Overall, this study presents an interesting and well-written technical paper, showcasing a low-cost alternative to conventional CO2 measuring devices. However, I miss a few critical details, that I think are needed to make it publishable.

First, I think a more detailed evaluation against LI-COR data is needed. Currently, there are few details and a more in-depth analysis of discrepancies, especially on systematic biases would be needed. Additionally, it would be good to see if there is temporal drift and to examine in more detail why there are certain times at which there is a relatively large mismatch.

Further, I missed a bit the discussion on how these methods could be applied in low-cost settings. E.g., how accurate are they, if no calibration data from a more expensive system is available? And how cheap are they really, if accounting for the time invested? I think it would be good to give an estimate of the time to assemble it and get it running from scratch. For people wanting to try this, these “time costs” may be very relevant.

Finally, I think it would be important to stress the limitations more clearly. Especially the geographic/climatic scope should be well described. For example, I could imagine that the device would work much worse in a more humid environment, if CO2 evolution after rainfall is a major issue (as you seem to indicate).

Detailed comments:

L46 it would be good to show some evaluation stats for you comparison to LI-COR.

L59 You might also mention the importance of Fs data for (agro)ecosystem and soil carbon models in calibration, validation, and development.

L145 Since you are talking about CO2 flux, it would be better to give soil organic carbon, not soil organic matter (usually SOM/1.72).

Section 2.5. I think you should do more than just a linear regression with LICOR data to check for the accuracy of your approach. For example, display of relative RMSE, analysis of whether there is systematic bias, and a slope of the regression different from zero (e.g., Gauch et. al, 2003, https://www.agronomy.org/publications/aj/abstracts/95/6/1442) are needed for proof of good performance. Also, it may be important to dissect where the systematic differences between sensors in Figure 3 b (raw data) stem from.

L229 I disagree that your displayed results validate the stability after 6 months. The only show how much correction was needed but not if there was a temporal trend in the correction needed (e.g., drift of one sensor from the other two). The latter would be interesting to analyze in detail. In the simplest term you could do this by plotting the sensor raw data over time. Maybe calculate correlation statistics for each month individually.

Figure 3: More information is needed in the caption so that this figure can stand on its own. E.g., over what time periods were reference and SCD30 measurements made. Additionally, it would be nice to code Fig 3 a by date of the measurements made.

Figure 4: also here, more information is needed so that the figure can stand on its own. E.g., abbreviations are not defined (Fs, Tsoil, SM). Panels c, d, and e should be amended with the exact dates that they refer to. Or are these averages? This is not clear for me.

L290 to 291 This part belongs to the Methods section.

Figure 5 a) is quite messy with all the lines, and hard to read. I suggest you move this into the supplement and show only the best-fitting method, here. Again, make sure to define all abbreviations so that that figure stands on its own.

L306 do you mean “correlated most strongly”?

Figure 6. Please also report additional evaluation statistics, as suggested above. And again, please define all abbreviations to make the figure stand on its own.

Section 3.3 is a bit short and I missed a clear recommendation where your system is to be used and where it could have limitations. For example, does it only work well in arid regions (your mismatch mainly occurring on rainy days could suggest this). Additionally, what methods could be used to improve the reliability of your system? What cheaper methods to double check your results could you recommend? Could your system even be applied in a manually built chamber to conduct the chamber method?

L345 this is the first time you mention maintenance requirements. It should be in the results/discussion if you want to have it in the conclusion.

L350 I think you should still mention the limitations and potential geographic/climatic suitable ranges (i.e., you only tested it in a very arid environment).

This manuscript presents the development, deployment, and evaluation of a low-cost, open-source soil CO₂ sensing system and its application in an arid environment to estimate CO₂ fluxes using the gradient method. In general, the manuscript is well written, and the study addresses a significant barrier in ecosystem carbon monitoring by proposing an accessible alternative to expensive flux chamber and eddy covariance methods. The open-source technical documentation and code for building the sensors is commendable, and the work has high potential for practical uptake and replication, especially in under-resourced regions.

However, the manuscript would benefit from a more detailed and critical interpretation of the flux estimation results, particularly regarding the limitations of the gradient method under dynamic soil moisture conditions. While the manuscript shows that the Buckingham model provided the best agreement at this site with chamber-based fluxes, the discussion does not fully explore the reasons behind its better performance. The authors note that discrepancies in other models may stem from their development under different soil conditions, citing previous literature, but they stop short of analyzing why Buckingham model's simpler formulation, based solely on air-filled porosity, aligns well with the arid, sandy-loam soil used in this study. A more detailed discussion comparing the assumptions and empirical bases of each model in relation to the site's soil texture, structure, and water content would strengthen the interpretation. Specifically, elaborating on why models like Campbell and Sadeghi underestimate fluxes at this site, likely due to their development in structured or clay-rich soils and their strong attenuation of diffusion when air-filled porosity is low.  Or why other models, like Millington or Marshall, tend to overestimate flux in coarse, dry soils by overemphasizing the role of air-filled porosity. These insights could assist researchers in choosing suitable diffusion coefficient models for various environments. This is particularly relevant in situations where users do not have access to flux chambers for validation. In such cases, selecting diffusion models based on soil texture and structure could improve results.

While Figure 5 presents a useful comparison of modeled and measured fluxes during dry conditions, the manuscript does not provide a similar analysis for wetter periods. This is a missed opportunity. Figure 6c shows that the gradient method performs poorly following rain, yet it remains unclear how different models would perform once a new diffusion gradient is re-established under persistently moist conditions. From the study it is unclear to me whether flux chamber measurements were collected during consistently wet conditions. Consider including or discussing modeled fluxes from different diffusion models during a wet period, analogous to the dry-period comparison in Figure 5. This would help assess how Buckingham’s relative performance changes under wetter conditions, and whether other models (e.g., Campbell) may become more appropriate. Consider reviewing Yan et al, 2021(https://www.mdpi.com/2071-1050/13/19/10874) evaluation of soil property effects on gas diffusion models.

Specific comments:

Lns 228-229.  Consider reminding readers which waterproofing design corresponds to each sensor number (Falcon vs thin coating).

Ln 266. Y-axis for 4c-4e incorrectly shows flux units instead of CO2 concentration.

Lns. 272-274. Was CM data also collected during a wet period? It would also be good to see diffusion model results when soils are wet.