Selecting appropriate vegetation indices for detection and prediction mapping of landslides in South Korea

Landslides are becoming more frequent and severe due to increasing summer heavy rainfall and typhoons caused by climate change. As a result, the importance of research to predict and detect landslides is also growing. In this study, suitable vegetation indices for detecting and predicting landslides in South Korea were identified based on 1,500 landslides, provided by the National Disaster Safety Institute from 2011 to 2017. Various vegetation indices listed in the Index Data Base (IDB) of the Institute for Crop Science and Resource Conservation (INRES) in Germany were reviewed and selected for construction. Five vegetation indices were constructed using Landsat-7 imagery: Normalized Differential Vegetation Index (NDVI), Simple Ratio (SR), Soil Adjusted Vegetation Index (SAVI), Atmospherically Resistant Vegetation Index (ARVI), and Renormalized Differenced Vegetation Index (RDVI). Additionally, the Maxent model was applied to predict landslides using each vegetation index, and the accuracy was measured by comparing the ROC-AUC values. The results showed that the SR vegetation index detected landslides most effectively. Furthermore, the accuracy comparison of the Maxent model revealed that the model using vegetation indices had higher ROC-AUC values compared to the model without vegetation indices, with the SR-based Maxent model yielding the most accurate results. This study provides remote sensing data necessary for creating landslide prediction maps. However, future research should consider factors influencing vegetation indices other than landslides and reflect the extent and magnitude of landslide damage.