Seasonal cycle of the salinity barrier layer revealed in the northeastern Gulf of Guinea

The region located in the far northeast of the Gulf of Guinea (NEGG), eastern tropical Atlantic, remains poorly documented due to a lack of available in situ ocean data. Heavy rainfall and intense river discharges observed in this region induce a strong salinity stratification that may have a significant impact on the mixed layer depth and on sea surface temperatures, through the so-called barrier-layer effect. By using recent in situ data and climatological outputs from a numerical simulation, we reveal the existence of a barrier layer in the NEGG and describe its seasonal occurrence. In the NEGG, the barrier layer limits the mixed layer depth. From January to March, significant values for the barrier-layer thickness are observed mostly due to the horizontal advection of fresh water. From April, vertical mixing along with vertical advection increase the sea surface salinity; hence, the barrier-layer thickness decreases and reaches its minimum in July. During the rest of the year, values for the barrier-layer thickness are again high, mostly under the influence of the Niger River discharge and precipitation, with the highest values recorded in October, when the river discharge and precipitation are at a maximum.

In a classically stratified ocean, the surface layer is virtually homogeneous in temperature, salinity and density. Thus, the layer that is homogeneous in density (also called the mixed layer) is determined by the pycnocline, which is at the same depth as both the halocline and the thermocline. However, in some parts of the ocean, particularly in tropical oceans where strong salinity gradients may be present, a salt stratification may be observed within the isothermal layer. When the halocline and the thermocline no longer coincide, the mixed layer depth is then determined by the halocline. The barrier layer (BL) is a layer that is homogeneous in temperature but stratified in density, and is located between the top of the thermocline and the base of the mixed layer (Godfrey and Lindstrom 1989;Sprintall and Tomczak 1992;de Boyer Montégut et al. 2007;Mignot et al. 2007).
The BL is a physical phenomenon observed in the three tropical oceans. De Boyer  have shown that the BL is almost permanent in tropical/ subtropical regions, such as the western tropical Atlantic and western Pacific oceans, the Bay of Bengal, the equatorial eastern Indian Ocean and subtropical basins, as well as at high latitudes, such as the Labrador Sea and in parts of the Arctic and Southern oceans. Those authors also showed that the BL appears seasonally in northern subtropical basins, the southern Indian Ocean and the Arabian Sea. Strong precipitation and river runoff have been shown to contribute to BL formation in the Bay of Bengal (Shetye et al. 1996;Varkey et al. 1996;Rao and Sivakuma 2003). Ando and McPhaden (1997) have shown that the main mechanism for BL formation in the western part of the tropical Pacific basin is the strong precipitation observed in that region. The contribution of strong precipitation to BL formation was also confirmed in the western tropical Atlantic basin (Tanguy et al. 2010).
Salinity, through the BL process, can modify heat exchange between the mixed layer and the interior ocean. The BL can reduce cooling through the entrainment of

Introduction
Open Access article distributed in terms of the Creative Commons Attribution License [CC BY 4.0] (http://creativecommons.org/licenses/by/4.0) surface water. Salt stratification limits the depth of the mixed layer and therefore the amount of solar flux entering the lower ocean layer (Lukas and Lindström 1991;Maes et al. 2002). Thus, positive sea surface temperature (SST) anomalies can persist for a long time-period in the presence of a BL (e.g. Maes et al. 2002). The BL can also limit nutrient transfer into the euphotic zone, with limiting effects on biological productivity (Prasanna Kumar et al. 2004).
The northeastern Gulf of Guinea (NEGG), eastern tropical Atlantic (1°-5° N, 5°-10° E), remains very poorly documented due to a lack of in situ ocean data for the region, where strong precipitation and important river runoff (mostly from the Niger River) are observed. According to Dai and Trenberth (2002), the Niger River's discharge (annual mean 7 × 10 3 m 3 s -1 ) is the 12th-largest in the world. These freshwater inputs could be expected to lead to a salinity stratification that is likely to generate a BL, similar to that observed off the Amazon River mouth in the western part of the Atlantic basin (Sprintall and Tomczak 1992;Pailler et al. 1999). In the NEGG region, strong precipitation also plays an important role in sea surface salinity (SSS) variability Da-Allada et al. 2014a;Camara et al. 2015).
By analysing in situ measurements obtained during the French oceanographic cruise EQUALANT, carried out in the boreal summer of 2000, Guiavarc'h (2003) provided evidence of a BL in the NEGG, especially around 3° N, 6° E. More recently, by using the climatology of de Boyer Montégut et al. (2007), Breugem et al. (2008) suggested that the BL does not exist in the NEGG in either the boreal winter (December to February) or boreal summer (June to August). However, very few temperature/salinity data were available for the region (see Figure 2 of de Boyer Montégut et al. 2007). Nonetheless, recent in situ data are now available, which, combined with the results of a high-resolution numerical model, allowed a focused study to be conducted in this region. The aim of the present study was thus to verify and confirm the existence of a BL in the NEGG, describe its seasonal cycle, and discuss its potential causes.

