진학프로석·박사 채용 접수 플랫폼

• The structure you will be joining 

Unité Mixte de Recherche - 234 GET, Géosciences Environnement Toulouse is a public laboratory dedicated to fundamental and applied research in geosciences and environmental sciences. It is a member of the Observatoire Midi-Pyrénées and employs over 250 people. GET is a Joint Research Unit (UMR) with four supervisory bodies: CNRS, IRD, University of Toulouse and CNES.

Its research activities can be divided into four main fields: (1) space and in situ observation of the Earth, (2) Earth evolution and dynamics, (3) fluid-rock-living interactions, and (4) anthropogenic activities and pollutants (air, water, soil and trophic chains). GET is the only research laboratory in the Occitanie region working in Earth Sciences and Environmental Sciences. It is at the heart of the “Environment - Resources” strategic axis of the regional plan for higher education, research and innovation (SRESRI) defined by the Occitanie Region. It is a pillar of the University of Toulouse's “Universe, Planets, Environment and Space” (UPEE) cluster. GET also contributes to teaching in the University of Toulouse's “Geosciences” discipline.

• 
  
• An attractive mission

Inland water bodies, including rivers, lakes and reservoirs, are essential components of terrestrial ecosystems. The quality of these inland waters affects all aspects of ecosystem and human well-being, including health, economic activities, ecosystem health, biodiversity and the global carbon cycle. 

New optical remote sensing sensors enable water quality to be monitored on a large scale with high temporal resolution. These images offer unprecedented capabilities for observing, understanding and anticipating the dynamics of water quality variables such as turbidity, chlorophyll, colored dissolved organic matter and suspended solids.

• 
  
•  Your future team 

With the arrival of Sentinel-2 and Sentinel-3 data and recent advances in artificial intelligence (AI), there is an unprecedented opportunity to develop new physics-driven deep learning methodologies for estimating the inherent optical properties of water. More specifically, new physics-constrained self-supervised learning methodologies can be developed to estimate the probability distributions of the intrinsic parameters of the physical model describing the interaction of light with optically active components in the water column. 

In this context, the goal will be to develop generic physics-driven deep learning methodologies for estimating water quality parameters using satellite remote sensing. 

More specifically, we plan to develop an unsupervised representation learning methodology exploiting multimodal satellite data to infer probability distributions of intrinsic parameters in simplified equations derived from radiative transfer modeling of light-water interaction. The algorithms thus developed will be validated by comparing them with a large number of in-situ data acquired during various field campaigns covering a wide range of geographical areas, or with global databases freely available in the literature. 

Several tasks have been identified :

Develop new deep learning algorithms for the inversion of water quality variables. 

Integrate large satellite datasets (time series, multiple regions). 

Validate and analyze results with in situ databases according to observed variables describing water quality. 

Publish research results in international journals. 

Present work at scientific conferences.