the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 20 Mar 2025
Evaluating parallax and shadow correction methods for global horizontal irradiance retrievals from Meteosat SEVIRI
Abstract. The transition towards an energy supply with a high share of renewable energy calls for accurate global horizontal irradiance (GHI) observations. Satellite-derived GHI covers large geographical areas, making it an excellent data source for nowcasting solar power generation and the validation of weather and climate models. To obtain a good match between satellite-derived GHI and surface observations of GHI, a precise geolocation of the satellite GHI is an essential factor in addition to the accuracy of the retrieval. The geolocation of satellite retrievals is affected by parallax, a displacement between the actual and apparent position of a cloud, as well as by a displacement between the actual position of a shadow and the cloud casting that shadow. This study evaluates different approaches to correct Meteosat Spinning Enhanced Visible and Infrared Imager (SEVIRI) retrievals for parallax and cloud shadow displacements using ground-based observations from a unique network of 99 pyranometers deployed during the HOPE field campaign in Jülich, Germany, in 2013. The first method provides geometric corrections for the displacements calculated using retrieved cloud top heights. The second method relies on empirical collocation shifting. Here, the collocation shift of the satellite grid is determined by maximizing the correlation between the satellite retrievals and ground-based observations. This optimum shift is determined either based on daily or time step averaged correlations. The time step averaged collocation shift correction generally yields the most accurate results, but the drawback of this method is its reliance on ground-based observations. The geometric correction, which does not have this limitation, achieves the most accurate results if a combined parallax and shadow correction is performed. At higher spatial resolutions, the GHI retrieval accuracy becomes increasingly sensitive to the applied correction. At the SEVIRI standard native resolution of 3 x 3 km2 (SR), the root mean square error (RMSE) is reduced by 6.3 W m-2 (6.0 %), going from the uncorrected to combined geometrically corrected retrieval. At a threefold higher resolution (HR), this difference increases to 11.7 W m-2 (10.8 %). Cloud heterogeneity also strongly influences the sensitivity to geolocation accuracy and spatial resolution. Variable cloud regimes exhibit a higher sensitivity to geolocation corrections compared to less variable regimes. As an illustration, the mean improvement in HR RMSE between the uncorrected retrieval and the time step optimal shift is 13.3 W m-2 (10.9 %) for the stratocumulus cloud regime, while for the cirrus cloud regime the improvement is negligible.
-
Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
-
Preprint
(2443 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(2443 KB) - Metadata XML
- BibTeX
- EndNote
- Final revised paper
Journal article(s) based on this preprint
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2024-4139', Anonymous Referee #1, 04 Apr 2025
In this paper, the authors validate two corrections to satellite-based ground radiation
retrievals using ground measurements: parallax correction and shadow correction. This
is more involved than it looks at first glance and the paper contains some results that
many might overlook. It presents some evidence that the parallax error and shadow error
partially cancel out each other for geostationary-based radiation measurements, and that the combined
error may therefore increase if only parallax correction is applied. The paper also shows
that results improve when using a value smaller than the retrieved CTH, at least in some
cloud regimes. These result should be stressed more clearly, including in the abstract.The science in this paper is good, but some conclusions should be presented more clearly.
I have just one significant worry on the validation of the empirical collocation shift correction.
Below is a list of suggestions before this paper can be accepted with minor revisions.
AbstractThis abstract undersells the paper. It should prominently mention that
applying parallax correction alone can worsen validation results (as
shown in Figure 4) and that using a height smaller than the retrieved
CTH may be better than the full height.Page 1, line 7: This line is a bit confusing. The geolocation of the shadow (on the ground),
as visible from the satellite, should not be affected by parallax. The 1-D assumption
of "shadow is straight below the cloud" should be mentioned here, an assumption that is of
course erroneous in most cases.Introduction
Page 3, line 64: same here, the shadow location would be retrieved
accurately by just measuring the radiance from the shadowy pixel?
But that's not what is done?3.2 Shadow correction
Page 6, line 160: this 1-D assumption should be pointed out earlier (introduction and perhaps abstract)
as otherwise the reader may be confused why there is an error in the shadow location, as a simple radiance
retrieval without any assumptions will have the correct geolocation for the shadowed pixel (if the shadow
is on the ground and not on a lower cloud).3.4 Empirical collocation shift correction
Page 8, line 195: variations in CTH are also not accounted for.
