Prior to further analysis, all time series (SST, SSS, SLA, and the zonal and meridional wind components U and V) were linearly detrended and filtered to remove high-frequency variability. Detrending was carried out to minimize the impact of long-term changes, particularly warming and associated sea level rise. A second-order Butterworth low-pass filter with a cutoff period of 20 d was applied. Butterworth filters are characterized by a maximally flat frequency response in the passband, providing a smooth transition between retained and attenuated frequencies without spurious oscillations (Butterworth, 1930; Thomson and Emery, 2014). The 20 d cutoff was selected to suppress synoptic-scale variability while retaining intraseasonal to interannual signals. This choice reflects the effective temporal resolution of satellite SLA and SSS products, which do not reliably resolve variability at periods shorter than ∼ 20 d (Ferrari et al., 2017; Ballarotta et al., 2019). A second-order filter was chosen as a compromise between frequency selectivity and temporal fidelity, minimizing edge effects while efficiently attenuating high-frequency energy.

Seasonality was removed using a fourth-order Butterworth band-stop filter with cutoff periods between 150 and 400 d. Higher-order filters provide a sharper separation of frequency bands but at the cost of stronger edge effects and increased distortion near the cutoff frequencies. The use of a fourth-order filter allowed for an efficient suppression of the seasonal band. Because such filters may introduce transient effects near the beginning and end of the records, particular care was taken in the interpretation of low-frequency variability and in subsequent spectral analyses. Alternative approaches to remove the seasonal cycle were also explored, including the subtraction of a daily climatology and harmonic analysis. While these methods effectively remove the mean seasonal signal, they retain residual variance within the seasonal frequency band. In contrast, the band-stop filtering approach ensures the removal of variability across the entire seasonal range. To assess the sensitivity to preprocessing, we tested alternative filtering approaches, including low-pass cutoffs of 10 and 30 d, a 20 d Lanczos low-pass filter, a second-order band-stop filter, and alternative seasonal cycle removal methods based on a smoothed daily climatology or annual and semi-annual harmonics. The main spatial patterns and relationships discussed in this study remained robust across these different preprocessing choices (Figs. S1–S5 in the Supplement). Hereafter, the resulting filtered time series are referred to as SSTa, SLAa, SSSa, Ua, and Va, corresponding to anomalies of SST, SLA, SSS, zonal wind, and meridional wind, respectively.

To analyse the space-time variability of SSTa, SSSa, SLAa, and wind anomalies (Ua and Va) over the continental shelves around southern South America, we use Empirical Orthogonal Functions (EOF). EOF analysis decomposes the data into a set of orthogonal spatial patterns (EOFs) and associated temporal coefficients, called Principal Components (PCs), that maximize the explained variance of the dataset. The PCs describe the temporal evolution of each mode and are mutually uncorrelated at zero lag. These properties do not imply statistical independence or a one-to-one correspondence between EOF modes and independent dynamical processes (Monahan et al., 2009). For each variable, we computed the first three EOF modes of variability. The analysis focuses on the first two modes, which represent the dominant regional patterns of variability. The third mode explains only a small fraction of the total variance and was therefore not considered further. Additionally, we analysed the PC time series of the retained modes to investigate their temporal variability and explore their covariability.

To analyse the frequency-domain variability of the leading modes of each variable (PC1 and PC2 derived from the EOF analysis), power spectral density estimates were computed using Welch's method. This approach segments the time series into overlapping sections, applies a windowing function to reduce spectral leakage, and then averages the periodograms of each segment to obtain a more stable estimate of the spectrum (Welch, 1967; Thomson and Emery, 2014). A Hann window was applied to each segment. The segment length was set to one-fourth of the total time series length nperseg=N4, with a 50 % overlap between segments. This configuration provides a compromise between frequency resolution and statistical stability and is frequently used for identifying broad spectral bands of variability rather than narrow spectral peaks. For visualization purposes, spectra are presented in a variance-preserving form. Confidence intervals at the 95 % level were obtained using the chi-square distribution with effective degrees of freedom computed following Thomson and Emery (2014), as defined in Eq. (1). 1ν=2P,P=83⋅Nnperseg2 where ν is the number of effective degrees of freedom used to define the chi-square confidence intervals, P is the effective number of periodogram estimates averaged, N is the number of data points in the time series, and nperseg is the length of each segment (nperseg=N/4). The factor 8/3 accounts for the variance reduction associated with the Hann taper and the 50 % overlap between consecutive segments (Thomson and Emery, 2014).

