Abstract
After a volcanic eruption, urgent field surveys should be conducted because the
susceptibility of debris flows increases. However, under the condition of volcanic activity,
it is difficult to survey eroded areas near the craters. Thus, some indirect surveys using
topographical information are necessary to sophisticate the disaster prevention system in
volcanic regions. This study estimated a deep learning model to predict eroded areas by
inputting the angle and Laplacian of a 100m times 100m area, discussed the quantitative
accuracy of the estimated model. The analytical results revealed that the deep learning results
smoothed by median filter could reproduce the training data with high accuracy and precision.
