1 Introduction

The Indian Ocean is characterized by seasonally reversing monsoon winds north of 10° S, which force a seasonal turnover of the upper ocean circulation (Shankar et al. 2002; Schott et al. 2009). From May to September, the southwesterly winds force the surface Summer Monsoon Current (SMC) to flow eastward. In contrast, from November to February, the Winter Monsoon Current (WMC) flows westward north of the equator, in response to the northeasterly winds. Both monsoon currents (SMC and WMC) force the exchange of surface waters between the highly contrasted Arabian Sea (to the west) and Bay of Bengal (to the east). Specifically, the WMC carries fresher Bay of Bengal waters into the Arabian Sea, while the SMC brings saltier Arabian Sea water into the Bay of Bengal (Jensen 2003).

Salt transported by the currents plays a key role in the tropical climate through its effects on upper ocean stratification. Salinity can indeed limit the thickness of the mixed layer by creating a barrier layer and therefore constraint the ocean-atmosphere interactions (Lukas and Lindstrom 1991; Sprintall and Tomczak 1992; Pailler et al. 1999). The barrier layer has been shown to be important in the dynamics of the Indian Ocean using a regional coupled model (Masson et al. 2005; Seo et al. 2009), an ocean general circulation model (Durand et al. 2007), but also based on observations (Vinayachandran et al. 2002; Rao and Sivakumar 2003; Mignot et al. 2007). Furthermore, salinity is thought to be a possible indicator of changes in the hydrological cycle (Webster 1994; Yu 2011; Terray et al. 2012; Da-Allada et al. 2014a).

Previous studies have investigated the observed seasonal variability of sea surface salinity (SSS) in the tropical Indian Ocean with the available datasets. Based on measurements of SSS collected from ships of opportunity along six shipping tracks, large seasonal SSS variability is found over the western Indian Ocean, while the eastern part of the basin shows little variability, except for regions of intense local rainfall and river runoff like the Bay of Bengal (Donguy and Meyers 1996). The analysis of climatological salinity data (Rao and Sivakumar 2003) extended the previous study and revealed that the northern Indian Ocean exhibits a larger SSS variability than the southern Indian Ocean.

The north Indian Ocean mixed-layer salinity (MLS) budget based on the same SSS data together with other measurements (atmospheric flux and currents) quantifies the relative contribution of the atmospheric freshwater flux and SSS horizontal advection (Rao and Sivakumar 2003). According to these authors, the horizontal advection and freshwater flux are both important to explain the SSS variability during the summer monsoon, while horizontal advection clearly dominates the salt budget during the winter monsoon.

However, this study was limited by data availability and did not investigate all the processes, such as vertical advection, entrainment at the mixed-layer base, and diffusion terms. These terms may significantly contribute to the seasonal SSS balance, as was shown in other basins of the Pacific and the Atlantic oceans (Vialard and Delecluse 1998; Vialard et al. 2002; Qu et al. 2011; Vinogradova and Ponte 2013; Hasson et al. 2013; Kolodziejczyk and Gaillard 2013; Da-Allada et al. 2013, 2014b).

To better quantify the contribution of each term of the salinity balance in the Indian Ocean, a coupled ocean-atmosphere model with no relaxation to SSS observations was used (Vinayachandran and Nanjundiah 2009). This study showed that the freshwater input to the ocean and its redistribution by ocean circulation are the crucial processes to the salt budget. These authors also showed like in the observations that the SSS tendency is mainly due to horizontal advection during winter, whereas both advection and freshwater fluxes are necessary to explain the SSS tendency during the summer. However, the model SSS was found to be underestimated (1–1.5 pss bias) in comparison with the observed climatology. This was attributed to the poor representation of orography in the atmospheric component of the coupled model.

