Different in situ and satellite datasets were used to evaluate the model.

Sea surface temperature (SST) in situ measurements are based on four coastal stations equipped with ONSET sensors installed along the NGG coast at Sassandra in Côte d’Ivoire (6.08°W; 4.95°N), Takoradi in Ghana (1.73°W; 4.88°N), Cotonou in Benin (2.67°E; 6.52°N) and Lagos in Nigeria (3.41°E; 6.42°N) in the framework of the Research Program in Physical Oceanography in West Africa (PROPAO). The in situ ONSET sensor hourly data series are available over the period 2008–2012 (Sohou et al., 2020). We also used the monthly version of the Moderate Resolution Imaging Spectroradiometer (MODIS-AQUA, L3, v2019, https://oceandata.sci.gsfc.nasa.gov ) SST, with a spatial resolution of 4 km over the time period 2007-2017. We prefer to use this infrared SST product rather than blended satellite products that can underestimate SST cooling in coastal upwelling regions (Meneghesso et al., 2020). Sea surface salinity (SSS) satellite data were collected from the Soil Moisture Active Passive (SMAP) v5 product, with a 0.25° and 8-day resolution, which benefits from a correction of systematic biases near the coasts and adjustment of the long term mean in dynamical areas like river plumes (Dossa et al., 2021). We also used SSS measurements collected in situ from underway thermosalinographs (TSG) during eight cruises or crossings of French research and commercial ships in the Gulf of Guinea during the summer season between 2000 and 2018 (Gaillard et al., 2015; Bourlès et al., 2019; Alory et al., 2021).

As for the sea surface height (SSH), it was evaluated using the CMEMS (Copernicus Marine Environment Monitoring Service) product, combining data from several satellite missions interpolated daily on a quarter degree (1/4°) resolution grid (Le Traon et al., 1998; Ducet et al., 2000).

In addition, we used monthly averages of in situ National Oceanic and Atmospheric Administration (NOAA) surface drifters data for the assessment of the zonal surface current (Laurindo et al., 2017).

Finally, subsurface temperature and salinity were extracted from the World Ocean Atlas (WOA; Locarnini et al., 2018; Zweng et al., 2019). We used the seasonally averaged climatological product, with 0.25° horizontal resolution and a 5 m vertical resolution in the upper 100 m.

The numerical setup is based on the Nucleus for European Modeling of the Ocean (NEMO; Madec, 2016) model. The model solves in a Cartesian coordinate system the primitive equations discretised in an Arakawa C grid. The regional configuration covers the tropical Atlantic at 1/36° horizontal resolution (10°S - 6.5° N and 10.5° W - 13° E), with 50 vertical levels (12 levels within the first 20 m and 22 levels in the upper 100 m). At the surface, the model is forced with the DFS 5.2 product (Drakkar Forcing Set version 5.2, Dussin et al., 2016) and the nonsolar heat fluxes are provided by bulk formulae (Large and Yeager, 2009). A river flow based on data from Dai and Trenberth (2002) is prescribed near the river mouths as a surface freshwater flow, associated with vertical mixing over a characteristic 10 m depth, which decreases exponentially river waters has not been prescribed in the model. It is similar to that of the oceanic surface waters. A reference simulation (“REF”) was carried out over the period 2005-2017, as well as a twin simulation (“noRIV”) over the period 2008-2017, where river runoff is not taken into account. In the latter, vertical mixing near the river mouths is maintained to ensure that differences between the simulations are only due to changes in freshwater flux. Both simulations follow a 2-year climatological spin up. Their monthly mean seasonal climatology will be compared over the common 2010-2017 period. The mixed layer depth (MLD) is defined as the depth at which the density has increased by 0.03 kg. m^-3, with reference to the density at 10 m. This criteria is recommended by De Boyer Montégut et al. (2004) and has been previously used in several studies in the region (e.g. Jouanno et al., 2011; Da-Allada et al., 2014; Jouanno et al., 2017; Alory et al., 2021; Da-Allada et al., 2021; Kanga et al., 2021). In particular, N’Guessan et al. (2019) show from visual inspection of more than 500 density profiles that it is the best suited criteria to determine the MLD in the Gulf of Guinea. The mixed layer is located between 9.5m and 12m in “REF” and between 10m and 13.5m in “noRIV” simulation, which means it extends over 9 or 10 model depth levels. These simulations were used for the first time in this work and have not been validated before.

