A5088300736- Meteorological Phenomena and Simulations
- Numerical methods for differential equations
- Advanced Numerical Methods in Computational Mathematics
- Differential Equations and Numerical Methods
- Climate variability and models
- Atmospheric and Environmental Gas Dynamics
- Electromagnetic Simulation and Numerical Methods
- Differential Equations and Boundary Problems
- Solar Radiation and Photovoltaics
- Model Reduction and Neural Networks
- Electromagnetic Scattering and Analysis
Columbia University
2025
ETH Zurich
2023-2024
University of Geneva
2019
As climate modellers prepare their code for kilometre-scale global simulations, the computationally demanding radiative transfer parameterization is a prime candidate machine learning (ML) emulation. Because of computational demands, many weather centres use reduced spatial grid and temporal frequency calculations in forecast models. This strategy known to affect quality, which further motivates ML-based parameterizations. paper contributes discussion on how incorporate physical constraints...
The Strang splitting method, formally of order two, can suffer from reduction when applied to semilinear parabolic problems with inhomogeneous boundary conditions. recent work [L. Einkemmer and A. Ostermann, SIAM J. Sci. Comput., 37, 2015; 38, 2016] introduces a modification the method avoid based on nonlinearity. In this paper we introduce new correction constructed directly flow nonlinearity which requires no evaluation source term or its derivatives. goal is twofold. One, only one...
This paper continues the exploration of \gls{ml} parameterization for radiative transfer \gls{icon}. Three models, developed in Part I this study, are coupled to More specifically, a UNet model and bidirectional \gls{rnn} with \gls{lstm} compared against random forest. The parameterizations \gls{icon} code that includes OpenACC compiler directives enable \glspl{gpu} support. coupling is done through Infero, by ECMWF, PyTorch-Fortran. most accurate physics-informed normalization strategy...
As climate modellers prepare their code for kilometre-scale global simulations, the computationally demanding radiative transfer parameterization is a prime candidate machine learning (ML) emulation. Because of computational demands, many weather centres use reduced spatial grid and temporal frequency calculations in forecast models. This strategy known to affect quality, which further motivates ML-based parameterizations. paper contributes discussion on how incorporate physical constraints...
This paper continues the exploration of Machine Learning (ML) parameterization for radiative transfer ICOsahedral Nonhydrostatic weather and climate model (ICON). Three ML models, developed in Part I this study, are coupled to ICON. More specifically, a UNet bidirectional recurrent neural network (RNN) with long short-term memory (LSTM) compared against random forest. The parameterizations ICON code that includes OpenACC compiler directives enable GPUs support. coupling is done through...
We show that the Strang splitting method applied to a diffusion-reaction equation with inhomogeneous general oblique boundary conditions is of order two when diffusion solved Crank-Nicolson method, while reduction occurs in if using other Runge-Kutta schemes or even exact flow itself for part. prove these results source term only depends on space variable, an assumption which makes scheme equivalent whole problem. Numerical experiments suggest second convergence persists nonlinearities.
Modelling the transfer of radiation through atmosphere is a key component weather and climate models. The operational scheme in Icosahedral Nonhydrostatic Weather Climate Model (ICON) ecRad. ecRad accurate but computationally expensive. It operationally run ICON on grid coarser than dynamical time step interval between two calls significantly larger. This known to reduce quality prediction. A possible approach accelerate computation fluxes use machine learning methods. Machine methods can...
The Strang splitting method, formally of order two, can suffer from reduction when applied to semilinear parabolic problems with inhomogeneous boundary conditions. recent work [L .Einkemmer and A. Ostermann. Overcoming in diffusion-reaction splitting. Part 1. Dirichlet SIAM J. Sci. Comput., 37, 2015. 2: Oblique conditions, 38, 2016] introduces a modification the method avoid based on nonlinearity. In this paper we introduce new correction constructed directly flow nonlinearity which requires...
<p>Atmospheric radiative transfer, which describes the evolution of radiation emitted by Sun, Earth's surface, clouds, and greenhouse gases, is an essential component climate weather modeling. In models, transfer approximated parameterizations. Theoretically, however, with sufficient computing power, electromagnetic equations could be solved, but in practice this out reach.  The current operational solver Icosahedral Nonhydrostatic Weather Climate Model (ICON)...