Lan et al., utilized five months of EC measurements across three sites in European cities and conducted comprehensive evaluations of friction velocity, sensible heat, latent heat, and CO2 flux using three different software packages. They found that the major difference among the three software packages lies in CO2 flux, primarily due to the spectral correction schemes.

The paper is well-organized and includes technical details that fit well into the scope of AMT. I recommend minor corrections before it can be accepted in AMT.

My major comment pertains to the conclusion of this paper. The author suggests that different spectral loss correction methods contribute to varying CO2 fluxes in the calculation. However, the paper does not specify which spectral loss correction method is more accurate. Does the author have any recommendations on the preferred spectral loss correction method and software to use? Can the author propose a more standard way of computing CO2 flux to reduce uncertainties in CO2 flux calculations?

Figure 10 illustrates a significant bias in fluxes between eddy4R and TK2, which is attributed to the unstable stratification during the daytime. Could the author possibly incorporate the PBLH or other dataset to further support this statement?

Even though the context suggests that the largest deviation is found in vatical velocity, it is hard to discern from Figures 5,6,8. Would it be better to show the distribution of differences instead of scatter plots?

EC method is the state-of-art direct measurement method for heat, vapor and CO2 fluxes. This paper conducted a comparison of three widely used software using 5-month ICOS-Cities tower observations and found good agreement on the mean values and standard deviations. The results also showed that the fluxes were of a good but not yet perfect agreement among three software packages due to the low-frequency spectral treatment. The topic is interesting and important. 

My comments are as follows:

1.    It is important to set the EC instruments in the inertial sublayer for city-scale, it is better to add the normalized height (with the average building height) in Table1 for the tall towers used in this paper. I am also wondering if the observation height will impact the results because the largest scale of the turbulence captured by the EC maybe different;

2.    10% difference in fluxes estimation may lead to quite great bias for long-term observations. May the authors give any suggestion on the selection of the postprocess software, or which is the best one for long-term measurements? This may be difficult when using the field observation,but can be done with some artificial perfect ‘ideal’ data.