the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 21 Apr 2026
New operational perspective to identify aerosol in real-time with a pioneering algorithm (CONIOPOL) based on single wavelength polarization lidar (CL61)
Abstract. Air quality monitoring and climate studies require continuous, vertically resolved observations to characterize aerosols and their impact on radiation, cloud microphysics, and atmospheric composition. In this study, we present CONIOPOL (CONIOlogy + POLarization), an automated depolarization-based classification algorithm developed with the polarized Automatic Lidar Ceilometer (ALC), CL61 (Vaisala Oyi, FIN) installed at Uccle, Belgium. The algorithm combines linear depolarization ratio (LDR), attenuated backscatter, and cloud-base height retrievals to distinguish between aerosols, clouds, and precipitation, and to further classify aerosol subtypes.
One full year (February 2024–January 2025) of observations was analyzed to retrieve and evaluate the seasonal and vertical distributions of major aerosol categories, with results compared against Copernicus Atmosphere Monitoring Service (CAMS) model forecast outputs. The CONIOPOL algorithm successfully identified different types of aerosol – including dust, smoke, hygroscopic, and mixed aerosols – demonstrating strong temporal and vertical coherence with CAMS simulations. In particular, dust and smoke plumes detected above 1000 m showed a good agreement with it.
Despite its spectral limitations, the single-wavelength lidar provides continuous, high-resolution, and climatologically consistent aerosol classification, offering valuable insights into the seasonal evolution of aerosol types over mid-latitude Europe. These findings underscore the potential of depolarization-capable ALCs for long-term aerosol and air quality climatology, bridging temporal gaps between satellite, in situ, and multi-wavelength lidar observations.
- Preprint
(2593 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 26 May 2026)
- RC1: 'Comment on egusphere-2026-948', Anonymous Referee #1, 29 Apr 2026 reply
-
RC2: 'Comment on egusphere-2026-948', Anonymous Referee #3, 05 May 2026
reply
General comments
The authors introduce a new algorithm to identify aerosol types from the measurements of the new Vaisala CL61 ceilometer with depolarisation capabilities automatically and in near real-time. The algorithm CONIOPOL is described in detail how the use of the backscatter profile, the depolarisation, and the cloud base height, by filtering, smoothing, and a threshold-based classification and clustering method is combined to allow the monitoring of aerosol and clouds and their seasonal variability.
The algorithm design and the necessary procedures as calibration procedures are described and verified using measurements from a one-year record in 2024-2025 of the CL61 instrument at Uccle, a long-running, well recognised atmospheric monitoring station in Belgium. The algorithm is able to distinguish a wide range of aerosol types, such as ash dust, mix aerosol, ice crystals, classifies hygroscopic aerosols, and detects different precipitation states.
This manuscript is well organised and offers a valuable contribution to the use of automatic classification of aerosol by adding the depolarisation ratio for networks of low-cost instruments as E-Profile and should be published after addressing the suggestions below.
Specific comments:
The LDR is defines several times in the introduction for linear depolarisation ratio. In one case it is also used for volume depolarisation ratio (line 155) and in one case for particle linear depolarization ratio (line 585). Please clarify what is used in the algorithm.
Fig. 8: and paragraph 6.1.1.: It is unclear what is meant by occurrence, if it can exceed 100%? 5% occurrence means 5% of all observed categories? During which time frame?
Fig. 9: The text of the axes and captions should be readable without turning around the graphs.
Line 460 – 467: Intercomparison with CAMS: please explain in more detail how the CAMS values are assessed conditionally on CONIOPOL detections in both time and altitude, and why this leads to multiple use of the CAMS values.
Fig. 10 and 11: Please explain more clearly, how these box plots show a good or bad agreement between CAMS and CONIOPOL results. It is not that obvious.
Citation: https://doi.org/10.5194/egusphere-2026-948-RC2 -
RC3: 'Comment on egusphere-2026-948', Anonymous Referee #2, 11 May 2026
reply
In this article, the authors present an algorithm called CONIOPOL, that aims to classify aerosols based on measurements from single-wavelength polarization lidar. They apply this algorithm to one year (feb 2024-feb 2025) of ground-based lidar data at 910.55 nm from the CL61 instrument operated at Uccle (Belgium). They document the output of the classification and compare it with CAMS data.
