We utilized four months of glider observations sampling across the continental shelf in the southern Sea of Oman. The Seaglider carried out repeated sampling along an 80 km transect between 23.65° N, 58.65° E and 24.25° N, 59° E in 2015 (Fig. 1a). Observations were projected onto a straight across-shelf transect (Fig. 1a, black line) and median-binned onto a 0.5 m (vertical) and 2 km (horizontal) grid. The origin of coordinates was taken at the 300 m isobath. Binned sections were interpolated vertically (1 m), then horizontally (4 km) to fill minor gaps, and smoothed horizontally with a 6 km running mean. The 6 km running mean is selected as a compromise between filtering out submesoscale variability (≲O(10 km)) and preserving mesoscale structures (≳O(20 km), comparable to the local Rossby radius). Mixed layer depth was defined using a potential density threshold of 0.125 kgm-3 relative to the surface (Font et al., 2022). Spice (τ; kgm-3) was computed from Conservative Temperature (Graham and McDougall, 2013; McDougall, 2003) and Absolute Salinity (McDougall et al., 2012) following the TEOS-10 routines (gsw_spiciness0, McDougall and Barker, 2011). Potential density and spice are referenced at 0 dbar. Eddy kinetic energy (EKE) was estimated from two independent sources: (1) the depth-averaged currents (DAC) derived from the glider flight model as EKE=(DAC-DAC‾)2, where DAC‾ is the mean DAC during the glider campaign (Frajka-Williams et al., 2011); and (2) from sea surface height-derived surface geostrophic velocities from satellite (CMEMS, 2025). EKE is defined as u′2+v′2; where u′=u-u‾ and v′=v-v‾ are respectively the zonal and meridional velocity time anomalies, where u and v are the surface geostrophic velocities at the glider transect location and u‾ and v‾ are the mean surface geostrophic velocities during the glider-sampling period. Mesoscale eddies in the region were identified using the ToEddies dataset, which applies an eddy detection algorithm to satellite-derived Absolute Dynamic Topography (ADT) fields (Laxenaire et al., 2024). The dataset provides daily information on eddy properties, including polarity, center location, and spatial extent (Laxenaire et al., 2024; Ioannou et al., 2024).

In addition, we used a total of 3565 Argo profiles of temperature, salinity, and pressure between 2000 and 2023 in the Sea of Oman (22.5–26° N, 56–61° E, Fig. 1a). Only profiles flagged as “good” by the CORIOLIS Data Centre were used (Gaillard et al., 2009). Argo float coverage spans the entire domain, with an average density of 28 profiles per 0.25°×0.25° grid cell (Fig. 1a) and a relatively uniform monthly distribution (Fig. 1c). Argo profiles within a 200 km distance from the across-Sea of Oman transect (Fig. 1a, orange dashed line) were selected. This strategy ensures sufficient monthly sampling coverage in this sparsely observed region to construct an across-gulf monthly climatology. Moreover, to avoid the influence of shallow profiles on the continental shelf, profiles shallower than 1000 m were excluded, so the transect remains representative of the environment where mode water forms and persists. Each profile was then orthogonally projected onto the nearest point along the transect (the orange dashed line in Fig. 1a), which provides its along-transect coordinate. Profiles were vertically interpolated onto a uniform 2 m pressure grid, and all projected profiles were median-binned into 3 km horizontal bins along the transect. Averaging was performed on pressure levels. Monthly climatologies were produced by taking the median across all profiles within each (depth, distance) bin for each month between 2000 and 2023. This gridded product is then used as input for the σ-τ water-mass transformation calculations.

We defined the mode water core as the layer bounded by the 25 and 25.25 kgm-3 isopycnals, which effectively capture the low-stratification and low-potential vorticity typical of mode water. This density-based definition closely aligns with a more traditional identification based on potential vorticity (Feucher et al., 2022; Herraiz-Borreguero and Rintoul, 2011; McCartney, 1982), as demonstrated by a strong correlation between potential vorticity and the potential density threshold method (see Fig. S1 in the Supplement, mode water thickness correlation between methods is r2=0.96). Using a fixed isopycnal threshold simplifies the interpretation of water mass transformation processes within the σ-τ framework (see Sect. 2.2), while remaining consistent with dynamically meaningful definitions of mode water.

