https://orcid.org/0009-0005-0917-778X
Figures
Abstract
The Agulhas Current is a vigorous western boundary current that flows off the southeastern coast of Africa, linking the Indian Ocean to the Atlantic Ocean as part of a broader inter-ocean current system. The Agulhas Current is characterized by intense heat fluxes between the ocean and atmosphere, as well as complex multiscale ocean dynamics. Dynamically unstable currents in this region generate mesoscale eddies, which propagate into the South Atlantic and interact with the Benguela Upwelling system and the Atlantic Meridional Overturning Circulation. These unique features make the Agulhas region an ideal site for studying air-sea interactions in climate models and examining the relative roles of atmospheric and oceanic weather in driving upper-ocean variability. In this study, we investigate how horizontal resolution in climate models affects their ability to represent thermal air-sea interactions over the Agulhas Current region, by comparing several simulations of state-of-the-art models. We identify ocean- and atmosphere-driven regimes using a covariance analysis of sea surface temperatures and turbulent heat fluxes. Our findings suggest that a minimum ocean model resolution of approximately 25 km is necessary to capture the signature of ocean dynamics, leading to a better alignment with theoretical and observational results. Furthermore, we identify a transition scale between ocean-driven and atmosphere-driven regimes within the 2°–5° range: when this critical length scale is exceeded, the ocean-driven behavior is filtered out and the atmosphere-driven regime, which becomes relevant at larger scales, dominates. While both oceanic and atmospheric resolution play a role, we find that increasing the horizontal resolution in the ocean component yields a comparatively larger improvement in the representation of air-sea flux variability.
Citation: Busatto J, Bellucci A, Adduce C, Yang C (2025) The impact of horizontal resolution on the representation of thermal air-sea interaction of the Agulhas system in coupled climate-models. PLOS Clim 4(9): e0000680. https://doi.org/10.1371/journal.pclm.0000680
Editor: Juan A. Añel,, Universidade de Vigo, SPAIN
Received: June 17, 2024; Accepted: August 1, 2025; Published: September 15, 2025
Copyright: © 2025 Busatto et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: The minimal data set required to replicate all study findings reported in the article are available for download at this DOI: 10.5281/zenodo.15699168.
Funding: This work is funded by European Copernicus Marine Service Contract GLO_RAN_Lot6, Evaluation of Ocean reanalysis in the tropical Ocean.
Competing interests: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
1. Introduction
The Agulhas Current (AC) is a western boundary current that flows along the South-Eastern coast of the African continent in the Indian Ocean between 27°S and 37°S, carrying warm waters poleward [1,2]. Once it reaches the tip of Africa, it retroflects toward the East in the Agulhas Return Current [3] following the subtropical front along the Antarctic Circumpolar Current. During the retroflection, the ocean current sheds anticyclonic eddies into the South Atlantic Ocean (Agulhas Rings) that propagate North - Westward carrying warm and saline water into the cold and fresher Atlantic basin [4–6]. The waters escaping from the Indian Ocean form the so-called Agulhas Leakage (AL). The Agulhas System is characterized by turbulence and nonlinear effects. Not only the leakage is dominated by mesoscale vortices (Agulhas Rings in this case), but the main current is also affected by meandering events [7,8], the so-called Natal Pulses, that bring the core of the current offshore and enhance the mixing effects of the different water masses, increasing the depth at which the mixing is allowed, rising the kinematic steering level [1].
The AL is modulated by several factors: bathymetry has a key role in the physical constriction of the current [9], perturbing the AC trajectory and forcing the current to intersect shallow regions and changing its path; meandering events, such as Natal Pulses [10,11], trigger ring shedding by themselves or by merging with Rossby wave-like meanders in the Agulhas Return Current; Southern Hemisphere westerly winds are shown to have a strong impact on the AL transport [12]. Submesoscale flows increase the strength of the AL, as demonstrated by comparing submesoscale-resolved and non-resolved simulations [13], highlighting the importance of resolution in numerical models for accurately representing ocean dynamics.
Being one of the main sources of saline and warm waters in the Southern Atlantic, the AL has a crucial role in the climate system and the Atlantic Meridional Overturning Circulation (AMOC) [14,15] generating buoyancy anomalies in the South Atlantic that induce dynamical responses in the AMOC [16,17]. Furthermore, the warm tropical waters carried by the current stimulate convection in the overlying atmosphere with direct consequences for regional weather systems [18], and extreme events, such as rainfall and droughts, are strongly related to the air-sea flux patterns [19].
In this context, it is important to investigate the representation of the processes occurring at the air-sea interface, particularly in climate models that have been widely used in climate studies of the Agulhas region. In terms of air-sea interaction, the ocean is not a purely passive element, affected by the variable atmospheric system. Due to its longer time-scale variability, it acts as a memory for the coupled system. The conceptual models of atmospheric-driven climate variability described by Hasselmann [20] hold over most of the mid-latitude ocean surface but fail where the mesoscale activity in the ocean is stronger, as it happens in the Agulhas region and, generally, over regions affected by the dynamics of western boundary currents. In his paper, it is shown that long-term climate variability can arise from the integrated effect of short-term weather fluctuations. By treating the climate system as a slow-response component driven by random atmospheric “noise”, it is demonstrated that such variability follows a red spectrum, matching observations. Furthermore, it is stated that intrinsic randomness limits climate predictability, even under ideal conditions.
