Abstract: To improve the accuracy of assessing seismic landslide susceptibility, we proposed a landslide susceptibility assessment model from the perspective of multimodal data fusion. With the Lushan earthquake as a case study, we extracted fifteen causative factors of landslide from multi-source data and divided them into four modalities. With the slope unit being the assessment unit, we compared and evaluated the present method and the logistic regression model by using receiver operating characteristic (ROC) curve and zonal landslide hazard point density. Results demonstrate that the area under the ROC curve is 0.86, and the hazard point density of high and very high susceptibility zones is 2.24 points/km², indicating that the present model is well applicable to the study region. Also, the present model has higher accuracy and better performance than the logistic regression model, suggesting that the presented model has a better applicability in the region.

Key words: landslide, susceptibility assessment, accuracy, multimodal classification model, causative factors, Lushan earthquake