Global changes in oceanic mesoscale currents over the satellite altimetry record

Oceanic mesoscale eddies play a profound role in mixing tracers such as heat, carbon and nutrients, thereby regulating regional and global climate. Yet, it remains unclear how the eddy field has varied over the past few decades. Furthermore, climate model predictions generally do not resolve mesoscale eddies, which could limit their accuracy in simulating future climate change. Here we show a global statistically significant increase of ocean eddy activity using two independent observational datasets of surface mesoscale eddy variability (one estimates surface currents, and the other is derived from sea surface temperature). Maps of mesoscale variability trends show heterogeneous patterns, with eddy-rich regions showing a significant increase in mesoscale variability of 2–5% per decade, while the tropical oceans show a decrease in mesoscale variability. This readjustment of the surface mesoscale ocean circulation has important implications for the exchange of heat and carbon between the ocean and atmosphere. Mesoscale eddy variability has increased in eddy-rich regions by 2–5% per decade but decreased in the tropical ocean over the satellite record (1993–2020). These changes will impact ocean–atmosphere heat and carbon exchange, with implications for regional and global climate.

C hanges in the climate system over recent decades have warmed the upper ocean and modified the wind stress, heat and freshwater fluxes that drive ocean circulation 1,2 . These changes have the capacity to modify the ocean circulation, including the overturning circulation 3,4 , basin-scale gyres 5,6 and boundary currents 7,8 . Changes in climate can also affect mesoscale processesfor example, through changes in wind stress forcing over the Southern Ocean 9 . The oceanic mesoscale incorporates motions that occur at spatial scales from approximately 10 to 100 km. These motions include both steady flows (such as jets and recirculations) and time-varying flows (such as meanders and coherent vortices). Generally, time-varying mesoscale flows are also referred to as eddies. Mesoscale eddies are ubiquitous in the global ocean and feed back onto all scales, from regional processes 10 up to the global meridional overturning circulation 3 . Moreover, these eddies transport and mix tracers such as heat, salt and nutrients 11,12 . Understanding the evolution of the mesoscale circulation is therefore crucial to formulating better predictions of our changing oceans.
Kinetic energy (KE) quantifies the magnitude of ocean currents 9, [13][14][15] . KE is proportional to the square of the velocity and is commonly separated into the mean KE (MKE, computed from the time-mean velocity field) and the KE of the time-varying velocity (known as the eddy KE (EKE)). The EKE is dominated by mesoscale variability and is a substantial fraction of the total KE 16,17 . A recent study has inferred a global increase in KE anomaly from ocean reanalyses and Argo floats 15 . However, these reanalyses and observations do not have the spatial resolution required to resolve the mesoscale field. Moreover, the ECCO ocean state estimate shows a slight speed-up of the currents, with a weak trend of surface KE 18 . In contrast, satellite observations resolve the mesoscale field at latitudes between 60° S and 60° N, and suggest that EKE in the Southern Ocean and northeastern Pacific has a robust increasing trend 9, [19][20][21] . However, the global multidecadal trends of mesoscale eddy activity from satellite observations are yet to be quantified.
Mesoscale flows have a footprint in both sea surface height (SSH) and sea surface temperature (SST). EKE can be directly inferred from SSH via geostrophy, and mesoscale eddies act to strain and shear the temperature field, meaning that regions of high EKE are associated with strong mesoscale SST gradients. Observed SST gradients can therefore be considered a proxy for mesoscale eddies [22][23][24] .
In this study, we examine the evolution of mesoscale eddies using satellite observations of SSH and SST over the satellite altimetry record . We use two independent datasetsnamely, altimeter SSH and National Oceanic and Atmospheric Administration Optimum Interpolated SST (v.2.1) 25 -to estimate EKE and SST gradients, respectively (Methods). These fields are then temporally smoothed using a running average of 12 months to eliminate the seasonal cycle. The trends and the significance of each field are computed using a linear regression and a modified Mann-Kendall test 26 (see Methods for further details). Mesoscale variability is spatially heterogeneous; thus, we explore the trends of mesoscale eddies both globally and regionally.

