An improved cloud index for estimating downwelling surface solar irradiance from various satellite imagers in the framework of a Heliosat-V method
- MINES ParisTech, PSL Research University, O.I.E. - Centre Observation, Impacts, Energy, 06904 Sophia Antipolis, France
- MINES ParisTech, PSL Research University, O.I.E. - Centre Observation, Impacts, Energy, 06904 Sophia Antipolis, France
Abstract. We develop a new way to retrieve the cloud index from a large variety of satellite instruments sensitive to reflected solar radiation, embedded on geostationary as non geostationary platforms. The cloud index is a widely used proxy for the effective cloud transmissivity, also called clear-sky index. This study is in the framework of the development of the Heliosat-V method for estimating downwelling solar irradiance at the surface of the Earth (DSSI) from satellite imagery. To reach its versatility, the method uses simulations from a fast radiative transfer model to estimate overcast (cloudy) and clear-sky (cloud-free) satellite scenes of the Earth’s reflectances. Simulations consider the anisotropy of the reflectances caused by both surface and atmosphere, and are adapted to the spectral sensitivity of the sensor. The anisotropy of ground reflectances is described by a bidirectional reflectance distribution function model and external satellite-derived data. An implementation of the method is applied to the visible imagery from a Meteosat Second Generation satellite, for 11 locations where high quality in situ measurements of DSSI are available from the Baseline Surface Radiation Network. Results from our preliminary implementation of Heliosat-V and ground-based measurements show a correlation coefficient reaching 0.948, for 15-minute means of DSSI, similar to operational and corrected satellite-based data products (0.950 for HelioClim3 version 5 and 0.937 for CAMS Radiation Service).
Benoît Tournadre et al.
Status: closed
-
RC1: 'Comment on amt-2020-480', Anonymous Referee #1, 21 Mar 2021
Tournadre et al. developed a new way to estimate downwelling surface solar irradiances (DSSI) from satellite images for Heliosat-V. Similar to previous Heliosat algorithms, the cloud index is needed in the DSSI estimation. In this new method, the maximum and minimum reflectances needed in the cloud index calculations are simulated using radiative transfer model instead of taking from archives of satellite images. The authors have demonstrated that the DSSI derived using this new method have good agreement with the CAMS and HelioClim3 DSSI. The new method is very promising. It has the advantage to be applied to both geostationary and polar orbiting satellites to get a global consistent DSSI data set using the same algorithm. The long term global DSSI data set will be interested by the solar energy and climate related communities. The authors have described the algorithm and results clearly. I think it is a good paper for AMT.
Specific comments
1) Line 19, ‘ plus a diffuse component due to scattering caused by the atmosphere (clouds, gases, aerosols) … ‘
Please also add ‘absorption’ in the sentence. In Fig. 2 you showed the gas absorptions by O2, O3, H2O.
2) Line 22 ‘renewable solar energy industries, …’
Is 'renewable' needed here?
3) Line 41-42. This sentence can be combined with the paragraph from Line 43.
4) Line 50, Please add the following paper in the reference list because they also use cloud properties to derive DSSI.
Retrieval and validation of global, direct, and diffuse irradiance derived from SEVIRI satellite observations
Greuell J. F. Meirink P. Wang https://doi.org/10.1002/jgrd.50194
5) Line 100-101 , ‘the upper boundary variables Xmax and Xmin’
Change to ‘the upper and lower boundaries …. ‘
6) Line 165 . This paragraph describes the MACC reanalysis used in the LUT. It is not clear if the MACC reanalysis has day, monthly or yearly AOD and which AOD is used.
7) Lines 184-185 can be combined with the paragraph below it.
8) Line 212 ‘ Heliomont’ Is it a typo?
9) Line 233 “ant”, typo?
10) Table1. What are the cloud base heights?
Please also add a table for the clear-sky LUT, including the BRDF, aerosols settings etc.. It is not complete if only having the table for the cloud LUT.
11) Line 258 ‘….for solar zenith angles lower than 80°’
Why do you use solar zenith angle until 80 degree in the validation? In the LUT, the solar zenith angle is until 85 degree. Is it possible to extend the solar zenith angle until 90 degree in the LUT?
12) Line 268 ‘However reflectance in the near infrared 0.8 µm channel are significantly higher, so is the absolute value of STD.’ Readers might want some explanations why the reflectances at 0.8 micron channel is larger than the 0.6 micron channel. Actually it is explained in the discussion section. This happens also in other paragraphs in the results section.
