Start of funding 01.07.2024

A Deep-Learning Approach to 3D Cloud Remote Sensing

Prof. Dr. Bernhard Mayer
Ludwig-Maximilians-University of Munich
Lehrstuhl für Experimentelle Meteorologie

Dr. Linda Forster

Near Earth Tracking Systems and Applications



This project introduces a novel deep-learning (DL) approach for cloud remote sensing, aimed at improving climate predictions. Traditional methods often yield biased results due to oversimplified cloud models. The new DL model, which is based on a Convolutional Neural Network (CNN) architecture, predicts cloud properties from multi-angle satellite imaging data, taking into account the three-dimensional nature of clouds. This method enhances accuracy, provides uncertainty estimates, and saves computational effort. Integrating these predictions into a tomographic reconstruction process improves the efficiency and scalability of cloud observation, setting a new standard in atmospheric remote sensing.