In this study the regional climate model, CRCM5, is employed to dynamically downscale a large ensemble of climate change simulations to investigate the nature of downscaled responses to the modeled North Atlantic Oscillation (NAO) and its influence on future European climate. By employing a large ensemble, the authors are able to evaluate future downscaled responses associated NAO inter-annual variability in addition to mean changes. The authors set out four key questions related to, documenting the properties and fidelity of the modeled NAO in both the GCM and RCM; the associated screen temperature and precipitation responses in both models; and how such properties change under future external forcings.
In my initial review of this manuscript, I asked the authors to provide a more detailed investigation of the RCM's ability to faithfully reproduce the NAO signal in the driving data and I provided a detailed suggestion on how that might be done. The authors have done an excellent job of addressing this issue in the revised manuscript. However, in reading through the new manuscript I continue to find it difficult to read due to back-and-forth referencing (including far too many acronyms), which I mentioned in my minor comments of the first draft.
In particular, the authors have front-loaded Section 2 with discussion and derivation of diagnostics which are then used in later sections of the paper. However, the discussion in those later sections is conducted as if that earlier material was just presented. It was not, and the reader must constantly stop and hunt down the meaning of acronyms and variables. This makes it very difficult to read the paper. It could easily be resolved by bringing the motivation, derivation, and application of diagnostics and ideas together. I have identified a few places but it is problematic in more than these examples. I don't want to hold the authors to another major revision as I did not raise this so strongly in my first review. As a consequence, I offer my comments as suggestions to the authors and leave it to them to follow through. I would highly recommend that the authors attempt to improve the flow of this paper, however, as I feel it is a very good piece of work and I wouldn't want to see that obfuscated by the manner in which it was presented. With this understanding, I therefore recommend publishing this work with minor modification. My detailed comments follow.
Minor Comments:
1) p.1 l.11 "Reproductions" -> "Reproduction"
2) p.2 l.23 "Atmospheric modes" -> "Such large-scale atmospheric modes"
3) p.2 ll.54-56 "One way to trigger internal variability in GCM simulations is to
perturb the initial conditions of the model, leading to several realizations of weather sequences under identical external forcing which also allow to derive a robust distribution of NAO index values." perhaps reword to, "One way to sample realizations is to perform an initial-value ensemble in which multiple simulations are performed with identical external forcings but perturbed initial conditions. Such an initial-value ensemble would allow a more robust distribution of NAO index values to be sampled."
4) p3. l.78 "until 2099" -> "that extend until 2099"
5) p.4 l.94 "regarding" -> "of"
6) p.5 l.97 At this point I didn't recall what IMS represented. Even though it was defined on the previous page under key question c, I had to stop and go back to find it. It is only used once in this paragraph and not used again for 5 pages. The economy of saving three words is not worth the break in flow here. This is a problem that exists all over the manuscript and it does significant harm to the readability of the paper. I would encourage the authors to greatly reduce the use of such acronyms - particularly when used so intermittently.
7) p.5 ll.100-104 Care should be taken here. Just because a globally integrated quantity from an initial-value ensemble of one model spans a similar range as a multi-model ensemble does not mean they are interchangeable. Different models have different physics packages and so could have very different regional behavior and that fact is potentially masked by a globally integrated diagnostic.
8) p.5 l.118 "Figure 1" -> "In the Appendix, Figure 1"
9) p.5 ll.120-124 Perhaps reword "The GCM... PR sum values" to "Displaying opposite bias to CRCM5, the GCM overestimates (underestimates) mean winter nSAT in the norther (southern) part of the domain, whereas winter PR sum is underestimated in the eastern half of the domain and overestimated on the western side of the Alps. As this study will focus on responses in nSAT and PR induced by the NAO (see Section 2.2.4), aside from regions with particularly high PR sum values, it is found that such NAO responses are generally insensitive to these biases."
10) p.5 ll.127-128 "...dispersion. So the following analyses were not confined to..." -> "...variability. So, in addition to analyses of"
11) p.8 ll.199-200 "As a next step, the monthly difference between driving data and the RCM data was taken for each time step and member" -> "As a next step, time series of the difference between monthly mean driving data and the RCM data was taken for each member"
12) p.8 l.205 "between driving" -> "between monthly mean driving"
13) p.8 ll.208-209 "allows to derive a measure relative to the inter-annual variability of the SLP pattern on a given location. Low RMS∗ values indicate a low error." -> "provides a measure relative to the inter-annual variability of the SLP pattern in a given location. Low RMS∗ values in a particular region indicate a low error and so a good reproduction of the SLP variations in that region of the RCM."
