the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 09 Aug 2023
Shortwave reflected energy from NISTAR and the Earth Polychromatic Imaging Camera onboard the DSCOVR spacecraft
Abstract. We describe a new method for estimating the total reflected shortwave energy from the Earth Polychromatic Imaging Camera (EPIC) and compare it with direct measurements from the NIST Advanced Radiometer (NISTAR) instrument (Electrical substitution radiometer) – both are onboard the Lagrange-1 orbiting Deep Space Climate Observatory (DSCOVR). The 6 narrow-band wavelength channels (340 to 780 nm) available from EPIC provide a framework for estimating the integrated spectral energy for each EPIC pixel. The Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) and the SCIAMACHY instrument provide spectral information away from the EPIC wavelengths, particularly for wavelengths longer than 780 nm. The total area-weighted reflected shortwave energy from an entire EPIC image is compared with co-temporal Band B Shortwave reflected energy observed by NISTAR. Our analysis from March to December 2017 shows the two are highly correlated with differences ranging from -10 to 10 Watts m-2. The offset bias over the entire period is less than 0.2 Watts m-2. We also compare our EPIC energy maps with the Clouds and the Earth’s Radiant Energy System (CERES) Single Scanner Footprint (SSF) Shortwave (SW) reflected energy observed within 3 hours of an EPIC image. Our EPIC-AVIRIS SW estimate is 5–20 % higher near the EPIC image center and 5–20 % lower near the image edges compared with the CERES SSF.
- Preprint
(18396 KB) - Metadata XML
- BibTeX
- EndNote
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2023-638', François-Marie Bréon, 27 Aug 2023
This paper presents a clear description of a method to extrapolate spectraly the narrowband measurements from the EPIC instrument onboard the DSCOVR spacecraft. The comparison over another instrument onboard the same satellite shows a high agreement while the comparison against the CERAS measurements, onboard another spacecraft shows large discrepencies. These results point to inaccuracies in the ADM, the tabulated factors to derive a flux from the radiance measurements.
The paper is interesting and very clear. It can be accepted as is. However, I strongly suggest the authors add some conclusion regarding the ADM in the abstract as this is an information that may be of interest for a wider community
Citation: https://doi.org/10.5194/egusphere-2023-638-RC1 -
RC2: 'Comment on egusphere-2023-638', Anonymous Referee #1, 29 Aug 2023
General Comments
The paper’s objective is to compare what is referred to as “shortwave reflected energy” from NISTAR and EPIC. EPIC is a narrowband imaging camera with 18 km resolution and NISTAR is a broadband radiometer that takes frequent measurements of the entire viewable sunlit portion of the Earth from the L-1 point. To compare the two, EPIC narrowband radiances are converted to “shortwave reflected energy” (or broadband shortwave, SW) using a best-fit algorithm that selects a spectrum from a set of only 10 candidate scenes from 20-m AVIRS measurements to “fill in” the missing spectral regions of EPIC (after normalizing to the EPIC channels). EPIC and NISTAR “shortwave reflected energy” values are compared and show consistent results. The paper goes on to use CERES SSF data within 1 and 3 h of the EPIC imaging time to compare single images over the Pacific and over Africa on June 1, 2017. The CERES measurements are adjusted to account for EPIC and CERES viewing geometry differences using CERES TRMM angular models.
After carefully reading the manuscript, my recommendation is to reject the paper. There are far too many technical issues with the analysis and far too many missing details provided about the methodology.
The spectral filling algorithm applied to EPIC uses spectra from only 10 homogeneous 20-m resolution AVIRIS measurements for scenes over North America to represent the spectral shapes of 18-km resolution EPIC pixels over the entire sunlit portion of Earth during all daylight hours. Furthermore, the AVIRIS scenes are restricted to solar zenith angles between 41-58 deg, whereas EPIC samples a full range of solar zenith angles. There is no justification given why 10 AVIRIS scenes are sufficient. Amongst the 10 AVIRIS scenes, there is only one overcast cloud case (“solid cloud spectra”). It is unrealistic to assume that a single cloud can represent upwelling spectral radiances for all clouds. Furthermore, there is no description of the one cloud case: is it a liquid or ice cloud? Is it optically thin or thick?
The methodology used to select the best-fit spectrum to fill in missing spectral regions for EPIC is overly simplistic and unphysical. For example, to infer a broadband shortwave radiance from 18-km EPIC pixels over ocean, the 20-m AVIRIS spectra from a single predetermined “solid cloud” and a single “clear ocean” scene is combined to fill in spectral radiances between EPIC bands. If EPIC observes a cloud that is much thicker/thinner than the “solid cloud” case in the AVIRIS look-up table, the weights assigned to the “clear” and “solid cloud” spectra in the fit to EPIC data will be inconsistent with the true cloud fraction, which could introduce large biases in spectral filling/extrapolation and consequently in the integrated shortwave radiance. This is important since there are no EPIC bands larger than 800 nm, yet 50% of the integrated shortwave radiance lies at wavelengths > 800 nm. Similarly, over land, the paper does not combine spectra from different surface types, even though the EPIC resolution is 18 km and therefore will often contain multiple surface types, and thick and thin clouds, etc. The impacts of these limitations on the results are not explored in the paper even though it would be possible using radiative transfer model calculations (e.g., simulation of the methodology or OSSE).
The comparisons with CERES are also problematic. It does not appear that time differences between EPIC and CERES observations are accounted? The caption says that CERES SSF reflected “energies” are determined from the CERES SSF flux multiplied by the CERES anisotropy factor at DSCOVR viewing and illumination conditions. If true, this is not the correct way to calculate this. The CERES flux needs to be at the same solar zenith angle as the anisotropic factor. However, the CERES SSF flux and DSCOVR measurement can differ by 3 h. Furthermore, clouds change in 3 hours. That is not accounted for in the comparisons either?
It should be noted that the Su et al papers, which apparently is a motivation for this paper, use a far more thorough approach for the EPIC/DSCOVR vs CERES comparisons. Many instruments from LEO and GEO are used to resolve the diurnal cycles observed by EPIC/DSCOVR to provide an apples-to-apples comparison. That is not the case here.
Specific Comments:
- 2: “This work describes a new approach to derive shortwave reflected energy from the calibrated EPIC data that is independent of CERES fluxes, CERES Angular Distribution Models (ADM) or MODIS observations”.
What exactly is “shortwave reflected energy”? Please specify if it is a SW radiance or SW flux. If it’s a SW flux, angular correction is necessary since EPIC measurements narrowband radiances, not fluxes. I suspect the authors are referring to fluxes and assume the SW radiation is isotropic, so that their flux is given by the product of pi times radiance. However, this is never actually stated in the paper.
- 2: “Our method uses additional spectral information from the EPIC 780 nm channel…”
Additional to what? Please clarify.
