Ocean Science PhD: Models, AI and Observations of Mixing and Deep-Ocean Ventilation

Ocean Science PhD: Models, AI and Observations of Mixing and Deep-Ocean Ventilation

Executive assessment

This PhD links physical-ocean observations with numerical modelling and scientific machine learning to ask how mixing inside deep fracture zones transforms dense abyssal waters and contributes to the return flow of the global overturning circulation. The proposed work combines long-term tracer evidence, turbulence measurements, an open fracture-zone configuration of MITgcm and a physics-informed AI emulator.

The opportunity's value is the physical question and the uncommon bridge between observations and modelling, rather than AI as a label. Exeter’s team brings together tracer analysis and water-mass modelling, while recent field and simulation studies show that the Brazil Basin remains an active research area. Scope is the main risk: the PhD could require curating tracer and turbulence data, building an open MITgcm domain, diagnosing mixing and overturning, improving parameterization, and training and validating a surrogate. The award covers up to 3.5 years. Before applying, establish which data and model components already exist and which result defines the thesis core.

What the project proposes

The project starts from the deep Brazil Basin and the Rio de Janeiro Fracture Zone, a passage linking the Brazil and Angola Basins. The description reports that a 22-year tracer-release record provides evidence of roughly 2,000 m of abyssal upwelling through this passage. The research question is how local turbulence, topography and stratification create water-mass transformation and tracer transport, and how those processes connect to basin-scale overturning.

The advertised work has three linked layers:

1. Interpret tracer observations and turbulence measurements to quantify mixing, transport and water-mass change.
2. Extend an established MITgcm framework from canyon simulations to an open fracture-zone geometry and test how topography, stratification and mixing affect tracer dispersion and circulation.
3. Train a low-cost, physics-informed emulator on ensembles of simulations, with the aim of rapid sensitivity tests, uncertainty quantification and evaluation of mixing parameterizations.

The project names Marie-José Messias as lead supervisor, Neill Mackay as an Exeter co-supervisor and Herlé Mercier as an external collaborator, and identifies Physical Geography as the PhD field. It does not specify the exact data files, grid and domain design, model version, ML architecture, whether a cruise is expected, or the sequence and scope of work packages. Clarify what the stated “22-year” record comprises: the 2024 BBTRE paper describes a 50-month experiment that began in 1996, which may be one component of a broader evidence base rather than the same record.

Research setting and supervision

Exeter has a relevant Geography and Earth Systems profile. The University’s REF 2021 table reports 55% 4-star and 36% 3-star activity for Geography and Environmental Studies, and 63% 4-star plus 27% 3-star for Earth Systems and Environmental Sciences. Exeter’s Geography page lists the subject in the global top 40 of the 2026 QS subject ranking and reports sixth in the UK for 4-star Geography and Environmental Studies research. These are institution-level signals, not project ratings.

The Coastal and Ocean Futures group describes work spanning ship and autonomous observations, CFC/SF6 and other ocean tracers, numerical models, ocean physics and machine learning. The TICTOC team identifies Messias as responsible for transient-tracer studies. Mackay’s 2024 paper on an optimal transformation method uses water-mass theory and tracer distributions to examine circulation, mixing and ocean-carbon fluxes. This gives the project a credible mix of observation and quantitative modeling experience. Mercier’s named role adds external oceanographic collaboration, though the official project brief should be used to confirm the precise co-supervision arrangement.

The public evidence does not confirm an incoming student’s access to the advertised tracer series, processed microstructure measurements, a research vessel, a particular HPC/GPU allocation or a prepared training-data set. Those are decisive feasibility questions. The broad group infrastructure is a positive setting signal, but it is not proof that a specific resource is allocated to this project.

Scientific frontier and feasibility

Brazil Basin observations make the problem unusually well anchored. The 2024 BBTRE observations paper describes a tracer-release and turbulence program in the deep basin. Earlier tracer studies estimated deep diffusivity from the plume’s long-term spread. A 2017 fracture-zone study used moorings and microstructure profilers to show that a sill-accelerated two-layer exchange can coexist with turbulence shaped by mean flow, tides and near-inertial waves. A 2018 study proposed that wave–mean-flow interaction can drive abyssal upwelling within fracture zones. Separate canyon dye observations in 2024 offer a direct measurement strategy for cross-density transport, although that site and geometry differ from the project’s Rio de Janeiro passage.

