Nearly half the Pacific Ocean comprises abyssal zones that have experienced persistent warming in the past 30 years . The bottom mixed layer (BML), a well-mixed region directly above the seafloor in the abyssal ocean is a critical interface where turbulent mixing facilitates exchange between the deep ocean interior and the seafloor, influencing water–mass transformation and global circulation. Dynamics within the BML are affected by internal wave activity , and near-boundary turbulence , impacting diapycnal mixing, deep-sea food web connectivity, and heat transport . Additionally, the BML region may contribute to abyssal mixing, as the interplay of turbulent processes through internal tides and stratification here facilitates diapycnal mixing , thereby helping to drive the meridional overturning circulation (MOC) . MOC is sustained by the abyssal flow of North Atlantic Deep Water (NADW) and Antarctic Bottom Water (AABW), which transport dense water masses equatorward and poleward from their formation regions through the deep ocean. The pathways involved in the MOC have been broadly identified, with general consensus on their origins, particularly in the ventilation regions adjacent to Antarctica (AABW) and the Labrador Sea (NADW) . While the broad pathways of these abyssal waters are accepted, the detailed mechanisms by which they return to the surface through diapycnal mixing and upwelling remain active areas of research . Regions of intense abyssal mixing over mid-ocean ridges and narrow inter-basin channels are key sites where stratification governs regional variability in the BML, current pathways and internal tide generation, connecting closely to abyssal pathways, while seafloor topography remains the primary control on BML thickness at the global scale . However, the relative importance between topography, its spatial scales, and the dynamic processes within ocean basins dictating the BML thickness remains unclear .

The strength of deep MOC in the Pacific Ocean has been historically underestimated due to a lack of data and its complex topographies, making simulations more challenging . Broadly, AABW enters the Pacific Ocean along the eastern side of the Tonga–Kermadec Ridge , then narrows through the Samoan Passage before bifurcating to the west and north towards the Japan Trench (Fig. ). North of the Samoan Passage, there is also bottom water transport to the east and south of the Hawaiian Ridge. Around the Hawaiian Ridge, energetic baroclinic tides are generated over the rough seafloor, contributing to distinct differences in the eastern and western regions of the Pacific Ocean, with larger dissipation in the western Pacific . AABW transforms into North Pacific Deep Water (NPDW) while reaching the North Pacific. It is then further transformed through deep ocean mixing while traveling south and reinforcing subsurface stratification and linking to deep convection in the Southern Ocean .

While rough and variable topography can modulate the BML thickness (e.g. fracture zones and seamounts ) regions of the seafloor with broadly similar depths or geomorphology can nevertheless exhibit vastly different BML thicknesses due to differences in ocean dynamics such as boundary currents or abyssal transformations . Along continental shelf regions, it is on the order of 40–70 m in the South China Sea and 5–15 m along the Northern California Shelf . The mean BML thickness over different latitudes in the North Atlantic Basin has been reported as 30–60 m . Across the Drake Passage, it was found to be over 100 m, similar to Gulf Stream regions . These differences may also be due to methodology. For example, the region immediately north of the Puerto Rico Trench in the North Atlantic (21° N, 66° W) has a BML thickness reportedly ranging from 80–800 m Fig. 9 in, 60–100 m Fig. 2b in and 80 m Fig. 1 in with the variation likely attributed to different methodology and spatial interpolation.

The transfer of mass and momentum between the ocean interior and the seafloor occurs via the BML. Yet in most large-scale ocean circulation models, it is generally unrepresented. As a result, robust parameterizations of bottom boundary processes are essential . initially proposed a vertically integrated, one-dimensional framework to estimate diapycnal mixing rates averaged across the ocean interior. However, it has been subsequently found that mixing in the bottom boundary is inherently three-dimensional, shaped by turbulent processes influenced by topography and internal wave dynamics . Simple bottom boundary layer parameterizations assume local, steady-state velocity shear and stratification relationships to simulate turbulent mixing and momentum transfer vertically, potentially neglecting variability in the BML thickness . Recent improvements have incorporated wave-driven turbulence and terrain-following schemes , some of which include profiles of diffusivity and viscosity in their parameterization . Nevertheless, additional observations along the ocean's bottom boundary remain crucial, not only for validating models, but for resolving the spatial and temporal variability in the BML structure that underpins interior-seafloor exchange .

