Overall this is very good work and consistent with previous findings cited here, that lower cost trace gas sensors have utility if properly calibrated and corrected, but each sensor must be independently corrected as they can vary in their usefulness. I think my only scientific comment (and I do not think it is required to be addressed for publication) is that PBL height was mentioned in passing in Section 3.7.2 e.g. "generally shallower boundary layers". If it's not too much trouble, and if the data exists, it may be good to show some PBL height observations from an urban and rural location nearby to strengthen this argument. It could be purely vegetation based in terms of the diurnal variation, but could it also be due to urban/rural PBL variations too? Just something to consider, overall very nice work!

This paper reports the initial results of establishing a CO₂ observation network using the low-cost NDIR-based Vaisala GMP343 sensors at 17 sites across the city of Munich.

The study focuses on two main aspects: (1) how the accuracy of the sensors used in the network was improved, and (2) how the system was applied to capture high-resolution spatial and temporal CO₂ variability within the city.

For the first aspect—sensor accuracy—the study compares the sensitivity of the Vaisala GMP343 sensors to three environmental variables (humidity, pressure, and temperature) against a Picarro reference instrument. Among these variables, temperature had the greatest impact on the NDIR sensors. In the second-generation network, an additional temperature stabilization enclosure was introduced to address this issue. As a result, the RMSE decreased from a maximum of 2.6 ppm in the first-generation system to less than 1 ppm in the second generation, achieving the target accuracy. In summary, the study aimed to enhance NDIR sensor accuracy primarily by controlling the temperature factor.

For the second aspect, the monitoring sites were categorized into three zones—urban, suburban, and rural—to examine spatial variability in CO₂ concentrations. At one specific site (MAIR), a Hampel filter was applied to remove the influence of nearby ventilation outlets. Although the filter effectively removed some peaks, it was not entirely successful in eliminating all local pollution signals. The study found that the classified zones showed clear diurnal and seasonal differences: during summer, rural and suburban sites exhibited greater diurnal variability than urban sites due to photosynthetic activity. This pattern persisted in winter, though the diurnal amplitude was considerably smaller.

Overall, the study is well conducted, but several areas require revision or clarification before publication. Please refer to the comments below:

(Page 6, Line 123)

The paper states that the intake line was extended up to 50 m, with a flow rate of about 0.5 LPM. Is this flow rate sufficient for such a long sampling line? Please provide a proper justification or reference.

(Page 9, Line 215)

Calibration was performed only at two points—400 ppm and 520 ppm—for slope/intercept correction. Can linearity across a wide and long-term concentration range (350–600 ppm) be ensured with only two calibration points? Since actual CO₂ levels in different urban zones may fall outside this range, would additional multi-point calibration or slope tracking be necessary?

(Page 9, Line 203)

The use of the Wagner equation to calculate water vapor saturation pressure for deriving dry mole fractions seems appropriate. However, since the water vapor data came from an external instrument, that instrument itself likely has some uncertainty. Would this not affect the accuracy of the dry CO₂ mole fraction? Please discuss this potential limitation.

(Page 10, Line 230)

Using a long analysis window may risk classifying short-term traffic plume signals as “outliers.” However, such short-term and abrupt fluctuations are key features of urban CO₂ dynamics. Applying too long a window could remove meaningful short-term events as noise. Please provide additional justification or discussion on this issue.

(Page 21, Figure 10, Lines 416–427)

To control excessive local pollution, the study applied the Hampel filter used in previous studies. While this method effectively removes extremely high peaks, it does not fully eliminate local contamination. The paper notes that the filter captured the ventilation effects but did not perform particularly well. Moreover, since this station is used as a background site, placing the sensor so close to a ventilation outlet seems questionable. Please provide further explanation or justification for this site configuration.

Review of the manuscript egusphere-2025-4157

ACROPOLIS: Munich Urban CO2 Sensor Network

By Aigner et al

Summary. This manuscript presents the setup and testing of a network of 20 mid-cost CO2 concentration sensors with the aim to set up an urban network of CO2 concentration observations to identify urban gradients in CO2. The manuscript fits the scope of the journal. The study is impressive and one of the first that illustrates the details in setting up, maintaining and interpreting such urban CO2 network. So I recommend to accept this manuscript after some revisions.

