Comparison of SAFNWC/MSG Satellite Cloud Type with Vaisala CL51 Ceilometer-Detected Cloud Base Layer Using the Sky Condition Algorithm and Vaisala BL-View Software
Round 1
Reviewer 1 Report
Comparison of sky condition of Vaisala CL51 ceilometer with SAFNWC-MSG satellite cloud type in National Atmospheric Observatory in Kosetice in the Czech Republic
by Milan Šálek and Beáta Szabó-Takács
The authors describe a one year intercomparison study of cloud information retrieved from eye observations (SYNOP), satellite data (MSG/SEVIRI) and a ceilometer (Vaisala CL51) at Kosetice/Czech Republic. Although this topic is not new in the field of cloud (parameter) observations, the article contains relevant information about the performance of the CL51 instrument, despite the fact that arguments are unluckily formulated. There are however several severe issues with this paper which need to be clarified before it can be published. I therefore recommend a major revision before publication.
General remarks:
1) Usually a scientific paper raises a question or a hypothesis that is further answered or proofed/rejected by the study. What is the goal of this paper and could you formulate the open question that is answered by this study ? This should go into the abstract.
2) The paper suffers from missing (new) scientific content. A simple intercomparison of ceilometer data with satellite data and SYNOP records for just one year at just one place is rather limited and conclusions drawn from such a study are not applicable in general. I therefore urge the authors to add results of more ceilometers at other places to the study and to enhance the time frame. There are e.g., also Cl51 instruments at Cheb, Churanov, Doksany, Karlov, Kocevolice, Pec pod Snezkou, Pribyslav, Serak, Usti nad Orlici.
3) There are, in my view, wrong conclusions and arguments à see under specific remarks. These must be clarified before the paper can be published.
4) Although not being a native speaker myself I recommend a major revision of the English language. There are too many unclear statements and possibly misleading expressions (or wrong usage of expressions) and erroneous grammar structures in this paper.
Specific remarks:
1) L 61, first sentence: A ceilometer is at first an instrument for detecting clouds. Please change the order of arguments. First the cloud parameters then all the other stuff.
The detection of the boundary layer (height) by a ceilometer is another issue but there is still no standard algorithm available for doing this. Every manufacturer cooks its own soup and most NMHSs just apply these algorithms (for possible problems originating from firmware algorithms see e.g., the paper by Kotthaus et al., AMT, 2016). Furthermore, the boundary layer height retrieved from a ceilometer is typically retrieved from the gradient(s) in the backscatter profile, thereby differing form the boundary layer height calculated by a model. So different scientific communities have different things in mind when they talk about the “boundary layer”.
Aerosol backscatter profiles aren’t mentioned (I recommend adding this !) but this capability is probably more important than the retrieval of the boundary layer height, since one retrieves a physical quantity from a calibrated ceilometer (the attenuated backscatter coefficient).
2) L 67, cloud type: As far as I remember the cloud type is defined being something like an ice cloud, a water cloud, a mixed-phase cloud, a stratiform or a convective cloud. It might be better to use cloud layer (type, height) instead of cloud type, in order to avoid misunderstandings with the common meaning in the community.
3) Using the abbreviation “LMH” for “high clouds above medium or low-level clouds” isn’t very plausible. More to the point would be “HML” or “HaML”.
4) L 86ff: A SYNOP record of 0 octa really means “no clouds” in the entire upper hemisphere whereas clouds not overpassing the ceilometer are simply not recognized. So a ceilometer may tell you “no clouds” whereas the SYNOP record says X octa, and both are right !. I’m therefore not sure what you want to say in L.89f. Which cloud information was used as “ground truth” ? The one from the ceilometer or the one from the SYNOP record ? Please rethink and reformulate the paragraph.
