the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 12 Feb 2025
Errors in stereoscopic retrievals of cloud top height for single-layer clouds
Abstract. Multi-angle stereoscopic methods are a promising means for retrieving high-resolution cloud volumes and their temporal evolution. Stereoscopic retrievals assume that light emerges from localized points on a surface. We assess the errors introduced by this assumption using synthetic measurements at various wavelengths, solar-viewing geometries, and spatial resolutions generated by applying a 3D radiative transfer model to an ensemble of 841 (8 km)2 cloud fields of varying fractional cover, cloud-top bumpiness, microphysics, and optical depth. We show that stereoscopic retrievals of cloud top height (CTH) have biases that vary from -175 m to +20 m as the cloud-edge extinction profile becomes sharper and absorption increases, all when mean visible cloud optical depth is greater than 5, and with little dependence on instrument resolution between 50 m and 250 m. Stereo CTH fields are smoother than the truth when CTH variability is concentrated at small spatial scales, viewing angles are oblique, and absorption is weak. We attribute this effect to both the smoothing effect of multiple scattering which is stronger at wavelengths with weak absorption, and the ill-posed nature of the retrieval in the presence of non-uniform CTH over the stereo matching window. The standard deviation of stereo CTH errors increases from 25 m to 200 m as the standard deviation of CTH increases to 200 m over the (8 km)2 domain. More than 50 % of stereo retrievals from two different 50 m resolution stereo viewing pairs of (0°, +38°) and (-38°, 0°) are consistent to within 30 m over (500 m)2 regions for clouds with standard deviation of CTH less than 200 m. We analysed airborne lidar observations and found that 75 % of shallow cumulus and all stratocumulus have standard deviations of CTH less than 200 m over 8 km transects. These results support the application of time-differenced stereoscopic cloud top height retrievals for the remote sensing of high-resolution cloud dynamics as well as macrophysics.
- Preprint
(9572 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 19 Mar 2025)
-
RC1: 'Comment on egusphere-2025-20', Yongbo Zhou, 17 Feb 2025
reply
General comments:
The manuscript thoroughly assessed the stereo-opacity bias in an idealized case, i.e., the observations were represented by model equivalents simulated by SHDOM, a widely used radiative transfer solver, based on 3D cloud fields derived from a stochastic generator (Loveridge & Di Girolamo, 2024). The findings are robust since the stereo-opacity bias was explored under different circumstances covering different cloud coverage, cloud-top bumpiness, etc. The authors presented a solid study which helps the understanding of the stereoscopic retrievals by satellite imagery from visible to infrared wavelength. Therefore, I suggest the manuscript be published after some specific comments were properly addressed.
Specific comment 1:
Introduction, line 48-49. “This is due to their ability to achieve high (< 50 m) resolution and precision retrievals of cloud boundaries from multiple viewing angles”. I checked the two references and found that the satellite data were gridded at a resolution of 2.9 m (or 58.1 m) in Castro et al. (2020) and of 50 ~ 200 m in Dandini et al. (2022b). My doubt is that the resolution of the retrieved CTH should be more related to the raw satellite data rather than stereoscopic retrieval technique. In my opinion, the spatial resolution of satellite instruments provides the high limit to the resolution of the retrieved products.
Specific comment 2:
Methodology, Section 2.2. The periodic horizontal boundary conditions were used for the 3D radiative transfer simulations by SHDOM. It is a useful approximation to account for horizontal photon transports periodically near the boundaries of a finite domain, since an adequately large domain would cause heavy computational burden. However, the radiative transfer simulations of incoming and outgoing photons in the horizontal directions should be less accurate near the domain boundaries. I was wondering whether the grids near the domain boundaries were discarded or not, could the authors please give an explanation of why?
Specific comment 3:
Methodology, Section 2.3. The stereo matcher is a crucial step of the stereoscopic retrieval technique. The authors already introduced this part very well. However for a potential reader (like me) who is unfamiliar to this technique, this part is hard to understand. Therefore, it would be really helpful if the authors could provide an example showing how the stereo matcher works.
Specific comment 4:
Results, Section 3.4, line 491-494. I am confused how the results in Fig. 3 & 7 could led to the conclusion that two, time-differenced sets of stereo retrievals may be highly precise when detecting a change in cloud top height over a short time interval. An explanation would be really helpful and appreciated.
