Satellite-derived coastal bathymetry, run-up hazard and seagrass mapping: from test areas to pioneering global maps

This project aims to use Big Data analytics with Sentinel-2 multispectral satellite imagery to develop an automated process for (i) mapping of coastal bathymetry, to depths of ca. 30m, along with (ii) the distribution of sea grass (effective for carbon storage) within that coastal zone.

With satellite-derived bathymetric maps we can then go on to map the wave run-up hazard from storm surges or tsunami, guiding disaster risk reduction strategies for potentially impacted coastal communities.

Satellite-derived mapping and quantifying of the distribution of sea grass will inform countries about a key component of their national Blue Carbon inventory, information that will assist them in bidding for climate resilience funding (e.g. from the World Bank).

This study builds on research into Satellite-Derived Bathymetry (SDB) carried out for the UK Space Agency-funded CommonSensing project (Jan 2019 to March 31st 2021), in conjunction with UoP’s Global Earth Model (GEM) group, the Centre for Applied Geoscience (CAG) and the Centre for Blue Governance (CBG).
During the past year, we have developed this methodology using test areas with detailed sonar and lidar bathymetry data, in Fiji and Vanuatu – from which two peer reviewed journal articles are being drafted, one on the data analytics, the other on the coastal risk applications. Recently we have been able to test our preliminary methodology, developed around Pacific islands, against the situation along the coast of West Africa, using published maps of coastal bathymetry and sea grass distribution.

We will use this TRIF funding to strengthen our ‘proof of concept’ pilot studies, which will improve our chances of success in various funding bids, later this year, to NERC (Standard Grant application, aiming to up-scale from test areas to global mapping); STFC (Innovation Partnership Scheme and IAA funding for the data analytics aspects); Leverhulme Trust funding for the coastal disaster risk reduction aspects; as well as for ESA and H2020 calls involving satellite-based coastal mapping and monitoring (e.g. via the NEANIAS consortium for the bathymetry data processing). We also plan to publish two further peer-reviewed papers: on the improved methodology for the automated satellite-derived bathymetry and on coastal habitat mapping applications.

The aim of this TRIF project is thus twofold: to develop our preliminary research, strengthening our case for ensuing funding proposals; and to publish at least two peer-reviewed papers about our findings. To achieve that aim, we have the following objectives:

1.Improve the accuracy of our existing bathymetry method, with removal of satellite image artefacts.
2. Test our satellite-derived bathymetry (SDB) methodology developed in the Pacific, against the published bathymetry of our new West Africa sites (main problems are Sahara dust storms, and a larger tidal range).
3. Modify our SDB methodology for mapping and monitoring the extent of sea grass in the West Africa test areas.

We will hire a short-term postdoc with a strong data analysis background (for instance, a recent graduate of the ICG, or via the DISCnet CDT) for 3 months, to take forward methods pioneered in our CommonSensing work on satellite-derived bathymetry. This postdoc will work closely with the Global Earth Modelling group, which will help build capacity in this group for future projects and funding bids. We are currently using two years of Sentinel-2 multispectral satellite data over hundreds of thousands of km2 in the Pacific. The multispectral optical data has been processed through a shallow neural network to determine the coastal sea-floor depth around Fiji, Vanuatu, and the Solomon Islands, which will be used to predict the run-up hazard from storm surge or tsunami waves.

We plan to take this pioneering work in two new directions with the help of a short-term postdoc. Our data has shown the ability to detect areas of sea grass, which are significant carbon sinks. We will extend our methods to use deep convolutional neural networks to reliably detect and measure the extent of these areas. This will simultaneously improve the reliability of our bathymetry and the resulting run-up hazard prediction, as seagrass is an important confounder in depth estimation. We will test this in 2 areas of the West African coast for which there are published maps of bathymetry and sea grass distribution (Banc d’Arguin, Mauritania; Joal, Senegal).