5) L96ff: A few more explanations about the temporal collocation would be helpful. Readers don’t want to be forced to read another paper before it becomes clear how the collocation was performed. To say it with a genius of past times: “Everything should be made as simple as possible, but not simpler“
6) An explanation of the spatial collocation is missing! MSG/SEVIRI pixels over Europe are already large, a few square kilometres, and therefore it is worth to mention if a single pixel is used for the intercomparison or a 3x3 or 5x5 matrix ? Please add that information.
7) What about MSG/SEVIRI’s pointing stability/accuracy, what about the parallax problem, especially for the geographical assignment of vertically extended clouds relative to the ground ? Nothing is said about but it may affect the entire study. Please add this information to the paper since it is essential for interpreting and judging the results.
8) Eqs 1) to 6): Sum from 1 to 5 or from 1 to 6 ? There are six “cloud types”.
9) In table 1: “low medium height” à better to say “low medium high”
10) Eq 5) array sij needs to be defined and explained. A “penalty array” can be anything.
11) L 144: “The low, medium and the high above medium and/or low clouds types were overestimated by ceilometer compared to satellite observation.”
This is a rather strange interpretation of results. The ceilometer is by far more sensitive to clouds than the satellite instrument, at the spatial scale of the ceilometer of course. So it is much better to conclude that the satellite instrument underestimates at least the low and medium clouds (at the place of the ceilometer), which is presumably simply a matter of the pixel size !, and the satellite instrument might have problems detecting low/mid level clouds under the presence of highs clouds. On the other hand, depending on the pixel size the satellite instrument should be able to detect high clouds (split window method) above lower clouds better than a ceilometer. It depends largely on the optical thickness of low/mid-level clouds if the backscatter signal penetrates through these clouds or not. This paragraph needs rethinking and must be rewritten.
12) L 146: “Despite this large bias, the probability of detection of medium clouds was larger in the summer period (0.69) compare to its value in the winter time (0.23).”
I think this is a trivial result which is driven by the fact that low clouds are more often observed during the winter time thus hampering the detection of mid-level clouds. Please reformulate the paragraph, add more substance to the findings, and discuss the results.
13) L 156: “The low clouds were detected the most accurately in the winter period (95%).”
Same trivial statement as above. Fog situations will occur much more frequently during winter time which makes it easy to detect them from ceilometer backscatter data. Either you get a distinct signal from a low-level cloud or you get just nothing useful (extremely noisy backscatter), when the instrument stucks in the fog.
14) L 156 ff: “The significant false detection in M cloud type category partly can be explained by the fact that the very low clouds may be classified as medium clouds in case of strong thermal inversion (and thus of low cloud top temperature) by SAFNWC/MSG.”
So the false detection of mid-level clouds by a ceilometer is a result of the misclassification of the NWC-SAF software, thus the satellite data ? Again, it is unclear which measurements are taken as the reference: Is it the ceilometer data or is it the satellite data ? What is the spatial scale we are interested in ? The limited space where a ceilometer is representative for or the larger spatial scale of the satellite instrument ?
It is the satellite instrument that has a problem here ! Rethink the arguments and reformulate the paragraph.
15) L 162ff: “The 12% occurrence of ceilometer detected high cloud in the cloud-free category of the satellite can be originated by the elusive cloud signal of CL51 when a transparent vapour level exists in upper air."
Sorry, this is completely odd and you dray the wrong conclusions: The strong backscatter of cloud droplets or ice crystals (high clouds) would certainly be recognized by the ceilometer’s detector. The fact that the satellite retrieval identifies a scene as cloud-free does not necessarily mean that there are no clouds covering the footprint. All satellite instruments and especially those with relatively large footprints such as the SEVIRI instrument are hampered and affected by sub-pixel cloudiness. And the ceilometer will detect these sub-pixel clouds when they pass over ! The CL51 has indeed a water vapour problem which manifests however mostly for aerosol retrievals while the cloud detection will not be affected.
Another likely explanation for the “underestimation” of high clouds are the firmware settings of the CL51. Have you analyzed the firmware (version) and the related settings ? If not, please do so and report it anyway here.
