A Shared Semantic Vocabulary as a Path Forward for Harmonized Inventory Data

Matching inventory data (processes, products, elementary flows) is currently an enormous challenge for the LCA and MFA communities. Though AI shows some promise in this area, it is limited by the sparse information provided in most glossaries and the weirdness of industrial ecology nomenclature. We propose and demonstrate a pragmatic way to generate a common vocabulary through three key insights.

First, we can build on existing databases which are already widely used in both science and trade as our foundation for describing the world. Specifically, the European Commission publishes annually the Combined Nomenclature (CN), a set of product codes based on HS but with more detail. These codes are used in all trade in the European Union, are published in a computer readable format, and come with translations to all EU official languages. Similarly, ChEBI, the Chemicals and Entities of Biological Interest database, has a great amount of detail for tens of thousands of substances, It has been published for decades and allows for open discussion and submission of new data.

Second, we can use semantic data structures to allow for a flexible data schema. Semantic data allows for new types of information to augment these published concepts without having to define new XML schemas or parsing software. It also allows for a web of knowledge to link our terms to a broader set of resources and potential consumers. CN is already published in the Simple Knowledge Organization System (SKOS) semantic ontology, and ChEBI can be adapted. Using this approach and ontology, we can add additional terms to these existing database to get the level of precision we need in our work. The new terms would preferably be based on existing international standards such as ISO.

Finally, SKOS is a taxonomy - a hierarchical system of organizing knowledge. This hierarchy allows our inventory models to match the correct level of specificity of given demanded product, and have the best available provider of that product be determined at run-time. In other words, we don't have to normalize our inventory models against a published database's nomenclature, but can normalize against a widely used and understood taxonomy based on international standards. This approach allows us to calculate against multiple background databases, and allows for automatic re-linking when new and more specific data becomes available.