Environment

This work package covers all dimensions of the environment, by addressing important AI-based issues concerning ocean, atmosphere and land. Our goal is to carry out integrated research dedicated to the analysis of data generated by various (remote and in situ) sensing techniques that are currently used to assess and monitor the state of biodiversity and ecosystem health.

The increasing availability of sensing data offers both opportunities (i.e., a wealth of information) and challenges (i.e., data handling and analyses) to study environmental changes at various temporal and spatial scales. Those encompass local (e.g., focal soil pollution, local meteorological phenomena), regional (e.g., marine pollution, plant invasion) and global (e.g. climate warming, atmospheric deposits) changes. Interestingly enough, AI offers a technological leap opportunity for exploiting data coming from different environmental backgrounds. Challenges are difficult, but underlying issues to be addressed are well-identified, corresponding to central themes of MAIA:

• data quality and reliability over time (interoperability, completion, fusion of heterogeneous data); some of these notions being related to acceptability;
• robustness of models by efficiently coupling, combining, calibrating approaches (models) coming from Physics, Mathematics, and AI-based systems; Hybrid and explainable AI will be at stake here.

Three main objectives will be addressed:

• Leveraging AI-based image analysis for improving ocean-color remote-sensing. The use of satellites to monitor the color of the ocean requires some signal processing With AI, an important step can be taken on various classification tasks, image analysis (e.g., fusion of satellite images with different spectral and spatial resolutions), and time-series analysis (while predicting key bio-geochemical variables).
• Combining ML and fundamental physical principles and approaches such as hydrothermodynamic laws, first principles molecular dynamics and high-resolution rotational–vibrational spectroscopy to study atmospheric dynamics, air composition for different applications including breathomics analysis for medical diagnosis, pollutant monitoring, and measurements of gas phase exchange at the atmosphere-ocean interface.
• Monitoring the health status of forest and agricultural lands at large spatial scales by combining remote sensing with field data. This is a timely issue to detect compositional and physiological responses to environmental changes. Aside from their inherent interest in assessing ecosystem health and resilience, these responses can serve as proxies for human exposure to particular pollutants. This requires the automatization of certain tasks, such as pattern recognition, species identification, and plant disease detection.