the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 02 Jul 2026
A Threshold-Based Method for Cloud Base Height Detection using Ceilometers: Application to Long-term observations in deriving Cloud Vertical Structure
Abstract. Ceilometers are widely used for cloud base height (CBH) detection, primarily through proprietary manufacturer algorithms. Although these algorithms are routinely employed at airports and meteorological stations, their suitability for climatological applications is often limited under complex atmospheric conditions. In this study, we propose an improved threshold-based detection (TBD) method for CBH retrieval that is applicable to both calibrated and non-calibrated ceilometer return signals. The detected cloud layers show good agreement with collocated observations from space-borne active sensors. A comparative analysis between the manufacturer’s algorithm and the proposed TBD approach demonstrates significant improvement in CBH separation. The method is further shown to be adaptable to ceilometers of different makes operating under diverse environmental conditions. The TBD approach is applied to long-term observations (April 2020 – October 2025) from a CL51 ceilometer deployed in the coastal urban environment of Kolkata, eastern India, to investigate cloud processes and characterize cloud vertical structure. A parameter termed normalized cloud occurrence is estimated for single-, double-, triple-, and all-layer cloud cases to qualitatively examine cloud vertical distribution. The close similarity between the occurrence patterns of single-layer and all-layer clouds indicates the dominance of single-layer clouds over the study region, while multi-layer cloud occurrences provide additional insight into cloud vertical structure. Seasonal and diurnal analyses reveal the persistent presence of low-level clouds (< 2 km) throughout the day across all seasons. The CBH of low-level clouds gradually increases after 09:00 local time, peaks during 12:00–15:00, and subsequently decreases, likely driven by solar-heating-induced convection. Such convection facilitates vertical cloud development up to 8–12 km, depending on the season, except during winter. Additionally, a persistent elevated cloud layer near 4 km is observed, likely associated with temperature variations around the 0 °C isotherm. The derived cloud vertical structure has important implications for understanding cloud radiative forcing and improving atmospheric model predictions.
- Preprint
(1124 KB) - Metadata XML
-
Supplement
(921 KB) - BibTeX
- EndNote
Status: final response (author comments only)
-
CC1: 'Comment on egusphere-2026-3317', Daniel Klaus, 17 Jul 2026
-
AC4: 'Reply on CC1', Ravi Kiran Varaha, 10 Sep 2026
I appreciate the authors' efforts in presenting this study and agree that the topic is relevant for the AMT readership. However, after reading the preprint, I have several concerns regarding the methodology, comparison data and the interpretation of the results. In my view, the following issues should be addressed before the conclusions can be considered sufficiently supported.
We would like to thank for going through our manuscript and providing comments/suggestion along with detailed information, references. Please find below point-by-point replies for the comments/suggestions.
· Major comment 1 – Assets
In the section “Code and data availability” (p. 25) the authors note that both the code and the data will soon be made available in a public repository. This should be done as soon as possible and is essential to clarify which CL51 data were used for the cloud base height (CBH) – as set out in Major comment 2. In addition, this should make it possible to verify the actual origin of the satellite-based CBHs (see Major comment 4).
The TBD algorithm, data shall be provided along with the revised manuscript. We reiterate that L2 data (16sec temporal and 10m vertical) from CL51 was utilized in the analysis.
· Major comment 2 – CL51 cloud base heights
In Sect. 2.3.4., the authors describe the manufacturer’s algorithm for CBH detection in the CL51, specifically the Sky Condition Algorithm (SCA). However, the Vaisala CL51 provides instantaneous cloud base detections (“cloud hits”) independently of the SCA.
Although we mentioned about the Sky Condition Algorithm (SCA) in the original manuscript, cloud hits from CL51 were used for comparison with TBD method. The statements regarding SCA are now removed in revised manuscript.
The CL51 cloud detection algorithm is applied to a single backscatter profile (typically every 16 s but depending on output configuration) and searches for significant increases in backscatter that meet its cloud detection criteria kept secret. According to the instrument documentation (Vaisala, 2026a) the backscatter profile is processed and cloud bases are detected directly, with up to three CBHs reported simultaneously.
We agree detailed information on manufacturer’s algorithm is not readily available.
The CL51 SCA is documented by Vaisala (2026b) and is used, for example, in the study by Šálek et al. (2019). It takes those cloud hits accumulated over time (moving 30 min, where latest 10 min weighted twice) and determines the heights and cloud cover of at most five layers. The Combine module is the stage that converts the many cloud clusters produced by the Clusterize module into a smaller set of candidate cloud layers. Rather than simply merging adjacent clusters, it performs a moving-window search for the altitude ranges that contain the greatest cloud cover. The key idea is not clustering cloud hits – that has already been done. Instead, it is aggregating nearby clusters into meteorologically meaningful cloud layers by maximizing total cloud cover within an adaptive vertical window. This produces a concise list of candidate layers, which the subsequent Select module then refines (rounding heights, merging very close layers, assigning high-cloud cover, and limiting the output to at most five reported layers).
We thank the reviewer for providing the valuable information on SCA.
I encourage the authors to clarify which CBHs they used from the CL51. The comparison with the proposed threshold-based detection (TBD) method must be based on the cloud hits.