Meridional advection was found to be the mechanism which controls the seasonal cycle of the MLS in the southwestern tropical Indian Ocean by Halkides and Lee (2011). They used an ocean general circulation model (OGCM) with a relaxation term toward the observed SSS climatology to compensate for errors in the forcing or in the model physics. In this region characterized by a pronounced thermocline ridge, in the heat budget based on observations, Foltz et al. (2010) found that the seasonal cycle of sea surface temperature (SST) is driven by a combination of the net surface heat flux, horizontal heat advection, and vertical turbulent mixing. Note that a similar result was found in the heat budget using an OGCM in the western Arabian Sea by de Boyer Montégut et al. (2007). In the eastern part of the Arabian Sea (with a meridional separation at 65° E), they concluded that the SST seasonal cycle is dominated by surface forcing, while the oceanic processes play a second role. They showed that salinity stratification plays a significant role in maintaining the high winter SST in the eastern part of the Arabian Sea. In the SSS case, as the relaxation term used in Halkides and Lee (2011) also could be sometimes significant in some regions, the conclusions about the mechanisms of SSS variability may not be fully closed.

The recent development of the Argo array in the Indian Ocean has improved the coverage and resolution of available temperature and salinity profiles. In the present study, we use a combination of in situ and satellite products to provide more insight into the main mechanisms that modulate the MLS seasonal evolution in the tropical Indian Ocean. In particular, we reveal here the role of the freshwater flux due to precipitation which controls the seasonal evolution of the MLS in the southwestern tropical Indian Ocean and was missed by previous studies.

This paper is organized as follows. Description of the methodology and data is given in Section 2. Section 3 presents the results, and finally, discussion and conclusion of the new results are presented in Section 4.

2 Methodology and data

2.1 Methodology

Following Da-Allada et al. (2013), the mixed-layer salinity evolution equation (Eq. 1) can be written as follows:

$$ \frac{\partial {S}_{\mathrm{m}}}{\partial t}=\underset{\mathrm{FWF}}{\underbrace{\frac{\left(E-P\right){S}_{\mathrm{m}}}{h_{\mathrm{m}}}}}\kern0.5em \underset{\mathrm{HADV}}{\underbrace{-\overrightarrow{\mathbf{u}.}\overrightarrow{\nabla}{S}_{\mathrm{m}}}}\kern0.5em \underset{\mathrm{ENT}}{\underbrace{-H\left({w}_{\mathrm{e}}\right)\frac{\left({S}_{\mathrm{m}}-{S}_{h_{\mathrm{m}}}\right)}{h_{\mathrm{m}}}}}+\underset{\mathrm{DIFH}}{\underbrace{K{\nabla}^2{S}_{\mathrm{m}}}}+\varepsilon $$
(1)