The BML thickness is most commonly defined as the thickness above the seabed at which a variable (typically conservative temperature or density) deviates from the seafloor value by a specified threshold (also known as the “threshold method”). While we use a hydrographic definition of the mixed layer, it is important to note that this does not necessarily coincide with the dynamically active mixing layer defined by turbulence; distinguishing these layers is increasingly recognized as essential when inferring mixing intensity and water–mass transformation. While we refer to this as the bottom mixed layer, its thickness may reflect varied bottom boundary processes, not exclusively active mixing. It is inhomogeneous throughout the world's oceans, not only because of the varying depth and roughness of the seafloor but also because of the influence of differing oceanographic processes in each region. Different oceanographic conditions require varying thresholds to calculate the BML thickness. For instance, weakly stratified abyssal regions necessitate small density thresholds (1×10-3 kgm-3), while highly turbulent areas are better suited to larger threshold values Fig. 1 in, or sensitivity in the threshold value may be directly related to instrument noise . To overcome sensitivity and subjectivity in threshold (or gradient) selections, approaches like the relative variance and integrated methods have been developed to provide more robust, non-arbitrary BML thickness estimates. The BML thickness serves as a useful proxy for characterizing diapycnal upwelling , nutrient transfer , sediment transport and the development of bottom boundary conditions and parameterizations in ocean models; therefore, the methods and outputs require diligent evaluation of their physical validity across spatial and temporal scales.

Our first objective is to evaluate BML thickness methodological approaches suitable for data-poor regions of the ocean. We focus on four methods: the threshold method, the threshold-gradient method, a relative variance method, and a split-and-merge algorithm to derive the BML thickness and critique their use and relevance in an abyssal basin setting. The second objective is to understand the BML variability across the central and eastern Pacific abyssal ocean, including assessment of the connection to internal tidal energy dissipation and bottom water pathways. To achieve these objectives, we collected surface-to-seafloor temperature-pressure profiles across the central and eastern Pacific Ocean and complemented them with publicly available repeat hydrographic datasets. As we show in the sections that follow, these data reveal new insights into the dynamics of the BML and its connection to broader abyssal processes.

The Trans-Pacific Transit (TPT) Expedition occurred over six individual legs, with the duration of each leg approximately 21 d, between June 2023 and January 2024 on Research Vessel (R/V) Dagon . The vessel covered 20° longitude and 30° latitude over the central and eastern abyssal Pacific Ocean, including the Molokai and Clarion Clipperton fracture zones (Fig. ). Bathymetry and backscatter intensity data were acquired throughout the expedition using a hull-mounted Kongsberg EM124 multibeam echosounder. At each site, three autonomous landers were deployed in a roughly 2 km equilateral triangle, acquiring a total of 73 surface-to-seafloor profiles of temperature and pressure (at a frequency of 2 Hz secured to the landers). These data are referred to here as “TPT profiles”. Conductivity, temperature and depth (CTD) sensor profiles from repeat Global Ocean Ship-Based Hydrographic Investigations Program (GO-SHIP) hydrographic sections, P16 and P02, provided observations within the study regions along meridional and zonal transects, respectively (see locations in Fig. ). The GO-SHIP sections were used in two ways: first, they were used to complete Gaussian mixture modelling (GMM), which was applied to the TPT profiles to generate modelled practical salinity (SP) profiles (see Methods section on GMM); second, they provided additional locations to derive the bottom mixed layer (BML) thickness. The BML derivation, followed by a random forest regressor, was applied to the GMM-derived TPT profiles and GO-SHIP profiles as one dataset and is detailed in the following sections.

Density profiles over the central and eastern Pacific Ocean provide an inhomogeneous outlook of BML thickness variations at abyssal depths across plains and topographic features. Basin-scale expeditions such as the TPT voyages are frequently multidisciplinary in scope, with competing demands on vessel time. Incorporation of these profiles with the repeat GO-SHIP profiles provides increased understanding of the BML. Using RF methods, we found that bottom depth, total internal tide dissipation and slope are the highest-performing features to predict the BML thickness in this region. Because several predictors share spatial structure (e.g., bottom depth and total dissipation), the RF highlights associations rather than uniquely isolating independent physical drivers, and this limitation should be considered when interpreting the feature importance results .

In the central and eastern Pacific abyssal ocean, the thickness of the BML was inhomogeneous with an average value of 226±172 m. The BML was calculated using the (Θ) profiles through the threshold method (0.003 °C) for the study region. There is no accepted standard methodology for calculating the BML thickness. It often depends on user-defined numerical values within those methods and depends on the region of interest. The integrated method proposed by , was used to calculate the BML depth globally, providing an average Pacific Ocean BML thickness of 64 m . We found that for our abyssal ocean context, using an integrated approach that combines multiple methods and calculates a QI to get the highest “quality” BML thickness generated spurious results for profiles within three kilometres of one another, making it difficult to compare BMLs estimated with different methods. Although the QI was calculated, visual interpretation was necessary to confirm the results, mirroring the approach taken within the integrated method where visual identification was still needed . The variability in BML thickness is not unexpected given the variation in topographic features across the region, likely changes in friction velocity, and a wide longitudinal and latitudinal range . Profiles from the TPT Expedition broadly followed the same spatial patterns as those from the hydrographic sections and show similar spatial variations in BML thickness as .

