University of Edinburgh logo British Antarctic Survey logo National Oceanographic Centre logo University of Leeds logo


Remote sensing and machine learning techniques for improved nearshore wave prediction

Image of Guadeloupe, Dominica and Martinique islands from a Medium Resolution Imaging Spectrometer [Source: ESA].
Scientific background and motivation: About 40% of the world’s population lives within 100 km from the coast. Present-day storm surge (often associated with large waves), aggravated by predicted sea level rise and driven by changes in extreme wave climate due to global warming [1] are serious hazards for coastal communities, ecosystems and infrastructure. To manage and mitigate current and future coastal hazards the characterisation of wave properties in the nearshore region, including the surf and shoaling zones, is essential. A vital part of the prediction and mitigation of coastal hazards are accurate observations of nearshore wave dynamics, however most ‘traditional’ oceanographic techniques require extensive resources to deploy, maintain and analyse. Additionally, the complex nature of nearshore hydrodynamics can be difficult to measure reliably with many in situ techniques, and most available measurements (e.g., wave buoys) are located further offshore. Satellite remote sensing represents an exciting opportunity to leverage internationally funded sensor platforms to understand the global coastal ocean at the spatial and temporal scales necessary to meet the challenge of understanding our changing environment. However, these sensor platforms are not without limitations, being restricted by e.g., local weather, flight orbits and revisit time. Innovative recording and data processing methodologies have arisen in the past decades, such as video imagery of nearshore wave patterns and machine learning [2,3] that present the opportunity to further improve and utilise satellite remote sensing of nearshore waves through hybrid data assimilation techniques: the blending of large amounts of data from multiple sources. The combination of these approaches provides the potential for a powerful tool to produce better estimates of nearshore wave variables, which might be otherwise difficult to monitor and predict.

Aim and objectives of the PhD: This PhD will develop a methodology to accurately estimate nearshore wave conditions from available offshore observations; combining in situ measurements, satellite imagery and altimetry data, numerical modelling and potentially shore-based optical imagery in order to improve nearshore wave monitoring and forecasting. The PhD will help reduce uncertainty in the operational prediction of wave conditions and in future wave climate projections. The project will answer the following research questions:

  • How can remote sensing be combined with in situ measurements and model data to predict nearshore wave parameters accurately and at high resolution using artificial intelligence?
  • What can this new method tell us about the effects of different marine conditions on nearshore wave transformation?

Methodology: The project will develop a methodology to estimate nearshore wave parameters at high resolution from offshore records [4,5] and satellite data [6], using machine learning techniques [7], and considering available outputs from numerical models [8,9]. The case study area will be Martinique, an ecologically diverse French island of the Eastern Caribbean, which is exposed to powerful Atlantic swell and hurricane hazard [10]. Convolutional neural networks will be applied to link multispectral satellite imagery (e.g., Sentinel-2 or Landsat 8, used to extract wavelength, direction and wave height estimates [2], with potential for 2D mapping) and satellite altimetry data (e.g., CFOSAT or Sentinel-1, providing quality wave spectrum products, but only along-track). The results will be related to model outputs and offshore in situ data using other machine learning techniques (e.g., random forests or artificial neural networks for regression) to produce predictions of nearshore wave parameters. Local video imagery from Basse-Pointe (northern Atlantic coast) may be used at later stages of the project to validate the methods and contribute to increased resolution of the final product. The student will be trained on the use and interpretation of in situ and satellite data, numerical model outputs, and machine learning techniques.


  1. Belmadani, A., Dalphinet, A., Chauvin, F. et al. Clim Dyn (2021).
  2. Almar, R., Bergsma, E.W.J., Catalan, P.A. et al. Remote Sens (2021).
  3. Al Najar, M., Thoumyre, G., Bergsma, E.W.J. et al. Mach Learn (2021).
  4. CANDHIS network (in French):
  5. NOAA NDBC network:
  6. European Space Agency, Sentinel Satellites:
  7. Rodriguez-Delgado, Bergillos, Medina-Lopez. Science of the Total Environment (2021), 789, 144039,
  8. WAMDI group. J Phys Oceanogr (1988), 18, 1775-1810.<1775:TWMTGO>2.0.CO;2
  9. Tolman, H.L. J Phys Oceanogr (1991), 21, 782–797.<0782:ATGMFW>2.0.CO;2
  10. Krien, Y., Arnaud, G., Cécé, R., Ruf, C., Belmadani, A. et al. Remote Sens (2018).