ニューラルネットワークによる土石流の発生限界降雨の評価

抄録

In the previous study, the neural networks with back-propagation method (BP) were applied to predict the occurrence of debris flow. It was also found that this model is useful to estimate the critical rainfall. In this study, LVQ (Learning Vector Quantization) is introduced to improve the accuracy of the prediction. The LVQ and BP are applied to the debris flow at Unzen and Sakurajima Volcanoes. Comparison between the results by both methods confirms that LVQ has an advantage in prediction. The BP model is used to find the critical condition at Unzen Volcano. Validity of the method is demonstrated by the theory of the occurrence criteria of debris flow.