On the other hand, the quantitative evaluation method establishes a statistical probability model based on mathematical knowledge to explore the relationship between landslides and influencing factors ( Yilmaz, 2009 Tsangaratos et al., 2016). Therefore, the accuracy of the qualitative method primarily depends on the experience and knowledge of the experts and the quality and completeness of the collected landslide data. The qualitative evaluation methods include landslide cataloging and is knowledge-driven based on the empirical evidence of experts to identify and analyze the geomorphology, topography, and geological conditions of a specific area, thus, being able to subjectively analyze the landslide susceptibility within the area ( Kayastha et al., 2013 Myronidis et al., 2016). The LSM methods can be divided into qualitative and quantitative evaluation methods ( Aditian et al., 2018). In the past 30 years, with the rapid development of geographic information systems and computing technologies, various modeling methods of landslide spatial analysis and LSM have been proposed and widely used ( Carrara et al., 2010 Reichenbach et al., 2018). Previous studies have shown that landslide susceptibility mapping (LSM) can reduce the risk of landslides and provide an essential basis and scientific support for decision-makers to deal with landslide disaster risk management and land use policies ( Fell et al., 2008 Nourani et al., 2014). Therefore, an effective solution that can reduce and mitigate the damage caused by landslide disasters needs to be urgently developed. Landslides have brought significant loss to both the economy as well as people’s lives worldwide ( Huang and Fan, 2013 Froude and Petley, 2018).

They are one of the most destructive geohazards and cause catastrophic consequences. Landslides are widely distributed around the world, especially in the areas with more active geological activities. Therefore, the optimized model can generate a high-quality landslide susceptibility map.Ī landslide is defined as the movement of a mass of rock, earth, or debris down a slope ( Cruden, 1991). Thus, hyperparameter optimization is of great significance in the improvement of the prediction accuracy of the model. However, the random forest model has a higher predictive ability than the extreme gradient boosting decision tree model. The results show that the AUC validation data of the Bayesian optimized random forest and extreme gradient boosting decision tree model are 0.88 and 0.86, respectively, which showed an improvement of 4 and 3%, indicating that the prediction performance of the two models has been improved. Both models were evaluated and compared using the receiver operating characteristic curve and confusion matrix. The hyperparameters of the random forest and extreme gradient boosting decision tree models were optimized using a Bayesian algorithm, and then the optimal hyperparameters are selected for landslide susceptibility mapping. The landslides were randomly divided into training data (70%) and validation data (30%). In addition, 14 landslide influencing factors are selected, and 734 landslides are obtained according to field investigation and reports from literals. Taking Wuqi County in the hinterland of the Loess Plateau as the research area, using Bayesian hyperparameters to optimize random forest and extreme gradient boosting decision trees model for landslide susceptibility mapping, and the two optimized models are compared. Landslides are widely distributed worldwide and often result in tremendous casualties and economic losses, especially in the Loess Plateau of China.

College of Geological Engineering and Geomatics/Key Laboratory of Western China Mineral Resources and Geological Engineering, Chang’an University, Xi’an, China.

Shibao Wang Jianqi Zhuang* Jia Zheng Hongyu Fan Jiaxu Kong Jiewei Zhan

Xgboost random forest