This manuscript had two major goals. First to create an improved algorithm for extracting breakup and freeze-up dates using microwave-based temperature brightness data from satellites on a shallow, and vegetated lake. And second, to re-construct a long ice phenology timeseries extending back in time before the availability of satellite-based estimates. The novelty of the study appeared to be the development of the DMTT algorithm to extract freeze-up and break-up dates. The DMTT algorithm appears to be better suited for mixed pixels more commonly associated with lakes with vegetated cover or complex shorelines.

The authors could use more clarity in their introduction and methods sections, specifically to both highlight the novelty of the study, summarize what has been done in the past, and differentiate better the different types of data. Regarding novelty, I think the authors should spend more time comparing the DMTT and MTT (e.g. expanding more on the improvements made). I believe this will help readers understand which algorithm to use given their own lake data. Further, the authors state that their goal was to understand shallow lakes with complex shorelines and mixed pixels, however they chose 5 pixels from the CETB dataset which only contains pixels that are mostly water. Also, the pixels are not fine resolution enough to capture the complex shoreline. The authors could spend more time justifying the use of this type of data for ice phenology research, especially on shallow lakes with complex shorelines/vegetation, or de-emphasize the mixed pixels and focus more on the shallow lake aspect. In the same vein, I suggest the authors spend some time in their introduction qualifying why microwave-based methods are more accurate or better than optical based methods. One suggestion would be to discuss if microwave data can improve upon the errors associated with missing the freeze-up or break-up dates due to cloudy pixels, which is ~5 days for the MODIS method (Zhang and Pavelsky, 2019) or whether microwave data better corresponds to in-situ data, if available. Finally, the authors could more clearly differentiate what data were used and where they came from. I liked the idea of Figure 2, but I found it hard to follow. I would suggest starting with raw data on the left and moving to more derived products on the right (or the same suggestion but top to bottom). All external validation data should be clearly marked versus the data use for testing and training in the RF model. I put more detailed comments about clarity throughout the methods section.

Specific comments:

Line 10 is this multi-source? Can you be more specific? Is this ERA5 data or other types of data?

Lane 23 consider “global lake ice loss is a prominent feature associated with climate change” or simplifying this beginning sentence so it is in more active voice.

Line 25 what do you mean by ecological balance?

Line 26 can you elaborate on how ice cover impacts the momentum and mass exchange between the atmosphere and water?

Line 28 see also Sharma et al. 2019, consider wording similar to: Many studies have evaluated lake ice phenology trends in the northern hemisphere, but shallow and vegetated lakes remain largely unexplored due to their lack of observational data

Sharma, Sapna, Kevin Blagrave, John J. Magnuson, Catherine M. O’Reilly, Samantha Oliver, Ryan D. Batt, Madeline R. Magee, et al. 2019. “Widespread Loss of Lake Ice around the Northern Hemisphere in a Warming World.” Nature Climate Change 9 (3): 227–31. https://doi.org/10.1038/s41558-018-0393-5.

Line 32 The sentence “shallow lakes are more sensitive to climate change then deep lakes because of …., because of their potential ecological effects on algal blooms” does not seem like a complete sentence. Are you trying to say that the fact that these lakes get algal blooms mean that they are more sensitive to climate change. Lake ice and ice formation seem to clearly stem from climate change sensitivity, but I feel that algal bloom formation is related to several factors including temperature, ice, nutrients, etc.

Line 34 what do you mean by microlevel, microscopic, lake-specific, etc.?

Line 37 Consider Hampton et al. 2017 (or citations within) if you want to make a broad statement of the need more lake ice research irrespective of shallow lakes, otherwise make this statement more explicitly about shallow lakes

Hampton, Stephanie E., Aaron W. E. Galloway, Stephen M. Powers, Ted Ozersky, Kara H. Woo, Ryan D. Batt, Stephanie G. Labou, et al. 2017. “Ecology under Lake Ice.” Edited by James Grover. Ecology Letters 20 (1): 98–111. https://doi.org/10.1111/ele.12699.

Line 47 what is the temporal resolution?