16) L 167ff: “The most of the LMH, 99% in winter and 97% in summer, was detected during night-time by ceilometer which resulted in very poor POD and large FAR.”
So the ceilometer does not detect “high clouds above other clouds” during night-time.
Fig 1 shows that in most cases when the satellite detects “high clouds above other clouds” the ceilometer detects lows clouds. Well, this is not surprising. The reason is most likely and simply the viewing geometry combined with dense clouds (beam blocking effect) closer to the ground (e.g., a wintertime stratiform cloud layer close to the ground). Fig. 2 shows the summer period and again the result is plausible since cloud layers are, on average, higher than during the winter. Consequently the majority of ceilometer measurements indicate mid-level clouds (for class LMH) and again, due to the viewing geometry, the satellite may see cloud layers above.
I think it is misleading to call this “a poor POD” because it is mostly the differing observation geometry.
17) L 170ff: “The small amount of total cloudiness (< 3 octas) was determined as a fractional cloud by ceilometer. This could be only clouds in mid- or high level which contained ice on the top of the cloud.”
Sorry again, this is confusing. Fractional clouds from a ceilometer as you call it are assigned if the coverage is below 3 octa. This has nothing to do with ice caps on top of high clouds. I don’t understand this argument.
18) Figures 1 and 2 deserve a lot more explanation than given. The y-axis shows “Frequency in %” and this quantity is not defined/explained. It seems to be POD x 100 ?
Fig 1 x-axis: LHM --> LMH or even better HaLM
19) L 203: “These results suggest that CL51 have a strong noise in high cloud detection.”
I don’t believe this conclusion at all. Besides the above-mentioned firmware problem, which influence on the cloud detection in upper levels is unknown, it is much more the problem of different cloud structures at different spatial scales that are seen by the satellite and the ceilometer. The split window technique applied to SEVIRI data for detecting high and optically thin clouds works on pixel basis and the satellite pixel is huge compared to the needle pin of the ceilometer. So especially the results for JJA and high clouds aren’t surprising (see also item 15 and the sub-pixel cloudiness problem). During winter time I expect that the threshold algorithm of the NWC-SAF assigns too often high clouds in the colder winter atmosphere. The satellite sensors detect the brightness temperature (of clouds) and later in the retrieval the height is assigned to cloud layers by assuming corresponding T-p-height profiles. Height profiles in wintering atmospheres are often quasi-isotherm, thus making it difficult to properly assign a height level to a brightness temperature.
20) Without going into details I believe that the entire conclusions section must be revised and arguments must be newly formulated and properly discussed. See all critical points above.
21) L. 195ff: “The difference between the SYNOP and CL51 and SAFNWC/MSG observation can be stemmed 196 from the fact that the SYNOP observation applied a time integration technique for estimating areal average”.
To my knowledge manned SYNOP observation of clouds are done at a certain time, e.g. 7, 14, 21, the classical “Mannheim hours”, whereas a ceilometer signal is typically integrated over time. How should then the SYNOP measurement be integrated over time ?
Comments for author File: 
Comments.pdf
Author Response
Point1: Usually a scientific paper raises a question or a hypothesis that is further answered or proofed/rejected by the study. What is the goal of this paper and could you formulate the open question that is answered by this study ? This should go into the abstract.
Response 1: Thank you for your suggestion. I modified the absract and the manuscript considering your suggestions and the questions of the revised manuscript are:
· Is the high cloud types based on SAFNWC Cloud Type data agree with CL51 detected high cloud layer thanks to the larger range of detection?
· How the range-variant smoothing window of BL-View improves the cloud base height detection?
