Sea surface salinity extremes over the global ocean

Sea surface salinity (SSS) extremes, characterized as events surpassing a certain threshold percentile, pose a threat to stenohaline organisms worldwide. This study presents the first global mapping of SSS extreme metrics and investigates their underlying drivers using daily reanalysis data. Our key findings have revealed that mesoscale eddies drive SSS extremes over most of the global ocean with peaks in duration ranging from 5 to 10 d and peaks in intensity ranging from 0.2 to 0.3 g kg−1. Freshwater fluxes (FWFs) and mean currents are generally associated with the occurrence of prolonged and intense SSS extremes in tropical and extratropical oceans, respectively. FWFs related to interannual climate modes contribute to the asymmetric distribution of high and low SSS extremes in the central Pacific Ocean and the western Indian Ocean. These results highlight the distinct impacts of different local drivers on the mean states of SSS extremes.


Introduction
Sea surface salinity (SSS) in the global open ocean spans a range from 32 practical salinity scales (pss, its unit uses practical salinity units (psu) after 1978, which is approximately 32 g kg −1 in absolute salinity) to 37 pss (Liu andWei 2021, Yu et al 2021).The upper end of this range is typically observed in areas characterized by high evaporation and strong ocean convergence (Gordon et al 2015), whereas the lower end is found in proximity to river outlets, regions impacted by melted ice, and areas with high precipitation (Yu et al 2011).Generally, there exists limited variability in SSS across the global ocean, with an average of 0.18-0.22g kg −1 between 50 • S and 50 • N (Liu et al 2022).
Due to the influence of ocean currents, airsea freshwater fluxes (FWFs), and other dynamic processes, SSS anomalies can occasionally exceed their typical variability, thereby triggering changes in coupled air-sea systems, such as typhoon intensity (Grodsky et al 2012).Abnormal salinity anomalies in the open ocean are typically characterized by their frequencies.Over a decadal time scale, 'Great Salinity Anomalies' (see supplementary texts for definitions) have been documented as extreme freshening events (0.15-0.2 psu) in the subpolar North Atlantic Ocean (Dickson et al 1988, Belkin et al 1998, Biló et al 2022).On an interannual scale, during record-breaking El Niño events, the tropical Pacific Ocean experiences extreme sea surface freshening exceeding 1 psu (Gasparin and Roemmich 2016).In the Amazon-Orinoco plume, SSS exhibits significant variability (>0.5 psu) associated with the North Atlantic Oscillation and the El Niño Southern Oscillation (ENSO) (Da and Foltz 2022).Intraseasonal SSS anomalies with amplitudes ranging from 0.2 to 0.35 psu have been observed in the Bay of Bengal and linked to the Madden-Julian Oscillation (Subrahmanyam et al 2018, Shoup et al 2019).
Additionally, tropical cyclones have been shown to induce notable changes in SSS over time spans of ten days to a few weeks, leading to either significant increases (up to 5.92 psu, Xu et al 2020) or decreases (0.16 psu, Steffen and Bourassa 2018).
Recent studies have introduced a different approach to defining abnormal events for oceanic variables.For instance, in the context of marine heatwaves, Hobday et al (2016) characterized a sea surface temperature (SST) extreme event as when the daily temperature exceeds the 90th percentile threshold corresponding to each calendar day over a prolonged period.This definition has been widely employed to comprehend the connections between SST extremes and climate dynamics (e.g.Holbrook et al 2019, Rodrigues et al 2019, Lee et al 2023).In contrast, the definition and investigation of extreme SSS events over the global ocean remain relatively unexplored.
While numerous studies have examined the physical drivers of SSS on various time scales, such as seasonal and interannual (Yu 2011, Vinogradova and Ponte 2013), our understanding of the drivers behind SSS extremes and their associated metrics remains limited.Mesoscale eddies, which account for a substantial portion of oceanic kinetic energy (Storch et al 2012), are potentially influential in triggering ocean extremes (Bian et al 2023).Therefore, it is crucial to consider the impact of mesoscale eddies when investigating the drivers of SSS extreme metrics.In this study, we examine SSS extreme metrics and their drivers by analyzing daily SSS anomalies obtained from the Global Ocean Reanalysis and Simulations 12v1 (GLORYS12v1) dataset (Jean-Michel et al 2021).Specifically, we compare the roles of mean currents, mesoscale eddies, and air-sea FWFs in shaping SSS extreme metrics.