Linear relationships between principal components and between wind and oceanic variables were quantified using the Pearson correlation coefficient. Because temporal filtering reduces the number of statistically independent observations, an effective sample size (Neff) was estimated for each pair of time series by accounting for their autocorrelation structure, following the framework discussed by Bretherton et al. (1999). The effective sample size was computed using Eq. (2). 2Neff=N1+2∑k=1Krxkryk where N is the number of overlapping observations, and rxk and ryk are the autocorrelation functions of the time series at lag k. The summation, whose upper limit is K, was truncated at the first lag for which either autocorrelation became non-positive to avoid overestimating persistence.

Statistical significance was then evaluated using a two-tailed Student's t-test with Neff-2 degrees of freedom (Wilks, 2011). Critical correlation thresholds were derived from the corresponding t distribution, and correlations exceeding this threshold were considered statistically significant. Given the relatively large effective sample sizes, the interpretation focuses primarily on moderate-to-strong correlations, which are more likely to reflect dynamically meaningful relationships.

Composite maps were constructed to examine the spatial expression of surface ocean anomalies during positive and negative phases of dominant wind modes. Composites were defined based on periods exceeding one standard deviation of the principal component of zonal and meridional wind anomalies for at least five consecutive days.

The mean SST field around southern South America exhibits a broad meridional gradient, with temperatures below 5.5 °C in the southernmost part of the domain and exceeding 14 °C around 40° S (Fig. 2a). In the Atlantic, north of ∼ 50° S, the mean SST field also exhibits a zonal gradient, characterized by differences between warmer coastal and colder offshore waters. In contrast, along the Pacific continental shelf north of 46° S, SST is lower on the shelf than in the open ocean, whereas the opposite occurs south of 46° S, where shelf waters are warmer than offshore waters. This pattern may be linked to the meridional wind divergence zone (see Fig. 3b). North of 45° S, the prevailing southerly winds induce coastal upwelling and northward flow over the shelf, favouring cooler waters near the coast. South of 45° S, the northerly winds induce downwelling and southward advection of warmer waters over the shelf.

On the Atlantic PCS, zonal SST gradients are much sharper in the northern sector compared to the south. SST is relatively high over the shelf but decreases sharply beyond the continental slope, where cold subantarctic waters are advected northward by the Malvinas Current (SST < 9.5 °C, Fig. 2a and c). The cold core of the Malvinas Current is located between the 200 and 2000 m isobaths, with SST increasing farther offshore. A secondary tongue of relatively cold waters is apparent over the outer continental shelf south of 47.5° S. An additional tongue of cold water extends northward to around 45° S near the Atlantic coast, likely associated with the Patagonian Current (Brandhorst and Castello, 1971; Palma and Matano, 2012). The spatial pattern of SST standard deviation (SD) in the Atlantic (Fig. 2d) displays a pronounced meridional gradient reaching values up to 4 °C, while it ranges between 1 °C and 2 °C over the Pacific shelf. It is important to note that these SD fields were computed from time series that had been previously low-pass filtered to remove high-frequency variability. Therefore, the reported values primarily reflect variability at intraseasonal to longer timescales. After removing the seasonal variability, the SD of SSTa is much smaller, typically around 0.5 °C over most of the domain, except over the northern Atlantic shelf near the continental slope where it reaches ∼ 1 °C (Fig. 2g). These patterns reflect the greater amplitude of seasonal variability, particularly in the northern part of the Atlantic domain (see Rivas, 2010).

