Optimizing Crop Management with Reinforcement Learning and Imitation Learning

Abstract

To increase crop yield while minimizing environmental impact, we present an intelligent crop management system that optimizes nitrogen fertilization and irrigation simultaneously via reinforcement learning (RL), imitation learning (IL), and crop simulations using DSSAT. We first use deep RL to train management policies that require a large number of state variables from the simulator as observations (denoted as full observation). We then invoke IL to train management policies that only need a limited number of variables that are measurable in the real world (denoted as partial observation) by mimicking the actions of the RL-trained policies under full observation. Simulation experiments using maize in Florida demonstrate that our trained policies under both full and partial observations achieve better outcomes than a baseline policy. Most importantly, the IL-trained management policies are directly deployable in the real world as they use readily available information.