In this study, we use version 2.2.4 of the Simple Ocean Data Assimilation (SODA 2.2.4) reanalyses [16]. This version corresponds to their first assimilation simulation spanning over 100 years and utilises wind data from the 20Crv2 satellite. The ocean model is based on the Parallel Ocean Program (POP) physics, with an average resolution of 0.25˚ × 0.4˚ over 40 levels. The observational data include in-situ and satellite data. The results are presented as monthly averages, projected onto a uniform grid of 0.5˚ × 0.5˚ × 40 levels covering 1901-2008 period. Data are available for download: https://www.cen.uni-hamburg.de/en/icdc/data/ocean/easy-init-ocean/soda-224.htm.

Observational data from the Hadley Centre Sea Ice and Sea Surface Temperature (HadISST) are also used. This dataset includes a combination of complete global monthly fields of SST and sea ice concentration on a 1˚ × 1˚ latitude-longitude grid, from 1870 to the present day [17]. HadISST1 data are compiled from SST measurements derived mainly from ship-based observations and, since 1982, from a combination of in situ measurements and satellite-derived SSTs. The dataset can be accessed and downloaded: https://www.metoffice.gov.uk/hadobs/hadisst/data/download.html.

We also used observed precipitation data from the Climate Research Unit Time Series (CRU TS). CRU TS is a monthly 0.5˚ × 0.5˚ grid-based terrestrial climate dataset available from 1901 onwards and covering all land areas except Antarctica [18]. It is based on the interpolation of monthly climate anomalies derived from an extensive network of weather stations. In the CRU TS v4 version, the station time series are first converted into anomalies relative to the 1961-1990 period, and then interpolated using angular distance weighting. The interpolated anomalies are then converted into actual climate values using the CRU CL v1.0 climatologies. The data are publicly available for download: https://crudata.uea.ac.uk/cru/data/hrg/.

Reanalyses from the National Centers for Environmental Prediction (NCEP) in collaboration with the National Center for Atmospheric Research (NCAR), known as the NCEP/NCAR Reanalysis, were also used to calculate the MABL. They cover the period from 1948 to present with a monthly temporal resolution. The data are provided on a global grid of 2.5˚ × 2.5˚ (longitude × latitude) and comprise 17 pressure levels ranging from 1000 to 10 hPa. Data are available for download: https://apdrc.soest.hawaii.edu/datadoc/ncep_mon.php.

2.2.1. Marine Atmospheric Boundary Layer Height

In this study, the MABL top is defined as the altitude where the gradient of the potential temperature between 700 and 2500 m is maximal as in [19]. This involves identifying the altitude at which the potential (or virtual) temperature gradient is at its maximum, marking the transition between the turbulent layer and the free atmosphere [20][21]. The MABL index is then constructed by spatially averaging the MABL over the equatorial Atlantic region (20˚W - 5˚W, 2˚N - 4˚S) and finally standardized.

2.2.2. Mixed Layer Depth and Thermocline Depth

Oceanic MLD relies primarily on two main categories of methods: threshold methods and gradient methods. In this study, the method based on a density threshold of 0.03 Kg m/m² is used as it provides a better estimate of the MLD [22][23]. The depth of the surface mixing layer is defined as the depth z at which the potential density difference Δ σ θ ( z ) = σ θ ( z ) − σ θ ( z 0 ) in the upper ocean exceeds a specified threshold value. Here z 0 is a reference depth (generally in the range z 0 = 0 at the ocean surface to 10 m), and σ θ ( z ) = ρ θ ( z ) − 1000 Kg∙m⁻³ is the density anomaly for the measured potential density ρ θ .

Thermocline depth (THD) is estimated from the 20˚C isotherm using data from the SODA dataset. This approach is based on the assumption that the 20˚C isotherm is a good proxy for the thermocline position in tropical regions [24].

2.2.3. Statistical Tools

1) Anomaly

An anomaly calculation involves measuring the deviation of a variable from a reference value, often a climatological mean. By subtracting the long-term mean from the observed value, an anomaly is obtained, which facilitates comparisons between different periods and the analysis of trends or extreme events. Anomalies are calculated by removing the monthly climatology from each variable in order to eliminate the seasonal cycle. A detrending procedure is then applied to the anomalies to remove long-term trends that could influence the analyses. Finally, the same reference period (1948-2008) is used for all the variables considered.

