Predictive models for the estimation of riverbank erosion rates

Abstract Riverbank erosion is a complex soil-water interaction process, highly dynamic and constantly changing. Consequently, the estimation of riverbank erosion rate requires an in-depth understanding between the riverbank properties and the hydraulic characteristics of the river. Given the complexity of predicting riverbank erosion rate and the limitation of existing analytical solutions, this study developed empirical models that include the analyses of basal erosion and bank failure using 358 erosion pin measurement collected from River Bernam in Selangor, Malaysia. Field measurements of the river and sediment data were performed following strict international standard protocols and instrumentation, such as gauging and wading technique, survey pole and SEBA F-1 current meter. Based on 50 years of record, the mean annual flood in River Bernam is 10 m3/s. This study includes extensive analysis between each measured variable in representing the factors influencing riverbank erosion rates, development of empirical predictive models in quantifying riverbank erosion rates using statistical approach and improves predictive performance. Model parameterization was performed using sensitivity analysis and comparison of measured riverbank erosion rates with flow-induced variables yielded the strongest correlation, whereas other variables were found to be less significant. These variables were evaluated based on Pearson’s correlation coefficient, trend of the scatter plots and degree of determination. Novel findings from the sensitivity analysis of this study are one of the substantial factors in deriving the most influential parameters which constitutes to the rate of bank erosion. Multiple linear, non-linear and logarithmic functions were employed in the development of new predictive model. This study concluded that the developed empirical equations using logarithmic-transformation is the best predictor. Logarithmic-transformation equations show the highest percentage of accuracy and degree of determination, i.e. up to 93.5% and 0.783, respectively, and the data are between the limit of discrepancy ratio 0.5