The capability of reproducing physical phenomena in climate models is crucially dependent on their spatial (and temporal) resolution. Where small-scale dynamics are strong, the resolution has to be consistently high to resolve the underlying processes [21]. In the simplified framework of energy balance models, theoretical predictions for the covariance between sea surface temperature (SST) (and its tendency) and surface heat fluxes have been first proposed by Frankignoul and Hasselmann [22], and further elaborated by Barsugli and Battisti [23], Wu et al. [24], Bishop et al. [25], and Tsartsali et al. [26]. Two different regimes are found: an atmosphere-driven regime, characterizing the open ocean regions and consistent with the conceptual model described in Hasselmann [20], and an ocean-driven regime, dominating over dynamically active regions, such as the western boundary current systems. An atmosphere-driven regime occurs where the high-frequency atmospheric variability drives changes in surface ocean properties that have longer time scales of variation. On the other hand, in an ocean-driven regime, intrinsic variability associated with mesoscale turbulence (the so-called “ocean weather”) and current dynamics cause variations in ocean physical quantities that affect the atmospheric system.
Recently, Bellucci et al. [27] inspected the regimes of air-sea interactions by comparing several climate model simulations with different horizontal resolutions over the eddy-rich Gulf Stream region, following the approach outlined in Bishop et al. [25]. They found that coupled models (both eddy-parameterized and eddy-permitting) can discriminate between ocean- and atmosphere-driven regimes. However, the increase in model resolution leads to a better representation of SST and turbulent heat fluxes (THF) cross-covariance patterns, and the major improvements can be largely ascribed to a refinement of the oceanic model component. Following the approach outlined in Bishop et al. [25] and Bellucci et al. [27], we investigate the air-sea interaction in the Agulhas Region (15°S–55°S, 20°W–40°E) using climate models with different horizontal resolutions. The objective of our study is to understand the impact of horizontal resolution on the representation of air-sea thermal exchanges over this peculiar area of the global ocean. The Agulhas and Gulf Stream regions share similar eddy-rich dynamics. However, the two boundary currents have different spatial and topographic contexts: the Gulf Stream extends for over 3000 km and is influenced by shelf and slope bathymetry near Cape Hatteras, which strongly constrains ring formation [28], while the Agulhas Current flows along the South African coast for about 1000 km and interacts dynamically with the Agulhas Bank and Plateau. These topographic features modulate retroflection behavior and eddy-shedding [9], including phenomena such as early retroflections when the current attaches to the Plateau.
These different bathymetric influences, affect the scale and characteristics of mesoscale turbulence. This implies that numerical models may require different horizontal resolutions to adequately resolve air-sea interactions in each system. Therefore, investigating the impact of model resolution in the Agulhas region could also inform the broader effort of improving climate model representation of boundary currents and eddy dynamics.
2. Data and methods
2.1. Models and observations
For our analyses, we used a set of global climate model simulations from different models, performed according to the High-resolution Model Intercomparison Project (HighResMIP) experimental protocol [29]. HighResMIP is a Coupled Model Intercomparison Project Phase 6 (CMIP6) -endorsed model intercomparison effort designed for the systematic investigation of the impact of horizontal resolution on the model representation of processes relevant to the climate system. The HighResMIP setup consists of atmosphere-only and coupled ocean-atmosphere simulations performed at standard (100 km or coarser) and higher (25 km or finer) resolution. Numerical results used in this work are from climate simulations of the 1950–2014 historical period (“hist-1950”) forced with time-varying historical forcings (greenhouse gases, aerosol, land surface forcings, solar irradiance, ozone concentrations). In this study, we analyzed simulations with four different models (see Table 1).
Within the HighResMIP framework, the models’ resolution is modified in two different ways: in the case of EC-Earth3P, the horizontal resolution is increased in both the ocean and the atmosphere component, while with CMCC-CM2 and MPI-ESM1–2 only the atmospheric horizontal resolution is increased. For the HadGEM3-GC31 model, four configurations have been considered, using different combinations of ocean and atmosphere resolutions (see Table 1).
Model results are verified against observational estimates. For THF, two different satellite-based datasets are considered, accounting for the uncertainty in observational estimates: OAFlux [35] and J-OFURO3 [36]. OAFlux (Objectively Analyzed air-sea Fluxes for the global oceans), is provided by the Woods Hole Oceanographic Institution and covers the 1958–2020 period with a horizontal resolution of 1°. J-OFURO3 is the most recent version of the Japanese Ocean Flux Datasets with the Use of Remote Sensing Observations with a nominal resolution of 0.25°, covering the 1988–2017 period. The SST data used in this study comes from the J-OFURO3 dataset, which provides an ensemble median product derived from multiple global SST sources, combining satellite and in situ data (see Table 3 in Tomita et al., [36] for a full list). This approach enhances the robustness of the SST estimates by minimizing the influence of extreme values in individual datasets.
