The interaction between ocean and atmosphere plays a crucial role in the climate system. This interaction involves heat and freshwater fluxes, which affect temperature and salinity variations in the upper-oceanic mixed layer. Therefore, sea surface temperature and salinity (SST, SSS) are two key climate variables, and understanding the balance of processes determining their variability is a key requirement to accurately simulating the climate system.

In the eastern tropical Atlantic, seasonal climate variability is mostly associated with the West African monsoon (WAM) and strongly linked with SST and SSS variations. SST is characterized here by a strong seasonal cycle, which influences the regional climate, particularly the large-scale atmospheric circulation and rainfall over the ocean and African continent (Carton and Zhou, 1997; Foltz et al., 2013; Kushnir et al., 2006). In the Gulf of Guinea, the seasonal formation of the Atlantic cold tongue (ACT), associated with equatorial upwelling, is the main feature associated with SST seasonal variations (Chang et al., 2006; Peter et al., 2006; Caniaux et al., 2011). This seasonal cooling creates an intense meridional SST front that enhances southern trade winds and shifts the Intertropical Convergence Zone (ITCZ) northward, which triggers the WAM (Philander and Pacanowski, 1981; Picaut, 1983; Waliser and Gautier, 1993; Caniaux et al., 2011). In the Angola–Benguela frontal zone off southwestern Africa, SST variability also has an impact on coastal precipitation (Reason and Rouault, 2006). Conversely, the atmospheric conditions in the eastern tropical Atlantic impact SST, as trade winds are the main driver of eastern boundary upwelling systems found along the coasts of Angola–Namibia and Senegal–Mauritania.

Like SST, SSS is an essential climate variable. Its variations are closely linked to the global hydrological cycle, as freshwater exchanges between the ocean and atmosphere control its mean large-scale distribution (Durack and Wijffels, 2010; Bingham et al., 2012): regions of high SSS are dominated by evaporation, while regions of low SSS are dominated by precipitation. In the eastern tropical Atlantic, the intense precipitation under the ITCZ has a strong impact on seasonal variability of SSS. For example, off Senegal and Guinea, SSS decreases in boreal fall following the period when the ITCZ reaches its most northern position, which leads to a maximum in precipitation and runoff from Senegal and Gambia rivers (Camara et al., 2015). The eastern tropical Atlantic is indeed characterized by low SSS plumes due to strong river discharges, including the Congo River (the second-largest outflow in the world after the Amazon River) and the Niger River in the Gulf of Guinea. SSS variations are also associated with upwelling. In the equatorial Atlantic, SSS increases during the development of the Atlantic cold tongue (ACT) in boreal spring–summer (Schlundt et al., 2014; Da-Allada et al., 2014). Summer upwelling at the northern coast of the Gulf of Guinea also increases coastal SSS (Alory et al., 2021). Off Angola, relatively high SSS is observed during the upwelling season in August, while low SSS appears in March and October–November (Kopte et al., 2017; Awo et al., 2022). In the Benguela upwelling system further south, the maximum in SSS appears in April and the minimum in October (Junker et al., 2017).

SST and SSS have already been the focus of several studies in the eastern tropical Atlantic. Studies only based on observations (in situ and/or satellite), and others combining observational data and model data, show a variety of physical processes contributing to the heat and salt budget with a different balance from one region to another (Foltz et al., 2003; Foltz and McPhaden, 2008; Da-Allada et al., 2013, 2014).

