Research on landslide hazard assessment in data-deficient areas: a case study of Tumen City, China

With the development of economy, the urbanization process is accelerated and the infrastructure construction is increased, which leads to the widespread occurrence of landslides in mountain areas all over the world. However, due to the complex geological environment or some other reasons, the lack of landslide-related data in some mountainous areas makes it more difficult to predict landslides. At the same time, the existing models have different prediction effects in different regions, and it is difficult for a single model to objectively and accurately evaluate landslide hazard. The purpose of this research is to complete the landslide hazard assessment (LHA) in data-deficient areas by proposed a combination model with help of remote sensing (RS) and geographic information system (GIS) technology. Firstly, 146 landslides and 10 LHA conditioning factors in Tumen City were obtained by using RS, GIS and field investigation. To increase the amount of model training data, 386 landslides (including 146 landslides in Tumen City) in some areas of Yanbian Korean Autonomous Prefecture with similar landslide conditions to Tumen City were obtained. Secondly, three combination models for LHA are proposed, which make full use of the effective information provided by logistic regression (LR), artificial neural network (ANN) and support vector machine (SVM), and the evaluation effect and applicability of the three combination models are discussed. Finally, the three combination models and three single models of logistic regression (LR), artificial neural network (ANN), support vector machine (SVM) are analyzed and compared through the overall accuracy (OA), confusion matrix and landslide density. The results show that it can effectively complete the landslide hazard assessment in data-deficient areas with help of RS and GIS, and the three combination models proposed in this research are superior to the other three single models, and the evaluation effect of the LA-SVM combination model is the best.