Point 2: The paper suffers from missing (new) scientific content. A simple intercomparison of ceilometer data with satellite data and SYNOP records for just one year at just one place is rather limited and conclusions drawn from such a study are not applicable in general. I therefore urge the authors to add results of more ceilometers at other places to the study and to enhance the time frame. There are e.g., also Cl51 instruments at Cheb, Churanov, Doksany, Karlov, Kocevolice, Pec pod Snezkou, Pribyslav, Serak, Usti nad Orlici.
Response 2: I agree with you that the comparison with more CL51 ceilometers would give a possibility to draw general conclusions but unfortunately, our department does not have enough budget for getting in CL51 data from other places recently. To compensate this matter I modified the manuscript and I also applied BL-View software produced cloud base layer data to investigate if the range-variant smoothing of aerosol backscatter improves the cloud layer detection? We cannot enhance the time because the ceilometer was applied on a campaign and we have a large amount missing data before 2017.
Point 3:There are, in my view, wrong conclusions and arguments à see under specific remarks. These must be clarified before the paper can be published.
Response 3: We modified the manuscript and CL51 cloud base layer data were applied as reference instead of satellite cloud type data. This change modified the results and new conclusions are drawn.
Point 4: Although not being a native speaker myself I recommend a major revision of the English language. There are too many unclear statements and possibly misleading expressions (or wrong usage of expressions) and erroneous grammar structures in this paper.
Response 4: We have sent the revised manuscript to language proofing to check and correct the unclear statements.
Specific remarks:
Point 5: first sentence: A ceilometer is at first an instrument for detecting clouds. Please change the order of arguments. First the cloud parameters then all the other stuff.
The detection of the boundary layer (height) by a ceilometer is another issue but there is still no standard algorithm available for doing this. Every manufacturer cooks its own soup and most NMHSs just apply these algorithms (for possible problems originating from firmware algorithms see e.g., the paper by Kotthaus et al., AMT, 2016). Furthermore, the boundary layer height retrieved from a ceilometer is typically retrieved from the gradient(s) in the backscatter profile, thereby differing form the boundary layer height calculated by a model. So different scientific communities have different things in mind when they talk about the “boundary layer”.
Aerosol backscatter profiles aren’t mentioned (I recommend adding this !) but this capability is probably more important than the retrieval of the boundary layer height, since one retrieves a physical quantity from a calibrated ceilometer (the attenuated backscatter coefficient).
Response 5: We accepted your suggestion. The introduction is modified.
„The ceilometer is an essential instrument for detecting the cloud cover and cloud base height based on attenuated aerosol backscatter profile measurement. Vaisala CL51 ceilometer has measurement range 0-15000 m [13]. The profile enables to discriminate the cloud from other obstruction (e.g. precipitation) by a fairly straightforward threshold criteria. Its sky condition algorithm provides an observation about cloud amount and cloud layer height on five different layers only every 5 min from 16 s measurements based on data collected during the last 30 minutes. The last 10 minutes are double weighted to make the algorithm more responsive to variations in cloudiness. “
Point 6: cloud type: As far as I remember the cloud type is defined being something like an ice cloud, a water cloud, a mixed-phase cloud, a stratiform or a convective cloud. It might be better to use cloud layer (type, height) instead of cloud type, in order to avoid misunderstandings with the common meaning in the community.
Response 6: We agree with you just partly. The cloud type can defined based on cloud layer (https://scied.ucar.edu/webweather/clouds/cloud-types). But I modified the sentence to use as „cloud type categories according to Vaisala CL51 sky condition algorithm detected cloud base heights”
Point 7: Using the abbreviation “LMH” for “high clouds above medium or low-level clouds” isn’t very plausible. More to the point would be “HML” or “HaML”.
Response 7: We accepted your suggestion and I changed the abbreviation to HaML.
Point 8: A SYNOP record of 0 octa really means “no clouds” in the entire upper hemisphere whereas clouds not overpassing the ceilometer are simply not recognized. So a ceilometer may tell you “no clouds” whereas the SYNOP record says X octa, and both are right !. I’m therefore not sure what you want to say in L.89f. Which cloud information was used as “ground truth” ? The one from the ceilometer or the one from the SYNOP record ? Please rethink and reformulate the paragraph.
