Tropical landslide susceptibility mapping using differential interferometry SAR and Statistic model in Bawakaraeng and Lompobattang Mountain, Sulawesi

As a tropical country Indonesia naturally characterized by high rate of rainfall and heavy cloud cover as well as lies in a complex tectonic settings. This situation makes Indonesia is prone to natural disaster. Comprehending natural disaster area is essential to prevent and mitigate people from further damage that might occur before and after such event. Mapping this area is one way to comprehend the situation when disaster strikes. Remote sensing data have widely been used along with GIS to create susceptibility map. The objective of this study is to map landslides susceptibility by integrating optical satellite images of Landsat ETM with Japanese Earth Resource Satellites (JERS-1) Synthetic Aperture Radar (SAR) data complemented by statistic logistical regression model using a Geographical Information Systems (GIS) platform. The study area covers landslide incidences that occurred in Bawakaraeng and Lompobattang Mountain areas of South Sulawesi, Indonesia. Change detection analysis is used to extract thematic information and the technique of Differential SAR Interferometry (DInSAR) was employed to detect slight surface displacement before the landslide event. The DInSAR processed images would be used to add as one weighted analysis factor in creating landslide susceptibility map in the statistic model with parameters of lithology, distance from the road, distance from the river, distance from the fault, land use, curvature, aspect and slope. The result indicates that during the JERS-1 SAR data acquisition period of 1993-1998, there has been slight movement of the slope prior to event of the landslide and from the statistic model we found the best model was selected among twelve trials that were chosen from equal number of landslide and non-landslide pixels. The success rate, determined from the AUC of training data set, was found to be 0.866, which means that model has accuracy of 86.6 % accuracy to predict future landslides. The prediction rate, calculated from the AUC of the validation dataset was found 0.855, which means a prediction accuracy of 85.5%. The close similarity of the success rate and prediction rate values showed how the logistic regression model is reliable in predicting future landslide with a good level of accuracy