New entropy-based method for variables selection and its application to the debris-flow hazard assessment

Abstract

We propose a new data analyzing scheme, the method of minimum entropy analysis (MEA). New MEA method provides a quantitative criterion for selecting relevant variables to model the studied physical system. This method can be easily extended to the analyses of various geophysical/geological data, where many relevant or irrelevant available measurements may obscure the understanding of the highly complicated physical system like the occurrence of debris-flows. After demonstrating and testing the MEA method, we apply this method to a dataset of debris-flow occurrences in Taiwan and successfully identify three relevant variables, i.e. the hydrological form factor, landslide area, and number of landslides, to the occurrence of observed debris-flow events by the 1996 Typhoon Herb.