Landslide susceptibility modeling and mapping at Dien Bien province, Vietnam using Bagging based MLP neural network

Abstract

Abstract In this article, the main aim is to build landslide susceptibility map at the Dien Bien province (Vietnam) using a hybrid machine learning model including BG-MLP which is a hybridization of Bagging and Multilayer Perceptron (MLP) neural networks. For this purpose, 665 past landslide events together with 665 non-landslide locations and 10 landslide influencing parameters including geology, normalized difference vegetation index (NDVI), distance to roads, distance to rivers, topographic wetness index (TWI), slope, curvature, aspect, distance to faults and elevation were collected and used for generation of datasets for model’s development and validation. To validate the predictive capability of the model, area under the ROC curve and other popular statistical indices were used. Results presented that BG-MLP (AUC = 0.81) has a good performance in modeling and mapping landslide susceptibility at the study area, especially its performance is better than single MLP model (AUC = 0.78). Thus, it can be concluded that BG-MLP is powerful tool that can be employed for assessment of susceptibility of landslides in other landslide prone regions of the world. Map of landslide susceptibility created from this study would be useful for decision making and land use planning in reducing the harmful impacts of landslides.