1 INTRODUCTION

Salinity is a fundamental property of sea water. Sea surface salinity (SSS) is a variable closely associated with freshwater fluxes between the ocean and atmosphere and serves as an indicator of the global water cycle (Yu, 2011; Vinogradova and Ponte, 2013). In subtropical oceans, basin-scale horizontal SSS maxima are formed by excessive evaporation over precipitation (E-P > 0) (Qu et al., 2011; Bingham et al., 2014; Johnson et al., 2016). These high SSS waters subduct to the ventilated thermocline by wind-driven Ekman downwelling (O'Connor et al., 2005) and spread throughout the world ocean (e.g., Cannon, 1966; Tsuchiya, 1968). The subtropical SSS maxima in the Pacific and Atlantic Oceans have been extensively studied by oceanographers (Qu et al., 2011, 2013a, b; Li and Wang, 2012; Katsura et al., 2013; Bingham et al., 2014). In comparison, the counterpart in the Indian Ocean has not been comprehensively studied yet. Due to the unique land-sea distribution, subtropical SSS maximum in a "traditional" sense is formed only in the South Indian Ocean (SIO). The Arabian Sea SSS maximum is a tropical SSS maximum rather than a subtropical one, although it is also generated under excessive evaporation. The SSS maximum of the SIO plays a fundamental role in regional climate. First, due to the large evaporation rate, it serves as a major source of latent heat and water vapor for the atmosphere. Second, the subtropical high-SSS water is a key component of the shallow overturning circulation of the Indian Ocean (Schott et al., 2002; Lee, 2004). It spreads equatorward as a vertical salinity maximum in the thermocline (Warren, 1981; Wijffels et al., 2002; Katsumata and Fukasawa, 2011) and affects salinity stratification and mixed layer heat budget of the tropical Indian Ocean (e.g., Qu and Meyers, 2005; Li and Wang, 2015; Kido and Tozuka, 2017). With these regards, the subtropical SSS maximum in the SIO is worthy of comprehensive investigation to complement our knowledge of ocean and climate dynamics.

In-situ ocean salinity observations are relatively sparse in the SIO as compared to the Pacific, Atlantic, and North Indian Oceans, and as a result the investigation of the Indian Ocean salinity has been greatly hindered by the lack of observation. However, existing observational data indicate complicated spatial structure and prominent SSS variabilities on various timescales in the SIO (e.g., Qu and Meyers, 2005; Roemmich and Gilson, 2009; Durack and Wijffels, 2010; Du et al., 2015). The recent advent of satellite SSS measurements, such as the Soil Moisture Ocean Salinity (SMOS), Aquarius/SAC-D, and Soil Moisture Active Passive (SMAP) (e.g., Lagerloef et al., 2008, 2012; Entekhabi et al., 2010; Reul et al., 2012), provide unprecedented ability of monitoring SSS in both global and regional perspectives. Using satellite SSS data, many new knowledges of SSS variability over the Indian Ocean have been achieved (e.g., Guan et al., 2014; Menezes et al., 2014a; Vargas-Hernandez et al., 2015; Li et al., 2015, 2017), which effectively improved our understanding of salinity-related ocean dynamical processes.

The annual-mean SSS map based on Aquarius Version-4.0 Combined Active-Passive retrieval (CAP) Level 3 data (Yueh et al., 2013, 2014) for September 2011-May 2015 is shown in Fig. 1a. The structures of the high-SSS water in the subtropical SIO are dramatically different from that based on gridded Argo data of the same period (Fig. 1b). In Aquarius data, we see three disconnected SSS maxima in the western, central, and eastern basins with SSS > 35.6, whereas Argo data suggest a much smoother SSS distribution and one single large SSS maximum in the central basin between 70°E-100°E. It is possible that gridded datasets based on historical observational data fails to represent the western and eastern maxima due to insufficient data sampling and large-radius smoothing, which now have been well resolved and revealed by satellite SSS measurements. Another possibility is that Aquarius SSS may be erroneous in the coastal areas and present fake maxima. This issue should be addressed with care. Here we aim to provide a thorough investigation for the three regions of high SSS values in the SIO. We describe their spatial-temporal characteristics and examine their relationships with surface freshwater fluxes (E and P) and local ocean dynamics.