Global mesoscale eddy trends
Over the past three decades, ocean thermal expansion and the melting of land ice have led to an increase in SSH 1,27 (Fig. 1a). This SSH increase can be observed in all ocean basins, but there is also regional variability (Fig. 1b). SSH gradients are proportional to the surface geostrophic flow, from which we can compute velocity anomalies and EKE (Methods). The time-mean EKE highlights eddy-rich regions including boundary currents and their extensions, the Antarctic Circumpolar Current (ACC), and the equatorial band (Fig. 1d). These oceanic eddy-rich regions show statistically significant trends over the satellite altimetry record from 1993 to 2020 ( Fig. 1f and Extended Data Fig. 1), which suggest a regional long-term adjustment of the ocean mesoscale eddy field. Moreover, the global surface area-integrated EKE has a positive trend of ~1.2% per decade (0.09 ± 0.04 PJ per m per decade; statistically significant at the 95% confidence level). The spatial structure of EKE trends is Global changes in oceanic mesoscale currents over the satellite altimetry record Josué Martínez-Moreno 1 ✉ , Andrew McC. Hogg 1 , Matthew H. England 2 , Navid C. Constantinou 1 , Andrew E. Kiss 1 and Adele K. Morrison 1 Oceanic mesoscale eddies play a profound role in mixing tracers such as heat, carbon and nutrients, thereby regulating regional and global climate. Yet, it remains unclear how the eddy field has varied over the past few decades. Furthermore, climate model predictions generally do not resolve mesoscale eddies, which could limit their accuracy in simulating future climate change.
Here we show a global statistically significant increase of ocean eddy activity using two independent observational datasets of surface mesoscale eddy variability (one estimates surface currents, and the other is derived from sea surface temperature). Maps of mesoscale variability trends show heterogeneous patterns, with eddy-rich regions showing a significant increase in mesoscale variability of 2-5% per decade, while the tropical oceans show a decrease in mesoscale variability. This readjustment of the surface mesoscale ocean circulation has important implications for the exchange of heat and carbon between the ocean and atmosphere.
highly heterogeneous, although its zonal average shows notable net tendencies, with increasing trends observed polewards of 25° S and 40° N (Fig. 1e,f). A strengthening of the EKE field is a direct indication of an increase in mesoscale currents.
SST is an independent dataset relative to SSH but is also influenced by mesoscale eddies and has better temporal and spatial resolution than SSH. SST has increased on multidecadal timescales due to climate change 28,29 (Fig. 2a), with a heterogeneous global spatial pattern modulated by interdecadal climate variability 28 (Fig. 2b). The time-mean SST gradients again highlight eddy-rich regions, such as boundary currents, their extensions and the ACC (Fig. 2d). These regions with large SST gradients also exhibit some of the largest positive SST gradient trends, while the subtropical gyres and the tropics mostly exhibit decreasing trends (Fig. 2e,f and Extended Data Figs. 1 and 2). The global area-integrated SST gradient magnitude has increased at a rate of 3.9 ± 1.33 × 10 6 °C m per decade or 0.2% per decade (95% confidence level) relative to the global time-mean area-integrated SST gradient magnitude (1.7 × 10 9 °C m). Moreover, SST gradients are enhanced by stretching and straining due to mesoscale eddies. Further analysis shows that mesoscale SST gradients (Extended Data Fig. 3; length-scales smaller than 3°; Methods) dominate the observed trends, increasing at a rate of 5.37 ± 0.94 × 10 6 °C m per decade (0.4% per decade; statistically significant at the 95% confidence level). This analysis confirms that the observed SST gradient trends are a consequence of the mesoscale eddy field stirring the temperature field.
EKE and mesoscale SST gradients show analogous spatial and temporal responses in the boundary currents and their extensions, the ACC, and the tropics. Note that eddy-rich regions such as the Kuroshio Current, the Agulhas retroflection, the Gulf Stream and the East Australian Current show large changes in mesoscale SST gradients colocated with some of the largest EKE changes (Fig. 3). Even though these fields do not match perfectly, we quantified the areas of same-sign trend for each of these four regions (Extended Data Fig. 4). We find that the increasing and decreasing trends of SST and EKE match to a good extent for the Kuroshio Current, the Agulhas retroflection and the East Australian Current (61-65% of same-sign area agreement). The spatial patterns of these independent satellite products further suggest an intrinsic response of the mesoscale eddy field to a changing and variable climate.