- 2: “We use calibrated radiances (Herman et al., 2018, Geogdzhayev and Marshak, 2018) from each EPIC pixel”.
What are the absolute calibration uncertainties of EPIC radiances for the different spectral bands used in this study?
- 2 & 3: The paper clearly states that calibrated EPIC and NISTAR radiances are used in this paper. Therefore, why are all the figures in flux units (Wm-2)?
- 3: “Since the filter has a SW transmission less than one, a calibration factor must be applied to the measured radiances, to produce unfiltered radiances. The archived Level 1B BandB SW
radiances used in this study are converted to unfiltered values by multiplying them by 1.15075 (1.0/0.8690) to account for photons absorbed by the filter.”
How is this ratio derived? Is there a reference for it? What is the absolute calibration uncertainty of NISTAR?
- 3: “The NASA ER-2 is the AVIRIS platform of choice for our study since it is closest to the TOA and has a spatial resolution of 20 meters.”
What does “closest to the TOA” mean? Does it mean ER-2 flies high enough that there’s no need to account for atmospheric scattering/absorption above the aircraft? If so, please justify this assumption.
Also, the spatial resolution of AVIRIS is 20 m while EPIC is 18 km. How is it justified to merge data with such markedly different spatial resolutions?
- 3: “We sifted through the many AVIRIS flights and selected nadir viewed spectra from 10 homogeneous scenes: 1) a solid cloud scene with near 100% cloud fraction; 2) a cloud-free ocean scene; 3 through 10) eight different scenes over land based on the measured Normalized Difference Vegetation Index NDVI (see Table 1).”
Are 10 AVIRIS scenes over North America with solar zenith angles between 41-58 deg sufficient to represent the spectral shapes of all EPIC pixels during all daylight hours? This is not justified in the paper.
Among the 10 AVIRIS scenes, there is only one overcast cloud case (“solid cloud spectra”). It is unrealistic to assume that a single cloud can represent upwelling spectral radiances for all clouds. What is the justification for this assumption? Furthermore, there is no description of the one cloud case: is it a liquid or ice cloud? Is it optically thin or thick?
With this oversimplified for the representation of clouds, the errors propagate to the derived EPIC anisotropy (Figures 6g and 7g). The spatial variations of the derived “EPIC anisotropy” resembles the spatial distribution of clouds (RGB pictures on the EPIC webpage). Therefore, the large spatial variation of the “EPIC anisotropy” is due to errors introduced during the spectral composite process and is not a representation of the actual anisotropy spatial variation.
Page 4: “Our method assumes that the spectra from a given EPIC pixel can be modeled by a linear combination of the spectra from selected pairs of these homogeneous scenes. Over ocean the EPIC pixel is modeled by a linear combination of the solid cloud scene with the clear-ocean scene. Over land the combination is the solid cloud scene with one of the eight cloud-free land scenes. To account for pixels over lakes or rivers the solid cloud scene can be linearly combined with the clear-ocean scene.”
This methodology is overly simplistic and unphysical. For example, to infer a broadband shortwave radiance from 18-km EPIC pixels over ocean, 20-m AVIRIS spectra from a single predetermined “solid cloud” and a single “clear ocean” scene are combined to fill in spectral radiances between EPIC bands. If EPIC observes a cloud that is much thicker/thinner than the “solid cloud” case in the AVIRIS look-up table, the weights assigned to the “clear” and “solid cloud” spectra in the fit to EPIC data will be inconsistent with the true cloud fraction, which could introduce large biases in spectral filling/extrapolation and consequently in the integrated shortwave radiance. As noted in the paper, there are no EPIC bands larger than 800 nm, yet 50% of the integrated shortwave radiance lies at wavelengths > 800 nm. Similarly, over land, the paper does not combine spectra from different surface types, even though the EPIC resolution is 18 km. Furthermore, there are no AVIRIS spectra for snow, ice or urban EPIC pixels. The impact of these limitations on the results are not explored in the paper even though it would be straightforward to do so using radiative transfer model calculations (e.g., simulation of the methodology or OSSE).
- 5: “Spectra from similar homogeneous scenes observed by the SCanning Imaging Absorption SpectroMeter for Atmospheric CHartographY (SCIAMACHY) instrument are merged with the AVIRIS spectra to construct a complete spectrum from 275 to 2500 nm for each homogeneous scene”.
How exactly is this merging done? AVIRIS and SCIAMACHY are not matched in time, and the spatial resolution of AVIRIS is 20 m while it is 32 x 215 km for SCIMACHY when in nadir mode. How can one possibly find similar homogeneous scenes between AVIRIS and SCIAMACHY? How do the authors find “homogeneous” clear and cloudy SCIAMACHY scenes at all when the spatial resolution is so coarse? Why is SCIAMACHY even used instead of AVIRIS in the UV region?
- 5: “Before comparing or merging two different spectra one needs to account for differences in viewing and illumination geometry. We generate look-up tables of radiances using the VLIDORT (Vector LInearized Discrete Ordinate Radiative Transfer package, Spurr, 2006) over the different wavelengths (275-2500nm), illumination (0-90 deg), viewing (0-90 deg) and azimuthal angle (0-180 deg) ranges.”
What scenes are used in the VLDORT calculations? Are they consistent with those in the AVIRIS look-up table? This would require knowing the cloud properties for the overcast case and the surface brdf and aerosol properties for the clear cases. Where does that information come from? Much more detail is needed about how this correction is made.
- 5: “To convert to a new viewing geometry, we scale the observed radiance from EPIC, AVIRS or SCIAMACHY by the ratio of the VLIDORT radiance at the new geometry with the VLIDORT radiance at the observed geometry.”
How accurate is the methodology to account for viewing/illumination geometry between EPIC, AVIRIS and SCIAMACHY?
- 5: Figure 2. Why are the y-axes in this figure in Wm-2 nm-1? EPIC provides radiances, not fluxes. Shouldn’t the units be Wm-2 nm-1 sr-1? Since there is no information about applying an anisotropic correction for the EPIC radiances, my suspicion is that the authors assumed without justification that the Earth is isotropic and simply multiplied the EPIC radiances by pi. However, it is well known that the correction for anisotropy at direct backscatter angles observed by EPIC are very large (e.g., Su et al papers). Why is this not discussed in the paper?
- 5: Why is the methodology used to construct composite spectra for EPIC described in the caption of Figure 2? This should be in the main text in a methodology section in addition to or instead of in the figure caption.
- 5: Figure 2 caption: The coefficients for the linear combination are 0.05 for solid cloud and 0.93 for the clear desert scene. Why don’t the coefficients add to 1.0? Normally, when combining clear and cloudy areas to represent an area-average, the following is used:
I = f*I_cld + (1-f)*I_clr, where f=cloud fraction.
