DEVELOPMENT OF A CORRELATION-BASED INVERSION METHOD FOR AEROSOL REMOTE SENSING

Abstract

Accurate characterization of aerosol microphysical properties from diverse remote sensing platforms is essential for quantifying aerosol-radiation and aerosol-cloud interactions. However, aerosol retrieval remains a highly ill-posed inversion problem, particularly when operating under limited information content. To address this challenge, this dissertation introduces two distinct applications that improve state-of-the-art retrieval techniques by incorporating well-characterized a priori constraints. The first application develops a surface model parameter conversion algorithm that bridges a priori information from semi-empirical, Ross–Li-based surface model parameters to physical, Rahman–Pinty–Verstraete (RPV)-based surface parameters. Validated through Directional Hemispherical Reflectance (DHR) and Bidirectional Reflectance Factor (BRF) comparisons, the algorithm demonstrates a highly successful conversion capability. Further retrieval tests indicate that the conversion algorithm introduces minimal impact to the aerosol inversion, providing a robust mechanism for surface a priori initialization. The second application presents the Correlation based Inversion Method for Aerosol Properties (CIMAP) framework, an aerosol inversion strategy that utilizes Principal Component Analysis (PCA) to implement intrinsic correlation constraints among aerosol properties. By retaining a targeted number of Principal Components (PCs), CIMAP substantially reduces the state vector dimensionality, enhancing both numerical stability and computational efficiency. The framework is evaluated under two configurations: CIMAP-A: Utilizes ground-based AERONET observations across three distinct regions (southeast coast of the US, northern China, and southern Africa). Using only 7–8 PCs out of 30, CIMAP-A achieves a processing speed acceleration of over 80% while maintaining high accuracy relative to official AERONET inversion products, yielding Mean Absolute Difference (MAD) of 0.005, 0.019, 0.039, 0.003, 0.013 μm, 0.02 μm and for Aerosol Optical Depth (AOD), Single Scattering Albedo (SSA), Real and Imaginary parts of Refractive Index (RIR and RII), effective radius of fine and coarse modes of particle size distribution (PSD), respectively. CIMAP-P: Applies the framework to multi-angle, multi-spectral, space-borne polarimetric observations using POLDER data, integrating the surface conversion algorithm. Validation against collocated AERONET references shows strong agreement, with AOD MAD ranging from 0.015 to 0.042 and SSA MAD from 0.048 to 0.070 across reference wavelengths and 0.032 μm and 0.402 μm for the effective radius of fine- and coarse-mode PSD, exhibiting performance competitive with multiple Generalized Retrieval of Aerosol and Surface Properties (GRASP) algorithm configurations. Limitations of the CIMAP configurations are discussed, alongside potential expansions toward globally covering observations and possible improvements to constraint construction. Ultimately, the flexible and computationally efficient architecture of CIMAP, with its surface initialization capability, positions it as a highly promising inversion framework for maximizing the rich information content offered by next-generation polarimetric missions.