Page 9, Figure 3, latitude shift: why is there no latitude shift (south/north) around 12 noon?
Shouldn't there be a shadow in the north?Page 10/11, Figures 3 (both panels) and 4(right panel): please add a thin grid line at 0, like you have in figures 5, 7, 8.
4.1.3 Pixel-based and area-based corrections
Page 12, line 258: I think "uncertainties" here should be "errors", shouldn't it?
4.1.5 Resolution sensitivity
Page 12, line 283: You have derived the emipirical collocation shift
correction using pyranometer measurements and now you are using
comparisons with pyranometer measurements to validate them. It would
seem your validation is not independent of your reference.4.2 Separation into cloud regimes
Page 13, figure 13: bit surprised by seeing the clear sky regime here, as
there will be no parallax or shadow error. But that actually provides a
baseline for improved analysis of the rest of the results. If you have
ca. 50 W/m² RMSE just from other sources, couldn't you subtract that
error estimate from all the other figures so you get an estimate that
covers only parallax and shadow effects? Or at least indicate this in
the other figures for context.Page 13, figure 13 caption: Note that the "natural color RGB" is known as
the "day land cloud RGB" in the USA and perhaps some other communities.
Adding this name (in addition to "natural colour" common in europe")
may help some readers.4.3 Diurnal Cycle
Line 19, page 398: "growing importance at increasing spatial resolutions", here one could mention FCI.
5.1 Generalizability of results
Page 19, 414: this could be tested with IODC. Not suggesting you need to do it for this study, but you could mention it at least.
5.2 Remaining mismatch errors
Page 21, line 473: this could be tested with LEO, such as Sentinel data or even Landsat (if calibration is good enough).
Page 21, line 479: the comment on 3D is a bit oddly formulated.
Typos:
page 19, line 391, missing space in "theassymmetric".
page 24, line 563: Github → GitHub
Citation: https://doi.org/10.5194/egusphere-2024-4139-RC1 - AC1: 'Reply on RC1', Job Wiltink, 20 May 2025
-
RC2: 'Comment on egusphere-2024-4139', Anonymous Referee #2, 05 Apr 2025
General comments:
The manuscript "Evaluating parallax and shadow correction methods for global horizontal irradiance retrievals from Meteosat SEVIRI" by Wiltink et al. presents correction methods for satellite-based GHI retrievals.
Two approaches are used here: purely geometric (parallax and shadow corrections) and empirical collocation shifts.
The paper is overall well written and the scientific topic investigated is justified and matches well with AMT.
The structure makes sense overall, although some main findings could be stressed a bit more clearly (e.g. the need to lower the cloud height to achieve the lowest RMSE and the counter-acting or opposing effects of two correction methods).
In the "Conclusions and outlook" section I miss a bit to bring the findings of this study into a broader context. E.g. what are the implications/suggestions for the usage of satellite derived GHI in predictions for PV yields? Is the difference of the satellite retrievals of about 100 W/m2 w.r.t. the pyranometers considered critical in view of typical total solar irradiance values? The presented correction methods seem to be able to reduce the errors between the satellite retrievals and the ground based retrievals of GHI by about 10-15 W/m2. This could be emphasized more clearly.
If the authors could expand a bit more regarding these points mentioned above it would be appreciated, however, I will not insist on that.