A t is the anomaly,

X t is the observed value,

X ¯ r e f is the climatological mean over the reference period (1948-2008).

2) Correlation and Linear Regression

The 20-year moving correlation between oceanic and atmospheric parameters allows us to examine how interactions between the ocean and the atmosphere evolve over time. It highlights slow variations, regime shifts, and the strengthening or weakening of climatic links. This approach thus provides a dynamic view of ocean-atmosphere interactions, which is essential for understanding long-term climate variability and anticipating its impacts [25]. In this study, we use Pearson’s correlation coefficient based on the following estimator [26].

where:

r t is the moving correlation at time t

ω is the window size (number of observations),

x are y two independent variables,

x ¯ t is the mean of x over the window,

y ¯ t is the mean of y over the same window.

The regression is the slope given by:

A 20-year moving window is used to analyze the temporal evolution of relationships by computing statistics over successive overlapping sub-periods, thereby revealing low-frequency variability and the non-stationarity of the processes [27]-[29]. The first period (1948-1971) contains 24 points and the second period (1972-2008) contains 37 points.

The statistical significance of the results is assessed using appropriate statistical tests. For moving correlations and linear regressions, a bootstrap procedure based on a Monte Carlo technique involving 300 permutations is applied. Only results that are statistically significant at the 95% confidence level are retained.

Figure 1 shows the seasonal cycle of the MABL height, the MLD depth and the SST gradient. The MABL remains relatively stable between January and May (700 - 750 m), a period characterised by more stable atmospheric stratification, indicating low turbulent flux between the ocean and the atmosphere [30][31]. From June MABL rises rapidly, reaching a maximum of around 1100 m in August. This trend is preceded by a decrease in the SST gradient, which reaches its minimum in July, suggesting a time lag of around one month between changes in SST and the response of the MABL. Indeed, the reduction in the SST gradient promotes cooling of the ocean surface and an increase in turbulent heat fluxes to the atmosphere, thereby enhancing instability and the thickening of the MABL [2][4][32].

Figure 1. Seasonal cycle of MABL, MLD, gradySST (averaged between 20˚W - 5˚W, 2˚N - 4˚S).

The MLD reaches a minimum in April (17 m) and a maximum in October (34 m). From April to October, its gradual deepening is consistent with the increase in heat flux, and thus a strengthening of the surface wind and the development of Atlantic cold tongue consistent with the work of [33]-[35]. Between October and April, there is a decrease in the MLD; this decrease is likely linked to a weakening of the southerly wind and a reduction in vertical mixing (thin MABL), as suggested by [36].

Figure 2 shows the trend in the 20-year moving average of MABL height and MLD during the JJA season in the Equatorial Atlantic Cold Water Tongue region (20˚W - 5˚W; 4˚S - 2˚N) for the 1948-2008 period. The use of a moving average makes it possible to filter out interannual variability and highlight low-frequency climatic fluctuations associated with ocean-atmosphere interactions in tropical regions [37][38]. Over 1948 to the mid-1950s period, MLD was slightly positive whilst the MABL remained relatively stable, indicating a relatively balanced state of the ocean-atmosphere system in the equatorial cold tongue. Between the mid-1950s and the mid-1960s, variability remained moderate, with positive fluctuations in the MABL while the MLD gradually decreased, possibly due to variations in the intensity of the equatorial trade winds. A clearly negative phase then emerges between the mid-1960s and the early 1970s, a period during which both variables reach their lowest values. Stronger stratification limits vertical mixing and results in a thinner mixing layer [33][39], and sea surface cooling stabilises the atmospheric boundary layer and reduces its thickness [2][4]. From the early 1970s, both variables show an upward trend. The period 1973-1980 corresponds to a transitional phase characterised by a gradual rise towards less negative values, possibly linked to a weakening of upwelling and a relative warming of the ocean surface. In 1980s, the MABL increased more sharply and the MLD returned to positive values.

Figure 2. 20-year moving average of the MABL and the MLD (averaged between 20˚W - 5˚W, 2˚N - 4˚S).