Datasets
To analyze extreme SSS events and compute salinity budget terms, we utilized daily mean ocean salinities and ocean velocities obtained from the GLORYS12v1 dataset (Jean-Michel et al 2021).GLORYS12v1 is a reanalysis configuration based on the Nucleus for European Modeling of the Ocean model.GLORYS12v1 offers a horizontal resolution of 1/12 • and employs 50 vertical levels with increasing spacing as depth increases.The dataset encompasses the time period of 1993-2020.
To evaluate the net FWFs from the salinity budget equation, we obtained precipitation and evaporation data from the fifth-generation European Centre for Medium-Range Weather Forecasts reanalysis dataset (ERA5, Hersbach et al 2020).ERA5 incorporates comprehensive four-dimensional data assimilation techniques that assimilate in situ and satellite observations.ERA5 provides a spatial resolution of 0.25 • and an hourly temporal resolution.Hourly data were converted to daily values.For this particular study, we utilized ERA5 data from 1993 to 2020, ensuring consistency with the GLORYS12v1 period.

Definition of SSS extremes and a set of metrics for their characteristics
The characterization of high and low SSS extremes in this study follows the methodology described in Liu et al (2023).Daily SSS anomalies are derived by removing a 30 day running mean smoothed 1993-2020 climatology and removing any linear trends.If the anomalies are above (below) a seasonally varying 90th (10th) percentile threshold for each calendar day for a minimum of five consecutive days, then they are considered as high (low) SSS extremes (supplementary figure 1 for details).Their metrics are calculated over 60 • S−60 • N. The mean intensity of an individual SSS extreme event is determined by calculating the average salinity anomaly observed throughout the event.Duration is defined as the number of days from the onset to the demise of an SSS extreme event.The total count of SSS extreme events encompasses the number of such events observed from 1993 to 2020.

Budget analysis
Mixed-layer salinity budget analyses were performed to explore the relative contributions of different physical processes to SSS extremes (Liu et al 2023).Mixed layer salinity is defined as the vertically averaged salinity between the sea surface and the bottom of the mixed layer, which can represent the variations in SSS.The budget equation (modified from Bian et al 2023, Liu et al 2023) is as follows: where subscript a denotes the variable vertically averaged between the sea surface and the bottom of the mixed layer, and subscript −h denotes the quantity at the bottom of the mixed layer.The overbar denotes the mean flow (or a large scale flow) signals obtained by a 3 • × 3 • horizontal running mean filter, and the prime denotes the mesoscale eddy (definition of mean flows and mesoscale eddies followed from Bian et al 2023) fields derived from the deviation from the running mean.S is the salinity, and h is the mixed layer depth, which is defined based on a 0.03 kg m −3 density threshold (de Boyer Montégut et al 2004).u and v are horizontal velocities.w is the vertical velocity, which is derived from the equation of volume continuity.E and P are evaporation and precipitation, respectively.
The left-hand side of equation ( 1) denotes the salinity tendency (Salt Tend).The first right-hand side terms (in parentheses) in equation ( 1) denote the horizontal salt flux convergence by mean flows (SFC-M), the second right-hand side terms denote salt flux convergence by mesoscale eddies (SFC-E), the third right-hand side terms denote entrainment (ENT), and the fourth right-hand side terms denote FWFs.The fifth term denotes the residual, which is the difference between Tend and the sum of SFC-M, SFC-E, ENT, and FWF.The residual includes the vertical eddy diffusive salt fluxes or horizontal eddy diffusive salt fluxes and calculation error.

