Shallow water bathymetry mapping using Support Vector Machine (SVM) technique and multispectral imagery

TLDR

Satellite imagery combined with image processing offers a time‑ and cost‑effective alternative to traditional bathymetry methods. The study applies a nonlinear SVM to derive shallow water bathymetry for Sint Maarten Island and Ameland Inlet by integrating echo‑sounding data with Landsat 7 ETM+ and Landsat 8 OLI reflectances at 30 m resolution. The SVM model was trained on 80 % of echo‑sounding points using a radial‑basis kernel, with hyperparameters tuned via cross‑validation, and its predictions were compared against linear and ratio transform models. The model achieved 8.26 % error for Sint Maarten Island and 14.43 % for Ameland Inlet, demonstrating comparable or superior performance to traditional linear and ratio transform approaches for shallow depths.

Abstract

Satellite imagery along with image processing techniques prove to be efficient tools for bathymetry retrieval as they provide time and cost-effective alternatives to traditional methods of water depth estimation. In this article, a nonlinear machine learning technique of Support Vector Machine (SVM) is used to derive shallow water bathymetry data along Sint Maarten Island and Ameland Inlet, The Netherlands, by combining echo-sounding measurements and the reflectance of blue, green, or red bands of Landsat Enhanced Thematic Mapper Plus (Landsat 7 ETM+) and Landsat 8 Operational Land Imager (OLI) imagery with 30 m spatial resolution. In the analysis, 80% of data points of the echo-sounding measurements are used for training and the remaining 20% data points are used for testing. The model utilizes the radial basis kernel function (nonlinear) and the other training factors such as the smoothing parameter, penalty parameter C, and insensitivity zone ε are selected and tuned based on the learning (i.e. training) process. The overall errors during test phases for Sint Maarten Island (1–15 m) and Ameland Inlet (1.00–3.50 m) are 8.26% and 14.43%, respectively, reflecting that the model produces significant estimations for the shallow depths ranges, considered in this study. The results obtained are also compared statistically with those estimated from the widely used linear transform model and ratio transform model, which establish a linear relationship between the water depth and band reflectances. Based on the results, it is evident that SVM provides a comparable or better performance for shallow depth ranges and can be used effectively for deriving accurate and updated medium resolution bathymetric maps.