In-situ data
Salinity and temperature data were obtained during the French EQUALANT-2000 and EGEE oceanographic cruises. The EQUALANT-2000 cruise was carried out as a French contribution to the international CLIVAR (CLImate and Ocean -VARiability, Predictability and Change) project (see http:// www.clivar.org). Based on three meridional sections carried out in the Gulf of Guinea (along 10° W, 0° E and 6° E), this cruise was implemented on board the RV Thalassa, from 25 July to 20 August in 2000 (for details see Bourlès et al. 2002). We used data that were collected at 6° E between 0.33° N and 3.5° N. CTD profiles were carried out every 0°30ʹ in latitude (every 0°20ʹ in the equatorial band between 1° S and 1° N) along all three sections, from 12 to 17 August in 2000. The vertical resolution of the CTD data was 1 m.
In the framework of the program 'Etude de la circulation océanique et du climat dans le Golfe de Guinée' (EGEE) (Bourlès et al. 2007), the French oceanographic component of the African Monsoon Multidisciplinary Analysis (AMMA) program (Redelsperger et al. 2006), six oceanographic surveys were carried out in the Gulf of Guinea from June 2005 to September 2007. We used CTD data that were collected along 6° E every 0°30ʹ of latitude (every 0°20ʹ in the equatorial band between 1° S and 1° N), on 7 and 8 June 2007 during the fifth EGEE cruise (EGEE5) and from 3 to 5 September 2007 during the sixth EGEE cruise (EGEE6). The vertical resolution of the CTD data was 1 m (for details, see Kolodziejczyk et al. 2014).
To investigate the potential causes of the BL, we also used freshwater-input data, namely for precipitation and Niger River runoff. We used precipitation data from ERA-Interim (ERAI) (Dee et al. 2011) and from the Global Precipitation Climatology Project (GPCP) (Adler et al. 2003). Both datasets were available monthly from 1979 to present, with a resolution of 0.75° for ERAI and 2.5° for GPCP. Niger River runoff was determined from altimetry, using the method developed by Papa et al. (2010).
Sea surface salinity (SSS) data were obtained from the SMOS satellite and from the In-Situ Analysis System (ISAS) (Gaillard et al. 2016). The SSS product derived from the SMOS data was L3_DEBIAS_LOCEAN_V3, with improved adjustment of land-sea biases close to the coast (Boutin et al. 2018) and thus suitable for studies close to river plumes. This product was distributed by the Ocean Salinity Expertise Center (CECOS) of the CNES-IFREMER Centre Aval de Traitement des Données SMOS (CATDS) at IFREMER, Plouzané, France. The datasets were 9-day composites at a spatial resolution of 0.25° × 0.25° and were available for the period 2010-2017 (https://www.catds.fr/Products/Available-products-from-CEC-OS/CEC-Locean-L3-Debiased-v3). The ISAS product consisted of climatological gridded fields of temperature and salinity based mainly on autonomous Argo profiling floats and CTD profiles and it was available for the whole period 2002-2016, with a spatial resolution of 0.5° × 0.5° (https://archimer.ifremer.fr/doc/00309/42030/).
The NOAA_OI_SST_V2 dataset, also called the Reynolds product, provided by the NOAA/OAR/ESRL PSD, Boulder, Colorado, USA (https://www.esrl.noaa.gov/psd/), was also used to compare the model seasonal cycle of SST with observations. It was produced on a grid of 0.25° × 0.25°, and was available daily from 1981 to present.

