An Integration of Logistic Regression and Geographic Information System for Development of a Landslide Hazard Index to Land Use: A Case Study in Pingtung County in Southern Taiwan

In Taiwan, mountainous areas account for approximately two-thirds of the total area. The steep terrain and concentrated rainfall during typhoons cause landslides, which pose a considerable threat to mountain settlements. Therefore, models for analyzing rainfall-induced landslide hazards are urgently required to ensure adequate land use in mountainous areas. In this study, focusing on Pingtung County in southern Taiwan, we developed a landslide hazard index (<i>I_RL</i>) to land use. Using FORMOSA-2 and SPOT-5 satellite images, data were collected before and after four typhoons (one in 2009 and three in 2013). The ArcGIS random tree classifier was used for interpreting satellite images to explore surface changes and disasters, which were used to analyze slope disturbances. The product of the maximum 3-h rolling rainfall intensity and effective accumulated rainfall was used as a rainfall trigger index (<i>I</i>_RT). Considering environmental and slope disturbance factors, an index of slope environmental strength potential (<i>I_SESP</i>) was developed through logistic regression (LR). Landslide hazard to land use was estimated using <i>I</i>_RT and <i>I_SESP</i>. The average coefficient of agreement (Kappa) was approximately 0.71 (medium to high accuracy); the overall accuracy of slope environmental strength potential analysis was approximately 80.4%. At a constant <i>I_SESP</i>, <i>I</i>_RT increased with the increasing hazard potential of rainfall-induced landslides. Furthermore, <i>I</i>_RT and <i>I_SESP</i> were positively correlated with landslide occurrence. When large <i>I_SESP</i> values occur (e.g., fragile environment and high land development intensity), small <i>I</i>_RT values may induce landslides.