where S m is the salinity average in the mixed layer, t is the time, E is evaporation, P is precipitation, h m is the mixed-layer depth (MLD), u m is the horizontal surface velocity vector averaged over the mixed layer (having u and v components defined positive eastward and northward, respectively), \( {S}_{h_{\mathrm{m}}} \) is the salinity at the base of the mixed layer, \( {w}_{\mathrm{e}}=w+\frac{\partial {h}_{\mathrm{m}}}{\partial t} \) is the entrainment velocity (at depth z = −h m) which corresponds to the difference between the vertical velocity w (positive when upwards and estimated from the horizontal currents through the continuity equation) at the mixed-layer base and the mixed-layer deepening rate, H (w e) is the Heaviside step function (H (w e) = w e if w e > 0 and H (w e) = 0 if w e < 0), and K (set to 500 m2 s−1 as in Yu 2011 and in Dong et al. 2009) is the horizontal diffusivity. Note that the river runoff is not quantified in this study because the studied areas are far from coastal regions. The left-hand side (lhs) term of Eq. 1 represents the MLS tendency, and the right-hand side (rhs) terms represent the surface freshwater flux (FWF), horizontal advection (HADV), vertical physics in the form of entrainment (ENT), horizontal diffusion (DIFH), and the sum of all unresolved physical processes (especially vertical turbulent mixing) and the accumulation of errors from the other terms calculated (ε), respectively. The horizontal advection is also decomposed into Ekman and geostrophic components. The Ekman velocity is calculated as \( {u}_{\mathrm{e}}=\frac{1}{\sigma_{\mathrm{o}}f{h}_{\mathrm{m}}}\left({\tau}^y-{\tau}^x\right) \) where τ x is the zonal wind stress, τ y is the meridional wind stress, f is the Coriolis parameter, and σ o (set to 1027 kg m−3, Li et al. 2013) is the reference density of seawater. The geostrophic velocity is deduced from the difference between total horizontal velocity and Ekman velocity. The ENT term contains subsurface vertical processes occurring at the base of the mixed layer: vertical advection from below the mixed layer (related to the effects of thermocline heaving) and entrainment mixing (when the mixed layer deepens). In addition, the entrainment terms may be sensitive to accuracy of the horizontal currents, salinity gradient at the base of the mixed layer, and the MLD (Kolodziejczyk and Gaillard 2013; Da-Allada et al. 2013). The vertical turbulent mixing (associated with small-scale turbulent Reynolds terms) cannot be resolved by the dataset because it results from vertical spatiotemporal variability (Reynolds terms) occurring at smaller scale than the vertical resolution of the data sampling. We neglected this term as we are not able to evaluate and its contribution thus appears in the residual and contributes to the MLS budget imbalance. The lhs of Eq. 1 (MLS tendency) can be obtained either from observations or computed as the sum of the rhs (diagnostic). In this study, the lhs term of Eq. 1 is referred to as the MLS tendency and the sum of forcing terms in the rhs of Eq. 1 as the diagnosed MLS tendency.

In order to compute the contribution of the different terms of the mixed-layer salinity balance to the salinity tendency, the following variables are needed: mixed-layer and subsurface salinity, mixed-layer depth, freshwater flux, wind stress, and the surface currents. The vertical velocity at the mixed-layer base is estimated from horizontal currents using the continuity equation.

2.2 Data

The mixed-layer and subsurface salinity are provided by the In Situ Data Analysis System (ISAS, Gaillard et al. 2009) monthly gridded fields of temperature and salinity optimally interpolated mainly from Argo profiles and a few other CTD observations including marine mammals and moorings (Gaillard et al. 2009). The ISAS product is provided with the monthly errors in salinity and temperature associated with each grid point. This error is given as a percentage of a priori variance at each point which depends mainly on the sampling. The annual mean error on MLS (Fig. 1a) is less than 70 % in the open ocean and could reach 100 % near shore due to poor data coverage in these areas. This study focuses on two regions where error on MLS is less than 70 % to compute the salt budget: the South-central Arabian Sea (SCAS, 1.5°–13° N, 56°–70° E) and the South-western tropical Indian Ocean (STIO, 5°–12° S, 50°–75° E). The number of Argo profiles on each box for each month between 2004 and 2012 is shown in Fig. 2. In the two boxes, we have about 40 profiles per month on average with a maximum of 120 profiles in mid-2004 in the SCAS and about 90 profiles in early 2007 in the STIO. The choice of these two regions is not only based on the MLS error but also on the MLS variability (see below). The ISAS fields are computed on a grid with a 0.5° × 0.5° horizontal resolution and 152 depth levels. The vertical spacing is 5 m in the first 100 m and then 10 m down to 200 m. We used the ISAS13 monthly climatology which is an average over the period 2004–2012.

Fig. 1
figure 1

Annual mean for a error on mixed-layer salinity (MLS) in percentage of explained variance, b mixed-layer salinity, c evaporation (OAFlux-E) minus precipitation (GPCP-P), and d seasonal standard deviation of MLS. Boxes shown in a and d indicate the areas of this study. Annual mean for each product is calculated from monthly averaged values spanning the 2004–2012 period. Units are practical salinity scale for b and d; and millimeters per day for c

Full size image
Fig. 2
figure 2

Monthly distribution of the available Argo profiles in the SCAS and the STIO regions between 2004 and 2012

