Modelling suspended sediment concentration in coastal Ireland using machine learning

Coastal environments are highly dynamic, making monitoring of suspended sediment concentration (SSC) both challenging and essential. SSC serves as an indicator of coastal processes, storm impact, water quality and ecosystem service delivery. However, direct measurement of SSC is costly, logistically difficult and spatially limited. Although remote sensing offers a promising alternative by estimating SSC from surface reflectance, it requires calibration and is often constrained by site-specific applicability. This study presents a machine learning framework for national-scale SSC estimation using Landsat-8 and Sentinel-2 imagery, calibrated with 147 in situ SSC samples. Several models were evaluated, with XGBoost yielding the best performance (R2 = 0.72, RMSE = 17 mg/L). SHapley Additive exPlanations values were used for model interpretability. Visible and infrared bands, along with geographic features, were identified as key predictors, reflecting the importance of coastal typology in shaping the SSC-reflectance relationship. The model’s value was demonstrated through a 10-year spatio-temporal analysis of SSC in Wexford Harbour. Seasonal patterns showed higher estuarine mixing in winter, while high SSC events coincided with rainfall and strong winds, indicating responsiveness to meteorological drivers. These findings highlight the potential of integrating remote sensing and machine learning for scalable, interpretable and cost-effective SSC monitoring.