Comparative Analysis of Data-Driven Models for PredictingTotal Sediment Discharge: A Case Study of Kor and SivandRivers in Iran

Estimating sediment discharge in rivers is crucial for effective river engineering and water resources management, particularly in scenarios where sediment congestion poses environmental and flooding risks. Traditional field methods for measurement are time-consuming and expensive, leading to the development of empirical equations with varying accuracy. Recent advancements in machine learning (ML) have introduced more accurate and cost-effective models for sediment discharge estimation. This study evaluates the performance of ML models, including Linear Regression (LR), Artificial Neural Networks (ANN), K-Nearest Neighbors (KNN), Random Forest Regressors (RFR), and eXtreme Gradient Boosting Regressor (XGBR), using a dataset of flow discharges from the Kor and Sivand Rivers in Fars, Iran. The findings demonstrate noteworthy performance across all ML models, with KNN emerging as the top-performing model, followed by XGBR and ANN. While RFR and LR ranked lower, they exhibited satisfactory performance with R۲ values exceeding ۰.۸. Reliability analyses indicate good reliability across all models in both training and test datasets, with KNN showing ۱۰۰% reliability for the training data and ۷۱% reliability for the test data. Future research could explore alternative ML models or ensemble techniques and incorporate additional variables for improved accuracy and efficiency. This study underscores the potential of ML models in accurately estimating total sediment discharge in rivers and emphasizes the importance of careful model selection and evaluation for reliable performance assessment in river engineering applications.

کلیدواژه ها:

Kor River ، Machine learning ، Sediment discharge ، Sivand River ، XGBoost.