To examine the causes of the surface warming induced by the Niger River, a mixed-layer heat budget is used, in both the REF and noRIV simulations, which can be written as follows (Jouanno et al., 2011; Jouanno et al., 2017; Da-Allada et al, 2021):

With

where, T is the model potential temperature,   ρ 0 the surface reference density (set to 1,021 kg.   m − 3 as in Li et al., 2013), Cp the heat capacity (set to 3984 J k g − 1   ° C − 1 as in Wade et al., 2011), h MLD, Q * the nonsolar surface heat flux (sum of the net longwave radiation, the sensible and latent heat fluxes), Q s the net shortwave radiation, f z = − h the fraction of the shortwave radiation that reaches the MLD, (u, v, w) the velocity components, k z the vertical diffusion coefficient for tracers and D l ( T ) the lateral diffusion operator. The total temperature trend on the left-hand side of equation (1) is thus controlled from left to right by zonal and meridional advection (XAD and YAD respectively), lateral diffusion (LDF), vertical advection (ZAD), vertical diffusion (ZDF), entrainment (RES) at the base of the mixed layer, and finally the net air-sea heat flux (ATMOS). We group oceanic processes into two categories: horizontal processes as HORIZ (HORIZ=XAD+YAD+LDF) and vertical processes as VERTIC (VERTIC=ZAD+ZDF+RES). All terms in equation (1) are computed explicitly in the model, except for the entrainment term (RES) which is estimated as a residual. For the investigation of the causes of the variation in meridional advection, we used the following decomposition:

where Δ represents monthly seasonal variations and< > represents annual means.

The meridional current is decomposed into geostrophic and Ekman components. The geostrophic component is calculated as follows:

Where SSH is sea level and f the Coriolis parameter. The Ekman velocity is obtained from the difference between the total velocity and the geostrophic velocity. The Brunt Vaïsala frequency N² (T, S) is decomposed into its components to describe the temperature stratification N²(T) and the salinity stratification N²(S). N² (T, S) is calculated as follows (e.g. Da-Allada et al., 2021):

where g is gravity, α is the coefficient of thermal expansion, β is the coefficient of haline contraction, ∂ T ∂ z and ∂ S ∂ z , the vertical temperature and salinity gradients. The shear parameter S h 2 is used to analyse the vertical shear of horizontal currents. S h 2 is calculated as follows (e.g., Da-Allada et al., 2021):

This section evaluates the ability of the model to reproduce the main characteristics of the NGG. Only a lower (1/12°) resolution regional NEMO version was previously validated (Alory et al., 2021).

a. Seasonal cycle of SST

The SST of the model was evaluated in comparison with the ONSET sensor data of the PROPAO program ( Figure 1 ). The model reproduces well the average seasonal cycle of the SST of the ONSET sensors. However, we observe differences in SST between the model and the in situ observations, which are more pronounced in boreal summer (up to 2°C at the Sassandra station). These differences could be due to small-scale near-shore dynamics not simulated in the model. The model is slightly warmer than the in situ measurements for all stations but the one in Nigeria where it is slightly colder, especially in summer. The later could be due to the fact that the temperature of the river waters, potentially warmer than the ocean, is not taken into account. These differences reach a maximum during the July-August-September (JAS) upwelling season, which timing is very consistent between the model and Onset sensors.

b. Summer climatology of SST, SSS, SSH and zonal surface current.

The climatological mean summer SST for both the model and the MODIS product for the 2007-2017 period are compared in Figures 2A, B . The model reproduces the main features of the NGG SST. The upwelling cells west and east of Cape Three Points are well represented. Both satellite and model show that the eastern cell is colder than the western cell downstream of Cape Palmas, as found by Djakouré et al. (2014). However, the model shows a cold bias in the coastal upwelling area relative to the satellite data, contrary to its warm bias relative to coastal stations. This suggests the satellite data may have deficiencies near the coast.

The model SSS is compared with the SMAP product in Figures 2C, D . The model reproduces roughly the SSS observed by the SMAP satellite, in particular the SSS minimum extending from 4°E to 6°E and associated with Niger River plume (Alory et al., 2021). However, salinity differences are observed, particularly in coastal upwelling area, East of capes where differences reach + 2 psu in SMAP product. Moreover, coastal freshening associated with rivers in western part of NGG are weaker in SMAP.

