Before examining the spatial patterns, it is useful to clarify how variability is defined at each timescale. Seasonal variability is isolated by removing the time-mean from monthly fields (demeaned). Interannual variability refers to year-to-year variations on timescales of 1–7 years, obtained by removing both the long-term trend and mean seasonal cycle from monthly-mean fields (detrended and deseasonalized). Figure 3 shows where forcing controls salinity variability versus where ocean circulation redistributes it. Forcing-dominated regions exhibit statistically significant positive correlations (r > 0.50 seasonal, r > 0.14 interannual; p < 0.1) between salinity tendency and local E-P. Circulation-dominated regions show weak, non-significant correlations (p > 0.1), indicating advection and mixing redistribute freshwater before forcing signatures accumulate. Significant negative correlations (r < -0.50 seasonal, r < -0.14 interannual; p < 0.1) occur where circulation processes oppose surface forcing. With an 11-year record (2012–2022), the degrees of freedom for interannual statistics are limited, and the significance threshold of r > 0.14 (p < 0.1) reflects this. The broad spatial patterns of interannual correlation are nonetheless consistent with results from longer records (Vinogradova and Ponte, 2017; Yu, 2023), supporting the robustness of the patterns shown.

Forcing-dominated regions (33 % of ocean area) concentrate in subtropical gyre interiors (20–40° N/S), semi-enclosed seas, and monsoon regions. Subtropical gyres maintain strong correlations (r = 0.6–0.7) at seasonal timescales and moderate correlations (r = 0.3–0.5) at interannual timescales, where weak advection (1–3 cms-1) (Yu, 2023) allows E-P signals to accumulate. Semi-enclosed seas exhibit strongest forcing control (r > 0.7 seasonal) because geometric constraints limit advective escape. Mediterranean evaporation (∼ 10 cmmonth-1; Fig. 1b) creates 0.3–0.5 psu seasonal amplitude correlating at r = 0.85 with E-P (Skliris et al., 2018). Monsoon regions show timescale-dependent behavior: Bay of Bengal exhibits strong seasonal correlation (r = 0.5–0.7) but weak interannual correlation (not significant). Monsoon precipitation (∼ 22 cmmonth-1 during June–September; Rao and Sivakumar, 2003) creates 2–4 psu freshening on 3-month timescales shorter than lateral spreading timescales (6–9 months), enabling local accumulation. At interannual timescales, ENSO-driven current anomalies introduce variance comparable to precipitation anomalies, degrading correlation (Delcroix and Hénin, 1991; Hasson et al., 2014).

Circulation-dominated regions (55 %–66 %) include western boundary currents, equatorial upwelling zones, and subpolar gyres. Western boundary currents exhibit weak correlation because strong advection redistributes freshwater faster than forcing accumulates (Hogg and Johns, 1995). Correlation degrades from subtropical gyre interiors toward western boundaries where Gulf Stream and Kuroshio advection dominates. Equatorial upwelling zones exhibit significant negative correlation (r = -0.4 to -0.6 seasonal; 5 %–10 % of ocean area) where upwelling brings high-salinity subsurface water to the surface during precipitation seasons, notably in eastern equatorial Pacific (Maes et al., 2014). Subpolar gyres and Southern Ocean show weak correlation despite large precipitation because strong currents, deep winter mixing (200–800 m) (de Boyer Montégut et al., 2004), and energetic mesoscale eddies redistribute freshwater (Müller et al., 2019) before forcing signatures accumulate.

Natural decadal variability operates on timescales comparable to 10–30 year analysis periods, complicating the detection of forced salinity signals. Decadal-scale trends reveal contrasting behavior between atmospheric forcing and ocean salinity response. Analysis of 1993–2010 shows the ocean water cycle (E-P) pattern amplified by approximately 5 %, with spatial correlation r≈ 0.5 between climatological E-P and observed trends (Vinogradova and Ponte, 2017), matching Clausius–Clapeyron expectations for atmospheric moisture response to warming (Held and Soden, 2006). Surface salinity patterns, however, amplified by less than 1 % globally, with near-zero spatial correlation between climatological patterns and observed trends (Fig. 4). Ocean basins show scattered responses: some amplify climatological patterns while others reverse them. This weak salinity pattern amplification, despite coherent atmospheric forcing, indicates that ocean processes operating on decadal timescales prevent surface salinity from tracking atmospheric forcing coherently.

