Regression models for sediment transport in tropical rivers

Area of Science:

• Hydrology
• Environmental Science
• Geomorphology

Background:

• Sediment transport in tropical rivers is crucial for river basin management but poorly predicted by existing models.
• Tropical river characteristics differ significantly from temperate regions, necessitating specialized models.
• Traditional methods like multiple non-linear regression (MNLR) show limitations in accuracy for tropical sediment transport.

Purpose of the Study:

• To evaluate the effectiveness of machine learning models for predicting sediment transport in Malaysian tropical rivers.
• To compare the performance of evolutionary polynomial regression (EPR), multi-gene genetic programming (MGGP), and M5 tree model (M5P) for this application.

Main Methods:

• Implemented three machine learning models: EPR, MGGP, and M5P.
• Utilized formulated variables derived from revised equations specific to Malaysian rivers.
• Assessed model performance using various statistical criteria.

Main Results:

• Evolutionary polynomial regression (EPR) demonstrated the best prediction accuracy among the tested machine learning models.
• MGGP and M5P also showed potential but with lower accuracy compared to EPR.
• Machine learning models excelled in predicting high data values but showed limitations with lower data values.

Conclusions:

• Machine learning, particularly EPR, offers a more accurate approach to modeling tropical river sediment transport compared to conventional methods.
• Further research is needed to enhance the accuracy of machine learning models for predicting lower sediment transport values.
• Improved sediment transport prediction is vital for effective integrated river basin management in tropical regions.