BioDivData will provide curated, annotated, and deployment-oriented biodiversity image datasets for AI- based recognition of grassland plant species. The use case builds on DAKIS research, UAV-assisted grassland monitoring, and multi-domain knowledge transfer. It is designed to support the training and benchmarking of robust AI models for biodiversity monitoring, High Nature Value grassland assessment, and results-based agri-environmental schemes such as Eco-Scheme 5. Introduction Reliable and scalable plant recognition is becoming increasingly important for biodiversity monitoring in agricultural landscapes. In species-rich grasslands, indicator plant species can provide practical evidence for biodiversity value and support results-based agri-environmental schemes. However, AI-based plant recognition remains limited by data scarcity, class imbalance, heterogeneous imaging conditions, and weak benchmark structures for testing model robustness in real-world deployment settings. Methodologies The aim is to provide benchmark-ready biodiversity image datasets and workflows for evaluating AI-based grassland plant recognition under realistic deployment conditions. • Create/curate multi-source RGB image data for grassland plant species and provide expert- validated species annotations • Develop standardised train/test splits for in-domain and cross-domain model evaluation. • Evaluate AI model robustness under deployment-related distribution shifts • Test cross-domain learning strategies Output • Open or shareable benchmark datasets for AI-based grassland plant recognition. • Baseline AI models for object-detection-based plant recognition. • Workflows for generating species occurrence maps or detection summaries from UAV and terrestrial imagery. Impact • Support High Nature Value grassland assessment through automated indicator-species detection. • Facilitate digital verification of result indicators for agri-environmental schemes such as Eco- Scheme 5. What I can offer • Expert-validated multi-domain datasets for grassland plant detection. • Experience with UAV-assisted biodiversity monitoring and deep learning for indicator-species detection. • Baseline object-detection models and evaluation protocols for cross-domain robustness. • Use-case connection to DAKIS and biodiversity-focused decision-support systems. What I need • Grassland datasets and partners interested in testing AI model generalisation across deployment scenarios. • Future EU or national project partners working on AI-based biodiversity monitoring

Source: Annual AgData Use Case Summit 2026 booklet.