In the following analysis, monthly mean SST and THF fields are used. Since the focus of this study is on the monthly-scale variability, the seasonal cycle and long-term trends have been removed before the analysis. Model outputs and observations cover different periods. For our analysis, we selected the overlapping periods (from January 1988 to December 2014) to maintain consistency between all the datasets.
2.2. Covariance analysis
In this study, we follow the approach outlined in Bishop et al. [25] to identify ocean-driven and atmosphere-driven regimes of air-sea thermal interactions over the Agulhas region. Bishop et al. [27] used a stochastic energy balance model driven by noise forcing to provide a zero-dimensional representation of the heat budget in the ocean-atmosphere coupled system:
where “a” and “o” subscripts relate to the atmosphere and ocean variables, and T indicates the temperature. Heat fluxes are parametrized as differences in ocean and atmosphere temperature, with α and β being exchange coefficients normalized by the respective heat capacities of the atmosphere and ocean. Finally, the ter-s γ_o 〖,γ〗_a describe the radiative damping, and the N terms are Gaussian noise terms that describe the stochastic forcing associated with atmospheric weather and ocean mesoscale turbulence.
Bishop et al. [25] show that when only the stochastic forcing associated with the atmospheric weather is active in the energy balance equations (i.e., No = 0 in Eq. 2) and hence an atmosphere-driven regime is established, monthly SST tendency-THF lead-lag correlation exhibits a symmetric pattern, with a negative minimum at lag zero. In this case, the SST-THF correlation pattern is anti-symmetric with positive values for negative lag (SST leads for negative lags). On the other hand, when only the stochastic forcing associated with ocean intrinsic variability is accounted for in the ocean’s heat balance equation (i.e., Na = 0 in Eq. 1), implying a purely ocean-driven regime, a positive and symmetric SST-THF pattern is obtained. In this case, the SST tendency-THF lead-lag correlation is anti-symmetric with positive values for negative lags.
To interpret the physical meaning of the covariance and correlation patterns, one has to consider how heat fluxes and surface temperature are linked. When the atmosphere controls the upper ocean variability, an increase of the THF leads to a cooling of the surface ocean temperature (hence, a negative SST tendency), implying a negative correlation between SST tendency and THF at the zero lag. On the other hand, when the ocean surface variability is entirely governed by the intrinsic oceanic processes (represented by the stochastic noise term N_(o) in Equation 2), the SST-THF correlation pattern features a positive maximum at the zero-lag. As an example, the latter can be understood as the ocean dynamics (through heat transport convergence) generating a positive SST anomaly, being damped by positive (i.e., from ocean to atmosphere) THF anomalies.
3. Results
3.1. SST-THF cross-covariance
As in previously quoted studies [25,27], cross-covariance maps between SST (or SST tendency) and THF are used to illustrate whether the drivers of SST variability are related to intrinsic variability in the ocean or the atmosphere. First, we analyzed the cross-covariance between SST and THF for (−1, 0, + 1) -month time lags in observations (J-OFURO3 and OAFlux) and model simulations with different horizontal resolutions in the Agulhas current area (Figs 1 and 2, respectively). Note that, in the following analysis, a positive lag indicates that SST (and tendency) leads over THF. Moreover, SST tendencies are calculated using a bi-monthly central difference.
Positive lag means SST leads over THF.
Positive lag means SST leading over THF in observations and model simulations. Low horizontal resolution (top panel) and high horizontal resolution (bottom panel) results.
In observational records (Fig 1A and 1B), SST and THF show their strongest positive covariance at zero lag, particularly in the Agulhas Retroflection and the ring-path areas. There, at +1 and −1 month lags, the SST-THF covariances decrease symmetrically compared to the zero-lag value. The positive SST-THF covariances indicate that the SST variability is mainly due to ocean-driven processes, according to the interpretation provided by Bishop et al [25]. This interpretation, however, requires not only a positive covariance at lag 0, but also a symmetric decrease at lag ± 1 — a distinguishing feature of an ocean-driven regime. In contrast, atmosphere-driven regimes display an anti-symmetric pattern with negative values at lag + 1 (see Fig 1C and 1F). Indeed, in the Agulhas Retroflection region, currents flow southward from the Indian Ocean along the eastern coasts of Africa and interact with the cold waters from the Antarctic Circumpolar Current, generating turbulence and eddies that propagate into the South Atlantic, carrying warm and salty waters. Thus, observations clearly show that the region dominated by the Agulhas current dynamics is governed by an ocean-driven air-sea interaction regime. In the open ocean, instead, the SST-THF covariance exhibits the typical asymmetric pattern around the zero lag, with positive (negative) values for lag-1 (+1), indicative of an atmosphere-driven regime [25]. Concerning climate model simulations, the configurations with lower horizontal resolution, reproduce the observed patterns of atmosphere- and ocean-driven regimes, exceptions made for EC-Earth3P-LR and HadGEM3-GC31-LL (Fig 1C and 1F): these models fail to represent the ocean-driven regime signal in the Agulhas region, displaying an anti-symmetric pattern in regions where the ocean-driven regime is expected. Their high-resolution counterparts (Fig 1, bottom panel), on the other hand, feature a more realistic behavior, clearly indicating that an adequately high resolution in climate models is needed to reproduce the ocean-driven regime, which is affected by mesoscale and smaller-scale dynamics over eddy-rich areas.
