Aerosols are the source of the largest uncertainties in our climate models, blurring our outlook of the future. This has been attributed to the complexity of measuring their properties, which vary over time and space. Atmospheric circulation spreads aerosols across the globe from a point source, which makes satellite-based observations lucrative. At present, there are several aerosol observing missions that deliver aerosol data products in a consistent and operational manner; these missions report several aerosol properties that are important for reducing the contribution of uncertainties to our climate models. What is missing, however, is an operational data product that measures the height of these aerosols at a global scale. Earlier attempts at this use data derived from lidar instruments in space; an example being the Cloud-Aerosol Lidar with Orthogonal Polarisation (CALIOP) instrument, which uses lasers to measure atmospheric composition. In the case of aerosols, the amount of backscattered electromagnetic radiation at each atmospheric layer gives an idea of the amount and height of aerosols. The mobility afforded by space-based instruments gives space lidars a leg up over ground-based lidars. However, the coverage of such lidar instruments is merely near-global. This has to do with the fact that while lidars in space can circle the entire globe, their footprint on the ground is very narrow, in the order of several hundred meters to a few kilometers: this is an inherent limitation of the measurement principle. Consequently, a specific patch on Earth is revisited in periods that can range several days. An alternative to space based atmospheric lidars are space based spectral imagers. These are essentially cameras that take snapshots of the Earth, capturing the light and splitting its different electromagnetic frequencies into the scale of nanometers using very precise prisms and detection techniques. The advantage of these instruments over lidars is that they have a very large footprint, covering several thousand kilometers of area in a single _y-by. This allows for daily to even sub-daily coverage of the Earth, as each snapshot covers larger and sometimes overlapping areas. The challenge is to estimate aerosol height using spectral signatures of the Earth’s atmosphere in an operational environment that can handle data coming in from the satellite at a rate of several million pixels every few minutes. This dissertation focuses on delivering the aerosol height data product operationally using computer algorithms. The logic of aerosol height estimation using these so-called spectral snapshots of the atmosphere differ from that using lidars; the instrument does not provide data for different atmospheric layers. This has to be inferred using the chemistry of the oxygen molecule. O2, the second most abundant gas in our atmosphere, has a unique spectral signature in the near-infrared region, comprising of electromagnetic radiation around 765 nm. The chemical structure of the oxygen molecule allows it to absorb some of these radiations, creating a structure of absorption bands. This spectral signature deepens as more light is absorbed by the oxygen: this happens as photons penetrate deeper and deeper into the earth’s atmosphere, unless they hit a barrier. If the photons bounce back from an aerosol layer at a very high altitude, the amount of absorption by oxygen would be low. This ‘depth’ of absorption gives clues on how high an aerosol layer might be present. Computer models can reconstruct this oxygen absorption structure onto a simulated spectrum. One of the control parameters within the model is the height of an aerosol layer. The generated spectral signature of a simulated atmosphere resembling the atmosphere of a pixel in the snapshot from space-based hyperspectral imagers is then compared to the measured spectral signature. This usually results in a non-zero difference, which is caused by errors in the model. These errors can be minimised by using computer algorithms and mathematical information retrieval techniques, resulting in a modeled atmosphere closer to the measurement by changing the height of the aerosol layer, resulting in an aerosol height estimate. In this dissertation, computer algorithms inspired from mathematical models of brain neural networks as well as information retrieval techniques such as least squares are used…@en

The Aerosol Optical Depth (AOD), a measure of the scattering and absorption of light by aerosols, has been extensively used for scientific research such as monitoring air quality near the surface due to fine particles aggregated, aerosol radiative forcing (cooling effect against the warming effect by carbon dioxide CO2 ), aerosol long-term trend analysis and the climate change on regional and global scale. Aerosols vary greatly over time and space. This is because of the short lifetime of aerosols (a few hours to a week), and also because of the heterogeneous distribution of sources and the variable effectiveness of atmospheric mixing though turbulence. To monitor aerosols, observations by space-borne instruments have a huge advantage (nearly global coverage daily) over ground-based measurements (point observation). Global quantitative aerosol information has been derived from satellite measurements for decades. The MODerate resolution Imaging Sepctroradiometer (MODIS) AOD product is proven to be mature and is extensively applied in different scientific fields. The current AOD product generated with the collection 6 (C6) Dark Target (C6_DT) algorithm over land is still suffering from errors or biases due to parameterization, assumptions, modeling, and retrieval techniques as well as ill-posed problems, presenting large uncertainties, including regional bias, angular effects and a large number of unphysical negative values. Chapter 1 discusses the challenges and limitations in the current satellite aerosol retrieval algorithm. Owing to the use of static aerosol properties (predefined aerosol models and fixed vertical profile over the globe), the MODIS algorithm may give serious errors since aerosols can change over time and are distributed very diversely at different altitude levels. To quantify these errors, in Chapter 3 the sensitivity of AOD retrieval to the variation of aerosol vertical profiles and types with the MODIS algorithm is evaluated by a set of experiments. It was found that the AOD retrieval shows a high sensitivity to different vertical profiles and types. As suggested by the sensitivity study, it is necessary to investigate the impact of dynamical aerosol properties in a real case. To do this, an adaptive development of the MODIS C6_DT algorithm was implemented to consider realistic aerosol vertical profile in the retrieval (Chapter 4). MODIS and Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) measurements were used. Inferred from CALIPSO data, the vertical profile was applied into the new algorithm to generate an accurate Top Of the Atmosphere (TOA) reflectance for the retrieval. The AOD retrieval was compared between C6_DT and the new algorithm with cases of heavy smoke and dust. The difference in the retrieval was significant between C6_DT and the new algorithm, which demonstrated that C6_DT would give large errors in the retrieval for these cases. In the MODIS algorithm, the assumption of the surface with isotropic reflection (Lambertian) is inconsistent with the well-known fact that the surface has a strong anisotropic reflection (non-Lambertian), and could lead to large uncertainties in estimating the surface contribution to satellite measurements, with resulting errors in the AOD retrieval. Chapter 5 describes a newly developed algorithm (BRF_DT) by considering non-Lambertian surface reflectance characterized by Bidirectional Distribution Reflectance Function (BRDF), where the surface reflection is described by four reflectance properties — bidirectional, directional-hemispherical, hemispherical-directional, and bihemispherical reflectance and coupled into the radiative transfer process to generate an accurate TOA reflectance. In addition, a parameterization of spectral relationship inherited from C6_DT was applied to constrain the surface BRF. The remaining three components are determined by MODIS BRDF/albedo product. As shown by sample plots and histograms as well as analysis and comparison against AERONET measurements, the AOD retrievals were significantly improved by BRF_DT especially for areas with heavy aerosol loading. For the case of areas with light aerosol loading, the parameterization of spectral surface BRF should be further refined to yield a better retrieval. Chapter 6 shows that a new parameterization was derived for the BRF_DT algorithm (called BRF_DT2) by using 3 years of BRF data from AERONET-based Surface Reflectance Validation Network (AS-RVN). The contribution to the TOA reflectance dominated by the surface BRF was well estimated. As a result, negative retrievals and angular biases were significantly reduced in BRF_DT2. A summary of the current and future research of satellite aerosol retrieval is introduced in Chapter 7. @en