Development and evaluation of a shortwave full-spectrum correlated k-distribution radiative transfer algorithm for numerical weather prediction.

摘要

The Full Spectrum Correlated k-distribution (FSCK) method, originally developed for applications in combustion systems, is adapted for use in shortwave atmospheric radiative transfer. By weighting k-distributions by the solar source function, the FSCK method eliminates the requirement that the Planck function be constant over a spectral interval. Consequently, integration may be carried out across the full spectrum as long as the assumption of correlation from one atmospheric level to the next remains valid. Errors resulting from the lack of correlation across the full spectrum are removed by partitioning the spectrum at a wavelength of 0.68 mum into two bands. The resulting two-band approach in the FSCK formalism requires only 15 quadrature points per atmospheric layer. This represents a 40--90% reduction in computation time relative to existing correlated k-distribution models.; The two-band FSCK approach is developed for general atmospheric conditions through the use of tabulated gas k-values, with nongray cloud absorption coefficients added on-the-fly. A two-part evaluation of the FSCK calculations is presented. First, the two-band FSCK results are compared with line-by-line (LBL) benchmarks alongside results from an earlier radiative transfer model intercomparison study. The median of 24 1-D models included in the intercomparison has a clear-sky mean bias error of -0.27 K/day relative to LBL benchmark heating rates, while the operational FSCK model has a mean bias error of +0.04 K/day. In a second set of calculations, two-band FSCK results are compared with those from six popular state-of-the-art operational and research radiative transfer models. The clear-sky RMS heating rate errors for three empirically-based models range from 0.78 to 6.28 K/day, while RMS errors for three correlated k-distribution models range from 0.85 to 2.83 K/day. For the same clear-sky case the FSCK RMS error is 0.57 K/day. Cloudy-sky cases show that the correlated k-distribution models overestimate in-cloud heating, while the FSCK approach with nongray cloud absorption is closer to the benchmark.