Interannual Variability and Multiyear Trends of Sea Surface Salinity in the Amazon-Orinoco Plume Region From Satellite Observations and an Ocean Reanalysis

,

sea level in the tropical Atlantic (Giffard et al., 2019). The abundant nutrients associated with the freshwater flux can strongly impact ocean productivity in the plume and beyond (Gouveia et al., 2019;Subramaniam et al., 2008).
The Amazon-Orinoco river discharge has strong seasonality. Amazon outflow ranges from 10 5 m 3 s −1 in November to 2.4 × 10 5 m 3 s −1 in June, whereas the Orinoco discharge ranges from 10 4 m 3 s −1 in March to 7 × 10 4 m 3 s −1 in August ( Figure S1 in Supporting Information S1). Combined, the two rivers discharge a minimum of 1.3 × 10 5 m 3 s −1 in November and a maximum of 3 × 10 5 m 3 s −1 in August. The mean Orinoco River discharge is about 15% of the Amazon's, but the Orinoco can contribute up to 25%-30% of the total Amazon-Orinoco discharge from August to October due to the seasonal phase shift between the two rivers.
In addition to the strong seasonality of river discharge, the strong western boundary currents in the northwestern Atlantic have strong impacts on the seasonal variability of the sea surface salinity in the plume region (Coles et al., 2013;Lentz, 1995) and more broadly in the northwestern basin (Foltz et al., 2004(Foltz et al., , 2015. The Amazon water is mainly transported northwestward to the Caribbean and North Atlantic in February-May by the North Brazil Current (NBC). From June to January, Amazon freshwater is carried by the NBC retroflection, which feeds the eastward North Equatorial Countercurrent (NECC) (Coles et al., 2013;Lentz, 1995;Muller-Karger et al., 1988). The Orinoco water mainly affects the eastern Caribbean (Lopez et al., 2013). The size of the Amazon-Orinoco river plume varies strongly throughout the year and is highly correlated with seasonal variations of river discharge (Zeng et al., 2008).
There have been studies reporting that North Atlantic climate variability is largely controlled by the North Atlantic Oscillation (NAO, George & Saunders, 2001;Hurrell & Deser, 2010) and El Niño Southern Oscillation (ENSO) (Ropelewski & Halpert, 1987;Tyaquiçã et al., 2017;Zeng, 1999;Zeng et al., 2008). The NAO represents the variation of the meridional gradient of atmospheric pressure between the Azores high and the Icelandic low. The variation of atmospheric pressure in the Azores high also controls the variation of the meridional pressure gradient between the Azores high center and the tropical Atlantic (George & Saunders, 2001), which can affect the strength of trade winds in the tropical North Atlantic and associated mixing, evaporation, precipitation, and large-scale wind-driven ocean circulation. In addition, anomalous zonal atmospheric circulation associated with ENSO is known to strongly influence precipitation in the Amazon-Orinoco catchment and over the ocean where the rivers discharge (Enfield & Mayer, 1997;Ropelewski & Halpert, 1987;Zeng, 1999;Zeng et al., 2008).
Findings from previous studies also suggest strong interannual variability of the Amazon-Orinoco river plume and attribute it to multiple factors, such as river discharge (Molleri et al., 2010;Tyaquiçã et al., 2017;Zeng et al., 2008), large-scale circulation (Coles et al., 2013;Foltz et al., 2015), eddies (Chérubin & Richardson, 2007;Fournier et al., 2017;Reverdin et al., 2021), wind, and ITCZ position (Fournier et al., 2017). Some of these studies use satellite ocean color to infer the variability of the Amazon plume . Besides the complication of separating colored dissolved organic matter (CDOM) and chlorophyll from surface reflectance, the variability of the plume water derived from ocean color is strongly affected by clouds and aliased by other phenomena, such as upwelling. Other studies use satellite SSS (Fournier et al., 2017;Grodsky et al., 2014) and ocean models (Coles et al., 2013) with limited time records of 2-4 years, which are too short to characterize interannual variability of the Amazon plume and its causes. In this study, we characterize the Amazon-Orinoco river plume interannual variability using an 11-year record of satellite observations. The causes of the variability of the plume are also determined using simple mixed layer salinity sensitivity analyses based on an ocean reanalysis, and connections to large-scale climate variability are investigated.

