The Equatorial Atlantic is a fascinating region of the Atlantic Ocean that extends on either side of the equator between 5˚N and 5˚S (Figure 1). It is a crucial transition zone connecting the North Atlantic and the South Atlantic, playing a key role in regulating the West African climate [8]. From an oceanographic perspective, this zone is particularly dynamic and complex, with a current system comprising major currents such as the northern and central branches of the South Equatorial Current (nSEC and cSEC), which drive surface waters westwards [16]. Just below this flows the Equatorial Undercurrent (EUC), a veritable underwater river moving in the opposite direction, towards the east. Furthermore, due to the Coriolis force which cancels itself out at the equator but deflects the waters just to the north and south the surface waters diverge. This causes deep, cold and nutrient-rich waters to upwelling, a phenomenon particularly evident in the GoG during the summer [8].

The temperature data used in this study are taken from the Copernicus Marine Environment Monitoring Service’s global ocean reanalysis (product GLOBAL_ MULTIYEAR_PHY_001_030, dataset cmems_mod_glo_phy_my_0.083deg_P1D-m) at 5 m. These daily mean fields, based on the MERCATOR GLORYS12V1 configuration from Mercator Ocean, provide a global simulation at a high spatial resolution of 1/12˚ (approximately 8 km). These data were extracted via the Marine Data Store’s MyOcean portal from the Copernicus Marine at https://data.marine.copernicus.eu/viewer/expert/. They have been the subject of several studies in the Atlantic, such as those by Qingyang Song and Hidenori Aiki [23], and are sufficiently well characterised in this oceanic region by the work of Drévillon et al. [24].

Figure 1. Map of the Atlantic with a close-up of the tropical Atlantic. The area outlined in red represents the study area, showing the average daily temperature on 12 August 2012, with the northern and central branches of the South Equatorial Current shown as solid lines flowing westwards and the Equatorial Counter-Current shown as a dotted line flowing eastwards. South Equatorial Current (nSEC and cSEC) and Undercurrent (EUC).

2.3.1. Calculation of the Cooling Surface

To determine the cooling surface area, the methodology developed by Caniaux et al. [8] was applied. This approach involves integrating, on a daily basis, the area of all grid cells where the SST is below a predefined threshold value, using the formula below. This thermal threshold is set at 25˚C, a characteristic value as it is lower than the average ocean temperature measured in June in the eastern equatorial Atlantic, a key period for the development of upwelling [8]. According to these same authors, the cumulative area thus obtained must exceed a critical threshold set at 0.40106 km².

where US is the upwelling area, H_e is the Heaviside function (H_e = 1 when SST < 25˚C and 0 otherwise), A is the domain from 17˚W to 12˚E and from 2˚S to 2˚N, and SST is the daily mean temperature.

Secondly, the daily surface anomaly is simply the difference between the observed surface value and the climatological value for the corresponding calendar day. This climatology is established over the entire study period (1993-2025). Thus, a positive anomaly indicates an upwelling area greater than the daily normal, whilst a negative anomaly indicates a deficit relative to this reference.

2.3.2. Empirical Orthogonal Functions

To analyse the spatio-temporal variation in the upwelling area, we used Empirical Orthogonal Function (EOF) analysis. This method, which is an extension of Principal Component Analysis (PCA) to spatio-temporal fields, has established itself as a robust statistical tool in oceanography and climatology. It enables the dimensionality of large, multidimensional datasets to be reduced by decomposing the signal into orthogonal modes of variability [25]. These modes, ranked according to their contribution to the total variance of the system, allow spatial structures and their associated temporal amplitude coefficients to be isolated [26]. In practice, most of the physical variance is concentrated in the first three or four modes, which capture the dominant dynamic processes, whilst higher-order modes are generally regarded as background noise. Readers may find further details on this method in the work of Shields et al. [26].

