Modelling suspended sediment concentration in coastal Ireland using machine learning.

Area of Science:

• Environmental Science
• Remote Sensing
• Coastal Geomorphology

Background:

• Monitoring suspended sediment concentration (SSC) in dynamic coastal environments is crucial but challenging due to limitations of traditional methods.
• Remote sensing offers a potential solution for SSC estimation, but requires calibration and can be site-specific.
• Existing methods for measuring SSC are costly, logistically complex, and spatially restricted.

Purpose of the Study:

• To develop and validate a national-scale machine learning framework for estimating SSC using satellite imagery.
• To assess the performance and interpretability of different machine learning models for SSC prediction.
• To demonstrate the application of the developed framework for spatio-temporal SSC analysis in a case study area.

Main Methods:

• Utilized Landsat-8 and Sentinel-2 satellite imagery for national-scale SSC estimation.
• Calibrated the models using 147 in situ SSC samples.
• Employed the XGBoost machine learning algorithm and SHapley Additive exPlanations for model interpretability.

Main Results:

• The XGBoost model achieved the best performance with R² = 0.72 and RMSE = 17 mg/L.
• Visible and infrared bands, along with geographic features, were identified as key predictors for SSC estimation.
• A 10-year spatio-temporal analysis in Wexford Harbour revealed seasonal patterns and correlations between high SSC events and meteorological drivers (rainfall, wind).

Conclusions:

• The integration of remote sensing and machine learning provides a scalable, interpretable, and cost-effective approach for monitoring coastal SSC.
• The developed framework accurately estimates SSC and captures its relationship with coastal typology and meteorological factors.
• This approach enhances our ability to understand coastal processes, water quality, and ecosystem health through improved SSC monitoring.