In all instances, the RF identified the bottom depth, slope, total internal wave energy dissipation, TRI and low-mode wave-wave interactions as the most important predictors of the BML thickness in this Pacific Ocean abyssal setting. These results are physically intuitive, with the bottom depth constraining the maximum possible BML thickness, background stratification and the vertical extent available for turbulent mixing, which is consistent with past research . Importantly, dissipation patterns in the de Lavergne et al. (2019) dataset are not set by depth, but by the distribution of internal tide energy sources, scattering pathways, and nonlinear wave–wave losses. Deep basins may accumulate low-mode energy and therefore show elevated dissipation, but this reflects remote energy propagation and decay rather than a mechanistic link to depth itself. Therefore, even where depth and dissipation appear spatially aligned in our RF model, this similarity arises from shared basin-scale structure rather than depth acting as a direct driver of turbulent energy loss.

The total internal wave energy dissipation value aggregates all internal tide dissipation mechanisms that drive turbulence . On abyssal plains, where local topographic features are sparse, a substantial amount of this energy would likely originate remotely and dissipate gradually through sustained mixing events . Therefore, the inclusion of low-mode wave-wave interactions as a predictor is especially significant. This variable refers to nonlinear energy transfers among long-wavelength internal tides. In regions with high low-mode wave-wave dissipation, this could lead to persistent near-bottom mixing that expands the BML thickness. This suggests that the BML thickness on the abyssal plain is of remote and sustained forcing origin, rather than high-mode breaking events . Although the study region lies predominantly within the abyssal plain, the terrain is not uniform. As shown by and Fig. b and c, the region is interspersed with multiple features of abyssal plains, abyssal hills, and seamounts, creating heterogeneity in the TRI and slope. The observations north of Hawaii highlight where higher TRI and slope intersect with the smallest values of total internal tide dissipation and low-mode wave-wave interactions.

The TRI captures local bathymetric complexity at 50 km scales, which enhances bottom drag and internal tide scattering, even where mean slopes may be weak, supporting thicker BMLs by maintaining sustained and patchy mixing close to the boundary layer . The slope of the topography within our study region is primarily of a subcritical regime; therefore, internal tides will refract and reflect weakly, allowing for persistent low-mode energy to mix over broader regions, rather than localized mixing. Between 4 and 15° N there is a small region where the low-mode critical slope dissipation increases (Fig. g), and the BML thickness is large, suggesting the critical slope may be more important here. Nested within the same region is the highest total dissipation of the study region (4–7° N) where there is high local high-mode dissipation, low-mode wave-wave interactions and low-mode critical slope dissipation. The TRI and slope are small over this region, culminating in the BML thickness being slightly above the average. Similar connections to the slope and the TRI have been identified in the North Atlantic Ocean and the South China Sea .

Our results have highlighted differences in what factors drive the BML. Despite limited and sparsely located data points, it is clear that the BML thickness is a culmination of processes, both local and remote. The limited and spatially inconsistent data points meant we were unable to further the model to predict the BML without the RMSE at times equating to the predicted BML thickness. This correspondence between dissipation and hydrographic BML thickness further suggests that, in this region, the threshold-defined BML is not merely a passive hydrographic feature but may effectively capture the vertical extent over which mixing is dynamically active. Such alignment between hydrography and turbulence is rarely shown explicitly for the abyssal ocean and may point to an underappreciated sensitivity of BML structure to the local dissipation field. Despite the three highest importance features remaining consistent, the nine iterations of random_state values do not visually provide a clear picture of regions that are consistently lower or higher performing than others. This variability, combined with shared spatial patterns among some predictors, further underscores that the RF approach here is more diagnostic than predictive and should not be interpreted as uniquely isolating mechanistic controls. In addition, the reasonably high variation in r2 and RMSE values suggests that more observations are required for there to be less sensitivity in the results to which random selection of the data is chosen to train the model. The RF herein should be used to understand drivers but cannot be used predictively, which would require additional data points. However, in regions where there is a more dense and equal spread of CTD profiles, the publicly available datasets from and GMRT bathymetry should be considered for usable predictive relationships. Prediction of the BML thickness was not in the scope of this study; however, we have shown the usefulness of publicly available datasets. Predictive relationships of the BML thickness would be useful for identifying regions of interest for internal wave-driven mixing at the ocean's bottom boundary, hydrodynamic model parameterisations and disentangling spatiotemporal variability in BML thickness within a given region.

