A Multi-Spatial, Multi-Temporal, Semi-Analytical Model for Bathymetry, Water Turbidity and Bottom Composition using Multispectral Imagery

TL;DR

This paper presents a semi-analytical model that estimates bathymetry, water turbidity, and bottom composition from multispectral satellite imagery, adapting physics-based methods for lower spectral resolution and validating with real-world data.

Contribution

The authors extend the HOPE model to multispectral imagery by incorporating temporal and spatial assumptions, enabling effective bathymetry estimation from Landsat-8 data.

Findings

Model achieves R^2 = 0.85 against sonar data

Mean absolute error of 1.17 meters

Mean relative error of 7.52%

Abstract

In this paper we introduce a semi-analytical model for bathymetry, water turbidity and bottom composition; which is primarily based on the physics-based model, HOPE, of Lee et al. Unlike the model of Lee, which was originally designed to use hyperspectral imagery, our model is specifically designed to use multispectral satellite imagery. In particular, we adapt to the greatly decreased spectral resolution by introducing temporal and spatial assumptions on the depth and water turbidity. We validate the extensions to the Lee et al model with a 260 km2 case study in the area of the Murion Islands off Western Australia, where we compare the atmospherically-corrected LANDSAT-8 derived bathymetry against a 2011 single-beam sonar survey by Transport Western Australia. The model validates well against the single-beam sonar survey, with R^2 = 0.85, a mean absolute error of 1.17 m and a mean relative error of 7.52%. This indicates the model could be widely applicable to LANDSAT-8 imagery.