Spatial patterns of high and low SSS extreme metrics
Over the global ocean, the total counts, duration, and mean intensity of both high and low SSS extremes are not spatially uniform based on results from 1993 to 2020 (figure 1).More frequent high and low SSS extremes mostly occur in the western boundary extension region, the equatorward side of the Antarctic Circumpolar Current (ACC), and tropical oceans.Thus, they can probably be attributed to eddy dynamics (Chelton et al 2007, Hayashida et al 2020) and tropical instability waves (Lee et al 2012).In contrast, less frequent events, below 50 counts, are observed in the western and southeastern South Pacific, northeastern North Pacific, and poleward side of the ACC (figures 1(a) and (b)).
Interestingly, regions with higher counts tend to exhibit shorter durations and vice versa.Such observations align with the finding that the total number of days with high and low extreme SSS occurrences has a similar magnitude at each grid point (see supplementary figure 2).This equality is achieved by multiplying the duration by the total counts.Ideally, at least 10% (1022 d) of the total observational period of 10 227 d from 1993 to 2020 should be occupied by extreme SSS events.
The maximum intensity of both high and low SSS extremes exhibits a distinctive spatial distribution across the global ocean.Specifically, these regions that coincide with major river outlets, the Intertropical Convergence Zone, and western boundary currents are characterized by pronounced SSS variability on multiple time scales, as evidenced by satellite and in situ observations (Liu andWei 2021, Liu et al 2022).Notably, the total counts, duration, and intensity (despite the difference in sign) of high and low SSS extremes exhibit a strikingly symmetric spatial pattern across most of the global ocean (figure 1).
Nevertheless, there is an apparent asymmetry in the occurrence of extreme SSS events.Specifically, in the western tropical Indian Ocean and central tropical Pacific Ocean, the total count of high SSS extremes is twice that of low SSS extremes (figure 1(c)).However, it is noteworthy that low SSS extremes tend to last longer and exhibit a more intense magnitude compared to high SSS extremes, surpassing 50% of their duration and intensity in these regions (figures 1(f) and (i)).This asymmetrical pattern in the metrics of marine extreme events can be attributed to the underlying distribution of SSS anomalies (Schlegel et al 2021), highlighting the non-Gaussian nature of SSS extremes (Bingham et al 2002).

The role of local drivers in the SSS extreme metrics 3.2.1. The dominant driver in the occurrence of SSS extremes
The local driver for a specific SSS extreme event can be determined by integrating the anomalies of each term from the start time (the first day when SSS exceeds and falls below the 90th/10th percentile) to the peak time (when SSS anomalies show maximum magnitude during an extreme event) of SSS extremes.Bian et al (2023) have defined the start time as the local minimum/maximum point immediately preceding the onset of a marine extreme event.
We have compared the outcomes obtained using these distinct methodologies and determined that the primary findings remain largely unchanged (figures not shown), despite variations in the definition of the start time.On a global scale, mesoscale eddies (SFC-E, figures 2(a) and (b)) predominantly contribute to both high and low SSS extremes.In tropical oceans, mean currents take on a dominant role (figures 2(e) and (f)), while near the coast, the influence of the ENT process becomes apparent in generating high SSS extremes (figure 2(i)).FWFs and residual effects generally have a lesser impact, accounting for less than 0.5 of the total counts at each grid point across most of the global ocean (figures 2(c), (d), (k), and (l)).

The impact of SSS budget terms on SSS extreme durations
To examine the influence of each SSS budget term on the duration of extreme SSS events, we have computed the composite of duration when each term dominates (as depicted in figure 2).Additionally, we have divided this value by the average duration to illustrate the impact of each term in the SSS budget on the mean duration (figure 3).The duration of SSS extremes associated with FWF is twice as long as the mean duration across tropical oceans (figures 3(a)    The probability density distribution (PDD) of duration for both high and low SSS extremes is consistent across different drivers (figures 3(c), (f), (i), and (l)).All PDDs exhibit peaks between 5 and 10 d, with the duration associated with mesoscale eddies showing higher values, surpassing 0.4.FWF and mean currents play a more significant role in longer durations, with values exceeding 0.2 between 20 and 50 d.

The impact of SSS budget terms on SSS extreme intensities
The intensity of SSS extremes driven by FWF exceeds the mean intensity by 1.5 times in the eastern tropical Pacific Ocean and western subtropical oceans (figures 4(a) and (b)).Mean currents contribute to stronger intensity in the extratropical region but exhibit a scattered distribution (figures 4(d) and (e)).For SSS extremes caused by mesoscale eddies and ENT, the duration is comparable to the mean intensity at each grid point across most of the global ocean (figures 4(g), (h), (j), and (k)).However, in tropical oceans, mesoscale eddies result in lower intensity for low SSS extremes (figure 4(h)).
The PDD of intensity for both high and low SSS extremes is generally consistent across most drivers, except for ENT (figures 4(c), (f), (i), and (l)).Low SSS extremes associated with ENT exhibit peaks between 0.1 and 0.2 g kg −1 , while high SSS extremes exceed 1 g kg −1 in magnitude.The PDD of intensity associated with FWF and mesoscale eddies shows peaks at approximately 0.2-0.3g kg −1 , whereas low SSS extremes related to mean currents exhibit peaks at 0.3-0.4g kg −1 .Consequently, low SSS extremes connected with mean currents display a stronger intensity.