Regional analysis reveals the physical basis for weak global pattern amplification. Pacific basin-average salinity decreased over 1970–2002, exceeding internal variability estimates and showing detectable anthropogenic influence, while Atlantic basin-average trends remained within internal variability range (Terray et al., 2012). This basin-scale contrast reflects differences in how circulation and natural variability compete with forcing accumulation. North Atlantic salinity over 1993–2012 exhibits large multiyear variability driven by NAO modulation of the North Atlantic Current, with circulation-driven anomalies dominating over surface freshwater forcing (Stendardo et al., 2016). Strong western boundary currents and active decadal modes redistribute North Atlantic freshwater on timescales comparable to 10–20 year forcing accumulation. The Pacific shows clearer forced trends during 2005–2015 (basin-average 2.2 × 10⁻³ psuyr-1) following earlier freshening in 1994–2005 (Li et al., 2020), suggesting forced signals can emerge where circulation redistribution operates more slowly.

Salinity budgets quantify the circulation-forcing competition. Pacific upper 200 m analysis for 2005–2015 shows surface freshwater flux and ocean advection produce opposite effects, each with magnitudes approximately twice the net salinity tendency (Li et al., 2020). This opposing relationship indicates active redistribution rather than passive forcing integration. Where precipitation dominates, advection exports excess freshwater; where evaporation dominates, advection imports freshwater from elsewhere. Horizontal advection and diffusion control subtropical redistribution while vertical diffusion and entrainment control tropical and high-latitude redistribution (Lyu et al., 2025), creating regionally varying pathways that prevent globally coherent response to coherent forcing. Despite these complications, salinity exhibits higher signal-to-noise ratios than atmospheric variables such as precipitation or E-P, because salinity integrates high-variance atmospheric forcing over time while spreading anomalies spatially. As a result, detecting anthropogenic influence in SSS requires fewer than three ensemble members for SSS versus substantially more for precipitation or E-P (Terray et al., 2012).

The Interdecadal Pacific Oscillation shift from positive to negative phase around 1998–1999 contributes approximately 30 % of Pacific salinity variance, exceeding 60 % in equatorial regions (Vinogradova and Ponte, 2017). NAO variability similarly drives North Atlantic salinity through circulation changes independent of local forcing (Stendardo et al., 2016). At decadal timescales, forcing accumulation, circulation redistribution, and natural variability all operate on comparable timescales, preventing any single process from dominating. The result is coherent atmospheric forcing (+5 % pattern amplification) producing weak salinity response (< 1 % pattern amplification), showing that 10–30 year periods are insufficient for forcing signals to overwhelm circulation redistribution and natural variability. Basin differences arise where local conditions favor one process. Pacific shows partial forcing dominance where redistribution is slower, Atlantic shows variability dominance where energetic currents operate faster. Longer integration periods are required for forcing to accumulate decisively beyond redistribution timescales, as examined in Sect. 3.3 for multi-decadal scales.

Over multi-decadal timescales (50+ years), the influence of freshwater forcing on ocean salinity becomes clearly detectable, although these patterns remain obscure at shorter timescales (Figs. 3 and 4). Spatial correlation between 1950–2000 salinity trends and climatological SSS patterns reaches r = 0.7–0.8, indicating fresh regions systematically freshened while salty regions salinified over 50+ year periods (Durack and Wijffels, 2010; Durack et al., 2012). This pattern correlation contrasts sharply with near-zero values at decadal timescales (Fig. 4b and d), showing that forcing accumulation over 50+ years produces spatially coherent salinity response despite circulation redistribution. The 145-year record of ocean salinity measurements, obtained from HMS Challenger and SMS Gazelle expeditions (1870s) through modern observations, demonstrates that pattern amplification has been detectable since the early industrial era, but with marked acceleration. Rates increased from ∼ 0.166 gkg-1century-1 (1870s–1950s) to ∼ 0.306 gkg-1century-1 (1950s–2010s), a 54 ± 10 % acceleration (Gould and Cunningham, 2021). Spatial correlation between regional salinity changes across these periods (r=0.64) indicates persistent forcing-driven patterns since the 1870s, though with non-linear intensification. Recent decades show further acceleration, with post-1991 rates nearly doubling those of 1960–1990 (Cheng et al., 2020; Douville and Cheng, 2024).

