Cloud Detection with Historical Geostationary Satellite Sensors for Climate Applications

Round 1

Reviewer 1 Report

This paper presents the philosophies that behind the methodology used for developing the EUMETSAT CM SAF long-term COMET Cloud Fractional Cover (CFC) CDR for climate applications. In general, the paper is well written and provides a valuable addition to the existing CM SAF technical reports and evaluation studies.  The ideas given in this paper are also useful for the development of other long-term CDRs using historical remote sensing data.  The reviewer recommends to accept it for publication after minor revision. 

Generally speaking, the English is good. But the sentences in this manuscript are not easy to read. Many sentences are long and/or lack of punctuations. 3-line sentences are common; some sentences are 4-line long.  Suggest to break down the long sentences to shorter ones add/or add more punctuations. 

 

Table 1: (1) it seems one score (STTscore) is missing compared to reference #28, Section 4.  (2) The descriptions for STRcore and TVscore are not accurate. (3) It is difficult to understand the functions of these scores independently (without reading reference #28). Suggest to provide more information about these scores in the table. 

 

Figure 5:   it is not straight forward to tell the filled and unfilled orange cycles apart in c) and d).  Suggest to replace the unfilled orange cycles with another symbol/color.

 

 

Lines 111 – 113: MVIRI, MFG, MSG have already been defined in the Introduction.

Line 163 – 166:  please clarity the meaning of this sentence. 

Line 302: please define term “ECV”.

Line 323:  suggest to change “panel a” to “Figure5a”.

Lines 505 & 513: suggest to change “in 1996” to “before and after 2016”.

Author Response

This paper presents the philosophies that behind the methodology used for developing the EUMETSAT CM SAF long-term COMET Cloud Fractional Cover (CFC) CDR for climate applications. In general, the paper is well written and provides a valuable addition to the existing CM SAF technical reports and evaluation studies.  The ideas given in this paper are also useful for the development of other long-term CDRs using historical remote sensing data.  The reviewer recommends to accept it for publication after minor revision.

Generally speaking, the English is good. But the sentences in this manuscript are not easy to read. Many sentences are long and/or lack of punctuations. 3-line sentences are common; some sentences are 4-line long.  Suggest to break down the long sentences to shorter ones add/or add more punctuations.

A: Thank you very much for your helpful review. We shortened 29 sentences by splitting them into two or three sentences. And we made sure that no sentence in the manuscript is longer than 3 lines.

Table 1: (1) it seems one score (STTscore) is missing compared to reference #28, Section 4.  (2) The descriptions for STRcore and TVscore are not accurate. (3) It is difficult to understand the functions of these scores independently (without reading reference #28). Suggest to provide more information about these scores in the table.

A: The paper is correct. STT_score was never implemented in the code. We have now removed it from the ATBD (ref. 28) and it is updated on the CM SAF website within the next week. We have also added a short description of the individual scores as part of Table 1 to simplify reading and understanding.

Figure 5:   it is not straight forward to tell the filled and unfilled orange cycles apart in c) and d).  Suggest to replace the unfilled orange cycles with another symbol/color.

A: Indeed. We have exchanged the unfilled orange circles for clear sky with filled green circles. The difference in color (green for clear sky, orange for cloudy) aids interpretation.

Lines 111 – 113: MVIRI, MFG, MSG have already been defined in the Introduction.

A: Indeed. We have removed the duplicate definitions.

Line 163 – 166:  please clarity the meaning of this sentence.

A: Sentence rewritten: "Our calibration strategy differs slightly from what is used by Rossow (2017) for the new ISCCP H series since they did not have new inter-calibration coefficients for geostationary sensors at hand. In ISCCP H radiances of geostationary sensors have been tied by the ISCCP team to radiances of polar sensors while we could use a stable geostationary FCDR."

Line 302: please define term “ECV”.

A: Done (Essential Climate Variable).

Line 323:  suggest to change “panel a” to “Figure5a”.

A: Done.

Lines 505 & 513: suggest to change “in 1996” to “before and after 2016”.

A: Done (before and after 1996).

Reviewer 2 Report

This is a nice document that describes the detection of cloud in geostationary satellite measurements in a manner that is consistent over decadal timescales and includes several successive satellite missions with changes in observing instrument characteristics. The cloud interference is a serious issue in many remote sensing applications, and there are a variety of approaches to dealing with it. In this work, the limited spectral information content of available observations is compensated with the efficient use of the high temporal resolution of data. The paper is well structured and, in most places, easy to follow, although the use of English language could be improved throughout the text (I mean adding commas in certain  places and avoiding sentences starting with either "And" or "But"). I would like to recommend accepting the manuscript with minor revisions, as I have only a few suggestions to make.

p3, l79: I would suggest replacing "ToA clear-sky temperature" by "ToA clear-sky brightness temperature". Furthermore, the authors should consider mentioning radiative transfer modelling (RTM) as a method to estimate ToA clear-sky brightness temperature using NWP input.

