Using Complex Citations to Close a Provenance Gap and Open an AI Highway Between IPCC and CMIP

Many of the figures in the Intergovernmental Panel on Climate Change (IPCC) Sixth Assessment Report (AR6) are derived from the data of more than 100 climate model simulations. However, recording over 100 data citations for a single figure would overwhelm traditional publication methods. The complex citation, named by the Research Data Alliance (RDA), is an idea to combine many metadata references into a single citable object. We present an implementation solution for the IPCC complex citation use case which uses the CERN publication platform, Zenodo. Our Zenodo innovation uses a widely available resource to provide traceability, credit, impact metrics and a learning conduit for Artificial Intelligence (AI). It has the potential to be adopted as best practice by the data publication community. It is an example of how innovation that has come about through the effort to increase IPCC data transparency can benefit the wider scientific community. A Zenodo record is created for each figure, it establishes traceability by collating references to the figure’s input datasets, and credit via Zenodo’s related work/ DataCite’s relation features. The Zenodo record acts as a node in a knowledge graph landscape, a machine actionable network of connections which maximises the potential of AI to make use of environmental data. Credit: Zenodo’s related work feature links to existing data objects via their Persistent Identifiers (PIDs), in this case the CMIP6 data citation Digital Object Identifiers (DOIs). CMIP6 data citation DOIs are assigned at a dataset collection level which is sufficient granularity to assign credit. Traceability: The specific CMIP6 dataset granules used in the figure are identified by their CMIP6 data reference syntax (DRS). Files containing the mappings between CMIP6 data collection DOIs and the specific data granules used in the figure are placed in the Zenodo record in csv and json formats. Impact: The Zenodo record is issued with its own DOI via DataCite. This DOI connects the Zenodo record to the wider knowledge graph landscape. Modellers can demonstrate the impact of sharing their data in terms of specific science outputs. AI: Complete chains of machine actionable metadata “from source to plot” provide a conduit for people and AI to learn how environmental data can be used. Complex citations provide a mechanism to document explainable AI data usage. There is still work to be done to establish full visibility of credit referenced within a Zenodo complex citation record. We hope to engage the community by presenting our prototype solution to the complex citation challenge which was developed in consultation with the Research Data Alliance (RDA) Complex Citation Working Group and the NERC Environmental Data Service (EDS).