Agro-forestry environments

This case study tackles food security, sustainable agriculture, and forest resilience under climate change by developing GenAI methods for trait-based phenotyping of crops and forest species. Applications include disease detection, biomass estimation, and resilience monitoring. The pilot assembles multimodal datasets (in-situ, UAV, satellite EO, and FM embeddings) enriched with morphological and phenological traits from farmers, botanists, and citizens. Textual corpora links traits to phenotypes, guiding GenAI training. The modelling pipeline integrates three modality-specific modules via a pretrained LLM with Retrieval-Augmented Generation (RAG), enabling phenotype inference from traits combined with environmental context (location, weather, soil).

The case study also develops diagnostics to monitor performance and generalisability. Validation spans European and African agro-ecosystems (OFVi programme). Outputs evolve from AI-ready datasets (M12) and an open-source multimodal phenotype/trait model (M18) to an operational pipeline for large-scale agriculture and forestry applications. Practical engagement of farmers, foresters, and SMEs is planned via cooperatives and established networks (AnaEE, eLTER), so that domain needs can inform outputs and adoption pathways.