Spatial patterns of ocean mesoscale trends
EKE and mesoscale SST gradient trends both indicate a net strengthening of the global mesoscale activity. However, both  datasets reveal heterogeneous patterns of increasing and decreasing trends. Thus, to further understand the spatial variability, we first focus our analysis on individual area-integrated regions: namely, the Southern Ocean (by which we mean south of 35° S) and the Pacific, Indian and Atlantic Oceans north of 35° S (Fig. 4d). This analysis reveals that the Southern Ocean and the Pacific Ocean are largely responsible for the global area-integrated trends and variations of EKE and mesoscale SST gradients; the trends in the Indian and Atlantic Oceans are much smaller (Fig. 4a,b). The Southern Ocean shows a statistically significant increase for both the EKE and SST gradients, where the observed changes have been attributed to the strengthening of the wind stress since the early 1990s 9 . The Pacific Ocean SST gradient decreases significantly, with the EKE signal also decreasing, albeit below the 95% significance level (Fig. 4c,e). The large uncertainty in the Pacific EKE trend (the orange error bars in Fig. 4c) is a consequence of the pulses in the time series during 1997 and 2015, both being El Niño onset years. These large, anomalous interannual signals dominate the uncertainty of the global EKE trend.
El Niño events are associated with a strengthening of the North Equatorial Countercurrent and the northern branch of the South Equatorial Current 30,31 , particularly during extreme eastern Pacific El Niños, such as those that occurred during 1997-1998 and 2015-2016 (ref. 32 ) (the grey bars in Fig. 4a,b). During such El Niño events, the equatorial currents generate transient circulation anomalies that extend over the equatorial band (9° N-9° S). After a scale decomposition of the velocities, we observe that these EKE pulses correspond to features located in the equatorial band and have scales larger than the typical mesoscale eddy size 11 (approximately 10 to 100 km; Methods and Extended Data Fig. 5a,c). Thus, equatorial currents during El Niño events modulate the equatorial EKE response, and the strong interannual variability conceals EKE trends over the equatorial region.
To further investigate the effect of El Niño events on the mesoscale, we remove the equatorial regions (9° S-9° N) and repeat the global trend analysis for EKE and SST gradients. The global area-integrated extra-tropical EKE and SST gradient trends increase, while the corresponding relative uncertainties decrease; namely, 7% of the area is statistically significant above the 95% confidence level; for the spatial distribution, refer to Extended Data Fig. 1c). c, Zonally averaged time-mean of SST gradient magnitude (|∇SST|). d, Map of time-mean of SST gradient magnitude. e, Zonally averaged SST gradient magnitude trend. f, Map of SST gradient magnitude trend (81.6% of the area is statistically significant above the 95% confidence level; see Extended Data Fig. 1d). Note that the spatial pattern of SST gradient maps is independent of the temporal extent of the SST gradient record used to compute the SST gradient trends (Extended Data Fig. 2).
EKE trends are 1.8% ± 0.25% per decade, and mesoscale SST gradient trends are 1.6% ± 0.09% per decade (see the striped bars in Fig. 4c,e), both significant at the 95% confidence level. It is thus clear that mesoscale activity in the Pacific, and particularly in the equatorial region, is strongly influenced by interannual variability.

Eddy-rich regions become richer
The observed changes in EKE and SST of whole ocean basins integrate over large, heterogeneous regions with opposing trends. For example, the Pacific Ocean aggregates the strengthening of the equatorial currents in the equatorial Pacific Ocean during El Niño events, boundary currents and the broader-scale oceanic gyres. These dynamical regions are not unique to the Pacific Ocean; the Atlantic and Indian basins also span diverse dynamical regions. We therefore further decompose the ocean into dynamical regions ( Fig. 5d): namely, (1) the ACC and surrounds, (2) the boundary currents and their extensions, (3) the equatorial regions and (4) the subtropical ocean gyres (see Methods for the dynamical region definitions). The remaining regions are aggregated into a fifth group. We then investigate the variability and trends within each of these dynamical subregions.