We reiterate that our TBD method utilizes profiles of calibrated total attenuated backscatter coefficients derived from 16 sec temporal resolution and 10 m vertical resolution (L2 data from Vaisala CL51) range corrected signals.
Our cloud base heights (TBD method) were compared against CL51 cloud base hits reported in L2 data.
· Major comment 3 – TBD method
As discussed and demonstrated by Van Tricht et al. (2014), the choice of the backscatter coefficient threshold for determining the CBH depends on the specific application. Under Arctic conditions with low aerosol loading, they found an optimal value of 3 ∙ 10−7m−1sr−1 for the attenuated backscatter coefficient (𝛽att). According to ISO 28902-4 (2026), particularly Tables 1 and 5, the threshold range of 𝛽att is 2 ∙ 10−4m−1sr−1 ± 15 % (1.7 ∙ 10−4m−1sr−1 … 2.3 ∙ 10−4m−1sr−1) for water clouds below 10,000 ft (3,048 m). In Sect. 2.3.5, the authors must comprehensibly explain how they determined their threshold value of 1 ∙ 10−5m−1sr−1 and how it can be justified.
Note that we arrived at this threshold value of total attenuated backscatter coefficient based on the visual inspection of cloud base from TAB contour plots. However, we have now performed statistical analysis on ‘threshold value’. The Results are updated in the revised manuscript.
The proposed TBD method requires a cloud layer thickness of at least 50 m, which corresponds to 5 range bins at a vertical resolution of 10 m. This value is arbitrary and certainly not universally applicable. A cloud layer can be only a few meters thick and still be considered a cloud. This is common for shallow lifting fog, thin stratus, or small patches of cumulus. Van Tricht et al. (2014) increased their minimum cloud thickness from 50 m to 90 m during the AMT interactive discussion stage (www.atmos-meas-tech-discuss.net/6/C4550/2014/). Is it a coincidence that your value is also 50 m?
We performed sensitivity studies on cloud thickness value including the cloud layer separation and revised the manuscript accordingly.
The requirement that the CBH must be consistently detected in at least four consecutive profiles within the range of 𝑡 ± 2 cannot be applied to real-time measurements, as future time steps are unknown.
As mentioned already, the proposed method is not intended for real-time cloud base detection. We want it to be implemented on recorded data, and on long-term observations to understand Climate variability/Trend analysis. However, we performed sensitivity studies on temporal/spatial consistency and revised the manuscript accordingly.
I wonder whether the independent precipitation measurements from the KCON’s automatic weather station (e.g., supplementary Fig. S1) were used to formulate points 5 and 6 in Sect. 2.3.5. The authors claim that a cloud thickness of more than 1.5 km clearly indicates a precipitation event and exclude the corresponding backscatter profiles. First, their “visual inspection of several representative cases” is subjective and difficult to follow. Second, this criterion can be considered a necessary but not sufficient condition for many precipitation events. Third, modern ceilometers can reliably detect CBHs during feeble and moderate rain. Only during obscuration conditions (e.g., heavy rain) a vertical visibility (VV) hit may be reported by the ceilometer. Even a CBH with greater uncertainty during precipitation is better than no value, because otherwise the cloud amount would be systematically underestimated.
Precipitation condition is updated in revised manuscript, consistent with the precipitation information from collocated Laser Precipitation Monitor (LPM).
The fact that most near-surface fog layers over the Indo-Gangetic Plain form within the 0 m to 200 m altitude range (lines 264-265) by no means justifies simply ignoring all ceilometer cloud detections up to 200 m. How do the authors intend to capture, for example, rising fog or low-level stratus clouds? At this point at the latest, the TBD method is no longer generally applicable.
To generalize the algorithm we considered the data below 200 m and revised the manuscript accordingly.
Major comment 4 – Comparison with satellite data
In Sect. 2.3.6, the authors refer to the dataset “CAL_LID_L1-Standard-V5-00”. This CALIOP Level 1B data product contains a half orbit (day or night) of calibrated and geolocated single-shot lidar profiles, including 532 nm and 1064 nm attenuated backscatter and depolarization ratio at 532 nm. Cloud characteristics such as cloud top height (CTH), CBH, and vertical structure (cloud layers) are derived from Level 1 data and provided in Level 2 products like “CAL_LID_L2_01kmCLay-Standard-V5-00”. As mentioned in Major comment 1, it is unclear where the CALIOP value of 2837 m for the CBH from the satellite flyover on 09 August 2020 comes from. The corresponding CBH value is missing for the CALIPSO flyover on 21 March 2023, both in the text and in the supplementary Fig. S10.
Thank you for the comment. We have shown the time-height plot of Level-1B attenuated backscatter from the “CAL_LID_L1-Standard-V5-00” product. However, the cloud-layer information (base, top) was obtained from “CAL_LID_L2_01kmCLay-Standard-V5-00” Level-2 1-km Cloud Layer product.
For comparison with our TBD detected CBH, on 09 August 2020, the CALIOP CBH of 2837m was obtained from the Level-2 1-km Cloud Layer product at the nearest overpass latitude–longitude point to our measurement site.