Full size image

The MLD is derived from the monthly temperature and salinity fields of this climatology using density criteria which have been used by several authors (e.g., de Boyer Montégut et al. 2004; Dong et al. 2009; Yu 2011). Based on the sensitivity tests discussed in the Appendix, we find that the criterion of a 0.125-kg-m−3 density change relative to the near surface value (selected at 10 m because of better data sampling than 0 or 5 m) is more appropriated to close the MLS budget in our studied area; thus, it is used for the reference experiment. The MLD obtained (with ISAS13) is compared with the product of de Boyer Montégut et al. (2004) in the result part. The subsurface salinity \( {S}_{h_{\mathrm{m}}} \) is chosen as the salinity 15 m below the MLD (as in Ren et al. 2011). This choice is motivated by the sensitivity tests described in the Appendix. We find that the MLS budget is sensitive to the choice of the salinity at the mixed-layer base and to the density criteria used to compute the MLD. The different results obtained with the sensitivity tests are summarized in the Taylor diagrams (Taylor 2001) which provide statistical parameters (standard deviation, root mean square difference (RMSD), and correlation coefficient, Fig. 10 in the Appendix).

The net freshwater fluxes are obtained from atmospheric reanalysis and satellite datasets. Three datasets available over the 2004–2012 period are considered. For the reference experiment, evaporation is from the Objectively Analyzed air-sea Fluxes (OAFlux-E) dataset (Yu et al. 2008) which is a monthly value at 1° resolution, and monthly precipitation fields are from the Global Precipitation Climatology Project (GPCP-P) given with a resolution of 2.5° grid (Adler et al. 2003). The OAFlux-E and GPCP-P were linearly interpolated on the MLS grid for a consistent analysis across all fields. The two other products tested in this study are described in the Appendix.

The advection is estimated with two different products. We used the Ocean Surface Current Analysis Real-time (OSCAR) surface current of Bonjean and Lagerloef (2002) obtained from satellite sea level, wind stress, and SST, using a diagnostic model. The OSCAR product available on a 1/3° × 1/3° × 5 day is selected in this study using the sensitivity tests shown in the Appendix (Fig. 10). The second other current product tested is described in the Appendix. All the tests performed in this study allow us to estimate the error bars on diagnosed MLS tendency term. The wind stress obtained from ERA-Interim reanalysis is used to estimate the Ekman velocity.

All the data used in this study are taken from the same 2004–2012 period and linearly interpolated on the MLS grid for consistent analysis. Based on sensitivity tests, we used the OSCAR current, OAFLUX-GPCP freshwater flux, salinity at 15 m below the mixed layer, and the MLD based on density criteria 0.125 kg m−3 for the reference experiment with ISAS MLS to estimate the seasonal salt budget.

3 Results

The large spatial variations of the hydrological forcing in the various areas of the tropical Indian Ocean induce very distinct patterns in the annual mean of MLS as shown in Fig. 1b, c. Between 10° N and 10° S, regions of low MLS (around 34.5 pss) are observed under the Intertropical Convergence Zone (ITCZ) region. In the northern Arabian Sea (north of 10° N), a high value of MLS is observed (larger than 36 pss) as a result of strong evaporation. In contrast, in the northwestern coastal regions of the Bay of Bengal, the MLS is minimum due to strong river discharges into the ocean (Ganges, Brahmaputra, Mahanadi, and Godavari rivers). South of 10° S, low values of MLS are also observed although evaporation dominates in this region which indicates that ocean dynamics might be of importance. In particular, the westward-flowing South Equatorial Current (SEC) located south of 10° S (Fig. 3a, b) may explain these low salinities. Note also that the eastward-flowing SMC represented by July around 2°–6° N and the westward-flowing WMC represented by January located around 3°–6° N are visible in Fig. 3a, b.

Fig. 3
figure 3

Oscar surface currents (a, b) and ISAS (color) and de Boyer Montégut et al. (2004) (contours) product of the mixed-layer depth (c, d) for January and July. Unit is meters per second for a and b and meters for c and d. Contour intervals are 20 m

Full size image

MLS shows larger variability north of the equator than in the southern part (Fig. 1d) due to strong seasonality in the local hydrological forcing and strong advection of surface saline and freshwaters (Donguy and Meyers 1996; Rao and Sivakumar 2003). The largest MLS is observed near the eastern coast of India and Bangladesh.

