Aerosol climatology – from remote sensing measurements to deep learning (CLARA)

Project number: PN-III-P2-2.1-PED-2019-2278. 

Implementation period: 17 August 2020 - 16 August 2022


  • National Institute of Research and Development for Optoelectronics (INOE), Remote Sensing Department (coordinator)
  • Politehnica University of Bucharest (UPB), SpeeD Research Laboratory

Project coordinating staff

  • Dr. Camelia Talianu - coordonator proiect (National Institute of Research and Development for Optoelectronics (INOE), Remote Sensing Department)
  • Prof. Dr. Corneliu Burileanu - responsabil proiect (Politehnica University of Bucharest (UPB), SpeeD Research Laboratory)


Aerosols, clouds, precipitations, solar and terrestrial radiations constitute a tightly coupled system for the Earth climate, governed by the thermodynamics and the dynamics of the atmosphere. The aerosols influence the Earth climate through aerosol-radiation interactions (effective radiative forcing – ERFari) and aerosol – cloud interactions (effective radiative forcing – ERFaci). The magnitude of the impact of aerosols on Earth climate remains highly uncertain because of poor knowledge of aerosol characteristics and their temporal changes. A coherent measurement strategy over whole life cycle of aerosols is necessary to improve the knowledge, requiring large and dense networks of observation instruments, properly inter-calibrated and validated. Long-term observations over different climatic regions are needed, with coherent definition of measured quantities and unitary formats of the data. Individual instruments or networks of single-purpose instruments alone can not provide all the information necessary for a complete, precise characterization of aerosols properties, their transport and their impact on climate. Consequently, the ground-based aerosol measurement networks (EARLINET, AERONET-EU, CLOUDNET) and a ground-based monitoring capacity for short-lived trace gases have been integrated in a pan-European research infrastructure, ACTRIS (Aerosols, Clouds, and Trace Gases Research Infrastructure Network). 

The instruments from the ground-bases and space-based networks can not make continuous measurements covering all aerosols trajectories, the aerosol sources are also not always known, therefore a detailed modelling of their evolution is needed to provide values in intermediate locations, to identify their sources and to study their dispersion over large areas. An effort to integrate the data from measurements of the atmosphere and state-of-the-art modelling was started as part of the European Union's Earth Observation Programme Copernicus.


The goal of the project is to develop an advanced aerosol climatology for Romania, starting from a pilot region centered on Bucharest area, covering the full period (2010 – present) of aerosol remote sensing and in-situ measurements from Romanian Atmospheric Research 3D Observatory (RADO,, Bucharest-Magurele, using an expert analysis complemented with state-of-the-art deep learning (DL) techniques. The RADO measurements for this area will be supplemented by satellite measurements and improved modelling results.

An aerosol climatology focused on biomass burning aerosols based on expert analysis using a synergistic approach of the satellite data, lidars measurements and transport modelling was started at RADO in 2014 by the PT-PCCA-2013-4-1798 project MOBBE  - “Computational Model for Prediction of the Biomass Burning Emissions and their Impact” (

In the proposed project, the PT-PCCA-2013-4-1798 project MOBBE concept will be improved, extended to the most common tropospheric aerosols and complemented with a machine learning analysis. The project proposes innovative concepts, including synergistic expert analysis and deep learning techniques, fulfilling also the Romanian Smart Specialization Strategy on Analysis, Management and Security of Big Data. Extending the biomass burning aerosol climatology to all tropospheric aerosols, the project will improve significantly the characterization of aerosols and their impact on climate, air quality and cloud-precipitation processes, moving from experimental level (MOBBE) to a higher technological maturity level (validated technology in the laboratory). The validated results will be made available for use in other studies of the impact of aerosols on climate. The project outcomes will also contribute conceptually to the Copernicus effort to assimilate the ground-based remote-sensing measurements in Atmosphere Monitoring Service (CAMS).