For 21 March 2023, the observed cloud layer showed a weak and inhomogeneous pattern, and the TBD method did not detect a CBH. Thus, we have not mentioned the corresponding CALIOP CBH value. Instead, we presented the time-height plots of CALIOP Level-1B attenuated backscatter and CL51 attenuated backscatter, which show a cloud layer between 8 and 10 km in both observations.
Previous comparisons between CALIOP and ground-based ceilometer or lidar observations (e.g., Dupont et al., 2010) indicate that CBHs agree within a few hundred meters under suitable observing conditions, although larger discrepancies occur owing to sampling differences, horizontal distance, cloud heterogeneity, and attenuation effects. Kim et al. (2011) concluded that CALIOP provides reliable CBHs for thin high-altitude cirrus clouds containing small-sized ice particles, but the cloud profiling radar (CPR) onboard CloudSat has low sensitivity to these clouds due to the use of different wavelengths, 532 nm and ∼3.2 mm (94 GHz), respectively. Further, for multilayer clouds CALIOP has difficulties in determining the cloud vertical structure for thick clouds underlying thin cirrus clouds due to the signal attenuations, whereas the CPR detects the CTH and CBH of both the cloud layers. The current comparison – with two very brief (about 8 s) CALIPSO flyovers and with just one equally brief flyover by the CPR onboard of EarthCARE on 21 June 2024 – is not meaningful enough and is not convincing. While CALIOP can detect CTH with relatively high reliability, the same is true for a ceilometer with regard to CBH. However, the derived cloud thickness – which is based on different assumptions – is generally more of an estimate. In case of Kolkata, a CALIPSO flyover that comes close may still be 10 – 50 km away, depending on the orbit. The detection of the cloud layers by CL51 was undoubtedly carried out simultaneously, but at a considerable horizontal distance. The CALIOP comparison serves merely as a baseline test, but it is by no means an evaluation that proves the TBD method produces “better” CBHs than the internal CL51 algorithm.
The satellite comparison section is revised by highlighting the limitation in both the active sensors.
The authors’ statement (line 308) – “A detailed study on inter-comparison will be taken up as a separate study.” – is unacceptable. It must be part of this article, as this is of fundamental importance to assess the suitability, robustness, quality, accuracy and added value of the proposed TBD method. In addition, the authors should discuss in more details the strengths and weaknesses of the comparison data used (different wavelength, satellite-based vs. surface-based, etc.) and specify any errors or uncertainties in order to better quantify the quality of the internal CL51 algorithm and the TBD method. I am aware of the difficulty of finding a suitable reference dataset. Even if CALIOP can detect CBHs in multilayer situations, Mülmenstädt et al. (2018) developed a technique to estimate the CBH of the lowest cloud in each column. Using ground-based ceilometer data the predicted CBHs and their uncertainty were found to be biased by less than 10 %. This dataset is freely available under https://doi.org/10.1594/WDCC/CBASE at “Deutsches Klimarechenzentrum” (DKRZ). Unfortunately, it only covers the years 2007 and 2008. Since both the source code for the Cloud Base Altitude Spatial Extrapolator (CBASE) algorithm is available and the CALIOP Vertical Feature Mask (VFM) can be downloaded as input data through 30 June 2023, the dataset could be expanded. I highly recommend it for the comparison with CL51 since this reference dataset would be a better “ground truth” for CBH.
Our CL51 data is available since April 2020, while CALIPSO observations are available only until June 2023. Within this short overlapping period, only a couple of CALIPSO overpasses were available within 1° of our measurement site. Considering the limited temporal overlap and number of coincident overpasses, we have revised the satellite comparison section accordingly.
· Major comment 5 – Interpretation and conclusions
The present study proposes a TBD method with a new threshold value for 𝛽att and application of sub-sequent plausibility criteria. However, it is essentially an adaptation to their specific use case and the shortcomings of the CL51 – neither of which is universally applicable. The actual main problem is that there is no universally valid and accepted definition of the CBH or of a cloud in general (e.g., Spänkuch et al., 2022; ISO 28902-4, 2026).
Thanks for pointing out that the manufacturer’s threshold values are also not universally applicable.
The authors concluded that the internal CL51 algorithm: A) “… detects multiple cloud-base heights within the same physical cloud layer (Fig. 5)”. This could simply be a bug in the CL51 SCA (see Major comment 2) that causes the clusters associated with CBH1 and CBH2 not to be merged correctly. The clustering of the "cloud hits" itself can lead to a single physical cloud layer being erroneously split into two.; B) “… produces ambiguous CBH estimates during precipitation events …” This should not be the case with feeble to moderate rain (see Major comment 3) and is perhaps a bug or artifact of the CL51.;
As these cases are repetitively found in data, we believe they are not a bug or artefact of CL51.
- C) “… sometimes reports very low CBH values that are more likely associated with fog layers …” This is a conjecture and needs to be verified with independent measurements. In conditions of visually dense fog, the CL51 should ideally report a VV hit rather than a CBH hit.
Revised TBD method takes care of fog cases.
