Short summary of your UC: DivAg-AIM develops trustworthy multimodal AI for field assessment and management support in diversified agricultural systems and communicates findings in natural language to stakeholders. Introduction: DivAg-AIM builds on the patchCROP landscape laboratory in Brandenburg, Germany, a small-field diversified cropping system designed to assess whether agricultural diversification can reduce agricultural inputs while maintaining crop productivity and resilience. This requires synthesizing multimodal information including field observations, predictive model outputs, and agronomic knowledge into coherent context-aware assessments that account for causal relationships. While recent advances in AI chatbots suggest the potential for automated synthesis, they struggle to synthesize multimodal information into actionable insights and may produce weakly grounded or hallucinated responses, thereby constraining trust in AI. Predicting field productivity from drone imagery is further constrained by small data, incomplete labels, and high spatial heterogeneity, often requiring specialized models. Methodologies: To address these challenges, we extend an AI chatbot with the recent framework Model Context Protocol (MCP), thereby providing structured and traceable access to tabular soil observations, deep learning-based yield prediction from UAV imagery, and agronomic literature. The chatbot uses these sources for grounded agronomic natural-language reasoning. For crop yield prediction, self-supervised learning supports joint representation learning across crop types and management practices under small-data conditions. Output: Outcomes include an MCP-based agricultural AI chatbot integrating soil observations, self- supervised crop yield prediction, and agronomic literature for grounded agronomic assessment with traceable justifications linked to literature and field evidence. Impact: Expected impacts include stronger evidence for sustainable crop management, transferable AI workflows for multimodal agricultural information synthesis, and improved transparency and trust in AI-based decision support. The UC is embedded in a German-Japanese collaboration to examine the value, limitations, and cross-regional transferability of the approach. What I can offer: Trustworthy multimodal AI for agriculture, including MCP-supported information synthesis, UAV-based crop yield prediction, and representation learning under small, heterogeneous, and weakly labeled data conditions. What I need: Scaling across crop types, landscapes, and datasets, including benchmark development, cross-regional evaluation, and biologically informed representation learning, and foundation-model approaches. Further exchange on stakeholder-centered evaluation, DSS/FMIS integration, agentic AI for trustworthy agricultural decision support.

Source: Annual AgData Use Case Summit 2026 booklet.