The IASI (Infrared Atmospheric Sounding Interferometer) instruments onboard the Metop satellites provide high-resolution infrared observations of the atmosphere, enabling the detection of several trace gases and atmospheric events such as volcanic eruptions and wildfires. Current approaches mainly focus on known and previously characterized species and events. However, the large volume of IASI observations may contain rare or poorly characterized spectral signatures that remain undetected.

This internship aims to explore unsupervised machine-learning approaches for the detection and characterization of such atypical spectral signatures. The methodology will first be evaluated on well-documented events and then applied to observations without predefined events, with a particular interest in unusual emission events and potential industrial sources.

Internship tasks

• Extract and quality-control a subset of IASI Level-1 observations.
• Preprocess and explore the spectral data.
• Develop and compare unsupervised methods for detecting atypical spectral signatures.
• Evaluate the methods on selected known events.
• Investigate atypical spectral signatures and identify those that may correspond to real atmospheric or emission events and warrant further analysis.

Prerequisites

• Master’s level (M2) in data science, signal processing, atmospheric physics, or a related field.
• Background in machine learning, with a good understanding of unsupervised learning and anomaly detection methods.
• Good programming skills in Python and familiarity with Linux.
• Experience with satellite or atmospheric data is a plus.
• Good scientific reading and writing skills.