13) Lines 272 – 275. Figure 7 shows the results compared to measurements at the PAY and CAM SMS stations. Please provide some information about the surface type of the stations used in the figure. When it is clear-sky, the surface type, aerosols are more import.
14) In line 272, Figure 7 should be Figure 6.
Since BRDF is an important feature in the clear-sky LUT, it would be nice to show a figure at PAY, CAM, SMS with diurnal cycle for a clear-sky day. Please use 0.6 and 0.8 channel both when there are green grass on the ground surface.
15) Line 280 ‘Figure 6’ should be Figure 7.
16) Fig. 7 Why the simulated reflectances have better agreements with measured reflectances at SMS than at CAM?
It might not be due to the calibration of MSG because it would have the same bias in the full disk image. It seems the ice cloud LUT has similar diurnal cycle to the 99 percentiles of the measurements but the simulated reflectances are larger than the measurements at CAM. It could be at CAM the cloud are less brighter than at SMS. Does it suggest the simulated maximum reflectance should depend on location?
17) Line 320.
Fig. C1. Why there are some outliers with large reflectances in McClear? Is it due to the model or the aerosol data? I would expect the outliers on two sides of the 1:1 line.
18) The authors did not mention direct irradiances in the paper. Are there any plans about the DNI?
- AC1: 'Reply on RC1', Benoit TOURNADRE, 01 Mar 2022
-
RC2: 'Comment on amt-2020-480', Anonymous Referee #2, 25 Mar 2021
General comments:
The explicit strength of the “original” Heliosat approach (referred to as Heliosat-o in this review) is that the retrieved cloud index (“cloud transmission”) is completely based on observations. No simulations or external data are needed to retrieve the cloud index (cloud transmission) but the observed radiances are used. This includes the retrieval of Xmin (“clear sky reflection”) & Xmax (“calibration”). Heliosat-o and the resulting radiation data are well validated and established (e.g CM SAF, ISE, University of Oldenburg and Bergen, Satellight….) and already close to the accuracy of well maintained ground based stations. Of course, there are some limitations linked with the Xmin retrieval, as listed by the authors (L85). However, some of the mentioned handicaps are already partly resolved (e.g. shadow correction method by University of Oldenburg) or on average of relative small effect (e.g. long lasting clouds occur usually in the North-West during wintertime. This means high COD and low SZA. Hence, low solar irradiance and thus low absolute errors induced by uncertainties in Xmin). In my opinion there is a high likelihood that the simulation of Xmin adds more handicaps and uncertainties than it resolves. Thus, the central question is: Is there an overall benefit, concerning accuracy and precision, if the observational-based Xmin retrieval is replaced by simulations. Why should the simulations lead on average to more accurate results than using observations ? The authors mention “Simulations consider the anisotropy of the reflectances caused by both surface and atmosphere, and are adapted to the spectral sensitivity of the sensor. The anisotropy of ground reflectances is described by a bidirectional reflectance distribution function model and external satellite-derived data”. Simulations might consider it, but to my experience they induce also additional uncertainties, e.g. the uncertainty induced by using 3rd party surface albedo data can easily lead to a bias of several per cent. Further, as for RMIN, clear sky situations are needed to retrieve the surface albedo, thus concerning long lasting clouds the same handicap is shared. The needed BRDF (ADM) functions induce further uncertainties and add complexity. A more complex method providing overall a lower accuracy would be of no significant value. The effect of SAL (surface albedo) and BRDF is already considered by observational-based Xmin for the same sensor and viewing geometry, no need for simulation.
Major concerns:
In my opinion the authors fail to show the advantage of combining the Heliosat relation (equation) with simulations of the radiances in order to get Xmin (“clear sky reflection”). If radiances (reflectances) are simulated than why not simply using one of the several RTM based LUT approaches or ECMWF. By the way, using BRDFs simulations to estimate radiances observed by satellite is already applied since decades in RTM based LUT approaches, thus this Is not a new idea. Where is the benefit to use the Heliosat relation (equation 1) when the special strength of Heliosat is disminished by using simulations ? These questions are not appropriately addressed in the manuscript. The authors mention that a motivation for the approach is the use of polar orbiting satellites, but again what is the advantage compared to RTM based LUT approaches (using COD&reff or TOA Albedo).