How does the TBD method perform with very low clouds having CBHs of less than 1,000 ft (ca. 300 m) when all CBHs below 200 m are generally disregarded (see Major comment 3)? The conclusion that the TBD method leads to “a clear improvement in CBH detection” is not sufficiently supported by the presented results – such as the comparison with satellite-based measurements (see Major comment 4). Even though the TBD method is technically applicable to other ceilometer data, it does not necessarily lead to a general improvement for every location like Munich (Fig. 8) or Lindenberg (Fig. S13).
We initiated the revised TBD method from first bin onwards considering the clouds below 200m in the revised manuscript.
The authors attempt to quantify the differences in CBH between the internal CL51 algorithm and the TBD method by comparing the respective differences (CBH2 – CBH1) between the second and first reported cloud layers. Figure 9 shows monthly mean differences using consecutive CBH detections. This procedure raises doubts: A) CBHs recorded by the CL51 below 200 m and during precipitation are intentionally excluded from the TBD method and are therefore not accounted for in the analysis.;
We would like to bring to your notice that, in the original manuscript, we have also ignored cloud hits from CL51 below 200 m and also during precipitation. It implies, corresponding data was involved in comparison between two methods. Note that the concept of 200m does not exists in the revised manuscript.
- B) If the lower cloud layer dissipates between time 𝑡 and 𝑡 + 1, then CBH2 at time 𝑡 corresponds approximately to CBH1 at time 𝑡 + 1. If an additional cloud layer appears between two existing ones at time 𝑡 + 1, then CBH2 at time 𝑡 roughly corresponds to CBH3 at time 𝑡 + 1. The difference (CBH2 – CBH1) for consecutive time steps therefore does not always consider the corresponding cloud clusters; C) Calculating a mean value (hourly or monthly) does not correct A) and B).
Thank you for bringing this interesting point to our notice. We have addressed this issue in the revised manuscript by revising the TBD method.
Significant deviations from the CL51 are inevitable, as the TBD method deliberately disregards precipitation events and CBHs below 200 m. The calculation of a normalized cloud occurrence cannot improve this apples-to-pears comparison. Why don't the authors also remove the respective events from the CL51 data?
We reiterate that we have considered corresponding data only for comparative analysis between algorithms. That means TBD ignores <200 m, the corresponding CL51 is also ignored. Precipitation data is removed in TBD implies the corresponding data from CL51 is also ignored. Note that the concept of 200m does not exists in the revised manuscript.
Overall, are you confident that your conclusions regarding the CL51's performance can be generalized to other ceilometers, even those from other manufacturers?
Our conclusions are purely based on CL51 data (only). However, we believe our method can be applied to all ceilometers.
Deficiencies in CBH detection are attributable not only to the internal algorithm but also to the optics and components used. I recommend examining the individual backscatter profiles. Figure 9 in Wagner et al. (2024) suggests that the CL51 can exhibit a 𝛽att profile that is atypical of a single cloud layer, potentially leading to very small separations between CBH1 and CBH2.
In the revised version, we have worked on individual backscatter profiles only.
Minor comments
Lines 11 – 12: “Although these algorithms …” In aviation, guidelines, requirements, and restrictions, for example from the International Civil Aviation Organization (ICAO), European Union Aviation Safety Agency (EASA), Federal Aviation Administration (FAA) or Civil Aviation Authority (CAA), are far stricter than for scientific studies. Nevertheless, the development and improvement of any ceilometer firmware is definitely not restricted to aviation.
The sentence is removed in revised version of the manuscript.
Lines 65 – 66: “The cloud base is generally identified …” This is only one possible definition of the CBH, not a general one (see Major comment 5).
The word ‘generally’ is removed in the revised version of manuscript.
Lines 97 – 98: “Standard CBH detection algorithms are primarily designed …” That is not quite correct! Ceilometer manufacturers certainly aim for CBH detection to work under all weather conditions. Any threshold approach is usually combined with gradient analysis. The CHM15k preprocesses the backscatter profile using altitude-dependent averaging to improve the signal-to-noise ratio, after which cloud features are identified from the averaged profile using backscatter thresholding and gradient-based boundary detection (OTT HydroMet, 2022). The CS136 (SkyVUE8) essentially requires a sufficiently steep increase in the slope of the extinction profile and applies an extinction threshold corresponding to a meteorological optical range (MOR) of 1,000 m (Campbell, 2018). Weaknesses in the manufacturers' algorithms were gradually addressed – sometimes following consultation – and the firmware was improved. The CHM15k, for instance, detects cirrus clouds quite well, but exhibited weaknesses regarding low-level clouds prior to firmware version 0747.
Quoted sentence is removed in revised version of the manuscript.
Lines 485 – 487: “… manufacturer-provided proprietary algorithms, which are primarily optimized for aviation applications …” Why should modern ceilometers not be suitable for climatological studies? A firmware upgrade or a ceilometer replacement often affects a time series more than a consistent bug.
Quoted sentence is removed in the revised version of the manuscript.
Summary
The present study addresses an interesting and important problem and has the potential to make a valuable contribution. In my view, however, the current version does not yet provide sufficient methodological detail, and several of its central conclusions are not fully supported by the evidence presented. Addressing the points discussed above would substantially strengthen the manuscript.
Thanks for your valuable insights; we refined the manuscript accordingly.
We once again thank for providing potential solutions for improving manuscript content further.