I recommend publication of the manuscript after the following minor revisons:
Specific comments:Line 33:
I think what is meant here is "Geostationary Operational Environmental Satellites (GOES)" and not Global Earth Observing System (GEOSS)?Line 189:
What does "any direction" mean here? Fixed along either N, S, W, E direction or along the SEVIRI footprints or along any polar angle? Please specify.Figure 3 and Line 202:
Is there a shading for the 5th to 95th percentile missing?Line 242:
Does this mean that the retrieved Hc, that is used as input to calculate the GHI, is on average over-estimated? (since Hc has to be reduced to get the smallest RSME?)Line 248:
Several cloud retrieval algorithms use a simplified Lambertian cloud model which is more representative of a "cloud mean height" than a "cloud top height". Have the authors considered to apply such cloud models instead of using Hc as input in order to mitigate this effect of ignoring the cloud's vertical extent?Figure 5:
If I understand correctly, a negative difference in panel c) indicates that the HR retrievals have a lower RMSE than the SR retrievals and that the HR-based daily or time-step corrections are better for the regimes CR2 and CR5-CR9. However, for CR1, CR3 and CR4 the SR-based corrections seem to be better. Do I understand that correctly? If so, then the term HR RMSE "improvement" in the Figure caption might be misleading since in some cases the improvement is with the SR and not the HR. Please clarify.Lines 340-341:
The phrases "mainly optimal" and "less optimal" sound strange as "optimal" is an absolute term. Maybe the following sentence sounds more reasonable: "In other words, the optimal shift is mainly suitable for the more variable regimes with lower retrieved clouds and, therefore, slightly less suitable for the more homogeneous regimes with higher retrieved clouds." Please consider to re-phrase this part.Line 362:
While this is true for the two extreme cases (uncorrected and fully corrected), it seems that the SR RSME is smaller than HR RMSE at relative cloud top heights above 60% and larger than HR RMSE at lower relative cloud top heights. Although not being statistically significant it might still be worth mentioning if the authors consider this relevant.Lines 365-366:
I don't understand how the statement of this sentence can also be deduced from Figure 5b? Can this be backed up by giving the exact RMSE numbers of the 40% Hc from Fig 7 versus the daily shift RMSE from Figure 5b?Line 387:
I don't understand how this can be seen from Figure 3a), where, to me, the shifts seem even larger in the morning. Did you mean to refer here to Figure 8a) instead of 3a)? Please clarify.Lines 457-460:
Maybe it could be useful for the reader if a Table (or text) is added in Section 2 that specifies the wavelegth ranges of the pyranometers and those of the SEVIRI bands that have been used. Or even show a Figure comparing the spectra for one collocated example case?Lines 461-463:
Could a rough estimation of the temporal mismatch error in a worst-case scenario be added here? E.g. by moving a high cloud for 5 minutes with a high wind speed into a fixed direction and then estimating the induced mismatch?Lines 464-472:
Same as above for the temporal mismatch errors, could a rough estimation for the order of magnitude of the spatial mismatch errors be added here?Lines 478-484:
Some interesting studies on 3D cloud effects have been done here (although with an application to trace gas retrievals instead of GHI, but still it might be interesting for the authors): https://amt.copernicus.org/articles/15/1587/2022/, https://amt.copernicus.org/articles/15/5743/2022/, https://amt.copernicus.org/articles/15/3481/2022/Technical corrections:
Line 356:
Remove the comma after Figure 7: "findings, Figure 7, shows"Line 358:
Remove the dot after Hc: "The median Hc. of the cirrostratus".Line 391:
Add a blank between "the" and "asymmetric".Line 425:
For instance, subtropical land regions --> For instance, in subtropical land regionsCitation: https://doi.org/10.5194/egusphere-2024-4139-RC2 - AC2: 'Reply on RC2', Job Wiltink, 20 May 2025
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2024-4139', Anonymous Referee #1, 04 Apr 2025
In this paper, the authors validate two corrections to satellite-based ground radiation
retrievals using ground measurements: parallax correction and shadow correction. This
is more involved than it looks at first glance and the paper contains some results that
many might overlook. It presents some evidence that the parallax error and shadow error
partially cancel out each other for geostationary-based radiation measurements, and that the combined
error may therefore increase if only parallax correction is applied. The paper also shows
that results improve when using a value smaller than the retrieved CTH, at least in some
cloud regimes. These result should be stressed more clearly, including in the abstract.The science in this paper is good, but some conclusions should be presented more clearly.
I have just one significant worry on the validation of the empirical collocation shift correction.
Below is a list of suggestions before this paper can be accepted with minor revisions.
