The Salinity of Coastal Waters as a Bellwether for Global Water Cycle Changes

While the global water cycle has been studied previously based on land and open ocean studies, here we use coastal satellite sea surface salinity (SSS) data to show that in global aggregate, SSS variations near coasts are strongly correlated with global water cycle variability driven by El Niño Southern Oscillation (ENSO). This is a significant finding as we demonstrate that open ocean SSS variability is not as sensitive to ENSO and global water cycle variability as the coastal oceans at interannual timescales. Aggregated global coastal SSS could therefore be used as a proxy for detection of changes in the large‐scale cycling of water between the oceans and continents. Moreover, we identify major potential “hotspots” on land and in the coastal ocean that tend to drive global coastal salinity variability, and which may consequently be most sensitive to future physical and biological impacts of water cycle changes on the coastal oceans.


10.1029/2023GL106684
2 of 8 100 km of the coast (Seas, 2011).The coastal ocean is a place that may actually be extremely rich in providing signals of change for the global water cycle, including observing ocean responses to changing hydrologic extremes such as flood events (Fournier, Reager, et al., 2016).In contrast to the open ocean that responds to large-scale changes in patterns of E-P, the coastal ocean is highly sensitive to subtle changes in rivers, occurring at concentrated regions and over shorter durations.
An obstacle to monitoring the coastal oceans is that the observational capacity for river discharge has traditionally been reliant upon a sparse, aging in situ observing network with limited measurement simultaneity, including a sparsity of river gauges (Vörösmarty, 2002).However, there now exists more than a decade of satellite sea surface salinity (SSS) observations, providing new opportunities to investigate the linkages between terrestrial hydrology and coastal zones over a longer record, and to characterize typical behavior for major river systems.Remote sensing enables us to accurately measure SSS starting from 40 km off the coast from space using NASA's Soil Moisture Active Passive (SMAP) mission (Meissner et al., 2018) and ESA's Soil Moisture and Ocean Salinity (SMOS) mission (Reul, Quilfen, et al., 2014;Reul, Fournier, et al., 2014).Freshwater plumes can be accurately tracked in the coastal ocean using measurements of SSS (Fournier et al., 2015;Fournier, Lee, & Gierach, 2016;Fournier, Reager, et al., 2016;Fournier, Vandemark, et al., 2017;Fournier, Vialard, et al., 2017;Gierach et al., 2013;Houndegnonto et al., 2021).While the extent of river plumes is modulated by a suite of factors, including river discharge (Q), oceanic advection, winds, eddies, climate modes (Fournier, Vandemark, et al., 2017;Fournier, Vialard, et al., 2017), seasonal SSS variability near river mouths is typically dominated by Q, and therefore by inland hydrological processes (Fournier et al., 2015, Fournier, Lee, & Gierach, 2016;Fournier & Lee, 2021).As the climate warms, the global water cycle is expected to change (Held & Soden, 2006), with complex impacts for land hydrology (Huntington, 2006) and its influence on the coastal ocean.This makes the coastal ocean a prime candidate for monitoring signals of global change.
Several previous studies have looked at the linkages between coastal SSS and the terrestrial water budget (runoff, land precipitation, land water storage) at the scale of large river basins (Fournier, Reager, et al., 2016;Gierach et al., 2013;Houndegnonto et al., 2021;Liang et al., 2020).Byrne et al. (2023) highlights the strong influence that land precipitation and consequent runoff has on the coastal ocean at seasonal and interannual time scales on the southwestern coast of the United States.In this study, we provide the first aggregated analysis on linkages between coastal SSS and terrestrial water budget globally to explore the appropriateness of coastal SSS as a potential indicator of global water cycle change.Previous studies (Fournier, Lee, & Gierach, 2016;Fournier & Lee, 2021) have demonstrated the inadequacy of in situ and model salinity fields to reproduce the interannual variability observed in satellite salinity products, likely due to in situ sampling representation and river forcing representation in state-of-the-ocean high-resolution global ocean assimilation products.Hence, in this study, we apply satellite observations exclusively to investigate their viability as a unique observational metric.Now that we have more than a decade of satellite SSS observations, analysis of this relationship at interannual time scales is possible.It is well established that El Niño Southern Oscillation (ENSO) drives a major mode of the aggregated global terrestrial water cycle variability (e.g., precipitation on land and aggregated Q) (Cheon et al., 2021;Phillips et al., 2012).If the coastal oceans are a sufficient tool for monitoring global water cycle change, then we would expect to see a measured response to the ENSO signal in coastal SSS observations at the appropriate time scales.In presenting coastal oceans as a candidate for global water cycle monitoring, some driving science questions follow: First, how does coastal salinity vary compared to the open ocean salinity?Second, how does coastal salinity relate to the terrestrial water cycle and ENSO?Finally, what are the hotspot regions influenced the most by coastal salinity variations at interannual time scales?