Also, how is the “best-fit” to observed EPIC radiances determined? Are the EPIC and AVIRIS radiances normalized to a common scale or does the fit use absolute values of EPIC and AVIRIS radiances. If the latter is case, this is problematic. A quick look at the Green et al. (1998) paper indicates that the calibration accuracy of AVIRIS is about 4%, and it’s probably similar for EPIC. The calibration uncertainty combined with the limited set of ten AVIRIS spectra and the need to use VLIDORT to account for differences in viewing geometry likely results in unrealistic combinations of clear and “solid-cloud” scenes/weights.
Is equal weight assigned to each EPIC band or are the EPIC bands weighted by incoming solar irradiance (i.e., energy weighted)?
Along the same lines, the authors appear to treat the AVIRIS measurements as the “absolute truth” and scale the SCIAMACHY radiances to match AVIRIS between 400-450 nm without any justification. It would be helpful to know the scaling factors between AVIRIS and SCIAMACHY to understand how consistent these two instruments are and if the scaling factors vary with time and location.
- 5-6: Figures 2 and 3 show examples where the linear fit appears to come close to the EPIC values. However, the reader has no idea how well this approach works in general and what the overall uncertainty of the method is. Can the error in the fit (without adjustment) to EPIC somehow be quantified across all observed scenes and illumination conditions? Can the error in the fit to EPIC radiance be used to estimate error across the entire SW spectrum?
- 7: “Averaging all pixels for an image yields an adjustment of +0.8%. or ~2.0 Wm -2 and a 1s standard deviation of 16 Wm-2.”
The units are for flux, not radiance. Is this just pi x radiance?
- 7: The authors acknowledge that the integrated energy that needs to be “filled in” beyond 780 nm can be 50% or more of the total. They further acknowledge that the accuracy of their approach rests on how well they can use the 780 nm EPIC channel to predict contributions beyond 780 nm. Yet, linear combinations of two land scenes are not allowed: they only use one spectrum for a thick solid cloud and a single land scene from 20 m AVIRIS data to “fill in” EPIC pixels with a spatial resolution is 18 km at nadir that increases with viewing angle. Furthermore, if the EPIC scene is snow, ice, or “urban”, there are no AVIRIS spectra available. It’s not at all clear how those spectra are “filled in”.
These assumptions/approximations are problematic, particularly since 50% of the energy is missing.
- 7 “Another source of error occurs when the water vapor column of an EPIC pixel is different than the column at the time of the AVIRIS observation. Four strong water vapor absorption bands at 0.94, 1.12, 1.39 and 1.89 microns (Ramaswamy and Freidenreich, 1991) contribute to the spectral shape in the near-IR for all the homogeneous land AVIRIS spectra (Figure 1)”.
The 10 AVIRIS scenes all occur over different parts of North America. Furthermore, the one “solid cloud spectra” cannot possibly represent all clouds. The water column above deep convective clouds is very different from that of trade cumulus.
p.7: “To investigate the ability of the 780mn EPIC measurement to predict the EnIR, we consider all the nadir spectra from 10 AVIRIS flights over a variety of land surfaces (Table 2)…This suggests that if our algorithm finds a land AVIRIS spectra that is consistent
with the spectra of the EPIC pixel, the EPIC 780nm radiance alone will be able to predict EnIR with a 1s error of ~5 Wm-2 (fourth column).”
Only 10 cases are used to test how well the methodology works globally. All appear to be over N. America, and many appear to be located close to those used in Table 1 to train the algorithm. There are no cases over the tropics, high latitudes, etc., nor are any additional cloud cases considered other than the single one used in Table 1. At 18 km, many EPIC pixels will consist of broken clouds. That too is ignored with this very limited set of test cases. Such a limited set of cases is insufficient to assess “the ability of the 780 nm EPIC measurement to predict the Enir” (or ~50% of energy they do not measure).
Figure 5: The annual cycle looks odd. The time series from February to December lacks the well-defined Earth-Sun distance variation. The magnitude of the SW flux is also unrealistically large.
Figure 6 Caption: How is the time difference between EPIC and CERES accounted for? The caption says that CERES SSF reflected “energies” are determined from the CERES SSF flux multiplied by the CERES anisotropy factor at DSCOVR viewing and illumination conditions. If true, this is not the correct way to calculate this. The CERES flux needs to be at the same solar zenith angle as the anisotropic factor. However, the CERES SSF flux and DSCOVR measurement can differ by 3 h. The sentence that follows makes this even more confusing: it says the EPIC-AVIRIS “energy” is interpolated onto the SSF pixels observed within 3 h of the EPIC image time. It seems there may be a serious problem with the calculation. At the very least, the explanation of the methodology (in the caption, no less) is extremely confusing.
Figures 6d and 7d, showing color coded scene types for “solid cloud over ocean”, “clear-sky over ocean, desert and vegetation” indicate huge areas that are white, yet there is no explanation of what the white areas represent. Also, it seems highly unlikely that all of Africa and Europe were entirely either “clear-sky vegetation” or “clear-sky desert” on June 1 Hour 10 (Fig. 7d).
Citation: https://doi.org/10.5194/egusphere-2023-638-RC2 -
RC3: 'Comment on egusphere-2023-638', Anonymous Referee #3, 04 Sep 2023
This was a difficult read. I had to read it several times and I still have unanswered questions about the approach and the validation. I am not sure what the overall purpose of the paper is. The introduction states the goal of the paper, which is to present a new EPIC narrowband to broadband approach, that is independent of CERES fluxes and CERES ADMs. The new EPIC approach was lacking in detail, some of which was embedded in the figure captions, with no supporting validation. The EPIC derived broadband fluxes were then compared to NISTAR and CERES fluxes. The NISTAR instantaneous full disk fluxes had a greater daily “GMT or longitude position cycle” flux range over the day than EPIC, giving no confidence that the EPIC broadband fluxes are capturing the cloud and clear-sky land flux variations across the globe. The authors then compare the EPIC broadband to the CERES radiances, which used the CERES ADMs to convert the CERES fluxes to the EPIC observed angles and conclude that the CERES ADMs are inadequate. This does not validate the EPIC narrowband to broadband radiances. If the authors instead used their VLIDORT based ADMs to correct the CERES fluxes to the EPIC angles and the resulting comparison accounted for angular radiance differences than I would be convinced. The VLIDORT LUT was a key component in the EPIC narrowband to broadband approach to account for angular differences.
I reject the paper based on the following comments.
General comments
I have reviewed my extensive list (more than 50) of SW narrowband to broadband papers from the 1980s and most approaches had access to narrowband and broadband collocated coincident data in order to derive empirical and theoretical enhanced visible channel coefficients and provided the validation to derive the uncertainties. Previous authors suggested increasing the number of identified scene types (surface, atmospheric and cloud conditions), finer angular bins, more channel wavelengths, to reduce the uncertainties.
