Objective This use case aims to support the update and maintenance of LPIS (Land Parcel Identification System) by leveraging AI-based change detection tools capable of automatically identifying and characterising modifications in agricultural landscapes and parcel-related features over time. Context and methodology A supervised deep learning model for change/non-change detection has been developed and validated on French agricultural remote sensing data. The training dataset covers approximately 500 km² of high-resolution aerial orthophotography (20 cm), annotated over multi-year temporal windows (3-year intervals). The model detects changes affecting: • agricultural parcels, • landscape éléments, • and broader agricultural land-use structures. The system has been tested on French contexts and demonstrates robust performance under national LPIS update conditions. Proposal from IGN The French mapping agency (IGN) proposes to scale this work through European collaboration by: • Sharing curated annotated datasets (aerial orthophotos + harmonised annotations of agricultural landscape features) • Publishing the trained AI model and documentation including inference pipeline, training configuration, and evaluation methodology • Supporting external partners (AgData ecosystem) to test the model on French benchmark data, and their own national datasets with the possibility of domain adaptation • Capitalising on feedback from partners to improve model robustness and feed into WP5 activities (ontology and dataset harmonisation) Expected outputs and deliverables The work will result in a fully reusable and interoperable AI package, including: • a clean, documented inference codebase • a harmonised label schema aligned with WP5 ontology work • a curated dataset published on Hugging Face • user and integration documentation for operational deployment • guidelines for model adaptation to national contexts Expected outcomes and impacts This initiative will enable AgData partners to: • run and evaluate the model on their own national datasets • adapt the model to local agricultural and agro-climatic conditions • contribute feedback for continuous improvement and interoperability At European scale, the expected impact is: • strengthened capacity for automated LPIS updates • accelerated development of FAIR geospatial AI assets • creation of a shared ecosystem of reusable models and annotated agricultural datasets

Source: Annual AgData Use Case Summit 2026 booklet.