Integrating principal component analysis with statistically-based models for analysis of causal factors and landslide susceptibility mapping: A comparative study from the loess plateau area in Shanxi (China)

• Published in: Journal of cleaner production Vol. 277; p. 124159
• Main Authors: Tang, Yaming, Feng, Fan, Guo, Zizheng, Feng, Wei, Li, Zhengguo, Wang, Jiayun, Sun, Qiaoyin, Ma, Hongna, Li, Yane
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 20.12.2020
• Subjects: Causal factor; GIS; Landslide susceptibility; Loess plateau area; Principal component analysis; Statistically-based model; GIS; Statistically-based model; Landslide susceptibility; Loess plateau area; Causal factor; Principal component analysis
• ISSN: 0959-6526; 1879-1786
• DOI: 10.1016/j.jclepro.2020.124159

Landslide susceptibility assessment is an important task in urban planning and risk management. For mountainous areas where multiple types of landslides occur, the differences between landslide types should be considered when mapping landslide susceptibility. The main objective of the present study is to compare the prediction performances of three statistically-based models, namely, information value (IV), logistic regression (LR), and weight of evidence (WOE) models, in susceptibility assessments of different landslide types and to optimize the combination of causal factors. The Loess Plateau area in Shanxi Province, China, was selected as the study area. We prepared a landslide inventory by mapping the detailed spatial distribution of 234 loess landslides and 232 rockfalls in the area. Subsequently, 11 causal factors were considered in the construction of an initial index system for the susceptibility assessments, and corresponding thematic layers were generated in a GIS environment. Principal component analysis was utilized to obtain the relative importance of these layers in causing loess landslides and rockfalls, and nine factors with relatively high weights were selected as the final landslide influencing parameters. The relationships between these parameters and landslide distribution were analyzed using the three aforementioned methods and training samples (75% of landslides). The outcomes were extended to the entire study area to determine landslide susceptibility, and the testing samples (25% of landslides) were used to verify the predictive ability of the models. The results illustrated that rainfall and land use were essential in predicting both loess landslide and rockfall occurrences. The WOE and LR models showed the best performance in rockfall and loess landslide susceptibility assessments, respectively. Both models had an accuracy of more than 80%. If both landslide types were considered as an entire group or if all 11 factors were included in the model, the accuracy decreased.