These differences are probably due to the difficulty of the satellite to observe salinity gradients near the coast (Dossa et al., 2021). Model SSS is also in good agreement with the TSG transects, with differences that may be due to interannual variability inherent to the sparse sampling (Alory et al., 2021).

The SSH of the model was also evaluated by comparison with the CMEMS altimetry data ( Figures 2E, F ). As expected, low SSH between longitudes 7°W and 2°E, is largely associated with cooling in the upwelling area. SSH increases eastward from 2°E with a maximum in the Niger River plume, due to steric variations associated with both temperature and salinity changes (Alory et al., 2021).

Finally, the zonal surface current is compared between the model and the drifters ( Figures 2G, H ). The model reproduces very well drifters’ observations. The eastward Guinea Current intensifies between the Capes (7°W-2°W) then weakens until it disappears around 4°E.

c. Vertical sections of temperature and salinity

In addition to the surface data, south-north vertical temperature and salinity sections of the model are compared to the WOA climatological data along 0°E ( Figures 3A, B ). The model reproduces well the vertical temperature section. The warm surface waters are isolated from the cold bottom waters by the thermocline (approximated by the 20°C isotherm).

There is also an upward movement of the thermocline from the open ocean to the coast which is consistent with coastal upwelling. Also, the model reproduces well the vertical salinity section ( Figures 3C, D ), with a slope of the halocline (in this case the 35.5 psu isohaline) similar to the thermocline slope. However, the steepening of the thermocline and halocline seen in the model between 5°N and the coast, where upwelling is particularly strong, is not seen in WOA, probably due to a poor density of observations near the coast.

Figures 4A, B show the summer SST climatologies of the REF and noRIV simulations, respectively. Differences between these climatologies show a clear warming effect induced by the Niger River, with a maximum of around 0.5°C between 2°E and 4°E near the coast ( Figure 4C ). This warming confirms the results of Alory et al. (2021), although it was stronger and with a larger extension in their lower resolution simulation (1/12°).

In the following, we investigate the heat budget in the box bordered by 2°E - 4°E; 5.5°N – 7°N ( Figure 4C ), where the warming reaches its maximum.

Processes influencing the temperature of the mixed layer can be classified into three categories: horizontal processes (zonal and meridional advection, lateral diffusion), vertical processes

(vertical advection, vertical diffusion and entrainment) and atmospheric processes (solar and nonsolar fluxes). Figure 5A shows their relative contribution to the total temperature tendency of the REF simulation mixed layer. Atmospheric and vertical processes are the main contributors to the mixed layer temperature trend.

Atmospheric processes warm the upper ocean all year long except in June, with a maximum contribution larger than 0.1°C.day^-1 in August. Vertical processes have always a cooling effect, peaking below -0.1°C.day^-1 in August too. Horizontal processes have a relatively small contribution, with absolute values always lower than 0.04°C.day^-1. The maximum cooling in June-July is mainly due to vertical processes as expected in a coastal upwelling region, but atmospheric processes contribute (in June) and also horizontal processes.

Figure 5B presents the REF-noRIV difference of the simulations with and without the Niger River. It shows that the river affects the 3 categories of processes. The river attenuates the cooling associated with horizontal and vertical processes, leading to warming. It also reduces the warming associated with atmospheric processes, which leads to cooling.

The warming prevails over the cooling most of the year. Thus, the net effect of the Niger River is to warm coastal waters, exclusively through oceanic processes (horizontal and vertical). The decomposition of each category of oceanic processes, into its different components, will allow to identify the terms of the heat balance through which the Niger River warms the coastal waters.

a. Horizontal processes

The contributions of zonal advection (XAD), meridional advection (YAD) and lateral diffusion (LDF) to horizontal processes (HORIZ) in the REF simulation are shown in Figure 5C . With minima below -0.1°C.day^-1 reached in July-August, zonal advection is the main horizontal process influencing the temperature of the mixed layer, partly compensated by meridional advection, which reaches 0.075°C.day^-1. Lateral diffusion has a weak contribution throughout the year. The REF-noRIV difference in horizontal processes ( Figure 5D ) reveals that meridional advection is the only horizontal process through which the Niger River warms the coastal waters. The seasonal signal of meridional advection shows two peaks, with the highest peak during the upwelling season in August (~0. 027°C.day^-1).