The subject is important, relevant to contemporary problems, and well within the scope of ACP. The correct identification of aerosol nature, and its documentation as a function of altitude (especially in the boundary layer) is key to making progress on documenting air quality and on quantifying the indirect influence of anthropogenic pollution on climate through aerosol-clouds interactions like the Twomey effect. To reach that goal, fruitful approaches include the development of reliable, well-documented algorithms to classify aerosols that can be applied systematically to easily-deployable operational lidar measurements. The continuous datasets obtained this way provide information essential to build long-term climatologies.
The manuscript is well-written, the figures are generally well-designed and convey a clear message. The algorithm is described with precision, the impact of many choices is discussed and well documented. There are several points that should be clarified before publication.
Main comments
My main issue with the article is that I can't understand the origins of the values of depolarization ratio used as thresholds for the classification. The article mentions (L. 256-288) a UK field campaign with as reference only this document :
https://www.vaisala.com/sites/default/files/documents/WEA-MET-WhitePaper-CL61-B212267EN-A.pdf
This document is not a peer-reviewed article, it reads more like an advertisement for Vaisala. It does not present the field campaign, explain which experiments made it possible to link specific depolarization ratios with specific aerosol types, or discuss the reliability of CL61 retrievals. Most importantly, I could not find in this document the numbers used in the CONIOPOL algorithm present in figure 1. One figure in the report shows the approximate ranges of typical LDR values for different particle types, but it is not clear where the ranges come from, or even to which wavelength they apply. The document refers to another "full report" being available from Vaisala on request. As long as the experiments and studies which led to these values, which are key to the classification, are not available and peer-reviewed, it will not be possible for me to recommend publication. This effort should involve relating the uncertainties of depolarization ratio measurements with the classification output.My second concern is that I had trouble following some of the reasoning : the article makes the point that LDR by itself does not contain enough information to uniquely identify several aerosol types (for instance distinguish volcanic ash from dust), but still argues that distinct categories of aerosols are mandatory and proceeds to classify separately aerosol types that have overlapping LDR distributions. I would appreciate it if the authors could clarify their reasoning here (see also several minor comments addressing this point). I do not see the point of uniquely labelling detections as ash or volcanic dust, while recognizing that this classification requires additional information (back-trajectory analysis, geographic source attribution) to be interpreted correctly. This opens the door to misinterpretations of the algorithm output. Uncertainties on aerosol identification should be visible in the output, and the algorithm should refrain from providing certain classifications when certainty is not possible.
Another issue with the article is that in my view, comparison with output from CAMS is not akin to validation. If we assume that the representation of aerosols in CAMS is close to reality, there is no need for lidar-based aerosol classification anymore, since we can just use CAMS. Actually, the point of lidar-based aerosol classification is to provide constraints to CAMS reanalyses through assimilation. Comparing output of CONIOPOL with CAMS can provide reassurance that lidar-based classifications are overall reasonable, but these classifications should provide additional information compared to CAMS (otherwise, what's the point?), and thus differences with CAMS should be expected. In my point of view, actual validation involves comparison of retrievals with in-situ measurements, for instance. Of course it is significantly harder and expensive to obtain in-situ data, and I do not expect the authors to provide new measurements during the review process. They should, however, refrain from identifying the comparison exercise with CAMS as validation, discuss it as useful comparison, and recognize its possible limitations.
The title focuses on the aerosol identification part of CONIOPOL, while the article shows the algorithm classifies atmospheric features between aerosol types, liquid/ice clouds, and precipitation. In my view, the impact of the article would be improved by focusing solely on aerosols and removing the cloud/precipitation results.
Finally, the CONIOPOL algorithm takes among its input cloud base detections, and apparently assumes they are perfect. The impact of cloud misdetections is not discussed. The algorithm assumes that cloud detection and aerosol detection are independent, which is not necessarily the case. Cloud detections, however, can include aerosols and vice-versa. In other words, the signal ranges of clouds and aerosols do overlap, and misclassifications do happen, for instance in presence of optically thick aerosol layers. The study should try to evaluate the sensitivity of results to errors in the cloud detection and cloud base retrievals.
Minor comments
- L. 13: Thanks for teaching me the word "coniology"!
- Figure 1 : why do the colors mean in the diagram? (red, orange, etc.)
- Figure 1, legend: "cloud base height" is not a lidar measurement. It is a retrieval, with all the attached uncertainties and interpretation issues.
- L. 50: The article should explicitly address the difference between volume depolarization ratio (which is affected by the ratio between signal contributions from particles and molecules) and particulate depolarization ratio (only due to particles), and be explicit which one it is referring to. "Linear" depolarization ratio is often assumed to refer to the particulate depolarization ratio (as in Burton et al. 2015) in opposition to the plain "depolarization ratio" often used to refer to the volume depolarization ratio. CALIOP measures volume depolarization ratio, while the HSRL ATLID measures separately Mie and Rayleigh backscatters and thus provides the particulate depolarization ratio.
