Author(s)

Uttam Paudel^1*, Takashi Oguchi², Yuichi Hayakawa²

¹Department of Natural Environmental Studies, Graduate School of Frontier Sciences, The University of Tokyo, Kashiwanoha, Kashiwa, Japan.
²Center for Spatial Information Sciences, The University of Tokyo, Kashiwanoha, Kashiwa, Japan.

Landslide susceptibility (LS) mapping is a requisite for safety against sediment related disasters, and considerable effort has been exerted in this discipline. However, the size heterogeneity and distribution of landslides still impose challenges in selecting an appropriate scale for LS studies. This requires identification of an optimal scale for landslide causative parameters. In this study, we propose a method to identify the optimum scale for each parameter and use multiple optimal parameter-scale combinations for LS mapping. A random forest model was used, together with 16 geomorphological parameters extracted from 10, 30, 60, 90, 120, 150, and 300 m digital elevation models (DEMs) and an inventory of historical landslides. Experiments in two equal-sized (625 km²) areas in Niigata and Ehime, Japan, with different geological and environmental settings and landslide density, demonstrated the efficiency of the proposed method. It outperformed all other single scale LS analysis with a prediction accuracy of 79.7% for Niigata and 78.62% for Ehime. Values of areas under receiver operating characteristics (ROC) curves (AUC) of 0.877 and 0.870 validate the application of the multi-scale model.

Multi-Resolution, Landslide Susceptibility, DEM, Random Forest

Paudel, U. , Oguchi, T. and Hayakawa, Y. (2016) Multi-Resolution Landslide Susceptibility Analysis Using a DEM and Random Forest. International Journal of Geosciences, 7, 726-743. doi: 10.4236/ijg.2016.75056.