2 DATA

Gridded satellite observational data and Argo data of SSS are used in this study. The Aquarius satellite was launched on the SAC-D spacecraft in June 2011 (Lagerloef et al., 2008). It provides SSS along-track measurements with an average spatial resolution of ~130 km and a target retrieval accuracy of 0.2 on monthly timescale. In this study we mainly analyze the Version-4.0 Aquarius Combined ActivE-Passive (V4 CAP) Level 3 product (Yueh et al., 2013, 2014), which provides 1°×1°, monthly SSS data from September 2011 through May 2015. CAP algorithm retrieves SSS, wind speed, and wind direction from Aquarius' radar (active) and radiometer (passive) data (Yueh et al., 2014) using a cost-function minimization. The CAP product for Aquarius has been distributed through the Physical Oceanography Distributed Active Archive Center (PO.DAAC) of the National Aeronautics and Space Administration (NASA). We also employed gridded Argo data, which are constructed and distributed by the Asia Pacific Data-Research Center (APDRC) of the International Pacific Research Center (IPRC) (Lebedev et al., 2007). The IPRC Argo data provide 1°×1°, monthly ocean temperature and salinity data on standard depth levels. SSS of this dataset is based on in-situ salinity measurements by Argo float near the sea surface which is typically at ~5 m.

To validate SSS data of Aquarius and gridded Argo product, we also analyze Argo profile data from January 2005 through December 2017 from the Coriolis Global Data Acquisition Center of France. These data have undergone automatic preprocessing and quality control by the Argo science team (Owens and Wong, 2009). Obvious bad records, such as temperature falling outside the 0-35℃ temperature range or 28.0-38.0 salinity range, were excluded. Additionally, profiles with no valid data in the upper 6 dbar were excluded. There were 152 672 usable profiles in the SIO (25°E-125°E, 50°S-10°S) (Fig. 1c). In the three high SSS regions, there are on average 30-50 profiles available for each 1°×1° grid box. But in the western region, data is rather sparse, particularly near the South Africa coast. We also computed the profile distribution during the September 2011-May 2015 period (Fig. 1d), which is too sparse for analysis. Therefore, we will use the Argo data of 2005-2017 for comparison with Aquarius SSS. In this study SSS of each Argo data profile is defined as the mean salinity of 0-6 dbar.

Other ocean and atmosphere datasets analyzed in the study include the 1/3 ×1/3 ocean surface current data from Ocean Surface Current Analysis-Real time (OSCAR) product (Bonjean and Lagerloef, 2002; Johnson et al., 2007), the 1°×1° precipitation rate of the Global Precipitation Climatology Project (GPCP) version 1.2 (Huffman et al., 2001), the 1°×1°, monthly evaporation rate of the objectively analyzed air-sea fluxes (OAFlux) (Yu and Weller, 2007), the 0.25°×0.25° 10-m winds from the Cross-Calibrated Multiplatform (CCMP) ocean wind vector v2.0 (Atlas et al., 2008), and surface geostrophic current from the Archiving, Validation, and Interpretation of Satellite Oceanography (AVISO) (Le Traon et al., 1998; Ducet et al., 2000). All these data are resampled into monthly mean fields for the period of September 2011-May 2015 to match Aquarius SSS data.

