GeoMachine Demo: Exploring Ocean Data with LLM Agents and EDITO Model Workflows
Date Published
We recently released a short demonstration video that shows where our work with GeoMachine is at now. In this update, we bring together two strands that have been developing in parallel: LLM-powered agents to assist with data exploration, and integration with EDITO model workflows that support process execution and visualisation.
The video begins with a dataset discovery scenario inside a thematic “machine” similar to the way we have organised sustainable fisheries contexts in past projects such as EcoScope. Navigating marine and spatial data can involve many layers, collections, and catalogues. In practice, the first challenge a researcher often faces is knowing where to start. In the demo, we use an agent to make this first step more conversational. Instead of browsing through metadata pages, we ask the agent to locate suitable datasets and suggest reasonable initial comparisons.
Once the agent has identified relevant catch data, we explore a question on Mediterranean warming impacts using anchovy catch as an example. Here the agent assists with sampling observations across three decades and summarises spatial patterns that emerge. Complementary environmental layers are suggested as potential next steps to test emerging ideas. The purpose of the agent in this context is not to replace formal statistical analysis but to reduce friction in the early stages of enquiry, helping users move from orientation to structured exploration more quickly.
In the second half of the demo, we switch focus to model execution. Many ocean science workflows involve processes such as particle tracking, transport simulation, or connectivity analysis. We show how an agent can be used to run a process drawn from the ILIAD context — an OpenDrift-style simulation — via the EDITO architecture. The agent initiates the workflow, executes the model run, and brings the outputs back into the GeoMachine interface for visualisation, for example showing transport pathways between aquaculture sites.
This combination of dataset interrogation and process orchestration reflects where we see practical workflows currently headed. Researchers want interfaces that help them reduce routine barriers in data discovery, contextualise their questions early, and connect exploration with analysis and simulation when appropriate. The developments shown in this demo aim to bring these stages into closer alignment without obscuring the underlying data, models, or assumptions.
GeoMachine has been used across a range of marine data contexts, including Iliad and EcoScope, and continues to evolve through its role in EcoTwin. The work described here is part of that ongoing trajectory, focused on improving how researchers engage with complex datasets and computational processes in practice.