This paper did not cite one narrowband to broadband approach. The approach presented did not follow any previous (successful) approach.
The world is 65% cloudy. Very little information about the cloud spectra is given in the text, no cloud properties other than cloud fraction is used to derive the spectra. Only one AVIRIS cloud spectra at one angular geometry is used. So little effort was made to account for cloud type spectral differences. For example, Fig. 2 in Li et al 2018 shows the various cloud spectra that differ. Fig 5 in Gao et al 2003, show the spectra between cirrus and cumulus clouds differ dramatically especially over water absorption bands. It is amazing how more reflective cumulus clouds are in the SWIR bands compared to cirrus.
Li, H.; Zheng, H.; Han, C.; Wang, H.; Miao, M. Onboard Spectral and Spatial Cloud Detection for Hyperspectral Remote Sensing Images. Remote Sens. 2018, 10, 152. https://doi.org/10.3390/rs10010152
Gao, B.-C., and Y. J. Kaufman, Water vapor retrievals using Moderate Resolution Imaging Spectroradiometer (MODIS)near-infrared channels, J. Geophys. Res.,108(D13), 4389, doi:10.1029/2002JD003023, 2003
VLIDORT is used to account for hyperspectral differences in view and solar zenith angles. No cloud information (height, thickness, ice/water phase, optical depth), atmospheric profiles, land spectra and surface BRDF (AVIRIS land spectra was based only one angular condition), was given in the text in the VLIDORT LUT.
I believe the success of this approach relies on the VLIDORT LUT accuracy, since the AVIRIS spectra (limited to one angular condition) was adjusted to the observed angular EPIC observations. 50% of the flux contribution is from wavelengths greater than 780nm all dependent on a simple spectral adjustment and the VLIDORT LUT.
Section 3.2. I cannot determine how the column water vapor is accounted for in the NIR water vapor absorption bands above low clouds or clear-sky ocean. Only clear-sky land was mentioned.
The NASA EMIT hyper-spectral reflectances would be a great way to validate this narrowband to broadband approach, especially VLIDORT. The EMIT sensor is on the international space station samples the entire range of solar zenith angles and reflected solar spectra (0.4µm to 2.5µm) mainly over clear-sky land. I encourage the authors to validate their EPIC narrowband to broadband approach using EMIT data.
Page 8 line 8 “The archived Level 1B BandB SW radiances used in this study are converted to unfiltered values by multiplying them by 1.15075 (1.0/0.8690) to account for photons absorbed by the filter.” No reference or explanation how this number was obtained.
The EPIC full disk flux comparison with NISTAR in figure 5 showed that the NISTAR instantaneous full disk fluxes had a greater daily “GMT or longitude position cycle” flux range over the day than EPIC, giving no confidence that the EPIC broadband fluxes are capturing the cloud and clear-sky land flux variations across the globe. I am assuming that NISTAR is the truth dataset for this comparison.
It just feels like the paper was put hastily together based on the following.
Is it AVIRS or AVIRIS, the paper goes back and forth, please pick one acronym, and stick with it. What are energies? radiances or fluxes? many plots do not have units, hard to tell if radiances or fluxes are plotted. The algorithm information was dispersed in the text, appendix, and figure captions. The algorithm was not coherently outlined in a flow chart detailing inputs and outputs. Page 8 line 13 “NDIV threshold”
Figure 6 caption. Please state whether this was Terra-SSF or Aqua-SSF, or NPP-SSF, Is the time, “Hour 00 Minute 45”, in GMT time. Lack of units 6a, 6b, 6c, on the plots. Given that CERES instruments are both in the 10:30 and 1:30 local time orbits, the authors could have plotted the CERES orbit that is to the right of the EPIC center.
Fig. 6 and Fig. 7. The term “reflected energy product” is confusing, there are no units given for either Figure. Are the SSF fluxes plotted in 6a converted to radiance at the EPIC angles?
Figure 6f. The units are in %, why not the same units in 6a or 6f? This will let the reader establish what regions are contributing to the overall EPIC-AVIRIS minus CERES SSF partial disk flux.
Fig. 6 and Fig. 7. The partial disk EPIC pixel level broadband comparisons against CERES fluxes. EPIC was 10 less than CERES for Fig. 6, and 10 greater than CERES for Fig. 7, but the patterns on Fig. 6f and Fig. 7f were similar. If the radiances are weighted by EPIC angle, then the edge contribution is less than the center, for Fig. 6 where EPIC-AVIRIS has a greater flux than CERES in the center (fig. 6f) but yet the overall disk flux is less than (CERES mean 299.5 and EPIC mean 289.3). Is there an explanation?
Figure 6. Why are there white patches embedded in the dark blue in the eastern most orbit and in 6i in the EPIC view angle plot?
Fig.7 For this case Africa is mostly observed in clear-sky conditions. (see Fig. d Africa is yellow and green). For the CERES orbit that is within 30 minutes of EPIC to the left of the black dot, the anisotropic factor varies gradually and is mostly green over land. The orbit to the right and left have strong anisotropic gradients. This is also evident in Fig A-1a and Fig A1b. Does the greater time difference explain the noisy radiances?
Page 10 line 13 “We use anisotropies from the Edition 2 Angular Distribution Models to convert the archived CERES flux to the energy that EPIC would observe based on the camera’s viewing geometry” What was the source of the cloud fraction, optical depth, and cloud phase needed to determine the CERES TRMM ADM scene type?
Page 10 line 13. There are certain locations on the EPIC disk where the CERES and EPIC observations have matching angles. This would be the locations to compare the CERES radiance with the EPIC-AVIRIS radiance without the use of an ADM. I encourage the authors to validate their EPIC narrowband to broadband approach using matched EPIC and CERES angle radiance pairs.
I would highly encourage the authors to use the MODIS channel radiances that are analogous to EPIC and aggregate them into the CERES footprint and compare their narrowband to broadband approach to the CERES flux. This would be a great way to validate the EPIC derived broadband fluxes.
Figure 8. I would also like to see a Fig 8d that has the June 2017 CERES minus EPIC relative radiance difference. There are a lot of horizontal features in Fig. 8a and 8b, that just do not seem to part of the Fig. 8c. The Fig. 8a and Fig. 8b colorbar just does not make sense. There are no relative fluxes less than 160 or greater than 480. Readjust the colorbar between 160 and 480 with 10 colors to increase the radiance resolution.
Appendix, Fig. 6 and 7. and section 5. Is Fig. A-1 “EPIC anisotropy” the ADM factor needed to convert the CERES flux to match the EPIC derived broadband flux. The VLIDORT LUT was used to determine the broadband radiances at all other EPIC angular conditions from the single AVIRIS angular observation. The VLIDORT LUT is an ADM. Why is the VLIDORT ADM not compared with the “EPIC anisotropy”? If the VLIDORT ADM matches the EPIC anisotropy that would validate the EPIC narrowband to broadband approach.