3 SSS MAXIMA IN THE SOUTH INDIAN OCEAN

Among the three SSS maxima in the SIO (Fig. 1a), only the central maximum is recognizable in gridded Argo data (Fig. 1b). It exists between 60°E-95°E, at the latitudes of 35°S-25°S. The eastern (100°E-115°E, 38°S-28°S) and western (25°E-40°E, 38°S-20°S) maxima in Aquarius data are close to the African and Australian coasts, respectively. To confirm the existence of the two maxima, we compare Aquarius SSS and gridded-Argo SSS with Argo profile data within the high-SSS band of 35°S-28°S (Fig. 1e). Although subjected to large variability, Argo profile data do exhibit higher mean SSS in the eastern basin than in the central basin, which supports the existence of an eastern SSS maximum. The disconnection between the eastern and central maxima is likely due to large SSS variability around 90°E. Frequently occurring low SSS values there lower the mean SSS and lead to the separation of the central and eastern maxima. This issue will be discussed later. The SSS maximum to the west of 40°E is absent in Argo profiles. SSS values there are evidently lower than those in the central and eastern regions. Aquarius mean SSS even exceeds the upper bound of Argo data. For a further confirmation, we also examined the mean SSS map of SMOS Level-2 data (Reul et al., 2012) for the same period and did not see the western maximum neither. The SMOS SSS is close to Argo data in the western region (figure not shown). It seems that Aquarius SSS values in the western region are possibly overestimated due to technical reasons, and there may be no robust large-scale SSS maximum formed there in reality. Since none of the three datasets (Aquarius, Argo, and SMOS) is perfect, this issue is left open here.

In fact, both Aquarius and gridded Argo dataset show discrepancies to the bin-averaged Argo profile-based SSS. Distribution of gridded Argo dataset is too smooth and fails to represent detailed structures such as the eastern maximum, whereas Aquarius tends to overestimate SSS near the western and eastern boundaries possibly owing to land contamination (Tang et al., 2014). It is also noticeable that in the open ocean, Aquarius SSS tends to be lower than Argo salinity due to rainfall effect on "skin" salinity, and this makes the eastern SSS maximum even more eye-catching in Fig. 1a. We also have to claim that to some extent, the comparison made here is "apple-to-orange". Aquarius measures the salinity of the skin layer (several centimeters), while the SSS provided by Argo is typically at ~5 m. Data sampling and smoothing/interpolation procedures of the two datasets also cause large discrepancies between them. To obtain more reliable results, the combination of satellite and in-situ observation is necessary for our analysis. Although the western and eastern regions are possibly subjected to overrepresentation of Aquarius, the two regions are less investigated than the central SSS maximum region. Exploring the SSS variability of the two regions can complement to our knowledge of the SIO high-salinity water. Therefore, in the following, we will discuss SSS processes separately for the three regions.

Annual-mean distributions of precipitation and evaporation are shown in Fig. 2a and b. The three high-SSS regions are generally enveloped within the subtropical band of low precipitation (< 2 mm/d) and high evaporation (> 3 mm/d) rates in the SIO. Apparently, the excessive evaporation over precipitation (E-P > 0) in the subtropical SIO is essential for the formation of SSS maxima. However, a careful comparison suggests that the high-SSS values are in fact existing south of the evaporation maximum and low precipitation minimum. This indicates that in addition to surface freshwater flux, ocean dynamical processes may play an important role in determining the large-scale distribution of SSS, as the case for the SSS maximum in other ocean basins (e.g., Qu et al., 2011).

Surface wind distribution (Fig. 2c) suggests that the central and eastern SSS maxima are within the anti-cyclonic wind circulation, with easterly trade winds to the north and westerly winds to the south. Strong wind shears give rise to positive wind stress curls (WSCs) that induce upper-ocean convergence. Surface current data of OSCAR (Fig. 2d) confirms the convergence of meridional currents in this region, which is in favor of the accumulation of high-salinity water and formation of local SSS maxima. The eastward-flowing Subtropical Countercurrent in the basin interior (Siedler et al., 2006) and southward-flowing Leeuwin Current along the Australian coast (Feng et al., 2003) may also contribute to high-SSS water convergence for the eastern maximum. The western region is in the southwestern corner of the Indian Ocean basin and along the South Africa coast. It overlies with the high local evaporation rates, and the southwestward flowing Agulhas Current along South Africa coast (Beal and Bryden, 1999; Beal et al., 2011) also acts to transport high-SSS water from the high evaporation, low precipitation ocean interior areas. Therefore, it is likely that formation of the three high-SSS regions involve both freshwater flux forcing and ocean dynamics (such as ocean current advection). It is difficult to isolate effect of each factor from the annual-mean distribution, and instead we can investigate seasonal variations and examine the roles of freshwater flux and ocean dynamics in maintaining the high SSS levels.