p3, l116-121: Please check the wavelength bounds of each channel here (they are not all correct).

p9, l302: ECV is an acronym for Essential Climate Variable, right? Is it sufficiently widely understood not to spell it out?

p11, l350: I could imagine synoptic variability to occasionally produce up to 5K skin temperature differences between successive days. Is there any risk of confusing such temperature variations with cloud signal and thus making false alarms by the cloud detection scheme?

p13, l426-428: To me, a "bug" sounds more like a technical problem than inability to detect low stratus in difficult continental winter conditions. I would suggest re-phrasing this as a "systematic failure" or something similar (while a "bug fix" could be re-phrased as a "performance improvement").

p15, l 497-500: This sounds somewhat speculative and is potentially wishful thinking. I would suggest putting it into context by stating how much there is tolerance (safety margin) for radiative effect that comes from failing to detect cloud.

p16, l 532-533: Note that there is not necessarily that much clear-sky diurnal cycle at higher latitudes in winter

Author Response

This is a nice document that describes the detection of cloud in geostationary satellite measurements in a manner that is consistent over decadal timescales and includes several successive satellite missions with changes in observing instrument characteristics. The cloud interference is a serious issue in many remote sensing applications, and there are a variety of approaches to dealing with it. In this work, the limited spectral information content of available observations is compensated with the efficient use of the high temporal resolution of data. The paper is well structured and, in most places, easy to follow, although the use of English language could be improved throughout the text (I mean adding commas in certain  places and avoiding sentences starting with either "And" or "But"). I would like to recommend accepting the manuscript with minor revisions, as I have only a few suggestions to make.

A: Thank you very much for your helpful review. We shortened 29 sentences by splitting them into two or three sentences. And we made sure that no sentence in the manuscript is longer than 3 lines. We have also rewritten three sentences starting with "But" and one starting with "And".

p3, l79: I would suggest replacing "ToA clear-sky temperature" by "ToA clear-sky brightness temperature". Furthermore, the authors should consider mentioning radiative transfer modelling (RTM) as a method to estimate ToA clear-sky brightness temperature using NWP input.

A: Done (replacement). We have also added a reference for estimating ToA BT with RTM and NWP input (e.g. done in the CC4CL algorithm).

p3, l116-121: Please check the wavelength bounds of each channel here (they are not all correct).

A: This was a very bad typo. The wavelength bounds were off by a factor 10 for the WV and IR bands. Thanks for pointing this out.

p9, l302: ECV is an acronym for Essential Climate Variable, right? Is it sufficiently widely understood not to spell it out?

A: Done.

p11, l350: I could imagine synoptic variability to occasionally produce up to 5K skin temperature differences between successive days. Is there any risk of confusing such temperature variations with cloud signal and thus making false alarms by the cloud detection scheme?

A: Yes, indeed. We have added a few sentences on this issue in the discussion where we talk about the influence of similar synoptic changes in the visible spectrum:  "A temperature drop of e.g. 5 K from one day to the next could generate a false cloud signal until the daily clear sky inversion has caught up with the changed ambient conditions. This problem is mitigated by running the cloud detection twice as described above. A first guess cloud detection with last day's clear sky inversion provides new clear sky brightness temperatures for estimating current day's clear sky diurnal cycle for a final CFC estimate. Potentially false cloud signals are also minimized by the two variability scores which are insensitive to constant shifts between the all-sky and the clear-sky signal."

p13, l426-428: To me, a "bug" sounds more like a technical problem than inability to detect low stratus in difficult continental winter conditions. I would suggest re-phrasing this as a "systematic failure" or something similar (while a "bug fix" could be re-phrased as a "performance improvement").

A: We understand your concerns and have changed the wording accordingly. Since most of the development of such a code is rather IT-based we may have become very geeky in thinking on the way. But indeed, we're up for precision and performance (although real code bugs have been fixed as well by our site-based development).

p15, l 497-500: This sounds somewhat speculative and is potentially wishful thinking. I would suggest putting it into context by stating how much there is tolerance (safety margin) for radiative effect that comes from failing to detect cloud.

A: Yes, it is good to give a few orders of magnitude in order to explain what we mean. We have extended the text as follows:  "If for a winter inversion the low stratus cloud (e.g. with reflectance 0.6 and brightness temperature 276 K) has almost the same signature as the underlying surface (e.g. with reflectance 0.55 and brightness temperature 273 K), the cloudy signal can be safely considered as realistic input to the clear sky inversion. With the spectral capabilities at hand, cloud and clear sky look the same. The missed cloud will however yield a too low CFC estimate as discussed before."

p16, l 532-533: Note that there is not necessarily that much clear-sky diurnal cycle at higher latitudes in winter

A: Yes, absolutely. Especially for brightness temperature the clear sky diurnal cycle has a much lower magnitude for e.g. snow covered surfaces in high latitude winter. However, this is also the case for water surfaces at lower latitudes where COMET works well. The magnitude of the diurnal cycle is less important for the model inversion than the number and quality of the cloud screened input data.