Data
We investigate variability of the Amazon-Orinoco plume using satellite observations of sea surface salinity and output from an ocean reanalysis. For satellite data, we use sea surface salinity data from the European Space Agency Sea Surface Salinity Climate Change Initiative Project (ESA CCI SSS, https://data.ceda.ac.uk/neodc/ esacci/sea_surface_salinity/). In this product, level-2 and level-3 data from the SMOS, AQUARIUS, and SMAP missions are combined with corrections for land contamination, radio frequency interference, latitudinal dependence bias, and long-term bias. We use the monthly product version V03.21, which covers the period 2010-2020.
To assess the role of different forcings in the variability of the plume, we use the Simple Ocean Data Assimilation (SODA) ocean reanalysis, version 3.4.2 . The forcings that can impact plume SSS include precipitation, evaporation, vertical mixing, and horizontal advection. Vertical mixing and horizontal advection can be inferred from SODA, but the SODA archive does not include precipitation and evaporation. We obtain these forcings directly from the ERA-Interim data set (Berrisford et al., 2011), which was used to force SODA . For river discharge, we obtained the Amazon River discharge data at Obidos station and Orinoco River discharge data at Ciudad Bolivar station from the HYBAM data set (available at http://hybam. omp.obs-mip.fr/).
To investigate the relationships between the plume's interannual variability and large-scale climate variability, we use the NAO index and Ocean Niño Index (ONI) produced by NOAA's Climate Prediction Center and distributed by NOAA's Physical Science Laboratory (https://psl.noaa.gov/). Both of these climate phenomena are known to strongly control the climate variability of the tropical Atlantic Ocean (Czaja et al., 2002;Enfield & Mayer, 1997).

Plume Index Computation
Previous studies have used different salinity thresholds to determine the plume extent, such as 34.7 psu (Zeng et al., 2008), 35 psu (Grodsky et al., 2014), or 35.5 psu (Fournier et al., 2017). In this study, we first define a region of high SSS variability that is bounded by the 0.5 psu SSS standard deviation contour computed from the merged satellite daily data (red line in Figure 1), representing the 80th percentile of SSS variability throughout the tropical North Atlantic (0-35°N). This means that the total variability in the region bounded by this contour is in the top 20% of the variability observed throughout the entire tropical North Atlantic. The region extends from 0° to 23° N and from 75°W to west Africa with a narrow band of high SSS variability (20°-40°W, 5°-11°N) that can be attributed to the NECC and Inter-Tropical Convergence Zone (ITCZ; black contour in Figure 1). The black contour encloses the upper 20% of precipitation variability in the tropical North Atlantic. The low variability of SSS within the black contour and between 1° and 5°N implies that the impact of ITCZ precipitation on SSS is not as strong as the advection of low-salinity water by the NECC. To minimize the impact of the ITCZ on the plume variability and to simplify the salinity balance analysis in the plume region, we define the plume region to be within the red rectangle in Figure 1 (70°W-42°W, 0°N-23°N). The plume index in this study is therefore defined as the spatial mean of SSS over the rectangle.

Impacts of Different Forcings
Potential forcings that affect interannual variability of the Amazon-Orinoco plume include river discharge (R), local evaporation (E) and precipitation (P), horizontal advection (adv), and vertical mixing. Previous studies (Dong et al., 2009;Ren et al., 2011) often use a salinity budget equation to attribute interannual variability of Figure 1. Tropical North Atlantic sea surface salinity variability over the 2010-2020 period obtained from satellite observations. The 0.5 psu standard deviation contour (in red) encloses the area in the upper 20% of sea surface salinity variability. The superimposed precipitation contour (in black, 5 mm/day) encloses the upper 20% of precipitation variability. The red rectangle defines the plume region for this study, and the white rectangle indicates a non-plume region used for comparison. The magenta arrows illustrate the North Brazilian Current, North Equatorial Current, and North Equatorial Counter Current. mixed layer salinity to different components, such as fresh water flux, horizontal advection, vertical velocity induced by Ekman pumping, and diffusivity. However, because those terms include the synergy effects between different forcings, the contribution from each term is not due purely to the forcing of consideration. For example, the impacts of the surface freshwater flux on mixed layer salinity ([E-P]S/h, where E, P, S, and h are evaporation, precipitation, mixed layer salinity, and mixed layer depth, respectively) are proportional not only to the variability of the freshwater flux but also to the variability of the mixed layer depth, which is controlled by vertical entrainment mixing, and to the observed mixed layer salinity, which is controlled by all forcings. In this study, we assess the impact of each forcing on the plume SSS interannual variability using simple mixed layer salinity balances in which only one forcing at a time varies interannually.
