Aerosol Absorption from Global Satellite Measurements in the Ultra-Violet: From Qualitative Aerosol Index to Quantitative Aerosol Absorptive Properties

Atmospheric aerosols are solid or liquid particles suspended in the air. The majority of them are produced by natural processes, including sea salt from oceans, mineral dust from (semi-)arid regions, carbon containing particles from wildfires, and sulfates and ash from volcanic activities. Anthropogenic aerosols are produced by industrial activities, power generation, transportation, agriculture, and human-induced biomass burning events. Depending on the meteorological conditions, aerosol particles can stay in the atmosphere for several hours to several months and can be transported over long distances, causing adverse effects on human health, visibility and climate.This thesis focuses on the aerosol optical properties, particularly the light absorption of the aerosol particles that has significant effects on the Earth’s climate system. This thesis starts with a general introduction of atmospheric aerosols, including its sources, categories, physical properties and measurement techniques (Chapter 1). Next, the Ultra-Violet Aerosol Index (UVAI) is introduced, which is calculated from satellite measurements of the radiance at two wavelengths in the UV. UVAI contains information of aerosol absorption, and it has a very long andalmost continuous data record starting in 1978. Direct use of UVAI is challenging because it is not a geophysical quantity, but a numerical index. The objective of this thesis is to derive quantitative properties on aerosol absorption from the UVAI (e.g. single scattering albedo, absorption aerosol optical depth) that can be directly used in aerosol radiative transfer assessments. Two types of methods have been developed, i.e. physically-based methods and statistically-based methods. The first compares the observed UVAI to the one simulated by radiative transfer models. The second uses Machine Learning algorithms trained by existing data sets.The physically-based methods have been applied to quantify aerosol absorption of several large scale wildfires (Chapter 2 and 3). An important challenge of these method is that assumptions have to be made on the aerosol micro-physical properties, leading to significant uncertainties in the results, whereas theMachine Learning-based methods can avoid this kind of assumptions. Chapter 3 investigates the feasibility to quantify aerosol absorption from UVAI using a Machine Learning algorithm. Despite the higher computational efficiency and better results, the application of such data-driven methods is still restricted by the limited data on the aerosol vertical distribution. Therefore, in Chapter4, a database of aerosol height is created from a chemistry transport model. This database is applied in Chapter 5, where a Deep Neural Network method is used to derive the quantitative aerosol absorptive properties from the OMI/Aura UVAI for the period from 2006 to 2019. In comparison to ground-based observations, the results of the Deep Neural Network agree better than satellite retrievals and also better than chemistry transport model simulations.This thesis demonstrates the feasibility of deriving quantitative aerosol absorptive properties from the satellite retrieved UVAI.We use traditional radiative transfer simulations meanwhile investigating the new possibilities of data-driven methods in aerosol remote sensing. Although the retrieval results are encouraging, there remain limitations and challenges which need to be addressed. These are discussed in Chapter 6 with corresponding suggestions and prospects. Despite the challenges, it is expected that a synthetic database of global aerosol absorption can be derived fromUVAI observations provided by multiple satellite products. Such a data set will make great contributions to quantify the effect of absorbing aerosols on the climate system.Atmospheric Remote Sensin