This paper describes an intercomparison of the properties of cirrus measured by in-situ probes vs those retrieved from a ground-based infrared spectrometer that spans from the mid-infrared to the far-infrared (i.e., from 1500 cm-1 to 400 cm-1).  Th authors use a ceilometer to ascertain the base and top of the cirrus clouds, and to estimate their optical depth.  Lastly, their infrared retrieval framework also retrieves thermodynamic profiles, which they compare against ERA5 model output and retrieved profiles from a collocated ground-based microwave radiometer.

The paper achieves what it sent out to do, namely, to perform an intercomparison of derived cirrus properties, which includes an evaluation of the assumption of the habit of the ice particles, and thermodynamic profiles.  It is well written, and the references included are sufficient.

My main concern is that this intercomparison evaluate 11 samples that were collected within a 30-minute period on a single day at a single location.  Thus, I must ask: how representative are these results?  Would these same results hold for different atmospheric state conditions (i.e., temperature and humidity profiles) or cirrus conditions (i.e., optically thicker clouds, cirrus generated by other means, etc).  This is further complicated by the sky image shown in Figure 2, which suggests that their in-situ aircraft is leaving a contrail (based upon the flight pattern on the right and the clear spiral signature in the sky image).  Can we assume the ice habit in the contrail is consistent with the habit of the background cirrus?  How much is the contrail impacting either the radiometric observations made by the FINESSE or the in-situ obs (when aircraft starts its next circle)?  Additionally, the ice optical depths sampled here are small; less than 0.2 – so again, do the conclusions (e.g., the importance of the far-infrared channels) hold when in cases when the optical depth is larger?  Before this paper should be accepted, I would hope that the authors can address these concerns well.

Minor concerns:

Fig 15 (question 1): why do the INCAS points change mean values and have different error bar ranges in the 4 panels?  This is most easily seen looking at the bottom-most panel against any of the other three?

Fig 15 (question 2): Why are the FINESSE error bars so much larger for cases 8-11 when HBR is assumed relative to when you assume ½ HBR+ ½ SBR?  This does not make sense to me.

You are using a Bayesian retrieval framework, which is able to provide degrees of freedom for signal.  For these very small optical depths, what is the DFS for the retrieved effective radius?  Does it change when you assume different habits?  Does it change when you only use the midinfrared vs when you use both the midinfrared and far infrared?  (This would be a pretty convincing point to make, if the far infrared is indeed important).

Fig 17: caption errors: INCAS measurements are green dots, and only using the mid-IR is the violet diamonds

You spend very little time talking about the thermodynamic retrieval and its accuracy.  Personally, I do not feel it adds anything to the paper; indeed, it is more distracting.  If you elect to keep that portion in the paper for the next iteration, I would want to see much more discussion about it (recognizing that you don’t have strong sources of truth).  But, in particular, your statement on line 468 about “excellent agreement…inside the cirrus” needs to be tempered.  Does the FINESSE retrieval actually have any information content (i.e., degrees of freedom for signal) at that altitude?  I would be very surprised if it did.  And if the DFS is very small, then the agreement in temperature within the cirrus is more of a happy coincidence (provided by the statistics of the prior dataset used in the constraint, which you did not discuss at all) then real skill by the retrieval. 

This paper addresses a critical issue for passive remote sensing of clouds from infrared spectral measurements. Previous measurements (e.g., Bantges et al 2020) indicated inconsistency in the ice droplet properties in the current database, which leads to difficulty in achieving radiance closure across the visible-MIR-FIR spectra in the measurements. This paper revisits this inconsistency issue using the measurements from a new campaign with a different deployment strategy, i.e., measuring cloud spectra from ground as opposed to airborne. I found this study has a well-motivated objective and a great potential to help found methods utilizing future FIR satellites such as FORUM for cloud remote sensing. However, I also found the current manuscript has several major issues. The presentation and interpretation of some results especially raised concerns, which should be addressed before it is published.

A major comment is that this work, as it currently stands, shouldn’t be mistaken as a retrieval test that proves or disapproves the feasibility of retrieving cloud microphysical properties from the FIR spectral measurements. In essence, this is a radiance closure test, showing sensitivity to ice habit as well as other potentially retrievable parameters. If the authors intend to establish this as a retrieval work, the paper needs to more systematically and comprehensively test whether and how the cloud microphysical properties (size, habit, etc) can be “unambiguously” determined from macrophysical ones (optical depth, temperature, etc) and from other atmospheric states (temperature, humidity, ozone, etc). This should be done using both synthetical data with perfectly known truth to quantify such quantities as averaging kernels and degrees of freedom for signals (DFS) and using independently measured cloud data to validate retrievals from actual spectral measurements. For example, the use of atmospheric profiles from ERA5 as both initial guess and validation is unfit to represent an actual retrieval. Moreover, although varying the multiple parameters is shown to lead to reasonable agreement between simulated and measured spectra, there is no proof that there is enough information content to determine them altogether in a real retrieval. In fact, some results, such as the noticeable differences in the sizes “retrieved” when different habits are assumed (Fig 15), suggest there is substantial degeneracy. If this (retrieval) isn’t the intention, the paper should be clear about it and avoid misleading claims such as “first retrieval” in Abstract (Line 1) and throughout the paper. It would be more appropriate to state the tests as “adjusting”, as opposed to “retrieving”, the relevant cloud parameters.

Another major comment is that the paper doesn’t provide a clear answer to the motivating questions. For example, is the pan-spectral inconsistency pointed out in earlier works (e.g., Bantges et al) now reconciled, e.g., by using the ground measurements or by considering a HBR habit? Given that some coauthors here were those who championed the earlier works, I would very much like to see such answers, which would help put this work in context and better identify its value. To these questions, combing Fig 16 with Fig 18/19 and showing how the residuals can (or cannot) be minimized by using different residuals would be helpful.

Regarding the ground-based measurement strategy, I would appreciate more discussions on its advantage and disadvantages compared to the airborne approach. For example, the cirrus as visualized in Fig 2 indicates clearly spatial inhomogeneity and, as noted in the paper, varies in measurement time; how does this affect sampling consistency and representativeness of retrieval (e.g., the optical depth) from spectra collected over a finite (200-m) FOV? Reflections on these questions and/or suggestions for future campaigns would be especially useful. A naïve question is: did the airplane affect the cloud fields and exacerbate the inhomogeneity issue, as indicated by the contrails?

Fig 16: There seems to be equally discernible signals in MIR (around 800 cm-1), which raises the question on the FIR benefits claimed. Fig 17 doesn’t show convincingly different performance between the two results (with vs without FIR). Even if it did, this wouldn’t make strong evidence given the aforementioned degeneracy and strong sensitivity to habit. To elucidate this point, a formal retrieval assessment based on DFS as suggested above would be more quantitative and convincing.