|Review of revised version of|
Nonlinear response of tropical lower stratospheric temperature and water vapor to ENSO
by Garfinkel et al.
For the revised version, Garfinkel et al. have addressed many of the concerns raised by reviewers. In particular, the revised version is more focused, the main points are clearer, and limitations are discussed. Some minor comments are listed below that can be addressed easily (the list below is not comprehensive, also, generally the text would deserve some polishing).
After reading the revised manuscript multiple times, my understanding is that the essence of the paper is as follows: the free running GCM produces statistically signigicant, and visually compelling non-linear relations between ENSO index and lower stratospheric properties (temperature, water vapor). The AGCM runs, forced with SSTs of the last four decades, produce at least some non-linear response - the weaker statistics are presumably because the last four decades only had one or two sufficiently strong (in terms of ENSO index) El Ninos, whereas the free running GCM calculations have a sufficiently large number of very strong ENSOs, as seen in Figure 2. Finally, because the AMIP GCM calculations can be run many times (ensembles), their response is statistically more robust than what was actually measured (i.e. the non-linearity in the observational record as shown in Figure 4 hinges essentially on one data point). While this chain of arguments requires quite a leap of faith (when looking at the observations displayed in Figure 4 in isolation, the notion of non-linearity is rather disturbing), I consider the results from the coupled ocean-atmosphere GCM fairly convincing, and therefore recommend publication of the paper. Because of the paramount importance of the coupled ocean-atmosphere model calculation for the central argument of the paper, I think the authors should address the question whether the GEOSCCM coupled to MOM5 produces an ENSO sufficiently similar to reality. Iin this respect, the fact that it produces stronger ENSOs than observed as noted on P4/L24 is not comforting - but this could also be an artifact related the definition of the NINO3.4 index, and GEOSCCM-MOM5 having (most likely) a bias in the mean state. For the revised version, I'd encourage the authors to provide more information about ENSO biases in their model, and reassurance that they do not critically affect the postulated non-linearity.
P4/L31: Observed SSTs: It is worth noting that some care is required regarding "observed" SSTs as these datasets are subject to similar problems as the atmospheric temperature record; and right around the first strong ENSO event 1982/83, for example, the HURRELL and HADSST1 dataset diverge substantial in their tropical mean (See Figure 1c, Flannaghan et al., JGR, doi:10.1002/2014JD022365., 2014). Given that the Indian Ocean signal is order 0.5K, it would be worthwhile to check whether this signal is consistent among different SST datasets.
P7/L8: Add "NH" or "boreal" before "winter".
P7/L24: "tha*t* might ..."
Figure 5: caption - if I am not mistaken this is an AMIP run, please label as in text (P8L8) "AGCM GEOSCCM"