This study is an efficiency comparison between four methods for the production of landslide susceptibility maps (LSMs), which include random forest (RF), artificial neural network (ANN), and logistic regression (LR) as the machine learning (ML) techniques and frequency ratio (FR) as a statistical method. The study area is located in the Southern Hiroshima Prefecture in western Japan, a locality known to suffer from rainfall-induced landslide disasters, the most recent one in July 2018. The landslide conditioning factors (LCFs) considered in this study are lithology, land use, altitude, slope angle, slope aspect, distance to drainage, distance to lineament, soil class, and mean annual precipitation. The rainfall LCF data comprise XRAIN (eXtended RAdar Information Network) radar records, which are novel in the task of LSM production. The accuracy of the produced LSMs was calculated with the area under the receiver operating characteristic curve (AUROC), and an automatic hyperparameter tuning and result comparison system based on AUROC scores was utilized. The calculated AUROC scores of the resulting LSMs were 0.952 for the RF method, 0.9247 for the ANN method, 0.9016 for the LR method, and 0.8424 for the FR. It is also noteworthy that the ML methods are substantially swifter and more practical than the FR method and allow for multiple and automatic experimentations with different hyperparameter settings, providing fine and accurate outcomes with the given data. The results evidence that ML techniques are more efficient when dealing with hazard assessment problems such as the one exemplified in this study. Although the conclusion that the RF method is the most accurate for LSM production as found by other authors in the literature, ML method efficiency may vary depending on the specific study area, and thus the use of an automatic multi-method LSM production system with hyperparameter tuning such as the one utilized in this study is advised. It was also found that XRAIN radar-acquired mean annual precipitation data are effective when used as an LCF in LSM production.

The rainfall-induced landslide disasters in July 2018 in Southwestern Japan yet again exemplified the severity of slope failure-related damage and the need for improvement of early warning systems. The Japanese Meteorological Agency (JMA) uses a method based on a threshold value of soil water index (SWI), a conceptual measurement that represents saturation of slope soil. The current SWI early warning system uses 60-min rainfall data on a 5-km² mesh and does not take into consideration other landslide conditioning factors such as slope angle and geology. This study calculates SWI values during the July 2018 disasters in Kure City (Hiroshima Prefecture) using 1-min XRAIN rainfall data in a 250-m mesh to investigate the relationship between SWI and landslide occurrence. It was found that the SWI threshold of 124 mm used in the JMA early warning system for the area was surpassed in all cells. A new SWI threshold calculation method taking slope angle and geology into consideration and produced with machine learning is proposed, comprising power lines for different geological units at a two-dimensional graph where points located above the threshold line represent landslide risk. It is judged that this method would provide a more accurate early warning system for landslide disasters.

In this study, we use GIS and other analytical platforms to analyze the landslide distribution pattern in the July 2018 heavy rain disasters in the southern part of Hiroshima Prefecture in Japan in conjunction with chronological XRAIN (eXtended Radar Information Network) radar-acquired localized rainfall data in order to better understand the relationship between rainfall characteristics and landslide probability. An analysis of event rainfall from the July 2018 disasters determines that landslide-inducing rainfall started from 8:30 AM on 5 July and continued until 7:30 AM on 7 July, accumulating to up to 368 mm in total precipitation, and that there were two intensity peaks, one around 7:30 PM on 6 July, and another one around 4:30 AM on 7 July. These two events are associated with particularly high landslide activity, which indicates that landslide activation is related to peak-intensity rainfall combined with accumulated continuous precipitation. The XRAIN data were also used together with landslide reports to calculate the intensity–duration (i.e., I-D) rainfall threshold for the area. The mean annual precipitation in the whole study area ranged between 2025 mm and 3030 mm, with an average value of about 2300 mm. The spatial distribution of rainfall throughout the sampled years indicates that rainfall is remarkably localized, with higher values concentrated on elevated areas. However, it was also observed that the maximum precipitation volumes are not so closely related to landslide occurrence, and the highest landslide activity was found in intermediate precipitation class zones instead. Correlating the localization patterns of event precipitation and mean annual precipitation using Pearson’s correlation coefficient, we found an r value of 0.55, which is considered a moderate correlation between the two datasets (i.e., event precipitation and mean annual precipitation).