Intensification of the global water cycle and evidence from ocean salinity: a synthesis review

The ocean plays a dominant role in the global water cycle. It is the center of action for global evaporation and precipitation and supplies the moisture that falls as continental precipitation. It also acts to some extent as nature's rain gauge, as it tells us about the long‐term changes in the global water cycle through monitoring of the changes in ocean surface salinity. As climate warms, the global water cycle is expected to intensify as a result of the strong nonlinear dependence of water vapor pressure (moisture‐holding capacity) on temperature. Such change is of great concern, as it has profound socioeconomic impacts throughout the globe. Despite the evidence of an intensified global water cycle, two important questions remain: What is the pattern of the warming‐induced intensification of the water cycle? and What is the rate of intensification? Our article provides a synthesis review of recent progress in diagnosing and understanding the changes in both the global water cycle and ocean salinity in recent decades. Targeted numerical ocean model experiments are also reviewed to provide insights into the response of salinity to the changes in evaporation‐minus‐precipitation flux, meltwater runoff, and ocean warming.


Introduction
The ocean is at the heart of the Earth's global water cycle: it is the source of 86% of global evaporation and the receiver of 78% of global precipitation. 1,2 The 8% imbalance in the freshwater exchange over the ocean surface reflects the fact that it serves as an important, remote source of continental precipitation. 3,4 The water vapor transported from the ocean into the atmosphere and then to the continents influences where and when floods and extreme weather occur on land. [5][6][7] The continental precipitation flows over the ground as surface runoff, some of which eventually reenters the ocean. This surface runoff together with meltwater runoff from snow and ice reestablishes the balance of the freshwater budget in the ocean.
The cycling of freshwater between evaporation (E), precipitation (P), and runoff (R) leaves a strong imprint on ocean salinity. 8,9 P and R bring freshwater into the ocean, diluting the salt content of the surface water. E from the ocean's surface leaves salt behind, increasing salinity. The exchange of freshwater between the atmosphere and ocean acts in concert with ocean circulation and mixing, raising the sensitivity of the salinity distribution to the balance between E, P, and R. On a time-mean basis, surface waters of high salinity are generally located in the subtropical regions where E exceeds P, and surface waters of low salinity are located in the tropical and high-latitude regions, where P and/or R exceeds E ( Fig. 1A and B). Thus, ocean surface salinity integrates the complex multifactorial variations in the water cycle, acting to some extent as nature's rain gauge. 10,11 Growing evidence indicates that the global water cycle 12,13 has responded to climate warming in the past half century, 3,[14][15][16] demonstrated by an increase in global atmospheric moisture content, 17 the frequency of heavy rain events, 18,19 evaporation, 20 and evapotranspiration. 21 Ocean salinity observations doi: 10.1111/nyas.14354 support this intensification, showing that surface salty waters have become saltier in the subtropical oceans and surface fresh waters fresher in the tropical rain belt. [22][23][24][25][26] Despite the strong evidence, two of the most important questions remain: What is the pattern of the warming-induced intensification of the water cycle? And what is the rate of intensification?
Accurate quantification of the changes in E and P is central to addressing these questions, but these water cycle components are notoriously difficult to observe. E cannot be directly remotely sensed as it cannot be detected by electromagnetic radiation. Estimation of E has large uncertainty because it is computed from bulk parameterizations using near-surface meteorological variables and SST as input. [27][28][29] P has long been inferred from visible infrared and passive microwave (PMW) retrievals. 30 The launch of the Global Precipitation Measurement mission in 2014 has significantly improved global observations of P. 31 Still, the potential of ocean surface salinity as a rain gauge to provide an alternative measure of changing rainfall and evaporation is thus appealing. The trends in ocean surface salinity appear to hold important insights into the long-term variations of the water cycle and the effects of climate change. Changes in surface salinity also impact surface seawater density, which can, in turn, alter surface and interior ocean circulation.
This article provides a synthesis review of recent progress in diagnosing and understanding the changes in both the global water cycle and ocean surface salinity. Targeted numerical ocean model experiments are reviewed to provide some insights into the response of salinity to changes in mean evaporation-minus-precipitation (E−P), meltwater runoff, and ocean warming. However, the effect of surface runoff owing to freshwater discharge from global rivers is not included. River input impacts the ocean locally 32 but has a limited contribution to the ocean's overall freshwater budget. 33,34 It is expected that our knowledge of global surface runoffs and their interactions with coastal and open oceans will be significantly advanced with the planned launch of the Surface Water and Ocean Topography (SWOT) Mission 35 in 2021.

