The new generation of atmospheric composition sensors such as TROPOMI is capable of providing spectra of high spatial and spectral resolution. To process this vast amount of spectral information, fast radiative transfer models (RTMs) are required. In this regard, we analyzed the efficiency of two acceleration techniques based on the principal component analysis (PCA) for simulating the Hartley-Huggins band spectra. In the first one, the PCA is used to map the data set of optical properties of the atmosphere to a lower-dimensional subspace, in which the correction function for an approximate but fast RTM is derived. The second technique is based on the dimensionality reduction of the data set of spectral radiances. Once the empirical orthogonal functions are found, the whole spectrum can be reconstructed by performing radiative transfer computations only for a specific subset of spectral points. We considered a clear-sky atmosphere where the optical properties are defined by Rayleigh scattering and trace gas absorption. Clouds can be integrated into the model as Lambertian reflectors. High computational performance is achieved by combining both techniques without losing accuracy. We found that for the Hartley-Huggins band, the combined use of these techniques yields an accuracy better than 0.05% while the speedup factor is about 20. This innovative combination of both PCA-based techniques can be applied in future works as an efficient approach for simulating the spectral radiances in other spectral regions.

Aerosol typing is essential for understanding atmospheric composition and its impact on the climate. Lidar-based aerosol typing has been often addressed with manual classification using optical property ranges. However, few works addressed it using automated classification with machine learning (ML) mainly due to the lack of annotated datasets. In this study, a high-vertical-resolution dataset is generated and annotated for the University of Granada (UGR) station in Southeastern Spain, which belongs to the European Aerosol Research Lidar Network (EARLINET), identifying five major aerosol types: Continental Polluted, Dust, Mixed, Smoke and Unknown. Six ML models – Decision Tree, Random Forest, Gradient Boosting, XGBoost, LightGBM and Neural Network- were applied to classify aerosol types using multiwavelength lidar data from EARLINET, for two system configurations: with and without depolarization data. LightGBM achieved the best performance, with precision, recall, and F1-Score above 90 % (with depolarization) and close to 87 % (without depolarization). The performance for each aerosol type was evaluated and dust classification improved by ∼ 30 % with depolarization, highlighting its critical role in distinguishing aerosol types. Validation against independent datasets, including a smoke case and a Saharan dust event, confirmed robust classification under real and extreme conditions. Compared to NATALI, a neural network-based EARLINET algorithm, the approach presented in this work shows improved aerosol classification accuracy, which emphasize the benefits of using high-resolution multiwavelength lidar data from real measurements. This highlights the potential of ML-based methods for robust and accurate aerosol typing, establishing a benchmark for future studies using multiwavelength lidar at high-resolution data from EARLINET.

Current atmospheric composition sensors provide a large amount of high spectral resolution data. The accurate processing of this data employs time-consuming line-by-line (LBL) radiative transfer models (RTMs). In this paper, we describe a method to accelerate hyperspectral radiative transfer models based on the clustering of the spectral radiances computed with a low-stream RTM and the regression analysis performed for the low-stream and multi-stream RTMs within each cluster. This approach, which we refer to as the Cluster Low-Streams Regression (CLSR) method, is applied for computing the radiance spectra in the O₂ A-band at 760 nm and the CO₂ band at 1610 nm for five atmospheric scenarios. The CLSR method is also compared with the principal component analysis (PCA)-based RTM, showing an improvement in terms of accuracy and computational performance over PCA-based RTMs. As low-stream models, the two-stream and the single-scattering RTMs are considered. We show that the error of this approach is modulated by the optical thickness of the atmosphere. Nevertheless, the CLSR method provides a performance enhancement of almost two orders of magnitude compared to the LBL model, while the error of the technique is below 0.1% for both bands.