WP 1: Data collection and selection of analysis cases

WP 2: Aerosol physics model and classification

WP 3: Classic expert analysis of aerosols

WP 4: DL analysis of aerosols: architecture

WP 5: DL analysis of aerosols: evaluation and optimization 

Related Links

EARLINET - European Aerosol Research Lidar Network to Establish an Aerosol Climatology
AERONET - Aerosol Robotic Network
ACTRIS - European Research Infrastructure for the observation of Aerosol, Clouds, and Trace gases.
CAMS - Copernicus Atmosphere Monitoring Service.
WINDY - Meteorological and Air Quality Service.
RADO - Romanian Atmospheric 3D Observatory Database


Scientific results: Presentations 


European Lidar Conference – ELC2020 (Granada, Spain, 18 – 20 Noiembrie 2020, on-line): 

  • Exploring The Use Of Warping Networks To Improve The Lidar Optical Retrievals
    by Doina Nicolae, Camelia Talianu, Livio Belegante, Dragos , Ene, Victor Nicolae (ELC2020-DoinaNicolae.pdf, oral presentation)
  • Investigation of Aerosol Layer Height Algorithms to Identify the Transition Layers Using Synergy of Lidar, Ceilometer and Radiometer Measurements
    by Camelia Talianu, Mariana Adam, Florica Toanca, Simona Andrei (ELC2020_poster_Talianu.Camelia_01.pdf poster presentation) 


European Aerosol Conference EAC 2021 (Birmingham, United Kingdom, Aug. 30 - Sept 03, 2021, on-line):

  • Quantification of black carbon anthropogenic changes over Bucharest during COVID-19 lock-down based on similarity of aerosol transport patterns
    by Camelia Talianu, Jeni Vasilescu, Luminita Marmureanu and Livio Belegante (EAC2021.pdf oral presentation)
44th International Conference on Telecommunications and Signal Processing (TSP) (Brno, Czech Republic, 26 - 28 July 2021):
  • Automatic Pollen Classification Using Convolutional Neural Networks
    Mihai Boldeanu, Horia Cucu, Corneliu Burileanu, Luminița Mărmureanu (on-line presentation)

 European Lidar Conference – ELC2021 (Granada, Spain, 16 – 18 Noiembrie 2021, on-line): 

  • Aerosol layer identification and segmentation from lidar and ceilometer profiles using unsupervised deep learning
    by Mihai Boldeanu, Cristian Manolache, Horia Cucu, Corneliu Burileanu, Camelia Talianu (S02P13_Talianu.pdf poster presentation)

30th International Laser Radar Conference - ILRC 30 (virtual conference, June 26th – July 1st, 2022)
    by Adam Mariana, Fragkos Konstantinos, Belegante Livio, Andrei Simona, Talianu Camelia, Luminita Marmureanu, Antonescu Bogdan, Ene Dragos, Nicolae Victor, Solomos Stavros, Amiridis Vassilis
14th International Conference on Communications COMM22 (online in perioada 16 – 18 Iunie 2022)
  • "Unsupervised Deep Learning Models for Aerosol Layers Segmentation" (oral presentation)
    by Cristian Manolache, Mihai Boldeanu, Camelia Talianu, Horia Cucu
1st ACTRIS Science Conference (online in perioada 11–13 Mai 2022)
  • "Long-range transported smoke from Ukraine as seen in Bucharest by ACTRIS instruments" (oral presentation)
    by D. Nicolae, A. Nemuc, C. Talianu, J. Vasilescu, L. Belegante, F. Toanca, L. Marmureanu, B. Antonescu, C. Radu, E. Carstea, S. Andrei, K. Fragkos, M. Adam, V. Nicolae, R. Parloaga, C. Marin, A. Ilie, A. Tilea

Results: List of peer-reviewed publications for project (ISI)