Data and methods
Our review includes data analysis and ocean model simulations. Data analysis focuses on the changes in ocean water cycle and salinity occurring in the past three decades spanning from 1988 to 2017. The period is marked by the availability of increased and improved satellite observations. In particular, the PMW observations from the series of Special Sensor Microwave Imager (SSM/I) and the Special Sensor Microwave Imager Sounder devices onboard Defense Meteorological Satellite Program satellites 36 have allowed, for the first time, the retrieval of surface meteorological variables that are needed for the parameterization-based computation of E. 28 PMW observations are also superior to previously used visible infrared observations for estimating instantaneous precipitation, resulting in improved spatial and temporal variability of satellite-based P products. 37 Reanalyses of the global atmosphere using data-assimilating models have shown marked improvements when PMW observations are included. [38][39][40] Four sets of E-P fields are used here to construct a synthesized view of the past changes in the ocean water cycle by consolidating common patterns among the products. One is a combined product, using E from the second-generation Objectively Analyzed air-sea Fluxes (OAFlux2) products at 0.25°resolution 29 and P from the Global Precipitation Climatology Project (GPCP). 30 The other three products are from the latest atmospheric reanalyses: the fifth generation of atmospheric reanalysis from the European Centre for Medium-Range Weather Forecasts (ERA5), 41 the Japanese 55-year Reanalysis (JRA55), 42 and the Modern-Era Retrospective analysis for Research and Applications, version 2 (MERRA-2). 43 Historical in situ salinity and temperature observations are spatiotemporally inhomogeneous. 44 Long-term changes in the sea surface salinity (SSS) have been estimated using both linear trend analysis 22,23,25,45,46 and mean differences between the Argo era (2003-present) and historical ocean climatologies. 47,48 Two monthly gridded objective analyses have also been constructed from historical archives of subsurface ocean temperature and salinity profiles. They are the Ishii dataset 49 and version 4 of the Met Office Hadley Centre "EN" dataset (EN4). 50 Droghei et al. 51 suggested that datasets constructed from in situ observations may be sufficient for studies on the large scales (∼1000 km) and at low temporal resolution (∼seasonally).
Diagnosis of the long-term changes in the water cycle and salinity is based on linear trends in the annual mean fields. These trends were estimated as linear fits against time and then multiplied by the 30-year analysis period (1988-2017) to show the total change. To determine the statistical significance of the trends, Student's t-test was performed with an overall two-tailed confidence interval of 90%. Climate varies over all spatial and temporal scales from the diurnal to interannual to multidecadal and millennial. Detecting long-term changes in mean conditions requires long-term averages to overcome the effects of natural variability (climate noise). It is recommended by the World Meteorological Organization 52 that a minimum duration of 30 years is necessary for determining the statistical properties of climate change. The 30-year analysis period in our review meets this minimum requirement. However, it may not be long enough to suppress the signal associated with longer-period climate variability, such as the Atlantic Multidecadal Oscillation (AMO). Unfortunately, the uncertainties of E-P products before the 1980s have much larger uncertainties. This makes it difficult to distinguish whether any difference in trends between the 30 years and a longer period going back to the 1970s or earlier is caused by the difference in actual trends over different periods or is due to the uncertainties in the E-P before the 1980s.

Mean E-P and its predicted changes
The spatial distribution of net E-P flux (Fig. 1A) captures the essence of the water cycle operation over the ocean: the variation of E-P with latitudes is a result of the atmospheric transport and condensation of water vapor evaporated from the ocean. The most evaporative regions of the global ocean are in the subtropics, where dry, subsiding air of the subtropical high-pressure centers feeds the trade winds that induce strong evaporation and low precipitation (E > P). The wettest regions on the planet are in the deep tropics, where the trade-wind convergence forces the water-laden air to rise in altitude, forming clouds and falling as rain in the Intertropical Convergence Zone (ITCZ) near the equator in all the ocean basins and the South Pacific Convergence Zone (SPCZ) in the western South Pacific. The higher latitudes (poleward of 45°north and south) are also net precipitation zones (P > E), where atmospheric moisture convergence is supplied by midlatitude storm systems and fronts. In general, the patterns of E-P are strongly zonal, except for the North Indian Ocean, where E dominates in the Arabian Sea and P dominates in the Bay of Bengal and to a lesser extent in the South Pacific.
