Using Machine Learning to Predict Suspended Sediment Transport under Climate Change

Abstract Sediment transport, an important element of the erosion‒sedimentation cycle, can be very high during extreme flood events and can cause hydromorphological changes within river networks. Therefore, improved sediment transport predictions are needed to establish sediment management at the catchment scale. A machine learning model (i.e., XGBoost) and a sediment rating curve method were tested for predicting the suspended sediment load in the Sora River catchment in Slovenia. The evaluation of the models based on the historical data for 2016–2021 revealed that XGBoost outperformed the sediment rating curve model and resulted in a lower bias (i.e., approximately 15%). The XGBoost model was used to predict future suspended sediment load dynamics. Three representative concentration pathway (RCP) scenarios (RCP2.6, RCP4.5, and RCP8.5) and several climate change models were used. The rainfall–runoff model was set up, calibrated, validated and applied to simulate future daily discharge data, as this was the required input for the XGBoost and sediment rating curve models. The simulation results indicate that suspended sediment load is expected to increase in the future in the range 15–20% under both the RCP4.5 and RCP8.5 scenarios. Additionally, the number of days with a suspended sediment concentration (SSC) greater than 25 mg/l, which is often used an indicator of inadequate water quality, is expected to increase by 2–4%, whereas some models indicate an increase of up to 8%. Erosion and sediment management mitigation measures need to be applied in the future to ensure adequate water quality and good ecological status of the river.