Response 8: The total cloud cover was calculated by adding the cloud cover values (in octa) in each level in the case of ceilometer measurement. If the total cloud cover value was 8 or larger the sky was determined as overcast. For example if the cloud cover values were 2, 7, 3, 0, 0 octa on the five layers, respectively the total cloud cover was 12 hence the sky was overcasted. When the total cloud cover was 0 it was determined cloud free. The cloud cover determination was the same as in the case of SYNOP because at NAOK the SYNOP data are from CT25K ceilometer. The comparison with SYNOP data is skipped in the revised version.
Point 9: A few more explanations about the temporal collocation would be helpful. Readers don’t want to be forced to read another paper before it becomes clear how the collocation was performed. To say it with a genius of past times: “Everything should be made as simple as possible, but not simpler“
Response 9: We accepted your suggestion. The method is detailed and a figure is added to demonstrate it.
“To reach a precise comparison we applied the time collocation method is detailed in [15]. The SEVIRI produces images every 0, 15, 30, 45 min of each full hour. The instrument scans from east to west and from south to north, and a full disc scan takes roughly 12 min. The remaining 3 min of a scan is used to retract the instrument to the nominal starting position. The Czech Republic is scanned approximately 10 min after the start. Therefore, the nominal repeat cycle time added by 10 min (RC + 10) was considered to be the real cloud type data acquisition time. We added 5 min window (RC + 10 + 5) to match the CL51 double weighted value of the last 10 min. Schematic colocation time is illustrated in Fig 2. For example for the 11:15 UTC repeat cycle the real cloud type image acquisition time is taken at 11:25 UTC in Košetice. The ceilometer double weighted value is received at 11:30 UTC, therefore we added 5 min window to search that time.
Fig.2: An illustration of the selection of ceilometer data for SEVIRI 1115 UTC repeat cycle. The first line shows the SEVIRI nominal repeat cycle time of scan, the Košetice line shows the actual scanning time for NAOK area and CL51 line shows the averaging process an 1130 UTC by sky condition algorithm.”
Point 10: An explanation of the spatial collocation is missing! MSG/SEVIRI pixels over Europe are already large, a few square kilometres, and therefore it is worth to mention if a single pixel is used for the intercomparison or a 3x3 or 5x5 matrix ? Please add that information.
Response 10: The spatial collocation was performed by an R script. In this software, the satellite geostationary (hereafter Geos) projection was defined (with proj4 syntax) and the coordinates (latitude and longitude) of the NAOK, where the CL51 is installed, were transformed into Geos projection with spTransform function. From the resulted raster and the particular pixel we found the SAFNWC CT values for CL51 location. It is added to the manuscript.
Point 11: What about MSG/SEVIRI’s pointing stability/accuracy, what about the parallax problem, especially for the geographical assignment of vertically extended clouds relative to the ground ? Nothing is said about but it may affect the entire study. Please add this information to the paper since it is essential for interpreting and judging the results.
Response 11: This information is added to the introduction. According to [9], compared to radar high cloud top detection, the MSG SEVERI is able to detect the low-level and high-level clouds but has difficulty detecting mid-level clouds. However, this problem was largely explained by limitations of the validation method rather than by incorrect MSG Cloud Types (CTY) assignments.
Point 12: Eqs 1) to 6): Sum from 1 to 5 or from 1 to 6 ? There are six “cloud types”.
Response 12: Thank you for your information. It was a typo. It is corrected.
Point 13: In table 1: “low medium height” à better to say “low medium high”
Response 13: Your suggestion is accepted. It is corrected.
Point 14: Eq 5) array sij needs to be defined and explained. A “penalty array” can be anything.
Response 14: Your suggestion is accepted. The si,j is detailed in appendix.