  • Ana Denisa Urlea,Nicu Barbu,Simona Andrei,Sabina Ştefan, " Simulation of Vesuvius volcanic ash hazards within Romanian airspace using the Hybrid Single-Particle Lagrangian Integrated Trajectory Volcanic Ash numerical model", MeteorologicalApplications,28(3),e2001. (2021)
  • Mihai Boldeanu, Horia Cucu, Corneliu Burileanu, Luminița Mărmureanu, "Automatic Pollen Classification Using Convolutional Neural Networks", TSP 2021 Conference Proceedings, IEEE Catalog Number: CFP2188P-ART, ISBN 978-1-6654-2933-7, DOI: 10.1109/TSP52935.2021  
  •  Mihai Boldeanu, Horia Cucu, Corneliu Burileanu, Luminița Mărmureanu, "Multi-Input Convolutional Neural Networks for Automatic Pollen Classification", Appl. Sci. 2021, 11, 11707. (2021)



  • V. Nicolae, S. Stefan, A. Nemuc, "Changes in the aerosol properties during pandemic restrictions in Romania", Romanian Journal of Physics 67, 809 (2022).

  • M. Boldeanu, M. González-Alonso, H. Cucu, C. Burileanu, J. M. Maya-Manzano and J. Buters, "Automatic Pollen Classification and Segmentation using U-nets and Synthetic Data," in IEEE Access, 2022, doi: 10.1109/ACCESS.2022.3189012

  • Adam Mariana, Fragkos Konstantinos, Belegante Livio, Andrei Simona, Talianu Camelia, Luminita Marmureanu, Antonescu Bogdan, Ene Dragos, Nicolae Victor, Solomos Stavros, Amiridis Vassilis, "Methodology for lidar monitoring of biomass burning smoke in connection with the land cover", Remote Sens. 2022, 14, 4734.



Project results

  • A synergistic methodology used to develop a complex aerosol climatology (CLARA) based on new methods of aerosol analysis based on an expert analysis and on DL technology
  • A web-based database containing 10 years of regional aerosol climatology, compiled from multi- spectral lidar data and supplemented by data from other remote sensing instruments existent at RADO, field observations, CAMS data-aerosols and transport modeling.

The synergistic methodology used to develop a complex aerosol climatology (CLARA) is schematically presented in Figure 1. This methodology is based on the synergy of the best available remote sensing technologies at MARS aerosol remote sensing laboratory combined with CAMS - "Copernicus Atmosphere Monitoring Service" (Inness A. et al., 20191), NATALI aerosol-typing model (Nicolae D. et al., 20182) and FLEXPART atmospheric transport model version 10 (Pisso I. et al, 20193)..

Figure 1: CLARA Synergistic methodology block diagram

A synthetic view of the aerosol parameters provided by each instrument is given in the following table:


Instruments / services

Attenuated backscatter due to aerosol (PABC)

Lidar; Ceilometer;

Aerosol extinction coefficient (PEC)

Lidar; CAMS global atmospheric composition

Aerosol backscatter coefficient (PBC)

Lidar; Ceilometer;CAMS global atmospheric composition

Aerosol optical depths (AODs)

Lidar; Photometer; CAMS global atmospheric composition

Ångström exponents (AEs)

Lidar; Photometer; CAMS global atmospheric composition

Lidar ratios (LRs)

Lidar; Photometer; CAMS global atmospheric composition

Aerosol mixing ratios

CAMS global atmospheric composition

The data from the instruments and CAMS (Copernicus Atmosphere Monitoring Service) are collected automatically and subjected to a process of checking their quality and consistency ("Data collection and preprocessing" panel). Consistent and high-quality data are selected, processed and prepared for expert analysis and DL analysis ("data processing panel").

The data are synchronized in time and space and processed with the algorithms and models developed in CLARA. The results are analyzed using expert analysis and DL-based analysis ("data analysis" panel). The DL analysis consists in the identification of aerosol layers from the attenuated backscatter profiles. An example of segmentation and identification of aerosol layers using neural networks of the W-net type is presented in figure 2.


