Use of a neural network-based longwave radiative transfer scheme in the ECMWFatmospheric model

TitleUse of a neural network-based longwave radiative transfer scheme in the ECMWFatmospheric model
Publication TypeMiscellaneous
Year of Publication1999
AuthorsChevallier, F, Morcrette, J-J, Chéruy, F, Scott, NA
Secondary TitleTechnical Memorandum
Number276
Pagination21
Date PublishedMarch
PublisherECMWF
Place PublishedShinfield Park, Reading
Type of WorkTechnical Memorandum
Abstract

Although important uncertainties remain concerning the far wing absorbing line shapes and the effect of clouds, a high standard of accuracy has been achieved by the scattering line-by-line models for the modeling of the LW radiative transfer (e.g. Moncet and Clough, 1997). however, the definition of an approach that would enable computation times suitable for climate studies and a satisfactory accuracy, has proven to be a challenge for modellers. A fast radiative transfer model is tested at ECMWF: NeuroFlux (Cheruy et al., 1996; Chevallier et al, 1998b). It is based on an artificial neural network technique (the Multi-Layer Perceptron: Rumelhart et al., 1986) used in conjunction with a classical cloud approximation (the multilayer grey body model: Washington and Williamson, 1977). The accuracy of the method is assessed through code-by-code comparisons, climate simulations and ten-day forecasts with the ECMWF model.