From Model-Based Parameterizations to Lookup Tables: An EOF Approach

Abstract

Abstract The goal of this study is to transform the Harrington radiation parameterization into a transfer scheme or lookup table, which provides essentially the same output (heating rate profile and short- and longwave fluxes at the surface) at a fraction of the computational cost. The methodology put forth here does not introduce a new parameterization simply derived from the Harrington scheme but, rather, shows that given a generic parameterization it is possible to build an algorithm, largely not based on the physics, that mimics the outcome of the parent parameterization. The core concept is to compute the empirical orthogonal functions (EOFs) of all of the input variables of the parent scheme, run the scheme on the EOFs, and express the output of a generic input sounding exploiting the input–output pairs associated with the EOFs. The weights are based on the difference between the input and EOFs water vapor mixing ratios. A detailed overview of the algorithm and the development of a few transfer schemes are also presented. Results show very good agreement (r > 0.91) between the different transfer schemes and the Harrington radiation parameterization with a very significant reduction in computational cost (at least 95%).