2.3.3. Detection of MHW Periods

To analyse the impact of MHWs on the upwelling area, we determined the MHW period using the method developed by Hobday et al. [3], which has become the international standard in oceanography [4]. It is based on a rigorous statistical approach that enables events to be identified from daily temperature time series spanning at least 30 years [27]. This approach, which has already been validated in the tropical Atlantic for the study of MHWs, is based on the definition of a moving seasonal climatology and a temperature threshold set at the 90th percentile of the local historical distribution [5]-[7]. For the present study, the impact of MHWs on the upwelling area is examined over the fifteen days preceding, during and the fifteen days following the MHW period. The choice of this time window is based on the work of Garfinkel et al. [28], which demonstrates that such a duration allows for the optimal capture of the responses of oceanic parameters induced by MHWs. This approach is also consistent with the methodologies used in recent studies of the region, which analyse oceanic and atmospheric conditions before, during and after MHW events in order to understand their impacts [5][6]. In order to confirm the statistical robustness of the impact of MHWs on the ACT, a composite approach is applied over the period 1993-2025, supported by a non-parametric bootstrap analysis (2.000 resamples) of the detected MHW events, providing 95 per cent confidence intervals for each phase.

The Hovmöller plots (Figure 2) show the seasonal and interannual variability in the spatio-temporal extent of the cooling (expressed as area: 10⁶ km²) within the cold water tongue region (17˚W - 12˚E and 2˚S - 2˚N) from 1993 to 2025.

Figure 2(a) shows an extremely pronounced and regular seasonal pattern of the cold-water tongue throughout the 33 years of observation. It indicates that the cooling begins to manifest itself tentatively in May (M), intensifies rapidly in June (J) to reach its maximum spatial extent in July (J) and August (A), with cooling areas regularly exceeding 70.10⁶ km² to 80.10⁶ km². A relaxation phase is also observed in September (S), followed by an almost complete disappearance of the surface signature between October (O) and April (A). This behaviour is consistent with the classic dynamics of upwelling induced by the seasonal strengthening of the trade winds. Figure 2(a) reveals significant interannual variability, resulting in years of intense upwelling during the period 1993-2006. Conversely, several years, such as 2007, 2008, 2016 and 2021, show a dramatic weakening of the cooling surface area, with values rarely exceeding 50 - 60 × 10⁶ km², indicating weak or aborted upwelling. Figure 2(b), which shows the evolution of the area as a function of longitude and time, highlights a marked east-west asymmetry in the extent and intensity of the cooling. Figure 2(b) reveals that the centre of the cooling is located between 15˚W and 5˚W. Indeed, in this zone the cooling is at its maximum, covering an area of 7.000 to over 10.000 km². Moving eastwards towards 3˚W to 10˚E, the extent of the cooling decreases drastically (<4.000 km²). The interannual variability shown in this figure indicates that the period 1993-2005 was one in which the western core was particularly active. During this period, two major peaks in extent were observed: in 1993-1997 and, in particular, between 2000 and 2002, when the area exceeded 10.000 km². Furthermore, a second period, 2006-2015, corresponds to a transitional phase during which the core remained stable around 10˚W, with a notable secondary peak in 2013-2014. However, the overall spatial extent is slowly beginning to contract horizontally compared with the previous decade. A break is observed in 2016, with the core at 10˚W almost completely disappearing between 2016 and 2023 halving the cooling area, which falls below 6.000 km². From 2024 to 2025, a rebound is observed, corresponding to a sudden resurgence of cooling towards the end of the 2025 period, centred on 10˚W.

Figure 2. Hovmöller plot of the upwelling surface (US) of the ACT by month-year (a) and longitude-year (b) during the period 1993-2025.