Regional patterns of E-P show quantitatively consistent amplification across independent analyses at 4 % °C-1–8 % °C-1 (Durack and Wijffels, 2010; Helm et al., 2010; Skliris et al., 2016), in line with Clausius–Clapeyron predictions for atmospheric moisture response to warming (Held and Soden, 2006). Tropical and subpolar fresh regions (climatological P > E) freshened 0.2–0.4 psu over 1950–2000, while subtropical salinity maxima (climatological E > P regions) increased 0.1–0.3 psu (Curry et al., 2003; Durack and Wijffels, 2010). However, pattern amplification exhibits marked asymmetry: freshening in low-salinity regions proves far more robust across ocean basins with consistent spatial patterns and strong emergence from natural variability, whereas salinification in subtropical gyres displays greater sensitivity to region definition and weaker signal-to-noise ratios (Cheng et al., 2020; Douville and Cheng, 2024). The Mediterranean Sea and Baltic Sea provide particularly clear evidence where geometric constraints can enhance forcing signals. In the Mediterranean, salinification (0.10–0.15 psu) has been linked to 10 %–15 % evaporation increases (Skliris et al., 2018), while in the Baltic, freshening (0.10–0.20 psu) is consistent with increased precipitation and runoff (Meier et al., 2006; Lehmann et al., 2022).

Attribution studies identify the different physical processes driving the observed pattern amplification. Observed multi-decadal patterns have less than 5 % probability of arising from internal variability alone when major climate modes are removed, with approximately two-thirds of variance attributable to external forcing (Pierce et al., 2012; Terray et al., 2012). CMIP5 simulations show the emergence of anthropogenic influence. Historical runs incorporating all forcings reproduce observed spatial patterns and amplification magnitude, whereas natural-forcing-only simulations show no significant trends (Cheng et al., 2020). Mechanistic decomposition using ocean model experiments reveals three contributing processes (Zika et al., 2018): direct surface freshwater flux changes from water cycle intensification, ice mass loss contributions (relatively minor), and ocean warming effects on stratification. Ocean warming proves particularly important because warming-induced stratification inhibits vertical mixing, effectively preserving and amplifying surface salinity patterns. Warming explains approximately half of surface salinity pattern changes from 1957–2016, with water cycle intensification of 3.6 % °C-1 ± 2.1 % °C-1 accounting for the remainder. These changes exceed 2σ natural variability (> 95 % confidence) (Zika et al., 2018), showing human influence on ocean salinity patterns since 1960 (Pierce et al., 2012; Terray et al., 2012).

The physical mechanisms enabling multi-decadal pattern emergence operate through timescale separation and stratification enhancement rather than circulation suppression. Surface E-P anomalies create near-surface salinity changes that subsequently subduct along isopycnals into low-latitude subsurface layers via subtropical gyre ventilation (Durack et al., 2012), advect poleward and downward through intermediate water formation driving broad 300–2000 m freshening (Helm et al., 2010), and interact with ocean warming to enhance high-latitude stratification and reduce vertical mixing that would otherwise erode surface salinity contrasts (Zika et al., 2018). At multi-decadal scales, circulation contributions remain important (Jarugula et al., 2025). The Atlantic-Pacific salinity contrast increased 5.9 ± 0.6 % over 1965–2020, yet direct E-P-R forcing explains less than half this change, with the remainder from circulation adjustments including thermocline heaving, cross-basin moisture transport, and gyre intensification (Singh et al., 2016; Friedman et al., 2017; Zika et al., 2021; Lu et al., 2024). What distinguishes multi-decadal from shorter timescales is not that circulation stops, but that forcing accumulates over 50+ years while basin-scale redistribution operates on 5–20 year timescales. Because forcing persists much longer than redistribution and mixing processes, it can build up a coherent signal despite ongoing circulation. Forcing controls only 30 % of ocean area at seasonal-interannual scales (Fig. 3) and produces less than 1 % pattern amplification at decadal scales (Fig. 4), but generates strong pattern correlations (r = 0.7–0.8) and 54 % acceleration post-1950s at multi-decadal scales. This progression shows that forcing fingerprints emerge when forcing timescales far exceed redistribution timescales, consistent with the rain gauge paradigm over sufficiently long periods and revealing ocean warming as a critical amplifying mechanism.