The mean E-P pattern is expected to intensify in a warming climate according to the Clausius-Clapeyron (CC) relation. 12 Under constant relative humidity, the CC relation states that saturation water vapor pressure (the water-holding capacity of the air) increases at a rate of approximately 7% per degree Celsius of warming (hereafter 7%°C −1 ). This rate of increase in atmospheric moisture content has been confirmed during the past few decades. 17,53 Since precipitation and evaporation are tightly linked to the vapor pressure of water, it is predicted that the dry subtropical regions of net evaporation will become drier and the wet tropical and higher latitudes of net precipitation will become wetter. 13 This "dry gets drier and wet gets wetter" (DDWW) paradigm can be derived from the following atmospheric moisture budget equation: 54 where w = 1 g ∫ p s 0 qd p is the integrated precipitable water within the troposphere, q is the atmosphericspecific humidity, g is gravity, and v is the horizontal wind vector. The first term on the left-hand side represents the change in storage of total moisture and is small on monthly and longer time scales. Hence, long-term changes in surface net E-P flux are determined primarily by the long-term changes in the divergence of the vertically integrated moisture flux. Using δ to denote the change in each variable induced by the warmer climate, Eq. (1) can be expressed as: which states that the two major contributions to δ(E -P) are changes in the moisture content of the atmosphere (the first term on the right-hand side (rhs)) and changes in the atmospheric circulation (the second term on the rhs). Assuming that the large-scale atmospheric circulation remains unchanged (i.e., δv ≈ 0) and δq increases by ∼7%°C −1 (i.e., δq ≈ q × 7%/°C), Eq. (2) yields the following result: Equation (3) reveals an important relation: the change in E-P scales with E-P, with the scaling coefficient being the CC rate of approximately 7%°C −1 . Accordingly, the pattern of E-P would remain the same, but the magnitude would become more extreme leading to amplification of the differences between dry and wet regions. 13 Climate models show that the enhancement of the mean E-P pattern is a robust feature in anthropogenic forcing simulations. Mean precipitation generally increases in the tropics and extratropics and decreases in the subtropics, and precipitation extremes increase across almost the entire globe. 13,19,55,56 Seager et al. 57 deemed that the DDWW is essentially a certainty in future changes in the water cycle, as long as relative humidity remains approximately constant. It is gen-erally expected that changes in relative humidity are small on large scales, 58 although locally, there may be some change in relative humidity caused by the changes in large-scale circulation patterns. 53 Despite the success of the DDWW paradigm in predicting the large-scale hydrological response to climate warming, many studies have articulated the need to extend the DDWW hypothesis to account for a broader range of observed and projected changes in the global water cycle. Of particular interest is the assumption that the large-scale atmospheric circulation will remain unchanged. Discrepancies exist in using the DDWW paradigm to explain an observed large-scale decline in P in the subtropical dry zones. 59, 60 Held and Soden 13 suggested that the "dry-get-drier" in the subtropics is largely accomplished by a decline in P. Thus, all else unchanged, P will be reduced in regions of P < E, including the subtropical dry margins of both the tropical rain belt and the midlatitude storm tracks (Fig. 1A). However, most climate model studies have shown that the decline in P is primarily driven by a poleward shift of subsidence associated with the expansion of the Hadley cell 61 and a poleward shift of the storm tracks. 62 In contrast to the thermodynamic "dry-get-drier" mechanism, this less known dynamical response acts mainly to reduce P on the poleward side of the subtropical P minima, and not on the dry margins of the tropical rain belt. 60 Despite the successful prediction of the subtropical decline in P by both the "dry-get-drier" and poleward expansion mechanisms, the subtropical area affected by the two mechanisms differs owing to the differences in interpreting the role of the large-scale atmospheric circulation.
Discrepancies from the DDWW paradigm also occur in the E-P response in climate models. The rate of change in modeled E-P is generally less than predicted by CC, at around 2%°C −1 (Refs. 12 and, 63-67), although a rate of total water vapor increase of 7%°C −1 has been confirmed by observations. 17,53,68 The discrepancy is often explained in terms of the thermodynamic energy constraints on the water cycle, which force changes in global P to accord with the availability of energy flux rather than with changes in temperature. 12,63,69 Broadly, the global mean P in climate model projections is dominated by three terms of the global energy budget: the radiative cooling at the top of the atmosphere, the net short and longwave radiative energy flux at the surface, and the surface evaporative energy flux (mostly restricted to the oceans). Two processes are regarded as the main limiting factors for capping the hydrologic sensitivity to anthropogenic warming at a rate of 2%°C −1 . One is the limited availability of atmospheric surface forcing that can be partitioned into E. The other is the short response time scales of the atmosphere that relate to the dissipation of the latent heat released from the formation of precipitation. 67,70,71 Estimated changes in E-P and uncertainties In the past, three decades since 1987, satellitederived products along with improved atmospheric reanalyses have been significant sources of data for assessing and understanding the operation of the global water cycle in the present climate conditions. During this period, global temperatures have continued to rise. The satellite-derived ocean surface air temperature (T a ) record obtained by the OAFlux2 analysis 72 shows that near-surface air temperature has steadily increased, at a rate of 0.15 ± 0.03°C per decade since the late 1980s ( Fig. 2). At the same time, the average global sea surface temperature (SST) has also increased, at a similar rate of 0.14 ± 0.03°C per decade. Both trends are statistically significant at the 90% confidence interval based on Student's t-test, and notably, match in shortterm detail. These trends are consistent with those deduced from the global land and ocean surface temperature dataset constructed by the Goddard Institute for Space Studies (GISS). 73,74 The GISS dataset shows that the average global temperature on Earth has increased by about 0.8°C since 1880. Two-thirds of the warming has occurred since 1975, at a rate of roughly 0.15-0.20°C per decade.