Special cases: prolonged and intensified low SSS extremes over the central Pacific and the western Indian Ocean
Two specific regions, namely, the central tropical Pacific and the western tropical Indian Ocean (black boxes in figure 1), were chosen to investigate the asymmetry in duration and intensity between high and low SSS extremes.In the central Pacific Ocean, the area-averaged SSS anomalies reveal the occurrence of two prolonged and highly intense negative SSS events (figure 5(a)), spanning from July 1997 to August 1998 and from June 2015 to June 2016, respectively.Similar cases are found in the western Indian Ocean, with two extended and intense events occurring from September 1997 to October 1998 and from October 2019 to September 2020 (figure 5(b)).By removing these two events from the time series, the asymmetry in intensity and duration between high and low SSS extremes diminishes (figures not shown), underscoring the significant role that these individual events play in shaping the mean state of SSS extreme metrics.
Budget analysis reveals that FWFs play a significant role in the freshening of the sea surface during extreme SSS events in both the central Pacific and western Indian Oceans (figure 5(c)).In contrast, the contributions from other terms are either less important or exhibit an opposing effect on SSS tendency.The extreme events in the central Pacific Ocean coincide with the two strongest El Niño events on record (Paek et al 2017, Liu et al 2023), and the two prominent low SSS extremes in the western Indian Ocean are associated with record-breaking Indian Ocean Dipole (IOD) events (using the Dipole Mode Index (DMI) as its proxy, Lu and Ren 2020).This implies that the interannual climate mode may have an important role in influencing the development of SSS extremes through FWF.Composite analysis also confirmed the El Niño and IOD related FWFs drives the strong freshening over the central Pacific Ocean and the western Indian Ocean (see supplementary figure 3).

Discussion and conclusion
Our analysis focuses on examining the characteristics of global SSS extremes between 1993 and 2020 using daily reanalysis data.The results highlight a notable symmetry in the total counts, duration, and intensity of high and low SSS extremes across most of the global ocean (figure 1).On a global scale, mesoscale eddies (figures 2(a) and (b)) predominantly contribute to the occurrence of high and low SSS extremes.High and low SSS extremes associated with different drivers mostly exhibit duration peaks ranging between 5 to 10 d, and peaks in intensity, ranging from 0.2 to 0.3 g kg −1 (figures 3 and 4).Within the tropical Indian and Pacific Oceans, an asymmetrical pattern emerges, with a skew toward longer durations and higher intensities for low SSS extremes.The freshwater fluxes connected to the ENSO and IOD give rise to longer durations and stronger intensities of low SSS extremes (figure 5).
This analysis highlights the significant role of FWFs and ocean dynamics, particularly mesoscale eddies, in shaping the spatial pattern of high and low SSS extremes.The separation between mesoscale eddies and mean flows investigated in this study was achieved using a 3 • spatial filter methodology based on the approach put forth by Bian et al (2023).Previous studies have employed alternative spatial scales in order to differentiate between mesoscale eddies and large-scale motions, such as a 6 • filter (Bonaduce et al 2021) and a range of 10 • -20 • (Delman and Lee 2021).Given the broad nature of the mesoscale peak observed in the spectra of meridional velocity (Delman and Lee 2021), a separation scale that encompasses a larger range could be more suitable for distinguishing between mesoscale eddies and large-scale motions from a global perspective.However, these scales could include instability waves and other mesoscale dynamics.The spatial filter employed in this study is more effective in distinguishing mesoscale eddies and other motions compared to filters of larger scales.However, it still does not ensure a completely precise demarcation.Therefore, caution must be exercised when interpreting the outcomes of this study.
The duration of extreme SSS events mostly occurs over a period of 5-10 d.Storms occurring over 1-2 weeks may also induce abnormal surface freshening due to excessive rainfall or salinification caused by subsurface ENT.Saline sea surface conditions resulting from storms have been observed in the vicinity of major river discharges, such as the Amazon or the northern Bay of Bengal (Reul et al 2021).However, the summed intensity between high and low SSS extremes exhibits negative values (figure 1(i)), indicating a distinction in the causes of SSS anomalies associated with extremes compared to those driven by storms.
This work is an important extension of the diverse body of existing knowledge on marine extremes (e.g.marine heatwaves by Hobday et al 2016).This study marks the first exploration into characterizing the spatial patterns and underlying causes of SSS extremes across the global ocean.Extreme salinity fluctuations have the potential to induce alterations in coupled air-sea systems (e.g.Grodsky et al 2012), and they are one of the key components in the compound marine extremes associated with air-sea interaction (Liu et al 2023).The result of this study could serve as a foundation for the investigation of abnormal changes in the coupled air-sea systems.Moreover, it is important to acknowledge that SSS extremes can have far-reaching impacts on marine ecosystems.For example, stenohaline organisms have restricted salinity tolerance.Both higher and lower salinity extremes in the ocean can result in elevated mortality rates among a wide range of marine species (Zhang et al 2019, Evans and Kültz 2020, Oliveira 2021, Tuwo et al 2021, Roibás-Rozas et al 2022).Thus, results in this study could improve our understanding of the ocean's habitability.