As explained in the sensitivity test presented in the Appendix, the MLD is an important variable for the salinity balance. In boreal winter (represented by January), the MLD used for the computation is generally shallower than 60 m, except north of 10° N where MLD is larger (Fig. 3c). In boreal summer (represented by July), the largest MLD (>70 m) is mainly located south of 10° S and around 10° N (Fig. 3d). It should be noted that this MLD is on average greater (13 m in winter and 30 m in summer) than the widely used MLD product of de Boyer Montégut et al. (2004) which is based on a 0.03-kg-m−3 density criteria (not shown). The patterns of the differences between these two MLD have large space variability and can reach 20 m in winter and 40 m in summer. The differences found here are comparable to those observed in the tropical Pacific Ocean by Hasson et al. (2013) when they compared their MLD based on a 0.01-kg-m−3 density criterion with that of de Boyer Montégut et al. (2004). Since a better closure of the salt budget is obtained in the regions of the Indian Ocean that we are studying with the reference MLD (Fig. 10), this criterion was preferred to the one defined by de Boyer Montégut et al. (2004).

The two specific open ocean regions selected in this study (SCAS and STIO) are characterized by large seasonal MLS variability. Note that in the heat budget, the Arabian Sea was separated into two parts with a meridional separation at 65° E to illustrate the different mechanisms which control the SST balance in the eastern and western parts of the Arabian Sea (de Boyer Montégut et al. 2007). In this study, no significant change has been noted in the MLS budget by changing the SCAS box eastward limit from 65° E to 70° E. The STIO box was designed to follow Halkides and Lee (2011) to perform strict comparisons with our results. Extending eastward, the STIO box toward the eastern tip of the SSS maximum does not significantly change the results (not shown). MLS exhibits a well marked annual cycle in the two focused regions (Fig. 4). The STIO MLS is maximum (minimum) in September (March), 1 month before the maximum (minimum) in the SCAS. STIO MLS shows slightly lower amplitude (0.7 pss) than the SCAS MLS region (0.8 pss).

Fig. 4
figure 4

Seasonal cycle of MLS (in pss) averaged in the SCAS and STIO boxes

Full size image

In the SCAS region, the diagnosed MLS tendency from the sum of the rhs terms of the salinity balance equation (Eq. 1) matches the observed one within error bars (r = 0.97 at the 99 % significance level) (Fig. 5a). These two estimates reach their maximum in June and their minimum in January. The dominant terms of the salinity balance are the horizontal advection terms, mainly driven by the meridional component (Fig. 5b, c). This term shows a strong seasonal cycle compared to the other terms in the salinity balance. This term is negative from November to April with the maximum freshening effect in January caused by the maximum northward velocity in the presence of positive meridional MLS gradient (Fig. 6b). This is explained as follows. During the winter monsoon, the northeasterly trade winds are reinforced, driving strong northward Ekman transport over the SCAS (Beal et al. 2013). So, the relative freshwater in this region is transported northward into the SCAS and results in a strong decrease observed in meridional Ekman advection (Fig. 5c). During the rest of the year (May to October), meridional advection remains positive due to southward advection of haline surface water from the Arabian Sea. During the summer monsoon, the southwestward trade winds strengthen and lead to a strong southward Ekman transport (Beal et al. 2013) which carries the Arabian saline water from the north into the SCAS (Fig. 5c). The June peak of meridional advection is responsible for the annual peak in observed and diagnosed MLS tendencies. Zonal advection shows a semiannual cycle but has weaker amplitude than the meridional advection annual cycle. It is negative during the winter monsoon (when it reinforces the meridional component) because of the westward WMC which carries fresher Bay of Bengal water into the Arabian Sea (Fig. 6a) and during the summer monsoon due to the eastward SMC in the presence of positive zonal MLS gradient (then it moderates the meridional component). The maximum freshening effect in this term appears in February and it remains positive during the rest of the year. The seasonality of horizontal advection is thus mainly associated to the monsoon winds in this region (Yu 2011; Beal et al. 2013). The seasonal cycle of atmospheric freshwater flux is small except from January to April. During this period, it is positive due to an excess of evaporation (Fig. 6c) and reduces to 0.1 pss/month of the freshening effect of the horizontal advection. The entrainment term is also small and positive throughout the year and brings salty water from the interior into the mixed layer. It shows the strongest contribution (0.1 pss/month) in May due to the maximum entrainment velocity (Fig. 6d) and to the increase in vertical salinity gradient as a result of precipitation observed in May in this region (Fig. 6c). The thermocline shoaling appears to be too weak to enhance the vertical gradient and thus upward advection of salt (Fig. 7a), while the deepening of the mixed layer could contribute slightly to the entrainment term during the April–July period (Fig. 7a). The small difference between diagnosed and observed MLS could result from residual vertical mixing which is not properly resolved. Horizontal diffusion is negligible in this region.