Changes in E and P over the global ocean in response to this warming are given by linear trend estimates performed for four different time series and converted to a percentage of change per degree C (%°C −1 ) (Fig. 3A-D). Despite large year-to-year fluctuations, both E and P are consistent in depicting a rapid intensification of the ocean water cycle in the 1990s but deviate from each other after 2000. Nonetheless, trends during the 30-year period are upward in the four time series shown, and these trends are statistically significant at the 90% confidence interval except for P from GPCP. The latter has a trend of 2 ± 3%°C −1 , which is nonsignificant using Student's t-test.
The three products, OAFlux2, JRA55, and MERRA5, have a similar increase rate for E, ranging between 8 and 9%°C −1 with the uncertainty of about 3-4%°C −1 . This rate of change is slightly higher than the CC rate but within the statistical margin of error. By comparison, ERA5 has an exceptionally large increase in both E and P, at a rate of about 18 ± 3%°C −1 , and so does the P in JRA55, at about 14 ± 3%°C −1 . The rate of increase in P in the other two products is considerably weaker, at 2 ± 3%°C −1 for GPCP and 5 ± 3%°C −1 for MERRA2. The products have a better agreement in the rate of increase for E than for P.
Linear trends in the E-P time series reflect a balance of the trends between E and P ( Fig. 3E-H). Three products, OAFlux2-GPCP, ERA5, and MERRA2, have positive trends over the 30-year period, showing that the ocean water cycle has intensified correspondingly to the warming. On the other hand, JRA55 is the only product that has a negative trend, driven by a faster increase in P. In the three products, where the E-P trend is positive, the rate of increase in E is larger than that of P. The rates of change of E-P are not converted to percentage per degree Celsius. The reason is that the conversion needs to use the average global E-P budget as scaling, but the four products have large differences in budgets, from 13 to 26 cm/year, which is a major source of bias for the conversion. As discussed in Schanze et al. 34 and Yu et al., 75 both satellitebased and reanalyzed E-P products have difficulties achieving a balanced budget over the global ocean.
Spatial patterns of E-P in the four products ( Fig. 4A-D) show that the common features of the changes are the intensification of the ITCZ rainfall in the tropical Atlantic and Pacific as well as in the sub-Antarctic Pacific sector, and the equatorward expansion and intensification of the subtropical dry zone in the South Pacific and South Atlantic. Overall, the trend patterns of OAFlux2-GPCP are similar to those of ERA5, although the two products disagree with the rate of change in global means. Differences among the four products are large. For instance, there is a weakening of the subtropical E-P maxima in North Atlantic in MERRA, but a strengthening in JRA55 and ERA5. OAFlux2-GPCP shows barely any change in this region. The trend patterns in JRA55 are dominated by an excessively large intensification of the tropical rain band, which explains the negative trend in average global E-P.
The better agreement in the decadal changes in E than in P underlies the fact that the two water cycle components are governed by very different dynamical processes. E is a continuous and slow process, while P is highly intermittent, spatially strongly varying, and governed by both large-and small-scale processes, such as cumulus convection. 2 The reanalyzed P is sensitive to the use of different cumulus parameterizations. 76 In addition, the 2.5°8 gridded GPCP monthly product is constructed from blending in situ station observations and satellite-derived information, 30 and thus has a large uncertainty due to limitations in algorithms and its coarse spatial resolution, which limits the ability to resolve small-scale rain events. 37 Additionally, the temporal coverage of ∼30 years may be too short so that trends are likely affected by natural decadal modes of variability. These difficulties contribute to large deviations in long-term trends of P among products and hinder accurate assessments of the changes in the global water cycle.

