Estimation of debris-flow volumes by an artificial neural network model

Abstract

Estimating debris-flow volume is an essential requirement for hazard assessment in mountainous areas. A number of studies have been conducted on the estimation of debris-flow volumes using empirical, statistical, and physically-based methods. Despite such efforts, the prediction of erosion and deposition processes in mountainous terrain remains a challenging task due to the complexity of the issue and lack of data. In this study, data on a total of 30 debris-flow events observed in Mt. Umyeon, Seoul, Korea, which consist of eight GIS-based geomorphological and hydrological datasets, were collected from technical reports, scientific journals, aerial photographs, and DEM. The collected datasets were applied for a correlation analysis to ensure independency among the variables in a dataset and to avoid the over-fitting the data with a model. Using an Artificial Neural Network (ANN) technique, an estimation model of debris-flow volume was developed and tested by randomly selecting a validation dataset. The final ANN model consisted of one hidden layer with 5 neurons and exhibited training, and testing correlation coefficient (R) values, and mean square error (MSE) value of 0.92, 0.88 and 0.25, respectively. In order to verify the applicability of the suggested model, the performance of the model was compared with an existing regression equation. The results showed that the suggested ANN model possessed greater predictive capabilities and could serve as a reliable tool for estimating the debris-flow volume. However, additional factors that affect debris-flow volume should be studied for the development of a higher performance model.