Overview
ISO 19178-1:2025 - Geographic information - Training data markup language for artificial intelligence - Part 1: Conceptual model defines a conceptual UML model (TrainingDML-AI) for describing training data used in Earth Observation (EO) AI/ML. The standard targets maximized interoperability and usability of EO imagery training data, covering dataset metadata, labels, tasks, provenance, quality, identifiers, versioning and licensing. It provides a structured vocabulary and class model to make EO training datasets discoverable, reproducible and machine-actionable.
Key topics and technical requirements
- UML conceptual model: A formal information model (TrainingDML-AI) expressed with UML to standardize how training datasets and annotations are represented.
- AI/ML task taxonomy: Specification of supervised learning task types relevant to EO - scene-level, object-level, and pixel-level tasks - and their associated labels.
- Dataset and data item metadata: Requirements to record permanent identifiers, version, licence, dataset or training data size, and the measurement/imagery sources used for annotation.
- Labeling and annotation: Structured description of AI_Label, AI_Labeling, and related classes to capture label semantics, geometry, and annotation provenance.
- Quality and provenance: Mechanisms to describe data quality (errors, representativeness, quality measures) and provenance (agents, labelling procedures, change sets).
- Extensibility and conformance: Guidelines for extending TrainingDML-AI and conformance rules to ensure consistent implementations.
- Data dictionary and ISO dependencies: Mappings to established ISO geographic metadata classes (e.g., ISO 19115-1, ISO 19157-1, ISO 19101-1) to promote integration with existing geospatial metadata ecosystems.
Practical applications and users
- Who uses this standard:
- EO data providers and catalog managers
- Remote sensing and GIS teams preparing annotated datasets
- ML engineers and data scientists training EO models
- Dataset curators, benchmark organizers, and platform developers
- Standards bodies and data stewards integrating EO metadata
- Practical benefits:
- Create interoperable EO training datasets that are reusable across projects and platforms
- Improve dataset traceability, reproducibility, and legal clarity via identifiers, licensing and versioning
- Support robust model development and evaluation through standardized quality and provenance metadata
- Facilitate dataset discovery, federation and cross-organization data sharing
Related standards
- ISO 19115-1 (geographic metadata)
- ISO 19157-1 (data quality)
- ISO 19101-1 (feature concept)
These referenced ISO standards provide metadata and quality classes used by TrainingDML-AI for consistent integration with geospatial information systems.
Keywords: ISO 19178-1, TrainingDML-AI, training data markup language, EO imagery, Earth Observation, AI/ML training data, dataset metadata, data provenance, data quality, supervised learning.