PCE-PINNs: Physics-Informed Neural Networks for Uncertainty Propagation in Ocean Modeling

2021/05/05 by Björn Lütjens, Lütjens, Björn, Catherine H. Crawford +5 · 1 citation
Computer Science · Earth and Planetary Sciences · Physics and Astronomy · #Computational Physics and Python Applications #FOS: Computer and information sciences #Machine Learning (cs.LG) #Model Reduction and Neural Networks #Seismic Imaging and Inversion Techniques

paper · pdf · doi:10.48550/arxiv.2105.02939

openalex publication_date 2021/05/05 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Climate models project an uncertainty range of possible warming scenarios from 1.5 to 5 degree Celsius global temperature increase until 2100, according to the CMIP6 model ensemble. Climate risk management and infrastructure adaptation requires the accurate quantification of the uncertainties at the local level. Ensembles of high-resolution climate models could accurately quantify the uncertainties, but most physics-based climate models are computationally too expensive to run as ensemble. Recent works in physics-informed neural networks (PINNs) have combined deep learning and the physical sciences to learn up to 15k faster copies of climate submodels. However, the application of PINNs in climate modeling has so far been mostly limited to deterministic models. We leverage a novel method that combines polynomial chaos expansion (PCE), a classic technique for uncertainty propagation, with PINNs. The PCE-PINNs learn a fast surrogate model that is demonstrated for uncertainty propagation of known parameter uncertainties. We showcase the effectiveness in ocean modeling by using the local advection-diffusion equation.

Citations

• Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations
• Embedding Hard Physical Constraints in Neural Network Coarse-Graining of\n 3D Turbulence
• Physics-Informed Generative Adversarial Networks for Stochastic Differential Equations

Cited by

• Deep Polynomial Chaos Expansion

Related

• Graph Convolutional Neural Networks as Surrogate Models for Climate Simulation
• Towards Optimally Weighted Physics-Informed Neural Networks in Ocean Modelling
• The Importance of Architecture Choice in Deep Learning for Climate Applications
• Climate Modeling with Neural Diffusion Equations
• CondensNet: Enabling stable long-term climate simulations via hybrid deep learning models with adaptive physical constraints