Point 15: “The low, medium and the high above medium and/or low clouds types were overestimated by ceilometer compared to satellite observation.”
This is a rather strange interpretation of results. The ceilometer is by far more sensitive to clouds than the satellite instrument, at the spatial scale of the ceilometer of course. So it is much better to conclude that the satellite instrument underestimates at least the low and medium clouds (at the place of the ceilometer), which is presumably simply a matter of the pixel size !, and the satellite instrument might have problems detecting low/mid level clouds under the presence of highs clouds. On the other hand, depending on the pixel size the satellite instrument should be able to detect high clouds (split window method) above lower clouds better than a ceilometer. It depends largely on the optical thickness of low/mid-level clouds if the backscatter signal penetrates through these clouds or not. This paragraph needs rethinking and must be rewritten.
Response 15: This suggestion is accepted and the paragraph is rewritted.
Point 16: “Despite this large bias, the probability of detection of medium clouds was larger in the summer period (0.69) compare to its value in the winter time (0.23).”
I think this is a trivial result which is driven by the fact that low clouds are more often observed during the winter time thus hampering the detection of mid-level clouds. Please reformulate the paragraph, add more substance to the findings, and discuss the results.
Point 17: “The low clouds were detected the most accurately in the winter period (95%).”
Same trivial statement as above. Fog situations will occur much more frequently during winter time which makes it easy to detect them from ceilometer backscatter data. Either you get a distinct signal from a low-level cloud or you get just nothing useful (extremely noisy backscatter), when the instrument stucks in the fog.
Response 16, 17: The calculation is changed and we draw a new conclusion. Originally, the satellite CTY data were applied as reference and the CL51 cloud layer detections were compared to satellite measurement for checking if the hight cloud layers are detected more accurately due to the larger backscatter range. We accepted your suggestion that the ceilometer is more sensitive to clouds, hence the ceilometer data are used as reference in the revised manuscript. We received new results hence these paragraphs are changed.
Point 18: “The significant false detection in M cloud type category partly can be explained by the fact that the very low clouds may be classified as medium clouds in case of strong thermal inversion (and thus of low cloud top temperature) by SAFNWC/MSG.”
So the false detection of mid-level clouds by a ceilometer is a result of the misclassification of the NWC-SAF software, thus the satellite data ? Again, it is unclear which measurements are taken as the reference: Is it the ceilometer data or is it the satellite data ? What is the spatial scale we are interested in ? The limited space where a ceilometer is representative for or the larger spatial scale of the satellite instrument ?
It is the satellite instrument that has a problem here ! Rethink the arguments and reformulate the paragraph.
Response 18: In SAFNWC cloud produc manual is reported that the very low clouds may be classified as medium clouds in case of strong thermal inversion (and thus of low cloud top temperature) by SAFNWC/MSG. We concluded that this weakness can be contributed to the disagreement with CL51 medium level cloud detection. In the revised version is mentioned that it is not possible to draw a conclusion from medium level cloud comparison owing to the different geometrical observation i.e. ceilometer detect the cloud base upward from the surface while satellite viewing downward from the top of the atmosphere.
Point 19: “The 12% occurrence of ceilometer detected high cloud in the cloud-free category of the satellite can be originated by the elusive cloud signal of CL51 when a transparent vapour level exists in upper air."
Sorry, this is completely odd and you dray the wrong conclusions: The strong backscatter of cloud droplets or ice crystals (high clouds) would certainly be recognized by the ceilometer’s detector. The fact that the satellite retrieval identifies a scene as cloud-free does not necessarily mean that there are no clouds covering the footprint. All satellite instruments and especially those with relatively large footprints such as the SEVIRI instrument are hampered and affected by sub-pixel cloudiness. And the ceilometer will detect these sub-pixel clouds when they pass over ! The CL51 has indeed a water vapour problem which manifests however mostly for aerosol retrievals while the cloud detection will not be affected.