Section 4.1 examined how horizontal advection competes with forcing to create spatial and temporal patterns. The vertical dimension introduces a fundamentally different competition between vertical mixing timescale τvmix and horizontal advection timescale τadv. At the surface where τvmix is short (days for wind mixing, weeks for convection) (Kraus and Turner, 1967; Niiler, 1975; Marshall and Schott, 1999), salinity responds quasi-instantaneously to E-P-R forcing. Below the mixed layer where stratification suppresses vertical exchange, τvmix increases dramatically (decades to centuries for diffusive mixing through the permanent pycnocline) (Ledwell et al., 1993; Wunsch and Ferrari, 2004), while τadv remains moderate (years for subtropical cells, decades for gyre circulation) (Fernandez et al., 2015). This reversal in timescale ratios creates a vertical regime boundary where salinity transitions from recording contemporary forcing to preserving formation-era conditions.

Section 4.1 and 4.2 described regimes in which controlling timescales are cleanly ordered, so salinity behaves primarily as a recorder. When τf≫τadv, salinity integrates persistent freshwater forcing and large-scale patterns resemble the forcing patterns (rain gauge behavior). When τf≪τadv or τvmix≫τadv, salinity is exported and preserved along pathways, mapping circulation and formation histories (passive tracer behavior).

At submesoscales O(1–10km) this ordering breaks down because the relevant processes operate on comparable hours-to-days timescales. Mixed-layer instabilities, frontogenesis, and surface forcing all evolve fast enough that no single process sets salinity before the others act (Boccaletti et al., 2007; Thomas and Ferrari, 2008; McWilliams, 2016). Salinity cannot simply integrate forcing because forcing varies on the same timescale as advection redistributes anomalies. It cannot passively inherit upstream properties because the upstream field is itself rapidly rearranged. Instead, salinity often becomes dynamically consequential by sharpening horizontal buoyancy gradients and thereby strengthening ageostrophic secondary circulations. These circulations can drive vertical velocities w ∼ O(10–100md-1), ventilating the upper pycnocline far more efficiently than mesoscale stirring at w ∼ O(1md-1) (Mahadevan and Tandon, 2006; McWilliams, 2016).

High-resolution Saildrone observations during a 2019 North Atlantic crossing reveal this thermohaline control shift (Fig. 7). At mesoscales exceeding 100 km, density gradients in the Gulf Stream and open ocean are predominantly temperature-controlled with Turner angle Tu > π/4 and density ratio Rρ > 1, where the density ratio Rρ=αΔT/βΔS and Turner angle Tu = arctan(Rρ) together quantify the relative contributions of temperature and salinity to density, with ΔT and ΔS being temperature and salinity differences over equal distances of 350 m along the Saildrone track, α and β being the thermal expansion and haline contraction coefficients, respectively. The continental shelf, however, shows salinity already contributing significantly at these larger scales due to freshwater inputs and coastal fronts. Near 10 km Tu approaches or drops below π/4 and Rρ approaches or falls below 1 in all three regimes, indicating a shift toward equal or salinity-dominated density gradients, though the transition is more pronounced in the open ocean and continental shelf regimes. In the Gulf Stream winter regime, Tu approaches π/4 but does not clearly cross below it, consistent with a transition toward equal T-S contribution rather than full salinity dominance (Yu, 2026). The transition clusters near the first baroclinic deformation radius (Rd = 28 km), consistent with submesoscale dynamics emerging as the flow approaches the breakdown of purely geostrophic adjustment (McWilliams, 2016).

The sharpness of this transition is regime-dependent. In the Gulf Stream where thermal wind balance strongly dominates, Tu drops abruptly from greater than 5π/12 to less than π/4 over just 10–20 km. On the continental shelf where salinity already plays a larger role, the transition spreads over 50+ km. This regime dependence emphasizes that 10 km marks where local nondimensional ratios (Rossby number, frontogenetic sharpening versus damping, mixing versus advection) cross order unity, not a fixed geometric constant (Callies and Ferrari, 2013; McWilliams, 2016).