ROMS framework
The numerical simulation results employed were from the Regional Ocean Modeling System (ROMS) framework (Shchepetkin and McWilliams 2005) with a configuration specially adapted to the Gulf of Guinea (Herbert et al. 2016). This model solves the three-dimensional primitive equation of Navier-Stokes following the Boussinesq and hydrostatic approximations. The version of the model used allowed grid refinement (AGRIF code), as follows: a zoom called a 'child' simulation, with a fine mesh (resolution of 1/15°), was nested in a wider domain, called a 'parent' simulation, that was at a lower resolution (1/5°). The wider domain extended from 60° W to 15.3° E, and from 17° S to 8° N, and the finer domain was from 10° W to 14.1° E, and from 17° S to 6° N. The vertical coordinate was discretised into 45 sigma levels, with the vertical S-coordinate surface and bottomstretching parameters set, respectively, to q s = 6 and q b = 0, to keep sufficient resolution near the surface (Haidvogel and Beckmann 1999). In the first 150 m, the model had 23 sigma levels. The vertical S-coordinate H c parameter, which gives the approximate transition depth between the horizontal surface levels and the bottom terrain-following levels, was set to H c = 10 m (Herbert et al. 2016). The Global Earth Bathymetric Chart of the Oceans (GEBCO) 1-resolution dataset was used for the topography (www.gebco.net). The parent simulation was forced at its borders by monthly fields of salinity and temperature, provided by World Ocean Atlas 2009 (WOA09) climatological fields with a spatial resolution of 1° × 1°. Atmospheric forcing on the sea surface (heat, freshwater and wind stress) was provided by Comprehensive Ocean-Atmosphere Data Set (COADS) climatology with a spatial resolution of 1° × 1°. The flows of the major rivers (Amazon, Congo, Niger, Ogooué, Sanaga and Volta rivers) were prescribed monthly from the climatology of Dai and Trenberth (2002). In addition, the model used surfacerestoring for salinity.
The simulation was integrated for 15 climatological years, with outputs averaged every 2 days. A statistical equilibrium was reached after 6 years of spin-up. Hence, the model analyses were based on the monthly model outputs averaged from year 7 to year 15.

Criteria for determining the barrier-layer thickness (BLT)
Our computation of the barrier-layer thickness (BLT) was based on the following formula: BLT = ILD -MLD, where ILD is the isothermal layer depth, and MLD is the mixed layer depth. We defined the ILD as the depth where the temperature was 0.5 °C below the SST at the reference depth of 10 m (Monterey and Levitus 1997;Spall et al. 2000;Foltz et al. 2003;de Boyer Montégut et al. 2004;Da-Allada et al. 2015). The MLD was determined using the density (σ) criterion ∆σ = 0.03 kg m -3 (de Boyer Montégut et al. 2004;Da-Allada et al. 2013); this corresponds to the depth where the density was equal to σ + ∆σ, where σ is the density at the reference depth of 10 m. The reference depth was chosen at 10 m in order to avoid the strong diurnal cycle that occurs in the first few meters of the ocean (de Boyer Montégut et al. 2004;Da Allada et al. 2015). These criteria for ILD and MLD allowed improved capture of the vertical salinity gradient. Da-Allada et al. (2014a, 2017 used the same criterion for MLD (∆σ = 0.03 kg m -3 ) to compute the MLD in the Gulf of Guinea.
We also performed sensitivity tests on the MLD and BLT computations. As their values depend on the choice of criteria, we computed different values of MLD and BLT based on different criteria of density, temperature and reference depth. The MLD and BLT monthly errors were estimated as the standard errors of the MLD and BLT computations (Appendix).

