General comments:

The manuscript presents a new instrument for the ground based remote sensing of atmospheric NO2. The sensitivity to the target species is provided by the principle of the gas correlation spectroscopy (GCS), but the main objective of this instrument is to become a quantitative imager of 2D fields of NO2. This NO2 imaging capability introduces this new instrumental concept in the very small family of imaging instruments capable of retrieving quantitative information on UV-VIS species (SO2 cameras, AOTF-based NO2 camera, and I-DOAS instruments to my knowledge). As such, this manuscript bears a strong original content, and can be considered as the seminal paper for a new class of NO2 imaging instruments.

The paper is well structured: it first explains the principles of the GCS, then presents a prototype and some results obtained in the laboratory, followed by the analysis of data acquired during an outdoor campaign at a coal-firing power plant. The quality of the text is very good: descriptions are exhaustive, the English is of very good level, and most figures are clear. The reader will not only find all the needed details to fully grasp the instrument principle, and what it can do, but he will also discover a honest account of real-life conditions performance, which is very much appreciated.

In the following, I'm listing a number of questions/remarks on scientific aspects. This review ends with the few typos/technical issues that I could spot.

Specific comments/questions:

My comments follow the structure of the paper:

- Introduction: The first part of the section lacks a few references about the harmfulness of NO2. Also, I believe that NO2 exposure limits recommended by the WHO have recently been lowered for NO2. Please check this and update if needed. Also, it might be interesting to list the previous remote sensing instruments which have employed the GCS method. I know about the satellite instrument HALOE, but others, even ground based might have existed.

- Section 2.1: In the beginning of the section it could be wise to simply remind that the model assumes that the pressure and temperature dependence of the cross-section is neglected. Also, I appreciate, in eq.(4) that the authors didn't overlook the temporal variability by applying an integration over time. But as no parameter is exhibiting any temporal dependence, I think it is ok to replace the integral by a product with the exposure time.

- Section 2.1, line 123 and below: You are rightfully stressing the delicate alignment of the two optical channels. Actually, as your processing algorithm contains a normalization of one image by the other, the consequence is that the derived quantity gets more sensitive to parameters which might not be as identical as you expect. For instance, the pixel response non-uniformity (PRNU) map of the two sensors could be quite different, and therefore not get properly cancelled by the normalization. Therefore I think it is good to acknowledge the potential differences across the two channels by adding a subscript in eq.(7)-(8) and further. In particular, I would add the subscript "c" to \eta and \mu in eq.(8). With this notation, the reader will be in a better position to appreciate the attempt of canceling the optical effects with the normalization performed in eq.(12).

- Section 2.2, line 145: The assumed range of NO2 SCDs should be justified.

- Section 2.2, line 161: From my own experience, relying on manufacturer data is risky. For quantitative measurements as the ones you intend to do, I would recommend to measure the detector parameters, and the optics transmittance by yourself, with the needed spectral resolution (which is very often very coarse in manufacturer's datasheet).

- Section 2.3: This section seems to be an attempt to get rid of the spectral integrals inherent to the method. In my opinion, the price to pay to achieve an analytical, integral-free expression for the effective optical thickness is too high. The reader can also be disturbed as the objective of this attempt is not fully clear. I would recommend to remove this section, in favor of stressing again that the measurements are not spectrally-resolved, and therefore a direct access to the target air mass SCD is not possible.

- Section 3, line 258: I was a bit surprised by the reported dark signals. It is quite strange to report values at temperatures well above the normal working conditions (e.g. 20°C). Also, the dark signal precision (24+/-9) is quite poor, much worse than the shot noise. This questions the accuracy of the dark current removal. But as the exposure times are orders of magnitude smaller than the second, I guess it is not really an issue. However, the text does not address the characterization of the ADC gain, which might be pixel-dependent, and detector-dependent. Perhaps providing the steps that you apply during the "L1" processing for the data would help the reader to understand exactly what is done.

- Section 4.1, eq.(25): This equation is not correct, as the ratio of the J-terms, followed by the log does not remove the spectral integrals contained in each J-term. At best, you obtain a biased estimate of the column. The importance of this bias can be assessed with a numerical experiment. Negligible or not, this aspect should be addressed.

- Section 4.2.1: The vertical gradient that you observe like in figure 12 (a) is intriguing. You argue that it could come from the inhomogeneity of the reference measurements, but this does not really hold as the term J_ref(i,j)/J_c,ref(i,j) largely cancels broad variations of the sky signal across the FOV. While you have assessed the false signal induced by H2O, and O4, you haven't addressed the impact of not modelling the Rayleigh and aerosol extinctions. As the spectral integration is not removed in eq.(12), it is not excluded that the extinction by the Rayleigh scattering, given its wavelength-dependence (15% decrease over the filter bandwidth) is not completely canceled in the normalization. Given that the air mass increases as the pointing elevation decreases, this could perhaps explain the vertical gradient you are observing...

