwelling during TC approach [50]. The ENSO correlation lag plots may indicate regions that could see MHW conditions months in advance. Positive residuals may also explain the recent normalization of historical marine heat extremes [2]; a shift in extreme conditions from relatively rare to common could result from global increases in the OHC100, and increased occurrence of extremes could result from increased ZAO, MAO, and OHCT contributing to enhanced variance in the OHC100. The locations of eddy-like features in the residuals may indicate where oceanic mesoscale eddies could be contributing to changes in MHWs [15].

PLs have dynamical similarities to TCs [18,103] that make many arguments applying to TCs also applicable to PLs in the context of ENSO and OHC100 impacts. In the PL and TC literature alike, the thermodynamic disequilibrium between the surface (typically over the ocean) and the upper troposphere is used to evaluate environmental favorability for formation and intensification [104]. PI theory has likely not been applied to PLs to date because of conflicting conclusions about what controls PL intensity in the literature. However, it is important to separate conceptually the influence of ocean temperature on a particular PL from the influence of ocean temperature on PL occurrence and intensity on longer timescales. Warm WBCs are known to be important for TC intensification, but even at high latitudes, they influence PL climatology [105]. Since the WBCs exhibit the greatest rates of increasing OHC100, which is related to the SST, this is likely to drive up the probability of formation of PLs in regions impacted by WBCs. Additionally, if PI theory applies to PLs as well as TCs, as has been hypothesized [18,103], these PLs may become more intense over time as well. SST anomalies may not significantly impact particular PLs, but increased fluxes to the atmosphere due to ocean dynamics do create conditions favorable for PL intensification [106]. While, for PLs, positive SST anomalies may not increase the probability of PL formation as much as low air temperatures, the SST has a controlling influence on where PLs may form [105]. MHWs also occur in regions that may be impacted by PLs in both hemispheres. For a particular subset of MHWs, those due to stationary zonal wavenumber-4 (W4) anomalies [107], the locations of the peak positive anomalies in the Southern Ocean—between Australia and Africa, central Pacific east of New Zealand, Western Atlantic near the convergence of the Brazil and Malvinas Currents, and Western Indian Ocean regions—saw local maxima in positive OHC100, ZAO, MAO, and OHCT residuals.

5. Conclusions

The ENSO correlations of the OHC100, ZAO, MAO, and OHCT change in lag space, reflecting the different speeds with which components of the complex ENSO signal propagate. Lagged correlations of the OHC100 with the ENSO show strong correlations even months before and after changes occur, perhaps revealing sources of predictability for the OHC100 and its effects on TCs months in advance. The geographic distribution of the residuals of these variables reveals not just an increasing global OHC100 but also increasing heat budget terms over much of the world ocean. This partitions the effects of a warming ocean into ENSO-like effects with competing feedback to TC characteristics and local effects that increase support for strong TCs. The large-scale ocean warming may drive more Niño-like conditions with possible feedback to the atmosphere that could suppress TCs, such as increased vertical shear of horizontal wind and weakened large-scale circulation. However, more local features of the warming patterns may not contribute to ENSO-like feedback and may simply add to the energy source and thermal disequilibrium that drive TCs. It is also possible that the large-scale ENSO-like effects and local effects propagate and evolve differently—and not necessarily following the common “long and slow” versus “small and fast” dichotomy. For instance, even though the local effects occur on smaller spatial scales, in this case, they were quantified evolving over decades. There will be an opportunity to evaluate the sensitivity of these changes to the selection of the study period as longer and more reliable data-assimilative reanalyses are produced, extending the available reanalysis with realistic atmospheric forcing and with minimal errors from the early expendable bathythermograph era prior to 1994. Perhaps this will start with an ocean reanalysis coupled with the upcoming ERA6 atmospheric reanalysis.

The tropics are ENSO-dominated, and the residual is important near WBCs, at high latitudes, and in shallow coastal regions. The hypothesized mechanisms of ENSO propagation in the ocean are well-represented in lagged correlations. While a substantial fraction of the OHC100 is related to the ENSO, the residual drives change on smaller spatial scales. Mesoscale processes and global ocean circulation are critical to the difference between ENSO impacts and residual impacts. WBCs have the most intense positive OHC100 residuals, which could have an intensifying effect on landfalling TCs in North America and Asia. Interannual variability in TC characteristics influenced by the OHC100 may follow the spatial patterns of the ENSO correlation. Residuals will increase the availability of heat from the ocean to drive TCs even if unrelated ENSO feedback works to reduce the intensity or frequency or change the geographic TC distribution. Intense and rapidly intensifying TCs are likely to increase in frequency under the influence of increasing residuals, with MHWs serving as the mechanism that feeds ocean heat accumulation back to the atmosphere. The global changes in the OHC100 and ocean dynamics may impact other forms of extreme weather as well, leading to the changing frequency or intensity of PLs and explaining the rise of TC-like “medicanes” in the Mediterranean Sea. Ocean dynamics are important not just for mediating ENSO effects globally but for exerting independent control on local processes that feed back to TC intensity.

Future work could include the following:

• Incorporating analysis of atmospheric variables like surface and upper-troposphere temperature and winds from a compatible reanalysis such as ERA5 could provide further insights into how ENSO-like or “residual-like” ocean-driven extreme weather impacts were during the study period—this would inform what these impacts would be like if they continued into the future.
• The symmetry of OHCT, ZAO, and MAO responses to the ENSO at various lags could be confirmed—since the El Niño phase is known to last longer and have greater intensity than the La Niña phase, correlations may disproportionately represent the former, but on the other hand, the correlations are fairly strong, which may support a symmetric response to both phases.
• Further investigation of the OHC100 leading the SST in correlation with the ENSO and vice versa could be conducted—does this relate to the local direction of forcing between the ocean and atmosphere?