典型矢量辐射传输模式计算精度与效率的初步比较

Preliminary comparisons of the typical polarized radiative transfer models: precision and efficiency

辐射传输模式是建立遥感反演方法和气候模式中辐射参数化的重要工具,尤其是全偏振的矢量辐射传输模式对于精确理解地气系统中的辐射过程至关重要。PolRadtran/RT3(polarized radiative transfer)、SOSVRT(vector radiative transfer based on successive order of scattering)和VDISORT(vector DIScrete ordinate radiative transfer)是基于不同物理原理求解矢量辐射传输的三个代表性数值模式。对这三个模式进行计算时间和计算精度的比较,发现,基于逐次散射法的SOSVRT计算效率最高,计算时间基本不随流数的增加而增长,但随单层光学厚度的增大,其计算时间有较为明显的增加,在米散射情况下,光学厚度从0.5增加到1.0时,其计算时间增加了1倍;基于倍加累加法的RT3和基于矩阵特征矢量求解方法的VDISORT计算效率较低,尤其是采用大流数计算时,RT3和VDISORT的计算时间随流数的增加迅速增长,特别是在瑞利散射条件下,波长为400nm,流数为40时,其计算时间分别为SOSVRT的23倍和7倍。但是,两模式随光学厚度增加计算时间却无明显的增加。在计算精度方面,3个模式比较接近,只是VDISORT在大流数的情况下会有震荡现象。

Radiative transfer models are key tools in the remote sensing and parameterization of climate radiative forcing, while polarized radiative transfer models can provide more accurate insights into the radiation processes in the earth–atmosphere system. PolRadtran/RT3 （Polarized Radiative Transfer, based on the adding-doubling method）, SOSVRT （Vector Radiative Transfer, based on successive order of scattering）, and VDISORT （Vector DIScrete Ordinate Radiative Transfer, a polarized version of DISORT based on the inverse of matrix method）, are three of the most common radiative transfer models, each with polarization based on different physical principles. A comparison of their accuracy and efficiency reveals that SOSVRT is the most efficient, with the time consumed remaining almost invariable with the increase of stream numbers, but increasing with the optical depth of the layered atmosphere. For example, the time consumed for an optical depth of 1.0 was found to be two times that for an optical depth of 0.5 for the Mie scattering atmosphere. The efficiencies of RT3 and VDISORT in modeling polariza-tion with a large stream number were found to be low. For example, under the Rayleigh scattering atmosphere at 400nm and a stream number of 40, the time consumed was 23 times and 7 times as much as that of SOSVRT, respectively. The computation time for the two models was found not to be sensitive to the optical depth, but increased greatly with the increase in stream number. All three models were found to be of the same order of accuracy, but VDISORT showed a fluctuating result for simula- tions with large streams.