Subsequently, SST tendency-THF covariance is inspected. In Fig 2, observational datasets J-OFURO3 and OAFlux (Fig 2A and 2B) show symmetry in open ocean areas, with a negative minimum for zero lag. The Agulhas Current region features the expected anti-symmetric pattern around the zero lag. Fig 2 shows results from climate model simulations for low and high-resolution configurations (top and bottom panels, respectively). For high-resolution climate-coupled models, at the zero lag (middle row in every panel), the SST tendency-THF covariance is close to zero. As stated by Bishop et al. [25], the heat exchange model described in equations 1 and 2 returns an anti-symmetric correlation function, i.e., a zero correlation value at lag zero. Our results are hence consistent with the theoretical analysis of Bishop et al. [25]. The observed antisymmetric covariance signal indicates an ocean-driven regime. In the open ocean, a symmetric SST tendency-THF lead-lag correlation pattern is found, indicative of an atmosphere-driven regime, consistent with observations. Compared to their high-resolution counterparts, low-resolution configurations (Fig 2, top panel) present weaker signals in the eddy-rich Agulhas Region, while the observed atmosphere-driven regime over the open-ocean areas is realistically reproduced. These results confirm the poor representation of ocean-driven covariance patterns in EC-Earth3P-LR and HadGEM3-GC31-LL models (Fig 2C and 2F).
Since EC-Earth3P-LR and HadGEM3-GC31-LL failed to reproduce the ocean-driven regime, we investigated the impact of their coarse ocean resolution by analyzing their SSH climatology and variance (as indicators of large-scale circulation and eddy activity). The 1993–2014 mean SSH of the EC-Earth3P and HadGEM3-GC31-LL simulations has been compared to the AVISO L4 product. The comparison shows that both the respective low- and high-resolution simulations capture the dynamic topography of the AC (see Fig A, Fig B and Fig C in S1 Text). However, in both cases, the strength of the circulation inferred from SSH gradients is notably weaker than observed in altimetry data. Furthermore, SSH variance in the two coarse resolution models is about two orders of magnitude weaker than in AVISO altimetry data. A key difference between the low- and high-resolution simulations of the AC is the absence of the large meanders associated with the Agulhas Retroflection in low-resolution models features that are well captured in high-resolution configurations.
Thus, despite the evident biases in the simulated AC structure and magnitude (affecting both low and, to a lesser degree, the high-resolution models), the low-resolution simulations reproduce the Agulhas system.
3.2. Local correlation
In this section, we focus on lead-lag SST (and SST tendency)-THF correlations over specific areas, representative of an eddy-rich and an open ocean region, respectively. Compared to the analysis of covariances, discussed in the previous section, here we consider a wider range of time lags.
To investigate the air-sea interaction in an eddy-active region, we select a 2° longitude-latitude box centered at 15.0°E, 41.5°S, located within the eddy-rich sector of the Agulhas system. This area is expected to exhibit a strong ocean-driven regime signal. Due to the coarse resolution of the HadGEM3-GC31-LL model, a larger 5° box is used instead for this configuration, in order to include a sufficient number of grid points. The selected boxes are shown as a yellow square in Fig 3F.
Blue lines represent SST-THF correlation, red lines represent SST tendency-THF correlation. Thicker solid lines are for observation (J-OFURO3), and thinner solid lines for high horizontal resolution configurations while thinner dashed lines are used for low horizontal resolution configurations. For simplicity of the figure, we only present observations from J-OFURO3 for comparison. In panel F, two boxes are shown to include the wider area used to sample HadGEM3-GC31-LL model.
In this eddy-rich region, all high-resolution models display a maximum SST–THF correlation at zero lag. In particular, CMCC-CM2-VHR, EC-Earth3P-HR, HadGEM3-GC31-HH, and HadGEM3-GC31-HM (Fig 3A, 3B, 3C and 3D) best reproduce the observed (and better resolved) J-OFURO3 correlation pattern. Similar results are found for the SST tendency–THF correlation, which shows the anti-symmetric pattern expected from theory.
HadGEM3-GC31-MM and MPI-ESM1–2-XR in Fig 3C and 3E also reproduce the observed maximum at zero lag for SST-THF correlation (respectively 0.61 and 0.58), but with values lower than the higher resolution observational data (J-OFURO3, 0.79).
