I have reviewed the paper entitled ‘Improved records of glacier flow instabilities using customized NASA autoRIFT applied to PlanetScope imagery’ submitted for consideration for publication in The Cryosphere. Using five test glaciers, this manuscript presents improvements to optically derived glacier velocities using a modified version of the NASA autoRIFT and the inclusion of PlateScope (PS) imagery. Overall, enjoyed reading this manuscript and congratulate the authors on a well written and clearly presented manuscript. The authors provide a detailed description of the processing steps, and their test cases clearly demonstrate the improvement in terms of the temporal density of velocity estimates that can be used to understand variations in glacier flow for surge-type glaciers. The only major comment that I have to improve this manuscript is to encourage the authors to include a comparison between the PS and in situ derived velocities. For example, Figure A3 presents in situ derived GPS displacements, which seem to match up with at least some of the PS derived velocities. This would provide a nice comparison/documentation of how well the PS derived velocities perform in very fast flowing situations. Other than this, I only have minor comments listed below to be addressed. 

L2/3: ‘glacier velocity records are sometimes incomplete due to gaps in the cloud-free satellite image record’ – here you may want to tweak the sentence and specify that cloud and night impact glacier velocity records derived from optical imagery. This is not an issue for glacier velocities derived from SAR imagery (although those that are derived from SAR have their own nuances as well). 

L12: ‘dramatic changes’ can you quantify this here?

L63: I believe that custom autoRIFT should be in quotation marks. 

L72: ‘in the Yukon’ change to ‘in Yukon’ (‘the’ is not necessary)

L70/75: Can you provide a sentence here that rationalizes the choice of these five glaciers? 

L161: Can you expand on this? How many images were manually filtered out? How user intensive is this process? I expect that it is rather time consuming.

Figure 2: All the Matlab figures should have colourmap labels.

L163/164: Why is a DEM needed? Please add a sentence to describe why this is needed for completeness. How were the use of different DEMs determined? For example, for the Svalbard site, ArcticDEM was available, so why was it not used instead of GMTED2010? Maybe it doesn’t matter, but there should be some justification for the DEMs used at each location. How much to the resolution of the DEM matter in terms of the quality of the final velocity maps?

L269/270: Can you clarify this sentence; you are comparing median stable surface errors from custom autoRIFT to maximum (SSE?) errors from ITS_LIVE? So, are you comparing the same thing? 

Figure 6: It feels to me that the scaling for a)b) and c)/d) and e)/f) should be done independent of each other. For example the maximum of c) and d) should be set to 20 m/day and for e) and f) it should be set to (maybe) 15 m/day. I know that you want consistency to compare across these three glaciers, but to me the more important aspect to showcase is the custom autoRIFT performance at each site. At present, with the current scaling applied, the reader cannot see the variations across the glacier in c) and e) because all the variation falls within the purplish colours.

In this paper, the authors present a customized/modified version of NASA autoRIFT, along with the inclusion of PlanetScope (PS) imagery, to produce improved records of glacier flow (and flow instability) for five surge-type glaciers. In nearly all cases, the customized pipeline and additional imagery provided a greatly improved time series of glacier flow speeds, as well as clear improvements to the consistency and structure of the derived speeds. I found the manuscript both well-written and easy to read, and think that it should be published after consideration of minor suggestions/revisions.

l. 7,26: spell out ITS_LIVE at first use (in both abstract and introduction)

l. 28: should the reference here be Gardner et al. (2018), rather than 2023?

l. 48,100, elsewhere: to override the annoying way that LaTeX parses non-standard author names, wrap the name in two curly brackets, e.g. {{Planet Labs}}

l. 89: how many images does this correspond to? Can you give a range, similar to the file sizes?

l. 98: What is the purpose of this geotiff - does it need to be, as mentioned later (l. 103-105), a re-interpolated satellite image of the area, or is it enough to just have a "blank" image with the desired resolution and extent? If the latter, would it not just be easier to specify an extent and spatial resolution?

l. 105: include the version of GRASS that you have used and a citation (e.g., https://zenodo.org/records/10817962)

l. 130: why would the download step fail based on the number of images?

l. 137-144: this seems to assume that the georeferencing errors do not vary spatially throughout the images. Is this really the case (I suspect not, as it will be related in part to the DEM used for orthorectification), or is it simply "good enough" for most applications to make this assumption?

l. 144: this is more of a thought/suggestion for future development, rather than for the manuscript. Instead of choosing the second image for co-registration, you could instead have the user provide (or download) a reference image (e.g., Sentinel-2), then co-register each of the PS tiles to that reference image. This way, you aren't left making the assumption that the "first" image or swath in the pair is correctly georeferenced.

Figure 2: can you indicate in the caption what the calculated offsets for the coregistration step were?

Table 1: Do you have an idea/indication of the impact of DEM quality/resolution on the final output? Additionally, you have used both GMTED2010 and GTOPO30, although GMTED2010 is meant to be a replacement/updated for GTOPO30, which is quite coarse in many areas, especially mountainous regions. Is it just the case that the differences between those two datasets for South Rimo are negligible?

l. 206-210: can you give an indication of the improvement/error reduction that is achieved using the optional coregistration steps?

Figure 5: Can you include an indication of the ITS_LIVE coverage here, perhaps as a line plot alongside the histograms? Or will it be the same as the S2/LS bars, as this may only show the potential pairs rather than the successful pairs?

l. 252-268: would it be possible to include a table summarizing the statistics derived for each time series for each glacier from the custom autoRIFT results and the ITS_LIVE record? In other words, one row for each glacier and two sets of columns showing the peak speed, average speed during the surge initiation (if possible), median SSE, and number of observations derived from your results and ITS_LIVE based on the points shown in Figures 6 and 7? I think this would help quickly illustrate the clear improvements in coverage that your approach provides.

l. 269-270: this appears to be a comparison of the median errors against the maximum errors - it would be better to compare like against like here.

Figures 6,7: I find the improvement in resolution hard to see in the right-hand column - perhaps increasing the size of the markers for the ITS_LIVE observations would help better illustrate where they are absent?

l. 293-295: some of these boundaries appear to be between different tiles, rather than swaths - is this a radiometric issue, or a georeferencing/orthorectification issue?

l. 307-309: can you quantify how much higher the SSE was for the S2 velocity map?

l. 312-315: would it be possible to illustrate the improved feature resolution in a supplemental image that shows the small-scale crevasses in a PS image vs an S2 or Landsat image?

Figure A3: can you also show the S2 map alongside the PS map here, as a reminder?