Summary

In their work, Park et al. motivate and present a new method to derive NO2 VCDs from low cost sensors that can be used on airplanes without regular maintenance. The Modified Wavelength Pair (MWP) method uses measured ratios at two wavelengths in combination with radiative transfer simulations to estimate the NO2 VCD. While the method has some limitations in terms of precision and accuracy, the approach with low cost and low maintenance is a good addition to ground based or satellite borne measurements.

The paper is generally well written, the scientific approach is motivated and well described. The presentation of the results and satellite comparisons leave some room for improvement. Therefore, some revisions and technical corrections are listed below.

General/Major revisions

As the Hyperspectral Image Sensor (HIS) is used specifically for the new Modified Wavelength Pair approach, it should be compared to other low-cost sensors. For example: How does the method compare to NO2 camera (Dekemper et al., 2016) in precision, detection limit and versatility?

With a wavelength resolution of 1.4nm (FWHM) it should be possible to perform a DOAS analysis. Comparing a classical DOAS approach to the newly developed MWP method would help to understand advantages and applications of both approaches. This should at least be included in the supplement.

Application/comparison: While the comparison to TROPOMI looks really nice (especially in the easier-to-read figures S9 and S10), it would be really interesting to see a comparison to ground based DOAS instruments – if any are available. TROPOMI is known two underestimate the NO2 VCD in heavily polluted regions and thus a comparison to ground-based instruments (maybe during GMAP/SIJAQ) would make it easier to evaluate the results of the MWP approach.

Figures 3 to 8 are generally difficult to understand. A detailed list of suggested/required improvements for each Figure will be attached at the end of this review.

The used TROPOMI data product needs some description as to how gridding for the comparison to the MWP data results was done. Details like cloud filtering need to be considered as they could strongly influence the NO2 VCD results (e.g. Boersma et al., 2004, Eskes et al. 2020).

Minor revisions

Line 90: “is known to be non-linear when the CCD counts exceed approximately 80% of the saturation level” – do you have a source for this claim?

Eqn. (3): Maybe switch the last term with the second to last term to get a fully logical chain of equations:

VCD_Obs = R_Obs […] = R_RTM […]

Eqn. (4): What is VCD_rtm? Should it be VCD_obs?

Line 231: In the MWP method: What makes A and B so special – wouldn’t the whole equation also work for any wavelength pair independent of type (ascending or descending). Considerung Table 2 and Figure 2 – would the wavelength pairs A2 and A3 not work similarly good as A2 and B2?

Line 267: The authors describe the use of NCEP data set as the initial conditions for the WRF model. Since ERA5 data set was used as input for the RTM simulations, wouldn’t it be more consistent to use the same data for the WRF model.

Line 273: How strong would variations in the assumed stratospheric NO2 affect the retrieved NO2 VCD?

Table 4: The uncertainties introduced into the RTM calculations by the assumptions given in 3.2 (ERA5 pressure, temperature, mixing ratios and CTM data for NO2 vertical profiles) seem to be missing in the error estimation description. A sensitivity study of the quantities used in the RTM calculations would be helpful.

Technical revisions

Line 90: “80 % of the staturation level, and the CCD” -> “80 % of the saturation level, and thus the CCD”

Line 186 “the wavelength pair” -> “each wavelength pair”

Line 209: “Then the Eq. (2) can be reformulated by replacing […]” – I think it should be Eq. 1.

Suggested/required improvements for the Figures

All figures

• Tooltips are in Korean

Figure 1

• The figure description should mention what red/gray/blue boxes stand for – I guess its red: input data, gray: retrieval step, blue: result?!
• I don’t think that there should be an arrow from “NO2 VCD retrieval” to “HIS raw spectra”. I guess the NO2 VCD retrieval is the whole thing depicted in this figure? Suggestion: Remove the arrow.
• Figure and description make clear that there is a strong dependency on input variables. How independent is the method and what is the uncertainty caused by errors in the assumption? (see also comment/minor revision on Table 4)

Figure 3 – Summarizing the figure in the text when referred to would be quite beneficial in understanding the content a lot faster – also some reasoning why some dependencies are found would help:

• None of the investigated uncertainties show a dependency on PBLH, altitude, solar zenith angle, viewing zenith angle or relative azimuth angle.
• The uncertainty contribution of the Albedo increases for small albedos (< 0.15) – is this caused by an absolute uncertainty of the albedo which becomes relatively large for small values of the albedo?
• The uncertainty contribution of the Albedo and the SF uncertainty slightly increase with increasing NO2 VCDs. Is this connected to the probability of the SF is contaminated by NO2 increases with NO2 VCDs? Why does the albedo uncertainty depend on the NO2 VCD?
• The figure could generally use a more consistent description mentioning all used abbreviations. The authors should provide more information on “wvl” and “Instrument noise”.