A particularly important frontier signal is the peer-reviewed Brazil Basin study published on 27 October 2026, the day before this opportunity’s advertised posting date. It compares 1, 3 and 6 km CROCO simulations with in-situ observations. Higher resolution produces more eddies and stronger modeled mixing; the simulations can match observed near-bottom values while overestimating mixing in the stratified interior. The paper attributes part of the error to smoothed bathymetry, which weakens modeled internal tides while leaving too much mixing associated with eddies. It does not use the project’s advertised MITgcm configuration and does not study precisely the same passage. Still, it makes the central validation issue concrete: increasing resolution or adding an emulator does not automatically reduce structural model error.

MITgcm’s documented configuration tools support configurable grids, geometry, parameterizations, diagnostics and verification experiments. An open fracture-zone domain nevertheless needs careful choices for bathymetry, boundaries, forcing, resolution, numerical stability and comparison data. A quasi-realistic Brazil Basin modeling paper from 2022 offers a process-model precedent. The 2026 CROCO study provides a cautionary contrast about the consequences of resolution and bathymetric filtering.

Ocean-mixing machine-learning work is active, but much of the retrieved precedent concerns upper-ocean boundary layers or regional circulation. A 2022 deep-neural-network study uses independent large-eddy-simulation test data; a 2024 regional-model paper embeds a learned vertical-mixing scheme; and the 2026 NORi paper combines a physics-based closure with neural ordinary differential equations and tests long integrations. These works support the plausibility of physics-plus-ML methods, while showing what the proposed PhD must do to be credible: define the target quantity, preserve physical constraints, reserve independent observational/model cases, test long-run stability and report where extrapolation fails. Their surface/regional results do not establish that an abyssal fracture-zone emulator will work.

Feasibility judgment: scientifically strong, technically high-risk unless the team has an inherited model baseline, curated data and compute allocation. A realistic 3.5-year path would first secure and reproduce the observations/model configuration, then answer one bounded physical question, and treat the surrogate as a later objective with agreed validation criteria. Ask whether machine learning is required for the thesis outcome or is an exploratory extension.

Funding, deadline and costs

The EPSRC Doctoral Landscape Award is a competitive cohort-wide competition for September 2027 entry. Exeter lists up to 16 awards across its eligible projects, not 16 awards for this topic. The project application deadline is 12:00 noon on 8 December 2026. The official route is the application link in the relevant project PDF. The scheme guidance allows applications to up to three projects; if doing so, identify the preferred project in the personal statement.

The current scheme page states up to 3.5 years of full-time funding, a £21,805 annual stipend at the 2026/27 rate and a £5,000 Research Training Support Grant across the award term. The 2027/28 rate may be revised. Verify how the three-and-a-half funded years align with PhD registration and any final writing period. The guidance says international fee-status applicants can compete for a limited share of full awards (up to 30% of the cohort); successful international awardees have the tuition-fee difference covered by an Exeter waiver. Visa, Immigration Health Surcharge and relocation costs remain the student’s responsibility. Residence rules are detailed; confirm fee status directly if recent residence or immigration history could affect it.

For the application, Exeter requests a project-specific personal statement, CV, full transcript(s), two references and English evidence where applicable; documents should be in English and qualifications need certified translations. The central DLA guidance gives references a 15 February 2027 deadline, while official departmental award pages for the same 2027/28 competition (Refs 6012 and 6014) state 8 January 2027. The project-linked Geography award page (Ref 6015) could not be retrieved in this audit, so 8 January is not confirmed for this project. To avoid a missed-document risk, plan for 8 January and obtain written confirmation from PGR Admissions; submit the main application by noon on 8 December. The central guidance also retains stale 2026 interview/outcome dates, so confirm the current schedule.

Exeter’s budgeting guidance estimates £1,300–£2,000 per month for basic living costs, based on April 2026 local spending. Compared with the current £21,805 stipend, this implies a simple annual margin from about +£6,205 at the low budget estimate to −£2,195 at the high estimate. This arithmetic excludes personal costs, relocation, visa/IHS, rate changes and any unfunded registration period. A private room is listed at £135–£250 per week before bills; self-catered university halls are £150–£270 per week including bills. International applicants should prepare a separate one-off-cost budget.

Applicant fit and preparation

The scheme expects a First or Upper Second (or international equivalent) in a relevant science, engineering or technology subject; a Lower Second may be considered alongside a master’s degree or substantial relevant non-academic experience. The project description welcomes physical oceanography, physics, mathematics, computer science, environmental science, engineering, data science and related quantitative fields. Quantitative and computational strengths matter; prior oceanography is helpful but not stated as essential.