Seasonal SSS variations from Aquarius and gridded Argo data are compared in Fig. 3. In Aquarius, the western and eastern SSS maxima emerge in the first half of the year (Fig. 3a). In the second half of the year, a zonal SSS minimum (< 35.6) occurred between 90°E-105°E which is also in favor of the separation of two maxima. The gridded Argo field is too smooth to resolve these details, but the freshening in austral spring is also suggested (Fig. 3b). Aquarius SSS is rather noisy near the western boundary, showing two peaks at 30°E and 34°E that can be seen throughout the year. In the eastern and western regions, Aquarius SSS is higher by 0.13 and 0.19 respectively than Argo SSS (Fig. 3c and e). Albeit with lower mean salinity, seasonal variations of SSS are quite consistent between the two datasets in the three regions. Argo-based variability is weaker in the central region (Fig. 3d). According to Aquarius data, the eastern and central regions reach the highest salinity in February-March and lowest salinity in August-September. The variations in the western region are roughly the opposite, showing the higher salinity in October- November and lowest salinity in February-March.

Seasonal variations of the three regions are associated with basin-scale SSS variability of the SIO (Fig. 4). The higher SSS values in the central and eastern maxima in April are associated with positive seasonal SSS anomalies covering the subtropical central and eastern SIO basin and also the tropical southeast Indian Ocean (Fig. 4b). By sharp contrast negative SSS anomalies are seen in the tropical South IO from the equator to 20°S. Negative SSS anomalies extend southwestward along the African coast to the western region. The SSS anomaly pattern is reversed in October, yielding negative anomalies in the central and eastern regions and positive anomalies in the western region (Fig. 4d). Seasonal variations of the freshwater flux (E-P) are essential for the generation of SSS variability (Fig. 4e-h). Largest variations of E-P occur in the tropical South IO between the equator and 20°S, reflecting the strengthening/weakening of the atmospheric intertropical convergence zone (ITCZ) in the Indian Ocean. In that region, SSS anomalies are largely attributable to E-P changes. However, SSS variations in the three subtropical maximum regions are not well explained by the local E-P changes. For example, Figures 3 and 4 suggest persistent SSS decrease in austral winter in the central and eastern region, but positive E-P anomalies were dominant over this region in spring and summer (Fig. 4f and g). To clarify the dynamics controlling SSS seasonal cycles, an in-depth investigation involving mixed layer budget analysis is conducted in the following section.

4 SEASONAL PROCESS

The above analysis broadly describes the spatial-temporal characteristics of the three SSS maxima and their seasonal variations. In this section, we examine the roles played by different processes in their seasonal variability. A mixed-layer freshwater budget is performed for the three regions. The method is widely used in recent studies to understand SSS variability on various timescales (e.g., Qu et al., 2011; Yu, 2011; Li et al., 2013; Li and Han, 2016), since the mean salinity of the surface mixed layer (MLS) is a good proxy for SSS in temporal change,

where ∂MLS/∂t is the temporal tendency of MLS computed with Argo salinity data. SFF is the surface freshwater flux forcing term computed with OAFlux evaporation and GPCP precipitation through

where H is the mixed layer depth (MLD) computed with as the depth at which the potential density increase ∆σ from the surface value (at 5 m) is equivalent to a temperature decrease by ∆T=0.5℃ (de Boyer Montégut et al., 2004),

where T₅, S₅, and P₅ are temperature, salinity, and pressure at 5 m (the shallowest level of IPRC Argo data). ADV stands for horizontal advection (ADV) by ocean currents and is calculated as

where u=(u, v) is surface horizontal current vector taken from OSCAR data, and MLS=(∂MLS/∂x, ∂MLS/∂y) is the horizontal gradient of MLS. ENT is vertical entrainment term used to quantify the effect of water exchange between the surface mixed layer and the water below, and it is calculated as (Stevenson and Niiler, 1983),