Low-resolution model configurations, as expected, show lower correlation values (in modulus) compared to the high-resolution ones. Compared to other models, EC-Earth3P-LR and HadGEM3-GC31-LL in Fig 3A and 3D display different behavior, revealing a pattern that is typical of the atmosphere-driven regime (see Fig 4).
In Fig 4, we show the same analysis performed over an open ocean region, centered over the longitude-latitude box centered at 16.0°W, 20.0°S. This location is shown as the red square in Fig 4F. Here, the ocean is not affected by strong mesoscale activity, and therefore, the model of Hasselmann [20] is expected to hold. The atmosphere has a dominant effect on ocean temperature variability shown in the lead-lag correlation plots in Fig 4. The analysis shows a relatively small impact of model resolution on SST-THF and SST tendency-THF correlation. This weak sensitivity is expected, since in the analyzed models, the atmospheric component (relevant for the representation of the atmosphere-driven regime) adequately resolves the internal variability in both low and high-resolution configurations.
To quantify the degree of realism of the modeled lead-lag correlation patterns over the AC region (the area that is most sensitive to model resolution), we use the symmetry index (SI hereafter) as defined in Bellucci et al. [27]. We calculated the SI on the reference box centered at 15.0E, 41.5S (as in Fig 3). The SI index quantifies the discrepancy between the modeled (or observed) correlation patterns and a perfectly symmetric pattern, as expected from theoretical results based on stochastic energy balance model results [25]. The SI is obtained as the absolute value of the difference of the SST-THF correlation at lag + 1 and −1, normalized by the correlation at lag 0, as follows:
where indicates the SST-THF correlation at lag i (in months). Based on the definition above, for a perfectly symmetric shape (as expected from theory), SI is expected to be 0. Thus, small (large) SI values indicate models with a realistic (unrealistic) representation of lead-lag correlations in the AC region. Results are reported in Fig 5. In every model, we find that high-resolution configurations have lower SI when compared to their low- resolution counterparts. Exceptions have to be made for EC-Earth3P-LR and HadGEM3-GC31-LL, which, as shown in Fig 3A and 3D, display a highly asymmetrical shape, with a local maximum at lag-1. This causes the SI index to feature very high values (> 50). For the above reasons, the SI values corresponding to EC-Earth3P-LR and HadGEM3-GC31-LL are not shown in Fig 5. Note that if the correlation function is perfectly anti-symmetric, i.e., C_0 = 0, the SI value is ill-defined due to a singularity. The SI values for MPI-ESM1–2 for LR and HR are indistinguishable, confirming the primary role of the oceanic resolution over the atmospheric resolution.
The SI is computed from the lagged SST–THF and SST tendency–THF correlations to quantify the symmetry of the curves around zero lag. Models such as EC-Earth3P and HadGEM3-GC31-LL display strongly anti-symmetric behavior; thus, their SI values fall outside the expected range (above 1) and are not shown. For HadGEM3-GC31, the mid-resolution configuration is labeled as HM. (O refers to the horizontal resolution of the ocean component, and A to that of the atmosphere.).
The smallest values for the SI are associated with the models featuring the highest resolution on the ocean: HadGEM3-GC31-HH, EC-Earth3P-HR, and CMCC-CM2-VHR. These results from the models are consistent with the SI of the observational data (J-OFURO3). Furthermore, HadGEM3- GC31 models clearly show the gradual reduction of SI values, due to the progressive increase of ocean resolution (100 km - 50 km - 10 km) over the atmosphere one (50 km – 50 km – 100 km) for MM - HM - HH configurations. Note that HadGEM is the only model with the setup that allows us to disentangle the relevant impact of model resolution in both ocean and atmosphere configurations. Confronting these values with the resolutions of the models shown in Table 1, it is notable that the greatest differences between low and high-resolution configurations occur when the ocean horizontal resolution is increased, meaning that it is the ocean dynamics to mostly affect the realism of correlations in the AC region.
3.3. Transition scale
In this section, we analyze the dependence of SST-THF and SST tendency-THF lead-lag correlations on the spatial scale, following an approach outlined in Bishop et al. [25], and later applied to the Gulf Stream region in HighResMIP simulations by Bellucci et al. [27]. We apply a circular top‐hat convolution kernel filter to the SST and THF fields for increasing values of the filtering length scale [37]. Then, for each length scale, the lead-lag correlation over the previously used eddy-rich box (centered at 15°E 41°S) is calculated, generating 2D correlation patterns depending on time lag and smoothing length scale.
Figs 6 and 7 show results from this analysis applied to SST-THF and SST- tendency-THF correlations for observations (model simulations).
Low-resolution observational data in the left panels and high-resolution data in the right panels. The region selected is 15°E 41.5°S. Both J_OFURO and OAFLUX are included as well (top panel).
Low-resolution observational data in the left panels and high-resolution data in the right panels. The region selected is 15°E 41.5°S. Both J_OFURO and OAFLUX are included as well (top panel).