Across Sect. 4.1–4.3, the same message keeps resurfacing: salinity behavior does not depend on absolute spatial or temporal scales (10 km versus 100 km, or days versus years); what matters is how forcing, advection, and mixing timescales are fast or slow relative to each other. In practice, three comparisons decide how to read a salinity signal. First, does freshwater forcing change slowly enough for anomalies to accumulate where they are made (τf/τadv)? Second, once salinity anomalies subducted below the mixed layer, are they erased by vertical exchange or preserved as water-mass properties (τvmix/τadv)? Third, are we in a regime where the process hierarchy collapses because everything happens on similar timescales? Table 3 summarizes these boundaries and what they look like in data. The four regimes are synthesized in Fig. 8 as a conceptual diagram organized by these timescale ratios, with gradient shading reflecting the gradual nature of transitions that depend on local conditions rather than fixed spatial or temporal thresholds.

The rain-gauge and passive-tracer limits are two ends of the same spectrum, but their physical basis is worth making explicit. This framework also makes the “rain gauge vs. passive tracer” behavior easier to interpret. In the rain-gauge limit, τf/τadv≫ 1, forcing persists over a spatial domain large enough that circulation recirculates water through the same forcing regio repeatedly before anomalies can escape. The rain-gauge limit therefore involves both a temporal and a spatial dimension: forcing must not only persist long enough relative to advection, but must also be spatially coherent over the circulation domain. In subtropical gyres, for example, recirculating water repeatedly samples the same evaporative forcing region, so the relevant forcing is the spatially integrated E-P over the gyre rather than the local instantaneous flux. This spatial integration amplifies the forcing signal relative to what would be expected from local E-P alone, which helps explain why subtropical salinity maxima are so robust and persistent. In the passive-tracer limit, τf/τadv≪ 1, anomalies are displaced from their formation region in both space and time. Laterally, water parcels are advected far from where they were forced before forcing changes significantly, so present-day salinity at a given location reflects conditions from a distant upstream source rather than local E-P. Vertically, once anomalies subduct below the mixed layer, they are insulated from further surface contact by the permanent thermocline and preserved along isopycnal pathways for years to decades (τvmix/τadv≫ 1). Spatial displacement between formation and observation regions is therefore the diagnostic signature of passive-tracer behavior, just as spatial coherence between salinity patterns and E-P is the signature of rain-gauge behavior.

The submesoscale case is not just another point on the same continuum; it is where the interpretation changes. When the timescales converge, salinity cannot be treated as a record of forcing or a record of upstream conditions because both are evolving on the same hours-to-days window. Through differential damping (Sect. 4.3.2), salinity contributes comparably to density and enables vertical velocities O(10–100md-1) that ventilate the upper pycnocline, an order of magnitude stronger than mesoscale processes.

Two aspects of the regime diagram remain least well constrained. The τf/τadv ∼ O(1) transition on decadal scales, where substantial forcing amplification produces a muted salinity pattern response, still lacks a clean mechanistic attribution, as gyre adjustment, coupled variability, and changes in ventilation geometry are all plausible contributors. Additionally, although the dynamical submesoscale regime is increasingly better observed, it remains poorly represented in many climate models; how salinity-driven submesoscale overturning feeds back onto basin-scale stratification and biogeochemistry is still an open question.

The timescale competition framework clarifies three aspects of how salinity relates to forcing, circulation, and predictability.

Why spatial mismatch is informative. Interpreting salinity as a water-cycle indicator often focuses on whether it maps onto E-P-R. The seasonal–interannual result (forcing-dominated behavior over ∼ 30 % of area; Fig. 3) has been interpreted as evidence that salinity is unreliable across most of the ocean. But that interpretation assumes the only purpose of salinity is to mirror forcing. Sections 3 and 4 show that mismatch provides information. Where salinity departs from local forcing, water parcels have been transported or mixed before forcing accumulates locally. Those departures reveal quantities not directly observable, such as transit times, source regions, and coherence of pathways. Examples include thermocline salinity correlating with subtropical surface conditions from several years earlier, subsurface salinity maxima recording the last surface contact of a water mass, and river plume far-fields whose phase lags trace boundary-current routes. The “70 % mismatch” is signal, not noise.

