Bathymetry modelling of the eastern Tendrivska Bay (Ukraine) using Sentinel-2 remote sensing data

The aim of the study was to create a bathymetric model of the eastern Tendrivska Bay based on Sentinel-2 remote sensing data. For this purpose, the method of log-transformed spectral band ratios was used. Initially, cloud-free Sentinel-2 scenes from 28 dates in 2015–2018 were selected for the study. The preliminary processing of remote sensing data included atmospheric correction using the DSF (Dark Spectrum Fitting) algorithm, clipping to the region of interest, and merging identical visible spectrum bands from two adjacent tiles that fully covered the eastern Tendrivska Bay. High-frequency noise was removed by applying a 7x7 pixel window median filter to each band. In the final stage of remote sensing data preparation, the log-transformed spectral band ratios were calculated and exported to raster files using a mask that clipped the final images to the bay boundaries. The study calculated and evaluated three types of models: BG (blue to green bands log-ratio), BR (blue to red bands log-ratio), and GR (green to red bands log-ratio). Field depth measurements from 79 sites were used to train the models through regression analysis. Model validation was performed using multiple k-fold cross-validation. The BR and GR models showed the best results for bathymetric modeling of the eastern Tendrivska Bay. In contrast, BG models were unsuitable due to their low accuracy. The final bathymetric model of the eastern Tendrivska Bay was obtained by averaging 12 individual models (7 BR and 5 GR models) that showed the best results in the regression analysis. The spatial variation of the model’s vertical accuracy, assessed by the standard deviation raster, indicated that vertical accuracy was slightly lower for relatively deep areas and areas with intensive water dynamics. This is due to the greater variability in water transparency in such areas. Overall, the obtained bathymetric model is characterised by high spatial resolution and vertical accuracy, making it suitable for geomorphological studies. It will also be useful as a spatial variable for modeling the distribution of aquatic organisms and waterfowl in the bay using machine learning methods. The experience of creating a bathymetric model for the eastern Tendrivska Bay showed that this method is quite simple and accessible, allowing it to be recommended for modelling the seabed topography of other shallow water bodies.