ABSTRACT
Reservoir sedimentation is a critical, ongoing issue in managing water resources sustainably. While conventional two-dimensional models are computationally efficient, they miss key three-dimensional processes, such as thermal stratification. Three-dimensional models provide a more accurate physical representation but require extensive computational resources, making them impractical for large-scale applications. This research creates a computational framework that combines High-Performance Computing, Artificial Intelligence, and advanced 3D multiphysics simulation to bridge this gap.
A two-dimensional hydro-morphodynamic model (R-Iber) was rebuilt for Graphics Processing Units, resulting in computational speed-ups of one to two orders of magnitude. The accelerated model supported training a Deep Neural Network surrogate, enabling a 100,000-run Monte Carlo analysis for robust model calibration and uncertainty quantification. In parallel, a comprehensive three-dimensional multiphysics model was developed in the Kratos framework to simulate the 3D fluid-thermal problem.
The integrated approach was used for the Riba-roja reservoir system. It measured how thermal stratification affects sediment trapping efficiency. Results show that combining HPC, AI, and multiphysics modeling leads to practical and actionable methods for sustainable reservoir management.
PhD Advisors:
- Prof Ernest Bladé
- Prof Antonia Larese
PHD CANDIDATE
Mr Danial Dehghan is a PhD candidate in Civil Engineering at CIMNE’s Machine Learning and Models in Hydro-Environmental Engineering research cluster.