Aerosol extinction coefficient estimation over the European troposphere and lower stratosphere using CALIOP profiles

Abstract

The aerosol extinction coefficient (AEC) is a critical parameter in atmospheric research, providing valuable insights into aerosol concentration, composition, and their effects on solar radiation, air quality, and climate change. While the Cloud-Aerosol LiDAR with Orthogonal Polarization (CALIOP) onboard the Cloud-Aerosol LiDAR and Infrared Pathfinder Satellite Observation (CALIPSO) satellite offers high temporal continuity in vertical profiling, its AEC retrievals rely on multiple assumptions —such as fixed lidar ratios, layer homogeneity, and pre-defined aerosol models—which introduce uncertainties and limit retrieval accuracy. To address these limitations, this study proposes a deep learning-based method utilizing a ResNet architecture to estimate and retrieve AEC profiles more accurately. The model is trained using CALIOP data and ground-based measurements from European Aerosol Research Lidar Network (EARLINET) stations, enhancing predictive performance and generalization. The proposed model's performance was evaluated across multiple EARLINET stations, CALIOP Level 2 (L2) products, and two major aerosol events—a European dust storm and aged volcanic ash over north Europa—demonstrating robustness across diverse atmospheric conditions. Comparisons of total column Aerosol Optical Depth (AOD) and LiDAR ratio (LR) profiles derived from the estimated AEC with CALIOP L2 retrievals and EARLINET measurements highlighted the model's superior accuracy and generalization. Specifically, the model showed excellent agreement with EARLINET AOD (R² = 0.98, RMSE = 0.01), significantly outperforming CALIOP (R² = 0.21, RMSE = 0.06). Moreover, the model provides vertically resolved LR profiles from 0 to 15 km, whereas CALIOP L2 offers limited and often fixed LR values due to missing AEC data and restrictive assumptions. Notably, the backscatter, AEC, and LR profiles produced by the model consistently outperformed CALIOP L2 retrievals when validated against EARLINET Raman measurements. Additionally, AOD estimates showed strong agreement with EARLINET data, achieving R² and RMSE values of 0.98 and 0.01, respectively, compared to CALIOP's 0.21 and 0.06. The analysis of LR values for the significant aerosol events aligned well with the physical characteristics of these phenomena, underscoring the model's ability to capture complex aerosol behavior across vertical layers of the European troposphere and lower stratosphere.