- L. 200: multiple scattering within the cloud is generally not noticeable in ground-based lidar measurements
- L. 215-225: Around this section, it would be helpful to document the variability of depolarization ratios in various SNR and target situations at the spatial and temporal averaging used by the authors. Figure 2 shows that the authors trust their values of depolarization are accurate within ±0.02, which is quite precise. Since the CONIOPOL algorithm is threshold-based, documenting the variability of depolarization would help the readers understand the reliability of classification output.
- L234-237: I understand that the threshold of attenuated backscatter used to separate aerosols from clouds is itself based on distributions of attenuated backscatters observed with and without clouds. This sounds circular to me. I don't understand how the presence of clouds is decided in the first place? Is it based on other observations, or on the lidar measurements themselves? If so, cloud detection must be based on some kind of signal threshold. Why not use it to inform the design of the CONIOPOL algorithm?
- L. 278 and below: "LDR... cannot uniquely identify aerosol type on its own (...) aerosols are thus classified into an explicit category". I am confused by this section, and in particular this paragraph. First it states that LDR cannot be used to explicitly identify aerosols, since different aerosols can generate overlapping LDR distributions. It then concludes that aerosols are classified using explicit categories, which I understand to be in total contradiction with what was explained before. I understand that methodological choices are made to minimize misclassifications, but minimizing misclassifications is not the same as eliminating misclassifications. I find this section even more puzzling further down, when it addresses the fact that "dust and volcanic ash exhibit overlapping LDR values" but somehow argues that it is still possible (and even mandatory) to treat them as distinct categories. From these elements I would argue that the algorithm should, instead, treat them as the same category ("dust or volcanic ash"). Please make your point clearer here.
- L. 289-294: overlapping signals for different aerosol species could be taken into account by attributing a probability for each species, for instance depending on the LDR distribution of a given aerosol type. The overlapping part could also be labelled as "volcanic ash or dust", while the clearly distinct parts could be labelled "ash" and "dust".
- L. 337-345: The article does not include the necessary information to understand what is being explained here. What exactly is the input of DBSCAN? It is "applied to a 2D grid" of what? What is the output of this algorithm? Is the DBSCAN algorithm part of the design of the CONIOPOL algorithm? If so, why isn't it mentioned in Figure 1? What is the principle of that DBSCAN algorithm? Why is it helpful?
- L. 399-402: I don't understand the point being made here.
- L. 454: "model simulations" my understanding is that CAMS forecast are not mere model simulations, but analyses that merge (through numerical assimilation) model simulations with observations of the atmosphere. One goal of CONIOPOL retrievals could be their assimilation in CAMS.
- Figure 8: I don't understand how the frequencies shown here were calculated. Is it the ratio between the number of profiles containing a given particle type, divided by the number of profiles sampled over the course of a given month? If so, the extremely small number of ice crystals is surprising. Ice clouds are fairly frequent throughout the year.
- Figure 9: It would have been interesting to see the evolution of the cumulative fraction of ice within clouds (compared to ice + liquid) with temperature. One would expect it to remain constant with time.
- Fig. 10 B and D : these subplots are not particularly useful. The small number of values they show could be provided directly in the text, or integrated within subplots A and C.
- Fig. 11: I do not see any significant difference between the values shown in the left column (detection of smoke /marine) and the values shown in the right column (no detection). Please clarify in the text what you mean by "clear differences emerge" or find a way to make these differences clearer visually.
- L. 684 : "EarthCARE's ATLID polarisation lidar will provide vertically resolved particle-type information that single-wavelengths ALC cannot retrieve" -- ATLID itself is single-wavelength. Its added value lies in its high spectral resolution, which is able to separate particulate from molecular backscatter (and thus simplifies lidar ratio retrievals). ATLID being in orbit since june 2024, its aerosol type retrievals are available today, and comparisons between the CONIOPOL output presented here and ATLID products could be made today by identifying a few ATLID overpasses of Uccle.