In summary, a more thorough discussion and description of the pros and cons of the presented method compared to established methods should be added (Heliosat-o and RTM LUT approaches). Uncertainties of BRDF and SAL should be discussed, more information on SAL source should be added. Also the solar zenith angle dependency of SAL in relation to BRDF should be discussed in more detail. Further, the potential improvements should be proven and discussed thoroughly by comparison with established high quality data sets, which are using the original observational-based Heliosat-o approach and with other data sets from external sources, e.g. ECMWF. Please note, comparison with Helioclim might be not a real benchmark for improvements, see e.g. Posselt et al, Remote Sensing of Environment Vol 118, 2012, pp, 186–198. Respective open data sets are available for inter-comparison. Concerning polar orbiting satellites, results should be compared to the ECWMF radiation data set.
I think that simulations of Rmin has been already used for the so called “Heliosat-2” version. Thus, the novel aspects of the approach should be reflected in more detail relative to “Heliosat-2” as well. By the way, calling a method with Rmin simulation still Heliosat is quite confusing. Rmin simulation breaks with the basic idea of Heliosat, thus using the name Helioat should be avoided in order to avoid misleading interpretations. Overall the discussion should be modified to be more balanced and reflected , lessons learnt in other projects and communities should be considered.
Specific comments.
- Please change the title, improved is not prooven, see general comments.
- 70 „raw satellite numerical counts (Pfeifroth et al., 2017; Perez et al., 2002)“;
Here and throughout oft he manuscript. Misleading citations. Raw satellite counts has been used already decades before within the Heliosat community. Please modify accordingly. In general ATBD, PUMs are grey literature. Please check the citations and replace them with peer reviewed articles where possible.
- 80 “In this paper, we aim at finding an alternative to the need for archives of satellite imagery.”
This is misleading, as long as radiances are needed using actual and/or 30 day is not a serious problem and not worth mentioning.
- 140 “Kc = 1−n introduced by Darnell et al. (1988)”
I think it is a well known and established that a modification for higher n is needed and respective modifications are published, please refer them.
- 190 “Cloud-index methods in the literature use various ways to estimate the TOA reflectances in overcast conditions (Perez et al., 2002; Lefèvre et al., 2007; Pfeifroth et al., 2017).”
Pfeifroth et al. 2017, again misleading citation. Please refer to the original peer-reviewed publications . In general ATBD, PUMs are grey literature. Please check the citations and replace them with peer reviewed articles where possible.
- 65 Xmin is ued later on rho_clear please unify.
- AC2: 'Reply on RC2', Benoit TOURNADRE, 01 Mar 2022
-
RC3: 'Comment on amt-2020-480', Anonymous Referee #3, 27 Mar 2021
The comment was uploaded in the form of a supplement: https://amt.copernicus.org/preprints/amt-2020-480/amt-2020-480-RC3-supplement.pdf
- AC3: 'Reply on RC3', Benoit TOURNADRE, 01 Mar 2022
Status: closed
-
RC1: 'Comment on amt-2020-480', Anonymous Referee #1, 21 Mar 2021
Tournadre et al. developed a new way to estimate downwelling surface solar irradiances (DSSI) from satellite images for Heliosat-V. Similar to previous Heliosat algorithms, the cloud index is needed in the DSSI estimation. In this new method, the maximum and minimum reflectances needed in the cloud index calculations are simulated using radiative transfer model instead of taking from archives of satellite images. The authors have demonstrated that the DSSI derived using this new method have good agreement with the CAMS and HelioClim3 DSSI. The new method is very promising. It has the advantage to be applied to both geostationary and polar orbiting satellites to get a global consistent DSSI data set using the same algorithm. The long term global DSSI data set will be interested by the solar energy and climate related communities. The authors have described the algorithm and results clearly. I think it is a good paper for AMT.
Specific comments
1) Line 19, ‘ plus a diffuse component due to scattering caused by the atmosphere (clouds, gases, aerosols) … ‘
Please also add ‘absorption’ in the sentence. In Fig. 2 you showed the gas absorptions by O2, O3, H2O.
2) Line 22 ‘renewable solar energy industries, …’
Is 'renewable' needed here?
3) Line 41-42. This sentence can be combined with the paragraph from Line 43.
4) Line 50, Please add the following paper in the reference list because they also use cloud properties to derive DSSI.
