Stage | Modélisation du climat et de son changement global | Hydrologie
Niveau requis : M1 ou M2
Date de début : 2 février 2026
Durée de la mission : 4 à 6 mois
Date limite des candidatures : 31 décembre 2025
Gratification : 649 euros
Hydrodynamics – Remote sensing – Flooding
AST-Hydro
Internship 4-6 months (Master 1/Master 2)
Expected starting date 02/2025 to 04/2025 depending on duration
Salary: 649 euros/month
Contact: ricci@cerfacs.fr, sylvain.biancamaria@utoulouse.fr
Context
Water is a scarce resource both nationally and globally. In France, 18 million people live in areas at risk of flooding when rivers overflow their banks. The frequency of summer droughts has doubled compared to the 1975–2005 period, and Europe has experienced a more than 1,000% increase in natural disasters between 1979 and 2019.
The « Axe Scientifique Transverse Hydrologie Spatiale » (AST-Hydro, https://www.omp.eu/ast-hydrologiespatiale/) at the « Observatoire Midi-Pyrénées » (OMP) is a consortium that brings together eight research laboratories committed to advancing the use of satellite data for hydrological applications. Satellite remote sensing in hydrology can be applied to the direct estimation of fluxes, water storage components, and water quality indices. It can also be integrated with hydrological models through various approaches, where satellite data are employed for model forcing, calibration, data assimilation, and validation. Several members of the consortium share a research interest in how remote sensing observations can be used to characterize flood extents during major flood events across both local and large catchments, in regions with varying levels of in-situ instrumentation worldwide. In this context, the AST-Hydro consortium is offering a Master's internship focused on these topics.
The internship will be hosted at CECI (Climat, Environnement, Couplages et Incertitudes), with strong collaborations with Centre National d’Etudes Spatiales (CNES, https://cnes.fr/), other labs in AST-Hydro ( Laboratoire d’Etudes en Géosciences et Océanographie Spatiale LEGOS, https://www.legos.omp.eu/ and Laboratoire en Géosciences environnement Toulouse GET, https://www.get.omp.eu/) as well as partners in Luxembourg (Luxembourg Institude of Science and Technology LIST, https://www.list.lu/ and the private company RSS-Hydro https://www.rsshydro.lu/).
CECI plays a significant role in national and international efforts to create innovative and robust solutions that provide reliable forecasts for water management. Together with numerous academic and private partners, CECI works on combining in-situ measurements, satellite data, numerical models and data science to develop algorithms, products, and services for the reanalysis and forecasting of water levels and discharge, from local to global scales. Building on over a decade of research in data assimilation and sensitivity analysis for hydrodynamic modeling at CECI—supported by funding and collaborations with CNES, EDF/LNHE, and Météo-France—our research focuses on demonstrating the value of remote sensing observations for improving short- to medium-term flood extent predictions. Number of publications have demonstrated the merits of combining Water Surface Elevation (WSE) maps generated by numerical modeling with flood extents derived from remote sensing Sentinel 1 images.
Objective of the internship
Deriving river water and flood extent maps is an active area of research. In recent years, several fully automated methods for mapping water extents from satellite imagery have been developed. Most of these approaches rely on algorithms trained on manually generated water mask databases. They convert satellite backscatter data—such as those acquired by Sentinel-1—into binary masks distinguishing wet and dry pixels. The Sentinel-1 mission consists of two polar-orbiting satellites operating continuously, day and night. Using C-band synthetic aperture radar (SAR) imaging, Sentinel-1 can capture high-resolution images under all weather conditions. Although these water masks offer a global perspective on hydrological states, particularly during flood events, they remain imperfect and limited by low revisit frequency. Automatically satellite-derived water masks can be evaluated against simulated water masks generated by two-dimensional hydrodynamic models. On one hand, such comparisons reveal uncertainties in satellite products, particularly in areas affected by vegetation cover, urban environments, and humid or wetland regions. On the other hand, simulated water masks are subject to uncertainties arising from factors such as river and floodplain topography, friction coefficients, and boundary inflows. Since both datasets contain inherent errors, recent studies have proposed methods for directly assimilating flood extent maps into flood forecasting chains and hydraulic models. This integration enables a more reliable representation of water extents, thereby paving the way for improved operational applications.
As part of this internship, we aim to explore various software solutions that provide algorithm and databases for generating river water extents (prior to data assimilation), with a focus on flooded surface maps, from Sentinel data. These algorithms rely on databases built from observed flood events and employ a range of learning approaches. First, we propose to investigate open-access databases available for training purposes, such as OPERA from PODAAC, FLOODS from the Joint Research Centre (JRC) or GIEMS-D3 from LERMA (Observatoire de Paris). Then a comprehensive review of state-of-the-art algorithms for automated flood extent generation will be conducted, followed by an assessment of their performance on the selected datasets. Specifically, the FloodML algorithm developed by CNES, which generates inundation maps from Sentinel-1 time series for event-based assessments, and the HASARD algorithm developed at LIST as part of the European Copernicus Global Flood Monitoring program, will be evaluated. The accuracy and reliability of both FloodML and HASARD will be assessed across selected catchments and flood events in France (Garonne, Chinon, Saint-Omer), the United States (Ohio), and Cambodia. The possibility to widen the comparison to other algorithms such as FloodSENS (RSS-Hydro), ExportEO (iCube/SERTIT) or other algorithms hosted by GFM (TU-Wien, DLR) will be investigated.
The potential for transfer learning to other sensors—such as Sentinel-2 for optical data and COSMO-SkyMed, TerraSAR-X, or IceEye (if available) for very high-resolution SAR data—will also be explored. Further improvements may be proposed. For example, enhanced detection of flooded vegetation and urban areas is expected, along with the potential use of exclusion maps to more effectively remove snow-covered and irrigated areas where SAR data may be less reliable.
Your main activities will include:
- Image Processing
- Programming in Python, Fortran and/or C++
- Shallow water Modelling
Conduct and supervision of the internship
Under the supervision of Sophie Ricci (CECI), in collaboration with Sylvain Biancamaria (LEGOS)
The profile we are looking for
You have developed the following competences:
- Very good command of Python in general and using geographic data.
- Advanced skills in applied mathematics.
- Experience in shallow water, image/signal processing will be considered as an asset.
- Good oral and written communication skills
You have the following skills:
- Strong motivation to carrying out a research project
- Team working ability.
- Very good sense of organization.
You are pursuing a master degree or engineer diploma or equivalent in the following areas: fluid mechanics, applied mathematics, data science or environmental modelling with a strong interest in environmental sciences and earth sciences.
How to apply
Please contact
ricci@cerfacs.fr
sylvain.biancamaria@utoulouse.fr
omp.hydro-spatiale-animation@utoulouse.fr
