This is a very interesting work that extends the diagnoses of the important OLR-Ts relationship in ENSO from broadband to spectral radiation. The development and application of a spectral kernel dataset is a notable strength (despite some questions detailed below). The paper shows that spectral data makes it possible to view the lagged OLR-TS relation with more clues relatable to the geophysical variables at the process level, which makes an important finding. I would recommend publication if the following comments were addressed.

The Introduction gives a good literature review to motivate this work, with a rather complete collection of relevant previous works, despite some misses, for example, Huang and Ramaswamy 2008 (https://doi.org/10.1029/2008GL034859) which was one of the earliest data-based diagnosis of spectral OLR-Ts relationship.

On the other hand, I found the paper does not make as strong connections with the previous works when discussing the results, which potentially impedes the revelation of the novelty or difference in the findings here. For example, much emphasis of the paper is on the lagged OLR-Ts relation, for which the spectral signatures are related to geophysical drivers (Figs 2/3 and texts around Line 200). However, these findings were similarly made based on broadband kernel-decompositions (e.g., Fig 3 of Kolly & Huang). Given the objective of this paper is to demonstrate the advantage of spectral information, such comparisons can be used to discuss what additional information is brought in by the spectral data.

A main, technical comment is on the kernels themselves, which are computed based on monthly profiles (Line 120) – a simplification known to introduce biases (e.g., see Huang and Ramaswamy 2009, https://doi.org/10.1175/2009JCLI2874.1). It remains to be demonstrated how well the kernels produced here explain the total spectral OLR changes. Although it is qualitatively discussed (Line 210), the biases are not quantified – better to show the residuals in closure tests of different lags. Another suggestion is to compare the kernels produced here to other kernels. For example, the broadband radiative sensitivity values spectrally integrated from the spectral kernels should reproduce broadband ERA5 kernels truthfully computed from instantaneous profiles (Huang and Huang 2023, https://doi.org/10.5194/essd-15-3001-2023). Such comparisons can be made with respect to maps of vertically integrated kernel values and with kernel-reproduced lagged OLR-Ts relationship (Fig 1). This would help provide a measure of the uncertainty in the results.

Given the results are exclusively for clear-sky. It is better to indicate this in the paper title.

Lastly, there are numerous grammatical errors. The English writing needs to be thoroughly proofed/edited.

The authors investigate the radiative response to ENSO from a spectrally resolved perspective. The underlying idea is that spectrally resolved OLR holds information about individual contributions to radiative changes from temperature, humidity, and ozone, which cannot be resolved by broad-band OLR. The authors compute spectral kernels combining ERA5 data with a radiative transfer model. Comparing the spectral kernels with the satellite observations allows to identify changes in surface temperature, air temperature, humidity, and ozone, and to further decompose these changes into contributions from different atmospheric levels. The authors find the changes one would expect from ENSO in the spectrally resolved OLR product.

I read this paper with great interest and recommend publication of the study under the condition that the comments below are addressed. I divide my comments roughly into major, minor, and suggestions that I don’t require to be addressed. I have some expertise with radiative feedbacks and tropical dynamics, but almost no experience with satellite observations.

“major”

The authors missed the opportunity to strictly decompose the observed spectral OLR into contributions from the individual feedback processes using spectral fingerprints, as was done for example in Huang et al. 2010 (https://doi.org/10.1029/2009JD012766). While I would be very interested to see this, it would be exaggerated to ask the authors to perform this analysis, and the paper contains valuable insights even without this. However, I still mention this here, because it implies some limitations to the interpretations that the authors give.

• l. 9 – 10: This is the first example where it matters. I strongly suggest to reword this sentence in the abstract because there was no strict mathematical “decomposition”. Rather, the satellite observations were compared by eye to the kernels to identify the feedback contributions.

• Huang et al. 2010 should be cited in the introduction as an example for how spectral OLR can be used to identify individual feedbacks

• l. 69, l. 329: I don’t quite understand what is meant here by “diagnostic”. If it refers to the decomposition into feedbacks using spectrally resolved OLR, then this should not be referred to as new, because it has already been done by Huang et al.

• l. 115 – 122: It is not clear to me how the kernels are computed from the data. In particular, I would like to know how the derivatives were computed. A (finite) partial derivative needs radiation computations from at least two different atmospheric states, and I don’t understand from the description which these are. I assume that in each partial derivative computation only one variable was perturbed, but how / to what value?

