קולוקוויום בחוג לגאופיזיקה: Utilization of machine learning techniques to retrieve aerosol and cloud properties from remote sensing measurements
Michal Segal-Rozenhaimer, TAU
Machine learning is a growing field that deals with huge amounts of data in order to create various inferences. Its applicability ranges from facial recognition, self-driving cars and natural language processing. The amount of data that is inherent to remote sensing measurements (spatial and spectral) makes it suitable for learning inferences by such techniques. In this talk I will focus on two techniques, namely neural-networks (NN) and convolutional NN (CNN), and will show their applicability in the retrieval of cloud and aerosol properties from two different types of remote sensing platforms. The first part of the talk will cover the development process of a cloud and above cloud aerosol algorithms designed specifically for observations over the South-East Atlantic Ocean, made by the Research Scanning Polarimeter (RSP), a highly accurate multi-angle polarimetric scanner, which was flown during the NASA ORACLES (ObseRvations of Aerosols above CLouds and their intEractionS) campaign in 2016-2017 and 2018. The second part of the talk will focus on utilization of CNNs to derive cloud and cloud shadow masks for the high-resolution multi-spectral WorldView-2 and Sentinel-2 satellite platforms. In this part I will cover the advantages of the new algorithm over the existing ones and its ability to be relatively transferable between the two different platforms, a topic that is important given the plethora of new platforms and the need to unify their observations.
מארגני האירוע: ד"ר רועי ברקן וד"ר אסף ענבל