Retrieval and validation of global, direct, and diffuse irradiance derived from SEVIRI satellite observations
Greuell J. F. Meirink P. Wang https://doi.org/10.1002/jgrd.50194
5) Line 100-101 , ‘the upper boundary variables Xmax and Xmin’
Change to ‘the upper and lower boundaries …. ‘
6) Line 165 . This paragraph describes the MACC reanalysis used in the LUT. It is not clear if the MACC reanalysis has day, monthly or yearly AOD and which AOD is used.
7) Lines 184-185 can be combined with the paragraph below it.
8) Line 212 ‘ Heliomont’ Is it a typo?
9) Line 233 “ant”, typo?
10) Table1. What are the cloud base heights?
Please also add a table for the clear-sky LUT, including the BRDF, aerosols settings etc.. It is not complete if only having the table for the cloud LUT.
11) Line 258 ‘….for solar zenith angles lower than 80°’
Why do you use solar zenith angle until 80 degree in the validation? In the LUT, the solar zenith angle is until 85 degree. Is it possible to extend the solar zenith angle until 90 degree in the LUT?
12) Line 268 ‘However reflectance in the near infrared 0.8 µm channel are significantly higher, so is the absolute value of STD.’ Readers might want some explanations why the reflectances at 0.8 micron channel is larger than the 0.6 micron channel. Actually it is explained in the discussion section. This happens also in other paragraphs in the results section.
13) Lines 272 – 275. Figure 7 shows the results compared to measurements at the PAY and CAM SMS stations. Please provide some information about the surface type of the stations used in the figure. When it is clear-sky, the surface type, aerosols are more import.
14) In line 272, Figure 7 should be Figure 6.
Since BRDF is an important feature in the clear-sky LUT, it would be nice to show a figure at PAY, CAM, SMS with diurnal cycle for a clear-sky day. Please use 0.6 and 0.8 channel both when there are green grass on the ground surface.
15) Line 280 ‘Figure 6’ should be Figure 7.
16) Fig. 7 Why the simulated reflectances have better agreements with measured reflectances at SMS than at CAM?
It might not be due to the calibration of MSG because it would have the same bias in the full disk image. It seems the ice cloud LUT has similar diurnal cycle to the 99 percentiles of the measurements but the simulated reflectances are larger than the measurements at CAM. It could be at CAM the cloud are less brighter than at SMS. Does it suggest the simulated maximum reflectance should depend on location?
17) Line 320.
Fig. C1. Why there are some outliers with large reflectances in McClear? Is it due to the model or the aerosol data? I would expect the outliers on two sides of the 1:1 line.
18) The authors did not mention direct irradiances in the paper. Are there any plans about the DNI?
- AC1: 'Reply on RC1', Benoit TOURNADRE, 01 Mar 2022
-
RC2: 'Comment on amt-2020-480', Anonymous Referee #2, 25 Mar 2021
General comments:
The explicit strength of the “original” Heliosat approach (referred to as Heliosat-o in this review) is that the retrieved cloud index (“cloud transmission”) is completely based on observations. No simulations or external data are needed to retrieve the cloud index (cloud transmission) but the observed radiances are used. This includes the retrieval of Xmin (“clear sky reflection”) & Xmax (“calibration”). Heliosat-o and the resulting radiation data are well validated and established (e.g CM SAF, ISE, University of Oldenburg and Bergen, Satellight….) and already close to the accuracy of well maintained ground based stations. Of course, there are some limitations linked with the Xmin retrieval, as listed by the authors (L85). However, some of the mentioned handicaps are already partly resolved (e.g. shadow correction method by University of Oldenburg) or on average of relative small effect (e.g. long lasting clouds occur usually in the North-West during wintertime. This means high COD and low SZA. Hence, low solar irradiance and thus low absolute errors induced by uncertainties in Xmin). In my opinion there is a high likelihood that the simulation of Xmin adds more handicaps and uncertainties than it resolves. Thus, the central question is: Is there an overall benefit, concerning accuracy and precision, if the observational-based Xmin retrieval is replaced by simulations. Why should the simulations lead on average to more accurate results than using observations ? The authors mention “Simulations consider the anisotropy of the reflectances caused by both surface and atmosphere, and are adapted to the spectral sensitivity of the sensor. The anisotropy of ground reflectances is described by a bidirectional reflectance distribution function model and external satellite-derived data”. Simulations might consider it, but to my experience they induce also additional uncertainties, e.g. the uncertainty induced by using 3rd party surface albedo data can easily lead to a bias of several per cent. Further, as for RMIN, clear sky situations are needed to retrieve the surface albedo, thus concerning long lasting clouds the same handicap is shared. The needed BRDF (ADM) functions induce further uncertainties and add complexity. A more complex method providing overall a lower accuracy would be of no significant value. The effect of SAL (surface albedo) and BRDF is already considered by observational-based Xmin for the same sensor and viewing geometry, no need for simulation.
