|I would like to thank the authors for there comprehensive response to all reviewers and the improvement of the paper. |
Most of your arguments are reasonable and I can understand your points.
However, I don't see the point why you are not following reviewers 2 suggestion:
"Furthermore, I would suggest to present two figures where the mean difference and respectively the Stddev is plotted against slope (e.g. binned to 0.05) for the whole DEM, observed and interpolated pixel, respectively. In addition, this Figure should include the same analysis for the other 3 DEMs. Such kind of figure would clearly show the difference between observed and interpolated pixels as well as the stated improvement of the new DEM against existing DEMs."
You argue that: "you don't feel that this way will provide a robust indication of the accuracy of a DEM, as the slope distribution of the evaluation dataset is not representative of the slope distribution of the Antarctic ice sheet"
But on the other hand some lines above in your response you mention:
"The analysis with respect to slope presented in Table 3 accounts for 95% of the total ice sheet area, and gives a good depiction of the accuracy of the overall product"
I still think that providing such a figure will give useful information, especially in contrast to other DEMs. I think it is important to show where the DEM is better than the others and this seems to be related to the complexity of the topographie and hence slope - and this is worth to show.
Table 3 is more a comparison of the different modes. I agree that a split up in observed to unobserved points is not necessary, due to the high coverage of observed pixels. All your arguments (e.g. we stay with the data resolution and don't introduce interpolation errors) are in your reviewer response - why not include this in form of an additional figure and some text in the manuscript?
If you think to have too many figures than skip Figure 3 and include the subfigures and labelling in Fig2.
At P6 L26 you mention the correction of temporal fluctuations in backscattered power. You don't explain how exactly you apply this correction. It is also not included in your equation. Please add more details in the supplements.
To answer your questions to reviewer2:
For the comparison I used the
preliminary version of the CryoSat-2 DEM available for download to the community and the same references as given in your preliminary response (DiMarzio et al., 2009).
As correctly stated the ICESat DEM has a posting of 500m. For providing the figure I used simply QGIS with its implemented raster calculator.
Doing so, the difference DEM was not resampled.
However, to double check I run the GMT grdsample tool and subtracted both DEMs. Additional I run IDL bilinear interpolation as well to generate the difference map. In all three cases the pixel shift is visible as shown in my review.
It's nice to see that the new version seems to be of better quality and without the pixel shift.
Lot's of people are using this preliminary version already, since this version of your DEM was highly advertised in the media, at conferences and widely distributed to the community via ESA webpages as well as at conferences.
To my opinion this bears some problems, and needs to be clearly addressed in the paper.
Therefore, I would like to see two things in a published paper.
1. Please make very clear in the main text that a preliminary version, which was widely distributed, should not be used anymore and explain the improvements.
2. Please show the difference between preliminary and updated version in the supplements (Difference MAP to point out the pixel shift)
I hope that you can reproduce a similar pattern as I observed. I think this is important for the community.