- Section 4.2.1: You are reporting a column of 2.72x10^18 in the cell, whereas it was explained that the best performance is achieved with a cell of 4x10^18. I also had understood that this was the value used in the prototype. Why such a deviation with respect to your ideal case? Is it caused by using eq.(25)? Why didn't you estimate this value with the DOAS instrument?

- Section 4.2.1, figure 14: It is very difficult for the reader to clearly distinguish the plume from the background. Please consider a color scale which would help the reader to observe the abundance of NO2 in the plume.

- Section 4.2.2, 2nd paragraph: This paragraph spans 41 lines, and contains a very technical discussion. Please consider splitting in more paragraphs, or even better, divide in subsections. A subsection dedicated to your analysis of the influence of the choice of reference area would make sense for instance.

- Section 4.2.5: The comparison with the MAX-DOAS measurements delivers results which are a bit deceiving, but it also seems like you attempted to compare data of very different temporal resolution. Can't you revisit your comparison in order to make sure that you have the highest temporal consistency? In other words, one would expect to have used a different set of camera images for comparison with each MAX-DOAS point. Working with a global average doesn't put you in the best conditions... In addition, you report an average plume signal of 1x10^16, which is below your detection limit. This is striking for the reader, and would deserve a little explanation.

Technical comments:

- General comment about significant decimal: Everywhere in the text, values are reported with too many decimals compared to the uncertainty of the measurements. This starts already in the abstract, but affects the manuscript globally. This particularly concerns the columns in molec./cm^2, and also the mass fluxes.

- Figures 5, 6, 7, 20: those figures have issues with labels or annotations.

- L4: gasses -> gases

- L17: "momentary" ?

- L19: "...is highly portable and cost efficient at building..." -> "...is highly portable for building..."

- L28: add a comma after "time"

- Figure 1: Are you sure that (a) is not the filled cell, and (b) the empty one? This is what I would have assumed based on the brownish color of cell (a)...

- L64: "Lamber-Beer" -> "Lambert-Beer"

- I found the unit "phe" quite uncommon. Why not using "e-" for representing photo-electrons?

- L118: I would not speak of "spectral channels" when referring to the two optical channels of the NO2 camera.

- L119: "... functions as a measure ..." -> I think it is more correct to say that it is a quantity exhibiting a monotonic sensitivity to S.

- L143: add a comma after "For this"

- L147: In don't think that t_exp has been defined before

- L151: double "of"

- L365: remove the comma after "showed"

- L384: "two cameras inside of the instrument" -> "two cameras inside the instrument"

- L386: "These displacements manifest as strong..." -> "These displacements manifest themselves as strong..."

- L392: you introduce the subscript "i" for your columns, whereas I guess that everywhere before, "i" was a row index (like in matrix algebra convention)...

- L619: this value of 1/2 FPS is new. I had 1/12 in mind from the paper. Please check

- L623: Split the paragraph after "expected."

- L630: Split the paragraph after "18.2%."

General evaluation

This manuscript introduces a novel measurement approach for the fast imaging of strong NO2 emission sources, such as stack plumes observed in power plants. The proposed NO2 camera is an application of the gas correlation spectroscopy, that was successfully used in the past e.g. for CO measurements in the infrared spectral range. The method is here extended to NO2 measurements in the visible spectral range. The study is largely exploratory in scope and concentrates on (1) a theoretical analysis of the measurement principle including estimates of the expected performances, (2) the verification of model predictions using a proof-of-concept instrument demonstrating the feasibility of the technique, (3) results from first measurements in the field at a large German power plant, and (4) comparisons with simultaneously recorded MAX-DOAS measurements. The proposed approach is very attractive since it is simple in concept and potentially inexpensive, which opens possibilities for future deployment at larger scale. In its current state, however, the system remains very experimental and a number of technical difficulties are still to be solved before such a camera can be ready for routine measurements in the field. Nevertheless I found the manuscript very interesting. The simple theoretical model is convincing and addresses several aspects of the measurements performances, such as sensitivity, selectivity, detection limit, etc. Model estimates are found to be in good agreement with actual measurements performed using the proof of concept instrument, which validates the approach. Also first measurements in the field show convincing results, for a stack plume of moderate strength. It also illustrates the main technical limitations of the current instrumental design. It also provides an interesting discussion on emission flux estimates performed using the camera, which arguably represents a promising application for future developments/applications. The last chapter on the comparison with MAX-DOAS measurements is however disappointing and somehow confusing. The authors struggle in a lengthy discussion to explain the potential reasons for a lack of agreement between both techniques, which result from a suboptimal operation of the DOAS system, a lack of time synchronisation and also calibration issues. In its current state, this comparison does not bring much to the study. I therefore strongly recommend to remove it and concentrate on an optimization of the experiment for a future publication. This reservation being made, I found the manuscript innovative, well written and definitely suitable for publication in AMT.

