Review of “An algorithm to retrieve peroxyacetyl nitrate from AIRS” by Laughner et al. (2025)

This paper presents an effort to retrieve peroxyacetyl nitrate (PAN) from AIRS observations using the TROPESS framework and the MUSES algorithm. Given the atmospheric importance of PAN as both a tracer and a reservoir of NOx, the development of a long-term dataset from AIRS, operational since 2002, is highly relevant and potentially valuable. The work builds on previous developments applied to CrIS, particularly those described in Payne et al. (2022), and aims to establish a consistent retrieval methodology across both instruments. The manuscript is well written and, overall, technically sound.

However, while the technical implementation appears robust, I find the scientific contribution limited in its current form. The novelty primarily lies in extending the CrIS-based PAN retrieval strategy to AIRS, but the manuscript stops short of fully exploiting this opportunity. In particular, the analysis is largely confined to comparisons with CrIS over a few case studies. The broader potential of the AIRS PAN dataset to provide scientific insight remains underexplored. The scientific impact would be significantly enhanced if the authors included additional analyses, such a preliminary global climatology or seasonal cycle of AIRS PAN, assessments of long-term or interannual variability, or regional investigations beyond biomass burning plumes.

In addition, several methodological aspects, particularly the use of machine learning for quality filtering based on CrIS, require further clarification. I also have questions about the applicability of the AIRS PAN product in the pre-CrIS era.

Despite these concerns, I believe this paper could make a valuable contribution to the field if it addresses the questions and suggestions outlined above and in the detailed comments below. Strengthening the contextual framing, expanding the scientific analysis, and clarifying key methodological choices would substantially improve the manuscript’s impact.

Major Comments

The authors note that low, warm clouds over oceans can be misinterpreted as PAN. Yet, similar clouds exist over land (e.g., tropical forests). Could the authors clarify why this misinterpretation would be less problematic over land?

Why are these low, warm clouds an issue for AIRS but apparently not for CrIS? Are CrIS retrievals performed “above clouds”? If so, wouldn’t that introduce a bias in retrieved PAN due to lack of surface contribution? More clarification on this aspect is needed.

In Figure 3, some PAN features seen by AIRS (e.g., near 45°N, 145°W and 50°N, 130°W) are absent in CrIS. Are these retrievals cloud-contaminated? A short discussion of these discrepancies would improve interpretation.

A central element of this study is the application of a machine learning-based quality filter, implemented as a decision tree, to identify reliable AIRS PAN retrievals by comparing them with co-located CrIS PAN retrievals. I have several questions about its implementation and implications:

•    The paper would benefit from a clearer and more detailed description of how the decision tree was designed, trained, and applied. Specifically, is the nearest CrIS PAN value used only during the training phase, or is it required systematically for each AIRS retrieval at the application stage?

•    If the quality filtering process requires CrIS data on a systematic basis (i.e., for each AIRS retrieval), then the utility of the AIRS PAN retrievals becomes restricted to the CrIS era (i.e., post-2012). This undermines one of the main potential advantages of using AIRS — the opportunity to generate a long-term PAN time series starting from 2002.

•    By tailoring the AIRS quality filtering strictly based on CrIS, there is a risk of overly aligning the two datasets. This may introduce biases or lead to the rejection of potentially valid AIRS PAN retrievals in cases where CrIS retrievals are biased, noisy, or simply absent. For instance, even over land, the quality filtering seems to restrict useful retrievals close to strong emission sources or fires.

The current implementation applies the AIRS AVKs to the CrIS retrievals to enable direct comparison. However, in my understanding of Rodgers, applying the AVKs from one instrument to retrievals from another is generally appropriate only when the second instrument has significantly higher vertical resolution and information content. In that case, it can reasonably serve as a "truth" profile. However, both AIRS and CrIS PAN retrievals have limited vertical sensitivity, with DOFS that would typically be well below 1, indicating no vertical information. 

