Optimizing Surface-Wave Methods for Quick-Clay Characterization

Quick clays are a type of sensitive marine clay that pose a serious geohazard in many glaciated areas, including Sweden. Because they can rapidly lose their strength under stress, it is important to identify them early. Surface-wave methods are well suited for this task, since quick clays typically have low shear wave velocities. A key step in these methods is picking good dispersion curves from seismic data, which is often done manually. This is both time-consuming and prone to error. In this project, a convolutional neural network trained on hand-labeled data from a non–quick clay site in Denmark is tested on new data from a known quick clay site. The main purpose is to investigate how well the model generalizes to different geological conditions. The results show that the model works well overall, though it performs worse in cases with low signal-tonoise ratio or overlapping modes. If improved, this type of model could become a useful tool in the process of automating dispersion curve picking, which would in turn make surface wave methods more efficient in large-scale quick clay surveys.