AbstractThis abstract undersells the paper. It should prominently mention that
applying parallax correction alone can worsen validation results (as
shown in Figure 4) and that using a height smaller than the retrieved
CTH may be better than the full height.Page 1, line 7: This line is a bit confusing. The geolocation of the shadow (on the ground),
as visible from the satellite, should not be affected by parallax. The 1-D assumption
of "shadow is straight below the cloud" should be mentioned here, an assumption that is of
course erroneous in most cases.Introduction
Page 3, line 64: same here, the shadow location would be retrieved
accurately by just measuring the radiance from the shadowy pixel?
But that's not what is done?3.2 Shadow correction
Page 6, line 160: this 1-D assumption should be pointed out earlier (introduction and perhaps abstract)
as otherwise the reader may be confused why there is an error in the shadow location, as a simple radiance
retrieval without any assumptions will have the correct geolocation for the shadowed pixel (if the shadow
is on the ground and not on a lower cloud).3.4 Empirical collocation shift correction
Page 8, line 195: variations in CTH are also not accounted for.
Page 9, Figure 3, latitude shift: why is there no latitude shift (south/north) around 12 noon?
Shouldn't there be a shadow in the north?Page 10/11, Figures 3 (both panels) and 4(right panel): please add a thin grid line at 0, like you have in figures 5, 7, 8.
4.1.3 Pixel-based and area-based corrections
Page 12, line 258: I think "uncertainties" here should be "errors", shouldn't it?
4.1.5 Resolution sensitivity
Page 12, line 283: You have derived the emipirical collocation shift
correction using pyranometer measurements and now you are using
comparisons with pyranometer measurements to validate them. It would
seem your validation is not independent of your reference.4.2 Separation into cloud regimes
Page 13, figure 13: bit surprised by seeing the clear sky regime here, as
there will be no parallax or shadow error. But that actually provides a
baseline for improved analysis of the rest of the results. If you have
ca. 50 W/m² RMSE just from other sources, couldn't you subtract that
error estimate from all the other figures so you get an estimate that
covers only parallax and shadow effects? Or at least indicate this in
the other figures for context.Page 13, figure 13 caption: Note that the "natural color RGB" is known as
the "day land cloud RGB" in the USA and perhaps some other communities.
Adding this name (in addition to "natural colour" common in europe")
may help some readers.4.3 Diurnal Cycle
Line 19, page 398: "growing importance at increasing spatial resolutions", here one could mention FCI.
5.1 Generalizability of results
Page 19, 414: this could be tested with IODC. Not suggesting you need to do it for this study, but you could mention it at least.
5.2 Remaining mismatch errors
Page 21, line 473: this could be tested with LEO, such as Sentinel data or even Landsat (if calibration is good enough).
Page 21, line 479: the comment on 3D is a bit oddly formulated.
Typos:
page 19, line 391, missing space in "theassymmetric".
page 24, line 563: Github → GitHub
Citation: https://doi.org/10.5194/egusphere-2024-4139-RC1 - AC1: 'Reply on RC1', Job Wiltink, 20 May 2025
-
RC2: 'Comment on egusphere-2024-4139', Anonymous Referee #2, 05 Apr 2025
General comments:
The manuscript "Evaluating parallax and shadow correction methods for global horizontal irradiance retrievals from Meteosat SEVIRI" by Wiltink et al. presents correction methods for satellite-based GHI retrievals.
Two approaches are used here: purely geometric (parallax and shadow corrections) and empirical collocation shifts.
The paper is overall well written and the scientific topic investigated is justified and matches well with AMT.
The structure makes sense overall, although some main findings could be stressed a bit more clearly (e.g. the need to lower the cloud height to achieve the lowest RMSE and the counter-acting or opposing effects of two correction methods).
In the "Conclusions and outlook" section I miss a bit to bring the findings of this study into a broader context. E.g. what are the implications/suggestions for the usage of satellite derived GHI in predictions for PV yields? Is the difference of the satellite retrievals of about 100 W/m2 w.r.t. the pyranometers considered critical in view of typical total solar irradiance values? The presented correction methods seem to be able to reduce the errors between the satellite retrievals and the ground based retrievals of GHI by about 10-15 W/m2. This could be emphasized more clearly.
If the authors could expand a bit more regarding these points mentioned above it would be appreciated, however, I will not insist on that.