Fig. 12 shows that both AIRS and CrIS exhibit heterogeneous and situation-dependent vertical sensitivities (their AVKs diverge markedly when surface temperature decreases). Given this, the assumption that AIRS AVKs alone can transform CrIS data into something comparable is questionable. Ideally, a symmetric or "two-way" treatment accounting for both sets of AVKs would be required for this inter-comparison (yet this is practically challenging and still not guaranteed to yield equivalence in a formal sense).

I find it unfortunate that the discussion and analysis of the AIRS PAN product is currently limited to land. Such limitation significantly reduces its utility in key applications, such as tracing fire plumes, where a large fraction of the signal occurs over oceans. In the case of the Australian bushfires, for example, nearly the entire plume over the ocean is lost. 

Lines 132-142: This section is difficult to follow without prior knowledge of the MUSES algorithm. I recommend expanding the explanation with more technical details to make it more self-contained and accessible to readers unfamiliar with previous TROPESS-related publications.

Section 3.4: Although I understand that deriving uncertainty estimates for retrieved quantities from satellite measurements is challenging, I remain unconvinced by the authors’ approach. The reported uncertainty value (0.5 ppb) is based solely on the difference in NESR between AIRS and CrIS. However, the uncertainty should realistically vary significantly with factors such as thermal contrast, cloud coverage, PAN abundance, and others.

Section 3.4: I find the discussion on vertical sensitivity rather brief. For example, what is the typical DOFS of the AIRS PAN retrievals in fire plume regions versus remote areas? How do these values compare to those from CrIS?

Minor comments

Lines 27-31: Do the authors have an estimate of what fraction of the total APNs signal in the retrievals corresponds specifically to PAN? Given its longer lifetime relative to other APNs, could one expect its share to increase in aged plumes or background air. 

Section 2.1 would benefit from more technical information about the AIRS instrument, especially in relation to its suitability for PAN retrieval (spectral resolution, radiometric noise characteristics (especially compared to CrIS), spatial sampling and footprint size).

Lines 150-153: The manuscript mentions a "global survey" sampling approach with TROPESS products. It would be important to clarify what proportion of soundings are included in the final products. For instance, is it 1 out of 2 soundings, 1 out of 3, etc.? This has implications for data representativity.

CCl₄ is not mentioned in the strategy table (Table 2), yet it has notable absorption features in the thermal infrared that could affect PAN retrievals, especially the spectral baseline. Is CCl₄ explicitly fitted in the retrieval process? If not, how is its temporal variability accounted for?

Lines 183-186: Could the use of different a priori profiles across regions introduce discontinuities in the retrieved PAN abundances at regional boundaries?

Lines 211-213: Are the threshold criteria used for AIRS the same as for CrIS? If so, is this appropriate given the different instrument characteristics (spectral resolution, sensitivity, etc.)?

Line 232 ("the filtering approach failed..."): Could stricter filtering criteria resolve this issue?

Lines 302-304: These statements could benefit from clarification in the case of the Australian Bush Fires. Much of the plume appears to be missing over the ocean, and the soundings over land seem relatively noisy. The observation that AIRS shows no PAN enhancement, similarly to CrIS, should be interpreted with caution. The agreement between the two instruments in this case does not necessarily validate the accuracy of the AIRS retrievals, especially in light of the limited data coverage.

In Fig. 8 (and similar figures), it is difficult to assess the differences in spatial sampling and resolution between the CrIS and AIRS soundings. Including a zoomed-in view might help better illustrate these differences.

Lines 321–326: Could you clarify whether the intention is to recommend that the AIRS PAN product be used primarily as 10° × 10° spatial averages? If so, this seems quite restrictive.

Lines 327–330: Would it be feasible to implement a similar H₂O bias correction for the AIRS PAN retrievals as is done for CrIS?

Typos / technical comments

Line 1: “an approach…”

Line 43: “tropospheric column”

Line 72: Please provide the AIRS spectral bands in wavenumbers for consistency with section 2.2.