Satellite-derived SSS mean values range from 32.4 to 34.7. In general, lower salinities are observed near the coast, increasing offshore (Fig. 2b). However, other notable spatial patterns emerge in the mean field. In the Pacific shelf, the lowest salinities are found along a coastal band with a minimum close to 47.5° S, consistent with previous observational studies (Dávila et al., 2002; Brun et al., 2020; Saldías et al., 2024). While satellites provide unprecedented temporal coverage, they do not provide reliable data within 40–100 km from land or within the fjords (Vinogradova et al., 2019; Jarugula et al., 2025), where salinity is known to be lower than 32 (e.g. Castillo et al., 2012; Schneider et al., 2014; Brun et al., 2020) due to significant continental freshwater input. Data availability in the Pacific portion of the PCS is limited to the Cape Horn region, where salinities are approximately 34. In the Atlantic shelf, the lowest salinity values (< 33) are observed around the eastern mouth of the Strait of Magellan, in agreement with the in-situ salinity distribution (Brun et al., 2020). However, the satellite-derived SSS does not provide data within the Strait of Magellan. Despite the limitations of satellite-derived SSS, away from the coastal band the observed pattern over the region aligns well with in-situ data from Brun et al. (2020).

The SD of SSS (Fig. 2e) exhibits increased variability near the coast, particularly along the Pacific shelf break, with values exceeding 0.4. The enhanced nearshore variability on both sides of the continent is likely associated with the combined effects of the limited resolution of satellite-borne radiometers in coastal areas and the influence of freshwater input from continental runoff. In the Pacific, significant seasonal cycles of freshwater input are associated with the melting of continental ice and substantial regional precipitation, particularly in the southeastern portion of the shelf (Dávila et al., 2002; Saldías et al., 2024). When the seasonal cycle is removed (Fig. 2h), the spatial pattern of enhanced nearshore variability persists, although overall SD values decrease. Notably, an SD maximum persists along the Pacific slope between around 47, and 51° S, where SSS SDs still exceed 0.25. This persistent variability suggests that salinity variability in this region is not only driven by seasonal forcing.

The ADT mean is shown in Fig. 2c, along with the associated mean geostrophic velocities. ADT is generally higher over the Patagonian shelf than in adjacent offshore regions. A minimum in ADT is noticeable in the Antarctic Circumpolar Current in the northern Drake Passage and along the Malvinas Current. Consequently, in the Atlantic, the largest geostrophic velocities, which reach ∼ 50 cm s⁻¹, are associated with the Antarctic Circumpolar Current south of Cape Horn and the Malvinas Current near 43° S (Fig. 2c). In the Pacific sector, the Cape Horn Current exhibits moderate geostrophic velocities (∼ 30 cm s⁻¹) flowing southeastward, around Cape Horn, and toward the Atlantic.

The SD of SLA (Fig. 2f) shows a clear meridional gradient across the Atlantic shelf, with higher variability (> 6 cm) in the mid- and inner shelf north of 47.5° S. In the Pacific, no dominant pattern emerges across the entire shelf; however, somewhat higher variability is observed over the Cape Horn shelf. When the seasonal signal is removed (Fig. 2i), SD values generally decrease, as expected, yet the overall spatial structure is similar to the SD of the unfiltered data. The persistent high values over the Cape Horn shelf and along the coast of the Atlantic shelf indicate that non-seasonal processes also contribute significantly to SLAa variability over the PCS. The possible impact of non-seasonal wind variability in modulating the SLA is explored in Sect. 4.3.

The mean winds in the region are predominantly westerly, with lower magnitudes (∼ 5 m s⁻¹) in the northern part of the domain and exceeding 8 m s⁻¹ in the core of the westerlies south of 52° S (Fig. 3a and b).

