|Review of the manuscript “Surface temperature response to the major volcanic eruptions in|
multiple reanalysis data sets” submitted by Fujiwara et al. for publication in ACP
The authors apply multi-linear regression techniques to identify responses in near-surface air temperature to three large volcanic eruptions that occurred during the 20th (Agung, El Chichon, Pinatubo) in several reanalysis data-sets. The topic is interesting and certainly very appropriate for ACP. To my understanding there are two major results, a) anomalies after large eruptions are very similarly represented in all analyzed data sets, b) the anomalies are patchy and globally relatively small compared to earlier analyses, e.g. the authors suggest that the maximum response to the Pinatubo eruption is in the order of 0.1 to 0.15 K when averaging over 60S to 60N. While I think that result a) is convincingly presented and useful, the authors failed to convince me that their technique provides a more reliable estimate for the responses than earlier studies. I would suggest publication only after a major revision discussing the results much more carefully in comparison to earlier studies. Furthermore, I think that while the paper is overall very clearly written, most of the figures could be substantially improved to better support the points that the authors would like to make.
Concerning the comparison to earlier studies I appreciate that the authors compare several different analysis methods in appendix A. I think it is a very important conclusion that “differences among the different methods are generally much greater than the differences among different reanalysis data sets”. But I don’t think the appendices provide convincing support for the authors believe that their estimates are “more appropriate because [they] have more thoroughly considered potential confounding factors outside of the volcanic eruptions themselves”. Indeed, with their technique the authors include 11 SST indices for which they make an effort to be orthogonal. However, a general issue is that the volcanic eruptions should have an impact on SSTs themselves and the larger the number of indices in the MLR, the more likely a projection of volcanic signals onto them seems to me. In addition, there is some literature discussing the possibility of an effect of volcanic eruptions on ENSO, which, if it exists, would make it necessary to question the inclusion of ENSO indices in such an MLR at all. The authors admit the problem early in the text, saying that their method “assumes that the zonally-symmetric volcanic aerosol forcing does not project substantially onto strongly asymmetric modes of variability like ENSO”. But they don’t discuss later on, in particular not in the conclusions, how this would affect their results, and how they come to their earlier mentioned “believe” despite these issues. One possibility to make a step into the direction of further evaluating their method would be to leave out the volcanic periods in the MLR. But even with such a test I don’t see that the issue of a clear identification of the volcanic signals can be solved and a fair comparison of different methods can be reached using only reanalysis data sets. So I would rather suggest that the authors openly discuss the existing problems than to state believes. This should start already in the introduction where several earlier studies have been mentioned that have analyzed volcanic signals, but it is not made clear what potential weaknesses of these studies were and how they could be overcome. I also would like to see this discussed in the conclusions: How to make further progress concerning this issue that would go beyond believes.
A further issue I have concerns the presentation of spatial patterns of anomalies after different volcanic eruptions. Of course, one can’t expect the responses to be the same because of the different characteristics of the different eruptions. However, any comparison and possibility to identify common signals is made impossible by the choice to show anomalies for different seasons. I understand that the authors want to present periods with the largest anomalies, but I’d ask them to try analyzing the same seasons and depending on the outcome show these results or just make a statement if there aren’t any similarities.
To be honest, I don’t expect similar patterns because my guess is that even for similar eruption characteristics, patterns of single events may look very different because they are not dominated by the response to volcanoes but internal variability. This could be tested, e.g. by comparing the residuals after the eruptions to residuals outside of volcanic episodes. If indeed responses may not be dominant, I’d also ask the authors to reconsider their use of language, which is unfortunately not consistent. Sometimes they are careful in their wording, but more frequently they are talking about “responses to volcanic eruptions” while in some places they mention that the patterns would include both, responses (with all the uncertainty mentioned above) and internal variability.
My final major point concerns the figures containing spatial patterns, of which I find the isolines very hard to read. E.g., to make the important Fig. 9 easier to read I think there are better ways than just to use green color above some threshold. In general I’d ask the authors to present clearly the information they want to convey to the reader and simply not show what they consider unimportant. When it comes to showing differences between different methods or data sets, it may be more informative to actually show difference plots than absolute fields.
One final minor comment concerns the very last statement in the conclusion section. Of course, one needs to carefully analyze potential signals of SRM, but when it comes to modelling one can easily avoid complications discussed here in the context of the single realization of reality.