Overall good research and worthwhile of publication with revisions. Further details needed about the experiment setup and how correction coefficients were established. Given the current details it would be difficult for a reader to replicate this work. Additionally, minor spelling and language changes are needed.

Despite these notes, having a 30-month analysis alongside a CRDS for low-cost sensors is impressive, significant, and worth publication. I was excited to see this work, and long time-frame pieces like this can further the research community.

Comments:

Line 49: When referencing Picarro or ABB-LGR, might want to reference measurement techniques those instruments use (CRDS, etc.)

Line 50: Awkward phrasing for cost comparison, might want to use a direct cost range instead

Line 58: Which cities in particular are being referenced?

Line 76: “deployed to co-locate” is awkward, should this be “colocated” instead?

Line 83: Section 2 title is awkward - application to what?

Line 87: references “higher-quality” while majority of paper references higher precision

Line 94: What is the JJJ network? This is the only time it is mentioned

Line 102: Check capitalization for Bao

Lines 107 - 110: How does air move in the box? Is it passive? Later experiment mentioned a pump used to move air into the box but nothing mentioned here about that. Is the system designed to be used passively or will it have to be used with a pump when deployed in the field?

Lines 154 - 157: More details are needed about how the sensitivity correction is done - a reader can’t replicate a similar experiment based on what is provided here. Was this done with an environmental chamber or on lab bench sampling ambient conditions? How long was this comparison done for, and at what temperature steps?

Lines 193 - 196: More details are also needed about how the experiment works. Does the SENSE-IAP system have it's own pump? Is the outflow from the main pump being directed into the SENSE-IAP systems and the Picarro siphoning off them?

Line 194: What kind of dryer is being used?

Figure 4b: This figure is confusing - Which is the pump in the figure and where is it located? Also which is the 4-way valve?

Figure 6: This figure is good but I found the order confusing. Figure 6 is showing corrected results while 7 shows uncorrected results. It was difficult for me to follow the analysis with the shuffling back and forth.

Line 238: “with [a] Picarro”

Line 240: “effectively corrects the impacts of diurnal environmental changes” - need to back this up with either a figure or analysis from a previous section

Lines 267 - 273: It was not clear to me if the RMSE evaluation was done on the same period used to do the drift corrections or another independent dataset.

Line 296: “manufacturer”

Line 364: Capitalization of names

This is an interesting evaluation of the SenseAir K30 sensors, co-located with Picarros, over a 30 month timescale. This is a valuable study, but the manuscript needs significant revision to properly categorize sensor types and adjust all comparisons accordingly, address scalability concerns, strengthen the literature context, and better acknowledge limitations. It should be accepted for publication in AMT if the following can be addressed.

The authors fail to distinguish between “low-cost” and “mid-cost” CO2 sensors. The Vaisala CarboCap GMP 343 sensor mentioned in line 59 and used in networks like ZICOS-M (https://acp.copernicus.org/articles/25/2781/2025/) and BEACO2N is at a significantly higher price point compared to the SenseAir K30 used in this paper and the Carbocaps are now usually called a “mid-cost” sensor to distinguish them for the LCS like SENSE-IAP. Please adjust the introduction to distinguish between sensors in the $10s-$100s USD (low-cost), sensors in the mid-cost range ($1000s), and reference grade sensors (typically $10,000s).

Additionally, the accuracy and precision statistics given for other sensors should be for sensors at a similar price point to the sensors used in the paper. Vaisala CarboCap is not a comparable sensor (lines 64-68). Focus literature review on truly comparable low-cost NDIR sensors.

What is China’s dual carbon goal (line 72)? Please add context for the international reader.

Line 95: is “homology” the intended word? Homogeneity maybe?

What exactly is meant by background noise level (line 113)?

Figure 4 map labels are too small to be legible and is difficult to locate for readers unfamiliar with Beijing city. Also the latitude labels are cut off on the left side.

Line 189-190 is confusing and grammatically incorrect. Same with like 193-194.

Why can you say that hanging on an open window is the same a field-deployment? If the instruments are even partially indoors surely they are more temperature controlled than a true field deployment?

What is the recommended length of co-location with reference instrument for determining the correction coefficients? It would be helpful to include this in addition to the recommendation of a 3-month calibration interval.

Line 296: “manufacture” -> “manufacturer”

Line 324-325: missing spaces around ±

Please add additional discussion of the limitations of this study and how the findings may or may not translate to other LCS and environments. Are the recommendations made only for K30s?

Have the authors explored alternatives to a co-location for drift correction every 3-6 months? This may not be feasible for large deployments. Did the authors explore the performance of a remote calibration strategy at all? How would the proposed calibration scale (in cost/time) to, say, 100s of sensors deployed?

Please add additional literature review describing the existing methods for LCS calibration.

The data availability statement in my opinion does not follow the best practice for open research. Why have the authors not made their data and calibration codes readily available to all readers in an online repository?