Detailed comments

Pg. 1, abstract: the first two sentences of the abstract could be omitted from the abstract. Such general information is generally provided in the introduction of the paper.

Pg. 2, l. 54: I find the formulation “immanent asynchrony of the push-broom scheme” a little bit obscure in the present context. I suppose you mean that because of the need to scan in one spatial dimension, the information is recorded sequentially, which can lead to image deformation effects. Please confirm or clarify.

Pg. 3, l. 74: the instrument concept requires that the NO2 contained in one of the cells remains stable during measurements. Doesn’t this requirement imply that the cell temperature must be stabilized to maintain the NO2/N2O4 ratio at a constant value? This question should maybe be addressed in the section dealing with instrument model calculations and uncertainties of the method.

Pg. 4, l. 93-96: are the units of radiances and irradiances important for this particular application? Certainly not for the measurement itself which is based on intensity ratios. But maybe this information is needed for the instrument model calculations. Please clarify whether absolute radiance values are used in this study.

Pg. 4, l. 100: I suppose that the wavelength dependence of the quantum efficiency indicated here is a property of the silicium-based detectors used for the measurements, which explains the limited spectral range (UV-Vis-NIR). Note that the use of the sun as a light source also limits the applicable spectral range.

Pg. 13, l. 269: the fact that the adjustment of the alignment of the two cameras is scene-dependent represents a major limitation for operation in the field. Can you further develop the reason why this is the case? I understood from the last sentence of the conclusions that the use of another instrumental design could solve this issue. It would be nice to introduce this possibility with a bit more details in the main part of the manuscript.

Pg. 15, Table 1: Is there any particular reason why the uncertainty on the DOAS measurements of cell 2 so much larger than for other cells?

Pg. 15, Fig. 9: this figure would gain being enlarged a little bit. Especially panel (a) is difficult to interpret.

Pg. 18, Fig. 12: again panels (a)(b) and (c) in this figure are very small and difficult to read. I suggest separating them from the two other panels and creating two separate figures. Since this figure shows the first illustration of an actual plume measurement with the camera, it deserves to be displayed in a more prominent way.

Pg. 19, l. 367: again these results demonstrate that the stability of the NO2 concentration in the reference cell is important, which suggests that an active stabilization of the temperature of the cell is needed (to constrain the NO2/N2O4) ratio.

Pg. 22, Fig. 22 and related discussion in pg. 23: the need to manually define the mask used to estimate the background ‘out-of-plume’ signal is also an important limiting factor for the technique. Do you see a possibility to overcome this difficulty either through an instrumental modification or by means of a more elaborated processing technique? If yes, it would be interesting to further discuss this question, maybe in a short section dedicated to perspectives for improvement of the technique.

Pg. 25, l. 465: at the end of the sentence, refer to section 4.2.4 where the question of the NO2/NOx ratio is explicitly analysed.

Pg. 28, section 4.2.5: as already pointed out in my general comments, I strongly recommend to remove this section from the paper. My feeling is that it brings confusion and does not help consolidating the measurements obtained with the camera. I would suggest replacing it by a small section outlining the possible improvements that can be envisaged for the instrument and eventually the data evaluation.

Pg. 34, l. 1: in fact, if I understand correctly, the camera was operated at a reduced resolution of 1300 x 600 pixels (accounting for the windowing applied to reduce the read-out time).

Pg. 34, 623: ‘… a detection limit of about 2e16  molec/cm2 is expected…’. Here I would add that this was confirmed by measurements using the proof-of-concept instrument.

Spelling, typos:

Pg. 2, l. 44: remove ‘either’

Pg. 6, l. 128: change ‘In reality this latter condition need not be perfectly filled’ by ‘In reality this latter condition does not need to be perfectly filled’

Pg. 7, l. 151: ‘of’ is duplicated between ‘choice’ and ‘particular’

Pg. 10, Fig. 5: there seems to be a confusion of the ‘S’ and ‘Sc’ notations in this figure. To my understanding, the x-axis of panel (a) should be labelled as ‘Sc’ as well as the legend of panel (b). Please check and adjust as needed.

Pg. 10, Fig. 6: the species of which cross-sections are disp^layed do not show up properly in the legend where the applied scaling factors are given.

Pg. 13, l. 258: I suppose that the sensor temperature is fixed at -50°C, and not +50°C as indicated here.

Pg. 22, Fig. 16, end of first line: replace ‘The left two’ by ‘the two left’