Consider the plume as a box with fixed horizontal boundaries (the red rectangle in Figure 1) and a flexible bottom boundary that varies spatially and temporally with the mixed layer depth ( Figure 2). Exchanges of water and salt through the surface and the lateral and bottom boundaries will result in an increase or decrease in the average mixed layer salinity. Since salinity is approximately constant vertically within the mixed layer, mean changes in salinity over the whole volume of the box are representative of changes in the spatial-mean SSS.
Different sensitivity analyses are conducted for different forcings to assess their partial impacts on the plume SSS. In a sensitivity analysis, only the forcing of consideration and the plume SSS can vary interannually; other background conditions, including the plume volume and salinity at the base of the plume, which represent the impacts of all available forcings, are set to their climatological values. Consider that at time t, the mixed layer ( Figure 2) has volume ̂ (in m 3 ), salinity ̂ , and a total salt mass of ̂ (in kg). The salinity balance at t + dt due to forcing F, with other forcings and background volume remaining climatological, can be formulated as where and ̂ are the mixed layer salinity at time (t + dt) and (t), respectively. The initial value of ̂ at t = 0 is the actual plume SSS in January 2004. ̂ represents the amount of salt in the mixed layer at time (t), = −̂ represents the net change of salt mass due to all climatological forcings, including F, from (t) to (t + dt), and represents the net change of salt mass due to anomalous F from (t) to (t + dt). When F is purely climatological, the third term on the right side of Equation 1 is zero, and Equation 1 reduces to where ( ) = = represents the climatological background variation of the mixed layer salinity. When F is not climatological, ( ) contains the cumulative effects of anomalous F up to time (t), and this is then the background condition for anomalous F at (t + dt). The third term on the right hand of Equation 1 represents the effects of anomalous F at (t + dt), which causes to differ from ( ) .
For river forcing, the freshwater volume gained from anomalous river discharge of increases the total volume of the plume box and raises the sea level. After equilibrium, the volume of water in the mixed layer does not change, leading to a loss of salt in the mixed layer of ( ) because the freshwater with zero salinity displaces the same volume of existing ocean water. Hence, This effect is easy to imagine when the anomalous river discharge is positive. When it is negative, the formula is the same because Equation 3 is an adjustment to the climatological river discharge that is already accounted for in the background salinity ( ) (Equation 2).
Similarly for precipitation: For evaporation, freshwater lost due to evaporation on the surface decreases the thickness of the mixed layer. Constant vertical entrainment mixing (i.e., the mixed layer volume) then introduces water at the base of the mixed layer with climatological salinity with a salt gain of , hence For advection, changes in circulation can result in convergence or divergence inside the plume region, which can change both the total salt mass and volume of the plume box. Given net advected volume anomaly of and net advected salt mass anomaly of , if there is convergence in volume ( > 0), after equilibrium, the excessive volume is displaced from the mixed layer so that the salinity advected is the background salinity ( ) : In contrast, if there is volume divergence ( < 0), entrainment will introduce base water into the mixed layer: where and are the integrated net volume and salinity flux (kg month −1 ), respectively, along the boundary of the plume region: where are surface zonal and meridional velocity, respectively, and , are the climatological mixed layer depth and sea surface salinity along the boundary, respectively. By fixing the mixed layer depth and sea surface salinity along the boundary at their climatological values, interannual variability of advection in the plume region is purely due to the variability of surface currents, avoiding the impacts of vertical mixing, precipitation, river discharge, and evaporation outside the plume region on the advection term.