Another likely explanation for the “underestimation” of high clouds are the firmware settings of the CL51. Have you analyzed the firmware (version) and the related settings ? If not, please do so and report it anyway here.
Response 19: This conclusion is skipped. It is not possible to analyze the firmware version. We have only one CL51 ceilometer. Presently, we cannot compare the detection with other CL51 with different firmware version.
Point 20: “The most of the LMH, 99% in winter and 97% in summer, was detected during night-time by ceilometer which resulted in very poor POD and large FAR.”
So the ceilometer does not detect “high clouds above other clouds” during night-time.
Fig 1 shows that in most cases when the satellite detects “high clouds above other clouds” the ceilometer detects lows clouds. Well, this is not surprising. The reason is most likely and simply the viewing geometry combined with dense clouds (beam blocking effect) closer to the ground (e.g., a wintertime stratiform cloud layer close to the ground). Fig. 2 shows the summer period and again the result is plausible since cloud layers are, on average, higher than during the winter. Consequently the majority of ceilometer measurements indicate mid-level clouds (for class LMH) and again, due to the viewing geometry, the satellite may see cloud layers above.
I think it is misleading to call this “a poor POD” because it is mostly the differing observation geometry.
Response 20: We accepted your suggestion. As it was mentioned above we changed the comparison and we mentioned in the revised version that the SAFNWC/MSG had a weakness in separating the low or mid-level clouds from the high one.
Point 21: “The small amount of total cloudiness (< 3 octas) was determined as a fractional cloud by ceilometer. This could be only clouds in mid- or high level which contained ice on the top of the cloud.”
Sorry again, this is confusing. Fractional clouds from a ceilometer as you call it are assigned if the coverage is below 3 octa. This has nothing to do with ice caps on top of high clouds. I don’t understand this argument.
Response 21: The satellite fractional cloud (it is coded with number 19) is defined when sub-pixel water cloud was detected. The small amount of total cloudiness (< 3 octas) was determined as a fractional cloud by ceilometer. We concluded that when the ceilometer detected fractional mid or high level clouds ,what covered the sky less than 3 octa, which contained dominantly ice partices on the top the satellite did not define these clouds as fractional cloud. But int the revised version of manuscript we did not include „fractional cloud” category due to the differing observation geometry.
Point 22:Figures 1 and 2 deserve a lot more explanation than given. The y-axis shows “Frequency in %” and this quantity is not defined/explained. It seems to be POD x 100 ?
Fig 1 x-axis: LHM --> LMH or even better HaLM
Response 22: We accepted your suggestion. The equation is explained in the manuscript (Eq 6). The x-axis is changet to HaML.
Point 23: “These results suggest that CL51 have a strong noise in high cloud detection.”
I don’t believe this conclusion at all. Besides the above-mentioned firmware problem, which influence on the cloud detection in upper levels is unknown, it is much more the problem of different cloud structures at different spatial scales that are seen by the satellite and the ceilometer. The split window technique applied to SEVIRI data for detecting high and optically thin clouds works on pixel basis and the satellite pixel is huge compared to the needle pin of the ceilometer. So especially the results for JJA and high clouds aren’t surprising (see also item 15 and the sub-pixel cloudiness problem). During winter time I expect that the threshold algorithm of the NWC-SAF assigns too often high clouds in the colder winter atmosphere. The satellite sensors detect the brightness temperature (of clouds) and later in the retrieval the height is assigned to cloud layers by assuming corresponding T-p-height profiles. Height profiles in wintering atmospheres are often quasi-isotherm, thus making it difficult to properly assign a height level to a brightness temperature.
Point 24:Without going into details I believe that the entire conclusions section must be revised and arguments must be newly formulated and properly discussed. See all critical points above.