• l. 155: I found no further investigation of this, can you point me to where this is done?

• All figures:

  • “slope” is not a very descriptive y-label. l. 110 – 112 state that this slope is referred to as “ENSO feedback”, so maybe “feedback” in Fig. 1, or “spectral feedback” in Fig. 2-4 could be an option?

• Figs. 2, 3, 4:

  • The color map should have linearly increasing lightness and brightness. The current color bar artificially suggests strong gradients where there are none. For example, with increasing values the colors get darker until 0.455 in Fig. 2 and then brighter again, creating false perceptions of the actual values. Some hues even seem to repeat. I find https://doi.org/10.1038/s41467-020-19160-7 to be a very helpful resource for picking a color map.

  • Figs. 2 and 3 should use the same value range so that they can be accurately compared to each other

• l. 182: The fact that there is a transition from analysing satellite observations to analysing something derived from the spectral kernels and reanalysis should be made clear.

• Section 3.4, Fig. 4, also l. 284-286: How sure can we be that this signal comes from ozone? While this spectral band surely represents ozone where ozone is optically thick, the positive values in the Eastern Pacific could also originate from increased surface emissions through the atmospheric window if ozone is not optically thick there. A fingerprint analysis might have been able to clearly separate between ozone emissions and surface emissions. Given that this has not been done, can we be sure how much of this is ozone signal and how much is surface signal? My doubt is intensified by the fact that the 905 cm^-1 channel which purely represents surface emissions also shows increases in the Eastern Pacific.

• l. 272: The humidity changes from ENSO are complicated and may also arise from the generally warmed atmosphere. Without further analysis or citations it is not totally clear to me that the positive water vapor feedback comes from the ascending branch of the Walker circulation. In particular in the dry regions, radiation is also extremely sensitive to small humidity changes. Furthermore, even though this study treats clear-sky regions, cloud (un)masking can play a role for apparent water vapor feedback.

• l. 277-279: Where does this statement come from? Why tropospheric cooling? Doesn’t an increased OLR in the FIR imply a warmed troposphere? The sentence as a whole is confusing to me. Cooling cannot maximize in a spectral region, only radiation can.

• Major edits are needed with respect to writing and grammar

• I suggest to indicate that clear-sky OLR is studied (and not simply OLR) in the title

“minor”

• l. 48: reproduced by models, I assume?

• l. 77: most of the paper is written in terms of wavenumber, but here wavelength is used. I suggest to stick to one, preferably wavenumbers

• l. 123 – 134: This is all fine but was hard for me to understand. It could be due to my lacking expertise of satellites. For example, I only understood in the Discussion section that FORUM is a satellite mission, and even a quick google didn’t help me understand what the “synthetic” in “synthetic radiance” refers to (although I could figure it out eventually). This paragraph could be improved with small edits.

• l. 161: “maps”: they are not maps

“suggestions”

I don’t require these to be addressed.

• l. 195: This is really counter-intuitive, because a positive value corresponds to a negative feedback. I see how this arises from the fact that OLR is defined positive outward, which makes sense. Still, can this confusion be overcome? Maybe by showing -dOLR/dT in the figures, or by referring to net downward LW instead of OLR, or by explicitly stating that a positive value implies a negative feedback?

• Fig. 2 first panel would profit from a line at y=0

• l. 329 – 333: Without finger printing, it is not clear to me how this method can be used for a quantitative analysis to evaluate climate models. Furthermore, when comparing Figs. 2 and 3 it seems that ERA5 does a good job, except in the FIR where the satellite anyway doesn’t really know what’s going on, because it’s extrapolated. So how would this add anything beyond directly comparing the feedback components due to surface temperature, air temperature, water vapor, etc. between ERA5 and climate models? These are directly available from the output.

• l. 319 – 323: it could be made clearer that “The sum of the Planck surface, atmosphere, water vapor and ozone feedback” refers to ERA5 and the “observed signal” to the satellite.