Citation: https://doi.org/10.5194/egusphere-2026-948-RC3
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 215 | 89 | 18 | 322 | 20 | 16 |
- HTML: 215
- PDF: 89
- XML: 18
- Total: 322
- BibTeX: 20
- EndNote: 16
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
General comments:
The manuscript presents a method on target categorization of various aerosol, cloud and precipitation types with a focus to aerosol types, utilizing the attenuated backscatter coefficient (ATB) and volume depolarization ratio (VDR) at 910 nm from a Vaisala CL61 ceilometer located at Uccle, Belgium. The classification is based on fixed thresholds defined in a decision tree. The results are then evaluated using aerosol mixing ratios from CAMS model for the various aerosol types. The manuscript is timely and relevant, polarization-based classification algorithms are genuinely useful and single-wavelength lidars with polarization capability are increasingly utilized by the community. Despite the fundamental limitations of this approach, using 1 ATB + 1 VDR for target classification can still provide long‑term insights and help build a coherent climatology of aerosols and clouds, which can be further applied to observational networks such as EUMETNET E‑PROFILE and ACTRIS. The manuscript is well organized, supported by relevant literature, and offers a comprehensive discussion of the algorithm’s strengths and limitations. Nonetheless, several passages are overly speculative, and some sections contain unnecessary repetition. The manuscript is suited for publication in Atmospheric Measurement Techniques and can be published after addressing the comments/questions listed below.
Specific comments:
1. The algorithm uses a decision tree to classify the various targets. At the moment, there is no inclusion or even discussion on the uncertainties associated with the ATB and VDR, the two essential parameters used in the classification, and their effect in the classification. These parameters undoubtedly influence the results of the target classification, and therefore, a comprehensive sensitivity analysis is needed. The authors mention that calibrating the CL61 instrument is essential. This can be done within 5-10 % uncertainty (O’Connor et al., 2004, Hopkins et al., 2018, Looschelders et al., 2025). In addition, temporal averaging (and potentially vertical averaging) modifies the measured parameters and may interfere with the hard‑set thresholds used in the classification. This raises two important questions: How sensitive is the classification to calibration biases and temporal averaging, and which parameter is the most sensitive? The manuscript would benefit from explicit recommendations and a clear discussion of its limitations from this perspective.
O'Connor, E. J., A. J. Illingworth, and R. J. Hogan, 2004: A Technique for Autocalibration of Cloud Lidar. J. Atmos. Oceanic Technol., 21, 777–786, https://doi.org/10.1175/1520-0426(2004)021<0777:ATFAOC>2.0.CO;2.
Hopkin, E., Illingworth, A. J., Charlton-Perez, C., Westbrook, C. D., and Ballard, S.: A robust automated technique for operational calibration of ceilometers using the integrated backscatter from totally attenuating liquid clouds, Atmos. Meas. Tech., 12, 4131–4147, https://doi.org/10.5194/amt-12-4131-2019, 2019.
Looschelders, D., A.Christen, S.Grimmond, et al. 2025. “Inter-Instrument Variability of Vaisala CL61 Lidar-Ceilometer's Attenuated Backscatter, Cloud Properties and Mixed-Layer Height.” Meteorological Applications32, no. 5: e70088. https://doi.org/10.1002/met.70088.
2. CAMS is used for validation, but the manuscript does not include studies showing known mismatches between CAMS and lidar measurements in smoke and dust cases. Furthermore, the comparison is carried out by partitioning the atmosphere into two regions: one below 1 km and one above from which conclusions are drawn. I would suggest including a more detailed and statistical meaningful comparison between the model aerosol mixing rations and the lidar-based target classification. One possible improvement would be to include heatmaps showing the hit/miss ratios for the different aerosol and cloud categories, add a seasonal comparison, and segment the atmosphere in more detail, for example, to capture the high-level clouds as well. Although it is difficult to find an aerosol target classification available for the validation, the authors could consider published long-term aerosol classifications within Europe from EARLINET-ACTRIS networks. Cloudnet could provide some insights into the cloud/precipitation typing. Have the authors considered using the target classification from the closest Cloudnet station?
3. Figure 1 appears rather early, without prior discussion of the choice and origin of the values shown, and before the requirements of the algorithm have been introduced. I recommend moving Figure 1 and the introduction of the typing scheme after Section 4.3.
4. Line 166: The minimum useful height in CL61 instrument is 40-50m above ground level according to Roininen et al. (2025) and Le et al., (2026) while there are a few studies reporting an artifact located between 80 and 150m that is not alwayspresent (Filioglou et al., 2025, Ve et al., 2026). How the selection of 33.6m is justified?
Roininen, R., Tuononen, M., Bircher-Adrot, S., Rüfenacht, R., O’Connor, E., Buxmann, J., Kotthaus, S., Van Hove, M., Diémoz, H., BELLINI, A., Fenner, D., Looschelders, D., Geiß, A., Wagner, F., and Mattis, I.: SOP CL61, Zenodo, https://doi.org/10.5281/zenodo.14833761, 2025.