The analyses are performed in σ-τ coordinates since: (1) potential density is the natural coordinate to study ocean dynamics and provides information about the diapycnal nature of the mechanisms driving the volume change; and (2) spice is interpreted as a measure of thermohaline variability along isopycnals, reflecting isopycnally-compensated temperature and salinity changes associated with the spreading of distinct water masses (Jackett and McDougall, 1985; McDougall and Krzysik, 2015).

Based on the constructed Argo monthly climatology and the ocean glider dataset, we performed a volume budget in σ-τ coordinates for the interior ocean (i.e., below the mixed layer). The water-mass volume change dV/dt results from the sum of water-mass transformation, ∑U(σ,τ), and the exchange flux across the domain's geographical limits Ψ (Fig. 2a and Eq. (1), Evans et al., 2014; Portela et al., 2020b). U is the water-mass transformation due to isopycnal and diapycnal mixing fluxes, whereas its convergence or divergence is represented by the sum (∑) of the interior fluxes across the geographical limits of a given a σ-τ class (Donners et al., 2005; Walin, 1982). The isopycnal and diapycnal transformation rates represent the net mixing of water masses along and across density surfaces, respectively, resulting in an irreversible property change. Ψ represents the net exchange across the northern boundary of the domain, defined as the northernmost bin of the glider and climatology transect. Since the southern end is shelf-constrained, cross-section exchange there is assumed to be negligible. Note that an underestimation of the exchange flux Ψ could result in an apparent overestimation of the local transformation rate U. However, since both diapycnal and isopycnal transformations are estimated using the same underlying framework, there is no reason to expect a preferential bias toward either component. A diagram illustrating how changes in water characteristics are represented in σ-τ space is shown in Fig. 2a. The processes that modify the volume of a σ-τ class are depicted in geographical coordinates in Fig. 2b and c. The σ-τ class highlighted with a square in Fig. 2a is marked with dots in Fig. 2b and c. In summary, changes in volume within a given σ-τ class can be attributed to the convergence or divergence of interior fluxes across σ or τ boundaries, or via exchange fluxes across the domain boundaries within the same σ-τ class. Unlike interior mixing, exchange fluxes do not imply a transformation of water mass properties.

The method used here to compute water-mass transformation in the σ-τ coordinates is based on Portela et al. (2020b) and Evans et al. (2014). The volume change within a given σ-τ class can be expressed as: 1dVdt(σ′,τ′)=Uσ(σ′,τ′)+Uτ(σ′,τ′)+Ψ(σ′,τ′) where 2aUσ(σ,τ)=∫σ′=σΠ(τ,τ′)⋅uσ⋅dA and2bUτ(σ,τ)=∫τ′=τΠ(σ,σ′)⋅uτ⋅dA; where uσ and uτ are the diapycnal and isopycnal (diaspice) velocity components, respectively, and dA is the cross-sectional area of the transect occupied by water within the specified σ-τ class. Π is defined as 3Π(σ,σ′)=Π(τ,τ′)=1if σ∈σ′ and τ∈τ′0otherwise. where σ′=(σ±Δσ/2) and τ′=(τ±Δτ/2) and determines if a water parcel falls into a chosen σ-τ class. The volume integration was made in σ-τ classes with density intervals of 0.05 kgm-3 between 23–27.5 kgm-3 and spice intervals of 0.1 kgm-3 between 4–8.5 kgm-3. This choice reflects a trade-off between adequately representing each water mass class and ensuring that the mode water layer is well resolved.

Using Eq. (1), a set of linear equations can be generated to equate the volume trend and the interior water-mass transformation in σ-τ coordinates as dV/dt=Ax, where dV/dt is the observed change in volume of each sigma-spice class, divided by the relevant time interval; A is the matrix of coefficients of the linear equations; and x is the vector of the resulting diasurface transformations and exchange flux. This system is solved by means of a least squares regression for the unknown transformation and exchange flow. The detailed methodology can be found in Sect. S1 in the Supplement. The solution x was then decomposed into the transformation across spice and density surfaces and the exchange flux across the geographical domain.