Why salinity prediction depends on ocean state representation, not just atmospheric forcing. Salinity is a conservative variable in the sense that, unlike temperature, it lacks a direct negative feedback with the atmosphere: evaporation and precipitation are no controlled by the salinity state itself, so salinity anomalies are not restored toward a reference value the way SST anomalies are through surface heat fluxes. Indirect feedbacks can exist, such as a warm SST anomaly driving excess evaporation and creating a positive salinity anomaly in subtropical regions, or precipitation responses to SST further modulating salinity in the tropics. These coupling are generally secondary to the direct atmospheric restoring that acts on temperature, but they are not negligible and can reinforce salinity anomalies in some setting. This asymmetry creates memory. A model can predict rainfall anomalies correctly yet still miss salinity because the salinity at any location reflects both local forcing and the water arriving from upstream (Fig. 5). If the model misrepresents ocean currents or mixing, it delivers the wrong water to that location even when atmospheric forcing is correct. Conversely, salinity can be predictable even when rainfall is not, because water arriving today may have been formed years earlier in a distant region. This matters for practical prediction, as seasonal forecast systems can show good performance in precipitation and SST yet fail in salinity because their circulation state, pathway geometry, or mixing parameterizations are biased. Salinity forecast accuracy is therefore a coupled test: it requires both accurate forcing and ocean transport and mixing representations.

Why different studies reach different conclusions about forcing control. Basin-scale studies emphasizing robust pattern amplification and regional studies emphasizing advection and mixing are not disagreeing about physics; they examine different regime boundaries. At O(1000km) and multi-decadal scales, forcing evolves slowly enough that circulation and mixing redistribute anomalies repeatedly and forced patterns emerge. At O(100km) and seasonal scales, water travels large distances while forcing is evolving; local E-P-R competes directly with supply from upstream. Both are correct within their domains. The distinction is not only temporal but also spatial. At basin scales, circulation redistributes anomalies within the same broad forcing regime, so the spatial pattern of salinity trends eventually reflects the spatial pattern of forcing. At regional scales, water can cross regime boundaries during transit, carrying properties formed under different forcing conditions into regions with very different local E-P-R.

Current satellite SSS products have an effective resolution of roughly 40–50 km (Vinogradova et al., 2019; Reul et al., 2020). That scale is sufficient to map basin-scale patterns and many forced signals, but it smooths the features that often set the balance between forcing, advection, and mixing. The result is a persistent observational gap at O(10km), the range in which salinity often transitions from a largely passive indicator of freshwater fluxes to an active dynamical driver that shapes density, stratification, and vertical exchange.

The importance of O(10km) is dynamical, not simply a matter of detail. At these scales, three controls converge (McWilliams, 2016). First, horizontal density gradients become steep enough to drive ageostrophic secondary circulations and vertical motions of order 10–100 md-1, directly coupling lateral structure to vertical exchange. Second, salinity contrasts between adjacent water masses become large enough that salinity can rival temperature in setting density. A 0.5 psu difference can produce a density effect comparable to 2 °C, so neglecting salinity in the density budget becomes dynamically consequential. Third, freshwater-driven stratification reaches an intermediate regime where it is strong enough to inhibit routine wind mixing yet weak enough that storms can episodically overcome it. The fate of freshwater anomalies, whether they persist, mix downward, or are exported laterally, is therefore decided on storm timescales by the pre-existing frontal and stratification structure. Beyond controlling whether anomalies mix downward or are exported laterally, submesoscale processes do not merely influence but also modulate the intensity of surface salinity signals themselves. Submesoscale restratification following a mixing event can trap freshwater near the surface, amplifying surface salinity anomalies rather than diluting them. This feedback between submesoscale dynamics and near-surface stratification means that O(10 km) processes do not merely redistribute existing signals but can actively intensify them, with implications for the amplitude of surface salinity variability observed from satellites.

These controls are strongly nonlinear, and this is why coarse resolution is limiting. At 40–50 km, fronts, filaments, and freshwater lenses are blended into smooth gradients that do not represent either the sharp structures where buoyancy suppresses mixing or the well-mixed interiors where winds dominate. The relevant physics depends on the sharpness and vertical structure of the gradients, not on their spatial average. Storm response depends on the initial state. A sharp halocline at 10–20 m can resist deepening and instead sharpen under shear, whereas a more gradual stratification profile allows mixing to deepen the mixed layer and entrain freshwater from below. Freshwater over saltwater creates stratification inhibiting mixing; less mixing allows more freshwater to accumulate, strengthening stratification further (Drushka et al., 2016). This positive feedback operates only when gradients are sharp. Doubling the gradient does not double the stratification effect but can shift the regime from mixing-dominated to stratification-dominated.

