Offre de stage à IMT NORD EUROPE : Application of Deep Learning Models to Predict Particle Number Size Distributions at the ATOLL Platform

This internship proposes to apply the open-source deep learning framework to the ATOLL platform (ATmospheric Observations in liLLe,). The model integrates air-parcel historical trajectories, meteorological and chemical reanalysis data into recurrent neural networks (Long Short-Term Memory, LSTM, and Bidirectional LSTM, BiLSTM) to predict aerosol Particle Number Size Distributions (PNSDs). At ATOLL, continuous Scanning Mobility Particle Sizer (SMPS), Aethalometer (AE33), and Aerosol Chemical Speciation Monitor (ACSM) observations, combined with Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2) and Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) data, will provide the basis to test this approach in a continental urban/suburban environment in north France. The student will adapt the model to local datasets, fine-tune network parameters, and evaluate its transferability and interpretability using SHapley Additive exPlanations (SHAP) analysis. The work aims to identify the main meteorological and emission factors controlling ultrafine particle variability and to assess whether trajectory-aware deep learning can complement conventional source-apportionment and dispersion modelling. Expected outcomes include improved prediction of particle size dynamics, enhanced understanding of aerosol–meteorology interactions at ATOLL, and a demonstrator for integrating Artificial Intelligence (AI) into the AREA Work Package 1 program on advanced air-quality monitoring. Keywords: deep learning, aerosol size distribution, ATOLL, ultrafine particles