Major concerns:
In my opinion the authors fail to show the advantage of combining the Heliosat relation (equation) with simulations of the radiances in order to get Xmin (“clear sky reflection”). If radiances (reflectances) are simulated than why not simply using one of the several RTM based LUT approaches or ECMWF. By the way, using BRDFs simulations to estimate radiances observed by satellite is already applied since decades in RTM based LUT approaches, thus this Is not a new idea. Where is the benefit to use the Heliosat relation (equation 1) when the special strength of Heliosat is disminished by using simulations ? These questions are not appropriately addressed in the manuscript. The authors mention that a motivation for the approach is the use of polar orbiting satellites, but again what is the advantage compared to RTM based LUT approaches (using COD&reff or TOA Albedo).
In summary, a more thorough discussion and description of the pros and cons of the presented method compared to established methods should be added (Heliosat-o and RTM LUT approaches). Uncertainties of BRDF and SAL should be discussed, more information on SAL source should be added. Also the solar zenith angle dependency of SAL in relation to BRDF should be discussed in more detail. Further, the potential improvements should be proven and discussed thoroughly by comparison with established high quality data sets, which are using the original observational-based Heliosat-o approach and with other data sets from external sources, e.g. ECMWF. Please note, comparison with Helioclim might be not a real benchmark for improvements, see e.g. Posselt et al, Remote Sensing of Environment Vol 118, 2012, pp, 186–198. Respective open data sets are available for inter-comparison. Concerning polar orbiting satellites, results should be compared to the ECWMF radiation data set.
I think that simulations of Rmin has been already used for the so called “Heliosat-2” version. Thus, the novel aspects of the approach should be reflected in more detail relative to “Heliosat-2” as well. By the way, calling a method with Rmin simulation still Heliosat is quite confusing. Rmin simulation breaks with the basic idea of Heliosat, thus using the name Helioat should be avoided in order to avoid misleading interpretations. Overall the discussion should be modified to be more balanced and reflected , lessons learnt in other projects and communities should be considered.
Specific comments.
- Please change the title, improved is not prooven, see general comments.
- 70 „raw satellite numerical counts (Pfeifroth et al., 2017; Perez et al., 2002)“;
Here and throughout oft he manuscript. Misleading citations. Raw satellite counts has been used already decades before within the Heliosat community. Please modify accordingly. In general ATBD, PUMs are grey literature. Please check the citations and replace them with peer reviewed articles where possible.
- 80 “In this paper, we aim at finding an alternative to the need for archives of satellite imagery.”
This is misleading, as long as radiances are needed using actual and/or 30 day is not a serious problem and not worth mentioning.
- 140 “Kc = 1−n introduced by Darnell et al. (1988)”
I think it is a well known and established that a modification for higher n is needed and respective modifications are published, please refer them.
- 190 “Cloud-index methods in the literature use various ways to estimate the TOA reflectances in overcast conditions (Perez et al., 2002; Lefèvre et al., 2007; Pfeifroth et al., 2017).”
Pfeifroth et al. 2017, again misleading citation. Please refer to the original peer-reviewed publications . In general ATBD, PUMs are grey literature. Please check the citations and replace them with peer reviewed articles where possible.
- 65 Xmin is ued later on rho_clear please unify.
- AC2: 'Reply on RC2', Benoit TOURNADRE, 01 Mar 2022
-
RC3: 'Comment on amt-2020-480', Anonymous Referee #3, 27 Mar 2021
The comment was uploaded in the form of a supplement: https://amt.copernicus.org/preprints/amt-2020-480/amt-2020-480-RC3-supplement.pdf
- AC3: 'Reply on RC3', Benoit TOURNADRE, 01 Mar 2022
Benoît Tournadre et al.
Benoît Tournadre et al.
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
497 | 322 | 21 | 840 | 14 | 11 |
- HTML: 497
- PDF: 322
- XML: 21
- Total: 840
- BibTeX: 14
- EndNote: 11
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1