Bridging agronomic science and context specific farm-level advisory through generative AI for rice systems in India

Area of Science:

• Agricultural Science
• Artificial Intelligence
• Data Science

Background:

• Agriculture faces a data paradox, generating vast amounts of information but struggling to translate it into actionable farm-level insights.
• Traditional advisory systems lack the scale and specificity for climate adaptation and food security.
• Generative AI (GenAI) offers a potential solution to bridge this gap.

Purpose of the Study:

• To present GenAI as a transformative interface for accessing advanced agricultural science through natural language.
• To review the development of crop-specific Large Language Models (LLMs) for agro-advisory systems.
• To examine the integration of LLMs with Knowledge Graphs and biophysical simulators for enhanced accessibility.

Main Methods:

• Review of GenAI applications in agriculture, focusing on LLMs and multimodal analysis.
• Analysis of hybrid integration frameworks combining LLMs with Knowledge Graphs (KGs).
• Examination of integrating LLMs with process-based simulators (e.g., DSSAT, APSIM) and vision-language models with imagery data.

Main Results:

• Crop-specific LLM agro-advisory models (e.g., SeedLLM-Rice, IPM-AgriGPT) outperform general models by using specialized corpora.
• Hybrid LLM-KG frameworks provide factual grounding, while integration with simulators enables natural language access to biophysical modeling.
• Fusion of imagery data with vision-language models allows for real-time, context-aware diagnostics for smallholder farmers.

Conclusions:

• GenAI holds potential for equitable precision agriculture, particularly for rice farmers in India.
• Addressing data sovereignty, infrastructure gaps, and implementing human-in-the-loop systems are crucial for GenAI's success.
• Ensuring scientific rigor and social inclusion are paramount for realizing the benefits of GenAI in agriculture.