Emulation to simulate low-resolution atmospheric data

Abstract

Climate model development, testing, and analysis involve running the model extensively to tune the subgrid- scale parameters that provide closure to the system. This process demands substantial time and computational resources even for typical spatial resolutions and becomes in feasibly expensive for high-resolution studies. This paper presents alternative, computationally feasible methods to emulate the simulations within acceptable error bounds. This strategy can be easily implemented to obtain an ensemble of model runs. The paper outlines three approximation strategies: (1) interpolation with Lagrange basis functions, (2) least-squares (LS) approximation, and (3) interpolation with radial basis functions. As a proof of concept, a suite of relevant physical quantities are evaluated at unknown grid points of parameters, space, and time. The values obtained by emulation are compared against the simulated values to check the feasibility of the method. The advantages and shortcomings of the above-mentioned approximation schemes are discussed, including the savings of time and computational resources.

Keywords:

• climate modelling
• low-resolution data sets
• stochastic collocation method
• interpolation
• radial basis functions

2000 AMS Subject Classifications::

• 41A10
• 47N50
• 65C30

Acknowledgements

This research was funded through the US Department of Energy Office of Biological and Environmental Research (BER) project No. 3ERKPE746, ‘Ultra High Resolution Global Climate Simulation to Explore and Quantify Predictive Skill for Climate Means,Variability and Extremes.’ This research used resources of the Oak Ridge Leadership Computing Facility at the Oak Ridge National Laboratory, which is supported by the Office of Science of the US Department of Energy under Contract No. DEAC05-00OR22725. The authors would also like to thank Dr Adrian Sandu from Virginia Tech. for all the support and useful feedback.