Project institution: Lancaster University
Project supervisor(s): Prof Adrian Friday (Lancaster University), Dr Kirsty Pringle (University of Edinburgh), Dr Carolynne Lord (UK Centre for Ecology & Hydrology), Dr Kelly Widdicks (UK Centre for Ecology & Hydrology), Prof Mike Berners-Lee (Small World Consulting)
Research area: Sustainability Solutions in Engineering, Environmental, and Social Sciences
Industry partner: Small World Consulting
Core research question: Exascale computing is the latest capability in environmental science’s growing array of ‘digital research infrastructures’ (DRIs), offering promise of exploring new frontiers of knowledge via larger scale data analyses, larger models and AI methods that exploit a new scale of hardware acceleration to achieve results at a scale not previously possible. As exascale implies, vast compute comes with similarly vast implications for the environment due to its direct and indirect associated operational emissions, possible water use, and a growing reliance on embodied materials as exascale computing is created, embedded into research methods, and at its end of life. This raises a core dilemma: how should we consider the environmental and social costs in the creation, embedding and disposal of exascale compute facilities, and balance these for gains in environmental science knowledge?
Exascale exists in relation with new hardware capabilities including GPU acceleration and highly parallel compute facilities. Yet, efficiency gain can come with an overall increase in cost as we have seen for AI data centres, where efficiency improvements are more than offset by increased overall use and other systemic ‘rebound effects’. This project focuses on examining this core tension: how does large scale environmental modelling and AI for Earth Systems relate to underlying efficiency of GPU and parallel hardware vs. achieving results of value. How might we stimulate sufficient and responsible adoption of the lowest cost and most appropriate methods for a scientific enquiry to minimise overall growth and work within limits.
Given the growing footprint and significance of digital systems on planetary systems, it’s imperative that future environmental scientists are equipped with new
understanding and principles to support the responsible adoption and use of computational methods, especially in the era of large-scale deep learning, LLMs and agentic AI. This PhD will help strengthen the Exascale cohort with the knowledge and skills to view new digital methods exactly with this critical lens making a positive contribution to the overarching project in terms of responsible practice for the future of computing in environmental science, exascale, and beyond.
Methods Used:
The PhD project will be developing technical tools and practical solutions that support the environmental science community in using exascale infrastructures more sustainably. We will examine the use of computational methods for environmental science at exascale, and bridge between understanding of the efficiency and footprint of methods including GPU accelerators and parallel computer architectures. We will be working with model developers and practitioners to understand how they think about and address energy and resource utilisation, and its relationship to wider sustainability and to findings of scientific value. At a technical level, we will explore both efficiency and utilisation of exascale hardware features by scientific software, especially the use of GPU vector operations, AI acceleration and ‘carbon efficient methods’ for achieving scientific results at the lowest environmental cost. Drawing on qualitative methods such as interviews, and workshops, we also wish to explore to what extent we can improve carbon literacy in green software engineering practice for scientific outcomes; exploring the trade-off between environmental impacts of computation and scientific results (for instance, toward ‘sufficiency’ (Bates et al. 2026)), and how to support scientists to embrace sufficient methods.
Two teaser, transdisciplinary research projects in this domain have been outlined below, each drawing together methods from computer science, environmental data science and from social science.
Teaser Project 1: ‘Scientific software and sustainable exascale Earth System modelling’
Objectives:
- Develop expertise in GPU hardware acceleration of earth systems models, AI model emulation and energy and resource profiling;
- Examine the energy and carbon performance of earth and environmental science software, relating to GPU acceleration and underutilisation of concurrency;
- Work with green software engineering and scientific software communities to improve software, toolchains, feedback and practices to enable lower carbon Earth System models and data
science/ AI methods; - Develop technical mechanisms to reduce computational waste and lower the footprint of large scale HPC (e.g., ‘green’ job queuing systems) for different exascale computing facilities, testing
and evaluating these with environmental scientists; - Produce practical guidance, training and policies to accelerate better and more sustainable use of exascale computing.
Teaser Project 2: ‘Lean/ green data science for the future of exascale Earth System modelling’
Objectives:
- Develop expertise in GPU hardware acceleration of earth systems models, AI model emulation and energy and resource profiling;
- Evaluate the purpose, outcomes and accuracy of common Earth System models, exploring the role of AI emulation in its potential to reduce model complexity and resource footprint;
- Build a benchmark framework for evaluating the energy use, time of use marginal emissions (scope 2) and embodied costs (scope 3) of Earth Systems modelling and AI in exascale HPC;
- Explore sufficient outcomes (e.g. ‘Green AI’, (Schwartz, 2020)) to drive more responsible and leaner use of exascale computing with users and developers;
- Develop additions to scientific toolkits to integrate accounting for resource use (e.g., contributing sustainability enhancements to environmental software toolkits), offering environmental
scientists more efficient/robust and less impactful models and methods; - Produce practical guidance, training and policies for exascale-ready model and method development that embodies sustainability and reduces waste and lowers potential for rebound
effects.
