Workflow

The Climate Knowledge Graph (CKG) workflow turns an unstructured climate science corpus into structured, queryable, and publishable knowledge. Beginning with IPCC AR6 sources, the process moves through harvesting, knowledge modelling, Wikibase import, annotation, analysis, and publication outputs. The result is a FAIR, linked data environment that supports both research and publication-ready outputs.

Schematic

flowchart LR
    A[Identify AR6 sources<br/>authors, glossary, references] --> B[Harvest corpus<br/>web and PDF content]
    B --> C[Create initial data model]
    C --> D[Import structured data into Wikibase]
    D --> E[Map report navigation<br/>MediaWiki categories and links]
    E --> F[Enrich entities<br/>authors, glossary, acronyms, bibliographic data]
    F --> G[Annotate and validate<br/>community review and linking]
    G --> H[Analyse and visualise<br/>Wikibase data and graphs]
    G --> I[Publish outputs<br/>API, HTML, PDF, notebooks, LLM interfaces]
    H --> J[FAIR linked open data]
    I --> K[Shareable publications and services]

Workflow summary

Working with the IPCC Sixth Assessment Report (AR6) corpus, the project follows a sequence of stages that convert source material into structured knowledge and reusable publication outputs.

1. Identify relevant report sources and metadata, including authors, glossary terms, acronyms, references, and bibliographic material.
2. Harvest the report corpus from web and PDF sources.
3. Design an initial knowledge graph data model suited to the report structure and research use cases.
4. Import the data into Wikibase as the central knowledge graph store.
5. Link the report content to media and navigation structures in MediaWiki for browsing and discovery.
6. Harvest and import additional supporting data such as authors, glossary terms, acronyms, references, and bibliographic records.
7. Annotate the report using the imported data and related entities.
8. Support community annotation and review through established knowledge-sharing communities and linked-data workflows.
9. Map relevant data to Wikidata and related knowledge structures where appropriate.
10. Perform analysis, visualisation, and data exploration over the resulting graph.
11. Generate publication outputs from the graph using computational publishing workflows.
12. Deliver searchable, shareable, and machine-readable outputs such as REST APIs, RDF data, HTML publications, and LLM-ready knowledge services.

Main stages and outputs

Stage	Purpose	Outputs
Source identification	Gather report materials and core metadata	Authors, glossary, acronyms, references, bibliographic fields
Corpus harvesting	Collect raw report content from web and PDF inputs	Structured text and document corpus
Knowledge modelling	Design the graph schema for report and entity relationships	Initial ER/data model
Wikibase import	Load core entities and metadata into the knowledge graph	Structured graph data
Navigation and linking	Connect report content to browsable structures	MediaWiki categories, report links, entity navigation
Enrichment	Add supporting data and metadata	Glossary, author records, bibliographic details, references
Annotation and review	Add semantic meaning and community validation	Linked report content, collaborative annotations
Analysis and visualisation	Query and explore knowledge graph data	Graph analysis, summaries, figures, derived insights
Publication	Turn graph content into shareable outputs	REST API, HTML, notebooks, PDF-like layouts, machine-readable data
FAIR data services	Make knowledge reusable and interoperable	RDF export, API access, linked open data, publication services

Publication model

The publication layer uses a computational publishing service that turns Wikibase content into HTML and other formats. This approach supports multi-format publishing, including PDF-like layouts via paged media CSS, while keeping the underlying knowledge in the graph so it can be reused for search, analysis, and machine-readable outputs.