High-resolution DEMs in the study of rainfall- and earthquake-induced landslides: Use of a variable window size method in digital terrain analysis

Abstract

We undertake digital terrain analyses of rainfall- and earthquake-induced landslides in Japan, using high-resolution orthoimagery and Light Detection and Ranging (LiDAR) DEMs. Our aims are twofold: to demonstrate an effective method for dealing with high-resolution DEMs, which are often too detailed for landslide assessments, and to evaluate the topographic differences between rainfall- and earthquake-induced landslides. The study areas include the Izumozaki (1961 and 2004 heavy rainfalls), Niihama (2004 heavy rainfalls), Houfu (2009 heavy rainfalls), and Hanokidachi/Kurikoma-dam regions (the 2008M 7.2 Iwate–Miyagi Nairiku earthquake). The study areas include 7,106 landslides in these five regions. We use two topographic attributes (the slope gradient and the Laplacian) calculated from DEMs in varying window sizes. The hit rates for statistical prediction of landslide cells through discriminant analyses are calculated using the two topographic attributes as explanatory variables, and the landslide inventory data as the dependent variable. In cases of surface failure, the hit rates are found to diminish when the window size of the topographic attributes is too large or too small, indicating that an optimal scale factor is key in assessing shallow landslides. The representative window sizes are approximately 30m for shallow landslides; the optimal window size may be directly related to the average size of landslides in each region. We also find a stark contrast between rainfall- and earthquake-induced landslides. Rainfall-induced landslides are always most common at a slope gradient of 30°, but the frequency of earthquake-induced landslides increases exponentially with slope gradient. We find that the Laplacian, i.e., the attributes of surface convexity and concavity, and the slope gradient are both important factors for rainfall-induced landslides, whereas earthquake-induced landslides are influenced mainly by slope steepness.