Author(s): Tao Yamamoto; So Kazama

Linked Author(s): So Kazama

Keywords: No Keywords

Abstract: In order to maintain the river environment, there is a need to predict the change of riverbed morphology. Physically based computational models are widely used for prediction. Yet, the more precise model has heavier computational cost, therefore we tried to develop a model that can predict long-term riverbed deformation with low computational load by applying machine learning. In this work, as a first step, we began developing a model to predict the 1D riverbed deformation based on measured bed deformations. We built a simple artificial neural network, where the riverbed elevation change was used as objective variable, and the spectrum of river discharge and riverbed slope as explanatory variables. The prediction accuracy was 0.47 m in RMSE and this is within the range of measured riverbed deformation. However, the coefficient of determination was low and it was not sufficient to predict the trend of riverbed deformation. We have developed an early-stage model for predicting riverbed deformation from measured data, and future challenges have been identified.

DOI:

Year: 2024