Advancing Seaweed Cultivation: Integrating Physics Constraint in Machine Learning for Enhanced Biomass Forecasting in IMTA Systems

Monitoring seaweed growth rates and biomass is crucial for optimizing harvest strategies in aquaculture systems. While such a task can be managed manually on a small farm with a few seaweed tanks, it is not feasible on a commercial farm with 1000s of tanks. To this end, an Internet of Things (IoT) based seaweed growth monitoring system is being developed at Harbor Branch Oceanographic Institute (HBOI) at Florida Atlantic University (FAU). Using the Integrated Multi-Trophic Aquaculture (IMTA) system at HBOI as the test site, the project aims to develop a solution that allows farm managers to monitor seaweed growth remotely using automated sensors. An important component of this IoT solution is the machine learning-based prediction model. This study introduces an advanced Long Short-Term Memory (LSTM)-based approach for forecasting seaweed growth and biomass. In particular, an algae growth mathematical model driven by readily available environmental and aquaculture conditions has been integrated as a physical constraint in the LSTM model. This design addresses a principal challenge in this study—the lack of continuous ground truth measurements, as the biomass is recorded only at discrete intervals (e.g., initial, weekly partial harvests, and final harvest). The LSTM models are trained and evaluated for their predictive performance using experimental and synthetic data. Compared with the LSTM models with MSE loss function alone, the results showed that the model with a loss function under physics constraint achieved a significantly lower error in predicting seaweed growth. The rationale behind choosing LSTM over other state-of-the-art models is presented in the paper. This study highlights the potential of integrating machine learning with physical models to optimize seaweed cultivation and support sustainable aquaculture practices. The proposed methodology can seamlessly extend to the remote sensing data in other aquaculture settings.