Climate-smart agricultural AI helping rice farmers manage crop disease
Why is AI decision support important for smallholder rice farmers?
The project creates a route for translating university research and agricultural expertise into practical decision support for smallholder farmers. Its intended value is to make specialist guidance more accessible in settings where connectivity, cost and fragmented knowledge constrain access to advice.
The emphasis on locally relevant evidence, expert validation and offline capability is designed to address limitations of generic AI tools. It also reflects OpenAg’s wider commitment to agricultural models developed with universities and experts, equitable access and AI that can work without continuous internet connectivity.
Beyond rice, the project aims to establish a reusable development pipeline that other universities, researchers and agricultural partners can adapt for additional crops and regions. Its significance therefore lies both in the intended farmer-facing tool and in the collaborative infrastructure for developing further agricultural AI applications.
The project aligns with NCIT’s stated focus on building digital tools and platforms that support informed, inclusive nature and climate action and turn insight into impact at speed and scale. In this context, it contributes through:
- Research translation: Connecting academic expertise with practical agricultural challenges.
- Inclusive innovation: Designing around smallholder needs, affordability and limited connectivity.
- Climate resilience: Aiming to support locally relevant decisions on water management, crop health and changing growing conditions.
- Collaborative delivery: Bringing together researchers, AI specialists and regional agricultural partners.
- Scalable impact: Developing an approach intended for reuse across crops, locations and partner networks.
How are we building agricultural AI for areas with limited internet connectivity?
The achievement described to date is the establishment of an interdisciplinary collaboration and technical approach for a rice-focused model intended to provide a reusable foundation for other crops, rather than a completed or field-validated deployment.
The collaboration combines specialist agricultural knowledge with a “small and useful” AI approach, adapting an existing compact language model rather than building one from scratch.
The development pathway comprises:
- Curating rice-specific evidence: Bringing together agronomic literature, irrigation protocols, soil management guidance and pest and disease information, alongside expert-developed scenarios.
- Structuring scientific knowledge: Developing a rice knowledge graph linking soil conditions, nutrients, pests and crop growth stages to support context-aware responses.
- Specialising the model: Applying parameter-efficient fine-tuning to adapt a compact model to rice-specific questions and management decisions.
- Designing for limited connectivity: Compressing the model for potential use on smartphones and other low-resource devices without continuous internet access.
- Validating recommendations: Combining quantitative benchmarking with agronomist review to assess accuracy, relevance, completeness, safety and alignment with farming practice.
- Preparing pathways to scale: Working with regional partners and existing farmer-facing platforms to support future testing, dissemination and adaptation.
These activities describe the proposed technical approach; the supplied material does not specify which development stages have been completed.
Who are the partners developing the OpenAg Rice Pathogen SALM?
- University of Exeter: Contributes expertise in rice agronomy, machine learning, knowledge graphs and mathematical biology. Key contributors include:
- Dr Richard Lindsay: Defines agronomic requirements, validates expert question-and-answer scenarios and reviews recommendations.
- Professor Ivana Gudelj: Provides supervision, strategic support and expertise in mathematical biology and complex systems.
- Professor Jia Hu: Leads machine learning, model fine-tuning, optimisation and internal evaluation.
- Dr Hang Dong: Leads knowledge graph design and integration of structured scientific knowledge.
- Faezeh Safari: Contributes to knowledge graph development.
- Fei Wu: Contributes machine learning expertise.
- Josh Ayers: Supports project coordination and management.
- Nature & Climate Impact Team: Connects digital technology and AI expertise with climate and nature impact, including through Dr Aseel Ali, Research Impact Fellow on Digital Tech & AI.
- Institute of Applied Artificial Intelligence and Robotics (IAAIR): OpenAg’s founding organisation and lead technical partner, providing AI research and development, GPU computing infrastructure, reusable training and deployment pipelines, and senior technical oversight.
- Hamara Foundation: Contributes regional agronomic knowledge, field manuals, locally relevant reference material and structured validation feedback, with potential delivery through its farmer-facing platform.
- World Food Bank: Provides partner networks and potential distribution pathways to support wider dissemination.
Technical specifications and project reach
The supplied material provides technical specifications and potential reach, but no verified farmer-level outcomes or completed deployment results.
- The proposed base AI model for the Rice Pathogen SALM is between 3 to 7 billion parameters in size.
- The AI model uses low-precision deployment, such as 4-bit quantisation, so it can function on lower-resource devices like smartphones without continuous internet connectivity.
- The project partners with the Hamara Foundation, an organization that connects to 150,000 small-scale farmers across sub-Saharan Africa.
- Partner networks provide potential dissemination pathways to reach tens of thousands of rice farmers across multiple global regions.
- The project plans to evaluate the model using quantitative benchmarking and structured agronomist review.