Line 139: “species”

Line 149: “affecting air quality”?

Line 283: “We tested”

Line 286: “the decision tree’s size gave it”

Line 287: “these somehow uncommon cases”

Figure 7: Consider moving panel (a) to an earlier figure where the regional view is introduced, to facilitate interpretation and cross-comparison.

Line 313: “Democratic Republic of the Congo”

Caption of Fig. 12: “The kernels shown are”

Please review all references, as I’ve noticed typos in, e.g., Clarisse et al. (2011) and MODIS (2017).

Review of “An algorithm to retrieve peroxyacetyl nitrate from AIRS” by Laughner et al.

This paper discusses the challenges of adopting an existing retrieval algorithm to measurements made by a different instrument for the sake of extending the current record of IR PAN retrievals by more than a decade; 2002–present instead of 2015–present. Specifically, the authors present their tests and analyses to determine whether the retrieval of PAN from AIRS can match the quality of PAN retrievals from CrIS for the sake of a consistent long-term record. Overall, the paper reads fairly well and I commend the authors for discussing some of the challenges caused by instrument differences (e.g., noise, spectral range and resolution, etc.), as this potentially broadens the audience for this work. The work presented here is appropriate for publication in AMT after major revisions. Overall, however, the paper fails to convince me that AIRS measurements can support a PAN product. My biggest concern is with the conclusions the authors draw from a small set of qualitative comparisons that do more to raise questions than address the primary goal of this work.

Major concerns:

• The title is misleading since the primary focus of the paper is not in the discussion of a novel algorithm for the retrieval of PAN from AIRS, but rather in how an existing algorithm can be adopted for a new set of instrument measurements. I strongly recommend adjusting the title to more accurately reflect the goal and content of the paper.
• There needs to be a sentence contrasting AIRS with CrIS, especially as far as instrument noise and spectral range goes, to help the reader understand the goal of this paper and why retrieving PAN from AIRS is more challenging than PAN from CrIS.
• Line 5: Here the authors state that they retrieve PAN from AIRS but omit all retrievals “from low, warm clouds over ocean”, but this is misleading because in Section 3.2, Line 245, the authors conclude that the AIRS PAN product needs to exclude all retrievals over ocean since they struggle to isolate only those cases with interference from low, warm clouds. The abstract needs to correctly reflect their conclusions. Moreover, it will help the reader (and promote the validity of this work) if the authors state in the abstract that the CrIS PAN product does not need the same type of land/ocean filtering as the AIRS PAN product.
• Line 5: “…we…develop a decision tree quality filter trained to predict whether a PAN value retrieved from AIRS…” The title should reflect this primary goal and outcome. Suggested new title: A quality filter for PAN retrievals from AIRS.
• Line 7: “We show that AIRS is capable of retrieving PAN plumes…” I’ve studied the figures and reread the paper, but remain unconvinced that the authors succeeded in demonstrating this. At best, the results show just how challenging it can be to design an algorithm for retrieving trace gases from two instruments as disparate as AIRS and CrIS.

