Job offer

Historically, space-borne aerosol observation initially focused on dark surfaces, such as oceans. It subsequently expanded to brighter land surfaces (e.g., the GRASP algorithm, Dubovik et al., 2011), but only under cloud-free conditions. The remote sensing of aerosols above clouds is a recent development (Waquet et al., 2009). The reference product AERO-AC, available through ICARE, was designed for global use. It was developed using data from the POLDER/PARASOL polarimeter, complemented by specific data from the MODIS instrument. AERO-AC provides key properties of aerosols detected above clouds, such as optical depth, the Ångström exponent (related to particle size), and particle absorption (Waquet et al., 2020). However, restrictions remain regarding the properties of the clouds located beneath the aerosol layers. Retrievals are currently only available for optically thick liquid water clouds that exhibit spatially homogeneous properties and 100% fractional coverage.

Current limitations restrict our ability to monitor aerosol properties within fractional cloud cover, which is nevertheless very frequent on a global scale. This constraint also reduces our capacity to track pollution events, particularly in France and Europe, and global extreme phenomena. This limitation is mainly due to the use of "plane-parallel" radiative transfer codes (1D codes). These codes do not account for the three-dimensional structure of clouds or the spatial variability of their properties. While more accurate, the use of three-dimensional (3D) radiative transfer codes is currently limited by significant computational time. Presently, even operational 1D algorithms rely on Look-Up Tables (LUTs) and a limited number of aerosol models to maintain manageable processing speeds. Artificial Intelligence (AI) methods are well-suited for studying aerosols in complex cloudy scenes. They facilitate the integration of multiple parameters (aerosols, clouds, altitudes, etc.), significantly reduce calculation times, and, in the case of Convolutional Neural Networks (CNNs), for example, allow for a more efficient exploitation of satellite imagery. This research project will focus on developing an AI-based algorithm specifically for studying aerosols within both total and fractional cloud cover. The training phase will be based on radiative transfer calculations performed with the 1D and 3D codes available at the LOA (Laboratoire d'Optique Atmosphérique). These simulations will rely on realistic, heterogeneous cloudy scenes containing aerosols. The objective of this work is to develop the neural network architecture tailored for measurements from ESA's future 3MI instrument, recently launched in Autumn 2025. This instrument combines the advantages of POLDER and MODIS in terms of spectral, directional, and polarized data. The research will compare 1D and 3D methods to propose a globally applicable algorithm. This algorithm will process aerosols above clouds as well as complex fractional cloud cover. Over land surfaces, it will incorporate surface reflectance databases produced by the GRASP algorithm. Ultimately, this work will enable better global monitoring of aerosols and a deeper understanding of aerosol-cloud interactions.

PhD or equivalent

Research Field Environmental science

Education Level PhD or equivalent

Research Field Environmental science