Observed amplification of the mean SSS
Long-term changes in the SSS [22][23][24][25][45][46][47] have been constructed from historical salinity databases by many studies using various approaches. Three studies, Boyer et al., 23 Durack and Wijffels, 25 and Hosoda et al., 47 focused particularly on multidecadal trends of near-surface salinity on a global scale, although they differ in analysis periods and methodologies of processing historical data archives. 77 Despite the differences, the patterns of salinity change from the three studies showed remarkable similarities with the climatological mean salinity patterns (Fig. 5). The long-term SSS trends are negative (surface freshening) in the P dominated regions, such as the Indo-Pacific warm and fresh pool, and the regions under the ITCZ, and positive (surface salinification) in the E dominated subtropical regions. This amplification of the mean SSS pattern has reinforced the marked contrast between the fresher Pacific and the saltier Atlantic, strengthening the existing gradients of the SSS spatial distribution. 77 The long-term increase of the subtropical SSS maxima and decrease of the tropical SSS minima provide an independent line of evidence demonstrating that the water cycle has responded to climate warming in a way as predicted by the DDWW paradigm. 12,13 Using the observed salinity trends to fingerprint the amplified water cycle has gained increasing prominence in recent years. 26,[77][78][79][80][81][82][83] However, the relationship between SSS and E-P is often complicated by upper-ocean dynamics by virtue of the salt conservation equation. 80,84 Even in regions Black contours express the associated mean SSS for the analysis period. Broad-scale similarities exist between each independent estimate of long-term change, suggesting that an increase in the spatial gradients of salinity has occurred over the observed record. However, regional-scale differences are due to differing data sources and analysis methodologies.
that are dominated by heavy rainfall, such as the ITCZ across the entire tropical ocean, or regions dominated by strong evaporation, such as the subtropical evaporative regions, the change of SSS is controlled primarily by processes internal to the ocean, such as advection and mixing. Yu 84 pointed out that E-P forces salinity anomalies but does not damp them, which is unlike surface heat flux that serves as both forcing and damping mechanisms for SST. 85 Salinity anomalies tend to be more persistent than SST, and are more strongly influenced by upper-ocean circulation and mixing. 86,87 Vinogradova and Ponte 81 examined the use of salinity as a fingerprint of the recent intensification of the global water cycle. They found that the decadal trends in global SSS patterns over the recent shorter period (∼20 years) do not follow the expected patterns of amplified salinity contrasts, as the seas do not always become saltier if it rains less, nor do they always become fresher if it rains more. It appears that variability in SSS alone does not track the changes in E-P, at least on subannual-bidecadal time scales. Hence, the rain gauge concept warrants a more evolved formulation [80][81][82][83][84] requiring taking all oceanic processes (e.g., horizontal and vertical advection, and mixing) into consideration.

Understanding the changes in the subtropical SSS maxima
The subtropical SSS maxima (hereafter SSS-max) 88 are found to be especially valuable as a subject for studying the relationship between ocean salinity, ocean dynamics, and the global water cycle. The SSS-max areas are situated in the middle of the subtropical gyres (Fig. 1B), which are well away from the influence of western or eastern boundary currents, or equatorial current systems, such as the Equatorial Countercurrent or the Equatorial Undercurrent. They are located near the mean locations of the subtropical atmospheric high-pressure zone, meaning weak winds. 89 Being local maxima in SSS, horizontal gradients are small and thus horizontal advection terms in the salt conservation equation are small, too. 90 This leads to the fact that large-scale oceanic mean and eddy flows are weak and that the contribution of ocean dynamics to the salinity balance at the surface is small. Surface freshwater flux forcing, or E-P, has been found to be the major term in the upper-ocean salinity balance [90][91][92][93][94][95] in these regions and thus changes in E-P can have a large impact on their shape, size, position, and configuration of these SSS-max areas. In addition, the SSS-max are the source of the so-called "River of Salt," 96,97 which is a high salinity tongue at the depth of 50-300 m that propagates equatorward with a distinct signature visible for a long distance into the interior. 98 This water mass is often given the generic name "Subtropical Underwater" (STUW) 99 and is formed by subduction of the SSS-max in later winter.
Recent studies by Yu et al. 100 and Liu et al. 101 showed that the STUW in the North Atlantic could carry the information about how the surface forcing has been changing, as the STUW properties are determined by its last contact with the atmosphere in the formation region. Yu et al. 100 found that an increase in the size of the SSS-max in the subtropical North Atlantic in the past 50 years 22,23,25 was caused by a poleward expansion of the feature's center in association with the widening of the Hadley circulation. [59][60][61]102,103 As a result of the SSS-max expansion, the ventilation zone of the STUW (marked by the location of the 36.7 isohaline) has shifted northward, by roughly 0.35 ± 0.12°l atitude over a 34-year (1979-2012) period, and also expanded westward into the Sargasso Sea ( Fig. 6A  and B). This led to an increase in the subduction at a rate of 0.29 ± 0.07 Sv per decade and the subsequent increase in the thickness of the STUW, most pronounced on the northern and western edges of the high-saline core (Fig. 6C and D). The companion paper by Liu et al. 101 suggested that the decadal change in subduction associated with the expansion of the SSS-max appeared to be related to the AMO, an indicator of decadal climate variability within the Atlantic basin. Thus, changes in the global water cycle related to decadal climate variability were translated into changes in SSS and then into the interior variability of the ocean.
