Description du sujet de thèse

The aim of this Ph.D thesis project is to design and test a new tool for predicting weather conditions conducive to tornado risks, linked to severe tornado warnings with 1 to 2 days lead time over the French mainland. It will involve the development of a real time probabilistic prediction algorithm optimized with respect to past weather events, using operational numerical weather predictions that model forecast uncertainties and thunderstorm physical processes related to tornadoes. The forecasts will be validate with respect to past tornadoes. Their robustness with respect to decadal trends in atmospheric conditions will be studied.
The tool used will be an innovative diagnostic of tornadic thunderstorm risks, optimized by machine learning over predictors derived from the AROME ensemble prediction system operational at Météo-France. It will be developed in three steps:
- diagnostic optimization over circa 10 years of operational prediction data and detailed observations over the AROME-Franbce geopgrahical area
- cross validation using samples of severe tornado reports over Europe and the USA, with high resolution numerical simulations
- time permitting, a complementary study of trends over longer time periods will be carried out.
Three factors will be critical for reaching accurate forecasts : the quality and quantity of observed tornado events data, the predictability of severe tornadic conditions over 1 to 2 days, and the identification of tornado-prone conditions such as supercell storms. Studied predictors may also include output from the future AROME-750m and AI-based weather prediction systems that might be relevant in weather prediction operations around 2030.
More scientific detail can be obtained from the named scientific supervisor.

Contexte de travail

The CNRM lab is presented here: https://www.cnrm.meteo.fr/
The host team is presented here https://www.cnrm.meteo.fr/spip.php?article367&lang=fr
The selected person will be part of a research team at Météo-France, and registered as preparing a doctorate at the Toulouse University (Ecole Doctorale SDU2E) following an M2 Master thesis or equivalent curriculum.
Relevant abilities include :
- being able to carry out scientific work under supervision in a research lab, to report on the work in writing and in oral presentations. The person should have completed an internship of level M2 Master or engineering school.
- relevant knowledge of atmospheric physical processes (e.g. thunderstorm dynamics), numerical modelling and statistics
- an ability to independently develop and perform scientific data processing and visualization using Python. Previous experience with modern machine learning tools (such as pytorch) will be useful.
- use of English: a key objective of the Ph.D work is to successfully write at least one scientific article in English and to present the work in international conferences.

Contraintes et risques

Most of the work will involve the use of a desktop computing environment.