Fig. 5
figure 5

For the SCAS region: a observed and diagnosed MLS tendencies with the shaded areas indicating error estimates (see Appendix) for these terms. b Individual contributions to the salt budget equation for horizontal (zonal + meridional) advection (HADV in blue), entrainment (ENT in dashed blue), freshwater flux (FWF in pink), and horizontal diffusion (DIFH in light blue). c Decomposition of the horizontal advection into Ekman (dashed green) and geostrophic components (dashed blue)

Full size image
Fig. 6
figure 6

For the SCAS region: latitude-time section for the SCAS region. a Zonal MLS gradient (pss m−1) and zonal current in contours (m s−1). Contour intervals are 0.15 m s−1. b Meridional MLS gradient (pss m−1) and meridional current in contours (m s−1). Contour intervals are 0.05 m s−1. c Difference between evaporation and precipitation (mm day−1). d Salinity gradient near the mixed-layer base (pss) and the positive entrainment velocity in contours (10−6 m s−1). Contour intervals are 2.5 m s−1

Full size image
Fig. 7
figure 7

Vertical profile of the seasonal salinity averaged over the two focal areas: a SCAS and b STIO. Solid black lines show the mixed-layer depth (MLD) and the 20 °C isotherm (a proxy for central thermocline depth)

Full size image

In the STIO, the seasonal cycle of the diagnosed MLS tendency correctly reproduces the observed MLS tendency (r = 0.97 at the 99 % significance level). However, the two estimates do not exactly match within error bars (Fig. 8a), except for the July–October period. In this region, the closure of the MLS budget appears to be very sensitive to ocean surface currents (Fig. 10). Using drifter currents instead of OSCAR, the observed and diagnosed MLS tendencies match within error bars (figure not shown). The main difference observed in the salt budget when using drifter currents manifests by a larger contribution of the entrainment term that reaches 0.07 pss/month instead of 0.04 pss/month with OSCAR. The diagnosed MLS tendency reaches its maximum in July, 1 month after the observed estimate. These two estimates are negative from September to March and positive during the rest of the year (April to August). In contrast with the SCAS region, during October to April, the freshwater flux and entrainment terms play an important role in MLS balance as zonal and meridional advection tend to compensate each other except in winter (Fig. 8b, c). During this period, the freshwater flux is dominated by precipitation (Fig. 9c) and contributes significantly to decrease the MLS and to increase the salinity vertical gradient at the mixed-layer base. The entrainment term, which contributes positively to MLS tendency, may result from the difference between the MLS and subsurface salinity because the entrainment velocity is weak (Fig. 9d). On the other hand, the thermocline vertical motion is rather weak to enhance the vertical gradient during this period (Fig. 7b). Thus, the salinity entrainment from the subsurface may respond to the heavy rainfall during the boreal winter and then slightly persist until May–June. Interestingly, the budget remains unclosed only during the period of significant entrainment term contribution (October–June) and of shallow MLD (Fig. 7). We know that the vertical mixing is not captured by the entrainment term. So, this residual includes at least the vertical mixing. During May to September, the MLS tendency is mainly driven by the meridional advection. Like in the SCAS, the southward advection of haline water from the north increases the MLS in the STIO (Fig. 9b). This term is positive in this period with maximum increasing effect in June due to the maximum southward current in the presence of positive meridional MLS gradient. Indeed, salinity increases northward and the meridional current is strong and southward in summer due to the monsoon forcing of Ekman transport (Fig. 8c). The zonal advection is slightly negative in this period due to westward current and negative zonal MLS gradient (Fig. 9a), and this term contributes to reduce the meridional advection. Freshwater flux is dominated by evaporation (Fig. 9c) during May to September and increases the MLS tendency. Note that although there is clearly the deepening of the mixed layer during May–September, its contribution to the entrainment term is weak except in May–June. As for the SCAS, our estimation of the horizontal diffusion is negligible.

