Predictive model for river bank erosion using dimensional analysis and ensemble method [EMBARGOED]

Abstract

Riverbank erosion is a natural process of removal of earthen materials from the bank surface. The process of riverbank erosion that is induced naturally resulted in formation of landforms such as valleys, canyons, and productive floodplain. However, riverbank erosion can also be considered a hazard when the process occurs at an alarming rate causing loss of land. The extent of erosion depends on many factors including the flow characteristic, bank and river geometry, and soil erodibility. The aim of this study is to establish a predictive model for river bank erosion that incorporate soil erodibility parameters based on the dimensional analysis and regression methods. Several models were trained using the Regression Learner Application in MATLAB software. Among the models trained, stepwise linear regression, ensemble boosted tree, and ensemble bagged tree are the three model types with the best fit. The training results for these three models were extracted and compared. Model 2, which is an ensemble boosted tree model was selected as the best model with Root Mean Square Error (RMSE) of 3.70E-08 and coefficient of determination, R2 of 0.55. Based on the training results for all models, it can be seen that models that include parameters from soil characteristics and properties category perform better than the models without the soil characteristics and properties parameters. It can be concluded that the soil characteristics and properties parameters can enhance the accuracy of riverbank erosion prediction model. The model produced in this study will be helpful to analyze and predict the rate of riverbank erosion for river in Malaysia and assists in the development of bank stabilization solution.