Vertical (diapycnal) mixing is related to the rate of change of the mixed layer depth, which can be due to wind-induced turbulent mixing, current shear, buoyancy forcing, or horizontal advection of the mixed layer depth. The first three processes entrain subsurface water with different salinity relative to the surface layer and are therefore capable of changing the mixed layer salinity. The advection of shallower mixed layer depth into the plume region may also trigger vertical mixing if, for instance, wind-induced mixing in the plume region remains the same. Hence, changes of volume and salinity at the base of the mixed layer from (t) to (t + dt) control variations of the plume salinity, not the volume anomaly compared to climatological volume: Vertical velocity induced by horizontal divergence/convergence due to Ekman pumping or other processes is not included in the vertical mixing component. In the absence of turbulent mixing, vertical velocity (upwelling/downwelling) only raises or deepens the pycnocline, which changes the mixed layer depth locally without changing the mixed layer salinity. In reality, if the vertical mixing length is deeper than the raised mixed layer in the upwelled water region, the upwelled water can be mixed up to the mixed layer and can affect the mixed layer salinity. Here, this term is attributed to advection because it is the forcing that creates the potential for a change in mixed layer salinity, and vertical mixing can act on that potential. In our simulations for forcings other than vertical mixing, we assume that vertical mixing is climatological and is constant for a given time step. Therefore, upwelled water is always introduced to the mixed layer and downwelling does not affect mixed layer salinity.
We limit these analyses to the Argo period from 2004 to 2017 when SODA benefits from the assimilation of more observations. The data and the time step used in these sensitivity analyses are monthly.

Interannual Variability of SSS in the Plume Region
The Amazon-Orinoco plume exhibits strong interannual variability with a range in the annual mean plume SSS anomaly of −0.17 psu in 2010 to 0.26 psu in 2016 based on satellite data (black solid line in Figure 3a). The interannual range is similar in SODA with a minimum annual mean of −0.24 psu in 2010 and maximum annual mean of 0.18 psu in 2016. The discrepancies in the values between observations and SODA may be partly due to the fact that SODA does not assimilate satellite SSS. To put these values into perspective, the interannual range of SSS anomalies to the northeast of the plume region (the white rectangle in Figure 1) is about three to four times smaller: −0.07 to 0.09 from observations and −0.06 to 0.07 from the SODA reanalysis ( Figure S3 in Supporting Information S1). These differences in the varying ranges between the plume and non-plume regions, confirmed  Figure 1) from the merged satellite SSS data set (a) and the Simple Ocean Data Assimilation (SODA) reanalysis (b). Dry season is from January to March, and the flood season is from July to September, chosen based on climatological mean plume salinity (see Figure S2 in Supporting Information S1). The trends are computed over the period 2010-2016.
by both satellite observations and reanalysis, may illustrate the important role of river discharge and strong ocean dynamics in the plume region.
In some years, the plume SSS has similar anomalies in the flood and dry seasons, for example, 2010 and 2016 (Figures 3a and 3b). In other years, such as 2012 and 2015, the plume SSS has a reduced annual cycle of SSS with higher than normal salinity in the flood season and lower salinity in the dry season. In contrast, during 2011, the plume SSS has an amplified annual cycle with lower SSS in the flood season and higher SSS in the dry season. The correlation between observed plume SSS anomalies in the dry and flood seasons over the period 2010-2020 is only 0.48 and is not significant. These differences in the phases of wet and dry season SSS suggest that either the magnitude or type of forces that drive the plume SSS interannual variability is not the same throughout the year.
There are also significantly increasing trends in plume SSS of 0.89 and 0.75 psu per decade for the wet season and annual mean, respectively, during 2010-2016 ( Figure 3a). The plume SSS shows an increasing trend over the dry season of 0.48 psu per decade, which is significant at 90%. These trends are all significant at 99% in the SODA reanalysis (Figure 3b).