Globally, there is a significant increase in EKE and SST gradients; however, each dynamical region shows a different response (Fig. 5). For example, the ACC region shows a significant increase in both EKE and SST gradients at rates of 5.1% and 3% per decade (Fig. 5c,e), consistent with an increase in eddy activity with strengthening wind stress, as demonstrated in previous studies 9, 13,20 . Boundary currents and their extensions collectively show a similar net response, in which EKE and SST gradients both increase at rates of 2.5% and 8.1% per decade, respectively. Individually, SST gradients increase in all boundary currents. However, whereas EKE in the Agulhas retroflection, the East Australian Current, the Leeuwin Current and the Malvinas Current has significantly increased, the Gulf Stream and the Kuroshio Current do not show a significant net strengthening 33,34 (Extended Data Fig. 6); instead, regions of increase and decrease tend to cancel each other out in an area integral. This cancellation is particularly evident for the Kuroshio Current ( Fig. 1f and Extended Data Fig. 6b). The response seen in the Gulf Stream and Kuroshio Current is consistent with a poleward shift in these currents 7,35,36 and a readjustment to climate modes 37 . Note that a poleward shift cannot be captured by our static climatological definition of the boundary current regions (Methods). A shift of the boundary currents will thus result in an increase in EKE and SST gradients outside our dynamical definitions (that is, in regions poleward of the boundary currents and ocean gyres; see the white regions in Fig. 5d). The observed strengthening of these remaining regions is comparable to that of the ACC and suggests a poleward shift of the boundary currents, particularly in the Northern Hemisphere.
The equatorial and subtropical gyre regions exhibit statistically significant negative SST gradient trends and statistically non-significant negative EKE trends (Fig. 5), suggesting a reduction of the mesoscale eddy variability in the equatorial region and the interior of the subtropical gyres (Fig. 2). The equatorial region is dominated by interannual variability, where large changes corresponding to El Niño events occur in both the EKE and SST gradient time series. The significant decrease in SST gradients in the subtropical gyres could result from the displacement of the extratropical atmospheric circulation 38,39 and the expansion of the tropics 40 . The decreasing SST gradient trends in the tropics could be due to a homogenization of the tropical surface SST gradients. In the future, the SST gradients induced by mesoscale stirring are expected to reduce as the surface ocean becomes more thermally homogeneous. However, a longer record is required to separate the mesoscale response from interannual-decadal climate variability.
We have analysed the available satellite altimetry record of SSH and SST to reveal a significant global intensification of the mesoscale eddy field over the satellite record. While the observed global surface percentage increases per decade may seem small (~0.5% per decade), like ocean heat content and sea-level trends, even small fractions of a percentage change correspond to a large energy perturbation of the Earth system. For example, if we assume that the mesoscale flow extends vertically to around 500 m in depth (a reasonable assumption given the vertical structure of the gravest mode 41 ), then the observed EKE trends integrated over 500 m imply a significant increase of 0.78% in the global volume-integrated mesoscale energy budget (1.3 × 10 19 J) 42 over the satellite altimetry record after removing the El Niño signal (Extended Data Fig. 5b). This percentage is equivalent to 1.03 × 10 17 J, the same order of magnitude as the global internal tide energy budget (10 17 J) 42 . EKE in eddy-rich regions exhibits even larger significant strengtheningfor example, 2% in boundary current extensions and 5% in the ACC (Fig. 5). Increased mesoscale activity observed in both SST and mesoscale SST gradients is most apparent in regions where eddies are already strong. These eddy-rich regions are critical for carbon and heat uptake by the ocean 43,44 , and they are known to be sensitive to climate modes that are readjusting in a changing climate-for example, the strengthening of the westerly winds 9 linked to recent increasing trends in the Southern Annular Mode 45,46 . With ongoing future projected changes in the westerly wind belt, it is expected that mesoscale activity in the Southern Hemisphere boundary currents and the ACC will continue to increase over the coming decades. Current generation models used for climate projections (CMIP6) do not generally resolve mesoscale eddies 47 ; thus, important climatic adjustments driven by changes in the eddy field are likely to be missing from these climate projections.
The mesoscale evolution described here cannot be captured by coarse-resolution reanalysis products and sparse Argo float observations, as previously used in other studies 15 . Here we have used eddy-permitting satellite observations to reveal a potential decrease in EKE over the tropics, while reanalysis products suggest that the tropics is where KE anomalies have increased the most 15 . The KE anomaly differs from EKE because it contains the time-mean flow in addition to the time-varying components. The source of differences between Hu et al. 15 and our results can be assessed by an analysis of the ocean KE anomaly using satellite observations. The KE anomaly trends are almost identical to the EKE trends presented above (compare Fig. 1 and Extended Data Fig. 7), yet strikingly different from those obtained by Hu et al. 15 (their Fig. 2a). This suggests that the difference between our results and those of Hu et al. 15 arises from the inability of reanalyses and the Argo dataset to resolve the mesoscale, due to coarse resolution and sparse sampling, rather than from the definition of the KE anomaly. An alternative explanation of the differences between our study and Hu et al. could arise from the KE anomaly trends in Hu et al. being depth-integrated (0-2,000 m), while our EKE calculations are derived solely from surface diagnostics. In addition, as we have demonstrated, the tropics are strongly influenced by interannual variability (such as El Niño), and yet the KE time series from coarse reanalysis data does not detect the two extreme El Niño events observed in the satellite record. Data resolution and subsurface ocean dynamics are therefore the likely causes of the discrepancy between our eddy-permitting analysis and the results from prior work using reanalysis products 15 .
