Novel Sea Wave Measurement Methods from Mono and Stereo Aerial Photography, Developed in a Photorealistic Virtual Environment

One of the biggest unresolved challenges in offshore wind farm maintenance is the real-time prediction of wave-induced motions and forces, enabling repair vessels to approach turbines safely. Even in sea states normally beyond the wind turbine accessibility limits, there are often quiescent periods lasting tens of seconds, during which the approach is possible. The end goal of this project is to develop a method for the immediate and accurate prediction of these quiescent periods, based on optical images acquired by drones. More generally, the processing of drone-based optical images will offer potential for cost-effective, high-resolution wave field measurements over unprecedented scales.

At IFP Energies nouvelles, innovative methods are being developed to measure and forecast ocean surface waves across a wide range of spatiotemporal scales [0], with applications in marine renewable energies and oceanography. In this PhD project, new techniques will be developed to “scan” wave fields from images captured by drones equipped with simple cameras, either through single-image acquisition (a single drone), or stereo acquisition (two drones flying close together).
These drone-based approaches will provide access to wavefield measurements over wider areas than fixed stereo systems, giving access to wavelengths ranging from a few tens of centimetres to several hundred metres. They would also be easily adaptable to prevailing conditions in terms of lighting and wave characteristics.
There is nowadays a large corpus of public and private wave imaging data, with and without ground truth [1,2]. Still, this data does not seem to capture all the possible variations and uncertainties of lighting and motion that are found in the wild. Hence, drone-based approaches require dedicated developments to take account of variations and uncertainties in lighting and camera positioning. Furthermore, modern single-image approaches [3] rely on deep learning methods and, therefore, require the creation of suitable training datasets.
To address these challenges in an original and cost-effective way, the real data will be augmented with simulations in a photorealistic virtual environment, which will be developed by the PhD student. The resulting synthetic datasets will be used to train and develop both single-image and stereo measurement methods, capable of retrieving surface displacement together with associated uncertainty estimates.
Several field measurement campaigns at sea will also be carried out to adjust the simulator parameters based on real environmental conditions, enrich the training database, and quantitatively validate the developed approaches using in-situ measurements.
The PhD student will work on a promising and highly interdisciplinary topic at the intersection of data science and machine learning, visual computing, hydrodynamics and oceanography. Thus, the candidate will develop expertise in fluid mechanics for sea state simulation, computer vision for estimating ocean conditions from images, and computer graphics for simulating realistic acquisitions to support the training of models with scarce ground truths.

Keywords : Applied maths. Computer science. Visual computing. Machine learning. Oceanography. Hydrodynamics.

References
[0] Mérigaud, A., Zhu, J., & Tona, P. Assessing the predictability of random ocean waves. Journal of Fluid Mechanics, 1007, A47. (2025). https://doi.org/10.1017/jfm.2025.84
[1] Guimarães, P.V., Ardhuin, F., Bergamasco, F. et al. A data set of sea surface stereo images to resolve space-time wave fields. Sci Data 7, 145 (2020). https://doi.org/10.1038/s41597-020-0492-9
[2] Gomit, G., Chatellier, L. & David, L. Free-surface flow measurements by non-intrusive methods: a survey. Exp Fluids 63, 94 (2022). https://doi.org/10.1007/s00348-022-03450-5
[3] Matsuki, H., Murai, R., Kelly, P. H., & Davison, A. J. Gaussian splatting SLAM. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition (2024). https://doi.org/10.1109/CVPR52733.2024.01708
[4] Smith, R., Dias, F., Facciolo, G., & Murphy, T. B. Pre-computation of image features for the classification of dynamic properties in breaking waves. European Journal of Remote Sensing, 56(1), 2163707 (2023). https://doi.org/10.1080/22797254.2022.2163707
[5] Aira, L. S., Facciolo, G., & Ehret, T. Gaussian splatting for efficient satellite image photogrammetry. In Proceedings of the Computer Vision and Pattern Recognition Conference (pp. 5959-5969) (2025). https://arxiv.org/html/2412.13047v2
[6] Ouerghi, E., Ehret, T., Facciolo, G., Meinhardt, E., Marion, R., & Morel, J. M. Tightening up methane plume source rate estimation in EnMAP and PRISMA images. Atmospheric Measurement Techniques, 18(18), 4611-4629 (2025). https://doi.org/10.5194/amt-18-4611-2025

• Academic supervisor    : Prof. Frédéric Dias and Dr. Enric Meinhardt-Llopis (Centre Borelli). The Centre Borelli is the ideal environment for the supervision of this research, due to its proven experience in both fluid mechanics and image processing [4-6].
• Doctoral School    ED 574 - Ecole doctorale de mathématiques Hadamard (EDMH)
• IFPEN supervisor Dr Alexis MERIGAUD    
• PhD location    IFPEN, Rueil-Malmaison, France  
• Duration and start date    3 years, starting in 2026, November 2
• Employer    IFPEN
• Academic requirements    Master’s degree (Master 2) required, ideally including a strong background in visual computing and machine learning. Interest in Fluid Mechanics.
• Language requirements    English level B2 (CEFR) 
• Other requirements    Very good programming skills (in Python for instance).

To apply, please send your cover letter and CV to the IFPEN supervisor indicated here below.