The New Agronomists: Language Models are Experts in Crop Management

TL;DR

This work introduces a language-model–driven reinforcement learning framework that couples Deep Q-Networks with a DistilBERT-based policy to optimize daily nitrogen fertilization $N_t$ and irrigation $W_t$ decisions in crop simulations via Gym-DSSAT/DSSAT. By translating numerical crop-model states into descriptive sentences, the LM-based RL agent gains richer state understanding and achieves superior performance across maize case studies in Florida and Zaragoza, with substantial economic profit gains and reduced environmental impact under various reward designs. Key findings include strong cross-location gains, robustness to measurement noise, and clear advantages over traditional MLP-based RL and baselines, as well as insights from ablation studies showing the benefits of joint optimization and the dangers of over-parameterized architectures. The work demonstrates the practical potential of language models as agronomist-like decision-makers and provides open-source code to spur further development toward real-world deployment and more sophisticated language-model planners in agriculture.

Abstract

Crop management plays a crucial role in determining crop yield, economic profitability, and environmental sustainability. Despite the availability of management guidelines, optimizing these practices remains a complex and multifaceted challenge. In response, previous studies have explored using reinforcement learning with crop simulators, typically employing simple neural-network-based reinforcement learning (RL) agents. Building on this foundation, this paper introduces a more advanced intelligent crop management system. This system uniquely combines RL, a language model (LM), and crop simulations facilitated by the Decision Support System for Agrotechnology Transfer (DSSAT). We utilize deep RL, specifically a deep Q-network, to train management policies that process numerous state variables from the simulator as observations. A novel aspect of our approach is the conversion of these state variables into more informative language, facilitating the language model's capacity to understand states and explore optimal management practices. The empirical results reveal that the LM exhibits superior learning capabilities. Through simulation experiments with maize crops in Florida (US) and Zaragoza (Spain), the LM not only achieves state-of-the-art performance under various evaluation metrics but also demonstrates a remarkable improvement of over 49\% in economic profit, coupled with reduced environmental impact when compared to baseline methods. Our code is available at \url{https://github.com/jingwu6/LM_AG}.

Figures (2)

• Figure 1: Framework and pipeline of the intelligent crop management system using LM-based RL
• Figure 2: Cumulative reward versus episodes for policy training under RF1