Mechanisms of the Variability
To investigate the causes of interannual variability of the plume SSS, we look at the forcing variability in 2011 and 2015. These years represent amplified (2011) and reduced (2015) seasonal cycles of plume SSS that were observed in both observations and SODA ( Figure 3). Figure 4a shows that the mean plume SSS in the dry season (January-March) of the two years differs slightly (∼0.1 psu), and this gap rapidly increases to about 0.45 psu in the flood season. Figure 4b-4f illustrates the potential forcings responsible for this variability of the plume SSS, including river discharge (R), precipitation (P), evaporation (E), horizontal advection by the ocean circulation (ADV, represented by zonal current Ucur), and vertical mixing using mixed layer depth as a proxy. The mixed layer depth shows a close relationship with the mean wind stress in the plume region (Figure 4b), highlighting the important role of wind-induced vertical mixing in the region (e.g., Rugg et al., 2016). Vertical mixing, advection, and precipitation have opposite signs between the years, reflecting a clear separation in their magnitudes (Figures 4b, 4c and 4e). Stronger precipitation and weaker vertical mixing (thinner mixed layer) support lower plume salinity in 2011 than in 2015. The positive anomaly of zonal current in 2011 (i.e., anomalous eastward flow) may represent a weakening of the NBC, NEC, and/or the strengthening of the NECC (Figure 1). This will be investigated later in this section. Evaporation and river discharge also show mirror-like patterns, but their magnitudes in the two years are mixed (Figures 4d  and 4f). Figure 5 shows the annual mean anomaly maps of the plume SSS and forcings in 2011 and 2015. In agreement with the spatial mean in Figure 4, the clear contrasts of salinity as well as wind stress, mixed layer depth, precipitation, and advection between the two years are seen almost everywhere in the plume region (Figures 5a-5h and 5). The wind stress fields in the two years are different not only in magnitude but also in the spatial patterns inside the plume region (Figures 5c and 5d). In 2011, southeasterly anomalous wind occurred over the northeastern part of the plume region, whereas west to southwesterly anomalous winds occupied the southwestern part. In contrast, strong northeasterly positive anomalous wind dominated the whole plume region in 2015. This strong northeasterly anomalous wind probably originated from the subtropical eastern Atlantic (Figure 5d). The mixed layer depth also shows close spatial variation with the wind stress field (Figures 5c-5f). The largest differences in plume salinity anomalies are in the southeastern portion of the plume box, where there is a large low-salinity patch in 2011 under weakened NBC, NEC (positive anomaly), and NECC (negative anomaly) conditions (Figures 5a  and 5g). The weakening of the current systems may favor a more pronounced northward spreading of freshwater discharge originating near the equator since it allows the fresh water to stay within the plume region for a longer period of time. The shape of the large patch of low-salinity water is similar to that of the shallow mixed layer region (Figures 5a and 5e), suggesting that either weaker vertical mixing (thinner mixed layer) supported lower SSS or lower SSS created stronger stratification that limited mixing or both. In addition, this large low-SSS patch coincides with an area of lower than normal evaporation in 2011 and higher than normal evaporation in 2015 (Figures 5a, 5i and 5j). In the other parts of the plume, evaporation does not show large differences between the years. Precipitation also contributed to the low-salinity patch with higher than normal precipitation in 2011 and lower than normal precipitation in 2015 (Figures 5k and 5l). However, the strongest contrast of precipitation in the two years is in the northwestern part of the plume box.
All of the forcings except evaporation and river discharge contributed to the strong contrast in plume SSS between 2011 and 2015. However, it is unclear which forcings contributed most to the plume SSS variability. To quantify this, we evaluate the partial effect of each forcing on the plume SSS using the sensitivity analysis described in Section 2.3. The overall results of the analysis over the extended period from 2004 to 2017 (e.g., time series of each forcing term including the seasonal cycle) can be found in the supporting document ( Figure S4 in Supporting Information S1). Figure 6 illustrates the partial effects of the forcings on the plume SSS anomalies for the years 2011 and 2015.
We can see that both evaporation and vertical mixing result in lower/higher plume salinity in 2011/2015 in agreement with the actual plume SSS variability in SODA (Figures 6a, 6e and 6f). However, the differences in these years caused by these forcings are minor compared to the SSS differences ( Figure 6f). River discharge shows an opposite effect, contributing to higher plume salinity in 2011 compared to 2015 ( Figure 6c) and thus, it cannot be the driver of the plume salinity contrast. Precipitation better reproduces the sign and magnitude of the plume SSS difference between the years with a plume salinity gap ( We also tested with another case study of 2010 versus 2016, which represents the minimum and maximum mean plume SSS anomaly, respectively. Advection is also the best match in terms of reproducing the magnitude of the difference in plume SSS ( Figure S5 in Supporting Information S1) for those years. This converges to the findings of previous studies (Coles et al., 2013;Lentz, 1995) on the important role of advection for seasonal plume variability. Table 1 Figure 7). Although advection does not reproduce these phases perfectly, it is the only forcing that shows two distinguishable phases over the same periods. During the decreasing phase of 2004-2010, the plume SSS driven by advection has a decreasing trend of −0.37 psu.decade −1 (significant at the 95% level) compared to −0.62 for SSS in SODA. During the increasing phase of 2010-2017, the trends are 0.85 and 0.58 psu decade −1 for the advection effect and SSS in SODA, respectively, all significant at the 95% level. None of the other forcings reproduces similar significant decreasing/increasing trends over the periods. This once again confirms the dominant role of advection in driving the plume SSS at interannual and decadal time scales.