Model validation Mean states of sea surface salinity (SSS) and temperature (SST)
The spatial distributions of the annual mean SSS from the ROMS model and from the SMOS and ISAS products are shown in Figure 1. The model and both the SMOS and ISAS products showed very similar patterns, with high SSS (~36) found in the southwestern region, and low SSS (~31) found in the northeastern region of the Gulf of Guinea, near the African coast. This region is under the influence of the Niger River discharge and high precipitation, enhanced by the presence of Mount Cameroon. The model exhibited another region with a low SSS value in the vicinity of 5° S, close to the coast between 10° E and 12° E, attributable to the Congo River discharge. This region of low SSS was also detected by the SMOS product. In contrast, the ISAS product did not exhibit such low SSS values in this region. Also, the SMOS product captured the low SSS values in the northeastern region better than the ISAS. This difference may be explained by the low number of in situ data available to build the ISAS product ( Figure 2 in Gaillard et al. 2016).
The model reproduced the spatial distribution of the observed SST reasonably well (Figure 2). Both the model and the observations exhibited an SST maximum (>27 °C) north of the equator, from 5° W to the African coast. However, the model seemed to slightly underestimate the SST, as was visible from 5° W to 2° W along the equator. South of the equator, a weaker SST (<27 °C) was recorded from 5° W to 12° E, both by the model and from observations. These comparisons showed that the ROMS model produced a satisfactory simulation of spatial distributions of SSS and SST.

Seasonal variability in SSS and SST
Seasonal cycles of SSS and SST in the NEGG area (1°-5° N, 5°-10° E), deduced from the ROMS model and from the observations (SMOS and ISAS for SSS, and Reynolds for SST), are shown in Figure 3. Both SSS and SST exhibited seasonal variation in the NEGG. The SSS seasonality simulated by the model presented maximum values between July and September and minimum values between January and February. The model output represented an underestimate compared with the ISAS observations from December to September, with a bias of 2.4 in February, but an overestimate from September to November, with a bias of 0.2 in October. When compared with the SMOS observations, the model underestimated SSS from January to July (with a bias of 0.6 in April) and overestimated SSS during the rest of the year (with a bias of 1.8 in November). The lag of ~1 month between the model output and the maxima and minima in the observations might be due to uncertainties in freshwater forcing. The model reproduced the seasonality of observed SST reasonably well throughout the year. Maximum SST values (>27.5 °C) were found from January to April, and minimum values (<27.4 °C) from May to December. However, the model presented a cold bias (of −0.7 °C in April, and −0.9 °C in August) throughout the year.
Despite the biases, which were relatively small, the model reproduced the seasonality in SSS and SST adequately and was thus considered to be appropriate for the current study.   Hence, the ROMS model adequately reproduced the vertical distribution of salinity and temperature in June along 6° E from 0.5° N to 2.5° N. EGEE6 observations and model output at 6° E in September -Data from EGEE6 were available from 3 to 5 September 2007. The salinity vertical section (Figure 8a) showed an important fresh layer north of 2.5° N from the surface down to a depth of 20 m, with values as low as 32. This local freshwater layer was also present in the model output ( Figure 8b). However, the salinity values from the model were higher (~33.5). Also, both the model output and the observations suggested a weak decrease in salinity above ~30 m between 0° N and 1° N.
Both the EGEE6 observations and the model output exhibited a warm surface layer (>25 °C) north of 1° N, extending down to the thermocline (Figure 9). The model simulated a more diffuse and shallower thermocline (i.e. at ~50 m depth) than was reflected in the observations (at ~60 m depth). However, both model and observations exhibited the same thermocline shape, deeper at ~2° N and shallower close to the coast in the north. Hence, the model adequately reproduced the spatial (horizontal and vertical) distribution of SSS and SST in September along 6° E, and thus its use was considered appropriate to verify the existence of a BL and to analyse its spatial and temporal variations.

Evidence of a barrier layer in the northeastern Gulf of Guinea
Below, we compare the output of the ROMS model to observations that were available in August and September only.   The in situ profiles of EQUALANT-2000 exhibited a surface layer of 20-m thickness, homogeneous in salinity, temperature and density (Figure 10a). In the model, this homogeneous surface layer exhibited a thickness of 12 m (Figure 10b). The in situ observations showed two clear salinity vertical gradients, at 20 m and 40 m (Figure 10a). At 20 m depth, the observed temperature underwent a slight change from 26° to 25.5 °C. In the same layer, the density profile also presented clear gradients, at the same depths as the salinity gradients. According to the definition of BL (see above), it is apparent that the observations revealed a BL of 20-m thickness, at between 20 and 40 m depth (Figure 10a). In the model, the salinity profile also showed vertical gradients around 15 and 30 m, although less marked than in the EQUALANT-2000 observations. However, these gradients were also apparent in the temperature and density profiles, clearly indicating the presence of a BL of 15-m thickness between 15 and 30 m (Figure 10b). In addition, the monthly mean of the BLT obtained from the model was compared with that obtained from the ISAS at 3.25° N and 6° E in August. ISAS exhibited a BLT of 21 m (Figure 10c), a difference of 6 m when compared with the model.

