Weak-layer position effects in stratified landslides based on a generative adversarial network surrogate model

[Objective] Changes in the spatial position of weak layers can significantly affect the sliding behavior and post-failure depositional evolution of stratified landslides. However, the controlling effects of weak-layer position on the coupled responses of depositional morphology, runout distance, and deposit thickness remain insufficiently understood. This study aims to investigate the characteristics and governing patterns of depositional responses of stratified landslides to variations in weak-layer position.[Methods] Based on a previously established rapid landslide-deposition prediction framework, a generative adversarial network (GAN) surrogate model was employed as a tool for multi-scenario analysis. Six three-layer stratified landslide scenarios with different weak-layer positions were constructed using MatDEM. Paired scenarios were further designed to systematically examine the differences in depositional morphology, runout distance, and deposit thickness as the weak layer migrated from the upper part to the middle part of the landslide body.[Results] The surrogate-model predictions showed good agreement with the MatDEM reference results, with mean SSIM and IoU values of 0.915 and 0.935, respectively, indicating that the model effectively preserved the overall outline and spatial distribution characteristics of the post-failure deposits. When the weak layer was located in the upper part, the post-failure deposits were mainly characterized by shallow traction and overall spreading. In contrast, when the weak layer was located in the middle part, interlayer differentiation, structural reorganization, and localized thickening became more pronounced. Overall, variations in weak-layer position exerted a more consistent influence on the redistribution of deposit thickness than on frontal runout, and deposit thickness exhibited a more consistent response than runout distance.[Conclusions] Weak-layer position is an important structural factor controlling the depositional evolution of stratified landslides, and deposit thickness can serve as a sensitive indicator for identifying weak-layer position effects. The GAN surrogate model can improve the efficiency of multi-scenario analysis while preserving key depositional characteristics, providing an efficient approach for investigating depositional evolution patterns of stratified landslides.