AI- and data-driven crop rotation planning

Abstract

Crop rotation planning is the process of deciding the types and the temporal succession of plants on agricultural areas to increase soil quality, crop yield, and pest/weed resistance. The data sources and modalities available for crop rotation planning are very diverse and the domain lacks solely data-driven approaches. In this paper we used literature- and NDVI-measurement-based successor crop suitability matrices and crop-specific attributes such as contribution margin and nitrogen demand as input for training an DQN-based reinforcement learning agent to generate crop rotation sequences. Practitioners and crop rotation experts validated the generated crop rotation sequences and concluded that most of the sequences are realistic, comply with existing crop rotation rule sets, and can be applied in practice.