Which Current System Is Most Important for Amazon-Orinoco Plume Variability?
We have seen that interannual and longer term variability of the Amazon-Orinoco plume SSS is strongly controlled by advection, and the dynamics in the plume region are complicated with the presence of three different major current systems, including the NBC, NEC, and NECC ( Figure 1). Hence, the next question is: Which current system contributes most to the plume variability?
To answer this question, we first determine the lateral boundary of the plume region that is most important for the total net salinity flux. Figure 8a shows the interannual variability of the integrated salinity flux along different plume boundaries and the associated net flux computed using Equation 7.
We can see that the salinity flux along the eastern boundary has the strongest variability and follows closely the net salinity flux. The correlation between the eastern boundary flux and the net flux is 0.63, which is significant at 95%, whereas the fluxes along the other boundaries do not have significant  correlations with the net flux. This remains true for climatological values: the eastern boundary flux provides the largest flux into the plume region and is the only component that has a significant correlation (0.83) with the net salinity flux ( Figure S6a in Supporting Information S1). Therefore, the eastern boundary, which includes the variability of the NEC and NECC (Figure 1), is the most important source of variability for the mean Amazon-Orinoco plume SSS.
Next, we locate where in the eastern boundary the exchange of water and salt is most crucial to the net salinity flux. Figure 8b shows a correlation map between the horizontal salinity flux annual anomalies and zonal surface current annual anomalies computed over the 2004-2017 period. Along the eastern boundary of the plume, there is a region of highly significant correlations (at 95%) between 4° and 6° N. This is the region where the NECC is located (Garzoli & Katz, 1983;Richardson & Reverdin, 1987). The negative correlation in this region means that a stronger NECC results in a weaker net salinity flux (less salt advected into the region) since the NECC brings salt and water out of the plume region. This patch of significant correlations associated with the NECC stretches westward to the NBC retroflection at around 50° W and 8°N (the black triangle in Figure 8b). Note that the magnitude of the salt flux does not directly change the mean plume salinity, which also depends on the difference between the salinity of the water in the plume and at its boundaries. For example, if the salinity of the plume is the same as salinity outside the plume, the salt flux does not change the plume's salinity. Figure 8c shows the variability (std) of the horizontal salinity flux along the eastern boundary with the highest interannual variability between 0° and 11°N. The actual latitude range of the NECC reported in previous studies (Garzoli & Katz, 1983;Richardson & Reverdin, 1987) is 3°-10°N (see also Figure S6c in Supporting Information S1). This range of latitudes contributes more than half of the total variability along the eastern boundary of the plume (Figure 8c). The integrated salinity flux along this range of latitudes of the NECC (the magenta line in Figure 8a) is also highly correlated with the eastern boundary flux and the total net salinity flux with coefficients of 0.75 and 0.68, respectively (significant at 99%). This means that the total variability of advection in the plume region is largely driven by the NECC. The NECC salt flux also has a significant correlation (95%) with the mean plume SSS anomalies of 0.58 over the 2004-2017 period. The positive correlation between NECC salt flux and the plume SSS means that a stronger NECC leads to higher plume SSS and conversely, a weaker NECC leads to lower plume SSS. This is probably due to the fact that the NECC tends to transport low-salinity water out of the plume region toward East Africa (Figure 1), reducing the residence time of low-salinity water inside the plume region.