In the northern tropical Atlantic far from the coast, the heat budget is largely driven by surface heat fluxes, essentially solar radiation varying with the cloudy ITCZ position (Carton and Zhou, 1997). Off northwestern Africa in the Senegal region, the seasonal cycle of SST is associated with coastal upwelling modulated by the seasonal variations of alongshore winds. In the equatorial zone, the mixed-layer heat budget is mainly controlled by surface heat fluxes, of which mainly the solar radiation is important (Carton and Zhou, 1997; Foltz et al., 2003; Yu et al., 2006). The dominance of the surface heat flux is highlighted by a recent study based on PIRATA buoy data off the Equator, in particular at the 6∘ S, 8∘ E position. At this latitude, seasonal variations in SST governed by solar flux and latent heat flux are shown to be associated with the meridional migration of the ITCZ and the formation of low-level marine stratocumulus (Scannell and McPhaden, 2018). However, other oceanic processes can also play an important role in the seasonal heat budget. These include zonal advection and vertical mixing, which are important during the formation of the ACT (Foltz et al., 2003; Hummels et al., 2014; Schlundt et al., 2014). However, the influence of vertical mixing in the heat budget remains low in the eastern (from 0∘ E to the coast of Africa) compared to the central equatorial Atlantic (Jouanno et al., 2011; Hummels et al., 2013). This is explained by the strong stratification in the eastern equatorial Atlantic due to the transport of low-salinity and warm waters from the Gulf of Guinea, which contributes to reducing the vertical mixing. The low salinity is associated with the intense precipitation and important freshwater intakes from the Niger and Congo rivers in the Gulf of Guinea. Off Angola, annual variability of SST is driven by meridional wind stress, causing coastal upwelling, and thermocline depth variations forced by remote equatorial effects (Carton and Zhou, 1997).

In the northeastern tropical Atlantic, including in the Senegal region, the salt budget is controlled by freshwater fluxes. The net mixed-layer salinity variations are, however, weak because of the compensation between the atmospheric and oceanic terms (Camara et al., 2015). In the eastern equatorial Atlantic and Gulf of Guinea, horizontal advection and vertical mixing play a dominant role in determining the seasonal cycle of salt budget, which cannot be explained by freshwater fluxes only (Tzortzi et al., 2013; Da-Allada et al., 2013, 2014). This dominance of zonal advection and vertical processes extends southward and in the Angola coastal region (Camara et al., 2015).

In addition, it can be noted that various approaches have been used to estimate the heat and salt budgets in the tropical Atlantic, in particular the advection terms. Foltz et al. (2003) analyzed the mixed-layer heat balance at PIRATA mooring locations, where they computed the heat advection from monthly gridded climatologies of near-surface horizontal velocity, based on ship drifts and Lagrangian drifters, and SST gradient fields based on a combination of ship, buoy, and satellite data. Wade et al. (2011) used a similar SST product but satellite-derived currents to estimate the heat advection every 10 d at positions of Argo profiles. Then monthly averages in nine boxes covering the Gulf of Guinea were used to study the seasonal cycle of mixed-layer heat as observed by Argo. Da-Allada et al. (2013) developed an original mixed-layer model of the tropical Atlantic at monthly, 1∘ resolution, with salinity driven by observation-based climatological freshwater budget terms, and several in situ and satellite surface currents products were tested for advection to identify processes driving SSS seasonal variations. With the same objective, Da-Allada et al. (2014) and Camara et al. (2015) used slightly different tropical Atlantic configurations of the NEMO (Nucleus for European Modeling of the Ocean; Madec, G., 2014) OGCM (ocean general circulation model), with mixed-layer salinity budget terms computed online, on the model spatial grid (0.25∘), and at each time step (20 min). We know from analyzing online diagnostics that the nonlinear advective terms cannot be neglected. Observational data, however, often have lower temporal and/or spatial resolution than would be necessary to fully capture the nonlinear terms. Moreover, it is much more difficult to estimate subsurface vertical terms than surface terms with observations; this leads to a non-negligible residual term, which interpretation is problematic. Online computation in an OGCM is practically the only way to close a mixed-layer heat and salt budget.

Nonlinear terms are associated with turbulence. In eddy-resolving models, they represent mesoscale activity related to eddies and tropical instability waves (TIWs). The characteristic size of eddies is given by the Rossby radius, which increases equatorward and has a minimum value of 30–40 km in the 30∘ S–30∘ N subtropical band (Chelton et al., 1998). TIWs have been shown to play an important role in the heat budget in the central equatorial Atlantic (Foltz et al., 2003; Grodsky et al., 2005; Jochum et al., 2005; Peter et al., 2006; Lee et al., 2014; Heukamp et al., 2022; Tuchen et al., 2022). In the tropical Atlantic, eddies have been detected mostly in coastal upwelling systems, where they are likely to affect mixed-layer heat and salt budgets too (Djakouré et al., 2014; Aguedjou et al., 2019).