Good application evidence would include a computational research project, reproducible code or data workflow, numerical methods, fluid mechanics or geophysical modeling, and an example of checking a model result against observations. Explain one physical mechanism you would test and how you would distinguish model calibration from independent validation. Strong candidates should show interest in both physical interpretation and careful uncertainty analysis, not just model speed or AI novelty.

Before applying, review tracer advection/diffusion, stratification, turbulent mixing, ocean overturning, MITgcm configuration and train/validation/test design. A small reproducible sensitivity analysis on advection–diffusion or model parameter variation could demonstrate preparation, but the public brief does not require a coding portfolio or a particular model experience.

Career value

The project can build transferable skills in ocean observation analysis, numerical modeling, scientific Python, geophysical fluid dynamics, uncertainty quantification and physics-informed machine learning. The UK National Careers Service describes oceanographers as analyzing observations and building circulation models, and notes that employers often value postgraduate study. Met Office science careers include scientific, scientific-software and scientific-consultant routes. Plausible destinations include academic oceanography, marine research institutes, climate-science organizations, scientific software, environmental data science and consultancy.

These sources show relevant paths, not job demand, graduate placement rates or guaranteed employment. Academic positions are competitive. A portable code/data portfolio, peer-reviewed outputs and collaboration experience will help make the project’s methods legible to employers beyond physical oceanography.

Questions to ask before accepting

1. Which exact tracer releases, sections and turbulence measurements comprise the advertised 22-year evidence? Where are the data, permissions and processing code?
2. Is the thesis centered on the Rio de Janeiro Fracture Zone pathway or the wider Brazil Basin mixing problem? What would count as the core successful result?
3. What MITgcm version, domain, bathymetry, grid resolution, tides, forcing and boundary conditions are already available? What must the student build?
4. What HPC/GPU allocation, storage and technical support are committed, and what ensemble size is realistic?
5. Which data or simulations are held out for independent validation of the model and the emulator? How will out-of-distribution performance be assessed?
6. Is new ship or cruise work expected? If so, how much and which budget covers travel and preparation?
7. What is the formal registration period, how does it align with 3.5 funded years, what 2027/28 stipend will apply, and what are the reference instructions?

Research limitations

The application PDF is the authoritative source for the specific project. The central DLA guidance supplies scheme-level terms and eligibility. Published papers establish the field context but do not guarantee access, a particular thesis task or a supervisor's resource allocation. The funding rate quoted is the current 2026/27 figure and may change for 2027/28. This assessment is based on information available on 3 November 2026; verify the application page and funding details before submission.

Evidence and Sources

1. Official project description and application PDF
2. Exeter EPSRC Doctoral Landscape Award overview
3. EPSRC DLA funding, eligibility and application guidance
4. Exeter Geography PhD programme
5. Exeter REF 2021 subject results
6. Coastal and Ocean Futures group members
7. Coastal and Ocean Futures research portfolio
8. TICTOC transient-tracer research team
9. Exeter Marine Ocean Systems theme
10. The Brazil Basin Tracer Release Experiment: Observations (2024)
11. Long-term Brazil Basin tracer diffusivity study (2012)
12. Deep fracture-zone turbulent mixing study (2017)
13. Abyssal upwelling in mid-ocean-ridge fracture zones (2018)
14. Tracking a passive tracer through Southern Ocean water masses (2020)
15. Observed diapycnal upwelling in a sloping submarine canyon (2024)
16. Deep Brazil Basin submesoscale-permitting simulations (published 27 October 2026)
17. Mackay et al., optimal transformation method for the ocean carbon budget (2024)
18. Dynamics of eddying abyssal mixing layers over sloping rough topography (2022)
19. Official MITgcm configuration and model documentation
20. Machine learning for vertical mixing in the ocean surface boundary layer (2022)
21. Data-driven mixing parameterization in a regional ocean model (2024)
22. NORi physics-plus-ML ocean boundary-layer parameterization (2026)
23. Exeter 2026 living-cost guide
24. UK National Careers Service: Oceanographer
25. Met Office science career routes
26. Exeter EPSRC DLA award page: Department of Earth and Environmental Sciences, Ref 6012
27. Exeter EPSRC DLA award page: Department of Engineering, Ref 6014
28. Exeter Geography EPSRC DLA award page, Ref 6015 (project-specific date to confirm)