Mean wind magnitudes are generally higher over the Pacific compared to the Atlantic, displaying a broad maximum along the Pacific coast south of 50° S and the northern Drake Passage (∼ 7 m s⁻¹). Throughout most of the domain, the mean meridional wind is northerly and weaker than the zonal wind, particularly over the Atlantic (>-1.5 m s⁻¹), while along the Pacific coast south of 50° S it reaches ∼ -4 m s⁻¹ (Fig. 3b).

The spatial pattern of the zonal wind standard deviation (Fig. 3c) shows maximum values over the Pacific shelf south of 52° S, reaching up to 3.5 m s⁻¹ over the southern Pacific PCS. Over the Atlantic shelf, variability is lower, ranging from 1.5 to 2.8 m s⁻¹, with local maxima around 53° S. The SD of the meridional wind component (Fig. 3d) exceeds 2.5 m s⁻¹ along most of the Pacific shelf, while values over the Atlantic shelf range between 1 and 2 m s⁻¹. Although the meridional wind component is generally weak compared with the zonal component, the standard deviations are similar (Fig. 3c and d). When the seasonal cycle is removed (Figs. 3e and f), SDs decrease, but some spatial features persist. The zonal wind variability remains highest south of 52° S over the Pacific, where the strongest zonal winds are observed. Over the Atlantic shelf, it peaks at ∼ 2.5 m s⁻¹ south of 52.5° S and reaches ∼ 2 m s⁻¹ farther north. The SD of the meridional wind is fairly uniform (∼ 2–2.5 m s⁻¹) along the Pacific shelf, especially near the coast, suggesting the influence of the Andes on the local wind variability. Over the Atlantic shelf, the meridional wind variability is comparable to that of the zonal wind in most of the domain, except south of 52° S, where zonal variations are slightly stronger.

These standard deviations indicate that, over most of the domain, variations in the zonal wind component are associated with westerly winds of varying intensity. In contrast, meridional wind deviations often exceed mean values, particularly over both continental shelves, implying frequent and significant reversals in wind direction. These frequent meridional wind reversals may effectively modulate the intensity of the along-shelf circulation, with northerly winds expected to reinforce the southward mean flow over the Pacific shelf and weaken the northeastward mean flow over the Atlantic shelf (Fig. 2c), and vice versa.

We analysed the variability of SSTa, SSSa, and SLAa using empirical orthogonal function (EOF) analysis, focusing on the continental shelf and upper slope (depths shallower than 2800 m). The EOFs help identify the statistically dominant spatial patterns and their time variability. For clarity, hereafter we refer to the spatial patterns of each variable as EOF modes (e.g., EOF1-SSTa, EOF2-SSTa), while the associated time series are denoted as principal components (PCs; e.g., PC1-SSTa, PC2-SSTa).

EOF1-SSTa explains ∼ 54 % of the variance, displaying a monopole pattern across the PCS and adjacent slopes, with the largest variations (∼ 0.5 °C) observed near the outer shelf in the Atlantic, particularly north of 47.5° S (Fig. 4a). EOF1-SSTa also shows enhanced low-frequency variance at periods of approximately 1–4 years (Fig. 4c) and a strong intraseasonal signal between 60–150 d. Although EOF1-SSTa is significantly correlated with SSTa over most of the PCS (> 0.6), it primarily reflects variability on the Atlantic continental shelf, where correlations exceed 0.8 (see Supplement Fig. S6). EOF2-SSTa explains ∼ 11 % of the variance and displays a dipole pattern centred around 47.5° S, dividing the Atlantic PCS into two distinct regions with opposite SSTa (Fig. 4b). In contrast, the SSTa associated with EOF2 has the same sign along the western coast of South America, matching the southern part of the Atlantic dipole.