In this paper, we exploit for the first time a recently produced tropical Atlantic mixed-layer heat and salt budget observation-based climatology. Moreover, we use a joint observation–model approach that has rarely been used: we compare the mixed-layer heat and salt budget terms estimated from observations to those simulated by a high-resolution OGCM simulation in the eastern tropical Atlantic. A sensitivity test to the spatiotemporal resolution at which advection terms are computed in the model is conducted. This comparison should allow providing a high-level model validation, isolating the contribution of mesoscale advection in the mixed-layer budgets, and quantifying the uncertainty in the different budget terms. We particularly focus for the mixed-layer budgets on the upwelling regions where oceanic processes are expected to be dominant. The observational product, the model, and the methodology used are presented in Sect. 2. Section 3 contains the results of observation–model comparison regarding mean heat and salt budgets, as well as their seasonal variability in selected regions. In Sect. 4, a discussion and conclusion are presented.

In this paper, we examined the dominant physical processes controlling the seasonal variability of mixed-layer heat and salt budgets in selected coastal regions of the eastern tropical Atlantic, namely the Senegal, equatorial, Angola, and Benguela regions. First, we used both a regional configuration of the NEMO model and the PREFCLIM observation-based climatology to analyze the spatial variations of the annual mean mixed-layer heat and salt budgets in the eastern tropical Atlantic (see Figs. 4 and 7, respectively). The model outputs were resampled to the PREFCLIM time–space resolution to compare maps of the mean processes contributing to mixed-layer heat and salt budgets, according to both sources. Second, we analyzed the seasonal variation of the mixed-layer temperature and salinity, their related tendencies, and potential driving processes: heat and freshwater flux, horizontal heat and salt advection, and other processes estimated from observations as a residual but explicitly resolved in the model for the selected regions. As the PREFCLIM climatology does not capture the mesoscale physical processes, we relied on the high-resolution model outputs to evaluate their contribution to the mixed-layer heat and salt budget.

For the preliminary validation, the results have shown that the model consistently reproduces the mean features of observed mixed-layer depth, temperature, and salinity in the eastern tropical Atlantic (see Figs. 1, 2, and 3, respectively). The existing differences between modeled outputs and the PREFCLIM climatology can be explained by the different heat and salt flux products that are used for forcing the model or for estimating the PREFCLIM budget terms. There are also differences in the method to define the MLD. The PREFCLIM climatology uses the algorithm of Holte and Talley (2009), whereas the model uses the density criterion (0.03 kg m-3 relative to the density at 10 m depth) recommended by de Boyer Montégut et al. (2004). For example, the observed strong positive bias of mixed-layer salinity in the model relative to observations around 12∘ S is associated with a positive bias in MLD. Except for these differences, the model and the PREFCLIM climatology capture the shallow nearshore mixed-layer depth along the Equator and the western coast of Africa where our selected regions are localized.

For the secondary validation, we have used both the model and the PREFCLIM climatology to analyze the annual mean of heat and freshwater flux as well as horizontal heat and salt advection. The model heat and freshwater fluxes largely agree with the PRECLIM climatology (Figs. 4a–b and 7a–b, respectively), except for differences in a few regions that can again be due to different flux products or MLD biases. There are important differences between the model heat and salt advection terms computed either offline (Figs. 4f and 7f, respectively), at PREFCLIM spatiotemporal resolution (2.5∘, monthly), or online (Figs. 4g and 7g, respectively) at high spatiotemporal resolution (0.25∘, 20 min) with subsequent monthly resampling at 2.5∘. These differences are explained by the high-frequency variability related to mesoscale and sub-mesoscale dynamics, which is included in the second case but not in the first case. In particular, including mesoscale dynamics leads to warming around the Equator and off the African coast in the southern Atlantic. West of 10∘ W and slightly north of the Equator, this is consistent with warming of the equatorial cold tongue by horizontal advection linked to TIWs (Jochum et al., 2004; Grodsky et al., 2005; Peter et al., 2006). This also suggests a similar role of eddies in the Angola and Benguela upwelling systems, where a large number of eddies have been detected (Aguedjou et al., 2019). Regarding salinity, mesoscale advection tends to freshen the Gulf of Guinea, probably by westward export through eddies of fresh waters from the Niger and Congo plumes (Houndegnonto et al., 2021).

