Probabilistic Characterization of Sediments Using a Combined Geotechnical and Geophysical Approach

Abstract

Reliable characterization of seabed surface sediments is critical for offshore engineering, naval applications, and understanding coastal sediment dynamics. Traditional geotechnical methods provide accurate point measurements but lack spatial continuity, while geophysical surveys offer broad coverage but yield indirect properties that are often difficult to link quantitatively to engineering parameters such as sediment strength or erodibility. This dissertation addresses these limitations by developing and integrating novel probabilistic frameworks utilizing data from Portable Free Fall Penetrometers (PFFP) and Chirp sub-bottom profilers for enhanced shallow-water sediment assessment. First, a probabilistic machine learning model, combining Random Forest and a Bayesian 1D Convolutional Neural Network, is developed to classify sediments based on full PFFP deceleration profiles. This approach moves beyond deterministic methods by providing robust classifications across four behavior types with quantified uncertainty bounds. Second, acknowledging the complementary strengths of geotechnical and geophysical data, a novel data fusion framework based on Gaussian Process Regression and Bayesian methods is introduced. This framework quantitatively integrates the probabilistic classifications derived from sparse, high-accuracy PFFP measurements with continuous, lower-certainty classifications obtained from Chirp sonar data via geophysical inversion. The result is a unified, spatially continuous sediment profile along survey transects with significantly reduced and quantified uncertainty compared to using either data source alone, demonstrated through field case studies. Third, the research establishes a direct, data-driven link between rapid in-situ PFFP measurements and sediment erodibility, specifically the critical shear stress (tauc). A two-step probabilistic model is developed, first correlating PFFP deceleration with sediment grain composition (fines-sand ratio) and subsequently linking this composition to tauc values derived from laboratory erosion tests. This framework provides the first probabilistic estimates of erodibility directly from PFFP data, highlighting the controlling influence of the fines-sand ratio and revealing significantly higher variability in the erodibility of mixed sediments. Collectively, this dissertation delivers an integrated suite of probabilistic tools that leverage PFFP and Chirp sonar data for rapid, spatially comprehensive, and uncertainty-aware characterization of surficial seabed sediments. These advancements enhance the interpretation of PFFP data, enable robust fusion with geophysical surveys, and provide novel means to assess sediment erodibility, contributing significantly to improved site investigations and predictive capabilities in dynamic coastal and estuarine environments.