| dc.description.abstract | In this study, factors used in the susceptibility analysis includes the watershed area, the length of main stream, the shape factor, relative height of watershed, slope of main stream, hypsometric integral, area of watershed slope great than 20°, area of landslide, and total rainfall of a storm event. These factors were calculated for each unit of a sub-watershed which is defined by an overflow point. Then these factors were used as input into discriminant analysis, logistic regression and fuzzy neural network to evaluate sub-watershed’s debris flow susceptibility in Shihmen Reservoir Watershed. This study focus on two different kind of cases: historical debris flow susceptibility analysis and the AERE typhoon induced debris flow susceptibility analysis.
In the historical debris flow susceptibility analysis, the potential debris flow streams were acquired from the Soil and Water Conservation Bureau, Taiwan (WCB). The overall accuracy of discriminant analysis, logistic regression, and fuzzy neural Network are 76.8%, 78.6%, and 94.6%, respectively. Fuzzy neural network has the highest overall accuracy in historical debris flow susceptibility analysis.
The debris flows induced by AERE typhoon are base on the changes detected from satellite images before and that after the typhoon event. The overall accuracy of discriminant analysis, logistic regression and fuzzy neural network are 94.1%, 100.0%, and 98.5%, respectively. Logistic regression method has the best performance in AERE typhoon induced debris flow susceptibility analysis.
From the comparison of historical debris flow susceptibility analysis and AERE typhoon induced debris flow susceptibility analysis, we can find that AERE typhoon induced debris flow susceptibility analysis has better accuracy. This is because the AERE typhoon induced debris flow susceptibility analysis has more definite debris flow data and also because the rainfall data as an import factor are considered. On the other hand and in the historical debris flow susceptibility analysis, most of debris flow streams are not well defined and mixed with the consideration of property lose, therefore , a less accuracy was presented in the result of analysis. | en_US |