Prediction of iceberg-seabed interaction using machine learning algorithms

Azimisiahchaghaei, Hamed (2023) Prediction of iceberg-seabed interaction using machine learning algorithms. Doctoral (PhD) thesis, Memorial University of Newfoundland.

[img] [English] PDF - Accepted Version
Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission.

Download (21MB)


Every year thousands of icebergs are born out of glaciers in the Arctic zone and carried away by the currents and winds into the North Atlantic. These icebergs may touch the sea bottom in shallow waters and scratch the seabed, an incident called “ice-gouging”. Ice-gouging may endanger the integrity of the buried subsea pipelines and power cables because of subgouge soil displacement. In other words, the shear resistance of the soil causes the subgouge soil displacement to extend much deeper than the ice keel tip. This, in turn, may cause the displacement of the pipelines and cables buried deeper than the most possible gouge depth. Determining the best burial depth of the pipeline is a key design aspect and needs advanced continuum numerical modeling and costly centrifuge tests. Empirical equations suggested by design codes may be also used but they usually result in an over-conservative design. Iceberg management, i.e., iceberg towing and re-routing, is currently the most reliable approach to protect the subsea and offshore structures, where the approaching icebergs are hooked and towed in a safe direction. Iceberg management is costly and involves a range of marine fleets and advanced subsea survey tools to determine the iceberg draft, etc. The industry is constantly looking for cost-effective and quick alternatives to predict the iceberg draft and subgouge soil displacements. In this study, powerful machine learning (ML) algorithms were used as an alternative cost-effective approach to first screen the threatening icebergs by determining their drafts and then to predict the subgouge soil displacement to be fed into the structural integrity analysis. Developing a reliable solution to predict the iceberg draft and subgouge soil displacement requires a profound understanding of the problem's dominant parameters. Therefore, the present study started with dimensional analyses to identify the dimensionless groups of key parameters governing the physics of the problem. Two comprehensive datasets were constructed using the monitored characteristics of the real icebergs for draft prediction and experimental studies for the subgouge soil displacements reported in the literature. Using the constructed database, 14 ML algorithms ranging from neural network-based (NN-based) to three-based methods were sequentially used to predict the iceberg draft and the subgouge soil displacement. The studies were conducted both in clay and sand seabed. By different combinations of the input parameters, several ML models were developed and assessed by performing sensitivity analysis, error analysis, discrepancy analysis, uncertainty analysis, and partial derivative sensitivity analysis to identify the superior ML models along with the most influential input parameters. The best ML model was able to predict the iceberg drafts alongside the subgouge soil features with the highest level of precision, correlation, and lowest degree of complexity. A set of ML-based explicit equations were also derived from the wide range of field and experimental measurements for the estimation of iceberg drafts, subgouge soil deformations, and ice keel reaction forces, which outperformed the existing empirical equations. The study resulted in developing a set of tools that can be used for both a cost-effective screening of the threatening icebergs and the prediction of the corresponding subgouge soil displacements. The outcome of the study can effectively contribute to a significant reduction of iceberg management costs and greenhouse gas (GHG) emissions through the mitigation of the marine spread operation.

Item Type: Thesis (Doctoral (PhD))
Item ID: 16132
Additional Information: Includes bibliographical references
Keywords: iceberg draft, iceberg-seabed interaction process, dimensionless analysis, machine learning, simulation, subsea assets
Department(s): Engineering and Applied Science, Faculty of
Date: October 2023
Date Type: Submission
Digital Object Identifier (DOI):
Library of Congress Subject Heading: Icebergs--North Atlantic Region; Ice calving--North Atlantic Region; Machine learning; Ocean bottom--North Atlantic Region

Actions (login required)

View Item View Item


Downloads per month over the past year

View more statistics