A sustained O(10km) SSS observing capability (Colliander et al., 2024) would turn these ambiguities into testable mechanisms. By tracking individual fronts and lenses through forcing events, it would quantify whether salinity stratification typically resists storm mixing or is routinely eroded, whether frontal structures sharpen through instabilities or broaden through wind stirring, and where and when the salinity contribution to density exceeds that of temperature. This is also the scale at which we can determine how much salinity variance, and therefore buoyancy variance, is currently unresolved. If the missing fraction is small, today's products and coarse models may be adequate for many climate diagnostics. If it is substantial in key regions or seasons, then parameterized submesoscale processes are not a detail but a leading source of uncertainty in regional stratification, mixing, and coupled feedbacks.

This capability matters most where freshwater forcing and gradients are strongest. River-influenced shelves and boundary currents, tropical convergence zones with patchy rainfall, western boundary current extensions with intense fronts, and marginal ice zones where freshwater and heat fluxes interact all experience large freshwater fluxes per unit area where salinity anomalies are strongest and gradients are sharpest. In these settings, O(10km) SSS would resolve the nearshore-to-offshore transition from forcing-dominated plume cores to circulation-dominated export pathways, enabling quantitative estimates of export fractions, pathway timescales, and seasonal regime switching that current observations often describe only qualitatively. Improving to O(10km) is not simply higher resolution but direct access to the mechanisms that control persistence, export, and vertical penetration of freshwater anomalies.

Horizontal resolution addresses only part of the regime-boundary problem. Two additional observational gaps determine whether salinity anomalies remain surface-confined or become climate-relevant interior signals. Both involve transitions where competing processes cross critical thresholds, and both remain poorly constrained because current observing systems capture mean states rather than the events that drive transitions.

The first gap concerns mechanisms of vertical penetration. Subsurface salinity maxima at 100–200 m depth record surface forcing from years earlier (Fig. 6), demonstrating that anomalies can penetrate and persist. Yet penetration depends on a buoyancy competition. Wintertime cooling removes buoyancy and destabilizes the water column, while freshwater input adds buoyancy and stabilizes it. Where haline stratification is weak, as in North Atlantic subtropical mode waters, cooling deepens mixed layers through hundreds of meters and subducts surface anomalies. Where haline stratification is strong, as in the North Pacific, cooling cannot erode the fresh cap and surface anomalies remain shallow or are exported laterally. Adjacent regions with similar atmospheric forcing show markedly different penetration depths, indicating that pre-existing subsurface structure and event timing matter as much as seasonal-mean forcing (Yeager and Large, 2007).

The critical unknown is whether penetration occurs through gradual seasonal deepening or through episodic storms whose mechanical energy briefly overwhelms stratification (Dohan and Davis, 2011; Whitt and Taylor, 2017). If penetration is episodic, then a few intense storms per winter disproportionately set subduction and subsurface pathway memory. If gradual, then cumulative seasonal forcing sets penetration depth. The distinction matters for projections because climate models parameterize vertical mixing differently for gradual versus episodic processes, and future warming may strengthen haline stratification faster than it weakens thermal stratification. Testing these mechanisms requires high-frequency time series capturing individual storm responses with co-located measurements of winds, buoyancy fluxes, and vertical structure across the seasonal cycle.

The second gap concerns high-latitude regime transitions. Arctic and subpolar seas switch between long stratified periods when freshwater accumulates near the surface and brief convective episodes when mixing homogenizes hundreds of meters within days (Marshall and Schott, 1999). This switching controls whether freshwater anomalies influence deep water formation and overturning or remain surface properties. Arctic freshwater content varies by 10 000 km3 over decades, potentially affecting Atlantic overturning, but attribution requires understanding whether freshwater remains near-surface or penetrates to depth (Proshutinsky et al., 2009). Current observations are biased toward ice-free summer periods when stratification is strongest, missing winter convective events that may determine whether surface anomalies ventilate or remain trapped. Year-round sampling through under-ice platforms, moorings, and autonomous vehicles would capture regime shifts and test whether high-latitude regions are approaching thresholds where convective mixing weakens or freshwater trapping strengthens under continued warming.

