Physics-Informed Multi-Modal Localization of Stressed Plants in the Internet of Plants

Abstract

Plants communicate through diverse modalities, forming what is increasingly recognized as the Internet of Plants (IoP). Two critical signaling pathways in this network are chemical signaling and acoustic signaling, both capable of transmitting information over long distances. Despite their biological significance, the combined use of chemical and acoustic signals for stress source localization remains largely unexplored. This study presents a physics-informed framework for local- izing stressed plants by fusing volatile organic compound (VOC) and acoustic emissions under realistic environmental conditions. A biologically grounded dataset is generated using COMSOL Multiphysics simulations, which model signal propagation across varied environmental configurations. To reflect the decentralized nature of plant communication, agent plants are modeled as bio-sentinels, consistent with the distributed sensing paradigm envisioned in the IoP. To estimate the source location, a Physics-Informed Neural Network (PINN) is developed to integrate physical emission models with sparse, noisy observations from agent plants. For benchmarking, a Maximum A Posteriori (MAP) estimator and a Multi-Layer Perceptron (MLP) are also implemented. The models are evaluated across four environmental scenarios. Additionally, the robustness of the PINN to parameter uncertainty and the benefits of multi-modal signal fusion are assessed. Results demonstrate that the proposed PINN achieves strong localization performance under environmental variability.