September profiles
An equivalent comparison as for August, above, was made for September, but with observations from EGEE6 recorded on 3 September 2007. The EGEE6 data provided evidence of a BL of ~32-m thickness (Figure 11a). The monthly average data from the model for September suggested a BL of 14-m thickness (Figure 11b). The important BLT value recorded from the EGEE6 data might be associated with the desalinisation that occurred during the same month around the same position (Figure 8a). Furthermore, we compared the monthly mean of the BLT obtained from the model with that from the ISAS data at the position 3.5° N, 6° E in September. The BLT from the ISAS data was 10-m thicker than that simulated by the model (Figure 11c). The difference might be linked to the spatial distribution of the BLT in the model, as it might be extended westward during this period of the year.

Mean state and seasonal cycle of the BLT in the NEGG
Here we describe the BLT mean state and its seasonal spatial distribution as computed from the model monthly average output, with a particular focus on the NEGG region.

Annual mean of the BLT
The annual mean of the BLT computed from the model output and the annual mean of the precipitation are presented in Figure 12. The presence of a substantial BLT (>10 m) was evident in the area 1-6° N, 3-10° E, with thicknesses in the NEGG of up to 17 m at around 2° N, 7° E. In the same region the mean precipitation from the ERAI data was at a maximum, with values of up to ( (Figure 12b). In its mean state, the region where the BL was present corresponded to the region where low SSS values (see Figure 1) and significant precipitation were observed. Just as low values of SSS are related to precipitation and river runoff, so too are BLT values induced by precipitation and river runoff in this region.

Monthly variation in distribution and thickness of the BLT
To analyse the extent and seasonal variations of the BL in the Gulf of Guinea, monthly maps of the BLT, from January to December, over the region 6° N, 5° W-5° S, 12° E, were computed from the ROMS simulation output ( Figure 13). The extent of the BL in the Gulf of Guinea showed considerable variation. From January to April the BL extended over a large part of the Gulf of Guinea, and the highest values of the BLT (>20 m) were observed in the NEGG in February and March. The extent of the BL decreased from May until July and remained in the most northeastern region close to the coast. From August the extent of the BL increased again with the thickness at >22 m, until October when a robust BL extended over the whole NEGG, from around 5° W to 10° E at ~1° N. In October the BL was also observed along the coast south of the equator, with thickness values of ~10 m. In November   20  21  22  23  24  25  26  22  23  24  25  26  22  23  24  25  26   32  33  34  35  36  33  34  35  36  33.5  34  34.5  35  35.5  36   14  16  18  20  22  24  26  14  16  18  20  22  24  26  14  16  18  20  22  and December the BL still extended over a large part of the NEGG but with decreasing thickness values.

Seasonal cycle of the barrier layer and its impact on the mixed layer depth
Seasonal cycles of the SSS, the MLD (with estimated error) and the BLT (with estimated error) obtained from the ROMS model are shown in Figure 14. SSS, as previously described (Figure 3)