In the above, S_-H and w_-H are the salinity value and vertical velocity at the mixed layer base (-H). w_-H is computed with horizontal current u using the conservation of mass, w_-H=H·∇·u, where ∇·u=∂u/∂x+∂v/∂y is the divergence of surface current. ∂H/∂t is the local MLD tendency, u·∇·H is the MLD change due to horizontal advection. We also used Ekman pumping velocity to replace w_-H in Eq.5, and the results do not change radically. R is the residual term representing the unresolved processes such as lateral mixing induced by eddies and diapycnal mixing at the bottom of the mixed layer.

Figure 5 shows the monthly time series (Fig. 5a-c) and climatological seasonal cycles (Fig. 5d-f) of Argo and Aquarius ∂SSS/∂t, in accompany with SFF, ADV, and ENT. For the gridded Argo data, ∂SSS/∂t and ∂MLS/∂t are highly consistent, except for a slightly larger amplitude of ∂SSS/∂t. It can be seen that in the eastern and central regions, SFF (blue) can to a large extent explain seasonal variations of ∂SSS/∂t (gray bar). Effects of ADV is small for the central region but rather large in the eastern and western regions. ENT has little contribution in all the three regions. For the western region, there is large discrepancies between Argo and Aquarius in ∂SSS/∂t. It is possible that the budget analysis based on Argo data cannot capture the complicated SSS dynamics in this region. Hereafter we mainly focus on the central and eastern regions. It is however noticeable that ADV shows large seasonal variability and it is able to explain some ∂SSS/∂t anomalies in the western region. In comparison with the eastern and central regions, ocean dynamics likely plays a more important role for the western region.

Figure 5 indicates that SFF is essential for regulating SSS seasonal cycles of the eastern and central maxima. However, recalling Fig. 4 suggests that evaporation and precipitation cannot explain SSS seasonal variations in the two regions. In Eq.2 we can see that SFF is determined by evaporation, precipitation, and MLD. In Fig. 6 we examine the roles played by them in SFF seasonal cycle. Evaporation and precipitation are both important in determining E-P (Fig. 6a-c). Note that E-P seasonal cycles are not consistent with those of SFF. Also worthy of noticing is that E-P is positive in all the three regions, exerting a forcing of salinity increase throughout the year. In the subtropical SIO MLD shows strikingly large seasonal variability, undulating between 20-30 m in austral summer and 80-120 m in austral winter (Fig. 6d-f). This is primary induced by the strong surface warming (cooling) in austral summer (winter) by solar radiation change. The shallow MLD in summer amplifies the positive E-P forcing effect and favors salinity increase (positive ∂SSS/∂t), while the deep MLD in winter attenuates the forcing effect and leads to salinity decrease (negative ∂SSS/∂t). This effect is important for the eastern and central SSS maxima.

To test this hypothesis, we attempt to isolate the effects of E-P and MLD on SFF following Li et al. (2016), by simply recomputing SFF by assuming constant E-P or MLD over the year. For example, SFF_EMP uses monthly E-P (identical to SFF in Eq.2) and annual-mean MLD, and as such it quantifies the seasonal change of SFF induced by E-minus-P; on the other hand, SFF_MLD uses annual-mean E-P and monthly MLD so that it quantifies SFF change induced by MLD change. Figure 6g & h clearly shows that SFF_MLD is quite consistent with SFF, while SFF_EMF does not contribute much to SFF. In the central and eastern regions, the large seasonal change of MLD is the primary cause for the seasonal variability of SFF and thus SSS.