These gaps highlight that resolving regime boundaries requires observing not just finer spatial scales but also the temporal variability that drives transitions between states. The processes controlling vertical penetration operate on storm timescales; the processes controlling high-latitude freshwater pathways operate on seasonal cycles that include winter conditions that remain systematically undersampled.

Climate model ensembles reproduce basin-scale pattern amplification yet diverge sharply in regional projections at the regime boundaries emphasized in this review. Tropical differences trace to how models represent shallow overturning and the efficiency of exporting subtropical anomalies into the equatorial band (Wang et al., 2023). Barrier layer changes depend on how models balance precipitation increases against wind-driven mixing (Pang et al., 2023; Vogt et al., 2025). High-latitude outcomes depend on whether models trigger or suppress episodic convection and how they handle ice-ocean freshwater exchange (Haine et al., 2023). These divergences concentrate where observations are currently sparse or averaged over the critical scales.

Reducing projection uncertainty requires constraining regime-boundary physics rather than perfecting basin-mean trends. O(10km) SSS observations would constrain front-resolving buoyancy structure at the surface, while coordinated vertical observing would constrain penetration mechanisms and high-latitude switching. Together, these would enable model evaluation to move beyond climatology to process-based metrics including phase lags between forcing and salinity response, depth and strength of subsurface salinity maxima, persistence statistics of freshwater lenses, and the contribution of salinity to density at fronts.

One path forward involves state-aware parameterizations. Rather than applying uniform mixing rules tied mainly to grid spacing, models could diagnose when a grid cell is front-like with strong gradients, lens-like with strong near-surface stratification, or quiescent with weak gradients, and adjust subgrid physics accordingly (Fox-Kemper et al., 2008). This approach would encode the framework's core insight that different balances apply in different regimes.

The scale- and regime-dependent framework developed here extends beyond salinity. Salinity is conservative enough to preserve formation history as water parcels move, changing primarily through boundary freshwater fluxes and mixing rather than continuous air–sea exchange. This conservation makes salinity an effective tracer of pathways and memory, and it explains why the same framework applies naturally to other quasi-conservative tracers such as nutrients, dissolved gases, and isotopes (Talley, 2008). Water-cycle monitoring benefits from complementary constraints on the same forcing. At large scales and long timescales, freshwater fluxes alter both ocean mass through ocean bottom pressure and sea level and salt concentration through salinity (Liu et al., 2025). Where advection is slow relative to forcing accumulation, these constraints reinforce each other and can be used to evaluate and refine E-P-R products. Where advection is fast, both mass and salinity become mixed signals of forcing and circulation, and their joint interpretation becomes regime aware rather than purely local. Observing strategies that combine gravimetry, altimetry, and high-resolution SSS offer a practical route to separating water-cycle change from redistribution by circulation, provided the key regime boundaries are resolved.

Coupled feedbacks operate through stratification. Freshwater input strengthens stratification and suppresses vertical exchange; reduced exchange can trap heat and biogeochemical properties near the surface, further modifying density structure and mixing. This feedback is inherently coupled because temperature and salinity jointly set density, density regulates mixing, and mixing sets the evolution of both fields (Zika et al., 2018; Liu et al., 2023). At basin scales the coupling can appear weak because different forcings dominate the mean budgets, but at fronts and freshwater lenses it becomes decisive on day-to-week timescales, precisely where nonlinearities are strongest. O(10km) salinity observations target the scales where coupling is most active and where small errors in stratification and mixing can propagate upscale through biases in heat uptake, ventilation, and tracer transport.

Physical-biological connections are similarly conditioned by regime boundaries, particularly by the persistence of physical structures relative to ecosystem response times. Barrier layers and freshwater lenses create stratified surface habitats that persist for days to weeks, long enough for phytoplankton to respond but potentially too brief for higher trophic levels. Subsurface pathways transport nutrients and other tracers on multi-year routes, creating predictability only if mixing does not erase source signatures. In both cases, ecosystem predictability relates to how coherent pathways are and how fast mixing acts relative to transport. Advancing these connections involves observations that resolve the transition regimes where pathway coherence breaks down, where stratification and mixing flip between states, and where biological variability can be linked mechanistically to the evolving physical environment.