Response 23, 24: We accepted you suggestion above that the ceilometer is more sensitive to clouds than the satellite and we applied the ceilometer detection as reference data. Moreover, we applied also the BL-View calculated cloud base heights to investigate if the range-variant smoothing of backscatter profile improves the high cloud detection. . Although ceilometer measurements were the reference data, if the satellite-detected high cloud type agreed with the ceilometer high cloud base layer, we concluded that the CL51 was improved considering high cloud detection, because the satellite instrument is sensitive to high clouds. The new conclusions you can find in the revised manuscript.
Point 25: “The difference between the SYNOP and CL51 and SAFNWC/MSG observation can be stemmed 196 from the fact that the SYNOP observation applied a time integration technique for estimating areal average”.
To my knowledge manned SYNOP observation of clouds are done at a certain time, e.g. 7, 14, 21, the classical “Mannheim hours”, whereas a ceilometer signal is typically integrated over time. How should then the SYNOP measurement be integrated over time ?
Response 25: According to my best knowledge the SYNOP data originate from CT25K ceilometer in NAOK. According toWMO Report on Direct and Indirect Measurement of Clouds (www.wmo.int/pages/prog/www/OSY/Meetings/ET-AWS/Doc8.doc). Although, the shape or genera of clouds can detect by human view but the information about heigh of the clouds most probably originate from ceilometer detection. As it was mentioned the SYNOP data are not used in the revised manuscript.
Reviewer 2 Report
I attached the manuscript with my suggestions.There are minor changes that I consider to be done before publishing the paper, especially in the state-of-the-art related to the performance of CL51.
Also please add a paragraph including references related to the algorithm of the cloud detection for CL51. Have you done any sensitivity studies related to this algorithm? What are the errors related to the height detection for each layer?
Comments for author File: Comments.pdf
Author Response
Point 1:There are minor changes that I consider to be done before publishing the paper, especially in the state-of-the-art related to the performance of CL51.
Response 1: Thank you for your suggestions. I accepted them. I also made a major revision considering the other reviewer’s suggestions.
Point 2: Also please add a paragraph including references related to the algorithm of the cloud detection for CL51.
Response 2: We accepted the suggestion. I add the description of the sky condition algorithm based on Module of Vaisala CL51. According to my best knowledge this algorithm has not been published in any journal paper before.
Point 3: Have you done any sensitivity studies related to this algorithm?
Response 3: No I did not do any sensitivity test related to this algorithm. But in the revised version I also compared the satellite cloud types with the cloud layer heights based on range-variant smoothing of backscatter profile. The result show that the smoothing increased the SNR and improved the hight cloud base detection of CL51.
Point 4: What are the errors related to the height detection for each layer?
Response 4: One error can stem from the fact that the ceilometer is more sensitive to the clouds than the satellite. Although, I applied the satellite data as reference to check if the CL51 can detect the hight cloud correctly thanks to the larger range of profile. In the revised version I used CL51 cloud layer data as reference.
The satellite detects the clouds from the cloud top temperature hence it is sensitive to hight clouds but presumably can hardly separate the low or medium clouds from the hight one, while the ceilometer can detect multi-layer cloud bases. These facts can related to the LMH and also the hight cloud detection errors.
The uncertainties or errors in medium cloud detection originate from the the different geometrical observation i.e. ceilometer detect the cloud base upward from the surface while satellite viewing downward from the top of the atmosphere and the ceilometer detected medium cloud types could be observed in the upper part of the medium cloud layer and these cloud tops can be occurred in the hight layer. In addition, the detection of the vertically developing clouds (e.g. Cumulonimbus) which base occurred in low- or midlevel and the top in high level can also cause the differences.
We skiped the comparison of fractional clouds because the error most probably originated the difference between the size of FOV of ceilometer and the pixel size of satellite image. It is possible to occurred a fraction of cloud inside the pixel area but outside the FOV.
Round 2
Reviewer 2 Report
The authors tried to accomodate all requests of the reviewer. The paper have been improved.
it can be published after minor spelling corrections
Author Response
The manuscript was undergo extensive English editing.