Another area that has been recently examined is the South Pacific SSS-max (Fig. 7), which has been the subject of two recent studies, Hasson et al. 92 and Bingham et al. 104 Both studies looked at interannual variability and compared the shape and configuration of the SSS-maxima with the E-P forcing. The midlatitude South Pacific is different from the North Atlantic, mainly to the much broader extent. The South Pacific has no regular ITCZ but does have the SPCZ 105 extending to the southeast from the western boundary that confines the SSS-max to the eastern part of the basin. 88 The South Pacific SSS-max was shown in both studies to migrate interannually. It underwent significant migration during the 2011-2016 period (Fig. 7A), about 10°eastward and 2-3°northward (toward the equator). Hasson et al. noted similar migration, but in the opposite direction and at a slower rate, about half the speed. 92 At the same time, the SSS-max increased in area by nearly a factor of 2, though this is not obvious from Figure 7B, and in salinity, by about 0.1. Though this change seems much more rapid than that in the North Atlantic shown by Yu et al., 100 it should be noted that this is interannual, as opposed to decadal, and the time scales are different. Melzer and Subrahmanyam 106 studied the SSS-max in all ocean basins and found decadal variability similar in magnitude and the extent to those shown in the studies highlighted here.
Bingham et al. 104 looked at a number of possibly related factors that might have caused the shifting noted in Figure 7B, but could not  100

(A) Mean SSS in the subtropical North Atlantic in the 1980s (black contours) and the 2000s (magenta contours) with the SSS mean differences between the two decades in the background (colors). (B) The total shift in the positions of the selected isohalines located north of 25°N, with error bars denoting the 95% confidence interval estimates. (C) The climatological mean thickness of the STUW. (D) The differences in mean STUW thickness between the 2000s and 1980s.
definitively point to any one in particular. They did note no interannual changes in E-P that might have been responsible, or changes in the configuration of the SPCZ. The most likely culprit was interannual changes in Ekman transport and Ekman pumping causing changes in the underlying subtropical gyre. Thus, in the case of the South Pacific, changes observed in SSS are associated with oceanic processes possibly forced by changing winds. A significant relationship was found between the area of the SSS-max and the Pacific Decadal Oscillation, which, like the AMO is an indicator of decadal variability within the Pacific basin. This illustrates the complex interplay between atmospheric forcing through E-P and winds, ocean circulation, and the global water cycle occurs on all climate time scales from seasonal to interannual to decadal and longer.
The SSS-max regions have been studied in other ocean basins, though not as thoroughly. We briefly mention one more, the North Pacific. Terray et al., 107  The North Pacific SSS-max also has shorter-term variability. Katsura et al. 93 studied the interannual variability of the two maxima separately by averaging SSS in boxes containing them. They found SSS increased significantly in these areas between 2004 and 2010. This was strongest in the western box, with a magnitude of about 0.25 during that time. This is a much higher rate of change than the decadal numbers cited above. The interannual variability of this SSS-max region is large compared with decadal. Katsura et al.'s analysis did not indicate a specific cause of the increase. They showed, for example, evaporation, entrainment, and advection, in the region, but none of those had strong enough interannual variability to cause such a large change in SSS. No matter the cause of such changes, interannual variability in SSS was transmitted into the interior ocean as STUW and observed to the west at later times. In particular, large fresh anomaly, formed at the surface in 2004, propagated to the western boundary over the following 3 years or so.
Linking salinity to E-P Several modeling studies have shown that the processes responsible for amplification of the ocean salinity pattern are not as straightforward as a simple response to changes in the E-P field as ocean warming and ice melt also play a role. Considering the changing E-P field first, Skliris et al. 26 carried out an assessment of multiple reanalysis and synthesis datasets (NCEP/NCAR, NCEP-DOE, 20CRv2, ERA-Interim, COREv2, and OAFlux) but found it difficult to establish robust changes in E-P (e.g., Fig. 3). The most reliable evidence for change was found in the southern hemisphere subtropical gyres, particularly in the Southeast Pacific, where precipitation rates are low and increasing evaporation is the dominant term, leading to a strengthening of net E-P. Changes in tropical precipitation showed a much greater degree of disparity between the different datasets available reflecting difficulties in robustly representing changes in the intense and localized rainfall in the relatively narrow tropical convergence zone.