At seasonal timescales, the monthly mixed-layer heat and salt tendency terms in the selected regions are very weak in both the PREFCLIM climatology and the model compared to individual terms contributing to the heat and salt budgets that tend to compensate for each other, as also found in previous studies (Da-Allada et al., 2013, 2014; Camara et al., 2015).

Surface heat fluxes, especially the solar flux, dominate the seasonal mixed-layer heat budget in the Senegal, Angola, and Benguela regions (Fig. 5b, h, and k, respectively). In the Senegal region, this result, and the secondary contribution of oceanic processes such as vertical diffusion and zonal advection, which add to latent heat flux to drive the observed winter cooling, confirms previous studies (Carton and Zhou, 1997; Yu et al., 2006).

In the equatorial region, the heat flux remains positive and nearly constant throughout the seasonal cycle. This shows the dominance of the shortwave flux that warms the mixed layer from September to April, although this warming weakens between November and December. Although our selected box slightly differs from previous regional studies, this result is in agreement with earlier studies (Peter et al., 2006; Wade et al., 2011). The variability of mixed-layer temperature, in particular the observed spring–summer cooling during the formation of the ACT, is mainly controlled by vertical heat diffusion (Fig. 5e), confirming other studies (Yu et al., 2006; Jouanno et al., 2011). Recently, Scannell and McPhaden (2018) also confirmed the role of turbulent vertical mixing from a PIRATA buoy located at the southeastern edge of the ACT. While it does not compensate for the cooling effect of vertical diffusion, zonal heat advection is positive all year long in the equatorial region, the only one among analyzed regions where it is so. This is the consequence of a negative zonal temperature gradient as the mixed-layer temperature decreases toward the coast, advected by westward currents associated with the SEC. When associated with meridional heat advection, this leads to a positive horizontal heat advection throughout all year except in the month of May. We note a nearly similar variability of horizontal advection in Wade et al. (2011), although this term is negative in their study except for June–July, when we both observe a positive maximum. This difference can be linked to either products used or the criterion used to define the MLD (temperature vs. density criterion) and maybe to the slightly different boxes of study. Jouanno et al. (2017) also found, like us, a permanent warming effect of horizontal advection in an equatorial box shifted west compared to ours using the same model configuration. In our box, according to the model this warming is largely due to mesoscale advection, probably by eddies as TIW activity sharply decreases east of 15∘ W (Foltz et al., 2003; Peter et al., 2006; Tuchen et al., 2022). There is overall agreement on the dominant role of vertical mixing to cool the mixed layer during ACT formation. This vertical mixing due to vertical diffusion is explained by the strong vertical shear between the Equatorial Undercurrent (EUC) and the SEC in our selected region, as discussed in previous studies (Wade et al., 2011; Jouanno et al., 2011; Hummels et al., 2013; Schlundt et al., 2014). The vertical mixing can also be enhanced by the effects of TIWs on vertical shear (Heukamp et al., 2022).

In the Angola region, the dominant role of heat flux in the mixed-layer heat budget was also found in previous studies (Carton and Zhou, 1997; Yu et al., 2006). In this region, the incoming shortwave flux warms the mixed layer from August until March against the action of latent heat flux. The competition between the shortwave flux and the latent heat flux is also mentioned in Scannell and McPhaden (2018), although at the 6∘ S, 8∘ E position, the horizontal advection remains weak in their study. The cooling observed between April and July is due to the decrease in solar radiation, probably due to cloud cover, with the added effect of zonal heat advection. We observe the warmest temperature in March, and the minimum is reached in August, which corresponds to the upwelling season (Ostrowski et al., 2009; Kopte et al., 2017). Although the contribution of zonal heat advection remains weak compared to the solar flux, the variations of zonal heat advection are in phase with the variation of heat tendency throughout the year.

In the Benguela region, the heat tendency variations are roughly in phase with the heat flux variations. The heat budget is mostly driven by the shortwave flux, as found previously in the neighboring southern Angola upwelling system. The cooling that occurs from March to August can be associated with cloud cover, which reduces the incoming solar flux, and also a small contribution of oceanic processes. The observed coldest temperatures correspond to the July–October upwelling season (Hagen et al., 2001; Muller et al., 2014).