Materials and Methods
The coastal ocean is typically defined by the waters that lie above the continental shelf.However, in this study, we define "coastal ocean" as the ocean waters within the first 200 km from the coast, waters that are influenced by river inputs, which can extend much further than few tens of kilometers from the coastline, as seen in several studies (Fournier et al., 2015(Fournier et al., , 2019;;Fournier, Lee, & Gierach, 2016;Fournier, Reager, et al., 2016;Fournier, Vandemark, et al., 2017;Fournier, Vialard, et al., 2017) but at least up to 200 km for major rivers.We tested the sensitivity of the analysis of this study to different threshold buffer lengths (ranging from 100 to 500 km in 100 km increments) and we concluded that this threshold selection does not affect our results in a significant way. Figure 1 actually shows that at interannual and seasonal time scales the coastal SSS variations considering the 10.1029/2023GL106684 3 of 8 different threshold only affect the amplitude of the variabilities.In our time series, we still show the coastal SSS considering the different thresholds of 100, 200, 300, 400, and 500 km from the coast.Our quantitative results though only consider the threshold of 200 km from the coast.
The seasonal time series in the following studies are computed as the average of all the monthly data per month from 2011 to 2022.The interannual time series are computed using a low pass filter on the signal with a window of 12 months, after removing the seasonal cycle of the signal.
In order to measure the influence of ENSO on land precipitation and coastal salinity, we perform a least fit at each geographic location (pixel) using a Hilbert transformation of the Multivariate ENSO Index (MEI) to estimate the coefficients a, b, and c in the Equation 1, as performed in Phillips et al. (2012).The Hilbert transformation is a tool in Fourier analysis.In this case, it shifts each Fourier components of the MEI signal by 90°.If there's no lag between the MEI and coastal SSS or land precipitation, at a given location, the coefficient c (see Equation 1) tends to 0.
With A standing for land precipitation or coastal salinity, describing the part of these parameter that can be attributed to the influence of ENSO.The last term corresponds to the imaginary part of a Hilbert transformation of the MEI, meant to capture land precipitation or coastal SSS variations out of phase with the MEI.More information on this method used in Earth Sciences can be found in Phillips et al. (2012) where they investigate the influence of ENSO on water storage, and in Salisbury and Wimbush (2002).
In this study, we try to isolate regions of interest or hotspots that have a significant influence on the globally-averaged coastal seasonal and interannual time series.These regions are highlighted in Figure 2a.

Data
We use the multi-mission Optimally Interpolated SSS data set.This is a level 4 product on a 0.25° spatial and monthly temporal grid starting in September 2011 (Melnichenko et al., 2021).The product is the monthly mean of the level 4 OISSS data set using three satellite missions: the NASA Aquarius/SAC-D, SMAP and ESA SMOS using Optimal Interpolation with a 7-day decorrelation time scale.This data set is produced by the International Pacific Research Center (IPRC) of the University of Hawaii at Manoa in collaboration with the Remote Sensing Systems, Santa Rosa, California, and distributed by the PO.DAAC (https://doi.org/10.5067/SMP10-4UMCS).
We exclude the high latitudes, above ±65°N.
We also use the monthly V06 Integrated Multi-satellitE Retrievals for GPM (IMERG) Final Precipitation (P) L3 product that is on a 0.1 × 0.1° spatial and daily temporal grid (Huffman et al., 2019).The precipitation estimates are computed from the various precipitation-relevant satellite passive microwave (PMW) sensors comprising the GPM constellation.This product is distributed by the Goddard Earth Sciences Data and Information Services Center (GES DISC; https://disc.gsfc.nasa.gov/datasets/GPM_3IMERGM_06/summary?keywords=gpm%20imerg).We exclude the high latitudes, above ±65°N.
We use different global river discharge data sets that are available from 2002 to 2022.These data sets use an ocean mass balance approach that is described in H. A. Chandanpurkar et al. (2017).The data was created using ocean altimetry (AVISO/DUACS), ocean mass from gravimetry (GRACE), ocean steric information (EN4), combined with ocean precipitation (GPCPv2.2,CMAP), ocean evaporation (OAFLUX), and also estimates of P-E calculated from ocean atmospheric moisture budget (MERRA-2, ERA-5).The discharge includes runoff from all land masses including the ice sheets.The data was created by H. Chandanpurkar, and is an updated version from H. A. Chandanpurkar et al. (2017) until 2022.There are eight different data sets that have been averaged in this study and a spread among the data sets was computed.