Grist et al. 110 further developed a reanalysisbased E-P analysis using a temperature-salinity (T-S) framework that reduces the complexity arising from spatial variability. They found three temperature-dependent E-P change regimes. At less than 10°C, corresponding primarily to the Southern Ocean, the region of net precipitation becomes fresher. For temperatures greater than 27°C, corresponding to the tropics there is an increase in net precipitation and a shift of this region in T-S space toward lower salinities. In the midrange of 15-27°C, corresponding to the subtropics, increased net evaporation was observed consistent with Skliris et al. 26 The observationbased results were found to be broadly consistent with changes in a 1% increase CO2 run of the HadGEM3 climate model although the model also experienced a general shift of the main E-P cells in T-S space toward warmer temperatures. Overall, they found that 1979-2007 was a period of hydrological cycle intensification, resulting in a broadening of the global ocean salinity distribution. They also noted that the broadening occurred mainly in the tropical/subtropical temperature range. By contrast, at lower temperatures, the water mass distribution shifted to lower salinities rather than broadening to both lower and higher salinities consistent with ocean processes redistributing high temperature freshening to lower temperatures.
Amplification of ocean salinity and its causes have been the subject of renewed interest 79,83 following the development of a new method based on water mass transformation theory for determining the amount of change in the water cycle. 82 Skliris et al. 79 employed this method to infer the water cycle change from observed variations in salinity using three global datasets covering the period 1950-2010 and spanning a range of depths. Earlier analysis using surface ocean salinity observations suggested that the global water cycle had amplified by 8 ± 5%°C −1 of global mean surface temperature rise over 1950-2000. 77 This result was consistent with expectations (7%°C −1 ) based on the increased capacity of air to hold water as it warms following the CC equation but inconsistent with results obtained from state-of-the-art climate models that had values about a factor of 2 smaller. Zika et al. 82 found a smaller value for the amplification of 3.0 ± 1.3%°C −1 using full depth rather than surface salinity observations resolving the earlier disagreement with climate models for which the corresponding value is 4.3 ± 2.0%°C −1 .
Zika et al. 83 addressed the problem of the stronger amplification of the SSS than the full-depth field. By employing a set of targeted experiments with different forcings using the NEMO (Nucleus for European Modelling of the Ocean) model at 1°resolution, they were able to show that ocean warming has a significant impact on the surface salinity pattern amplification in addition to the effects of changing surface freshwater flux arising from the strengthening water cycle. The importance of including surface temperature changes in replicating the three-dimensional pattern of observed salinity change is also shown by Lago et al. 111 The warming acts to increase near-surface stratification, prolonging existing salinity contrasts and causing surface salinity patterns to amplify further. This process accounts for about half of the observed change in the SSS with the changing water cycle impact and a minor contribution from ice mass loss explaining the rest.
The results of the Zika et al. 83 model experiments are shown in Figure 8 (panels A-F will correspond Ongoing research is concerned with the potential influence of vertical salt fluxes on the ocean salinity distribution 112 and the explanation for basin scale/marginal sea ocean salinity anomalies. 113,114 Liu et al. 112 use the ECCO ocean state estimate to explore vertical salt fluxes and find that in the upper ocean, the vertical salt flux can be similar in magnitude to the air-sea freshwater flux. Further research is needed to establish how robust these results are to the choice of the ocean state estimate. Regional variations in ocean salinity have long been the subject of interest, for example, major freshening anomalies in the North Atlantic in the 1970s and 1990s. 8,115 Most recently, a significant fresh anomaly has developed in the subpolar gyre of the North Atlantic since 2012 and its causes and relation to a colocated cold water anomaly 116 are under investigation. In the Pacific, Li et al. 114 find using Argo that increased salinity in the upper 200 m since 2005 is primarily due to a reduction in precipitation. Marginal seas provide an opportunity at regional scales to study the impact of the changing water cycle on salinity. For example, an analysis of Mediterranean Sea salinity trends from observations over 1993-2016 shows a rapid salinification of this basin in recent decades linked to the intensification of net evaporation. 112 Subsequent work using water mass transformation theory applied to observations over a longer 60-year time scale (1950-2010) also finds that increased evaporation plays a key role and that long-term salinification results from regional water cycle changes rather than variations in the salt transport at the Straits of Gibraltar. 117

Discussion and future perspective
Significant advances have been made in documenting the evidence for the hydrological response to the global warming of recent decades and in estimating and understanding the rate of intensification in the freshwater entering and leaving the ocean surface. In particular, historical ocean salinity data in combination with climate model simulations have demonstrated that the global patterns of SSS changes are consistent with the "dry-gets-drier and wet-gets-wetter" paradigm, and the evidence supports an intensifying global water cycle as a consequence of the increase in atmospheric water vapor transport. Ocean salinity has emerged as a key indicator of the past changes in the ocean water cycle, shaping up our view of salinity as nature's rain gauge for the E-P over 71% of the Earth surface.
