Making environmental measurements machine-comparable: semi-automatic decomposition of observed-property descriptions into the I-ADOPT semantic framework, tested on environmental sources the methods had never seen

Environmental datasets often describe what has been measured using short text labels, such as observed property names. These labels are usually understandable to domain experts, but they can be harder for machines to compare, search and reuse consistently across datasets. This project investigates whether natural language processing and semantic methods can support the semi-automatic decomposition of environmental observed properties using the I-ADOPT Framework.

The work will explore how observed-property descriptions can be broken down into structured components such as the measured property, object of interest, matrix, context object, constraints and statistical modifiers. The project will begin with an interpretable rule-based baseline, then examine whether vocabulary matching, ontology-based methods and lightweight NLP techniques improve the consistency and usefulness of the decompositions. Candidate methods will be evaluated against example or gold-standard decompositions, with particular attention to component-level accuracy and error analysis.

The expected outcome is a practical comparison of approaches, supported by documented notebooks, a reusable code repository and, if feasible, a small prototype tool. The project aims to support better discovery, integration and reuse of environmental data by reducing the gap between free-text observed-property labels and machine-interpretable semantic descriptions.