The SST-THF correlation patterns (Fig 6) show a transition from a symmetric to an antisymmetric shape when the filter kernel diameter is increased. For SST tendency-THF correlation patterns, instead, a transition from an anti-symmetric to a symmetric functional shape is found for increased kernel size. These transitions reveal the effect of removing the impact of smaller scales (through the filtering process) on the co-variability between surface temperatures and heat fluxes, starting from a typical ocean-driven regime (for no filtering) and gradually revealing the underlying atmosphere-driven regime when the effects of ocean mesoscales are progressively filtered out.
Note that for EC-Earth3P-LR and HadGEM3-GC31-LL models, this process is less evident since, even for a low degree of filtering applied, they fail to reproduce a realistic ocean-driven regime over the AC region.
Following the approach outlined in Bishop et al. [25] we estimate the critical length at which the above transition takes place, as the filtering scale where the SST-THF curve (which is typically positive) intersects the (modulus of) SST tendency-THF correlation curve, at lag 0. The length scale at which the two correlation functions intersect each other identifies the transition scale.
The definition of a transition scale provides an estimate of the spatial dimension needed in a filtering method to cancel out the ocean signal on the air-sea heat exchange processes. By doing so, it is possible to isolate the atmosphere-driven regime signal. In the study area under exam, intersections occur between 2.5° and 5° (see Fig 8).
Ec-Earth3P-LR and HadGEM3-GC31-LL are missing since they do not show a regime transition. The region selected is 15°E 41°S. (O represents ocean model resolution, and A represents Atmospheric model resolution).
As for the symmetry index, the greatest differences between low and high-resolution model configurations are related to the increase of the ocean horizontal resolution. This is clear when comparing low versus high-resolution configurations for model couplets where only the resolution of the atmospheric component is changed (CMCC-CM2, MPI-ESM1–2, and Had-GEM3-GC31-HM and -MM). Their transition length scales are similar. In the first two cases, the same ocean grid is shared by low and high-resolution configurations, highlighting the relatively minor role of the higher atmospheric resolution. The highest transition scale value is found for the HadGEM3-GC3-HH model, which is also the model with the finest ocean grid resolution available among the models included in the study. This model also shows the largest discrepancy from J-OFURO3 estimate, and substantially deviates from all other model-based estimates of the transition scale. A tentative explanation for this peculiar behavior may be provided by invoking the impact of the ocean-atmosphere resolution ratio on the representation of air-sea feedback. In our multimodel ensemble, the HadGEM-GC31-HH model is the one featuring the widest discrepancy between atmospheric (50 km) and ocean (10 km) grid resolution. According to several authors [26,27,38] when the difference in resolution between the two grids exceeds a certain threshold, the realism of the modelled air-sea interaction may show little or no improvements, or even undergo an overall deterioration. The latter case particularly applies to the HadGEM-GC31 model hierarchy (which includes the HH configuration). Specifically, Moreton et al. [38], using three different configurations of this model, found evidence that increasing the oceanic resolution at a constant atmospheric resolution can deteriorate the representation of air-sea interactions. According to these authors, the regridding of SST from the ocean to the atmosphere grid causes an underestimation of air-sea feedback by 20%–80%, and this bias increases with the atmosphere-to-ocean grid resolution ratio. This may explain the apparent “outlier” behavior of HadGEM-GC31-HH in the representation of the transition scale.
Bishop et al. [25] show that the transition between the two regimes in the Agulhas region occurs at a filtering scale of about 2°. In contrast, our results indicate that the regime shift occurs at a generally higher scale—ranging from 2° to 5°—with most models exhibiting a transition at around 3°. J-OFURO and OAFLUX datasets, on the other hand, provide a 3.4° - 6.0° range, yielding a measure of the observational uncertainty, which is grossly consistent with the uncertainty displayed by the HighResMIP models inspected in this study.
4. Summary and conclusions
This study investigates how the horizontal resolutions in coupled climate-model affect the representation of air-sea interactions in the AC region. The research builds upon the methodology outlined in Bishop et al. [25] and Bellucci et al. [27], who examined the scale-dependent nature of air-sea interactions in mid-latitude regions.
Scale-Dependent Air-Sea Interaction. Air-sea interactions are scale-dependent in the Agulhas region. At smaller spatial scales, oceanic processes, such as mesoscale eddies, play a significant role in modulating sea surface temperature (SST) variability. Conversely, at larger spatial scales, atmospheric processes become more dominant. This aligns with the conclusions of Bishop et al. [25], who highlighted the transition from ocean-driven to atmosphere-driven SST variability at scales less than 500 km.
Impact of model resolution. By comparing our results for the AC with those from Bellucci et al. [27] for the Gulf Stream (GS), a consistent picture emerges: the impact of horizontal resolution on modelled air–sea interactions is qualitatively similar across both current systems. Specifically, increasing ocean resolution—from eddy-parametrized (100 km) to eddy-permitting (25 km) and beyond—markedly improves the realism of simulated ocean–atmosphere co-variability. As in Bellucci et al. [27], our results indicate that ocean model resolution plays a critical role, while the contribution of increased atmospheric resolution appears secondary in shaping the spatial structure and temporal coherence between sea surface temperature and turbulent heat fluxes.
