Projectes de transferència

Recent work shows the added value of scatterometer sea surface wind data for correcting numerical weather prediction (NWP) model output local biases (due to several geophysical processes unresolved by the model). The rationale of this method is that when the scatterometer wind data are accumulated over short periods of time, it is possible to overcome sampling errors and maintain some of the scatterometers most beneficial features. The so-called scatterometer corrections (SC) are based on temporal windows of a few days centered around the NWP forecast time. A particularly concerning problem in NWP is the well identified wind vector biases linked to atmospheric stability effects, moist convection, ocean currents, etc. The aim of this work it to employ deep learning methods built upon NWP forecast fields of ocean vector winds and associated ocean surface and atmosphere parameters to predict SC, leading to corrected and hence better-quality NWP surface wind and stress fields for both atmospheric and oceanic applications. These SC may for example be employed in data assimilation, seasonal forecasting and of course in model parameterization studies. A preliminary development was successfully carried out in the context of the OSI_VSA22_01 study. A follow-on work is now proposed, aiming at further developing and enhancing the derived ML models.