Ecological Systems and Devices Annotation << Back Machine Learning on Microcontrollers for Biological early Warning Systems in Aquatic Environments Grekov A.N., Vyshkvarkova E.V., Bayandina Yu.S. Traditional water quality monitoring, based on periodic chemical-physical analysis, suffers from data latency and an inability to detect new pollutants. Biological Early Warning Systems (BEWS), which use the behavioral responses of aquatic organisms as integral stress indicators, offer a promising alternative. However, the widespread adoption of classical BEWS is hindered by their high energy consumption and limited autonomy. This review explores the potential of Tiny Machine Learning (TinyML) technology deployed on microcontrollers to create a new generation of energy-efficient, scalable, and fully autonomous BEWS. We analyze methods for overcoming hardware constraints, including quantization, pruning, and distillation of machine learning models for classifying behavioral patterns and detecting anomalies directly on the device. Practical implementations using fi sh, bivalves, and zooplankton as examples are examined in detail, and key challenges, such as data scarcity and the need for adaptation to changing environmental conditions, are outlined. The integration of TinyML and BEWS paves the way for creating distributed networks of intelligent living sensors for the continuous monitoring of aquatic ecosystems. Keywords: environmental monitoring, anomaly detection, TinyML, BEWS, microcontrollers, behavioral Monitoring. DOI: 10.25791/esip.12.2025.1561 Pp. 13-22.