The volume of water bounded by a given tracer iso-surface (e.g., here σ or τ ) can change due to mixing across the tracer iso-surface or due to volume transport into or out of the domain. However, it can also change by air-sea buoyancy fluxes if the tracer surface outcrops at the ocean surface (Evans et al., 2014, 2023; Groeskamp et al., 2019). To assess the role of surface forcing, we applied the water mass transformation framework in temperature-salinity (T-S) space (Evans et al., 2014, 2023). Using glider and Argo T-S observations, together with ERA5 air-sea buoyancy fluxes (Hersbach et al., 2020), we diagnosed the surface transformation of distinct T-S classes, thereby identifying which classes are actively transformed by air-sea fluxes. These classes were mapped into σ-τ space and compared with the σ-τ domain of the mode water (Fig. S2), showing no overlap. The classes influenced by surface fluxes lie well above the density range of the mode water (Fig. S2), indicating that surface buoyancy forcing does not influence the observed mode water transformations. Moreover, we exclude a subduction term through the base of the deepest mixed layer (in contrast to Portela et al., 2020b). This is justified because we start the water-mass transformation analysis well after mode waters are capped and the σ-τ classes lie well below the surface mixed layer under strong stratification (O(10-3)s-2).

The water-mass transformation framework was applied to 20 repeated ocean glider sections, each 60 km long and extending from the surface to 500 m depth, at the location indicated in black in Fig. 1a. To minimize the influence of shelf processes on the transformation, sections begin 10 km from the 300 m isobath, where depths exceed 500 m. The temporal resolution is approximately 3 d between successive sections. The first transect included in the analysis starts on 25 March, roughly a week after restratification, to minimize the influence of potential subduction/exchange with the surface mixed layer. Moreover, the framework is also applied to the across-gulf monthly climatology to obtain a climatological average of water mass transformations in the region. As restratification of the surface mixed layer typically completes by mid-March (Font et al., 2022), we restrict the analysis of monthly transformations to the period from April onwards. The residual errors for both datasets, computed as (|dV/dt-Ax)|/|(dV/dt)|, were on average of order 10-5, which is on the higher end of values reported in previous studies (e.g., Portela et al., 2020b).

We examine the seasonal-scale evolution of mode-water properties and transformations using the monthly Argo climatology (April–July). This climatological view captures the broad, low-frequency changes as the mode water evolves through late spring and early summer, but necessarily smooths over shorter-term variability. Mode water forms between February and March, and lingers as a homogeneous layer below the seasonal pycnocline (Fig. 1d and e), forming a boundary between the mixed layer above and the Arabian Sea oxygen minimum zone below (<60µmolkg-1) (Fig. 1e). Climatologically, the low stratification characteristic of mode water in the Sea of Oman is present until June (Fig. 1d), but it can persist longer inside eddies (Font et al., 2025; Liu and Li, 2013). Over seasonal timescales, the transformation framework reveals a net shift toward mintier (i.e., fresher and colder) and lighter waters, driven by negative isopycnal and diapycnal transformations (Fig. 3). Isopycnal transformation (Fig. 3a) indicates mixing into colder, fresher classes, while diapycnal transformation (Fig. 3c) reflects mixing into lighter classes. On average, the isopycnal transformation dominates (-0.015±0.009m2s-1), with a magnitude approximately three times larger than the diapycnal transformation (-0.005±0.008m2s-1) (Fig. 3a, c, and f). The signs of the transformation terms reflect the direction of the fluxes between density and spices classes, whereas the net Δσ and Δτ reflect the evolution of the volume-weighted mean properties. Thus, there is mode water volume formed due to transformation from lighter and mintier waters (red shading, Fig. 3b), but a large destruction predominantly in the lighter and spicier classes of mode water (blue shading, Fig. 3b). This fact, alongside a thinning of the mode water layer (ΔThickness=10 m per month, Fig. 3d), results in mean mintification and densification of mode water between March and June (Δτ=-0.03kgm-3 per month, Δσ=0.0013kgm-3 per month, Fig. 3e).

