Integrating machine learning into a fully coupled current-wave-sediment model: Characterizing particle size in the settling process in estuaries of the great barrier reef, Australia

Abstract

Accurate prediction of sediment settling is critical for management of coastal ecosystems, but complex estuarine processes that influence sediment deposition and erosion present a major modelling challenge. This study introduces a proof-of-concept framework that integrates machine learning (ML) into environmental simulations to improve accuracy and efficiency by modelling dynamic sediment flocculation processes and their influence on particle size, enabling a more precise determination of settling velocity. Environmental factors influencing in-situ sediment particle size were used to train a regression model based on coeval measurements of three key parameters: salinity, shear rate and suspended sediment concentration (SSC). This regression model was developed using ML and integrated into a fully coupled current-wave-sediment model to simulate the flocculation response to these three parameters. The integrated model framework demonstrates its reliability and accuracy when evaluated against the in-situ measurements, satellite-derived SSC for the Fitzroy Estuary (Great Barrier Reef), and a parametric flocculation model that only relates settling velocity to SSC. We present an example of the ML-based approach outperforming a parametric model by capturing nonlinear particle-hydrodynamic interactions while maintaining computational efficiency, enabling high-resolution SSC simulations. This work demonstrates an advancement for hybrid modelling using rapidly evolving ML applications, offering a scalable tool for sediment transport and water quality management.