Figure 2:  Example of 6 spatio-temporal profiles of measurements taken at different time points superimposed on the aerosol layers (marked with different background colors) identified in the latent layer of a W-net model.

For each identified layer, the optical properties, mixing ratios and mass concentrations are calculated. The allocation of aerosol sources is done with the FLEXPART dispersion model (Flexible particles dispersion model) and the aerosol type is identified with the NATALI model (Neural network Aerosol Typing Algorithm based on LIdar data, available at . The results are saved both as an image and as an ASCII file and stored in the database. The database was created in MYSQL and is operational on Linux platforms. To be able to access it, a web interface was developed using elements of PHP, Java Script and Python.The web interface to the database is presented in figure 3 and can be accessed from the web address:


Figure 3:  Web database interface


The methods developed in the project were implemented in software packages written in Python. The software package for segmentation and identification of aerosol layers based on W-net neural networks is available at The software packages for data collection and expert analysis are integrated at RADO and can be provided upon request to academic users or non-profit organizations focused on atmospheric and climatological studies dedicated to atmospheric aerosol.

 Prezentare succintă a rezultatelor obținute în cadrul proiectului

Aerosolii sunt particule în suspensie produsi local din surse naturale si antropice (aerosoli primari) sau pot fi formati ca aerosoli secundari prin procese chimice, cu proprietăti date; aerosolii sunt transportati pe distante scurte până la distante lungi (mii de km), de obicei în miscări turbulente. Datorita variabilitătii ridicate a surselor de aerosoli, distributia geografică a aerosolilor este foarte neuniformă si dependentă de timp.

Climatologia aerosolului obtinuta prin proiectul CLARA, este sustinută cu date observationale si date furnizate de CAMS (Copernicus Atmosphere Monitoring Service) complementata cu modele de transport a particulelor de aerosol într-un mod automat pentru a furniza informatiile in timp real. In figura 4 sunt prezentate principalele tipuri de aerosol definite in CLARA. Clasificarea este prezentata pe diagrama depolarizare (forma particulelor) versus raport lidar(compozitia particulelor).



Figura 4 Tipuri aerosol definite in CLARA folosind diagrama depolarizare (DEP550) vs raport lidar (LR550)


In figura 5 este prezentat un exemplu de distributie a regiunilor cu surse de aerosol obtinute din modelul de dispersie FLEXPART pentru statia pilot RADO (Romanian Atmosperic Research 3D Observatory) de pe platforma Magurele.


Figura 5: Relatia sursa aerosol – receptor (statia pilot RADO – Magurele) calculata ca timp petrecut de o masa de aer peste o anumita regiune


In figura 6 sunt prezentate exemple de distributie a straturilor de aerosoli pe verticala


Figura 6 Straturi de aerosoli inregistrate la statia pilot RADO – Magurele


1 Inness, A, Ades, M, Agustí-Panareda, A, Barré, J, Benedictow, A, Blechschmidt, A, Dominguez, J, Engelen, R, Eskes, H, Flemming, J, Huijnen, V, Jones, L, Kipling, Z, Massart, S, Parrington, M, Peuch, V-H, Razinger M, Remy, S, Schulz, M, Suttie, M, The CAMS reanalysis of atmospheric composition Atmos. Chem. Phys., 19 (2019), pp. 3515-3556, 10.5194/acp-19-3515-2019

2 Nicolae, D., Vasilescu, J., Talianu, C., Binietoglou, I., Nicolae, V., Andrei, S., and Antonescu, B., A Neural Network Aerosol Typing Algorithm Based on Lidar Data, Atmos. Chem. Phys., 18, 14511–14537,, 2018

3 Pisso, I., Sollum, E., Grythe, H., Kristiansen, N. I., Cassiani, M., Eckhardt, S., … Stohl, A. (2019). The Lagrangian particle dispersion model FLEXPART version 10.4. Geoscientific Model Development, 12(12), 4955-4997.

 This project is funded by: