Satellite derived bathymetry for times of absent in situ data

Abstract

Knowledge of seafloor topography (bathymetry) is increasingly important as coastal environments are unprecedentedly stressed by climate change and anthropogenic pressure. The bathymetry of shallow nearshore waters is yet marginally monitored due to costly and time-intensive survey techniques. Methods to obtain satellite derived bathymetry (SDB) have become increasingly valuable. Mapping temporal change is however challenging, because the majority of these methods remain heavily dependent on situ observations. This thesis introduces an SDB approach to estimate temporal bathymetric changes, which omits the need for synchronous in situ data. The approach is based on a reference image correction method that enables direct comparison of multitemporal imagery and temporal extrapolation of a conventionally-trained bathymetry estimation model (BEM). Research focused on pre-processing multispectral imagery, developing a bathymetry estimation model and estimating bathymetry for times of absent in situ data. The proposed method is demonstrated with a case study in the Dutch Wadden Sea; a site characterised by dynamic morphology, high turbidity and homogeneous bottom type. A log-linear estimation model is obtained by linear regression on in situ observations and the three visible bands of Sentinel-2 imagery. Scarcity of high-quality Sentinel-2 imagery is managed by combing multiple images into a six-month composite. The availability of two sets of vaklodingen in situ observations allowed for training and testing two bathymetry estimation models (BEM 2016 and BEM 2019) and for cross-validating the depth estimates after a three-year extrapolation of these models. Bathymetry is estimated for times of absent in situ data by temporal extrapolation of the two estimation models. The extrapolation showed estimation of shallow bathymetric structures in up to four metre water depth with an RMSE of approximately one metre. Additionally, the migration direction of these bathymetric structures is successfully estimated. Within the tested three-year time frame, predictive power did not decrease. These results imply that estimation performance is governed by composite quality and predictive power of the bathymetry estimation model. The limited influence of temporal extrapolation on estimation performance suggests that the availability of high-quality satellite imagery and one set of non-synchronous in situ observations is sufficient to estimate bathymetry for times of absent in situ data. The proposed method potentially provides a tool for mapping temporal bathymetric changes of nearshore zones across the globe.