Analysis of PlanetScope Dove Digital Surface Model Accuracy Using Geometrically Simulated Images

Round 1

Reviewer 1 Report

 Obtaining Digital Surface Models (DSMs) of sufficient quality from these images is challenging due to their lower 12 resolution and weaker stereo geometry relative to stereo mode satellites. To determine the stereo geometries required for obtaining quality DSMs from PlanetScope Dove imagery, the authors applied a new methodology for generating simulated stereo images of varying geometries using adjusted orientation parameters obtained by a self-calibrating bundle adjustment and validated by comparing the resulting rigorous sensor and rational function models. The results provide a basis for automated stereo imagery selection to enable large-scale DSM production from PlanetScope Dove imagery. This is a good attempt to obtain the DSM from the images.

Main comments:

The organization of the manuscript is problematic, particularly the Section 4. I recommend to relate the discussions directly to the simulation results. Also, a Conclusion section should be added.

Moreover, in Section 3, many simulation results are provided, but the scientific descriptions of the results are poor.

There are abundant parameters used in the paper, but the description of the parameters is not enough. E.g., “the rotation angles of ?, ?, ?”, please give a clear word description in the paper.

There are not clear problems in English expressions.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

The explanation of equation (1) is not entirely clear. For example, what is "T", etc.

The explanation of equation (7) is also incomplete. Eg what is "R" or "Q"?

On the figure 3 I cannot find the k rotation marked.

In the row 413 the "stimulated images" are originally simulated?

What was the basis of the reference DSM prepared by the USGS? Does the reference model involve trees in the area? From the point of view of the foliage, didn't too much time pass between the reference surface and the photogrammetrically derived surfaces?

Figure 10 is quite far from its first mention.

Does "CA" mean Convergence Angle?

In the Discussion, we do not get comparative information about the obtained results, in relation of other literature. A more comprehensive discussion is needed.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 3 Report

 

This paper aimed to investigate the accuracy of DSMs generated from simulated PlanetScope Dove digital images, which in order to get some findings for stereo images selection. I do believe that the idea of the article is of practical significance.

Here I have some comments on the methodology and result analysis.

1.     The major comment to methodology is that the retrieval of orientation parameters is not necessary for the accuracy analysis. Using a DOM which generate from PlanetScope Dove images or other source and reference a DSM can performance the simulation and accuracy analysis.

2.     The accuracy analysis of the article is not systematic enough. What are the main sources of errors of the DSMs generated from simulated images? What is the distribution pattern of DSM error size? How did it happen? These issues need to be elaborated on in more detail in the article. Perhaps some theoretical analysis can be done and cross validated with the simulation analysis in the article.

3.     As mentioned in the article, matching accuracy is crucial in stereo selection, and the lighting conditions in image photography are an important aspect that affects matching accuracy. The simulation analysis in the article did not analyze this aspect for convenience, which limits the reference value of the article. In addition, The DSM errors caused by differences k angle (Fig 13) interpreted as the insufficient matching accuracy. Can we improve the match method to compensate it? Such as doing the image match on resampling epipolar images.

 

 

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 4 Report

First of all, I would like to thank the Editorial Board of Remote Sensing for giving me the opportunity to participate in the review process of this manuscript.

The paper is about the generation of digital surface models (specifically, about the altimetric accuracy of the generated models) from images simulated from PlanetScope Dove images. In this sense, it can be considered that the work fits perfectly with the topics of this journal, and also addresses a topic of great interest such as the generation of digital surface models from images captured by sensors installed on satellites.

On the other hand, it should be noted that the structure and length of the article can be considered as correct.

The problem I find in this article is the design of the applied method itself. As the authors recognize, the geometry of the PlanetScope captures is not the most suitable for the generation of models from photogrammetric methods. It is evident that the capture scheme is radically different from that used by classic stereoscopic satellites such as Airbus' SPOT and Pleiades, and Maxar's WorldView family, which capture both images by tilting the sensor within the same orbit, with time differences of seconds (which also guarantees an important radiometric continuity in the stereoscopic pair or trio). It is evident that PlanetScope is not intended for this and the proposal is to try to obtain this type of product (digital surface models) from these images. It is understood that trying to take advantage of the lower cost of PlanetScope images, although they present problems of lower coverage and radiometric differences between the potential images to form a stereoscopic pair.
The scheme is based on the generation of simulated images from digital images, but for this, it is required to have a reference digital surface model (e.g. SRTM). This fact is the one that really, in my opinion, invalidates the methodology, since finally what is being incorporated into the stereoscopic pair is the altitude derived from the global model. In reality, the transformation proposed is the generation of a new image in which a relationship between XY (planimetric coordinates), Z (derived from the global model), and a simulated image to which a certain position (XoYoZo) and certain orientation angles (omega, phi, kappa) are applied. It is evident that when we try to generate the model from the simulated images, what we will actually be generating is the same global model used for its construction, with small logical differences, of the radiometric differences that may affect the stereo matching process.
Therefore, although the subject of the paper is interesting, I consider the methodology to be incorrect, and the article is not suitable for publication in Remote Sensing.

In my opinion, the English is correct.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Round 2

Reviewer 3 Report

The article has been improved after modification. I recommend publishing it.

Author Response

I would like to thank for the valuable feedback and comments

Reviewer 4 Report

No significant changes have been applied in the second version of the paper so the sense of my assessment remains the same.

The authors state that the objective of the work is not the generation of models, but the analysis of the optimal characteristics to generate synthetic models. However, it is necessary to take into account that this sensitivity analysis -without assessing its real usefulness- will in any case be marked by the DEM used for the generation of the simulated images.
The optimal characteristics for this type of satellite image acquisition is a topic that is already well known and applied.

 

Author Response

As explained in section 2, the radiometric difference between images can lead to significant variation in the quality of DSMs, even when pairs have identical stereo geometries. To mitigate the radiometric uncertainty and focus only on the stereo geometry ( particularly for convergence angle of less than 10 degree in PlanetScope Dove images), the proposed method of utilizing geometrically simulate images is the optimal approach, so that this paper utilizes the simulated images to provide more rigorous result based on the stereo geometries. Therefore, the sensitivity results presented in the paper are relative to the stereo geometry and do not represent absolute accuracy of DSMs, as discussed in section 4 and 5. These findings offer valuable insights into the selection and configuration of stereo pairs for generating PlanetScope Dove DSMs using real images.

We will continue to research the generation and accuracy validation of PlanetScope Dove DSMs with real images, but it requires sufficient amounts of images and a development of multiple-pair stereo matching algorithm as described in section 5.

We will be pleased to have the papers related to PlanetScope Dove satellite for further research, as the reviewer’s comment of “The optimal characteristics for this type of satellite image acquisition is a topic that is already well known and applied”