• In my opinion, the spectral kernels are the coolest contribution of this paper to the scientific community. It would be amazing to have them publicly available. :)

I read this manuscript with great interest. The study addresses the radiative response to ENSO from a spectrally resolved perspective, which provides valuable insights into the contributions of individual processes. By combining CERES broadband observations with spectrally resolved measurements from AIRS and IASI, and by employing radiative kernels derived from ERA5 reanalysis, the authors disentangle the roles of temperature, water vapor, and ozone in shaping the clear-sky OLR variability associated with ENSO. The spectral dimension is a particularly valuable addition, as it helps separate overlapping processes that cannot be distinguished in broadband analyses.

I found the introduction to be well written and comprehensive, setting a solid foundation for the work. However, the remainder of the manuscript often lacks rigor, and the quality of the English requires substantial revision.

My comments are structured in three categories: General, major, and minor, as detailed below. I would recommend publication of this manuscript provided that the authors carefully address the issues raised in this review.

General comment:

While the study offers interesting insights into the spectral perspective of ENSO-driven OLR variability, I found that several aspects of the analysis lack the level of rigor and quantitative support.

First, the comparisons presented in the manuscript are mostly qualitative. For example, differences between instruments or between kernels and observations are described verbally but rarely quantified. Providing numerical values (e.g., absolute differences, percentage differences) or plots of differences (e.g. for Fig. 2 and Fig. 4) would significantly strengthen the robustness of the analysis.

Second, unless I overlooked it, the authors do not discuss the impact of the difference in overpass time between AIRS and IASI as a potential explanation for the discrepancies observed in the atmospheric window regions. Even when restricted to ocean-only observations, such differences have already been reported in the literature (e.g., Whitburn et al., 2020), and one likely contributing factor is indeed the distinct overpass time of the two instruments. This issue should be explicitly acknowledged and discussed.

Third, it is important to note that Metop-A started drifting from its orbit in June 2017. The manuscript does not discuss the potential impact of this drift on the results. Considering Metop-B data from 2018 onwards could help avoid potential biases. Addressing this point would increase confidence in the robustness of the conclusions.

Major comments:

• Lines 104-105: The manuscript mentions “[…] removing […] the linear trend for the whole period.” Could the authors provide more details on the method used to remove the linear trend? In particular, it would be important to confirm and explicitly emphasize in the manuscript that this step is intended to remove the effect of increasing greenhouse gas concentrations over the study period.
  
  
  
  At the same time, the authors should discuss the implications of this methodological choice. Assuming a uniform linear trend may not fully capture the real evolution of the climate system, which can be non-linear (e.g., acceleration of warming in recent decades). There is also a risk that part of the signal interacting with ENSO could be inadvertently removed. While detrending is a reasonable approach to isolate interannual variability, the rationale and limitations of this choice should be clearly stated.

