Coherently with listed problematics, we propose to describe the research themes of the CLIME project around two components having a ``research'' orientation (themes 1 and 2), with the aim of facilitating transfer towards applications (theme 3).
Visual processing
The goal is to find visual description modes of phenomena observed on satellite data.
This research lies more or less in the continuity of image processing research done in the previous AIR project:
Image processing and physical models
The goal in this activity is to perform image processing for the estimation of physical parameters (e.g. atmospheric models parameters), taking into account the whole set of physical information describing the image formation process (in the case of atmospheric remote sensing: description of the atmosphere and radiative transfer). Practically, one gets rapidly confronted with inverse problems, most of them being ill-posed. A first important research direction lies in the definition of regularization constraints coming from spatial neighbourhood relationships, instead of tackling the inverse problem separately for each pixel. Another promising approach lies in the consideration, during the process of analyzing an image sequence, of the physical model governing the evolution of the observed phenomenon. In this manner, one seeks to establish conservation constraints based on the Navier Stokes equations for example, or on the advection-diffusion equation etc. Lastly, a third direction aims at evaluating quantitatively the quality of image processing results. That latter direction will be contemplated within the framework of data assimilation. To this end, data obtained by image processing will be assimilated within models, and the quality of the data assimilation results will serve as criterion for evaluation.
This activity is one of the present major stakes in environmental sciences. It matches up the setting and the use of assimilation data methods, notably variational methods (4D-var). An emerging point lies in uncertainties propagation in models, notably through ensemble prevision methods.
While the modeling activities are not in the framework of the present project proposal, they are able to contribute to it: CEREA is running its own models: POLAIR (photochemical pollution forecasting at continental and regional scales) and MERCURE (urban scale).
Building on the scientific capabilities developed in the previous directions, the ambition of the common project lies in the participation to the realization of software chain tools for impact assessment and environmental crisis management. Such software chains put together static or dynamic databases, data assimilation systems, forecast models, processing methods for images and environmental data, complex visualization tools, scientific workflows...
A privileged partner on that subject, inside INRIA, is the SMIS project. In the future, collaborations should also be defined, with research projects specialized in virtual reality.
Concerning external collaborations, the challenge is to rally ``operational'' partners on the building of such software chains (e.g.: an industrial actor for the ``industrial accident'' problematic, an air quality or meteorological agency for atmospheric pollution).