There are several possible causes of the observed trends in mesoscale activity, including (1) changes in winds (such as wind curl and wind stress), (2) changes in stratification, (3) changes in large-scale horizontal temperature gradients and (4) changes in the shear of the ocean currents. These forcing agents can impact the eddy activity via a combination of processes-for example, the non-local intensification of winds, the outcropping and tilting of isopycnals and the strengthening of baroclinic and barotropic instability. Atmospheric reanalyses show distinct and inconsistent wind stress trend patterns, while records of in situ measurements of isopycnal tilt and baroclinic growth rate are too short and too sparse to provide evidence of a dynamical mechanism driving the observed increase in eddy activity. Longer observational records with higher temporal and spatial frequency are thus required to better understand the increase in EKE observed from satellites. In addition, as discussed above, trends computed from coarse-resolution ocean reanalysis products 15 with parameterized mesoscale eddies differ from the mesoscale trends we detect from eddy-resolving satellite altimetry. An in-depth analysis of the dynamics leading to the observed mesoscale eddy evolution should therefore be explored in more detail using truly eddy-resolving global ocean models or eddy-resolving reanalysis products.
Our study has documented a major global-scale reorganization of the ocean's mesoscale KE observed over the past three decades. These observed adjustments in the mesoscale field have the potential to affect ocean circulation at regional and global scales and to modify the transport and redistribution of tracers, such as heat, carbon and nutrients. Our findings thus have major implications for   48 ). Note that the y axis is discontinuous in a and b. c, Linear EKE trends for each basin. d, Ocean basins; the equatorial region (9° S-9° N) is marked by the dashed lines. e, Linear mesoscale SST gradient trends. In c and e, standard errors are shown with orange bars, and statistically significant trends (above the 95% confidence level) are solid bars, while non-significant trends are translucent. The regions are distinguished by colour: global (solid black), Southern (blue), Indian (magenta), Pacific (green) and Atlantic oceans (grey) and each region separately without the equatorial region (striped bars).
ocean readjustment to a changing climate, as the enhancement of the mesoscale ocean currents may feed back on the sequestration of anthropogenic heat and carbon into the ocean.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/ s41558-021-01006-9.    a and b, the solid curves are 12-month running averages for each region, the dashed lines correspond to statistically significant time-series trends and the dashed-dotted lines show insignificant time-series trends. c, Linear EKE trends for each dynamical region. d, Definition of the ocean regions. e, Linear mesoscale SST gradient trends. Note that in a, the top curve that corresponds to the tropics has a different scale than the rest. In c and e, standard errors are shown with orange bars, and statistically significant trends (above the 95% confidence level) are solid bars, while non-significant trends are translucent. The regions are distinguished by colour: tropics (yellow), ACC (cyan), boundary currents and their extensions (magenta), subtropical ocean gyres (green) and the rest (grey).

Methods
Observational products. The data used in this study include SSH, geostrophic velocities and SST. The EU Copernicus Marine Service Information (CMEMS) gridded multimission SSH and geostrophic velocities have a horizontal resolution of 1/4° (although the effective resolution may be coarser in some regions 11 ). Currents in the equatorial region (5° S-5° N) are estimated using an equatorial β-plane approximation of the geostrophic equations 49 . National Oceanic and Atmospheric Administration Optimum Interpolation SST has a horizontal resolution of 1/4° (ref. 25 ). This dataset is constructed by combining observations from different products (such as satellites, ships, buoys and Argo floats).