Fig. 8
figure 8

Same as Fig. 5 except for the STIO region

Full size image
Fig. 9
figure 9

Same as Fig. 6 except for the STIO region. Note that here contour intervals are 0.08 m s−1 for a. Meridional current unit is 10−1 m s−1 and contour intervals are 0.25 m s−1 for b. Contour intervals are 1 m s−1 for d

Full size image

4 Discussion and conclusion

In this paper, we investigated the mechanisms that contribute to the seasonal cycle of the mixed-layer salinity budget in the tropical Indian Ocean using a combination of satellite products and in situ observations for the period 2004–2012. We focused this study on two particular regions characterized by important MLS variability: the SCAS (1.5°–13° N, 56°–70° E) and the STIO (5°–12° S, 50°–75° E). The seasonal cycle of the directly observed MLS tendency is well reproduced by the diagnosed MLS tendency in the two regions. It is especially the case in the SCAS where the processes which are dominating the MLS budget are better captured with the coverage, resolution, and quality of the available observations, as previously shown in the tropical Atlantic (Da-Allada et al. 2013).

In the SCAS region, in agreement with previous studies based on observations and models (e.g., Rao and Sivakumar 2003; Vinayachandran and Nanjundiah 2009), our results show that horizontal advection driven by the seasonally reversing monsoon winds plays a crucial role in the MLS seasonal cycle. We find that the contribution of this term is mainly explained by the meridional advection associated with the Ekman transport which is northward during the northeast monsoon and southward during the southwest monsoon. Atmospheric freshwater flux dominated by evaporation (except in May and in October) acts to increase the MLS. The contribution of this term in the salt budget is small in the second half of the year. The vertical entrainment term due to the increase in vertical salinity gradient as a result of freshwater flux and the deepening of the mixed layer contributes to slightly increase MLS mainly in boreal winter and spring. During the rest of the year (from July to December), the contribution of the entrainment term is very weak. Horizontal diffusion is negligible in this region.

In this study, we find that the entrainment term which was neglected by Rao and Sivakumar (2003), although small, significantly contributes to close the salt budget in the SCAS. This term shows the greatest contribution (0.1 pss/month) in the salinity balance in May. Contrary to Vinayachandran and Nanjundiah (2009) who found a significant negative contribution of freshwater flux in the salinity balance in November in the SCAS region (see their Fig. 6), we find here, on the basis of observations, that this term is nearly equal to zero at that time of the year. This likely explains why these authors found a lower SSS in the model than in observations (see their Fig. 1).

Several studies have shown the impact of the barrier layer on winter-spring variability of the southeastern Arabian Sea (6°–15° N, 68°–77° E; e.g., Masson et al. 2005; Durand et al. 2007). In the SCAS box, we observed that the vertical salinity gradient across the mixed-layer base is relatively weak and leads only to a small contribution in the entrainment term. Indeed, the chosen SCAS does not include the barrier layer located more eastward in the eastern tropical Indian Ocean.