---END---
Citation: https://doi.org/10.5194/egusphere-2026-3317-AC4
-
AC4: 'Reply on CC1', Ravi Kiran Varaha, 10 Sep 2026
-
RC1: 'Comment on egusphere-2026-3317', Anonymous Referee #1, 26 Jul 2026
The study addresses an important aspect of cloud observations by estimating CBH from the Vaisala CL51 ceilometer. The proposed Threshold-Based Method for CBH detection appears promising for long-term cloud observations. However, the method still relies on visual inspection to confirm cloud detection, which limits its applicability for fully automated, continuous observations, particularly at remote or unattended sites. Additionally, the algorithm appears to have limitations in detecting cloud bases below 200 m, which may lead to the omission of low-level cloud events and affect the completeness of long-term cloud climatology.
Comparison with satellite data is not very convincing especially in case of multilayer clouds.
Is TBD method applicable to all ceilometers at different locations or region specific to KNOC
Citation: https://doi.org/10.5194/egusphere-2026-3317-RC1 -
AC1: 'Reply on RC1', Ravi Kiran Varaha, 10 Sep 2026
The study addresses an important aspect of cloud observations by estimating CBH from the Vaisala CL51 ceilometer. The proposed Threshold-Based Method for CBH detection appears promising for long-term cloud observations.
We would like to thank the reviewer for going through our manuscript carefully, appreciating the actual content and providing valuable suggestions/comments. Please find below point-to-point replies to the reviewer comments.
However, the method still relies on visual inspection to confirm cloud detection, which limits its applicability for fully automated, continuous observations, particularly at remote or unattended sites.
The proposed method is not intended for real-time cloud base detection. We want it to be implemented on recorded data, and on long-term observations to understand Climate variability/Trend analysis. However, we performed sensitivity studies on temporal/spatial consistency and revised the manuscript accordingly.
Additionally, the algorithm appears to have limitations in detecting cloud bases below 200 m, which may lead to the omission of low-level cloud events and affect the completeness of long-term cloud climatology.
In the revised manuscript we initiated the TBD method from the first bin itself so the clouds below 200m will not be missed.
1) Comparison with satellite data is not very convincing especially in case of multilayer clouds.
As it is well known, satellite observations are limited by frequency of overpasses having both advantageous and dis-advantages. However, it is still worth to make comparison in absence of other observations. These limitations are mentioned in the revised version of the manuscript.
2) Is TBD method applicable to all ceilometers at different locations or region specific to KNOC
TBD method is developed to be applied to all ceilometers. Note that this method is already applied to different locations and makes in the original manuscript itself.
We once again thank the reviewer for providing potential solutions for improving manuscript content further.
---END---
Citation: https://doi.org/10.5194/egusphere-2026-3317-AC1 -
AC5: 'Reply on RC1', Ravi Kiran Varaha, 10 Sep 2026
The study addresses an important aspect of cloud observations by estimating CBH from the Vaisala CL51 ceilometer. The proposed Threshold-Based Method for CBH detection appears promising for long-term cloud observations.
Response: We would like to thank the reviewer for going through our manuscript carefully, appreciating the actual content and providing valuable suggestions/comments. Please find below point-to-point replies to the reviewer comments.
However, the method still relies on visual inspection to confirm cloud detection, which limits its applicability for fully automated, continuous observations, particularly at remote or unattended sites.
Response: The proposed method is not intended for real-time cloud base detection. We want it to be implemented on recorded data, and on long-term observations to understand Climate variability/Trend analysis. However, we performed sensitivity studies on temporal/spatial consistency and revised the manuscript accordingly.
Additionally, the algorithm appears to have limitations in detecting cloud bases below 200 m, which may lead to the omission of low-level cloud events and affect the completeness of long-term cloud climatology.
Response: In the revised manuscript we initiated the TBD method from the first bin itself so the clouds below 200m will not be missed.
1) Comparison with satellite data is not very convincing especially in case of multilayer clouds.
Response: As it is well known, satellite observations are limited by frequency of overpasses having both advantageous and dis-advantages. However, it is still worth to make comparison in absence of other observations. These limitations are mentioned in the revised version of the manuscript.
2) Is TBD method applicable to all ceilometers at different locations or region specific to KNOC
Response: TBD method is developed to be applied to all ceilometers. Note that this method is already applied to different locations and makes in the original manuscript itself.
We once again thank the reviewer for providing potential solutions for improving manuscript content further.
