Review for AMT of Wilcox et al. manuscript entitled “Deep convective cloud system size and structure across the global tropics and subtropics.”

Summary: 

The authors develop a database of convective cloud systems (using MODIS as a reference) with important characteristics mapped to each identified system (including some microphysics-related [e.g., effective radius from MODIS], and some environmental variables [e.g., CAPE from MERRA-2]).  The database would be of use to the community for studies of what processes might drive system sizes (one could imagine co-locating other predictor variables to their database to support such a study), and their database would be potentially useful for evaluating climate model features and informing targets for climate model convective parameterization development, particularly given the length of the dataset, which is long enough to cover important large scale modes of tropical variability and a sampling of their phases (e.g. ENSO, MJO). 

Prior to publication, I have two comments warranting major revision related to recommended additions to the database, although I lay out steps toward making such additions quite tractable and hopefully easier to implement.  Considering such additions would make this database more useful to the convection community. Secondly, there are some enhancements needed to descriptions and alternate interpretations that I think are warranted in parts (mostly elaborated upon in Minor Comments below).

Major Comments:

System flagging (#1) – what do the authors do for systems whose boundaries reach the MODIS swath edge?  Are those systems flagged?  Such systems are always going to be under-reported in size, since some unknown (greater or equal to 0) fraction extends beyond the viewing area, which therefore will bias future science analyses.  If MODIS pixels touch the edges of the swath granule (in the swath or in the pixel direction), I think inclusion of an “edge” flag (binary: e.g., 0 for not touching, 1 for touching) would be useful.  That way, users who would like to use this database can decide their own comfort level for using systems fully within the viewing area for statistical analysis versus not using systems if they are not confident that their size is accurately reported.  By including a flag as opposed to removing, the developers are not forcing users to make a decision.

System flagging (#2) – regarding use of MERRA-2 outputs (CAPE, etc.) mapped to the identified systems. Have the authors plotted latitude-longitude maps of MERRA-2 convective systems (e.g., map of OLR or rainfall rates?) alongside maps of identified convective systems from MODIS?  For any day & time, they often do not resemble each other (unless it is an O(1000km) system or mid-latitude system).  MERRA-2, although it assimilates observational data, more often than not, does not simulate individual systems at the same time/location (unsurprising since MERRA-2 convective systems are dependent on their convective parameterization). Many times, MERRA-2 produces a convective system where the observations do not indicate one exists (or vice versa).  For mesoscale environments quickly modified by convection (particularly diagnostics influenced by the planetary boundary layer [PBL] characteristics), this issue might impact CAPE computations (or anything related to T and Qv) since not having a convective system in MERRA-2 leads to the PBL remaining “undisturbed” and characterized by a buoyancy metric that differs from reality.  I think a) it should first be determined if an equivalent convective system was identified in MERRA-2 via some determination of whether rainfall was beyond some threshold over the MODIS convective system area and/or OLR was below some threshold for the same time/space locations as the observed MODIS system, and b) a MERRA-2 flag should then be derived such that if a convective system is not found in MERRA-2 at the same time/place: e.g., 0 is reported for no equivalent system existing in the reanalysis; and 1 if MERRA-2 itself is simulating a system going on at the same time/place an observed system is evolving.  Having this flag allows users to have confidence in MERRA-2 environments (or to use their own filtering) if it can be known in advance that the MERRA-2 environment is at least approximately resembling observed mesoscale convective environments.  

Minor Comments:

1. Lines 48-50: The authors use MODIS (sun-synchronous orbiter), and I do not think systems are tracked (as can be done with GOES) in this database; writing the word “Lagrangian” is confusing, unless I am missing something, in which case clarification is needed.  
2. I am aware of the literature suggesting how cloud top distributions for effective radius (Re) approximate a Re(z) height profile into a cloud; however, this view is not unanimously agreed upon for all convection environments globally in the tropics. Aircraft are the only source of “validation”, and not only are these data sparse, aircraft usually only infer sizes adjacent to convection and not inside convection (due to flight restrictions in general, with a few exceptions).  It is unnecessary speculation to call cloud top distributions a profile, particularly in convection of varying life stages.  Therefore, I recommend relabeling the “profiles” in section 2.4 and figure being shown as “cloud top distributions.”  Then, the discussion text noting how this may be equivalent to cloud top PDFs can remain for interpretation purposes. From a perspective of introducing an observational database and its utility, it is not clear to me why this assumption has to be pre-supposed for discussing the database and plotting a preliminary result.  Spatial variations as a function of cloud top heights themselves are very useful too, independent of the assumption that they represent profiles.    
3. Regarding Fig. 5b, the sensitivity to averaging – to my eyes, those differences are comparable to day/night differences over land in Fig. 5a (or in other words, they appear large). Eyeballing, I see a factor of 5 at the low end, and 2 at the high end.  Thus,  I do not understand the comment on lines 242 about them being similar with slight over-sampling of large clouds. What about the small end too?  Since the authors intend for this dataset to be used as a model evaluation benchmark, and in light of this dependence on averaging, how do the authors propose to apples-to-apples facilitate this database to be used for comparing to a range of models, since averaging does have this impact?  Can averaging impacts be incorporated into an uncertainty estimate somehow?
4. The discussion of convective lifecycles in the conclusions section was important, but I think this discussion should be introduced much earlier during the “interpretation of database results” parts of the manuscript. For example, lines 183-185 (“The goal…”) suggest environment, regional variations, land-ocean differences, but convective life stage also matters here; and, it also matters for cloud top Re differences. Another example: for lines 266-274 (sentences beginning with “These results…”), talking about a size of the cold areas relative to the total size of the cloud and potential relation to mesoscale organization/environments, a recent study (Elsaesser et al. 2022; https://doi.org/10.1029/2021JD035599) showed that convective fraction (which definitely relates to the fraction of cold cloud below 220K discussed here) is very connected to lifecycle, as shown in Figure 2 of that paper.  Furthermore, lifecycle also has implications for how Fig. 7 and 8 in this manuscript would be interpreted, because one could imagine only plotting up the maximum sizes along the path of any system as a function of shear/CAPE instead of any instantaneous size.  In short, life cycle discussion should be brought in at various parts during discussion of interpretation, instead of first mentioning it in the conclusions. 

Minor Typos:

I did not focus on grammar nor typos, though I did catch a few:

L285: “…interacting with over convective elements”; should “over” be “other”?

L287: “these quantities are certainly not predictive of these quantities” ; rephrase to avoid confusion, and avoid double use of “quantities.” 

This papers introduces a dataset focused on atmospheric deep convective cloud built using various satellite observations and a specific object-oriented methodology. Before going further it seems to me this is more a paper relevant for ESSD than for AMT. Indeed the method is not new but its application to MODIS is ! For with this in mind, and so assuming this is more a data paper than a method paper I have the following positive comments.

This study aims to provide process oriented observational data to help development and evaluation of numerical atmospheric models and possibly parameterization of deep convection. In this respect the paper is very convincing. The new dataset is a timely addition and a nice consolidation of previously available of existing equivalent geostationnary and polar oribter based datasets. The long span of hte proposed record is indeed a strong asset of the present dataset. The extension to the AMSRE 89GHz information is also a nice new features (that probably needs to be put more forward in the manuscript). 

The paper is clearly written and well referenced. Indeed the selected illustrations , like the joint PDF of CAPE, shear and size for different regions are very convincing of the possibilities of the dataset for its endeavor. Yet I would welcome a little more cautionnary notes of the use of reanalysis based on assimilation and parameterized GCM for deep convection processes studies.