Accelerating Radiative Transfer Calculations in the Thermal Infrared Through Principal Component Analysis of Inherent Optical Properties

Principal component analysis (PCA) is a linear mathematical transformation that reduces data dimensionality and enables easy identification of data variability; PCA converts a correlated mean-subtracted data set into a series of principal components (PCs). PCA can be performed on inherent optical properties to enhance the computational efficiency of radiative transfer calculations. Prior work has demonstrated the efficacy of this technique for the computation of radiances and Jacobians in the solar scattering regime, and for estimating broadband fluxes. Here, we extend the methodology to the thermal emission regime. We perform simulations for a nominal (model-based) aerosol scattering scenario, and for a scenario with enhanced aerosol scattering that is representative of strong pollution. The PCA-based approach provides top-of-the-atmosphere radiance results that are typically within 0.1% of numerically exact computations, while enabling a three-orders-of-magnitude speedup. We also investigate the accuracy levels achieved using variable numbers of PCs for different gas absorption optical depth regimes. Results show that more PCs are required for optical depths close to unity, while low and high optical depth scenarios can be simulated accurately with only one PC.

主成分分析（Principal Component Analysis，PCA）是一种线性数学变换，可降低数据维度并便于辨识数据的变异性；该方法可将一组经去均值处理且存在相关性的数据集转换为一系列主成分（PCs）。主成分分析可应用于固有光学特性，以提升辐射传输计算的计算效率。已有研究表明，该技术在太阳散射场景下的辐射亮度与雅可比（Jacobian）计算、宽带通量估算中均表现出良好效果。本研究将该方法拓展至热发射场景。我们分别针对标准（基于模型的）气溶胶散射场景，以及代表强污染情形的增强型气溶胶散射场景开展了模拟实验。基于主成分分析的方法所得到的大气顶（Top-of-the-Atmosphere）辐射亮度结果，通常与数值精确计算结果的偏差在0.1%以内，同时可实现三个数量级的计算加速。我们还探究了针对不同气体吸收光学厚度场景，使用不同数量的主成分（PCs）所能达到的精度水平。研究结果表明，光学厚度接近1的场景需要使用更多的主成分，而低光学厚度与高光学厚度场景仅需单个主成分即可实现高精度模拟。