EOF1-SSSa explains ∼ 14 % of the total variance and exhibits a monopole over the Atlantic PCS and a dipole along the Pacific shelf break, with anomalies opposite to the Atlantic south of ∼ 50° S and of the same sign to the north (Fig. 4d). The associated temporal variability spans multiple timescales, with an apparent low-frequency contribution at periods longer than approximately 2 years as well as variability near ∼ 100 and ∼ 60 d. EOF2-SSSa accounts for ∼ 7 % of the variance and is primarily linked to intraseasonal variability, exhibiting a clear spectral peak near ∼ 100 d and meridional dipoles centred at approximately 50° S in the Atlantic and 47.5° S in the Pacific (Fig. 4e). Because SSSa variability is spatially heterogeneous and distributed across multiple modes, no single EOF captures a large fraction of the local variance; consequently, correlations between PC1-SSSa, PC2-SSSa, and the SSSa field are not statistically significant over most of the domain (see Fig. S6c and d). PC2-SSSa exhibits statistically significant correlations only in the southern Pacific PCS, along the coast, where the largest EOF2 amplitudes (exceeding 0.1) and enhanced intraseasonal variance are observed (Fig. 4e–f). Thus, despite explaining a modest portion of the total variance, EOF2-SSSa captures the most energetic intraseasonal salinity fluctuations in the southern Pacific. EOF1-SLAa explains ∼ 28 % of the total variance and displays a monopole pattern over the entire domain, with the largest variations over the Atlantic and Cape Horn shelves and somewhat smaller variations over the Pacific shelf and the southern Atlantic continental slope (Fig. 4g). Most of the variance associated with this mode lies within the 30–90 d period range (Fig. 4i). The geostrophic velocity anomalies associated with EOF1-SLAa are small, reaching ∼ 1.8 cm s⁻¹ in the Malvinas Current region and up to 2.6–3.3 cm s⁻¹ in the Cape Horn sector. Compared to the MDT, these values represent ∼ 3 % of the climatological velocities in the Malvinas region (61 cm s⁻¹), ∼ 9 % over the Cape Horn shelf (27.7 cm s⁻¹), and up to ∼ 22 % in localized regions of maximum gradients. EOF2-SLAa, explaining ∼ 10 % of the variance, displays a dipole centred north of 47.5° S on the Atlantic shelf and variability along the southern Pacific shelf of the same sign as the southern Atlantic region (Fig. 4h). The related geostrophic anomalies can reach ∼ 1.5 cm s⁻¹ in the northern Atlantic PCS. When compared to the MDT velocities in this region (2.6 cm s⁻¹), this accounts for ∼ 60 % of the mean, underscoring the regional significance of this mode despite its relatively low explained variance.

To analyse the wind variability, we conducted EOF analyses for the anomalies of the zonal and meridional components (Ua and Va). As with the other variables, we removed the mean, trend, high-frequency variability and seasonal cycle before the EOF analyses. The first two modes of both wind components capture approximately 77 % of the total variance, with the first mode explaining more than half of the variance in each case. Correlation maps between the PCs and the corresponding anomaly fields further show the spatial patterns associated with these modes (Fig. S7).

EOF1-Ua explains ∼ 50 % of the variance and shows a monopole with maximum variance centred around 54° S in the southern part of the domain (Fig. 5a). This centre extends zonally, with the highest intensities over the ocean near the continent, particularly in the southeast Pacific. EOF2-Ua, accounting for ∼ 27 % of the variance, displays a meridional dipole centred around 52.5° S, with greater variability at the northern and southern edges of the domain, with somewhat weaker anomalies than EOF1-Ua (Fig. 5b). The variance associated with both EOF modes is mostly centred around intraseasonal time scales, though EOF1-Ua presents more energy at 30 d, and EOF2-Ua around 60 d (Fig. 5c).

