The paper titled 'The Chalmers Cloud Ice Climatology: Retrieval Implementation and Validation' by Amell et al. highlights the ability of a neural network to estimate cloud ice properties, including ice water path and ice water content, using only the 11-micron IR channel from geostationary satellites. This innovative approach shows promise for creating a long-term quasi-global record of IWP and IWC. While the paper is well-written and includes multiple statistical examples demonstrating the efficacy of the machine learning technique, it lacks maps and curtain plots illustrating the geographic representation of CCIC retrievals. To convincingly demonstrate the representativeness of these estimates, such examples are essential. The authors could add for example:

• Monthly global IWP maps showing the CCIC against the cloudsat estimates (no need to subsample the CCIC, just show that the global distribution is as expected)
• Percentage difference in these types of maps.
• A latitude – altitude cross section of cloud ice fraction, again comparing versus the cloudsat one.
• global maps of IWC at different levels showcasing the variation with height
• Cross sections of IWC through different longitudes

I understand that this is considerable work and hence I suggest major revisions.

Specific comments:

The title is not representative of the context of the manuscript. There is no mention of constructing a climatology or anything of that sort.

Line 41: “For the study of processes on annual and decadal scales it is therefore necessary to find ways to make better use of observations with a long record of availability”. The authors should mention that several IWP records exist with annual and even decadal scales, such as the ones from MODIS, Aura MLS, Odin SMR, CloudSat, etc. As currently written, the introduction implies that such records do not exist.

Further since CCIC provides IWC, the authors could compare partial IWP versus those records matching their respective altitude coverage.   The comparison versus the campaigns is limited to a few periods and it is limited geographically.

Line 17: “considerable skill” is a qualitative description please provide a more quantitative description.

Line 20: “first order”  is a qualitative description please provide a more quantitative description.

Line 45: please describe the rationale behind only using the 11micron channel. Presumably additional channels could provide more information.

Line 55: “Estimates of TIWP differ widely between”. Please give the ranges, this would allow you to later show how well (or bad) CCIC estimates are.

Line 87: why not use lat lon info as well? And day of the year?

Line 108: The use of “2D” here is confusing since the authors are talking about profiles, I suggest deleting it.

Line 113 – Line 119: A schematic of this entire procedure will be appreciated. Also, what is the treatment for the uncertainties in 2C-ICE

Line 136: “Training scenes of 384x384 pixels”. Is the geographical size of this scene important? Is there an impact for using smaller or bigger scenes? Why this particular size.

Line 136: “The process involved randomly selecting a pixel with valid reference data as the starting point and then adding a random zonal offset.”  I don’t really understand this please clarify

Line 140: lower -> coarser

Figure 2 caption: Why is the brightness temperature normalized?

Line 155:   Good is the enemy of great. The authors should explore tuning of these parameters or rephrase this sentence to state that minimal tuning was required.

Figure 3, and 4 captions: specify the locations. for example, Aug 2015, and Jul-Aug 2018 flights were over the US and the US nearest oceans.  Or, Flights took place over the Olympic Peninsula in the Pacific Northwest of the United States,

Line 229: Which of the four periods is the Darwin campaign?

Line 237: Specify frequency

Line 247: Which year?

Figure 5: panels g and h should say Cloud classification and cloud classification (retrieved) respectively.

Figure 6 caption should mention cloudsat somewhere, as well as the period use for this comparison.

Section 3.2.1.  It is not clear which period this comparison cover.

Line 297: This should be shown as a separate subsection to emphasize its importacnce: Zero order comparison of IWP and IWC (for example)

Figure 7, 8 and 9 caption: Add cloudsat somewhere

I think the whole classification is barely working and the authors should just not show any of those results.

Figure 13 is missing the conditional mean line

Line 417: CCIC is not representing the diurnal variability well, it is really flat.

Line 438: provide an estimate of the uncertainties associated with the estimates of the ice hydrometeors.

Line 456: “Despite these encouraging results, CCIC should still be considered a proof of concept. CCIC’s principal objective remains to explore the potential of modern deep-learning techniques to expand the observational climate record of ice clouds”.  This should be mentioned upfront in the abstract and the introduction.

Table A1: “Cloudy pixel” Cloudsat or retrieved? how come the cloudy pixel is 40ish while the no cloud is 97%, please clarify

See attached PDF