---END---
Citation: https://doi.org/10.5194/egusphere-2026-3317-AC5
-
AC1: 'Reply on RC1', Ravi Kiran Varaha, 10 Sep 2026
-
RC2: 'Comment on egusphere-2026-3317', Anonymous Referee #2, 30 Jul 2026
Review of "A Threshold-Based Method for Cloud Base Height Detection using Ceilometers: Application to Long-term observations in deriving Cloud Vertical Structure"The authors present a threshold-based cloud base height detection method and apply it to long-term groundbased ceilometer observations. The resulting cloud base height from the detection method presented here differs from the method used by the ceilometer manufacturer. A comparative analysis is performed and long-term statistics of normalized cloud base height occurrence are presented. The study is well-written and the data processing and cloud base detection method is described clearly. Below are general comments that should be addressed in a revised version of the manuscript.General comments1) There is an inconsinstency about cloud base and cloud top (or cloud vertical structure) in the title and throughout the manuscript. The authors discuss cloud tops based on a distribution of cloud base height. The ceilometer provides information on cloud base only and we cannot infer the cloud vertical extent and cloud top without additional information from radiosondes (relative humidity) or ideally cloud radar. The discussion should be updated and account for the lack of cloud top height information in the groundbased ceilometer data that does not allow any interpretation on the vertical extent of clouds. Finally, it is not clear how we can use cloud base height information in radiative transfer simulations as mentioned in the final paragraph.2) It is not clear why the authors state that the method can be applied to uncalibrated ceilometer signals. While I understand that the method simply relies on a threshold, with additional criteria on spacing etc, there is no unique property of this method that corrects instrument biases or drift. This makes it unsuitable for uncalibrated data, unless the stability of the sensor is known - in which case there is also calibrated data available presumably.3) Supplementary material should not be used to provide 14 additional figures the reader gets pointed at throughout the article. These types of figures should be included in the main text or the appendix. Given the limited space, the authors should decide which ones are really needed, as I do not see a need for a supplement here.4) A very simple climatology of normalized cloud base occurrence is shown for two completely different algorithms. It would have been useful to get also information on the absolute cloud base occurrence, as this is relevant for the surface energy balance and water cycle.5) Are any additional data collected on the site? The results chapter shows interesting views on the diurnal cycle that could be complemented with information on lifting condensation level, radiation data, radiosonde data, rainfall, etc. This will help with the interpretation of the statistics that is currently presented in section 3. Ideally, the information from Figure S1 gets shown together with cloud base occurrence.6) The evaluation of the approach is not convincing. There is only a few hours of case studies and a comparison with CALIPSO (which provides cloud top information rather than cloud base). It would be good to show long-term statistics from the new and the existing method.7) How are the thresholds for the TBD method chosen? Obviously, these seem to be guessed based on individual cases and are likely not optimal. It would be interesting to have a sensitivity study to learn, which parameters are most important for the cloud base statistics.Figures:Most figures need a more descriptive caption. Sometimes it is unclear what the authors are showing. In Figure 4, the right panel is not described at all and the reader has to guess what it is. Only Figure 6 appears to have a detailed caption. Also, I strongly encourage the authors to make sure the figures are of appropriate quality for a publication. Units change for the same physical quantity, colorbars get shuffled, height ranges differ within the same figure, or the figures are simply of very low resolution.Citation: https://doi.org/
10.5194/egusphere-2026-3317-RC2 -
AC2: 'Reply on RC2', Ravi Kiran Varaha, 10 Sep 2026
The authors present a threshold-based cloud base height detection method and apply it to long-term ground based ceilometer observations. The resulting cloud base height from the detection method presented here differs from the method used by the ceilometer manufacturer. A comparative analysis is performed and long-term statistics of normalized cloud base height occurrence are presented. The study is well-written and the data processing and cloud base detection method is described clearly. Below are general comments that should be addressed in a revised version of the manuscript.
We would like to thank the reviewer for going through our manuscript carefully, summarizing it and providing the valuable suggestions/comments. Please find below point-by-point reply to the reviewer comments.
- General comments
1) There is an inconsistency about cloud base and cloud top (or cloud vertical structure) in the title and throughout the manuscript. The authors discuss cloud tops based on a distribution of cloud base height. The ceilometer provides information on cloud base only and we cannot infer the cloud vertical extent and cloud top without additional information from radiosondes (relative humidity) or ideally cloud radar. The discussion should be updated and account for the lack of cloud top height information in the ground based ceilometer data that does not allow any interpretation on the vertical extent of clouds.
Thanks for bringing this to our notice. The word ‘cloud vertical structure’ is reframed as ‘vertical distribution of cloud base heights’ in the suggested lines.
Finally, it is not clear how we can use cloud base height information in radiative transfer simulations as mentioned in the final paragraph.
The statement in final paragraph is now revised as “Cloud base height information is essential for carrying out aerosol-cloud interaction studies “.
2) It is not clear why the authors state that the method can be applied to uncalibrated ceilometer signals. While I understand that the method simply relies on a threshold, with additional criteria on spacing etc, there is no unique property of this method that corrects instrument biases or drift. This makes it unsuitable for uncalibrated data, unless the stability of the sensor is known - in which case there is also calibrated data available presumably.
We revised our method focusing (exclusively) on calibrated ceilometer return signal.
3) Supplementary material should not be used to provide 14 additional figures the reader gets pointed at throughout the article. These types of figures should be included in the main text or the appendix. Given the limited space, the authors should decide which ones are really needed, as I do not see a need for a supplement here.
Thank you for the suggestion. We moved the relevant figures to the main text (revised manuscript) while arranging others in Supplementary that is mandate for any journal.
4) A very simple climatology of normalized cloud base occurrence is shown for two completely different algorithms. It would have been useful to get also information on the absolute cloud base occurrence, as this is relevant for the surface energy balance and water cycle.
Thanks for nice suggestion. Absolute cloud occurrence is also shown in the revised manuscript.
