Application of environmental variables in statistically-based landslide susceptibility mapping: A review

Abstract

Environmental variables are crucial factors affecting the development and distribution of landslides, and they also provide vitally important information for statistically-based landslide susceptibility mapping (SLSM). The acquisition and utilization of appropriate and the most influential environmental variables and their combinations are crucial for improving the quality of SLSM results. However, compared with the construction of SLSM models based on machine learning, the acquisition and utilization of high-quality environmental variables have received very little attention. In order to further clarify the research status of the application of environmental variables and possible development directions in future research, this study systematically analyzed the application of environmental variables in SLSM. To this end, a literature database was constructed by collecting 261 peer-reviewed articles (from 2002 to 2021) on SLSM from the Web of Science and CNKI platform (www.cnki.net) based on the keywords of “landslide susceptibility” and “environmental variable.” We found that existing methods for determining environmental variables do not consider the regional representativeness and geomorphological significance of the variables. We also found that at present, environmental variables are utilized generally without the realization and understanding of their spatial heterogeneity. Accordingly, this study raises two major scientific issues: 1) Effective identification of important environmental variables required in SLSM. 2) Effective representation of the spatial heterogeneity of environmental variables in SLSM modeling. From the perspective of the identification of dominant variables and their geospatial pattern of heterogeneity, targeted solutions for future research are also preliminarily discussed, including the method for identifying dominant variables from qualitative and quantitative perspectives and SLSM model construction considering the specific geospatial patterns. In addition, the applicability and limitation of the mentioned methods are discussed.