PhD title : Tropical Cyclone classification and prediction using multimodal Physics-informed Artificial Intelligence methods
Supervisors : Alexis Mouche, Pierre Tandeo, Bertrand Chapron, Ronan Fablet, John Knaff
Funding: IFREMER, AI Chair OceaniX
Start date : 15/03/2021
Satellite acquisitions are used to study Tropical Cyclones (TCs) since the 1960s. Today, satellite observations of TCs comprise both high spatial resolution data with low temporal resolution (such as SAR imagery) and low spatial resolution data with a better temporal resolution (such as radiometer and scatterometer data). Thus, the idea of the PhD is to exploit both the high spatial resolution of the SAR data (though at low temporal resolution) and the good temporal resolution of radiometer/scatterometer (though at low spatial resolution) to better describe the TC structure. More precisely, the objective is to use high spatial resolution satellite images (SAR imagery) of TCs to "enhance" low spatial resolution images (radiometer, scatterometer,…) so that these “enhanced” low spatial resolution images alone (i.e without the help of high spatial resolution acquisitions) allow to retrieve the TC dynamics, i.e changes in the TC structure (stretching/contraction of the Radius of Maximum Winds, axisymmetrization) as well as changes in intensity (intensification/weakening).
In order to exploit large sets of multimodal data (i.e using satellite acquisitions of TCs with different spatial and temporal resolution), the thesis will rely on Artificial Intelligence (AI) methods combined with dynamical/physical considerations, referred to as "Physics-Informed AI".