Figure 4 The figure is generally quite hard to read/understand. Here are some suggestions:

• When using text within the figures only use “white” as a text color.
• There is no description of red boxes or blue circles – maybe unify marked regions with one color (e.g. white)
• The legend in the upper left panel is valid for all panels – it should be moved out of that panel, so this becomes clear.
• The individual panels should be named. Right now, the upper left figure is the “label legend” for all 4 panels – maybe it is enough to show a bigger version of the upper left figure?
• Panels (a), (b) and (c) look peculiar. What gridding routine was used for the satellite data and which filters (e.g. clouds) were applied?
• It is not fully clear which data were averaged to obtain the displayed NO2 distribution.

Figures 5, 6, 7

• The results of the airborne measurements are hard to read.
• Generally, the same problems as in Figure 4 apply.
• The figures need to be larger/better resolved to be readable.
• Are the shown days cloud-free or are cloud data just not flagged? Interpretation of NO2 columns retrieved from satellite instruments are difficult to interpret in cloud contaminated scenes, it should be made clear whether the measurement days were cloud free or a cloud flag needs to be applied.

Figure 8

• If the TROPOMI data are not cloud filtered, this should also be done here.

References

Boersma, K.F.; Eskes, H.J.; Brinksma, E.J.: Error analysis for tropospheric NO₂ retrieval from space. J. Geophys. Res. Atmos. 2004, 109. https://doi.org/10.1029/2003JD003962

Dekemper, E., Vanhamel, J., Van Opstal, B., and Fussen, D.: The AOTF-based NO₂ camera, Atmos. Meas. Tech., 9, 6025–6034, https://doi.org/10.5194/amt-9-6025-2016, 2016.

Eskes, H., J. van Geffen, K. F. Boersma, K.-U. Eichmann, A. Apituley, M. Pedergnana, M. Sneep, J. P. Veefkind, and D. Loyola. 2020. “Sentinel-5 precursor/TROPOMI Level 2 Product User Manual Nitrogen Dioxide.” V.4.0.0 (S5P-KNMI-L2-0021-MA).

General Comments

This paper describes a simplified retrieval for NO2 from solar backscatter measurements, based on wavelength pair ratios (on/off absorption spectral lines)—the Modified Wavelength Pair (MWP) method—designed for use with low-cost hyperspectral sensors that lack the measurement stability of satellite and more-expensive airborne instruments. This technique is applied to the Hyperspectral Imagining Sensor (HIS), which was flown on aircraft over three significant pollution sources in Korea. An analytical uncertainty analysis is included. Results are compared with TROPOMI retrievals, and sub-satellite-grid-scale differences are discussed, in the context of geophysical variations.

The manuscript is well written, thorough, and generally clear. I recommend minor revisions.

Specific Comments

Paragraph starting in L48: Mention TEMPO and other, geostationary spacecraft, which are also achieving fairly good spatial resolution information.

Table 1: Please include some information about the f-number, etendue, and/or SNR of the system (under a given set of circumstances), that would indicate the optical throughput of the system.

Section 3.1: Please reference the use of wavelength pairs in other retrievals. For example, wavelength pairs were long used in the retrieval of total ozone.

Section 3.1: By combining pairs of wavelengths, one with the shorter wavelength having the stronger absorption (“Type A”) and the other with longer wavelength having the stronger absorption (“Type B”), you in effect partially cancel the bias from spectrally changing surface reflectivity (since the forward model does not include reflectivity spectral dependence). I suggest you say that explicitly. It would make the mathematical discussion easier to understand.

Throughout the entire manuscript, I recommend using r (small r) for the correlation coefficient, to reduce confusion with the radiance ratio R (capital R).

