Interoperable agricultural digital twins with reinforcement learning intelligence
Abstract
Digital twins and artificial intelligence are increasingly explored to support decision-making. In this work, we introduce a modular and interoperable architecture that combines digital twins with reinforcement learning for adaptive decision-making in complex environmental systems. We apply this approach to smart farming, where efficient resource use is critical to balance productivity with environmental impact. Our contributions are threefold: (a) the augmentation of agricultural models as digital twins—specifically the crop growth model WOFOST and the plant disease model A-scab—that assimilate field data to reflect current crop conditions; (b) the integration of reinforcement learning agents that generate recommendations for pesticide and fertilizer application—the first to demonstrate interoperable reinforcement learning-integrated digital twins in operational agriculture; and (c) the development of a FIWARE-based interoperability layer that integrates a diverse set of (edge) components. We demonstrate our approach in two pilot studies—apple scab management and nitrogen application in winter wheat—showcasing its potential for real-world application in diverse agricultural contexts and its transferability to other domains.
Published as:
M. Kallenberg,
H. Baja,
M. Ilić,
A. Tomčić,
M. Tošić,
I. Athanasiadis,
Interoperable agricultural digital twins with reinforcement learning intelligence,
Smart Agricultural Technology, 12:101412,
2025, Elsevier BV, doi:10.1016/j.atech.2025.101412.
You might also enjoy (View all publications)
- BloomBench: A Multi-Species Benchmark for Evaluating the Generalization of Fruit Tree Phenology Models
- Corn yield estimation under extreme climate stress with knowledge-encoded deep learning
- Causal machine learning methods for understanding land use and land cover change