Rain nowcasting with Radar reflectivity maps

Encadrants : Dominique Béréziat
Disponible : OUI

Spécialité : IMA/DIGIT
Nombre d'étudiants : 4
Description : Rainfall forecasting is an important environmental issue (flood anticipation, agricultural management, …). Here we are interested in a short-term forecast (time horizon up to 2 hours). We use Radar data which are robust rainfall indicators. The algorithm will be based on deep learning approaches with an up-to-date architecture.
Pré-requis : pytorch
Travail demandé : - review literature. In particular, investigate the field of video prediction, and also relevant metrics, such as SSIM - choose one or two models, implement, train, and study the performance The supervisor will provide a series of recent papers.