Paragraphs starting in L182: This section is confusing. In L183, what does “k-fold of the value” mean? The definition of k spectral dependency factor—is not introduced until L195. R is then discussed—I think meaning the radiance ratio—but since reflectivity is also being discussed, it is tempting to think R means reflectivity. In L184, “The same k value can be assumed for the wavelength pairs...”—why can that be assumed? In L193, “relation between the biased VCD estimates”—biased in what way? Why are they biased? Do you mean they include measurement errors?

L202, Eq (4): Equating VCD_True with VCD_rtm,A and VCD_rtm,B is confusing. If the modeled VCDs were “true,” observations wouldn’t be needed. Maybe I don’t understand what “True” means in this context?

L232: “three independent NO2 VCD”—I understand what you mean, but they aren’t really “independent.” “Different” may be more accurate.

L234: “to increase the signal-to-noise ratio” and L236: “spectral binning of ±2 original spectral bins”—It would be good to capture this in Table 1, so the reader doesn’t look at the table and this and think you have 5x spectral oversampling. The table would be more useful if it reflected the sampling/binning and SNR that are used in the retrieval.

L254: “highly resolved CTM data”: Highly resolved in what way? Horizontal spatial? Vertically? Also, I would not say models produce “data”—they produce “output.” Measurements are data.

Paragraph staring in L366: Q is not clearly defined. What does “transpose the ratio between R_rtm,A and R_rtm,B as Q” (L369) mean? No Q ppears in Eq. (11).

L472: “mean value”—How well did the three different VCDs agree? It would be useful to know how similar they are, and if they’re different, why.

L484: Footprint is 400 m x 400 m, so this is essentially the entire swath width, correct? It would be helpful to state that.

Figures 4/5/6/7: How were the TROPOMI data downscaled?

L575: “Before comparing the collected set of collocated HIS and TROPOMI NO2 VCDs, bias offsets were incorporated into the HIS NO2 VCDs”—How well do HIS and TROPOMI agree in general, before bias are removed? I am interested in inherent retrieval bias as well as the representativeness of the TROPOMI footprint. In the Abstract and Summary, it is stated that 0.106 DU is the absolute error of the measurement, in comparison to TROPOMI. Is that true? It seems like some bias has already been removed.

Technical Corrections

The English usage is generally good. I include several suggestions that I noted when reviewing the paper, along with technical comments, below.

Line 12:

“The high spatial”—remove “The”

“(VCDs) were measured”—replace “measured” with “retrieved” (VCDs are not measured, strictly speaking)

“from the airborne”—remove “the”

L24: “The typical”—remove “The”

L29: “different observation geometries under complex vertical wind fields”—The winds don’t change the geometry, per se. Better to say something like “different pollution distributions...”

L37: “pollutant”—change to “pollutants”

L65:

“calibrations”—change to “calibration”

“retain”—change to “maintain”

L75: “East Asia, where the”—remove “the”

L78: “calibrations”—change to “calibration”

L91: “latest (spectral rows)”—change to “spectral rows at the edges”

L92: “grating is used with a concave mirror”—Is it a concave grating, or is there a mirror in addition? Please clarify. Also, no where do you say if HIS uses reflective or transmissive optics. I am assuming reflective.

L100: “Unlike the”—change to “Despite these”

L122: “DOAS fitting”—change to “the DOAS fitting”

L127: “compartments”—change to “components”

L152: “following the convention”—remove; this is confusing and unnecessary

L164: “condition”—change to “conditions”

L175: three dots before “VCD” (mathematical symbol meaning “because”)—remove; not necessary

L240: “The UVSPEC”—remove “The”

Table 3 headings: “Unit”—change to “Units”

Table 3 Solar Zenith Angle and Flight Altitude: Are variable, but would you report the range used? (the column has the other ranges)

L253: “legitimate”—change to “realistic”

L258: “23 vertical grid”—23 layers over what altitude/pressure range? 70 hPa top noted in L274, but it should be stated here. Same comment for L266.

L292: “inferring”—change to “implying”

L315: “should be calibrated”—change to “was calibrated”

L329: “clean pixel”—define what this means when it’s first used (here)

L375: “premises”—change to “circumstances”

L400: “Assuming a NSR”—change to “Assuming an SNR”

L586: “premises—change to “circumstances”

L609: “As an extent”—do you mean “To a certain extent”?

L623: “volatile”—change to “unstable”

L630: “succeeding”—change to “principal”