Fast radiative transfer models (RTMs) are required to process a great amount of satellite-based atmospheric composition data. Specifically designed acceleration techniques can be incorporated in RTMs to simulate the reflected radiances with a fine spectral resolution, avoiding time-consuming computations on a fine resolution grid. In particular, in the cluster low-streams regression (CLSR) method, the computations on a fine resolution grid are performed by using the fast two-stream RTM, and then the spectra are corrected by using regression models between the two-stream and multi-stream RTMs. The performance enhancement due to such a scheme can be of about two orders of magnitude. In this paper, we consider a modification of the CLSR method (which is referred to as the double CLSR method), in which the single-scattering approximation is used for the computations on a fine resolution grid, while the two-stream spectra are computed by using the regression model between the two-stream RTM and the single-scattering approximation. Once the two-stream spectra are known, the CLSR method is applied the second time to restore the multi-stream spectra. Through a numerical analysis, it is shown that the double CLSR method yields an acceleration factor of about three orders of magnitude as compared to the reference multi-stream fine-resolution computations. The error of such an approach is below 0.05%. In addition, it is analysed how the CLSR method can be adopted for efficient computations for atmospheric scenarios containing aerosols. In particular, it is discussed how the precomputed data for clear sky conditions can be reused for computing the aerosol spectra in the framework of the CLSR method. The simulations are performed for the Hartley–Huggins, O${}_2$ A-, water vapour and CO${}_2$ weak absorption bands and five aerosol models from the optical properties of aerosols and clouds (OPAC) database.

Aerosol typing is essential for understanding atmospheric composition and its impact on the climate. Lidar-based aerosol typing has been often addressed with manual classification using optical property ranges. However, few works addressed it using automated classification with machine learning (ML) mainly due to the lack of annotated datasets. In this study, a high-vertical-resolution dataset is generated and annotated for the University of Granada (UGR) station in Southeastern Spain, which belongs to the European Aerosol Research Lidar Network (EARLINET), identifying five major aerosol types: Continental Polluted, Dust, Mixed, Smoke and Unknown. Six ML models – Decision Tree, Random Forest, Gradient Boosting, XGBoost, LightGBM and Neural Network- were applied to classify aerosol types using multiwavelength lidar data from EARLINET, for two system configurations: with and without depolarization data. LightGBM achieved the best performance, with precision, recall, and F1-Score above 90 % (with depolarization) and close to 87 % (without depolarization). The performance for each aerosol type was evaluated and dust classification improved by ~30 % with depolarization, highlighting its critical role in distinguishing aerosol types. Validation against an independent dataset from a Saharan dust event confirmed robust classification under real and extreme conditions. Compared to NATALI, a neural network-based EARLINET algorithm, the approach presented in this work shows improved aerosol classification accuracy, which emphasize the benefits of using high-resolution multiwavelength lidar data from real measurements. This highlights the potential of ML-based methods for robust and accurate aerosol typing, establishing a benchmark for future studies using multiwavelength lidar at high-resolution data from EARLINET.

Aerosol typing is essential for understanding atmospheric composition and its impact on the climate. Lidar-based aerosol typing has been often addressed with manual classification using optical property ranges. However, few works addressed it using automated classification with machine learning (ML) mainly due to the lack of annotated datasets. In this study, a high-vertical-resolution dataset is generated and annotated for the University of Granada (UGR) station in Southeastern Spain, which belongs to the European Aerosol Research Lidar Network (EARLINET), identifying five major aerosol types: Continental Polluted, Dust, Mixed, Smoke and Unknown. Six ML models – Decision Tree, Random Forest, Gradient Boosting, XGBoost, LightGBM and Neural Network- were applied to classify aerosol types using multiwavelength lidar data from EARLINET, for two system configurations: with and without depolarization data. LightGBM achieved the best performance, with precision, recall, and F1-Score above 90 % (with depolarization) and close to 87 % (without depolarization). The performance for each aerosol type was evaluated and dust classification improved by ~30 % with depolarization, highlighting its critical role in distinguishing aerosol types. Validation against an independent dataset from a Saharan dust event confirmed robust classification under real and extreme conditions. Compared to NATALI, a neural network-based EARLINET algorithm, the approach presented in this work shows improved aerosol classification accuracy, which emphasize the benefits of using high-resolution multiwavelength lidar data from real measurements. This highlights the potential of ML-based methods for robust and accurate aerosol typing, establishing a benchmark for future studies using multiwavelength lidar at high-resolution data from EARLINET.