In fact, you could take two EPIC images that are spaced up to 3-hours apart. Then use the VLIDORT ADM to convert one EPIC image angular geometry to the other EPIC image angular geometry and see if the EPIC derived radiance differences are mitigated in the illuminated portion. I encourage the authors to validate their EPIC narrowband to broadband approach by performing this consistency check using the VLIDORT ADM.
Citation: https://doi.org/10.5194/egusphere-2023-638-RC3
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2023-638', François-Marie Bréon, 27 Aug 2023
This paper presents a clear description of a method to extrapolate spectraly the narrowband measurements from the EPIC instrument onboard the DSCOVR spacecraft. The comparison over another instrument onboard the same satellite shows a high agreement while the comparison against the CERAS measurements, onboard another spacecraft shows large discrepencies. These results point to inaccuracies in the ADM, the tabulated factors to derive a flux from the radiance measurements.
The paper is interesting and very clear. It can be accepted as is. However, I strongly suggest the authors add some conclusion regarding the ADM in the abstract as this is an information that may be of interest for a wider community
Citation: https://doi.org/10.5194/egusphere-2023-638-RC1 -
RC2: 'Comment on egusphere-2023-638', Anonymous Referee #1, 29 Aug 2023
General Comments
The paper’s objective is to compare what is referred to as “shortwave reflected energy” from NISTAR and EPIC. EPIC is a narrowband imaging camera with 18 km resolution and NISTAR is a broadband radiometer that takes frequent measurements of the entire viewable sunlit portion of the Earth from the L-1 point. To compare the two, EPIC narrowband radiances are converted to “shortwave reflected energy” (or broadband shortwave, SW) using a best-fit algorithm that selects a spectrum from a set of only 10 candidate scenes from 20-m AVIRS measurements to “fill in” the missing spectral regions of EPIC (after normalizing to the EPIC channels). EPIC and NISTAR “shortwave reflected energy” values are compared and show consistent results. The paper goes on to use CERES SSF data within 1 and 3 h of the EPIC imaging time to compare single images over the Pacific and over Africa on June 1, 2017. The CERES measurements are adjusted to account for EPIC and CERES viewing geometry differences using CERES TRMM angular models.
After carefully reading the manuscript, my recommendation is to reject the paper. There are far too many technical issues with the analysis and far too many missing details provided about the methodology.
The spectral filling algorithm applied to EPIC uses spectra from only 10 homogeneous 20-m resolution AVIRIS measurements for scenes over North America to represent the spectral shapes of 18-km resolution EPIC pixels over the entire sunlit portion of Earth during all daylight hours. Furthermore, the AVIRIS scenes are restricted to solar zenith angles between 41-58 deg, whereas EPIC samples a full range of solar zenith angles. There is no justification given why 10 AVIRIS scenes are sufficient. Amongst the 10 AVIRIS scenes, there is only one overcast cloud case (“solid cloud spectra”). It is unrealistic to assume that a single cloud can represent upwelling spectral radiances for all clouds. Furthermore, there is no description of the one cloud case: is it a liquid or ice cloud? Is it optically thin or thick?
The methodology used to select the best-fit spectrum to fill in missing spectral regions for EPIC is overly simplistic and unphysical. For example, to infer a broadband shortwave radiance from 18-km EPIC pixels over ocean, the 20-m AVIRIS spectra from a single predetermined “solid cloud” and a single “clear ocean” scene is combined to fill in spectral radiances between EPIC bands. If EPIC observes a cloud that is much thicker/thinner than the “solid cloud” case in the AVIRIS look-up table, the weights assigned to the “clear” and “solid cloud” spectra in the fit to EPIC data will be inconsistent with the true cloud fraction, which could introduce large biases in spectral filling/extrapolation and consequently in the integrated shortwave radiance. This is important since there are no EPIC bands larger than 800 nm, yet 50% of the integrated shortwave radiance lies at wavelengths > 800 nm. Similarly, over land, the paper does not combine spectra from different surface types, even though the EPIC resolution is 18 km and therefore will often contain multiple surface types, and thick and thin clouds, etc. The impacts of these limitations on the results are not explored in the paper even though it would be possible using radiative transfer model calculations (e.g., simulation of the methodology or OSSE).
The comparisons with CERES are also problematic. It does not appear that time differences between EPIC and CERES observations are accounted? The caption says that CERES SSF reflected “energies” are determined from the CERES SSF flux multiplied by the CERES anisotropy factor at DSCOVR viewing and illumination conditions. If true, this is not the correct way to calculate this. The CERES flux needs to be at the same solar zenith angle as the anisotropic factor. However, the CERES SSF flux and DSCOVR measurement can differ by 3 h. Furthermore, clouds change in 3 hours. That is not accounted for in the comparisons either?
It should be noted that the Su et al papers, which apparently is a motivation for this paper, use a far more thorough approach for the EPIC/DSCOVR vs CERES comparisons. Many instruments from LEO and GEO are used to resolve the diurnal cycles observed by EPIC/DSCOVR to provide an apples-to-apples comparison. That is not the case here.
Specific Comments:
- 2: “This work describes a new approach to derive shortwave reflected energy from the calibrated EPIC data that is independent of CERES fluxes, CERES Angular Distribution Models (ADM) or MODIS observations”.
What exactly is “shortwave reflected energy”? Please specify if it is a SW radiance or SW flux. If it’s a SW flux, angular correction is necessary since EPIC measurements narrowband radiances, not fluxes. I suspect the authors are referring to fluxes and assume the SW radiation is isotropic, so that their flux is given by the product of pi times radiance. However, this is never actually stated in the paper.
- 2: “Our method uses additional spectral information from the EPIC 780 nm channel…”
Additional to what? Please clarify.
- 2: “We use calibrated radiances (Herman et al., 2018, Geogdzhayev and Marshak, 2018) from each EPIC pixel”.
What are the absolute calibration uncertainties of EPIC radiances for the different spectral bands used in this study?
- 2 & 3: The paper clearly states that calibrated EPIC and NISTAR radiances are used in this paper. Therefore, why are all the figures in flux units (Wm-2)?
- 3: “Since the filter has a SW transmission less than one, a calibration factor must be applied to the measured radiances, to produce unfiltered radiances. The archived Level 1B BandB SW
radiances used in this study are converted to unfiltered values by multiplying them by 1.15075 (1.0/0.8690) to account for photons absorbed by the filter.”
How is this ratio derived? Is there a reference for it? What is the absolute calibration uncertainty of NISTAR?
- 3: “The NASA ER-2 is the AVIRIS platform of choice for our study since it is closest to the TOA and has a spatial resolution of 20 meters.”
What does “closest to the TOA” mean? Does it mean ER-2 flies high enough that there’s no need to account for atmospheric scattering/absorption above the aircraft? If so, please justify this assumption.