We will now examine the mechanisms that govern the decadal variability of the MABL. Figure 3 shows the temporal evolution of the moving correlations between MABL height and several atmospheric and oceanic parameters over the period 1948-2008. The indices are averaged over the equatorial cold-water tongue region of the Atlantic, bounded by 2˚N - 4˚S and 20˚W - 5˚W. Moving correlations are calculated using a 20-year time window in order to analyse long-term variability. The analysis is restricted to the JJA season. This approach allows the identification of dominant relationships as well as any long-term variations therein.

Figure 3. 20 years-moving correlation between MABL and MLD, SST, SST gradient, and meridional wind THD over Equatorial Atlantic (averaged between 20˚W - 5˚W, 2˚N - 4˚S. Circles indicate statistically significant correlation at 95% confidence level.

Over the entire period, weak and non-significant positive correlations were observed between MABL and MLD, as well as between the MABL and the meridional wind (V). These results suggest that the influence of surface ocean dynamics and meridional circulation on the MABL is not direct, but occurs via other variables, particularly thermodynamic ones, such as SST and air-sea turbulent fluxes [2][32].

By contrast, the correlation between MABL and SST is generally negative and significant throughout the entire period studied, with the exception of a few intervals where it is weaker and non-significant. This persistent link highlights the fundamental role of SST in controlling the stability and vertical structure of the MABL, particularly through turbulent heat fluxes, as suggested by [4][5].

However, the relationship between the MABL, the meridional SST gradient (dSST/dy) and the thermocline (THD) is non-stationary. Indeed, while the correlations between the MABL and dSST/dy, as well as between the MABL and THD, were negative and significant during the period 1948-1971, they became positive and insignificant from in 1970. These temporal breaks suggest the existence of a climate regime shift affecting air-sea coupling mechanisms. Such changes are consistent with the large-scale climatic reorganisations documented in the 1970s, notably the climate shift of 1976-1977, which is largely associated with lasting changes in atmospheric and oceanic circulation in tropical and subtropical basins [40]-[42]. These reorganisations may have altered the structure of oceanic temperature gradients, the intensity of air-sea fluxes and, consequently, the sensitivity of the MABL to local thermodynamic forcings.

In summary, if MABL variability was strongly controlled by local mechanisms in the equatorial Atlantic notably dSST/dy and THD prior to the 1970s, this variability appears to have been driven by remote factors from 1970 onwards. The weakening of the correlations between the MABL and the SST gradient or THD after the 1970s suggests a transition towards a regime in which the MABL is more strongly controlled by large-scale atmospheric forcings [43][44]. In the next section, we will attempt to understand the climate shift through SST signals across the two periods.

Figure 4 displays the regression fields for SST anomalies associed to the depth-based MABL index (see the Methods section for details on how the index is calculated) for two distinct periods: 1948-1971 and 1972-2008.

Over the period 1948-1971, a significant cooling was observed across the entire equatorial Atlantic, with the greatest cooling occurring along the equatorial cold-water tongue, and a barely significant warming in the northern tropical Atlantic. These anomalies form an SST dipole in the tropical Atlantic associated with the Atlantic Meridional Mode, as described by [45]. Negative SST anomalies peak at lags −1 and 0 (MJJ and JJA); this corresponds to the phase of intensification of the cold tongue. During this season, the southerly trade winds promote the upwelling of the thermocline and strengthen upwelling in the eastern Atlantic, leading to a cooling of surface waters and amplifying negative SST anomalies [46][47]. In the Pacific, negative SST anomalies are appearing across the entire equatorial Pacific; this anomaly becomes significant from lag −4, well after the negative anomaly in the Tropical Atlantic: the MABL variability is driven by the SST gradient in the Tropical Atlantic (the meridional mode), hence the strong correlation between the MABL and the meridional SST gradient during this period (see Figure 3).

Figure 4. Linear regression of the SST on the MABL index. Black dots represent areas that are statistically significant at 95% confidence level. (a) 1948-1971, (b) 1972-2008.

By contrast, over the period 1972-2008, the MABL is evident across all basins: the tropical Atlantic, the Indian Ocean and also the Pacific. The pattern of SST anomalies shows a notable change: significant positive anomalies of around 0.4˚C appear in the tropical Pacific, between lag −9 and lag −6, peaking at lag −8. This SST pattern is associated with the warm phase of El Niño-Southern Oscillation (ENSO), El Niño in the Pacific, which, according to the literature, began to become significant from the 1970s [48]. In the equatorial Atlantic, the negative SST anomaly still appears, but this anomaly is at its maximum at lag 0 in JJA. ENSO-related SST anomalies alter tropical convection and the structure of global atmospheric circulation, notably the Walker circulation [49][50].

