AgGym: An agricultural biotic stress simulation environment for ultra-precision management planning

TL;DR

AgGym introduces an open-source, modular cyber-agricultural system for simulating biotic-stress spread and yield losses to enable ultra-precision, RL-driven management planning. It couples a configurable infection-spread model with a yield-loss function and a supervisory RL framework, allowing localized pesticide decisions under realistic field conditions. Validation against satellite NDVI data and multi-year farm datasets demonstrates the model’s ability to reproduce observed spread patterns and yield responses, while deep RL policies (notably TRPO) reduce chemical usage and improve yield recovery compared with conventional schedule-based spraying. The modular design and public release aim to engage the research community in building field-specific models, enhancing decision support for sustainable crop protection at intra-field scales.

Abstract

Agricultural production requires careful management of inputs such as fungicides, insecticides, and herbicides to ensure a successful crop that is high-yielding, profitable, and of superior seed quality. Current state-of-the-art field crop management relies on coarse-scale crop management strategies, where entire fields are sprayed with pest and disease-controlling chemicals, leading to increased cost and sub-optimal soil and crop management. To overcome these challenges and optimize crop production, we utilize machine learning tools within a virtual field environment to generate localized management plans for farmers to manage biotic threats while maximizing profits. Specifically, we present AgGym, a modular, crop and stress agnostic simulation framework to model the spread of biotic stresses in a field and estimate yield losses with and without chemical treatments. Our validation with real data shows that AgGym can be customized with limited data to simulate yield outcomes under various biotic stress conditions. We further demonstrate that deep reinforcement learning (RL) policies can be trained using AgGym for designing ultra-precise biotic stress mitigation strategies with potential to increase yield recovery with less chemicals and lower cost. Our proposed framework enables personalized decision support that can transform biotic stress management from being schedule based and reactive to opportunistic and prescriptive. We also release the AgGym software implementation as a community resource and invite experts to contribute to this open-sourced publicly available modular environment framework. The source code can be accessed at: https://github.com/SCSLabISU/AgGym.

Figures (10)

• Figure 1: A high-level overview of interaction components with AgGym. The crop model and historical weather information inputs to AgGym provide a realistic field setup for initialization for threat dynamics simulation. After initialization, the deep RL agent trains via interacting with AgGym until a sufficient policy that maximizes the actual yield is achieved.
• Figure 2: A) Three possible infection regions with High, Medium, and Low probability, indicate three levels of neighbors. Infection in these neighboring areas can occur under three conditions: B) Probability of infection without prior spraying. C) Probability of infection with prior spraying. D) Probability of re-infection after prior spraying and recovery.
• Figure 3: Yield loss as a function of infection severity and time elapsed since the initial day of infection. Another factor influencing yield loss is the specific time point during the growing season (R1-R7) when the infection occurs.
• Figure 4: A) Comparison of infection spread during the growth season using Normalized Difference Vegetation Index (NDVI) between simulation data and real data.(Comparative Visualization of Disease Spread Dynamics during the growth season through NDVI and Distribution Functions.) B) Disease and severity indices information related to different agricultural fields in Iowa counties from 2018 to 2020. The MG I, II, III along with the curved lines show the regions in Iowa where these maturity groups are grown. C) R-squared values for yield loss predictions without pesticide applications and D) with different Grades of pesticide application.
• Figure 5: (A) Spread dynamics of infection during different growth stages. (B-i) Number of infected plots and (B-ii) pesticide applications on the simulation day, shown separately. The darker a section, the more efficient and expensive the pesticide application.