PhD thesis | Climate Modelling And Global Change | Hydrology
Required Education : M2
Start date : 1 October 2026
Mission duration : 36 mois
Deadline for applications : 30 June 2026
Salary : 2500 euros brut mensuel
TOULOUSE – PhD (CDD 36 mois vortex-IO/CERFACS)
Contacts : Sébastion Villon (villon@cerfacs.fr), Sophie Ricci (ricci@cerfacs.fr)
Scientific Background and Challenges
Real-time hydrological forecasting is a major challenge for water resource management, flood prevention, and the safety of hydraulic infrastructures. In a context of increasing climate variability and intensifying extreme events, improving the reliability and transferability of hydrological forecasts has become both a scientific and operational priority.
Although conceptual and physically based hydrological models have proven effective, they generally require basin-specific calibration, limiting their applicability to poorly gauged or ungauged catchments. Over the last decade, deep learning approaches, particularly Long Short-Term Memory (LSTM) networks and their variants (EALSTM, hybrid CNN–LSTM models, and Transformers for time-series forecasting), have demonstrated their ability to develop regional or even global rainfall–runoff models with performances comparable to, or exceeding, those of individually calibrated models.
However, several scientific challenges remain open, including:
- Spatial generalization to underrepresented hydroclimatic regimes;
- Effective integration of catchment descriptors (topography, land cover, geology, and soils);
- Systematic evaluation of meteorological forcing sources (reanalyses, satellite products, numerical weather predictions);
- Rigorous quantification of predictive uncertainty for operational decision-making.
Within this context, we offer a CIFRE PhD position in partnership with CERFACS, entitled:
“Regionalizable Real-Time Hydrological Forecasting Based on Artificial Intelligence: Model Architectures and Dataset Optimization.”
Objectives
The primary scientific objective is to design, develop, and evaluate deep learning architectures capable of delivering reliable and transferable hydrological forecasts across different catchments, including ungauged basins, both in France and internationally. The candidate will contribute to the entire workflow—from data exploration to operational deployment—and will directly support vorteX-io's forecasting systems.
The successful candidate will join the Value-Added Hydrological Products Team at vorteX-io. The industrial scientific supervisor will be Rémy Lopez (CTO), while academic supervision will be provided by Sébastien Villon and Sophie Ricci (CERFACS).
In accordance with the CIFRE framework, working time will be shared between vorteX-io and CERFACS. During the first year, the expected distribution will be approximately 50% at CERFACS and 50% at vorteX-io, with possible adjustments in subsequent years depending on project progress.
Deep Learning Model Development
The candidate will:
- Design, train, and compare several multi-basin rainfall–runoff architectures (LSTM, EALSTM, hybrid CNN–LSTM models, Transformers);
- Integrate spatial catchment information (topography, land cover, geology, soil characteristics), notably through convolutional encoders applied to raster datasets;
- Optimize training strategies, including hyperparameter tuning, loss functions adapted to floods, droughts, and seasonal variability, and approaches for handling basin heterogeneity and data imbalance.
Meteorological and Hydrological Data
The candidate will:
- Identify and assess meteorological forcing datasets and their impact on model performance, including:
- Global reanalyses (ERA5),
- Satellite products,
- Global numerical weather prediction systems (ECMWF),
- Higher-resolution regional products (COMEPHORE, AROME);
- Build multi-basin datasets to evaluate generalization to ungauged basins;
- Conduct sensitivity and interpretability analyses to understand the influence of input variables and catchment attributes.
Validation, Uncertainty Quantification, and Operational Deployment
The candidate will:
- Evaluate model performance using hydrological metrics such as NSE, KGE, and bias, with particular attention to floods and extreme events;
- Quantify predictive uncertainty through model ensembles, Monte Carlo dropout, and quantile regression approaches;
- Validate the models over priority regions:
- Europe,
- Morocco (catchments feeding the Al Wahda and Idriss I reservoirs),
- Cambodia (Mekong Basin);
- Ensure compatibility with operational forecasting constraints, including forecast lead times, computational cost, and storage requirements.
Research and Scientific Dissemination
The candidate will:
- Publish scientific papers and participate in international conferences;
- Maintain active scientific monitoring of advances in computational hydrology and AI;
- Contribute to the integration of research outcomes into vorteX-io products.
What We Offer
- The opportunity to contribute to a project addressing one of the most critical environmental challenges worldwide: water management;
- The chance to join a highly skilled, multidisciplinary, and supportive team;
- An innovative work environment encouraging initiative and professional development.
vorteX-io and CERFACS are committed to fostering an inclusive workplace based on equal opportunities and diversity.
Engineering degree or Research Master’s degree (MSc, Bac+5) in:
- Artificial Intelligence,
- Machine Learning,
- Applied Mathematics,
- Hydrology and Water Sciences,
- Computer Science with specialization in Data Science;
- Eligibility for admission to a doctoral school is required for CIFRE funding.
Technical Skills - Strong proficiency in Python and machine learning/deep learning ecosystems (PyTorch or TensorFlow);
- Experience with time-series analysis and geospatial data processing;
- Knowledge of deep learning architectures for time series (LSTM, Transformers);
- Familiarity with Git and version control;
- Experience with GPU computing environments (HPC or cloud platforms);
- Professional-level English, particularly for scientific writing.
Personal Qualities - Scientific rigor and critical thinking;
- Ability to work independently on long-term research projects;
- Intellectual curiosity for hydrology and environmental sciences;
- Excellent communication skills in both French and English;
- Strong interest in multidisciplinary teamwork.
Additional Assets - Experience or specialization in hydrology, meteorology, or environmental sciences;
- Familiarity with hydrometeorological datasets;
- Knowledge of geospatial formats (NetCDF, GeoTIFF, shapefiles);
- Experience with AI coding assistants (Claude Code, GitHub Copilot, Cursor);
- Prior experience in rainfall–runoff modeling or flood forecasting.