There are much larger differences between PREFCLIM and NEMO in the seasonal variations of the mixed-layer salinity (Fig. 8a, d, g, j) compared to temperature. These seasonal variations generally have a larger amplitude and lower minima in the model, as also seen in Fig. 3. This can be due to several factors. First, although the PREFCLIM product benefited from newly available hydrographic data in the Senegal, Angola, and Namibia coastal waters, the data density is still low in the equatorial Gulf of Guinea (Dengler and Rath, 2015), the freshest waters associated with heavy rain as well as the large Congo and Niger River plumes; hence, the largest difference in seasonal salinity variations is in the equatorial box (Fig. 8d). Poor data density can also be associated with a seasonal bias that may prevent capturing the full seasonal cycle. Second, hydrographic profiles, notably those from Argo floats, do not sample the salinity minimum found in the upper few meters of the ocean in regions highly stratified by rain and river plumes, which induces SSS estimations higher than those observed from satellites (Boutin et al., 2016; Houndegnonto et al., 2021). This leads to overestimation of mixed-layer salinity too. Third, while the NEMO model configuration has high vertical resolution in the upper few meters and homogeneous spatial coverage, the way it reproduces mixed-layer salinity highly depends on its freshwater forcing, including river runoff, and its own dynamics that are of course not perfect. Despite these differences between observations and the model, the comparison is instructive.

The Senegal region is the only one among the four analyzed regions where the salt budget is clearly controlled by the surface freshwater fluxes, with an added runoff effect (Fig. 8b). From March to October the observed freshening can be explained by the combined effect of precipitation and Senegal and Gambia rivers inputs. From October to November, zonal salt advection adds its contribution to existing freshwater inputs to freshen the mixed layer. Vertical salt diffusion, with an additional contribution of meridional and vertical salt advection, tends to increase mixed-layer salinity and partly compensates for the previous freshening effect. Although our selected regions are slightly different, these results are consistent with those of Camara et al. (2015). However, in our study, evaporation plays a dominant role in increasing salinity for the rest of the year, even if there is also a weak contribution of oceanic processes. This disagrees with the study of Camara et al. (2015), wherein the contribution of evaporation to the mixed-layer salt budget is very weak compared to our results. This contradiction can be explained by different model configurations as described in Da-Allada et al. (2017). Camara et al. (2015) use for model forcing an older version of the DRAKKAR Forcing Set (DFS4) compared to the one (DFS5.2) used in the present study, and their model has fewer vertical levels than ours (46 vs. 75). They also use a smaller density criterion (0.01 kg m-3 for Camara et al., 2015, vs. 0.03 kg m-3 in this study to estimate mixed layer depth) and an additional restoring term for salinity in their model.

In the equatorial region, the seasonal variability of mixed-layer salinity is mainly due to oceanic processes as shown in other studies (Da-Allada et al., 2013, 2014). In our study, we found vertical salt diffusion and zonal salt advection as dominant oceanic processes (Fig. 8f). From October to December and March to July, zonal advection is the most important freshening contribution, stronger that precipitation. It is explained by the westward South Equatorial Current (SEC), which transports low-salinity waters from the Gulf of Guinea associated with the Niger and Congo River plumes (Houndegnonto et al., 2021). The major role played by vertical salt diffusion in increasing the salinity in the mixed layer, demonstrated in previous studies (Da-Allada et al., 2014, 2017), is confirmed by our results for boreal spring–summer in particular. This strong vertical salt diffusion is the consequence of the vertical shear between the westward SEC and the eastward EUC (which transports high-salinity waters) but can be reduced by the strong salinity stratification caused by the Niger and Congo River plumes (Jouanno et al., 2011). Note that vertical diffusion, however, is strongly compensated for by zonal advection, above all in May. These results agree with the other studies covering the region despite slight differences in the limits of selected boxes (Berger et al., 2014; Da-Allada et al., 2014, 2017; Camara et al., 2015). Although the contribution of surface freshwater fluxes and runoff remains weak in our study, its seasonal variations follow those described by Da-Allada et al. (2017), with some time lag. During the period when the ITCZ is close to the Equator between November and April, the freshwater flux is dominated by precipitation and decreases the mixed-layer salinity, whereas the rest of the year, the freshwater flux is dominated by evaporation and increases salinity.

