FLARE-GMM: an automatic aerosol typing model based on Mie–Raman–fluorescence lidar measurements with LILAS

Abstract

This study presents the development of an automated aerosol typing model utilizing Mie–Raman–fluorescence lidar data collected by LILAS (Lille Lidar for Atmospheric Study), located on the ATOLL (ATmospheric Observations at LiLLe) platform in Lille, France. The proposed model, FLARE-GMM (Fluorescence Lidar-based Aerosol REcognition from Gaussian Mixture Model), employs a Gaussian mixture model trained on a dataset spanning from early 2021 to the end of 2023. FLARE-GMM is able to distinguish the predominant aerosol type in a given layer between dust, urban, and biomass burning aerosols by using the PLDR (particular linear depolarization ratio) and the fluorescence capacity, as well as relative humidity, all measured with LILAS. To ensure accurate model training, cases were manually selected to include only pure aerosol layers, as mixed aerosols are not accurately modeled by GMM. Following the training phase, the model’s performance was evaluated by investigating extreme events in which the aerosol type is not ambiguous. This approach was also completed with the use of a test dataset, on which FLARE-GMM was compared to NATALI (Neural Network Aerosol Typing Algorithm Based on Lidar Data), another automatic aerosol typing model based on neural networks using lidar data. The results demonstrated that FLARE-GMM shows promise in accurately identifying aerosol types, indicating its potential for classifying aerosols in a variety of situations. Finally, FLARE-GMM was used to estimate the aerosol types present in Lille’s atmosphere throughout the entire dataset from early 2021 to the end of 2023. A statistical analysis of these results was conducted, further underscoring the model’s capability in automated aerosol classification.