I recommend publication of the manuscript after the following minor revisons:
Specific comments:Line 33:
I think what is meant here is "Geostationary Operational Environmental Satellites (GOES)" and not Global Earth Observing System (GEOSS)?Line 189:
What does "any direction" mean here? Fixed along either N, S, W, E direction or along the SEVIRI footprints or along any polar angle? Please specify.Figure 3 and Line 202:
Is there a shading for the 5th to 95th percentile missing?Line 242:
Does this mean that the retrieved Hc, that is used as input to calculate the GHI, is on average over-estimated? (since Hc has to be reduced to get the smallest RSME?)Line 248:
Several cloud retrieval algorithms use a simplified Lambertian cloud model which is more representative of a "cloud mean height" than a "cloud top height". Have the authors considered to apply such cloud models instead of using Hc as input in order to mitigate this effect of ignoring the cloud's vertical extent?Figure 5:
If I understand correctly, a negative difference in panel c) indicates that the HR retrievals have a lower RMSE than the SR retrievals and that the HR-based daily or time-step corrections are better for the regimes CR2 and CR5-CR9. However, for CR1, CR3 and CR4 the SR-based corrections seem to be better. Do I understand that correctly? If so, then the term HR RMSE "improvement" in the Figure caption might be misleading since in some cases the improvement is with the SR and not the HR. Please clarify.Lines 340-341:
The phrases "mainly optimal" and "less optimal" sound strange as "optimal" is an absolute term. Maybe the following sentence sounds more reasonable: "In other words, the optimal shift is mainly suitable for the more variable regimes with lower retrieved clouds and, therefore, slightly less suitable for the more homogeneous regimes with higher retrieved clouds." Please consider to re-phrase this part.Line 362:
While this is true for the two extreme cases (uncorrected and fully corrected), it seems that the SR RSME is smaller than HR RMSE at relative cloud top heights above 60% and larger than HR RMSE at lower relative cloud top heights. Although not being statistically significant it might still be worth mentioning if the authors consider this relevant.Lines 365-366:
I don't understand how the statement of this sentence can also be deduced from Figure 5b? Can this be backed up by giving the exact RMSE numbers of the 40% Hc from Fig 7 versus the daily shift RMSE from Figure 5b?Line 387:
I don't understand how this can be seen from Figure 3a), where, to me, the shifts seem even larger in the morning. Did you mean to refer here to Figure 8a) instead of 3a)? Please clarify.Lines 457-460:
Maybe it could be useful for the reader if a Table (or text) is added in Section 2 that specifies the wavelegth ranges of the pyranometers and those of the SEVIRI bands that have been used. Or even show a Figure comparing the spectra for one collocated example case?Lines 461-463:
Could a rough estimation of the temporal mismatch error in a worst-case scenario be added here? E.g. by moving a high cloud for 5 minutes with a high wind speed into a fixed direction and then estimating the induced mismatch?Lines 464-472:
Same as above for the temporal mismatch errors, could a rough estimation for the order of magnitude of the spatial mismatch errors be added here?Lines 478-484:
Some interesting studies on 3D cloud effects have been done here (although with an application to trace gas retrievals instead of GHI, but still it might be interesting for the authors): https://amt.copernicus.org/articles/15/1587/2022/, https://amt.copernicus.org/articles/15/5743/2022/, https://amt.copernicus.org/articles/15/3481/2022/Technical corrections:
Line 356:
Remove the comma after Figure 7: "findings, Figure 7, shows"Line 358:
Remove the dot after Hc: "The median Hc. of the cirrostratus".Line 391:
Add a blank between "the" and "asymmetric".Line 425:
For instance, subtropical land regions --> For instance, in subtropical land regionsCitation: https://doi.org/10.5194/egusphere-2024-4139-RC2 - AC2: 'Reply on RC2', Job Wiltink, 20 May 2025
Peer review completion
Journal article(s) based on this preprint
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 258 | 201 | 19 | 478 | 15 | 42 |
- HTML: 258
- PDF: 201
- XML: 19
- Total: 478
- BibTeX: 15
- EndNote: 42
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Hartwig Deneke
Chiel C. van Heerwaarden
Jan Fokke Meirink
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(2443 KB) - Metadata XML