To read similar articles, subscribe to our Newsletter
The Earth Systems Predictability (ESP) Forum took place in May with support and major contributions from ESA Φ-lab. The event saw academics, institutional representatives and industry specialists come together to explore how the combination of Earth observation (EO) and artificial intelligence (AI) can inform data-driven decision-making on climate issues. The initial findings of the forum have now been published.
“The impact of our failure to care for our planet is now self-evident on a daily basis, underscoring the fundamental role of Earth observation in increasing awareness, helping to forecast trends and guiding policy on the climate crisis. AI can and does assist us with the considerable computational challenges associated with modelling Earth systems. Creating a sustained dialogue between experts in their respective fields will undoubtedly contribute to scaling and developing our predictive capabilities, and the ESP Forum was an exciting and valuable step towards that goal.” – Jonathan Bamber, Professor of Glaciology and Earth Observation at the University of Bristol and member of the ESA Advisory Committee for Earth Observation (ACEO).
The current catastrophic wildfires and extreme weather in southern Europe have sparked renewed discussions on the need for climate mitigation. Clearly climate change resilience planning and rapid disaster response will continue to be a central focus for society over the coming decades, driving the need for increasingly reliable and comprehensive Earth systems predictability. Initiatives like ESA’s Digital Twin Earth will provide a progressively more accurate picture of natural and human activity on our planet, but a gap exists between the knowledge gained from such models and the necessary action that should result from that knowledge.
Organised by Trillium Technologies in partnership with ESA Φ-lab and Oxford University, the ESP Forum was set up to build a cohesive vision on Earth systems predictability between data scientists, data users and decision makers. In a series of preliminary and main workshops, an interdisciplinary cohort came together to define the opportunities and challenges of developing ESP through the twin enablers of AI and satellite-derived data. The subject matter was divided into the three principal themes of Twinning and Simulation, Integrating Knowledge and Decision Intelligence, with participants at each session tasked with examining the practical steps required to establish and maintain a functioning and trusted ESP system.
The preliminary conclusions of the gathering have now been published. For the Twinning and Simulation theme, the forum emphasised that ESP technologies that combine EO data, fast simulations and robust machine learning models have a great potential for enhancing decision making at all levels, but these technologies must be deployed carefully to encourage adoption, scaling and impact. EO tools need to be accessible to a broad range of stakeholders, including for instance local populations and indigenous peoples, who in turn must be involved in creating shared knowledge tools. A holistic approach towards the simulation of interconnected systems will allow users to visualise consequences and outcomes more fully.
The Integrating Knowledge group discussed the role of Foundation Models and Large Language Models (LLMs), which offer a tremendous opportunity to build accessible and democratic expert systems for ESP. A key recommendation was to build LLMs that are continuously updated concept engines, with continuous learning/unlearning as a driver. The subject of stakeholders was also addressed by this group, including the proposal that semantic layers should be investigated so that jargon is not a barrier to communication, together with creating a broad advisory committee for knowledge systems in order to engage marginalised groups and vulnerable populations.
Under the Decision Intelligence theme, the participants stressed that AI should be seen as a joint decision-making paradigm, although integrating humans into hugely complex, multivariate and rapid decision scenarios will require innovation in how we interface with AI-derived recommendations. Disciplines from ESG (Environment, Social and Governance) best practice may be usefully applied to AI4EO, but the nuanced considerations that these areas require have so far been beyond the capabilities of machines. However, the technology is improving and therefore so is the ethical reach of the decisions that AI can support. The group concluded that ultimately, extending this moral remit to the decisions society needs to make to avoid the worst effects of climate change is where the greatest potential goal for AI4EO lies.
The full provisional findings are detailed on the forum webpage.
Kirsten Dunlop, CEO of Climate-KIC and keynote speaker for the Integrating Knowledge theme, emphasised the importance of the timing of the event and its findings: “We are at a critical juncture in terms of on the one hand, the rapid rate of global temperature rises and on the other, the exponential increase in data and knowledge available for decision making. With its broad spectrum of expert contributors, the ESP Forum was able to identify both the potential and the pitfalls of Earth system models for raising awareness and enabling more effective decision making in climate action.”
“Given our remit of transformational innovation and advanced computational research in Earth observation, the ESP Forum was a natural fit for us,” commented Head of ESA Φ-lab Giuseppe Borghi. “We were delighted to have contributed to the event and the discussions, and I’m convinced that the conclusions and directions to be set out in the final report will provide the building blocks for a true symbiosis of Earth system prediction and fact-based climate policy making.”
To know more: ESA Φ-lab, Trillium Technologies, Oxford University, ESP Forum
Images courtesy of Trillium Technologies