Le, V., O'Connor, E. J., Filioglou, M., and Vakkari, V.: Operational performance of the Vaisala CL61 ceilometer for atmospheric profiling, EGUsphere [preprint], https://doi.org/10.5194/egusphere-2025-6331, 2026.
Filioglou, M., Tiitta, P., Shang, X., Leskinen, A., Ahola, P., Pätsi, S., Saarto, A., Vakkari, V., Isopahkala, U., and Komppula, M.: Lidar estimates of birch pollen number, mass, and CCN-related concentrations, Atmos. Chem. Phys., 25, 1639–1657, https://doi.org/10.5194/acp-25-1639-2025, 2025.
5. Line 173: The cloud base height reported in the ceilometer files uses the signal strength and strength of the change to determine the presence of a cloud. While in liquid clouds the condition is easily met, in optically thinner clouds, like cirrus, the cloud base height is shifted upwards, or the cloud can even go undetected if the visibility is not compromised (see Figure 7). In addition, a challenge which is also present in multi-wavelength lidar-based target classifications, is that intense aerosol layers (volcanic ash, dust, pollen) may be falsely classified as clouds, specifically as ice clouds. Although the authors chose to limit the analysis to the lowest 5 km, thereby excluding most high‑level clouds, it would be important to address the long‑term bias this constraint may introduce by the algorithm, if any. For example, in Figure 7 there are some high‑level clouds showing variable LDR values at 23.30 UTC with no detected cloud base. Nevertheless, the algorithm still classifies these bins as clouds. This should be clarified.
6. Figure 2: Could the authors clarify how many height bins within the liquid cloud layer were considered, and what altitudes are represented in the figure?
7. Line 215: Earlier in the text and again at the end of this paragraph, it is stated that the profiles are temporally averaged over 30s, yet Line 215 indicates that a 30min window is used for noise estimation. Why was such a long time interval chosen?
8. Lines 256-261: The gauge and CL61 observations have temporal resolutions of 1 h and 10 min, respectively. The authors attribute part of the discrepancy to these differences in temporal resolution. However, why were the CL61 observations not averaged to match the gauge’s 1‑hour resolution, rather than speculating about whether the mismatch arises from this source? A brief justification would strengthen the interpretation.
9. Section 4.4: LDR plays a central role in the classification, and published values at 910 nm are still limited. It would improve clarity in this section if the authors explicitly indicated which values refer to 910 nm and which are taken from studies at other wavelengths.
10. Line 289: Pollen types are also overlapping (Cao et al., 2010).
Cao, X., Roy, G. A., and Bernier, R.: Lidar polarization discrimination of bioaerosols, Opt. Eng., 49, 116201, https://doi.org/10.1117/1.3505877, 2010.
11. Line 295: How is the fresh vs aged smoke considered in the algorithm? Could the authors provide some insights?
12. Line 355: Which microphysical properties you are referring to?
13. Line 356: Are temperature profiles available for this case? The liquid layer is most probably in sub-freezing temperatures as there is a melting layer at around 3 km modifying the LDR values (see for example, at 4 and after 6 UTC). In this context, some solid precipitation is being misclassified as dust (around 4 UTC). Also, some thin precipitation bits are missed in the classification but that is inevitable given their high temporal variability.
14. Sect 6.1.1: It is unclear how the percentages are calculated and even more importantly why they can exceed 100%.
15. Line 472: ‘…..above 1000m’. What is the upper limit of this comparison? Is it the 5 km as mentioned earlier or higher?
16.Line 549: ‘The observed difference in total dust mixing ratio detected by CONIOPOL…atmospheric processes.’ The algorithm itself is not providing the dust mixing ratio. Rephrase the sentence.
Technical corrections:
The same acronym is defined more than once in the manuscript. For example, the LDR acronym has two different definitions. It is the Linear Depolarization Ratio (Line 15, Line 50, Line 87) or the Volume Linear Depolarization Ratio (Line 156). The Volume Linear Depolarization ratio is used here, and the text should be updated throughout to avoid any confusion with the particle linear depolarization ratio.
Line 39: Reference is missing
Line 46: Reference is missing
Line 93-94: Indeed! Two great examples are CALIOP and currently ATLID!
Line 185: “Calibrated in manufactory” → calibrated by the manufacturer or calibrated at the Vaisala factory.
Lines 309-401: Unclear sentence. Consider rephrasing.