Fostering FAIRification and AI-Readiness of Agrosystems Research Data

As data-intensive research methods, including the use of Large Language Models and extensive simulation techniques, gain prominence, the necessity for curated high-quality datasets becomes crucial in numerous research disciplines. This applies in particular to agrosystems research, where advanced modelling and simulation typically involves automated access, evaluation, integration and processing of vast amounts of data from different areas including Earth observation, plant breeding, integrated pest management, and agricultural robotics. Notably the FAIR principles have become a broadly recognized framework for improving the discoverability, access, and sharing of appropriate data and research outputs across scientific domains. At its core, the FAIR principles aim to enhance the ability of software agents or, in short, machines to discover relevant data and process it autonomously. The so-called machine-actionability emphasizes that data and its associated metadata should be easily locatable and processable by machines, even if the data is new to the system. By adhering to the FAIR principles, researchers can publish their datasets in a manner that promotes reusability for future research. Beyond just making datasets FAIR for consumption, the FAIR principles are increasingly important for publishing software and computational workflows that handle and generate these datasets. Implementing these principles throughout the entire research data lifecycle — from access and processing to generation and publication — ensures documentation of data quality metrics, thereby enhancing traceability and transparency in scientific research. Several platforms and repositories are already working towards supporting the FAIRification of workflows and the data they produce. For instance, WorkflowHub is a repository where computational workflows can be accessed and downloaded as machine-actionable Research Object Crates (RO-Crates), adhering to the Workflow RO-Crate profile. We present workflow.earth, our FAIR workflow platform designed to execute machine-actionable workflows and mobilize resulting research data from domains such as agrosystems research and agrobiodiversity. Utilizing RO-Crates as a lightweight packaging format for both data and workflows, the platform enhances reusability across different domains while preserving data provenance. Currently implemented within the Destination Earth Data Lake, a key component of the European flagship initiative Destination Earth (DestinE), workflow.earth contributes to DestinE's core objective to create a highly-accurate digital model of the Earth based on a set of thematic digital twins (DTs) such as the Climate Adaption DT, Weather-Induced Extremes DT, and the Biodiversity DT. Workflow.earth powers a digital twin that enhances the resilience of domesticated crops against environmental changes by leveraging the broader genetic diversity of crop wild relatives for crossbreeding. Our use case illustrates how agrosystems research, biodiversity and Earth observation data can be integrated in a FAIR manner to address significant challenges such as biodiversity loss in the Anthropocene.