Electromagnetic and Cone Penetration Test Data Fusion on Levee Characterization

The research combined geophysical (HEM) and geotechnical (CPT) data to simulate subsurface conditions during levee risk analysis in an area with complex geology. Owing to complexities such as saltwater intrusion, machine learning techniques (Random Forests, neural networks) were applied in order to combine these sets of data and estimate soil behavior and lithology. Spatial consistency was applied during data preprocessing. The constructed models predicted the major soil types (clay, silt, sand) with high accuracy, consistent with the established stratigraphy. The models performed poorly in the case of mixed types. The study delivers interpretable and continuously spatial models that, in the end, provide an enhanced tool during the working lives of the engineers involved in levee safety and design.