In the STIO region, from October to April, the freshwater flux is mainly due to precipitation and dominates the MLS budget, although the vertical entrainment term linked to the increase in salinity stratification as a result of heavy rainfall and the deepening of the mixed layer is not negligible. In contrast with the SCAS region, precipitation plays a major role in the salt balance of the region. During the rest of the year (May to September), like in the SCAS, horizontal advection dominated by meridional advection drives a seasonal evolution of the MLS as freshwater flux and entrainment show a weak contribution.

In the STIO, our results, based on observations, differ from the Halkides and Lee (2011) model study. While they concluded that the seasonal cycle of MLS is dominated only by meridional advection, we find in the present study that the freshwater flux is an important contribution in this region; in particular, precipitation plays a major role in MLS budget during the boreal winter. It should be noted that the freshwater flux used in our study differs from the one they used. Their model was forced by the freshwater flux derived from the National Center for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) reanalysis product, so the precipitation minus evaporation could be the source of the difference. Based on sensitivity tests, we find that using the same reference experiment current (OSCAR), salinity at the mixed-layer based (S_h15) and density criteria (h_d0125) with NCEP/NCAR product, the RMSD is twice larger than that obtained with OAFLUX-GPCP in the salt budget. Halkides and Lee (2011) also considered a different time period: the SSS seasonal cycle is built on the 1993–2008 period which includes interannual events like the 1997–1998 EL Niño Southern Oscillation (ENSO; e.g., Vialard et al. 2002) and the 1997 Indian Ocean Dipole (IOD; e.g., Masson et al. 2004). The present study is based on the 2004–2012 period and includes also interannual events like 2009–2010 ENSO (e.g., Kim et al. 2011) and the 2010–2011 IOD (Durand et al. 2013). Note that it is a period with more La Niña than El Niño. Although we only investigated the seasonal cycle, the choice of the period could also be a source of bias of the seasonal cycle, and estimating the seasonal budget over a longer time period would reduce this potential bias. Indeed, during strong interannual events (ENSO, IOD), the contribution of each term of the salinity balance could be changed and therefore bias the mean seasonal cycle. As in the SCAS region, the entrainment term neglected by Rao and Sivakumar (2003) appears necessary to close the salt budget, and in November, the negative contribution of freshwater found by Vinayachandran and Nanjundiah (2009) is twice the size of our estimate.

As in the tropical Pacific (Hasson et al. 2013) and Atlantic (Da-Allada et al. 2013, 2014b) oceans, we find that all terms of the MLS equation have to be taken into account to close the salinity budget. As in the subtropical southeastern and northeastern Pacific (Ren and Riser 2009; Kolodziejczyk and Gaillard 2013) and in the equatorial Atlantic (Da-Allada et al. 2014b; Berger et al. 2014) which showed that vertical mixing plays a major role in the salinity balance, this term (neglected in this study) also appears to be important in this study and particularly in the STIO region.

The STIO region is characterized by a unique open ocean upwelling during boreal summer (Xie et al. 2002) and the SST in the STIO exerts an important influence on global climate through its impact on the Indian monsoon and Northern Hemisphere atmospheric circulation (Foltz et al. 2010; Schott et al. 2009). So, understanding the seasonal variability of upper ocean properties in this region is crucial for climate studies. New SSS measurements collected with soil moisture and ocean salinity (SMOS) are able to detect the signature of IOD (Durand et al. 2013) in this region. The use of the new SMOS SSS dataset or SSS from Aquarius satellite combined with in situ observations should improve significantly the resolution of the SSS seasonal cycle, especially by resolving the mesoscale variability (Hernandez et al. 2014; Kolodziejczyk et al. 2015) which was not possible with the current resolution of Argo array, and complement the SST seasonal cycle described by several studies (e.g., Foltz et al. 2010). A better understanding of the role of the different terms of the MLS budget as exposed in the present study will permit to evaluate and improve local and global ocean models and lead to increase their predictive skills.