Also, the spatial resolution of AVIRIS is 20 m while EPIC is 18 km. How is it justified to merge data with such markedly different spatial resolutions?
- 3: “We sifted through the many AVIRIS flights and selected nadir viewed spectra from 10 homogeneous scenes: 1) a solid cloud scene with near 100% cloud fraction; 2) a cloud-free ocean scene; 3 through 10) eight different scenes over land based on the measured Normalized Difference Vegetation Index NDVI (see Table 1).”
Are 10 AVIRIS scenes over North America with solar zenith angles between 41-58 deg sufficient to represent the spectral shapes of all EPIC pixels during all daylight hours? This is not justified in the paper.
Among the 10 AVIRIS scenes, there is only one overcast cloud case (“solid cloud spectra”). It is unrealistic to assume that a single cloud can represent upwelling spectral radiances for all clouds. What is the justification for this assumption? Furthermore, there is no description of the one cloud case: is it a liquid or ice cloud? Is it optically thin or thick?
With this oversimplified for the representation of clouds, the errors propagate to the derived EPIC anisotropy (Figures 6g and 7g). The spatial variations of the derived “EPIC anisotropy” resembles the spatial distribution of clouds (RGB pictures on the EPIC webpage). Therefore, the large spatial variation of the “EPIC anisotropy” is due to errors introduced during the spectral composite process and is not a representation of the actual anisotropy spatial variation.
Page 4: “Our method assumes that the spectra from a given EPIC pixel can be modeled by a linear combination of the spectra from selected pairs of these homogeneous scenes. Over ocean the EPIC pixel is modeled by a linear combination of the solid cloud scene with the clear-ocean scene. Over land the combination is the solid cloud scene with one of the eight cloud-free land scenes. To account for pixels over lakes or rivers the solid cloud scene can be linearly combined with the clear-ocean scene.”
This methodology is overly simplistic and unphysical. For example, to infer a broadband shortwave radiance from 18-km EPIC pixels over ocean, 20-m AVIRIS spectra from a single predetermined “solid cloud” and a single “clear ocean” scene are combined to fill in spectral radiances between EPIC bands. If EPIC observes a cloud that is much thicker/thinner than the “solid cloud” case in the AVIRIS look-up table, the weights assigned to the “clear” and “solid cloud” spectra in the fit to EPIC data will be inconsistent with the true cloud fraction, which could introduce large biases in spectral filling/extrapolation and consequently in the integrated shortwave radiance. As noted in the paper, there are no EPIC bands larger than 800 nm, yet 50% of the integrated shortwave radiance lies at wavelengths > 800 nm. Similarly, over land, the paper does not combine spectra from different surface types, even though the EPIC resolution is 18 km. Furthermore, there are no AVIRIS spectra for snow, ice or urban EPIC pixels. The impact of these limitations on the results are not explored in the paper even though it would be straightforward to do so using radiative transfer model calculations (e.g., simulation of the methodology or OSSE).
- 5: “Spectra from similar homogeneous scenes observed by the SCanning Imaging Absorption SpectroMeter for Atmospheric CHartographY (SCIAMACHY) instrument are merged with the AVIRIS spectra to construct a complete spectrum from 275 to 2500 nm for each homogeneous scene”.
How exactly is this merging done? AVIRIS and SCIAMACHY are not matched in time, and the spatial resolution of AVIRIS is 20 m while it is 32 x 215 km for SCIMACHY when in nadir mode. How can one possibly find similar homogeneous scenes between AVIRIS and SCIAMACHY? How do the authors find “homogeneous” clear and cloudy SCIAMACHY scenes at all when the spatial resolution is so coarse? Why is SCIAMACHY even used instead of AVIRIS in the UV region?
- 5: “Before comparing or merging two different spectra one needs to account for differences in viewing and illumination geometry. We generate look-up tables of radiances using the VLIDORT (Vector LInearized Discrete Ordinate Radiative Transfer package, Spurr, 2006) over the different wavelengths (275-2500nm), illumination (0-90 deg), viewing (0-90 deg) and azimuthal angle (0-180 deg) ranges.”
What scenes are used in the VLDORT calculations? Are they consistent with those in the AVIRIS look-up table? This would require knowing the cloud properties for the overcast case and the surface brdf and aerosol properties for the clear cases. Where does that information come from? Much more detail is needed about how this correction is made.
- 5: “To convert to a new viewing geometry, we scale the observed radiance from EPIC, AVIRS or SCIAMACHY by the ratio of the VLIDORT radiance at the new geometry with the VLIDORT radiance at the observed geometry.”
How accurate is the methodology to account for viewing/illumination geometry between EPIC, AVIRIS and SCIAMACHY?
- 5: Figure 2. Why are the y-axes in this figure in Wm-2 nm-1? EPIC provides radiances, not fluxes. Shouldn’t the units be Wm-2 nm-1 sr-1? Since there is no information about applying an anisotropic correction for the EPIC radiances, my suspicion is that the authors assumed without justification that the Earth is isotropic and simply multiplied the EPIC radiances by pi. However, it is well known that the correction for anisotropy at direct backscatter angles observed by EPIC are very large (e.g., Su et al papers). Why is this not discussed in the paper?
- 5: Why is the methodology used to construct composite spectra for EPIC described in the caption of Figure 2? This should be in the main text in a methodology section in addition to or instead of in the figure caption.
- 5: Figure 2 caption: The coefficients for the linear combination are 0.05 for solid cloud and 0.93 for the clear desert scene. Why don’t the coefficients add to 1.0? Normally, when combining clear and cloudy areas to represent an area-average, the following is used:
I = f*I_cld + (1-f)*I_clr, where f=cloud fraction.
Also, how is the “best-fit” to observed EPIC radiances determined? Are the EPIC and AVIRIS radiances normalized to a common scale or does the fit use absolute values of EPIC and AVIRIS radiances. If the latter is case, this is problematic. A quick look at the Green et al. (1998) paper indicates that the calibration accuracy of AVIRIS is about 4%, and it’s probably similar for EPIC. The calibration uncertainty combined with the limited set of ten AVIRIS spectra and the need to use VLIDORT to account for differences in viewing geometry likely results in unrealistic combinations of clear and “solid-cloud” scenes/weights.
Is equal weight assigned to each EPIC band or are the EPIC bands weighted by incoming solar irradiance (i.e., energy weighted)?
Along the same lines, the authors appear to treat the AVIRIS measurements as the “absolute truth” and scale the SCIAMACHY radiances to match AVIRIS between 400-450 nm without any justification. It would be helpful to know the scaling factors between AVIRIS and SCIAMACHY to understand how consistent these two instruments are and if the scaling factors vary with time and location.