5) Are any additional data collected on the site? The results chapter shows interesting views on the diurnal cycle that could be complemented with information on lifting condensation level, radiation data, radiosonde data, rainfall, etc. This will help with the interpretation of the statistics that is currently presented in section 3. Ideally, the information from Figure S1 gets shown together with cloud base occurrence.
Thanks for the suggestion. As suggested. complimentary information is used to interpret the diurnal cycle of CBH.
6) The evaluation of the approach is not convincing. There is only a few hours of case studies and a comparison with CALIPSO (which provides cloud top information rather than cloud base). It would be good to show long-term statistics from the new and the existing method.
As it is well known, satellite observations are limited by frequency of overpasses having both advantageous and dis-advantages. However, it is still worth to make comparison in absence of other observations. These limitations are mentioned in the revised version of the manuscript.
Note that long-term statistics from the new and existing method is already shown.
7) How are the thresholds for the TBD method chosen? Obviously, these seem to be guessed based on individual cases and are likely not optimal. It would be interesting to have a sensitivity study to learn, which parameters are most important for the cloud base statistics.
Sensitivity analysis is performed to justify the threshold value of total attenuated backscatter implemented in TBD method.
- Figures
8) Most figures need a more descriptive caption. Sometimes it is unclear what the authors are showing. In Figure 4, the right panel is not described at all and the reader has to guess what it is. Only Figure 6 appears to have a detailed caption. Also, I strongly encourage the authors to make sure the figures are of appropriate quality for a publication. Units change for the same physical quantity, colorbars get shuffled, height ranges differ within the same figure, or the figures are simply of very low resolution.
Thank you for the suggestion. All the figures are now updated in the revised manuscript.
We once again thank the reviewer for providing potential solutions for improving manuscript content further.
---END---
Citation: https://doi.org/10.5194/egusphere-2026-3317-AC2
-
AC2: 'Reply on RC2', Ravi Kiran Varaha, 10 Sep 2026
-
RC3: 'Comment on egusphere-2026-3317', Anonymous Referee #3, 04 Aug 2026
This paper develops a threshold-based cloud base height (TBD) retrieval algorithm for multi-model ceilometers and applies the method to five years of coastal urban observations in Kolkata to analyze cloud vertical structure. The work collects valuable long-term tropical data, compares the new scheme against the manufacturer’s built-in algorithm, and conducts cross-validation with CALIPSO/EarthCARE satellite measurements and European ACTRIS ceilometer datasets, which brings practical value for ground-based cloud observation research. Nevertheless, the manuscript contains a fundamental observational limitation that undermines its core argument: ceilometers are primarily designed to detect cloud base height instead of complete cloud vertical extent. Laser signals are strongly attenuated within liquid hydrometeors, so the instrument cannot reliably capture cloud top height and full vertical cloud layers. The authors overstate the ability of ceilometer data to resolve full vertical cloud structure throughout the title, abstract and discussion. Besides this central conceptual concern, the paper suffers from insufficient uncertainty analysis, limited statistical intercomparison with satellite data, and incomplete validation of fog/precipitation filtering logic. Substantial revisions to methodology, results interpretation and discussion are mandatory before it can be considered for publication in ACP.
- Ceilometers only capture backscatter signals at the lowest cloud base and lose signal penetration inside thick liquid clouds, making them incapable of accurately constraining cloud top altitude and full vertical cloud thickness. The manuscript repeatedly claims its TBD method can retrieve comprehensive cloud vertical structure, which is misleading to readers. The authors must drastically moderate all overemphasized conclusions about vertical cloud extent, add prominent, repeated caveats on laser attenuation bias, and clarify that only cloud base heights are robust retrievals while multi-layer vertical information carries large unquantified errors. All figures interpreting full vertical cloud profiles need revised explanatory text highlighting this critical limitation.
- The TBD method filters fog (<200 m) and precipitating cloud layers via fixed thickness thresholds, but the authors only provide single case examples without bulk statistical evaluation across the 5-year dataset. No quantitative metrics (hit rate, false positive rate, misclassification ratio) are calculated to compare the fog/precipitation discrimination performance between the native CL51 algorithm and the new TBD scheme. The revision must add statistical screening performance derived from the full observation record to prove the TBD filter’s advantages.
- The intercomparison against CALIPSO and EarthCARE only presents several discrete overpass snapshots, without matching hundreds of collocated ground-satellite profiles for statistical analysis. No correlation coefficients, bias or RMSE between TBD-derived CBH and spaceborne cloud base retrievals are provided. The authors must build a large collocated matching dataset and add quantitative statistical comparison to solidify cross-sensor validation.
- Sensitivity tests are missing for key TBD threshold parameters. The TBD method sets fixed rules for minimum cloud layer thickness, temporal continuity window and allowable CBH vertical variation, but the manuscript does not test how adjusting these thresholds alters the derived vertical cloud separation distances shown in Figure 9. Without sensitivity experiments, it is unclear whether the larger inter-layer gaps from TBD reflect real atmospheric cloud structures or artificial algorithm bias. Threshold sensitivity analysis must be supplemented in the revised methodology.