Results
The first part of our investigation is to understand better the variability of coastal SSS at seasonal and interannual scales and mostly how it compares to the variability of the open ocean SSS. Figure 1  After studying the global seasonal and interannual variability of coastal and open ocean SSS, it is important to understand if there are regional hotspots where these variabilities are higher.In Figure 2a, we show the map of the amplitude of the SSS and land precipitation seasonal cycle at each pixel.Overall, the amplitude of the seasonal cycle is much larger in the coastal ocean (reaching almost 5 psu) than it is in the open ocean.The seasonal SSS amplitude is also larger in some hotspot regions such as near river mouths such as the Amazon River, Congo and Niger rivers in the Gulf of Guinea, Ganges-Brahmaputra River in the Bay of Bengal and around the Maritime Continent.On land, the amplitude of the seasonal precipitation is larger in the tropics in some hotspot regions including in the Brahmaputra, Amazon, Congo and Niger basins as well as over Borneo, regions adjacent to high SSS seasonal amplitude coastal regions.The different hotspot regions are displayed on Figure 2a. Figure 2b show the season of SSS minimum and land precipitation maximum seasonal cycle.In areas of high land precipitation variabilities (Figure 2a), the season of peak is in boreal summer.As expected, the resulting peak in seasonal SSS variations is also during boreal summer as freshwater from land to reach the coastal ocean takes about a month or 2 (Fournier & Lee, 2021).
We then map the interannual variabilities in order to highlight the regional hotspots and for that we use the standard deviation as a metric for interannual variabilities.Figure 2c represents the maps of the standard deviation of the interannual variabilities of SSS and land precipitation at each pixel.River mouth regions represent hotspot regions of strong interannual SSS variabilities (Gulf of Mexico, Northwestern Tropical Atlantic Ocean, Gulf of Guinea, Bay of Bengal, La Plata River mouth), with standard deviations reaching more than 1 psu while the standard deviations are close to zero in the open ocean.On land, the standard deviations of the interannual land precipitation are larger in some hotspot regions in the tropics and can reach 35 mm/month in the Mississippi Basin, 80 mm/month in India, 55 mm/month in the Amazon Basin, over 90 mm/month in the Maritime continent, and 45 mm/month in the La Plata Basin, all regions adjacent to large standard deviations in interannual SSS variabilities.Most of the SSS interannual variability is located in the coastal regions, adjacent to land regions of high P interannual variability variances.
We then try to understand the link between the interannual variability of coastal SSS and the terrestrial water cycle, as well as one of its primary drivers.Figure 3 shows the relationship between ENSO and the interannual variations of global land precipitation, global river discharge and global coastal SSS.To quantify the variability and intensity of ENSO we use here the MEI.In these time series, during the exceptional El Niño event of 2015 (maximum MEI), we can notice a decrease in the interannual global land precipitation followed by a decrease in interannual global discharge, and an increase in the coastal SSS signal.This signal in global coastal SSS in response to ENSO variability has never been shown before.We can see in Figure 3 that the signal in global coastal SSS, river discharge and land precipitation associated with El Niño is very well correlated with the MEI, with the MEI being the driver.The maximum correlation between the interannual variability of the MEI and (a) the global land precipitation is about −0.46 with no time lag (p-value of 0.01), (b) the global river discharge is about −0.76 (p-value of 2.10 −5 ) with no time lag, consistent with Syed et al. (2010) that shows that interannual variabilities are mainly driven by ENSO, and (c) the global coastal SSS ranges from 0.73 (p-value of 6.10 −5 ) (for SSS within 300 km from the coast), to 0.74 (p-value of 4.10 −5 ) (for SSS within 200 km from the coast), to 0.75 (p-value of 3.10 −5 ) (for SSS within 400 km from the coast), to 0.77 (p-value of 1.10 −5 ) (for SSS within 100 km or 500 km from the coast), with no time lag.Typically, freshwater from land to reach the coastal ocean takes at least a month, or more for larger rivers such as the Amazon, Congo or Ganges-Brahmaputra rivers.Therefore, a lag of at least a month is to be expected between land P and the corresponding changes in SSS.Correlations are slightly lower at 1 to 2-month time lags (lower by 0.02 at 1-month time lag and by 0.04 at 2-month time lag) between MEI and discharge, precipitation or SSS at interannual time scales but still significant (based on the p-value).Here as we are looking at interannual timescales, the effect of seasonal lags gets reduced to be negligible.A similar analysis was conducted with other climate modes such as the Indian Ocean Dipole with the Dipole Mode Index and North Atlantic Oscillation (NAO) with the NAO index.However, the correlations between any of the additional interannual indexes and coastal SSS, land precipitation or discharge at global scale, are very low and not significant compared to the correlations with the ENSO index, showing the evidence of that the influence of ENSO at a global scale is more dominant.
After establishing the high significant correlations at interannual time scales between coastal SSS or land precipitation and MEI globally, it is important to understand if there are particular regional hotspots where the influence of ENSO is larger.Therefore, to measure the influence of ENSO on the interannual land precipitation or SSS, we compute a least square fit at each geographic location using a Hilbert transformation of the MEI (see Section 2 for more details).As pointed out in Chao and Chung (2019), results from a Hilbert transform by itself can be difficult to interpret, however, this method still shows some possible regions of high significant correlation between land precipitation or SSS and ENSO.Results are shown in Figure 4.The hotspots in the ocean corresponds to major river plume regions such as the Amazon River plumes, the Gulf of Mexico, the Bay of Bengal, the La Plata River plume, the Gulf of Guinea as well as the Maritime Continent and the Tropical Pacific Ocean.Most of these hotspots are located in coastal regions.On land, the hotspots are located in Tropical regions, mostly in the Amazon basin and Maritime Continent.