At around 18° N, bottom water passes through the Horizon Passage (170° W) and flows around Hawaii , intersecting where the BML was small and there was increased stratification in the water column above. This can be demonstrated by Θ-SA profiles with increased fractions of North Pacific Intermediate Water (NPIW) from 16–19° N (Fig. a) and more saline and cooler water at the seafloor within the BML, aligning with Antarctic Bottom Water (AABW) properties (Fig. b). While the complete profiles between 0–2 and 10–15° N displayed visually similar water mass characteristics, the properties of the BML were distinct (X marks in Fig. ) with the equatorial seafloor BML fresher and warmer, indicating NPDW, compared to 10–15° N closer to the properties of AABW, and 16–19° N the most saline and coolest . This region of water mass and inter-basin exchange highlights the difference between stratified regions of bottom water pathways (Fig. ) compared to low ocean interior stratification south of this region (curved Θ-SA in Fig. b, red and blue) and less variation in Θ-SA space (Fig. a) . In essence, a more strongly stratified ocean interior likely suppresses mixing by reducing the turbulent diffusivity, even in regions where the turbulent kinetic energy dissipation may be high. Therefore, the buoyancy gradient remains difficult to overcome, resulting in a thinner BML in AABW regions compared to regions of NPDW at the equator .

Consistent with previous analyses , there are multiple processes influencing the BML thickness at abyssal basin scales. While the RF provides a quantitative approach to dissect the variations in BML thickness based on the features, the profiles are a single snapshot of the water column at that point in time. As exemplified by in the Clarion–Clipperton Fracture Zone, they were unable to define the diffusion processes of the suspended sediment within the BML, as additional short-term processes such as internal gravity waves were highlighted as likely influencing the results. In our case, the TRI and slope are single values over a 50 km buffer region, not including proximity and direction from features such as the Hawaiian Ridge, which may influence the formation of the BML and the water column in different ways; hence the inclusion of internal tidal dissipation from de Lavergne . Considering the broader consequences of BML dynamics for deep ocean mixing and overturning circulation, temporal variability in the BML over abyssal depths should be considered in future studies. For example, a mooring configuration both within the BML thickness and above, in a region of increased internal tide energy dissipation south of Hawaii and then at a similar depth to the north-east of Hawaii where the BML thickness is higher and dissipation is lower, while intersecting with a region of water mass transport. These locations transition from flat abyssal plains to the Hawaiian Islands, each with distinct BML patterns and drivers across ∼6° of latitude. Increased observations of both direct mixing and far-field or wave-wave energy dynamics are required in this relatively dynamic, yet undersampled region of the abyssal Pacific Ocean.

The BML is crucial for understanding diapycnal transport, which causes significant upward movement of deep-sea waters in the undersampled abyssal ocean. This research highlights the importance of abyssal seafloor regions, which are not typically categorized as dynamic, shifting in space and time. Through the application of four BML detection methods, we find that the commonly used threshold method provides the most consistent and interpretable estimates of BML thickness across large spatial scales. However, we have highlighted the necessity to test each method-specific parameter. We also show that GMM offers a useful approach for predicting essential ocean variables, such as salinity here, using publicly available data. The RF revealed BML thickness variation related primarily to bottom depth, followed by total internal tide energy dissipation and topographic slope.

Global studies of the BML using multiple methods seldom focus on the potential of variability over time , while others aim to use a small range in time to comprehend processes such as sediment dispersal within the BML reaching the conclusion of significant temporal variability long noted in literature . The relative contributions of the mechanisms that control the BML thickness across the abyssal ocean and basin boundaries requires further investigation through increased continuous observations and modelling efforts with reduced interpolation. The role of abyssal circulation pathways and internal tide driven mixing is at the forefront of current research e.g., within which the formation of the BML forms a key component of the processes. Therefore, the present study highlights and encourages sustained observations of abyssal regions over the bottom boundary and ocean interior above. Such observations are particularly important around rough topography, specifically in the central and eastern Pacific, where the abyssal ocean is frequently overlooked. At present, the temporal scales of BML variability remain poorly understood. Determining these scales is essential for characterising how the BML is mediated by abyssal water–mass transformation and circulation.