The present work presents the results obtained using an original open-source low-cost sensor (LCS) system developed to measure tropospheric O3 in a remote high altitude alpine site. Intra and inter-comparison between sensors and reference (Thermo 49c) had been conducted for seven months. The authors analyzed the effects of environmental factors on sensors' performances. The work is novel but similar works have been reported in recent a decade. It may be the first work in an alpine site, but it isn't the first work in a remote environment. In general the data analysis discussing for one sensor is enough for this work such as in figure 4, 5 and 6. The manuscript is a little wordy and is large in page scale. All tables are too simple to be understood by the reader and more detailed information have to be added. The RH and temperature are the important factors to sensors performances so the authors can be improve the conditions to sensor. More discussion can be focused on the point.

There are specific comments as the followings.

In figure 1 it is better to show the sampling site in a larger scale of map in bottom panel.

The section of 4.2 can be deleted from the paper since it is unnecessary information.

The section of 3.1 Laboratory calibration of LCSs can be more concise since the section is unimportant in the work. The authors can be pay more attention on the field employment.

In section 4.2 line 345, this sentence is  complicated to reader and can be simpler.

This work depicts the results of a harsh environment long term colocation experiment focused on performance evaluation of Ozone LCS versus a high precision and accuracy regulatory grade instrumentation. Surprisingly accurate results are obtained with a very simple open source architecture based on redundant EC sensors targeted to O3 using lab calibration. Field calibration allowed for further amelioration. Their experimental design allowed them also to

a) detect drift during two consecutive lab calibration in May and July 2018.

b) evaluate the possibility of calibration transfer correcting the response from one sensor with coefficient derived from other sensors calibration.

c) finally, reliability of the LCS sensor outperformed the regulatory grade one, which failed several times due to typical operational and technical issues.

 I enjoyed the paper, it is well organized and the technical content is certainly relevant with validation encompassing several seasons.

However, it could be important to highlight that the selected sensor are actually  sensitive  to NO2 (also) with a sensitivity that is very similar to the one expressed towards the intended target, O3. Actually Alphasense suggests to couple this sensor with an NO2 targeted one and to subtract the NO2 estimate from the NO2+O3 estimation provided by the original sensor. From author's results, influences from low NO2 concentrations at high altitude is clearly found to be negligible.

In the paper, the description of the auxiliary electrode working principle is  imprecise. Actually, auxiliary electrodes in Aphasense sensors are just architectural twins of working electrodes which are screened from the atmosphere. As a result they react only to temperature providing a way to correct from temperature dependance subtracting the response elicited by temperature itself on both electrodes. Unfortunately geometry and fabrication variance does not allow for building an exact replica of the working electrode so correction is not exactly accurate and field calibration is still necessary.

It could be interesting to the discuss the cost and the long term accuracy of the field calibration which in the proposed version needs to use two months in different extreme seasons to obtain the presented results. Recent papers have analyzed the robustness to changing environment of field calibration, which is to say that they lose accuracy when concept drift arise.  Is the relative stability of the alpine environment beneficial for field calibration robustness?

Moreover calibration transfer “absolute” results are very interesting but they could be slightly overoptimistic from the numerical point of view since, in my understanding, authors are using the entire dataset to derive the calibration coefficients and then they validate the performance of the other so calibrated LCS in the very same conditions in which the calibration have been derived. If I am correct I would like this to be discussed in the paper.