Citation: https://doi.org/10.5194/egusphere-2026-3317-RC3 -
AC3: 'Reply on RC3', Ravi Kiran Varaha, 10 Sep 2026
This paper develops a threshold-based cloud base height (TBD) retrieval algorithm for multi-model ceilometers and applies the method to five years of coastal urban observations in Kolkata to analyze cloud vertical structure. The work collects valuable long-term tropical data, compares the new scheme against the manufacturer’s built-in algorithm, and conducts cross-validation with CALIPSO/EarthCARE satellite measurements and European ACTRIS ceilometer datasets, which brings practical value for ground-based cloud observation research.
We thank the reviewer for going through our manuscript carefully, summarizing it and providing deeper insights for the improvement of our manuscript. Please find below point-by-point replies to the reviewer comments.
Nevertheless, the manuscript contains a fundamental observational limitation that undermines its core argument: ceilometers are primarily designed to detect cloud base height instead of complete cloud vertical extent. Laser signals are strongly attenuated within liquid hydrometeors, so the instrument cannot reliably capture cloud top height and full vertical cloud layers. The authors overstate the ability of ceilometer data to resolve full vertical cloud structure throughout the title, abstract and discussion.
Thanks for bringing this to our notice. The word ‘cloud vertical structure’ is reframed as ‘vertical distribution of cloud base heights’ as also suggested by another reviewer.
Besides this central conceptual concern, the paper suffers from insufficient uncertainty analysis, limited statistical intercomparison with satellite data, and incomplete validation of fog/precipitation filtering logic.
TBD method is thoroughly revised (including uncertainty analysis) accounting for fog and precipitation. Limitations in satellite inter-comparison are highlighted.
Substantial revisions to methodology, results interpretation and discussion are mandatory before it can be considered for publication in ACP.
Results and discussion are revised thoroughly as suggested.
1) Ceilometers only capture backscatter signals at the lowest cloud base and lose signal penetration inside thick liquid clouds, making them incapable of accurately constraining cloud top altitude and full vertical cloud thickness. The manuscript repeatedly claims its TBD method can retrieve comprehensive cloud vertical structure, which is misleading to readers. The authors must drastically moderate all overemphasized conclusions about vertical cloud extent, add prominent, repeated caveats on laser attenuation bias, and clarify that only cloud base heights are robust retrievals while multi-layer vertical information carries large unquantified errors. All figures interpreting full vertical cloud profiles need revised explanatory text highlighting this critical limitation.
The limitation in cloud top estimation is highlighted in the revised manuscript. The words ‘cloud vertical structure’ reframed as ‘vertical distribution of cloud base heights’. The corresponding discussion is also changed.
2) The TBD method filters fog (< 200 m) and precipitating cloud layers via fixed thickness thresholds, but the authors only provide single case examples without bulk statistical evaluation across the 5-year dataset. No quantitative metrics (hit rate, false positive rate, misclassification ratio) are calculated to compare the fog/precipitation discrimination performance between the native CL51 algorithm and the new TBD scheme. The revision must add statistical screening performance derived from the full observation record to prove the TBD filter’s advantages.
TBD method now starts from first range bin (profile up to 200 m is also considered), fog, and rain are discriminated in the profile. Statistical evaluation of fog/precipitation discrimination among the two methods is included in the revised manuscript.
3) The intercomparison against CALIPSO and EarthCARE only presents several discrete overpass snapshots, without matching hundreds of collocated ground-satellite profiles for statistical analysis. No correlation coefficients, bias or RMSE between TBD-derived CBH and spaceborne cloud base retrievals are provided. The authors must build a large collocated matching dataset and add quantitative statistical comparison to solidify cross-sensor validation.
We don’t think the available number of satellite overpass are sufficient in number to estimate the statistical factors such as correlation coefficient, RMSE between TBD-derived CBH and space borne cloud base retrievals. Therefore, statistical analysis is not included the revised manuscript. However, we think it is relevant to show availability of satellite data in this part of the world where satellite correlative data is sparse.
4) Sensitivity tests are missing for key TBD threshold parameters. The TBD method sets fixed rules for minimum cloud layer thickness, temporal continuity window and allowable CBH vertical variation, but the manuscript does not test how adjusting these thresholds alters the derived vertical cloud separation distances shown in Figure 9. Without sensitivity experiments, it is unclear whether the larger inter-layer gaps from TBD reflect real atmospheric cloud structures or artificial algorithm bias. Threshold sensitivity analysis must be supplemented in the revised methodology.
As suggested we have performed the sensitivity analysis in the revised manuscript.
We once again thank the reviewer for providing potential solutions for improving manuscript content further.
---END---
Citation: https://doi.org/10.5194/egusphere-2026-3317-AC3
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 294 | 90 | 31 | 415 | 56 | 23 | 16 |
- HTML: 294
- PDF: 90
- XML: 31
- Total: 415
- Supplement: 56
- BibTeX: 23
- EndNote: 16
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
The present study addresses an interesting and important problem and has the potential to make a valuable contribution. However, I believe that the current version would benefit from additional methodological detail and stronger support for several of its central conclusions. Addressing the points discussed in the attached PDF would, in my opinion, substantially strengthen the manuscript.