14) p.9 l.228 "IMS" see my point 6)
15) p.9 ll.227-234 This is poorly worded. The authors are mixing ideas of stationary and non-stationary systems to discuss issues related internal variability. For stationary dynamical systems, one can define/identify internal variability by looking a the statistical properties of a time series relative to its long-term time average. For a non-stationary system, one can define/identify internal variability by looking a the statistical properties of an initial-value ensemble of time series relative to its ensemble mean at any specific time. The degrees of freedom that go into each evaluation depends on the length of the time series in the former and the number of ensemble members in the latter.
16) p.10 l.248, l.251 The terms ERA-I data and REF realization are the same I believe. This is very confusing as the equivalence between the two was drawn 12 pages earlier! Why REF? Why not OBS. This would be much more obviously connected to ERA-I. This again falls back to my point 6). Also l.254 I couldn't remember what IMS was again here.
17) p.10 l.260 l.261 My point 6) again. I had to hunt a long time to find where SOIC, MOIC and "n" were defined. The first two weren't even in the text - they were in the caption to Fig. 1! Again, this breaks the flow of the paper and makes it very difficult to read and understand. Are these acronyms really necessary? For "n" perhaps remind the reader what it is when using it here - even a text search of a PDF isn't much help with a single letter!
18) p.10 ll.254-260 "For further ... in Leduc et al. (2019):" In addition to comments 16 and 17 above, I found this description very hard to follow. In its place I offer the following. "The independence of the 50-member ensemble is critical to interpreting the inter-member spread as a proxy for internal variability. In evaluating this, it is important to recall that the 50-member LE was constructed in two parts. First, independent atmosphere/ocean states in 1850 were used to launch 5 historical simulations and integrated forward until 1950. Second, in 1950, each of these 5 ensemble members were used to launch 10 individual simulations by applying a small perturbation to the atmosphere and integrated forward until 2099 thereby producing the 50-member large ensemble. As a consequence, for this study, members between each of the 5 groups of 10 are expected to be highly independent while members within each group of 10 are perfectly correlated in 1950 and progressively increase their independence beyond their 1950 starting point. To evaluate whether the 10 members within each of the 5 groups had become sufficiently independent by the two 30-year periods of interest (1981-200 and 2070-2099), correlations were applied to two groups following Leduc et al. (2019):"
19) Fig.3 Perhaps plot the normal distribution in this figure for reference.
20) p.12 l. 297 "RMS*" -> "RMS* (eq. 2)" it's been many pages since it was discussed. Which brings up an important point. Why not motivate the need for this diagnostic, then discuss the derivation of RMS* at the place where the diagnostic is used - here. This goes for most of what is presented in Section 2 and then is used in later sections of the manuscript. It would make the paper far more readable. The constant back-and-forth looking for things being discussed is quite disruptive. The authors are making it very challenging for themselves by this choice of presentation and they are not doing a good job of meeting this challenge.
21) p.12 ll.298-300. "A value of RMS∗ ≥ 1 indicates that the root-mean squared error between the RCM and driving data is larger than the temporal variability in the driving data. In this case, the large-scale SLP pattern may not be seen as being correctly represented in the RCM data." To make this point clearer, perhaps reword to, "An O(1) value of RMS* would indicate a poor reproduction of the SLP signal in the RCM because the RMS difference between the RCM and driving data SLP is of the same order as the variability of the SLP in the driving data itself. Values of RMS*<<1, on the other hand, would indicate a good reproduction of the SLP signal in the RCM because it suggests that the RCM is tracking the variability in the driving data."
22) p.12 ll.300-302 With 21) above you could now change, "The large-scale SLP pattern over the entire ClimEx domain, which also includes the CEUR, NE, BY and SE domains, is reasonably well represented: with RMS∗ < 1 in most parts of..." to "With this understanding it can be seen that the large-scale SLP pattern over the entire ClimEx domain, which also includes the CEUR, NE, BY and SE domains, is reasonably well represented in most parts of..."
23) p.13 ll.311-313 "nSAT", "PR", "\alpha_1" see my point 6) and 20). I've run out of steam highlighting these and I leave it to the authors to identify and correct the many that remain beyond this point in the paper
24) p.13 l.311 "... the ERA-I data are reproduced in their general properties under current ..." -> "... the ERA-I data are generally reproduced under current ..."
25) Figs. 6-7 One quantity is displayed in (a)-(f) and a different quantity is displayed in (g)-(j). The titles on the plots in (c)-(f) are identical to those in (g)-(j). This is confusing. Also, it would be helpful if the mean (columns 1 and 3 in rows 2 and 3) and sd (columns 2 and 4 in rows 2 and 3) plots were placed beside each other to facilitate comparison of identical quantities between the GCM and RCM. |
- Printer-friendly version
- Supplement