These datasets have a quasi-global coverage (65° S-65° N) and span 27 years, from January 1993 to March 2020. The SST product is available for a longer duration, but we analyse only the period of overlap with the altimetry record (Extended Data Fig. 2). Anomalies were computed with respect to the record's climatology. We have verified that using a different period for defining the climatology does not change the observed trends in the anomaly fields. We have also verified through wavelet analysis of the time series at individual points that there is no evidence that steps in the record occurred due to improved technology in satellite missions and oceanographic observations. KE decomposition. KE density is decomposed into the energy density contained by the steady flow (time-mean) and that contained by the transient flow (time-varying). In other words, the surface geostrophic velocity components are split using a Reynolds decomposition into their time-mean (u, v) and time-varying components (u ′ = u − u, v ′ = v − v), with bars denoting time-averages over the whole record. The terms u′ and v′ are the anomalies of the surface geostrophic velocities provided by CMEMS, which are proportional to the SSH gradients (via the geostrophic approximation and equatorial β-plane approximation). The KE is therefore decomposed as: where we approximate the density of the seawater by the constant ρ 0 = 1,025 kg m −3 . The energy contained in the time-varying component of the flow is known as the EKE, while the MKE is the energy of the time-mean flow.
Maps of EKE in this study correspond to the time-mean EKE, defined as where the units of EKE (x, y) are J m −3 . Time series correspond to the surface area-integrated EKE (globally or over specific regions): where A refers to the area of each geographical or dynamical region, angle brackets 〈 〉 denote the area integral and 〈EKE〉(t) has units of J m −1 , as it is multiplied by the grid area. Furthermore, the velocity field was decomposed into mesoscale (u m , v m ; scales smaller than 3°) and large-scale (u ls , v ls ; scales larger than 3°). To decompose the velocity field, we first compute u ls and v ls by a spatial convolution with a constant 3° × 3° kernel K: and the mesoscale − → u m is defined as: The mesoscale and large-scale EKE can then be computed using these velocity fields.
SST gradients. Analogous to mesoscale and large-scale EKE, SST gradients are decomposed into mesoscale (SST gradients with scales smaller than 3°) and large-scale (SST gradients with scales larger than 3°). To decompose the SST gradients, we first compute large-scale SST by using a spatial convolution with a constant 3° × 3° kernel K and a 12-month running average-that is, where the tilde e I denotes a 12-month running average. The mesoscale SST is then defined as The gradients of the large-scale and mesoscale SST are computed afterwards. The SST gradient magnitude is: with analogous expressions for SST m and SST ls .
Computations of SST gradient time-series and time-mean SST gradient trend maps are analogous to those of EKE-for example, for the area-integrated SST gradients: Trends, significance and uncertainties. Linear trends are calculated using a linear least-squares regression for spatially integrated time series. For the trend maps, the fields are first coarsened to a 1° × 1° grid, and then the linear trends are computed for each grid point. All the observed trends for EKE and SST gradients (time-series and trend maps) are assessed using a Theil-Sen estimator, while the statistical significance uses a modified Mann-Kendall test 26 . This statistical test takes into account autocorrelations within the time series. Finally, the reported uncertainties in Figs. 4c,e and 5c,e correspond to the standard error using the effective sample size from the Mann-Kendall test-that is, the standard deviation of the time series divided by the square root of the effective sample size.
Geographical and dynamical regions. Geographical regions consist of the following ocean basins (Fig. 4d): the Southern Ocean (south of 35° S), the Indian Ocean, the Pacific Ocean and the Atlantic Ocean. These ocean basins were defined to capture ocean processes at all scales. The ocean basin mask can be obtained from the repository https://doi.org/10.5281/zenodo.3993823, which contains all the data used for this study (refer to data availability; file-name ocean_basins_ and_dynamical_masks.nc). Dynamical regions were defined from the climatological mean SSH and the MKE (Fig. 5d). We defined a mask for each dynamical region; then we extracted and masked each dynamical region in the following order, to avoid any overlap between regions: 1. The equatorial region is defined as the region between 9° S and 9° N. 2. The boundary currents and their extensions are defined as regions with MKE above the ~99th percentile (2.8 σ). Note that the Peruvian and Californian currents are weaker (below the 99th percentile of MKE); therefore, according to our definition, they do not qualify as boundary currents. 3. The subtropical gyre masks depend on each ocean basin: the Pacific Ocean gyres correspond to mean SSH above the 0.65 m contour, the Atlantic Ocean gyres correspond to mean SSH above the 0.36 m contour and the Indian Ocean gyres correspond to mean SSH above the 0.60 m contour. All these values were tuned to approximately capture the same extension as the theoretical estimation of ocean gyres according to the Sverdrup balance. 4. The ACC and its surrounding areas is defined as all remaining regions left between 35° S and 60° S. The dynamical regional mask can be obtained from the repository containing all the data used for this study (refer to data availability; file-name ocean_basins_ and_dynamical_masks.nc).