Speaker Details

Speaker Company

Pierre Lepetit

I began my career as a math teacher. I taught at Clermont-Ferrand University before joining Météo-France in 2015 as an engineer. There, I worked on the cleaning of meteorological radar data by deep learning. In 2018, I began to work on webcam images at the LATMOS, as a posted worker. I defended a thesis on that subject in 2021. Back in Météo-France, I worked one year on the scoring of NWP and the detection of breaks in temporal series. Since 2022, I’m a data scientist at the Direction des Systèmes d’Observation, where I’m building up a webcam data processing chain.


Automatic monitoring of snow and visibility on webcam images

Webcam images often contain valuable information about weather conditions. In particular, the monitoring of snow cover and horizontal visibility would benefit from automatic processing of available webcam images. This presentation will explain how recent machine learning methods might be adapted to extract information from fixed outdoor webcam images. This challenge is complicated by the scene diversity, noisy images and unnotified camera switches. However, these issues can be efficiently handled through specific deep learning approaches. Once converted into qualitative classes or uncertainty intervals and mapped, our model outputs help to characterize and monitor fog and snow events.