Generation of segment data using 30-m DEMs as the base material of future hazard maps

Geomorphological maps delineate homogeneous terrain elements which contain topographies frequently encountered by natural disasters, and can reveal the vulnerability of land. VS30 (time-averaged shear wave velocity for the top 30 m) is an important index to estimate the amplification of ground shaking due to seismic waves, and the NEHRP seismic site classes are categorized based on VS30 and indicate the category of ground strength. However, such interpreted or surveyed data does not generally cover the whole country, therefore use of proxies is necessary. In this study, we generated segment data from raster data of topographic measurements using DEMs and created two maps: a geomorphological map which was a replica of expert-driven maps and a classification map of the NEHRP site classes. For this experiment, we used high-precision 30-m DEMs of all over Japan. The creation of these maps was done as follows. First, we calculated several geometric signatures that were correlated with VS30. The selected signatures were: 1st order measurements (slope, HAND (Height Above the Nearest Drainage)) which did not emphasize the effects of artificial terrain noise and HAND that can represent low topography of plains, and the surface texture, which was designed to reduce the effects of artificial terrain noise. Second, we segmented the raster data of the geometric signatures because of training data imbalances, the small amount of the training data, and scale issues of the topography.　The segment data was used for material of the classifications. Assignment of the legend was performed by grouping segment data by machine learning. The method used in this research is considered to be widely applicable as basic information of hazard maps, even in foreign countries where DEMs of the same level of accuracy are prepared. This work was supported by JSPS KAKENHI Grant Number JP18H00769.