Coupled model performance: Coupled models that incorporate both high-resolution oceanic and atmospheric components are better equipped to capture the complex dynamics of air-sea interactions in the Agulhas region. These models can distinguish between ocean-driven and atmosphere-driven regimes, providing a more nuanced understanding of the processes at play.
In conclusion, taken together with the companion analysis on the GS presented in Bellucci et al. [27], our findings support a robust and generalizable pattern: high-resolution ocean models are essential for realistically simulating air–sea coupling in eddy-rich current systems such as the GS and AC.
Supporting information
S1 Text. Absolute Dynamic Topography from AVISO observations and sea surface height (zos) from EC-Earth3P-LR and HadGEM3-GC31 (LL, MM). Cliatology, variance, and de-trended variance are compared to evaluate the representation of mesoscale eddy activity in the models.
https://doi.org/10.1371/journal.pclm.0000680.s001
(DOCX)
Acknowledgments
This work is funded by European Copernicus Marine Service Contract GLO_RAN_Lot6, Evaluation of Ocean reanalysis in the tropical Ocean. We also thank the anonymous reviewers for their constructive comments, which helped improve the quality of the manuscript and the clarity of our work.
References
- 1. Beal LM, Chereskin TK, Lenn YD, Elipot S. The Sources and Mixing Characteristics of the Agulhas Current. Journal of Physical Oceanography. 2006;36(11):2060–74.
- 2.
Lutjeharms JRE. The Agulhas Current. Berlin, Heidelberg: Springer; 2006. doi: https://doi.org/10.1007/3-540-37212-1
- 3. Lutjeharms JRE, Ansorge IJ. The Agulhas Return Current. Journal of Marine Systems. 2001;30(1–2):115–38. doi: https://doi.org/10.1016/s0924-7963(01)00041-0
- 4. Lutjeharms JRE, Van Ballegooyen RC. The Retroflection of the Agulhas Current. J Phys Oceanogr. 1988;18(11):1570–83.
- 5. Lutjeharms JRE, Valentine HR. Evidence for persistent Agulhas rings southwest of Cape Town, South Africa. S Afr J Sci. 1988;84:781–3.
- 6. Bryden HL, Beal LM, Duncan LM. Structure and Transport of the Agulhas Current and Its Temporal Variability. J Oceanogr. 2005;61(3):479–92.
- 7.
Lutjeharms JRE. The Role of Mesoscale Turbulence in the Agulhas Current System. Elsevier Oceanography Series. Elsevier. 1989. p. 357–72. doi: https://doi.org/10.1016/s0422-9894(08)70196-x
- 8. Sandalyuk NV, Belonenko TV. Three-Dimensional Structure of the mesoscale eddies in\vspace {.3pc} the Agulhas Current region from hydrological and altimetry data. Russ J Earth Sci. 2021;21(4):1–19. doi: https://doi.org/10.2205/2021es000764
- 9. Speich S, Lutjeharms JRE, Penven P, Blanke B. Role of bathymetry in Agulhas Current configuration and behaviour. Geophysical Research Letters. 2006;33(23).
- 10. van Leeuwen PJ, de Ruijter WPM, Lutjeharms JRE. Natal pulses and the formation of Agulhas rings. J Geophys Res. 2000;105(C3):6425–36.
- 11. Tsugawa M, Hasumi H. Generation and Growth Mechanism of the Natal Pulse. Journal of Physical Oceanography. 2010;40(7):1597–612.
- 12. Durgadoo JV, Loveday BR, Reason CJC, Penven P, Biastoch A. Agulhas Leakage Predominantly Responds to the Southern Hemisphere Westerlies. Journal of Physical Oceanography. 2013;43(10):2113–31.
- 13. Schubert R, Gula J, Biastoch A. Submesoscale flows impact Agulhas leakage in ocean simulations. Commun Earth Environ. 2021;2(1).
- 14. De Ruijter WPM. The Agulhas Current. EoS Transactions. 2008;89(10):98–98.
- 15. Beal LM, De Ruijter WPM, Biastoch A, Zahn R, SCOR/WCRP/IAPSO Working Group 136. On the role of the Agulhas system in ocean circulation and climate. Nature. 2011;472(7344):429–36. pmid:21525925
- 16. Biastoch A, Böning CW, Schwarzkopf FU, Lutjeharms JRE. Increase in Agulhas leakage due to poleward shift of Southern Hemisphere westerlies. Nature. 2009;462(7272):495–8. pmid:19940923
- 17. Zahn R, Lutjeharms J, Biastoch A, Hall I, Knorr G, Park W, et al. Investigating the Global Impacts of the Agulhas Current. EoS Transactions. 2010;91(12):109–10.