Conclusions
Now that we have more than a decade of satellite SSS observations, we are able to study longer time scale variations of salinity.The major finding of this study is that coastal SSS variations at interannual time scales are sensitive to changes in the terrestrial water cycle and ENSO (Figure 3), unlike open ocean SSS variability that demonstrates significantly different variability (Figure 1).While the water cycle has been studied mainly based on land studies or open ocean studies, we show that coastal SSS can also be used to monitor the water cycle at interannual time scales.
ENSO is projected to change (Cai et al., 2015) in intensity/frequency with climate change.Following our major finding, we can expect that changes in ENSO would then have potential impacts on the coastal ocean.More generally, with global warming, a change in the water cycle is expected and therefore, as the coastal ocean is largely impacted by the terrestrial water cycle,  changes in the impacts on the coastal ocean are expected.The resulting impacts of changes in the water cycle on the coastal ocean via changes in runoff can have major consequences on physical, biological, optical and chemical Earth system processes at larger scales (Stephens et al., 2020).For example, consequences of changes in supply of freshwater to the coastal ocean and beyond can have impacts on air-sea interactions such as convection and rainfall (Shenoi et al., 2002), intensification of hurricanes (Grodsky et al., 2012;Reul, Fournier, et al., 2014;Reul, Quilfen, et al., 2014), and global ocean circulation (Lee et al., 2019).Also, the supply of nutrients, sediments, organic and inorganic matter into the coastal ocean by discharge has been associated with altering biogeochemistry, and ecological and biological activities (Fournier et al., 2015;Hickey et al., 2010) with subsequent impacts on fisheries and ecosystems (Shore et al., 2021) and on carbon cycle fluxes (Körtzinger, 2003) in the greater regional ocean.Therefore, because the coastal ocean is sensitive to changes affecting the land, it is of important to monitor the terrestrial water cycle changes and this study shows that salinity could be used as a proxy to infer changes affecting the land at interannual scales.
Finally, we isolated major hotspots (shown in Figure 2a) in the coastal ocean that are driving the coastal salinity variability at interannual scales and its relationship with the terrestrial water cycle and ENSO (Figure 4).These regional hotspots might be more sensitive to future changes affecting the terrestrial water cycle and ENSO.Future related studies could therefore focus on these hotspots.