The previous analysis was based on monthly climatologies and, therefore, mesoscale and smaller scale contributions to the transformation are largely averaged out. In contrast to the climatological perspective, the high-resolution glider time series resolves submonthly variability and therefore reveals the episodic, intraseasonal changes in mode water structure and transformations that are absent from the monthly climatology. This allows us to directly quantify short-lived events associated with transient processes, such as mesoscale eddies. Over shorter timescales, the high-resolution time series provided by the ocean gliders offers a unique view that highlights the large variability of mode water thickness and depth due to eddy presence and topographical interactions with the shelf break (Figs. 1e and 3d). We observe a transition toward mintier waters throughout the study period (Δτ=-0.04kgm-3 per month), albeit with notable temporal variability as indicated by the sign changes in the isopycnal transformation (Fig. 3e and f). Until early April, waters become spicier (isopycnal transf.>0). Then, they shift to mintier conditions (isopycnal transf.<0), which last until the end of April. After that, waters get spicier again. This behaviour reverses in mid-May, and the period ends with mintier waters (isopycnal transf.<0). A similar pattern is observed in the diapycnal transformation. Over the sampled period, waters get slightly lighter (Δσ=0.001 kgm-3 per month) but with large variability (Δσ=±0.02 kgm-3, Fig. 3e). Densification occurs until early April (diapycnal transf.>0), followed by a period of lightening (diapycnal transf.<0), and a shift to densification after mid-May (diapycnal transf.>0, Fig. 3e). Notably, the isopycnal and diapycnal transformations are well correlated through most of the period until they begin to diverge in late April, when the isopycnal transformation changes sign approximately two weeks before the diapycnal transformation (Fig. 3f).

Despite the variability, the mean magnitude of the glider-based transformations is of the same order (10-2–10-3m2s-1) as the climatological transformation, with twice the contribution from isopycnal to diapycnal transformation, yet larger extremes for the latter (isopycnal: -0.004 ± 0.027m2s-1; diapycnal: 0.002 ± 0.029m2s-1, Fig. 3f). We suggest that the high variability in both positive and negative diapycnal transformations is averaged out in the monthly climatological sections.

Applying the water-mass transformation framework to high-resolution glider sections enables quantification of transformation variability across different temporal resolutions and scales. To assess how sampling period and resolution affect the ability to reproduce the high-resolution glider mean, the 3 d transformation time series (Fig. 3f) was sub-sampled over a continuous range of intervals from 3 d to monthly (Fig. 4, scatters). For each interval, we generated multiple realizations of a mean by shifting the starting time iteratively by 3 d steps – for example, a 6 d sampling interval yielded two mean estimates, while a 30 d interval produced 10. This approach provides a distribution of possible means for each effective sampling resolution. To illustrate how smoothing influences variability, we also applied rolling means of increasing window length to the 3 d series (shown as violin plots in Fig. 4). As the window size increases, extreme values in both isopycnal and diapycnal transformation are progressively damped (Fig. 4). The spread in the mean transformation (black markers) increases as resolution decreases (up to monthly climatology resolution, orange markers). This spread reflects the reduced likelihood of recovering the 3 d mean. If each transformation were insensitive to the sampling frequency, its magnitude would remain close to the high-resolution mean regardless of the sampling period. From an Argo monthly climatological perspective, the probability of recovering the 3 d sampled mean within one standard error is about 60 % for the diapycnal transformation and 40 % for the isopycnal transformation. Using the glider sub-sampled timeseries, this likelihood drops with coarser sampling: from 75 % at 6 d intervals to less than 50 % at sampling frequencies longer than 12 d. These results highlight the importance of collecting data at least weekly to capture the influence of high-frequency variability. For example, sampling only once every 30 d reduces the chance of capturing the 3 d mean by about 85 % compared to 3 d sampling.

