Amélioration de la connaissance et de la prévision des vents de vallée en conditions stables : expérimentation et modélisation statistique avec réseau de neurones artificiels

Abstract : The aim of this thesis is to study the valley flows under stable stratification to participate to the atmospheric dispersion calculations improvement under such conditions. These conditions are studied because vertical motions are reduced under stable conditions, which diminishes mixing and dispersion of compounds emitted close to the surface. Furthermore, the winds are strongly dependent on topography, which makes their calculation very hard with mesoscale NWP models. The region studied is the Cadarache valley, in south-east of France, which hosts the Cadarache CEA center. The first objective was to improve our knowledge on stratification conditions and local winds at the valley scale thanks to the KASCADE 2017 experiment. A non-dimensional valley depth allowed us to determine whether above and inside valley winds are coupled or not. Each of the two situations has been related to the temperature and wind heterogeneity in the valley. When winds are not coupled, the stratification in the valley behaves like a marginal cold pool. Slope and valley winds associated to this stratification regime have been analysed. The second objective was to improve the wind definition, inside the valley and close to the surface, from a larger scale information. A statistical nowcasting of this wind was performed, based on above valley wind observations and the knowledge of the stratification at the valley scale, with an artificial neural network (ANN). Since the method has proved to be efficient, it was thus applied to operational forecasts done on the area with the WRF meteorological model at a 3-km horizontal resolution. This constitutes a statistical downscaling process, applied to simulations where neither the topography nor the land use are adequately resolved at the local scale of the valley. This allows to improve significantly the valley wind forecast.
Complete list of metadatas

https://tel.archives-ouvertes.fr/tel-02293020
Contributor : Williams Exbrayat <>
Submitted on : Friday, September 20, 2019 - 2:28:08 PM
Last modification on : Friday, October 11, 2019 - 8:22:36 PM

Files

Identifiers

  • HAL Id : tel-02293020, version 1

Collections

Citation

Florian Dupuy. Amélioration de la connaissance et de la prévision des vents de vallée en conditions stables : expérimentation et modélisation statistique avec réseau de neurones artificiels. Océan, Atmosphère. Université Toulouse 3 – Paul Sabatier, 2018. Français. ⟨tel-02293020⟩

Share

Metrics

Record views

38

Files downloads

15