EOF1-Va captures ∼ 62 % of the total variance, displaying a monopole with maxima in the southern Pacific region near the coast, where anomalies reach up to 1.5 m s⁻¹, and another centre near the Malvinas Islands (Fig. 5d). The variance associated with EOF1-Va is mostly intraseasonal, with the largest variance in the 30–40 d range (Fig. 5f). EOF2-Va explains ∼ 15 % of the variance, showing a meridional dipole centred around 50° S and displaying larger variations over the Pacific region (Fig. 5e).

The wind spectra show their largest peaks at periods of approximately 20–40 d, with additional but somewhat lower variability between 40 and 100 d (Fig. 5c and f). These periods overlap with the broad 30–90 d band of PC1-SLAa (Fig. 4i), associated with the monopole pattern strongest over the Atlantic and Cape Horn shelves (Fig. 4g). PC1-SSTa, which primarily represents variability over the Atlantic shelf, has enhanced variance at 60–150 d, overlapping the longer intraseasonal periods present in the wind spectra (Fig. 4a and c). The SSSa spectra show a prominent peak near 100 d, particularly in PC2-SSSa, and little variance at periods shorter than 40 d (Fig. 4f). The temporal resolution of the SSS product and its implications for this comparison are discussed in Sect. 5.2. At interannual periods, the wind PCs show comparatively little spectral energy, whereas PC1-SSTa retains low-frequency variance comparable to that observed at intraseasonal frequencies (Figs. 4c, 5c, and f). The strongest spectral overlap between the wind and ocean modes therefore occurs at intraseasonal periods. The following section examines their temporal relationships through correlations between the PCs.

Tables 1 and 2 summarize the statistical relationships among principal components of variability in meridional and zonal wind components, temperature, salinity, and sea level. Temporal autocorrelation was accounted for when assessing correlation significance by estimating an effective sample size for each pair of time series (Sect. 3.3.3). The statistical significance of the correlations was assessed, and those significant at the 95 % confidence level are highlighted in bold (Tables 1 and 2).

The principal component of EOF1-Ua (PC1-Ua), which primarily represents the strengthening or weakening of the westerlies, is significantly correlated with PC1-SLAa (r=0.41), suggesting that stronger westerlies are associated with higher sea level throughout the domain. In contrast, PC1-Ua is negatively correlated with PC2-SLAa (r=-0.25), suggesting that stronger westerlies are associated with increased SLAa over the northern Atlantic and reduced SLAa elsewhere. However, note that the anomalies linked to EOF2-SLAa are weaker, and the correlation between PC2-SLAa and the zonal wind variations is lower. PC1-Ua also shows a weaker but significant negative correlation with PC2-SSSa (r=-0.28), pointing to a possible connection between intensified westerlies and increased coastal salinity along the southern Pacific margin.

PC1-Va, which explains 62 % of the variance in meridional winds, is weakly and negatively correlated with PC1-SSTa (r=-0.24), indicating that enhanced southerly winds are associated with surface cooling throughout the region – particularly near the shelf break over the northern sector of the Atlantic shelf – while northerly winds are associated with surface warming. PC1-Va is further linked to salinity variability, showing positive correlations with both PC1-SSSa and PC2-SSSa (r=0.20 and r=0.36, respectively), and exhibits a strong negative correlation with PC2-SLAa (r=-0.74). These relationships support the hypothesis that meridional wind anomalies and the associated cross-shelf Ekman transport and induced pressure gradients modulate the along-shelf geostrophic flow and the salinity distribution. However, other patterns are not consistent. For example, the decrease in SLA over the Atlantic shelf south of 50° S is opposite to the expected response to southerly winds (see Figs. 4h and 5d), suggesting that other forcing mechanisms may control these changes.

