PhD student, LISTIC, USMB
Contact
E-mail : Matthieu.gallet@ univ-smb.fr
Website: https: //matthieu-gallet.github.io/
Telephone: +33(0) 4 50 09 42 09
Fax: +33(0) 4 50 09 65 59
Office: A224
Address 1: LISTIC - Polytech Annecy-Chambery, BP 80439, 74944 Annecy le Vieux Cedex, France
Thesis
Group: LISTIC
Theme: AFuTé
Subject: Machine-learning for the characterization and analysis of cryospheric dynamics using
radar satellite imagery
Abstract: The texture of the cryospheric cover varies over time as energy is exchanged between the snowpack, the underlying ground and the atmosphere.
between the snowpack, the underlying ground and the atmosphere. While precipitation and meteorological
the properties of a surface snow layer, its evolution over time, as well as that of the underlying
its evolution over time, as well as that of the different layers superimposed to form the
snow cover, depends on weather conditions.
These are assessed in collaboration with Météo-France's Centre d'Étude de la Neige.
On the small scale of satellite vision (a few hundred square meters), we are interested in analyzing
spatial variations in textures (snow in all its states and ice), whereas on a large scale
(several hundred square kilometers), we'll be more interested in the presence or absence of snow cover
or the differentiation between dry and wet snow.
To date, the studies available on the subject have been site-specific (very limited geographical areas), and have made little use of the data currently available, let alone the potential of the new methods.
available data, let alone the potential of deep learning methods in this type of application.
of deep learning methods in this type of application.
The aim of the project is to propose convolutional neural models, operating in a deep learning framework
framework and based on time series of satellite images, in order to enable :
- Evaluation of the extent of cryospheric surfaces (semantic segmentation);
- Automatic recognition of observable textures (classification) on cryospheric surfaces.
cryosphere, for example: tracking of total snow, including dry snow, to differentiate between
different snow states; - Dynamic tracking :
- surface areas of interest (changes in the contours of the cryosphere element in space and
in space and time) and - contents of these surfaces (occupancy ratios on the same surface between different
components/features visible on a given surface) to deduce the spatial
the spatial evolution of observable surface states.
- surface areas of interest (changes in the contours of the cryosphere element in space and
Keywords: Cryosphere, Radiative transfer, Machine-learning, Radar
Supervisor: Abdourrahmane ATTO (LISTIC)
Co-supervisors: Fatima KARBOU (Météo-France), Emmanuel TROUVÉ (LISTIC)
Start of thesis: October 2021
Doctoral school: Environmental Engineering Sciences