Discussion
To our knowledge, and probably due to the lack of in situ data for the region, previous studies have not pointed out the presence and seasonal occurrence of any BL in the NEGG. In this study, we have revealed the existence of the BL and described its seasonal cycle in the NEGG by using a ROMS model and available in situ observations. Although Guiavarc'h (2003), using in situ data, had suggested the existence of the BL, its formation had not been well documented due to insufficient data. We computed the BLT based on temperature and density criteria, and our computation was performed with a sensitivity test using the criteria 0.5 °C for the ILD and 0.03 kg m -3 for the MLD. These criteria had been shown in previous studies to be most suitable for capturing the vertical salinity gradient (Tanguy et al. 2010;Da Allada et al. 2014a, 2017. Using a ROMS simulation, we have shown that the annual mean BL is present in a large part of the Gulf of Guinea, extending from 5° S to 7° N and from 5° W to 12° E. In particular, the northeastern area (1°-6° N, 3°-10° E), bordered by the coast of West Africa, exhibits a robust BL pattern with maximal BLT values. These large BLT values in the NEGG, where the BL is observed throughout the year, are induced mainly by the contributions of river discharge and precipitation. Hence, we focused on the NEGG region.
We have also shown that the BL in the NEGG exhibits a seasonal pattern, with the extent of the BL being at a maximum during February and October and at a minimum in July. The BL seasonally affects the MLD by limiting its depth, and the BLT and the MLD vary in an opposite phase. Also, in August/September, when the BL is confined to the NEGG, the SST is at a maximum and is associated with minimum values of SSS. Figure 15 shows seasonal cycles of the BLT in the NEGG, the Niger River discharge and precipitation from two products, ERAI and GPCP, which exhibited similar cycles. Here, we diagnose the relative contributions of the Niger River discharge and precipitation to seasonality in the BL. Precipitation was highest in October (~10.5 mm day -1 from ERAI, and 9 mm day -1 from GPCP), and lowest in January/February. This region is under the influence of the Inter-Tropical Convergence Zone (ITCZ) which affects the seasonal variability of rain in the NEGG. The Niger River discharge was low from January to July before increasing to a maximum in October and then decreasing again. The coincident low precipitation and Niger River discharge from January to March indicate that these cannot be responsible for the high BLT levels recorded during these particular months; hence, it appears that other processes influence the BL during this period. Da-Allada et al. (2014b) found that a seasonal decrease in the salinity of the mixed layer in the NEGG was a consequence of both freshwater flux (precipitation plus runoff) and horizontal advection. Therefore, the horizontal advection of fresh water might be the mechanism inducing high BLT values at the beginning of the year. The BLT then decreased from March to a minimum of 4 m in July, which is a period during which high precipitation was recorded (maximum in May and June) before a decrease, but the Niger River discharge remained low. A recorded increase in SSS from May to August in the Bight of Biafra is due to vertical advection and vertical mixing Da-Allada et al. 2014b). Consequently, these vertical processes contribute to a decrease in the BLT by increasing salinity of the mixed layer. The BLT in the NEGG increased from August, to a maximum of 18 m in October, and then decreased to 10 m in December, during which period precipitation and the Niger River discharge followed a similar trend. Given the coincident maxima in October in the Niger River discharge, precipitation and the BLT, it can be deduced that the strong BL from August to December is directly induced by precipitation and the Niger River runoff.
It appears that the mechanism that induces the BL is the presence of low-salinity surface water in the region. A similar result was found by Pailler et al. (1999) in the western tropical Atlantic. Using in situ and high-resolution CTD data, they suggested that fresh water from the Amazon River may induce a strong halocline that generates a BL. Also, Cronin and McPhaden (2002) noted that a BL can be formed through rainfall which, in the absence of strong mixing and surface heating, can induce BL formation between the fresh lens and the top of the isothermal layer. Ocean currents might cause a BL to be advected horizontally from its region of formation (Cronin and McPhaden 2002). In the northern Gulf of Guinea, two major currents are present: the Guinea Current and the South Equatorial Current (Hisard 1975;Hisard and Merle 1979). The eastward-flowing Guinea Current is observed between the coast and the northern branch of the South Equatorial Current, between 2° N and 4° N, depending on season (Donguy and Privé 1964). Variations of these currents might be one of the mechanisms responsible for the seasonal extension of the BL into the Gulf of Guinea.
In this study, in agreement with the studies of Guiavarc'h (2003) but contrary to Breugem et al. (2008), we observed a BL in the NEGG with a well-marked seasonal cycle. The BL could not be observed by previous studies due to a lack of in situ surface and subsurface data for this region. We combined recently available in situ data for the NEGG with climatologic simulation outputs. The ROMS model used surface-restoring for salinity, and it is noted that this could affect the simulated salinity results (e.g. Large et al. 1997;Behrens et al. 2013). Although the model reproduced the vertical salinity distribution well, it would be interesting to revisit our results by using a numerical model without surface-restoring.