To further investigate the possible relationship between SSTa, SSSa, and SLAa variability in Table 2, we present the correlations between their dominant principal components. PC1-SSTa is positively correlated with PC1-SLAa (r=0.28), which is consistent with a thermosteric effect. However, it also shows a somewhat higher correlation with PC2-SLAa (r=0.34), which is associated with a dipole pattern of SLAa (Fig. 4h). This is not consistent with a thermosteric response, but rather suggests that northward advection of cooler waters could drive temperature changes. In this context, SLAa gradients can induce changes in the geostrophic flow, particularly over the northern Atlantic shelf near the continental slope. Additionally, the second mode of SSSa shows weak but statistically significant negative correlations with both SLAa modes (r=-0.16 and r=-0.15). This mode is only relevant over the southern Pacific, where its spatial pattern significantly correlates with the observed salinity anomalies (see Fig. S6d). In this region, the inverse relationship with SLAa contrasts with the expected halosteric effect, suggesting that salinity-driven contributions to SLAa variability are minimal, despite the presence of marked salinity changes.

Altogether, these statistical relationships highlight a dynamical link between wind variability – particularly of the meridional component – and surface ocean variability over the PCS. In the following section, we use composite anomaly maps to further investigate the spatial patterns associated with the dominant modes of wind variability and to gain a deeper insight into how it modulates the circulation and water mass distribution over the PCS.

To assess the influence of wind variability on the surface properties over the PCS, we constructed composites of SSTa, SSSa, and SLAa for the positive and negative phases of the leading empirical orthogonal function modes of zonal and meridional wind variability. Phases were defined by periods during which PC1-Ua or PC1-Va exceeded ±1 standard deviation for at least five consecutive days, yielding approximately 1100 d per phase (∼ 700 d for salinity). Threshold values and annual distributions of events are provided in the supplementary material (Figs. S8 and S9). These composites describe statistical associations consistent with the proposed wind-driven mechanisms but do not establish causality on their own.

EOF1 of meridional wind accounts for ∼ 62 % of the total variance and exhibits a monopolar structure centred over South America (Fig. 5d). This mode captures regional meridional wind shifts, with southerly winds prevailing during the positive phase and northerly winds during the negative phase, affecting both the Pacific and Atlantic shelves.

During the positive phase of meridional winds (southerly winds; Fig. 6a), widespread surface cooling is observed over both shelves, most intensely over the outer Atlantic shelf north of 50° S (∼ -0.2 °C, Fig. 6a). The opposite response occurs during the negative phase (northerly winds, ∼ 0.2 °C, Fig. 6b). While anomalies extend over the entire shelf, they are weaker over the inner shelf, particularly near San Matías Gulf and southern Grande Bay. Notably, cooling during the positive phase tends to be stronger than warming during the negative phase.

Composites of SSSa show the strongest response over the Pacific slope south of 47.5° S. During the positive phase, low-salinity anomalies exceed -0.1 (Fig. 6c). The negative phase exhibits a somewhat weaker inverse pattern (Fig. 6d). On the Atlantic shelf, southerly winds correlate with slightly saltier conditions, and northerly winds with fresher waters, although anomalies are small (< 0.05).

Under southerly winds, SLAa increases markedly over the northern Atlantic shelf, exceeding 3 cm over the mid-shelf (Fig. 6e). This sets up a zonal pressure gradient that drives northeastward geostrophic velocity anomalies of up to 2 cm s⁻¹ in the mid and outer shelf. Around 50° S, SLAa signs reverse, presenting weak and spatially diffuse anomalies over the southern Atlantic shelf, resulting in negligible geostrophic velocity anomalies (Fig. 6e). In the Pacific, SLAa patterns are more pronounced. Under southerly winds, SLAa minima form over the Cape Horn Current region, accompanied by strong horizontal gradients both offshore and nearshore. Maximum geostrophic velocity anomalies reach ∼ 5 cm s⁻¹ along the slope, suggesting a slowdown of the southward mean flow. Nearshore, the anomaly fields display patterns that resemble northward counterflows; however, these may also reflect a reduced southward transport of the Cape Horn Current, rather than a complete reversal. Given the limited accuracy of gridded altimetry maps over the coasts, the near-shore response should be interpreted with caution.

