Use of webby FDOs in NFDI - Use cases in Biodiversity and Data Science

FAIR Digital Objects (FDOs) [1] aim to make it easier for researchers to access research artifacts by standardizing the use of identifiers, types, metadata, and type-based operations, moving from a passive approach to FAIR to an actionable way. There are different ways to implement FDOs [2], all of them starting from the Permanent Identifier (PID) assigned to a research artifact. Webby FDOs [3] have emerged as an alternative for those familiar with Linked Open Data and Semantic Web [4]. Webby FDOs rely on (i) FAIR Signposting [5] to clearly point machines towards practical information, and RO-Crates [6] to expose metadata and downloadable files. FAIR Signposting uses typed web links to navigate from a landing page to other more machine-friendly resources, e.g., type and metadata, while RO-Crates provide a lightweight method to package research outputs along with their metadata. The approach has already been adopted by several existing repositories covering a variety of domains including life sciences, data science and Earth and environmental sciences [7]. Senckenberg's AI4WildLIVE platform combines live biodiversity monitoring, Earth observation data (e.g. Copernicus data) and Senckenberg's large Citizen Science Community (>850 active participants in AI4WildLIVE [8]) to monitor and quantify ecosystem dynamics in response to increasing land use and deforestation. The platform provides an interconnected processing pipeline for annotations (i.a. taxonomic determinations) of data from camera traps involving both machine learning-based annotation and subsequent refinement by Citizen Scientists (Human-in-the-loop pipeline). AI4WildLIVE leverages on the Webby FDO approach using RO-Crate as packaging format for observations (capture events), components (sensors) and relations (annotations), handles as identifiers to persistently link the digital objects and FAIR Signposting to add a machine-interpretable mode of presentation of the digital objects to the platform's landing pages [9]. We will continue with the institutional implementations of webby FDOs and promote further adoption and alignment with community frameworks such as the FDO Forum [15]. The NFDI4DataScience registry for Machine Learning (ML) models, MLentory [10], aggregates and harmonizes metadata describing ML models hosted at various platforms, including Hugging Face [11] and OpenML [12]. MLentory has three main components, an Extraction, Transformation, and Loading (ETL) pipeline to process information from the original sources, a back-end built on top of FastAPI v0.115.11 [13], and a front-end using VueJS v3.15.1 [14]. Aggregated and harmonized metadata is exposed using Webby FDOs, type and metadata are indicated via FAIR Signposting, while the metadata together with downloadable files follows the detached mode for RO-Crates. MLentory uses the detached mode to point to the downloadable file in the original source rather than duplicating it. The detached mode is ideal for registris, such as MLentory, where the original data is hosted in an external repository.