Figure 1 .
Figure 1.Interannual time series from September 2011 to February 2022 of global (blue) open ocean sea surface salinity (SSS) and (red shades) coastal SSS (within 100-500 km from the coast).In the top left corner, the seasonal time series of global open and coastal ocean SSS are shown.For visualization purposes, the time mean was deducted from each time series.Globally-averaged coastal SSS variability is higher at both seasonal and interannual scales compared to open ocean SSS variability.
shows the global averaged interannual time series of both the global open ocean (500 km from the coast and further) and global coastal (from 100 to 500 km from the coast) SSS as well as the global averaged seasonal time series (in the top left corner).Overall, the amplitude of the interannual and seasonal SSS variabilities are much larger in the coastal ocean than in the open ocean.Also, the phase of the variability in each case is also different showing that the nature of SSS variability is itself different in the coastal ocean than it is in the open ocean.The interannual variabilities reach 0.11 psu in the coastal ocean (within first 200 km from the coast) and only 0.02 psu in the open ocean from 2014 to 2016 with an opposite response in SSS in the open ocean (with a lag of maximum correlation (−0.60 with a p-value of 0.001) of 6 months between the interannual variability of coastal and open ocean SSS).Similarly, while the amplitude of the global open ocean SSS seasonal cycle is small (0.02 psu), the amplitude of the global coastal SSS seasonal cycle is larger, about 0.16 psu.In the coastal ocean, the SSS reaches a global maximum (34.50 psu) around March and a global minimum (34.34 psu) around September.In the open ocean, the cycle is reversed with SSS reaching a minimum (34.97 psu) from February to April and a maximum (34.95 psu) from July to October.There is also a lag of maximum correlation (−0.97 with a p-value of 9.10 −8 ) of 6 months between the seasonal variability of coastal and open ocean SSS.This figure shows that the open ocean SSS does not have as much seasonality or interannual variability and that the coastal ocean is more sensitive to variations of SSS seasonally and interannually (probably associated with runoff as fresh tongues extend mostly from river mouths).The variance in the first 200 km from the coast is 31 times higher than the variance in the open ocean (200 km from the coast and further) at monthly time scales.At seasonal and interannual time scales, the variance in the first 200 km from the coast is respectively 54 and 23 times higher than the one in the open ocean.

Figure 2 .
Figure 2. (a) Maps of the seasonal cycle's amplitudes of sea surface salinity (SSS) and land precipitation at each pixel.The red contour delimits the coastal ocean from the open ocean as defined in this study (200 km from the coast) and different regions of interest are highlighted.(b) Maps of the season of SSS minimum and land precipitation maximum.The large seasonal variability of coastal SSS and land precipitation is concentrated in few hotspots.(c) Maps of the standard deviation of the interannual variabilities of SSS and land precipitation at each pixel.Most of the SSS interannual variability is located in the coastal regions, adjacent to land regions of high P interannual variability.

Figure 4 .
Figure 4. Map of the amplitude of the Hilbert transform between Multivariate ENSO Index (MEI) and the interannual sea surface salinity (SSS) or land precipitation.The high correlation between interannual coastal SSS or land precipitation and MEI is mostly concentrated in few hotspots (shown in purple and green).

Figure 3 .
Figure 3. Interannual time series of (a) the Multivariate ENSO Index (MEI), (b) global land precipitation, (c) global river discharge, and (d) global coastal sea surface salinity (SSS) (the y-axis of the precipitation and discharge times series are flipped for visualization purposes).The light green shading highlights the 2015-2016 El Niño.Land precipitation, discharge and coastal SSS are highly correlated with MEI at interannual time scales.