The Role of Large-Scale Climate Phenomena
Although advection in general and the NECC in particular are the most influential forcing of the plume SSS interannual variability, advection can only explain about 50% of the total variance of the plume SSS annual mean anomalies (Table 1). The strong contrast between the plume SSS in 2011 and 2015 was associated with distinct differences not only in advection but also in precipitation, mixed layer depth, evaporation, and river discharge (Figure 4). This suggests that there is a large-scale driver behind these consistent changes in multiple forcings.   Figure 9 shows correlation maps between the winter-time NAO index (November-January mean) and the annual means of anomalous forcings and plume SSS in the following year. First, over the plume region, the NAO has a positive correlation with wind stress (not highly significant, Figure 9b) and mixed layer depth (highly significant in the western part of the plume region, Figure 9e). The increase in mixed layer depth in the western basin is consistent with the signal found for negative phases of the Atlantic Meridional Mode (i.e., anomalously cold SST in the tropical North Atlantic relative to the South Atlantic) (Rugg et al., 2016). This means that a positive phase of the NAO (NAO+ hereafter) tends to increase the wind stress and mixed layer depth, which can potentially increase the plume SSS due to mixing with higher salinity at the base of the mixed layer. Second, NAO+ tends to decrease precipitation and evaporation over the Amazon-Orinoco catchment and plume region with highly significant negative correlations with precipitation and less significant negative correlations with evaporation (Figures 9c and 9d). This will result in increasing plume salinity because precipitation has much stronger variability than evaporation (Figures 4 and 6). Third, NAO+ tends to strengthen the westward NBC and NEC (negative correlation) and eastward NECC (positive correlation), which may also result in higher plume SSS because the strengthened current systems tend to reduce the residence time of low-salinity water in the plume region. Therefore, NAO + influences almost all forcings (except evaporation) in a way that increases the plume SSS (Figure 9a). The region of highly significant correlations between the NAO and plume SSS seems to coincide with the region of active NBC and NECC. Given the important role of advection revealed from previous results, this suggests that the largest impact of the NAO on the plume SSS is through advection. Figures 10c and 10d show that ENSO has significant negative correlations with evaporation and less significant negative correlations with precipitation over the Amazon-Orinoco catchment and plume regions. These correlations suggest that El Niño tends to decrease evaporation and precipitation, probably resulting in higher plume salinity due to stronger impacts of precipitation on plume SSS (Figures 4 and 6). The impacts of ENSO on precipitation and evaporation are qualitatively similar to those of the NAO. However, the NAO shows stronger correlations with precipitation than ENSO and thus the NAO may play a larger role in driving plume SSS via the surface moisture flux than ENSO. ENSO has significant negative correlations with wind stress, but unlike the NAO, it has spatially varying correlations with mixed layer depth, meaning that the impact of ENSO on vertical mixing is unclear. ENSO also has weak negative correlations with the NBC. Overall, ENSO has positive correlations with SSS over the plume region, meaning that El Niño tends to increase the plume SSS. However, ENSO's effects on the plume SSS are not as strong as the NAO's (Figures 9a vs. 10a).
To investigate the causes of the stronger influence of the NAO on plume SSS compared to ENSO, we consider seasonal and spatial variations of their correlations. Figures 11a and 11b show meridional migrations of the NAO's influence on wind stress and precipitation. The band of positive correlations between the NAO and wind stress is located outside the plume region in January (Figure 11a). This band migrates southward and reaches the equator in April. It then increases in magnitude until July and weakens in the following months. Similarly, the strong negative band of correlations between the NAO and precipitation starts in March outside the plume region ( Figure 11b). This band migrates southward, reaches the upper plume region in April, and remains there until August before extending southward to the equator in October-November. These migrations of the NAO's effects on wind and rainfall result in the NAO's strongest influence on the forcing fields occurring over the plume region in the flood season ( Figure S7 in Supporting Information S1). The lag of the NAO's effects on precipitation compared to wind stress needs further investigation. In contrast, there are strong negative bands of correlations between ONI and wind stress and precipitation over the plume region during the winter months (January-February, Figures 11c and 11d). These bands of correlation weaken in the following months. This means that it takes less time for ENSO's effects on the forcing fields to arrive in the plume region, and ENSO's strongest effects are observed in the dry season ( Figure S8 in Supporting Information S1).
The seasonal differences between the NAO's and ENSO's effects on the forcing fields over the plume region may be due to differences in the locations of the centers of forcing of the NAO and ENSO in relation to the plume region. First, the Azores high, where NAO effects initiate, is farther away from the plume region than the western tropical pacific, where ENSO effects initiate. Second, zonal propagation in the atmosphere is generally much faster than meridional propagation because of equatorial waves. This difference in phasing of the impacts of the NAO and ENSO on the plume region may also partly explain the low correlation between the plume index in the flood season and the dry season (Section 3.1, Figure 3).  Observed plume SSS also peaked in 2016 and then decreased in response to decreasing phases of the NAO and ENSO during 2016-2020. Over the period 2004-2017, the correlation between the NAO and SODA plume SSS is 0.70, which is significant at 99%, whereas the overall correlation between ONI and SODA plume SSS is 0.51 and less significant (at 90%). The respective correlations between observed plume SSS and NAO and ONI indices over the 2010-2020 period are 0.60 (95% significance) and 0.43 (not significant). This once again confirms a stronger influence of the NAO on Amazon-Orinoco plume variability.

