Research fellowship within the field of machine learning within data assimilation and forecasting systems
EUMETSAT & Danish Meteorological Institute, Copenhagen, Denmark
Closing date: 10 December 2023
EUMETSAT is now inviting applications from suitably qualified data scientists from its Member States for a Research Fellowship on an innovative and collaborative project about the potential benefits of machine learning (ML) within data assimilation (DA) and forecasting systems in Arctic regions. The Research Fellow will join the Numerical Weather Prediction (NWP) department at the Danish Meteorological Institute (DMI) and, as expected by the joint fellowship program, He/She will also collaborate with scientists from the Norwegian Meteorological Institute (Met Norway).
In recent years, there has been increased interest in producing more accurate weather predictions for Polar regions. To do so, due to the lack of conventional observations in these remote areas, NWP/DA systems have to significantly improve the use of satellite observations to produce the best estimate of the state of the atmosphere and the surface.
Within NWP/DA systems, the direct assimilation of microwave radiances is managed by a physical-based forward model (also named as ‘observation operator’) that is necessary to simulate the satellite observations. If the simulated and observed brightness temperatures are close enough, then the observations are assimilated. In ice-covered Arctic regions, the poor performance of operational observation operators (e.g., RTTOV) at simulating realistic brightness temperatures is related to the difficulty of representing the surface emission properties of sea ice and snow (more generally, the ‘sea ice/snow emissivity’). The Fellow’s work will focus on investigating state-of-the-art ML tools (e.g., ‘supervised’ and ‘unsupervised’ learning methodologies) which can improve the performance of the canonical forward model so that simulated radiances can better fit satellite observations. The Research Fellow will develop ML techniques that can be interfaced with the Met Norway NWP/DA limited area model (AROME-Arctic) which covers a large part of the Arctic region. The goal of the work is also to conduct data assimilation experiments to evaluate the performance of the derived ML model with respect to the use of the default observation operator. Statistics on the forecast scores will be computed to illustrate the improvements, associated to a case study eventually.
The fellowship is offered for one year, with the possibility of extension for up to two additional years.