- 18. Reason CJC, Jagadheesha D. Relationships between South Atlantic SST Variability and Atmospheric Circulation over the South African Region during Austral Winter. Journal of Climate. 2005;18(16):3339–55.
- 19.
Biastoch A, et al. The Agulhas Current System as an important driver for oceanic and terrestrial climate. In: von Maltitz GP, et al., editors. Sustainability of Southern African Ecosystems under Global Change. Ecological Studies, vol 248. Cham: Springer; 2024. doi: https://doi.org/10.1007/978-3-031-10948-5_8
- 20. Hasselmann K. Stochastic climate models: Part I. Theory. Tellus A: Dynamic Meteorology and Oceanography. 1976;28(6):473. doi: https://doi.org/10.3402/tellusa.v28i6.11316
- 21. Kirtman BP, Bitz C, Bryan F, Collins W, Dennis J, Hearn N, et al. Impact of ocean model resolution on CCSM climate simulations. Clim Dyn. 2012;39(6):1303–28.
- 22. Frankignoul C, Hasselmann K. Stochastic climate models, Part II Application to sea-surface temperature anomalies and thermocline variability. Tellus A: Dynamic Meteorology and Oceanography. 1977;29(4):289.
- 23. Barsugli JJ, Battisti DS. The Basic Effects of Atmosphere–Ocean Thermal Coupling on Midlatitude Variability*. J Atmos Sci. 1998;55(4):477–93.
- 24. Wu R, Kirtman BP, Pegion K. Local Air–Sea Relationship in Observations and Model Simulations. Journal of Climate. 2006;19(19):4914–32.
- 25. Bishop SP, Small RJ, Bryan FO, Tomas RA. Scale Dependence of Midlatitude Air–Sea Interaction. Journal of Climate. 2017;30(20):8207–21.
- 26. Tsartsali EE, Haarsma RJ, Athanasiadis PJ, Bellucci A, de Vries H, Drijfhout S, et al. Impact of resolution on the atmosphere–ocean coupling along the Gulf Stream in global high resolution models. Clim Dyn. 2022;58(11–12):3317–33.
- 27. Bellucci A, Athanasiadis PJ, Scoccimarro E, Ruggieri P, Gualdi S, Fedele G, et al. Air-Sea interaction over the Gulf Stream in an ensemble of HighResMIP present climate simulations. Clim Dyn. 2021;56(7–8):2093–111.
- 28. Richardson PL. Gulf Stream Trajectories Measured with Free-Drifting Buoys. J Phys Oceanogr. 1981;11(7):999–1010.
- 29. Haarsma RJ, Roberts MJ, Vidale PL, Senior CA, Bellucci A, Bao Q, et al. High Resolution Model Intercomparison Project (HighResMIP v1.0) for CMIP6. Geosci Model Dev. 2016;9:4185–208.
- 30. Cherchi A, Fogli PG, Lovato T, Peano D, Iovino D, Gualdi S, et al. Global Mean Climate and Main Patterns of Variability in the CMCC‐CM2 Coupled Model. J Adv Model Earth Syst. 2019;11(1):185–209.
- 31. Döscher R, Acosta M, Alessandri A, Anthoni P, Arsouze T, Bergman T, et al. The EC-Earth3 Earth system model for the Coupled Model Intercomparison Project 6. Geosci Model Dev. 2022;15(7):2973–3020.
- 32. Roberts MJ. MOHC HadGEM3-GC31-LL model output prepared for CMIP6 HighResMIP. Earth Syst Grid Fed [dataset]. 2017.
- 33. Roberts MJ. MOHC HadGEM3-GC31-HH model output prepared for CMIP6 HighResMIP. Earth Syst Grid Fed [dataset]. 2018.
- 34. Gutjahr O, Putrasahan D, Lohmann K, Jungclaus JH, von Storch J-S, Brüggemann N, et al. Max Planck Institute Earth System Model (MPI-ESM1.2) for the High-Resolution Model Intercomparison Project (HighResMIP). Geosci Model Dev. 2019;12(7):3241–81.
- 35.
Yu L, Jin X, Weller RA. Multidecade global flux datasets from the objectively analyzed air–sea fluxes (OAFlux) project: Latent and sensible heat fluxes, ocean evaporation, and related surface meteorological variables. OAFlux Project Tech Rep. Woods Hole Oceanographic Institution; 2008.
- 36. Tomita H, Hihara T, Kako S, Kubota M, Kutsuwada K. An introduction to J-OFURO3, a third-generation Japanese ocean flux data set using remote-sensing observations. J Oceanogr. 2018;75(2):171–94.
- 37. Aluie H, Hecht M, Vallis GK. Mapping the Energy Cascade in the North Atlantic Ocean: The Coarse-Graining Approach. Journal of Physical Oceanography. 2018;48(2):225–44.
- 38. Moreton S, Ferreira D, Roberts M, Hewitt H. Air‐Sea Turbulent Heat Flux Feedback Over Mesoscale Eddies. Geophysical Research Letters. 2021;48(20).