- 5-6: Figures 2 and 3 show examples where the linear fit appears to come close to the EPIC values. However, the reader has no idea how well this approach works in general and what the overall uncertainty of the method is. Can the error in the fit (without adjustment) to EPIC somehow be quantified across all observed scenes and illumination conditions? Can the error in the fit to EPIC radiance be used to estimate error across the entire SW spectrum?
- 7: “Averaging all pixels for an image yields an adjustment of +0.8%. or ~2.0 Wm -2 and a 1s standard deviation of 16 Wm-2.”
The units are for flux, not radiance. Is this just pi x radiance?
- 7: The authors acknowledge that the integrated energy that needs to be “filled in” beyond 780 nm can be 50% or more of the total. They further acknowledge that the accuracy of their approach rests on how well they can use the 780 nm EPIC channel to predict contributions beyond 780 nm. Yet, linear combinations of two land scenes are not allowed: they only use one spectrum for a thick solid cloud and a single land scene from 20 m AVIRIS data to “fill in” EPIC pixels with a spatial resolution is 18 km at nadir that increases with viewing angle. Furthermore, if the EPIC scene is snow, ice, or “urban”, there are no AVIRIS spectra available. It’s not at all clear how those spectra are “filled in”.
These assumptions/approximations are problematic, particularly since 50% of the energy is missing.
- 7 “Another source of error occurs when the water vapor column of an EPIC pixel is different than the column at the time of the AVIRIS observation. Four strong water vapor absorption bands at 0.94, 1.12, 1.39 and 1.89 microns (Ramaswamy and Freidenreich, 1991) contribute to the spectral shape in the near-IR for all the homogeneous land AVIRIS spectra (Figure 1)”.
The 10 AVIRIS scenes all occur over different parts of North America. Furthermore, the one “solid cloud spectra” cannot possibly represent all clouds. The water column above deep convective clouds is very different from that of trade cumulus.
p.7: “To investigate the ability of the 780mn EPIC measurement to predict the EnIR, we consider all the nadir spectra from 10 AVIRIS flights over a variety of land surfaces (Table 2)…This suggests that if our algorithm finds a land AVIRIS spectra that is consistent
with the spectra of the EPIC pixel, the EPIC 780nm radiance alone will be able to predict EnIR with a 1s error of ~5 Wm-2 (fourth column).”
Only 10 cases are used to test how well the methodology works globally. All appear to be over N. America, and many appear to be located close to those used in Table 1 to train the algorithm. There are no cases over the tropics, high latitudes, etc., nor are any additional cloud cases considered other than the single one used in Table 1. At 18 km, many EPIC pixels will consist of broken clouds. That too is ignored with this very limited set of test cases. Such a limited set of cases is insufficient to assess “the ability of the 780 nm EPIC measurement to predict the Enir” (or ~50% of energy they do not measure).
Figure 5: The annual cycle looks odd. The time series from February to December lacks the well-defined Earth-Sun distance variation. The magnitude of the SW flux is also unrealistically large.
Figure 6 Caption: How is the time difference between EPIC and CERES accounted for? The caption says that CERES SSF reflected “energies” are determined from the CERES SSF flux multiplied by the CERES anisotropy factor at DSCOVR viewing and illumination conditions. If true, this is not the correct way to calculate this. The CERES flux needs to be at the same solar zenith angle as the anisotropic factor. However, the CERES SSF flux and DSCOVR measurement can differ by 3 h. The sentence that follows makes this even more confusing: it says the EPIC-AVIRIS “energy” is interpolated onto the SSF pixels observed within 3 h of the EPIC image time. It seems there may be a serious problem with the calculation. At the very least, the explanation of the methodology (in the caption, no less) is extremely confusing.
Figures 6d and 7d, showing color coded scene types for “solid cloud over ocean”, “clear-sky over ocean, desert and vegetation” indicate huge areas that are white, yet there is no explanation of what the white areas represent. Also, it seems highly unlikely that all of Africa and Europe were entirely either “clear-sky vegetation” or “clear-sky desert” on June 1 Hour 10 (Fig. 7d).
Citation: https://doi.org/10.5194/egusphere-2023-638-RC2 -
RC3: 'Comment on egusphere-2023-638', Anonymous Referee #3, 04 Sep 2023
This was a difficult read. I had to read it several times and I still have unanswered questions about the approach and the validation. I am not sure what the overall purpose of the paper is. The introduction states the goal of the paper, which is to present a new EPIC narrowband to broadband approach, that is independent of CERES fluxes and CERES ADMs. The new EPIC approach was lacking in detail, some of which was embedded in the figure captions, with no supporting validation. The EPIC derived broadband fluxes were then compared to NISTAR and CERES fluxes. The NISTAR instantaneous full disk fluxes had a greater daily “GMT or longitude position cycle” flux range over the day than EPIC, giving no confidence that the EPIC broadband fluxes are capturing the cloud and clear-sky land flux variations across the globe. The authors then compare the EPIC broadband to the CERES radiances, which used the CERES ADMs to convert the CERES fluxes to the EPIC observed angles and conclude that the CERES ADMs are inadequate. This does not validate the EPIC narrowband to broadband radiances. If the authors instead used their VLIDORT based ADMs to correct the CERES fluxes to the EPIC angles and the resulting comparison accounted for angular radiance differences than I would be convinced. The VLIDORT LUT was a key component in the EPIC narrowband to broadband approach to account for angular differences.
I reject the paper based on the following comments.
General comments
I have reviewed my extensive list (more than 50) of SW narrowband to broadband papers from the 1980s and most approaches had access to narrowband and broadband collocated coincident data in order to derive empirical and theoretical enhanced visible channel coefficients and provided the validation to derive the uncertainties. Previous authors suggested increasing the number of identified scene types (surface, atmospheric and cloud conditions), finer angular bins, more channel wavelengths, to reduce the uncertainties.
This paper did not cite one narrowband to broadband approach. The approach presented did not follow any previous (successful) approach.
The world is 65% cloudy. Very little information about the cloud spectra is given in the text, no cloud properties other than cloud fraction is used to derive the spectra. Only one AVIRIS cloud spectra at one angular geometry is used. So little effort was made to account for cloud type spectral differences. For example, Fig. 2 in Li et al 2018 shows the various cloud spectra that differ. Fig 5 in Gao et al 2003, show the spectra between cirrus and cumulus clouds differ dramatically especially over water absorption bands. It is amazing how more reflective cumulus clouds are in the SWIR bands compared to cirrus.