The high-resolution observations provide an excellent dataset to analyze the temporal and spatial evolution of mesoscale eddy activity and associated water mass transformation within the mode water from mid-March to early June. The ToEddies Global Mesoscale Eddy Atlas (see Sect. 2.1, Laxenaire et al., 2024) reveals the presence and persistence of anticyclonic and cyclonic eddies in the Sea of Oman during this period (Fig. 5a), as well as the crossing of three anticyclones along the glider's trajectory (Fig. 5b). These mesoscale structures dominate the surface circulation during the sampling period and undergo gradual displacement and structural changes over time. The glider's path through these eddies (Fig. 5b) captures the dynamic interaction between the platform and the evolving eddy field. Enhanced EKE values (Fig. 5c) correspond to periods of intensified mesoscale activity and potential eddy-glider encounters. The agreement between the satellite-derived EKE and the glider-derived EKE confirms the presence of high-EKE periods during the campaign.

The glider-derived observations reveal the vertical imprint of these eddies on the thermohaline structure and associated water mass transformation. The observed vertical displacements of isopycnals and intrusions of anomalous water masses (Fig. 5e) are consistent with mesoscale stirring. These anomalies coincide with changes in isopycnal and diapycnal transformation along with mesoscale eddy activity (Fig. 5d), suggesting an active role of mesoscale dynamics in modulating vertical exchanges and interior ventilation. For instance, around 6 April, when an anticyclone is present, a marked increase in surface EKE (Fig. 5b) corresponds with simultaneous peaks in both isopycnal and diapycnal transformation rates of change (|ΔTransf.|, Fig. 5c and d), suggesting intensified stirring and mixing as the glider samples through the eddy structure.

Around 15 April (non-eddy case), the mesoscale eddy moves away from the glider transect, as indicated by both the ADT maps (Fig. 5a3) and the glider spice time series, where isopycnals flatten (Fig. 5e). The glider trajectory during this period does not intersect any prominent eddy (Fig. 5b), suggesting that it is transiting through a relatively quiescent background flow. Correspondingly, the transformation rates, both diapycnal and isopycnal, are stable and similar to the ones derived from the climatology (negative and O(10-2), Figs. 3f and 5d). This calm period contrasts with the more dynamic intervals before and after, suggesting that enhanced transformation is closely tied to eddy presence and activity. The transformation around 15 April thus serves as an ad hoc baseline, highlighting the mesoscale-driven nature of the stronger exchange events observed throughout the record.

A second event occurs around 27 April–3 May (Eddy case), when the glider crosses an anticyclone again (Fig. 5a4, b, and e) and isopycnal transformation increases (Fig. 5d), underscoring the dominant role of mesoscale processes in lateral tracer transport. This is followed by a pronounced diapycnal transformation event, likely driven by vertical exchanges at eddy boundaries and/or interaction with shelf bathymetry.

To investigate the spatial structure of water-mass transformation and its dependence on mesoscale eddy activity, we remapped the transformations in distance-pressure coordinates from the shelf. The transformation rates, when averaged over the repeated sections of the entire glider deployment (Fig. 6a and b) show weak diapycnal and isopycnal transformations. This mean view closely resembles those derived from the monthly climatological dataset, despite the climatology being composed of basin-wide Argo profiles (Fig. 3b), reaffirming that glider-based observations can reproduce the climatological view when averaged over long enough timescales, and that high-frequency variability is largely smoothed out in monthly means.

We compare the average transformation sections with two contrasting case studies: one with no eddy presence (Fig. 6d–f) and one with an eddy (Fig. 6g–i, the periods marked in Fig. 5e). The Non-Eddy Case (Fig. 6d–f) provides a baseline scenario where the glider transect occurs in a region with no identifiable eddy, as confirmed by the lack of coherent structures in the ADT maps (Fig. 5a3). Both isopycnal and diapycnal transformation rates are comparable to the climatological mean and are relatively weak within the mode water core (∼-0.01 and 0.005 m2s-1, respectively, Figs. 3, 6d and e).