Line 57 consider rewording the sentence that starts with “in addition…” This sentence marks an important transition from your text about creating phenology records using satellite data and creating a forecasting/back casting model using meteorological data. Here maybe you could emphasize past work using satellite (optical/microwave) as training data to create longer term time series

Line 68 using which data data?

Line 70 using the created data in step one as training data?

Line 80 Is there a citation associated with the evaporation and precipitation data?

Line 100 Am I correct that the phenology record using CETB was not used in the training of the random forest model, only the phenology record derived from optical satellite data?

Line 100 meteorological data from ERA5?

Line 113 What is the spatial resolution

Line 125 the ice phenology data originated from the cloud bitmask in the state_1km product of MODIS? How does that reduce the influence of clouds?

Line 131 Which part of Sentinel-2 and Landsat did you use to create your threshold? Can you elaborate more on the specifics of this method (e.g. Fmask on Landsat, and a threshold of X in the MODIS red band reflectance from the 250 m product)

Line 135 Did you use the phenology record directly from Huo et al. 2022? If so, then I would put that citation further up

Line 179: Did you mean “we could not use the omega-shaped analysis for this study because the “w” shape violates that method”? The sentence as it currently reads is a little vague as to why the omega series are not available

Line 180: Can you use this algorithm from all W-shaped series? How common is W-shaped series?

Line 190: I would add more information here about how your algorithm is different from the MTT since I feel like this is where the novelty of the study lies. Much of that information is in the appendices, and I would move that information into the main manuscript.

Line 193: How complex are the mixed pixels? The five that you show were mostly water and not near the edge. Do you still expect these pixels to be mixed?

Line 199: How does including rules like “above 0 degrees” and “below 0 degrees” make mixed pixel classification better? I think I understand your logic, but I would like it spelled out in the manuscript

Line 201: what does the bar symbol above Tb1 mean? What does b1 mean and b2 mean?

Line 202: If  means the average over 20 days, then how can there be a minimum?

Line 214: Were the multisource Sentinel, Landsat, and MODIS?

Lines 215 -220: I think some of this text belongs in the results section to test the validity of the DMTT model

Line 235: which phenology datasets did you use? Which meteorological datasets did you use?

Line 255: Did you validate or ground truth your model using in-situ data?

Figure 6: How did you choose 1982 as the threshold change? I would include significance values in the figure directly. For example, only drawing lines and indicating trends if the results are significant

Line 284: Do you have a citation for the statement “ice phenology is primarily determined by local meteorological conditions”?

Line 292: This comment about Lake Mendota feels more like a methods comment

Figure 7: I would recommend moving this to the supplementary text, I feel that the preceding text does a good job summarizing the information.

Line 331: Is there evidence that the preceding 12 months weather data could impact ice formation? I am not sure that each monthly correlation is necessary.

Line 330: I am not sure that I followed the logic regarding solar radiation and algae and attenuation. I was wondering if the authors would consider expanding on this mechanism

Figure 8: Are you correlating FUS and BUE and ICD with months that occur after ice cover? It seems that in BUE and ICD the months stop at March, which feels appropriate. But, for freeze-up start, the months from Dec to Mar occur after the event and cannot correlate with the event as they cannot influence the event anymore.

Section 5, Discussion: I would recommend starting this section with an overarching paragraph of main findings before delving into each subsection

Line 274: If the pixels chosen in the CETB database are such that they reduce the effects of mixed pixels, how do you know if your DMTT algorithm does a better job at overcoming mixed pixels.

Line 298: I would elaborate on this further. If most lakes in the Northern Hemisphere that have data are deep lakes, but shallow lakes are abundant? Then what about the comparison between the two makes it interesting? (I agree with you that it would be interesting, but I would like to know your reasoning)

Line 402: Can you qualify why you chose to compare Lake Ulansu with Qinghai Lake?

5.2 Comparison with other lakes: In general, I think this section needs to be better qualified, why did you choose the comparisons with Qinghai Lake or the northern European lakes? What are some major consequences of the similarities/differences? How do you think adding shallow lakes will change our contextual understanding of lake ice, globally? Or locally?

5.3 Conclusion: This section is extremely well written and summarizes the study well

Line 455: Is there a citation associated with this, or is this in comparison to the other lakes in section 5.3?

Technical corrections:

Line 54 add CERP acronym here