Figure 1 .
Figure 1.Global patterns of extreme SSS metrics and asymmetry in these metrics.(a) Total number of high SSS extremes from 1993 to 2020, (d) mean durations (days), and (g) mean intensity (g kg −1 ).(b), (e), and (h) are the same as (a), (d), and (g) but for low SSS extreme events.The black contours denote the 0.5 g kg −1 isoline of standard deviation derived from daily SSS anomalies with climatological seasonal cycles removed.(c) The ratio of the total counts of high-SSS extremes to low-SSS extremes (unitless for all ratios in this study).(f) The ratio of the mean duration of high-SSS extremes to low-SSS extremes.(i) The sum of the mean intensity between high-SSS extremes and low-SSS extremes divided by the magnitude of the intensity of these two extremes (i.e.SSSa high + SSSa low / SSSa high + |SSSa low | ).Regions with maximum asymmetry in counts, duration, and mean intensity are shown in black boxes over the central tropical Pacific Ocean and western tropical Indian Ocean.

Figure 2 .
Figure 2. The dominant local drivers of SSS extremes.The dominant driver for (a) high SSS extremes and (b) low SSS extremes at each grid point.The ratio of counts is explained by a specific driver in SSS extremes.(c), (e), (g), (i) and (k) High SSS extreme cases.(d), (f), (h), (j) and (l) Low SSS extreme cases.

Figure 3 .
Figure 3.The effect of each term from the salinity budget on the mean SSS extreme duration.The ratio of high/low SSS extreme durations caused by (a), (b) FWF, (d), (e) SFC-M, (g), (h) SFC-E, and (j), (k) ENT to the mean duration at each grid point.(c), (f), (i) and (l) Probability density distribution of SSS extreme durations induced by each term.

Figure 4 .
Figure 4.The effect of each term from the salinity budget on the SSS extreme intensity.The ratio of high/low SSS extreme intensity caused by (a), (b) FWF, (d), (e) SFC-M, (g, h) SFC-E, and (j), (k) ENT to the mean duration at each grid point.(c), (f), (i) and (l) Probability density distribution of SSS extreme intensity induced by each term.Low SSS extreme intensity is multiplied by a negative sign.

HFigure 5 .
Figure 5.The relationship between SSS extremes over the central tropical Pacific Ocean and western tropical Indian Ocean with potential climate drivers.(a) Time series of daily SSS anomalies averaged in the central tropical Pacific Ocean (170 • E-150 • W, 5 • S-5 • N).(b) The same as (a) but for anomalies averaged in the western tropical Indian Ocean (40 • E-70 • E, 8 • S-7 • N).(c) The accumulated anomalies for SSS tendency terms in individual abnormal events in (a) and (b).The blue line denotes the Niño 3.4 index.The green line denotes the DMI index (calculated from Saji et al 1999).The blue shadings denote the occurrence of low-SSS extremes.SSS anomalies are subtracted from the daily climatological seasonal cycle.The climate indices are multiplied by a negative sign for easier visual comparison.