Skip to Content

Estimating coastal bathymetry from both Sentinel-1 and Sentinel-2 using image processing and machine learning techniques

Bathymetry retrieval in coastal waters is a particularly interesting application of satellite Earth Observation. In fact, creating a bathymetry map from traditional data (shipborne sonar or airborne lidar) is both costly and time consuming, while satellite data can provide maps over large areas in a very short time. However the processing involved in bathymetry retrieval and the interpretation of the results is mastered only by a handful of experts today. This may be about to change, thanks to some recent work on the Coastal TEP.

The Coastal TEP is a cloud-based ‘one-stop shop’ that gathers coastal-zone satellite data, processing algorithms and computing power. Obviously coastal bathymetry has been one of the main topics of interest for the Coastal TEP project.

In the frame of the ESA project ECOBAW, Céline Danilo from Univ. of Trento, has implemented in the Coastal TEP an algorithm which retrieves coastal bathymetry using Sentinel-2 images. In collaboration with Prof. Farid Melgani at Univ. of Trento, this project aims to develop applications for estimating coastal bathymetry without any ancillary data (wave period or pre-information of the bathymetry) from both Sentinel-1 and Sentinel-2 using image processing and machine learning techniques.

The algorithm implemented in Coastal TEP, depends on a first estimation of coastal bathymetry with a coarse resolution and on the reflectance of the four spectral bands of the visible and near infrared domain. In the ECOBAW project, this first estimation is provided by a model based on wave propagation and described in an IEEE TGRS publication[1]. Such estimation is successively improved by means of an innovative approach using a Gaussian Process Regression (GPR) model. In particular, it uses these first estimations as target values to train a GPR model which has as input the four spectral bands. The GPR model learned is then able, given the reflectance of the four spectral bands, to estimate the water depth and a related confidence value (estimate variance).  The theoretical principle relies then on the light extinction with water depth. This approach is thus capable to improve the knowledge of the coastal water depth, especially regarding the resolution and the coverage of the training values. Besides, the algorithm can be applied to any cloudless image and is completely unsupervised and automatic.

After validation of her processor, Céline was able to integrate the processor on the Coastal. It is now available to other users as a “contributed” processor in the Coastal TEP catalogue. In this way, all users will be able to reproduce Céline’s results.

The good thing about Coastal TEP is that it provides an easy interface to integrate a processor, so algorithms developers can work autonomously with minimal support from the Coastal TEP team. Within some weeks, Céline has been able to process some Sentinel-2 images on the platform and compute a bathymetry map for the beach of Waimanalo in Hawaii.

This first success has generated many ideas for the future.

We could extend the capabilities of the current processing to be able to handle any geographical zone. Even if the model providing the first estimation is not yet available on the platform, any other kind of first estimations could be exploited as well for training the model. 

We could also provide other bathymetry retrieval algorithms and allow users to choose between different approaches. Indeed each algorithm has its benefits and drawbacks, so having a set of processors can improve the reliability of the EO-based retrieval. A benchmark of algorithms on a given site where reference in-situ measurements are available could make a lot of sense for the future. It would help provide guidelines and recommendations and help promote the use of EO-derived bathymetry maps.


[1] Céline Danilo and Farid Melgani, “Wave Period and Coastal Bathymetry Using Wave Propagation on Optical Images,” IEEE Transactions on Geoscience and Remote Sensing 54, no. 11 (2016): 6307–6319.

Polar TEP provides polar researchers with access to computing resources, data and software tools for polar research.