طبقه بندي هواويز با استفاده از داده هاي ليدار ماهواره اي بر مبناي روش هاي يادگيري ماشيني عميق

Abstract The classification of atmospheric aerosol types is fundamental for understan‎ding their roles in climate change, air quality, an‎d radiative–dynamic cycles. In this context, the aerosol extinction coefficient (AEC) serves as a key parameter in atmospheric research, providing valuable insights into aerosol concentration, composition, an‎d their impacts on solar radiation, air quality, an‎d climate. Although the CALIOP sensor onboard the CALIPSO satellite offers high temporal continuity in vertical profiling, its AEC retrieva‎ls rely on a set of algorithmic assumptions that inherently limit monitoring accuracy. To address this limitation, this study proposes a deep learning approach based on the ResNet architecture to estimate an‎d retrieve AEC profiles with higher accuracy. The model was trained using CALIOP data an‎d ground-based measurements from the EARLINET network to enhance its performance, predictive capability, an‎d generalization.The effectiveness of the proposed model was eva‎luated at multiple EARLINET stations by comparing it with CALIOP Level 2 (L2) products during two representative aerosol events—an intense European dust storm an‎d aged volcanic ash transport over northern Europe. The results demonstrated the model’s robustness under diverse atmospheric conditions. Comparisons of the columnar aerosol optical depth (AOD) an‎d lidar ratio (LR) derived from the estimated AECs with both CALIOP-L2 retrieva‎ls an‎d EARLINET observations confirmed the superior accuracy an‎d generalization of the model. In particular, the backscatter, AEC, an‎d LR profiles produced by the proposed model consistently outperformed those of CALIOP-L2 when validated against EARLINET Raman lidar measurements. The estimated AODs also exhibited excellent agreement with EARLINET data (R² = 0.98 an‎d RMSE = 0.01), whereas the corresponding CALIOP values were 0.21 an‎d 0.06, respectively. Furthermore, the LR values obtained for the examined events were fully consistent with the physical characteristics, types, an‎d classifications of the aerosols involved, demonstrating the model’s capability to capture the complex behavior of various aerosol types within the vertical layers of the European troposphere an‎d lower stratosphere. Overall, these results establish the proposed framework as a powerful tool for advanced aerosol monitoring an‎d for supporting regional (European-scale) climate–atmosphere studies. Keywords: Aerosol Extinction Coefficient, Satellite LiDAR, CALIOP, EARLINET, LiDAR Ratio, AOD

5 فصل

149419