Aims
This project develops AI surrogates to optimise peatland regeneration across multiple objectives, such as carbon sequestration, biodiversity habitat, wildfire risk reduction, flood attenuation, and community benefits, using a place-based, co-designed approach. Uniquely, the project will translate place-based knowledge into the surrogate itself through co-design with domain scientists and stakeholders so that diverse views shape model design and interfaces, aligning with “Doing AI Differently” (Hemment et al. 2025).
The work focuses on three tasks:
- Co-design a stakeholder Scorecard.
- Build and validate Flow Country emulators.
- Create a decision layer that ranks rewetting options under uncertainty.
Context
Peatlands like the Flow Country in Scotland are effective carbon sinks, storing significant soil carbon and aiding climate moderation while supporting biodiversity.
Models such as JULES-PEAT simulate peat processes and improve predictions of soil carbon but are computationally costly for extensive scenario analysis (Chadburn et al. 2022).
AI surrogate models, which emulate these complex simulations at lower cost, enable exploration needed for practical regenerative planning, following workflows such as “calibrate, emulate, sample” (Cleary et al. 2020).
Recent work shows sparse Gaussian processes can emulate JULES (Baker et al. 2022), but existing emulators overlook peatland hydrology and carbon feedbacks, which matter for restoration intervention design. To address this research gap, we will build “peat-aware” emulators to support decisions around Flow Country rewetting efforts. A second gap is that surrogates rarely embed stakeholder-defined, place-based decision criteria; this work makes a Regenerative Scorecard a core input to the emulator and decision layer.
Methodologies
Task (i): Co-design a Regenerative Scorecard (benefits, constraints, risk posture) with Stakeholder Steering Group (industry, government, NGOs, community organisations, and other contacts suggested by interdisciplinary supervisory team) to define and embed interpretive considerations (plural meanings, local knowledge, acceptable trade-offs) into model structure and interfaces.
Task (ii): Build and validate Flow Country emulators from an offline ensemble of JULES-PEAT simulations spanning key features (in collaboration with the interdisciplinary supervisory team). This involves training a sparse, multi-output Gaussian process (or Fourier Neural Operators or DeepONets) to map input parameters and forcing summaries to targets such as water-table depth, soil-moisture profiles, and carbon fluxes.
Task (iii): Create a decision layer using robust methods to rank interventions on the Scorecard under weather/parameter ensembles, exposing uncertainty bands and scenario narratives rather than single optima. Package this as a Regenerative Planning Toolbox for rapid scenario exploration.
Potential impact
By focusing on regeneration and co-design, this project delivers actionable decisions. The surrogates will underpin a Regenerative Planning Toolbox for rapidly testing rewetting options against the stakeholder-defined Scorecard, making decision uncertainty and trade-offs explicit. It will also open new research pathways in interpretable AI model-building and environmental/community justice with the UNESCO Centre for Water Law, Policy and Science, linking technology to governance and practice. The outcomes will be practical tools that accelerate science, inform policy, and develop skills at the interface of AI environment Net Zero while advancing the “Doing AI Differently” agenda.
Diversity statement
Our research community thrives on the diversity of students and staff which helps to make the University of Dundee a UK university of choice for postgraduate research. We welcome applications from all talented individuals and are committed to widening access to those who have the ability and potential to benefit from higher education.
How to apply
The application process is a 2-stage process:
- Email Dr Eric Hall for informal enquiries about the project as early as possible ahead of the deadline and establish suitability of candidacy, and any particular needs that are relevant to the project.
- Formal applications can be made via the Scholarship Application Form.
From the Supervisor-led selection process each project will generate a ‘Preferred Candidate’. Final appointment of Studentships will be made by formal interviews during the week commencing 26th January 2026, involving an interview panel of Programme Directors and current Regnr8-i scholars. Successful candidates should be available to start their Studentships in October 2026.
