Back to Search Start Over

Downscaling shallow water simulations using artificial neural networks and boosted trees

Authors :
Bakong, Killian
Guinot, Vincent
Rousseau, Antoine
Toulemonde, Gwladys
Littoral, Environment: MOdels and Numerics (LEMON)
Inria Sophia Antipolis - Méditerranée (CRISAM)
Institut National de Recherche en Informatique et en Automatique (Inria)-Institut National de Recherche en Informatique et en Automatique (Inria)-Institut Montpelliérain Alexander Grothendieck (IMAG)
Centre National de la Recherche Scientifique (CNRS)-Université de Montpellier (UM)-Centre National de la Recherche Scientifique (CNRS)-Université de Montpellier (UM)-Hydrosciences Montpellier (HSM)
Institut de Recherche pour le Développement (IRD)-Institut national des sciences de l'Univers (INSU - CNRS)-Centre National de la Recherche Scientifique (CNRS)-Université de Montpellier (UM)-Institut de Recherche pour le Développement (IRD)-Institut national des sciences de l'Univers (INSU - CNRS)-Centre National de la Recherche Scientifique (CNRS)
Hydrosciences Montpellier (HSM)
Institut de Recherche pour le Développement (IRD)-Institut national des sciences de l'Univers (INSU - CNRS)-Centre National de la Recherche Scientifique (CNRS)-Université de Montpellier (UM)
Institut Montpelliérain Alexander Grothendieck (IMAG)
Centre National de la Recherche Scientifique (CNRS)-Université de Montpellier (UM)
Projet SB / IRT Saint-Exupéry
Source :
Discrete and Continuous Dynamical Systems-Series S, Discrete and Continuous Dynamical Systems-Series S, In press, ⟨10.3934/dcdss.2022198⟩
Publication Year :
2022
Publisher :
HAL CCSD, 2022.

Abstract

International audience; We present the application of two statistical artificial intelligence tools for multi-scale shallow water simulations. Artificial neural networks (ANNs) and boosted trees (BTs) are used to model the relationship between low-resolution (LR) and high-resolution (HR) information derived from simulations provided in the learning phase. The two statistical models are analyzed (and compared) through hyper-parameters such as the number of epochs and the network structure for ANNs, and the learning rate, tree depth and number for BTs. This analysis is performed through 4 numerical experiments the input datasets of which (for the learning, validation and test phases) are varied through the boundary conditions of the flow numerical simulation. The performance of the ANNs is remarkably consistent, regardless of the choice made for the training/validation/testing set. The performance improves with the number of epochs and the number of neurons. For a given number of neurons, a single-layer structure performs better than multi-layer structures. BTs perform significantly better than ANNs in 2 experiments, with an error 10 to 100 times lower and a computational cost 5 to 10 times larger). However, when the validation datasets differ from the training datasets, the performance of BTs performance is strongly degraded, with a modelling error more than one order of magnitude larger than that of ANNs. Used in conjunction with upscaled flood models such as porosity models, these techniques appear as a promising operational alternative to direct flood hazard assessment from HR flow simulations.

Details

Language :
English
ISSN :
19371632 and 19371179
Database :
OpenAIRE
Journal :
Discrete and Continuous Dynamical Systems-Series S, Discrete and Continuous Dynamical Systems-Series S, In press, ⟨10.3934/dcdss.2022198⟩
Accession number :
edsair.dedup.wf.001..79c40a65a210f6267b88dc85d86b9374
Full Text :
https://doi.org/10.3934/dcdss.2022198⟩