In the western region, none of SFF, ADV, and ENT is able to well explain the SSS variability in Aquarius and Argo. This is likely subjected to complicated processes which is not resolved by Eq.1. For example, lateral diffusion by mesoscale eddies may be important. The annual-mean surface eddy kinetic energy (EKE) distribution reveals high mesoscale eddy activity in the western and eastern SSS maximum region (Fig. 7), particularly in the vicinity of the Agulhas Current region (Beal et al., 2011; Beal and Elipot, 2016) and the Leewin Current region (Feng et al., 2005). Mesoscale eddies generally tend to reduce the salinity value of an SSS maximum by mixing the high-salinity water with surrounding fresher water. As such, the time-varying mesoscale eddy activity (quantified by EKE) may exert a forcing effect on SSS variability, that is, high (low) EKE acts to reduce (raise) the local SSS value.

Figure 8 compares the seasonal cycles of EKE and SSS tendency in the three regions. Except for the central region where the mean EKE is low, EKE shows a good out-of-phase relationship with ∂SSS/∂t in the eastern and western regions. Positive EKE anomalies generally co-occur with negative SSS tendency anomalies (indicating SSS decrease). Therefore, lateral mixing by mesoscale eddies is likely a fundamental ocean dynamical process controlling SSS variability of the two regions.

Summarizing the results of this section, we come to the conclusions for SSS seasonal variability as follows. The central and eastern regions are primarily controlled by surface freshwater flux forcing term whose seasonal variability is however caused by MLD rather than E-P, and in addition the eastern SSS maximum is also affected lateral mixing of mesoscale eddies. For the western region, our budget analysis fails to resolve the processes controlling SSS variability, and mesoscale eddy mixing is likely one of the major processes affecting SSS seasonal cycle.

5 DISCUSSION AND CONCLUSION

SSS maxima are formed in the subtropical oceans due to positive mean evaporation minus precipitation and serve as the primary salt sources for the global ocean. In comparison with the counterparts in the Pacific and Atlantic Oceans, the SSS maximum of the Indian Ocean is rarely investigated. In this study, we utilize SSS data of Aquarius and Argo floats during September 2011-May 2015 to provide the first comprehensive description of the SSS maximum of the subtropical SIO and examine the ocean processes controlling its seasonal cycle. Aquarius data detect three disconnected SSS maximum regions (SSS>35.6) in the eastern (100°E-115°E, 38°S-28°S), central (60°E-95°E, 35°S-25°S), and western (25°E-40°E, 38°S-20°S) parts of the subtropical SIO basin, respectively. Such structure is however not seen in gridded Argo data. Analysis of Argo profile data confirms the existence of the eastern SSS maximum patch but is not supportive for the western one. Comparisons suggest evident overestimation of SSS by Aquarius in the coastal areas of the eastern and western boundaries. SSS is low in the tropics (Fig. 1) due to heavy rainfall. The fresher water, along with that brought in by the Indonesian Throughflow, flows southward along the west coast of Australia via the Leeuwin Current. This may serve as a plausible cause for the more southward position of the eastern maximum than the central one.

Seasonal cycles of the three high-SSS regions are then investigated, which show different characteristics and are likely controlled by different mechanisms. The eastern and central regions reach the highest salinity in February-March and lowest salinity in August-September, while the western SSS maximum in the Agulhas Current region reaches the highest salinity in austral spring and the lowest salinity in austral fall. A mixed layer freshwater budget analysis is employed to shed lights on the processes controlling the SSS seasonal cycles. It is found that in the central and eastern regions, the SSS tendency can be well explained by surface freshwater forcing (SFF). However, SFF change is not induced by changes of evaporation or precipitation. Seasonal cycles of SFF are not consistent with those of E-P. Instead, the large seasonal undulations of the MLD in the subtropical SIO is the major controlling factor. The shallow MLD (20-30 m) in austral summer amplifies the positive forcing effect of E-P and causes SSS rising, while the deep MLD (80-120 m) in austral winter attenuates the positive E-P forcing effect and leads to SSS decrease. For the western region, our budget analysis fails to resolve the processes controlling SSS variability. Large-scale ocean current advection (ADV), particularly that of the Agulhas Current, may be important there. Further analysis reveals that the subtropical SIO is characterized by high activity of mesoscale eddies, as quantified by EKE. EKE shows a good out-of-phase relationship with SSS tendency in both the western and eastern regions, indicating the important role played by mesoscale eddy-induced lateral mixing in SSS variability.