Discussion and Conclusions
In this study, we characterize the plume variability using mean SSS over a fixed plume region (Figure 1). Previous studies characterized the plume variability using plume size computed based on the area of water with SSS less than certain criteria (  test how close these area-averaged SSS and plume size indices are, we computed plume size with both satellite and SODA SSS using similar plume size criteria as in Fournier et al., 2017 (SSS < 35.5 psu). We found that the two indices are closely related with correlations of −0.84 and −0.69 for monthly climatological and interannual anomalies, respectively, using satellite SSS over the 2010-2020 period ( Figure S9 in Supporting Information S1). The correlations found using SODA SSS over the 2004-2017 period are also high: −0.85 and −0.93 for monthly climatological and interannual anomalies, respectively. This means that there is little difference between the two indices.
Some studies highlighted the important role of river discharge for plume size seasonal variations (Molleri et al., 2010;Zeng et al., 2008). We also found highly significant seasonal correlations between Amazon-Orinoco river discharge and plume size from satellite and SODA with coefficients of 0.95 and 0.94, respectively, when plume size lags by 3 months, in agreement with those studies ( Figure S10 in Supporting Information S1). This means that river discharge is the dominant driver of seasonal variability of the Amazon-Orinoco plume. However, on interannual time scales, the correlations between river discharge and plume size anomalies are low and insignificant with correlations of 0.25 and 0.16 based on satellite and SODA SSS, respectively. This suggests a minor influence of river discharge on interannual variability of the Amazon-Orinoco plume. However, other studies (e.g., Grodsky & Carton, 2018) have pointed out the importance of southern tributaries in conveying NAO effects to the interannual variability of the Amazon system discharge. Since it is still challenging for ocean reanalyzes including SODA to accurately account for the effects of continental runoff due to limited observations, the role of river discharge on the Amazon-Orinoco plume interannual variability may have been underestimated. However, given the similarities of interannual variability of the plume SSS from SODA and satellite data (Figure 3), SODA has reasonably reproduced the actual interannual variability of the plume SSS. This means that the missing runoff from the southern tributaries is unlikely to be important for the total variability of the plume SSS.
We conducted sensitivity analyses to compare the impacts of individual forcing on the plume SSS interannual variability and long-term tendency. In doing these tests, we ignored the synergy effect (interactions) among different forcings, which may result in underestimation or overestimation of the impacts of each forcing term. Further studies will be helpful for quantifying in more detail the contributions of different forcings to the interannual variability and trends of the plume SSS.
Using satellite observations of sea surface salinity, we showed strong interannual variability of the Amazon-Orinoco plume over the period 2010-2020 with a magnitude that is about five times larger than the surrounding region. This interannual variability of the plume SSS was found to be well reproduced in the SODA reanalysis. Using data from the SODA reanalysis and its forcings over an extended period from 2004 to 2017, we found that horizontal advection was the most important forcing of the interannual variability of the plume SSS. Exchange of water along the eastern boundary, especially within the active latitude range of the NECC, is most crucial to the variability of the advected net salinity flux and the plume SSS.
The variability of the forcings in the period was found to be strongly related to the NAO and ENSO. Over the plume region, the NAO appeared to be most influential on the forcing fields in the flood season, whereas ENSO had a stronger influence on the forcing fields in the dry season. These results identify NAO as the dominant driver of the plume SSS through its modulations of precipitation, vertical mixing, and especially horizontal advection. The difference in seasonality of the NAO's and ENSO's impact on the plume region partly explains the different variability of the plume SSS in the flood and dry seasons. The plume SSS also shows a significant upward trend during 2010-2016 as both the NAO and ENSO transitioned from a negative to a positive phase. It then shows a downtrend during 2017-2020 when both the NAO and ENSO are in decreasing phases. Under global warming, a possible increase in the magnitude and change in position of the NAO (Hu & Wu, 2004), combined with an increase in the frequency of extreme ENSO events (Cai et al., 2014), may result in an increase in the magnitude and a change in the phases of seasonal, interannual, and decadal variations of Amazon-Orinoco plume salinity with potential impacts on ocean-atmosphere interaction and biogeochemistry (Gevaudan et al., 2021;Subramaniam et al., 2008).