Practice

The hands-on activities span Sessions 2 and 3. Both work from the same workshop bundle, so that the second activity can be compared directly against the first. They can be followed live during the training or used afterwards as a reference for working with the GEO Knowledge Hub.

Download the bundle

The example files used by both activities, hosted on GitHub.

Before you start - what a good package looks like

Before creating your own content, explore a well-structured, reproducible Knowledge Package to see what “good” looks like, a clear description, complete metadata, and all the auxiliary materials needed to reproduce the application. The Digital Earth Africa packages are a useful reference, and a regionally relevant one.

Browse the GEO Knowledge Hub

Find and open published Knowledge Packages, such as the Digital Earth Africa applications, and study how they are structured.

Activity 1 - Publish a Knowledge Package

Session 2. Turn the bundle into a published Knowledge Package by hand, using the GEO Knowledge Hub web interface.

Map the files to resources

The first decision is not technical — it is deciding what each file is. A Knowledge Package groups several Knowledge Resources, each with its own metadata and its own persistent identifier:

Resource	Files	Type
Station observations and trend analysis	station-1.csv, trend-results-location.csv	Dataset
Model evaluation metrics	models-biases-temperature.csv	Dataset
Köppen zone processing script	koppen-modal-ssp585-2071-2100.js	Software
Data management plan	kalamkas-dmp.pdf	Publication
Results presentation	presentation-2025.pdf	User story

The Knowledge Package itself sits above these and carries the description of the study as a whole: what question it answers, who it is for, and how the pieces fit together.

Create and publish

1. Review the concepts - Knowledge Packages, Knowledge Resources and the user roles that govern who can publish what.

   Knowledge Packages & Resources

   The core sharing-unit concepts.

2. Create the Knowledge Package - follow the Knowledge Provider guide to create the package and add each resource from the table above.

   Creating a Knowledge Package

   Step-by-step creation flow in the web interface.

3. Describe each resource - apply the DataCite metadata schema. Record creators, licence, and the geographic and temporal extent of the work. State plainly which files are synthetic; a reuser who discovers that later will not trust anything else in the record.

   Metadata schema

   Reference for the metadata fields used by records.

4. Assign identifiers and tags - mint a DOI for the package and its resources, set the version, and tag the record, including country association, which is what makes the package discoverable through the national hub.

5. Publish and share - publish the package so it can be discovered, reused and replicated by the community, then import it into the National GEO Knowledge Hub and the South Africa Marketplace.

Tip

Prefer to automate? Every operation available in the web interface is also available through the REST API. See the REST API examples for end-to-end publishing recipes.

Activity 2 - Produce a Knowledge Package with AI assistance

Session 3 Same bundle, different route. Instead of writing every field by hand, use an AI assistant to draft the package, then curate what it produces.

1. Give the assistant the material - provide the bundle files and ask it to propose a package title, description, keywords, resource-level descriptions and DataCite metadata.

2. Curate the result - This step is not optional. Check every generated claim against the files themselves:

   • Does the described study area match what the script actually processes?
   • Are the creators, affiliations and licence correct, or invented?
   • Does the description present synthetic data as if it were observed?
   • Are the dates, model names and scenario labels accurate?

   An assistant is good at producing fluent metadata and unreliable at knowing what is true. Curation is where the knowledge provider adds the value.

3. Compare with Activity 1 - put the drafted package next to the one you wrote by hand. What did the assistant get right, what did it invent, and which fields still need a human?

Caution

Never publish AI-drafted metadata unreviewed. A Knowledge Package carries a DOI and is meant to be cited. An inaccurate description propagates into every reuse of that record.