Ocean dynamics are closely associated with large-scale wind forcing (Fig. 9). We provide discussion for this issue as below. Meridional displacements of the surface wind circulation are rather evident. In January the subtropical SIO is controlled by easterly winds (Fig. 9a), and meanwhile the tropical South Indian Ocean is dominated the ITCZ which causes negative SSS anomalies through enhanced rainfall (Fig. 4). These negative SSS anomalies penetrate to the western region, probably via the east Madagascar Current and the Mozambique Channel, leading to the SSS decrease by ADV (Fig. 5f). In July, the westerly winds shift to the north and strengthens the subtropical gyre of the SIO which is quite evident if looking at the westward South Equatorial Current between 20°S-10°S. In response, the Agulhas Current and East Madagascar Current are also enhanced (Matano et al., 2002), transporting more fresh water to our western region and again exerting a freshening effect (Fig. 5f). It is still unclear how surface winds and large-scale circulation modulates local EKE variability shown in Fig. 8. This issue requires instability and energy analysis to understand.

Another important issue we would like to discuss is how the eastern maximum separates from the central one. To our understanding, the lower mean salinity in the gap area is crucial. As revealed by Fig. 2a, the low SSS values occur mainly in austral spring. During these months, negative E-P anomalies there are stronger than those within the two SSS maxima (Fig. 4h). Mesoscale eddy activity is also high in this region (Fig. 7) due to eddy shedding from the Leeuwin Current (e.g., Fang and Morrow, 2003; Feng et al., 2005) and is enhanced in austral spring (Fig. 8a and 8b), which favors the mixing with fresh water in the north and south.

The budget analysis in Section 4 contain large uncertainties, due to various errors contained in the datasets, and the residual term is rather large, particularly for the western region. It can be seen in Fig. 5 that SSS tendency computed from Aquarius and Argo show evident discrepancies due to reasons outlined in Section 3, and ADV is rather noisy in the western region possibly due to altimeter sea level data error in coastal area. This study provides a preliminary description for the structure and seasonality of the SSS maximum regions of the SIO. Further investigation with more data and numerical models should be conducted to confirm our findings.

Here we reveal complicated mechanisms for the seasonal SSS variability in the subtropical SIO. Ocean dynamical processes, including mixed layer depth change, large-scale ocean circulation, and mesoscale eddies, play the fundamental role in the formation of the SSS maxima and their seasonal cycles. SSS variability cannot be simply regarded as the "slab-ocean" response to surface freshwater flux change. To gain an in-depth understanding of the variabilities of ocean salinity and global water cycle, more effort should be devoted to relevant ocean dynamical processes on various scales. There are many interesting topics for future researches, including its intraseasonal, interannual, and decadal variabilities and their relationship with climate change, its effect on the ocean stratification and circulation, its spreading in the Indian Ocean thermocline, and its connection with Atlantic Ocean through the Agulhas Leakage.

6 DATA AVAILABILITY STATEMENT

Aquarius Version-4.0 CAP Level 3 data are distributed through the Physical Oceanography Distributed Active Archive Center (PO.DAAC) of the National Aeronautics and Space Administration (NASA) and accessible through the FTP site ftp://podaac-ftp.jpl.nasa.gov/allData/aquarius/. IPRC Argo data are distributed by the Asia Pacific Data-Research Center (APDRC) through the website http://apdrc.soest.hawaii.edu/data/data.php?discipline_index=2. Argo profile data are provided by the Coriolis Global Data Acquisition Center of France through http://www.coriolis.eu.org; OSCAR sea surface current data are downloaded from http://www.oscar.noaa.gov/. GPCP precipitation rate data are available at https://data.noaa.gov/dataset/. OAFlux data are from http://oaflux.whoi.edu/. CCMP data are from http://podaac.jpl.nasa.gov/. AVISO SSH and surface geostrophic current data are downloaded from http://www.aviso.oceanobs.com/.