Skylight polarization patterns provide valuable information for atmospheric measurements, polarized remote sensing and navigation applications. However, efficiently and accurately modeling polarized radiative transfer in atmospheric scattering remains challenging. We propose a backward Markov chain Monte Carlo (B-MCMC) method to simulate skylight polarization patterns by constructing a Markov chain in parameter space to track photons from the sensor to the top-of-atmosphere (TOA). The results show that the B-MCMC model significantly improves the computational efficiency by a factor of 8-10 while retaining the computational accuracy compared with Monte Carlo simulations. The experiments show that in cloudy skies, the skylight polarization pattern is generally weakened, in the field of skylight polarization detection and application, long wavelengths should be used in clear weather and blue-violet light should be used in cloudy weather, which corresponds to a larger degree of polarization (DOP) and facilitates the acquisition of polarization information. Finally, the aerosol optical depth (AOD) has an important effect on the skylight polarization, as the AOD increases, the DOP decreases, and the decreasing trend will be more and more obvious, when the AOD is above 0.3, the maximum DOP will not exceed 0.5, which is verified by the division of focal plane (DOFP) polarization measurement device.

天光偏振模式为大气测量、偏振遥感和导航应用提供了有价值的信息。然而，有效、准确地模拟大气散射中的偏振辐射传输仍然具有挑战性。我们提出了一种后向马尔可夫链蒙特卡罗（B-MCMC）方法，通过在参数空间中构建马尔可夫链来跟踪从传感器到大气层顶部（TOA）的光子，来模拟天光偏振模式。结果表明，与蒙特卡洛模拟相比，B-MCMC模型在保持计算精度的同时，计算效率显着提高了8-10倍。实验表明，在多云天气下，天光偏振模式普遍减弱，在天光偏振检测及应用领域，晴天时应采用长波长，阴天时应采用蓝紫光，对应于更大的偏振度（DOP），有利于偏振信息的获取。最后，气溶胶光学深度（AOD）对天光偏振有重要影响，随着AOD的增加，DOP减小，并且减小的趋势会越来越明显，当AOD在0.3以上时，最大DOP不会增加。超过0.5，这是通过焦平面分割（DOFP）偏振测量装置验证的。