Evaluation of multibeam echosounder backscatter for classifying seafloor sediment using unsupervised and supervised technique at navigational channel / Muhammad Hafizuddin Mohmad Nazri

• Main Author: Mohmad Nazri, Muhammad Hafizuddin
• Format: Student Project
• Language: English
• Published: 2023
• Subjects: Geographic information systems
• Online Access: https://ir.uitm.edu.my/id/eprint/87958/1/87958.pdf; https://ir.uitm.edu.my/id/eprint/87958/

Capitalizing on the increased accessibility of bathymetry and backscatter data from multibeam echo sounders (MBES), this study explores the classification of seafloor sediments in a busy shipping channel. The dynamism of the seafloor, impacted by both natural and anthropogenic factors such as ship passage, demands robust techniques for accurate sediment classification. In this work, bathymetry and backscatter data obtained by the MBES Kongsberg EM2040 in 2021 were processed using software tools QPS Qimera, FMGT, and ArcGIS to derive both primary and secondary features. The former, limited in scope, were complemented by the latter that emerged from the integrated analysis of several data layers such as slope, BPI, rugosity, and backscatter properties. Principal Component Analysis (PCA) was employed to identify the most significant features for classification. For bathymetry derivatives, PCA generated 3 principal components where the combination explains a total of 95.08% for the total variance with Eastness (PCA1 – 35.67%), Northness (PCA2 – 68.69%), and Aspect (PCA3 – 95.08%). For backscatter derivatives, PCA generated 3 principal components where the combination explains a total of 99.66% of the variance with Mean (PCA1 – 92.88%), Second Moment (PCA2 – 98.11%), and Variance (PCA3 – 99.66%). Maximum Likelihood (ML) and ISO Cluster classification techniques were then applied, resulting in sediment maps with three distinct classes, similar to the results obtained from the Angular Range Analysis (ARA), employed as a proxy for ground truthing. The performance of the classification models was evaluated using Kappa and accuracy metrics, revealing that the ML Classification (Kappa: 0.087, Accuracy: 53.5%) outperformed the ISO Cluster Classification. Future work may consider ground truthing samples from the site to refine the classification further, employing techniques such as Random Forest. Finally, this research provides a new technique to classify the types of seabed sediment using backscatter which can determine and visualize using remote sensing and GIS technology.