Li, H.; Zheng, H.; Han, C.; Wang, H.; Miao, M. Onboard Spectral and Spatial Cloud Detection for Hyperspectral Remote Sensing Images. Remote Sens. 2018, 10, 152. https://doi.org/10.3390/rs10010152
Gao, B.-C., and Y. J. Kaufman, Water vapor retrievals using Moderate Resolution Imaging Spectroradiometer (MODIS)near-infrared channels, J. Geophys. Res.,108(D13), 4389, doi:10.1029/2002JD003023, 2003
VLIDORT is used to account for hyperspectral differences in view and solar zenith angles. No cloud information (height, thickness, ice/water phase, optical depth), atmospheric profiles, land spectra and surface BRDF (AVIRIS land spectra was based only one angular condition), was given in the text in the VLIDORT LUT.
I believe the success of this approach relies on the VLIDORT LUT accuracy, since the AVIRIS spectra (limited to one angular condition) was adjusted to the observed angular EPIC observations. 50% of the flux contribution is from wavelengths greater than 780nm all dependent on a simple spectral adjustment and the VLIDORT LUT.
Section 3.2. I cannot determine how the column water vapor is accounted for in the NIR water vapor absorption bands above low clouds or clear-sky ocean. Only clear-sky land was mentioned.
The NASA EMIT hyper-spectral reflectances would be a great way to validate this narrowband to broadband approach, especially VLIDORT. The EMIT sensor is on the international space station samples the entire range of solar zenith angles and reflected solar spectra (0.4µm to 2.5µm) mainly over clear-sky land. I encourage the authors to validate their EPIC narrowband to broadband approach using EMIT data.
Page 8 line 8 “The archived Level 1B BandB SW radiances used in this study are converted to unfiltered values by multiplying them by 1.15075 (1.0/0.8690) to account for photons absorbed by the filter.” No reference or explanation how this number was obtained.
The EPIC full disk flux comparison with NISTAR in figure 5 showed that the NISTAR instantaneous full disk fluxes had a greater daily “GMT or longitude position cycle” flux range over the day than EPIC, giving no confidence that the EPIC broadband fluxes are capturing the cloud and clear-sky land flux variations across the globe. I am assuming that NISTAR is the truth dataset for this comparison.
It just feels like the paper was put hastily together based on the following.
Is it AVIRS or AVIRIS, the paper goes back and forth, please pick one acronym, and stick with it. What are energies? radiances or fluxes? many plots do not have units, hard to tell if radiances or fluxes are plotted. The algorithm information was dispersed in the text, appendix, and figure captions. The algorithm was not coherently outlined in a flow chart detailing inputs and outputs. Page 8 line 13 “NDIV threshold”
Figure 6 caption. Please state whether this was Terra-SSF or Aqua-SSF, or NPP-SSF, Is the time, “Hour 00 Minute 45”, in GMT time. Lack of units 6a, 6b, 6c, on the plots. Given that CERES instruments are both in the 10:30 and 1:30 local time orbits, the authors could have plotted the CERES orbit that is to the right of the EPIC center.
Fig. 6 and Fig. 7. The term “reflected energy product” is confusing, there are no units given for either Figure. Are the SSF fluxes plotted in 6a converted to radiance at the EPIC angles?
Figure 6f. The units are in %, why not the same units in 6a or 6f? This will let the reader establish what regions are contributing to the overall EPIC-AVIRIS minus CERES SSF partial disk flux.
Fig. 6 and Fig. 7. The partial disk EPIC pixel level broadband comparisons against CERES fluxes. EPIC was 10 less than CERES for Fig. 6, and 10 greater than CERES for Fig. 7, but the patterns on Fig. 6f and Fig. 7f were similar. If the radiances are weighted by EPIC angle, then the edge contribution is less than the center, for Fig. 6 where EPIC-AVIRIS has a greater flux than CERES in the center (fig. 6f) but yet the overall disk flux is less than (CERES mean 299.5 and EPIC mean 289.3). Is there an explanation?
Figure 6. Why are there white patches embedded in the dark blue in the eastern most orbit and in 6i in the EPIC view angle plot?
Fig.7 For this case Africa is mostly observed in clear-sky conditions. (see Fig. d Africa is yellow and green). For the CERES orbit that is within 30 minutes of EPIC to the left of the black dot, the anisotropic factor varies gradually and is mostly green over land. The orbit to the right and left have strong anisotropic gradients. This is also evident in Fig A-1a and Fig A1b. Does the greater time difference explain the noisy radiances?
Page 10 line 13 “We use anisotropies from the Edition 2 Angular Distribution Models to convert the archived CERES flux to the energy that EPIC would observe based on the camera’s viewing geometry” What was the source of the cloud fraction, optical depth, and cloud phase needed to determine the CERES TRMM ADM scene type?
Page 10 line 13. There are certain locations on the EPIC disk where the CERES and EPIC observations have matching angles. This would be the locations to compare the CERES radiance with the EPIC-AVIRIS radiance without the use of an ADM. I encourage the authors to validate their EPIC narrowband to broadband approach using matched EPIC and CERES angle radiance pairs.
I would highly encourage the authors to use the MODIS channel radiances that are analogous to EPIC and aggregate them into the CERES footprint and compare their narrowband to broadband approach to the CERES flux. This would be a great way to validate the EPIC derived broadband fluxes.
Figure 8. I would also like to see a Fig 8d that has the June 2017 CERES minus EPIC relative radiance difference. There are a lot of horizontal features in Fig. 8a and 8b, that just do not seem to part of the Fig. 8c. The Fig. 8a and Fig. 8b colorbar just does not make sense. There are no relative fluxes less than 160 or greater than 480. Readjust the colorbar between 160 and 480 with 10 colors to increase the radiance resolution.
Appendix, Fig. 6 and 7. and section 5. Is Fig. A-1 “EPIC anisotropy” the ADM factor needed to convert the CERES flux to match the EPIC derived broadband flux. The VLIDORT LUT was used to determine the broadband radiances at all other EPIC angular conditions from the single AVIRIS angular observation. The VLIDORT LUT is an ADM. Why is the VLIDORT ADM not compared with the “EPIC anisotropy”? If the VLIDORT ADM matches the EPIC anisotropy that would validate the EPIC narrowband to broadband approach.
In fact, you could take two EPIC images that are spaced up to 3-hours apart. Then use the VLIDORT ADM to convert one EPIC image angular geometry to the other EPIC image angular geometry and see if the EPIC derived radiance differences are mitigated in the illuminated portion. I encourage the authors to validate their EPIC narrowband to broadband approach by performing this consistency check using the VLIDORT ADM.
Citation: https://doi.org/10.5194/egusphere-2023-638-RC3
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 346 | 126 | 42 | 514 | 40 | 40 |
- HTML: 346
- PDF: 126
- XML: 42
- Total: 514
- BibTeX: 40
- EndNote: 40
Viewed (geographical distribution)
| Country | # | Views | % |
|---|---|---|---|
| United States of America | 1 | 267 | 52 |
| China | 2 | 52 | 10 |
| Germany | 3 | 29 | 5 |
| France | 4 | 21 | 4 |
| Finland | 5 | 15 | 2 |
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
- 267