In summary, the difference observed between the two periods can be interpreted as a shift in the tropical climate regime that occurred around the 1970s. After the 1970s, the influence of the Pacific on the tropical Atlantic appears to have become more pronounced, which would explain the presence of a precursor ENSO signal in the Pacific prior to the development of the cold tongue in the Atlantic. Regression analysis of SSTs against the MABL index reveals a change in ocean-atmosphere coupling in the equatorial Atlantic between the two periods studied. Prior to 1970, the variability of the MABL, or the equatorial cold-water tongue, appears to have been dominated by local processes linked to the ocean-atmosphere dynamics of the equatorial Atlantic, namely the meridional mode. By contrast, after 1970, the variability of the cold tongue appears to be more strongly influenced by teleconnections associated with ENSO, suggesting a stronger interaction between the two tropical basins. In the following sections, we will analyse the atmospheric dynamics associated with MABL variability and its impact on rainfall in West Africa.

Figure 5 shows the regression of wind speed, wind direction and mean sea-level pressure (MSLP) anomalies on the MABL index over 1948-1971 and 1972-2008 periods. Remember that negative lags indicate that atmospheric anomalies lead MABL variability and positive when these anomalies lag the MABL index.

Over the period 1948-1971 (Figure 5(a)), an increase in wind tension is observed north of 16˚N. This anomaly, which appears two months before the peak of MABL variability (lag −2), is driven by positive MSLP anomalies and may be associated with the southern part of the Azores High.

Between 8˚N and 16˚N, offshore, the wind anomaly is negative, reflecting a weakening of the northeasterly trade winds. In contrast, the southeasterly trade winds strengthen north of the equator (between 15˚N and 30˚N) as one approaches lag 0, then persist for a month afterward (lag+1). This strengthening can be explained by a dipole of MSLP anomalies: negative north of the equator and positive south of it. The negative MSLP anomaly, near the continent, promotes the inflow of a westerly flow toward the West African continent, while the positive anomaly generates a southward wind anomaly in the cold tongue region, reflecting a decrease in the monsoon flow which disappears entirely at lags 0 and +1.

Over the 1972-2008 period (Figure 5(b)), which the exception of a signal indicating a weakening of the Azores High north of 16˚N (in contrast to the previous period, where a strengthening was observed), the structures of wind and MSLP anomalies remain broadly similar. However, the westerly flow is less intense, and the decrease in the monsoon flow is significantly more pronounced.

Figure 5. Linear regression of stress wind (colors and vectors) and MSLP (contours) anomalies on the MABL index for 1948-1971 period (a) and 1972-2008 period (b). Only MSLP anomalies significant at 95% confidence level are shown.

To examine the precipitation signals associated with MABL variability, we present in Figure 6 a regression of precipitation anomalies in West Africa against the MABL index for two distinct periods (1948-1971 and 1972-2008).

Across both periods, a dipole pattern of precipitation anomalies is observed: positive over the Sahel and negative over the Gulf of Guinea region. The increase in precipitation over the Sahel can be explained by the strengthening of the southeasterly trade winds as well as a more intense westerly flow, while the decrease observed over the Gulf of Guinea is linked to a weakening of the southwesterly monsoon flow. However, significant differences emerge between the two periods. The dipole is much more pronounced during the 1948-1971 period than during 1972-12008, and clearly opposite anomalies are observed over the western Sahel. Indeed, during 1948-1971, a maximum positive anomaly is observed over the coastal areas of the Sahel, whereas it becomes negative during 1972-2008. This maximum positive anomaly can be attributed to the strengthening of the westerly flow, which constitutes the main source of moisture for precipitation in the western Sahel [51][52]. In contrast, during 1972-2008, the precipitation anomaly becomes negative in this same region, while it remains positive over the rest of the Sahel.

Figure 6. Linear regression of precipitations anomalies on the MABL index for 1948-1971 period (a) and 1972-2008 period (b). The black contours indicate anomalies that are statistically significant at 95% confidence level.