Data assimilation strategies for operational NWP at meso-scale and implication for nowcasting

Thibaut MONTMERLE
CNRM-GAME (Météo-France/CNRS)
thibaut.montmerle@meteo.fr

The latest developments in supercomputers and observing systems now make it feasible to implement, for operational use, results from the most recent research in mesoscale models physics, dynamics, and data assimilation. The high spatial and temporal resolutions of these models allow realistic representations of different surface features or atmospheric phenomena such as convective systems or fog. Apart from technical issues inherent to numerical and observational data handling, the operational implementation of NWP systems at mesoscale raises many issues: i) operational models representing the surface conditions and the atmospheric flow at larger scales are needed in order to provide boundary conditions, ii) the model state has to be frequently corrected towards the latest available observations to start forecasts from the best initial conditions possible. This last point is the purpose of data assimilation (DA), which aims in retrieving the best initial state (or analysis) from a previous forecast and from various observations, the weight of these two entities being given by their respective error representation.

This keynote will firstly focus on the different implementation of DA strategies at mesoscale that are currently considered in different operational NWP centres. Pros and cons of sequential vs. variational approaches will be discussed in this context. The implication that such strategies could have to support nowcasting applications in terms of forecasts suitability and availability will be presented. A forecast becomes meteorologically relevant as soon as the initialization (e.g spin-up) processes have been controlled through a numerical wave filtering. The availability mainly depends on the choice of the cut-off time for observations, of the DA algorithm, of the cycling strategy and of the large scale forecast needed for the coupling. Feedback from three years of operational use of the French AROME model, that runs operationnaly with a 2.5 km horizontal resolution and that makes use of an incremental 3DVar DA system, will then be given, and recent works obtained in this framework will be displayed to illustrate these issues.




Abstracts