b. vertical processes

Figure 5E shows the contributions of vertical advection (ZAD), vertical diffusion (ZDF) and entrainment (RES), to vertical processes (VERTIC) in the REF simulation. Vertical diffusion largely dominates the vertical processes, with vertical advection and entrainment relatively negligible. The REF-noRIV difference ( Figure 5F ), shows that river-induced changes in vertical processes are also almost entirely due to changes in vertical diffusion. The river thus contributes to mixed layer warming by limiting the cooling induced by vertical diffusion.

As vertical diffusion and meridional advection are the heat budget terms through which the Niger River warms the coastal waters, we proceed to examine in more details these warming terms.

a. Vertical diffusion

Vertical diffusion is modulated by two mechanisms with opposite effects: stratification of the water column and vertical shear of horizontal currents.

Stratification isolates the surface waters, which are less salty and warmer, from the colder and saltier subsurface waters. On the contrary, vertical shear of the horizontal currents favours the upwelling of the subsurface water towards the mixed layer. To study the respective effects of these two mechanisms on vertical diffusion, we use the parameters N 2 (T), N 2 (S) and S h 2 detailed in Section 2, calculated for each of the two simulations. The impact of the Niger River on vertical diffusion is evaluated from the REF-noRIV difference in the three parameters mentioned above. Figures 6A, B show the temperature stratification in the water column over the upper 100 m in the REF simulation and the REF-noRIV difference.

In REF, the strongest temperature stratification is observed between March and May, with values reaching 7*10^-4 S^-2 around a depth of 30 m. The positive stratification values in the REF-noRIV difference map show that stratification is stronger with the river at this depth. The river enhances by almost 60% the maximum temperature stratification between 20 m and 40 m depth.

Indeed, in the absence of a river, only solar radiation warms the waters of the mixed layer. In REF, in addition to the solar radiation, the stratification is reinforced by the freshwater input from the river plume. It also causes the mixed layer and thermocline to rise by a few metres as observed in Figure 6B .

The vertical stratification of salinity parameter N²(S), calculated in the two simulations, reveals that the salinity stratification is almost entirely due to the river’s freshwater inputs. Indeed, the similarity of Figures 6C, D indicates that salinity stratification is very weak in the noRIV simulation. The salinity at the surface is almost the same at 100 m. But in the presence of the Niger River (REF), the water column becomes very stratified in the upper 20 m. The river contributes to the maximum salinity stratification about 91%. As the salinity stratification caused by the river is much higher than temperature stratification ( Figures 6B, D ), we conclude that the strong salinity gradient induced by the river’s freshwater plume is the main cause of NGG sea water’s stratification.

We have also diagnosed the influence of the Niger River on the vertical shear of the horizontal currents ( Figures 7A, B ). The shear remains almost the same in the mixed layer of both simulations, except between August and October, where it is reinforced by about 40% ( Figure 7B ). In the REF simulation ( Figure 7A ), no values are observed at the bottom of the mixed layer. On the other hand, negative values are observed at the bottom of the REF-noRIV mixed layer ( Figure 7B ). These values indicate the entrainment of subsurface water in the mixed layer of the noRIV simulation. The stratification induced by the fresh river water thus prevents the entrainment of the subsurface water into the mixed layer.

b. Meridional advection

The previous sections showed that meridional advection is one of the processes explaining the warming induced by the Niger River. In this section, we investigate the potential causes of its variation, namely meridional current and meridional temperature gradient ( Figure 8 ).

The computation of the correlation coefficients shows that the changes in the meridional advection are strongly linked to the variation of the meridional current. Indeed, meridional current is the main driver of the variation in meridional advection (correlation r=0.61 between

the two-time series), with a weak contribution from the meridional temperature gradient (r=-0.10). Thus, the Niger River contributes to the advection of warm waters towards the coast by reinforcing the meridional current.

Regarding the changes in the meridional current, they are related to the variations of the two components ( Figure 9 ). Between January and March and September to October, the geostrophic component dominates the changes in the meridian current while the Ekman component takes over from March to August and October to December.