While the referees see potential in this submission they also express some serious concerns about the superiority of this approach.
In order to demonstrate the advantage of your method you will need to define quantitative criterias.
Please provide an amended manuscript that will include a real assessment of the method superiority as well as the changes required by the reviewers.

While I see some efforts towards providing sone validation of sort, this manuscript is still lacking a clear validation scheme.
The results are significantly different to published crustal thickness maps.

I suggest to provide 2 kind of validations.

1- validation of the method. This aspect is now a bit clearer but the authors need to describe the efficiency of the approach (for example computing time, training sample size etc.).

2- validation of the results. The authors need to demonstrate that the crustal thickness model is correct and that the differences to published data are not artefacts.

I will allow this last round of revision at the condition that the authors take the reviewers comments seriously. The reviewers identified SERIOUS and JUSTIFIED problems in the manuscript. While they see the potential of this submission they do not favor publication at this stage. I concur.
1- demonstrate (validate) that deep learning is justified AND outperform more conventional ML
2- discuss critically the results

Reformulate your conclusions to reflect your own findings. Tune down your results and enhance the added value of the deep learning approach.