During the negative phase (northerly winds), SLAa displays a reversed pattern, forming a dipole in the Atlantic with reduced northeastward flow over the outer portion of the northern shelf, and elevated sea levels in the Pacific shelf, mostly south of 50° S and near Cape Horn, strengthening the Cape Horn Current (Fig. 6f).

Overall, SSTa, SSSa, and SLAa composites display spatially coherent responses to meridional wind variability. Anomalies consistently reverse sign between phases, and spatial patterns remain robust across different phase definitions. These structures are also supported by pointwise correlations between meridional wind and surface ocean anomalies (Fig. S10d–f). However, the SSSa signals are generally quite small, except in the southern Pacific outer shelf and upper slope regions.

EOF1 of zonal wind explains ∼ 50 % of the total variance and features a monopolar pattern centred over South America, with maximum anomalies near 53° S, especially over the Pacific (Fig. 5a). This mode represents variability in westerly wind strength, with stronger winds during the positive phase. Phase-averaged wind fields (arrows in Fig. 7a and b) show intensified westerlies during the positive phase, with anomalies reaching ∼ 3 m s⁻¹ near 53° S. In the phase-averaged wind fields, winds are generally stronger over the Pacific than over the Atlantic and exhibit a slight northerly component, particularly south of 45° S in the southeast Pacific. During the negative phase, westerlies weaken and are combined with northerly anomalies across most of the domain, with peak zonal anomalies of ∼ -4 m s⁻¹ in the Pacific and weaker than -1.5 m s⁻¹ in the Atlantic (Fig. 7b).

SSTa patterns in the PCS under zonal wind forcing are weaker compared to those associated with meridional winds (Fig. 7a and b). Intensified westerlies (positive phase, Fig. 7a) correspond to weak negative SSTa across most of the region, except in the southern Pacific PCS, where weak positive anomalies (< 1 °C) are present over the shelf south of 50° S. During the negative phase, the sign of the anomalies reverses in the Pacific; however, over the Atlantic the sign reversal is less evident, and the anomalies remain weak (< 1 °C).

SSSa composites display a clear dipole in the Pacific (Fig. 7c and d). During the positive phase, fresher anomalies occur north of ∼ 48° S and saltier anomalies to the south (Fig. 7c). This pattern reverses in the negative phase, with the strongest signals between 50–57° S along the slope. In the Atlantic, slight freshening is observed in the inner shelf during the positive phase and a weak salinity increase in the negative phase, although anomalies remain below 0.03 (Fig. 7d).

SLAa composites exhibit coherent patterns of opposite sign in each shelf. In the Atlantic, a positive SLAa monopole forms during intensified westerlies, peaking over the mid-shelf north of 52.5° S and east of Estados Island (Fig. 7e). Another variability hotspot appears over the Cape Horn shelf, with anomalies exceeding 3 cm. These regions exhibit sea level rise during the positive phase and a drop during the negative phase (Fig. 7e and f). Near the Strait of Magellan and along the coast of Grande Bay, SLAa are minimal and occasionally reverse sign compared to the neighbouring region, though uncertainties are high in this narrow, shallow and macrotidal region, where the accuracy of the altimeter retracking and of the tidal and dynamic atmospheric corrections degrades due to the proximity of the coast (Birol et al., 2025). Over the Pacific shelf north of 53° S, SLAa also reverses but is relatively small (< 1 cm), and its sign opposes that of the Atlantic and Cape Horn: sea level falls under stronger westerlies and rises when they weaken. Strongest SLAa gradients appear along the outer Atlantic shelf near the 200 m isobath (Fig. 7e and f), driving geostrophic current anomalies of ∼ 1 cm s⁻¹. These anomalies intensify the northeastward mean flow during the positive phase and reduce it during the negative phase.