Estimating suspended sediment load with multivariate adaptive regression spline, teaching-learning based...

Area of Science:

• Hydrology
• Environmental Engineering
• Water Resource Management

Background:

• Dam functionality is significantly impacted by reservoir sediment accumulation.
• Accurate estimation of riverine sediment load is vital for dam design and lifespan prediction.
• Direct sediment measurement is costly and often impractical at all monitoring stations.

Purpose of the Study:

• To evaluate and compare various regression models for estimating suspended sediment load (SSL).
• To identify the most accurate predictive model for SSL at gauging stations on the Çoruh River, Turkey.
• To assess the efficacy of streamflow data in conjunction with SSL prediction models.

Main Methods:

• Comparison of classical regression analysis (CRA) with advanced algorithms: artificial bee colony (ABC), teaching-learning-based optimization (TLBO), and multivariate adaptive regression splines (MARS).
• Utilized streamflow and historical SSL data as inputs for model training and testing.
• Employed two distinct training and testing dataset configurations to validate model performance.

Main Results:

• The MARS model demonstrated superior accuracy, with root mean square error (RMSE) values ranging from 35% to 39% across the two gauging stations.
• Further refinement using a different dataset configuration yielded significantly lower error rates for MARS, between 7% and 15%.
• Simultaneous measurement of streamflow and SSL was identified as the most effective parameter for accurate predictive modeling.

Conclusions:

• Multivariate adaptive regression splines (MARS) is the most accurate and reliable model for predicting suspended sediment load (SSL).
• Integrating streamflow data with SSL measurements significantly enhances the predictive capability of hydrological models.
• The findings provide valuable insights for optimizing dam management and infrastructure planning through improved sediment load forecasting.