Nevertheless, significant uncertainties still remain in the current estimates of E, P, and their balance over the global ocean and in the length of time series needed for assessing possible long-term trends related to climate warming. Maintaining and improving the space-based observing system is key to advancing research in this area. Improving the observational capabilities for global river discharges and freshwater exchanges at polar seas is also important. Balance of the ocean freshwater budget using E, P, river runoffs provides a physical constraint to help quantify uncertainties in E-P, but none of the present data products, sourced from either satellite or the latest atmospheric reanalyses, are able to fully close the balance. Thus, continuity in time series and reliability for long-term trends need to be improved.
Given the sensitivity of salinity to freshwater transport into and out of the ocean and the increasing availability of accurate salinity measurements over the globe, 118-120 the use of ocean salinity observations to fingerprint the changes in the hydrological cycle stands as a logical and practical alternative to satellite-derived rainfall and evaporation estimates. There are still many unanswered questions regarding the role of ocean dynamical processes in the relationship between SSS and the E-P forcing. As shown by climate model experiments, ocean warming and changes to the water cycle and salinity are closely entangled. 79,83 Our understanding of the impacts of the changing water cycle on ocean salinity must consider other factors, including ocean warming, 83 interbasin patterns of moisture convergence, 121 and the interactions between SSS and the interior ocean via mixing and subduction. 112 The latter requires further observations to extend the 2000-m depth limit currently imposed by Argo. Enhancement of the Argo array for deep and polar oceans and marginal seas is expected to improve our ability to use salinity to estimate the change in the water cycle.
The recent development of salinity remote sensing has greatly enhanced the capability to measure SSS, allowing the land-sea linkage of the integrated water cycle. 120 Ongoing and recent missions include the Soil Moisture and Ocean Salinity mission by the European Space Agency (2010-present) as well as the Aquarius mission (2011)(2012)(2013)(2014)(2015) and Soil Moisture Active Passive (SMAP) mission (2015-present) by the National Aeronautics and Space Administration. The upcoming SWOT satellite mission presents an exciting prospect for improved quantification of river discharges, and hence, improved estimation of the freshwater budget balance on a global scale. Presently, river flows are measured by river gauges that record water surface elevations at fixed points along river channels. 122 However, the sparsely distributed gauging stations are not sufficient for producing the spatial and temporal distribution of water stored on land surface and discharged into the ocean. Additionally, the number of gauging stations is declining worldwide for economic and political reasons. 122,123 SWOT is designed as a wide-swath interferometric altimeter for high-resolution two-dimensional mapping of terrestrial water surface height, inundation extent, and ocean surface elevation globally. 35 Preliminary results suggest that SWOT could allow accurate estimates of the Earth's medium-to-large, landbased water bodies on average every 21 days. 124,125 It is anticipated that SWOT will dramatically expand the current measurement-based knowledge that has been built upon the existing network of river gauges. It will also improve our ability to understand and model the consequences of climate change and variability on surface water storage volumes and their connection to precipitation, evaporation, infiltration, and runoff.
Satellite SSS in combination with near-surface salinity measurements from in situ platforms has the potential to constrain evaporation-precipitation estimates in coupled assimilation and improve coupled and ocean reanalysis products. 126 Since the E-P forcing on the upper ocean is mediated by ocean dynamics (advection and mixing) on regional scales, improving model physics in the upper ocean is important for maximizing the benefits of SSS measurements in simulating the E-P impacts on the upper ocean processes, especially those related to water mass transformation that are sensitive to the water cycle changes.
Climate models currently predict that the global mean surface temperature will continue increasing in the coming decades in response to anthropogenic climate warming. Because water is of central importance to the Earth system, the question of how the Earth's hydroclimate may change in the future as a result of anthropogenic impacts is one of the greatest challenges the world faces this century. Good adaptation strategies can be made only when climate models are capable of reproducing all key processes in the water cycle, including evaporation at the ocean surface, moisture transport by the atmosphere, cloud condensation and precipitation, freshwater transport by the ocean, and river runoffs. These processes, in turn, can affect ocean salinity and density and subsequently heat transport, release, and storage, which ultimately modify the role of the ocean in climate change. As more accurate satellite observations become available, along with higher-resolution and more comprehensive climate model projections, understanding the response of the water cycle to global warming and its impacts on society and ecosystem will continue to be a major research challenge.