Nonlinear multi independent variables in quantifying river bank erosion using Neural Network AutoRegressive eXogenous...

Area of Science:

• Hydrology and Water Resources Engineering
• Computational Fluid Dynamics
• Machine Learning Applications

Background:

• Riverbank erosion poses significant challenges in water resource management and infrastructure stability.
• Conventional methods often struggle to accurately predict the nonlinear dynamics of riverbank erosion.
• Understanding erosion rates is crucial for mitigating land degradation and managing river corridors.

Purpose of the Study:

• To propose and validate a novel Neural Network AutoRegressive eXogenous (NNARX) model for predicting riverbank erosion rates.
• To accurately estimate complex riverbank erosion behavior under varying flow conditions.
• To provide a more accurate predictive tool compared to conventional approaches.

Main Methods:

• Development and application of a Neural Network AutoRegressive eXogenous (NNARX) model.
• Utilizing a dataset of 203 training and 135 testing data points from Sg. Bernam, Malaysia.
• Establishing nondimensional parameters using the method of repeating variables as model inputs.
• Employing One-Step-Ahead time series prediction for model accuracy assessment.

Main Results:

• Model no. 6, with 5 independent variables and 10 hidden layers, demonstrated strong predictive performance.
• Achieved high accuracy with discrepancy ratios of 94% (training) and 90% (testing).
• Model no. 6 yielded R-squared values of 0.932 (training) and 0.788 (testing), indicating good model fit.
• Identified optimal near-bank velocities (0.2-0.5 m/s) for maximum erosion (1.5-1.8 m/year) and higher velocities (0.8-1.3 m/s) for lower erosion rates (0.1-0.4 m/year).
• Sensitivity analysis highlighted the ratio of shear velocity to near-bank velocity as the most influential factor (91% accuracy).

Conclusions:

• The developed NNARX model accurately predicts the nonlinear behavior of riverbank erosion rates influenced by flow variations.
• The findings offer valuable insights for advanced simulations of channel migration and land degradation.
• The study provides a robust tool for effective riverbank protection and management strategies.