Edge-Intelligence for Sustainable Pisciculture: Machine Learning Framework for Real-Time Environmental Monitoring and Weather-Driven Grow Forecasting

Modern aquaculture requires continuous physicochemical monitoring and automated health surveillance to prevent substantial economic losses caused by rapid environmental fluctuations and insidious disease outbreaks. Although cloud-based platforms can centralize monitoring and analytics, they often introduce unacceptable latency, consume significant energy, and depend on reliable network connectivity—constraints that limit their usefulness in remote or resource-constrained farming sites. To overcome these limitations, this paper proposes an edge-intelligence framework for sustainable aquaculture management that tightly integrates TinyML-driven environmental sensing, lightweight vision-based object detection, and AI-enhanced weather forecasting. The proposed architecture deploys low-power Internet of Things (IoT) nodes at pond and cage locations to perform localized acquisition of critical water-quality metrics. These nodes run TinyML models to perform preliminary data validation, anomaly screening, and compression at the edge, thereby reducing both upstream bandwidth use and energy consumption. Complementing this telemetry, an edge-optimized YOLOv11 vision model processes underwater and surface video streams in real time to perform object detection, behavioral tracking, and identification of visible disease indicators without disturbing the stock. The vision pipeline includes image enhancement and feature extraction steps to improve robustness under challenging aquatic lighting and turbidity conditions. Meteorological inputs are obtained from a weather API and processed by machine learning forecasting models to predict short- to medium-term changes in environmental drivers (for example, rainfall, wind, and ambient temperature) that directly influence pond conditions. By fusing these heterogeneous data streams—sensor telemetry, visual indicators, and AI-generated weather forecasts—the framework’s decision engine produces contextualized risk assessments and actionable recommendations. The engine prioritizes low-latency responses: upon detecting hazardous conditions or elevated disease risk, it issues alerts and intervention guidance to farmers through mobile applications or SMS, and can trigger local mitigation actions such as aeration control or feed adjustments .Extensive experimental evaluations demonstrate that the edge-centric design markedly reduces bandwidth consumption and processing latency compared with cloud-first alternatives while maintaining high diagnostic accuracy for both water-quality anomalies and behavioral abnormalities. The system’s modular, low-power components and robust multi-modal fusion make it suitable for scalable deployments across diverse aquaculture settings, offering a practical and deployment-ready foundation for precision, resilient, and sustainable farm management