Lea et al present a map of ridge and valley features based on satellite observations of the surface. They validate the inferences using the dense AGAP radar survey. This technique allows mapping beyond the radar survey, and, in theory, resolution at length scales shorter than the radar line spacing. The technique is clever and offers additional information to the radar data. The paper would be much improved from a more quantitative validation with objective definitions of ridges and valleys in the radar data as the qualitative assessment (i.e. Figure 6) is not particularly convincing. The technique, with additional quantitative validation, will be a useful addition to tools for inferring subglacial features in regions undersampled by radar data.

The technique is, in general, well described. It would be useful to expand Figure 4 to include both a valley as well as short length scale identifications of valleys and ridges, e.g. segment 10 10-16km. Figure 4 is a good example, but likely lacks many of the complications that arise between manual and automated interpretation.

The validation suffers from a lack of quantification of valley and ridges in the radar data. How a ridge and a valley are defined in the radar data is not clear. The authors provide statistics of the percentage of features that are correctly identified, but do not describe how the “true” features are defined. Figure 6 shows the inferred features, and the bed data, but not the “true” ridges and troughs defined by the radar data. Without a specific definition of ridges and troughs,  quantitative metrics of how many features were correctly and incorrectly identified do not convince me. In Figure 6, segment 10, 5 features are identified between 10 and 16km, but the features are very small in scale and more significant features between 3 and 10 km are not identified. It would be much more useful to have the “true” ridges and troughs marked, solid lines that indicate what the authors consider a successfully identified feature, and dashed lines for incorrect features. Further, the authors need to define a distance that the mapped feature must match the true feature – or possibly define a ratio of distance to amplitude of feature.

It is surprising to me how often the same features are identified successively (i.e. 2 or more ridges in a row). For instance, Figure 6, segment 9 has 3 ridges between 5 and 10 km without any valleys in between. And then the only valley is identified at a ridge at just over 15km. It seems like a sequence of multiple of the same feature in a row would be a good check that could be implemented into the automated routine to help improve its performance.

Figure 6 also reveals that the same scales of basal features are not consistently identified between segments. For instance, small features are identified in segment 10, but are not identified in Figure 9. The bed elevation is similar between them, so it’s unlikely that a greater ice thickness is obscuring the bed variations.

I would also like to see a more quantitative analysis of the acceptable distance between a “true” and a mapped feature. There are many instances in Figure 6 where there are potentially multiple features in the data that the mapped feature could be assigned to.

While most of my comments have to do with the validation, there are a couple of other areas where the paper could be improved. A morphological analysis of a few alpine regions that could be directly compared with GSM would be very informative. How do the Rockies compare quantitatively? Is GSM closer to the northern Rockies (i.e. Glacier NP) or the southern Rockies (ie. The San Juans in CO). Or I’m sure you come up with better analogs to determine the extent of glaciation that may have occurred.