A Digital Twin-Driven IoT Smart Aquaculture System with Explainable AI for Water Quality Prediction, Fish Health Monitoring, and Sustainable Aquaculture Management

Abstract

Aquaculture supplies more than 52% of global fish consumption yet remains beleaguered by water quality deterioration, epidemic disease outbreaks, feed inefficiency, and escalating environmental impact. Conventional monitoring relies on periodic manual sampling, which provides insufficient temporal resolution to detect rapid deterioration events that unfold within minutes to hours. This paper presents the Digital Twin-Driven IoT Smart Aquaculture System (DT-ISAS), the first operationally validated platform to unify five transformative technologies within a single architecture: a heterogeneous real-time IoT sensor network, a live-synchronised digital twin, an ensemble deep learning prediction engine, multi-modal fish health monitoring, and an integrated Explainable AI (XAI) layer. Seven water quality parameters dissolved oxygen (DO), pH, temperature, turbidity, ammonia, nitrite, and salinity are monitored continuously by 284 underwater sensor nodes across six commercial farms. A cloud-hosted digital twin synchronises with physical pond state at sub-800-millisecond latency, enabling real-time simulation and what-if scenario evaluation. An ensemble of Bidirectional LSTM, LightGBM, and Graph Neural Network models achieves 97.3% prediction accuracy. Multi-modal fish health classification fusing underwater computer vision (YOLOv8 + ResNet-50), behavioural acoustic analysis, and near-infrared spectroscopy achieves 94.8% accuracy with 89.6% early disease detection. SHAP and LIME-based XAI explanations, validated in a user study with 48 practitioners, influenced management decisions in 73.8% of cases. Reinforcement learning-based management optimisation delivers a 32.1% feed waste reduction and 47.8% decline in chemical treatment events. Eighteen-month validation across farms in India, Norway, Thailand, and Brazil confirms statistically significant improvements over eight state-of-the-art baselines. The system advances UN Sustainable Development Goals 2, 13, and 14 through quantified reductions in mortality, energy consumption, and ecological discharge.