Specific comment 5:
The results of this study are based on idealized cases with a lot of approximations, including the sea surface with a pre-defined reflectance of the MODIS 0.86 um channel is 0.0531 or 0.0594, the limited solar zenith angles (30°, 45°, and 60°), and stratocumulus clouds with flat cloud bases. I have two major concerns to extend the findings to real-world cases.
1) The robustness of the findings could be further strengthened by including more solar zenith angles since 3D radiative transfer simulations are highly dependent on the solar zenith angles. But I guess including more solar zenith angles could make the 3D radiative transfer simulations rather time-consuming. If so, I would suggest mention this limit in this part or other parts.
2) The study only discussed liquid water clouds. I am also curious how the findings would change if the stereoscopic retrieval was applied to ice clouds. For the climate research or precipitation processes, the macro-physical properties such as CTH should be also important. I would like to see some discussions on this topic in the revised manuscript.
Citation: https://doi.org/10.5194/egusphere-2025-20-RC1 -
RC2: 'Comment on manuscript by Loveridge and Di Girolamo', Anonymous Referee #1, 06 Mar 2025
reply
In my opinion, this is an excellent paper that will provide helpful guidance to the community in designing instruments and data processing algorithms for stereoscopic retrievals of cloud top height. The methodology is sound, and the presentation is of a high quality. A particular strength of the manuscript lies in the in-depth explanations and discussions it offers. Even so, the manuscript needs some minor improvements. Please find my specific comments below.
Contents:
Lines 134, 327-328, and Section 3.5: I wonder if it is important that the paper compares cloud top height variability statistics obtained differently for observed clouds than for simulated clouds. Specifically, for clouds observed during the CAMP2Ex campaign, the paper calculates the standard deviation of cloud heights encountered along an 8 km long straight line—whereas for simulated clouds, the paper uses the standard deviation of all cloud height values within 8 km by 8 km areas. It may be worth adding a brief note into Section 3.5 about whether the comparisons would look different if (instead of calculating an overall standard deviation of all cloud height values within a simulated field) we calculated cloud height standard deviation values for several individual 8 km long transects and then we used the mean of these standard deviation values to characterize cloud top height variations within each simulated cloud field.
Lines 133-149: It would help to clarify how the geometric and optical thicknesses of individual cloud columns are related to each other. Within an individual cloud field, do geometrically thicker clouds tend to have larger (or smaller) optical thicknesses and/or column-average extinction coefficients?
Lines 289-291: If my interpretation is correct, the wording should be changed to clarify that the sampling bias is the difference between the true heights of two sets of pixels (identified by colors other than gray or dark blue in Figs. 1b and 1d, respectively). This is needed because the current wording suggests that the sampling bias includes the differences between the non-zero height values shown in Figs. 1b and 1d. (The difference between the values in Figs. 1b and 1d would not provide the sampling bias, as Fig. 1d shows retrieval results that are affected by retrieval errors as well as sampling issues.)
Presentation:
Line 193: It would help to explain what four relative azimuth values are the possibilities when selecting an azimuth value for each cloud and solar zenith angle.
Lines 209-211: This sentence contains the words “utilize” and “which” twice; one of the occurrences of each word should be replaced.
Line 248: I suggest replacing “Errors statistics” by “Error statistics”.
Line 346: I recommend changing “As above” to something like “As in the left panel”.
Lines 480-481: I suggest replacing “comparative” by “comparable”. Also, it is unclear how providing a comparable amount of information implies that the information provided by the two predictors is independent.
Figure 10: The caption should clarify what the grey and black lines (at 30 and 45 m, respectively) represent.
Line 582: The reference Loveridge and Di Girolamo (2024) is missing from the reference list.
Lines 609-610: I recommend refining the wording, as extinction profiles do not have a magnitude.
Line 717: I recommend changing “cloud” to “clouds”.
Citation: https://doi.org/10.5194/egusphere-2025-20-RC2
Data sets
Simulated Stereo Cloud Top Height Retrievals Jesse Loveridge https://doi.org/10.5281/zenodo.14509808
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 100 | 19 | 6 | 125 | 3 | 1 |
- HTML: 100
- PDF: 19
- XML: 6
- Total: 125
- BibTeX: 3
- EndNote: 1
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1