Large-area slope stability analysis: Performance comparison of three-dimensional limit equilibrium methods

• This study is the first to compare 3D LEMs for large-area slope stability analysis. • The large-area analysis used actual terrain data considering heavy rainfall. • The 3D LEMs can predict landsides similarly despite different computational costs. • The similar predictions were achieved by adjusting the factor-of-safety threshold. 3D limit equilibrium methods (LEMs) for assessing slope stability over large areas have been applied in some studies (e.g., Tran et al., 2018 ; He et al., 2018). However, their use remains limited compared to computationally cheaper alternatives (e.g., statistical or simple physics-based models) due to higher computational demands and the fact that their practical accuracy has not been sufficiently examined. In addition, while the Hovland method is recognized as the most computationally efficient 3D LEM, its accuracy has been questioned (e.g., Azzouz and Baligh (1978); Hutchinson and Sarma (1985)), which has further hindered the broader adoption of 3D LEMs. This study compares three well-established 3D LEMs. It first compares the characteristics of these methods under simple slope conditions. Then, the comparison is extended to a large area with numerous slopes, considering actual terrain data and a heavy rainfall event. The findings reveal variations in factor-of-safety values across methods. However, receiver operating characteristic curve indicated no significant differences between their accuracies in predicting the actual landslide distribution. This suggests that the Hovland method, boasting the lowest computational cost (e.g., under 50 min) among the methods, can effectively pinpoint high-risk areas identified by the other methods (e.g., over 7 h) by simply adjusting the factor-of-safety threshold. To further generalize these findings, future work should additionally consider highly variable and complex inputs, such as the spatial distribution of groundwater tables.