• From my understanding of the manuscript, the study treats El Niño and La Niña phases equivalently by relying on a linear regression with the Niño 3.4 index to evaluate the radiative impact. However, El Niño and La Niña are not necessarily mirror images of each other: their amplitudes, spatial structures, and temporal lags can differ significantly. Could the authors clarify whether an analysis was performed separately for El Niño years and La Niña years? If so, were the results consistent? If not, at least a discussion of this potential limitation would be important to assess the robustness of the conclusions.
• Lines 127-128: The spectral flux is computed using the Gaussian quadrature method with only three angles. I wonder whether this choice provides sufficient accuracy for the purposes of the study. I suggest that the authors evaluate the potential impact of using a larger number of angles (e.g., 4 or 5) on at least one or two representative cases, to demonstrate that the results are not sensitive to this assumption.
• The manuscript primarily discusses the slopes of the lagged regressions, but the correlation coefficient (R) or the fraction of variance explained is almost never reported. Including R or R² would be important to assess how much of the variability in spectral fluxes is actually explained by ENSO. Reporting these values would strengthen the robustness and interpretability of the results.
• The slopes of the lagged regressions are expressed in W m⁻² 10 cm⁻¹ K⁻¹. While this is certainly relevant to evaluate the radiative response to ENSO, the analysis would benefit from also discussing the changes in relative terms, for example as a percentage or in terms of Brightness Temperature. Indeed, regions where the slope is largest often correspond to wavenumbers with the highest radiance, rather than where the ENSO response itself is maximized. Including such normalized metrics would provide a clearer and more physically meaningful interpretation of the spectral response.
• Lines 175-176: “Since AIRS and IASI differ only in these two spectral regions, the opposite biases could compensate each other […]”. Could the authors clarify whether this compensation actually occurs when the total (integrated) flux is calculated?
• The manuscript does not seem to discuss the uncertainty associated with the reconstructed radiances in the spectral ranges not covered by AIRS. Could the authors provide an evaluation or at least an estimate of these uncertainties? Such information is crucial to assess the robustness and reliability of the results in those spectral regions.
• Lines 223-224: To evaluate the spatial pattern of the OLR response to ENSO, the authors select a channel at 905 cm⁻¹ to extract information on surface temperature and the effect of water vapor. However, no significant H₂O absorption line is present at 905 cm⁻¹. Would it not be more appropriate to consider the OLR integrated over a spectral range within the window region? This would better capture the combined effects of surface temperature and any residual water vapor absorption.
• Lines 234–235: The manuscript states, “As already anticipated in section 3.2, the different temporal and spatial sampling between the two here becomes evident, as highlighted by the more scattered patterns of AIRS with respect to IASI.” However, in lines 100–101, it is mentioned that all datasets have been regridded to a common 2.5° × 2.5° grid using bilinear interpolation, and both datasets correspond to monthly mean fluxes. It is therefore unclear why AIRS would appear more scattered than IASI. Could the authors clarify the origin of this apparent discrepancy? For instance, is it due to residual effects from cloud screening, incomplete coverage, or some other factor?
• The Discussion section needs significant rewriting. Several passages are unclear or imprecise, and the analysis would benefit from more quantitative support. References to figures are often missing, making it difficult for the reader to link statements to the presented data; the authors should clearly indicate which figure panels support each point. Additionally, the last paragraph should include a reference to Whitburn et al., 2021, as similar points were discussed there.

Minor comments:

• Lines 18-19: The sentence “It contains both the signature of the increased concentration of radiatively active species […]” is slightly misleading. The spectral OLR contains the signatures of all radiatively active species, not only those whose concentrations increase. The current formulation is correct when referring specifically to OLR trends.
• Lines 87-88: The statement “AIRS fluxes are calculated exploiting the same angular distribution models of CERES […]” is not fully clear. CERES provides broadband OLR, while AIRS produces spectrally resolved fluxes at 10 cm⁻¹. Could the authors clarify how the CERES ADMs are applied to AIRS data in this context?
• Line 92: […] “from 15 to 1995 cm^-1”. Isn’t it from 10 to 2000 cm^-1?
• Line 95: Metop-C was launched in 2017 and not in 2013.
• It could be useful to include a plot of the Niño 3.4 index for the study period, so that readers can better visualize the ENSO variability. In addition, a brief discussion of the relevant ENSO phases during this period would provide valuable context for the analysis.
• Lines 154–155: The manuscript mentions that the difference is “likely due to AIRS’s clear-sky scene selection, as will be investigated further in section 4.” However, I could not find a further discussion of this point in section 4. Could the authors clarify or correct this statement?
• Lines 196–197: For clarity, it would be helpful to remind the reader that a negative slope corresponds to a decrease in the TOA OLR, which implies that more radiation is trapped and therefore corresponds to a positive feedback (and conversely).
• Figure 2: Why does the x-axis stop at 1515 cm⁻¹, whereas Figure 3 extends further? For consistency, it would be preferable to use the same spectral range in both figures unless there is a specific reason for this choice.
• Lines 205–215: This paragraph is not very clear and would benefit from some rewriting. In particular, the statement “the net signal in the FIR and MIR spectral region” is too vague. Could the authors specify the exact wavenumber ranges referred to and indicate clearly in which panel(s) of the corresponding figure this signal can be seen?
• Figure 3: For clarity, it might be better to move panels 1 and 2 to the bottom of the figure, as they are discussed later in the text. This would make it easier for the reader to follow the discussion in the correct order.
• A study of ENSO teleconnections beyond 30° N/S could be an interesting addition to the conclusion or perspectives. Extending the analysis to the extratropics might provide further insights into the global radiative response to ENSO.
• In the Conclusion, the authors state: “to the best of our knowledge, this is the first time that the radiative response to ENSO is analyzed using satellite-based spectrally-resolved measurements of the OLR.” This statement is not entirely accurate, as similar analyses, albeit to a lesser extent, were conducted in Whitburn et al., 2021.