• The authors list many other PAN studies and products, but omit mentioning other successful AIRS+CrIS long-term products. This effort to retrieve a trace gas species from AIRS and CrIS is not the first of its kind. Others have successfully addressed instrument differences between AIRS and CrIS (especially with respect to interference from clouds) to generate consistent long-term records for a host of other trace gas species. Perhaps the authors can contrast their approach to other AIRS+CrIS records to help the reader better understand the authors’ algorithm choices and subsequent challenges.
• Line 100: “..the OE algorithm calculates uncertainty from noise only.” As the authors well know, OE is a generalized retrieval framework, not a universal retrieval algorithm. The way that noise and uncertainty are quantified in practice vary significantly across the many OE products in operation today. I strongly encourage the authors to rephrase this statement (and similar ones throughout the manuscript) to clarify such characteristics as their own algorithm choices instead of attributing them to the OE framework in general. Again, it may be helpful for the authors to consult and mention other OE retrieval implementations that quantified noise, error and uncertainty in different ways that could help inform their results.
• First paragraph of Section 2.4: The summary of the TROPESS product presented here is confusing. Many of the phrases reads more like jargonn than scientific explanations, e.g., what is a “global survey sampling approach”? And, can the authors clarify what they mean with a “forward” and “reanalysis” stream? Why not process the full record (2002 to present) with “the latest version of the MUSES algorithm”? If two different MUSES algorithms are used to process the full record (2002 to 2021 versus 2002 to present), could the resulting PAN product really be considered a consistent record?
• Does the TROPESS MUSES PAN product from CrIS cover the full global range of CrIS measurements on a twice daily basis? This is not clear in the text.
• Line 212: How did the authors decide on a surface temperature threshold of 265 K?
• Lines 274–275: “different vertical sensitivity between CrIS and AIRS.” What exactly is the difference? There are many published texts contrasting and quantifying the main instrument differences between AIRS and CrIS. I strongly recommend that the authors add the appropriate citations as well as summarize a few of them in this manuscript, specifically with respect to instrument noise, spectral coverage and resolution.
• On page 15, the authors conclude that it is best to exclude AIRS PAN retrievals over ocean and deserts from the final product, but I wonder if this is sufficient given the results they present. How do the authors know that their PAN retrievals over land-based low, warm clouds are more accurate than over ocean-based low, warm clouds?
• Lines 313–315: The authors communicate that elevated PAN values are present in both the AIRS and CrIS products presented in Figure 8, but I fail to see this. The AIRS PAN product has a significant speckle effect (random distribution of high and low values) that is mostly absent in the CrIS PAN product. The CrIS PAN product indicates an elevated plume over the region centered on 10˚S, in contrast to much lower values throughout the rest of the mapped region. The AIRS PAN product, on the other hand, has a speckled distribution of PAN throughout the southern African region without any obvious featured plumes. As this work is currently presented, the conclusion is not supported by the results. I suggest the authors either rethink (and rephrase) their conclusion, or present results in support of their current statements. I have the same concerns for results communicated in Figure 9.
• Lines 315–320: While I appreciate the authors’ attempt to communicate the practical interpretation of their product in downstream applications, I feel this section is a bit muddled. Does the co-located CO product need to be from the same TROPESS MUSES suite, or can an independent CO product serve to confirm elevated PAN retrievals? 
• Line 344: Why would AIRS maximum sensitivity decrease more quickly as surface temperature decreases?
• Section 4: “AIRS and CrIS is < 0.1 ppb when averaged to a 10˚ x 10˚ box”, which suggests only spatial aggregation. Yet later in the paragraph the authors suggest that users choose to average 250 PAN retrievals over an unspecified “spatiotemporal window”. This is confusing (even misleading) as the authors do not present or discuss whether the AIRS PAN product quantifies small change over time. It appears the authors simply assume that averaging over time (days? Weeks?) will yield the same results as averaging over space.
• Line 377: “We recommend averaging 250 AIRS soundings which will result in a ~0.1 ppb error.” Why is this type of averaging not recommended for CrIS PAN retrievals? I.e., why does the CrIS PAN product not display the same speckled pattern? Also, on Line 352 the authors state that the 0.1 ppb value should not be interpreted as an overall error, yet here they state it as an overall error. Please clarify.

Minor comments:

• Line 95: “…GEOS-Chem profiles appended to the top.” This is not sufficiently descriptive. What do the authors mean by “append” and by “top”?
• Line 96: “aircraft free tropospheric PAN column averages” What does this mean?
• Figure 8: What does the box over the southwestern region represent?
• Lines 282–289: This discussion is confusing. E.g., “However, we found that either caused the filter to screen out soundings with enhanced Xpan.”, “…to account for these someone uncommon cases”, etc.
• Line 332: “The CrIS radiance noise is lower than the AIRS radiance noise.” Can the authors quantify this difference and provide references to text that demonstrate it?