Applying EOF analysis to our time series from 1993 to 2025 enabled us to extract the principal modes of variability of the ACT surface over this period. Of the modes obtained, only the first three were retained, as they account for the majority of the system’s total variance. The first mode (Figure 3(a), EOF1, 71.5% of variance) reveals an overall negative signal across the entire basin, with values ranging from 0 to –2.450 km² and maximum amplitude at the heart of the cold water tongue. This mode represents an overall contraction or suppression of upwelling. PC1 (Figure 3(d)) shows a dramatic interannual transition: prior to 2015, it indicates a signal with strongly negative values (1996, 2000), which, when combined with the negative spatial pattern, points to phases of very widespread upwelling. From 2016 onwards, PC1 becomes almost consistently positive, peaking at an absolute maximum in 2021. A positive PC1 multiplied by a negative EOF1 yields a strongly negative result, indicating a drastic and continuous reduction in the area of upwelling over the past decade. Furthermore, Mode 2 (Figure 3(b), 9.4% of the variance) reveals, through its spatial structure (EOF2), a very distinct zonal dipole. This dipole is characterised by negative anomalies in the west (−800 km²) and positive anomalies in the east, near the African coast (+1.200 km²). The PC2 (Figure 3(e)) of this mode captures high-frequency variability with marked oscillations in 1993, 1996, 2005, 2010 and 2021, with strong positive peaks, and in 1998, 2000 and 2002, with negative peaks. Furthermore, mode 3 (Figure 3(c), 4.3 per cent of the variance) has a spatial structure (EOF3) that shows a meridional banding pattern or a heart-shaped pattern. The centre of the ACT exhibits positive values of over 1.200 km², whilst the northern and southern edges (particularly in the south-west) show negative values exceeding −300 km². The PC3 (Figure 3(f)) associated with this mode shows high variability with extreme phases, particularly at the very start of the 1993 period, and marked troughs in 1999 and 2014.

To characterise the direct impact and temporal signature of MHWs on the ACT, surface upwelling anomalies were composite analysed according to three key phases (Figure 4). However, MHW events separated by ≤2 days were merged into a single episode; the others, including those with overlapping Before/After windows, were retained individually for the composites.

The first phase is the precursor phase, corresponding to 15 days before the MHWs, and it shows that the negative anomaly is often already present 15 days before the MHW, with the exception of the years 1994, 1996, 2009, 2010, 2013, 2014, 2022 and 2025. This indicates that the ACT was already weakening, creating conditions conducive to the onset of the MHWs. During the MHWs, the surface deficit of the cold-water tongue reaches massive proportions. Historical lows were recorded in 2007, with a record collapse of over −30.000 km², in 1999 at around −230.000 km², in 2003 at over −250.000 km² and, more recently, in 2021 at around −270.000 km². Values recorded after MHWs are almost always higher than those recorded during MHWs. This shows a tendency towards resilience: once the MHWs have ended, upwelling tends to resume and reduce its negative anomaly. However, recovery is rarely complete; values remain negative, often fluctuating between −50.000 and −100.000 km², indicating a thermal memory effect in the surface ocean, which takes time to dissipate the accumulated heat. Furthermore, prior to 2015, the significant events were distinct but spaced out over time (1999, 2003 and 2007).

Figure 3. Spatio-temporal trends in the main modes of variability of the Equatorial Atlantic CTD surface from 1993 to 2025.

Figure 4. Changes in the ACT area 15 days before the MHWs (grey curve plus circle), during the MHWs (red curve plus square) and 15 days after the MHWs (blue curve plus triangle) from 1993 to 2025. The horizontal dotted line indicates the zero value.

Figure 5. Composite of the ACT surface based on bootstrap resampling 15 days before the MHWs (grey box), during the MHWs (red box) and 15 days after the MHWs (blue box) during the period 1993-2025. The vertical lines represent error bars and the (p) values correspond to p-values.

In order to confirm the statistical robustness of the impact of MHWs on the ACT, a composite windowing tool combined with a bootstrap resampling method (95% CI, n = 2.000) was applied across the entire 1993-2025 period. The results (Figure 5) demonstrate with high statistical significance (p < 0.001) that MHWs are accompanied by a systematic collapse in the upwelling area. During the pre-event phase (Before), the average negative anomaly is already nearly −68.000 km². During the active phase (During), which corresponds to the peak of the impact, MHWs reduce the ACT by an average of nearly −110.000 km² over the entire study period. Finally, the post-event phase (After) shows a persistent anomaly of −56.000 km². Although there is a marked improvement compared with the active phase, the anomaly does not return to zero. Furthermore, for all three phases, the anomaly is significantly different from zero with a confidence level of over 99.9 per cent (p < 0.001). The error bars for the “Before” phase (from −47.000 to −90.000 km²) and the “After” phase (from −40.000 to −73.000 km²) overlap considerably; by contrast, the amplitude of the “During” phase extends much further down to −142.000 km².