Susceptibility maps in the field of sediment-related and landslide disasters, together with related analytical methods, were mostly made prior to the current digital era. For example, design debris flow volumes for facility planning were estimated based on analog geomorphological maps through expert judgement. River sources at the risk of erosion and failure are identified as zero-order channels in Japan. Quality of such analog extraction by human experts is neither stable, due to their skill levels, nor clear to the third party. As a result, designing and planning to debris flows are accepted only with limited confidence. In this study, zero-order channels in several river basins were extracted by human experts firstly. Then, the results were used as training data for machine learning. The machine learning, in turn, produced digital extraction for a river basin of Asakura county, Fukuoka, hit hard in 2017. The recall and precision rates of pattern recognition, obtained by digital extraction, were 88 and 64 % respectively. Quality of the machine learning was regarded as achieving an allowable level, practically, in comparison to expert extraction since the proposed method failed to pick up only in small non-critical catchments. Comparison with sediment-related thematic maps in two municipalities of the basin turned out to be 50 per square kilometer for debris flow hazardous channels, whereas that for landslide prone areas was 15 on average respectively. In addition, geomorphological dissection and its hierarchical order were analyzed in the same river basins following the original papers which had introduced the concept of zero-order channel. Discrepancy between the results obtained hereby and those observed in the original papers was understood to be due to geological and geomorphological structures. For practical application, difference of geological and geomorphological structures needs to be taken into consideration when the map of extracted zero-order channels was used in other areas nationwide. Lastly, reliability of the method by machine learning needs to be upgraded constantly. At the same time, further analysis on presumed relationship with geological and geomorphological characteristics is to be carried out properly.