As in the equatorial region, horizontal and vertical oceanic processes drive the mixed-layer salinity in the Angola region too (Fig. 8g–i), in agreement with previous studies (Camara et al., 2015; Awo et al., 2022). Meridional salt advection explains most of the variability in salt budget, particularly its semi-annual cycle, as it freshens the mixed layer in February–April and September–October, when the southward Angola Current brings low-salinity water from the Congo River plume (Gordon and Bosley, 1991; Awo et al., 2022). However, for the rest of the seasonal cycle, a combined action of meridional salt advection, vertical salt advection, and vertical salt diffusion increases the salinity of the mixed layer. Although vertical advection is stronger than vertical diffusion, they remain in phase throughout the cycle and act against the runoff and zonal salt advection. Seasonal variations in both vertical salt diffusion and advection are driven by changes in the vertical salinity gradient related to the semi-annual intrusion of low-salinity surface waters (Camara et al., 2015; Awo et al., 2022).

In the Benguela region, individual contributions of physical processes are relatively weak in comparison to the other regions. The mixed-layer salinity variability is partly controlled by freshwater fluxes, particularly evaporation (Fig. 8j–k). Zonal advection remains negative throughout the year, and from March to December, it acts to decrease salinity, against the action of evaporation that is reinforced by vertical salt diffusion between September and December. The increase in salinity corresponds to the upwelling season in the southern part of Benguela upwelling system in summer (Muller et al., 2014).

Although increasing resolution in oceanic models intends to produce more realistic simulations by explicitly resolving mesoscale variability, and models are the only way to estimate all terms of the heat and salt budget in the mixed layer, it is difficult to directly validate such model budgets with in situ data. One problem is that globally available in situ data can only explicitly resolve near-surface horizontal processes, particularly advection, not vertical processes that have to be estimated as a residual. A second problem is that in situ observation density does not allow estimating horizontal advection at the high resolution available from models. Therefore, to be properly compared with those available from observations, model horizontal advection terms must be computed offline at the spatiotemporal resolution of observations. Our results indeed show that the time-averaged spatial distribution of NEMO offline heat and salt advection terms compares much better to PREFCLIM horizontal advection terms than the online heat and salt advection terms. However, when examining the seasonal cycle of horizontal advection in selected boxes, NEMO offline terms do not always compare well with PREFCLIM, sometimes less than online terms. This suggests that temporal coverage of in situ observations is more critical than spatial coverage, particularly for salinity, and especially in coastal areas of Africa where Argo profiles are relatively scarce and in the equatorial region where Lagrangian drifters do not stay long due to Ekman divergence. Another possibility would be to estimate advection from satellite products of SST, SSS, and currents, the latter estimated from altimetry and satellite wind for their geostrophic and Ekman components, respectively (Bonjean and Lagerloef, 2002), which are available at a resolution of a few tens of kilometers and a few days. The new Surface Water and Ocean Topography (SWOT) mission (Morrow et al., 2019) should soon further improve the resolution of geostrophic currents. The Soil Moisture Ocean Salinity High Resolution (SMOS-HR) mission project (Rodriguez-Fernandez et al., 2022) would also help to capture SSS gradients. The often large differences between offline and online advection terms in the model suggest an important role of small-scale (<2.5∘, <1 month) variability, which includes mesoscale activity recently documented in the eastern tropical Atlantic (Aguedjou et al., 2019), including at the Equator where differences are particularly large. Although the model mixed-layer budget validation has some limitations in the present study, our results are generally in agreement with earlier studies of mixed-layer heat and salt budgets in the tropical Atlantic (Foltz et al., 2003; Jouanno et al., 2011; Wade et al., 2011; Da-Allada et al., 2013; Camara et al., 2015). Except for local studies wherein coordinated field measurements and mooring deployment can be combined to close a short-term mixed-layer budget with in situ observations (Farrar et al., 2015; Farrar and Plueddemann, 2019; Vijith et al., 2020), in most regions, one can only use models to close mixed-layer budgets and trust them to quantify the processes hidden in the residual unresolved by global observations.