Downscaling Using CDAnet Under Observational and Model Noises: The Rayleigh-Benard Convection Paradigm

• DOI: 10.48550/arxiv.2310.11945
• Publication Date: 2023-01-01
• Authors: Mohamad Abed El R...; Edriss S. Titi; Ibrahim Hoteit; Omar Knio
• Keywords: Computational Engineering, Finance, and Science (cs.CE); FOS: Computer and information sciences; FOS: Mathematics; 37 Earth Sciences; 3704 Geoinformatics; Dynamical Systems (math.DS); Mathematics - Dynamical Systems; Computer Science - Computational Engineering, Finance, and Science

Efficient downscaling of large ensembles coarse-scale information is crucial in several applications, such as oceanic and atmospheric modeling. The determining form map a theoretical lifting function from the low-resolution solution trajectories dissipative dynamical system to their corresponding fine-scale counterparts. Recently, physics-informed deep neural network ("CDAnet") was introduced, providing surrogate for efficient downscaling. CDAnet demonstrated efficiently downscale noise-free data deterministic setting. Herein, performance well-trained models analyzed stochastic setting involving (i) observational noise, (ii) model (iii) combination noises. analysis performed employing Rayleigh-Benard convection paradigm, under three training conditions, namely, with perfect, noisy, or downscaled data. Furthermore, effects noises, Rayleigh number, spatial temporal resolutions input on fields are examined. results suggest that expected l2-error behaves quadratically terms standard deviations also responds uncertainties similar theorized numerically-validated CDA behavior an additional error overhead due being map.