Last updated: September 2008
See also: 2008 workshop
This project is a redefinition of the previous project 'advanced interpretation of cosmo-model outputs'. It now concentrates on very high resolution models (1-3 km) and precipitation.
The priority project "INTERP" has three main focus points:
This project has the goal to provide and experiment methods for the interpretation of high resolution models. The classical methods, like Kalman filtering or MOS are now well settled and are not really specific to high resolution. One part of the project consists to experiment new methods like neural networks, boosting, Bayesian networks.
Going down to resolutions of the order of 2 km, leads to the problem of proliferation of gridpoints. Although giving more details, these are rarely at the correct place at the correct moment. Some type of aggregation is thus needed. Verifying these aggregated values led us to put the weight of the project on the verification of very high resolution models and to bridge towards the verification group.
The keyword of 'fuzzy verification' can be attached to this activity. One important task is to show if the very high resolution (~2km) models show to be better than high resolution (~7km) models, which is not trivial using classical score. We also would like that the project defines the 'best' products to be shown to the forecasters.
Another application treated in this project is hydrology. Feeding hydrological models with NWP models is a quite recent activity. Brute DMO precipitation values can be used, but one can imagine more sophisticated statistical values taken out from the multiplicity of values at gridpoints. Running the hydrological models with an EPS also starts to be experimented and show promising results.