Recommendation: Minor revisions are needed

Major remarks

1. The paper misses in my opinion the opportunity to explain why such a CO2 concentration network is needed. Most studies deal with CO2 fluxes, rather than concentrations in order to study the carbon budget of sources and sinks. Here you add a network of CO2 concentration observations, but the study does not say much about how this complements or support the CO2 flux budget estimations, or how it can help to study CO2 advection estimates (since you generate spatial gradients that are unique).
2. I think the manuscript can do more to justify the sensor network is more or less free from local influences. I fully agree with the strategy to find measurement locations like schools and hospitals and independent buildings to limit local influences, but at the same time the paper does not say/claim/justify one succeeded in doing so (which is not an easy task, I understand). This could be made more clear.

Minor remarks

Ln 209: What about extreme high RH? Does the system still work at RH between 95 and 100% or in fog and rainy conditions? These are usually troublesome. Have data been removed, and if so, how many?

Appendix A: please swap the x axis and the y axis, since the Picarro is your reference, and should be thus on the x axis (independent variable) and the system is your test case (dependent variable).

Figure 6: in the x axis MAE and RMSE need a unit. The caption should be elaborated since it is not clear what is the meaning of a dot (i.e. is one dot one sensor?), and it is not clear how the RMSE and MAE are calculated. I.e. is it the RMSE over the all hourly values, daily values, daytime values… Some more guidance in the caption is welcome.

I would like to see some more justification for the chosen error metric (e.g. based on https://gmd.copernicus.org/articles/15/5481/2022/ and underlying works). RMSE and MAE are likely common practise, but RMSE is not an unbiased error estimator (https://www.sciencedirect.com/science/article/abs/pii/S0020025521011567 ), which may mean your results are better than you present now in the manuscript.

Table 2, header: Mean bias, MAE and RMSE should have a unit

Table 2: Reword Mean bias to bias, since bias is by definition an mean (as long as you do not average over all systems – which you do not do here).

Figure 7: caption: reword “sensor temperature” to “hourly mean sensor temperature”.

Figure 7: revise y axis label. The graph now suggests the temperature measurement is accurate at 0.01 K, which is not the case. In your wording in ln 371, you also use only 1 decimal.

Ln 377: significant improvement. Please add the results to a statistical test that confirms this statement.

Figure 10, caption: Please add the height at which the wind speed was measured.

Figure 11, caption: reword to “Time series of observed CO2 concentrations…. “

Figure 12: profiles-> evolution. Profile is more reserved for vertical profiles.

Ln 467-469: can be removed, since a) it is 3 short-sentences paragraph (too short), but mostly it reads as a figure caption, so should belong to the caption of fig 12 and not in the main text.

Ln 471: This reflects enhanced photosynthetic uptake and higher boundary-layer heights. This statement is not confirmed with additional measurements. Are these available? I would say the concentrations are first of all lower because lower emissions in spring and summer than in winter. So a car traffic count or emission databases could support these.

Fig 13, caption: The results reveal distinct seasonal and spatial patterns in diurnal CO2 variability. This sentence should be removed, it is interpretation of the figure and thus needs to be in the main text.

Fig 13, caption: “The diurnal variation is defined as the daily maximum minus minimum CO2 concentration.” Do you mean “The diurnal variation is defined as the daily maximum minus minimum hourly CO2 concentration.” ?

Fig 13: y axis: the label says: Mean hourly diurnal cycle of CO2 variation. This is of course impossible (measuring an hourly diurnal cycle). I suggest to change to “Mean diurnal cycle of CO2 concentration (ppm) based on hourly mean observations”

Figure 13: The figure’s content is hyper-interesting and intriguing (and nicely plotted). But I was wondering whether an uncertainty estimate can be added to each (or a representative) label. E.g. for the urban station on the most left in the graph, the max diff between summer and winter is order 20 ppm. But if the error bar is 30 ppm (which I do not expect),then the differences between seasons are virtual. So if the error estimates are small, better to add them to show you have measured significantly different CO2 diurnal cycles between seasons. In fact you do in Fig 14!

Section 3.7.3: More justification is needed for the definition of the afternoon hours (12:00 - 18:00 local time). I do agree with your strategy to ignore nocturnal accumulation. However, in Munchen in mid winter the sunset is at 16:22 CET, which means you will have about 2 h of stratified atmosphere in your sample. Please explain why 10:00-16:00 local time, was not chosen as study period. Or whether that would have given other conclusions.