The Eddy case (Fig. 6g–i) captures a glider transect through an anticyclonic eddy. In contrast to the quiescent background conditions seen in the Non-Eddy Case, this eddy drives an order of magnitude enhancement of diapycnal transformation above and below the mode water core (∼0.05m2s-1), pointing to intense vertical exchanges associated with eddy-induced mixing. Isopycnal transformation is also intensified and spatially coherent along sloping density surfaces (∼0.04m2s-1). The elevated isopycnal transformation rates imply vigorous lateral mixing along density surfaces, enhanced by eddy-driven stirring and frontal slumping. Such strong transformation rates suggest that the eddy acts as an efficient conduit for water-mass transformation, especially within mode water density layers. This case highlights the dynamically active role of anticyclonic eddies in modulating both vertical and lateral mixing processes, and thus promoting the redistribution of heat and salt and contributing to the ventilation of the ocean interior.

The exchange term (grey bars in Fig. 6c, f, and i), calculated as the residual in the transformation budget, serves as a proxy for advective fluxes through the section and can highlight dynamically active regions where advection dominates over mixing. In the Eddy Case, this term is particularly pronounced, indicating a strong advective component related to eddy presence (0.59±0.48m2s-1, Fig. 6i). In contrast, the exchange flux is much weaker in both the mean section and the Non-Eddy Case (0.16±0.13m2s-1, Fig. 6c and f), suggesting a reduced role of advective processes. Eddies, both surface and subsurface, are well known to transport coherent parcels across long distances (e.g., Frenger et al., 2018), driving the strong exchange seen here. Altogether, the exchange term adds an important dimension to the volume budget, helping to quantify the net effect of non-mixing processes, and further supporting the view that mesoscale advection can be a dominant pathway for ventilation and altering stratification in the upper ocean.

The impact of the dynamical regime on transformation magnitudes within the mode water layer is striking (Fig. 6j–l). The passage of mesoscale eddies leads to a marked intensification of all three components of the volume budget within the mode water layer, reflecting enhanced isopycnal and diapycnal mixing as well as increased advective exchange (see summary schematic in Fig. 2). Compared to non-eddy conditions, diapycnal, isopycnal, and exchange fluxes increase by 61 %, 45 %, and 66 %, respectively (Fig. 6j–l). In the absence of eddies, mode water undergoes significantly weaker transformation – with transformation magnitudes similar to the climatological average (Fig. 6j–l). It is noteworthy that these estimates represent integrated values across the glider transect, and local transformation rates may vary substantially, enhanced or suppressed, depending on eddy position, intensity, and interactions with topography.

To further contextualize these differences, we compared the transformation rates under eddy and non-eddy conditions with the climatological mean derived from Argo profiles. Diapycnal transformation exhibits the largest relative increase during eddy periods, rising by 163 % compared to the climatological mean (Fig. 6j). This substantial enhancement likely reflects intensified vertical exchange driven by eddy-related processes that are absent from monthly climatologies. Even outside of eddy influence, the diapycnal transformation from the glider data remains elevated (∼0.13m2s-1, a 63 % increase relative to the climatological mean), likely due to the glider's ability to resolve finer-scale vertical structure and fluxes resulting from diapycnal mixing.

Isopycnal transformation also increases under eddy influence, with a pronounced peak in the mode water layer (∼0.16m2s-1), corresponding to a 43 % rise over the climatological mean (Fig. 6k). This points to enhanced lateral stirring and strain-driven intrusions along density surfaces, characteristic of mesoscale eddy activity. In contrast, during non-eddy conditions, isopycnal transformation remains weaker and closely matches the climatological estimate (∼0.11m2s-1). This suggests that eddy-related strain and intrusions play a major role in redistributing water masses and altering thermohaline structure within the mode water. In the absence of eddies, isopycnal transformation is reduced and comparable to the climatological mean.

The exchange component (Fig. 6l) similarly shows a pronounced peak in the presence of eddies (∼0.5m2s-1, 140 % increase relative to climatology), and is notably lower under non-eddy conditions (∼0.3m2s-1, 45 % above the climatological mean). The greater exchange fluxes observed in the high-resolution glider dataset reflect the contribution of strong mesoscale advection not captured in the monthly climatologies.